The discovery of a visual system: the honeybee 978-1-78924-089-4, 1789240891, 9781789240900, 9781789240917

Table of contents :
Content: The difficult birth of honeybee colour vision --
No way to untie the spell --
Innovation, deep thought and hard work --
The fundamentals of the insect compound eye --
How bees distinguish colours and modulation --
Feature detectors, cues resolution, preferences and coincidences --
Symmetry and asymmetry: signposts in route finding --
Bee vision is not adapted for pattern or shape --
The visual control of flight --
The route to the goal and back again --
What was not mentioned --
What we learned.

Citation preview

The Discovery of a Visual System: the Honeybee

The author with coloured patterns on square targets of standard size.

The Discovery of a Visual System: the Honeybee Light guides, optics, visual cues, optic flow, route finding

Adrian Horridge

The views and opinions expressed in this book are those of the author and do not necessarily reflect the official policy or position of CAB International. CABI is a trading name of CAB International CABI Nosworthy Way Wallingford Oxfordshire OX10 8DE UK Tel: +44 (0)1491 832111 Fax: +44 (0)1491 833508 E-mail: [email protected] Website: www.cabi.org

CABI 745 Atlantic Avenue 8th Floor Boston, MA 02111 USA Tel: +1 (617)682-9015 E-mail: [email protected]

© Adrian Horridge 2019. All rights reserved. No part of this publication may be reproduced in any form or by any means, electronically, mechanically, by photocopying, recording or otherwise, without the prior permission of the copyright owners. A catalogue record for this book is available from the British Library, London, UK. Library of Congress Cataloging-in-Publication Data Names: Horridge, G. Adrian, author. Title: The discovery of a visual system : the honeybee / Adrian Horridge. Description: Boston, MA : CABI, [2019] | Includes bibliographical   references and index. Identifiers: LCCN 2018049622 (print) | LCCN 2018055072 (ebook) |   ISBN 9781789240900 (ePDF) | ISBN 9781789240917 (ePub) |   ISBN 9781789240894 (hbk : alk. paper) Subjects: LCSH: Honeybee--Physiology. | Vision--Research. Classification: LCC QL568.A6 (ebook) | LCC QL568.A6 H57 2019 (print)   | DDC 595.79/9--dc23 LC record available at https://lccn.loc.gov/2018049622 ISBN-13: 978 1 78924 089 4 (hardback) 978 1 78924 090 0 (ePDF) 978 1 78924 091 7 (ePub) Commissioning Editor: Ward Cooper Editorial Assistant: Emma McCann Production Editor: Shankari Wilford Typeset by SPi, Pondicherry, India Printed and bound in the UK by Severn, Gloucester

Contents

About the author

vii

Preface

ix

Introduction

xi

  1  The Difficult Birth of Honeybee Colour Vision

1

  2  No Way to Untie the Spell

12

  3  Innovation, Deep Thought and Hard Work

28

  4  The Fundamentals of the Insect Compound Eye

44

  5  How Bees Distinguish Colours and Modulation

89

  6  Feature Detectors, Cues, Resolution, Preferences and Coincidences

106

  7  Symmetry and Asymmetry: Signposts in Route Finding

135

  8  Bee Vision is Not Adapted for Pattern or Shape

150

  9  The Visual Control of Flight

173

10  The Route to the Goal and Back Again

197

11  What Was Not Mentioned

222

12  What We Learned

231

Appendix Training and Testing Bees

253

Author Index

259

Subject Index

263

v

About the Author

Born in 1927, after wartime teenage years Adrian Horridge was admitted as a scholar of St John’s College, Cambridge in 1945, and passed the Natural Sciences Tripos with first class. He joined the Department of Zoology, completed a PhD on electrophysiology of primitive nervous systems, won a Senior Award from the Commissioners from the Great Exhibition of 1851, and a Research Fellowship at St John’s College. In 1956, he became a lecturer in Zoology at University of St Andrews, Scotland. Most of his early work was on marine animals found there and at Millport (west coast), Naples, Plymouth and on reefs of the Red Sea. In 1959, he was a Fellow of the Center for Advanced Studies in the Behavioral Sciences at Palo Alto, California, where Ted Bullock and Adrian wrote most of The Structure and Function of the Nervous Systems Of Invertebrates (1965). He returned as Director of the Gatty Marine Laboratory at St Andrews in 1961. After a period of intense research on invertebrate nervous systems, mainly on eyes of insects and Crustacea, he was elected to the Royal Society in 1969. In the same year, he moved to the Australian National University as one of four founder professors of the new graduate Research School of Biological Sciences. There, his research and training of others focused on insect vision. He supervised 51 PhD students altogether, 37 of whom became professors in 14 different countries. Three of his students, Meyer-Rochow, Simmons and Warrant, won three separate Ignoble Prizes: this must be unique. Eight of his students or post-docs have been elected to the Royal Society, and one to the Royal Society of Canada, four of them retired as Cambridge professors. Up to 2018, Adrian published 260 papers on neurobiology. This output, together with the two volumes with Ted Bullock, Structure and Function of the Nervous Systems of the Invertebrates (Bullock and Horridge, 1965), followed by another, Interneurons (Horridge, 1968), makes him one of the fathers of neurobiology of invertebrates. The books are still prized. vii

viii

About the Author

Adrian has also researched and written three books, several monographs and 12 papers on the sail trading fleets, fishing vessels and outrigger canoes of Indonesia and their history, functions, rigs and structures. These are the major reference source for Indonesian maritime ethnography and have recently become expensive collector’s items. In 2016, a version was published in the local language, Bahasa Indonesia. In sabbatical years, Adrian has twice been an Overseas Fellow of Churchill College Cambridge, and a Colonial Fellow of Balliol College, Oxford. He made many research visits to marine laboratories, and has dived on coral in all parts of the world. His son Mark and grandson Tor Lattimore are professors of mathematics. Remarkable practical innovations emerged unexpectedly from the research on insect vision in Canberra. By studying the light path in flies’ eyes, Allan Snyder (1972) realized how light signals entered most efficiently into light guides and travel long distances along them. This new understanding led directly to the World Wide Web of optical fibre. In 1985, Adrian realized that range to nearby objects was measured by the relative motion induced in the compound eyes by walking or flying insects (Horridge, 1987). With this measurement of range to nearby contrasting edges, insects have no need to distinguish every separate object. His group showed that bees use this optic flow to give them a third dimension (Srinivasan et al., 1989a, b). They carried it through to guidance principles for vehicles and put insect-style vision on drone aeroplanes and helicopters with a computer on board. Both Snyder and Srinivasan won the Australian Prime Minister’s Prize for Science and were elected to the Royal Society for these discoveries. At the age of 91, Adrian is still researching the vision of the honeybee as a guide to vision of arthropods in general, and as a model for reverse engineering of simple visual systems for robot tasks. Adrian Horridge was married for 59 years to Audrey (1930–2013), née Lightburne, a Cambridge and Oxford graduate. In her final job, Audrey was in charge of welfare of all 36,000 government-supported overseas students, with an office in every university in Australia. His son, Mark, also a Johnian, is a professor of macroeconomics and its computing techniques, based in Melbourne. Daughter Alison is an artist who home-schooled her family; Naomi juggles motherhood with writing, lecturing and museum curating. His youngest daughter, Meret, also a Johnian, is a published poet and teaches molecular biology in Zürich. Rebecca is a roaming radio journalist, specializing on environmental and social issues in Australia. His granddaughter, Dr Finnian Lattimore works in computer technology in the Australian National University. His grandson Tor Lattimore is a professor of ­analyticals, working on Deep Mind at Google, London. Two great-grandchildren look quite promising. Adrian Horridge Adrian-Horridge.org

Preface

When I retired in 1992, a research topic with little expense was essential, so for 15 years every spring and autumn I continued training bees in my old ‘shed of discovery’ in the grounds of the Australian National University (ANU). I would start work at 8 a.m. for 2 h training them, then prepare tests. My assistant would turn up at 10 a.m. and, between regular periods of continued training we would test them. We took turns to have lunch, shut shop at 4 p.m., 5 days a week during October to December, and March to May. The other months were either too hot or too cold for the bees. Usually, we started with a new group of bees each Monday morning and finished the experiment on Friday afternoon. I fed the bees early each morning at the weekend so that they kept their habit of coming for sugar. My aim was to train bees to every kind of display and test them to reveal exactly what they had detected and learned. My retirement saved a lot of money for ANU, but funding was never a problem for me. First, to pay for a part-time assistant, I had a grant as an Emeritus Fellow of ANU, then a series of grants from the Royal Society that came by a remarkable route. I asked the Secretary of the Royal Society if there were funds to hire an assistant. The reply was initially unfavourable, because UK government funds were for work based in the UK. Then it was suggested that I should apply to a fund ‘for the assistance of fellows’ that had been established 200 years ago to assist fellows who fell into poverty. The fund had accumulated ‘a large sum’. I certainly needed assistance, and my case fitted the rules, so I applied and was asked to reveal my income. By good fortune, all my retirement funds were in a portfolio of mining companies, which paid few dividends, and my wife Audrey had a separate income. So, I sent them a copy of my tax return, which indeed revealed a very small income, and they sent me £10,000 to pay for a part-time assistant. There was no shortage of interesting school leavers and backpackers, who had to be not allergic to bees, and who had never expected to learn how to train them. All funds that came from overseas to support research in Australia were subject to 10% tax unless the product was re-exported. So, I was obliged to publish the results in overseas journals. The grant was continued for a time, but one day a letter arrived to say that all the available funds had been spent on the refurbishment of the staircase in Carlton House. Research prospered and eventually ANU E Press published my book on black/white vision of the honeybee in 2009, while I was fully occupied caring for Audrey, my wife. It lacked chapters about colour because the new results were not yet comprehensible, and an attack of the shingles forced me to send the book to press although unfinished. Also, on the question of colour vision of bees, I had not been sufficiently provoked. To write anything, I needed first ix

x Preface

to feel jubilant about a discovery, furious about obsolete textbooks or persistent error, and comfortable with a new product. All that took time to appear, which it certainly did. After Audrey died in January 2013, work started in earnest on the colour vision of the bee, with the experiments moved to my garden. Others had terminated their efforts to understand bee vision of colour and pattern. My new results revealed a totally new type of colour vision. Trichromatic vision of hue, as in humans, was soon ruled out. There had to be another book because the topic was so rich in history, power struggles between professors, error and its concealment, refusals to acknowledge new facts, ignorance of the literature, interesting behaviour of editors and disgraceful duplicity of inept referees. Besides, the use of coloured display papers that were equiluminant to either the green or the blue receptors of the bee had to be fully investigated. In the end, the scientific advances were few and simple but powerful, but they were accompanied by a fascinating circus of blind alleys, other researchers trapped by their own earlier publications, periods of intense effort to find a new idea, and the joy of discovery. The stimulus lay in the variety show of the journey; while the basic facts of the new paradigm at the arrival were relatively obvious once discovered. I would like to thank the Fellows of St John’s College, Cambridge, who initiated my career with a scholarship in 1945, and later a Research Fellowship, and also the numerous distinguished staff of the Department of Zoology, who nourished me for 10 years with the aid of the Commissioners for the Great Exhibition of 1851, and later Churchill College, where I was also a Fellow. I am indebted to many historians and philosophers of science, who showed me the inadequate foundations of sciences that rely on intuitive guesswork. I especially thank the late Professor Ubisch, Director of the Zoological Institute in Münster from 1927 to 1933 when he was dismissed from his job. In the early 1960s, I purchased his collection of reprints that contained all the early German papers on bees. I am also indebted to the University of St Andrews, Scotland and the Australian National University, for a lifetime of freedom in research with working conditions that allowed time to read the literature thoughtfully. The new account of vision of colour was delivered in plenary lectures to the International Society of Invertebrate Neurobiology at Tihany, Hungary, in 2012, at the 7th European Congress of Apidology at Cluj, Romania, and the 25th International Entomological Congress in Orlando, Florida in 2016. I thank them all for the invitations. I have drawn freely from the illustrations used there and in my recent papers. I am also indebted to numerous helpers who have spent hours counting choices of trained bees; to secretaries and assistants, especially Ljerka Marcelja, who taught many students how to use microelectrodes and made recordings for our extensive electrophysiological survey of insect compound eyes. Thanks to the cartoonist, Stuart McMillen, for the sketches in Fig. 1.1. I also thank students and associates with their own skills, discoveries and words of wisdom; and especially my family who have endured years of neglect while I was training research students and testing bees.

Introduction

As a start, let me introduce an idea. The topic of bee vision of shapes, colours and visual sensing of the environment, together with the so-called superposition, or optical summation, in the eyes of nocturnal insects, somehow remained in a cataleptic state for about a century. Catalepsy is a trance state with suspension of consciousness. The story swirls around the cities of Münich in south-east Germany, and Vienna in neighbouring Austria. The Austro-Hungarian Empire in the late 19th century was a paradise for the privileged and the bourgeoisie (Coen, 2002). Our central character, Karl von Frisch, had two uncles, Karl and Sigmund Exner, both professors at the University of Vienna, and he had related family, Oscar and Richard Hertwig, who were professors of biology. These five are dominant actors in my early scenes, together with Carl von Hess, who arrived in 1910 from Würzburg to be Professor of Ophthalmology in Münich, where Richard Hertwig was Professor of Zoology. Hess was definitely an outsider. The Exner family had produced ten professors, and for decades had been advisors in education and administration to the Hapsburgs. Notable European academics were their students, including Schroedinger, Boltzman, Mach and Sigmund Freud. By the end of the century, however, the Exners became repeatedly involved in quarrels with German Nationalists, Christian Socialists and rising left-wing groups as the economy declined. Throughout this period, and still today, German universities were characterized by a tribal allegiance to the professor, who was a male autocratic public servant in control of every aspect of his department, appointing staff and fixing the programme. In general, this has changed little. In the period between 1886 and 1891, Sigmund Exner studied the optics of the cornea in several insects and published a picture of an erect image formed within the compound eye of a firefly. For a century this image has been accepted as a valid component of insect vision, suggesting that insects see images of things. Unfortunately, the image was in the wrong place, and fireflies have a light guide in each ommatidium, but no lens cylinders. Exner’s young nephew, Karl von Frisch described discrimination of colours and shapes of flowers by honeybees in a long paper that became the scientific validation of the common belief that bees see colours. This was contradicted by von Hess (1918) with a better experimental design, but von Hess was ignored. Von Frisch went on to discover the bee compass formed by scattered ultraviolet light in the sky, and the dances of returning forager bees. He won the Nobel Prize, and became a national hero. Accounts of the erect image and

xi

xii Introduction

the visual discriminations of colour and pattern had another explanation. They were reproduced in every textbook but the errors were not mentioned. Why the revision was so delayed is an interesting question. There was no better theory, and the work of von Hess was suppressed. After the war, there was silence about the frightful fate of the gypsies, insane, crippled, ethnic minorities and racial hybrids, and denial among those near to the crimes. At least four of my students were children of that generation, and many German visitors came to Canberra, but none would speak of it. Not until the 1990s was it possible to open the topic freely. Even then, referees were reluctant to consider my partial revisions of the orthodox story or discuss the work done in the period between von Frisch (1914) until the slow revival in Germany after 1945. In general, researchers in England and the USA had no idea what had happened. Referees became quite hostile and editors wary of publishing anything relevant. Even as I write, no publishers are prepared to commission a student textbook that includes the new developments and their history. Naturally, any revision would be counter-intuitive because bees really do appear to see things and colours. Revision was delayed, disputed and denied, by a combination of lack of new techniques, stubborn referees, misplaced cover-up, and what looks like a bad case of catalepsy. The topic of this book is the much-needed revision that puts insect vision as exemplified by the bee, and dim-light vision in other insects, back on track. The new paradigm for visual inputs of the compound eye is one tonic colour blue and two derivatives that are rates of change in blue and green receptor channels. I hope that the novel prospect will be of interest to engineers of robot vision. The rapid rise, persistence for a century, and eventual fate of Exner and von Frisch may also interest historians of science. It was inevitable that a book in the early part of the 21st century would upset the topics of ‘insect vision’ and ‘compound eye’ because electron microscopy and electrophysiology had surged ahead un-noticed while several topics had been held back by conservative forces. These topics are: (i) the legacy of Exner’s superposition image in the firefly that began in 1886; (ii) the supposed bee vision of colour and shapes of flowers dating from 1914; (iii) the action of light guides in almost every insect compound eye; and (iv) the measurement of range and distance flown by foraging bees. These are all major topics where mechanisms and principles can now be rewritten. In addition, there were descriptions of behaviour concerned with scanning behaviour, odour seeking, foraging, discrimination of place, route finding, directional signs on routes, and symmetry of flowers, that have previously been indifferently understood. You will meet them all within. Beside this saga of contending personalities, this is in fact a serious book that attacks the detail and gist of the compound eye of a typical day-flying insect. One of my strong considerations was to include the illustrations of the various eye types, and to present the actual experiments that demonstrated the inputs used by the trained bees and supported the main findings. To be comfortable at every point in the text, the reader should compare text with illustration and legend. These are the facts that displaced errors. I apologize for the incompleteness of my book, which is essentially an account of the need for change and recent work. You will still need another source for entomology of vision, classical histological and physiological details of the eye and optic lobes, for which the fly is better known than the bee (Strausfeld, 2012; Wolff and Strausfeld, 2016). Books and papers on bee eyes and insect vision are abundant, but all need revision. The reader will soon realize that throughout our research effort on insect vision from 1962 to 1992, everyone in my group believed that bees see patterns and hue of colours. At one point, three future professors put together a theory predicting the number of pictures that a fly could discriminate from each other (Snyder et al., 1977), but they no longer cite it. Others showed that every colour could be discriminated from every other colour (Backhaus et al., 1987). An orthodox review by three distinguished professors (Kelber et al., 2003) made no mention of the numerous unexplained anomalies discovered by Giurfa or



Introduction xiii

myself. Experiments on bee vision are still used to support the idea of cognition in bees: for example, by converting my term ‘learning by trial and error’ to ‘unhappiness’ (Menzel, 2012). Of course, like the rest of my colleagues, I believed for decades that bees see colours and patterns, but nowadays simple cues provide better explanations. The upsets of orthodox views was entirely the result of new experiments after a great deal of thought and experimental search. The bees themselves showed the way forward at every step. Already, two major innovations of practical importance have emerged directly from our studies of insect vision. In 1972, from the optics of fly eye, Allan Snyder discovered how the light gets into the light guides. The details were copied into the design of the fibres that carry light pulses for extreme distances in the World Wide Web (Chapter 4, this volume). In 1989, Srinivasan, Lehrer, Zhang and I discovered that insects detect a panorama of range by signals created by their own motion, not of objects (Srinivasan et al., 1989a, b), and Srinivasan later completed the installation of the resulting applications into mobile vehicles that fly with a computer on board (Chapter 9, this volume). The future will bring another important development: reverse engineering of insect vision, simply because insects are examples of extremely sophisticated and well-adapted visual mechanisms at exactly the right level of complexity to be copied into silicon for practical applications, quite unlike our own vision. Moreover, many neurons in an insect, perhaps all, are individually identifiable, and the wiring diagram is also constant, so that we can repeatedly return to precisely the same circuit, neuron or synaptic connection, with a variety of different techniques. Biologists, who wonder what other lower animals see, and how the panorama is detected, can now take a cue from new work on the bee, which is a representative insect, although sexless and herbivorous. Many other insects have specializations for courtship and capture of live food with visual adaptations that include extra photoreceptor types. As yet, we have no firm circuitry or formal connections in the bee beyond cues and their coincidences. The ways that preferences are established, modified and remembered remain a mystery. We have no idea how bees reach decisions. We have not identified the points where the layout of the two-dimensional image on the eye disappears into responses of line-labelled nerve cells, where odour signals are combined with vision, where traces of the image finally disappear in decisions such as ‘avoid’ and ‘attract’, or where meaningful visual cues are stored in memory. We do not even know whether memory is located at every synapse where the signal passes, or in special localized centres, or in both. A similar state of ignorance applies to every other animal brain, but the bee is one animal where progress is possible, perhaps with the aid of large tropical species. My conclusion at the end of the investigative scientific journalism was that the delays and persistence of error that I describe were little to do with National Socialism, anti-Semitism or the German character. My topic is far from the well-known high standards of German science, engineering and medical technology. Those efforts are subject to checks of quality and efficiency, but I use ‘what insects see’ as a relatively harmless example of an upset in science. There have been many others (Gratzer, 2000). In the case of bee colour vision or Exner’s image, no alternative paradigm was available, and the orthodox would not look for one. I first concluded that the causes were the lack of better ideas and experimental techniques, the interaction between professorial powers, fear of the loss of reputation and financial support, and too much respect for the discoveries of the previous generation. The giants’ shoulders were not for building on. On a different level, trust in reliable science (Nurse, 2015) gives way to vested interests. Location of power and trust in delegation are now the problems. Loss of trust follows on the heels of increased state control. There is universal inertia against innovation because it is expensive and requires thought, hard work and persistence, and then disrupts set ways. Forces for conformity are now a universal and growing danger that should be resisted. But worst of all, innovation implies an escape from bureaucratic control. It is

xiv Introduction

nothing to do with trust, as Nurse (2015) suggests. Trust can be misused, and in modern times, state power and control of the cash by non-scientific politicians and bureaucrats tends to hinder rather than help innovation and fundamental research. Nowadays it would be difficult to find funds for fundamental animal biology such as we had in St Andrews and Canberra. We were extremely fortunate to be able to seize the moment. The projects paid off handsomely, if you count only fall-out from the optics of light guides (Chapter 4, this volume), the $10 million from Fujitsu Computer Company, and much more from the USA, in return for our know-how about the piloting of drone aircraft (Chapter 9, this volume), and many large grants, not to mention, here and there on the way, eight Fellowships of the Royal Society, one of the Royal Society of Canada, and at least four of the Australian Academy of Science, two Prime Minister’s Prizes for Science and 37 professor’s chairs in 14 countries. Despite all the wobbles, shilly-shallying and necessary apostasy, this is a serious ­account of revision of the scientific backbone of insect vision. I hope it will inspire you to repeat the experiments and advance further. A hundred years of research has gone down river, gone under the bridge, and must now be rewritten. The time has come to rouse from catalepsy and write new texts and lectures.

References Backhaus, W., Menzel, R. and Kreißl, S. (1987) Multidimensional scaling of color similarity in bees. Biological Cybernetics 56, 293–304. Bullock, T.H. and Horridge, G.A. (1965) Structure and Function in the Nervous Systems of Invertebrates. Two volumes. Freeman & Co., San Francisco, 1719 pp. Coen, D.R. (2002) Vienna in the Age of Uncertainty: Science, Liberalism, and Private Life. University of Chicago Press, Chicago. Exner, S. (1886) Cylinder welche optische Bilder entwerfen. Pflügers Archiv gesampte Physiologie 38, 274–290. Exner, S. (1891) Die Physiologie der facettirten Augen von Krebsen und Insekten. Leipzig, Germany, Franz Deuticke. Translated by Hardie, R.C. (1988) The Physiology of the Compound Eyes of Insects and Crustaceans. Springer-Verlag, Berlin. Gratzer, W.B. (2000) The Undergrowth of Science: Delusion, Self-deception and Human Frailty. Oxford University Press, Oxford. Horridge, G.A. (1968) Interneurons. Their Origin, Action, Specificity, Growth, and Plasticity. Freeman & Co., San Francisco, 436 pp. Horridge, G.A. (1987) The evolution of visual processing and the construction of seeing systems. Proceedings of the Royal Society of London B 220, 279–292. Kelber, A., Vorobyev, M. and Osorio, D. (2003) Animal colour vision – behavioural tests and physiological concepts. Biological Reviews 78, 81–118. Menzel, R. (2012) The honeybee as a model for understanding the basis of cognition. Nature Reviews Neuroscience 13, 758–768. Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222. Snyder, A.W. (1972) Coupled mode theory for optical fibres. Journal of the Optical Society of America 62, 1267–1277. Snyder, A.W., Laughlin, S.B. and Stavenga, D.G. (1977) Information capacity of eyes. Vision Research 17, 1163–1175. Srinivasan, M.V., Lehrer, M., Zhang, S.W. and Horridge, G.A. (1989a) Motion cues provide the bee’s visual world with a third dimension. Nature, London 332, 356–357. Srinivasan, M.V., Lehrer, M., Zhang, S.W. and Horridge, G.A. (1989b) How honeybees measure their distance from objects of unknown size. Journal of Comparative Physiology A 165, 605–613. Strausfeld, N.J. (2012) Arthropod Brains: Evolution, Functional Elegance and Historical Significance. Belknap Press, Cambridge, Massachusetts, 848 pp.

Introduction xv

von Frisch, K. (1914) Der Farbensinn und Formensinn der Bienen. Zoologische Jahrbücher. Abteilung für allgemeine Zoologie und Physiologie der Tiere 35, 1–188. von Hess, C. (1918) Beiträge zur Frage nach einem Farbensinne bei Biene. Archiv für gesampte Physiologie 170, 337–366. Wolff, G. and Strausfeld, N.J. (2016) The insect brain: a commentated primer. In: Schmidt-Rhaesa, A., Harzsch, S. and Purschke, G. (eds) Structure and Evolution of Invertebrate Nervous Systems. Oxford University Press, Oxford, pp. 597–639.

Chapter 1 The Difficult Birth of Honeybee Colour Vision

Scientists . . . need to explain what is being done and why, ensuring that conflicts of interest are revealed, and that it is clear what knowledge is secure and what is not. [my emphasis] (Sir Paul Nurse, Anniversary Address, 2015)

We expect to see colours everywhere, and use them to distinguish, for example, ripe fruit. Not to see colours as constant would be very worrying, but we depend also on a peculiar property of human colour vision. White or coloured surfaces or lamps do not change colour with changes of intensity over many orders of magnitude: for example, from moonlight to bright sunlight (a factor of 107). A neutral filter is said to be neutral because it absorbs some of the light but leaves constant the composition of the light, as humans perceive it. Bees have excellent eyes, as anyone can infer when watching them forage. Throughout the 19th century, a number of competent scientists used coloured and neutral grey filters to test whether bees had colour vision (Forel, 1908; von Hess, 1912). They were aware that people who were totally colour-blind could distinguish between two colours by their relative brightness. They used the attraction of insects to light, called phototaxis, and found that bees walking out of a dark box preferred blue when the test colours ­appeared

equally bright, but the bees’ preference could be shifted away from blue when other colours were simply made brighter. Bees obviously confused colours with brightness in a way that was no different from colour-­ blind people. This seemed a straightforward demonstration that bees did not have true colour vision. A young boy, John Lubbock (later Sir John, later Baron Avebury, MP for Westminster), who lived next door to Charles Darwin, was urged by the old scientist to observe ­nature and experiment, when he left Eton School to join his father’s London bank. From 1874 onwards, John published his ­observations on ants, bees and wasps. He found that bees have an order for spontaneous preference: blue, then white, yellow, green and orange, while red was generally the least preferred. At the time, no one recognized that, apart from yellow, this was a rough scale of diminishing content of blue. He found that hungry bees could learn which of several colours indicated honey (Lubbock, 1881). He understood that observing bees on flowers told one very little. To discover what bees actually detected, it was essential to train them on a target, or discriminate between two targets, then test them. The essential technique was to move the colour and the reward with it while the training was in progress, so that the bees

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)

1

2

Chapter 1

­ ecame accustomed to search for the colour b in several places. Then, in subsequent tests with other colours, they made an effort to search; in fact, they were trained to search. For the first time, it was noticed that they took a much longer time to learn, because they could no longer use the fixed local landmarks or odours to head straight to the reward. Lubbock’s popular books reinforced the view that bees distinguish colours like humans do, but that was not the academic opinion at the time. In 1912, Carl von Hess, Professor of Ophthalmology at Münich University and Director of the Münich Eye Clinic, published a well-researched book on the comparative physiology of vision. As a result, he was regarded as a sound authority on the whole topic. He had newly arrived from Würzburg, a powerful professor at the height of his powers, which meant a lot in a German university at that time. For some years, at the Naples Marine Laboratory, he had collected data on a variety of animals, including fish. There his findings had been challenged by a young assistant called Karl von Frisch. Hess had shown that several species were colour-blind in the phototaxis response. Octopus and other cephalopods reacted similarly in his apparatus. They went to the brightest place irrespective of colour. In his book, as a true scientist, Hess began his account of honeybee vision with his own phototaxis experiments in which freely flying or walking bees could choose between two compartments with coloured lamps of controlled brightness. With colours of similar intensity, he validated the earlier demonstrations that bees in a dark box preferred blue, but on increasing the brightness they could change their preference to another colour. Later, Hess showed that in the escape response bees are attracted to ultraviolet (UV), which was a signpost to the escape route to the sky. Later still, Menzel and Greggers (1985) showed that when bees move out of darkness, they detect the total photon capture including green receptors. Therefore, total summation of receptors is possible, as well as specific attraction to UV, in the escape response and in the righting response, which is a directional

r­esponse to UV that keeps them in level flight. This shows how easily we may be confused, even before we have reached the foraging behaviour. In his book, Hess discussed much of the vast amount of information collected in the 19th century about visits of bees to flowers, and questioned the universal belief that they distinguished flower colours. All observations were about successes of the bees, so any reasonable theory of how they succeeded would have fitted the evidence. In all this mass of data on foraging, even by John Lubbock, there were no studies where colour and brightness were separately tested with foraging bees. Hess, however, found more secure evidence, which modern authors usually fail to quote. Between 1885 and 1906, in a long series of experiments with coloured artificial flowers, Felix Plateau, Professor of Z ­ oology at Ghent University and a son of the famous mathematician, found that shapes or colours of artificial flowers could be changed with little effect, and were not critical for the bees to find the reward. Plateau concluded that bees used odour, not colour, as the cue (Plateau, 1885–1899; cited in Forel, 1908). Auguste Forel, a medical professor and entomologist at Zürich, heavily criticized Plateau. Forel found that odours or vision were not essential guides, but the bees used surrounding landmarks that included flowers. The odours, colours and shapes of flowers were cues only in so far as they were consistent parts of the scene. Forel used Professor Plateau’s own data as evidence against his conclusions. There is no better way to demonstrate the weakness of an intuitive leap from the safe ground of data to an apparently obvious but incorrect conclusion. After 50 pages of fierce criticisms, Forel a­ ccepted Plateau’s data but not the false ­conclusions. Hess must be given credit for quoting in his book a suggestion by Forel that bees should be trained to go to a colour, and then tested with that colour versus other colours and against various shades of grey. The criterion was still based on performance and inspired by human vision. In fact, whether or not bees could distinguish colours from grey would not reveal much about bee



The Difficult Birth of Honeybee Colour Vision

­ ision; because they could not detect grey. v However, this experiment had to wait. In 1898, Albrecht Bethe, Professor of Physiology at the University of Strasbourg, showed that bees apparently did not recognize even familiar objects such as their hive, or a tree, as separate things. In fact, for centuries beekeepers had known that the bees did not recognize their own hive if it was moved only a metre sideways, whatever its colour. The fact that these distinguished professors experimented and argued for years should add some support for their main conclusion that the bees did not see the shapes or colours of flowers in the way that humans do. Hess also reviewed this research and concluded in his book that essential evidence for colour vision of bees was lacking. Later reviewers, if they mentioned Hess at all, said incorrectly that he referred only to the attraction to coloured lights. We now know that bees are not interested in the shapes or colours of flowers or of any other thing. They learn a few landmarks that bring them to the place where they find nectar or sugar solution. If the experiment involved a change in flower colour or shape, they scarcely noticed, as Plateau found, but if their familiar landmarks were moved or hidden, they went away and looked elsewhere, making it difficult to discover what they had learned. Too late to be in the book by Hess, Lovell (1910) showed that bees discriminated between colours irrespective of the intensity of illumination, and they could also learn to avoid a particular colour. Also in the USA, a remarkable Chicago schoolmaster, Charles Turner (1910) displayed a colour or pattern on the outside of a small box with a reward of sugar inside, and a different colour or pattern on a similar box with no reward. He changed the positions of the boxes at intervals to make the bees examine them irrespective of the exact place. The bees learned to look for the boxes and ignore the local landmarks. This experimental strategy made possible all the subsequent discoveries about bee vision of pattern and colour. On the other hand, bees rewarded at a fixed place learned the local landmarks

3

after a single reward. On their return they never made an error, but they did not reveal what they had learned because, when they were given unfamiliar tests, they went away or simply started to learn the landmarks afresh. Bees had to be trained to search before they could be tested with a variety of unfamiliar targets. In his book, Hess did not distinguish clearly between results from foraging bees and those from bees in the phototactic response. At the time, there was no clear understanding that a bee could distinguish colours in one kind of behaviour but be colour-blind in another situation. This was the missing thought that could have helped to distinguish different strands in the data. Karl von Frisch (1886–1982) was from an intellectual family based in Vienna in the Austro-Hungarian Empire. His maternal grandparents were Jewish converts to Catholicism. This generation of professors, Bethe, Chun, Anton Dohrn, Eimer, the Exners, Goldschmidt, Haeckel, von Uexküll, and Oscar and Richard Hertwig, left a huge legacy of wonderfully accurate descriptions of the anatomy and physiology of invertebrates, often based on research in Naples, bringing Mediterranean fauna into European classical zoology. Young Karl was appointed by a relative, Professor Richard Hertwig, as an assistant in the Zoology Department in the University of Münich in 1910 at the age of 24. Like Hess, he had investigated the colour vision of fish at the Naples Marine Laboratory, but he trained fish to come to a coloured target and then, as in the test for defects in human colour vision, he tested them with various shades of grey papers, finding that they distinguished the colour. Also at Naples, Hess found by the phototaxis method that they were colour-blind (not necessarily the same species). At the time, no one guessed that two different kinds of vision of colour could exist in one animal. From that time on, they were fierce antagonists: Frisch even visited Naples in 1913 just to prove that cephalopods have colour vision, and contradict von Hess, but he failed in that effort, and never mentioned it again (Dröscher, 2016).

4

Chapter 1

The conclusion by Hess, that bees do not have colour vision like humans, obviously stimulated Frisch to disprove the senior professor, a strategy that gives a young man a good start. He was able to use the family holiday chalet at the idyllic village of Brunnwinkl in the Austrian Alps for his own work, training bees and fish, then testing their senses. Several members of his family, including his uncles Sigmund (Physiology) and Karl Exner (Physics), both professors at Vienna, helped as his assistants there. Working through three successive summers, he was able to collect a vast amount of data by counting arrivals of trained bees at black or grey feeding tables. Almost certainly he knew of the work on colour vision of bees by Turner and by Lovell, and cited the book by Lubbock. He quickly adapted the techniques used by Hess and Turner (with a reference to them) and started a long series of tests on trained bees with his own unique method developed for fish, which for some strange reason, has never been repeated, even by his own students.

(A)

Von Frisch trained bees for several days to come to a single coloured paper laid flat on a table in sunlight and then tested with it placed on a panel of 15 grey papers (Fig. 1.1B). It was a crucial choice of experimental arrangement. If Frisch had trained bees to discriminate between two colours and then tested with all colours and grey levels separately, he would have immediately revealed a lack of colour vision of the human type. In 1914, it seemed obvious that the test for colour vision with a colour versus all grey levels would prove or disprove colour vision, but that was a test for defects in the colour vision of man. It was just bad luck that the bees distinguished green contrasts at edges and measured the average blue content of areas of grey. From the start, Frisch was absolutely convinced that bees trained on a certain colour distinguish the training colour from other colours, as research of others had shown: Dass sich die Insekten an gewisse Farben gewöhnen, auf sie ‘dressieren’ lassen und sie von anderen farben zu unterscheiden

(B)

6 cm

15 cm

Fig. 1.1.  Test patterns. (A) Hess trained bees on yellow or blue squares and tested with the same, or a succession of colours. (B) Von Frisch trained bees on a single coloured paper, and tested on an array of grey levels with the colour added.



The Difficult Birth of Honeybee Colour Vision

vermögen, darüber kann nach den Versuchen von Lubbock, Forel, Lovell, etc., kein Zweifel herrschen. [After the works of Lubbock, etc., there is no doubt that insects can distinguish colours from other colours.] (von Frisch, 1914)

Actually, he quoted earlier authors knowing that they had not done the tests with grey levels that he considered essential. He argued that if bees are colour-blind there is always at least one shade of grey that they confuse with any particular colour. So, if bees are trained to go to a coloured paper, they should be able to find it among a selection of similar papers of all shades of grey. In the first summer, 1912, Frisch started with two rewarded yellow papers (15 × 20 cm wide) laid flat on a black table in sunlight. Marked bees collected the sugar solution or honey, then left when it was all gone, but returned very quickly to exactly the same place when the reward was renewed. When the yellow paper was placed on an array of 15 grey papers, trained bees had no difficulty in flying directly to yellow, even though the sugar solution was odourless He regularly took clean papers and dishes, shuffled the arrangement of the grey papers, and even tried to deflect them with the odour of honey, but bees are attracted to honey only when they have been alerted and are searching for it. He noted that bees were attracted to the edges of yellow paper but he never made use of this observation. With blue paper, it was even easier to train the bees. In his final and most successful method, Frisch trained for up to 5 days with the rewarded colour under a glass cover (to prevent odour detection) with a reward of sugar solution in a small watch glass placed over the rewarded colour. In the tests, he had 15 different unrewarded papers (15 × 15 cm) selected from a grey scale from white to black. He made a four-by-four display of papers in the shape of a square, laid flat on a black table and covered by a glass plate. He, then tested the batch of trained bees with the colour placed on the scale of 15 grey levels alone or with grey levels mixed with various colours, to test whether the training colour could be discriminated from all grey

5

levels. All the papers were from commercially available series that were accurately reproduced. Bees could certainly be trained to recognize rewarded yellow or blue papers when placed on the array of 15 grey papers, but there were interesting unexplained anomalies. The most serious was that bees trained on mid-grey and tested on the whole grey series landed indiscriminately, as if they could not distinguish the grey level ­except at the extreme ends of the scale; let me repeat, trained on mid-grey, they could not return to land on mid-grey (Fig. 1.2A). Frisch had no explanation; he gave up training on the grey level series and turned to training on one colour mixed with various grey levels. When trained on blue-green (Fig. 1.2B), they landed on number 6 in the grey series twice as often as on the blue-green paper (650 times compared with 306). Trained on green (Fig. 1.2C), they landed on black or mid-greys. Frisch also worked through a range of 14 colours from the standard Hering series plus two of his own, grass green and chlorophyll, and published 124 tables of data (Fig. 1.2). He found that yellows and blues at each end of the series were easily distinguished from grey levels but was clearly concerned about green because he used five different types of green in attempts to demonstrate good discrimination. Bees trained on grass green preferred yellow and red (Fig. 1.2D) but he had no explanation. He repeated the conclusion that ‘blue-green was perceived by the bees like medium grey, and they could not discriminate it from other colours’: Es wurde schon erwähnt, dass die auf blaugrüne Papiere (No 10 und No 11) dressierten Bienen die Dressurfarbe aus der Grauserie nicht herausfanden. Auch wenn ihnen die Farbenserie vorgelegt wurde, schwärmten sie gänzlich ziellos über der Papieren her. [Bees trained on blue-green paper cannot find the training colour in the grey series. Also when tested with the colour series, they go aimlessly everywhere.] (von Frisch, 1914)

This clearly contradicts his main conclusion.

6

Chapter 1

(A) 150 120 90 60 30 0

Trained on mid-grey Test on grey levels Tables 1–5 in von Frisch White

Trained on blue-green Test on grey levels

Counts of choices made by tested bees

(B) 800 600 400 200 00

Tables 72–79 in von Frisch White

(C)

Black

Mid-grey

Mid-grey Trained on green Test on grey levels

200 150 100 50 0

Grey levels

White

Bluegreen (11)

Black Tables 69–71 in von Frisch Green (10) Black

Trained on grass green Test on colour series

(D) 800

Grass green

600

Tables 51–55 in von Frisch

400

Purple

200 0 Red

Yellow

Green

An explanation for his successes is that most colours differ from grey in the measures of blue content and green contrast at the edges. When trained on black and tested with the grey series, bees went to black, but when tested on the colour series or the grey series with addition of blue, they went to blue. After 6 days training on blue-green they had a poor score and responded best to mid-grey. Trained on blue or purple, the tested bees went to blue or purple. When trained on yellow versus mid-grey for 6 days, most selected yellow when tested, but when repeated, half went to yellow and half to black; but in another repeat most went to green, and those trained on the grey series alone went to black (Fig. 1.3). So yellow looked like black, but not every time. When trained on white, they went to white. To the

Blue

Fig. 1.2.  Anomalous collected data from von Frisch (1914). Blue-green (11) in (B) and green (10) in (C) are the standard coloured papers in the Hering series used by von Frisch.

trained bees, yellow-green and blue-green were equivalent to mid-grey. Again Frisch had no explanation. He gave up this type of experiment and turned to training on one colour among a mix of colours and grey levels, and testing on a range of colours (where the colours all had different levels of blue content). Large numbers of bees were counted; precautions against obvious possible errors were taken, and most of the scores look reliable. Credit should be given to Frisch for his honesty in presenting results that enable us to disprove his conclusions with his own data. Little did he realize that his anomalous results were exactly as predicted by a different theory. As will be seen, these anomalies were missed by later researchers, were never explained, and soon forgotten. Looking now at the data, it is surprising that



The Difficult Birth of Honeybee Colour Vision

Midgrey

Lost

Lost

Light grey

Dark grey TRAIN ON THESE

White Be

Blue

es

tr ai n e

Bluegreen White go to these Blue

Black

e d on th

Green

se

Red

Yellow Black

Lost

go to these

Green

Yellow

Red

Bees trained on a colour at each end of the spectrum tend to go to similar colours. Fig. 1.3.  A summary of the choices reported by von Frisch. (Drawn from numerical data in von Frisch, 1914.)

his final conclusion was ever accepted, but clearly he was absolutely convinced of the correctness of his conclusions, although somehow his long paper never appeared in translation. Now go forward in time to the excellent, and representative, textbook by Gould (1982). I quote ‘if they saw the world in black and white, they would confuse at least one of the shades of grey with the colored sheets. The bees scored on exclusively the coloured papers’. After 1918, all textbooks were orthodox and said much the same. Move forward again to an anniversary volume for Frisch, and we find similarly ‘if bees were only using an achromatic mechanism’, etc. ‘However, the bees reliably chose the colored stimulus among many shades of grey’ (Dyer and Arikawa, 2014). It is the duty of authors to check the main conclusions of their sources, but at that time no one suspected anything was wrong. We now know that the failure in tests to recognize the green-blue or mid-grey level

7

was because the bees could not separate different grey levels presented together in the four-by-four array. They took a measure of the average level of blue content measured across the whole array, and this average was the same as that in the mid-grey, or the blue-green. They measured the total green modulation in the array. Unknowingly, Frisch made two false assumptions. First, he believed that bees could see grey levels and secondly, to save time, he presented grey levels in one array. Bees have receptors for green-blue and UV, but none for grey or white. As shown later, bees detect total blue in the whole display as a single colour (actually a measure of the blue content), and they detect green or blue contrast at edges. In later work, Frisch frequently used blue for his rewarded colour, so the bees used their innate preference to fly towards blue, and omitted the process of search by trial and error. Also, he assumed that the bees learned the colour that was rewarded, and he made no controls or tests about that. He was not aware that white looked bluer than blue to the bees, that they learned to avoid the unrewarded papers, and relied on green receptor contrast at edges. Of course, the bees could distinguish other pairs of colours if given targets that contained different amounts of blue colour, and all colours contain some blue. Yellow, for example, has 12% of the blue content of white paper. Although a theory of trichromatic vision had been suggested for man by Maxwell in 1855, Frisch had no way of knowing that the bees had only green, blue and UV receptors. More importantly, he had no idea that they might have an entirely new kind of ­colour vision. In 1914, before the results were published, Frisch took his bees to the 24th Zoological Congress in Freiburg, which was open to the public, well publicized and reported in the press. The burghers all turned up expecting to find support for the colour vision that Hess had denied. A feeding plate was put down with several different grey papers and a blue paper. The travelling hive was opened, whereupon the bees descended in a swarm on anything blue, like hatbands, cravats and buttonholes among the audience,

8

Chapter 1

who were absolutely convinced about their colour vision, with little thought for the essential controls. Uncritical popular enthusiasm clearly hastened the acceptance of his ideas, and professors from Vienna and Münich added weighty support. After all, they had contributed some of the data. Later, Frisch admitted that he had been training these hungry bees for weeks to go to a blue reward. None bothered to mention Forel or Hess, or that bees preferred blue anyway. There were no referees, but Hess was quick to point out that Frisch’s bees could have discriminated by local comparison of brightness and previous knowledge of detail in the way that a colour-blind person would, and that there was no control against the bees marking the training target with an odour cue from their Nasonov gland. The UV reflectance of the coloured papers was also questioned, but these objections were weak or invalid. The press and the public were already convinced. In response to Frisch, Hess (1918) set up a significantly different training arrangement, with eight blue squares and eight yellow squares arranged in a four-by-four chequerboard (Fig. 1A). He trained on several blue squares, or alternately on yellow squares, and then tested on other colours, black or white, with no reward in the tests. He found that bees trained for 8 days on grey or blue paper would immediately go to blue paper that they had not seen before. Bees trained on yellow versus grey, and then tested with a variety of colours appeared to have learned nothing. Hess also trained bees continuously for 6 weeks on yellow squares versus blue, then found that they could not distinguish yellow from other colours or grey. His experiments seemed to disprove Frisch, but in fact could not, because he did not test what the bees ­detected in each training. Of course, his test showed that the bees had not learned the yellow colour as the cue, but he did not realize that they had learned to avoid the unrewarded blue, or use a different cue altogether, such as contrast at edges. Their positive responses in tests with other colours appeared very strange, because everyone believed the bees were conditioned to the reward on yellow, forgetting the possibility of learning by trial and error

to avoid the unrewarded colour. Hess was ­obviously very stressed by his data: Aus meinen eigenen einschlägigen Versuchen, die ich in grősserem Umfange durch viele monate fortfűhrte, sei hier nur erwähnt, dass Bienen, die ich 6 wochenlang täglich ununterbrochen auf Gelb ‘dressiert’ hat, als ich ihnen unter einer reinen Glasplatte farbige Flächcharen auf Blau, Purpur und Schwarz flogen, während dicht daneben liegende gelbe Felder nicht besucht wurden. (von Hess, 1918)

Hess pointed out that ‘bees trained to go to yellow versus blue would not later recognize yellow’, a point that was ignored and later suppressed by von Frisch and his students for nearly a century. Frisch and Hess had been enemies since the days when they obtained opposite results with fish colour vision in Naples. Supercilious criticism by Hess, and the blazing certainty of the youthful Frisch made collaboration impossible; lack of a useful philosophy of science made them powerless to think constructively. Neither of them understood their own results or repeated the experiments of the other. Frisch found that bees trained on yellow on a background of mixed grey levels would go to black, dark grey or yellow. Hess found that bees trained on yellow on his chequerboard of yellow and blue would go to any colour. When trained on yellow, Frisch bees went to the black end of the grey series, while Hess bees learned only to avoid blue, and were lost when tested with a variety of colours. If only they had agreed to discuss their results constructively, they might have concluded that yellow looked like dark grey or black to bees. If you reach chapters in this book on the mechanism of bee colour vision, you will realize that they were both partially correct. Unfortunately, neither used a method that would demonstrate what the bees actually detected. To do that, it is essential to test trained bees until all inputs have been identified. Actually, both authors clearly demonstrated that there was something very peculiar about honeybee colour vision, but they could not identify it.



The Difficult Birth of Honeybee Colour Vision

As a senior professor of ophthalmology, Carl von Hess must have known more about colour vision, and tests for it, than most scientists in Europe. His objections to Frisch’s conclusion deserved serious consideration. Hess had done a good many experiments, which appeared to be properly recorded, so there should have been a reinvestigation to find what the bees had actually detected. However, the topic became taboo. Frisch wrote raging illogical rejoinders in the public press. They were not liked by anybody, especially his influential zoological relatives, the Hertwig and Exner families. He had no scientific authority at that time. His complaints were weak and now bounce back upon him: Ich protestieere gegen diese in der ­Wissenschaft nicht ühliche Methode der Polemik, und ich kann verlangen, dass v. Hess so wegwerfende Redensarten, wie sie namenlich in seinen letzten ­Publikationen auf Kosten einer sachlichen Kritik meiner Versuche überhandnehmen, entweder im einzelnen begründet oder unterlässt. [I protest against the unscholarly methods of argument, the wandering logic and underhand methods; Hess should shut up or put up.] (von Frisch, 1914)

No doubt they were well matched in argument, but the quality of thought left much to be desired. Considering the superior position of a Münich medical professor in 1914 relative to a junior assistant in the Zoology Department, most of the excitement had little to do with the science, but the excitement gave the young Frisch a great reputation and early promotion to professor, although he was quickly banished to Rostock, a place far away by the Baltic. No doubt Hess continued to teach the contents of his own book in Münich, but he was written out of history. In the anniversary volume for von Frisch, we find: Hess’s mistake was to consider phototactic responses, which are exclusively mediated by light intensity, as a proof of general color blindness in bees. On the contrary, von Frisch adopted a different scenario, and

9

focused on appetitive foraging responses. . . . The result is meanwhile well known; bees always chose the color associated with sucrose and never confused that color with an achromatic cardboard. (Avarguès-Weber and Giurfa, 2014)

These authors, and everybody else, honestly believed what they had been taught. After the death of Hess in 1923, Frisch quickly returned to a favourable position in Münich Zoology Department, and later became the dominant professor on bee behaviour in Europe. In this bitter and acrimonious dispute, he eventually prevailed because only his work was quoted in the biological literature. As Founder and Editor of the principal journal, with no tradition of referees, Frisch was able to steer this branch of science and control many academic positions in this area, even long after his retirement. Several men who disagreed with him, Wolf, Jander and Esch were obliged to leave Germany to get a university appointment in the USA. A few further experiments by Frisch were intended to demonstrate colour vision or patterns (Fig. 1.4) but it is more likely that the bees detected differences in blue content and modulation (length of edge multiplied by the contrast at each edge). It is very interesting that Frisch never worked on colour vision again. My own view is that he was surprised by the resounding success at the Freiburg Congress, and kept his doubts to himself. Among later students were a few young men, notably Baumgärtner, and later Daumer and von Helversen, who made a thesis and published a paper on bee vision of colour with new results that did not compete with von Frisch, but they all appear to have abandoned a promising topic. Publications of young women who worked on bee vision up to 1939, Hertz, Lotmar, Friedlaender, Wiechert and Zerrahn, all contain anomalous data and doubts about the accepted ideas. Subsequently, however, following the demonstration of three types of photoreceptor (Autrum and von Zwehl, 1964), the idea of trichromatic colour vision was accepted without question. There were always cracks in the argument,

10

Chapter 1

These four, flat on a table, coloured green, yellow, blue, but (A) 10 cm

bees could not distinguish four shapes. (B) 5 cm

Discovery that amount of edge was distinguished. (C)

(D)

Displayed vertically, pairs were distinguished, but colour vision not required with two colours.

Fig. 1.4.  An abundance of convincing data. (A) Shapes of similar colour were not discriminated. (B) Flower patterns of differing length of edge were distinguished. (C–E) Different distribution of colours of unequal brightness was also discriminated. Von Frisch (1914) claimed that these examples demonstrated vision of colour and some patterns.

mainly because insect vision is adapted to detect edges but colour vision must be based on areas. In all the above, effort was always on the performance (i.e. whether the bees could do the task) with little attention to what the bees had detected and learned. Successful performances could never prove anything rigorously, but the popular assumption of colour vision was now justified by being in the scientific literature. Hess had some satisfaction in validating that phototaxis in the bee was UV sensitive and colour-blind, which was the first indication of different inputs for different visual responses. In his biographical memoir of Frisch, Bill Thorpe (1983) mentions that ‘Hess would not even consider Karl’s evidence, which in fact was conclusive, but tried to

smother him with the great weight of his scientific authority’. In my opinion that was incorrect: as shown above and by later discoveries, none of the evidence was conclusive. Thorpe had not read the papers carefully, but repeated what Frisch had said in conversation with him years later at Brunnwinkl. However, unfortunately for the advance of science, Hess died in 1923, and only the work of Frisch was quoted thereafter. I say ‘unfortunately’ because Frisch’s bees were usually successful in the tests, which tell us little about possible mechanisms, whereas Hess’s bees failed in many tests. Such failures tell us when a critical cue is lacking, and they lead to further investigations. It is interesting that Frisch would not accept Hess’ results because they were negative evidence, but that is exactly

(E)



The Difficult Birth of Honeybee Colour Vision

where later a­dvances originated. The tragedy for 100 years was the failure to look for an alternative paradigm. The rest of this book is mostly about the fall-out from this failure. In science, a paradigm is a comprehensive package of theory that is effective for human understanding of a topic, usually based on many solid observations or manipulative experiments. Thus, Newton’s statics and dynamics, Lavoisier’s chemistry of atoms and molecules and the Darwin– Wallace theory of evolution are important paradigms that clarified our understanding. From 1914 until 2014, the Frisch paradigm

11

about colour vision of bees, and by ­extension, other insects, was accepted. In my opinion, in German academic circles it was accepted without thought, criticism was discouraged, errors were covered up, and anomalies went unnoticed because few elsewhere actually read the German journals. The result was an attractive but misleading paradigm that was accepted and used for many studies of bees’ relations to coloured flowers. This story, and the persistence of the paradigm, convinced me that scientists prefer to follow accepted texts rather than trying new experiments, and they have difficulty in generating a new thought.

References Autrum, H. and von Zwehl, V. (1964) Spektrale Empfindlichkeit einzelner Sehzellen des Bienenauges. Zeitschrift für vergleichende Physiologie 48, 357–384. Avarguès-Weber, A. and Giurfa, M. (2014) Cognitive components of color vision in honeybees: how conditioning variables modulate color learning and discrimination. Journal of Comparative Physiology A 200, 49–461. Bethe, A. (1898) Dürfen wir den Ameisen und Bienen psychische Quälitaten zuschreiben? Archiv für gesampte Physiologie 70, 15–100. Dröscher, A. (2016) Pioneering studies on cephalopod’s eye and vision at the Zoologica Anton Dohrn (1883–1977). Frontiers in Physiology 7, 618. Dyer, A.G. and Arikawa, K. (2014) A hundred years of color studies in insects; with thanks to Karl von Frisch and the workers he inspired. Journal of Comparative Physiology A 200, 409–410. Forel, A. (1908) The Senses of Insects. Translated by Yearsley, M. Methuen, London. Gould, J.L. (1982) Ethology, the Mechanisms and Evolution of Behaviour. Norton, New York. Lovell, J.H. (1910) The color sense of the honey-bee: can bees distinguish colours? American Naturalist 44, 673–692. Lubbock, J. (1881) Ants, Bees and Wasps, (13th edn 1898). Kegan Paul, London. Menzel, R. and Greggers, U. (1985) Natural phototaxis and its relation to colour vision in insects. Journal of Comparative Physiology A 157, 311–321. Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222. Plateau, F. (1885–1899) Comment les fleurs attirent les insects. Recherches expérimentales. Bulletin Academie, Société royale belge 30, 466–488. (See papers listed by Forel (1908) p. 142.) Thorpe, W.H. (1983) Karl von Frisch. Biographical Memoirs of Fellows of the Royal Society 29, 196–200. Turner, C.H. (1910) Experiments on color-vision of the honey-bee. Biological Bulletin, Wood’s Hole 19, 257–279. von Frisch, K. (1914) Der Farbensinn und Formensinn der Bienen. Zoologische Jahrbücher. Abteilung für ­allgemeine Zoologie und Physiologie der Tiere 35, 1–188. von Hess, C. (1912) Vergeichende Physiologie des Geschichtssines. Gustav Fischer Verlag, Jena, Germany. von Hess, C. (1918) Beiträge zur Frage nach einem Farbensinne bei Biene. Archiv für gesampte Physiologie 170, 337–366.

Chapter 2 No Way to Untie the Spell

It is the overall strength of the evidence and argument that matters in science, not the hierarchical authority of the scientists involved. (Nurse, 2015)

After the death of Hess, Frisch quickly re-established himself in Münich and returned to work with bee navigation, not vision of colour or shape. The orthodox ­ paradigm for his new assistants was to assume that bee vision resembled human vision unless shown to be otherwise. Bees obviously saw things and colours, as anyone could understand as they watched bees fly about their business. Earlier research by Bethe, Plateau, Forel and Hess was ignored.

Efforts to Advance with the New ­Paradigm In 1923, Alfred Kühn, Professor of Zoology and Genetics at the University of Göttingen, and joint editor of a new journal with von Frisch, entered the fray with the aid of a huge arc lamp and a mercury vapour lamp to investigate the ultraviolet (UV) part of the spectrum; brave man. He trained bees to land on selected narrow bands in the spectrum and demonstrated that they distinguished four separate spectral regions between 300 nm and 650 nm. White flowers with no 12

UV reflection attracted bees. However, white that reflected UV did not attract, and all later researchers (except Hertz, 1939a, b) had continued to ignore this significant detail, which excluded trichromatic human-­ like colour vision. Kühn (1927) also found that bees responded to a grey ring on a yellow background as if it looked blue. Later, this finding was extended to other colour combinations, and given a name, colour opponency, but there were no tests of what the bees actually detected. Of course, grey reflected quite a lot of blue anyway, and the bees detected and measured modulation at the boundary, which almost all researchers failed to notice. After Goldschmidt lost his job in 1933 as Director of the Kaiser Wilhelm Institute in Berlin, Kühn took over as Director in 1937, and continued when it became the Max Planck Institute after 1946. Control of the research programmes by the same small group, the funding, and the journal (the Zeitschrift für vergleichende Physiologie) was complete. From 1914 until about 1984, most of the students who studied bee vision in Germany or Austria were concerned with topics initiated by von Frisch. They explored the performance of bees searching for food or returning to the hive. In topics involving colour, blue was usually the training colour. Tests of mechanism were few; the aim was

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)



No Way to Untie the Spell

to discover the means of navigation to food sources and return home, and to measure performance in training and find its limits. Assuming colour vision, Baumgärtner (1928) had inferred that a principal cue was the difference in colour, or chromatic contrast, an assumption taken from human psychophysics and repeated for decades. He found that bees detected single coloured spots subtending more than 3–5° on a white background. Later, this sharp resolution was attributed to detectors of green contrast. The maximum distance from which a blue target versus a yellow target was discriminated was proportional to the area, not the length of side, suggesting that the cue for colour depended on the total photon flux, which depended on the number of facets involved, as confirmed 70 years later and attributed to the measurement of blue content. Baumgärtner also found that resolution of coloured patches in the vertical direction was three times better than in the horizontal direction, a detail that Friedlaender (1931)

(A)

13

could not repeat. Baumgärtner explained it by his anatomical measure of the astigmatism of the interommatidial angle, but it was probably because scanning was in the horizontal plane. Much later, the bee retina turned out to be almost isotropic (Fig. 4.3, this volume). Bees discriminated a small yellow square from a blue square at various separations in front of the hovering bee (Fig. 2.1). Baumgärtner’s explanation was that the two eyes must collaborate to detect the polarity of the yellow/ blue combination, but recently it was shown that bees detect relative positions of blue content and green contrast at the edge of the yellow square, up to separation angles of 90°. Yellow and blue squares of side 5 mm could not be resolved at all at a range of more than 25 mm or angular width of 11°. At this range the total angular span between the blue and yellow squares exceeded the field of a single eye, suggesting that the detection of left/right polarity was done with two eyes or the bees were scanning.

(C) 90° 80° 60° 40°

2

18 mm Blue circle

5 mm

20°

29°

Vertical plane

1

7 cm

0 20°

(B)

40°

29°

60° 80° 90°

Horizontal plane 30° 20° 10°

0

10° 20° 30°

Fig. 2.1.  Detection of polarity of yellow and blue positions. (A) The training target, with (1) separated squares, (2) adjacent squares. (B) Only the squares close to the reward hole were discriminated (black squares), probably by detection of blue, and green contrast. (C) Measurements of the angular width of overlap of the visual fields of the two eyes above and below the head at various angles to the horizontal. The maximum overlap is at 50° from horizontal looking down. (Replotted from Baumgärtner, 1928.)

14

Chapter 2

In a section on resolution, Baumgärtner also showed that the minimum subtense to discriminate the colour of an area displayed in the vertical plane was 20° in sunlight, while for detection of any colour on a white background it was as little as 3–4°. This interesting anomaly was ignored and forgotten until revived by Giurfa in the 1990s (see Chapter 3, this volume) and later attributed to the switch between blue content and green contrast cues. With reference to pollen guides on flowers, Baumgärtner found the following order of effectiveness of small marks of various colours on various backgrounds: yellow on white – excellent; blue on white, yellow on blue, blue on yellow, white on yellow, white on blue – useless. The explanation is that white reflects most blue, and yellow reflects 12% blue compared to white, while yellow on blue gives a strong green contrast. Foliage in sunlight is usually neutral for bees because they adapt to it. Yellow or blue flowers on a background of green foliage modulate the bees’ green Experiment 1 train on blue; discs shuffled

r­eceptors and also display a measure of blue content less or greater than background; therefore bees ­easily locate them. Later authors repeated this work without reference to it. Frisch had found that bees recognized a few examples of flower-like shapes but failed with unnatural geometrical patterns. Baumgärtner found that an essential cue for discrimination between different numbers of petals was a central petal to land on, with others on each side. Bees would not land with a gap on the midline of their visual field. He also found that the arriving bee learns the position of blue colour below the reward hole (Fig. 2.2). He observed that the bees scarcely ever fly directly to the reward hole, but when offered a blue disc they always aim towards it then diverge to the hole [‘das die Bienen eigentlich niemals direkt das Flugloch anflogen, sondern immer auf die blauen Farbkreis zuhielten und von diesen dann nach dem Flugloch liefen’ (Baumgärtner, 1928)]. Again, this discovery was lost in the literature. Experiment 3 train on blue

88%, n = 122 63%, n = 160 Experiment 4 train on blue Experiment 2 train on blue; discs shuffled

98%, n = 82

94%, n = 217

Fig. 2.2.  An early study intending to show that bees recognize blue on a white background. Results in Experiments 1 and 2 show they learn it better in the lower part of the target. Adding more in the upper part of the target has little effect, as later authors confirmed. Results in Experiments 3 and 4 show that the bees were not using the position of the missing blue in the white of the background. The result now implies that black alone, covering white background, would be best detected in the upper half, as also later confirmed. The black circles are the reward holes, on three separate reward boxes that were regularly shuffled in position in each training or test. Percentage values indicate the percentage of bees visiting reward holes. (Drawn from numerical data in Baumgärtner, 1928.)



No Way to Untie the Spell

In 1931, Marianne Friedlaender, a student of Otto Köhler in the University of Königsberg, showed that a radial pattern ­ acted as a reference centre and could be identified as radial even after being moved on the display, otherwise simple patterns were not recognized after being moved (a  feature of bee vision called retinotopic memory). The discrimination of black, white or coloured panels presented in the vertical plane was much better for up/down interchange of panels than for left/right change. The bees learned the retinotopic positions of a blue and a yellow panel presented side by side then tested separately. For the first time many tests were made. Her experiments were remarkably apt; indeed, conclusions can be drawn from them. Students of von Frisch measured performance in the training, those at Königsberg tried to make sense of numerous tests of the trained bees. When trained on blue and yellow versus the mirror image, none of her tests conclusively isolated a particular panel or colour as the cue, but there are indications that each eye learned one cue (Fig. 2.3B, C, D). In a test with left sides only, preference was reversed (Fig. 2.3E). Friedlaender could not explain this anomaly at all, but it was later shown that the unexpectedly exposed white background appeared intensely blue to the trained bees, as shown by * signs in Fig. 2.3 (Horridge, 2015). By 1933, Gertrud Zerrahn, working with black and white patterns, concluded that it was difficult to train bees to discriminate between two targets when the rewarded one was less preferred spontaneously. Much later I repeatedly found that in this situation the bees learned to avoid by trial and error the preferred cues on the unrewarded target, so explaining many curious results, for example training on yellow by von Hess (Fig. 1.1A). Ruth Lotmar, a student of von Frisch, using coloured papers, showed in 1933 that UV reflection contributed to discrimination of papers and flowers. For example, bees detected the common red poppy (Papaver) by its reflected UV. Following Kühn, Lotmar distinguished four regions of the bees’ spectrum: violet, blue, yellow and green. Within

15

40 mm (A)

Train 100%

76.3%, n = 527 (B)

Blue in the white (C)

100% Test

96.6%, n = 90 100% Test

81.8%, n = 44 (D)

Blue in the white

100% Test

91.9%, n = 112 100%

(E)

Test

18.9%, n = 74 Fig. 2.3.  An early example with numerous tests and an anomaly. (A) Train with blue and yellow versus the mirror image. (B) Improved score with right-hand panels only. (C) Discrimination with blue panels only. (D) Discrimination with yellow panels only. (E) With left sides only, the preference was reversed, with no explanation. This anomaly was not further investigated. (Drawn from data in Friedlaender, 1931.)

these regions, differences in brightness or UV content were well discriminated, but small differences in colour were not noticed, as if UV was a specific channel that inhibited the rest, not part of colour space. When two coloured papers differed mainly in brightness, the darker was preferred. With sufficient training time, and UV present, the bees could learn to distinguish any pair of coloured papers, probably by the amount of stimulus to the blue receptors. Lotmar also noticed that bees preferred to

16

Chapter 2

land on edges of green contrast. She made no comment that some results were contrary to the orthodox theory, but they were never quoted in the later literature. Based on work by Hertz and Lotmar, I later conclude that the UV receptor is an independent channel that plays little part in discriminations of hue in flowers, because UV inhibits the response to blue receptors. The last major publication on this topic before the war was in some ways the most informative, though not the most imaginative. Elsbeth Wiechert (1938), another student of Otto Köhler at Königsberg, trained bees to small squares 4 cm × 4 cm, each composed of two black, grey, blue or yellow rectangular panels, displayed in the vertical plane, continuously shuffled in position on a large slowly moving vertical wheel. The criterion of success was the landing on the correct reward hole, so the size at the moment of decision was comparable to that of flowers. Blue on a white background was spontaneously preferred to yellow in the ratio 5.5/1 and grey (with blue content) to black by 10/1. The experiments lasted many days with enormous numbers of counts. After long training, the position of a yellow panel was learned as well as a blue one (Fig. 2.4B, C). The bees learned blue at the bottom on the rewarded target (Fig. 2.4D), but they had not learned the unrewarded target (Fig. 2.4E). Yellow looked black to the bees and would be detected by edge contrast only. When Wiechert rotated the rewarded target, test scores were reduced a little at first (Fig. 2.4D), but reaching about 50% (random) at a rotation of about 90° (Fig. 2.4E). This shows that the unrewarded pattern was ignored. Bees discriminated the positions of coloured, black or grey panels better with up/down comparison than with left/right [because the vertical position of blue content was a strong cue]. It is now difficult to understand why these pre-war efforts failed to train bees to distinguish two colours then test them to see what they had learned. Questions that would have led to the discovery of the actual cues were not asked until 60 years later. There was no alternative to the belief that bees saw the patterns and colours. Nevertheless, these bright s­ tudents

Train 100%

(A)

55°

Blue in the white

81.9%, n = 502 100% Test

(B)

Blue in the white 67.8%, n = 456 100% (C)

Blue in the white

Test

Blue in the white

78.8%, n = 208 Rotated by 75° 100% (D) Test 70.0%, n = 316 Rotated by 90° (E)

100% Test

46.7%, n = 107 Fig. 2.4.  Another early example with numerous tests. (A) Training, vertical presentation. (B, C) Blue or yellow panels alone are sufficient. (D) Rotation of the rewarded target alters the amount of blue below the reward hole. (E) They had not learned the unrewarded target. (Redrawn from Wiechert, 1938.)

managed to publish hints that all was not well in bee discrimination. Wiechert made two other discoveries. First, that the bees could distinguish a differ­ ence in rotation of a thin black horizontal bar by 30°, but failed to detect a 30° rotation from the vertical. We now know that, with a thin bar, the preferred cue was not orientation but a measure of total modulation generated in a scan. Secondly, she showed that



No Way to Untie the Spell

bees would discriminate all experiments with coloured panels just as well when they were laid flat as when presented vertically, so long as the bees were forced to approach along the same flight track at each arrival. Wiechert’s work was leading towards discoveries of the 1990s but she was ignored for at least 50 years.

17

over a total sensitivity range of only 1000fold (Wolf and Zerrahn-Wolf, 1935, 1936) but in human photopic vision the range is 1000 times that. It is hard to understand how two such different mechanisms could have evolved to detect the same colours in the same environment with eyes of similar F value. This and his many papers in the Journal of General Physiology were ignored.

Forgotten contributions of Ernst Wolf Experiments of Mathilde Hertz Ernst Wolf (1931) measured visual acuity of bees when flying freely. Then, having escaped from Heidelberg to Harvard, in 1933 Wolf measured the resolution of light intensity. As a response criterion, he usually used the movement of a standing bee in the opposite direction to a small motion of a regular grating. This was the effect of optic flow, not an optomotor response. In the lowest light level at which the bees would respond, the minimum period was 20° but it was 2° in 1000 times brighter light, corresponding to the range between bright moonlight and daylight. Later in the century, an enormous effort to analyse the motion perception system of insects assumed a correlation between adjacent facets, ignoring this large smooth transition that Wolf found in dim light, with correlations between ommatidia up to ten facets apart as light intensity was reduced (Hecht and Wolf, 1929). Alone and in later collaborations with his wife Gertrud Zerrahn, and with Selig Hecht at Columbia University, Wolf made many measurements of the bee that were mostly ignored in the German literature. Using the phototaxis response, he measured the discrimination between lights of differing ­intensity. Surprisingly, the minimum differ­ ence was 25% in bright light and 300% at an intensity two log units down. It is difficult to imagine a trichromatic system of colour vision with such poor intensity discrimination. Corresponding values for the human eye are a 2% difference in bright light and a 20% difference at three log units down (1000 times less). The rapid adaptation to a brightness change is also relevant. Over the range from sunlight to darkness, bees adapt

Mathilde Hertz worked independently in Berlin as assistant to Richard Goldschmidt, the Director of the Kaiser Wilhelm Institute, from 1925 to 1933 when they both lost their jobs. She discovered that bees distinguish patterns by ‘figural intensity’ (Hertz, 1933), defined as the total amount of edge (Fig. 2.5). It is not clear whether she understood that contrast was equally important. This was a fundamental discovery because total length of edge in the pattern, multiplied by contrast at each part of the edge and the spectral sensitivity for each receptor type, is an approximation to the stimulus at the receptor level, and now called ‘modulation’, and measured relative to a black/white edge of the same length in the same illumination (Horridge, 1997, 2003, 2016). Modulation at any moment is a scalar product (dot product) of the spectral content of sunlight (amount of (vertical) edge displayed, contrast against background at each part of the edge, and spectral sensitivity of the receptor channel). Rapid adaptation means that light intensity cannot be controlled, so all measurements must be relative. Horizontal scanning selects vertical edges. The scanning eye detects the product, therefore confounds together edge length and colour, which cannot then be separated in the signal. Bees also detect, locate and measure local concentrations of edges (Hertz, 1929, 1930, 1931). More edge means more modulation, so a flower with more contrast, or more edges is more easily detected. It is essential to understand that contrast and structure are combined in one signal in each channel all the way to memory.

18

Chapter 2

Four types of patterns Tangential

Radial

Distributed

Irregular

Increasing ease of discrimination

Increasing disruption and longer edge

Besides edge length as a cue, Hertz (1933) defined ‘global’ features as those related to the detection of a pattern as a whole, and called this attribute the ‘figural quality’. On a horizontal or vertical surface, bees distinguished radial patterns in general from circular ones, and from patterns that were neither radial nor circular (Fig. 2.5). To do this, they first have to detect local edge orientation, but they did not remember it in detail, apart from total edge length or modulation. There is nothing special or ‘cognitive’ about this generalization of spokes or circles, it requires only a matrix of connections along lines of facets at 60° to each other (Chapter 6, this volume). Hertz trained bees to three types of patterns, tangential, radial, and modulation only. The greater the length of edge, the more easily the three types were recognized and learned, and pairs were more easily discriminated from each other, the greater the difference in edge length (Fig. 2.5). On a horizontal surface, patterns such as squares versus triangles, chequerboards versus other textures, or random patterns,

Fig. 2.5.  Hertz identified types of pattern that were distinguished; here shown as columns. Discrimination was easier with greater pattern disruption, shown as higher rows. Figural quality (type of pattern) in the three different columns, and measures of modulation in the different lines across. Small numbers are edge lengths. (Redrawn from Hertz, 1933.)

were not distinguished, except by total modulation, or differences in area of black, presumably because modulation was a total sum or they failed to fit the matrix coordinates at 60°. Bee preference of radial and avoidance of circular patterns, but failure to distinguish the great variety within each group, took a long time to be recognized. In Canberra in 2006, Srinivasan still concluded that bees distinguish between shapes largely by their outlines, while in the same year, I  concluded that predesigned arrays of orientation detectors of green contrast detect radial and circular edges, and showed that patterns and shapes were not reassembled in the memory of the bee. Hertz also found that edges and areas were detected differently. Training on a colour was successful with a bright figure with few edges on a dark surround, but the bees located and measured modulation and ignored colour in patterns with numerous edges. With a coloured ring on a bright surround, it was possible to change the colours and trained bees would still respond.



No Way to Untie the Spell

In other work, Hertz found that edges were distinguished better when in constant positions, and recognition was restricted to places on the eye where the inputs had been learned; so-called retinotopic vision. When colour, position or pattern was changed during an experiment with fixed patterns, memory was lost and learning had to begin again. When training patterns were regularly moved or when learning was distributed on the eye by scanning in flight, bees learned to search for displaced patterns. Hertz made the significant discovery that bees did not detect white spots on a black background in sunlight when the white paper reflected UV (Hertz, 1937). She used paper loaded with powdered barytes (barium sulfate) and was able to calibrate the UV reflectance. Even 1% of UV was sufficient to inhibit detection of white. On a black surface in sunlight, no bees would land on a rewarded white ring (6 cm outside diameter (OD), 2.5 cm inside diameter (ID)) made of paper that reflected UV, but they quickly learned to land on paper with no UV reflection. We can now understand that von Frisch, Friedlaender and others were unable to train bees to discriminate different grey levels, even if adjacent, because they worked outside in sunlight rich with UV. The inhibitory effect of UV on white or blue, and the difference between responses to areas and edges were published at least twice in German and later in English (Hertz, 1939a, b). It was proof that simple trichromatic vision was impossible in bees but these results were never incorporated into the literature. Hertz moved to England in 1936. Her long fruitless search for vision of shape (‘figural quality’) was praised, but her analytical results, outlined above and recently found to be generally valid, were forgotten. Not knowing the receptor properties, she could not infer the inputs at the eye, but made an effort. Bertholf (1931) had distinguished green and UV receptor peaks, and Sander (1933) a blue and a green type of receptor from the phototaxis response. Hertz (1939a, b) accepted ‘bee colours’, and was able to assign wavelengths to their positions in a

19

Maxwell triangular colour space for the honeybee, but she was unable to understand the colour ‘white displaying UV’ that bees avoided. In Cambridge, she became openly critical: ‘These results and arguments affect, in many instances, the interpretation of previous experiments.’ What could she be talking about? When she learned that three receptor types, UV, blue and green, were probable in the bee, and there was no receptor for white, this latest experiment had demonstrated that von Frisch was mistaken. In his influential survey, Wehner (1981) agreed that Hertz was correct about figural intensity but he was unable to find any other parameters in bee pattern perception and ignored all her other results: Mathilde Hertz was among the earliest and certainly among the most eloquent proponents of Gestalt theory in insect vision. This is mainly borne out by her influential papers on the organization of the visual field in bees, especially by those in which the concept of ‘figural quality’ was discussed extensively, though vaguely. As Thorpe (1979, p. 75) has pointed out in retrospect, this concept added to ‘the complexity of an already involved German style which many foreigners found difficult to comprehend’, but once understood ‘extremely rewarding’. In her later papers dealing with the perception of movement and hue of colour, she followed a much more analytical line of research, which was taken up by Hassenstein (1950) and Daumer (1956). (Wehner, 1981: footnote p. 525)

This assessment is all misleading; that is the point of the quote; Hertz had a Jewish family name although she was a Christian, and in 1981 there was still no appreciation of her work. In a clear, readable discussion in 1933, she had demonstrated ‘figural quality’ as the discrimination of radial, circular and other patterns of edges as types, and found that within each type they were discriminated only by total length of edge not pattern. Her work on edges, pattern types (Fig. 2.5) and colour was fundamental for understanding bee recognition of place (not for ‘perception of movement or hue of colour’, nor was her work taken up by those mentioned).

20

Chapter 2

Simultaneously, but far away in Heidelberg, Gertrud Zerrahn (1933) also found that bees could learn to distinguish between spokes, gratings and tangential types of black/white patterns even though they displayed similar lengths of edge. When given a choice between different lengths of edge in patterns of the same type, they always preferred greater edge length. She concluded that the mechanism involved a measure of frequency of stimulus changes at the receptor level, an idea very close to ‘receptor modulation’. The reader will have noted that all the young ladies who worked on bee vision, Friedlaender, Hertz, Lotmar, Wiechert, Zerrahn, had Jewish family names, irrespective of their religion. Their frightful social situation in Germany after 1933 meant that they could not be employed in universities. Research on bees was almost entirely in the hands of these intelligent and well-educated young women who were supported by their families with the protection of von Frisch. They obviously learned the orthodox story of bee colour vision. They worked hard and to the point, but only along the lines defined by the Professor. It must have been awkward when anomalies turned up in their ­results, as all of them reported. All of them dropped hints obscurely in their text but none actually contradicted von Frisch, or claimed openly that bees did not have colour vision. All were well informed, but they lacked the tools or experience to discover what the bees actually detected or imagine a new paradigm. Their scientific education, hours of bee training, and careful accounts deserved a better fate. Hertz was the most outspoken: in 1929 she published anomalies in the learning of yellow, and in her 1939 paper written in Cambridge, UK, she repeated that UV inhibits the detection of white (Hertz, 1939b). In those years, Germany was a terrible place for all, irrespective of religion or family name. They watched the destruction of the Austro-Hungarian Empire and the ­increasing Bolshevic threat from the East, while their academic world was slowly ­destroyed (Chapter 11, this volume). Von Frisch survived until 1938 because he

Austrian, then he too had to take was  ­ steps to save his skin. The unsuspecting scientists Almost all of the pre-war work was forgotten or totally distorted by the time that research began again in 1965. There was a new generation with quite different attitudes to the past, with more measurements in their experimental designs, but with complete acceptance of the orthodox, human-like colour vision of bees. After von Frisch retired in Münich in 1958, his former colleagues moved to Goethe University Frankfurt, with his long-time assistant Lindauer, as professor. It must have been a competitive place, packed with brilliant future professors, who all had ultimately to find jobs elsewhere. In 1956, Daumer made critical measurements showing that by suitable adjustment of the intensities, bees could match the colour of any particular wavelength by a mixture of a wavelength on either side. This was supposed to prove that three receptor types were involved, with broad and overlapping spectral sensitivities, but would be equally true for monochromatic vision. Daumer calculated the relative peak sensitivities and supposed spectral sensitivities of the receptor types. He drew a Maxwell colour triangle for the bee, which resembled human vision shifted about 100 nm towards the UV. The result strengthened the belief that bees had a trichromatic system, but all his data was about receptor cells, and I believe that the match would be possible no matter what curves he calculated. The method does not distinguish deeper mechanisms or show that the bees detect colour as we do. Quite rightly, the 1972 paper by Otto von Helversen at Freiburg is frequently quoted because it is a model of careful calibration and experiment. By this time, the relative spectral sensitivities of the three ­receptor types had been measured directly (Figs 3.1A and 4.1A). Von Helversen first plotted the intensity–response curves at different wavelengths, and calculated the sensitivity at different wavelengths, confirming



No Way to Untie the Spell

the high value in the UV. Next, bees trained to come to one wavelength were tested one by one with a neighbouring wavelength to find the points of minimum sensitivity to a wavelength difference. The best discrimination, near 500 nm is only about 5 nm, and near 400 nm is about 7.5 nm (Fig. 4.8C). The results were in agreement with the trichromatic theory for bees and also with the new paradigm. When this curve is compared to the receptor spectral sensitivities, it is obvious that the best discrimination is where the spectral sensitivity curves of the blue and green receptors are steepest (i.e. changing most rapidly) and most sensitive to changes in wavelength. The trough in discrimination is at the peak where the blue receptor curve is flat (i.e. insensitive to changes in blue), but the discrimination curve says nothing about the mechanisms of interaction. Simultaneous colour contrast A student of von Helversen, Christa Neumeyer (1980) trained individual bees to land on a blue-green paper which was near the middle of a series of nine differently coloured papers from yellow to violet, all on a large grey background. The bees were then tested with the same nine papers on a blue or yellow background. Neumeyer assumed that the bees looked for the colour that they had been trained on, and inferred ‘that the blue surround changed the apparent hues of the test fields towards green (yellow) and the yellow surround towards blue-violet’ (Neumeyer, 1980). The positions and separations of the hues in colour space were calculated, assuming a ratio of 5.6: 2.6: 1 in the relative sensitivity to UV, blue and green. There was no mention of the blue content of the grey training background, although bright blue has similar reflection of blue photons to a grey that is 50% black. All the discussion was about colour, with no consideration of edges, although it appears that the changes in preference were related to the contrasts against background. Neumeyer also found that the blue and yellow backgrounds lowered the sensitivity of the bees to monochromatic light at 431 nm

21

(blue) and 535 nm (green), respectively. She explained the change in preference as an ­effect of the coloured surrounds, although that was the observation, not an explanation. However, surrounds only 5–10 mm wide had similar effects to the 50 cm surround. Separating the surround from the test papers by a smaller gap greatly reduced the colour contrast effect even when the surround was narrow. Change in size of the papers between 5 mm and 50 mm had small indefinite effects. These interesting results were explained by ad hoc suggestions based on human vision that were not tested. Effects due to unavoidable changes in green contrast at the edges, and to the effect of total blue content on the discrimination of blue, were not considered. However, when bees were trained under white light and then tested under yellow or blue light, the bees responded as if ‘the test fields appeared in almost unchanged hues, despite the fact that the light reflected by a given test field stimulated the three receptor types in very different ratios’ (Neumeyer, 1981). So the bees enjoy colour constancy when the illuminating colour is changed but lateral spread of adaptation was suggested for effect of background (Neumeyer, 1981). No satisfactory explanation of colour constancy was offered at the time, because it was assumed that three receptor types were involved. Neumeyer (1981) found that the colour as seen by the bee appeared to change when only the background was changed, but stayed the same when the illumination of both background and target was changed. The logical conclusion is that the bees measure contrast at the boundary, not colour. However, she invented two new phenomena, the large spatial interactions and the colour constancy, to explain the data. Where did the ideas come from? Copied from human vision, of course. Anthropomorphism was at work again. At the time it was thought that humans detect colour but now we know that we also detect edges first and fill in the colour later. Explanations were invented within the existing belief system, an example of how the orthodox paradigm made researchers blind to their own discoveries.

22

Chapter 2

In later work with landing of the bee on the target as the criterion, bees detected a star on a coloured background versus a plain background. This was called simultaneous contrast (Dyer and Neumeyer, 2005). Of course, the bees would measure the difference in total content of blue and green modulation at the rewarded star. The limit of discrimination of contrast was remarkably low, similar to that in human colour discrimination, and better than that predicted by calculation from the noise levels of the three receptor types in trichromatic theory (Vorobyev et al., 2001). This anomaly was apparently not investigated, although one might expect a measure of monochromatic emission, or green modulation, to be at least an order if magnitude more sensitive than discrimination of chromatic contrast. For successive contrast, the bees had to discriminate between two plain coloured targets. In this experiment there was no star. The need to remember one colour while they looked at the other was now suggested as the reason why the performance was not so good. Not a good argument, I would say; it was more likely that the total stimulus was smaller with no star. Retroactive interference and the concept of ‘context’ In 1967, Randolf Menzel, a student of Lindauer at Frankfurt, trained bees to come to a flat white surface. In tests with colours, the order of preference by naïve bees was UV, blue and then green. When illuminated with spectral light, he found that bees were 85% correct after only one successful reward with blue light at 413 nm or 428 nm. Green at 532 nm was learned more slowly. The colour and intensity above threshold of the unrewarded target had little effect. In 1969, with the same apparatus as before, Menzel trained bees to one colour then suddenly changed the colour. Learning the new situation and forgetting the old one starts as soon as the bee detects the unfamiliar stimulus. After training on either orange or blue, and switching to the other, the bees rapidly learned the new colour, but the more

rewards they had received on the first colour, the less the response to the second, as if there was a cumulative effect of the rewards to the first colour. After two changes from blue to orange, the bees failed to respond (Fig. 2.6A). This destruction of memory (Fig. 2.6B) was later called retroactive interference. Bees could be retrained if shut up in the hive for a day. Irmgard Kriston (1973), a student at Frankfurt, published some unusual results while Menzel was away working in Canberra. Menzel had shown that bees normally prefer to learn blue first and yellow last, when given a choice. However, Kriston discovered that when trained first on a grey target with no colour, bees were more attracted to yellow than any other colour. Colours that were normally attractive became repellent. The author had a problem to explain this result, and the result was forgotten, but we now know that the bees had learned green contrast and amount of blue. Koltermann (1971), also at Frankfurt, extended retroactive interference to apply to inhibition of any cue by a different state of the first cue, including odour and time of day. He offered no explanation of the inhibition. Later, in 1996, I used this to show that cues could have different states. Bees trained to discriminate the orientation of a thin black bar would respond progressively less as edges at other orientations were added to the same side of the target. In local regions of the eye, different orientations were mutually cancelled, so they were different states of one cue. Later I found that different states of any single cue such as colour, radial, tangential or edge orientation, cancelled each other, but two or more different kinds of cues presented alternately could be learned together (Horridge, 1998). Different cues really were distinct categories for the bee, but colours were not, they were levels of blue. Later, when I trained bees to discriminate between a 20° spot that alternated every 5 min between yellow and blue on a white background versus a plain white target, at first they learned nothing, as Menzel had found, but after 6 h training they performed quite well. However, the trained bees then



(A)

No Way to Untie the Spell

Train 440 v. 590 a

Score correct

90

Train 440 v. 590 b

c

Train 440 v. 590 e

d

60 30

0

10

20 Choices

(B)

Score correct

23

Train 444 v. 613

Train 361 v. 484

Train 532 v. 413

f

g

h

90

i

Test f at 444

Test f at 444

60 Test g at 361

30 0

10

20 Choices

accepted a grey spot as well as either of the coloured ones. They had learned to accept edges with green contrast as preferable to nothing. In the early 1960s, Lopatina and her group in the Pavlov Institute in Leningrad showed that bees trained regularly at a fixed time would not respond at other times. Koltermann (1971) found that bees trained to two scents or two colours at different times of day would recognize them best at that time. Apparently the link to time of day applied to all cues and to the foraging response to the dance. In 1978, Franz Bogdany, found that when there was a change to a different state, such as blue colour to yellow, or lavender odour to benzaldehyde, the bees’ sense of time was lost, as if odour, colour and time of day were linked. There was general agreement that training was much easier if the targets differed in colour. Clearly different states of the orientation cue mutually cancelled each other. Therefore

Fig. 2.6.  The effect of change of the training colour. (A) Slow failure of the ability to learn with changes of colour. (a) Initial training on 440 nm versus 590 nm. (b) Reverse the colours. (c, d, e) Successive reversals destroyed the ability to respond. (B) Learning to discriminate between two colours that were changed. (f) Train on 444 nm versus 613 nm. (g) Change to 361 nm versus 484 nm. (h) Change to 532 nm versus 413 nm. (i) Single tests demonstrate the loss of memory caused by change in the colours. (Redrawn from Menzel, 1969.)

retroactive interference tells us that blue, green and yellow are not different cues but different states of the same cue (i.e. that colours, black, white and grey are one colour for bees). Years later, this can be explained because bees have only monochromatic vision in shades of blue. For them, white is intense blue and a black area is not a stimulus; only its edges are measured. Electrophysiology had already demonstrated the neural machinery that would explain detection of modulation by each type of receptor. In fact, the neuron properties are more suitable to detect line-labelled modulation than a tonic, maintained photon flux. Retinula cells adapt rapidly to changes in intensity; lamina ganglion cells adapt even faster, and large-field neurons of the deep optic lobe are mostly phasic and reveal a great variety of antagonistic inputs from two or three receptor types. None of that seems to be adapted to the discrimination of colour as a palette of sensations,

24

Chapter 2

and within the optic lobe there is no sign of the colour triangle, or of neurons that could discriminate between colours. In 1981, Alexa Riehle, an electrophysiologist in Menzel’s lab at Berlin, declared ‘Color opponent neurons [of the bee] cannot differentiate between two monochromatic lights in the heterochromatic flicker test’, and she could offer no explanation of recognition of different colours. Moreover, in the deep optic lobe, neuron fields are large, which implies that they are adapted to quantitative measurement of total stimulus over a large solid angle, but not the separation of the original inputs. Flowers have colours that create contrast relative to background and to shadows. It seems likely that bees are not very interested in colour as such, but they detect sums of modulated responses of large-field neurons with antagonistic inputs. Like shape, tonic colour turns out to be intuitive for humans but of little interest for bees. Bees return to the rewarded place exactly 24 h after they were trained on a colour or an odour. When trained at different places or different times of day, however, the bees kept each group of memories separate. Similarly, bees learn a succession of cues in a maze because each is in a different place. In all the above work the trained bees headed for the coloured target and then searched for the reward, so the context is not as simple as a coincidence of cues; it is a sequence of coincidences. Koltermann (1971, 1974) concluded that the linkage of coincidental cues was the explanation of the integrated memory of a familiar place. The linkage is now called context, confirmed later in Canberra (Zhang et al., 2006). To suggest that context is an abstract property of the place for the bee is an example of an attractive error of thought, called misplaced concreteness. To the bees, the place is recognized by the coincidence of cues to form landmarks in different regions of the eye at the proper time of day, not by a cause called context. A common cell mechanism, the synthesis of protein causing changes at synapses, is more likely an explanation of learning of both coincidences and measurement of time. To claim

that sensitivity to context is an indicator of cognition in bees, illustrates how language can mislead us. Absolutely convinced that their theory of trichromatic colour vision in bees was correct, Backhaus et al. (1987) used 132 pairs of 12 calibrated colours. Bees were trained on one colour and tested with all colours in turn. The scores were correlated with the differences in the calculated stimulus for each colour pair at each receptor channel in a huge analysis of variables. ­Significantly, the bees did not make use of intensity differences. Two independent ­ variables in human colour vision that were factors in the Backhaus model, called hue (blue/green) and saturation (UV/blue minus green), accounted for all the scores. The results agreed with a hypothetical neuron model with three types of opponent inputs (Backhaus, 1991), but the model was far from reality. They did not realize that bees changed their preferred cues from green edges to blue areas as the pairs of colours changed and the data was no longer homogeneous. Altogether, this was a great quantity of good data ruined, and because the bees never failed to discriminate, and were not critically tested to see whether they learned edges or areas. Alternative theories were not even suggested. Still the spell was tied From the time I started on insect vision in 1961, to about 2009, with all my students, I  believed that bees have trichromatic colour vision. For a few years, I joined Autrum and von Frisch as an editor of the Journal of Comparative Physiology. I was thoroughly in favour of bee colour vision, and appointed Menzel and later Backhaus to my research group in Canberra. In one example, Menzel reviewed bee colour vision in 64 pages plus 13 pages of references that defined the accepted view (Menzel, 1979). Evidence and reviews of trichromatic colour vision in the worker bee came almost entirely from work done in Germany, but there was no mention of the contributions of Lubbock, Plateau, Bethe, Forel, von Hess, Baumgärtner, Wolf, Opfinger, Hertz,



No Way to Untie the Spell

Lotmar, Friedlaender, Wiechert, Zerrahn, Kriston, Bogdany, Koltermann, or anomalies in von Frisch. We were all clearly in a period of unsuspecting acceptance of trichromatic colour vision in bees and therefore selective attention to the literature. In a volume edited by Autrum, reviews like this became the basis for research by students and experts alike, and subsequently a main drag on progress, but we, and the whole world interested in bees (e.g. Gould, 1982), were unaware of any problem. While in Canberra, Menzel trained dark-adapted bees to turn right or left in a choice between a narrow bandwidth coloured light and a white one. Near threshold, bees were a little more sensitive to blue than to green but about 100 times more sensitive to white than to perception of colour. That anomalous experiment should be ­repeated. Several times in this period, bees were trained to go to a half black, half white rectangle or disc versus the same with the colours reversed. Similar half-discs were used in colour. Tests with one target rotated by various angles showed an effect attributed to the changed angle of edge orientations, and another from the shift in position of black. These tests could reveal nothing about mechanisms because they ignored the main factors, the blue content of white background that was unexpectedly exposed and a measure of green modulation by horizontal scans. The results were performance only, with few tests of what features the bees actually detected. None of those who used black/white patterns made use of earlier work showing that UV in sunlight inhibited detection of white (Hertz, 1937, 1939a, b). Recent work shows no

25

achromatic visual channel. Two black and white patterns are discriminated by a differ­ence in blue content if they differ in area or position of black, otherwise by differences in green modulation at edges. In this unsuspecting period, almost all work was observation of the performance in training, with no progress with tests of what bees actually detected. Several researchers from 1916 to 1939 had found indications, indeed proof, that bee vision of colour was not like that in humans, but no one sought an alternative explanation. We see many descriptions of learning of black/white patterns, but nothing new about colour discrimination. It was assumed that receptors detected black, white, grey levels and colours, and resolution of pattern depended on their angular spacing. Before 1964, hard-­ working bee researchers knew nothing of the field sizes or spectral sensitivities of the feature detectors, and it was impossible to analyse vision of colour without papers that were equiluminant for green or blue receptors. Von Hess (Fig. 1.1A) had contradicted von Frisch (Fig. 1.1B) with a better-designed experiment but he could not explain his own results. In short, new ideas were ­required. Most deplorable in my view was the anti-scientific policies of never referring ­ to publications that disagreed with orthodox conclusions, and of dissembling referees who blocked progress, causing many missed opportunities. With great damage to all, ‘the overall strength of the evidence and argument’ to which the President of the Royal Society referred in the quote (Nurse, 2015), had failed completely, thanks to ‘the hierarchical authority of the scientists involved’.

References Backhaus, W. (1991) Color opponent coding in the visual system of the honeybee. Vision Research 31, 1381–1397. Backhaus, W., Menzel, R. and Kreißl, S. (1987) Multidimensional scaling of color similarity in bees. Biological Cybernetics 56, 321–331. Baumgärtner, H. (1928) Der Formensinn und der Sehschärfe der Bienen. Zeitschrift für vergleichende Physiologie 7, 56–143. Bertholf, L.M. (1931) The distribution of stimulatory efficiency in the ultraviolet spectrum for the honey bee. Journal of Agricultural Research 43, 703–713.

26

Chapter 2

Bogdany, F.J. (1978) Linking of learning signals in honey bee orientation. Behavioral Ecology and Sociobiology 3, 323–336. Daumer, K. (1956) Reizmetrische Untersuchungen des Farbensehens der Bienen. Zeitschrift für vergleichende Physiologie 38, 413–478. Dyer, A.G. and Neumeyer, C. (2005) Simultaneous and successive colour signals in honeybee orientation. Discrimination in the honeybee (Apis mellfera). Journal of Comparative Physiology A 191, 547–557. Friedlaender, M. (1931) Zur Bedeutung des Fluglochs im optischen Feld der Biene bei senkrechter Dressuranordnung. Zeitschrift für vergleichende Physiologie 15, 193–260. Gould, J.L. (1982) Ethology, the Mechanisms and Evolution of Behaviour. Norton, New York. Hassenstein, B. (1950) Wandernde geometrische Interferenzfiguren in Insektenauge. Naturwissenschaften 37, 45–46. Hecht, S. and Wolf, E. (1929) The visual acuity of the honeybee. Journal of General Physiology 12, 727–760. Hertz, M. (1929) Die Organisation des optischen Feldes bei der Biene. Zeitschrift für vergleichende Physiologie 8, 693–748. Hertz, M. (1930) Die Organisation des optischen Feldes bei der Biene. Zeitschrift für vergleichende Physiologie 11, 107–145. Hertz, M. (1931) Die Organisation des optischen Feldes bei der Biene. Zeitschrift für vergleichende Physiologie 14, 629–674. Hertz, M. (1933) Über figurale Intensität und Qualitäten in der optische Wahrnehmung der Biene. Biologische Zentralblatte 53, 10–40. Hertz, M. (1937) Beitrag zum Farbensinne und Formensinne der Biene. Zeitschrift für vergleichende Physiologie 24, 413–421. Hertz, M. (1939a) Versuche über das Farbensinne der Biene. Naturwissenschaften 25, 492–493. Hertz, M. (1939b) New experiments on colour vision in bees. Journal of Experimental Biology 16, 1–8. Horridge, G.A. (1997) Pattern discrimination by the honeybee, disruption as a cue. Journal of Comparative Physiology A 181, 267–277. Horridge, G.A. (1998) Pattern vision by the honeybee (Apis mellifera), training on two pairs of patterns alternately. Journal of Insect Physiology 45, 349–355. Horridge, G.A. (2003) Visual resolution of gratings by the compound eye of the bee (Apis mellifera). Journal of Experimental Biology 206, 2105–2110. Horridge, A. (2015) How bees distinguish colors. Eye and Brain 7, 17–34. Horridge, A. (2016) Parallel inputs to memory in bee colour vision. Acta Biologica Hungarica 67, 1–26. Koltermann, R (1971) 24-Std-Periodik in den Langzeiterinnerung an Duft- und Farbsignalen bei der Honigbiene. Zeitschrift für vergleichende Physiologie 75, 49–88. Koltermann, R. (1974) Periodicity in the activity and learning performance of the honeybee. In: Barton Brown, L. (ed.) Experimental Analysis of Insect Behaviour. Springer, Berlin, pp. 218–227. Kriston, I. (1973) Die Bewertung von Duft- und Farbsignalnen als Orientierungshilfen an der Futterquelle durch Apis mellfera L. Journal of Comparative Physiology 84, 77–94. Kühn, A. (1923) Vesuche über das Unterscheidungsvermögen der Bienen und Fische für Spectrallichter. Nachrichten von der Gesellschaft der Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse 1, 1–6. Kühn, A. (1927) Über den Farbensinn der Bienen. Zeitschrift für vergleichende Physiologie 5, 762–800. Lotmar, R. (1933) Neue Untersuchungen über den Farbensinn der Bienen, mit besonderer Berücksichtigung des Ultravioletts. Zeitschrift für vergleichende Physiologie 19, 673–723. Menzel, R. (1967) Untersuchungen zum Erlernen von Spektralfarben durch die Honigbiene (Apis melifera). Zeitschrift für vergleichende Physiologie 56, 22–62. Menzel, R. (1969) Das Gedächtnisse der Honigbiene für Spektralfarben. Umlernen und mehrfachlernen. Zeitschrift für vergleichende Physiologie 63, 290–309. Menzel, R. (1979) Spectral sensitivity and colour vision in invertebrates. In: Autrum, H. (ed.) Handbook of Sensory Physiology, Volume VII/Part 6A: Invertebrate Visual Centres and Behaviour. Springer, Berlin, pp. 503–580. Neumeyer, C. (1980) Simultaneous color contrast in the honeybee. Journal of Comparative Physiology 139, 165–176. Neumeyer, C. (1981) Chromatic adaptation in the honeybee: successive color contrast and color constancy. Journal of Comparative Physiology 144, 543–553. Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222.



No Way to Untie the Spell

27

Riehle, A. (1981) Color opponent neurons of the honeybee in a heterochromatic flicker test. Journal of Comparative Physiology 142, 81–88. Sander, W, (1933) Phototaktische Reaktionen der Bienen auf Lichter verschiedener Wellenlänger. Zeitschrift für vergleichende Physiologie 20, 267–286. Srinivasan, M.V. (2006) Honeybee vision: in good shape for shape recognition. Current Biology 16(2), R56. DOI: 10. 1016/j.cub.2006.01.002 Thorpe, W.H. (1979) The Origins and Rise of Ethology. Heinemann, Praeger, London. von Helversen, O. (1972) Zur spektralen Unterscheidsempfindlichkeit der Honigbiene. Journal of Comparative Physiology 80, 439–472. Vorobyev, M., Brandt, R., Peitsch, D., Laughlin, S.B. and Menzel, R. (2001) Colour thresholds and receptor noise, behaviour and physiology compared. Vision Research 41, 639–653. Wehner, R. (1981) Spatial vision in arthropods. In: Autrum, H. (ed.) Handbook of Sensory Physiology, Volume VII/Part 6C: Vision in Invertebrates. Springer, Berlin, pp. 287–616. Wiechert, E. (1938) Zur Frage der Koordinaten des subjectiven Sehraumes der Biene. Zeitschrift für vergleichende Physiologie 25, 455–493. Wolf, E. (1931) Sehschärfeprüfung an Bienen im Freilandversuch. Zeitschrift für vergleichende Physiologie 14, 746–762. Wolf, E. (1933) The visual intensity discrimination of the honeybee. Journal of General Physiology 16, 407–422. Wolf, E. and Zerrahn-Wolf, G. (1935) The dark adaptation of the eye of the honeybee. Journal of General Physiology 19, 229–237. Wolf, E. and Zerrahn-Wolf, G. (1936) Flicker and the reactions of bees to flowers. Journal of General Physiology 20, 511–518. Zerrahn, G. (1933) Formdressur und Formunterscheidung bei der Honigbiene. Zeitschrift für vergleichende Physiologie 20, 117–150. Zhang, S., Schwarz, S., Pahl, M., Zhu, H. and Tautz, J. (2006) Honeybee memory: a honeybee knows what to do and when. Journal of Experimental Biology 209, 4420–4423.

Chapter 3 Innovation, Deep Thought and Hard Work

The birth of modern science coincides with the founding of the Royal Society in 1660. Our motto, ‘nullius in verba’ reflects an emphasis on the need to rely on ­demonstrated observation and experiment rather than established authority. (Nurse, 2015)

At last, a time arrived when anomalies piled up so high it was obvious that the orthodox concept of trichromatic colour vision was no longer useful for the honeybee, and no alternative was available. In a volume for the anniversary of the 1914 paper by von Frisch, an apparently authoritative review (Hempel de Ibarra et al., 2014) made no mention of the anomalies of bee vision of colour, and said nothing about feature detectors and cues really used in vision of colour, black or white. This chapter summarizes the gradual arrival of significant ideas that eventually led to a finale in 2014, just 100 years after the paper by von Frisch. Halfway through that century, in 1964 in Münich, Professor Hans Autrum and his assistant, Vera von Zwehl, measured spectral sensitivities of the three types of photoreceptors in the compound eyes of worker bees (Fig. 3.1A). I saw their huge apparatus with a carbon arc; I was there. This advance eventually opened many doors, but initially it was regarded as further evidence of trichromatic colour vision. In new reviews 28

and textbooks, those working on bee vision were reinforced in their belief in the orthodox account. No one seemed to notice that the bee ommatidium had receptors for UV, green and blue but nothing for grey, white or black. However, it became possible to calculate the relative stimulus to each receptor type from the emission spectrum of any coloured paper in sunlight relative to that of white paper (Table 3.1). These numbers give an excellent idea of what a bee detects with blue and with green receptors when looking at coloured papers used in training and tests in sunlight (in Canberra). Thanks to these calibrations, following far behind human psychophysicists in the use of equiluminant colours, a whole new paradigm could be explored in the bee.

Equiluminant Colours In all work outdoors, the composition of sunlight (Fig. 3.1A) must be taken into account. Visual pigments have broad spectral sensitivities (Fig. 3.1B), but still have excellent discrimination between neighbouring wavelengths (Fig. 3.1C), showing that they measure and remember relative intensities (of blue content). Coloured papers have broad emission spectra, and receptors ­detect

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)



Innovation, Deep Thought and Hard Work

29

Light intensity (photons/cm2/s/nm)

(A) Midday sunlight × 10 –12

200

Sunset × 10 –9 100 90° to sun × 5 × 10 –10 300

400

500

600 nm

Relative sensitivity

(B) 100%

Threshold difference (nm)

(C)

Ultraviolet

60%

Green

Worker bee

0 300

400

500

600 nm

25 Worker bee

17 nm 15

5 300

(D) 100% Percentage of maximum

Blue

60%

8 nm

5 nm 400

500

600 nm

Blue receptor sensitivity Relative emissions of coloured papers

Yellow Green Blue 0 300

400 500 600 Wavelength of the light (nm)

photon absorptions with no record of their wavelengths within their broad absorption spectra. Therefore, as illustrated by the areas of overlap under the curves (Fig. 3.1D), an

Fig. 3.1.  Data relating to spectra and sensitivity. (A) Spectral composition of sunlight at different times of day. (B) Spectral sensitivity curves of the photoreceptors of the compound eye of the worker bee, normalized to 100%. (C) Threshold differences in the discrimination of wavelength by trained bees. (D) Definition of equiluminance. The spectral sensitivity curve for the blue receptor, together with spectral emission curves for three papers that could be equiluminant for this blue receptor, and therefore indistinguishable by it alone.

infinite number of coloured papers and patterns can generate the same response to a single receptor type. These different colours are called ‘equiluminant’ for that receptor,

30

Chapter 3

Table 3.1.  The receptor stimulus from each paper relative to white copy paper (100%), and contrasts between two equiluminant pairs of papers. White displays by far the greatest stimulus because it spreads maximum emission over the whole wavelength range of each receptor type. Canson papersa White copy paper Hemp 374 Ultramarine 590 Green 576 Buff 384 Blue 595 Dresden Yellow Contrast 374/590 Contrast 384/595

Blue receptor (%)

Green receptor (%)

100 34.2 33.8 17.0 25.7 54.2 13.1 0.006b 0.36

100 56.3 20.7 22.3 41.7 40.0 78.1 0.46 0.02b

Colour names and numbers are of the manufacturer, Canson. For supplies of papers, see: http://www. canson-infinity.com/en/values.asp (accessed 1 November 2018). b These are the values of contrast at a boundary between the two colours identified by their code numbers in the table. a

29

Connecting slots

cm

Pattern on target Reward hole

Transparent baffles Feeder

Leg

Reward box

Choice chamber Bees fly in here

No reward

Fig. 3.2.  The Canberra Y-choice chamber used for all training and testing of trained bees since 1988. The two displays, together with the reward box are changed between the two sides every 5 min during training and testing, to train the bees to look at both targets and ignore side preferences. The transparent baffles, added in 1994 (appear in publications in 1996), force the bees to stop and look at the two displays, before they choose which side to go. The concentration of sugar solution in the feeder is adjusted to keep the bees coming without attracting recruits. For detailed instructions about the use of this apparatus, see Appendix, this volume.

but a different receptor channel would distinguish some of them. Equiluminant colours that put one channel out of action during training or testing in the absence of UV made possible the discovery of the signals carried by the remaining isolated blue or green channel. In 1980, my colleague Srinivasan joined me in Canberra from the psychophysics laboratory of Professor W. Miller at Yale. He moved to Zürich in 1983, where, with Miriam Lehrer, he used equiluminant flashes to show that green and blue receptor channels of the bee have different time constants in a behavioural response and are therefore separate. He also found that bees distinguish a heterochromatic flicker (two colours alternating) from a steady light of the same average colour, but because they move, and have compound eyes, bees cannot distinguish ­between a flickering light and a steady one of similar colour (Srinivasan and Lehrer, 1984). Back in Canberra, we designed a general-­ purpose choice chamber for all kinds of experiments on honeybee vision (Fig. 3.2). In our new apparatus, Srinivasan and Lehrer (1988) used an equiluminant heterochomatic regular grating with vertical bars, displayed versus another similar grating with horizontal bars, at various distances to measure spatial resolution of green or blue receptor channels separately. The bees used two separate colour channels in parallel. The lower limit of green contrast detected was about 5% and green contrast became saturated above 30%. Blue channel modulation was effective as an input, but with poorer resolution. The authors ignored length of vertical edge as an input and assumed that their bees detected bar orientation on both targets. In 2003, I revisited these details. Discrimination between a grating at 45° to the vertical and a similar one at −45° (with no difference in modulation or colour content) was possible with gratings that were equiluminant to blue receptors but not when equiluminant to green (Fig. 3.3B), because only green receptors detected edge orientation (Giger and Srinivasan, 1996). When trained on horizontal versus vertical equiluminant gratings with no green contrast (Fig. 3.3C),



Innovation, Deep Thought and Hard Work

Bees easily discriminate black/white gratings

31

Train with no green contrast (C)

(A) Train 100%

100%

55° 80.5%, n = 200

74.5%, n = 200

Train with no green contrast (B)

100%

(D)

Test 100%

47.7%, n = 300

72.0%, n = 200

With no green contrast, blue receptor channel is insensitive to orientation

With no green contrast, the cue was a measure of blue modulation irrespective of pattern

Fig. 3.3.  An illustration of the difference between the bee’s detection of modulation and orientation. Percentage values indicate the percentage of bees visiting reward holes. There is only one reward hole in each experiment, the other hole is unrewarded. (A) Orientation of black and white oblique gratings is easily discriminated, but the feature detector is not revealed. (B) With no green contrast, the remaining blue channel is insensitive to orientation. (C) With no green contrast, the vertical/horizontal cue is the difference in blue receptor modulation. (D) Bees trained on (C) distinguish the blue modulation difference irrespective of pattern. (Redrawn from Horridge, 2003.)

bees still discriminated modulation very well (Fig. 3.3C) as shown by testing them with equiluminant radial or circular patterns with a difference in modulation (Fig. 3.3D). Although his bees trained to go to midgrey did not distinguish mid-grey in a palette of grey levels (Fig. 1.2A), von Frisch introduced the idea that bees see grey levels as if they were similar to other colours, and everyone else followed without question. By analogy with human vision, it became routine for black/grey/white targets to be described as achromatic for bees, but bees do not have receptors for black or shades of white. For bees, grey levels are shades of blue with green contrast at edges, and ‘achromatic vision’ is an oxymoron for bees.

Effect of Spot Size on Detection of Colour In 1995, Miriam Lehrer and Bischof trained bees in Canberra to detect a grey or coloured disc on a contrasting background versus the background alone, not realizing there was a

difference in blue content. Training was successful because they worked indoors in a bee house with UV excluded, and with papers that reflected little UV. The minimum detectable angular size of a disc on a white background varied from 10° for grey down to 2° for blue. The training scores for each colour combination of disc and background increased monotonically with angular size of the disc, but the extra effects of each increase in size decayed with increasing size, and were not even proportional to edge length, even less to area. The scores for a given disc were not related to contrast (which may have been saturated), but were low for a grey disc and high for a blue or a violet disc. When colour contrast was present, a change in receptor specific contrast or total intensity contrast had little effect. Contrast of 30% was needed to detect grey on white, but much less when colour was present. For grey discs on white, Lehrer still had a closed mind. She used the term achromatic, meaning ‘no colour’, although she knew quite well that grey excites both green and blue receptor types. She believed that bees

32

Chapter 3

see things, including their colours, and described the task as detection, not realizing that it was an exercise in measuring preferences for targets that they could detect. It was obvious that these results did not agree with the orthodox von Frisch trichromatic theory, but she had insufficient data to revise the theory, and no explanation was offered. Martin Giurfa also studied colour preferences (Giurfa, 1991). He trained bees to come to small rewarded aluminium discs (with flat reflection of UV in sunlight), and tested with a selection of coloured papers. Violet was preferred to blue which was preferred to yellow, as Menzel had found with narrow band filters, but the scores did not fit the positions of the papers in the colour triangle (compare Lehrer, above) and aluminium was abandoned. Significant for me was the discovery that bees carefully reared in isolation had innate hue preference for UV/blue, then for green, not for yellow, irrespective of background, intensity or green contrast (Giurfa et al., 1995), but white with and without UV was not tested. In that year, we discovered that bees have innate preference for radial patterns (Lehrer et al., 1995) that were easily learned (Hertz, 1933).

‘Absolute Conditioning’

The minimum angular subtense detected was then independent of blue or UV contrast (because the cue was green modulation). They calculated ‘chromatic contrast’ and interpreted their results in terms of blue or green contrasts in separate receptor channels, but even these were not the modulation cues (that include edge length). The human eyes did not notice the changes in area of background and length of edge that the bees detected (Fig. 3.4B). Later, in 1998, Giurfa and Vorobyev added that targets subtending more than 30° (A)

Train by ‘absolute conditioning’

(B)

Test with more yellow

More edge (C)

At this point the confusion grows with every new experiment. In 1996, Giurfa and others trained bees in a Y-choice maze (with a plexiglass top) to detect a coloured disc on a grey background, versus a plain grey target. (We now know that this ‘absolute conditioning’ forced the bees to learn whether to choose the greater or less blue content.) It was difficult to train with grey on white, showing that UV was reflected from their papers. Trained bees were tested with discs of the same colours at different angular sizes (Giurfa et al., 1996) but they ignored changes in area of background. Minimum detectible subtense (angular diameter) of a coloured disc was 15° when green receptor contrast was lacking (only blue content was available as a cue), but 5° when there was contrast to the green receptors (that have high resolution).

Less blue

Train, large versus smaller spot

More edge

Less blue

Fig. 3.4.  Cues that honeybees use to detect and learn a pattern. (A) When trained on a spot on a background versus just the background, bees ignore the outer edges because they are identical (arrows). They detect just the difference in total content of blue and a measure of width between locations of green modulation. (B) When tested with a larger ­yellow spot, the wider the spot, the smaller the amount of blue and greater the edge length. (C) With one pattern versus another, bees look for the difference in blue content and positions and measures of green modulation.



Innovation, Deep Thought and Hard Work

were not detected with green contrast alone. This strange result was an artefact of the display arrangement, caused by the limited total size. With white/grey/black targets the lower limit was 4° to 5° and detection of colour began at 10–15°. There were no tests of which cues were detected. A theory based on ‘facilitation’ of the blue channel by the green channel was no more than the data expressed in different words, and the model with a centre-surround field to explain angular sensitivity was a special case for one size and type of training pattern. In 1997 Giurfa et al. found that detection was independent of which colour was target and which was background. Surely that suggests that location and total edge modulation was the cue. None of this would be expected according to von Frisch. Two explanations were suggested: first, that green and chromatic contrasts interacted to produce normal detection of colour, and secondly, two different channels differed in field size. Amount of blue was not considered. They did not realize that the cues were blue content and total modulation (Fig. 3.4). At the time, it was believed that a tonic green response participated in chromatic contrast. From the loss of acuity when green contrast was removed, they agreed that the detector of green contrast has the angular size of one facet, as found by many others. Their interpretation in terms of the colour of (A)

33

areas implied that the input was photon flux, but the use of contrast implied that edges were cues. In 2003, Hempel di Ibarra and Giurfa discovered that discrimination of closed shapes (e.g. a square versus a diamond) required only green contrast, but they failed to find the actual cue, which was total green modulation at vertical edges (Horridge, 2016). They admitted that the contribution of UV and blue receptors ‘to achromatic vision has not been convincingly established’. How could it be? They found that a minimum of 59 facets was sufficient to smooth receptor and synaptic noise for ­detection of colour. The influential ‘noise-­limited theory’ of chromatic vision by Vorobyev et al. in 2001, was useless because it was based on this minimum area, and because there is no chromatic vision in bees, The Giurfa group were unaware, or did not mention, that detection of motion was restricted to the green receptor pathway alone, so without green contrast at all, bees do not stabilize their eyes on the pattern. No one imagined that, without green contrast at the boundaries (Fig. 3.5B–D), bees merged together areas of different colour. The separated colours are merged together to make a new average colour that is indistinguishable from the original pattern (Fig. 3.5F–H). A paper in 2003 by Niggebrügge and Hempel di Ibarra confirmed that increasing chromatic contrast improved detection, but

(E) 55° 58.0% after 3 h training

(B)

(F)

(C)

(G)

(D)

(H)

Fig. 3.5.  These pairs of patterns were not discriminated. In (A) and (E), a horizontal scan provided no horizontal position relative to a landmark. In (B–D) there was no green contrast between blue and buff. In (F–H) the average of buff and blue was green with the same total emission. For emission spectra of the papers in sunlight, see Table 3.1.

34

Chapter 3

of their achromatic intensity (Fig. 1.2B–D). For a start, in a test they could not discriminate mid-grey placed among the whole set of grey levels because the set was averaged and equalled mid-grey (Fig. 1.2A). Avarguès-Weber and Giurfa (2014) also used colour to infer ‘cognition’ in bees and explained it by ‘top down effects’ from higher centres in the brain. They trained bees in three ways to similar training colours. In the first method, called ‘differential conditioning’, there was one display versus another with a colour difference (like Fig. 3.4C). In this situation, bees use the differences in location between green contrast and blue content. In the second training method, called ‘absolute conditioning’, one colour was displayed on a background versus only the background (Fig. 3.4A). In this task, there will always be a difference in blue content, which the bees learn as first preference. In the third method, punishment replaced the reward and the bees behaved differently in tests. The bees behaved differently because they switched learning cues depending on the display. Bees learn by trial and error, and therefore learn first to avoid one target, but the authors did not test for that. Avarguès-Weber and Giurfa (2014) explained the difference by a supposed ‘top Summary of consequences of confusion down effect from higher centres’, but differences were actually caused by changes in The introduction to a paper by Avarguès-­ availability of cues. Even though the proper Weber and Giurfa in 2014 is worth a quote: critical tests had been illustrated for years, I believe these scientists resorted to imaginHess’s mistake was to consider phototactic ary explanations and steadfastly omitted to responses, which are exclusively mediated test whether ‘cognition’ was merely the reby light intensity, as a proof of general color sult of changes of the available cues that blindness in bees. . . . On the contrary, Von Frisch . . . asked in each case, whether they never considered. To my mind, ‘cognithe rewarded color could be discriminated tion’ falls into the category of concepts that by the trained bees from different achromatic the ancient Greeks, about 2000 bc, started to cardboards, some of which shared the same replace with ‘mechanism’. achromatic intensity with the trained color. This group of results of a particular group of researchers illustrates two ways of losing These orthodox beliefs were universally ac- scientific trust. A group in various combincepted in the 1970s, but they were totally ations act as collaborators and as referees for misleading. Hess had also looked at feeding each other, bypassing criticism and effectbehaviour (Fig. 1.1A); von Frisch used grey ively fooling the editors. Then, incredible levels in a grouped array too complex for conclusions emerge in efforts to make the bees to distinguish; there was no achromatic data fit their orthodox but incorrect paradigm. vision of bees, and von Frisch did not show A very solid review by three distinthat bees see different colours ­irrespective guished professors (Kelber et al., 2003) from adding green receptor contrast made no further improvement. They concluded that the achromatic system (presumed to be black/ grey/bee white) was insensitive, even though all three receptor types were stimulated. How could that be? Recent work shows that a main cue was a difference in total blue content of the two patterns, and a change in size of part of a pattern could increase the blue difference or reduce it to nothing and reverse it. Also, the bees detected a mixture of a retinotopic memory, adapted background, amount of edge, angular widths between edges, and the exposure of unadapted regions to unexpected changes in size of areas of colour or black at each test, none of which were mentioned. With a change in size or a movement of a colour, preferences changed as the cue was changed, so that the data was not homogeneous. The total picture was incomprehensible; a remarkable number of unsuspecting co-authors were misled and these extensive results of this group were not mentioned in an orthodox review to which one of the same group of co-authors later contributed (Kelber et al., 2003).



Innovation, Deep Thought and Hard Work

different laboratories was quite unsuspecting that colour and modulation caused by pattern structure are necessarily and inextricably involved in separate green and blue receptor pathways, and that grey is a shade of blue for bees. It was an excellent orthodox review of trichromatic bee vision as still held at that time: workers such as J. Lubbock and K. von Frisch developed behavioural criteria for establishing that non-human animals see colour [exaggeration]. Colour is used for specific behaviours such as phototaxis and object recognition [incorrect] . . . Having established the existence of colour vision, research focussed on the question of how many spectral types of photoreceptors are involved [not true]. (Kelber et al., 2003; my comments in square brackets)

There was no advance in thought; all the anomalies listed above were omitted. Critical thought had been dead for 64 years since Hertz (1939a, b).

Recognition of the Hive Entrance by Newly Impregnated Queens In 2004, with no reference to any of the above anomalies, at Kiev in the Ukraine, Alexander Komissar made great efforts to improve the success rate of newly impregnated queens that must find the entrance to their own nucleus or be killed immediately. To improve recognition of the correct entrance, in 1971 von Frisch had recommended a selection of one or two from only four colours (white, blue, yellow and black) to be placed at each entrance, to distinguish it from other entrances (Fig. 3.6, left side). Without direct experimental evidence, von Frisch believed that blue would be distinguished from white. Painting white all hives was known to confuse returning queens, and he recommended zinc white, with no UV reflection with the addition of coloured markers, probably influenced by reading Hertz (1937, 1939a, b). Working with multiple mating hives developed for commercial production of

35

Karl von Frisch recommendations 1. White

2. Blue

3. Yellow

4. Black or red

Alex Komissar recommendations 1. White or blue, not together 2. New proposed colour aluminium 3. Yellow

4. Black or red

Komissar found two other satisfactory patterns for recognition by returning queens Fig. 3.6.  Marks painted at the hive entrance so that the returning queen can identify the correct entrance after her nuptial flight. Size: 4–8 cm each side.

queens, Komissar (2004) and others had found extensive drifting between entrances painted with colours recommended by von Frisch, with 40% fatal results. Of course, repeated training sessions were impossible, so each queen had to learn the correct colour as they emerged and must recognize it correctly on their return. In 1986, Wells and Wells found that foraging bees confused blue flowers with white ones if the reward was similar. In tests with no theory stated, Komissar (2004) found that without UV, queens failed to distinguish between blue and white or grey and he recommended the use of yellow, black, red and aluminium reflectors of UV. White should not be used alongside blue, and blue that reflects UV should be avoided (Fig. 3.6, right side). In all this work, the practical applications improved the survival of fertilized queens to near 100%, but the fundamental basis was ignored. Komissar was convinced that edges of the dark entrance played a part, but no one mentioned that the best results were obtained with vertical edges with strong green contrast near the hive entrance. Subsequently, in 2016, I found that blue colour is not well localized in the horizontal plane, but a reliable landmark is the relative

36

Chapter 3

position of blue colour to vertical edges with green contrast (Chapters 5 and 6, this volume). Probably, landmarks nearby are more effective than colours on the hive.

Train with equal panels, no green contrast Train (A) 100% 55° 51%, n = 250 after 5 h of training

Coloured Panels and Large Spots ­Invisible to the Bee In 1996 Giger and Srinivasan showed that detectors of edge orientation were confined to the green receptor input so, with no green contrast, orientation of edges was not detected. In 1999, I found that learning and recognition of patterns with radial or circular edges also required green contrast because discrimination of this type of pattern required detection and location of edge orientation (Horridge, 1999a). Furthermore, two large areas of buff and blue with no contrast to green receptors at the boundary between them (Fig. 3.7A, B), were not discriminated from their mirror image, but addition of a vertical line or black spokes (Fig. 3.7C, D) restored discrimination. This was the mother of all anomalies, a crucial observation for analysis of bee vision of colour. My incorrect explanation at the time was that green contrast was required for motion perception, so the bee eye was not stabilized on the target. In 2015, the essential feature for recognition of all patterns that displayed this feature, turned out to be spatial polarity between the area with most blue content and a boundary or landmark formed by green contrast (Fig. 3.7E, F) (Horridge, 2015).

Failure to remember the colour of a spot One would suppose that bees trained to a single coloured spot would remember its colour, but experiments prove otherwise. In 2006, bees were trained to discriminate between a buff 20° spot on a blue background, with no contrast to the green receptors, versus a plain blue target (Fig. 3.8A) (Horridge,

Train with 20° spots, no green contrast Train 100%

(B) 20° spots

58.4%, n = 250 after 5 h of training Train 100%

(C) 20° spots

64.4%, n = 500, grey background with black star 66.0%, n = 400, white background with black star minimum spot size was about 6° Train with 20° spots, no green contrast, plus bar (D)

100%

63.0%, n = 400 Train, mirror images with green contrast (E)

100%

(F)

96.5%, n = 200 Test 100%

86.0%, n = 200 They learned relative positions of blue colour and the green contrast at edges of black.

Fig. 3.7.  The crucial advance in understanding bee vision of patterns stimulating blue and green receptors. (A) Bees cannot distinguish polarity of patterns of two equal blue and buff panels when there is no green contrast at the central boundary, and equal green contrast at the outer edges. (B) Similarly, they fail with blue and buff spots on a grey background. (C, D) They distinguish the patterns when equal black bars are added to each target. (E) Similar panels of yellow and blue, with green contrast, are easily distinguished. (F) Exhaustive tests later revealed the strongest cue to be the left/right polarity relative to green contrast at black edges. (A–D redrawn after Horridge, 1999b.)



Innovation, Deep Thought and Hard Work

Train, fixed 20° buff spot, no green contrast (A)

Blue 595

Buff (B) Blue

Blue 595

100%

55°

80.0% ± 3.6%, n = 200 Test 100%

50% White (C) Blue

31.0%, n = 200

20% White (D) Blue

69.0%, n = 200

Test 100%

Test 100% 25% White

Buff (E)

Blue

53.0%, n = 180 Test 100%

Buff Blue

Buff 66.0% ± 3.1%, n = 200

Fig. 3.8.  Detection of a spot with no green contrast. (A) The bees were trained to a 20° buff spot on a blue background, with no green contrast, versus a plain blue target. (B) They strongly avoided a grey display (50% white) and preferred the unrewarded training target. (C) They preferred a dark grey display (20% white) and avoided the unrewarded blue target, showing that a measure of blue content was a cue. (D) With similar blue content, they failed. (E) Apparently they had learned the spot position (but actually it was the blue position). For data on coloured papers, see Table 3.1. (Redrawn from Horridge, 2012.)

2012). There was a large area of blue on both targets, so blue would be a poor cue. To simplify the experiment, the UV receptors were put out of action by working in the shade and by use of coloured papers that reflected negligible UV.

37

When tested versus a plain blue target the trained bees avoided a light grey target (Fig. 3.8B) but preferred a dark grey one (Fig. 3.8C), showing that they had learned to avoid the target with most blue content. The trained bees were unable to distinguish between the buff spot and a grey one with no green contrast, both on the blue background (Fig. 3.8D). Clearly, they had not learned the buff colour. The trained bees distinguished the training pattern from the same with the spot moved up (Fig. 3.8E). Taken together, these tests show that the trained bees had learned a measure of the total blue content, with no indication of detection of the spot colour. This result makes nonsense of the usual definition of colour vision and the von Frisch test for it.

A spot with no blue difference A similar experiment with no blue contrast and no blue difference gave a similar result (Fig. 3.9) (Horridge, 2012). The blue colour was ultramarine, darker than before, and the hemp yellow was also different, with a large green contrast. The trained bees preferred a grey spot, with no blue contrast, versus the plain blue target (Fig. 3.9B), because green contrast was available. They failed to distinguish the rewarded training target versus a similar spot on a green background (Fig. 3.9C) and they did not recognize a grey spot from a white spot, both on ultramarine background (Fig. 3.9D). They failed to detect a buff spot with no green contrast on a blue background (Fig. 3.9E), because the only cue they had learned was green receptor modulation. The two experiments above were completed in 2006 and not published until 2012. Even then, the interpretation was wrong but now corrected. Work on the discrimination of colour had been delayed by work on black/white patterns, but there was enough data by 2006 to show that a palette of trichromatic colour vision was impossible in the bee.

38

Chapter 3

Train no blue contrast, no blue difference (A) Hemp 374

Ultramarine 590

100%

Ultramarine 590

55°

90.0% – 3.0% (2.0%), n = 220 (B) Grey

Ultramarine 590

Test 100%

Ultramarine 590

76.0% – 2.9% (3.0%), n = 200 green contrast detected (C) Hemp 374

Ultramarine 590

Test 100%

Green

52.5%, n = 200 similar green contrast, blue not learned (D) Grey

Ultramarine 590

Test 100%

Ultramarine 590 White

51.0%, n = 200 similar green contrast, blue not learned (E)

Blue 595

Buff 384

Test 100%

Blue 595

49.0%, n = 200 no green contrast, the only cue

Fig. 3.9.  A hemp-coloured spot on ultramarine background with no blue difference. (A) Training patterns. (B) The bees responded to a grey spot (55% black) with no blue contrast, so they did not need the hemp colour. (C) They could not distinguish the training spot from the same spot on a green background. (D) They confused a dark grey spot with a white one. (E) They could not detect a buff spot on a blue background with no green contrast. Therefore the cue was saturated green contrast at the edge of the training spot. (Redrawn from Horridge, 2012.)

A Strange Case of Achromatic Vision In 1979, Bernhard Ronacher made a remarkable discovery. Bees were trained to discriminate between a black disc 50 mm

in diameter and one 19 mm diameter on a white background (Fig. 3.10A). Training was possible because UV intensity was low. The trained bees were tested with a 50 mm disc in different shades of grey versus a small black disc of various sizes (Fig. 3.10B). As the smaller disc increased in size, the discs first appeared more similar to the bees (i.e. approached 50% correct choices) then appeared more different. For each size of the smaller disc, there was always a grey level of the large disc that made it indistinguishable from the smaller black one (Fig. 3.10B at 50% line). I repeat: the large grey disc was indistinguishable from the smaller black one. This was the father of all anomalies. Of course, the same bees could be trained to distinguish between the discs that had been indistinguishable in the tests. Ronacher took the trouble to thank a great number of professors for their helpful suggestions, but they had not detected the problem. Instead of a thorough testing, new bees were trained with a different size of the small disc and the process repeated, producing a trading curve of all the 50% points for size versus grey level (Fig. 3.10C). Then training was repeated with different sizes of the large disc. A large amount of data was obtained, most of it for dark grey levels. Similar results with different absolute v ­ alues were obtained when other black shapes were used, implying that bees make an additional allowance for shape. It seemed obvious that bees measured two parameters, size and shade of grey, for which preferences were opposite and cancelled when equal. ‘Different perceptual parameters are postulated and treated as coordinates of an abstract vector space (perceptual space)’ (my translation from Ronacher, 1979). The concept was derived from similar studies on humans. Indeed, the trading curve implied that a physical relationship existed, but there were no tests to reveal it. There was a large amount of excellent data but these remarkable results were rarely quoted, being either not noticed, or quickly forgotten. Time passed and



Innovation, Deep Thought and Hard Work

(A)

39

Train in all experiments 50 mm

19 mm

versus

(B)

Test with large disc at various percentages of white 50 mm

Choice of larger disc (%)

100

19 mm Test

3%w

80

Grey disc preferred

15%w 60

50%

40 Grey disc avoided

20 0 0

10

20

30

40

50

Size of smaller disc (mm) Plot size of small disc versus percentage of white to give a trading curve at line of 50% Amount of white in the grey spot (%)

(C)

30 Larger small discs appear less white

20

10 < 50% > 50% 0

10

20

30

40

50

Size of smaller disc (mm)

in 1998, Ronacher revived the identical data with reference to the problem of ‘detecting and processing such characteristic properties of objects that support the correct classification and representation of objects in spite of these ambiguities of receptor responses’ (Ronacher, 1998). After 20 years of thought, he concluded that processing in bees and humans is similar because both use trading parameters, and pay selective attention to certain features. In other words, there was no explanation.

Fig. 3.10.  Bees confused spot size with greyness. (A) Training patterns: a 50 mm black spot versus and 19 mm black spot. (B) Results of tests with a large grey spot of 15% and 3% white (w) versus small black spots of different sizes. Note the 50% levels where different spots were not distinguished. (C) Formation of a trading curve of percentage white in large disc versus size of small disc. (Redrawn from data in Ronacher, 1979.)

The difficulty was that the kind of system in bee recognition was not known at the time. Following von Frisch (and everybody else), it was assumed that bees have ‘achromatic’ vision of black, white and grey levels, although von Frisch had actually found that bees did not recognize a mid-grey level in a panel displaying all shades of grey (Fig. 1.2B). Also, Autrum and von Zwehl (1964) had shown that honeybees have receptors for green, blue, and UV, but nothing for black, grey or white.

40

Chapter 3

How bees distinguish a large black spot from a small one In response to Ronacher’s enigma, bees were trained, also with low UV emission, to distinguish between a large and small black spot (Fig. 3.11A), then tested with two black lines on a white background, showing that they had measured the width of the large spot but not its height (Fig. 3.11B, C). They failed to distinguish when the larger spot was a grey of 50% white, so clearly had not learned to recognize the small spot (Fig. 3.11D). They also failed when tested with a large black spot versus a plain grey target that was 70% white (Fig. 3.11E) so had not learned to recognize the large spot either. They also failed when tested with yellow spots on a green background showing no contrast to green receptors, showing that green contrast was essential (Fig. 3.11F) but succeeded when there was green contrast (Fig. 3.11G). When tested with pairs of gratings equiluminant to blue but with different length of edge, they showed a preference for greater green modulation at the longer edge (Fig. 3.11H), but again failed with no green contrast (Fig. 3.11I). Therefore, in Ronacher’s experiment, the absence of UV allowed the bees to follow changes in total blue content of the background as the spot diameter changed, but other tests show that spot width and length of edge also contributed. The origin of the tipping point is as follows. Trained bees avoided the shorter edge and the smaller width of the smaller spot, and the greater blue content around the small spot. As white was added to the large black spot, its increasing blue content eventually reached the tipping point, and then the small spot was preferred. When we consider other 20th century experiments, it is apparent that training with a black shape on a plain white background versus the plain white background (Wehner, 1969, 1972; Horridge, 2009a, b; and many others) was a recipe for disaster because the initial difference in area of black, and therefore in total blue content, was an unsuspected preferred cue. However, in many experiments, pairs of black/

white patterns displayed equal white areas, so that a blue difference was not available. Trained bees then located and measured green channel modulation, vertical position and average edge orientation (Chapters 6 and 8, this volume).

The Emerging Mechanism of Colour Vision The starting points for further advances were my early discovery in 1966 that crabs used separate detector channels for the memory of positions of black edges and for black areas projected on the eye. Receptor channels for only green and blue were available in the bee because the axons of all six green-sensitive receptors ended at the lamina, as shown by Wakakuwa et al. in 2006. In the fly, second-order neurons on the green channel are adapted to detect contrast, as shown by Laughlin and Hardie in 1978. Also, UV inhibited detection of white, as shown by Hertz (1939b), and green modulation inhibited blue modulation (Srinivasan and Lehrer, 1988). Bees could use green or blue contrast when they distinguished coloured spots on a suitably coloured background, as shown above (Figs 3.8 and 3.9). The paper by Dittrich (1995) showed that trained bees detected a change in input as they walked across a boundary between two colours, but did not distinguish two colours presented simultaneously on neighbouring parts of the retina. The active movement of the bee was apparently essential for an effective discrimination. This was a strong indication that discrimination between two colours depends on seeing a moving boundary. Electrophysiology demonstrated that the neural machinery favours the detection of the modulation at each type of receptor, in that retinula cells adapt rapidly to changes in intensity; the lamina ganglion cells adapt even faster. The large-field neurons of the deep optic lobe are mostly phasic and reveal a great variety of antagonistic inputs from two or three receptor types, weighted in different proportions so that they detect modulation in different



Innovation, Deep Thought and Hard Work

41

Train, 20° versus 8° black spots (A) 100%

69.0%, n = 200 on 2nd day (B)

Test 100%

(F)

50%, n = 100

65.0%, n = 200 (C)

Test 100%

(G)

Test 100%

nbc 100% Test 63%, n = 200

50.5%, n = 200 (D)

ngc 100% Test

(H)

nbc 100% Test

Grey 50% 71%, n = 100

56.5%, n = 200 (E)

Test 100% Grey 70% white 57.0%, n = 200

(I)

ngc 100% Test 49%, n = 100

Fig. 3.11.  How bees actually distinguished the size of spots. (A) Training patterns. (B, C) Trained bees distinguished horizontal widths between vertical test bars but not heights. (D, E) They failed to distinguish because total blue content was similar. (F, G) They failed with spots with no green contrast (ngc) against background, but succeeded with no blue contrast (nbc). (H, I) They preferred the greater length of edge, but with ngc, and equal blue content, they failed. They had learned blue content, spot width and green modulation.

parts of the spectrum. Apart from the blue receptor axons that bypass the lamina, optic lobe neurons are more suitable to detect line-­labelled modulation than a tonic maintained photon flux. Moreover, within the optic lobe there is no sign of the colour triangle, or of neurons that could discriminate between colours. The spatial fields for ­colour-coded neurons are large, often whole eye field, which implies that they are adapted to quantitative measurement of a summed mixture, region by region, but not the identification of a spatial pattern. By 2006, new results implied that much old work would have to be revised, for example the effect of the colour of the illumination

upon another colour, and the discrimination of large areas of colour or black/white patterns. The part played by the UV receptors was also little understood. These results make us again wonder what the bees actually detect when they fly around. There was nothing to show that bees do anything more than use feature detectors and coincidences to recognize a place by cues. The visual input is combined with other senses to build up a total impression of the adjacent environment, which we humans might appreciate as the way that a bat or a ship guided by radar in a fog is aware of the surroundings. When the bees’ behaviour suggested a

42

Chapter 3

higher performance graced by an abstract noun, such as cognitive behaviour, concepts, cognition or detection of abstract properties such as shape, similarity, symmetry,

colour in general or texture, we simply looked for the appropriate tests to demonstrate the same few simple feature detectors and cues.

References Autrum, H. and von Zwehl, V. (1964) Spektrale Empfindlichkeit einzelner Sehzellen des Bienenauges. Zeitschrift für vergleichende Physiologie 48, 357–384. Avarguès-Weber, A. and Giurfa, M. (2014) Cognitive components of color vision in honeybees: how conditioning variables modulate color learning and discrimination. Journal of Comparative Physiology A 200, 49–461. Dittrich, M. (1995) Time course of color induction in the honeybee. Journal of Comparative Physiology A 177, 207–217. Giger, A. and Srinivasan, M.V. (1996) Pattern recognition in honeybees: chromatic properties of orientation analysis. Journal of Comparative Physiology A 178, 763–769. Giurfa, M. (1991) Colour generalization and choice behaviour in the honeybee Apis mellifera ligustica. Journal of Insect Physiology 37, 41–44. Giurfa, M. and Vorobyev, M. (1998) The angular range of achromatic target detection by honey bees. Journal of Comparative Physiology A 183, 101–110. Giurfa, M., Núñez, J., Chittka, L. and Menzel, R. (1995) Colour preferences of flower-naive honeybees. Journal of Comparative Physiology A 177, 247–259. Giurfa, M., Vorobyev, P., Kevan, P. and Menzel, R. (1996) Detection of coloured stimuli by honeybees, the role of chromatic and achromatic contrast. Journal of Comparative Physiology A 178, 699–709. Giurfa, M., Vorobyev, P., Brandt, R., Posner, B. and Menzel, R. (1997) Discrimination of coloured stimuli by honeybees; alternative use of achromatic and chromatic signals. Journal of Comparative Physiology A 180, 235–243. Hempel de Ibarra, N. and Giurfa, M. (2003) Discrimination of closed coloured shapes by honeybees requires only contrast to the long wavelength receptor type. Animal Behaviour 66, 903–910. Hempel de Ibarra, N., Vorobyev, M. and Menzel, R. (2014) Mechanisms, functions and ecology of colour ­vision in the honeybee. Journal of Comparative Physiology A 200, 411–433. Hertz, M. (1933) Über figurale Intensität und Qualitäten in der optische Wahrnehmung der Biene. Biologische Zentralblatte 53, 10–40. Hertz, M. (1937) Beitrag zum Farbensinn und Formensehen der Bienen. Zeitschrift für vergleichende Physiologie 24, 413–421. Hertz, M. (1939a) Versuche über die Farbensinne der Bienen. Naturwissenschaften 25, 492–493. Hertz, M. (1939b) New experiments on colour vision in bees. Journal of Experimental Biology 16, 1–8. Horridge, G.A. (1999a) Pattern vision of the honeybee (Apis mellifera) is colour blind for radial/tangential cues. Journal of Comparative Physiology A 184, 413–422. Horridge, G.A. (1999b) Pattern vision of the honeybee (Apis mellifera). The effect of pattern on the discrimination of location. Journal of Comparative Physiology A 185, 105–113. Horridge, G.A. (2003) Visual resolution of gratings by the compound eye of the bee (Apis mellifera). Journal of Experimental Biology 206, 2105–2110. Horridge, G.A. (2009a) Visual discrimination by the honeybee (Apis mellifera). In: Lazareva, O., Shimizu, T. and Wasserman, E. (eds) How Animals See the World. Oxford University Press, Oxford. Horridge, G.A. (2009b) What Does the Honeybee See? And How Do We Know? A Critique of Scientific Reason. ANU E Press, Canberra, 360 pp. Available at: http://epress.anu.edu.au/honeybee_citation.html (­accessed 1 November 2018). Horridge, G.A. (2012) The anti-intuitive visual system of the honey bee. Acta Biologica Hungarica 63(Suppl. 2), 146–161. DOI: 10.1556/ABiol.63. 2012. Suppl.2.2 Horridge, G.A. (2015) How bees discriminate a pattern from its mirror image. PLoS ONE 10(1), e0116224. DOI: 10.1371/journal. pone. 0116224 Horridge, G.A. (2016) Parallel inputs to memory in bee colour vision. (Plenary Lecture. 31 August 2015, at International Congress, International Society of Invertebrate Neurobiology, Tihany, Hungary.) Acta Biologica Hungarica 67, 1–26.



Innovation, Deep Thought and Hard Work

43

Kelber, A., Vorobyev, M. and Osorio, D. (2003) Animal colour vision: behavioural tests and physiological concepts. Biological Reviews of the Cambridge Philosophical Society 78, 81–118. Komissar, A. (2004) The peculiarities of the honeybee perception of the white and blue flat near-entrance marks. Journal of Apicultural Science 48, 5–11. Laughlin, S.B. and Hardie R.C. (1978) Common strategies for light adaptation in the peripheral visual systems of fly and dragonfly. Journal of Comparative Physiology A 128, 319–340. Lehrer, M. and Bischof, S. (1995) Detection of model flowers by honeybees, the role of chromatic and achromatic contrast. Naturwissenschaften 82, 145–147. Lehrer, M., Horridge, G.A., Zhang, S.W. and Gadagkar, R. (1995) Shape vision in bees, innate preference for flower-like patterns. Philosophical Transactions of the Royal Society of London B 347, 123–137. Niggebrügge, C. and Hempel di Ibarra, M. (2003) Colour-dependent target detection by bees. Journal of Comparative Physiology A 189, 915–918. Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222. Ronacher, B. (1979) Äquivalenz zwischen Größen- und Helligkeitsunterschieden im Rahmen der visuellen Wahrnehmung der Honigbiene. Biological Cybernetics 32, 63–75. Ronacher, B. (1998) How do bees learn and recognize visual patterns? Biological Cybernetics 79. 477–485. Srinivasan, M.V. and Lehrer, M. (1984) Temporal acuity of honeybee vision, behavioural studies using flickering stimuli. Physiological Entomology 9, 447–457. Srinivasan, M.V. and Lehrer, M. (1988) Spatial acuity of honeybee vision, and its spectral properties. Journal of Comparative Physiology A 162, 159–172. von Frisch, K. (1914) Der Farbensinn und Formensinn der Bienen. Zoologische Jahrbücher. Abteilung für allgemeine Zoologie und Physiologie der Tiere 35, 1–188. von Frisch, K. (1971) Bees, their Vision, Chemical Senses, and Language. Cornell University Press, Ithaca, New York. Vorobyev, M., Brandt, R., Peitsch, D., Laughlin, S.B. and Menzel, R. (2001) Colour thresholds and receptor noise, behaviour and physiology compared. Vision Research 41, 639–653. Wakakuwa, M., Kurasawa, M., Giurfa, M. and Arikawa, K. (2006) Spectral heterogeneity of honeybee ommatidia. Naturwissenschaften 92, 464–467. Wehner, R. (1969) Die Mechanismus der optischen Winkelmessung bei der Biene (Apis mellifera). Zoologische Anzeiger (Supplement) 33, 586–592. Wehner, R. (1972) Pattern modulation and pattern detection in the visual systems of Hymenoptera. In: Wehner, R. (ed.) Information Processing in the Visual Systems of Arthropods. Springer, Berlin, pp. 183–194. Wells, H. and Wells, P.H. (1986) Optimal diet, minimal uncertainty and individual constancy in the foraging of honeybees, Apis mellifera. Journal of Animal Ecology 55, 881–891.

Chapter 4 The Fundamentals of the Insect Compound Eye

Scientists have to be rigorous and honest. Such qualities are a useful corrective when scientists feel under pressure to generate particular results, . . . when junior scientists are pressurized by a senior colleague, or by the demands of their career. (Nurse, 2015)

The insect compound eye (Fig. 4.1A) looks a fearsome structure under a microscope, with hundreds of small facets that fill the surface and collect light. The hexagonal facets in the bee are 25–30 μm across, that is, about 500/mm2. When the surface is peeled off, cleaned and placed on a microscope slide, one sees an array of little images of the microscope lamp, or whatever else is in focus. This array led people to imagine that the insect sees an array of similar little pictures, but that is an artefact because the eye surface with many lenses is flattened on the slide, making the optical axes all parallel. As originally pointed out by Hooke (1665) the images, being of a single lens, are each rotated by 180°, so they could not fuse seamlessly together. In the eye, curvature of the cornea creates a two-­ dimensional array of optical axes in angular space over a very wide angle (Figs 4.2, 4.3). The axes are measured with calibrated gimbals (Fig. 4.4). Functionally similar to the mammal eye (Fig. 4.1B), the compound eye is just a way to obtain an image of the surrounding

44

panorama over large angles horizontally and vertically with even coverage. Hexagonal facets allow a maximum lens diameter and detection of motion and edge orientation at all angles, not just horizontal and vertical.

Gross Architecture of Insect Compound Eyes In 1891, Sigmund Exner published a short book on the structure and optics of insect eyes. I think it was promoted by the family firm, because Exner was an uncle of von Frisch, and the Hertwig branch of the family had zoologist professors at Munich and Vienna. Exner described only two main ­ groups of insect eyes, apposition and superposition, but ignored the fly and all the various interesting bugs, beetles, mayflies, moths and dragonflies.

The retina of large day-flying insects The prize for excellent early work must go to Max Sigmund Schultze (1868) at Bonn, who fixed insect eyes with osmic acid and teased out the cells with a needle. He showed that the solid corneal cones of the firefly Lampyris are surrounded by a dense layer

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)



The Fundamentals of the Insect Compound Eye

45

Dorsal

(A)

P Screening pigment

Receptor axons

32 31 30 29 28

Lenses

Nodal points Receptor layer

(B)

34 33

37 36 35

4647 4445 4243 4041 39 38

27 26

Optical

25 24 23

centre

22 21 20 19 18

Anterior

17 16 15

Nodal point

P

Lens Receptor rods and cones

14 13

Best

12 11 10 9 8 7 6 5

Optic tract

4 3 2 1

24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

Fig. 4.1.  An apposition compound eye (A) and a single-lens eye (B) are both devices that sample the light from the surrounding panorama in angular coordinates. The nodal point (P) is a useful fiction, defined as the point through which rays pass when they are parallel outside the eye and at the receptors. Resolution of detail depends on the receptor size and aperture of the lens. Detection of separation depends on the angle between adjacent receptors.

Fig. 4.2.  (A) Vertical section though the retina of a worker bee, showing the layout of the optical axes. (B) Horizontal section. These are not representations of tissue structures. (Redrawn from data in Baumgärtner, 1928.)

of pigment cells during the day, so their ­optics must then be separate. He showed fused rhabdoms of Pieris, the cabbage white butterfly and the open rhabdomeres of a hoverfly (Syrphidae). He even showed the solitary retinula cell at the tip of the cone in the eye of the water beetle, Dytiscus, and the way this cell migrates at night away from the surface towards the layer of much larger proximal receptor cells below (Fig. 4.43). Later texts omit these details. Exner examined a few large insects in detail. He was a physicist who concentrated on the optics but left out most of the other subject matter. Like all German accounts in

the 20th century, Exner writes as though unaware that another German professor, Georg Hermann Grenacher (1843–1923), working at the University of Rostock, had published in 1879 a great volume on microscopic structure of the eyes of arthropods. Grenacher distinguished three types of insect eyes, classified according to cone structure. The most abundant were the acone eyes, with a mobile cone that flattened towards the cornea in the dark, carrying with it seven separate rhabdomeres away from screening pigment; found among beetles, hemipterans, earwigs and lower orders. In daylight, the receptors move deeper, pulling the cone cells into a conical

Posterior Ventral

46

Chapter 4

Hexagonal facets

Receptor axes

ΔϕV 80° ΔϕH

60° 40°

Vertical

20° 0 20° 40° 60° 80°

Anterior 0

20°

Posterior 40° 60° Horizontal

80°

Fig. 4.3.  A map of the visual axes of a worker bee left eye in angular coordinates, constructed with the apparatus in Fig. 4.4, with the eye illuminated from the inside of the head. Each dot represents the angle of an ommatidial axis in two angular coordinates. The region of best spatial sampling is along the horizontal midline. Note that the interommatidial angle in the horizontal direction (top right) gives a false impression of the spatial resolution, and the projection to flat paper causes distortion at the edges. (After Giger, 1996 with original numerical data from Seidl, 1982.)

shape surrounded by pigment grains (Fig. 4.19). Next, he has the pseudocone eyes of flies, where no cone structure is seen except four tiny cells at the bottom of a fluid-filled space. He illustrates seven receptors in a fly, with the typical asymmetrical pattern of rhabdomeres (Fig. 4.18, inset). His third type were eucone eyes of butterflies, grasshoppers, many beetles, bees and wasps; in fact, most of the large

day-flying insects. In them, a transparent cone surrounded by pigment cells meets the fused rhabdom at a narrow point. The corneal-cone eyes of fireflies and many related beetles make the fourth type, eyes with a wide clear zone ­between the cones and the receptors. There are many intermediates and many eyes with unexplained arrangements of receptors at different levels (Tuurala, 1954). Grenacher discussed the importance of the diameter of the rhabdomeres. When they are small an acute eye can detect small dots and fine detail; when large they are more sensitive and useful in dim light, but have less resolution, like the silver grains or pixels in a camera. He also understood the concept that a greater aperture of a facet ­implies greater sensitivity but less ability to resolve separate details like blades of grass or petals (spatial resolution) because there must be fewer facets. All these insights were forgotten until the topic was reinvented by Kuno Kirschfeld in the early 1970s (Kirschfeld, 1974). In each ommatidium in apposition eyes like the bee (Figs 4.5, 4.6B), the transparent cone ends directly upon the distant tip of the rhabdom rod, which is a light-sensitive light guide, that absorbs light, that triggers the cascade of reactions that generate depolarization of each receptor cell membrane. A rhabdom is a fusion of 6–8 rhabdomeres, one from each receptor cell. Apposition ­optics is easy to comprehend in principle because each facet effectively is one pixel in the image. The optics is difficult to describe in detail because at the tip of the cone a narrow beam passes between dense pigment cells and into the end of the rhabdom, which is a light guide (Fig. 4.5), and light guide optics is a difficult topic. Where the light path is more than 2 μm wide (as in Fig. 4.9B), it can be simplified by using classical optics. In these respects, the bee retina is typical of large day-flying insects, such as dragonflies, grasshoppers, mantids, ants, butterflies and cockroaches. In apposition eyes of large day-flying insects each facet accepts rays from a narrow



The Fundamentals of the Insect Compound Eye

47

Camera

Narrow aperture

Goniometer stage

Eye

Fig. 4.4.  Apparatus for constructing a map of the axes of ommatidia in angle coordinates. The eye is illuminated by a lamp behind the camera or inside the head. The pseudopupil is photographed at measured angles along facet lines, with the pupil centre at intervals of five facets in every direction until the eye has been covered. This gives the angular position of every fifth ommatidial optical axis from which every axis can be derived. The worker bee eye is black and does not have a pseudopupil, except in the late pupa.

field about 2° wide. The curvature of the cornea focuses parallel axial rays upon the distal end of the fused rhabdom, which acts as a light guide (Figs 4.5H, 4.6B). The light path in the rhabdom must be long ­because only about 1–2% is absorbed per micrometre. The light with electric vector in a preferred direction, as shown (Fig. 4.6C), is captured by rhodopsin molecules regularly packed in the microvilli (Fig. 4.6A). This makes the ­receptor sensitive to the plane of polarization. The spin-off from fundamental work In 1970, Barry Ninham, a new professor of applied maths in the Australian National University (ANU), phoned me to ask my opinion about an applicant who wanted to collaborate with me on the physics of light guides in insect eyes. As I had just left ­behind in Scotland an excellent project in which we modelled a large locust ommatidium in wax and used radar waves instead of light, I was delighted. The candidate was

Allan Snyder, who was an excellent mathematician, a very imaginative researcher, and soon a great success. Allan invited Kuno Kirschfelt from the Max Planck Institute in Tübingen to work in my department in 1971–1972, to directly observe the light path in the transition region between the cone and the rhabdom in the fly ommatidium (see Figs 4.5H, 4.9 and 4.18). Allan knew little about flies’ eyes, Kuno little about light guides, but each was superb on his own topic. They discovered that the c­entral rhabdomere was so narrow, and the refractive index so little different from the surrounding tissue, that only the first mode of transmission along this thread was possible (Kirschfeld and Snyder, 1975). Effectively, this implied that from the cone into the rhabdomere there was longitudinal transmission along the middle, like a tidal bore (an eagre) advancing into an estuary. Having seen that it was possible, Allan calculated that appropriately manufactured light guides and light pulses would

48

Chapter 4

(A)

(B)

(G) C

C P

C P P

(C)

R

(D) (H)

(E)

9 9

P

(F) 9 R

Fig. 4.5.  The arrangement of cells in the retina of the bee. (A, G, H) Vertical sections through the retina. (B–F) Transverse sections of one ommatidium at different levels. Light passes first though the cornea, then the cone (C), then is focused on the distal end of the rhabdom, which is a long thin absorbing light guide formed by fusion of a component from each retinula cell (R). C, cone; P, pigment grains; R, receptor cell; 9, ninth cell.

behave in the same way, and they would transmit photon pulses (called solitons) faithfully for long distances without loss of energy. Bell Laboratories in New York had given up work on fibre optics in 1971 because Corning’s pure glass fibre transmitted less than a mile. In the same year, Kompfner at Bell Labs discovered the modulated laser, which made possible the transmission of code. Allan was aware of these developments, and was spurred on by the rapidly developing field. He calculated the best design for entry and transmission (Snyder, 1972). ‘From that start, telecommunications was revolutionized with millions of kilometres of fibre optic cable, constructed to Snyder’s

specifications, laid around the world’. This quote is from the address given much later at the award of the Australia Prize to Allan in 1997. The discovery had enormous consequence. In 1962 the IT network had consisted of 24 university nodes on the west coast of the USA; in 1972, the Pentagon went on line; by 1973 there were 36 nodes, but all still connected by telephone wires. By 1975, the light guide network was established, first to Norway, then further until it became the World Wide Web. Standard intercontinental e-mail protocol was established by 1982. Allan went on to make more discoveries in optics and mathematics, and was elected to the Royal Society in 1990.



The Fundamentals of the Insect Compound Eye

(A)

One microvillus R R R R R R R R

Central filament

High e R R R R R R R R

absorption Low e absorption

(C)

Light ray

e

(B) m Direction of light ray

One receptor cell

e m e m e m

Axon

Rhabdomere rod

e

Fig. 4.6.  Light is absorbed in rhodopsin (R) (the visual pigment protein), which is packed into each microvillus in an orderly way. (A) One microvillus, and the preferred direction of absorption of the electric vector of light. (B) The light path down the rhabdom can be considered as internal reflection, but below 2 μm in diameter light guide optics apply. (C) A light ray can be considered as alternating rings of electrical (e) and magnetic (m) energy, with the electrical one defining the plane of polarization.

The performance of the bee eye In their performance, bees’ ommatidia compare with human eyes, but are much smaller, so they catch less light. However, this does not make them less sensitive to light, it just ruins the resolution because the light captured by each receptor is a solid angle about 2.5° wide, which is 100 times that for human rod receptors. Sensitivity is the critical factor for the honeybee; if sensitivity were lost, the bee would not see at all. Sensitivity depends on the F number, which is the ratio of focal length to aperture (f/D in Fig. 4.9A). In fact, bee sensitivity is about

49

the same as in man, because the F number in man and bee lies between about 4 and 10, as in all typical diurnal eyes and cameras that operate in the range between bright and dim daylight. Because eyes, cameras and telescopes are optimized to make the best use of the light and are well focused, they have similar F values, and resolution of detail is inversely proportional to aperture. The aperture of a single facet in the bee is about 25μm compared with 2.5 mm for the human pupil, about a 1000-fold difference. As a result, the minimum period of a grating that the bee can detect in bright light is about 2°, compared with 0.02° in man. An important feature of honeybee vision is the very rapid rate of adaptation to light or dark, a tenfold change in less than a second. Rapid independent adaptation of each receptor channel is essential for the measurement of blue content and also of green modulation irrespective of changes in light intensity and of the angular velocity of the image relative to the eye as the bees scan in flight. The range of intensities over which human eyes function, from sunlight to moonlight, is about 10 million times. For bees, it is only about 1000-fold so bees must go home at sunset; perhaps it makes them bring their nectar home at bedtime. There are other bee species that fly in forests, or even at night, with eyes that have a much larger receptor response to each photon. Honeybee eyes are notably insensitive when light adapted. Over a 1000-fold range, the acuity of the bee eye falls until, at about the lowest intensity limit for flying, only stripes of period 20° can be resolved. As demonstrated by effective early radar carried in fighter aircraft, such poor resolution is still very useful if it is the only input available. Bees are not good at detecting changes of brightness as they scan, a point to remember when discussing vision of colour. In bright light, the least noticeable change in brightness for a bee is about 20% at best, but for man is about 1.5%. Near the lower limit in dim light for each species, it is 5% in man but over 100% change in the bee. It is usually assumed that the green and blue-­ sensitive

50

Chapter 4

(A)

Sky analyser

(D)

180°

X

90°

270°

(B)

(E)

(C)

100%

Sensitivity

90°

8% 7°

0° 180°

X

Sun position

270°

2%

30°

0°

Fig. 4.7.  The dorsal rim organ of the honeybee eye. (A) The position. (B) Orientations of the orthogonal pairs of the UV receptor cells. (C) The plot of the angular sensitivity to the plane of polarization has a wide skirt that integrates over a large part of the visual field. (D) The pattern of the e-vector in the blue of the sky, at right angles to the direction of the sun. (E) The orientations of the (central) detectors inferred from the behavioural experiments. (Redrawn from Labhart, 1988.)

r­ eceptors respond similarly when adapted to white light (sunlight), and white light was used for the measurements summarized above (Hecht and Wolf, 1929). Green and blue-­ sensitive receptor types adapt independently to colour, and contrasts at edges change their appearance when backgrounds change, but less so when the ambient light shifts in colour (see Chapter 2, this volume; colour constancy). Sensitivity to polarized light Sensitivity to the plane of polarization was a confused topic for decades. In 1911, Santschi, in Zürich, had shown that ants were sensitive. Earlier workers had found that the bee electroretinogram was insensitive to the plane of polarization, and inferred there was no analysis of the plane of polarization within the eye, causing confusion. At the same time, von Frisch’s group (1960) found that bees could use the polarization

pattern in the blue of the sky to direct their foraging route, and proposed that the mechanism lay in the orientation of the microvilli in the rhabdoms of the compound eyes. There was much discussion about whether the rhabdomeres were twisted or not, until finally the specialized detectors (Fig. 4.7A, B, E) were discovered, first in crickets, then in the bee (Labhart, 1980, 1988). The line of specialized ommatidia along the dorsal rim of each eye responds to the axis of polarization of the blue and UV sky light from the blue of the sky. This little organ acts as a collector that brings the spatial pattern of the sky polarization into the central complex in the brain (Homberg et al., 2011), and the bee can determine the position of the sun when it is obscured by a cloud or tree. Maybe the bee must rotate itself to pick up the maximum and minimum of the integrated e-­vectors. The polarization pattern by itself is ambiguous in the middle of the day because the solar half of the sky and the anti-­ solar half are similar (Fig. 4.7D). Therefore,



The Fundamentals of the Insect Compound Eye

with the sun behind clouds, the intensity gradient across the blue sky is also essential. Even with this equipment, bees tend to have a break from work at solar noon. At least one receptor cell in each ommatidium in the main part of the compound eye is also sensitive to the angle of orientation of the polarization plane of light falling on the eye, showing that the supposed cancellation of polarization sensitivity by a twist in the axis of the ommatidium is not complete. In a single receptor there is no way to separate the effects of polarization, intensity, colour and angle of incidence of the light, because in each receptor they are simply added. The idea of representing a separate kind of vision by detecting polarization, and then representing it as colours, ignores this ambiguity of inputs, and the fact that bee vision of areas is monochromatic in blue. The UV cells in each ommatidium are the presumed inputs for the righting response in flight and for providing the optimum exit direction in the escape ­response. When the sun shines, it generates a polarization pattern in the sky, which, together with the sun’s position, is the primary compass for bees (von Frisch et al., 1960), but the intensity and colour gradient across the sky in the morning and evening is used when the sky is overcast, although e-vectors are still detectable below cloud. The bee uses the sky compass to maximum precision by uniting the broad distribution of UV polarization axes across the visible sky (Fig. 4.7D) in conjunction with the position of the sun (Rossel and Wehner, 1987). Green receptors Green receptors have peak sensitivity near 550 nm, which is within the strongest photon flux of the spectrum of the sun (Fig. 4.8A, B) and corresponds to the strongest reflectance from green foliage, therefore maximum contrast against shadows at edges. They carry the heavy load of detecting image structure as the bee scans in flight, which is perhaps why there are six of them in each ommatidia, presumably serving different channels. The axons of all are short, ending in cartridges of

51

the lamina (Wakakuwa et al., 2006). At least two of them that detect direction of motion of a contrast each make many synapses with the initial axon segment of a second-order cell. Bees also detect something related to the total amount of modulation moving across the eye. This input is integrated over each foraging journey to measure distance travelled (see Chapter 9, this volume). These lamina synapses adapt rapidly, adding to the adaptation of the receptors, so that from this point on, all green-­ sensitive channels are first-order derivatives of the input and detect contrast as the eye scans, not the intensity of a continuous signal. The moving edge is the stimulus, so that besides the contrast across the edge, the signal is summed along the length of the edge. Right from the start, therefore, the signal in the receptor includes structure inseparable from contrast. Scanning by the eye is predominantly in the horizontal direction, so vertical edges are dominant. With green receptors alone, the bee can resolve a grating of period 2°, which is near the limits of the optics and the spatial sampling of the retina (Fig. 4.3). Blue-sensitive receptor cell The blue-sensitive receptor cell in each ommatidium has a long axon, that carries a (non-­ spiking) tonic voltage signal, into the middle layers deep in the medulla of the optic lobe. A tonic signal does not rapidly adapt or decay as the stimulus persists, and therefore can measure the emission from a reflecting area as the bee scans in flight. Bees are sensitive to the vertical position of blue, showing that numbers of these axons from each horizontal row of ommatidia, sum upon horizontal fibres in the medulla. The summation is presumed to measure blue content over an area covered by these connections. Bees also make use of phasic components of the blue input as the bee scans across edges, as shown by detection of edges and measurements of width by blue channels in patterns equiluminant to green receptors. The responses of bees to displays in training experiments show that the accurate measurement of blue content by blue receptors is sufficient for efficient discrimination of the range of flower colours relative to a

Chapter 4

(A)

Light intensity (photons/cm2/s/nm)

52

Midday sun × 10–12

200

At sunset × 10–9 100 At 90° to sun × 5 × 10 –10 300

400

500

Blue

Green

600

100 Relative sensitivity (%)

(B)

60 Ultraviolet

Threshold difference (nm)

0 (C)

25 17 nm

Worker bee

15 5

8 nm

300

400

5 nm 500 600 Wavelength (nm)

Fig. 4.8.  Colour and receptor data. (A) Spectral distribution of intensity in daylight, in photon units (photons/cm2/s/nm). (B) Normalized spectral sensitivity of the three types of receptors in the honeybee (from Autrum and von Zwehl, 1964). (C) The threshold minimum difference in wavelength that is best discriminated at wavelengths corresponding to the greatest slopes of the blue receptor curve, while the UV and green cells do not participate (redrawn from von Helversen, 1972).

green background illuminated by the same source (the sun) (see Chapter 5, this volume). Functions of the single UV-sensitive receptor cell The functions of the single UV-sensitive receptor cell in each ommatidium in the bulk of the compound eye of the bee are not yet known to be connected to memory. Many insects use UV as the direction in which to escape when disturbed, so upwards-directed ommatidia and UV sensitivity go together. Honeybees use the direction of the UV of the sky to maintain a flight posture, and somersault when illuminated by UV from below: for example, when they fly over a mirror. In many crepuscular insects and

those that mate or catch prey in flight, the dorsal part of the eye is more UV sensitive and often has larger facets: for example, in drone bees, owl flies, dragonflies and male Ephemeroptera. The use of UV doubles the resolution of small black targets seen against a bright sky, by halving the wavelength. Aquatic arthropods usually have some UV receptors. Maybe this was the primitive use of the dorsal part of the eye and UV ­receptors. Foliage and flowers reflect little UV, and it is doubtful that UV assists in selection of flowers during foraging because bees distinguish blue so much better (Fig. 4.8C). Lotmar (1933) found that blue receptors were always involved when the stimulus was UV. In fact, UV inhibits the detection of blue or white and repels bees, so that they refuse to



The Fundamentals of the Insect Compound Eye

land. For this reason, white flowers rarely reflect UV. On the other hand, some yellow and red flowers, such as poppies, reflect UV, but these colours appear black to bees and it is more likely that they are ­detected by blue receptors sensitive to UV and strong green contrast against a green background. In the usual training method, the differences in positions and measures of green contrast and blue content are quite sufficient for discrimination of any flower colour or pattern one can imagine on a green background.

have a cell sensitive to yellow that improves the detection of a yellow flower or a rival male. Dragonflies identify prey and have more than three receptor colour types. There is no evidence that these long wavelength receptors are connected to memory or combined with others to discriminate hue. The worker bee, however, is sexless and colour blind like many other herbivores, with emphasis on detection of green contrast at edges.

Compound eye resolution

Single yellow or red receptor A single yellow or red receptor (always cell 8?) occurs in some insects that catch prey, lay eggs on special food plants or when the ­females display markers to attract males, for example some wasps, flies, butterflies and many others. Some flies (e.g. housefly and dronefly) recorded by Roger Hardie (Horridge et al., 1975, 1976) (A)

Following ideas based on resolution of human eyes, for almost the whole of the 20th century the orthodox view was that the resolution of the compound eye was fixed, not by the field of view of a single receptor (Δρ), as in Fig. 4.9, but by the angular separation of the visual axes (Δϕ) as in Fig. 4.10. In vertebrate

(B)

(C)

D

n.p.

f

53

Lens aperture D Nodal point

d3 /f

d2 /f

d1 /f

Rhabdoms

Blur circle λ/D

d1

Blur circle

d2 s1 Δρ1 ϕ

d3 s2

s3

Δρ2

Δρ3

ϕ

Fields

ϕ

Fig. 4.9.  The effect of the of the rhabdom width d (μm) on the field of view of a single receptor (Δρ). (A) Rays focused on the distal end of the rhabdom produce a blur circle (Airy disc) which subtends an angle of λ/D at the posterior nodal point (n.p.), where λ is the wavelength of the light and D is the aperture diameter of the lens. The field of the receptor is produced as the blur circle moves across the end of the rhabdom, which subtends d/f radians (where f is the focal length) at the nodal point. (B) The optics is near optimal when λ/D = d/f. (C) The wider rhabdom increases sensitivity (s as in s1, s2, s3 is relative sensitivity) but reduces resolution of detail, which depends on the F number, f/D.

54

Chapter 4

Δρ

Two fields

Δϕ

fH

fV

Fig. 4.10.  Two angles that define different aspects of resolution; the field width of each ommatidium, Δρ, which limits detection of detail, and the ­interommatidial angle, Δϕ, which limits the ability to separate adjacent detail. In daylight, the optimum ratio is near Δϕ = Δρ. Waveform recording

Preamplifier

Light guide

Lamp

Filters

Electrode Micromanipulator

Eye

Vertical arm

Shutter control

Horizontal arm

Fig. 4.11.  Apparatus for measuring receptor responses and angular and spectral sensitivity of single receptor cells. The insect head (here greatly enlarged) is mounted on a micromanipulator which also holds the microelectrode. The stimulus comes from a pinhole at the end of a light guide which comes from a lamp with filters and shutter on a separate support. Electrical responses are detected and recorded top left.

eyes the receptors lie side by side so that their fields do likewise. In compound eyes, however, the fields can overlap, or be adjacent, or be separated so the panorama is sampled with redundancy, optimally or sparsely. The actual arrangement often gives the optimum compromise between sensitivity (achieved by overlap of fields and larger eyes) and spatial resolution (achieved by large lenses and small individual fields). In general, the smaller the eye the poorer the resolution of spots and bars, but the sensitivity of small eyes is often excellent because the focal length is very small and the F number is a ratio independent of size. Baumgärtner (1928), a student of von Frisch, measured the interommatidial angle of the bee in histological sections (Fig. 4.2). This orthodox astigmatic account was questioned by Friedlaender (1931) but she was ignored. The accepted data Δϕ was in error all through the last century, and misled several significant efforts, including Srinivasan and Lehrer (1988) and Giurfa et al. (1997). The raw data from which the map was calculated (Fig. 4.3) is in the unpublished PhD thesis of R. Seidl (1982), which was replotted in the ANU PhD thesis of A. Giger (1996), and in turn was replotted as contour maps by Land (1997). This is just a minor example of the difficulty of rediscovering correct data that was ignored To complicate matters, unlike the human eye, resolution of individual features depends on the field size of the feature detector. Modulation is the smallest at about 2° for the bee, independent of the interommatidial angle, orientation is 3–4°. A minimum small spot detected by green modulation is 5°, detection of blue needs a larger patch, and detection of spokes or circles depends on spatial summation of many detectors of orientation (Chapter 6, this volume).

Direct measurement of receptor sensitivity and resolution In the early 1960s, at the start of my career, it became possible to record from single cells with a microelectrode (Figs 4.11, 4.12).



The Fundamentals of the Insect Compound Eye

(A)

55

(B) –0.78 –1.04

50% linear

–1.42 –1.78 –2.16

10 mV

–2.46 –2.91 –3.22 –3.76 –4.26 6

5

4

3 2

1 0 1 2 3 4 5 6

Fig. 4.12.  (A) Successive depolarizing responses of a photoreceptor to brief flashes from a pinhole in a constant position on an axis, and of calibrated intensity on a logarithmic scale. (B) Responses of the same photoreceptor to brief flashes of constant intensity, as the point source was moved through the field of the receptor (scale in degrees). The field of the receptor in linear coordinates can be plotted from the response heights in (B) transformed with the anti-logs of the values in (A).

Ocelli Eye

Corpora pedunculata

Optic tubercle

Head

Retina Medulla Lamina Lobula Cerebrum

Thorax

Abdomen

Thoracic ganglia

Suboesophageal ganglion

Fig. 4.13.  A distorted diagram of a bee with the enlarged head on the left, showing the principal ganglia of the nervous system.

56

Chapter 4

LAMINA PROTOCEREBRUM

HORIZONTAL FIBRES

LOBULA

Motion feedback (inhibitory gate?)

MEDULLA

Lamina Second order neurons detect mainly modulation irrespective of steady intensities in each map. Medulla Separates local inputs into different local combinations, with some antagonistic interactions. Sums blue content over local areas. Detects spatial and temporal contrasts with exact timing. Detects coincidences between feature detector. Lobula Colour-blind large-field neurons detect direction and change of global motion in different vertical and horizontal seletions of the whole flowfield. Others detect average edge orientation in local regions or small object location within large fields and coincidences of cues in local regions. Neuron combinations form signatures that are related to the visual world and to visual tasks. Weighting of outputs from visual processing. Path to memory activated by a failure of search. Cerebrum Selection of combinations appropriate for orders to the motor centres, with weighted inputs to the ventral cord neurons.

LOCAL THORACIC NEURONS

NECK CONNECTIVE

MOTOR NEURONS

VISUAL FEEDBACK from movements caused by locomotion, opticflow fields, saccades and fixation.

RETINA

The visual world contains values of intensity, wavelength and polarization at each angle; distributions of contrasts, motions, features, cues and coincidences between all these. Retina F2 lens and wavelength 0.5 μm focused on receptors. Two-dimensional array of receptor cells. Measures intensity change, and separates angles, colours and polarization into maps.

Thoracic neurons Selection of appropriate action by context. Formation of action sequences that are initiated by descending combinations together with local sensory inputs.

Motorneurons Control muscle sequences in patterns.

MUSCLES

Fig. 4.14.  An artist’s impression of the connections of the optic lobe of the bee, with a summary of what happens at each level.



The Fundamentals of the Insect Compound Eye

There was a burst of activity on the bee and fly by the Autrum group at Münich, and another at St Andrews, Scotland, on locusts, water beetles, drone bees and other insects.

57

Retina LMC body Lamina columns

Gross Structure of the Optic Lobes Behind the retina is a series of neuropile layers, the lamina, medulla and lobula with decreasing numbers of channels as we go towards the cerebrum (Figs 4.13, 4.14, 4.15). The optic lobes are connected together and to the central body via optic tubercles, with a very small number of axons to the mushroom bodies (Fig. 4.16), implying that the visual memories are stored in the optic lobes, which issue ‘turn left/turn right’ or ‘approach/avoid’ commands.

Medulla columns

Lobula

Lifestyles and eye geometry are closely related

Tracts to protocerebrum and other side

Different lifestyles of insects led to a huge range of differences in the gross architecture of their compound eyes (Fig. 4.17). This topic in my view was so distorted by the heavy tradition of following Exner, while other excellent contributions were so ignored, that I feel impelled to present the recent work.

Retina

Optic tubercle

Lamina Medulla

Fig. 4.15.  A summary of the principal neurons in the regions of the optic lobe of the honeybee, and their connections, from the remarkable work of Cajal and Sanchez (1915). An immense amount of further detail of the neuron anatomy has accumulated since that time, but the way it functions is only slowly being discovered neuron by neuron. LMC, lamina–medulla cell.

Three ocelli

Central body Calyces of the mushroom bodies

Lobula Olfactory-globular tracts Antennal lobe Fig. 4.16.  An artist’s impression of the layout of the bee brain showing principal neuron tracts. Connections from the lobula to the optic tubercles apparently end at the central body, which initiates appropriate responses. Connections from the medulla to the calyces of the mushroom bodies are less certain, and probably carry orders such as ‘approach’ or ‘avoid’, but not visual information. Connections between the two sides carry mainly information about perceived relative motion of the surroundings. (Based on Hertel and Maronde, 1987.)

58

Chapter 4

(A)

(B) Dim light

Bright light

(C)

Fixed range strike

(D)

Fovea

(E)

Fixed direction strike Fig. 4.17.  Different arrangements of axes of ommatidia evolved for different lifestyles. (A) In bright light, facets can be small and more numerous, but sensitivity is less. (B) The opposite holds in dim light ­environments. (C) A dorsal fovea, with larger facets, smaller interommatidial angles, and more resolution of detail, is useful for detecting a prey or mate overhead. A forward-looking fovea is found in dragonflies, asilid flies that hunt flying insects, and many hunting wasps. Lateral foveas occur in some dragonflies. (D) Using two eyes to triangulate, the mantis strikes in different directions at a fixed range. (E) Using two eyes to triangulate, dragonfly larvae strike in a fixed direction at various ranges. Three parallel rays excite one cartridge 4 4 4 3

5

3

2

5

Cone cell Cornea

Retina Receptors Section 7, 8

Second order neurons

3 4 5 2 6 1 Section through ommatidium

Lamina cartridges 4

3 7

8

5

Fig. 4.18.  Neural superposition by bringing together terminals of six receptor axons to one post-­synaptic column of the lamina increases sensitivity without loss of resolution in eyes of some flies, perhaps all Brachycera. Note the tiny cone cells and clear fluid-filled space between them and the cornea (Horridge, 2005). The connectivity is very accurate (Meinertzhagen, 1976).



The Fundamentals of the Insect Compound Eye

Night

Day

Cone cells

Cornea P

59

P Principal pigment cells

Crystalline tract Accessory Dorsal pigment cells 4 5 3 Lateral Medial 6 2 1 Ventral (A)

(B)

Eight axons

Fig. 4.19.  In the acone eyes of Hemiptera, a diurnal rhythm of receptor and cone movements causes large changes in sensitivity and resolution. P, nodal point. (A)

(B)

Cornea F

Fig. 4.20.  (A) Artist’s impression of Exner’s image of a church tower taken through the eye of a firefly. (B) Exner’s diagram of rays in a superposition eye. F is the focal plane. Much was left to the imagination of the reader. The image was at the wrong level. The rays were hypothetical. In real eyes, rays that arrive obliquely at each facet in a parallel bundle never emerge as a perfectly parallel bundle, as shown in results of ray tracing (Figs 4.28, 4.30, 4.38).

60

Chapter 4

Fly and bug: two other types of insect compound eye Just in case you consider the bee eye to be typical of insects, I go out of my way to remind the reader of the immense variety of insect retina anatomy, which presumably is the result of natural selection for many ­unknown behaviour patterns. Eyes of Odonata, orders from Blattoidea to Phasmatodea, Lepidoptera and Hymenoptera mostly have solid apposition eyes like the bee, but the rest are a very mixed bag. Apposition eyes like those of the bee (Fig. 4.5) are typical of large day-flying insects that rely on sharp vision and are inactive at night. They are the best studied but they are not typical of all insects. In fact, there is no one typical insect vision because they have been one of the stress points and sources of diverse behaviour throughout the arthropods (Land, 1989). In pseudocone eyes of flies, including Drosophila, the rhabdomere of each receptor cell (numbered 1–6 in the inset Fig. 4.18) is separate from the others, and the axons form an intricate rearrangement so that they superimpose coincident parts of the image from six neighbouring facets, and increase sensitivity (Fig. 4.18). These six are green sensitive and have axons that end at the lamina. The central two are No. 7 UV and No. 8, blue sensitive, and have axons running deep into the medulla where they sum upon long tangential neurons. Acone eyes are characteristic of Hemiptera, and found among beetles, earwigs and some lower orders. They are little studied optically or with microelectrodes because the receptor cells move as the light intensity changes (Fig. 4.19). In the giant water bug Lethocerus, in the light-adapted day eye, the eight retinula cells have the same optical axis because there is a short light guide between the cone tip and the rhabdomeres. In the dark-adapted night eye, the extensions of the cone disappear and the rhabdomeres (30  μm wide) are exposed to the corneal aperture (50 μm wide). The individual rhabdomeres can be seen from outside the eye, and it was possible to measure the angles of 7–15° ­between adjacent axes, and show neural superposition (Ioannides and Horridge, 1975).

Clear Zone Eyes Clear zone eyes are characteristic of moths and many beetles that fly in the evening or at night. In all of them in the dark-­adapted night eye, there is a wide transparent space between the cornea and the receptor layer. I preferred to call them ‘clear-zone eyes’ because there are so many varieties and the superposition optics cannot be assumed, justifiably, as will be seen. Later, the great variety was described by electron microscopy, the optics of light guides was clarified, and receptor responses were measured with microelectrodes. Superposition in clear zone eyes In the dark-adapted superposition eye, light from a distant source passes through many facets and is concentrated on a small area of the rhabdom layer (Figs 4.34, 4.36). In most of them, in the day eye, pigment cells migrate into the clear zone and surround each light guide that runs from the cone tip to the thick layer of rhabdoms that form the main retina. In these eyes, in bright light, each ommatidium works as an isolated unit, and functions like an apposition eye. However, in dorsal eyes of male mayflies (Ephemeroptera), there are no pigment movements; cone cell and retinula cell extensions act as light guides across the clear zone, and there is no superposition at all. In eyes with true optic superposition, the curvature of the outer surface of the cornea and the graded refractive index in the cornea act as a lens focused inside the cone, where the graded refractive index acts as a second lens to form a telescope variously well focused. In the dark-adapted night eye, the cone tip and clear zone are free from pigment, so that rays from each single point in external panorama pass though many adjacent facets and are concentrated to a ­ small area on the rhabdom layer below (Figs 4.34, 4.36). Superposition eyes occur in many moths, skipper butterflies (Hesperiidae) and, famously, in fireflies (Lampyridae, which are cantharid beetles). Moths and butterflies have crystalline cones like those in most day-­ flying insects but in fireflies



The Fundamentals of the Insect Compound Eye

61

the water beetle eye. ­Similar lens cylinders occur in the dung beetle Anoplognathus. Despite the lack of confirmation, the belief in focused superposition and usefulness of the The Exner contribution erect image in fireflies was strongly supported in the orthodox papers and texts for another 50 In 1875, the Professor of Physiology at the years, while the wonderful superposition eyes University of Vienna, Sigmund Exner dis- of water beetles, skipper butterflies and some covered an erect combined image behind diurnal moths were unexplored. Next, Sigmund Exner ignored Grenachthe retina in a water beetle, Hydrophilus sp., but he knew nothing about lens cylinders at er’s description of the transparent crystalthat time. With a new design of refractom- line tract of high refractive index running eter (Exner, 1889), and help from his brother, from the cone tip to the rhabdom behind Karl, Professor of Physics at the same uni- every facet of the eye in the firefly, a neuropversity, he demonstrated that the non-­ teran and a couple of moths. Finally, he did homogeneous cornea over each ommatidium not know that the receptor layer of the fireof Hydrophilus was a lens cylinder with fly is double. In each ommatidium, there are ­refractive index gradient from 1.545 outside two receptor cells (cells 7 and 8) with thin to 1.565 inside (Exner, 1886), from which flat rhabdomeres at the end of the light the optics was inferred. These results on guide, below which is a second layer of water beetles were not mentioned in his book thick solid proximal rhabdom formed by six in 1891, where he published a picture of a long vertical receptor cells (cells 1–6) in church tower (Fig. 4.20) taken though the eye each ommatidium, with no optical separof a firefly, thereby creating several problems ation between them (Horridge, 1969). In this that persist in the literature. First, he had deep rhabdom layer, the lack of a separation cleaned out the cornea of the firefly so that of the rhabdomeres by pigment or tracheae pigment cells in the cone region were lost. means that absorption of oblique rays inI  imply that the image may be not so well creases sensitivity but reduces resolution. Exner observed lens cylinders in water focused through the mass of cells in the living eye. The cone tip is 0.3 mm from the outer beetles in 1886 and an image in firefly in surface. The image was 0.23 mm behind the 1891, but the night eye is specialized to decone tips, and the important point is that the tect a flashing point of light with maximum thick receptor layer extends between 0.26 sensitivity, not an image. The image caused mm and 0.43 mm from the cone tip, and a decades of confusion because, following Exner, it was believed that insects detect an fine focus cannot be formed on a thick layer. The observation of the lens cylinders image that must be reconstructed in the had been made on the water beetle not the optic lobes, otherwise it would have no firefly. Exner never claimed to have observed function. After pages of discussion on this lens cylinders in the firefly. In 1969, I found point, Forel (1908) came to no conclusion no lens cylinders in the eye of the firefly Pho- about the image, because he knew that bees turis, and in 1962, Kuiper found none in the detect detail but cannot recognize patterns, firefly Lampyris. Seitz (1969) found a gradi- or fly though a coarse net with large holes. Exner (1891, p. 145) inferred, but was ent in Lampyris and traced four rays but clearly could not make a convincing case for careful to say that he did not observe, the focused superposition. Exner illustrated a existence of an afocal combination of two series of layers in the firefly cornea that did lenses in each cone. At first, he thought that not fit at all with the gradients of a lens cylin- the curvature of the cornea was the first lens der. We confirmed this layered structure by and the cone tip was the second, and that electron microscopy in 1959. In Canberra, in the two formed a confocal pair that made our series of papers on this topic (Horridge, telescope optics and a combined image. 1969, 1971; Horridge et al., 1972), we were By experimenting with small light sources unaware of the marvellous lens cylinders of he found that the ‘Das Resultat ist wieder and some other groups of beetles, the cones are deep extensions of the cornea.

62

Chapter 4

ein Strahlencylinder, der demselben Weg ­zurückgeht, den er, von dem beleuchtenden Lichtpunct kommend nach dem Auge gegangen war’ [‘layering of the cone formed a lens cylinder so that the light was bent back towards the direction from which it had come’] (Exner, 1891). All this was inference. There was no geometrical tracing of ray paths with a pencil of parallel light. Karl Exner was a good physicist and Sigmund was convinced that there was no alternative. At the time, that was acceptable. As his diagrams show, Exner did not observe the final position of screening pigment in the light-­adapted eye, as shown on the left side of Fig. 4.21. Doubts were soon raised in an excellent histological study of pigment distribution in the light- and dark-adapted eye of many insects. Kirchhoffer (1908) carefully described the daily movement of pigment grains in the Lampyris eye, cell type by cell type. He illustrated six deep and long fused rhabdoms of each ommatidium (cells 1–6), and two small retinula cells with separate rhabdomeres (cells 7 and 8), and clearly took a dim view of Exner’s histology. Among his examples, he noted that superposition is absent in beetles that are active only by day, except tiger beetles (Cicindelinae) that have the typical pigment movements of nocturnal groups, not explicable by Exner’s theory. He continued: ‘Es liegt mir ferne, auf Grund dieses einen gegen die Richtigkeit sprechenden befundes, die ganze Theorie vom Zwecke der Pigmentverschiebung angreifen zu wollen’ [‘Far be it from me to attack the whole explanation of pigment movements because of one exception’] (Kirchhoffer, 1908). This is a subtle dig at Exner, who founded his whole theory on the basis of one species, Lampyris. Details of light guides, double retina and dark/light adaptation in the firefly retina, work by Kuiper and Kirchhoffer, and studies of firefly light guides were omitted by Kunze (1979) and Autrum (1981) in their orthodox reviews. All early authors, except Exner, were aware of extensions of the cone running as threads across the clear zone in most nocturnal insects, so that a large area of cornea was concentrated on a small volume of

(A)

(B)

(C)

LA

DA (D)

(E)

(F)

(G) Fig. 4.21.  Details of an ommatidium of the firefly Photuris versicolor (Lampyridae). In the light-adapted (LA) day eye on the left side, four cone extensions guide light through the pigment screen to the distal (E) and proximal (F) rhabdomeres. Light arrives first at the two distal rhabdomeres of cells 7 and 8 (E). At night, on the dark-adapted (DA) right side, the pigment grains move to surround the cone (right side of the figure) and the six large proximal rhabdomeres (F) are exposed to light crossing the clear zone. The depth of the proximal receptor layer increases sensitivity but, being thick, it could not resolve an image. (From Horridge, 1968a.)

r­eceptors nearer to the centre of curvature of the whole eye (Fig. 4.22). They have a high refractive index and act as light guides (see below). In the firefly, they separate at the level of rhabdomeres 7 and 8, as if to spill out light there. Light guides went out of fashion after Exner’s descriptions of superposition optics, but there is no doubt that crystalline tracts are realized in the eye of the firefly (Figs 4.21, 4.22). Kirchhoffer was forgotten



The Fundamentals of the Insect Compound Eye

(A)

(B)

(C)

25 µm Fig. 4.22.  Movement of pigment grains in the eye of the firefly Photuris. The extensions of the cone cells act as light guides to the two layers of rhabdomeres and receptor cells. (A) Light-adapted day eye. (B) Partially dark-adapted state, with pigment cell nuclei on the move. (C) Dark-adapted, with pigment grains between the cones. The rest of the structures do not move. (From Horridge, 1969.)

and this knowledge lost because Exner’s book became orthodox doctrine. Exner pointed out the difficulty of imagining the halfway stages in the evolution of a superposition eye, in that the structure would only be of advantage when perfectly evolved. He did not know that in all insect eyes, extensions of the cone cells reach the basement membrane and hold together the retina. Light guides crossing the clear zone in primitive insects (see below), provided an alternative light path long before optical superposition evolved in several insect groups. When I wrote Bullock and Horridge (1965, p. 1067), I refused to support Exner’s ideas. I  acknowledged that the superpos-

63

ition image had been observed, but there was no solid evidence for the curved ray paths that he proposed; and then: ‘Except for the example of Lampyris, there seems to be no reason to suppose that the superposition eye is anything more than a device to collect a greater amount of light by having a larger collecting surface available.’ I listed seven features that supported this view. My source was the research group in Gröningen, where Kuiper told me that the cone of Lampyris is laminated with growth layers, but is homogeneous under an interference microscope (J.W. Kuiper, University of Gröningen, 1961, personal communication). He later confirmed this statement (Kuiper, 1962, 1964). Later, Carricaburu (1975) shows interference microscopy of sections. Exner illustrated layers, but his figure does not agree with his supposed distribution of refractive index. In 1974, I suggested three ways of forming a superposition image, and surveyed our work on skipper butterflies, which have excellent true superposition. For eyes that were not yet analysed, I used the neutral term ‘clear-zone eyes’ because it did not assume a function for the clear zone or assume superposition. Later, we discovered many insects with light guide eyes and others with excellent superposition (see below). From my (1968a) letter to Nature on the American firefly Photuris, I quote: When a dark-adapted retina, which has been light fixed so that it is sufficiently firm to cut, is examined from behind, with the receptor layer removed, it is possible to focus up and down in the region of the crystalline threads. With a light in front of the eye, bright points of light can be traced along the crystalline threads by moving the focal plane of the microscope. It is directly observable that light entering the facets is channeled along crystalline threads in the dark-adapted eye. [my emphasis] (Horridge, 1968a)

How much more specific could I be, but to no avail. An aroused Seitz (1969) traced rays in cones of the firefly Lampyris and calculated that the ommatidia were afocal. But he used elementary formulae suitable only for a homogeneous medium. Kunze published briefly in

64

Chapter 4

1969, followed by several short r­ eviews. In a comprehensive review, he assumed that the Lampyris eye had perfect superposition, then worked out what the ­optics would be, and gave an orthodox account (Kunze, 1979), but these authors never mentioned the extensions of the cone cells. Carricaburu (1975, p. 244) describes how he cleaned a cornea of the firefly Lampyris, and he illustrated small inverted real images close to the cone tip at the back of the eye. He says for Lampyris, ‘there are no afocal systems, otherwise this inverted image would not be seen’. Carricaburu measured the refractive indices in Lampyris and found insufficient gradients to give the effects that Exner assumed, but he was not mentioned in the orthodox literature, and soon forgotten. My whole effort was ignored in the subsequent German literature. However, I alerted the Orthodox establishment and initiated a burst of activity on the eye of Ephestia, as described below. Afocal optics in insect ommatidia were not properly demonstrated until we discovered them in skipper butterflies in 1972 (Horridge et al., 1972) (Fig. 4.34).

Various Endopterygote Clear-zone Eyes These eyes are described as clear zone because this terminology does not assume a particular kind of optics before the investigation begins. Archichauliodes (Megaloptera) The most primitive endopterygote insects, the Megaloptera, fly only at dawn and dusk in dim light. We found the larvae under stones in a nearby river, and the adults emerged in the laboratory. The eyes appear to be used only for sex and egg-laying activity. The rhabdom, formed by eight retinula cells, is all proximal, and does not move on adaptation. In bright light, a crystalline tract is drawn out as extensions of cone cell processes (Fig. 4.23) and clearly acts as a longitudinal shutter. We have here an extendible light guide eye, with

(A)

(B) Cornea

Cone Pigment cell Crystalline tract Accessory pigment cell nucleus Distal receptor cell Proximal receptor cell Rhabdom

Basement membrane Axons

Fig. 4.23.  Ommatidium of Archichauliodes (Megaloptera) in dark (A) and light-adapted state (B). Adaptation to light is large-scale movement of cells. (From Walcott and Horridge, 1971.)

s­ hutter and mobile receptor cells, in the most primitive clear-zone eye we could find.

The questionable moth eyes There are plenty of dubious statements that moths have sharply focused superposition eyes but little hard data; the huge variety of large moths have never been examined (­except Tuurala, 1954). In his book, Exner (1891, p. 81) gives only a brief statement that ‘Schön sind die Bilder freilicht nicht . . . ich musste mich begnügen’, saying he must be content with a poor image. This was not surprising as the eyes were either frozen or fixed in alcohol. Exner noticed that the eyeshine was observable from the direction of the incident light, from which he inferred the lens cylinders (p. 145). This was the only experimental observation on the whole eye. Subsequently, several generations of scientists were stunned by the erect image behind the isolated cornea of the firefly.



The Fundamentals of the Insect Compound Eye

The flour moth, Ephestia This very common moth is usually in cultures that have not used their eyes for many generations. Cultured insects without natural selection are notorious for having abnormal anatomy and being blind or deaf. In our Canberra specimens, numbers of retinula cells varied from eight to 11 or more, and the pigment cells sheathing them were equally variable. Insects reared in cultures away from daylight also tend to lose their sensitivity to light. Maybe inbreeding explains some of the differences in the accounts. Examination of the retina of Ephestia reveals a curved cornea at each facet and a typical crystalline cone that has higher refractive index along the axis. In both day and night eyes, there is a continuous thread of high density rhabdom stretching from the cone to the main rhabdom, which is 0.045 mm deep. In the light-adapted day eye, a short light guide is pulled out from the cone cells and surrounded by pigment grains. It looks like an eye that is apposition by day, with a clear zone and a light guide at night. In 1969, Peter Kunze considered that ray tracing through models of the optics of the eye of the flour moth Ephestia illustrated very well the superposition theory, while at the same time, in my laboratory, we observed at an early stage in the light adaptation of a dark-adapted eye that each ommatidium had a small erect image in the clear zone close to the cone tip that was ‘neither bright nor distinct, . . . almost obscured by scattered light, and are often not seen at all’. There was no evidence that these erect images fused together to make true superposition. Electron microscopy clearly showed a short narrow light guide at the cone tip in the light-adapted eye and a dense rod of rhabdom structure between the cone and the deep rhabdom in the dark-­adapted day eye (Fig. 4.24). These structures shone bright in dark-adapted eyes cut just behind the cones, and shone longer than the small erect images as light adaptation continued (Horridge, 1971, 1972). In three species of American moths, Døving and Miller (1969) observed ‘a very

(A)

65

(B)

Four cone cells

Principal pigment cell

Accessory pigment cell

Retinula cell

Rhabdom Basal cell Tracheal tapetum

Fig. 4.24.  Ommatidium of Ephestia. (A) Dark-adapted night eye, with pigment grains at the periphery of the cone. (B) Light-adapted day eye, with pigment grains surrounding the receptor cells through the whole depth. Cross sections at each level are shown. Note the continuous thin thread of dense rhabdom along the axis from the cone tip to the deep receptor layer in both states. (From Horridge and Giddings, 1971.)

dim erect image only in the plane of section of the eye’ (i.e. at the cut surface) and inferred that ‘the image is formed from the conduction of light in the tracts’. To cap it all, wave modes at the cut ends of crystalline tracts in the eye of a moth Dysauxes were illustrated at the International Congress of Entomology in Canberra in 1972 (Carricaburu, 1975), but none of the orthodox followers of Exner were there. Kunze (1969) had found that when a narrow parallel beam is shone briefly anywhere into the eyeshine patch of a dark-­adapted Ephestia night eye, the whole eyeshine patch lights up. As a demonstration that

66

Chapter 4

some form of superposition was present, Kunze also found that the motion of an optomotor stimulus could enter via a single facet. These observations have no value because the same results would be obtained if the light was scattered in any way inside the eye. Kunze did not examine the angular distribution of the eyeshine in intact eyes of healthy moths. In dark-adapted Ephestia, the incident light can be moved 20° on either side before the patch disappears, suggesting that a parallel beam on the eye is spread on the reflecting layer over an angle of 40° relative to the eye centre. Later, we developed a method used to check optical devices by sending a thin pencil of light through each facet and measuring the exit rays (see below, Figs 4.29, 4.36). A better criterion of high resolution superposition is to examine the bright deep pupil at the side of an oblique beam directed on the living eye at various angles (Fig. 4.30), as we did in the diurnal agaristid moth Phalaenoides (Horridge et al., 1977), or use an ophthalmoscope (Land, 1984). Ray tracing in Ephestia Measurement of the refractive index gradients in the cornea and watery cone of a small moth presents a challenge, and the errors ­accumulate as one proceeds to trace a ray along its route. However, results illustrated here (Figs 4.25, 4.26 and 4.27) were accepted without question in reviews by Autrum (1981) and Kunze (1981). The small dimensions are a challenge for calibration of refractive index gradients. In Canberra, with great ­difficulty, we measured the index ­gradients in longitudinal and transverse sections of Ephestia cones with a two-beam interference microscope and oil immersion lens, then traced rays in flat planes on large pieces of paper. The results demonstrated the enormous scatter of rays (Fig. 4.28). Even in eyes of skipper butterflies known to have well-focused superposition optics, we could not avoid large errors and also obtained a huge scatter of traced rays (Fig. 4.38). It looks as if the efforts of Cleary et al. (1977) were designed

to defend the orthodox doctrine and save Exner’s reputation. It was not a case of data in and conclusion out; it was an unsubstantiated belief and a self-delusion. Chapters by Kunze (1979) and Autrum (1981) give only the Exner version, in an example of preference falsification (Kuran, 1997) in favour of the orthodox a­ ccount. A better method No physicist or manufacturer of optical equipment would analyse an unknown optical system, or check the precision of ­ their product, by measuring refractive indices and then drawing rays. The convenient method is to track the passage of a pencil of light through the system. We made an apparatus that would measure the exit angle of the eyeshine of a small group of facets (Fig. 4.29) that was easily modified to measure the incident and exit angles at each incident angle of a thin pencil of light passing through each cone. The data obtained from the angular distribution of eyeshine in the intact eye is crucial in demonstrating the distribution of a parallel beam on the receptor layer and the scatter in the superposition mechanism in the dark-adapted animal (Fig. 4.30). As the angle between the axis of the observing microscope and the direction of illumination is increased, the disc of eyeshine stays the same size, but moves round with the observer, and stays the same up to about 12° in any direction. Some eyeshine can be seen up to an angle of 20° (Y–D in Fig. 4.30). Indeed, Ephestia was a most ­unsuitable subject for the demonstration of a ­superposition eye. With this large angular spread of eyeshine measured in the dark-­ adapted intact animal, it is hard to understand how Kunze, Cleary and co. could conclude otherwise, or how they obtained their remarkable data that supported Exner. However, their calculations were the only ones to feature in the influential chapters edited by Autrum (1979, 1981). Judging by their eyeshine, some large night-flying insects (e.g. dung beetles and Agrotis spp.) have superposition eyes with



The Fundamentals of the Insect Compound Eye

21°

15°

9°

67

3°

Fig. 4.25.  Ray paths through the cornea and cone in the eye of Ephestia, calculated from the measured refractive index gradient of the cones. (From data in Cleary et al., 1977.) The greater the angle of incidence α, the greater the angle of exit β on the same side of the axis. Facet diameter is 0.017–0.019 mm and the cone is 0.012 mm wide at the widest place. It is curious that rays that enter parallel emerge parallel.

quite good focus. In those that fly in a narrow range of intensity, on light adaptation only the pigment around the cone migrates inward, but in species that experience a wide range of intensities, pigment also migrates into the clear zone (Warrant and McIntyre, 1996).

30

Angle to the normal inside β°

25

20

15

α°

10

β°

5

0

3

6

9

12 15 18 21

Angle to the normal outside a°

Fig. 4.26.  The published relation between α and β obtained by ray tracing (Fig. 4.25) using the measures of refractive index in cornea and cones. (From data in Cleary et al., 1977.) The refractive index data must have been very precise. The regression ratio shown by the line has a slope of 1.32, but the expected Exner line would be about 1.5 (as in Fig. 4.28).

Daymoth Phalaenoides glycine (Agaristidae) This black and white diurnal Australian moth is common where grape vines are grown. The moth flies slowly and only in bright sunlight. The ommatidium shows little change between the night and day states (Fig. 4.31). The eye as a whole has wonderful optics (Fig. 4.32) but recording with a microelectrode was difficult (Horridge et al., 1977, 1983a). Recording from the eye of Phalaenoides reveals three colour types of receptors, with peak λ near 380 nm and 520 nm, and a rare blue type near 475 nm. They all have similar field width (Δρ) at 50% sensitivity, with best value near 1.7°, which is amazing for a superposition eye. Land (1984) published

68

Chapter 4

c cc

Rh T Fig. 4.27.  Imagined ray paths across the clear zone in the eye of Ephestia, calculated from the measured refractive index gradient of the cornea and cones. (Redrawn from data in Cleary et al., 1977.) Similar convincing diagrams of the dark-adapted moth eye were printed in every text in the 20th century, with no good experimental support. c, cornea; cc, cones; Rh, rhabdom layer; T, trachea between rhabdoms. (A)

(C) +α α

α2

+α

α3 α4

+α α5 +β –β

β2

(B)

β3

β3

β

β5

+α

–β

20°

+

++

+

Ex α

ne

10°

×

rl

ine

× ××

× +

+

+

+

0° –15°

+

×××

×

++ +

+

× ×

×

× +

× × ××× × ×

× + × × +× ×

×+

× + ×+ + ×

–10°

–5°

0°

β

+

+× +×

+

+

× ×

+ +++ + + +

× × ××

× ×

+ × × ×

+

+ × ×

+

+

× × × × × +

+ +

× × × ×+ +5°

+ ×

× × + × ×+ × β

+10°

+15°

Fig. 4.28.  Our ray tracing data from Ephestia. (A) Ideal paths of rays in a superposition eye. (B) Ray paths in the cone, calculated from the measured refractive index gradients. At each angle of the incident rays, α, there was a large scatter of the emerging ray. (B) Most rays were not bent back towards the side whence they came, and the average was nowhere near the Exner line (C). Off-axis parallel rays could not be parallel on emergence from the cone tip; compare Figs 4.37, 4.38. This is a serious comment on the value of calculated ray tracing as a way to explore an unknown inhomogeneous optical system. (From Horridge, 1972.)



The Fundamentals of the Insect Compound Eye

69

Photocell traverse Image plane Eyepiece Angle adjustment

Photocell 45° mirror

Luminar objective Half-silvered mirror

Collimator Lamp Eye on head Sector wheel

Fig. 4.29.  The apparatus for measuring the angular distribution of eyeshine, as later modified to measure the angles of exit of pencils of light through individual cones. A parallel beam from a pinhole illuminated by a xenon arc lamp was reflected by a tiny sliver of half-silvered mirror on to the eye. A microscope was mounted on a cardan arm to measure the angle of rays from each ommatidium in the eyeshine. Measurements were quantified with a small photocell mounted at the image plane inside the microscope. The Luminar microscope objective had long focal length and was fitted with an adjustable diaphragm.

E

G

F

C

2α A

B 2α D 2α

2β

2β

2β 2β

V

W

X

22.5°

Y Z

45° 0

100 μm

Fig. 4.30.  Scale diagram of the eye of Ephestia, upon which is drawn the receptor surface VWXYZ, the eyeshine patch AB (inward arrows) and the angular distribution of eyeshine (outward arrows). The dotted area is the measure of light that reaches the receptors, and which causes the eyeshine. E, F and G are light rays with directional arrows. The values of 2α = 40°, and 2β = 30° can be read from the diagram. (From Horridge, 1972.)

70

Chapter 4

(A)

photographs taken through an ophthalmoscope that resolved the distal ends of the separate columns of receptors and their sheaths of trachea in the living eye. The receptors have similar absolute sensitivity to locust and blowfly, with a lowest detection threshold near 109/cm2/s, measured as parallel normal rays on the surface of the eye. This is 100 times brighter than moonlight. In brief, the optics is transparent and very well focused but useful only in daylight. This moth does not fly after sunset.

(B) c

DA

LA

ppc

cc ppc

apc rc

Skipper butterflies, Hesperiidae, Lepidoptera t apc

t

bcn bc

bc

Axons

tcn Axons Fig. 4.31.  The ommatidium of the agaristid diurnal moth Phalaenoides, showing negligible change between night and day. The fixed pigment absorbs rays that are not sharply focused on the retinula cell layer. The long rhabdomeres are optically isolated by trachea and pigment. apc, Accessory pigment cell; bc, basal cell; bcn, basal cell nucleus; c, cornea; cc, crystalline cone; DA, dark adapted; LA, light adapted; ppc, principal pigment cell; rc, receptor cell; rcc, receptor cell extensions; t, tracheole; tcn, tracheole cell nucleus.

50%

–4

Relative intensity (%)

100%

3.25°

0 Angle (°)

4

Fig. 4.32.  The relative intensity of the eyeshine of Phalaenoides plotted as a function of angle. Even after a second pass through the optics, the width at 50% was only 3.25°

Skippers fly in bright sunshine as if they lack sensitivity but they have a perfect form of superposition eye with no pigment movement or other signs of day/night adaptation (Figs 4.33, 4.34). Recording from the receptor layer reveals a sharp angular sensitivity curve, compatible with the angular distribution of eyeshine. The facet of maximum sensitivity could be found when the tip of a tiny light guide was moved over the surface of the cornea. In agreement with the size of the eyeshine patch, the tip could be moved about ten facets in any direction before the response disappeared (Horridge et al., 1972). To observe the passage of rays through individual cones in the living state, a fresh eye slice was hung under a cover slip in the apparatus in Fig. 4.36(A), so that a parallel beam could be constructed through it in either direction. The optics of the eye was then reconstructed (Fig. 4.36B). When we discovered the superposition eyes of the skipper butterflies (see below), we realized that we had a rigid eye with splendid optics that deserved proper investigation. First, we recorded from the receptors, then recorded the angular distribution of eyeshine (Fig. 4.35), then tracked pencils of light through the cones in situ (Figs 4.36, 4.37). This result can be compared to the scattered ray directions found by ray tracing in the same species (Fig. 4.38), where we knew that the eye was well focused.



The Fundamentals of the Insect Compound Eye

DA

LA (A)

c

(B) rce (C) (D)

rc

bc tc

Fig. 4.33.  The skipper butterfly, Trapezites symmomus. (A) A dark-adapted (DA) ommatidium, showing extensions of the retinula cells reaching to the tip of the cone. (B, C, D) The screening pigment around the cone eliminates unavoidable unfocused rays, but there was little change when light adapted (LA). bc, Basal cell; c, cone; rc, retinula cell; rce, receptor cell extension; tc, tracheal cell.

In skipper butterflies (Hesperioidea) the screening pigment around the cone and near the basement membrane is unaffected by light and the clear zone is always clear, making them convenient for demonstration of the optical superposition. We cut the ­living eye and observed the superposition ­directly with two beams of light passing through different groups of facets (Fig. 4.34).

Dung beetle Anoplognathus (Polyphaga, Scarabaeidae) This night-flying beetle was investigated by six methods, that all gave similar ­ results (Meyer-Rochow and Horridge, 1975). The

71

cornea is smooth and flat, therefore the optics is unchanged by rainwater. The inner side of the cornea is strongly curved like a lens. The crystalline cone is non-homogeneous and, in the dark-adapted eye, has a rounded tip that partially focuses a parallel beam across the clear zone on the receptor layer, 400 μm below (Fig. 4.39). When the dark-adapted night eye was ­ illuminated with a strong light, the cone tip was slowly pulled out to form a crystalline tract up to 60 μm long, while the transparent columns of the retinula cells contracted (Fig. 4.39B). The dark-adapted eye is partially focused. Following measurements of the ­refractive index in each part of the cones, ray tracing shows that a patch of ommatidia 40  facets wide admits light from a parallel beam that reaches the rhabdom (Fig. 4.39A). At each angle of incidence, there is a lot of scatter (±  10°) but the average follows the line expected by the Exner theory (Fig. 4.40). The width of the angular sensitivity curve at 50% sensitivity calculated from ray tracing was 20°; the angle indicated by the distribution of the eyeshine was 34°; the angle measured with a microelectrode was 20° when dark adapted, and 12° when light adapted. When living animals were tested in an optomotor drum revolved around them, they responded to stripe periods of 12° or 15° in the light-adapted state in bright light, and down to 20° or 25° in the night eye adapted to dim light. In these tests, the b ­eetle responded well but a white butterfly similarly tested could not see at all. It is an error to consider that these measures of resolution make the eye useless, because the dung beetle is active at night. It can see well enough in faint starlight to navigate by the line of the Milky Way across the night sky, just as fighter aeroplanes in 1940–1945 made use of simple airborne radar to locate enemy planes in the air with a poor angular resolution of about 30°. To home in on a target at night, absolute sensitivity is at a premium, and poor spatial resolution will eventually bring success. The intensity at the rhabdom layer was calculated from measurements of eyeshine (Fig. 4.41), and compared to the original intensity on the outer surface of the eye (Fig. 4.42).

72

Chapter 4

Two parallel beams 8° apart

Two spots 8° apart

Centre of eye 0° Fig. 4.34.  Trapezites symmomus. The view inside the clear zone though a cut in the top of the eye. Two spots of light were formed by two pencils of light at an angle of 8° to each other. Each spot covers an area of three by four ommatidia. (Drawn from a photograph taken by the author in Horridge et al., 1972.) 100

60

20

–6

–4

–2

Percentage of maximum

2.5° width of observing aperture

2° facet width

+2

+4

+6

Angle of observation (°) Fig. 4.35.  The angular distribution of eyeshine rays in the skipper Taractrocera papyria relative to the normal incident beam. The measurement of intensity has been integrated over the whole eyeshine patch. The width at 50% maximum was only 4°.

The water beetle Dytiscus (Adephaga, Dytiscidae) This was the first insect retina where we found distal and proximal rhabdomeres in the same eye, with a deep clear zone between,

with no trace of a light guide across it (Horridge et al., 1970). We d ­ escribed it as a new type of insect retina, not knowing at the time that it was already described by Schultze (1868) and Grenacher (1879). Exner (1886) discovered a gradient of refractive index in



The Fundamentals of the Insect Compound Eye

73

(A)

β

α α

(B) β

Fig. 4.36.  (A) A slice of the eye hanging under a cover slip and inner side illuminated by a parallel beam from the collimator and half-silvered mirror. (B) Reconstruction of a parallel beam from the data from (A).

+12° +8° +4°

–16°

–12°

–8°

+4°

+8°

+12°

+16°

–4° –8° –12°

Fig. 4.37.  Skipper Toxidia peroni. Actual angles measured with a pencil of light passed through the optics, as in Fig. 4.36.

74

Chapter 4

(A)

(B)

α

α 14°

10° Exner line

8°

β

6° 4° P

2° β

6°

4°

2°

0

2°

4°

6°

8° 10°

β

Fig. 4.38.  (A) Ray tracing in the eye of the skipper Toxidia peroni, based on best measurements of refractive index gradients in the cornea and cone. Rays heading towards P would be cut off by the tracheal sheaths around each rhabdom. (B) Diagram of the eye showing that the Exner line is based on the aperture of receptors surrounded by a tube of isolating tracheae. It is obvious from the angular distribution of eyeshine (Fig. 4.35), and from tracing thin pencils of light through the eye (Fig. 4.36), that this is a sharply focused eye. Therefore, calculated ray tracing introduces huge errors. (From Horridge et al., 1972.)

lens cylinders of the crystalline cones, but went no further, but Exner’s o ­ rthodoxy obfuscated the subject for almost a century. In the dark-adapted retina, pigment migrates to the periphery around the cone, and seven retinula cell bodies press closely against the cone tip. One of these contains a rhabdomere (Fig. 4.43). The others form a thick rhabdom layer at the basement membrane, with rhabdomeres in the shape of a cross. In the light-adapted retina, the cone extends to form a crystalline tract and carries pigment grains with it. Similar changes occur in Carabidae and probably in many beetles that are active by night and day. In 1970, we discovered that the retinula cells (Fig. 4.44) were large enough to give excellent recordings (Horridge et al., 1970). When he explored the surface of the eye of Dytiscus with the tip of a tiny light guide, Ben Walcott was able to show directly that the proximal rhabdom was reached by light entering an area of many facets, but the ­distal rhabdomere was excited only by light entering a single facet. The retinula cell with a distal rhabdomere behaved like an apposition eye, but the proximal rhabdomeres have very wide fields as if there is partial superposition optics with a very

poor focus. In cleaned cornea of water beetle eyes of this type, Exner (1891) found erect images, but in the living eye the focus would be spoiled by the thickness of the receptor layer. In Dytiscus, and some related water beetles, we find no evidence for light guide action, and abundant evidence of optical superposition achieved by a combination of corneal curvature and lens cylinders in the thick cornea. The focus cannot be good because the receptor layer is thick. In addition, the single distal rhabdomere at the cone tip acts like an appostion eye that is functional at all times. Consideration of beetle and moth retinas with and without optical separation of the proximal rhabdoms by stacks of pigment grains or tubular sheath formed by tracheae show two possible and overlapping lines of evolution. One is the separation of the rhabdoms of neighbouring ommatidia, with ­narrowing of the aperture of each, and consequently loss of sensitivity, but Δρ near 2–4°. The other is optical continuity laterally through the rhabdom layer with maximum sensitivity from use of oblique rays, but Δρ near 12–14° (Fig. 4.45). Dytiscus combines both.



The Fundamentals of the Insect Compound Eye

(A)

75

(B)

DA

Cornea

LA

Corneal cone Cone cell Cone Pigment cells

Crystalline tract nuclei Retinula cells

Clear zone

Rhabdom

brc bm Axons Fig. 4.39.  Ommatidium of the dung beetle Anoplognathus (Polyphaga, Scarabaeidae) in dark-adapted (DA) (A) and light-adapted (LA) (B) state, showing large changes at the tip of the cone, with formation of a short light guide. bm, Basement membrane; brc, basal reticular cell. (From Meyer-Rochow and Horridge, 1975.)

Macrogyrus (Adephaga, Gyrinidae) This common family of shiny beetles that skim the surface of freshwater have part of each eye looking upward and part looking down into the water. In the dark-adapted night eye, the cone has a rounded end free of screening pigment (Fig. 4.46A), but in the day there is a short crystalline tract leading into the end of the rhabdomere of the distal retinula cell. The change in the shape of the

cone is particularly marked in this animal. The apposition compound eye of the bee cannot match this high acuity over the whole natural range of intensities. In electrophysiological recordings, all cells were green sensitive with peak near 552 nm. Some units were sensitive to the plane of polarization, others not. In the day eye, angular sensitivity was as low as 3.0° at 50% sensitivity (Δρ); in the night eye, stimulated at low light level, Δρ was 4.1−4.4°,

76

Chapter 4

(A)

(B)

α

α 30°

20°

10° β

10°

30°

0

10°

20°

30°

β

Fig. 4.40.  Anoplognathus dung beetle, Scarabaeidae. The relation between the angle of the incident rays and the angle of the corresponding ray inside the retina. The large scatter agrees with all other methods of analysis of the optics and is therefore acceptable.

and by night because they can be detected by predators below and above. ‘The superb adaptation, with maximum sensitivity at little expense of resolution is in the night eye’ (Horridge et al., 1983b).

100

Eyeshine intensity (%)

80

The lacewing Chrysopa (Neuroptera) 60

50% 40

20

12° 0

20

10

0

Angle to illuminating beam(°) Fig. 4.41.  Anoplognathus. Total intensity of the eyeshine at different angles to the illuminating beam that was centred on the centre of the eye. The width of the whole distribution at 50% intensity was 24°.

which was very good for a small insect, but it must be remembered that gyrinid beetles on the water surface must be alert by day

This beautiful crepuscular insect presents us with an enigmatic eye that is clearly adapted to some unknown change of behaviour between day and night. The facets are small, 16–17 μm, with corneal radius of curvature also 16 ± 1 μm. At the boundaries between the facets, yellow light is strongly reflected (hence the name ‘golden eye’ because the cornea is layered with repeat period about 0.25 μm). In the light-adapted day eye, there is a long crystalline tract to the rhabdom of eight fused rhabdomeres (Fig. 4.47B). Two of the rhabdomeres, cells 7 and 8, fill the light path with rhabdomeres of microvillae that are similarly oriented across large areas of the eye, as if sensitive to the polarization plane for some useful purpose. In the dark-adapted night eye, the tract is withdrawn into the cone cells. Cell 7 is tightly attached to the tract end by desmosomes, but cell 8 stays with its rhabdomere attached to the main rhabdom (Fig. 4.47A). In the dark-adapted night eye, there is no structure that could guide



The Fundamentals of the Insect Compound Eye

20

77

Number of rays at cornea

Number of rays at receptor level

100 μm

Cones

10

50%

20°

Receptors 0

20 18 5°

15

12 16

8

4

0

4

8

12

15

18 20

12 8 4 0 4 8 12 16 Upper scale: cone number Lower scale: receptor number

Fig. 4.42.  Anoplognathus. The number of rays passing through the cornea then through cones across the patch which admits light from a parallel beam. The receptor layer admits rays over a smaller angle than the cone layer because each receptor has its own aperture. Note that the number of rays at the receptor layer just rises above the number at the cornea, showing a small gain in sensitivity in this example. (From Meyer-Rochow and Horridge, 1975.)

light to the ­proximal rhabdom. The proximal part of the rhabdom is coated with a sheath of tracheae that are enlarged by the basement membrane and are responsible for the strong eyeshine. The presence of a tapetum shows that sensitivity is important for this animal, but inevitably oblique rays spread laterally and reduce ­resolution.

The soldier beetle Chauliognathus ­(Cantharidae) This colourful small beetle congregates to breed on certain bushes in summer, offering no sign of the habits for which this peculiar corneal cone eye structure may be adapted, except that it is then very active in bright sunlight. Four cone cells at the tip of the cone form a short light guide that does not change in length with adaptation, but is surrounded by dense pigment in the light-­ adapted day eye (Fig. 4.48B). Three retinula cells form a long thin rhabdom that must be a light guide, otherwise it would not be that shape. Four retinula cells form two large lumps of rhabdom at the peripheral side of the clear zone, where they stay by day and night. An eighth cell near the basement membrane has no rhabdomere. Recording

revealed r­ etinula cell types at 360 nm, 450 nm and 525 nm and acceptance angle of 3°, presumably of the central rhabdom rod (Horridge et al., 1979). Similar off-axis rhabdomes have been described in the eye of the alder fly, Raphidia (Megaloptera), which has a long central rhabdom rod but typical crystalline cones (Ast, 1920). At the time, Ast was explicit that the two types of rhabdom in the same eye showed that it was both apposition and superposition, but his discoveries were forgotten. The rate of scientific advance certainly was slow.

The mayfly Cloeon (Baetoidea ­Ephemeroptera) In this common group of mayflies, the males have remarkable turbinate eyes on the top of the head, to enable them to catch females in flight by detecting them from below. As in Atelophlebia (see below), this is an eye that functions by light guides across the clear zone, as shown by directly observing the illuminated cut ends of the crystalline threads, 2 μm wide, which are a bundle of extensions of retinula and cone cells. Seven retinula cells stretch from the cone to the

78

Chapter 4

DA

LA

Cornea

Crystalline cone

Principal pigment cells Accessory pigment cells Distal rhabdomere and retinula cell nuclei

Retinula cell extensions

Proximal rhabdome layer

Basal retinula cell Eight retinula cell axons

basement membrane (Fig. 4.49), with distal rhabdom around the tip of the cone and a thick proximal rhabdom layer in all seven cells. As in Phalaenoides (Fig. 4.31) and Atelophlebia (Fig. 4.50), there is a sheath of tracheae around each proximal rhabdom, isolating it optically from its neighbours, and giving it a narrow aperture. In eyes of this type, the longer the light guide and wider the clear zone, the greater the c­ ollecting area of

Fig. 4.43.  Details of the structure of an ommatidium of the water beetle, Dytiscus, in dark-adapted (DA) and light-adapted (LA) states. (From Horridge, 1968b.)

the facet for each ommatidium, and the greater the optical gain. The mayfly Atelophlebia, Leptophlebiidae, Ephemeroptera The male of this common large mayfly Atelophlebia (wingspan 25 mm) has a double eye on each side. Unlike other mayflies



The Fundamentals of the Insect Compound Eye

(A)

(B)

79

Cell 8

Cornea Crystalline tract Receptor cells Clear zone

Electrode Rhabdom of cells 1–6 Cell 7 Eight axons per ommatidium

Fig. 4.44.  Details of the retina of Dytiscus. (A) Scale drawing with position of the distal rhabdomeres (dr) and the electrode in the proximal rhabdom layer. (B) Details of distal retinula cell, proximal cells and basal cell.

Anoplognathus

0.4

λ/D

Width of point spread function (λ/D) radians at 50% level

0.2

See ‘blur circle’ fig 4.9 Chrysopa

Diurnal moth Phalaenoides

0

Probable range for DA moths

Probable range for skipper butterflies

1

2

Ephestia

3

Aperture/focal length (1/F value) Fig. 4.45.  The trade-off between resolution and sensitivity is set by the physics of light and refractive index of organic matter. λ is the wavelength of the light, D is the aperture of the lens and F, the F value of the lens. DA, dark adapted.

80

Chapter 4

(A)

(A)

(B)

(B) LA

DA Cornea

DA

LA

Crystalline cone Principal pigment cell Receptor cells

Rhabdom of cell 8

Rhabdom cells 1–6

2 1

Retinula cell bodies 1–6

Retinula cell 7?

Retinula cell 8?

7

Retinula cell bodies 1–6 Cell 8

8

7 8

Retinula cell 7?

3 4

6 5

Proximal rhabdom

Receptor cell 7 8 Basement membrane Fig. 4.46.  An ommatidium of the whirligig water beetle Macrogyrus (Gyrinidae, Adephaga) in (A), dark-adapted (DA) and (B), light-adapted (LA) state. Comparison with Gyrinus shows that this beetle retina has evolved into an apposition eye.

(­Baetoidea) the dorsal eye of the male has square facets. In all male mayflies, the dorsal eye is sensitive only to UV, and designed for spotting the females in flight overhead. The ­arrangement within is very unusual. All eight retinula cells have their nuclei near the cone tip. Yellow pigment grains in the principal pigment cells surround the neck of the cone in the day eye and move away at night. The dorsal male eye is sensitive only to UV, which explains why the screening pigment is yellow. There is a wide clear zone crossed

7

Fig. 4.47.  Ommatidium of the lacewing Chrysopa (Neuroptera). (A) In the dark-adapted (DA) night eye, the distal receptor cell with its rhabdomere moves close to the cone tip. (B) In the light-adapted (LA) day eye, the cone tip is elongated into a light guide. (From Horridge and Henderson, 1976.)

by the extensions of the cone cells and retinula cells, with a thick proximal rhabdom of very strange shape in cross section that is formed by six of the cells (Fig. 4.50). This is surrounded by a sleeve formed by air-filled trachea and yellow pigment cells that separate the rhabdoms of each ommatidium. The trachea contribute to the high resolution because oblique rays cannot pass laterally through the proximal rhabdom layer. Retinula cells 7 and 8 have no rhabdomere, and an unusual



The Fundamentals of the Insect Compound Eye

(A)

81

(B) DA

LA

Cone extension

Cone extension

Principal pigment cell

1

3

Crystalline tract

1

3 2

Extra rhabdoms

1

Rhabdom

3

4

2 1

4

5 6

7

5 6

7

n

2 Cell 7

6

1 n

Cell 8 4 Basement membrane

3 4 5

3 2

7

5 6

8 1

8

7

Fig. 4.48.  Ommatidium of the soldier beetle Chauliognathus (Cantharidae). As in the firefly, the cone is an inward extension of the cuticle but the extensions of the cone cells are short and fixed. Three retinula cells form a thin rhabdom, 2 μm in diameter and 120 μm long, which must act as a light guide at all times. These three cells include UV-, blue- and green-sensitive ones. (A) In the dark-adapted (DA) night eye, three retinula cells with short thick rhabdomeres, up to 1000 μm3, are exposed, and presumably function in dim light. (B) In the light-adapted (LA) day eye, pigment grains surround the crystalline tract. (From Horridge et al., 1979.)

eccentric relation to the other retinula cells. The nuclei lie between the cones; the cytoplasm is filled with small inclusions, the cell body wraps around the rhabdom formed by cells 1–6, and is attached to them by desmosomes (Fig. 4.50). This arrangement strongly

suggests that cells 7 and 8 are sensitive to the total current flow when the rhabdom is depolarized by light. Electrical coupling was also suggested by the flow of the dye Lucifer Yellow between all the rhabdomeres of an ommatidium but not further across the clear

82

Chapter 4

70 μm

70 μm

Cone

120 μm

Retinula cell nuclei

Retinula cell extensions

12 μm

120 μm

Rhabdom

Tracheal sheath

Fig. 4.49.  Ommatidium of the mayfly Cloeon. (From Horridge, 1976.)

zone. Cells 7 and 8 have the largest axons to the lamina and directly to the medulla. It is possible that the dorsal eye of the male has no sensitivity to the plane of polarization, and probably also to motion. Recording with microelectrodes confirmed the sensitivity to UV light at 323 nm (Fig. 4.51A), peak sensitivity at 369 nm (Fig. 4.51C), and a field width of 2° at 50% sensitivity level (Fig. 4.51D). This is little more than the theoretical minimum for a facet ­aperture of 18 μm. The eye is not particularly sensitive in absolute terms, catching only about 2% of the incident photons, compared with a large fly or locust, which usefully catches 50%. However, the resolution is superb. A problem with many studies of clearzone eyes is that light guides action and also

a superposition image can be presumed or even demonstrated, but in this mayfly it is possible to exclude a superposition effect. A thin wire 35 μm in diameter on the arm of a micromanipulator was brought parallel and close to the eye surface, while intracellular recording was in progress (Fig. 4.52). When the wire shaded the sensitive facet, the electrical response fell to half height, showing that 80% of the incident stimulus had been excluded from that facet.

Very Small Compound Eyes We know almost nothing about the variety and problems of miniaturization. Efforts so far have concentrated on the largest and most striking examples, and the smaller the eye,



The Fundamentals of the Insect Compound Eye

83

Cornea Cone cell nuclei

a b c Retinula cell 7

5

4

6

3

rc 7 2

7

Retinula cell nuclei

1

Crystalline tract 5

6

4

7

3

Retinula cell 7

1

2

4 3

Tracheal sheath

7

7 t

5

6 7

2

1

Axons

t Fig. 4.50.  Ommatidium of the male Atelophlebia. (From Horridge et al., 1982.)

the greater the difficulty, the greater the curvature of each surface, and the greater the spherical aberrations. Consequently, data is lacking, so discussion is stillborn. In the apposition eyes of many very small insects, the focal length must be very short, and the

power required of the lens at the cornea makes the c­ ompound eye look like a pile of beads. In clear-zone eyes of nocturnal insects, focus is poorer in smaller eyes, and resolution is sacrificed for the sake of preserving sensitivity, because it is always ­better

84

Chapter 4

(A)

λ = 323 nm

1s

0 10 mV

402 nm

369 nm

(C)

336 nm

(B) Fig. 4.51.  Microelectrode recordings from a retinular cell in the dorsal eye of the male mayfly Atelophlebia. (A) Bumps attributed to single photon arrivals with a very low intensity stimulus. (B) Calibration with intensity steps, with wavelength 352 nm, each step is 0.2 log units. (C) Spectral sensitivity calibration, with peak near 369 nm. (D) Angular sensitivity calibration with steps of 1° movement of a pinhole light source at 352 nm. (From Horridge et al., 1982.)

1°

(D)

10 mV

λ = 352 nm

(A)

(B)

(C)

(D)

(E)

(F)

Fig. 4.52.  Intracellular records from the receptors of the dorsal eye of the mayfly Atelophlebia, male. (A) Final test to locate the recorded ommatidium. (B) Responses to movement of a wire across facets adjacent to the sensitive one. (C) Responses to calibrated intensity steps. (D, E) Responses as the wire is moved across the sensitive facet in steps of 25 μm. (F) Repeat of (C) with the wire closer to the eye. (From Horridge et al., 1982.)

to see something badly rather than nothing. In  small moth eyes, there is less space, so crystalline tracts cannot be long, and apertures cannot be wide, placing a limit on gains in sensitivity and resolution. In very small moths there is no room for a clear zone.

Conclusions In all these examples, we usually find 6 + 1 + 1 retinula cells, and a constant architecture of small grains in 16 pigment cells reaching the basement membrane, and



The Fundamentals of the Insect Compound Eye

l­arger grains in principal pigment cells surrounding the four cone cells. Six retinula cells are presumably green sensitive, and feed into processing channels that detect motion and structure from edges. The other two are probably a UV and a blue-sensitive cell. Insects that fly only by day have a solid retina in which the pigment grains around the rhabdom move when the illumination is changed. Insects that are active by day and in dim light present us with an interesting variety of movements of pigment and retinula cells. Pure superposition eyes without light guides exist, as in skipper butterflies. Pure light guide eyes also exist, as in mayflies. Eyes formerly believed to have superposition optics rely on crystalline tracts in the light-adapted day state. Screening pigment usually surrounds the threads and absorbs light by frustrated total internal reflection. In the dark-adapted night eye, the tracts are still in place, but oblique rays from the same or neighbouring facets flood into the eye. In some special Lepidoptera, these oblique rays are remarkably well focused across the clear zone at the receptor layer and fixed pigment absorbs unwanted rays. These eyes can have high resolution if the rhabdoms of each ommatidium are isolated by tubes of pigment, or air-filled tracheae. Otherwise, when the basal rhabdoms are not optically isolated, as in the firefly and water beetles, oblique rays increase sensitivity but resolution is lost. This cannot be avoided because rhabdomeres absorb only about 1% of light/μm, so they have to be up to 100 μm thick, or 100 μm long in the case of a central rod. Large water beetles studied have a dual mechanism with a single distal rhabdomere in each ommatidium and a clear zone with superposition and no light guides (Fig. 4.43). Some cantharid beetles have a rhabdom rod of four cells, and large lateral rhabdomeres in the other four cells (Fig. 4.48). In 20th century reviews, this diversity has been ignored in favour of the two extremes presented by Exner. In an otherwise very balanced and accurate review for the time, Land (1981) proposed ‘It is now

85

widely agreed that where there is a clear zone, there is nearly always a superposition image’. And he continues later: ‘Superposition . . . should be retained in preference to clear zone . . . except where some other mechanism has been clearly demonstrated to exist’. I suspect editorial pressure, which was not so obvious in other reviews (Land, 1985). In 2018, the term ‘superposition’ applies to a few special cases, but we are faced with a wonderful variety of clear-zone eyes, almost all of which have cone extension threads and apposition optics when light adapted. Despite possible editorial influence, in view of the rarity of an image, and demonstrations of light guides in almost all examples studied, this opinion will have to be revised because light guides are clearly preponderant and more ­primitive. Several methods of analysis were required because the examples are so diverse. Measures of refractive index gradients followed by ray tracing is the least satisfactory, and generates errors. Measures of angular distribution of eyeshine are a useful start. Electrophysiology is conclusive but often not possible because cells move. Techniques derived from the way optical equipment is calibrated are very informative, for example by measuring the angles between incoming and outgoing pencils of light. In some eyes the photoreceptors can be directly observed from outside the eye with an opthalmoscope. In 20th century reviews, eyes of crepuscular or nocturnal insects are usually assumed to have superposition optics, as described by Exner, but most are a­ pposition by day and poorly focused by night. Almost all eyes with a clear zone have in each ommatidium a light guide that increases sensitivity by increasing the surface area that catches light, and in some (e.g.  Atelophlebia) there is no superposition, only light guides. For years, old literature was neglected, and most of the wonderful variety and subtle designs were unknown. Most of this anatomical diversity cannot be fully explained even today.

86

Chapter 4

References Ast, F. (1920) Über den Feineren Bau der Facettenaugen bei Neuropteren. Zoologische Jahrbucher, Abteilung für Anatomie 11, 411–458. Autrum, H. (ed.) (1979) Handbook of Sensory Physiology, Volume VII/Part 6. Springer, Berlin. Autrum, H. (1981) Light and dark adaptation in invertebrates. In: Autrum, H. (ed.) Handbook of Sensory Physiology, Volume VII/Part 6A: Invertebrate Photoreceptors. Springer, Berlin, pp. 1–92. Autrum, H. and von Zwehl, V. (1964) Spektrale Empfindlichkeit einzelner Sehzellen des Bienenauges. Zeitschrift für vergleichende Physiologie 48, 357–384. Baumgärtner, H. (1928) Der Formensinn und der Sehschärfe der Bienen. Zeitschrift für vergleichende Physiologie 7, 56–143. Bullock, T.H. and Horridge, G.A. (1965) Structure and Function in the Nervous Systems of Invertebrates. Freeman, London. Cajal, S.R. and Sanchez, S.D. (1915) Contribución al conocimiento de los centros nerviosos de los insectos. Parte I. Retina y los centros opticos. Trabajos del Laboratorio de Investigaciones Biológicas del Universidad, Madrid 13, 1–168. Carricaburu, P. (1975) Examination of the classical optics of ideal apposition and superposition eyes. In: Horridge, G.A. (ed.) The Compound Eye and Vision of Insects. Clarendon Press, Oxford, pp. 236–254. Cleary, P., Deichsel, G. and Kunze, P. (1977) The superposition image in the eye of Ephestia kühniella. Journal of Comparative Physiology A 119, 73–84. Døving, K.B. and Miller, W.H. (1969) Function of insect compound eyes containing crystalline tracts. Journal of General Physiology 54, 250–267. Exner, S. (1875) Über das Sehen von Bewegungen und der Theorie des zuammengesetzten Augen. Sitzungberichte der Akademie der Wissenschaften in Wien; mathematische-naturlische Classen 72, (3), 156–190. Exner, S. (1886) Cylinder welche optische Bilder entwerfen. Pflügers Archiv für die gesampte Physiologie des Menschen und der Tiere 38, 274–290. Exner, S. (1889) Das Netzhautbild des Insectenauges. Sitzungberichte der Akademie der Wissenschaften in Wien. Abteilung 3, 98, 13–65. Exner, S. (1891) Die Physiologie der facettirten Augen von Krebsen und Insecten. Leipzig, Germany, Franz Deuticke. Translated by Hardie, R.C. (1988) The Physiology of the Compound Eyes of Insects and Crustaceans. Springer-Verlag, Berlin. Forel, A. (1908) The Senses of Insects. Translated by Yearsley, M. Methuen, London. Friedlaender, M. (1931) Zur Bedeutung des Fluglochs im optischen Feld der Biene bei senkrechter Dressuranordnung. Zeitschrift für vergleichende Physiologie 15, 193–260. Giger, A.D. (1996) Studies in honeybee vision. PhD thesis, Australian National University, Canberra. Giurfa, M., Vorobyev, P., Brandt, R., Posner, B. and Menzel, R. (1997) Discrimination of coloured stimuli by honeybees, alternative use of achromatic and chromatic signals. Journal of Comparative Physiology A 180, 235–243. Grenacher, H. (1879) Untersuchungen über das Sehorgan der Arthropoden, insbesondere der Spinnen, ­Insecten und Crustaceen. Vandenhoeck and Ruprecht, Göttingen, Germany. Hecht, S. and Wolf, E. (1929) The visual acuity of the honeybee. Journal of General Physiology 12, 727–760. Hertel, H. and Maronde, U. (1987) Processing of visual information in the centrally projecting visual interneurones in the honeybee brain. Journal of Experimental Biology 133, 301–315. Homberg, U., Heinze, S., Pfeiffer, K., Kinoshita, M. and El Jundi, B. (2011) Central neural coding of sky polarization in insects. Philosophical Transactions of the Royal Society of London B 366, 680–687. Hooke, R. (1665) Micrographia: or Some Physiological Descriptions of Minute Bodies Made by Magnifying Glasses. J. Martyne and J. Allestry, London. Horridge, G.A. (1968a) Pigment movement and the crystalline threads of the firefly eye. Nature, London 218, 778–779. Horridge, G.A. (1968b) The eye of Dytiscus (Coleoptera). Tisue and Cell 1, 425–442. Horridge, G.A. (1969) The eye of the firefly (Photuris). Proceedings of the Royal Society of London B 171, 445–463. Horridge, G.A. (1971) Alternatives to superposition images in clear zone compound eyes. Proceedings of the Royal Society of London B 179, 97–124. Horridge, G.A. (1972) Further observations on the clear zone eyes of Ephestia. Proceedings of the Royal Society of London B 181, 157–173. Horridge, G.A. (1974) Optical mechanisms of clear-zone eyes. In: Horridge, G.A. (ed.) The Compound Eye and Vision of Insects. Oxford University Press, Oxford, pp. 255–298.



The Fundamentals of the Insect Compound Eye

87

Horridge, G.A. (1976) The ommatidium of the dorsal eye of Cloeon (Ephemeroptera) as a specialization for photoreisomerization. Proceedings of the Royal Society of London B 193, 17–29. Horridge, G.A. (2005) The spatial resolutions of the apposition compound eye and its neurosensory feature detectors, observation versus theory. Journal of Insect Physiology 51, 243–266. Horridge, G.A. and Giddings, C. (1971) The retina of Ephestia (Lepidoptera). Proceedings of the Royal Society of London B 179, 87–95. Horridge, G.A. and Henderson, I. (1976) The ommatidium of the lacewing Chrysopa (Neuroptera). Proceeedings of the Royal Society of London B 192, 259–271. Horridge, G.A., Walcott, B. and Ioannides, A.C. (1970) The tiered retina of Dytiscus: a new type of compound eye. Proceedings of the Royal Society of London B 175, 83–94. Horridge, G.A., Giddings, C. and Stange, G. (1972) The superposition eye of skipper butterflies. Proceedings of the Royal Society of London B 182, 457–495. Horridge, G.A., Mimura, K. and Tsukuhara, Y. (1975) Fly photoreceptors II. Spectral and polarized light sensitivity in the drone fly Eristalis. Proceedings of the Royal Society of London B 190, 225–237. Horridge, G.A., Mimura, K. and Hardie, R.C. (1976) Fly photoreceptors III. Angular sensitivity as a function of wavelength and the limits of resolution. Proceedings of the Royal Society of London B 194, 151–177. Horridge, G.A., McLean, M. and Stange, G. (1977) A diurnal moth superposition eye with high resolution. Proceedings of the Royal Society of London B 196, 233–250. Horridge, G.A., Giddings, C. and Wilson, M. (1979) The eye of the soldier beetle Chauliognathus pulchellus. Proceedings of the Royal Society of London B 203, 361–378. Horridge, G.A., Marĉelja, L. and Jahnke, R. (1982) Light guides in the dorsal eye of the male mayfly. Proceedings of the Royal Society of London B 216, 25–51. Horridge, G.A., Marĉelja, L. and Jahnke, R. (1983a) Retinula cell responses in a moth superposition eye. Proceedings of the Royal Society of London B 220, 42–68. Horridge, G.A., Marĉelja, L., Jahnke, R. and McIntyre, P. (1983b) Daily changes in the compound eye of a beetle (Macrogyrus). Proceedings of the Royal Society of London B 217, 265–285. Ioannides, A.C. and Horridge, G.A. (1975) The organization of visual fields in the hemipteran acone eye. Proceedings of the Royal Society of London B 190, 373–391. Kirchhoffer, O. (1908) Untersuchungen über die Augen der pentameren Käfer. Archiv für Biontologie 2, 237–287. Kirschfeld, K. (1974) The absolute sensitivity of lens and compound eyes. Zeitschrift für Naturforschuing 29c, 592–596. Kirschfeld, K. and Snyder, A.W. (1975) Waveguide mode effects, birefringence and dichroism in fly photoreceptors. In: Snyder, A.W. and Menzel, R. (eds) Photoreceptor Optics. Springer, Berlin, pp. 29–41. Kuiper, J.W. (1962) The optics of the compound eye. Symposium of the Society for Experimental Biology 16, 58–71. Kuiper, J.W. (1964) On the optics of the superposition eye. Archives néerlandaises de zoologie 16, 171–173. Kunze, P. (1969) Eye glow in the moth and superposition theory. Nature, London 223, 1172–1174. Kunze, P. (1979) Apposition and superposition eyes. In: Autrum, H. (ed.) Handbook of Sensory Physiology, Volume VII/Part 6A: Invertebrate Photoreceptors. Springer, Berlin, pp. 441–502. Kuran, T. (1997) Private Truths, Public Lies. Harvard University Press, Cambridge, Massachusetts. Labhart, T. (1980) Specialized photoreceptors at the dorsal rim of the honey bee’s compound eye, polarization and angular sensitivity. Journal of Comparative Physiology A 141, 19–30. Labhart, T. (1988) Polarization-opponent interneurons in the insect visual system. Nature, London 331, 435–437. Land, M.F. (1981) Optics and vision in invertebrates. In: Autrum, H. (ed.) Handbook of Sensory Physiology, Volume VII/Part 6B: Invertebrate Visual Centers and Behavior. Springer, Berlin, pp. 471–592. Land, M.F. (1984) The resolving power of superposition eyes measured with an ophthalmoscope. Journal of Comparative Physiology A 154, 515–533. Land, M.F. (1985) The eye: optics. In: Kerkut, G.A. and Gilbert, L.I. (eds) Comprehensive Insect Physiology, Biochemistry and Pharmacology. Pergamon Press, Oxford. Land, M.F. (1989) Variations in the structure and design of compound eyes. In: Stavenga, D.G. and Hardie, R.C. (eds) Facets of Vision. Springer, Berlin, pp. 90–111. Land, M.F. (1997) Visual acuity in insects. Annual Review of Entomology 42, 147–177. Lotmar, R. (1933) Neue Untersuchungen über den Farbensinn der Bienen, mit besonderer Berücksichtigung des Ultravioletts. Zeitschrift für vergleichende Physiologie 19, 673–723. Meinertzhagen, I.A. (1976) The organization of perpendicular fibre pathways in the insect optic lobe. Philosophical Transactions of the Royal Society of London B 274, 555–596. Meyer-Rochow, V.B. and Horridge, G.A. (1975) The eye of Anoplognathus (Coleoptera, Scarabaeidae). Proceedings of the Royal Society of London B 188, 1–30.

88

Chapter 4

Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222. Rossel, S. and Wehner, R. (1987) The bee’s e-vector compass. In: Menzel, R. and Mercer, A. (eds) Neurobiology and Behavior of the Honeybee. Springer, Berlin, pp. 76–93. Santschi, F. (1911) Observations et remarques critiques sur le mécanisme de l’orientation chez les fourmis. Revue Suisse de Zoologie 19, 303–338. See also Memoires de la Societe Vaudoise des Sciences Naturelles 137 (1923). Schultze, M.S. (1868) Untersuchungen über die zusammengesetzten Augen der Krebsen und Insekten. Cohen, Bonn, Germany. Seidl, R. (1982) Die Sehfelder und Ommatidien Divergenzwinkel von Arbeiterin, Königin und Drohne der Honigbiene (Apis mellifera). PhD thesis, Technische Hochschule, Darmstadt, Germany. Seitz, G. (1969) Untersuchungen am dioptrischen Apparat des Leuchtkäferauges. Zeitschrift für vergleichende Physiologie 59, 205–231. Snyder, A.W. (1972) Coupled mode theory for optical fibres. Journal of the Optical Society of America 62, 1267–1277. Srinivasan, M.V. and Lehrer, M. (1988) Spatial acuity of honeybee vision, and its spectral properties. Journal of Comparative Physiology A 162, 159–172. Tuurala, O. (1954) Histologische und physiologische Untersuchungen über die photomechanischen Erscheinungen in den Augen der Lepidopteran. Suomalaisen Tiedeakatemian Toimituksia 24, 5–69. von, Frisch, K. Lindauer, M. and Daumer, K. (1960) Über die Wahrnehmung polarisierten Lichtes durch das Bienennauge. Experientia 16, 289–301. von Helversen, O. (1972) Zur spektralen Unterscheidsempfindlichkeit der Honigbiene. Journal of Comparative Physiology 80, 439–472. Wakakuwa, M., Kurasawa, M., Giurfa, M. and Arikawa, K. (2006) Spectral heterogeneity of honeybee ommatidia. Naturwissenschaften 92, 464–467. Walcott, B. and Horridge, G.A. (1971) The compound eye of Archichauliodes (Megaloptera). Proceedings of the Royal Society of London B 179, 65–72. Warrant, E.J. and McIntyre, P.D. (1996) The visual ecology of pupillary action in superposition eyes. Journal of Comparative Physiology A 176, 75–90.

Chapter 5 How Bees Distinguish Colours and Modulation

Scientists should be open, honest, rigorous, in their thinking and sceptical, especially of their own ideas. An effective scientific community should . . . encourage the constant challenge of data. (Nurse, 2015)

When considering how things work, humans tend to look first in the wrong direction, at performance, especially related to humans, and only look for mechanisms when shown the necessity and the means. From the swamps of anthropomorphism, we see the similarities between bee and human colour vision. We assume they are the same. We use the same words to describe them, then leap chasms of speculation and infer that bees have trichromatic colour vision. In his 1914 paper, von Frisch said as much, and for the rest of the century everyone had to agree. There was no alternative and the only show in town became a matter of national pride. The 20th century, the century of great advances in the sciences, saw little to be pleased about in honeybee colour vision (Chapters 1 and 2, this volume). In 2014, an expert straight from the school of von Frisch, Lindauer and Menzel, volunteered to me that the topic was a complete mess. It was impossible to think straight when the elegant models that ‘explained’ bee colour vision were literally superficial because nothing was known about what the bees a­ ctually detected

beyond the receptors. In the neural jungle were thousands of baffling neurons in parallel, passing through many layers of processing but recorded one at a time. It was like the stifling ignorance about planetary motion when Galileo was a boy. Nobody could think of a suitable experiment. Nobody was willing to break out, though a few freethinkers were kicked out. Table 5.1.  Calibrated values of the stimulus from papers. Relative receptor stimuli from the different papers relative to the white paper (100%), and contrasts between two equiluminant pairs of papers. White displays by far the greatest stimulus because white spreads maximum emission over the whole wavelength range of each receptor type. Canson papersa White copy paper Hemp 374 Ultramarine 590 Green 576 Buff 384 Blue 595 Dresden Yellow Contrast 374/590 Contrast 384/595

Blue receptor (%)

Green receptor (%)

100 34.2 33.8 17.0 25.7 54.2 13.1 0.006b 0.36

100 56.3 20.7 22.3 41.7 40.0 78.1 0.46 0.02b

To avoid confusion, the manufacturer Canson’s colour names and numbers are used. See: http://www.canson-­ infinity.com/en/values.asp (accessed 1 November 2018). b These are the values of contrast at a boundary between the two colours identified by their code numbers in the table. a

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)

89

90

Chapter 5

I thought that perhaps bee vision was just a simplified version of vision in man, where we knew that the inputs were phasic derivatives of receptor responses, and colour is not a simple sum of three inputs, but is hallucinated inside the head. However, ‘For all its weary years of thought: The starkest fights must still be fought, The most surprising songs be sung’ (from ‘A New Year’s Carol’ by James Elroy Flecker; Squire, 1918). So, I set to work. Endless sunshine and an even 17–25°C in spring and autumn makes Canberra, Australia, one of the best places in the world to study the vision of bees. My arms were neither rosy nor youthful but no matter, I would use some weary years of thought. First, you find the best material for the targets. All the hard work had been done by Miriam Lehrer, who taught us how to train and test bees, and by Srinivasan who calibrated for me coloured papers that are manufactured to a constant colour and available worldwide. Intensity of colour is measured in two ways. First, the light is absorbed in a black hollow sphere, called a bolometer, and the energy released as heat is measured as a rise in temperature. This gives the energy in the light in watts, or equivalent units. Animal eyes, however, catch the light with rhodopsins that absorb it as photons, absorbed one at a time by individual pigment molecules. In vision, the correct measure of the intensity is the photon flux, which is the number of photons at each wavelength arriving per second per square centimetre, or the number absorbed per second per receptor cell. One can buy equipment that does the job. Colours are usually considered to be different ratios of photon flux of different wavelengths. Recently I found that bees have a constant order of preference in the learning process, and that the preferred cue of the green receptor channel is the modulation of the signal, not the photon flux. So, with UV excluded, many of my experiments tested whether the bees used either the photon flux or the modulation of either the green or the blue receptors or both. All it takes is for someone to point the way, because the task is not difficult. A repetition of the Hess training experiment

(Fig. 1.1A), with a few additional tests, would have been sufficient. Hess trained bees on a four-by-four chequerboard of blue and yellow squares. When trained on the blue, they went faithfully to blue, but when trained on yellow, they had no particular preference for any colour. They were at a loss; and so was von Hess, and with him a century of error. Maybe other efforts were made, and lost data may be discovered in the archives. One of my early experiments was to train on blue then test with blue versus white. Surprisingly, the trained bees preferred white. At the time, it was a wake-up call but I could not distinguish between the several possible explanations. Another experiment was to train on a black square on a white or grey background, then test with the same square versus a coloured one. When tested, the trained bees always preferred the yellow end of the spectrum. Of course, this was exactly the discovery made by Kriston (1973) in a paper that was never cited in the literature. In the above experiments, bees preferred the greater content of blue in the white than in blue itself (Horridge, 2014), and when trained on black, they learned the width between the green modulation at the edges (Fig. 3.11B). What the bees actually detect is much easier to discern when we train with two ­papers that are equiluminant to the green or alternatively to the blue receptor channel, so that the part played by each channel can be examined separately (Fig. 5.1). Then we learn that they are separate all the way to memory. The experiments could have been arranged any time after 1964, when the spectral sensitivities of the receptor types were first measured (Fig. 3.1B), but no one doubted the orthodox story for a further 50 years. In the flies and bees, probably in all Endopterygota insects, axons of green-sensitive receptors 1–6 in each ommatidium end at the lamina, and only their phasic signals are effective. Studies of the channel by Laughlin and Hardie in Canberra in 1978 showed that responses of 2nd order neurons of the lamina were optimized to detect contrast irrespective of intensity. (Actually, they never detect contrast, only modulation.) Bees certainly detect, locate and measure



How Bees Distinguish Colours and Modulation

Train with equal contrast to green receptor channel

+ (A)

(B)

100%

_ 55°

91%, n = 200 Bees compared blue contents of each display. Test Test (D) 100% 100%

69%, n = 200

26%, n = 200 Trained bees avoided the greater blue content of white.

Trained bees avoided the greater blue content of the grey.

Test with buff versus 25% white

Test with buff versus black (C)

91

100%

26%, n = 200 Trained bees avoided the greater blue content of the buff.

(E)

100%

54%, n = 200 A grey level could be found that was confused with buff.

Fig. 5.1.  Discrimination of two colours equiluminant to the green receptors so green modulation was not learned. Percentage values indicate the percentage of bees visiting reward holes. (A) Training patterns, buff versus blue. (B–D) Irrespective of the colours, the trained bees avoided the greater blue content, and they could not recognize either training pattern. (E) A grey level that matched the blue content of buff.

modulation. They do not care which side of the edge is which, as was discovered by Friedlaender (1931) long ago. This is difficult to imagine until we realize that all symmetrical detectors with short time constants behave so. A symmetrical detector that scans an edge cannot detect the polarity because the direction of scanning is also known, although the edge position, modulation and orientations are located and learned. The result certainly limits what can be distinguished. In training and testing experiments, the green receptor channel measures the total of amount of modulation as: (vertical bar edge length multiplied by green contrast in each part of the edge). The bee never detects contrast alone because it is replaced by something that depends also on the structure in the whole visual field. Green modulation is retinotopic in the horizontal direction, and the compound eye is a device that measures angles between

vertical edges. Colours of grating bars, but not their edges, are smoothed together in the blue receptor channel in the horizontal direction and an average blue content is measured and located in the vertical direction, so all cues are extremely astigmatic. In this chapter, it is important to keep an eye on the illustration while following the description of it in the text. All the data was published in five recent papers (Horridge, 2014, 2015a, b, c, 2016).

Discrimination With No Difference in Green Modulation Let us begin the series of new experiments with a Canson buff paper versus a Canson blue of the same area, each on a black ­background, and with equal green stimulus and indistinguishable by green receptors

92

Chapter 5

(Table 5.1). Despite being restricted to the blue channel, and with UV excluded, bees discriminate very well (Fig. 5.1A). In brief, when trained on two equal areas of colours that were equiluminant to the green receptor channel, bees learned by trial and error to avoid the greater blue content of the unrewarded target (Fig. 5.1B–D). When a grey level of 25% white matched the buff (Fig. 5.1E). The trained bees failed to recognize either of the training patterns.

Blue Content and Green Contrast Are Measured A difference in blue content is usually accompanied by a difference in green modulation, but

bees separate them. Two blue squares of different sizes were discriminated by blue content and by green modulation at the vertical edges (Fig. 5.2A–D). When the shapes were adjusted to make blue content equal, the bees distinguished the green modulation (Fig. 5.2F–H).

Discrimination Between Buff or Yellow, Versus Green or Blue This is the oft-repeated experiment that seemed to demonstrate colour vision. However, when tested, trained bees show that they have measured the blue content and the green modulation, so both colour channels discriminated. On the other hand, when trained to distinguish buff and green or blue that differed

Train versus twice as much blue and more green edge (A)

100%

Train versus equal blue and more green edge (E)

100%

55°

(B)

(C)

91.5%, n = 200

83.0%, n = 200

Test versus black

Test with no green contrast

100%

15.0%, n = 200 They fail to recognize the rewarded pattern, and avoided blue. Test 100%

(F)

They failed. 47.0%, n = 200 They did not learn shape. Test with green contrast, no blue (G)

35.5%, n = 200 They fail to recognize the unrewarded pattern, and avoided blue. Test with equal amounts of green contrast and blue (D)

100%

100%

81.0%, n = 200 Test with black bars on white; little blue difference (H)

100%

100%

91.5%, n = 200 They failed. 48.5%, n = 200

They avoided green modulation.

Fig. 5.2.  Bees learn blue content independently of green modulation. (A) Training patterns. (B, C) Bees trained on (A) avoid the greater blue content. (D) They fail when the blue content is equal. (E) Bees learn to avoid the greater green modulation at the vertical edges of blue. (F) They fail when the green contrast is removed at those edges. (Shape difference is detected by retinotopic green contrast). (G, H) They avoid the greater green modulation irrespective of colour or pattern. Discrimination of width was not tested, but was weak, as shown by (B–D).



How Bees Distinguish Colours and Modulation

93

Trained with a blue difference, they avoided green modulation and greater blue content. (A)

(B)

+

_

88%, n = 200 Contrasts were measured against white background (arrows). Test Test (D) 100% 100%

40%, n = 100 Equal green contrast so they avoided greater blue content. (C)

100%

Test 100%

96%, n = 200 Avoided greater blue content and green contrast.

c

49%, n = 200 They fail with no green contrast and equal blue content. Test (E) 100%

25%, n = 100 When equiluminant to blue, they avoided greater green modulation.

Fig. 5.3.  Discrimination of yellow versus green or blue. The green has more blue content and more green contrast against the white background (Table 5.1). Arrows show the location of green contrast. (A) Training patterns. (B, C) Trained bees avoided the greater blue content. (D) With no green contrast and equal blue content, they failed to find a cue. (E) They avoided the greater green modulation.

in blue content and green contrast, against a  white background (Fig. 5.3A), they responded to the difference in blue (Fig. 5.3B), and green modulation at the outer edges (Fig. 5.3C). This was a common type of response. In Fig. 5.3(D), when tested with no green contrast and equal blue content, the trained bees were lost. The final test with a pair of gratings equiluminant to blue receptors, and with equal blue content (Fig. 5.3E), confirmed that they avoided the greater green modulation. Two equiluminant gratings with differing period provide a useful test for identifying and validating the learning of green or blue modulation.

Green Versus Grey and Black Versus Blue Next, we turn to a group of green, grey, black or white papers because, since the original experiments of von Frisch (Fig. 1.2B–D), they have been repeatedly reported as anomalous. When trained on green versus grey of 70% white, and then tested with green versus

black, they reversed their preference for green, and went to black or yellow that had lower blue content (Fig. 5.4A–C). The result depended completely on the amount of white, and therefore blue, in the training displays. When trained on black versus blue (Fig. 5.4D), green contrast was saturated on both targets, and therefore equal. When tested on white versus blue, they preferred blue and avoided white that had greater blue content than any other colour (Fig. 5.4E). When tested on hemp and ultramarine that were equiluminant to blue receptors, the only cue they had learned, a difference in blue content, was unavailable and they failed (Fig. 5.4F). In the next experiment, when trained on green versus white (Fig. 5.5A), bees avoided the greater blue content, as shown by testing with green versus grey or hemp yellow (Fig. 5.5B–C). The trained bees could not distinguish between green colour and panels composed of half buff and half blue with no green contrast, even though the blue was very obvious, because they failed to ­detect the boundary between buff and blue (Fig. 5.5D–E), and summed blue content

94

Chapter 5

Train green versus 70% white

Train black versus blue

100%

(A)

(D)

100%

55°

55°

91%, n = 200 They learn to avoid most blue.

95.0%, n = 200 They learn to avoid blue.

Test, green versus black

Test with white versus blue

100%

(B)

(E)

12%, n = 200

15.0%, n = 200 White displayed more blue.

They avoid most blue. Test, yellow versus any colour

Test, equiluminant to blue (F)

100% any other colour, grey or white

(C)

100%

100%

51.0%, n = 200 The only cue was removed.

91%, n = 200 Preferred yellow; to avoid blue.

Fig. 5.4.  (A) Training patterns of green versus grey of 70% white (= super blue) both on a white background, with saturated (and therefore equal) green contrast. (B) Trained bees prefer black because they learned to avoid the greater blue content in the grey. (C) Tested with yellow versus any other colour, they go to yellow because they avoid the greater blue content. (D) New training patterns, black versus blue. (E) When tested with white versus blue, they now avoid the greater blue content in the white. (F) When tested with colours equiluminant to blue receptors, they fail. Train on green versus white 100%

(A)

55°

94.0%, n = 200 after 2 h They learned to avoid the blue content of white. (B)

Test 100%

(D)

64.5%, n = 200 (C)

100% Test

Test 100%

52.0%, n = 200 (E)

100% Test

45.0%, n = 200

55.0%, n = 200

They avoided the most blue.

Buff plus blue = green

Fig. 5.5.  (A) Train on green versus white. (B, C) Tested on green versus grey or yellow, trained bees avoided the most blue. (D, E) Tested against equiluminant colours, trained bees could not recognize the green they were trained on.



How Bees Distinguish Colours and Modulation

95

Train, green versus black

(A)

A very easy discrimination, but what had they learned?

100% 55°

85.0%, n = 200

(B)

100% Test

The strongest input was at the edge of black, but is now on both targets.

47.5%, n = 200

(C)

100% Test

The strongest input is now contrast at the vertical edges of black.

86.0%, n = 200

(D)

100%

They preferred the yellow to the green because they avoided the greater green contrast at the edges of green.

Test 24.5%, n = 200

Fig. 5.6.  (A) Training patterns, green versus black. (B) The trained bees could not distinguish black from white with black edges. (C) Blue content was not a cue. (D) They avoided the greater green contrast and switched preference. Training with green versus randomized grey, black, white Train, green versus various greys

(A)

Green

100%

Grey, black, white sequence

55°

The bees might learn an average grey level (half white = some blue) and avoid green contrast at edges of dark displays.

89.5%, n = 200

(B)

100%

A result suggesting they learned both a greater blue content and green modulation at edges.

Blue

Green Test

96.0%, n = 200

(C)

100%

The result suggests they avoided greater green modulation, but blue content was not important.

Green Test

97.5%, n = 200

(D)

Green

This result shows they avoided greater green modulation against white, and blue content was not important.

100% Yellow Test

16.5%, n = 200 Train green versus grey, test with yellow against any other display

(E)

100% Green

Grey, 40% black 91%, n = 200

(F)

Yellow

100%

Test black, white or colour

They always avoided blue, went to yellow.

Fig. 5.7.  (A) Bees were trained on a sequence of black, grey and white. (B–D) They avoided greater green modulation and greater blue content. (E) When trained on green versus grey (60% white), they preferred yellow to any other colour because they had learned to avoid blue.

96

Chapter 5

over the whole of each target (compare with Fig. 3.5). Next, bees were trained on green versus black (Fig. 5.6A) and given a variety of tests to see what they had learned. They avoided black edges but the very strong green contrast overwhelmed the blue content of the white, although there had been a blue difference in the training patterns (Fig. 5.6B, C). They preferred yellow to green because they had learned to avoid the greater green contrast against white at the outer edges (Fig. 5.6D). Finally, in this section, bees were trained to go to green versus a shuffled sequence of black, grey and white (Fig. 5.7A). They avoided greater green modulation and greater blue content (Fig. 5.7B–D). When trained on green versus grey 60% white (Fig. 5.7E), they preferred yellow to any other colour because they had learned to avoid blue (Fig. 5.7F). In all these experiments, green behaved as a normal colour. Despite the variety of training and testing colours,

bees only made use of a difference in blue content and/or green modulation.

Bees Locate and Measure Total Content of Blue A blue square on a grey background of 40% white was discriminated from a plain grey target, also 40% white (Fig. 5.8A). The rewarded target had more blue content because the blue paper displayed the blue equivalent of 54.2% white but displaced a grey area of 40% white. A test with the same area of blue on each target, but with different length of edges that displayed some green and blue modulation (Fig. 5.8B) showed that a modulation difference had also been learned in the training. When the rewarded target was tested versus plain grey of 50% white, the additional blue in 50% white reduced the score

Train with excess blue content and low green contrast (A)

100% 55°

on 40% white

on 40% white 89%, n = 100

(B)

Test 100% 40

40

50

35%, n = 100 With equal blue content, they were attracted to green modulation. (C)

Test 100% 50

Greater blue on the left.

40

Prefer

60

33%, n = 100 Greater blue on the right, but attracted to green modulation.

Test 100%

(G)

36%, n = 100 Greater blue on the right although attracted to green modulation.

Test 100%

60

75%, n = 100 Greater blue on the left.

Prefer

50

Prefer 70%, n = 100

(F)

40

(D)

Test 100%

(E)

60

40

100% Test 40 62%, n = 100 With similar blue content attracted to green modulation.

Fig. 5.8.  Blue content is summed over the whole target. Numbers in bold show the percentage of white in the grey background squares. (A) Bees were trained with two grey targets of 40% white, one with a blue square that displayed 54.2% white and obscured some grey. (B) With equal areas of blue on 40% white, they preferred the target with most vertical edge. (C– F) In tests with grey backgrounds of selected content of white (as indicated in bold numbers) the trained bees preferred the target that displayed most blue content summed over the whole target. (G) When blue content was similar, they preferred green ­modulation.



How Bees Distinguish Colours and Modulation

97

Train with equal blue content and no green contrast (A)

100% 55°

89%, n = 100 Position of blue and blue modulation were the only cues. (B)

Test 100%

(E)

65%, n = 100 The detector had a large field not sensitive to vertical shift. (C)

100%

34%, n = 100 White was bluer than background; preference was reversed. (F)

Test

(D)

Test 100%

100% Test

72%, n = 100

69%, n = 100

Increase in green modulation had little effect.

Blue modulation was detected.

100% Test

(G)

100% Test

60%, n = 100

67%, n = 100

Reversal of contrast had little effect irrespective of colour.

Blue modulation was detected irrespective of contrast reversal.

Fig. 5.9.  Local position of blue and blue contrast was learned. (A) Bees were trained to locate a buff square on blue, with no green contrast. (B, C) Moving the squares up or adding green contrast had little effect. (D) Reversal of contrast had little effect so bees had located blue contrast, not blue content. (E) White squares reversed the preference by reversing the average position of blue content. (F, G). Relative positions of blue modulation had been learned.

(Fig. 5.8C), but plain grey of 60% white, made the bees reverse preference (Fig. 5.8D). Similarly, with 50% white on each background, the blue square was preferred (Fig. 5.8E). With 60% white on each background, the preference was again reversed (Fig. 5.8F) because the square was less blue than background grey. Despite the lack of blue, a buff square on 40% white (with similar green contrast) was preferred to plain 40% white (Fig. 5.8G), because some blue or green modulation had been learned. These results show that the blue content was measured over the whole target and the content of blue in the grey background was just as effective as the blue displayed in the blue square. The field size for measuring blue content must have been large because blue content in the small square was summed along with the rest of the target, not separated from background. In the training, the bees were obliged to measure total blue over

each whole target because one target was blank with no pattern at all. In e­ xperiments of this type, bees detected a difference of 5% in blue content on targets this size. The next experiment, however, shows that, in the ­absence of green contrast, bees could locate blue modulation.

Vertical Location of a Small Coloured Patch After a brief training, a small buff square at the bottom right corner of the rewarded target on a blue background was distinguished from a similar square at the bottom left corner of the unrewarded target (Fig. 5.9A). There was no difference in green contrast (Table 5.1). The squares could be moved upwards on their targets and still be detected, but by fewer bees (Fig. 5.9B). Addition of a

98

Chapter 5

(A)

Test with black spots

Train 100%

(D) 55°

(B)

90%, n = 200

70%, n = 100

Test with no green contrast

Test with blue spots

100% Test

(E)

72%, n = 100 When eqiluminant to green receptors, they used the position of blue content.

100% Test

39%, n = 100 When equiluminant to blue receptors, they used the only remaining cue, position of green contrast.

100% Test

24%, n = 100 The preference was reversed because the spots appeared more blue than adapted background.

Test with no blue contrast (C)

100% Test

Test with green spots (F)

100% Test

49%, n = 100 The green spots were equally attractive because the position of blue in the displayed white just balanced the position of green contrast.

Fig. 5.10.  Blue receptors adapt to blue content of white background. (A) Bees were trained to distinguish the location of black in the vertical direction. (B) When tested with patterns of buff and blue, equiluminant to green, they preferred blue at the bottom. (C) When equiluminant to blue they preferred the green modulation at the top, as in the training (arrows). (D) With black spots, they followed their training. (E) With blue spots, the preference reversed, because blue spots, ringed by green contrast, appeared bluer than adapted background white. (F) A spot colour can be found such that the average position of blue (in the combined white and green on each target) cancelled the position of green contrast, giving a null response.

narrow yellow margin to each square reduced the score (Fig. 5.9C) because green contrast inhibited blue contrast. Some recognition persisted even after reversal of the training colours (Fig. 5.9D), showing again that blue contrast had been learned. Replacement of buff squares by white reversed the preference (Fig. 5.9E), because the centre of blue content was shifted vertically. Positions of hollow blue squares on a buff background, and hollow buff squares on a blue background, were distinguished (Fig. 5.9F, G), confirming that locations of blue modulation and local blue content had been learned. These locations must have been learned relative to landmarks, probably strong green contrast at edges in the ­apparatus, or at the edges of the targets.

The Bees Learned the Position of Blue in the Vertical Direction The bees were trained on a black panel above a white one versus a white panel above a black one (Fig. 5.10A). A white background reflects blue (Table 5.1; in the table we find that the stimulus from blue paper in sunlight to the blue receptors is 54.2% of that of white paper, and to the green receptors is 40% of white) but inside a white apparatus the response to white adapts. This is an example of a task in which preference for one cue is balanced against another so that trained bees ­detect no difference between targets that are obviously different to naïve bees. When tested with targets that were equiluminant to the green receptors (Fig. 5.10B) the trained bees



How Bees Distinguish Colours and Modulation

detected blue where it had been learned in the white training areas. Tested with targets equiluminant to the blue receptors (Fig. 5.10C) they reversed their performance because they responded to the position of strongest green contrast at the outside edges of the ultramarine (arrows). With black spots, they behaved as expected (Fig. 5.10D). They had learned both cues. However, with blue spots, trained bees reversed their preference (Fig. 5.10E) because blue paper surrounded by green contrast in the position where they had learned the absence of blue was a strong stimulus. The two targets became indistinguishable to the bees when the spots were billiard green on a white background although the spots were in different positions (Fig. 5.10F), because the vertical difference in the position of green contrast cancelled the vertical difference in amount of white covered by green, as if (D) and (E) were added together, and preferences cancel. The trained bees detect the targets but they cannot decide between the conflicting inputs, because they look for the cues that are in their memory. Of course, these green spots would be discriminated if the bees were trained on them, starting afresh. Trained bees do not behave like untrained ones. A recent study claims that bees can learn whether a black pattern on a white background is above or below another pattern at a different height, irrespective of the pattern (Guiraud et al., 2018). However, there were no tests whether the bees actually detected nothing but the height of blue, as could be tested by the use of patterns that were equiluminant to the blue receptors. It is more likely that the bees detected the height of the blue in the white background.

Green and Blue Modulation Blue content is only half of what bees detect in an area surrounded by green contrast; the other half is the location and measure of stimulus put into the eye by a scan of the flying bee. Contrast is the change in stimulus detected as a photoreceptor scans across a boundary or an edge against background. The modulation is the re-

99

sponse proportional to the (angular) vertical component of total length of edges in view, because the bee scans in the horizontal plane. It is also proportional to the contrast at each part of the edge, which is detected by green and blue receptor channels acting separately. Contrast is a principal variable in human psychophysics because the contrast defines edges of objects and areas that we recognize and categorize. Human vision cannot avoid seeing contrast; it is impossible not to see the polarity at a single edge, and the panorama is inconceivable without directional steps between areas of colour, black or white. Bees respond to contrast with receptors, but they cannot separate out the effect of the edge length at the next stage, the summation of inputs, and they detect modulation. Therefore, they cannot see as humans do. This experiment demonstrates the fundamental importance of modulation as opposed to contrast. After training to prefer the pattern with less edge, with no green contrast (Fig. 5.11A), bees preferred a pattern with even less edge, and abandoned the rewarded target in the training (Fig. 5.11B). They learn a relative, not absolute, measure of blue modulation. When tested with an equal number of bars on each target (Fig. 5.11C), the trained bees were lost, although the period on each target was identical to that in the training. Therefore, they had learned a measure of modulation from the total length of vertical edge on each target, not the patterns or widths of bars. It must have been blue receptor modulation because there was no green contrast in the training targets. Using the same cue, the trained bees easily distinguished between targets with the same period but different numbers of bars (Fig. 5.11D), but were lost with gratings that displayed green contrast and equiluminant to blue receptors (Fig. 5.11E). When trained on similar gratings that were equiluminant to the blue receptors (Fig. 5.11F), the same tests gave similar results (Fig. 5.11G–J) because the properties of green and blue contrast are similar, but this time the trained bees were unable to distinguish gratings with no green contrast (Fig. 5.11J).

100

Chapter 5

Train, equiluminant to green, and no colour difference (A)

Train, equiluminant to blue and no colour difference (F)

100%

100%

55° 75.0%, n = 200 (B)

100% Test

76.5%, n = 200 (G) They preferred smaller period

100% Test

41.0%, n = 200 34.0%, n = 200 Trained bees avoided the greater modulation. (C)

Test with blue contrast 100% Test

Test with green contrast (H) Period ignored

49.5%, n = 200

100% Test 53.0%, n = 200

There was now no difference in modulation in either channel. (D)

Test 100%

(I) Period ignored

Test 100%

71.0%, n = 200 69.5%, n = 200 In each case, the trained bees avoided the greater modulation. Test with no green contrast

Test with no blue contrast (E)

100%

Each (J) channel acted separately

100%

c

48%, n = 100 49.5%, n = 200 In each case, they had not learned the other modulation cue. Fig. 5.11.  Bees measure modulation, which includes contrast multiplied by length of vertical edge. (A and F) One group of bees was trained on gratings of (A) buff and blue with no green contrast, or (F) ultramarine and hemp, with no blue contrast. (B and G) In tests, they avoided the target with most vertical edges, irrespective of pattern. (C and H) In each case, they failed to avoid the unrewarded target when tested with equal pattern period and equal number of edges. (D and I) In each case, they avoided the greater amount of modulation. (E and J) Bees trained on colours equiluminant to green failed when tested on colours equiluminant to blue; and vice versa, so the green and blue receptor channels are separate and each group of bees used only the training cue.

Green Modulation Inhibits Blue Channel Modulation A fresh group of bees was trained to distinguish between a vertical grating of buff on blue from the same grating horizontal, with both on a black background. There was no green contrast in the grating and no differ­ence in

contrast around its edges; but a difference in blue modulation was available. The bees learned rapidly to a high score (Fig. 5.12A). When tested with similar patterns in black on white, the trained bees failed, although the blue modulation was greater than in the training (Fig. 5.12B). Failure was caused by the appearance of strong green contrast,



How Bees Distinguish Colours and Modulation

101

Train, with no green contrast and no blue difference (C)

100%

Train, with no green contrast and no blue difference (A)

83.5%, n = 200 Test rotated

100% 55°

(D)

88.5%, n = 200 They learned only blue modulation. Test with black on white (B)

100%

49.%, n = 200 Test with green contrast added (E)

100%

100% Test

48.0%, n = 200 Blue modulation inhibited.

60.5%, n = 200 Blue modulation inhibited.

Fig. 5.12.  Green modulation cancelled blue modulation, and bees distinguished angular width between two vertical bars with blue modulation. (A) Training patterns with a difference in blue modulation but equal green contrast and blue content. (B) Failure when tested in black on white, because strong green contrast inhibited the use of blue modulation. (C) New training patterns. (D) The rotated patterns were not ­distinguished, because there was no cue. (E) Addition of a black vertical bar to each display reduced the discrimination by the position of blue modulation. Train with no green contrast (A)

100%

55°

91%, n = 200 Test with less blue difference (B)

100%

69%, n = 100 Less blue difference, smaller score. Test with no blue difference (C)

100%

79%, n = 100 Blue modulation is revealed.

Test with colours reversed (D)

100%

69%, n = 100 They avoid the greater blue. test, rewarded target versus yellow (E) 100%

23%, n = 100 Rewarded target is not recognized.

Fig. 5.13.  Blue modulation and blue content were learned simultaneously but separately. (A) A broad buff bar on blue displayed a position, a width, a difference in blue modulation and blue content, but no green modulation. (B) A thin bar was less effective because less blue difference. (C) With equal blue content, blue modulation was still available. (D) The greater blue content in a blue line on buff was avoided because it was stronger than the blue modulation cue. (E) Yellow was preferred to the rewarded target because the bees were trained to avoid most blue.

102

Chapter 5

which cancelled the cue that the bees had learned, and perhaps also by flooding the targets with a strong blue content. A fresh group of bees was trained to distinguish two widely spaced vertical blue bars on a buff background from the same bars placed closer together (Fig. 5.12C). These targets conveyed nothing when the bars were horizontal (Fig. 5.12D). The training targets were equiluminant to the green receptors; therefore, the bees looked for ­ retinotopic blue modulation. However, the addition of a single black vertical bar with green modulation at the centre of each training target destroyed the discrimination (Fig. 5.12E).

wide versus the plain background (Fig. 5.13B) or versus a horizontal 2° bar (Fig. 5.13C). They had detected the differ­ence in modulation by scanning, not the orientation because they could not learn orientation with the blue modulation pathway. When tested with a plain buff target versus a 4° blue bar on a buff background, with no green contrast (Fig. 5.13D), the trained bees avoided the small extra blue content. They ignored the blue contrast ­ because the bar had been moved. When tested versus yellow, they avoided blue (Fig. 5.13E) showing again that in the training they had learned to avoid the greater blue content.

With No Green Contrast, Blue and Blue Modulation Were Learned

Bees Detected Amount of Blue and Blue Modulation Separately

In this example, the bees were trained to distinguish a vertical buff bar 10° wide on a blue background, with no green contrast, from a plain blue target (Fig. 5.13A). The trained bees could easily distinguish a bar only 2°

A fresh group of bees was trained to distinguish a vertical buff bar (16° × 55°) on a blue background from a plain blue target with no green contrast (Fig. 5.14A). With reversed colours, the response was reduced be­cause Train 100%

(A)

55°

16°

95%, n = 100

Test with blue bar on buff (B)

100%

Test with no green contrast (D)

4° 69%, n = 100 Blue modulation opposes blue content.

74%, n = 100 They prefer blue modulation.

Test with no blue difference (C)

100%

49%, n = 100 They detect no cue.

100%

Test with no blue difference (E)

100%

53%, n = 100 They detect no cue.

Fig. 5.14.  In the absence of green contrast, a measure of blue and blue modulation were preferred. (A) A buff bar 16° wide on blue displays a difference in blue modulation and blue content but no green modulation. (B) Reversal of the colours reversed the preference for the targets but not the cues. (C) Ultramarine and yellow rectangles with no blue difference were not distinguished. (D, E) Tests with gratings with no green contrast and no blue contrast show that the trained bees avoided the greater blue modulation.



How Bees Distinguish Colours and Modulation

the two cues acted in opposition, but blue content difference was the stronger cue (Fig. 5.14B). With two patches of colour with no difference to the blue receptors, the bees failed (Fig. 5.14C). The trained bees responded to a difference in modulation of blue receptors (Fig. 5.14D), but not of the green receptor channel (Fig. 5.14E).

Bees Distinguished Width with Blue Modulation, and Amount of Blue Next, a group of bees was trained to distinguish a buff bar 6° wide from a similar bar 12° wide, both on blue backgrounds (Fig. 5.15A). The trained bees failed to distinguish between a 16° bar and a 12° bar (Fig. 5.15B), and had difficulty with a 6° bar and an 8° bar (Fig. 5.15C), all on the same blue

103

background, showing that whatever had been learned was accurately measured. Tests with each training target versus plain blue showed that something had been learned from each (not illustrated). In a test with thin buff bars 2° wide replacing the edges of the training bars, the trained bee distinguished very well, showing that an ­accurate location of the blue modulation at both edges of at least one bar was a preferred input (Fig. 5.15D). In a test with a 6° bar versus a pair of edges of the same width (Fig. 5.15E), or with one bar versus two bars (Fig. 5.15F), the trained bees avoided the extra modulation, and they also avoided the target with less blue content (Fig. 5.15G). To distinguish the bars in this experiment the bees located two positions of blue modulation. Intrinsic to the eye anatomy, they had available the angle between these inputs. They remembered the difference

Train, with no green contrast (A)

100% 55° 6°

Test 100%

(B)

16° (C)

(D)

(E)

48%, n = 100

63%, n = 100 100% Test

12° Test 100%

69%, n = 100 (F)

100% Test

6°

89%, n = 100

8°

100% Test

66%, n = 100 (G)

Test

100%

80%, n = 100 31%, n = 100 They learned width and less blue from negative target. Fig. 5.15.  Small differences of blue content were measured, and widths of two bars were compared using blue modulation on both targets. (A) Training patterns, equiluminant for green receptors. (B) The difference in width cancelled a difference in the blue content. (C) The trained bees went towards more blue and an expected width. (D) Widths between lines of blue modulation were compared. (E, F) Probably there were conflicting cues, blue content, width between blue modulation lines, and unexpected extra blue m ­ odulation. (G) The strongest cue was the less blue on the unrewarded target. They did not recognize the rewarded target.

104

Chapter 5

­ etween 6° and 12° separations of edges with b blue modulation and detected a small differ­ ence in blue content.

Colour Vision of Small Floral Guides A minimum measure for detection of a t­ arget with colour and green contrast ­difference is 4.0° ± 0.5° for the frontal part of the eye. For the ventral eye the minimum is 7.1° ± 0.5° (Giurfa et al., 1999), although we regularly measure a resolution of green modulation down to 2° (Horridge, 2005). Honeybees are easily observed touching honey guides down to 2  mm in diameter. Perhaps this is done with the frontal part of the eye, perhaps with both sides collaborating.

Colour Vision of Bees With Clamped Head At first, it was discovered that bees do not distinguish colour or pattern when the head is held firmly in a clamp relative to the display. Later, it was found that they detect blue. Finally, at least two research groups have found that bee vision is relatively normal if the display moves relative to the clamped eye (Hori et al., 2006; ­Dobrin and Fahrenbach, 2012; Riveros and Gronenberg, 2012).

Other Species of Bees Some investigations have been made using the test against grey papers, but so far there

is nothing with a significant number of tests of bees trained to discriminate between a variety of colours. Most likely, all insects will detect, measure, and locate blue and green contrast, and some will have additional yellow- or red-sensitive copies of cell 8 for detection of particular foods, food plants, mates or prey. The place of UV in vision is still not fully explored, but probably UV does not contribute to hue of colour. In the honeybee, UV inhibits detection of blue and white, but may assist nocturnal bees.

Summary In insect vision, the layout of a coloured pattern cannot be reconstructed because, right at the start, almost all spatial distribution of contrast in the image is thrown away by summation that reduces the information load. Adding to the essential reduction of information, the green receptor pathway in bees is not sensitive to differences in brightness (except near threshold). Bees certainly detect colour differences by blue content, and green and/or blue modulation, and width between vertical edges with green or blue contrast at vertical edges (Figs 5.11B, 5.15), providing a rich, consistent and ­memorable input as they scan. Informative ­studies of the part played by modulation differences in the foraging behaviour are hard to find because the relevant variable was thought to be colour, so contrast and lengths of edge were neglected in field studies of foraging bees.

References Dobrin, S.E. and Fahrenbach, S.E. (2012) Visual associative learning in restrained honeybees with intact ­antennae. PLoS One 7(6), e37666. Friedlaender, M. (1931) Zur Bedeutung des Fluglochs im optischen Feld der Biene bei senkrechter Dressuranordnung. Zeitschrift für vergleichende Physiologie 15, 193–260. Giurfa, M., Zaccardi, G. and Vorobyev, M. (1999) How bees detect coloured targets using different regions of their compound eyes. Journal of Comparative Physiology A 185, 591–600. Guiraud, M., Roper, M. and Chittka, L. (2018) High-speed videography reveals how honeybees can turn a spatial concept learning task into a simple discrimination task by stereotyped flight movements and ­sequential inspection of pattern elements. Frontiers in Psychology. DOI: doi.org/10.3389/fpsyg.2018.01347



How Bees Distinguish Colours and Modulation

105

Hori, S., Takeuchi, H., Arikawa, K., Kinoshita, M. and Ichikawa, M. (2006) Associative visual learning, color discrimination and chromatic adaptation in the harnessed honey bee Apis mellifera. Journal of Comparative Physiology A 192, 691–700. Horridge, G.A. (2005) The spatial resolutions of the apposition compound eye and its neurosensory feature detectors, observation versus theory. Journal of Insect Physiology 51, 243–266. Horridge, G.A. (2014) How bees distinguish black from white. Eye and Brain 6, 9–17. Horridge, G.A. (2015a) How bees distinguish a pattern of two colours from its mirror image. PLoS One 10(1), e0116224. DOI: doi.org/10.1371/journal.pone.0116224 Horridge, G.A. (2015b) How bees distinguish colors. Eye and Brain 7(1), 17–34. Horridge, G.A. (2015c) How bees distinguish patterns by green and blue modulation. Eye and Brain 7, 83–107. Horridge, G.A. (2016) Parallel inputs to memory in bee colour vision. Acta Biologica Hungarica 67, 1–26. Kriston, I. (1973) Die Bewertung von Duft- und Farbsignalnen als Orientierungshilfen an der Futterquelle durch Apis mellfera L. Journal of Comparative Physiology A 84, 77–94. Laughlin, S.B. and Hardie, R.C. (1978) Common strategies for light adaptation in the peripheral visual systems of fly and dragonfly. Journal of Comparative Physiology A 128, 319–340. Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222. Riveros, A.J. and Gronenberg, W. (2012) Decision making and associative color learning in harnessed bumble bees (Bombus impatiens). Animal Cognition 15, 1183–1193. Squire, J. (ed.) (1918) The Collected Poems of James Elroy Flecker. Martin Secker, Norwich, UK. von Frisch, K. (1914) Der Farbensinn und Formensinn der Bienen. Zoologische Jahrbücher. Abteilung für ­allgemeine Zoologie und Physiologie der Tiere 35, 1–188.

Chapter 6 Feature Detectors, Cues, Resolution, Preferences and Coincidences

To build trust in science, scientists have to be rigorous and honest in what they do, and be sceptical, particularly about their own ideas and hypotheses. (Nurse, 2015)

Between 1990 and 2006, experimental testing of trained bees in Canberra revealed how bee vision worked by coincidences of feature detector responses deep in the insect brain, reaching and reading the contents of memory. The aim was to list the nuts and bolts of bee vision. Fortunately, most of the work was done with a pattern on each target, and the patterns were identical except for features that were investigated. Several cues in black and white patterns were identified, and it was discovered that bees had no general reassembly of patterns. It was a revolution in the study of insect vision. Moreover, it showed that shapes, similarity of scenes, categories of objects, or classifications of shapes were meaningless for bees. Combinations of feature detectors were ­detected to form simple cues that were summaries of measurements and positions, not images, things or panoramas. Generalization between different patterns and recognition of abstract properties were explained by collections of nothing but a small variety of simple cues. In this respect, bees illustrate what we can expect for all insects. The most conspicuous features for them were 106

vertical edges of shadows, and blue content of patches of colour or white. These advances greatly improved the chances of identifying the neuronal circuits with the cues they carry. More recently, in the period 2012–2014, it was realized that bees have no receptors for white or black, and do not discriminate yellow or green as colours. In the foraging and feeding behaviour, they detected only blue colour content, and blue and green contrast at edges. UV receptors were not essential for discrimination of coloured patterns or recognition of place in the foraging behaviour. Bees detect nothing in black and white patterns except blue areas and edges of green contrast. In earlier work, I was unaware of that. Therefore, problems with a coloured area on a background versus a plain background (Fig. 3.4), or with two black spots of differing sizes (Figs 3.10, 3.11) were not solved until the recent rapid discovery of colour mechanisms. In 1996, in an early analysis of cues, I  rediscovered the finding of Hertz (1937) that bees detected thin bars (< 5° wide) by their edges but thick bars by the position and area of background that was obscured. It was essential to use targets with equal areas of black, which implied equal areas of white, and therefore equal blue content, so that the bees had to find other less preferred cues. As a result, strong green modulation

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)



Feature Detectors, Cues, Resolution

excited by black/white edges blocked modulation in the blue channel, so almost all my black/white patterns were detected via green channel modulation. The following account is therefore mainly of the way that the green receptor channel detects combinations of edges by green modulation and by average orientation, which is the least preferred cue. Before 1988, in almost all training experiments, bees made their choice with the natural external panorama around them. Target size was measured as length, not angle. Bees learned cues at an unknown range from a fixed target while they prepared to land on or near the reward. In this situation, the ­moment of decision was unknown. The bees learned in a very few trials but they failed to respond when tested with unfamiliar experimental patterns, because ­ they had learned a combination of cues for that situation that were fixed on the retina (retinotopic). If the changes in a test were too great, they simply went away, so it was impossible to discover exactly what they had learned. Despite the uncertainties, many examples of the surprising ability of the bee to learn almost any pattern and later recognize it, or distinguish between two patterns, were documented before 1988. Performance during training, however, does not reveal mechanism. For that, it is essential to repeat a great many tests of trained bees. Features are defined as outside the bee; feature detectors and their responses lie superficially inside the optic lobe. A cue is a sum of responses of similar feature detectors in a local region or over the whole eye, irrespective of pattern. Cues are located and measured, but locations of the original feature detector responses are lost. The field sizes over which detectors are summed are large, filling at least half the eye. Each kind of cue can have several possible states: for example, modulation is based on a measure of power in voltage oscillations in neurons as the eye scans, and therefore can change in frequency, amplitude and temporal pattern. Edge orientation can be at different angles to the vertical, and blue colour can differ in blue content and position. Each kind of cue has its own story. Coincidences

107

between cues, and the angles between, are also learned. They make possible the recognition of a signpost on a route, or a place where food was previously obtained. To reveal the feature detectors and cues, we had already a convenient Y-choice apparatus, in which flying bees were isolated in a controlled learning situation where they chose between two targets at a known distance (Fig. 6.1). The apparatus had a top of transparent polycarbonate to exclude UV light. Once inside, to learn positions of cues on the targets, they use a frame of reference provided by the vertical edges of the displays. In my early experiments, like almost all earlier researchers, I trained bees on various targets and looked at the relative learning performance. Certainly, some patterns produced a higher score than others. However, it was unsatisfactory to infer the cue intuitively. You may think you know what the bees learned, but testing them reveals surprising new insights. There is no point in training bees if you fail to test what they learned. For example, bees discriminated well when trained on a diagonal black bar on a white background, versus the same bar centred in the same place on the other target but sloping the other way (Fig. 6.2A). However, the trained bees had learned the task only at the expected place, and could not

29

cm

Connecting slots

Pattern on target Reward hole

Feeder Leg

Reward box

Transparent baffles Choice chamber Bees fly in here

No reward

Fig. 6.1.  Apparatus for training and testing bees. Bees fly in through the hole at the front, then have a choice between two simultaneously visible displays. They pass one of the baffles though the central hole or via the slot at the top, which also act as exit routes.

108

Chapter 6

Train and test, fixed (A)

(D) 100%

77.6% ± 3.2%, n = 300 (B)

Test 100%

Move down, 51.5%, n = 200 (E)

70.0% ± 3.5%, n = 200 (C)

Test 100%

53.0%, n = 200 They do not recognize rewarded target.

Test 100%

Test 100%

Add spot, 48.0%, n = 200 (F)

Test 100%

Destroy orientation, 52.0%, n = 200

Fig. 6.2.  With these patterns, the bees learned orientation but only in the training position. They learn nothing about black or shape. Percentage values indicate the percentage of bees visiting reward holes. (A) Training patterns. (B) They distinguish the edges alone. (C) They do not distinguish these two targets because the cue was identical. (D) Memory was retinotopic. (E) Addition of a large black spot after learning prevented discrimination because modulation and a tangential cue were added. (F) Edge orientation was cancelled by the steps.

distinguish between the same two bars moved to the bottom of the target (Fig. 6.2D). This was an illustration of retinotopic learning and memory. Tests also showed that they recognized the orientation of edges without the black (Fig. 6.2B), but could not recognize the rewarded black bar (Fig. 6.2C). They had learned the least preferred cue, orientation of the edges only, not the black area or its shape. Note, at that time there were no tests of which target was learned. In a training situation sometimes known as ‘absolute training’, the unrewarded target was blank, so the most obvious cue was always the difference in amount of blue in the white backgrounds, and the position of displaced blue (Fig. 6.3A, C and F). Of course, the bees learned well, but only the position of omitted blue (Fig. 6.3D, F), not orientation (Fig. 6.3C). They failed when the bar was tested versus a stepped bar or a rotated bar

(Fig. 6.3D, E). They had not learned orientation, no need. About 1998, I changed the training procedure. Every 5 min the patterns were shuffled by changing every possible cue except one (Fig. 6.4A, B). When the bees still learned the task, it was clear that they had learned the cue that was consistent during the training. The trained bees were then tested with an unfamiliar target that displayed the same cue. Of course, they responded positively. When trained with a bar at 135° on one ­target versus one at 45° on the other target (Fig. 6.4), the bars had to be in corresponding positions on the two displays, as if the bee learns exactly where to look as well as detecting the cue. Therefore, the comparison of one display with the other in the learning process was retinotopic, even when the memory was established at every place



Feature Detectors, Cues, Resolution

109

Train and test, fixed (A)

Test 100%

(D)

100%

84.3%, n = 300 (B)

51.5%, n = 200 Test 100%

(E)

100%

64.0%, n = 200

50.5%, n = 200 (F)

Test 100%

(C)

Test 100%

81.5%, n = 200

70.0%, n = 200

Fig. 6.3.  In this display, they learned only the position where black covered the background. (A) Training patterns. (B) Response was reduced by a vertical shift in position. (C) Rotation at the same place made no difference. (D, E) Orientation was not recognized. (F) Position of black was recognized in the vertical direction.

4 coincident locations (A)

1+

1–

2+

2–

4 non-coincident locations (B)

1–

1+

2+

2–

55°

3+

3–

3+

100%

4+

61.7%, n = 237

3– 100%

4–

4+

4–

50.7%, n = 219

Fig. 6.4.  Shuffling position of the pattern on the display was done by change of the patterns, 1, 2, 3, 4, every 5 min. (A) Learning was possible with the bars in corresponding positions on the two displays. (B) When the bar was in non-corresponding positions on the two displays, the bees failed, so the process of learning is retinotopic, as well as the memory.

110

Chapter 6

where learning occurred. The success of the randomization technique suggested many new experiments. After training bees, I tested them with the training target versus an unfamiliar target that displayed the same cues that I had found, and when they could not tell the difference, irrespective of the pattern or place on the target, I knew that all the cues had been identified. A cue is therefore learned and later detected irrespective of the pattern in which it is embedded.

The Modulation Cue Long ago, Hertz (1933) showed that bees distinguish between many patterns by the difference in total length of edge (Fig. 2.5), a finding later confirmed by all who examined this point, but it was never determined over what area of the display the total edge was measured. Both green and blue receptor channels detect modulation, which is the signal in the receptor caused by scanning an edge while in flight. Modulation is much more than contrast. It is the product of the emission spectrum of sunlight, multiplied by the length of edge resolved to the ­vertical, multiplied by the contrast at each part of the

edge, multiplied by the absorption spectrum of the receptor. All of that can be calculated for each receptor type when coloured patterns are prepared. Tables of data are in Chapters 5 and 7, this volume. Resolution tests, going back to Hecht and Wolf (1929), and repeated several times since, gave an absolute limit near 2° for the resolution of a regular grating, irrespective of the bar orientation. Measured with gratings (Fig. 6.5B), the modulation detectors had the resolution of a single ommatidium with a symmetrical inhibitory surround, which is smaller than that measured with an electrode in the receptor cell (Fig. 4.12) because there is lateral inhibition in the lamina. The modulation of a pattern is a preferred and common cue that is located relative to other landmarks and measured quantitatively. The modulation cue appears to include the total number of simultaneously excited modulation detectors in a local region of the eye. Bees do not distinguish shapes or identify black/white objects by their shape; they measure and locate concentrations of modulation, use them as landmarks, and relate them to other landmarks, as in the measure of width (Fig. 3.11B). Training on equally spaced grating patterns is very informative. With a grating at

Train no green contrast train, rotating 180° every 5 min (A)

100% 55°

47.7%, n = 300 + (B)

Train as before 100%

_

74.5%, n = 200 (C)

Test 100%

72.0%, n = 200

Fig. 6.5.  (A) With no green contrast, oblique gratings were not distinguished. (B, C) When trained on similar V/H gratings, modulation was distinguished irrespective of pattern.



Feature Detectors, Cues, Resolution

45° and one at 135°, with no green contrast (Fig. 6.5A), bees failed to discriminate even if the period was very large, because there was no difference in blue content, no modulation difference, and the orientation detectors had inputs only via the green receptors. Although green contrast was essential for discrimination of orientation, bees discriminated between a horizontal and a similar vertical grating in colour with no green contrast (Fig. 6.5B), because they made use of the difference in blue modulation. This was the first demonstration that modulation was discriminated via the blue channel when the green channel detected no difference. When trained on horizontal versus vertical gratings with no green contrast (Fig. 6.5B), bees learned to prefer the greater blue modulation irrespective of pattern (Fig. 6.5C). When trained on black/white gratings, the strong signal in the green channel inhibits a response in the blue channel, and test gratings of blue contrast alone are not discriminated. When trained on fixed gratings of period less than 10°, bees do not learn an eidetic image. Of course, trained or not, with its small brain, a bee cannot possibly see or remember the separate bars in the grating, but detects, locates and measures modulation and orientation cues without reconstructing the pattern. When trained on orientation of gratings and tested on bars or lines, cues of modulation or orientation may be detected, but when trained on bars with no modulation difference, and tested on gratings, they respond only if they had learned orientation.

Orientation discrimination with shuffled bars In 1990, van Hateren and others continually shuffled the bar positions and bar width while training bees to discriminate horizontal bars versus vertical bars. They used the Y-choice apparatus (Fig. 6.1) without baffles. The shuffle was to eliminate positions or width of edges or black as cues. It was claimed that retinotopic memories were refuted, but they were simply distributed to every location where they had been during the training. The authors also assumed that the orientation

111

of edges was learned, although Srinivasan and Lehrer (1988) had already found that in this situation the bees learned the difference in the modulation. Later, the difference in modulation was eliminated with oblique gratings at 45° and 135° to the vertical, so the only cue was orientation (Fig. 6.5A). Maximum size of orientation detectors, and training orientation only When trained with the positions of edges moved every 5 min, and the same length of edge on each target (Fig. 6.6A, D), modulation was useless as a cue, so bees learned the less preferred orientation. When tested with lines of squares 4° apart, they failed (Fig. 6.6B), but with squares 3° apart, they discriminated orientation as if the squares were not separated (Fig. 6.6C). Therefore, the feature detectors for orientation have a maximum length of 4°. Their short length accounts for the poor resolution of differences in edge orientations. Also, when trained with the positions of edges moved every 5 min (Fig. 6.6D), they failed to recognized the pattern they were trained on (Fig. 6.6E), but recognized orientation where it had occurred on the training target (Fig. 6.6F, G). Modulation as a cue irrespective of pattern Bees were trained to discriminate between coarse and fine vertical bars with bar positions shuffled to ensure that they were not used as retinotopic cues (Fig. 6.7A). The bees retained their discrimination of the modulation difference when tested with coarse versus fine sector and spiral patterns (Fig. 6.7B, C). The patterns were irrelevant because bees were interested only in the most preferred cue, modulation. Peripheral position of black was the preferred cue in large targets In Wehner’s (1968, 1971) work on discrimination of the rotation of a square cross, the patterns were huge, subtending 130° at the

112

(A)

Chapter 6

Train, rotating 1, 2, at top 1 1 100 %

Train, rotating 1, 2, at top (D)

1

100%

1

55° 2

2

2

76.7%, n = 300 (B)

100 % Test

2 74.0%, n = 200

(E)

100 % Test

4°gaps 54.0%, n = 200 (C)

100 % Test

(F)

52.5%, n = 200, because orientation same Train on (E) and test on (D) 52.5%, n = 200, orientation and pattern not learned. Train on (E) and test on (G)

3°gaps

(G) 66.5%, n = 200

Test 100%

65.0%, n = 200, They had learned the modulation difference in E. Fig. 6.6.  The orientation detector. (A, D) Training patterns with position of black bars shuffled to avoid retinotopic memories. (B) Trained bees failed to detect orientation with 4° gaps between squares. (C) They discriminated well with 3° gaps between squares. (E) With equal length of oriented edge, the training pattern was not recognized. (F) Bees trained on (E), and tested with (D), failed to recognize the training pattern or the orientation cue. (G) However, bees trained on (E) preferred a test pattern with less modulation which had been the preferred cue. (Modified from Horridge, 2003a.)

eye (Fig. 6.8A). The response of a trained bee to rotation of this cross was far from the sine-squared function of the angle expected in a small display. Unlike the orientation cue, which requires a difference of at least 45° to be distinguished, the smallest detectable rotation of a large cross was only about 4°, which was also the smallest detectable shift in any direction. Edges of very large bars could be cut into steps with no effect (Fig. 6.8B), so edge orientation was ruled out as a cue. Instead, the bees had learned the positions of separate areas of black in separate regions of the eye. Later, I showed that the bees detected the positions of black at the extreme periphery of the patterns (Fig. 6.8E–H). Because black is not a stimulus to the eye, bees used the position of the  missing blue obscured in the white

­ackground. Whether these large targets b ­require both eyes is not known. Mutual cancellation of orientation of orthogonal edges In 1994, Srinivasan et al. discovered that bees were not able to learn to discriminate between a smaller square cross (subtending 45° at the eye) and the same cross, rotated by 45° (Fig. 6.8D), even after pretraining on bars (Fig. 6.8C). They suggested that the detectors of the orientation of edges have an angular orientation sensitivity curve that is 90° wide at the 50% level of sensitivity and a huge field that covered the whole eye. The response of the detector to the rotation of an edge or thin bar would be a sine-squared



Feature Detectors, Cues, Resolution

113

Train and test on gratings with alternating phase

100% period 12.7° 6 cm

(A)

4 cm

8.4°

50°

88.3% ± 2.1%, n = 230 (B) Test on six sectors versus eight sectors, also alternating

100%

60.1% ± 4.9%, n = 98, p < 0.05 (C) Test on spirals period 10.5° versus 4.2° 100%

79.5% ± 3.8%, n = 113

Fig. 6.7.  (A) Training patterns with phase randomized to avoid retinotopic memories. (B, C) Greater modulation was avoided irrespective of pattern. (From Horridge, 1997a.)

Train, bees landing on reward hole

Train in Y-choice apparatus

(A)

(C) 100%

100% 130°

(B)

> 90% Test 100%

45° > 90%

(D)

Test 45°

130° Fail

> 90% (E)

Train and test

(G)

100% 19%

81% (F)

130°

Test 100%

19% 81% The central region is not required.

64° 100% 49%

51%

The central region alone is ineffective. (H)

64°

Train on (E), test on (G)

130°

Test 100%

70% 30% The tips alone provide sufficient cue.

Fig. 6.8.  Contrasting results with patterns subtending 130° and 45° at the eye. (A, C) A large cross subtending 130°, and a single bar subtending 45° at the eye were easily discriminated from the same rotated by 45°. (B) Bees trained to the large cross were undisturbed by the loss of the edge orientation. (D) A cross subtending 45° was not discriminated from the same rotated by 45°, even after pretraining on the bars in (C). (E–H) When trained to detect rotation of a very large cross, bees learned only the positions at the periphery. (Data sources: (A, B) Wehner, 1967; (C, D) Srinivasan et al., 1994; (E–H) Horridge, 1996b.)

114

Chapter 6

function that rises from zero to a saturation of 100% as the angle changes from 0° to 90°. The response to the rotation of an edge or bar at right angles would a cos-squared function. With two equal bars in the form of a square cross, the total response was constant so that rotation of the cross has no effect. It was assumed that the primary orientation detectors had a very large field, but when measured, they turned out to be only three facets long (see Fig. 6.23C–J discussed later in the chapter), so a different explanation is required. Two or more different orientations in a local region cannot be detected simultaneously (Figs 6.9, 6.10C) and therefore interaction of orientations destroys pattern and texture, but modulation at edges and residual average orientation persists. Neurons sensitive to orientation with fields over a whole eye were found in the deep optic lobe of the bee (Fig. 6.12). At that time, we all believed that responses of the edge detectors are strung together to make continuous lengths, as observed by humans, but later it became clear that the primary orientation detectors of the bee were only 3° long (Fig. 6.11) and they were summed in local areas in such a way that total modulation was measured and learned but different orientations cancelled. Later, many examples were found where different orientations were mutually cancelled when located within 30° of each other on the target (Fig. 6.9). This cancellation destroys any possibility that patterns were reassembled in the brain of the bee. Neither large nor small ­ displays supported the idea that whole shapes or patterns were recognized by relative orientations of edges, although this was the almost universal belief. Making a fixed edge fuzzy, even extremely fuzzy, has little effect on the orientation cue. A gradient from black to white spreading over 20° was still detected as an edge with an orientation, because edges are detected by scanning by the bee’s eye. The minimum size of orientation detectors When training on gratings have a period of less than about 10°, it is quite unnecessary

to shuffle the positions of the bars. Bees were trained to discriminate between two orthogonal arrays of equal oblique bars (Fig.  6.11A) and then tested on arrays of shorter bars of controlled length. The minimum length for orientation detection at the threshold is about 3° (Fig. 6.11B). This result implies that the feature detector of orientation consists of a group of seven adjoining ommatidia (Fig. 6.23C, D, E).

Angular sensitivity of the orientation ­detectors The feature detectors for orientation of an edge are only about 3° long, which limits the orientation sensitivity curve to a angle of about 90° at 50% sensitivity. This gives a sharp resolution but an approximate orientation of details of edges. They are never strung together to make a line in a way (as in human vision) that would improve the discrimination of the angle of orientation. Feature detectors for all orientations are summed over the whole eye into neurons that can be recorded in the deep optic lobe. These neurons have the same angular sensitivity function (Fig. 6.12) as the elementary orientation ­detectors (Fig. 6.23K, L). There are several types with different orientation axes, and their fields cover the whole eye. In 1994, Srinivasan and others suggested that neurons like these respond to average orientation, so that equal lengths of orthogonal edges anywhere in the field cancel. On the other hand, in discrimination experiments, edges at different orientations within the same local region of the eye are summed in a peculiar way. Equal lengths of edges at right angles cancel the orientation but retain the measure of total modulation. Bees do not detect separate orientations of edges in textures but they can distinguish local differences in modulation. The function of orientation detector neurons with a field over the whole eye (Fig. 6.12) is therefore in doubt, and also they were not clearly distinguished from motion detector neurons with large fields. Averaging or summation of orientation of edges at different angles



Feature Detectors, Cues, Resolution

Misunderstandings with orientation detectors

Train, rotating by 180° every 5 min (A) 100%

55°

91.5% ± 3.9%, n = 200 (B)

Test 100% 20°

78.5% ± 3.0%, n = 300 (C)

Test 100%

40°

75.0% ± 3.7%, n = 200 (D)

Test 100% 90°

60.5% ± 3.9%, n = 240 (E)

Test 100%

115

90°

52.5%, n = 200 Fig. 6.9.  (A–D) Bees were trained on horizontal versus vertical bars and then the targets were made more equal in stages. The preferred cue was the difference in modulation, which was progressively made equal. (E) With oblique bars, the only cue was the orientation, which was more easily cancelled.

means that bees cannot detect the structure of a pattern or construct a shape from the orientations of edges. The discrimination of the orientation cue was little affected when black in a pattern was exchanged for white (Fig. 6.13C, G), showing that the detectors of edge orientation are bilaterally symmetrical (Fig. 6.23C, D, E).

In all the earlier work where the criterion of success was the landing on the correct target, orientation was scarcely mentioned as a cue. The reported parameters were the position, area and length of edge, probably because the patterns subtended a large angle at the choice point. According to Schnetter (1972), ‘Wehner succeeded in isolating the position of the black areas as the relevant parameter for direction’ (Fig. 6.8A, B). That was true, but by accident because he chose to work with bars 130° long and 53° wide. This result illustrates the belief that the bees saw the patterns, and shows how the unsuspected details of the training method influenced the progress of the research. I drew a different conclusion. With very large targets, they used the coincidences of features in well-separated parts, as if they were viewing several landmarks. Wehner (1971) trained with a single oblique black bar, at 45° to the vertical, subtending 130° by 20°, versus a blank target. The trained bees discriminated correctly ­between the training bar and a similar bar at right angles. They discriminated rather poorly when both test bars were black on a white background, but very well between the training black bar on white at 45° and a white bar on black, also at 45° (Fig. 6.13E–H). We now know that the bees measured the peripheral distribution of blue in the white of the background as well as edge orientation. Jander et al. (1970), working in the same institute at Frankfurt, had trained wasps to learn a smaller oblique bar, also presented vertically. The trained animals distinguished an oblique black bar versus a vertical black bar of equal area, or even white bars on a black background (Fig. 6.13A–D). Jander interpreted these results in terms of orientated local edge detectors, but this was an ­unorthodox idea. There were three reasons why Wehner’s idea of an eidetic image could not possibly be correct: (i) it required a different set of constants for each range; (ii) it could not apply to the discrimination of a grating; and (iii) Wehner (1968) had already shown that the response to rotation of a thin bar was independent of bar

116

Chapter 6

Train, rotating by 180° every 5 min (A) 100% 55°

90.0%, n = 200 (B)

Test, fixed 100%

Fig. 6.10  (A) Training with shuffled bars versus a blank target. Note that blue content difference and orientation were possible cues. (B) Test with equal locations and lengths of edge to demonstrate an orientation cue. (C) Trained bees failed to distinguish the training target from a target with more total edge and area of black because the targets were equal in average orientation.

74.5%, n = 200 (C)

Test, fixed 100%

47.5%, n = 200

10 impulses/sec

Train on 10° bars (A)

100%

55°

10

Reverse phase, as the edges flicker Axis

69.0%, n = 300 Test on 3° bars (B)

Number of impulses/sec 10

100%

57.2%, n = 260 Fig. 6.11.  Technique to measure the length of the orientation detectors. (A) Oriented training patterns like these did not require shuffling. (B) Test with short bars showed a minimum length of 3° or detection of orientation. (From Horridge, 2003b.)

width. So, even as Wehner’s theory of detection of positions of black and overlaps emerged, it was attacked with a rival theory of orientation detectors. There was a conflict and Jander went off to America, but actually all the data is now explained

Motion, as the bee scans across edges Fig. 6.12.  The responses of a neuron recorded in the deep optic lobe of a worker honeybee to various orientations of a grating stimulus. The stimulus was a small target containing stationary phase flicker (smaller response) or local oscillatory motion (larger response). The neuron field covered the whole of one eye. The responses (solid lines) were plotted in angular coordinates. The angular width of the field that gave more than 50% response was about 90° (dashed lines).

by the bees’ ability to locate the position of black in Fig. 6.13(H), by the missing blue in the background, and by feature detectors for



Feature Detectors, Cues, Resolution

117

Train with fixed single bar versus white background (E)

(A) Wasp Jander (B)

Bee Wehner

bar 4 cm by 16 cm 100% Test

(F)

100% Test

97%, n = 647 (G)

85%, n = 204 (D)

100% Test

54%, n = 50

100% Test

77%, n = 229 (C)

bar 130° by 20°

100% Test 68%, n = 152

(H)

100% Test 96%, n = 232

Fig. 6.13.  Early contrasting results when the criterion was the insect landing on a single target. (A–D) The wasp (Vespa germanica). The cue could have been modulation or edge orientation. (E–H) The honeybee. (A and E) Training targets. (B and F) The training targets were recognized. (C and G) Reversal of contrast had little effect because the cue was at edges. (H) This result from very large patterns shows that the cue was location at the periphery (as in Fig. 6.8). (Data from: (A–D) Jander et al. (1970); (E–H) Wehner (1971.)

edge orientation, with a difference in preferences between wasp and bee. Wehner (1981) had a very confused account of pattern perception and nothing of substance about orientation detectors, but made no reference or mention of Jander’s orientation detectors. To explain his test result (Fig. 6.13G), Wehner (1971) switched to ‘analysers’ of generalized orientation and found that ‘an angle discrimination is more easily possible when the relevant analyser … is switched on by previous training.’ Bees ‘abstract from some special stimulus properties by generalizing the sensory input according to special cues, for example the direction of visual stimuli’ and ‘the information about the direction of the visual cue had been transferred to a new pattern configuration never seen by the bees during the training situation’. The word ‘­direction’ is ambiguous and could mean the direction of black as seen from the point of choice or the angle of orientation on a vertical surface as an abstract cue irrespective of the actual pattern. For some reason, Wehner

gave up research on pattern perception by bees. In Canberra after 1990, we looked at the detection of bar orientation at various distances from the target to compare global and local discrimination, confirming the above interpretation (Zhang and Horridge, 1992). In 1992, we knew nothing about the size of orientation cues, radial and circular features, or separate blue and green modulation. Inferences were premature because we also misunderstood the order of preferences for cues and jumped to conclusions with insufficient experimental support. Throughout this period, we thought that bees saw black and white displays, continuous line, edges lacking green contrast, and that bees had trichromatic detection of colour. In 1995, it was shown that detection of orientation as well as motion required green contrast. Exciting results with closing and opening parallax in motion perception also misled us into thinking that bees would discriminate the orientation of the boundary when a p ­ atterned edge moved against a patterned background.

118

Chapter 6

We wasted half a decade from 1990 to 1995 looking at performance instead of mechanisms. We also looked at detection of orientation in illusory edges (Horridge et al., 1992), but most of these 1990–1995 experiments were useless. The patterns were fixed and the bees could have learned the cues of position, modulation, area and edge length, and maybe orientation. The random textures we used for camouflage, with pixels 4 mm2, were not ­resolved at 27 cm range. Unfortunately, the scores were too high because the test patterns were presented for 10 min on each side, during which some bees made two visits and could learn which side to go. Also, it was not realized that radial and tangential edges were distinguished as separate cues, and it was thought that orientation on the left side of the target was discriminated separately from that on the right side, so many test patterns were inappropriate. All these errors of the day were uncritically accepted by refereed journals at the time, and still confuse the literature.

Adaptation of Flowers to Bees In his earliest experiments, von Frisch (1914) found that flower-like radial or concentric patterns of the same size (Fig. 1.4) were easily distinguished, but that a triangle, ­ square, disc and ellipse were not, and a chequered pattern of squares was not distinguished from one of triangles. He postulated that bee vision was adapted to the patterns of the flowers, forgetting that many insects have similar visual systems, but have no interest in flowers. Later Hertz (1933) and Zerrahn (1933) each found three classes of patterns, stars, circles and irregular blobs (Fig. 2.5) that are discriminated from each other even with similar length of edge, location or orientation. In 1970, Free found that bees preferred radial symmetry before bilateral symmetry, and then irregular patterns, also that bees landed at the edge of a plain target or on a spot at the centre of a circle. Flowering plants evolved millions of years after the insect visual system, and the retina, lamina and medulla of the optic lobes are remarkably similar in all insects

and even in Crustacea (Strausfeld, 2012), so the evolution of flower colours and shapes must have been influenced by existing ­insect avoidance of asymmetry (Chapter 7, this volume) and detection of contrast. Probably bees already detected radial spokes and circles or discs: for example, to learn to avoid spider webs or to enter a dark hole. The plants also solved the problem of growing symmetrical coloured flowers from apical meristem where cell division was based on a spiral.

Preferences for flower-like patterns All the early workers had made use of radially symmetrical shapes for training bees. On a horizontal surface, they could be approached from any direction. They showed salience (i.e. the bees found them easily on a flat featureless white table and would learn them readily). What the bees learned was a different matter. When two training patterns were equally symmetrical, the bees would not learn to discriminate symmetry, because both patterns showed it. In Canberra, Lehrer et al. (1995) found that they avoided circles (Fig. 6.14B). It was also noticed that when flying, bees land upon bilaterally symmetrical flowers, they line up with the direction of the axis (Jones and Buchmann, 1974). Much later it was shown that they score the flowers for degrees of symmetry. In 1994, following an old demonstration by Gertrud Zerrahn (1933) we built an apparatus with 12 compartments and trained marked bees to come to neutral patterns. After training on a plain black disc as the attractant pattern, bees preferred radial patterns to random or circular ones. When we randomized the modulation in the attractant pattern by using one of six regularly changed 50:50 black and white chequerboards, bees preferred patterns of low spatial frequency. They preferred any kind of radial pattern, and a vertical axis of bilateral symmetry, but avoided concentric circles (Fig. 6.14). No preferences were found with other patterns, although many were tried. Although we began by looking for preferences towards flower patterns, we



Feature Detectors, Cues, Resolution

119

(A) 40°

40° (B)

32%

27%

23%

18%

33%

23%

22%

22%

30%

27%

23%

20%

(C)

(D)

discovered a new system of radial or tangential feature detectors and cues.

Discrimination of sectors and circles Bees were easily trained with a pattern of black and white sectors or spokes, versus one with spirals or concentric circles, with no average orientation in either target. To control against differences in area of black, length of edge or location of black areas, the patterns were both randomized by substituting a different target every 10 min (Fig. 6.15A), so that nothing remained constant except the kind of pattern. Bees then discriminated unfamiliar patterns with radial or tangential contours, such as a cross versus a hollow square (Fig. 6.15D), and parts of circles or patterns of spokes. However, there was no proof here that these bees learned to recognize a circle, because they may have used their innate preferences for radials in the training and tests (Fig. 6.15). Although the referees failed to notice this inadequacy, we later validated our conclusion by rewarding randomized circles in

Fig. 6.14.  Spontaneous preferences of naïve bees given a choice of four patterns. (A) Training patterns that were regularly moved in position. (B) With 12 black bars. (C) With eight black bars. (D) Spontaneous preference for bilateral symmetry.

the training, although it took longer. Then we discovered that the bees preferred to ignore the rewarded circles anyway, and learned to avoid the randomized sectors. These results, and those of Hertz (1933) with patterns presented on a flat surface, led naturally to the proposal that bees have global filters for radial spokes, and other global filters for concentric circles, and that these filters respond to any pattern which coincides with their preference. How easy it was to imagine global filters for favoured shapes!

Directing recognition with a coloured spot There are detectors of radial features, and other detectors of concentric circles or tangents that are assembled from the same elementary feature responses as edges elsewhere. Indeed, the location of an added blue spot can influence bees to treat a bar as a radius or as a tangent, irrespective of location of the pair on the target. To demonstrate this, the positive target had a blue spot at the side of the bar; the negative ­target

120

(A)

Chapter 6

(C)

Train on random sectors and circles

100%

100%

45° 45°

85% 15%

(B)

(D)

74% 26% 100%

100%

45° 45°

80% 20%

61% 39% The radial and tangential cues are recognized.

Fig. 6.15.  (A) Shuffled training patterns, to spread the retinotopic memory. (B, C) Circles and sectors were both distinguished from a chequerboard pattern. (D) The discrimination was repeated with unfamiliar patterns, at a lower score.

had a similar blue spot at the end of the bar, otherwise the bar was the same in each target (Fig. 6.16A, B). Both targets were ­rotated by 90° in the same direction every 10 min between corresponding positions, 1, 2, 3 and 4, so that the location and orientation of the bar and spot were useless as cues, but the relative positions displayed the two states of the cue. To the bees, the blue spot was a centre or hub. After training, the bees were able to discriminate a different pattern of tangents versus radials without the spot (Fig. 6.16C). Whether the single black bar was accepted as a radius or a tangent depended on the relative position of the blue spot, which the bees accepted as the centre or hub of a circle or spokes.

Strategies for defining the number of cues Successive efforts progressively defined the limited number of cues in the repertoire of the bees. First, in 1995, the radial, circular or spiral patterns were rotated at intervals during training to remove cues derived from orientations, leaving radial and/or tangential cues. In 1998, these cues were found to be colour blind. In 1999, radial and tangential

patterns with radial symmetry based on three or six spokes were easily discriminated, but those based on four, five or seven spokes or sectors were not, suggesting that feature detectors for orientation lay along the lines of hexagonal facets on the eye. Next, the four bars of the square cross were rearranged to make many fixed patterns of two pairs of orthogonal bars, all the same size, area of black, length of edge, and average position of black on the target (Fig. 6.17). These patterns of four bars could not be discriminated from the same pattern rotated by 45° unless one of them had a vertical axis of bilateral symmetry. The bees could discriminate a rotation of the axis of bilateral ­symmetry by 90°, even if the test patterns were different from the training patterns. The patterns with two pairs of orthogonal bars could be roughly divided into three groups. The first group differed greatly in their content of radial, tangential or bilateral symmetry cues (Fig. 6.17A–D), and were easily distinguished from each other by the bees. The second group were quite different from each other but were not distinguished because they displayed similar cues (Fig. 6.17F–H). The third group also were different from each other but were distinguished with difficulty because radial



Feature Detectors, Cues, Resolution

121

Train and test with 1, 2, 3 and 4 in sequence (A)

(B) 1+

3+

1–

3–

2–

4–

55°

2+

4+

100%

67.7%, n = 239 (C)

Test 100%

50°

66.9%, n = 142 Fig. 6.16.  (A) Rewarded training pattern; a tangential feature detector versus a radial one irrespective of position. The same rewarded pattern was rotated to four positions, keeping the bar position constant relative to the blue spot. (B) Unrewarded patterns with radial position. The four radial patterns had the blue spot in the same position as the corresponding rewarded pattern. (C) The trained bees recognized the radial/tangential cue in quite different patterns. (From Fig. 4.7 in Horridge, 1997c.)

and tangential cues on the same target cancelled out or were the same on the two targets (Fig. 6.17I, J). Trained bees accepted any pattern displaying the expected cues but not if there was an unexpected cue. The bees measured the cues quantitatively. The actual pattern was irrelevant. These patterns and experiments illustrate the value of J.S. Mill’s (1843) rules of logic in the search for causative factors. Many pairs of patterns differed from each other but were not discriminated (Fig. 6.17J). When the bees had learned the orientation cue, they failed in tests to distinguish the rewarded training pattern from other patterns with the same average orientation and total length of edge. The bees looked for the orientation cue and found it equally in both patterns, irrespective of differences in layout. Authors who still supported the eidetic image in the 1990s were blind to numerous examples where patterns were different but not distinguished. The third strategy turned to the recognition of position. A fixed pattern composed

of two colours was discriminated from the same with the colours reversed in position (von Frisch, 1914; Gould, 1986). Bees easily detected a vertical shift of the centre of an isolated area of black or colour relative to the reward hole. They also discriminated the exchange of two colours in the left/right direction if there was green contrast.

Experiments with two bars at right angles, tangential versus radial Bees preferred the radial/tangential cues when available even when they were rotated around the centre of the target, but failed to respond to the general form of the global pattern. Bees were trained with two bars at right angles so the orientation cues were cancelled. When the two bars were alternated between radial and tangential (Fig. 6.18A) the bees could not learn to discriminate the obvious pattern difference because there was no consistent cue. They

122

Chapter 6

Patterns that differ in radial/tangential cues

O--

Train and test

Train and test 100%

R

(A)

R

(F)

T

100%

64% (B)

36%

50% 50%

Train and test

R

36%

48% 52%

Train and test 100%

R

62%

Train and test

(H)

100%

T

38%

53% 47%

Train and test

(D)

Train and test

(I)

100%

100% R

T

61%

T

39% (J) V

V

H

R

T

50%

Train and test 100%

(E)

T

T 64%

(C)

100%

T

45°

O+

T

Train and test

(G)

T

100% T

H

R

O+

T

O--

Patterns that are confused by bees

50%

Train and test T

T

T

100% T

T

T 69%

31%

49%

51%

Fig. 6.17.  (A–E) Discriminations and (F–J) failures to discriminate between patterns of four equal bars that were either oriented (H, V), radial (R), or tangential (T). This was an early search for further unidentified cues, and demonstrates that bees detect cues, not patterns. Success required different cues on the two targets; failure followed lack of a difference in cues, irrespective of pattern. ‘O’ and ‘O+’ (in A and B) are orthogonal orientations that cancel each other.

could not learn to discriminate because radial and tangential cues alternated, and there was no consistent cue. The bees could not detect the consistent global pattern of an arrowhead pointing to the left on one target and to the right on the other (Horridge, 1997b). Next, fixed bars in corresponding positions formed an arrowhead pointing upwards

on the rewarded target and downward on the other (Fig. 6.18B). The bar orientation was cancelled. When the patterns were moved down, the radial versus tangential cues reversed although the global arrowhead pattern was unchanged. From these experiments, it was clear that bees did not detect a global pattern as simple as the combination of two bars.



Feature Detectors, Cues, Resolution

(A)

(B)

Train with alternating radial and tangential bars, coincident presentation

Train with fixed locations

55° 1+

123

100%

1– 74.7%, n = 229

2+

Test with different locations

100% 2–

100%

49.9%, n = 86 35.3%, n = 93 Fig. 6.18.  (A) In this situation, bees learned nothing because the radial cue alternated with the tangential one, and the bees learned neither. (B) In this situation, bees learned the radial versus tangential cue although they were moved around the reward hole, and ignored the global V pattern and orientation of individual bars.

The twilight of pattern vision of the bee Miriam Lehrer was convinced that bees distinguished flower shapes among foliage irrespective of colour, and also that bees learned an association with the rewarded target. With Campan in 2006, she set to work to test that idea. Bees easily distinguished a disc from a triangle when both were painted with a random black/white pattern of 50% black (Fig. 6.19A). With central area removed (Fig. 6.19B), and with centres and some edge removed (Fig. 6.19C), the discrimination was unaffected. Moreover, the trained bees distinguished the training disc from the same with the centre removed (Fig. 6.19D). This experiment epitomizes all that was undesirable in thought and design. The strategy of the experiment was decided by the previous convictions of the researchers, not what the bee required; only two pairs of patterns were used; training was long to reach 80% correct; the patterns were initially randomly arranged, but for the bees they were constant patterns; the areas of the patterns, and their positions in the vertical plane were not the same; difference in modulation or black area may have been the cue; the observer had no idea whether the bees detected the random pattern, the shape,

the corners, a particular bit of white, or what; and there were no tests that would separate the cues. The experiment tells me that outline shape was not a cue. The final test here showed that the cue was not the circular shape by itself (Fig. 6.19D), but one conclusion was that the bees distinguished shape and generalized it to similar shapes, despite being partially obscured. The publication was an effort to demonstrate supposed performance; it was far from an analysis of mechanism. Analysis of cues in radial patterns Bees easily spot the difference between radial and tangential edges, revealing a mechanism in radial coordinates, and orientation is less preferred. Radial and tangential edge detectors are colour blind, with inputs only from the green receptors (Horridge, 1999b). Radial edges at 60° to each other were preferred. In tests, the trained bees accepted other patterns displaying the radial/tangential cues on either side of the target irrespective of the pattern (Horridge 1996a). Bees discriminate between a fixed ring of up to six spots or sectors, versus ring and

124

Chapter 6

Train with strationary patterns criterion, landing on target (A)

100%

79.7%, n = 1424 (B) 100% Test

79.5%, n = 200 (C)

100% Test

86.1%, n = 200 (D)

100% Test

77.3%, n = 200 The clue was not circular shape. Fig. 6.19.  Bees were tested for generalization of shape discrimination. Bees trained to distinguish between a disc and a triangle were correct when tested with partial patterns. (Data from Lehrer and Campan, 2006.)

the same ring rotated by half the angle between the spots (Horridge, 2000a, Fig. 6). The cue was the difference in modulation caused by a horizontal scan in flight. Unlike the situation with spokes, the number of axes of symmetry with spots was less important than location or size of the individual spots. When targets were shuffled by rotation during the learning process so that positions were shuffled, the cues must be presented as radial or tangential edges, not as spots or areas of black. Radially symmetrical patterns of spokes have salience for bees but they lose it when green contrast is removed, which shows the reliance on

edges. In conclusion, there were two separate pathways for a bee to detect a radial pattern, by the orientations of edges of spokes, or by the modulation and positions of spots or sectors. Feature detectors for radial and tangential edges By using the displays with no green contrast, in 1999 it was found that feature detectors for circles and spokes were restricted to the green receptor channel. When analysed by cutting the edges into short lengths or into square steps that were separately resolved, the feature detectors for radial and tangential edges were the same as those proposed for the orientation cue, 3° long, and therefore spanning three ommatidia in a row (Fig. 6.6B, C; Fig. 6.23K, L). A new cue, the position of the centre or hub Bees could also learn the position of the centre of a ring or of concentric circles when trained versus a blank or a neutral pattern (Fig. 6.20). There was no evidence for the idea that the bees detected the layout of whole rings or long curved edges, and much evidence against it. When trained with a pattern symmetrical about a point versus a blank target and then tested with two patterns at different heights (Fig. 6.20F), bees discriminated the expected position of the hub by as little as 5°, in some cases with unfamiliar test patterns (Fig. 6.21). A pattern of spokes or rings also stabilized the vision of the bees in the horizontal plane so that the position of a hub could then be learned (Fig. 6.20F and Fig. 6.21). Bees discriminated half of a pattern of radial spokes or concentric circles from the other half, cut either vertically or horizontally, and irrespective of scale, showing that radial and circular patterns were not detected by complete templates or global ­ ­filters. Bees detected edges as radial or circular by the coincidence of numerous local edge detectors converging to a hub from anywhere in the array, irrespective of the



Feature Detectors, Cues, Resolution

125

Train with fixed targets (D)

(A) 100%

100%

55°

95.0%, n = 200, after 3 h training (B)

Test

54.5%, n = 200 (E) Test 100%

100%

68.5%, n = 200

83.0%, n = 200 (C)

Test 100%

58.0%, n = 200

Test

(F) Test 100%

Move down 15°, 65.5%, n = 200 Move down 10°, 62.5%, n = 180 Move down 5°, 55.0%, n = 200

Fig. 6.20.  With radial spokes versus a blank target, bees detected the cues of ‘difference of blue content in white background’, ‘radial edges’ and ‘position of the hub’. (A) Training pattern. (B) With black on both sides, the score was reduced, so black contributed. (C) With the bars rearranged at the same orientations, they detected trivial difference. (D) They detected spokes on both targets, and no difference. (E) In contrast to (D), trained bees easily distinguish spokes at 60° or 30° to vertical from those at 45°, which do not excite the radial detectors. (F) Trained bees notice a difference in the position of the hub down to 5°. (From Horridge, 2006a.)

actual pattern (Fig. 6.22). The binding that defined the cue as radial or tangential was therefore not hard-wired but depended on the coincidences of responses of edge detectors anywhere in the local region of the eye. This is a diffuse mechanism with no template and no memory of the actual pattern. It also explains the mutual cancellation of radial and tangential edges or of orthogonal orientations. Because it depends on coincidences, such a system would give the impression of having taken a snapshot. There was no global template that detected the positions, angles, or numbers of spokes, a circle of a given size, or a right angle. Instead, there was a diffuse administration that would identify an incomplete or partially

obscured pattern and find the position of its hub (Fig. 6.22). In conclusion, radial and circular patterns were identified by the regional coincidences and convergence of edge orientation detectors, and the position of the hub was an additional cue. Every possible use was made of coincidences of the positions and vectors of the edge detectors in a local region, but there was no evidence of reassembly of an unfamiliar pattern. Defining the feature detectors graphically A model that suggests what cues look like is helpful in relating them to neurons and

126

Chapter 6

The position of the hub train on fixed patterns (C) (A)

Test 100%

100%

55°

85.5%, n = 240 (B)

58.5%, n = 200 (D)

Test

Test 100%

20° down, 49.0%; 10° down, 64.0% 20° up, 61.0%; 10° up, 75.5%, 4° up 55.0%

57.0%, n = 200

Fig. 6.21.  Bees learn the position of the hub of concentric circles. (A) Training patterns. (B) The trained bees notice a difference in the position of the hub, by as little as 5°. (C, D) With quite different patterns they still prefer the hub at the centre. (From Horridge, 2006a.)

neuron responses in the optic lobes. Several observations are crucial for the visualization of the feature detectors. First, they are in a green or a blue receptor channel at the level of the retina. Position and amount of blue by itself is a strong cue, and many whole patterns can be distinguished by position and measure of green modulation alone, but a cue can also be a coincidence of the position of blue relative to green or blue contrast. Secondly, modulation and orientation detectors are very small. Resolution of modulation is better than expected from the width of the receptor visual field or the interommatidial angle as if it is improved by lateral inhibition between neighbours at the level of the lamina. The minimum and the maximum size of the orientation detector is about 3°, as shown by use of short lengths, or short steps of an edge, or bridging of gaps in dashed lines (Fig. 6.11). The feature detector for modulation is in both green and blue channels but orientation is green only. It is high time that the optic lobe neurons be tested with appropriate stimuli to distinguish them in terms of known cues (Figs 6.23 and 6.24).

In every example of bee vision of blackwhite patterns analysed in the same way, the bees learned a similar coincidence of relatively simple inputs with no suggestion of vision of the pattern or colour. Black necessarily generated the maximum green and blue contrast at its boundary with any other colour, and the bees were quick to use this strong signal. Movement of a black area to expose background acted like blue because black is merely the absence of blue to the bees. Patterns on a background of white appeared to display no differences in contrast at their outer borders, because the contrasts were saturated, or because the apparatus was painted white inside so that the bees were adapted to white. The results revealed the kind of system in the recognition of pattern, coloured or not. In every experiment, bees detected the same few discrete inputs. They preferred to measure the monochromatic blue input and edges of buff, yellow or black to quantify the green contrast. The preferred retinotopic positions of the green contrast and blue well separated on the target, optimum for triangulation.



Feature Detectors, Cues, Resolution

127

(A)

Parallel orientations sum (B)

(C)

Radial orientations sum and point to a hub Tangents of a circle sum and point to a hub

In each context, bees learn one state of each kind of cue Bees were trained, first on one pair of patterns for 10 min then on a second pair for 10 min, and then back to the first, ­repeated for 2 h (Fig. 6.25). The pairs of patterns were selected to test the hypothesis that there is only one channel for each kind of cue, although each cue can be displayed in different states. It was observed that two different pairs of patterns that display different cues can be learned when trained alternately (Fig. 6.25A–D), but different pairs of patterns that display states of the same cue during alternate training interfere with each other and cannot be learned (Fig. 6.25E–H). The cues tested were: (i) orientation of bars or sectors; (ii) radial and tangential edges based on a symmetry of three or six; and (iii) position of a black spot.

Fig. 6.22.  Visualization of a ­distributed system of feature detectors to generate three separate cues. Note that all feature detectors are separate, but they interact in differet ways. (A) Array of three orientations of feature detectors in each column of the medulla. (B) Radials and tangentials. (C) Orientation feature detectors along an edge, but no continuity of the line.

Patterns in Fig. 6.25 all look different to human vision, but the cues are different in each pair in (A)–(C). In (E)–(H), the patterns are different states of the same cues, although these pairs look quite different. The positions of the spots in Fig. 6.25(H) are all different but the bees are unable to learn that they are positive in two positions and negative in the other two, when they are seen in the same context. Cues in the positive and negative targets continually change their state, making learning impossible. On the other hand, bees easily learn to discriminate when one alternated pair of patterns displays two states of one cue and the other pair displays two states of quite a different cue (Fig. 6.25A–D). The inference is that there is only one final processing channel for each kind of cue (Fig. 6.24) in each eye. The bees failed when they were faced with two simultaneous tasks involving the same cue in different states although all pairs

128

Chapter 6

Green sensitive receptors

Three edge orientation detectors

1°

(A)

–1 (D)

(C) –1 –1

(B) –1 –1

+6

+2 –1

+1 (E)

–1

Vertical (H)

Hub

Hub

orientation cue (L)

(K)

Orientation cancelled (M)

(N) Height of blue

Hub

One tangential cue

+1

Orientation cancelled width measured

(J)

Oblique

–1 +2

(G) Hub

orientation cue

3° v1

–1

–1

–1

Feature detectors

(F)

+1 +2 –1

–1

Modulation detector

–1

–1

+1

+1

–1

+1

–1

One radial cue

Left/right polarity cues

(O) (P)

Asymmetry along horizontal

Modulation gradients detected in scanning

Fig. 6.23.  Defining graphically the organizational structure of feature detectors and cues. (A–E) Groups of seven adjacent receptor channels sum to form feature detectors for (B) modulation, (C–E) orientation. (F–J) Groups of orientation detectors sum to form cues for net orientation in local regions. (K, L) Groups of orientation detectors sum to form cues for tangential and radial edges. (M) Polarity cues. (N) Blue height cue. (O, P) Asymmetry of modulation.



Feature Detectors, Cues, Resolution

Feature

Weak

Strong

Modulation

Area

Blue

Polarity

Blue

Width and height

129

Action of blue receptor input

Action of green receptor input

Active when green is not available

First preferred input from all targets

Tonic blue input summed, measured and located

Green contrast is a boundary; not sensitive to areas or colours

Retinotopic

Measures and locates angular width to 90°

Locates and measures angular width from tonic blue to green modulation, and polarity in horizontal direction

Angle Radial

Not sensitive

Active

Bilateral symmetry

Not sensitive

Active

Orientation

Not sensitive

Active

Tangential

Not sensitive

Active

Fig. 6.24.  Visualizing bee cues in human terms. Cues are represented by diagrams showing what human vision identifies. Bees can learn and recognize these cues.

were readily learned individually. They do not learn one of the pairs and ignore the other, which would improve their chance of a reward. Instead, they start to learn afresh each time the patterns are changed to a different cue. Clearly, each cue channel cannot learn two tasks at the same place. Of course, in a different context with different landmarks the bees may use the same cues for a different choice. Detection with and without memory of it In the experiments in which the bees discriminated gratings, they detected the modulation

and orientation differences but not the grating pattern. Similarly, they detected spokes or parts of circles as radial or tangential, and located the hubs but did not recognize the patterns. This helps to clarify the point that humans discriminate patterns but bees learn only cues. The essential first step for bees is the simultaneous detection of all edges in a local region but there is no memory at this stage. The number of excited edge detectors gives the total modulation, and their orientations are averaged (Figs 6.9, 6.10C, 6.23F–L). Spokes and circles are identified and their centres located (Figs 6.20, 6.21). Other cues,

130

Chapter 6

Easy combinations 1

Difficult combinations 2

(A) 50° 78.1%, n = 400

2

55.5%, n = 300

54.7%, n = 300

54.0%, n = 400

52.7%, n = 400

53.1%, n = 300

50.6%, n = 300

47.4%, n = 300

48.0%, n = 300

(E)

74.0%, n = 400

(B)

1

50°

(F)

66.2%, n = 350

63.7%, n = 350

(C)

(G)

67.1%, n = 350

71.6%, n = 250

(D)

(H)

66.9%, n = 350

70.3%, n = 350

Fig. 6.25.  (A–D) Evidence that bees can learn two different cues when trained alternately on two pairs of patterns (1 and 2). (E–H) Evidence that in similar training, they cannot learn two states of the same cue. The cues were orientation, radial, tangential, and position of black. Training was for 2 h.

such as the position of the centre of black, colour, and the area or size, are detected by other pathways in parallel and learned according to a scale of preferences. All cues are separately remembered if rewarded, but processing of the image stops at a few coincidences, far short of reassembly of the image.

Preferences for the Cues In Hertz’s earliest efforts, preferences were observed when untrained bees selected one pattern from a variety of them, and they learned most readily the patterns that they spontaneously preferred. Preferences were also revealed in differences in the rate of learning and the maximum score achieved. In most discriminations over the past decades the bees had no control over the choice of the images, and it was often not clear whether they learned the rewarded pattern, the difference between the rewarded and the unrewarded patterns, to avoid the unrewarded target, or to avoid unexpected cues

that were not in the training patterns at all. More to the point, it was not clear how many cues were learned in parallel and in what order. This situation was due to the lack of sufficient tests of trained bees. The preferences were not explored because most of the cues had not been described. Because the preferences were ignored, it was not possible to understand how patterns displaying several cues were discriminated in the early experiments. To reveal the preferences, bees were trained to discriminate between a rewarded target with one pattern on its left side and a different one on the right, versus a neutral pattern (Fig. 6.26). This arrangement gave the bees a choice of what to learn. Tests showed that they learned two or three cues with equal preference; or they preferred to avoid the unrewarded target. By working with different combinations of patterns, it was possible to put the cues into an order of preference, so that when two or more were displayed, the bees learned one first and more strongly. The order of preference during learning was:



Feature Detectors, Cues, Resolution

Train on fixed patterns

Train on fixed patterns (E)

(A) 100%

100%

55°

85.0%, n = 200 (B)

(C)

Test 100%

58.0%, n = 200 Test 100%

Test 100%

75.0%, n = 200

55°

83.0%, n = 200 (F)

Test 100%

75.5%, n = 200 (G)

Test 100%

72.5%, n = 200

74.0%, n = 200 (D)

131

(H)

Test 100%

59.0%, n = 200

Fig. 6.26.  Relative preferences for cues during the learning process. Bees were trained to discriminate between a rewarded target displaying two cues versus a neutral pattern of squares. (A) Train with modulation and a large spot. (B) Test against the mirror image, with a poor score showing that the rewarded target was not preferred at all. (C) Test, showing that either bees had learned the spot or the unrewarded target. (D) Final test showing that they had learned the unrewarded target. (E) Train with parallel bars and a large spot. (F) Test against the mirror image, with a high score showing that they had learned the rewarded target. (G) This test showed they used the spot as a cue. (H) Compared to the spot in G, bars were not preferred. (From Horridge, 2007.)

(i) area of black or blue; (ii) position of centre of area; (iii) total modulation; (iv) radial edges; (v)  positions of hubs; (vi) average local orientation; and (vii) tangential edges. Single black spots and strongly modulated patterns were easily learned. Large black spots were preferred over small ones. Tangential edges (circles) and horizontal edges were weak cues. S ­ ymmetry in a pattern of bars was preferred above the edge orientations that generated the symmetry. When a weak and a strong cue were presented together, the weak one was scarcely noticed. Various patterns such as a spot, a square cross, a group of small squares and many complex patterns provided as cues only the

area of black, modulation and position of the centre of black. The bees could learn not to avoid circular patterns. When two colours were presented side by side on the rewarded target or on separate targets, the bees could not learn both at the same time. They learned blue or white in preference to other colours, even if they had to learn to avoid the blue. When presented with a pattern on each target they ignored the cues that were displayed on both targets. They avoided any unexpected change on a target. In general, they learned to avoid the negative target unless strongly preferred cues were displayed on the rewarded one.

132

Chapter 6

Salience versus retinotopic cues In previous work, when a broad black bar or spot was moved more than 10° after training, the bees did not recognize it in its new place, showing that the bees had learned it in only one place. This was retinotopic all the way to memory. In other cases, even though the pattern was in a fixed place, the preferred cue was recognized after displacement, and therefore called ­salient. The sensitivity to displacement was related to the field size of the cue. Cues of orientation and the position of a black area were more retinotopic and therefore detected in smaller fields, and were not salient. Modulation was detected over larger displacements and therefore in larger fields, so that it was more salient. Small fields implied some failures to detect; large fields implied some failures to localize, but improved the salience. Each cue had its own compromise field size. The most salient cue was a small black spot.

Resolution by feature detectors In general, the sizes of the fields of the different cues have not yet been measured. The bees estimated the cues quantitatively and learned absolute size, relative size or angular size of a spot, depending on how they were trained. The minimum detected difference in modulation between two textured patterns was about 30% (Horridge, 1997a). When trained to the orientation of an edge, and then offered a choice between two others, they preferred the orientation that was nearest to the rewarded one, with minimum detectable difference about 30°. The widths of the angular sensitivity curves for the orientation of an edge (Srinivasan et al., 1994) or of the axis of bilateral symmetry (Horridge, 1996b) were about 90° at the 50% level, because the orientation detectors are short. The minimum detected difference in positions of an area of black or the centre of a radial hub in the vertical direction was about 10° (Horridge, 2006b).

Long training improved the precision of discrimination. It is probable that the field sizes, minimum difference and resolution limit depended also on the pattern. Bees are particularly effective in discriminating the transposition of two coloured panels in the vertical direction on the target, even with no green contrast, and easily discriminate differences of 6° (Gould, 1986; Horridge, 1999a, b, 2000b).

Avoidance of a cue not in the training It was accidentally discovered that in a test, discrimination is lost when a black spot is added at the same place on both targets. Later it was noticed that cues that were not displayed in the training cause the bees to fail in the tests. They act as if they are in the wrong place because they detect a cue that should not be there. As a result, a small black spot is characterized mainly by cues that are absent. The decision process makes full use of the available options provided by the repertoire of feature detectors. On arrival at a new place, the bees behave as if they have a list of cues marked as familiar or not, so increasing the variety of labels and useful landmarks. There is less effect when a cue is duplicated or when an expected cue is omitted from a test, because they had learned several. All these conclusions were logical inferences from a variety of tests.

Why do they learn more than one cue? The very high scores obtained when training on a single pattern versus a white target are partly due to the fact that the bees quickly distinguish the location and total amount of blue on each target. It is a very easy cue. However, subsequent tests showed in every case that the bees learned two more preferred cues in parallel, notably the measure and location of modulation and the position of the centre. High training scores are misleading at the start of the training when the bees learn the blue content. The high



Feature Detectors, Cues, Resolution

scores show that the bees have an easy choice, not that they see the patterns. Learning several cues with their positions in order of preference has two advantages. First, in the natural situation it is less likely that they mistake the place. Secondly, the more cues they learn, the more likely they are to find the reward although some part of the scene changed.

It was cues all the way By the 1960s, feature detectors were the best explanation of image processing in vertebrates, and were used as inputs in computer vision. By 1994, measures of modulation and orientation were observed in bee vision.

133

However, a different type of experiment was required to show that there was no additional memory of the pattern as distinct from the cues. Instead of shuffling the patterns to eliminate unwanted cues, the patterns now had to be fixed stationary on the targets to give the bees an opportunity to form an eidetic image (or retinotopic memory) of them. Then the trained bees were given a large variety of tests to see what they had learned. The most conclusive test is to show that when bees are trained to recognize a pattern, they fail to distinguish it from quite a different pattern that displays the same cues. No matter what was the pattern in the training, it was always possible to show that bees learned in order of preference from the small repertoire of cues that they inherited. It was cues all the way.

References Free, J.B. (1970) Effect of flower shapes and nectar guides on the behaviour of foraging honeybees. Behaviour 37, 269–285. Gould, J.L. (1986) Pattern learning by honeybees. Animal Behaviour 34, 991–997. Hecht, S. and Wolf, E. (1929) The visual acuity of the honeybee. Journal of General Physiology 12, 727–760. Hertz, M. (1933) Über figurale Intensität und Qualitäten in der optische Wahrnehmung der Biene. Biologische Zentralblatte 53, 10–40. Hertz, M. (1937) Beitrag zum Farbensinn und Formensehen der Bienen. Zeitschrift für vergleichende Physiologie 24, 413–421. Horridge, G.A. (1996a) Vision of the honeybee Apis mellifera for patterns with two pairs of equal orthogonal bars. Journal of Insect Physiology 42, 131–138. Horridge, G.A. (1996b) Pattern vision of the honeybee (Apis mellifera); the significance of the angle subtended by the target. Journal of Insect Physiology 42, 693–703. Horridge, G.A. (1997a) Pattern discrimination by the honeybee, disruption as a cue. Journal of Comparative Physiology A 181, 267–277. Horridge, G.A. (1997b) Vision of the honeybee Apis mellifera for patterns with one pair of equal orthogonal bars. Journal of Insect Physiology 43, 741–748. Horridge, G.A. (1997c) Spatial and non-spatial coding of patterns by the honey-bee. In: Srinivasan, M.V. and Venkatesh, S. (eds) From Living Eyes to Seeing Machines. Oxford University Press, Oxford, pp. 52–79. Horridge, G.A. (1999a) Two-dimensional pattern discrimination by the honeybee. Physiological Entomology 24, 1–17. Horridge, G.A. (1999b) Pattern vision by the honeybee (Apis mellifera) is colour blind for radial/tangential cues. Journal of Insect Physiology A 184, 413–422. Horridge, G.A. (2000a) Visual discrimination of radial cues by the honeybee (Apis mellifera). Journal of Insect Physiology 46, 629–645. Horridge, G.A. (2000b) Seven experiments on pattern vision of the honeybee, with a model. Vision Research 40, 2589–2603. Horridge, G.A. (2003a) The visual system of the honeybee (Apis mellifera), the maximum length of the orientation detector. Journal of Insect Physiology 49, 621–628. Horridge, G.A. (2003b) Visual resolution of the orientation cue by the honeybee (Apis mellifera). Journal of Insect Physiology 49, 1145–1152.

134

Chapter 6

Horridge, G.A. (2006a) Visual discrimination of spokes, sectors, and circles by the honeybee (Apis mellifera). Journal of Insect Physiology 52, 984–1003. Horridge, G.A. (2006b) Some labels that are recognized on landmarks by the honeybee (Apis mellifera). Journal of Insect Physiology 52, 1254–1271. Horridge, G.A. (2007) The preferences of the honeybee (Apis mellifera) for different visual cues during the learning process. Journal of Insect Physiology 53, 877–889. Horridge, G.A., Zhang, S.W. and O’Carroll, D. (1992) Insect perception of illusory contours. Philosophical Transactions of the Royal Society of London B 337, 59–64. Jander, R., Fabritius, M. and Fabritius, M. (1970) Die Bedeutung von Gliederung und Kantenrichtung für die visuelle Formunterscheidung der Wespe Dolichovespula saxonica am Flugloch. Zeitschrift für Tierpsychologie 27, 881–893. Jones, C.E. and Buchmann, S.L. (1974) Ultraviolet floral patterns as functional orientation cues in hymenopterous pollination systems. Animal Behaviour 22, 481–485. Lehrer, M. and Campan, R. (2006) Generalization of convex shapes by bees: what are shapes made of? Journal of Experimental Biology 208, 3233–3247. Lehrer, M., Horridge, G.A., Zhang, S.W. and Gadagkar, R. (1995) Shape vision in bees, innate preference for flower-like patterns. Philosophical Transactions of the Royal Society of London B 347, 123–137. Mill, J.S. (1843) A System of Logic, Ratiocinative and Inductive, Being a Connected View of the Principal ­Evidence, and the Methods of Scientific Investigation. Two volumes. John W. Parker, London. Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222. Schnetter, B. (1972) Experiments on pattern discrimination in honeybees. In: Wehner, R. (ed.) Information Processing in the Visual Systems of Arthropods. Springer, Berlin, pp. 195–200. Srinivasan, M.V. and Lehrer, M. (1988) Spatial acuity of honeybee vision, and its spectral properties. Journal of Comparative Physiology A 162, 159–172. Srinivasan, M.V., Zhang, S.W. and Witney, K. (1994) Visual discrimination of pattern orientation by honeybees. Philosophical Transactions of the Royal Society of London B 343, 199–210. Strausfeld, N.J. (2012) Arthropod Brains: Evolution, Functional Elegance and Historical Significance. Belknap Press, Cambridge, Massachusetts, 848 pp. van Hateren, J.H., Srinivasan, M.V. and Wait, P.B. (1990) Pattern recognition in bees, orientation discrimination. Journal of Comparative Physiology A 167, 649–654. von Frisch, K. (1914) Der Farbensinn und Formensinn der Bienen. Zoologische Jahrbücher. Abteilung für allgemeine Zoologie und Physiologie der Tiere 35, 1–188. Wehner, R. (1967) Pattern recognition in bees. Nature, London 215, 1244–1248. Wehner, R. (1968) Die Bedeutung der Streifenbreite für die optische Winkelmessung der Biene (Apis mellifica). Zeitschrift für vergleichende Physiologie 58, 322–343. Wehner, R. (1971) The generalization of directional visual stimuli in the honeybee, Apis mellifera. Journal of Insect Physiology 17, 1579–1591. Wehner, R. (1981) Spatial vision in arthropods. In: Autrum, H. (ed.) Handbook of Sensory Physiology, Volume VII/ Part 6C: Vision in Invertebrates. Springer, Berlin, pp. 287–616. Zerrahn, G. (1933) Formdressur und Formunterscheidung bei der Honigbiene. Zeitschrift für vergleichende Physiologie 20, 117–150. Zhang, S.W. and Horridge, G. (1992) Pattern recognition in bees, size of regions in spatial layout. Philosophical Transactions of the Royal Society of London B 337, 65–71.

Chapter 7 Symmetry and Asymmetry: Signposts in Route Finding

Scientists . . . need to explain what is being done and why, ensuring that conflicts of interest are revealed, and that it is clear what knowledge is secure and what is not. [my emphasis] (Nurse, 2015)

The biological significance of flower symmetry appears obvious. When offered several flowers of one species, bees appear to prefer the more symmetrical. This is interesting because apart from radial patterns like spokes, which they prefer, and tangential edges like circles or spirals, which they avoid, bees have no generalized detectors for shape. Naïve bees rarely show a spontaneous preference for other shapes or patterns. Trained bees look for simple cues that distinguish or identify each particular display (Chapter 6, this volume). However, no one considered whether bees were attracted to symmetry or whether they avoided asymmetry.

Bilateral Symmetry In the mid-1990s, it was discovered that untrained honeybees spontaneously preferred a vertical axis of bilateral symmetry in arbitrary unfamiliar targets irrespective of pattern, and almost any bilaterally symmetrical pattern of short black bars on a

white background could be discriminated from similar but unsymmetrical patterns (Horridge, 1996). A chevron pattern with its axis of symmetry vertical was discriminated from itself rotated by 180° or by 90° (Fig. 7.1). The chevron pattern is curious in having no net orientation because its two radial bars cancel the orientation of its two tangential bars, and more importantly, it gives a similar modulation pattern when scanned horizontally in either direction. I concluded ‘The result with the chevron suggests that bees have a filter beyond those for circles or radial patterns, or for average orientation, and that it is related to bilateral symmetry, which is already known to have a broad biological significance for bumblebees’. This was sent to press in March 1995 (Horridge, 1996). The widespread occurrence of symmetry in animals and plants, and the fast learning of it by bees, led me to postulate (quite incorrectly) innate mechanisms for detecting symmetry, but there was a difficulty: I could not model a bee visual filter that would detect the axis of symmetry irrespective of pattern. Detection of the axis of bilateral symmetry irrespective of pattern was demonstrated with other patterns of two pairs of orthogonal bars in the following way. To train the bees to ignore the actual pattern, they were trained with seven different bilaterally symmetrical patterns of four bars

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)

135

136

Chapter 7

Train and test

Train and test (D)

(A)

80%

50%

20%

Separate orientations on two sides (B)

Train and test

Insufficient difference in angle (E)

Train and test

45° 63%

37%

50% No vertical axis

Separate orientations on two sides (C)

Train and test

62% 38% Vertical versus horizontal axis

(F)

Train and test

51%

49%

No vertical axis

Fig. 7.1.  Results of training bees to discriminate between a chevron pattern and the same at a different orientation. One axis had to be vertical, and the angle more than 45°. Because no tests were done to see what the bees detected, the intuitive interpretations of 2009 have no value. Percentage values indicate the percentage of bees visiting reward holes. (From Horridge, 2009b.)

taken in succession, versus the same pattern rotated through 90° (only three patterns are shown in Fig. 7.2). The positive (rewarded) targets had the axis laid on its side, and the negative targets all had a vertical axis of ­bilateral symmetry (Fig. 7.2A–C). Training was therefore against the innate preference. In a few hours of training, bees learned to discriminate a vertical axis of symmetry irrespective of pattern. On successive days, with continued training, they were tested on the same seven patterns rotated through 180° (only three patterns are shown in Fig.7.2D–F). Although all the patterns in the tests were unfamiliar, the trained bees still picked out the asymmetrical one of each pair, and with an improved performance because they had additional training. These trained bees also preferred a vertical axis of bilateral symmetry in completely d ­ifferent unfamiliar patterns with different numbers of bars. The bees certainly detected something that these patterns displayed in common.

Filters that detect symmetry as a general property are unknown, and no simple cues used by bees to detect symmetry irrespective of pattern could be found. Furthermore, it became clear that this is the only example where bees seem to detect an abstract property of a pattern. In all other examples where bees trained to prefer a pattern have been carefully tested, one or two simple cues explained the discriminations. For the whole of the last century, learning of colour or pattern by bees was usually considered to be Freudian conditioning. Naïve bees apparently associated the display with the accompanying reward. Without further thought, in all the earlier work on discrimination of symmetry, bees had been trained with the symmetrical pattern rewarded, and there were no tests whether they had learned the unrewarded pattern. Working with simple black spots on white backgrounds, however, Ronacher (1992) noticed an influence of the unrewarded pattern.



Symmetry and Asymmetry

Train all pairs together and test separately (A)

100%

137

Also test each pair separately rotated by 180° (D)

100%

50° 67% (B)

33%

100%

68%

79% (E)

100%

32% 78%

(C)

100%

64%

21%

36%

(F)

22%

100%

73%

27%

Fig. 7.2.  Detection of a vertical axis of bilateral symmetry irrespective of pattern. (A–C) The bees were trained on seven bilaterally symmetrical patterns, taken successively in pairs for 10 min on each side in the Y-choice maze. (A, B, C) Only three of the patterns are shown here. The pattern with the vertical axis in each pair was not rewarded, so the training was against the preference. Training scores from each pattern were collected separately. (D–F) The trained bees were tested on the same seven pairs of patterns rotated through 180°. The same three are shown. The tests were done in random order between periods of continued training, which slowly improved the performance. (After Horridge, 1996.)

No one had seriously considered the possibility that bees might learn to avoid the unrewarded display. Without evidence, some assumed that, when distinguishing between two targets, bees learned the difference. Recently, it has become apparent that when there is no obvious preferred cue that is detected from a distance, such as location and amount of blue or strong green contrast, bees learning in the Y-choice maze (Fig. 6.1) have to learn by trial and error. They learn only when they approach first the unrewarded target and are frustrated because they search there but find nothing. They learn to look and distinguish or else they waste time and effort by first going the wrong way. Those that choose to go first to the rewarded display take sugar solution, having had no reason to learn anything. As they leave with a reward of sugar, they may turn back and look at the display and the landmarks, but the displays change places every 5 or 10 min, so in the Y-choice maze they must learn the display, not the place. In agreement with these

observations, I have found consistently over years of research that bees learn first the unrewarded target and avoid it, knowing nothing about the rewarded side when tested (Horridge, 2015a; 2016, Fig. 2b, c; Fig. 10b, c; Fig. 11b–m). Further examples are given later in the chapter (see Figs 7.11, 7.12). The mechanism is not such a puzzle as it is in man, because there was no evidence that bees detected symmetry. Recently, it was shown that when bees were trained to detect a symmetrical distribution of simple vertical lines from an asymmetrical one, they recognized the asymmetry in the pattern of green contrast about a vertical axis, not symmetry (Horridge, 2015b, Fig. 14). It is now time to ask how bees detect asymmetry, starting with mirror images that differ only in polarity.

An Algorithm for Recognition of Polarity Bees readily distinguish the difference in polarity between two mirror images, and

138

Chapter 7

recognize polarity and its direction in single images. Any coincidence of two different feature detector responses in a horizontal scan must produce a cue with polarity. In the training pattern (Fig. 7.3), there are two colours, buff and blue, that display no difference to the green receptors but a strong contrast to blue receptors. To provide green contrast, in a fixed position, there is a vertical black line that acts as a landmark; as revealed in the tests, the cue on each display was the horizontal spatial relation between a patch of blue colour and the landmark of green contrast, which was either to the right or to the left of the blue. The spatial relation in this coincidence of two cues acted as a directional signpost, directing incoming bees to right or left. The angle between these cues, together with measures of amount of blue and green contrast, further identified this coincidence. In discrimination of black/ white patterns, white has twice the blue content of blue (Table 7.1). Most asymmetrical displays have asymmetry of distribution of blue or white, even if only in the background

colour. Cues with similar polarity are abundant in the natural panorama, and must play a significant part in recognition of a route or goal.

The Landmark Must Be Fixed and the Blue Must Be Blue A group of bees were trained with just the algorithm versus its mirror image (Fig. 7.4A), then tested with the vertical bars in different positions, with the conclusion that even

Table 7.1.  Relative receptors’ excitations by the different papers on the same linear scale. Canson colours Blue 595 Buff 384 Dresden Yellow White copy paper

Blue receptor (%)

Green receptor (%)

54.2 25.7 13.1 100

40.0 41.7 78.1 100

Train wih no green contrast and a black bar (A)

(D)

100% 55°

80%, n = 200 (B)

100%

3° black

45%, n = 100 (E)

Test

62%, n = 100 (C)

100% Test

48.0%, n = 100

100% Test

100% Test

38%, n = 100 (F)

100% Test

81%, n = 100

Fig. 7.3.  The general algorithm for detection of polarity. (A) Training patterns with a mix of colours and a black landmark. (B) With blue alone, the score was reduced. (C, D) With buff alone, or with buff and a landmark, the trained bees failed. (E) With the polarity of the landmark reversed, the response was reversed. (F) The minimum stimulus to give a large response to the polarity. (After Horridge, 2015a.)



Symmetry and Asymmetry

Train (A)

139

Train (E)

100%

100%

55°

91.0%, n = 200 (B)

3° black,

100% Test

55°

6° blue (F)

100% Test

(G)

68.0%, n = 200 (D)

100% Test

51.5%, n = 200

100% Test

55%, n = 100

52.5%, n = 200 (C)

95%, n = 100

6° buff

100% Test

57%, n = 100 (H)

100% Test

54%, n = 100

Fig. 7.4.  Properties of the minimum algorithm for detection of polarity. (A) Training patterns. (B, C, D) Moving the landmark or position of blue spoils the score. (E) New training patterns. (F, G, H) Change of colour of blue or landmark, or removing the landmark, spoils the score.

a small shift of the landmark spoils recognition (Fig. 7.4B–D). Similarly, two targets with the same polarity are distinguished (Fig. 7.4E), but the blue must be blue and the landmark is essential (Fig. 7.4F–H). With asymmetrical mirror images, the cue was modulation of polarity There is now sufficient evidence to support the hypothesis that bees detect an asymmetrical distribution of green contrast. Appropriate training patterns were devised to test this hypothesis. I used yellow bars on black to display maximum green contrast and no total blue difference. The bees were trained to discriminate between a triangular pattern of yellow bars and its mirror image (Fig. 7.5A). There was little blue content, so the abundant green modulation provided the cue. The width and total green modulation was identical on the two targets, so the asymmetrical distribution

of green contrast was the only remaining cue. When tested with a symmetrical pattern with the same length of bars, recognition was little affected (Fig. 7.5B), showing that the unrewarded pattern had been learned. On the other hand, the bees failed to recognize the rewarded pattern when tested with the rewarded pattern versus the symmetrical bars (Fig. 7.5C). They recognized the difference in polarity of the patterns but had learned to recognize only the one that had provided no reward. With similar symmetry, colour and modulation, the cue was width The bees were trained to discriminate between a rectangle of vertical yellow bars versus the same bars arranged in a square block (Fig. 7.6A). The two patterns had equal total length of bar and area of yellow. They had their centres at the same height, and displayed no difference in blue content.

140

Chapter 7

Train on a triangle of bars versus its mirror image

Train on symmetrical yellow bars with equal green modulation and blue content

100% (A)

55°

(A)

100% 55°

78.0%, n = 200 after 3 h

73%, n = 200

Test: symmetrical pattern versus unrewarded one (B)

Test versus same width in a different pattern (B)

73.5%, n = 200 They avoided the unrewarded pattern.

70%, n = 200

Test: unrewarded one versus symmetrical pattern (C) 100%

52.0%, n = 200 They did not recognize the rewarded pattern.

Fig. 7.5.  Polarity of green modulation in the horizontal direction was learned on the unrewarded target. (A) Training patterns. (B, C) When tested versus a symmetrical pattern displaying similar green modulation, trained bees could distinguish the unrewarded pattern, but not the rewarded pattern they were trained on.

Both patterns were bilaterally symmetrical about the vertical midline, leaving little for a bee to distinguish them. When tested with the rewarded training pattern versus a couple of bars separated by the appropriate width, the trained bees retained the high score (Fig. 7.6B), but they could not distinguish between the two yellow bars and the unrewarded target (Fig. 7.6C). Therefore, they had learned to distinguish the widths, not the heights or number of bars or the shapes of the envelopes around the groups of bars. Bees learned the width of a symmetrical unrewarded pattern Bees easily discriminated between asymmetrical and symmetrical arrangements of yellow bars of the same total length (Fig. 7.7A) but the rewarded pattern could not be distinguished

100%

Test: two bars versus unrewarded pattern (C)

100%

49%, n = 200 With no memory of modulation or the original pattern, they failed. Fig. 7.6.  With no difference in symmetry or total modulation, the default cue was width. (A) A wide grating was easily distinguished from a tall grating of the same edge length and period. (B) Vertical bars of the same height and expected outer width provided a sufficient cue. (C) Trained bees failed to make a distinction when the widths were similar.

from its mirror image (Fig. 7.7B). However, the trained bees correctly distinguished the rewarded target from two yellow bars at the correct width (Fig. 7.7C), but failed to distinguish the two bars from the unrewarded pattern, because the cue was present on both targets. Therefore, they had learned only the unrewarded width. When the rewarded target was symmetrical and the unrewarded target was asymmetrical, it was usual to find some learning of both targets, as in Fig. 7.8(E) (see below), it was usual to find some learning of both targets. When the unrewarded target was symmetrical bees learned from both The bees were trained to discriminate between an asymmetrical pattern of yellow



Symmetry and Asymmetry

Train, triangle of bars versus rectangle of bars (A) 100% 55°

84.0%, n = 200 after 3 h Test: triangle of bars versus mirror image (B) 100%

63.5.0%, n = 200 Some recognition of the rewarded pattern Test: rewarded pattern versus two vertical bars (C) 100%

76.5%, n = 200 They preferred the asymmetrical pattern. Test: two vertical bars versus unrewarded pattern (D) 100%

47.5%, n = 200 They had not learned the symmetrical pattern.

Fig. 7.7.  Bees discriminated between an asymmetrical and a symmetrical pattern. (A) Training patterns. (B) Poor recognition of the rewarded pattern versus its mirror image. (C) Good recognition when two yellow bars at the correct width replaced the unrewarded pattern. (D) Failure to distinguish the two bars from the unrewarded pattern, because the bees had learned the width, which was present on both targets.

bars and a symmetrical arrangement of the same total length (Fig. 7.8A). When tested with the asymmetrical rewarded pattern versus its mirror image, with the same length of bars, recognition failed (Fig. 7.8B), so the ­unrewarded target was essential and the rewarded target was not recognized. When tested with a symmetrical pattern with the bars near the centre versus the unrewarded pattern, recognition was not ­

141

significant (Fig.  7.8C), showing confusion between two symmetrical patterns. When tested with two bars placed on the left versus two bars placed symmetrically at the original width, recognition returned (Fig. 7.8D). Further tests would be required to see which of several possible cues the bees had learned, possibly width, or retinotopic memory. In a second experiment, with the same training patterns, the symmetrical one was rewarded and the bees again learned quickly (Fig. 7.8E). In a test with the rewarded pattern versus the other with just three bars, the bees did not recognize the previously rewarded pattern (Fig. 7.8F), but with the unrewarded pattern versus its mirror image, they had no difficulty (Fig. 7.8G). This was the opposite result to the situation in Fig. 7.8(B), because in Fig. 7.8(G) they had learned to avoid the asymmetrical target, but in Fig. 7.8(B) they had learned to avoid the symmetrical one. Finally, in a test with the symmetrical rewarded target versus another symmetrical pattern (Fig. 7.8H), they were lost because they had learned the unrewarded one. In this experiment, they did not rely on recognition of the rewarded target at all, and learned to avoid the unrewarded one.

With equal width, modulation and colour, bees learned position of blue The bees were trained to discriminate between a square pattern of yellow bars and the same pattern inverted. There was no difference in total colour but they differed in the vertical positions of yellow areas and bars (Fig. 7.9A). This was a difficult task and they learned slowly. When tested with the upper parts of the patterns only, they were lost (Fig. 7.9B). Therefore, they had not learned a differ­ ence in length of edge or modulation of their green receptors, and they required an essential cue in the lower part of the display. When tested with the lower halves

142

Chapter 7

Train, with same total green contrast and no colour difference; asymmetry rewarded (A)

Train, with same total green contrast and no colour difference; symmetry rewarded (E)

100%

100% 55°

55°

4° bars (B)

92%, n = 200 100%

4° bars (F)

Test

They had not learned the polarity. 100% Test

100% Test

51%, n = 100

52%, n = 100

(C)

88.5%, n = 200

They had not learned the rewarded target. (G)

100% Test

57%, n = 100

81%, n = 100

Confusion between two symmetrical targets.

The unrewarded target was sufficient.

(D)

100% Test

(H)

100% Test

66%, n = 100

54%, n = 100

Whole pattern not required; also available: retintopic memory, average position of blue, width and asymmetry of modulation.

They had not learned the rewarded target concentration of modulation was retinotopic.

Fig. 7.8.  Bees discriminated between asymmetrical and symmetrical patterns of bars. (A) Training patterns with an asymmetry rewarded. (B) The bees failed when the unrewarded pattern was changed. (C) The bees failed when both displays were symmetrical. (D) Discrimination returned with two bars on the left versus two bars at the correct width. (E) Training patterns with a symmetrical rewarded pattern. (F) The bees failed when the unrewarded pattern was changed. (G) Discrimination returned when the unrewarded target was restored. (H) The bees did not recognize the rewarded pattern when both targets were symmetrical.

only, or with the lower yellow panels versus black, the scores were lower (Fig. 7.9C, D). However, when tested with yellow panels in the lower half versus a square of blue also in the lower part of the display, they switched preference to the blue (Fig. 7.9E). Therefore, they had learned to prefer the blue content of the yellow panels in the lower part of the pattern. Bees spontaneously prefer blue or the blue in white and they measure the position of blue or white in the vertical direction (Fig. 7.9).

­ herefore, for bees, yellow was a shade of T blue. The sensitivity to blue was high. In sunlight, as a stimulus to the blue receptor, Dresden Yellow generates only 13% of the stimulus from the blue in white paper, as shown in Table 7.1. Colour was equal on the two targets, so the bees had to learn the position of blue. This experiment confirmed that the vertical position of green contrast was not a preferred cue, and blue content was measured in the lower part of the display.



Symmetry and Asymmetry

Train on symmetrical targets with equal colour and modulation (A)

100% 55°

They learned slowly; 82%, n = 200 Test on upper halves only (B)

100%

They recognized nothing; 45.5%, n = 200 Test on lower halves only (C)

100%

They had learned something; 66.0%, n = 200 Test versus black (D)

100%

68.0%, n = 200 They had learned to go to yellow below the reward hole. Test versus blue (E)

100%

28.5%, n = 200 They preferred to go to greater blue content in lower half.

Fig. 7.9.  Symmetrical targets could be distinguished by the positions of green modulation and blue content in the unrewarded pattern. (A) Training patterns. (B) The upper halves were insufficient. (C) Lower halves displayed a cue. (D) One cue must have been on the unrewarded target. (E) The cue must have been the greater amount of blue in the yellow of the rewarded target.

With all else equal, bees learned vertical position of blue Naïve bees were trained to prefer a blue spot under four vertical yellow bars on a black

143

background, versus the same pattern inverted (Fig. 7.10A). When tested with yellow bars only (Fig. 7.10B), or with blue spots only (Fig. 7.10C) on a black background, performance was reduced but positive, showing that the positions of either the bars or blue areas were sufficient. A new group of bees were trained to the same patterns with yellow horizontal bars and a blue spot on a black background on each training target (Fig. 7.10E). When tested with yellow bars only (Fig. 7.10F), performance was lost, showing that the blue areas were essential. When tested with blue spots only (Fig. 7.10G), they performed better than in the training, so the cue was the position of blue (compare with Fig. 2.2). When tested with vertical bars on the training target recognition was destroyed by the unexpected presentation of strong green modulation that was not in the training (Fig. 7.10H). This cancellation of learning by stronger attraction of green contrast has been observed in similar situations elsewhere.

With all else equal, bees learned position of blue content of yellow Although bees do not detect yellow as a colour that is associated with food, bees were trained next to distinguish between entirely yellow mirror images on a black background (Fig. 7.11A). When tested with yellow bars only (Fig. 7.11B), or with yellow spots only (Fig. 7.11C) on a black background, the trained bees failed. When a blue spot on the unrewarded pattern replaced the yellow one (Fig. 7.11D), the score was reduced but still acceptable. However, when the rewarded pattern with a yellow spot was tested versus itself with a blue spot (Fig. 7.11E), the preference was reversed. This interesting result shows that ­ the trained bees had learned to prefer the rewarded training pattern because it displayed some blue in the yellow spot that was on the left of the bars. They were attracted to the spot with the most blue in the expected position relative to a landmark with green contrast, and used the pattern polarity to guide them to the correct arm of the choice apparatus.

144

Chapter 7

Train on single blue spot and yellow bars versus mirror image (A)

Train on single blue spot and yellow bars versus mirror image (E)

100%

100%

55°

55°

Slow to learn, 90.0%, n = 200 (B)

Test bars only

Trained in 90 min; 83.0%, n = 200 (F)

100%

Bars sufficient, 65.5%, n = 200

(C)

Test spots only 100%

Failed with bars alone, 54%, n = 200 (G)

79.0%, n = 200 They had learned blue position. (D)

Test 100%

71.0%, n = 200 Horizontal bars made no difference.

Test bars only 100%

Test spots only 100%

88.0%, n = 200 They had learned blue position. (H)

Test 100%

46.5%, n = 200 Vertical bars spoiled discrimination.

Fig. 7.10.  The vertical positions of blue and of vertical green modulation were cues. (A) Training patterns. (B, C) Yellow bars alone or blue spots alone were sufficient. (D) Rotation of the bars made little difference because the bees also used the position of blue. (E) New training patterns. (F, G) Horizontal bars alone were useless, but blue spots alone were effective. (H) Increase of green modulation destroyed the discrimination.

When trained with the yellow bars above the yellow spot on a black background versus the same pattern inverted, bees learned nothing although trained all day (Fig. 7.11F), showing that no cue was available.

Bees recognized asymmetrical green ­modulation alone To demonstrate the point with a simple shape, bees were trained to distinguish

mirror images that displayed no cue except polarity of asymmetrical green modulation, using yellow outlines of a triangle on a black background (Fig. 7.12A). They learned this task readily, and were not influenced by addition of a vertical yellow bar at the centre (compare Fig. 7.12B with Fig. 7.12E). The available cue can be represented as simple asymmetry of green modulation (Fig. 7.12C), with no differ­ ence in colour, height, blue modulation, or anything else known to be detected by bees.



Symmetry and Asymmetry

145

Train on yellow spot and bars versus mirror image (A)

Test 100%

(D)

100% 55°

65.5%, n = 200 Test 100%

88.5%, n = 200 (B)

100% Test

(E)

47.0%, n = 200 (C)

100% Test

57.5%, n = 200

12.5%, n = 200 Train 100%

(F)

48.5%, n = 200 after training all day

Fig. 7.11.  To the bees, a yellow spot was a shade of blue. (A) Training patterns. (B) Green modulation alone was insufficient. (C) With yellow spots only, maybe a small signal was detected. (D) Blue on the right had little effect. (E) They abandoned the rewarded pattern for a blue spot on the left. (F) Yellow modulation and yellow spot were not distinguished from the same inverted. Train with equilateral triangle of yellow bars versus mirror image

Train with blue equilateral triangle and yellow bar versus mirror image

100%

100% 55° (D)

(A) 77.5%, n = 200

83.5%, n = 200

Test with bar centred (B)

100%

70.5%, n = 200 No effect of the bar (C)

Basic asymmetry of modulation

Test with bar centred (E)

100%

50.5%, n = 200 Equal blue on each side, no asymmetry (F)

Asymmetry of blue and green contrast

Colour content and total modulation is equal in these pairs of patterns. Fig. 7.12.  The two types of cues identified as lefty/right signposts. (A–C) Edges of yellow triangles were easily discriminated by asymmetry of green modulation. (A) Training patterns. (B) The central yellow line had no effect. (C) The basic pattern of asymmetrical modulation. (D) Relative positions of blue colour and a landmark identified the direction of polarity. (E) With the yellow landmark bar at the centre of the blue area, discrimination failed. Polarity was the cue, not the shape of the triangle. (F) The basic pattern of polarity of blue content and green contrast.

146

Chapter 7

Bees learned asymmetry of green modulation and its position relative to blue In contrast to the above, bees were trained to distinguish a blue triangle and yellow bar from its mirror image (Fig. 7.12D). The bar was ineffective at the centre because the polarity disappeared (Fig. 7.12E). The basic polarity algorithm was the position of blue relative to a landmark of green modulation (Fig. 7.12F).

Earlier Results with Symmetrical Patterns In previous work, when bees distinguished between an asymmetrical pattern and its mirror image, the performance was recorded but no explanation in terms of cues was offered (Fig. 7.13A). Failure to discriminate between some pairs of symmetrical images (Fig. 7.13B) was never explained, except as being too difficult for bees. In previous discriminations between similar black shapes at different heights on a white background, the only cue previously identified was the difference in height of the

centres of black (Horridge, 2003). An equilateral triangle was discriminated from the same triangle inverted (Fig. 7.13C), but not when the centres of the two triangles were at the same height (Fig. 7.13D) (Horridge, 2009a). The conclusion that bees discriminated the height of the centre of gravity of the black will now have to be revised in the light of the new results. The cue for the triangles must have been the height of the centre of the blue content of the white background because black reflected no light. A special case of discrimination of an asymmetrical black triangle versus its mirror image (Fig. 7.14A) concluded that one cue was the asymmetric distribution of green modulation, but the possibility of other cues was not considered (Horridge, 2012), and there was no analysis in colour. A coincidence of a vertical edge at the side of an area of absence of blue must have been the most powerful cue. Revision of old conclusions, taking new discoveries into account, is now essential. Only two types of photoreceptors of the compound eyes are known to be involved in the feeding behaviour, with a peak in the green near 560 nm, and blue near 450 nm.

Train with spot and cross versus mirror image (A)

100%

Train in fixed positions (C) 100% 55°

75.3%, n = 147

65%, n = 200 after 6 h training Train with spot and cross versus same inverted (B)

100%

No learning after 6 h training

Test, equal blue above and below (D)

100%

50.9%, n = 151

Fig. 7.13.  With symmetrical patterns, the cue is not obvious. (A) Bees learned to distinguish with difficulty, and the cue for horizontal polarity was unknown. (B) They failed to distinguish these patterns after 6 h training; therefore had found no cue for vertical polarity. (C) Bees easily learned to distinguish the black triangle from the same inverted. (D) They failed to distinguish the shapes when the centres were at the same height. (Adapted from Horridge, 2003.)



Symmetry and Asymmetry

Train with black equilateral triangle versus mirror image (A)

100% 55°

78%, n = 200 Test with edges only (B)

100%

66.0%, n = 200 Test with skeleton orientation (C)

100%

73.5%, n = 200 Test training pattern versus cue (D)

147

fussy along the horizontal (Fig. 7.11C). The trained bees did not notice the horizontal position of the blue in the yellow spots in the training in Fig. 7.11(A), but they learned the polarity of blue in the yellow spots in relation to the vertical bars (Fig. 7.11D, E). When blue content was the same on each target, they distinguished a difference in vertical position of blue (Figs 5.8, 7.13C, D). These differences presumably relate to the known projections of blue and green receptor pathways on long tangentially arranged neurons of the medulla. These cues were significant for the detection of asymmetry. The strongest cue, however, was a horizontal relation between the centre of blue content and a landmark with green contrast, and the angle between (Fig. 7.10E). Next in significance was the asymmetry of green modulation (Fig. 7.10C). These cues indicated ‘turn left’ or ‘turn right’, and bees can also be trained to ‘accept’ or ‘avoid’ them.

100%

Importance of Polarity Cues in Route Finding 51.5%, n = 200 Fig. 7.14.  An example where a cue for asymmetry was inferred. (A) Training patterns. (B) A test with edges only showed that a black area was not essential. (C) A subtle test showed that horizontal and vertical edges in the correct positions were a sufficient cue. (D) The trained bees could not recognize the rewarded pattern or distinguish it from the cue. Distribution of blue in the white was also a factor (not tested).

Bees have no receptors for black or white. Therefore, bees trained on black/white patterns must have learned coloured cues. Recent studies show that the simple cues are the content of the colour blue, the modulation in the green receptor channel, and similar but less used modulation in the blue receptor channel. Combinations of simple cues form coincidences (Chapter 8, this ­volume). We know already that the position of green modulation in the vertical direction was not easily distinguished. Bees measured the position of blue in the vertical direction (Figs 7.9, 7.10C, G) but were less

The new findings show that the role of almost all cues previously thought to be concerned with finding the route or recognition of the goal must now be reconsidered. Clearly, the foraging bee makes use of two kinds of cue: those with polarity relate to route finding while preferences for symmetrical cues control choice of landing on the goal. Symmetrical patterns are similar to shop signs; they never deflect the customer to the shop next door. Similarly, advice to turn left or right is hardly the best signal indicating a flower. However, bees can learn to turn left or right when they detect a symmetrical cue such as a colour or black vertical line in a maze or on a route. Plotting their tracks ­reveals how they approach a symmetrical landmark then loop or turn in the direction of the reward (Fry and Wehner, 2002). Bees automatically learn first from the unrewarded target, because they learn to avoid everything that yields no reward. They need this at every point along their familiar route. They detect, learn and interpret the polarity that acts like a signpost,

148

Chapter 7

Train in fixed positions

Train in fixed positions (I)

(A) 100%

100%

55°

53.5%, n = 200 after 5 h training

75.3%, n = 147 Test with centres at the same height (B)

Train in fixed positions

Test 100%

(J) 100%

50.9%, n = 151 The cue was a difference in height.

57%, n = 200 after 3 h training Train in fixed positions

Train in fixed positions

(K)

(C) 100%

68.0%, n = 200

81.0%, n = 200 (D)

(L)

Test 100%

52.5%, n = 200 They fail with similar pattern of modulation. (E)

100%

55°

53.5%, n = 200 They did not recognize the cross. (M)

Test 100%

Train in fixed positions

Train in fixed positions (F)

(N) 100%

100%

55°

78.7%, n = 300

(H)

Test 100%

52.0%, n = 200 They did not recognize the ring. They learned black at the cross centre.

56.5%, n = 200 They had not learned the black disc.

(G)

Test 100%

Test 100%

76.0% , n = 200 (O)

Test 100%

52.5%, n = 200

67.0%, n = 200

They did not learn the rewarded pattern.

Orientation cue remains

Test 100%

73.0%, n = 200 Cue is revealed as modulation difference.

(P)

Test 100%

50.5%, n = 180 Similar orientation not distinguished.

Fig. 7.15.  Abbreviated results of nine experiments that identified typical cues on the unrewarded target, and whole patterns were not recognized by symmetry. (Adapted from Chapter 11 in Horridge, 2009b.)



Symmetry and Asymmetry

and they turn left or right. When no cue is available, or when they detect an avoidance cue, they may also turn round and return to the previous choice point. Bees at each choice point do not detect symmetry as a first preference. Bilateral or radial symmetry does not influence the choice of other cues (Fig. 7.15), so flying bees are not deflected; they carry on a straight track, and land on a symmetrical flower. There is no sign that they even detect the symmetry that humans so admire. As a result, flowers exhibit an immense variety of radial spokes, rings and central

149

spots within the floral symmetry. Simultaneously, the route to the reward is crowded with an irregular distribution of edges where the bees find an endless supply of asymmetrical green contrast at edges and other landmarks in the foliage. We now see how the unique visual system of the insect is adapted to this environment. Asymmetry is for the signposts on the route; symmetry is for the terminus. We have distinguished road signs and shop signs. Possibly, these rules can be used to devise ways of directing bees along a printed route in glasshouse agriculture.

References Fry, S.N. and Wehner, R. (2002) Honeybees store landmarks in an egocentric frame of reference. Journal of Comparative Physiology A 187, 1009–1016. Horridge, G.A. (1996) The honeybee (Apis mellifera) detects bilateral symmetry and discriminates its axis. Journal of Insect Physiology 42, 755–764. Horridge, G.A. (2003) Visual discrimination by the honeybee (Apis mellifera), the position of the common centre as the cue. Physiological Entomology 28, 132–143. Horridge, G.A. (2009a) Visual discrimination by the honeybee (Apis mellifera). In: Lazareva, O., Shimizu, T. and Wasserman, E. (eds) How Animals See the World. Oxford University Press, Oxford, pp. 165–190. Horridge, G.A. (2009b) What Does the Honeybee See? And How Do We Know? A Critique of Scientific Reason. ANU E Press, Canberra, 360 pp. Available at: http://epress.anu.edu.au/honeybee_citation.html (accessed 1 November 2018). Horridge, G.A. (2012) Visual discrimination by the honeybee. In: Lazareva, O., Shimizu, T. and Wasserman, E. (eds) How Animals See the World. Oxford University Press, Oxford, pp. 165–190. Horridge, G.A. (2015a) How bees distinguish a pattern of two colors from its mirror image. PLoS One 10(1), e0116224. DOI: 10, 1371, 1–23./journal. Pone.0116224 Horridge, G.A. (2015b) How bees distinguish patterns by green and blue modulation. Eye and Brain 7, 83–107. Horridge, G.A. (2016) Parallel inputs to memory in bee colour vision. (Plenary Lecture. 31 August 2015, at International Congress, International Society of Invertebrate Neurobiology, Tihany, Hungary.) Acta Biologica Hungarica 67, 1–26. Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222. Ronacher, B. (1992) Influence of unrewarded stimuli on the classification of visual patterns by honey bees. Ethology 92, 205–216.

Chapter 8 Bee Vision is Not Adapted for Pattern or Shape

Understanding how science is done increases trust in science, as it can be seen to be built on reliable data. (Nurse, 2015)

It is usually assumed that insects probably detect and remember vague, crude or fuzzy patterns or shapes, and that the onus of proof lies with those who would show that they do not. However, the opposite is the case. It is easy to show that bees detect cues (Chapter 6, this volume), and impossible to demonstrate that memory of shapes really is a memory of the shape and not of some cue that has not been identified. Wherever shape has been identified it has always been possible to find a simple cue.

Lack of Shape Discrimination is Hard to Prove Lack of generic shape discrimination cannot be proved. Previous chapters have shown that bees distinguish colours by their blue content and contrasts at their edges. They detect cues and their coincidences and combinations, and asymmetry, but those abilities do not prove that bees do not see the layout of patterns in the way that we humans do. We cannot imagine how we would not see the shape of an object that we examine and recognize, so you will find it hard to accept that 150

bees recognize and distinguish simple patterns and some shapes but do not see them. Having explained much of bee vision by the cues, however, we may have overlooked other ways to detect shape. The only way to find out is to investigate a variety of shapes. In the salient example when bees distinguished between a large and a small black spot, we found that the cues were not what you would expect. They were the width between green modulation at two edges, and the location and amount of blue in the white background (Fig. 3.11). In all examples where bees are tested for vision of shape, therefore, it is essential that the training patterns should display no difference in known cues: colour, areas of black on white, area and colour of background, lengths and positions of edges, width and average height of the black areas, and blue content. Otherwise, if any cue differences are displayed, and there were many examples in earlier work, the bees will grab the preferred cue, and the tests will not demonstrate vision of shape. Similar considerations apply to discrimination of photographs, panoramas, faces or paintings by Picasso; successful discrimination must be followed by critical tests of what the bees actually detected. We have at least three ways to demonstrate that bees do not remember shapes as whole shapes. First, we can show that, when they appear to discriminate shape, they actually

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)



Bee Vision is Not Adapted for Pattern or Shape 151

use simple cues that have been previously demonstrated. This is an alternative explanation and says nothing about a possible additional memory of shape. Secondly, there are many examples of patterns between which they cannot discriminate, and we can investigate why they fail. This method requires a large number of examples to be secure and is not relevant to normal behaviour. Becoming more critical, we can show that they discriminate between two shapes but cannot remember the rewarded or the unrewarded shape that they were trained on when tested versus a different pattern that displays the same cues (see the following figures discussed later in the chapter, Fig. 8.1F, G; Fig. 8.2B, D; and Fig. 8.4G, H). This positive test for absence of recognition of shape and pattern is the principal tool in this chapter. It is not negative evidence; it is positive evidence of failure to find any difference in patterns that look different to us.

One or Several Local Regions of the Eye There is abundant evidence that large patterns that extend more than 100° are discriminated by the spatial layout of their outer parts (Horridge, 1996), giving the impression that the whole shape is detected (Fig. 6.8), but so far we have only the demonstration that this happens in experiments, and both eyes may be involved. When the criterion of a successful choice is the bee landing on the display, the angular size at the moment of choice is unknown, but probably very large. It seems likely that a bee in flight views the panorama with different local regions of the eye that detect separate cues, but this topic has been impossible to investigate. Work by Hertz (1929, 1930, 1931) on retinotopic detection of small targets, by Wehner with very large displays in the late 1960s (Wehner, 1967, 1968), and by Ronacher and Duft (1996), Dafni et  al. (1997) and Collett and Zeil (1998), suggested that bees compare each test pattern with the training pattern reassembled in memory. The evidence was that recognition was improved the more the black areas of the test display

overlapped those of the training display. Of course, this trend would be expected whatever theory was adopted, so it was useless as evidence. Moreover, retinotopic matching is unsatisfactory as a recognition process because the size of the image depends on the range. As more and more evidence of cues appeared, the idea of image matching faded away. Finally, Efler and Ronacher (2000) found recognition of test patterns that showed no overlap with the training pattern. To avoid these problems, all my work was with patterns on targets subtending less than 55°, viewed at a distance of 30 cm in the standard Y-choice apparatus. On this scale, previous chapters have shown that bees detect asymmetry and sum totals of blue content and green modulation over the whole target. They identify groups of radial or tangential edges that are related to a hub. Edges with different orientations can be separately detected in the left and right sides of the target. The angular width between two vertical edges can be learned. The position of blue is measured in the vertical direction, and by its horizontal angular distance from a vertical edge. Although restricted by the use of a few cues, the display is retained in angular coordinates upon the eye, but by spatial relations between cues, not by spatial layout of pattern or shape. It is the cues that are retinotopic. However, the question persists: is there something more than cues that bees detect in patterns and shapes?

The balance of preferences between two targets Naïve bees demonstrate an order of preference for all displays presented to them, so one should test first for spontaneous preferences, and train against them. Training is a process of modifying preferences. When presented with a choice between two patterns, one of which is rewarded, the other not, cues displayed equally on both are not used because the bees learn them on one target and unlearn them on the other. Bees first learn from the unrewarded pattern because they learn by trial and error. When they make an error,

152

Chapter 8

target is moved around to make them search (Lubbock, 1881; Turner, 1910, 1911).

they are forced to take a second look and search again, but when a naïve bee goes first to the rewarded target, there is no incentive to look and learn. Bees learn first the most preferred unique cue, not a difference, even if they learn to avoid it on the unrewarded display. This can lead to a false conclusion that they have learned to prefer the rewarded pattern. The only way to find out is to test both patterns separately against a neutral target (Fig. 8.1B, C and Fig. 8.2B, E, F). They learn to associate the correct target with a reward only after extensive training. When they learn on a single target, they may learn nothing but the local landmarks, unless the

The difference between two coloured discs In the first example, before training, bees preferred a blue spot to a buff spot, both subtending 20° at the choice point, but were trained to go to the buff one (Fig. 8.1A). When tested against a plain white target, preference for the buff spot increased (Fig. 8.1B), because the newly exposed area of white on the unrewarded target displayed additional blue to avoid. The blue spot alone (Fig. 8.1C) was not avoided because a

Train on fixed patterns (A)

100% 55°

76.5%, n = 200 (B)

Test 100%

(E)

84.5%, n = 200 Newly exposed white means extra blue. (C)

Test 100%

Test 100%

72.5%, n = 200 They avoided the greater blue. (F)

Test 100%

48.0%, n = 200

50.5%, n = 200

Newly exposed white means extra blue.

No blue difference.

(D)

Test 100%

56.0%, n = 200 No blue difference.

(G)

Test 100%

52.0%, n = 200 No blue difference.

Fig. 8.1.  (A) Training patterns. (B) A higher score when the blue spot was removed. (C) The blue spot alone was not recognized. (D) Position in the training was not learned. (E) Colour was distinguished irrespective of pattern. (F, G) Colour was equal and the spots were not recognized. Percentage values indicate the percentage of bees visiting reward holes.



Bee Vision is Not Adapted for Pattern or Shape 153

Train on fixed patterns (A)

100% 55°

75.0%, n = 200 (B)

Test 100%

(E)

57.5%, n = 200 Equal green contrast and blue (C)

Test 100%

67.0%, n = 200 Blue difference and green contrast (F)

Test 100%

55.5%, n = 200 Equal blue content and green contrast

Test 100%

74.5%, n = 200 Blue difference and green contrast

69.0%, n = 200 Blue difference and green contrast (D)

Test 100%

(G)

Test 100%

64.0%, n = 200 Blue minus green contrast

Fig. 8.2.  Trained on a pattern versus a blank target, the bees learned only that something lies at the right place. (A) Training patterns. (B) The training pattern is not distinguished from black spots. (C) Black spots were sufficient. (D) Reversal of colours scarcely noticed. (E, F) A buff or blue spot is adequate. (G) In a forced choice, the innate preference for blue was unchanged by training. (After Horridge, 2007.)

difference in blue content was balanced by a change in average height and lack of green modulation. With the training spots at corresponding places on the targets, the bees did not learn their position, as shown by a test with equal green contrast and no blue difference (Fig. 8.1D). When tested with the original spot versus a scattering of 40 small spots of the same colour, with the same total area as the large spot (Fig. 8.1F, G), they did not remember either large spot. They had learned only the total content blue cue, difference in green modulation, and the width of the buff spot (not tested here). The trained bees distinguished very well between a scatter of 40 small buff spots and 40 small blue spots on white backgrounds

(Fig. 8.1E). The blue colour did not have to be in a large spot, implying that a colour learned from a flower nearby may be transferred to the same colour in many scattered flowers further away. There are obvious implications there for the evolution of bushes with many small flowers. The bees appeared to have learned very well. In the tests, however, they could not recognize either spot. In the next experiment, bees were trained with the blue and buff spots on the rewarded target (Fig. 8.2A). In a test, they could not distinguish the coloured spots from a pattern of black spots (Fig. 8.2B), and they responded just as well to black spots versus a white target (Fig. 8.2C). Trained bees could not distinguish the training target from its

154

Chapter 8

­ irror image (Fig. 8.2D). When tested with m the buff and blue spots separately against the white target, they preferred something to nothing and the blue more than the buff (Fig. 8.2E, F). When the blue spot was tested versus the buff one (Fig. 8.2G), they preferred the blue, as expected from their innate preference. When trained on Fig. 8.2(A), they had probably learned only to go to the most contrast. Please don’t conclude that bees cannot learn two colours. In Fig. 8.1, it was shown that they learned a difference between the same two colours. They also discriminate readily between patterns displaying two colours and a mirror image of the same patterns. The point is that they learn the minimum cue. The task in Fig. 8.2 required only that they distinguished something from nothing, which is a common situation. More clearly than with black and white shapes, these experiments show that the bees learn very little, sometimes as little as learning to avoid one preferred cue when trained to prefer a blank target. We are now in a position to examine claims made elsewhere that they really see shapes, symmetry, topology, or anything else that could be graced by an abstract noun. Fortunately, not many examples have been proposed recently.

Arrangements of separate spots Black spots display area width, modulation and position. Bees can distinguish two fixed spots from two similar spots at different positions (Fig. 8.3A), but three or four fixed spots are hard to distinguish (Fig. 8.3B, C). A large black spot on the left of a display is not distinguished from a similar one on the right side (Fig. 8.3D), unless a vertical black line acts as a reference mark (Fig. 8.3E). In this case, bees use a standard algorithm for polarity: whether greater blue content is left or right of a landmark of green contrast. They learn the polarity, but they have no record of which way they scanned the targets. When the training displays are randomly rotated during training, bees cannot even learn to distinguish between two spots

and three of the same total area, no matter how large the targets or how long they are trained (Fig. 8.3F, G, H). This seems to rule out counting of spots. Those on the left were fixed in position during the training; the pairs on the right were rotated between visits. When rotated, even two spots were not distinguished from three. Clearly, bee vision is not designed to distinguish patterns of spots.

Do bees count? There have been many reports that bees appear to count. Of course they do, in a way, because they detect a modulation difference of 10–12%, so they distinguish ten steps of amounts of modulation up to 100%. Previous claims that bees count omitted essential controls for the position of blue content and differences in modulation. Whether bees actually count is at present impossible to say. The trained bees measure blue content, amount of modulation, and asymmetry of modulation, but not numbers of items, so they only appear to count.

Blue content is summed and located but pattern lost In common situations, bees appear to discriminate pattern or shape when they do nothing of the kind. They do not see black because it emits no light and cannot be a stimulus. Bees do not identify two spots by their separate positions but simply lump them together and measure the green modulation and average vertical position of blue in the background (Fig. 8.4A–D). For bees, the number of spots makes no difference. The positions of the spots can be adjusted so that two targets that are indistinguishable to bees still look different to the human eye (Fig. 8.4B, D). The same applies to any pair of shapes or patterns. A difference in average position of black is detected by the corresponding difference in vertical position of blue in the background (Fig. 8.4E, F) or of the blue of the pattern itself (Fig. 8.4G, H).



Bee Vision is Not Adapted for Pattern or Shape 155

Train, equal edge and blue fixed positions (A)

Train 100%

Train, equal blue content randomly rotated (F) 100%

50°

50°

14° 48.0%, n = 460

70.3%, n = 300 (B)

Train 100%

(G)

100%

11.4°

67.2%, n = 500 Train 100%

57.7%, n = 800

10° (H) 100% Spot size

(C)

52.2%, n = 500

51.8%, n = 400 Train on black spots 20° versus 8°

Train on positions of spots (D)

Train 100%

(I)

48.7%, n = 600 trained all day

69.0%, n = 200 on 2nd day Test, making blue content equal

61.5%, n = 200 after training on (E) (E)

Train 100%

66.0%, n = 300

Train 100%

(J)

Grey 40% white

Test 100%

56.5%, n = 200

Fig. 8.3.  Patterns of spots. Each pair of patterns was trained separately (except J). (A–C) Positions of a few large spots were learned better than more spots with the same total area. (D) Left/right positions of a large spot were not discriminated, unless (E) a landmark was added. (F–H) In general, rotated patterns were not discriminated. (I) A large spot was readily discriminated from a small one. (J) Adding white to the large spot destroyed the cue, which was the difference in total blue content. (A–C, F–H from Horridge, 2000.)

Edges at right angles cancel orientation In 1967, Rudiger Wehner showed that when each pattern subtended 130° at the eye, bees discriminated a square cross from the same cross rotated by 45° (Fig. 6.8A) and the cue

was position of black, not orientation, as shown by cutting the edges into large steps (Fig. 6.8B). At that time, Wehner used words like ‘Winkelstellung’ and ‘Winkellage’ but Jander et al. (1970) used the equally vague ‘Kantenrichtung’ and ‘Gliederung’. Similarly,

156

Chapter 8

Train on fixed targets (A)

Train

100%

(E)

100% 55°

65.5%, n = 200

68.3%, n = 300 (B)

100% Test

(F)

48.0%, n = 200

100% Test

50.5%, n = 200

The bees learn average position not pattern. (C)

100% Test

Train with blue on black (G)

100%

32.5%, n = 200 73.0%, n = 200 (D)

Train new bees 100%

(H)

100% Test

48.8%, n = 1000

51.0%, n = 200

The bees cannot detect a difference.

Average position of blue is the cue.

Fig. 8.4.  The centre of the blue in the white over the whole display was remembered irrespective of pattern. (A) Training patterns. (B) Failure with the centres of white moved to the same level. (C) Small spots reversed the preference. (D) Training failed with two centres at the same level. (E) Train with two shapes at different levels. (F) Failure with centres at the same level. (G) Train with blue on black. (H) Test failed with the pattern centres at the same level. (After Horridge, 2003b.)

the word ‘orientation’ could apply to the edges or the bar as a whole. There were no experiments that distinguished between the variables of interest to the bees, position of blue area and angle of green contrasting edge to the vertical (as later demonstrated). Later, it was shown that bar width was a factor, and only the outermost areas of black were located in large targets (Fig. 6.8E–H). Imagine our surprise when, in 1994 Srinivasan et al., found that bees could not discriminate these two crosses when they subtended 45–55° at the eye, even when previously trained on an oblique bar versus

a vertical one to give them a clue (Fig. 8.5A, B). This was the first demonstration that objects at a distance and those in the landing situation were processed differently; it was a question of the angular size. The crucial point is that equal edges at right angles, or each group of three at 120° to each other in the optic medulla, produced zero orientation. This is a strange kind of average that emphasizes parallel edges but takes a strange average of other patterns of edges, retaining a measure of modulation. Oriented edges lose their orientation when cut into square steps (Fig. 8.5C, D). Minimum step



Bee Vision is Not Adapted for Pattern or Shape 157

Train in fixed positions (A)

100%

Train in fixed positions (C)

100%

50°

77.6%, n = 300

> 90% (B)

Test 100%

(D)

Test 100%

50°

49.0%, n = 200

52.0%, n = 200

Fig. 8.5.  (A) Training patterns. (B) Trained bees failed to detect the bars they had learned. (C) New training patterns. (D) Making steps destroyed the orientation cue.

size to be detected was very small, 3°, showing that orientation detectors are only three ommatidia long. This accounts for the poor angular resolution of 90° at the 50% response level. Therefore, when a pattern or shape displays several edges, there is a massive discrepancy between what is there and what the bee detects, because most of the orientation is averaged away, and perception of pattern or shape is impossible (Fig. 8.5D). Mutual destruction of two orientations proceeds gradually as opposing orientations approach within the display area (Fig. 8.6). This is one of the best ways to demonstrate orientation is reduced in patterns with edges at different angles and bees detect a peculiar kind of average orientation as a cue.

Train in fixed positions (A)

50°

77.6%, n = 300 (B)

100% Test

72.5%, n = 200 (C)

100% Test

64.0%, n = 200

Changes in the cue with range When a bee flies, objects like stems, twigs and leaves with edges at various orientations approach and recede in unpredictable ways, and the bee may need to remember the route. This is one more reason why it is useless to learn pattern, and to rely instead on average orientation of a very wide field and angles separating vertical edges. As the range changes, perceived orientation can disappear and even reverse. In 1992, we believed that bees learned horizontal edge orientation when trained on

100%

(D)

100% Test

56.5%, n = 200 Fig. 8.6.  The effect of proximity on the averaging orientation. (A) Training patterns. (B–D) As two orthogonal orientations come nearer, discrimination disappears.

randomly spaced and shuffled black/white horizontal bars versus similar vertical ones (Fig. 8.7A). When tested on broad bars

158

Chapter 8

Test at 40°

Train at 40°

(F)

(A)

40° 40° 90%

83%

10%

17%

(G)

(B) All tests

(C)

(D)

60°

(E)

53% 47% Test at 60°

(H)

83% 17% 53% 47% 23% 77% Test at 27 cm Test at Test at 9 cm subtending 40° 18 cm subtending 100°

100°

Test at 100°

23% 77%

Fig. 8.7.  (A) Training to an orientation cue with shuffled black stripes in random positions. (B–E) Tests with the composite bar at different ranges. (F–H) Realistic views of the test pattern displayed at different ranges.

composed of many small bars at right angles (Fig. 8.7B), the response depended on range (Fig. 8.7C, D, E). Targets increased in size as the bee approached, and an apparently single bar (Fig. 8.7F) was detected as nine separate bars (Fig. 8.7H). At the most distant position the period of the small bars was 4.6° (Fig. 8.7F), which was just resolved, and at the null point it was 6.7° (Fig. 8.7G), so resolution was a big factor, in the rotation of the cue and reversal of preference. When this experiment was published, based on our reading of human psychophysics, we claimed to examine ‘whether bees analyse patterns in terms of their local properties, global properties, or both’ (Zhang et al., 1992), but there were severe faults in the experiments. In 1995, before baffles were ­ introduced into the apparatus, bees could have detected the global orientation from a distance and the opposite orientation of small bars later in the flight path. Before the bees reached a range of 27 cm at the baffles, the small bars were not separately resolved.

Conversely, the bees probably detected little of the global pattern from a range of 9 cm because they had been trained to detect the orientation cue within a target subtending 45°. Probably they learned only a modulation cue from the unrewarded target; no tests of that were done. There were other problems at that time. Bees were allowed two visits on each side of the apparatus in each test, so they could improve their marginal success rate at the second visit. Also, vertical edges generated more modulation than horizontal edges because bees in flight scan in the horizontal plane. Luckily, our conclusions were cautious. ‘Although our experiments demonstrate the existence of local and global analysis, they do not shed light on the underlying processes.’ How could they, without tests? Indeed, it was later found that bees indeed preferred to learn the modulation cue rather than an orientation ­ ntrained wasps cue (Horridge, 2007), and u and bees preferred modulation to orientation (Jander et al., 1970; Lehrer et al., 1995).



Bee Vision is Not Adapted for Pattern or Shape 159

The words ‘global perception’ were consistent with human impressions that the bees detected the pattern as a whole. At first, we assumed the bees detected the global orientation of a whole grating. This confusion is exactly what the numerous tests are supposed to eliminate.

Differences in edge orientation in separate eye regions We planned ‘to see how many parts of a pattern could be discriminated separately, and whether discrimination was lost on rotation or inversion of the parts’ (Zhang and Horridge, 1992). A target was divided into four quadrants with a differently oriented grating of period 8° in each. These patterns confused subsequent researchers but bees in the Y-choice maze readily discriminated the rewarded training pattern from a similar pattern with the quadrants rearranged (Fig. 8.8A). We expected ‘some idea of how an array of numerous feature detectors, each individually ineffective, can collaborate together to make specific ensembles that fit the pattern sufficiently well’ (Zhang and Horridge, 1992). This was rubbish, partly because different orientations so close together cancel out each other, but the experimental design and data were also faulty. We did not test what the bees actually detected, and at the time we were unaware that radial and tangential edge cues were preferred. One side of our training targets consistently displayed more horizontal edge, and the other side more vertical edge (Fig. 8.8A, B). We were unaware whether bees processed asymmetry of modulation, as they scanned the target. Also, our bees had 10 min on each side of the apparatus, which allowed two visits in each test, which was sufficient for them to add a few points to their borderline scores. Notwithstanding that our conclusions were rubbish, Giurfa et  al. (1999) made similar patterns with orientation cues in four quadrants, but allowed their bees to approach close to the targets, which therefore

subtended very large angles at the final choice point. In this quite different situation, the positions of peripheral black areas were discriminated with a high score. It was assumed that the bees detected the bar orientations, and the cancelling between different orientations was ignored. Their training and test patterns also displayed preferred radial versus tangential edges (Fig. 8.8E) that were never mentioned. They never tested what the bees detected, and the successful recognition alone tells us nothing about mechanisms, especially when the targets were very large. More recently, the same group trained bees with similar patterns but with shuffled thickness and positions of bars, versus a similar pattern with different orientations (Stach et al., 2004; Stach and Giurfa, 2005). This time the targets subtended 37° at the point of choice. Discrimination depended on green contrast, and therefore on edges. In the training targets and tests, the orientation cues summed to zero on one side of both targets (Fig. 8.8F, G), but on the other side were radial versus tangential edges that bees could discriminate, but the authors did not mention this. In some tests, the pattern was reduced to one bar in each quadrant (as in Fig. 8.8D), in others the details were shuffled within each quadrant, but well-known cues remained. Trained bees discriminated with black and white reversed, as expected because the orientations were unchanged. There was no mention of modulation, or different orientations that mutually cancelled. The trained bees could discriminate ­related but unfamiliar patterns, which was described as ‘global’ vision, and ‘under such a differential conditioning the bees learn the patterns as a whole, not only their local cues’ (Stach et al., 2004). There was no evidence for this conclusion. The same data appeared again, and supposedly demonstrated ‘categorization based on sets of multiple features’ and the bees ‘were shown to assemble different features to build a generic pattern representation which could be used to respond appropriately to novel stimuli sharing the same basic layout’ (Benard et al., 2006), revealing their belief in spatial

160

Chapter 8

Percentages of correct responses after 2 h training (A)

Zang and Horridge (1992, Fig. 3)

(E)

67%

Giurfa, et al. (1999)

33%

87%

13%

Radial versus tangential (B)

Zang and Horridge (1992, Fig. 4)

(F) Stach et al. (2004)

39%

61%

80%

20%

Radial versus tangential (G)

(C) Zang and Horridge (1992, Fig .5)

61.5%

Stach et al. (2004)

38.5%

Horizontal

29%

Radial versus tangential

Horizontal

(D)

71%

(H) Vertical

63%

27%

Horridge (1996, Fig. 2e)

49.4%

50.6%

Fig. 8.8.  Patterns displaying a different orientation in each quadrant. The cue was probably asymmetry of modulation, as in Fig. 6.2, generated by a scan (not tested). In general, different orientations cancelled and the preferred cues were radial and tangential edges. See the explanation in the text. (From Horridge, 2009.)

reassembly in the brain. There were no critical tests of what the bees actually detected, and certainly no evidence for their claims of high level processing, especially their claim that the bees responded to ‘the perceived layout’. They did not notice the radial and tangential cues or the asymmetry in the patterns. Although my objections to my own data had been aired extensively in my book (­Horridge, 2009, pp. 291–293), the same data from Stach et  al. (2004) was used again to show how ‘simple feature detectors can enable complex feature generalization and stimulus location invariance’ in bees (Guiraud et  al., 2018). Like Giurfa, and the others, Chittka was a student of Menzel, who in turn was a student of Lindauer and von Frisch, all with orthodox beliefs that bees see pattern

l­ ayout. Compared to feeding brood, making honey and similar skills, pattern perception is the least likely activity to demonstrate cognition, but Guiraud et  al. (2018) began with the statement ‘Honeybees have remarkable visual cognitive visual abilities, enabling them to classify visual patterns’. They made no tests of what the bees detected, or evidence of classifying patterns, or whether the bees learned the unrewarded target. They set out a model that was composed of orientation detectors of intermediate field size feeding into a network that ­terminated on model memory cells of the corpora pedunculata (mushroom bodies) in the brain. Realizing this network in a computer, and running the program with



Bee Vision is Not Adapted for Pattern or Shape 161

s­elected data, showed that some pairs of patterns, as in Fig. 8.8 yielded the published discrimination results. Of course the model worked because it was designed to do so, but the conclusions are not valid. The model also suffers from two fatal design faults. The first is that each computation would only work for one range. At different ranges these patterns change appearance (Fig. 8.7). Secondly, according to experts, there is little visual projection to the mushroom bodies, which receive terminals mainly from the olfactory lobes (Wolff and Strausfeld, 2016). It is more likely that the visual memory lies in the processing columns of the medulla and decisions are made in the deep optic glomeruli. Olfactory inputs are correlated in the antennal lobes and mushroom bodies, and initiate compact orders to accept or avoid what they jointly detect. Bees rely more on olfactory than on visual information.

Triangles and squares Before cues were discovered, triangles and squares were frequently used to illustrate that bees could learn their shape. In 1997, however,

I found that bees trained to discriminate equilateral black triangles at different heights (Fig. 8.9A) failed when tested with the centres at the same height (Fig. 8.9B). The same applied to white triangles on a black background (Fig. 8.9C), so the cue was not shape. Bees are very slow to learn to discriminate between a black square subtending up to 40° at the point of choice and the same rotated by 45° (Fig. 8.9D) when the centres are at the same height. The cue is the difference in modulation caused by the vertical edges in a scan by the flying bee.

A diamond versus a square with no blue difference This recently introduced strategy makes use of yellow on black (Fig. 8.10). Bees ignore height, and the 12% blue content of the yellow, that are the same on the two targets. The green receptors detect strong green contrast at vertical edges when the bees in flight make a scan. A square diamond is easily distinguished from a square (Fig. 8.10A) and the trained bees distinguished the skeleton with edges

Train in fixed positions

Train in fixed positions (A)

(C) 55°

75.3%, n = 147 Difference in height of blue.

53.5%, n = 200 after 5 h training No difference in height of blue.

Test

Train in fixed positions

(B) (D)

50.9%, n = 151

57%, n = 200 after 3 h training

No difference in green modulation.

Small difference in green modulation.

Fig. 8.9.  Position (of blue in the background white) is discriminated but not shape. (A) Bees discriminate if the centres differ in position in the vertical direction. (B) They fail when the centres are at the same height (Horridge, 2003b). (C) They fail when black and white are interchanged. (D) They have difficulty in discriminating the rotation of a black square.

162

Chapter 8

Train, with green contrast no blue difference (A)

100% 55°

84.0%, n = 200 With equal areas of yellow, the only available cue was green modulation. (B)

100%

(F)

73%, n = 100 (C)

Test 100%

69%, n = 100 (G)

27%, n = 100 (D)

Test 100%

(H)

31.0%, n = 200

100% Test

55%, n = 100

Test no blue contrast 100% Test

100% Test

49%, n = 100

34%, n = 100 (E)

100% Test

Test

Test no green contrast (I)

100% Test

49.5%, n = 200

Fig. 8.10.  Shape distinguished by modulation difference. (A) Training patterns. (B–F) Trained bees avoided targets with most green modulation at vertical edges. (G–H) They fail to recognize the training shapes. (I) With no green contrast, they ignore a blue modulation difference.

only (Fig. 8.10B). Neither the diamond nor the square was recognized when presented versus a novel shape (Fig. 8.10C, G, H), but novel shapes were distinguished when one displayed vertical edges the correct distance apart (Fig. 8.10F). They distinguished between two novel shapes by the vertical edges (Fig. 8.10F). When shown two gratings, the trained bees avoided excess green modulation (Fig. 8.10D, E), and they failed when there was no green modulation difference (Fig. 8.10I).

To the bees, the yellow shapes would a­ ppear to be 12% blue, and the edges a bold green contrast, but blue content was the same on each target, so only the green modulation was learned. Mirror image triangles with a vertical edge In the past there have been claims that mirror images are variously favoured or confused



Bee Vision is Not Adapted for Pattern or Shape 163

in discriminations, but the trained bees were not tested for cues. Bees discriminated between a black triangle with one side vertical, versus its mirror image located at the same height (Fig.8.11A). The trained bees distinguished the triangles in white on a black background (Fig. 8.11B) and with edges only (Fig. 8.5C), so the cue was in the edge orientation (Fig. 8.11D). Areas or positions of points were not likely cues (Fig. 8.11D). With points only, tests were inconclusive (Fig. 8.11E), but a combination of vertical and horizontal black lines was effective (Fig. 8.11F), and included most of the input because trained bees could not distinguish these black lines from the rewarded target (Fig. 8.11G).

They could not recognize the rewarded target when it was presented versus a different pattern displaying the same cues in the positions where they had been trained to look for them. A disc versus a triangle of similar area and position This combination displayed no vertical edges, no difference in height or area, but bees learned to discriminate (Fig. 8.12A). Trained bees tested with the disc versus a random pattern of spots scarcely recognized the difference (Fig. 8.12B). Clearly, they had not learned to go to the disc. When tested

Train in fixed positions no blue difference (A)

100% 55°

78.0%, n = 200 (B)

Test 100%

(E)

66.5%, n = 200

58.5%, n = 200

No change in green modulation (C)

100%

Test 100%

Reduction in green modulation (F) Test

100%

Test 73.5%, n = 200

66.0%, n = 200 No change in green modulation (D)

100%

No change in green modulation (G) Test

100%

Test 69.0%, n = 200 No change in green modulation

51.5%, n = 200 No difference in green modulation

Fig. 8.11.  Mirror images of a triangle are discriminated when one edge is vertical. (A) Training patterns. (B, C, D) Trained bees discriminated when black and white were interchanged, when only edges were displayed, or when the triangles were smaller. (E) The positions of corners had little effect. (F) Average edge orientation on the two sides was a cue. (G) Failure to distinguish between the rewarded black triangle and the edge orientation.

164

Chapter 8

Train in fixed positions (A)

100% 55° 81.0%, n = 200

(B)

Test 100%

(E)

60.0%, n = 200

56.5%, n = 200 Rewarded disc not recognized (C)

Test 100%

(F)

66.0%, n = 200

(D)

72.5%, n = 200 Size matters a little.

Some green modulation difference Test 100%

52.0%, n = 200

Triangle was recognized. Test 100%

Test 100%

(G)

Edge orientation not recognized Test 100%

52.5%, n = 200 Not distinguished.

Fig. 8.12.  (A) Training patterns. (B) No preference for rewarded disc. (C) Avoidance of triangle. (D) Discrimination independent of size. (E) Some cue in edges only. (F) Exact layout of edges was not relevant here (contrast with Fig. 8.11). (G) Preference was lost with similar oblique edges.

with the spots versus the triangle, they avoided the triangle (Fig. 8.12C). The trained bees discriminated a smaller disc and triangle (Fig. 8.12D) and also the outlines (Fig. 8.12E), but could not distinguish the triangle from the same inverted (Fig. 8.12F), showing that they had not learned the locations of edge orientations. When white oblique lines were drawn on the disc, however, the bees could not distinguish it from the triangle (Fig. 8.12G). Therefore the cue was the modulation at the edges of the triangle, and they had not learned shape.

A ring versus a cross Previous authors assumed that bees discriminated these two shapes, as indeed they do, but bees find a small difference not noticed

in earlier experiments. Bees were trained to distinguish a rewarded black ring (inside diameter (ID) 18°, outside diameter (OD) 33.4°) versus a black cross of similar area (Fig. 8.13A). Initially they avoided the ring innately. Trained bees could not distinguish the ring versus a pattern of spots (Fig. 8.13B), suggesting that they had learned nothing about the ring. They failed when tested with the ring versus the cross with the black centre removed (Fig. 8.13C). When a black disc (OD = 28°) of similar area was tested versus the black cross (Fig. 8.13D), there was black around the reward holes on both targets, and again the trained bees failed. This result shows that they did not recognize the cross. With the pattern of spots versus the cross (Fig. 8.13E), with the cross minus its centre versus the black disc (Fig. 8.13F), and with the cross minus its centre versus



Bee Vision is Not Adapted for Pattern or Shape 165

Train, ring versus cross (A) 100%

55°

68.0%, n = 200 (B)

Test 100%

(E)

48.0%, n = 200

70.5%, n = 200

They did not recognize the ring. (C)

Test 100%

Cue was white at the centre. (F)

52.0%, n = 200

Test 100%

Test 100%

31.5%, n = 200

Same cue on each side. (D)

Test 100%

Cue was white at the centre. (G)

Test 100%

53.5%, n = 200

62.5%, n = 200

They did not recognize the cross.

Cue was white at the centre.

Fig. 8.13.  A cue on the unrewarded target. (A) Training patterns. (B, C) The ring was not distinguished from a pattern of spots or a hollow cross. (D) A solid black disc was not distinguished from the cross. (E, F, G) The cue is the black around the centre on one target but not the other, irrespective of the pattern. (After Horridge, 2006.)

the intact black cross (Fig. 8.13G) they performed as well as in the training, showing that cues were available although the patterns were so altered. Therefore, the necessary and sufficient cue was the black around the reward hole on the unrewarded target, and there was clearly no discrimination of shape as assumed by Zhang et al. (1995). A ring and a cross versus a plain white target When given a ring versus a cross (Fig. 8.14), bees used the difference in black at the centre as the cue, and learned neither pattern. In

the next experiment the bees were trained with a ring and a cross on one display, versus a white target (Fig. 8.14A) simulating an isolated landmark. The trained bees distinguished a pattern of spots versus a blank (Fig. 8.14B), but could not recognize the cross versus the spots (Fig. 8.14C), so they cared little for either training pattern. They had noticed the position of black on the target, however, as shown by testing with either the ring or the cross in a different position (Fig. 8.14D, E). In a separate experiment, they could not be trained to discriminate between the ring/cross pattern and the pattern of spots (Fig. 8.14F).

166

Chapter 8

Train on fixed patterns Train

(A)

100% 55°

85.0%, n = 200 (B)

Test

(D)

100%

Test 100%

78.5%, n = 200

61.5%, n = 200

Any black is sufficient. (C)

(E)

Test 100%

Test 100%

51.0%, n = 200

63.0%, n = 200

They do not recognize the training patterns.

They recognize the positions of centres.

(F)

Train 100%

49.0%, n = 200 after 4 h training No contrast difference, so no hubs.

Fig. 8.14.  Excellent recognition of the place, but failure of the bees to recognize either circular or radial symmetry presented together. (A) The bees readily learned the task. (B) The trained bees accepted 12 squares equally well. (C) They failed to discriminate the rewarded target from the 12 squares. (D, E) The training position of blue or pattern hubs had been learned. (F) In a new training, the bees could not discriminate the ring and the cross from the 12 squares.

This result illustrates how little they have learned, and that the training score was high because the task was easy. The bees detected little more than position and a difference in modulation. A ring versus a disc Learning was slow because the bees first avoided the ring, but the score reached 70% after 2 h of training (Fig. 8.15A). Trained

bees distinguished in tests with a difference in black around the reward hole irrespective of the patterns (Fig. 8.15B, C). When this cue was lacking, the bees failed with any shape (Fig. 8.15D, E). They probably also detected a modulation difference, which in the spot is half that in the ring. I could not repeat an earlier study claiming that bees trained on a rewarded ring versus a large round spot could transfer the discrimination to patterned targets raised over a patterned background, and discriminate



Bee Vision is Not Adapted for Pattern or Shape 167

Train on fixed patterns (A) 100%

55°

73.5%, n = 200 (B)

Test 100%

(D)

50.5%, n = 200

64.5%, n = 200 (C)

Test 100%

Test 100%

(E)

Test 100%

72.5%, n = 220

54.0%, n = 200

Bees detect the cue irrespective of shape.

The cue is the same on each. The ring was not recognized.

Fig. 8.15.  With a ring versus a spot, the cue was the black near the centre of the unrewarded target. (A) Training patterns. (B, C) As long as the cue was present, the shape was of no consequence. (D, E) Failure to distinguish between the rewarded pattern and a hollow square or a hollow cross, because the cue was lacking on both.

the shapes from the parallax as the eye moved (Zhang et al., 1995).

A thick black O versus a large letter S Bees were trained with a large black O (OD = 33.4° and ID = 18°) versus a large black letter S of the same area (Fig. 8.16A), as used by Chen et al. (2003) to show discrimination of topology. Naïve bees at first avoided the O and learned slowly. Tests were done only when the score was over 70%. Clearly, the bees had not learned to go to the O (Fig. 8.16B), but tested with a pattern of spots versus the unrewarded S, they avoided the S (Fig. 8.16C). The mirror image of the S was weakly discriminated from the S (Fig. 8.16D), suggesting an additional cue beside the black near the centre. Other tests were required to demonstrate cues. The trained bees failed with an

oblique bar versus the S (Fig. 8.16E). The O was then discriminated from two thin bars (Fig. 8.16F) but failed when the thin bars were turned through 90° (Fig. 8.16G). Therefore, the two cues already familiar from earlier work were the black near the reward hole and the edge orientation at the centre of the S. Recognition was not related to the topology of the shapes, contra Chen et  al. (2003).

Discrimination of the rotation of a sector pattern Until quite recently it was accepted that bees could be trained to remember the layout or the global aspects of a pattern. For example, ‘insects are able to compare a stored neural image . . . with a current neural image . . . has directly been shown in honeybees’ (Wehner, 1981). With reference to the

168

Chapter 8

Train in fixed positions (A) 100%

55°

76.0%, n = 200 (B)

Test 100%

(E)

53.0%, n = 200

50.5%, n = 180

No difference in available cues (C)

Test 100%

Test 100%

No difference in available cues (F) Test 100%

66.0%, n = 200 65.5%, n = 220 A cue is detected but the S and the O are not fully recognized. (D)

Test 100%

(G)

Test 100%

61.0%, n = 200 55.0%, n = 200 Orientation cue reversed and similar cue at the centre in each pair Fig. 8.16.  Cues were unrelated to the topology. (A) Training task. (B) Failure with the O versus a pattern of spots. (C) Test with the pattern of spots versus the S. (D) Test with the mirror image of the S versus the S. (E) Failure to discriminate an oblique bar from the S. (F) The O was discriminated from the thin diagonal bars. (G) The O was discriminated from thin bars rotated through 90°. The cues were therefore central black and orientation of the central bar of the S.

sector pattern (Fig. 8.17A), ‘The only factor that can account for the bees’ ability to discriminate . . . is the exact retinal position of the black and white sectors’ (Wehner, 1981, p. 476). Actually, for 25 years no factors were tested. To find the cue, bees were trained on two patterns of six sectors, one rotated by half a period relative to the other (Fig. 8.17A). The trained bees failed to recognize the rewarded pattern versus the same rearranged (Fig. 8.17B). They had not learned the position of the hub because it was the same on both training targets (Fig. 8.17C). They failed when the horizontal sectors were removed

from the training patterns (Fig. 8.17D) but they discriminated very well when only the horizontal sectors were displayed (Fig. 8.17E). Therefore, modulation at vertical edges of the horizontal sectors on the unrewarded target was a sufficient cue.

Patterns that bees do not distinguish The final nail in the coffin of bee pattern perception was a collection of obviously different patterns that bees do not distinguish (Fig. 8.18).



Bee Vision is Not Adapted for Pattern or Shape 169

Train on fixed patterns (A) 100%

55°

78.7%, n = 300 (B)

Test 100%

(D)

52.5%, n = 200

56.0%, n = 200

Rewarded pattern not recognized (C)

Test 100%

58.0%, n = 200 The cue is on both sides.

Test 100%

Vertical edges lacking on both sides (E)

Test 100%

73.0%, n = 200 Cues are revealed as vertical edges.

Fig. 8.17.  Discrimination of sectors. (A) Training patterns. (B) Trained bees failed to recognize the rewarded pattern versus the rearranged pattern. (C) The bees had not learned the position of the hub, and the cue was the same on both targets (arrows). (D) The trained bees failed when the horizontal sectors were removed. (E) They discriminated with the horizontal sectors displayed. The cue was therefore the vertical edges of the horizontal sectors on the unrewarded target, as in Fig. 8.10. (After Horridge, 2006.)

Randomizing Cue Positions in the Training When the rewarded cue was kept constant during the training while the other parameters were randomized, bees could be trained to choose a black disc at a certain range irrespective of the angular size of the disc (Lehrer et al., 1988). They could also select a disc of a certain absolute size irrespective of the apparent angular size (Horridge et al., 1992). The angular size, the absolute size, and the range, all turned out to be cues that could be learned separately when the others were randomized. The same strategy was used with vertical parallel bars on one target versus similar but horizontal bars on the other (van Hateren et al., 1990). The positions and widths of the bars were randomized during the training, so that the bees ‘made their decision on the basis of orientation only’ (Srinivasan et al.,

1993). For a time, these results suggested that the orientation was detected irrespective of position, and that ‘specific features of the pattern, such as bars and edges, are extracted and their orientation analysed as in the mammalian cortex’ (Srinivasan et al., 1993). As shown later, however, feature detectors and cues were recognized in all the places where they had occurred on the target during the training. With vertical versus horizontal edges, the bees preferred to learn the modulation difference and ignore the orientation (Horridge, 2003a, 2007). Anyone familiar with the 20th century literature on honeybee vision of pattern and shape knows that every step along the way was balanced precariously on a guess, leaning towards how humans saw the world. Most researchers did not look for cues, and were misled by the obvious ability of the bee to distinguish simple patterns. Most of my work on patterns, for example, and that of my

170

Chapter 8

Bees confuse these patterns

55°

White merges into background Black not located horizontally Edges display no difference. Score 58.0%, n = 200 after 3 h

White merges into background Black not located horizontally and total white equal vertically. Score 50%, n = 200

No colour difference or central polarity Green contrast at edges was saturated Targets were effectively black all over. Score 52.5%, n = 200

Equal summed colour content Equiluminant to green receptors Outer edges equal green contrast. Score 49%, n = 200

Equal summed colour content Equiluminant to green receptors Outer edges equal green contrast. Score 51.7%, n = 300

Equal summed colour content Equiluminant to green receptors Outer edges equal green contrast. Score 48%, n = 100

Equal summed colour content Equiluminant to green receptors Blue contrast conveys no orientation. Score 47.7%, n = 300

Equal summed colour content Equiluminant to green receptors Blue contrast conveys no orientation. Score 53%, n = 100

Fig. 8.18.  Bees cannot distinguish these and countless other pairs of patterns. Of course, bees detect these patterns, but each pair displays no difference in cues.

colleagues in the years 1990–1995, was mistaken in some way; some of it just rubbish.

Global versus Local Perception: a Dog’s Breakfast An old controversy in human psychophysics easily intruded into studies of bee vision, as a discussion whether bees see a few highlights

called cues, or the whole picture or pattern supposedly reassembled somehow in the brain, although never demonstrated. We have no idea how perception works in the human brain, but we know that the retinal input is duplicated to many parts of the primate brain. Bees did not evolve a type of vision with pattern constancy independent of range; they would have no use for that. They evolved simple feature detectors that distinguish adequate cues that are largely



Bee Vision is Not Adapted for Pattern or Shape 171

independent of range, and useful for directional decisions at signposts along the route (Chapter 7, this volume).

The Same Few Cues Were Used Every Time Once a way was found for defining the test set for each pair of patterns that were discriminated, it was possible to discover exactly what the bees had learned in each case. The choice of tests was the result of a long history of progressive understanding of how bee vision works. Each example yielded the same general conclusions. They learned to ignore cues that were the same on both targets,

and they remember one or more simple cues in order of preference, but nothing about the pattern plan or shape. For each pair of patterns that was detected, the bees learned a selection from the same small repertoire of cues. When a new pair of patterns was substituted, the bees were obliged to learn the new situation. In each context, therefore, they could learn only one task, but in a different context, there would be other tasks. We might ask why they evolved this kind of vision: it is because their visual world is composed of vectors in the sky, range in all directions, and signposts along their routes. Changes in range are unavoidable for a mobile animal, and disrupts vision, so bees use cues that are relatively ­independent of range.

References Benard, J., Stach, S. and Giurfa, M. (2006) Categorization of visual stimuli in the honeybee Apis mellifera. Animal Cognition 9, 257–270. Chen, L., Zhang, S.W. and Srinivasan, M. (2003) Global perception in small brains: topological pattern recognition in honey bees. Proceedings of the National Academy of Science, USA 100, 6884–6889. Collett, T.S. and Zeil, J. (1998) Places and landmarks: an arthropod perspective. In: Healy, S. (ed.) Spatial Representation in Animals. Clarendon Press, Oxford, pp. 18–53. Dafni, A., Lehrer, M. and Kevan, P.G. (1997) Spatial flower parameters and insect spatial vision. Biological Reviews 72, 239–282. Efler, D. and Ronacher, B. (2000) Evidence against a retinotopic–template matching in honeybees. Vision Research 40, 3391–3403. Giurfa, M., Hammer, M., Stach, S., Stollhoff, N., Müller-Deisig, N. and Mizyrycki, C. (1999) Pattern learning by honeybees, conditioning procedure and recognition strategy. Animal Behaviour 57, 315–324. Guiraud, M., Roper, M. and Chittka, L. (2018) High-speed videography reveals how honeybees can turn a spatial concept learning task into a simple discrimination task by stereotyped flight movements and sequential inspection of pattern elements. Frontiers in Psychology. DOI: doi.org/10.3389/fpsyg. 2018.01347 Hertz, M. (1929) Die Organisation des optischen Feldes bei der Biene. Zeitschrift für vergleichende Physiologie 8, 693–748. Hertz, M. (1930) Die Organisation des optischen Feldes bei der Biene. Zeitschrift für vergleichende Physiologie 11, 107–145. Hertz, M. (1931) Die Organisation des optischen Feldes bei der Biene. Zeitschrift für vergleichende Physiologie 14, 629–674. Horridge, G.A. (1996) Pattern vision of the honeybee (Apis mellifera); the significance of the angle subtended by the target. Journal of Insect Physiology 42, 693–703. Horridge, G.A. (2000) Visual discrimination of radial cues by the honeybee (Apis mellifera). Journal of Insect Physiology 46, 629–645. Horridge, G.A. (2003a) Discrimination of single bars by the honeybee (Apis mellifera). Vision Research 43, 1257–1271. Horridge, G.A. (2003b) Visual discrimination by the honeybee (Apis mellifera), the position of the common centre as the cue. Physiological Entomology 28, 132–143. Horridge, G.A. (2006) Visual discrimination of spokes, sectors, and circles by the honeybee (Apis mellifera). Journal of Insect Physiology 52, 984–1003.

172

Chapter 8

Horridge, G.A. (2007) The preferences of the honeybee (Apis mellifera) for different visual cues during the learning process. Journal of Insect Physiology 53, 877–889. Horridge, G.A. (2009) What Does the Honeybee See? And How Do We Know? A Critique of Scientific Reason. ANU E Press, Canberra. Available at: http://epress.anu.edu.au/honeybee_citation.html (accessed 1 November 2018). Horridge, G.A., Zhang, S.W. and Lehrer, M. (1992) Bees can combine range and visual angle to estimate absolute size. Philosophical Transactions of the Royal Society of London B 337, 49–57. Jander, R., Fabritius, M. and Fabritius, M. (1970) Die Bedeutung von Gliederung und Kantenrichtung für die visuelle Formunterscheidung der Wespe Dolichovespula saxonica am Flugloch. Zeitschrift für Tierpsychologie 27, 881–893. Lehrer, M., Srinivasan, M.V., Zhang, S.W. and Horridge, G.A. (1988) Motion cues provide the bee’s visual world with a third dimension. Nature, London 332, 356–357. Lehrer, M., Horridge, G.A., Zhang, S.W. and Gadagkar, R. (1995) Shape vision in bees, innate preference for flower-like patterns. Philosophical Transactions of the Royal Society of London B 347, 123–137. Lubbock, J. (1881) Ants, Bees and Wasps. (13th edn 1898). Kegan Paul, London. Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222. Ronacher, B. and Duft, U. (1996) An image-matching mechanism describes a generalization task in honeybees. Journal of Comparative Physiology A 178, 803–812. Srinivasan, M.V., Zhang, S.W. and Rolfe, B. (1993) Is pattern vision in insects mediated by ‘cortical’ processing? Nature, London 362, 539–540. Srinivasan, M.V., Zhang, S.W. and Witney, K. (1994) Visual discrimination of pattern orientation by honeybees. Philosophical Transactions of the Royal Society of London B 343, 199–210. Stach, S. and Giurfa, M. (2005) The influence of training length on generalization of visual feature assemblies in honeybees. Behavioural Brain Research 161, 8–17. Stach, S., Benard, J. and Giurfa, M. (2004) Local feature assembling in visual pattern recognition and generalization in honeybees. Nature, London 429, 758–761. Turner, C.H. (1910) Experiments on color-vision of the honey-bee. Biological Bulletin, Wood’s Hole 19, 257–279. Turner, C.H. (1911) Experiments on pattern-vision of the honey-bee. Biological Bulletin, Wood’s Hole 21, 249–264. van Hateren, J.H., Srinivasan, M.V. and Wait, P.B. (1990) Pattern recognition in bees, orientation discrimination. Journal of Comparative Physiology A 167, 649–654. Wehner, R. (1967) Pattern recognition in bees. Nature, London 215, 1244–1248. Wehner, R. (1968) Die Bedeutung der Streifenbreite für die optische Winkelmessung der Biene (Apis mellifica). Zeitschrift für vergleichende Physiologie 58, 322–343. Wehner, R. (1981) Spatial vision in arthropods. In: Autrum, H. (ed.) Handbook of Sensory Physiology, Volume VII/6C: Vision in Invertebrates. Springer, Berlin, pp. pp. 287–616. Wolff, G. and Strausfeld, N.J. (2016) The insect brain: a commentated primer. In: Schmidt-Rhaesa, A., Harzsch, S. and Purschke, G. (eds) Structure and Evolution of Invertebrate Nervous Systems. Oxford University Press, Oxford, pp. 597–639. Zhang, S.W. and Horridge, G. (1992) Pattern recognition in bees, size of regions in spatial layout. Philosophical Transactions of the Royal Society of London B 337, 65–71. Zhang, S.W., Srinivasan, M.V. and Horridge, G.A. (1992) Pattern recognition in honeybees, local and global analysis. Proceedings of the Royal Society of London B 248, 55–61. Zhang, S.W., Srinivasan, M.V. and Collett, T. (1995) Convergent processing in honeybee vision, multiple channels for the recognition of shape. Proceedings of the National Academy of Sciences, USA 92, 3029–3031.

Chapter 9 The Visual Control of Flight

One problem is those politicians, columnists, commentators who distort science. Sometimes they refuse to name who finances their activities. (Nurse, 2015)

When you think about their tiny brain compared to the complexity of their appendages and size, variety and performance, it is amazing that most insects fly around and appear to see quite well. Bees behave as if they are on a mission, stabilize their flight in a gust of wind, dodge obstacles, and land on their food flower. This is piloting, as distinct from navigation, or homing (Chapter 10, this volume). Bees control their piloting visually and have also a variety of hair sensillae and internal mechanoreceptors of joints and muscles that we cannot ignore. Earlier chapters show the visual input for piloting is from green contrast in the environment to modulation of the green channels in the optic lobe, and all joint and muscle responses are continually modified by learning while in flight. We can distinguish several kinds of piloting tasks, and analyse some. In the 1950s, motion perception became the basis of insect vision, with a huge diversion of resources to the optomotor response, when an insect responds to the rotation of a drum placed around it by turning in the same direction (Fig. 9.1A). More recently

we have clear-cut examples of detailed analysis of visual behaviour in free flight, when insects turn to fly upwind against the optic flow of surrounding objects. Optic flow is the map of the perceived velocity caused by motion of the eye relative to the surrounding panorama, in angular coordinates. There has been much discussion, even acrimony, about how insects in nature perceive the three-dimensional world because several mechanisms always operate in parallel, and it is hard to demonstrate the relevance of a response in a laboratory experiment. Also, for obvious reasons, insect vision is narrowly dedicated to the actual tasks required in the ecological context, and these must be included in the design of analytical experiments.

Responses to Light In classifications of movements that bring insects to their preferred places, responses directed by the direction of the light were called taxes and those that were undirected were called kineses. The old term ‘tropism’ vaguely covers all these and is more convenient. To understand mechanisms, classifications and definitions are useless. Measurements of performance in natural situations are useful to determine visual

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)

173

174

Chapter 9

resolution or speed of response, but explanations end at the receptor layer. Consequently, most experiments on vision of living insects fail to connect with the neuron anatomy or physiology. Nowadays we aim for an experimental analysis of mechanisms. Some insects can fly in complete darkness, but most, like locusts and dragonflies, cannot because they need a cue for staying the right way up. Many use the general brightness of the sky as a sign of ‘up’, called the dorsal light response, and some use the direction of the light beams rather than the diffuse intensity of the sky. Insects that swim upside down have a reversed righting reflex. The UV of the sky is poorly reflected from natural objects, so UV comes almost entirely from above. A disturbed bee heads towards the brightest UV part of the sky to escape. Bees use the UV of the sky to stay the right way up in flight, and turn a forward somersault when they fly over a mirror that reflects UV upwards. Usually, blue and UV receptors are more abundant on the dorsal part of the compound eye. Many insects have three small eyes called ocelli at the top of the front of the head, commonly sensitive to UV. In the dragonfly they are detectors of the average position of the sky and they stabilize flight in dim light. In some species they are partially focused laterally, towards the horizon. The large lens apertures and the summation of ocelli receptors upon the neurons below account for the extraordinary sensitivity to the position of the sky and horizon, enabling them to fly at night. In the locust, the ocelli detect the horizon even in starlight. In the bee their functions and interactions with the compound eyes are unknown. Little binocular vision among insects Cats and primates have families of binocular neurons in the brain with a variety of angular offsets from the visual axes of the two eyes – so-called disparity units. Combinations of disparity neurons measure range even off the midline. Insects have two eyes but lack the convergence of eye movements or vertebrate binocular mechanisms.

Jumping spiders, some beetles and other insects have a tiered retina with receptors at different focal planes, so that they might crudely distinguish different object distances. Binocular overlap of two eyes that are fixed in position provide a mechanism for measurement of range in mantids, dragonfly larvae and a number of predatory insects with two widely separated eyes. So far as we know, the triangulation is done by coincidences of inputs from corresponding facets on the two eyes upon the central body in the brain, not by a congruent mapping of the spatial array of each eye into the opposite optic lobe. In mantids, dragonflies, and perhaps all insects, direction of the gaze is detected by hairs of the neck as the head turns. The honeybee worker, like many herbivores, has eyes that look to the side and downwards, with a small overlap in front that sees the food. The dorsal rim of the eye has a line of specialized ommatidia for detection of UV light that is polarized along particular axes, from which the bee can be aware of the position of the sun when it is behind clouds (Chapter 4, this volume). The most skilful fast-flying insects that catch prey in the air commonly have broad binocular overlap of axes at the front and top of the eyes but a short baseline between them, so they must measure range by moving in flight (see below). Honeybee workers have a curious way of hovering close to a target as if staring at it with both eyes, called fixation. This is unlikely to be binocular vision, but may stabilize the eye for better detection of retinotopic cues (i.e. fixed in position on the eye). Finally, there remain a vast number of insects with a little binocular overlap, but they may triangulate downwards over short distances to see what they stand on. Certainly, the frequent occurrence of nectar guides on flowers suggests that bees see between their feet.

The optomotor response A stationary insect that is hovering in flight or just standing still is sensitive to the positions of landmarks in the vicinity, especially



The Visual Control of Flight

vertical edges, and it turns its head or whole body in the same direction as an unexpected displacement of the whole visual field around it. This is called the optomotor response. The insect almost recovers its position relative to its surroundings. When disturbed in flight by an unexpected gust of wind, large day-flying insects recover their former flight posture visually relative to their surroundings, and other responses are active at the same time. However, the optomotor response is very slow. A hovering fly will rotate in the same direction as a drum that is rotated slowly around it, but will not oscillate faster than one per 10 s (0.1 Hz) if the drum is moved to and fro. Clearly, the optomotor system is only a part of flight control because the motion of the flying insect is in the opposite direction to the perceived flow field. In some old accounts, the optomotor response supposedly keeps a floating insect on station in a river in spite of the water current that would wash it downstream. In this situation, water beetles, and large aquatic bugs, like fish, face upstream and swim forwards, keeping the apparent angular velocity the same on the two sides. In a narrow channel, they do not follow the motion at their side and turn into it; they avoid the sides and stay in the centre of the optic flow. Freely flying insects turn against the perceived direction of the surroundings and so head upwind. Similarly, in a tunnel that has any visible texture on the insides of the walls, bees fly upwind along the centre. Working on migration flights of mosquitoes in the natural environment John Kennedy and others knew very well before 1940 that the optomotor response was not involved (Kennedy, 1940). In a surprisingly early measurement of resolution, optomotor responses of Drosophila to the movements of gratings of different periods were plotted by Lotte von Gavel (1939) at different light intensities. The interesting point was that, in bright light, as the period of the grating was reduced, the response reversed at the period of the interommatidial angle, 9°. The reversal was correctly explained as a Moiré effect

175

between the interommatidial angle and the grating (see Fig. 9.2D discussed later). The response beyond the reversal point showed that the actual resolution at the limit was better than that predicted by the interommatidial angle. The reversal showed that the fly perceived the best correlation between the adjacent facets with the shortest delay, and did not detect the bars or actual direction of movement of the drum. At low light levels the reversal occurred at a larger period of the bars, showing that the spatial tuning of the motion detectors had increased up to 20°. Even at this early date, there was sufficient data to show that insects detected the output of the motion detector as a vector without pattern, not the image of the bars. In an earlier study, to explain how the period at the reversal point could increase at low light levels, Hecht and Wolf (1929) had championed the improbable idea of a wide variety of receptor field sizes and sensitivities. Indeed, longer lateral interconnections between sub-adjacent visual axes were later found. Lotte and her family managed to escape to Australia, but after 1945 her pioneering effort was ignored by the numerous researchers on this topic. The next step was taken by Hassenstein (1951), who held a potato beetle firmly by the thorax with the head free to move. The beetle was surrounded by a large drum that could be rotated and variously illuminated, and held a paper toy in the form of an endless strip with four points where it could choose left or right (Fig. 9.1A, B). An inner drum with fixed vertical slots could be adjusted to deliver to selected ommatidia a series of flashes calibrated in intensity, angular space and time. The response was in the perceived direction of the stimulus. Hassenstein showed that two adjacent or sub-adjacent vertical rows of ommatidia were sufficient to detect motion, but that was the limit of the detector’s span. The response to motion was in the opposite direction when one vertical row was dimmed and the next was brightened, and was proportional to the square of the contrast of the input. Based on this, the response was inferred to be a multiplication between successive changes at adjacent ommatidia.

176

Chapter 9

(A)

(B)

Open loop

(C)

Fly

(D) wp

Closed loop

_

Torque

gF

w

T

y Open . . Closed loop

gK

w = perceived angular velocity wp = angular velocity of surrounds Open loop T = gF.wp

Closed loop

(E)

T = gF . w y = gK . T

w = wp – y

T=

gF. gKwp 1 + gF . gK

so T is always less than wp Fig. 9.1.  Three early set-ups for study of the optomotor response: (A) beetle walking on endless paper ribbon with choice points; (B) rotating drum and fixed perforated screen within it; (C) fly on a pivot is free to rotate while flying and turns by saccades. (D, E) Systems analysis: (D) with open or closed loop; (E) relations between components, all of which can be measured using open and closed visual feedback loop. gF and gK are the gains (amplifications) in the boxes; T, torque; w, perceived angular velocity; wp, angular velocity of surrounds; y, the signal in that bit of the interaction loop.

The beetle, being fixed, with eyes almost clamped still, could respond only by making the correct decision at the next choice point on its path (Fig. 9.1A). In the systems analysis diagram (Fig. 9.1D), the feedback loop is shown as ‘open’ because the response is simply that of the eyes and memory of the signal, with no movement of the head.

These ideas were absorbed into a new Max Planck group headed by Werner Reichardt (1961), with a shift to the housefly. The head was fixed to the body and clamped to the torque meter to measure the open loop relation between the turning force exerted by the flying fly (torque), and the motion stimulus. The idea was to restrict the input to that of the drum and prevent a feedback



The Visual Control of Flight

stimulus from the turning effort of the fly. However, the fly could not make a saccade or get any visual feedback from its own efforts. When so restrained, both fly and beetle tried to turn in the same direction as the drum. Data was fed into extensive mathematical calculations with the assumption that motion was detected by autocorrelation between adjacent facets in time and angular space, but was useless because the fly could not see the effects of its own efforts. Martin Heisenberg tried to reason with Reichardt, to no effect. When I visited Reichardt in the early 1980s, it was clear that he had an unshakeable belief in the relevance of his complex equations. He was drinking schnappes at 9.30 a.m. in the morning and dominating his splintering research team. Varju left early, Gotz and Buchner did splendid work within the orthodox limits set by the boss. Kirschfelt opted out and kept silent. Ernst Pick was working on the very point that Wolf had discovered, the spatial resolution of motion detection at low light intensity, and committed suicide in the laboratory. When my laboratory in Canberra was reviewed in the 1980s, the Review Committee went to Reichardt for an opinion of our work. From our start on the optomotor system, we had been casting doubt on his conclusions. Back in Tübingen, the same experiments done with honeybees in restrained flight with fixed head (Fig. 9.2A) gave similar results, but the bees soon reverted to a state of learned helplessness (Fig. 9.2B) and the data was then useless. The data for insects with fixed head was useless mainly because they were unable to move their head. This was just one of the confusions that was never admitted or resolved by Reichardt’s group. In this system’s analysis (Fig. 9.1D, E), inputs were in terms of the angular velocity at the eye although at the time it was also known that they responded to the rate of passing of contrasts across the eye (called contrast frequency), not the period of the stripes or angular velocity of the edges (Fig. 9.2A). The optomotor response in the bee is slow, with a latency of 40–50 ms, and is tuned to low temporal frequencies, rising

177

rapidly to a peak near 10 Hz, then falling again to zero near 100 Hz (Fig. 9.2C), so an ambiguous response can arise from two different stimulus situations. The response adapts to a steady motion and then responds afresh at each unexpected change in contrast frequency. The increase in interommatidial angle suggests light intensity was low (see Fig. 4.3). For all these reasons, the optomotor response cannot account for the way insects fly in a constant direction at a preferred speed and height over the ground or how they total their successive turns and remember the direction of home. Reichardt had founded his own journal to publish mathematical results from his own group. He was told firmly to recognize the saccades, but would not listen, so years of expensive effort ignored the main sensory input, the saccades, and the direction of the response. Reichardt’s (1961) model of a unit motion detector was no more than autocorrelation, a mathematical expression representing the detection of motion of a contrast that requires a non-linear step such as multiplication. The real mechanism is a neural circuit, such as that suggested by Barlow and Levick (1965). However, well into the next century, Reichardt’s model was regarded as useful for understanding nervous systems, especially by researchers on bigger brains. Even some of my own colleagues repeated it in reviews, as if it were validated. The best prospect at present is by serial sectioning the optic medulla of Drosophila to reveal regularly repeated neuron connections. A little known feature of the optomotor response is that the motion stimulus is not required. When an arthropod (bee, locust, shore crab) is held in a stationary position but free to move, it learns the relative positions of major vertical green contrasts in its surroundings. The light is then switched off, and the surroundings moved in the dark. When the light returns, the head rotates so that the contrasts return to the same parts of the eye as before. This response, called optokinetic memory, may be just a demonstration of retinotopic learning of landmarks, but it offers a hint about how these animals detect the motion of objects

178

Chapter 9

(A)

(B)

Bee with fixed head in helpless posture

(D)

Response

10

Bee and sensors suspended within the drum

(C) 100

*

Response (%)

20°

* 0 + 0.1

* +

*°

+

0 –5

* 60° ° 6.2° + 2.9°

5

* +

+ °

° * + + * ° Response is to ° *+ frequency of *° passing edges *+ *+ 1

5°

°

10

10°

20°

40°

Angular period of the stripes ° *+

° 100

(E)

Successive positions of drum

Temporal frequency of stripes (Hz)

Arrows show motion of drum bars

Arrows show motion detected by bee

Fig. 9.2.  Experiments with the optomotor response of the bee. (A) The Reichardt apparatus with a clamped insect with fixed head. One coil holds the insect in a fixed position, the other compensates for the insect’s torque responses. (B) Honeybee in learned helpless posture. (C) Response depends on frequency of passing edges, not stripe period. (D) Near the lower limit of resolution the zero response indicates the interommatidial angle (in this case much too large). (E) Origin of the reversal of the response, as in (D), when the eye facets are closer than the bars on the drum. (All data from Kunze, 1961.)



The Visual Control of Flight

around them (called optic flow). They respond to the change in angular position of vertical contrasting edges. The composite eye of many ommatidia could act as a theodolite that takes bearings on surrounding concentrations of receptor modulation. The eyes of a shore crab and locust (also dogs and maybe birds) can track the motion of the sun or moon across the sky at 15°/h, which is half the angular speed of the hour hand of a clock. This implies that they easily detect the motion of edges of shadows caused by the rotation of the earth. Perhaps this is why the optomotor response detects direction of very slow movements. When walking or foraging in free flight, ants, bees, crabs, and many other arthropods, remember the direction of home at all times, partly by integrating their turns and partly with the help of external cues from the direction of motion of shadows and the sun. Details of the crab optomotor responses will be found in my earlier work (Horridge, 1966, 1968). Finally, the control of different muscles in the optomotor response of insects is learned, like all short-term postural control of legs, wings, neck or antennae. There is a stabilizing mechanism with directionally sensitive neurons that respond to average motion across large regions of the eye. The output to accessory flight-steering muscles is modified by learning, so that insects can learn to recover when part of the flight mechanism is damaged.

The exact neural circuit for biological motion perception Motion perception has been intensively studied by probing with electrodes, to date with no clear solution because the cells of the optic medulla are small, and there are many parallel circuits and feedback loops. The good news is that serial sectioning of the whole optic lobe, that tracks every neuron of the fly Drosophila and every synapse, is progressing well. Already, repeated local circuits that could be motion detectors have been recognized. This tedious work, at Janina Research Institute, Washington, DC,

179

can reveal whether some synapses are inhibitory or excitatory, and may reveal the wiring diagram. Currently, maps are limited to a few column widths in the medulla because the field of view is limited.

Reverse engineering of motion perception Your computer has either a mouse or a finger pad that controls the cursor on the screen. If you turn over your mouse these days you will see a glowing eye on the underside. This eye detects the direction and distance moved by the mouse over a textured surface. By the time that the computer mouse arrived, there had been many years of research on insect eyes to discover motion-detecting circuits and how they worked. The earliest computer mice had a ball held between three or four wheels. As the ball rolled over a surface, the turning of the wheels drove the cursor over the screen. Nowadays, the sequence of responses in several photocells in the single eye of your mouse detects and transmits motion in any direction.

Saccades Saccades are spontaneous jerks of the eye at intervals of up to a few seconds, so small as to be scarcely noticeable, and were ignored until recently. In 1963 David Sandeman noticed that the crab Carcinus makes a saccade when a contrasting object comes suddenly into view. In 1975, Land gave the same name to sharp turns of the housefly and Drosophila, between short straight sections of flight or in restrained flight (Fig. 9.1C). Saccades serve essential purposes in active vision, seeing by doing and generating movement in order to get the feedback. New experiments by Heisenberg and Wolf in the 1970s used an apparatus similar to Reichardt’s, but the insect now saw the stimulus movement caused by saccades and its own effort to turn its head (torque) (Heisenberg, and Wolf, 1984). Although the head was fixed, the fly could see the effects of its own efforts and respond to them.

180

Chapter 9

Each saccade arouses the edge detectors and gives the insect a snapshot of the angular distribution of landmarks, and could calibrate the motion detectors by making a voluntary motion of a standard size, and renew the retinotopic location of every ­ ­contrast. In flies Drosophila and Musca in flight, a voluntary turn is initiated by making a spontaneous fast saccade towards the direction of the intended turn usually towards a contrast in view. The fly’s effort to make a single saccade appeared on the recording as a brief pulse of torque that was converted to a drum

rotation up to 60° in the opposite direction (Fig. 9.3B). The optomotor response is too slow to respond to the movement initiated by the saccade, but it holds the head in the new position and the body follows the head through the turn. In this action, the saccade is the unit of self-directed turning. On the other hand, an unexpected imposed rotation of the whole visual field by as little as 0.1° causes an optomotor response with 40–50 ms latency and high gain. After the saccade, straight flight is resumed with the aid of nearby landmarks. This action is easily seen when watching a fly.

Torque meter

Wind tunnel

Torque meter electronics

(A)

Drum

Oscilloscope

Fly

FM recorder Lens Pattern

Bias generator

Lamp

Rotating mirror

Waveform generator Flight response processor

Motor (C)

50

180°

Stripe position ϕ°

90

0

0 –90

Time (s) –50

0

5

10

15

20

25

–180

Stripe position (°)

Yaw torque (10–10 nm)

(B)

1s

Saccades cause the turns Fig. 9.3.  Experiments with a fly that is free to rotate. (A) Apparatus. The fly is heading towards a vertical black bar that is controlled by the fly’s own torque output or by a slow oscillator. (B) The fly turns on a continuous track in a series of small saccades. (C) The track of the fly if it were not clamped in the wind tunnel. At each set of saccades, the insect returns to its former direction. (Modified from Heisenberg and Wolf, 1990.)



The Visual Control of Flight

These discoveries were made long before our work on flight of the bee against the optic flow in tunnels, and nothing similar has been found in bees, which detect cues in flight by continuous scanning in the horizontal plane. The smooth tracking by the head of a mantis follows a moving fly on a featureless background, but when tracking a prey against a patterned background the mantis is obliged to make a series of saccades in order to override motion feedback from edges in the background (Rossel, 1979). To fixate on a prey, principal eyes of jumping spiders make saccades by moving the retina inside the head. In most insects, including the bee, saccades have not been studied. If one wing is damaged so that turning in flight is continual, a fly readjusts to the new relation between the eyes and the torque. The visual input generated by a saccade informs the fly within 50 ms whether there has been a change in position of contrasts in the visual field apart from that which was expected, and it adjusts the output to muscles. Any low contrast in the visual field is sufficient for visual stabilization

(A)

T

a

on a straight course. There is no cancellation of the optomotor response to allow for voluntary turning, because it is required to help compensate for damaged wings. Peering or scanning before a jump Every country child knows that grasshoppers jump, but few observe that before they jump they make a subtle peering movement of the head to one side, keeping the head pointing in a constant forward direction by a compensatory bend of the neck (Fig. 9.4). The scan causes nearer objects to move sideways through a greater angle than objects further away. As shown by shifting the target during the scan to mislead the insect, it measures the change in angle to assess how far to jump. Presumably, when it grows at each moult, it recalibrates by learning. Humans also move the head sideways, but estimate range from the way that nearer objects appear to move relative to more distant ones in the background. Without this aid, we are not very good at estimating range of isolated objects.

(B)

(C)

t=0 R*= R

h

Stationary target R* = R = h/tan(a)

181

g

t

t

R R*

h

Moving target to left (target recedes) R = h/tan(g) g < a R* < R

R* b

R h

Moving target to right (target approaches) R* = h/tan(b) b > a R* > R

Fig. 9.4.  A grasshopper makes a sideways scan, keeping constant the direction of looking. (A) Normal situation. The new angle at the eye is a measure of range R* equal to R the jump distance. (B) The target T is moved a distance t during the scan and the jump R is too far. (C) The target T is moved the other way during the scan and the jump is too short. (Derived from Sobel, 1990.)

182

Chapter 9

Following an odour plume When searching for water, a flower, a scent mark, dung pile, carcass, male’s pheromone or animal sweat, many insects, including bees, fly upwind towards the odour source by visual control using feedback from optic flow. When they begin to forage, bees may fly across the wind searching for an odour trail, then turn upwind against the ground motion and track the odour to its source. Odour receptors initiate the search but they are not directional. When the insect loses the scent, it casts about or scans from side to side under visual control until it finds the odour again and turns upwind.

Scanning while in flight Many kinds of insects that fly by day, including bees and blowflies, weave from side to side as they go, and make a sideways scan as they approach an obstacle or possible landing place. In a bee training apparatus scanning is easy to observe, but hard to record or measure quantitatively. I have seen grown men brought to tears trying to make sense of responses of a bee scanning in tethered flight in the apparatus above (Figs 9.2 and 9.3). As described below, bees measure range and height above ground from the optic flow induced by scanning in flight, and continually measure the induced angular velocity of surrounding contrasts to monitor their range.

Measurement of Range In 1977, just in time to win a footnote in Rudiger Wehner’s review in 1981, I used a 19th-century idea of Helmholz, and suggested that insects measure surrounding relative motion as they move (Horridge, 1977). At the time, others guessed that the apparent size of objects of known size was the key to measurement of range. Perhaps they were also right. In the mid-1980s I started working on the peering behaviour of juvenile praying

mantis as they walk. They actively measure range when they look out for the next foothold. The Australian garden mantis Tenodera was particularly well behaved; ordinary bodily movements such as swaying while walking were just as effective as peering by the head. Edges with green contrast generated relative angular motion that was inversely proportional to the range. It was the only input that could measure range. Moving the target just as it made a scan, caused the step to be too short or too long. Here was a guidance system for any freely moving vehicle. There were at least three other groups looking at similar topics. Aloimonos (1993) at the University of Maryland was interested in active visual detection of surroundings by motion of a spherical eye. At Würzburg, Germany, Heisenberg and Wolf (1984) had revised our understanding of the way that flies (Drosophila) use fast saccades to change direction in flight and remember the positions of surrounding objects (Fig 9.3). Starting in the late 1930s, John Kennedy found that mosquitoes, locusts, aphids and moths tended to fly upwind at constant ground speed irrespective of flight height or terrain, when following an odour plume (Kennedy, 1940). Swarms of mayflies, midges or bees obviously keep station, even in a wind. When Kennedy and his student, David (1979a, b, 1982), analysed Drosophila flight in a tunnel, by 1982 they found a constant angular velocity of image movement irrespective of stripe period in their surroundings, and constant ground speed irrespective of wind speed. At an unexpected step in contrast frequency, the insects adjusted immediately to keep a constant ground speed, but at a gradual constriction in the tunnel they reduced ground speed, as later found for the bee. That was about the time that I visited them at Silwood Park, Imperial College London. Later, when I started at a different end of the problem, with range measurement by bees, their results had a strong influence on our programme. In 1986, when Mandyam Srinivasan returned from Zürich, it was easy to interest him in active vision for measurement of range. Srini brought with him the technique of training honeybees, which would respond



The Visual Control of Flight

(A)

Training

Test

(B)

Water

183

Reward

5 cm 5 cm Fig. 9.5.  Discrimination of range looking down by freely flying bees, irrespective of angular size, absolute size or position. Experimental set-up: (A) black paper discs (‘flowers’) of different diameters on stalks of various heights; (B) the ‘flowers’ at different ranges on parallel sheets of Perspex. The black discs were shuffled in position and height during training and tests. (Redrawn from Horridge et al., 1992.)

to a variety of tests, so that we could discover what they actually detected. Srini also brought Miriam Lehrer, the best bee trainer in the world, on the first of many long visits from Zürich. First, they measured the resolution of bee discrimination of parallel gratings, published in 1988, for which we invented the Y-choice maze for training and testing flying bees. From about 1985 onwards, ANU was short of cash to pay existing staff, and there was increasing pressure against blue-skies research. Our university administration urged us to collaborate across disciplines and direct research to attract funding. We formed a collaboration between professors from Biological Sciences, Physics and the Medical School, named the Centre of Visual Sciences. The university gave us an extension to our building and some extra post-­ doctoral positions for collaborative projects. More important, we could operate independently and apply for our own funds on Visual Sciences notepaper. I had met Zhang Shaowu at the Academia Sinica in Beijing, and invited him to Canberra with the new money from the Centre. Shaowu did experiment after experiment, Miriam Lehrer provided the techniques of handling the bees, Srini did the mathematics and progressively took over the management. It was hard to beat a team composed of a survivor of the Cultural Revolution, a Swiss Israeli, a Tamil Brahmin from Yale and a Cambridge-educated Yorkshireman, working with highly trained bees in the perpetual Australian sunshine. We even trained bees to fly inside a rotating drum and learn to ignore it while they

s­elected a moving target inside the drum. Once we had broken into a new chapter of ideas there were endless new experiments to be done. I deliberately pushed the work and publishing into Srini’s hands because he would soon have to take over when I retired. My own work was diverted into bees’ perception of patterns and colour. Our first critical experiment on range was to train freely flying bees to go for a reward on a black paper flower standing on a stalk of a certain height, in preference to several other similar paper flowers of different diameters on stalks of different heights. The bees flew above the flowers, which were at different ranges below them (Fig 9.5A). The positions and sizes of the flowers were shuffled at intervals so the range was the only possible cue. The bees easily learned this task. It was our first example of randomizing the cues that the bees should not learn, while keeping constant those that the bees should notice. The black discs emitted no light; therefore bees detected relative motion of their edges with green contrast. Later, we put the flowers at different ranges on parallel sheets of Perspex (Fig 9.5B). The 1988 finding, that bees measure downward range, would apply to all directions (Lehrer et al., 1988). We tested that idea thoroughly, with objects at different positions around the bee.

Experimental Demonstration that Bees Detect Angular Velocity In the next critical experiment, bees were trained to fly along the channel while their

184

Chapter 9

flights were followed with a video camera. The bees flew along a line that kept the angular velocity of the walls the same on each side; this was called the centering response. They would hug a stationary wall (Fig 9.6) and not turn into the moving wall as they would if ruled by the optomotor response. The significant variable was angular velocity of nearby contrasts relative to the surroundings, not spatial frequency at the eye, or speed through the air or over the ground. The world of the bee is a great solid angle filled, not with objects as shapes or colours that humans see, but with measures of the range of each edge that displays green contrast. The motion of the animal itself generates sufficient information to measure range of all contrasting objects in the vicinity. Flying bees move away from any surface with relatively faster horizontal movement of contrasts in either direction. They measure angular velocity but not direction. When the surrounding contrasts are moved vertically upwards, flying bees also rise up higher, as if they compensate for an inferred loss of altitude. This is the same direction as the optomotor response, but they do not respond to surrounding falling motion of contrasts (Srinivasan et  al., 1993). With panoramic vision of range, object recognition is not essential for piloting. A basic equation describes the motion of the surroundings as a function of the forward velocity V, the angle on the eye ϕ, and the range X. The mechanism functions in all three dimensions. There are three degrees of freedom in forward, sideways and vertical

motion, and three in rotation of the bee, as seen by the bee in each direction, each derived from the basic equation of optic flow (Fig 9.7). Six equations define the system in terms of the angular velocity at each point on the eye. As a bonus for a foraging insect, the observed optic flow integrated over a flight journey gives a measure of the distance travelled. In one way it was obvious that the animal used information about angular velocity in this way. However, another powerful if not fully accepted view was that insect motion perception relies upon a measure of induced contrast frequency, not angular velocity (Fig 9.2C). There was a long-running argument between the bee workers and the fly partisans, but it was an error to suppose that all insects use the same visual mechanism for control of flight. In the ventral cord of the locust and the bee, there are descending premotor interneurons from the brain that signal angular velocity at the eye, not the contrast frequency. Von Frisch assumed that bees measure distance by the energy used on the outward journey. He confirmed this with measurement of sugar used on the flight, but his bees flying uphill or carrying small burdens reported longer distances because they flew lower and registered more optic flow, not because they were working harder. The experiments of Harald Esch and Srinivasan in the 1990s showed that freely flying honeybees measure the distance travelled over the ground by integration of the optic flow. We have to recognize that bees must find their way to the foraging sites and then back home to the hive, so they evolved the

Moving O

T

R

Stationary

Fig. 9.6.  Bees flying along a channel between a moving belt and a stationary one equalize the average angular velocity on the two sides. O, opening (i.e. entry point); R, reward; T, tunnel.



The Visual Control of Flight

185

x

V ϕ

_

Angular velocity dϕ V sinϕ =– dt x

Expanding flowfield Fig. 9.7.  The relation between forward ground speed V, angle at the eye ϕ, and the range X where dϕ/dt is the rate of change in angle with respect to time (in degrees/second).

only possible mechanism for a flying insect, measuring total optic flow. Also, they have to search to find the signposts, type of flower or a familiar landmark, so bees scan as they fly. On the other hand, as an example of a small homeless fly, Drosophila in straight flight keeps landmarks in constant positions on the eye as it seeks a familiar odour of food, and has no need to scan, but always keeps a constant height and ground speed. Insects are very diverse in their habits. Probably there exist many other types of flight control, for example, catching moving prey by dragonflies, and vertical scanning by butterflies, but at present we have detailed studies of only four: locust, fly, bee and running by the desert ant. The general principles are straightforward. Forwards or sideways motion of an insect that is stabilized against rotation in flight causes an induced relative angular velocity of surrounding objects that is inversely proportional to range (Figs 9.7 and Fig 9.8A, B). When the flying insect makes a lateral scan of known or predetermined amount, distant objects appear to move little (Fig 9.8B). When the flying insect sees a regular striped pattern of constant spatial ­frequency (the reciprocal of the stripe period) at the side, the perceived angular veloc­ ity is inversely proportional to the range (Figs 9.7, 9.8C).

Range R = forward velocity (m/s) / ­angular velocity of an edge (°/s) Or, time to contact (s) = 1 / angular rate of expansion (°/s) The bee does not have a measure of its ground speed or wind speed. Sensitive mechanoreceptors at the base of the antenna detect the bending of the first joint due to the air speed in flight. Similarly, hairs on the head detect air speed, but these are insufficient to report speed over the ground. For an eye looking at a textured background at the side, or directly down: Angular velocity = (contrast frequency) / (angular spatial frequency) These calculations do not give a measure of distance travelled. For regularly repeated edges, as in a grating, the angular spatial frequency is the reciprocal of the angular period as seen from the eye. This relation implies that the absolute value of the average period in the surrounding panorama is part of the calculation, but the bee in the air does not have a measure of it. Yet the bee does measure her perceived velocity over the ground and integrates it to make a useful measure of distance travelled, as shown by many experiments. In forward locomotion, the angular velocity of passing contrasts is a measure of

186

Chapter 9

(A)

(C) R θ

Range R V

Induced angular velocity –dθ V sin θ = = dt R V = velocity over the ground

(B)

-

θ –V

V

Angular spatial frequency = R/P when θ = 90° Induced contrast frequency = V/P Induced angular velocity = V/R

(D) R

θ

Period P -

Opening parallax

V

Induced angular velocity =

–dθ V cos θ = R dt

+V

Closing parallax Parallax

Fig. 9.8.  The relations between the range R, the velocity V, and the angle to the longitudinal axis θ, for (A) forward flight, (B) sideways scanning, (C) frequency of passing regular stripes. Angles are measured in radians. (D) Opening and closing parallax. dϕ/dt is the rate of change in angle with respect to time (in degrees/second).

range of nearby objects, and total optic flow is a measure of perceived distance travelled. If this is the mechanism, it explains why insects in flight keep flight height and ground speed constant so they can notice a warning of a sudden change.

responded to the velocity over a range of 40–1000°/s of the angular velocity of the input, irrespective of the temporal or spatial frequency. They responded over the same range that the bee encounters during free flight and landing.

Neurons that code for velocity of optic flow

Measurement of turning in dead reckoning

Neurons that code for the directional optomotor response are very prominent in electrophysiological studies of optic lobe neurons, but they are useless for explaining the measurement of optic flow because their responses depend on the temporal frequency of the passing of edges across the receptive field. In the bee, they were recorded in the ventral cord (Ibbotson and Goodman, 1990; Goodman et al., 1991; Ibbotson, 1991, 2001a, b). The significant detail is there were 12 out of the 96 neurons recorded that

An important part of angular dead reckoning is the memory of the retinotopic position of an outstanding contrast or landmark on the eye, so that if disturbed, the animal can turn itself until the landmarks return to the same position as before. Many insects that have a home to return to have this ability. When this performance was analysed in the crab Carcinus, it was found that the positions of edges and areas were detected separately, and that vertical black/white edges were not necessarily distinguished from white/black edges (Horridge, 1966).



The Visual Control of Flight

Several crabs show that they are aware of the direction of their burrow at all times when they are out of it, no matter how they turn as they move. The mechanism is the measurement of the present position of a modulation on the retina relative to the memory of its initial position, and is not the same as optic flow.

The Measurement of Distance Flown: the Odometer A flying bee cannot possibly measure the distance flown over the ground, in yards or metres. They evolved an odometer that measures distance flown from the totalled estimates of optic flow integrated over the time of flight, to give a perception of distance travelled based on the angular velocity of passing contrasts. This was first demonstrated by use of a tunnel with a random pattern on the inside of the walls. Bees were trained to visit a reward at a fixed place along the tunnel; then the search position was recorded with the reward removed (Fig 9.9A). They learn the distance on the way in, and do not fly at constant speed. If the tunnel is replaced by a wider one, they fly further because the angular velocity is less, and with a narrower tunnel they fly shorter (Fig 9.9B). The bees’ estimate of distance was independent of the spatial frequency of the pattern on the inside of the walls. A confused bee that cannot find the reward returns to the previous landmark and tries again, showing that bees learn segments of the trip, from landmark to landmark. There was no effect on the search position when a wind blew along the tunnel, showing it was not a measure of the effort. In a variety of other experiments, bees were trained to fly along a variety of tunnels up to 10 m long, loading the odometer with less or more perception of distance flown, and the tunnel end could be moved so that they emerged at an unexpected place. They learned the flow on the way out, and in all three dimensions, and can also learn it on the return home. Later, Tautz et al. (2004) compared the distance flown with the distance

187

indicated by the dance of bees returning over different distances over ground or water with differing spatial frequencies. One millisecond of waggles coded for every 18° of angular motion in the optic flow at the side of the eye. This result, of course, completely negates von Frisch’s conclusion that bees measured distance by the effort of flying or the amount of sugar used as fuel (see Chapter 10, this volume).

Opening and closing parallax Flying bees live in a world of motion and relative motion rather than one of spatial layout of objects and edges that humans rely upon. Nothing shows the dependence better than the landing approach to an edge. Bee edge detectors cannot detect a difference between the two sides of a boundary because they ignore their scan direction. We trained bees to come to a platform where we could arrange various kinds of parallax at the edge and observed where the bees landed (Fig 9.10). They responded to closing parallax but would not land where there was opening parallax, so relative motion had polarity, not just flicker (Fig 9.8D) (Srinivasan et al., 1990).

Contrast frequency contrast and velocity parallax Perception of range from relative motion of the eye and also against background is a primary feature of insect vision. Large motion perception neurons are abundant in insect optic lobes. They are commonly reported to respond to local motion seen against a background of large-field movement that is caused by the insect’s own active movement (Rowell, 1971; Rowell et al., 1977). A familiar human ability is to notice and follow motion of a small object such as a football while running. Bees seeking a familiar goal can easily land on a moving arm that moves against a moving background. Contrast is a measure of the change at a boundary;

188

Chapter 9

Reward

Score, relative frequency

(A)

With added extension 100 Train and test

50

0 –100

100

0

Extension Here bees fly in

200

300 cm

Dishes in tunnel 0

300 cm Reward

Score, relative frequency

(B)

Train and test

Narrower tunnel

100

Wider tunnel 50

0 0

100

200

300

Landing point in tunnel (cm) Fig. 9.9.  The odometer: the bee measurement of the distance flown. (A) Distribution of landing places when trained bees searched after the reward had been removed. Without the extension, the bees searched around the correct distance, but with the extension added after the training, they searched at the expected distance from the entrance. (B) When tested with a wider or narrower tunnel after training, they flew too far or too short. There is an error in the data in (B): the widths of the curves should differ because bees flew different distances. (Redrawn from Srinivasan et al., 1996.)

contrast frequency is the rate of repetition of the contrast, so contrast frequency contrast is a measure of the change at the boundary between two moving surfaces, sometimes called velocity parallax.

Using very simple preparations in studies of the reaching and stepping responses of young mantids in motion, and also by placing a bee feeder at an edge where one patterned surface moved over another, it



The Visual Control of Flight

(B) Disc

h

Landings within disc (%)

(A)

189

100

Control disc

50 0

5 3 2 1 0 0 5 Disc height above background h (cm)

Fig. 9.10.  Bees detect a camouflaged disc that lies over a camouflaged background by the active use of parallax. (A) The apparatus. (B) The response depends on the height of the disc above the background (h). (From Srinivasan et al., 1990.)

was obvious that these insects detected velocity parallax. In 1987, I concluded: as long as it is moving, the insect sees well enough to fly around, catch mates or prey, avoid enemies and tree branches, land on a target and so on, with no picture of the outside world projected into its head in the sense that we see in ourselves or in a camera. The potential of this type of visual system for robot vision is immeasurable. (Horridge, 1987)

This idea led us directly into making and operating gadgets for the visually impaired that led to flying vehicles with a computer on board.

Control of Flight Flight height In important work, in 1982, Kennedy and his student, David, found that flying Drosophila in a vertical tunnel have a constant flight speed and kept constant angular velocity of image movement on the eye irrespective of spatial frequency of patterns in view (i.e. constant optic flow) (David, 1982). Bees have a preferred flight speed, and a height above ground depending on their location and the wind speed that depends partly on what they have previously experienced and learned at the same place, for example they fly a low route in the shelter of a wall or hedge to avoid a headwind (Heran,

1956). On familiar journeys, they have learned the familiar optic flow below them, and use it to hold a preferred flight height (Horridge, 2009). When bees were trained to fly into a large featureless tunnel, 1 m high and 1 m wide, they flew low near a white floor, but at a height of about 50 cm over a patterned floor (Fig 9.11A, C), at a height depending on the spatial frequency but irrespective of the pattern. When the spatial frequency of the pattern below changed, bees adjusted their height to maintain the same optic flow as before (Fig 9.11E, F, G). Probably many insects behave similarly, using lateral or ventral vision as available in open spaces. They learned to fly higher or lower if they anticipated obstacles in the way, and they learned to follow at speed a stretched black tape (Fig 9.11B). Given a choice of reward holes at different heights, at first they followed the blank floor (Fig 9.11C) even when the the reward was in the top hole, but after a few visits they learned to fly nearer the height of the anticipated reward. They fly low against a headwind (Fig 9.11D). When they had thoroughly learned the correct height to fly over a coarse pattern on the floor (Fig 9.11E), they responded to a sudden change of pattern period below them (Fig 9.11F, G), but soon relearned the height of the reward. Honeybees cannot recruit to a food source that is high above the ground, so the dance apparently contains no information about altitude, but they can learn to fly high or low.

190

Chapter 9

End view

(A)

Floor

Plan view 200 cm

50 cm

(B) Reward box Plan view End view

(C)

Floor

Side view

(D)

Side view

Plan views

Gauze fan

Side views

(E)

(F)

(G)

Fig. 9.11.  Experiments on flight height. Bees enter the tunnel from the left. (A) Plan view and end view of an experimental tunnel. (B) They fly along a black tape and are diverted. (C) Bees fly close to a blank floor but anticipate the height of a reward. (D) They fly close to the floor against a headwind. (E) Bees were trained over a uniform contrast period on the floor. (F, G) When tested with a discontinuity of pattern period, they compensated by a change in flight height that reduced changes in modulation. (From Horridge, 2009.)

Control of flight speed In familiar situations, bees have a preferred speed of passing the panorama or over the ground. They fly at greater airspeed against a headwind, and when familiar with the terrain, but slowly in an unfamiliar place

and faster when certain of their route. Outdoors, they fly fast when high up and slower when closer to contrasts or when exploring. As a bee flies along, it adjusts the perceived optic flow by what it detects and by what it learned at that place. In the tunnel experiments, bees fly slowly b ­ ecause



The Visual Control of Flight

closer walls appear to pass faster, but by the time measurements are made the bees have learned the dimensions of the tunnel. When there is a constriction in the tunnel, the bees slow up as they pass through, but the perceived angular velocity remains approximately constant. Because continuous learning is involved and the internal state of the bee is unknown, it is difficult to discover how the preferred speed is decided. The sides of the eyes are dominant in measuring the optic flow as the bee flies along, but the optic flow has a distribution around the eye, slow at the front but faster at the sides and below. The directional effects also differ with the region of the eye, as lateral motion at the front causes turning, at the side it affects only the speed, and ventral motion may make the bee move higher or turn and fly upwind. In experimental unmanned planes and helicopters, it was sufficient to control landing by the optic flow in a solid angle of 50° looking forward and downward, and to prevent collisions via an eye with 360° vision looking towards the horizon. In a large tunnel with white walls, floor and ceiling with zero contrast frequency (a white out), bees refuse to fly; they walk. Besides the visual control via the optic flow, bees and flies have head hairs and a special mechanoreceptor system at the base of the antennae, both of which are sensitive to the speed through the air. How the three systems interact is not known.

191

range estimation, have all been described as contributory mechanisms. Expansion of the image on the retina is a sign of an approach to something. Stationary hovering in a wind is fine tuned by detecting and moving away from every centre of expansion, combined with keeping landmarks at fixed positions on the eye. Hovering to fixate on a contrast may improve resolution, or detection of cues and motion. Some dragonflies have more than one fovea. In some situations when insects hover, the image is fixated with deliberation by the fovea and this behaviour is somehow related to the improved discrimination of a mate, flower or prey. Almost always these insects ‘turn and look’ at a specific object. However, it has not been shown that they partition their visual world into separate objects. Male hoverflies and dragonflies, hovering in wait for a passing female, fly in exactly the correct direction to intercept her. The response is simpler than it appears at first sight, because the line of interception is selected on the assumption (perhaps learned) that the target is a female of standard size flying at a predictable speed. Bees use landmarks at the side, or the lateral parts of a target to locate a reward that is in front of their eyes, but they turn to centre their vision on a spot of blue colour, a vertical edge, a radial hub, a source of parallax or an expected cue.

Landing Hovering In several groups of insects, adept fliers hover in flight while they examine an object visually, feed from a flower, lie in wait, or guard a nest entrance. It is no more of an achievement than flying; exactly the same parallel mechanisms are in action, including learning each familiar situation. In various hovering insects, a suitably tuned optomotor response for stability against unexpected perturbations, fixation upon a target to stabilize the direction of looking, and a shift sideways, or a measure of target size for

As a flying or swimming bee approaches an object, it slows down long before the legs are extended for landing. When an object approaches in its path, a fly unfolds its forelegs, lowers the other legs, then extends the forelegs to break the shock at contact. The necessary visual input is any strong addition into the flow field at the front of the eyes. There appear to be preset fixed or learned templates for the initial stages of landing, triggered by looming and net darkening. Flying flies held by the thorax go through these responses repeatedly. The motion perception for the landing response

192

Chapter 9

is tuned to a higher contrast frequency like that for optic flow, unlike the optomotor response. In several vision mutants of Drosophila, either the landing or the optomotor response is lacking, suggesting separate systems. Selective habituation of each also suggests that they have separate motion detector pathways. The leg extension at landing is best stimulated by fast motion of a single edge or by a spatial period greater than 20°, whereas the optomotor response is most sensitive to slower motion all round. A honeybee coming in to land on a flat patterned surface detects the increased angular velocity of the ground beneath, and turns to land at right angles to an edge. High speed photography shows that the average bee keeps the average angular velocity of the surface constant as it approaches, until the ground speed is zero at touch down. One input, the perceived angular velocity and one output, the ground speed, are sufficient. For most of the way, the angular velocity is maintained at 400–600°/s, but there is great variability. How this translates into other situations, like landing on a thin twig, is not known.

Measure of Range as an Aid for the Blind From the start we realized that measurement of angular velocity would be useful for measurements of range by vehicles and mobile robots. Srini and I visited a number of manufacturers, but none of them were interested. One day a member of the Royal Society for Guide Dogs for the Blind arrived to find out what went on in our Centre for Visual Sciences. She was surprised to find that we worked on insects, but her visit led to me giving a talk at the Guide Dog Association in Melbourne in 1986, about the possibility of making visual aids for visually challenged people out of silicon chips, copy­ ing the principles that we had learned from insect vision. The Guide Dogs had recently brought out from England a remarkable electronics engineer with PhDs in psychology and

physics. This was Tony Heyes, a Cambridge graduate who had been blind and had worked on silicon chip technology for aids for the blind since he recovered partial sight. Almost at the same time, a Canberra newspaper notice announced that government funds would be available to support collaborative research between academics and any company that could use new technology. The Guide Dogs had the status of a company, and they sent a letter to say that they would lend Tony to our project for a fraction of his time. I went along to the Federal Department of Education, Research and Development in Canberra, and found the bureaucrat in charge of handing out the money. He seemed to have the stuff coming out of his ears but was unable to find any relevant files on the department computer network, which had 29 drives but no explanations. So he had to telephone his aides for every bit of information, and eventually gave me a form to fill. A couple of weeks later the University Treasurer rang me to say that a third of a million dollars had been sent to ANU for our project, to appoint two staff and two research students with expenses. It was not the only time we took the University Treasurer by surprise. With the funds, we hired Peter Sobey and Martin Nagle. Peter had just finished a PhD in engineering at Adelaide on the inspection of sawn timbers by computer vision; he did the programming and Srini did the mathematics. Martin had been working on light detectors in charged coupled device (CCD) camera systems at the Mount Stromlo Observatory; he did the optical inputs. A Polish immigrant, Jan Dalkzinski, from Sweden with a degree in medical technology, applied for a technician job, but I gave him a scholarship to build the gadgets for humans to wear. Gert Stange also transferred into the group. So we had a gang of gadgeteers, and they made gadgets. Srini, Miriam and Zhang also continued the work on the bee behaviour. The general idea was to make handy little rangefinders for blind people to wear and on vehicles to detect obstacles, by putting the equations for relative motion into silicon circuits. The camera gave a picture



The Visual Control of Flight

in black and white, but when the camera was moved, the relative motion was converted into a colour code for range. We could not find any company interested in making these gadgets, because the market is very small and most blind people in the world are too poor and not able to manage the technology. This was in 1986–1987, long before self-driving cars appeared. Several gadgets were made and steadily improved, eventually to be put on self-steering wheeled vehicles running along the floor, with vision of optic flow and a computer on board. To try to simulate the 360° vision of insects, Sobey and Nagle made a video camera directed at the point of a cone of polished reflecting metal, to give an all-round view. The picture was transformed geometrically to fit a flat screen. The shape of the cone was improved and eventually patented. I think this was the only patent that came out of the project, and it was a side issue.

The Aid for the Visually Impaired Becomes a Weapon of War At this point a strange coincidence happened. In 1987, the Russian nuclear power plant at Chernobyl exploded, and the Japanese Government was concerned in case a similar catastrophe happened to one of their 30 reactors in Japan (in 2004, they had 70 of them). So they offered tax breaks and cash to their big companies for the development of equipment, including seeing robots that would enter a hot radioactive reactor instead of men. As we later discovered, they had no idea how to make a mobile seeing system. They were trying to copy human vision, which was a hopeless task for a mobile robot in those days of limited computer power. They did what the Japanese have done before: they sent engineers with pocket cameras around the world to look at useful developments in centres of excellence. ­ Eventually, a group from Fujitsu Computer Company arrived in the ANU and was told that we existed.

193

I happened to be away at the time, and had left in charge Ian Morgan, who worked on chicken retina, and had no idea what we were doing. He showed the Japanese our laboratory and gadgets (but not how they worked), and convinced them that we, and of course they also, had found the Holy Grail. A few weeks later, an invitation arrived for four professors from ANU to visit the Fujitsu research labs in Japan and be entertained by the Fujitsu Company. The other three became unpleasantly ill after eating decayed lobster at a feast with Geisha ladies. I must admit that the Fujitsu supercomputer construction factory was impressive and they had a team of robots that played football, but they revealed nothing about simple visual systems. We all gave talks about our work, but like Tar-Baby in the Brer Rabbit story, I said nothing about relative motion or how our system worked. Some weeks later, the Deputy Vice-­ Chancellor of ANU, Ian Ross, received an offer of AUS$5 million from Fujitsu for our know-how. He called me to his office and asked me if our gadget worked! When I assured him that they had seen it working, after a few seconds of thought he suggested that AUS$5 million was not enough, and we should increase the stakes. At the time, nobody except ourselves realized that our whole story was known to any one-eyed sports player: that relative motion of the eye provides a measure of range. The simplest ideas are worth the most. I continued to say nothing. There was nothing we could patent that could not have been pirated. In due course, Fujitsu paid AUS$10 million and sent a couple of engineers to collect the good oil. ‘What!’ they said, when they saw it. ‘We paid 10 million for this!’ With a moving window of 20 by 20 pixels, their system could calculate optical flow in real time up to velocities of six pixels per frame. Visual Sciences received as much funding as we needed, and ANU General Revenue took AUS$9.6 million. When I inquired where it went, Ian Ross ‘thought it was usefully used’ in a new IT building. The Research School of Biological Sciences got nothing except fame, or perhaps umbrage.

194

Chapter 9

Soon after, Srini was approached by the US Air Force (USAF) to put insect range-­ finding vision, self-steering and our visual piloting controls into small helicopters and later into model aeroplanes. We started with miniature helicopters with a range of a few kilometres, powered by a diesel running on nitromethane. We hired a landing field and took on more staff, with great success. Later we turned to model aeroplanes that flew through a forest. However, the USA had deployed military drone helicopters starting in March 1968, in Vietnam, some with a remotely controlled reconnaissance camera, even night vision, that were piloted by hand controls from a moving jeep, and pictures were relayed back to base. Optic flow was used to measure height, range to surroundings and distance travelled. A satellite navigation system started only in 1989. In 1990, the first American drone with flight control by a computer on board appeared, and in 1992 there was publicity about them. Miniature models were marketed; Black Widow, a MAV (military air vehicle) weighing 56 g, flew with armed payload in 1999. According to DARPA (Defence Advanced Research Projects Agency), 87 nations had military drones by 2014 (Jacobson, 2015). Research on robot vision by copying the vision of honeybees brought into ANU about AUS$20 million of grants, but it is doubtful if we actually contributed anything new or useful. The innovative ideas about optic flow had been initiated long ago by G.C. Grindley in a classified wartime study of landing aircraft at Farnborough, England, in the 1940s (Mollon, 1997). The gist of it was stolen by an American spy, who was seconded there, and who later published a popular book (Gibson, 1950) that made little impact. When I observed range measurement by mantids and locusts, I realized that a computation of range from angular velocity at the eye was possible, and deliberately appointed Srini to help and eventually take over the project. The opening up of the bee system happened naturally as we discovered bit by bit that bees were ideal for the analysis. Because the response at each point on the eye depends on the filter properties in vision, forward

velocity, and range of the surroundings, the angular velocity involves six fast-changing variables (three in roll, turn and climb; three in translatory motions) at each point on the eye. Simple! Fujitsu put a team of 20 engineers on the project and made a black box, called Ishtar, which simply speeded up the computation by parallel processing (Sobey et  al., 1992). Sadly, when a nuclear power station was hit by a tsunami, the Japanese were not ready. Fujitsu put vision of range into a few mobile robots and sold a few, but they recovered value in another way. The equations of the processing system could be applied in reverse to make virtual reality. In vision by relative motion, eye movement gave the correct ranges of surrounding objects. In virtual reality, the relative positions are known from the start and the apparent motion is calculated and presented on the screen. So we contributed to the computer games industry. Years later, at a vision conference at Bäckaskog royal castle, Sweden, I found myself sitting next to the USAF official from Washington who had funded our project. When I asked how the Americans had discovered us in a month or so after Fujitsu, he politely hinted that their office in Tokyo could write a cheque for a million dollars for industrial secrets. Lest you go away believing that bee vision was actually useful, I must add that once you had the idea and mathematics of optic flow, which was already published by a German nobleman, Hermann von Helmholz in the 19th century, bee vision was not much use to the military. Bees have negligible inertia and their system reacts very quickly, but inertia is the main problem with all fast or heavy vehicles, making the rapid change of direction or speed impossible. Also, when designing stand-alone vehicles with a computer on board, there are now so many sensors and gadgets available (night vision, GPS (Global Positioning System), accelerometer, indicators for rate of climb, airspeed, gravity, direction, altitude, time to contact and powerful memory) that the product is nothing like a bee. In my opinion, based on what



The Visual Control of Flight

actually happened, bee research was an expensive decoy meant to waste the time and attention of the other side. The research that we could do in Canberra had all been completed by 1992, when I retired and fortunately changed my topic to principles of insect pattern and colour vision, to be published 25 years later (Chapters 5–8, this volume). We had talked about our work overseas at IEEE (Institute of Electrical and Electronics Engineers) and DFG (Deutsche Forschungs Gemeinschaft) conferences, Peter Sobey and Srini published in Journals of Robotics. Soon, there were efforts in Sussex, San Francisco, Zurich, Marseilles and elsewhere to copy insect vision into simple machines. Sandini in Milan called his The Beebot, though he never worked on bees. In 1993, Srini was invited to give a talk at the Australian Defense Research Establishment at Adelaide, and from there he secured more funding to continue the robot vision work. Later, he was elected to the Royal Society, awarded a

195

distinguished professorship, and had his portrait in the National Portrait Gallery of Australia. None of this research could have been planned or explained beforehand in a grant application. All the money was given on the understanding that trained researchers would spend it as well as they could, on what appeared to them the best way forward. No questions were asked, and no explanations of what we did were ever offered. Later, we realized that no one in government or university management was aware at the time that we had a sensitive military project in a biological laboratory researching harmless bee vision. Innovation was achieved by bringing together a team of extremely well-prepared scientists of several disciplines, with technicians highly skilled in optical, camera, silicon chip and programming technologies, in a well-equipped research laboratory, and funding was sufficient but never overflowing. It was great fun based on an idea from insect vision.

References Aloimonos, Y. (1993) Active Perception. Erlbaum, Hillsdale, New Jersey. Barlow, H.B. and Levick, W.R. (1965) The mechanism of directionally selective units in rabbit’s retina. Journal of Physiology 178, 477–504. David, C.T. (1979a) Height control by free-flying Drosophila. Physiological Entomology 4, 209–216. David, C.T. (1979b) Optomotor control of speed and height by free-flying Drosophila. Journal of Experimental Biology 82, 389–392. David, C.T. (1982) Compensation for height in the control of ground speed by Drosophila in a new ‘barber’s pole’ wind tunnel. Journal of Comparative Physiology 147, 485–493. Esch, H.E. and Burns, J.E. (1995) Distance estimation by foraging honeybees. Journal of Experimental Biology 199, 155–162. Gibson, J.J. (1950) The Perception of the Visual World. Houghton Mifflin, Boston, Massachusetts. Goodman, L.J., Ibbotson, M.R. and Bidwell, N.J. (1991) Spatial, temporal and directional properties of motion-sensitive visual neurons in the honeybee. In: Goodman, L.J. and Fischer, R.C. (eds) (1991) The Behaviour and Psychology of Bees. CAB International, Wallingford, UK, pp. 203–226. Hassenstein, B. (1951) Ommatidienraster und afferente Bewegungsintegration. Zeitschrift für vergleichende Physiologie 33, 301–326. Hecht, S. and Wolf, E. (1929) The visual acuity of the honeybee. Journal of General Physiology 12, 727–760. Heisenberg, M. and Wolf, R. (1984) Vision in Drosophila: Genetics of Microbehavior. Springer, Berlin. Heisenberg, M. and Wolf, R. (1990) Visual control of straight flight in Drosophila melanogaster. Journal of Comparative Physiology A 167, 269–283. Heran, H. (1956) Ein Beitrage zur Frage nach der Wahrnehmungsgrundlage der Entfernungsweisung der Bienen. Zeitschrift für vergleichende Physiologie 42, 168–218. Horridge, G.A. (1966) Study of a system as illustrated by the optokinetic response. Symposia of the Society for Experimental Biology 20, 179–198. Horridge, G.A. (1968) Interneurons: Their Origin, Action, Specificity, Growth, and Plasticity. Freeman, San Francisco, California, 436 pp.

196

Chapter 9

Horridge, G.A. (1977) Insects which turn and look. Endeavour [new series] 1, 7–17. Horridge, G.A. (1987) The evolution of visual processing and the construction of seeing systems. Proceedings of the Royal Society of London B 220, 279–292. Horridge, G.A. (2009) What Does the Honeybee See? And How Do We Know? A Critique of Scientific Reason. ANU E Press, Canberra, 360 pp. Available at: http://epress.anu.edu.au/honeybee_citation.html (accessed 1 November 2018). Horridge, G.A., Zhang, S.W. and Lehrer, M. (1992) Bees can combine range and visual angle to estimate ­absolute size. Philosophical Transactions of the Royal Society of London B 337, 49–57. Ibbotson, M.R. (1991) A motion-sensitive visual descending interneurone in Apis mellifera monitoring translatory flowfields in the horizontal plane. Journal of Experimental Biology 157, 573–577. Ibbotson, M.R. (2001a) Characterising temporal delay filters in biological motion detectors. Vision Research 41, 2311–2318. Ibbotson, M.R. (2001b) Evidence for velocity-tuned motion-sensitive descending neurons in the honeybee. Proceedings of the Royal Society of London Series B, Biological Sciences 268, 2195–2201. Ibbotson, M.R. and Goodman, L.J. (1990) Response characteristics of four wide-field motion-sensitive descending interneurones in Apis mellifera. Journal of Experimental Biology 148, 255–279. Jacobsen, A. (2015) The Pentagon’s Brain. Little, Brown & Co., New York. Kennedy, J.S. (1940) The visual responses of flying mosquitoes. Proceedings of the Zoological Society of London A 109, 221–242. Kunze, P. (1961) Untersuchungen des Bewegungssehens fixiert fliegender Bienen. Zeitschrift für vergleichende Physiologie 44, 656–684. Land, M.F. (1975) Head movements and fly vision. In: Horridge, G.A. (ed.) The Compound Eye and Vision of Insects. Oxford University Press, Oxford, pp. 469–489. Lehrer, M., Srinivasan, M.V., Zhang, S.W. and Horridge, G.A. (1988) Motion cues provide the bee’s visual world with a third dimension. Nature, London 332, 356–357. Mollon, J.D. (1997) ‘On the basis of velocity clues alone’: some perceptual themes. Quarterly Journal of Experimental Psychology 50, 859–878. Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222. Reichardt, W. (1961) Autocorrelation, a principle for evaluation of sensory information by the central nervous system. In: Rosenblith, W.A. (ed.) Principles of Sensory Communication. Wiley, New York, pp. 303–317. Rossel, S. (1979) Regional differences in photoreceptor performance in the eye of the praying mantis. Journal of Comparative Physiology A 131, 95–112. Rowell, C.H.F. (1971) The orthopteran descending movement-detector (DMD) neurons, a characterization and review. Zeitschrift für vergleichende Physiologie 73, 167–194. Rowell, C.H.F., O’Shea, M. and William, J.L.D. (1977) The neuronal basis of a sensory analyser; the acridid movement detector system. I. The preference for small-field stimuli. Journal of Experimental Biology 68, 157–185. Sobel, E.C. (1990) The locust’s use of motion parallax to measure distance. Journal of Comparative Physiology 167, 579–588. Sobey, P., Sasaki, S., Nagle, M., Toriu, T. and Srinivasan, M.V. (1992) A hardware system for computing image velocity in real time. Proceedings of SPIE – The International Society for Optical Engineering 1823, 334–341. Srinivasan, M.V. and Lehrer, M. (1988) Spatial acuity of honeybee vision, and its spectral properties. Journal of Comparative Physiology A 162, 159–172. Srinivasan, M.V., Lehrer, M. and Horridge, G.A. (1990) Visual figure–ground discrimination in the honeybee, the role of motion parallax at boundaries. Proceedings of the Royal Society of London Series B 238, 331–350. Srinivasan, M.V., Zhang, S.W. and Chandrashekara, K. (1993) Evidence for two distinct movement-detecting mechanisms in insect vision. Naturwissenschaften 80, 38–41. Srinivasan, M.V., Zhang, S.W., Lehrer, M. and Collett, T.S. (1996) Honeybee navigation en route to the goal, visual flight control and odometry. Journal of Experimental Biology 199, 237–244. Tautz, J., Zhang, S.W., Spaethe, J., Brockman, A., Si, A. and Srinivasan, M.V. (2004) Honeybee odometry: performance in varying natural terrain. PLoS Biology 2, 915–923. von Gavel, L. (1939) Die kritische Streifenbreite als Mass für die Sehschärfe bei Drosophila melanogaster. Zeitschrift für vergleichende Physiologie 27, 80–135. Wehner, R. (1981) Spatial vision in arthropods. In: Autrum, H. (ed.) Handbook of Sensory Physiology, Volume VII/ Part 6C: Vision in Invertebrates. Springer, Berlin, pp. 287–616.

Chapter 10 The Route to the Goal and Back Again

Serial offenders who continue to distort science repeatedly . . . need to be counted robustly, because if they are not, they will undermine the whole scientific endeavour. (Nurse, 2015)

Finding the way to a familiar feeding ground and back again for shelter is understandably widespread among many groups of animals. From the most primitive insects, the bristletails to the most highly evolved colonial Hymenoptera, bees and wasps, we find many examples, such as cockroaches, dragonflies, crickets and some butterflies, that have territories where they learn routes, landmarks, hiding places and food sources. Horseshoe crabs walk to the coast where they lay eggs; spiders know their webs; crustaceans that live in burrows and emerge to forage, like many crabs, have a mechanism that indicates the direction of home. For all of them, whether their vision is poor or excellent, it also involves great numbers of odour receptors and a suitable memory centre. Arthropods that forage and return to a home have brain lobes with thousands of small neurons and dense regions of dendrites (mushroom or ellipsoid bodies) that look like memory banks. It is important to stress that these lobes have strong olfactory inputs; they are nothing like the regular array in an optic lobe or visual cortex. The honeybee has several connections from the antennal (olfactory) lobes into visual centres in the brain

but, unfortunately, we know almost nothing about them. It is now clear, however, that we will not explain memory of the outward route and home again by visual cues alone. Bees have long been a convenient example of foraging and homing because they are observable and available worldwide. They learn easily and relearn directions and distances up to a few kilometres. Experiments to illustrate this behaviour require little more than some sugar and a few dishes, but to explain the mechanisms requires a lot of thought. The first few weeks of life of an adult worker are spent in the darkness of the hive, nursing eggs and larvae, cleaning, transporting and feeding, even making wax honeycomb, all in the dark. When a worker bee first ventures outside, all the local varieties of pollen and nectar will already be familiar as odours. In learning to forage, however, the world of vision has to be learned, mostly by trial and error, and quite rapidly because the visual mechanism is anatomically formed, if not yet programmed. Odour continues to play a large part in finding food and water. Therefore, my story is no longer an analysis of peripheral visual inputs in terms of simple cues alone; it has become the behaviour of the whole brain. As a result, we will have to change terminology and designs of experiments. We move from critical tests of trained bees in a

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)

197

198

Chapter 10

completely controlled box, to a limited variety of manipulations out in the open, during which we can observe bees’ performance while engaged in route finding. The data is animal behaviour, not physiology, so mechanisms have moved from receptors and peripheral neurons to the whole animal, interactions with other bees, and the environment. Inevitably, some unsubstantiated inferences replace logical deductions. Before the days of radar and satellite navigation, ocean travellers had many ways to navigate: (i) by recognition of landmarks on a map; (ii) by compass directions; (iii) by dead reckoning which sums distances and directions noted in a log book; (iv) by observations of currents; (v) by sampling depth, temperature and the sea bed; and (vi) by finding a position and direction from the sun and stars, with a few extras such as the direction of ocean swell and the smell of land. Insects also have several interacting navigational mechanisms: • • • • •

cues from odours; a measure of distance over the ground; direction relative to the pattern of UV in the sky; the sun’s position; and recognition of places or landmarks.

Less certain mechanisms are the earth’s magnetic field and following other bees’ odour. This rich flow of sensory inputs requires an efficient, discrimination, learning and forgetting system, while circumventing obstacles with the general direction maintained, distance measured, and irrelevant things ignored. Before these inputs had been identified by laborious observation and experiment, it was an impossible task to explain bee navigation. Even then, in early studies with success attributed to one cause at a time, the bees’ alternative use of several mechanisms in parallel led to unnecessary controversy.

Early Observations Interest in this topic is very old. Aristotle observed that bees dance, and thought it was a wake-up call, and that the aroused

bees followed the dancer back to the food supply. Around 1880, the famous entomologist J.H. Fabre took marked mason bees in a box a few kilometres from their nest and noted which way they disappeared. Although 20 out of about 40 set off in the right direction, less than half found their way home, and the rest were lost. Fabre, an optimist, inferred that bees have an inner sense of direction! Darwin suggested that they used the earth’s magnetic field, but experiments with little magnets on bees failed. In a beautiful experiment, a physiologist with a critical mind, George Romanes (1885), took a hive of bees to his seaside holiday house. When the hive was opened, the bees explored the garden and all returned in the evening to the hive. The hive was then closed and the next day it was taken at most 250 m across a stretch of empty sand without landmarks, and again opened. This time every bee that emerged never returned. It is an excellent example of the power of an experiment in which the bees fail in a test. Romanes knew nothing about the sun compass of the bees. He correctly concluded that they have no mysterious sense of direction that would bring them home, that they rely upon local landmarks, and that they learn quickly. Modern critics also conclude that the bees did not compensate for the wind blowing over the featureless sand, and the sky compass was of little use because the bees had no reference landmark for the hive. In the next decade, Albrecht Bethe (1898), an indefatigable experimentalist, moved a hive sideways and put another in its place. The bees went to the new hive but soon came out again, then flew in again and out again, over and over. Eventually, a few bees that landed at the entrance of the new hive started a procession that walked home. When their hive was moved back by 2 m, a huge swarm of returning bees congregated in the air where the entrance would have been. When the hive was replaced, the whole swarm poured in. Bethe concluded that the bees could not recognize the hive visually, and were not guided by the sense of smell or hearing. To show that the bees are not guided by colour, he left large pieces of coloured cloth and paper near the hive or



The Route to the Goal and Back Again 199

covering it, and changed the colours, but the bees could still find their way. He cut down a large tree near the hive, but with little effect on their homing ability. These results showed that the bees did not have a picture memory of the scene or their destination. He then took marked bees from the hive up to 3 km away. When released, most headed in the correct direction and returned home. From all this, Bethe wrongly concluded that a totally unknown force guided the bees. The age-old way to find a hive of wild bees is to catch one, feed it with honey until full. Then, with a magnetic compass, note the ‘beeline’ it takes when released. Next, at another place, release another satiated bee. The hive will be found near the place where the two beelines cross. Bethe explained the beeline by the directing power of his new force but was stumped when he released some bees far away, finding that they flew high into the air and returned to the box where they were released. Of course, these bees had been moved to a place with no familiar landmarks, and started to make exploratory flights. Many others have since confirmed these observations, but it was performance that was studied, not how the bees succeeded. Researchers observed the newly emerged bees point towards the hive as they flew in exploratory circles above it, as if to keep it as a reference point. Forel (1908) concluded that they familiarized themselves quickly with the appearance and direction of landmarks. Their ability to learn and be adaptable was a stumbling point; it made bees look rational. If they were not rational, bees would never adjust their daily tasks to the changes of the natural world about them; so it was said. Even then, with almost no data, empiricists and intuitionists argued whether bees were automata or rational. Experimental transport of the hive to a new site always showed that visual exploration was essential before emerging experienced or juvenile bees could find their way back to the hive (von Buttel-Reepen, 1900). In 1905, Forel made experiments on a garden table where he ate breakfast in summer, and bees discovered marmalade, recently

introduced to Switzerland by English mountaineers. When the sweet stuff was placed under covers, bees quickly learned to be versatile in their search. However, it would be another 50 years before analysis began, and a century before we understood that honeybee vision could be rigid, and recognize at only one place, or adaptable and exploratory at another. In the century that followed, arguments that had started with Fabre, a creationist, Romanes a physiologist and intuitionist, and Bethe, a physician and mechanist, remained unresolved in France, Germany, the USA, Switzerland and eventually Australia. It is of interest that these arguments made little progress. It is amazing that so little was noticed by scientists who had been doing experiments for a century or so. There were many anomalies they could have noticed, many simple experiments they might have tried, but for a century the partial explanations were repeated. Having several methods of navigation available simultaneously means that experimental analysis must be carefully designed to avoid ambiguity. In hindsight, researchers were misled by the bees’ ability to switch between odours and vision, and between different cues, but progress had to wait for new techniques and new physical principles to be discovered. A lack of new and thoughtful observations was also a limiting factor.

Dead Reckoning and Alternative Inputs By the beginning of the 20th century, it was known that ants use visual landmarks (Fig. 10.1B), make directional odour trails, and have at least one sense that detects direction. In a classical experiment, Pièron (1904) allowed ants of the genus Messor on their way to the nest to walk on a piece of paper. With the ants on it unaware of the move, he moved the paper sideways. The ants continued exactly along the former direction (Fig. 10.1A), ignoring the shift of any odour trail. When they had walked for the distance that had previously brought them to their nest, they searched for the entrance

200

Chapter 10

(A)

Compass o

Train

(B)

Landmarks X X X

X X oX

Test

(D) Compensates o o

Test

Train

Distance o

Train Test

(F) Path integration o

(E) Size o

(C)

o Homeward

Train

Test

Train Test

(Fig. 10.1C). They clearly knew the remaining distance, and ignored landmarks, but their cue for the direction was unknown. Later work by Santschi (1911) showed that the sun’s direction was one factor, but there were others as described below. Many arthropods remember a record of their recent path. As they progress, bees and ants integrate every turn and distance, so they have a continual measure of the direction and distance of home, but forget the previous part of the calculation. Many animals, including humans, do it even in the dark, but we have no idea what the mechanism can be. Experienced bees and desert ants set out from the nest with an internal representation of the direction and distance to their goal. As they travel, they check their direction by the sun and landmarks, and measure the distance covered over the ground visually (bees) or by counting footsteps (ants). After foraging, bees fly directly home along a beeline that may be new to them (Fig. 10.1F). They can fly to their hive by dead reckoning, but must use landmarks if in a wind. Without considering the dead reckoning, many have claimed that bees have an internal map. Having deviated around an unexpected obstacle, desert ants and bees take a new direction directly towards the nest (Fig. 10.1D). Experienced bees cut out turns and shorten their path, and at all times know the

Outward

Fig. 10.1.  Diversity of options for ant and bee navigation. (A) When the path is shifted sideways, the sun compass retains the direction. (B) If the reward is shifted, the bees search in the place indicated by the landmarks. (C) They search for the reward after going the expected distance. (D) They compensate after being deflected by an obstruction. (E) They search for the reward at the place where the landmark has the expected angular size. (F) The beeline home.

direction of the hive or the feeding place. They continually update their direction from the sky compass (Wehner and Rossel, 1985; Wehner et al., 1996) and learn the relations between their path and the landmarks, particularly at the ends of the path (Collett and Rees, 1997). Their direction can be changed by an internal rotation of 180° to reverse the path either to food or to home. While returning home, some ants use vision to remember sufficient for the next trip, others use odour trails. In dead reckoning, they measure the angles through which they turn from the rotation of contrasts in the surrounding panorama, irrespective of landmarks or of the compass. The fly Drosophila flies in shade and is not known to learn landmarks. However, a Drosophila placed in a uniformly striped drum can learn to face one way when the light is blue and turn at right angles when the light is green. The motivation to learn is provided by temperature, which is controlled by the orientation of the fly. The regular stripes allow the fly to measure the angles through which it turns but they provide no fixed landmarks. When the colour of the light changes, the fly turns through the appropriate angle relative to the pattern, as if it integrates its angular velocity relative to the drum and at all times remembers its direction relative to its visual surroundings. Probably many insects do the same.



The Route to the Goal and Back Again 201

An important part of angular dead reckoning is the memory of the retinotopic position of an outstanding contrast or landmark on the eye, so that if disturbed, the animal can rotate itself until the landmarks return to the same position as before. When this performance was analysed in the crab Carcinus, it was found that the positions of edges and areas were detected separately, and that vertical black/white edges were not distinguished from white/black edges (Horridge, 1966), so the signal is the modulation at edges or position of black areas, not the pattern. This is a simple form of the visual recognition of a place with a memory of the retinal positions of two or more cues (Figs 6.2A, B; 7.4B, C, D; 7.8H). Dead reckoning is continually updated and previous versions are lost as new information is incorporated. We have no reason to suppose that the bee memory in dead reckoning is any more sophisticated than that. Memory of landmarks and vector paths relying on the sky compass is quite different from dead reckoning, which remembers only the latest direction of the journey and homeward. When using familiar vector directions, landmarks, signposts and distances, bees can learn how to head for any one of a number of different destinations (Wehner and Menzel, 1990).

The Motivational State of the Bee Researchers were often puzzled by conflicting results before it was realized that bees must be in the appropriate motivational state for study. Primarily, they must be forager bees, and known to be experienced or otherwise. Those caught at the nest entrance on their way out are motivated to take their accustomed route. Conversely, those caught fully satiated at the foraging place are motivated to head homewards. Only when they are captured as they arrive home have they no preferred flight direction. Not surprisingly, when bees in flight are caught, then displaced and released, they continue in the same direction as before, relying mainly on their sky compass and ignoring landmarks unless they realize they are lost.

Bees presented with a choice may hover and look. If they fail to find the reward when arriving at the expected place, they will hunt about for it (Fig. 10.1B, C). Bees are motivated to search and learn by trial and error only when they meet an unexpected difficulty or when presented with alternatives, one of which has the expected reward. Most learning involves active participation. However, bees fixed in the opening of a small tube learn passively in a single trial that a particular odour is associated with a reward, and they extend their mouthparts. With colour, similarly fixed bees learn only blue versus not blue, because green modulation from relative motion is lacking.

Landmarks By definition, a landmark cannot be recognized by its position; it must be recognized uniquely by its shape, size or colour, and then it indicates a position. Bees fly towards an obvious isolated landmark that they have learned, pass it, then on to another, and so to the goal, then directly back home. If they lose the route, they return to the previous local landmark and make their approach again from there. Some landmarks are like beacons upon which they rely. With these, if the landmark or the feeder is moved, the bees show that they have not learned the compass direction, and, not finding the goal, they cast around until they find it. Then they calculate a new compass direction towards home and remember the additional information needed to correct the error that they made. On overcast days when no sun or blue sky is visible, and dead reckoning is not much use in a strong wind, bees must detect familiar features in the local panorama, as they do inside mazes, tunnels and tents. In a strong wind, they hug the ground to take a more sheltered route. The orthodox view assumes that large and small landmarks are important guides for bees and wasps to find their foraging place and return home. However, the observations are of performance of the foraging insects, and the landmarks used are rarely identified, but are assumed

202

Chapter 10

to be there. In the only example that comes to mind, Bethe (1898) cut down a large prominent tree that he guessed was a landmark, with no effect. When a hive of bees is taken to an unfamiliar site, the emerging bees explore the immediate vicinity. If individual exploring bees are removed from the hive during this process, their ability to return depends on the number of hours that they have explored. As days pass, their familiar area expands. In his short textbook, Rabaud (1928) described the use of landmarks and of the position of the sun by visual ants, but made no mention of the dance of the honeybee, because at that time von Frisch had described the round dance as a mechanism for alerting the recruits to look for nearby food sources by their odour. Rabaud quotes results from Romanes, von Buttel-Reepen and others who inferred, but did not prove, that the bees learned the visual appearance of landmarks and updated their memory in successive journeys. This was the state of the art when von Frisch was struggling to understand the relation between the details of the dance and the direction and distance taken by the follower bees. As he found, landmark memories include directional vectors, as shown by directional dances on overcast days. There is abundant evidence, from artificial landmarks in tunnels, mazes, tents or open featureless fields, that every familiar landmark alerts the bee to the direction and distance to the next landmark. The track is therefore a chain of measured vector stages between landmarks, as well as a total vector and distance. In addition, dead reckoning helps. It is worth rereading his monographs (von Frisch, 1965, 1967a, 1971), to appreciate his enormous effort. Many observers noticed that shifting a single prominent landmark may have no effect, and concluded that the bees (and wasps) memorized the view of several conspicuous objects around the goal, particularly those on the skyline, and made the best visual match of the whole scene. I think there is little evidence of this complex and distant view. A few vertical edges are sufficient. If experimental landmarks are moved

while bees are in the hive, at the next flight under an overcast sky they follow the landmarks and go in the wrong direction, but still get rewarded. They return home by the direct route by dead reckoning but they have inferred the wrong direction of the sun from the displaced landmarks, so on arrival they dance as if the sun had been moved, and their recruits are misled. With a clear sky, they would use the sky compass. The memories of bees in a hive can be activated by an odour of a food source. When this happens, they exit the hive, fly in a remembered direction to the food source, or fly across the prevailing wind until they pick up the odour, then they fly up wind and search. The homing ability was broadened by Baerends (1941, 1959), who marked individual female digger wasps (Ammophila) that carried caterpillars back to their nests. A female could have four to six nests, with an egg in each, so that she was obliged to visit them in turn until they were filled and sealed. The wasps were familiar with landmarks over a large territory, and when holding a caterpillar and carried in a box to another place, they had no difficulty in taking the direct route to whichever of the nests was the former goal. It was 60 years before a similar ability of the honeybee to remember the routes to several foraging places was demonstrated, and even then only after the introduction of new radar responder technology that recorded the tracks of individual bees. In recent decades, many more researchers contributed interesting indications of the way that bees use landmarks. First, bees learn along tracks they use, not the whole surrounding district. If satiated bees are removed from a reward to the north of a hive and carried an equal distance to the south of the hive, they take a long time to get home, or get lost. If the goal is marked by a fixed retinotopic cue, such as the angle between two landmarks and their height, bees will respond to nothing else. On the other hand, bees can be trained to go to search around the hive and find a reward even if it is 10 m away in a direction that is changed every 10 min. They will search only when they have been trained to search with a shuffled cue or target position,



The Route to the Goal and Back Again 203

or when the only cue is an odour. If their hive is moved, they lose it, even though they learned the whole area. The performance and number of available routes depends strongly on what was previously learned, making the memory of the landscape an elastic concept. Caution is required because bees are extremely sensitive to a whiff of the odour from the Nasonov gland, with which other bees mark a food source, and to small differences in the distribution of light, sky polarization and brightness. However, they are easily trained to ignore this odour (or any other). When forced, they learn to use remarkably small landmarks, down to a single black spot. When the sky compass and all landmarks are removed, there is still behaviour suggesting sensitivity to the earth’s magnetic field. When bees occupy a new hive they orient new combs in the direction familiar to them, using the earth’s magnetic field (De Jong, 1982), and this sense also helps to keep combs parallel. Behavioural experiments show that flying bees detect the direction of the earth’s field. Sense organs in the abdomen contain magnetite (Lambinet et al., 2017).

When given a single stripe at the side of the target as a marker, however, they locate the correct level, and enter the hole on this level at the correct range from the stripe. When the indicator stripe is moved in a test, the bees return to the hole indicated by the new stripe. In locating themselves relative to a stripe, the bees measure range from relative motion, not from apparent stripe size, and transfer the information to both eyes. This use of markers in the peripheral field is colour blind and done by the green channel of receptors, like vision of motion and range. When using the Y-choice apparatus (Fig. 10.4A), the narrow entrance hole excludes naïve recruits, and even trained ­ bees will not fly through a familiar hole after it has been reduced in size. In 1995, I trained bees to pass between two black bars at each side on the baffle. Bees were trained with vertical bars in one arm of the apparatus but at an angle in the other. Left and right sides of the apparatus were interchanged every 10 min and there was no other cue. The bees quickly learned to fly between the bars at the rewarded orientation, and detected a difference of 15°.

Signs Along the Route Recently, I discovered that bees in flight are very sensitive to left/right asymmetry, which acts as a signpost that indicates ‘turn left’ or ‘turn right’ on a route (Fig. 10.2). In my Y-choice maze (Fig. 10.4A) the choice they make is the final left/right choice on their route (Chapter 7, this volume). Bees flying towards their goal don’t only look ahead. Miriam Lehrer demonstrated that bees locate themselves vertically relative to markers in their lateral vision when they are presented with a spatially complicated set of choices ahead of them (Lehrer, 1990). The bees were trained to look for a reward in one of 89 holes in a round target (Fig. 10.3A). When presented with this target  alone, they saw the holes with the front part of the eye but were unable to remember which region of the target to search, so entered the holes at random.

The Route is Divided into Stages Between Choice Points Many researchers in the field have noted how bees seem to treat the journey in stages. In a Y-choice maze, with baffles in the normal position (Fig. 10.4A), bees were trained to discriminate between a vertical and a horizontal coarse grating (Fig. 10.4B) on the targets, and then tested with the grating replaced by two vertical or horizontal black bars on each side of the baffles (Fig. 10.4C). Although this task looks simple to us, the trained bees were unable to use the orientation cues of the bars, because they were not at the expected choice point. However, the bees were easily trained to use the oriented bars alone (Fig. 10.4D). In fact, they preferred to learn the bars at the sides rather than the gratings in front. When trained with gratings and bars in place (Fig. 10.4E)

204

Chapter 10

(A)

(E)

(B)

(F)

(C)

(G)

(D)

Fig. 10.2.  Symbolic artificial route signs with polarity. Yellow on black stimulates green modulation. (A–C) Polarity of blue and green contrast. (D) Design for a route sign. (E–G) Asymmetry of green receptor modulation.

they learned nothing about the gratings (Fig. 10.4F). In each case, they used the simplest cue (Horridge, 1996).

Maze Learning by Bees Ants find their way through a maze by trial and error, assisted by directional odour marks that signal direction like an arrow on the road. Walking bees have to negotiate through their hive. They can search for a route to a reward and then return for more. A maze is an apparatus to reveal the performance, and only recently used to find out where the bees look, what cues they recognize, and how long they remember them.

Flying bees can use an odour or a visual cue to tell them which way to turn at a choice point in a maze (Weiss, 1953). When the cues were identified in the 1990s, Zhang Shaowu started an investigation of maze flying by bees using boxes that could be arranged side by side on the floor, with holes communicating between neighbours. There was only one correct way through a series of boxes (Fig. 10.5). Fresh boxes were used in tests to prevent the use of odour cues. Bees learn to fly though the hole marked by the correct cue, which can be an instruction to turn right or left at the following choice. At the end of the maze, they escape by a back door. In a maze, bees learn by trial and error to follow a trail marked by a constant cue



The Route to the Goal and Back Again 205

0

(A)

45°

Bar 90°

Disc

–90°

135° (B) 0 90°

Bar

1m

45°

135° 180°

Fig. 10.3.  (A) An array of holes, only one of which leads to the reward. (B) The position of the horizontal black bar at the side indicates the correct row of holes. Bees find the correct hole by learning its range from the bar. (After Lehrer, 1990.)

and then use the same cue in a different maze. Bees trained to use one colour can switch to another colour even in an unfamiliar maze because they have learned ‘any colour’ not ‘this colour’. If the maze has a fixed route, the bees soon learn to negotiate it correctly, even after all cues are removed. Cues can be indirect: for example, the bee can learn to turn right when the back wall is green but left when the back wall is blue, and bees trained to do this can thread novel mazes guided at each choice by the same colour cues, but not by different colours. Bees learn an unmarked maze slowly. They can also learn to choose to turn left and right alternately (Fig. 10.5A), or always to the left (Fig. 10.5B). Of course, they do not detect colours but odours, measurements of blue content, and edge modulation (Chapter 5, this volume). Maze learning by bees shows that they learn to turn through a given angle at that place and go a certain distance, and they can learn a sequence of choices. There is no reason to suppose that this ability is restricted to mazes. The experiments in which bees take a definite track through space show that

angular orientation can follow a learned motor sequence, as demonstrated by the persistence of turns in the track when all obstacles are removed. When a bee cannot find the next correct choice on a route, she goes back to the previous remembered choice point and starts again to relearn from there. Bees can also learn to turn left or right when one colour is placed before the entrance to a choice chamber, so they remember what they have just seen. In other experiments, a colour cue was placed on one side of a narrow tunnel through which the bees must walk, yellow to turn one way and blue to turn the other. The trained bees were tested with the cues on the opposite wall of the tunnel, and it was shown that in this situation the bees could transfer a colour cue from one side to the other (Zhang et al., 1998). The experiments became more sophisticated. A randomly changed cue, A or B, was exhibited outside the entrance to each choice chamber. The bees must look at the cue and then inside the choice chamber they must choose the hole with the same cue. In other experiments, other bees must not choose the same cue. Finally, one of two cues (e.g. horizontal or vertical bars) was displayed over the first hole as an instruction how to make the next choice (e.g. between blue and green) (Fig. 10.5D). The correct choice, in this case blue, then instructed the bees how to make the next choice, between circles and sectors. The correct choices are in the left column in Fig. 10.5(E). The trained bees were able to use a series of cues starting at any point in the sequence. When lost, they return to the last choice point they remember. In all these experiments, the mazes were fixed and the bees were successful, so it is hard to say what the bees detected and remembered because they were not tested. It was all performance, not analysis. The bees detect a cue and act on it. The bee learns the minimal cue. For example, it learns that there is a colour, not the colour, just as, in experiments with patterns, it learns less than the whole pattern. There is no case for inferring that the bees learned a ‘concept’ of sameness or difference because they were not tested to see what they had actually learned. Where the

206

Chapter 10

(A)

(B) Reward hole

Air

Train and test on targets alone 100% 45°

Target 27 cm

85.4%, n = 201 (C)

27

Baffle

cm

Choice chamber

49.6%, n = 55

Bees fly in here (E)

100%

Train and test with both

(D) Train and test on bars alone

100%

Test with bars alone

100%

74.9%, n = 172 (F)

100%

Test on targets alone

75.8%, n= 87

49.9%, n = 109 Fig. 10.4.  Orientation cues in front and at the side, in a two-stage maze. (A) The Y-choice apparatus with bars on the baffles. (B) The bees discriminate the gratings alone. (C) Bees trained on the gratings fail when tested with the bars. (D) They discriminate when trained on the bars alone or (E) on both bars and targets. (F) When trained on bars and gratings together, they had no need to learn the targets. Percentage values indicate the percentage of bees visiting reward holes. (Horridge, 1996.)

Bees fly in (A)

(C)

B G

Bees fly in (B)

Reward (D)

B

G

Successive choices

Fig. 10.5.  Mazes. (A) Zig-zag maze composed of cylinders standing on end. (B) A turn-left maze. (C) A maze for successive alternate choices. (D) Pairs of cues for successive choices in (C). B, blue; G, green. (Zhang et al., 1998.)



The Route to the Goal and Back Again 207

performance looks remarkable at first sight, analysis reveals that the bees learned a cue that was just adequate for the next choice, as in Fig. 10.4, not a general solution that would suggest some kind of insight. These experiments revealed an amazing sequential memory, but scarcely began to analyse what the bees actually detected.

The First Orientation Flights The first flight of a young bee from the hive is an orientation flight in larger and larger loops up and away, returning in a few minutes. Older bees that emerge from their hive to an unfamiliar scene make a new orientation flight. If carried away from the hive, bees that have taken only one orientation flight usually return, but those that had no orientation flight always get lost. However, without an orientation flight, newly mated queens must return to their own hive or be killed. After several orientation flights, some landmarks can be distinguished by colour, height, angular size, range, orientation and by their angular directions relative to each other, but not by compass direction from the point of choice. This implies that landmarks near the hive do not have an attached homeward direction, as they do when at a distance. When a hive is moved, all the bees in it must make new orientation flights. Experienced bees learn the direction of the hive from any point in their area. When the sky is clear they progressively shift their reliance from landmarks to the sun compass and optic flow over the ground. On overcast days, landmarks take precedence, and if the landmarks move, bees are either fooled or make new orientation flights. The orientation flight is a performance, not a mechanism. How their several navigation systems are coordinated has not been analysed.

Turn Back and Look Flights On leaving a nest or food site, bees turn back and look (TBL) to acquire or update

their memory of the location, as described by Turner (1908). He says: When a bee had discovered one of my honey producing artefacts and collected therefrom, it would make a flight of orientation and then fly home. […] After the association had been well established, the bees usually departed for home without making a careful flight of orientation. If however I had made a marked change in the position [my emphasis] of the artefact since the last visit of the bee, then a careful flight of orientation was always made. (Turner, 1908)

Bees leaving a new food source back away and swing from side to side in flight, along a series of successive short arcs, first looking towards the food source, then turning towards possible landmarks (Fig. 10.6A). They probably learn a succession of landmark sizes, directions and ranges that they use in reverse order to return to the food. Ground speed increases as the bee backs away. They do not zoom and loom. Flat shadows that have no motion parallax have no influence. Parallax may be a crucial cue. The relative locations and ranges of nearby landmarks are learned during one manoeuvre, but there is insufficient time to learn the pattern around the goal. At later visits, these actions progressively disappear (Fig. 10.6C, D). If the TBL manoeuvre is prevented, the size of the target becomes the cue (Lehrer, 1993). When the approaching bee centres its vision on a symmetrical target or on the colour of the goal, it can learn its size, elementary pattern and colour on the approach and then depart without making a TBL manoeuvre, but the bee can also learn the shape or colour of the goal in a TBL if the cue is put in place just before the bee leaves. If shown one colour, shape or size of target on arrival, and another on leaving, they retain the first cue only, so the TBL is not primarily for learning colour, shape or size, but for range, landmark size and depth. There is no evidence that the area around the goal is compared with a retinotopic memory. In the Y-maze (Fig. 10.4A) bees rarely make TBL manoeuvres showing that they already know where they are. Details of a TBL are more obvious in wasps than in bees.

208

Chapter 10

(A)

(C)

(B)

(D)

Fig. 10.6.  The turn-back-and-look (TBL) flight of the honeybee. (A) Arrows show the directions of looking. (B) The same track without arrows. (C) Track at a later visit. (D) Tracks when the bees are familiar with the place. (After Lehrer, 1993.)

Navigating with One Eye There are strange discrepancies between investigations of this topic, as if overlap of the visual fields of the two eyes was ignored. Early work showed that the ant Cataglyphis can navigate by the polarization of the sky or by landmarks with only one eye and can then find their way home with that eye. When trained to home with one eye and then tested on the other eye they can navigate by the polarization of the sky, but not by using landmarks, so landmark recognition is ipsilateral (Wehner and Müller, 1985). Similarly, bees can learn a colour or an orientation cue when presented to one eye but cannot transfer when tested on the other eye, and bees use the position of a laterally placed bar (Fig. 10.3B) but they do not recognize it

with the other eye. However, they can transfer relative motion of the cue and the eye between the two eyes, and also a colour when it is a cue for the direction of the next turn. The experiments were descriptions of the performance, not an analysis of mechanisms. Naïve bees trained to measure a distance along a tunnel with one eye can transfer the information to the other eye. They can also be trained to measure one distance along a tunnel with one eye and a different distance with the other eye. When one wall of the tunnel was blank and the other displayed a pattern, the bees estimated the distance of the reward along the tunnel almost as well as when both walls were patterned. There remains some uncertainty about how the eyes interact in the natural situation, because bees turn round for the return flight.



The Route to the Goal and Back Again 209

The celestial compass in ants Ants use the sun as well as landmarks to provide a reference direction. Species with large eyes can be expected to use landmarks, the sun compass, dead reckoning and odour trails. Between 1900 and 1920, Felix Santschi (1872–1940) described several examples and showed that the tracks were deflected when he changed the perceived position of the sun with a mirror. The best example, however, is the desert ant Cataglyphis. Santschi (1911) showed that when placed at the bottom of a featureless drum, this ant could use a small patch of the blue sky to take a direct route homewards. A ground glass screen spoiled the performance, but Santschi did not know it was polarization in the sky. Extensive analysis of this splendid animal in recent times by Wehner and his group at Zürich has shown that its navigation mechanism is very similar to that of the honeybee.

Exploration Around the Hive and the Round Dance Three millennia ago, Aristotle, and presumably many beekeepers since, thought that the dancing forager led recruits back to the food. From a famous little book by M. Maeterlinck, von Frisch already knew that this was not true. For some years, he believed that odours carried by returned foragers recruited new foragers, as described in Russian literature and known to beekeepers. However, while at Rostock in 1917, he discovered the round dance, and that marked bees would return to a dish containing sugar water that had no odour, even after their glands that secrete the attractant marker substance had been sealed. In 1923, he described the round dance (Fig. 10.7A) as an arousal signal for nectar, and showed that the dance of returned foragers excited recruits to go out and search nearby for food having the same odour. At the time, he missed the sun compass and thought that the figure of eight dance signified pollen. In 1944–1945, von Frisch observed that bees returning from an identified distant

field of red clover faced in a direction related to the positions of the sun and the field when they danced. There was no question of odour in this observation. After numerous experiments, he discovered how the bees used the sun as a compass to fix the direction on the earth’s surface; the dancing bees made an adjustment according to the time of day. They recognized the sun as a bright light free from UV polarization. Von Frisch described how the returned forager dances on the vertical comb in darkness in the hive or on a flat surface outside, in either a circle or a figure of eight double loop, with the straight piece in the centre giving the direction. On a horizontal surface, the direction of the central line in the figure eight points to the food source. On a vertical surface, the angle between the central line and the vertical is the angle between the direction of the food and that of the sun (Fig. 10.7C). On a vertical surface, hairs that are bent by the sagging at the neck joint detect the direction of gravity. Von Frisch (1965, 1967b) also discovered that the distance to the food source is somehow conveyed by the dance. Bees flying uphill signalled a greater distance and he concluded that they measured the effort used. This appeared to be confirmed when the blood sugar was measured. However, these conclusions were all upset when it was discovered that bees measure the total of the optic flow over the outward route (Chapter 9, this volume).

The Waggle Dance In 1996, David Sandeman, working with Tautz in the bee facility founded by Lindauer in Würtzburg, discovered that the dance on the surface of the comb is not a figure of eight, nor do other bees follow, as described in orthodox accounts (Tautz et al., 1996). In the darkness of the hive, the dancer faces in the appropriate direction but actually takes only one step while waggling her abdomen from side to side. The further away the food source, the longer the waggles last, and the more waggles there are. With a source 1200 m

210

(A)

Chapter 10

(C)

Sun

(B)

G G Hive

G G

Fig. 10.7.  The dance of the returning forager. (A) The round dance. (B) Potential recruits stand closely around the dancer, while others stand within the area of vibrations of the comb. The arrow shows the direction of the waggle of the dancer’s abdomen. (C) The direction of the central bar in the figure of eight relative to gravity (G) on the vertical surface indicates the direction of the food source relative to the direction of the sun.

away, this step takes 1.2–1.8 s to cover 8 mm. While waggling, the bee may then take another step, then run around and start again. In the hive, the dance is not seen in the dark. Number and duration of waggles and the direction are conveyed by vibrations through the comb and the legs of the follower bees, not by vision or following the dancer. In most groups of insects, there are examples of drumming or conveying signals by vibration in the substrate. In other situations, however, dancers make longer directional runs of a few centimetres over hard surfaces, or even over backs of other bees. Films of these dances in the light can be found on the World Wide Web.

The control exerted by the message in the dance is well illustrated when bees swarm. Scout bees return to the swarm and dance on any available surface, indicating direction and distance of a possible new home. The whole swarm moves only when all dancers agree. Foraging dancers indicate the direction used on the homeward flight, together with the total optic flow of the outward flight, and these two components are learned separately. For example, a segment of the outward flight along a narrow tunnel, which increases the optic flow for a given length of flight, increases the apparent distance to the goal, but has no effect on the direction of the



The Route to the Goal and Back Again 211

homeward flight. The bee sees its surroundings passing by, integrates the angular velocity over the whole length of the outward flight, and in the dance reports only the final total. This is exactly what the other bees need in order to fly for the right distance, as long as they take the same direction and have similar preferences for flight height and avoidance of obstacles. Odours are not essential for success. Of course, recruits require the sun or blue sky to follow the vector in the dance for the optic flow indicated, and then may need the odour from the dancer to discover the actual place. We must remember that experimenters use a small point source with sugar as a reward, but bees usually have a whole field as their target. Experienced bees can recall the whole track from the odour of the dancer alone. If new to the track, newly recruited bees quickly learn landmark vectors and their separation distances. Experienced bees may switch between the memory of the total vector and the expected landmarks, so omission of a landmark may have no effect. They can also ignore landmarks and take a beeline home. This ability to switch cues delayed the discovery of the cues. The calculation of direction allows for the movement of the sun, which is in opposite directions in the northern and southern hemispheres. The direction of compensation is learned once in a bee’s lifetime, and they get no second chance. When transported across the equator by air, experienced bees do not learn to compensate for the reversed direction, and get lost. A transported hive survives because young bees learn the task as they emerge. There are two components to navigation by the sun compass, one innate and one learned. Lindauer found that bees that had seen the sun only in the afternoons could immediately use the position of the sun if released in the morning, so they have an innate expectation where to find it in relation to the time of day. Bees are genetically programmed to expect the sun to be at a constant position in the eastern sky in the morning and at another position in the western sky in the afternoon, although the azimuth position of the sun actually rotates from east to west at an

average of 15°/h. Inexperienced bees, shut in a box make predictable errors when released. Around noon, bees come to training sites less frequently. To exclude the sky compass, experiments must be done indoors or under heavy overcast skies, as had been done by accident for a century or so, forcing the bees to use landmarks. Bees can learn a revised relation between the track of the sun and a new landscape (Towne, 2008). Recruits are influenced by the odour of the food source on the dancer. From earlier Russian work, Frisch knew that, even without a dancer, bees could be induced to leave the hive and search in the correct places when the odour of familiar flowers was blown into the hive. Frisch published quite a lot about this, and it was considered useful for directing the bees towards crops for pollination. This effort, and dealing with an outbreak of the bee disease nosema, may have preserved his life in the late 1930s.

The Bee Wars There were two parts to this skirmish. One part was an endless but useless discussion of whether the dance was in reality a language, as Frisch had suggested. The bees, being practical, were not interested in such matters. Soon after Frisch’s publications of the figure of eight dance, a fierce controversy broke out when an experienced Californian beekeeper and junior academic, Adrian Wenner (1967), reported that bees were able to locate the food source although exposed only to its odour in the hive, and that the bees that followed the dance were unable to locate the food source without assistance from its odour. Presumably, the former were familiar with the locality, the latter were not. Wenner found it impossible to publish until after the Nobel Prize was awarded to von Frisch, Lorenz and Tinbergen in December 1973, and was infuriated because his publications were important for obtaining tenure (Wells and Wenner, 1973). Of course, many had known about the odours before, but the design of Frisch’s experiments over many years was so varied and persistent, that he

212

Chapter 10

should be given the credit; in particular, he used sugar, which has no odour, and he did demonstrate the sky compass, the dance and a measure of distance. However, his account was definitely faulty in some details. Later, Harald Esch, one of his former students banished to the USA, showed that bees do not dance until they have returned with several loads, and that follower bees did not respond at once, but often attended half a dozen dances before they foraged (Esch et al., 2001). Even then, they do not necessarily fly in the indicated direction, but fly across the wind until they pick up the food odour and then track it upwind, as they would with no dance instructions. Others have shown that when the distance of the reward is suddenly changed, the dancers take some time to adjust their dance. Every decade from 1970 on, Adrian Wenner was at the centre of lively discussions in bee circles of the USA about the relation between memories of odours and the direction of the dance (Gould, 1976; Wenner and Wells, 1990; Kak, 1991; Wenner, 2002; Munz, 2005). However, apart from some efforts by James L. Gould (1976 onwards) there were no new ideas or results, little reference to the literature, and no tradition of training and testing bees in the New World. A lone voice made no impression against the orthodox bastion, because all the literature came from Europe. On the other side, in Europe after 1945, there was no breakthrough because no admission of a need for progress, no examination of anomalies, and a veto by stubborn referees. A wonderful experiment with a balloon broke the deadlock (Esch and Burns, 1995) and a great effort by Srinivasan and his students followed our first papers (Horridge, 1987; Lehrer et al., 1988) (Chapter 9, this volume). Recent discoveries of discrimination of spatial arrangements of odours have now crashed into all discussions about bee recognition of a feeding site. Odours have been mentioned 39 times, so far in this chapter alone, and it is time to recognize that they are the most significant cues for bees’ memory of routes and places. The first consideration is that all the early life of a honeybee is spent in

the darkness of the hive, relying on touch, taste and odours every moment, to perform complex coordinated activities like rearing brood or evaporating nectar. So, every bee is thoroughly familiar with the world of odours and their spatial arrangement before they go outside. Foraging bumblebees learn to distinguish different spatial arrangements of a single scent in a pattern, and they are better at detecting and learning patterns when they also have a visual cue (Lawson et al., 2018). They also learn a spatial arrangement of one modality quicker when they already learned it in the other. So far, work on bee navigation is entirely about learning performance, with no exhaustive tests of what the bees actually detected, and no tests were done on specially trained bees. That is a waste of trained bees. In my book, they are superb at identifying odours and can distinguish different distributions of several odours. They are more likely to detect modulation of odour, positions of edges of odours, measures of intensity and total odour, and asymmetry of odour distribution, just as they do with vision, using the same circuitry. In fact, I suggest, after learning where to go, they would not know which modality they had used to detect the place.

The Signposts in the Blue Sky In August 1869 at the top of the Aletschhorn in Switzerland, the English scientist J. Tyndall scanned the blue sky with his Nicol prism and found that the direction of maximum polarization of light is always perpendicular to the direction of the sun (Fig. 4.7D), because the dust particles that scatter the light are elongated and float horizontally. Scattering is inversely proportional to the fourth power of wavelength, and at large angles to the sun this scattered UV is useful for detecting small dark objects against the sky with the improved lens resolution that the shorter wavelength allows. This was eventually published in The Forum for February 1888, such was the speed of scientific advance. In drone bees that pursue the virgin



The Route to the Goal and Back Again 213

queen against the background of the sky, and in dragonflies that catch flying prey from below, the dorsal part of the eye is predominantly UV sensitive to improve resolution. While much of Europe was in ruins, and knowing nothing about Santschi (1911), von Frisch discovered that a small patch of blue skylight was sufficient to direct the orientation of the figure of eight dance. A physicist colleague in the faculty at the University of Graz, Han Benndorf, directed him to the polarization pattern of the sky, and in 1967, the observation was published (von Frisch, 1967a). Knowing the time of day and the direction of polarization, the bees had another compass besides the sun. Scientific theories often appear in imperfect form, and the mechanism within the eye took some trouble to unravel. In each ommatidium of cockroaches, butterflies, dragonflies and the honeybee, there is usually one retinula cell with sensitivity peak in the UV. These cells are probably responsible for UV-specific behaviour, such as the escape response towards the open sky. In 1945, however, properties of receptor cells were unknown.

Wild Theories of Compass Mechanisms Bees obviously detected the polarization plane, but how? Several mechanisms of selective scattering or reflection to convert the polarization pattern into an intensity pattern outside the eye were suggested in the late 1950s. Other mechanisms inside the eye, for example a Nicol prism in the cone, were suggested in the early 1960s; one was actually found in some fossil trilobites. These theories were forgotten when the intrinsic properties of the visual pigment rhodopsin were discovered. By illumination of vertebrate rods and cones from the side in tissue slices, it became clear that molecules of rhodopsin absorb more light polarized in one direction than in the perpendicular direction; that is, they are individually dichroic and are lined up (as in Fig. 4.6A, C). In the 1960s, early electron microscopy revealed the structure of insect

rhabdomeres as organelles made of parallel microvilli. It was generally accepted that rhodopsin molecules were oriented relative to the lipid membranes, and it was assumed that the whole compound eye detects the plane of polarization by having each retinula cell with rhodopsin molecules oriented in a different direction, with a pattern repeated in other ommatidia. To mimic this, von Frisch had models made with a mosaic of polaroid, so that when held up to the sky they showed different intensities in each piece. There was much discussion, and even some experiments, about how the responses of the various retinula cells could be integrated together. There was further discussion about how the rhodopsin molecules in the microvilli could be oriented to have more than double the sensitivity in the best direction than in the worst direction, because the value of only 2 is the maximum to be expected from rhodopsin molecules that lie randomly in the plane of the membranes. Later, the basal (ninth) retinula cell of the bee ommatidium was proposed as the sensor because it was sensitive to UV and has a high sensitivity to the plane of polarization (explained by filtering by cells above it). There was more discussion about twist in the bee rhabdom, which, if true, could abolish the sensitivity to the plane of polarization. Direct recordings showed that all bee retinula cells have some polarization sensitivity, but the ninth (basal) cells remained mysterious. In a number of publications in the 1970s, these discoveries were explained in detail. The explanations were based on known components, but they ignored the possibility of a different explanation and were all a wasted effort.

The Dorsal Rim Receptors Many insect groups have a specialized region along the dorsal edge of the compound eye (Fig. 4.7A) where the microvilli in the rhabdoms are aligned in cells with poor optics, large fields, blue or UV sensitivity, and high polarization sensitivity, first properly

214

Chapter 10

described in the cricket. During the 1980s, it became apparent that in the bee these ommatidia have poor spatial resolution but are sensitive to the plane of polarization in the UV. The critical experiment showed that bees could not dance or navigate correctly when both dorsal bands were covered (Labhart, 1980). The dorsal rim cells have angular sensitivity fields with a wide skirt that integrates the polarization over fields at least 45° wide (Fig. 4.7C). The ratio of sensitivity in the plane of the e-vector to that at right angles (see Fig. 4.7B), can be as great as 15 because rhodopsin molecules are lined up in the parallel microvilli and the stimulus is polarized. The receptors in the rest of the eye have a ratio less than 2.0. The 140 or so dorsal rim ommatidia of the honeybee (Fig. 4.7B) look upwards, and each contains nine long straight retinula cells. The axis of sensitivity to the polarization plane has a special pattern in the dorsal band with two types of cell at each place with orthogonal axes (Fig. 4.7B). In tests with the polarization pattern of the sky, if it is assumed that the group acts as a functional unit, the axes appear to be parallel (Fig. 4.7E). So, as Rudiger Wehner, Sam Rossel, Tom Labhart and colleagues in Zürich concluded during the 1980s, the dorsal rim cells act together as a fixed feature-detecting organ for an expected visual task. Observation of the bees suggests an internal analyser of neurons in the brain that read the compass direction from the sky without the need to rotate the whole body. Three such analyser neurons have been found in the central body of the locust (Homberg et al., 2011). They are binocular, with large fields. In all three, light polarized in one plane is inhibitory to light polarized in another, with peaks of maximum sensitivity 60° apart. Whatever the system, the function is clearly not to see the whole polarization pattern of the sky, but to extract a single output. The task is to show the direction to go, for which two other sets of data are required, the land coordinates from the distant landmarks, and the expected position of the sun at the time of day, both of which are learned by each individual bee. Bees learn the sky polarization patterns that

are useful to them, from coincidences in this distributed array of high-level neurons.

The Visual Estimation of Distance Flown Gould (1982) wondered why German bees signal 100 m by two waggles, Italian bees use four, and Egyptian bees require ten. Von Frisch believed that the bees measured the distance by the energy expended on the homeward flight because they reported a shorter distance when flying downhill than when flying uphill; an inference that became fixed in textbooks. Harald Esch, in Lindauer’s group at Frankfurt, never believed this orthodox opinion, but he had no opportunity to prove otherwise until long after he was banished to the USA. While Srinivasan was first experimenting with bees flying along tunnels in 1995, Esch and Burns published their now-famous experiment. They trained bees to fly to a food source on the ground. Then they raised it with a balloon. Later, they flew bees between the tops of tall buildings. As the bees flew higher above the ground, they reported a shorter homeward distance, showing that the perception of distance was dependent on the scene. They had a visual odometer, but it did not indicate a fixed unit of distance for each waggle in the dance. The next year, Srinivasan showed that bees flying along a tunnel towards a food source measured the distance by integrating the visually perceived angular velocity of the walls and floor, independent of the pattern on the walls (Chapter 9, this volume). Tests with a wind in the tunnel, or moving walls, showed that the bees summed the apparent angular velocity over the whole time of the flight, not the total number of edges, effort, or time passed. The next logical step was to calibrate the dance in terms of the perceived motion. In 1995 there was a new enthusiasm among grant-giving bodies for collaboration between distant laboratories, and Srini had plenty of funds, so it was easy to bring experts on the waggle dance to Australia and send our experts on bee tunnels to Europe,



The Route to the Goal and Back Again 215

resulting in a flurry of detail published by Srinivasan, Tom Collett, Harald Esch, Zhang, Aung Si (a Burmese student), Miriam Lehrer and several PhD students, Jane Bidwell, Marie Dacke and Tony Vladusich. Professor Jürgen Tautz and Axel Brochman visited from the Bee Group at Würzburg, and there were other colleagues at both antipodes. The following results are extracted from their work. In thin woodland on our bushy ANU campus, the bees gave about 1.6 ms of waggle dance/m travelled, and 1 m of flight in a tunnel 30 cm wide was equivalent to 25 m outside. In both situations, 1 ms of waggle encoded about 18° of image motion on the eye. A tunnel provided a convenient way to add a large deviation at right angles to the path to a distant goal. In the dance, the bees that were deviated by a tunnel still measured the total optic flow on the outward journey and the compass direction of the hive in a straight line from the food source. Bees would not fly through the tunnel on the return to the hive or switch between tunnels at right angles, showing that they followed their dead reckoning. When bees fly slowly against the wind or faster with the wind, they still measure the impression of the distance travelled over the ground from the optic flow. As the contrast is reduced, for example at dusk, the odometer continues to function normally down to contrasts of about 20% of that in sunlight. When flying over water, the odometer registers much less than the usual distance (Tautz et  al., 2004). The relation between number of waggles and distance in metres depended on the visual appearance of the surroundings. When desert ants travel up and down hill, they remember only the horizontal component of the distance walked. Bees flying in tunnels measure and remember the total optical flow, even in the vertical direction in a vertical tunnel. This agrees with the earlier finding that flying bees measure the angular velocity of passing contrasts irrespective of their direction of motion. Some species of stingless bees convey the height of a food source to the recruits by scent marks.

It was recently found that when the distance of the reward is suddenly changed at a constant direction, bees that find the new location do not adjust their dance immediately, but may take several visits to do so. Bees reset their odometer at a transition such as a tunnel entrance. They also reset their odometer at each landmark and the flight is divided into sections each with an identified beginning and length. A landmark placed in the tunnel before training improves the accuracy of measurement in the tunnel, and when the outside landmark is moved, the search place moves with it. The bees would not search beyond a landmark that was placed in the tunnel after training, just as they would turn away from a landmark that displayed an unfamiliar cue. Conversely, bees overshot the goal when a familiar landmark was removed. In a tunnel the landmark positions overruled the odometer positions from optic flow (Vladusich et  al., 2005), which is what happens in their natural terrain, but paper tunnels are far more convenient for manipulation of the visual scene and for experiments with positions or numbers of landmarks. After this flurry of new results in Canberra, Jürgen Tautz was interviewed in Würzburg by a newspaper reporter, Karin Blawat, who published her own version of the interview in the Süddeutschen Zeitung on 2 December 2009. She claimed that some of von Frisch’s longstanding conclusions about the nature of the dance, the indication of distance and direction to discover a food source, were not correct (quite right, they were not). In response, seven (maybe more) professors quickly contributed varied responses in a coordinated whinge to the same newspaper, perhaps not aware that they would be published. Professor Daumer (Münich) reiterated the history of von Frisch’s famous discovery at length, with a deliberate mention of von Frisch’s early work on flower odours. Professor Menzel (Berlin) poured scorn on the poor methodology of the new work (ours in Australia, and including that in England), demanded a redaction from the publishers of the newspaper.

216

Chapter 10

Then he gives an account of his own work, tracking bee flights with radar, which repeated the orthodox results, which he had published many times, and improved upon with the help of radar to follow the bees. Of course, radar does not see odours. Professor Loher (University of California) reminds us that von Frisch had been in favour of the odour trail, and made experiments that demonstrated it, and that in 1823 Unhoch had described the dance as a rejoicing, but beyond that, says nothing. Professor Markl (Constanz) allows the odour cues to sit alongside the sun compass in the dance, as many researchers had concluded. Professor Rosler (Würzburg) is the only one to accept the new results of Tautz, but with no mention of work in Canberra, Esch or Srinivasan. Professor Hoelldobler (Arizona), a German specialist on ants, expressed at length his poor opinion of Wenner’s work (actually the wrong topic) and recalls that Wenner failed to come when Lindauer invited him to Frankfurt (quite irrelevant). Hoelldobler also provided the familiar orthodox view but seems unaware of the new work on optic flow and the odometer. None seemed confident that Frisch was correct. By 2009, these professors should have been more informed about optic flow and Tautz’s detailed observations of the dance; and also more honest about recent discoveries elsewhere on optic flow. It was a concerted whinge to conceal the truth. They all blamed the journalist for the poor report, and the affair was never mentioned outside the newspaper. It was ever so. May I quote the first speech of Cicero, M.T. (65 bc) when the Roman Senate acted as a law court in the case In Catilinam: Do you not feel that your plans are detected? Do you not see that your conspiracy is already arrested and rendered powerless by the knowledge which everyone here possesses of it? What is there that you did last night, what the night before—where is it that you were—who was there that you summoned to meet you—what design was there which was adopted by you, with which you think that any one of us is unacquainted? (Yonge, 1856)

Practical Route Finding by Foraging Bees In the 19th century, Lubbock described how displaced wasps fly higher and higher until they recognize a distant large landmark, perhaps even a mountain, then head towards home. Bees might learn how many landmarks to pass, but normally they judge distance from the optic flow of perceived ground speed. The pattern of the horizon is also important for ants with good vision, but it might be only a few metres away. Bees that are recruited at the dance fly from the hive using the sky compass, learning new landmarks as they go. They aim correctly under an overcast sky and can infer the sun’s position from known landmarks. At the distance indicated, they search for the food scent picked up from the dancing bee. The dance does not code landmarks or a visual memory. As they search for an odour, they are not looking for a retinotopic memory. They look out for other bees to follow and they will land beside bees that have settled. The results from maze learning show that bees learn whole routes as a series of small cues, but Romanes’ (1885) experiment showed that the sky compass and dead reckoning are not sufficient by themselves on a windy featureless place; they need at least one landmark. Experienced bees in a busy visual environment use the sun compass if it is available and are not disturbed by the shifting of a landmark under a blue sky. They can use one set of local landmarks at one location and another set somewhere else, and learn to visit each foraging area at the appropriate time of day. They can select an odour cue according to time of day or location, but different colour cues only according to location. When some of the local landmarks that mark a goal are displaced, the bees search where the remaining landmarks lie in the expected positions relative to their own eye. Similarly, they perform better if they fly along a particular path regularly, and there is evidence of assistance from the horizontal vector of the earth’s magnetic field (Martin and Lindauer, 1977). Combs in a dark hive are apparently aligned according to the magnetic field (De Jong, 1982), and bees can be trained to come



The Route to the Goal and Back Again 217

to a local magnetic anomaly. A magnetite detector has been tracked down in the abdomen (Lambinet et al., 2017). Each bee repeatedly flies along the same track and the landmarks fall on the same regions of the eye each time. On the other hand, the sun and its polarization pattern move across the sky, so the bees must adjust for the time of day. A landmark is a local cue indicating which direction to fly relative to the sky compass at that place. Bees that are displaced to a new location under a clear sky continue in the direction they were going but quickly begin to search for familiar landmarks. Bees displaced under an overcast sky can recognize tall landmarks from unfamiliar directions, which is the only way they can head towards a food source or home. In these experiments, it is essential to distinguish at the hive and at the food source between bees returning and those setting out. From comparisons of the TBL with the behaviour in route finding, bees and wasps apparently take snapshots with the side of the eye and bounce from side to side of a memorized corridor. Experienced bees use the location and apparent size of landmarks. As Tom Collett found, if landmarks are made larger, bees search further away from them (Collett and Rees, 1997). If local landmarks are moved, the target is sought at a place from which all landmarks have the best approximation to their expected size, appearance and directions (Fig. 10.1B). Movement of one landmark relative to distant ones confuses the insects. Actively flying insects quickly discover how to bring individual landmarks into the desired arrangement as seen from the goal, even while flying in circles. Perhaps a single saccade can put a snapshot of all surrounding landmarks into the visual system. Along a familiar route, natural changes in landmarks are tolerated. The mechanism is flexible because there are usually alternative cues (Fig. 6.23) but they may turn back if they meet an unfamiliar landmark (Collett et  al., 1993; Collett and Rees 1997). Many careful observers noted that bees tend to aim for vertical edges and land on edges rather than surfaces. They scarcely ever fly directly to the reward hole, but when

offered a blue disc they always aim towards it, then diverge to the hole (Baumgärtner, 1928). A detailed study of bee approach tracks suggests that they recognize only one landmark at a time and then make a loop, or turn left or right towards the goal, ignoring a second landmark if not required (Fry and Wehner, 2005). The landmark need not be a cue with polarity, as shown by the ability to learn to turn left or right at a colour or symmetrical cue in a maze, or elsewhere. All work on navigation is compatible with the conclusion that bees identify landmarks and patterns in a Y-choice maze with the same feature detectors and cues, as in open flight, but they recognize them only in the context and places where they were in the training. Everywhere, odours are probably the most significant and generalized guides for bees, because odours are far richer in variety and meaning than the visual input. Juvenile bees work in the darkness of the hive for weeks, using only odours, and foraging bees are sensitive to an immense variety of them.

Cognitive Maps If a map is defined as any internal representation of an extended world, however rudimentary or small, then most active animals have it built into the their visual responses. However, a cognitive map is defined as an internal representation of the geometric layout of objects in the surrounding environment, such that an animal can place itself in the right place by recognizing landmarks or other cues, and then take a novel short cut to its goal. An animal with an internal map takes the shortest distance along a novel track to any goal in its territory. Success is simply a matter of the scale of the map. On a small scale in the Y-choice maze, bees quickly learn the layout of the whole apparatus and when they receive no reward at one target they quickly fly to the other target without going back to the last landmark, as if they know the relative locations of the targets. On a small scale, they learn a map of sorts when the location of the goal is randomized, which suggests that

218

Chapter 10

training with the reward in many possible places is essential to make the bees learn a large-scale map. In a maze, the bees that take the wrong turning soon learn to take the alternative turning as if they have an appreciation of the structure of the maze. Bees can learn to go to one feeding place in the morning and another in the afternoon. If bees that are departing from the hive are taken to the wrong feeding site and allowed to feed, they set out in the correct direction to the hive, showing that they associate the landmarks they see with the compass direction to the hive. If they are taken to the wrong place and not fed, they usually do not fly off in the direction of the other place, but they usually return to the hive and then set off again, or they go scouting around for another food source, then fly their beeline home. For wasps, the detailed studies of Baerends and many others before him showed that wasps carrying food could head directly to their goal when displaced to anywhere in their territory, probably by use of landmarks learnt in numerous exploratory flights. In 1986, Gould produced evidence that displaced bees were able to ‘make use of novel and efficient routes on the basis of map-like cognitive representations’ of local landmarks. (It was retinotopic memory of patterns.) In the following year, he went too far in claiming that recruits interpreted the dance of a returned forager in terms of their own internal maps based on landmarks (Gould, 1987). More likely, they flew out and picked up the correct odour. The idea of an internal map was so objectionable that two professors, not known for previous amicability, collaborated to throw it out. In 1990, they jointly reported new experiments in Europe and the USA in which marked satiated forager bees continued along their compass directions when displaced (Wehner and Menzel, 1990). These bees eventually searched around or flew up high and circled before returning home, so were apparently using odours on the wind or distant landmarks as backup. It was concluded that bees use local landmarks close to their usual tracks, but they reveal memories of distant landmarks when lost.

The present opinion seems to be that when bees make orientation flights they are learning to associate the directions of landmarks with the sun-compass direction of home. A ‘beeline’ home does not prove the existence of a two-dimensional map in the brain of the bee. Departing bees, going either way, have a strong internal signal for distance, which is all used up in arriving bees. Then, after finding the food site, sometimes by scent or by seeing other bees feeding, the new bees learn the local landmarks at the food site and also associate them with the direction of home. When the foragers were fed at one site in the morning and a different site in the afternoon, they were able to take a novel short cut back to the hive when displaced from the hive, but not when displaced from the feeding sites. There was no evidence of an internal map. Notwithstanding this earlier conclusion, in 2000, Menzel and others found evidence of memory of a wider area within which bees could return to the hive from any point (Menzel et al., 2005). In 2005, using radar responders on the bees, they measured the time it took for bees to return home after being displaced (Riley et  al., 2005). Bees that had been trained to a feeder that was regularly shifted in any direction, at distance of 10 m from the hive, all returned quickly, but bees familiar with only one flight track took longer times. Finally, they brought into use a method for continually tracking the positions of a bee in flight by attaching a radar transponder (Riley et  al., 2005). After being displaced under an overcast sky, bees can use familiar landmarks to take a novel short cut to where they ought to be. The results revealed ‘a rich, map-like organization of spatial memory in the navigating honey bees’ (Menzel et  al., 2006). The displaced bees can choose between at least two goals. They can take straight and rapid flights directed either to the hive or first to the feeding station and then to the hive. In the featureless landscape used, moveable tents acted as landmarks, but apparently the varied textures of the ground provided sufficient information for navigation when the landmarks were shifted. All this work ignored the bees’ learning by trial and error



The Route to the Goal and Back Again 219

where not to go, and was still looking only at performance, not mechanisms. To my mind, these results show that the bees and wasps learn as much as they need in order to know the direction of home, and they recognize places wherever they have learned them, as they do with small cues on a target. When the destinations are shuffled, they learn the positions of possible places to look, exactly as in the Y-choice apparatus. They are able to extend the scale of the exploratory flights, and build up a memory of the vector directions to two or more goals from a larger number of landmarks, as was inferred by Baerends for an experienced

wasp taking a caterpillar to one of several nests from any point in her territory. Also, bees fly through complicated mazes. We have no idea how or where this memory of a sequence is coded in bee optic lobes or brain. However, now that the superficial stages of visual perception are known in outline, it should be possible to release well-trained bees back into a foraging area and see how they behave, especially in relation to cues at decision points. We are close to being able to send pretrained pathfinder bees along printed routes in directions and to targets of our choice in glasshouse agriculture.

References Baerends, G.P. (1941) Fortpflanzungsverhalten und Orientierung der Grabwespe Ammophila campestris. Tijdschrift Entomologie 84, 68–275. Baerends, G.P. (1959) Ethological studies of insect behaviour. Annual Review of Entomology 4, 207–234. Baumgärtner, H. (1928) Der Formensinn und der Sehschärfe der Bienen. Zeitschrift für vergleichende Physiologie 7, 56–143. Bethe, A. (1898) Dürfen wir den Ameisen und Bienen psychische Quälitaten zuschreiben? Archiv für gesampte Physiologie 70, 15–100. Collett, T.S. and Rees, J.A. (1997) View-based navigation in Hymenoptera: multiple strategies of landmark guidance in the approach to a feeder. Journal of Comparative Physiology A 181, 47–58. Collett, T.S., Fry, S.N. and Wehner, R. (1993) Sequence learning by honeybees. Journal of Comparative Physiology A 1272, 693–706. De Jong, D. (1982) Orientation of comb building by honeybees. Journal of Comparative Physiology 147, 495–501. Esch, H.E. and Burns, J.E. (1995) Distance estimation by foraging honeybees. Journal of Experimental Biology 199, 155–162. Esch, H.E., Zhang, S., Srinivasan, M.V. and Tautz, J. (2001) Honeybee dances communicate distances measured by optic flow. Nature, London 411, 581–583. Forel, A. (1908) The Senses of Insects. Translated by Yearsley, M. Methuen, London. Fry, S.N. and Wehner, R. (2005) Look and turn: landmark-based navigation in honey bees. Journal of Experimental Biology 208, 3945–3955. Gould, J.L. (1976) The honey bee dance–language controversy. Quarterly Review of Biology 51, 211–244. Gould, J.L. (1982) Ethology, The Mechanisms and Evolution of Behaviour. Norton, New York. Gould, J.L. (1986) Pattern learning by honeybees. Animal Behaviour 34, 991–997. Gould, J.L. (1987) Landmark learning by honeybees. Animal Behaviour 35, 26–34. Homberg, U., Heinze, S., Pfeiffer, K., Kinoshita, M. and El Jundi, B. (2011) Central neural coding of sky polarization in insects. Philosophical Transactions of the Royal Society of London B 366, 680–687. Horridge, G.A. (1966) Perception of edges versus areas by the crab Carcinus. Journal of Experimental Biology 44, 247–254. Horridge, G.A. (1987) The evolution of visual processing and the construction of seeing systems. Proceedings of the Royal Society of London B 220, 279–292. Horridge, G.A. (1996) The relation between pattern and landmark vision of the honeybee (Apis mellifera). Journal of Insect Physiology 42, 373–381. Kak, S.C. (1991) The honeybee dance language controversy. The Mankind Quarterly Summer, 357–365. Labhart, T. (1980) Specialized photoreceptors at the dorsal rim of the honey bee’s compound eye, polarization and angular sensitivity. Journal of Comparative Physiology A 141, 19–30.

220

Chapter 10

Lambinet, V., Hayden, M.E., Reigl, K., Gomis, S. and Gries, G. (2017) Linking magnetite in the abdomen of honeybees to a magnetoreceptive function. Proceedings of the Royal Society of London 284. DOI: 10.1098/rspb.2016.2873 Lawson, D.A., Chittka, L., Whitney, H.M. and Rands, S.A. (2018) Bumblebees distinguish floral scent patterns, and can transfer these to corresponding visual patterns. Proceedings of the Royal Society of London B 285. DOI: 10.1098/rspb.2018.0661 Lehrer, M. (1990) How bees use peripheral eye regions to localize a frontally positioned target. Journal of Comparative Physiology 167, 173–185. Lehrer, M. (1993) Why do bees turn back and look? Journal of Comparative Physiology A 172, 544–563. Lehrer, M., Srinivasan, M.V., Zhang, S.W. and Horridge, G.A. (1988) Motion cues provide the bee’s visual world with a third dimension. Nature, London 332, 356–357. Martin, H. and Lindauer, M. (1977) Der Einfluss des Erdmagnetfeldes auf die Schwereorientierung der Honigbiene (Apis mellifera). Journal of Comparative Physiology A 122, 145–187. Menzel, R., Greggers, U., Smith, A., Berger, S., Brandt, R., et al. (2005) Honeybees navigate according to a map-like spatial memory. Proceedings of the National Academy of Sciences of the USA 102, 3040–3045. Menzel, R., De Marco, R.J. and Greggers, U. (2006) Spatial memory, navigation, and dance behaviour in Apis mellifera. Journal of Comparative Physiology A 192, 889–903. Munz, T. (2005) The bee battles: Karl von Frisch, Adrian Wenner and the honey bee dance language controversy. Journal of the History of Biology 38, 535–570. Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222. Pièron, H. (1904) Du rôle du sens musculaire dans l’orientation des fourmis. Bulletin de l’Institute génerale de Psychologie 45, 221–229. Rabaud, E. (1928) How Animals Find Their Way About. (Translated from French by Myers, H.). Kegan Paul, London. Riley, J.R., Greggers, U., Smith, A.D., Reynolds, D.R. and Menzel, R. (2005) The flight paths of honeybees recruited by the waggle dance. Nature, London 435, 205–207. Romanes, G.J. (1885) Homing faculty of Hymenoptera. Nature, London 32, 630. Santschi, F. (1911) Observations et remarques critiques sur le mécanisme de l’orientation chez les fourmis. Revue Suisse de Zoologie 19, 303–338. See also Memoires de la Societe Vaudoise des Sciences Naturelles 137 (1923). Tautz, J., Rohrseitz, K. and Sandeman, D.C. (1996) One-strided waggle dance in bees. Nature, London 382, 32. Tautz, J., Zhang, S.W., Spaethe, J., Brockman, A., Si, A. and Srinivasan, M.V. (2004) Honeybee odometry: performance in varying natural terrain. PLoS Biology 2, 915–923. Towne, W.F. (2008) Honeybees can learn the relationship between the solar ephemeris and a newly experienced landscape. Journal of Experimental Biology 211, 3737–3743. Turner, C.M. (1908) The homing of the burrowing bees (Anthophoridae). Biological Bulletin 15, 247–258. Vladusich, T., Hemmi, J.M., Srinivasan, M.V. and Zeil, J. (2005) Interactions of visual odometry and landmark guidance during food search in honeybees. Journal of Experimental Biology 208, 4123–4135. von Buttel-Reepen, H. (1900) Sind die Bienen Reflexmaschinen. Verlag Arthur Giorgi, Leipsig, Germany. von Frisch, K. (1965) Tanzsprache und Orientierung des Bienen. Springer, Berlin. (Translated into English (1967) The Dance Language and Orientation of Bees. Harvard University Press, Cambridge, Massachusetts.) von Frisch, K. (1967a) The Dance Language and Orientation of Bees. Harvard University Press, Cambridge, Massachusetts. von Frisch, K. (1967b) A Biologist Remembers. Translated by Gombrich, L. Pergamon Press, Oxford. von Frisch, K. (1971) Bees, Their Vision, Chemical Senses, and Language. Cornell University Press, Ithaca, New York. Wehner, R. and Menzel, R. (1990) Do insects have cognitive maps? Annual Review of Neurosciences 13, 403–414. Wehner R. and Müller, M. (1985) Does interocular transfer occur in navigation by ants? Nature, London 315, 228–229. Wehner, R. and Rossel, S. (1985) The bee’s celestial compass – a case study in behavioural neurobiology. Fortschritt für Zoologie 31, 11–53. Wehner, R., Michel, B. and Antonsen, P. (1996) Visual navigation in insects: coupling of egocentric and geocentric information. Journal of Experimental Biology 199, 129–140. Weiss, K. (1953) Versuche mit Bienen und Vespen in farbigenlabrinthen. Zeitschrift für Tierpsychologie 10, 29–44. Wells, P.H. and Wenner, A.M. (1973) Do honey bees have a language? Nature, London 241, 171–175.



The Route to the Goal and Back Again 221

Wenner, A.M. (1967) Honey bees: do they use the distance information contained in their dance manoevre? Science 155, 847–849. Wenner, A.M. (2002) The elusive honey bee dance ‘language’ hypothesis. Journal of Insect Behavior 15, 859–878. Wenner, A.M. and Wells, P.H. (1990) Anatomy of a Controversy: the Question of a ‘Language’ Among Bees. Columbia University Press, New York. Yonge, C.D. (1856) The Orations of Marcus Tullius Cicero. Vol. II. Henry G. Bohn, London. Zhang, S.W., Lehrer, M. and Srinivasan, M.V. (1998) Eye-specific route learning and interocular transfer in walking honeybees. Journal of Comparative Physiology A 182, 745–754.

Chapter 11 What Was Not Mentioned

Those in the public domain who distort science to support their particular political, ideological, or religious beliefs ultimately damage trust in science and deny the benefits that science can bring to society. (Nurse, 2015)

In all ages and all cultures, there is an abundance of that which is not mentioned. It may be superficial or harmless, like bodily functions, or deeply resentful and disastrous, like failures of human rights, extortion of the weak, elderly or disabled. My topic, ‘what bees see’, is a midget on this vast scene, but carries the serious message that we must be able to trust in science at all times because therein lies our sustainability on this planet, with our health and technologies that allow us to survive in modest comfort. This is of such importance that it was the topic of the Anniversary Address of the President of the Royal Society in 2014. Despite all the trouble and research done to reach this point, I really don’t care a fig about the reputation of von Frisch or von Hess, though they get a good deal of publicity; or any other reactionary old professor, for that matter. I do care about the loss of the Haldane principle: that doctors control the health and hospital system, head teachers control schools, professors the universities, farmers and horticulturalists control the agricultural services, and so on, simply because experienced specialists know what 222

they are talking about. Of course, funding must be voted by a democratic government at each budget, and those in control of it must be voted into office for fixed periods by democratic committees of their peers and those they serve, and certainly not by bureaucrats. For some decades now, the situation has been moving away from Professor Haldane’s ideal and autocratic management has become intolerable in many parts of the world wherever pressure to perform creates a risk to validity of all scientific results. The very best scientists now look for positions in collegiate or wealthy endowed institutions where they can be guaranteed freedom for a decade for their programme, with continued funding, excellent equipment and numbers of students selected by them. They escape from Deans and Deputy Vice-Chancellors of Research. In the case of the bee visual behaviour, however, three academic generations of powerful professors either sincerely believed the wrong theory, or feared that if they revealed errors the loss of their reputation would be followed by loss of funding and position, or even their life. Objectors were denied a platform (Blackford, 2019). However, science depends on new data that drives out old errors, and then new theories can arise. This clan of (mainly) orthodox German professors had to keep quiet because

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)



What Was Not Mentioned 223

they could not dream up better theories of the optics, vision of colour and measurement of range by bees. The resulting scientific calamity (on a very small scale) began with overmuch trust in wobbly work by the previous generation, which was confused and frozen in place by the rise of National Socialism in Germany. After about 1970, new research had to agree with old conclusions and anomalies were ignored. The effect was a disastrous loss of confidence among all interested in analysis of insect vision, and eventual collapse of whole Max Planck divisions at Zeewiesen and Tübingen. I felt it in Australia. Four of my students had direct family connections back to the concentration camps. Several German students and staff were glad to come overseas for experience in comparative academic freedom in Canberra. In my view the active stifling of innovation and discovery could only happen in a relatively closed society, centrally governed from the top. However, that is the growing situation that scientists are now facing for a different reason in many countries that used to have open collegiate scientific research communities. The old international trust in the validity of science, including mathematics, scientific medicine and economics, is subject more and more to top-down control. The stronger the bureaucratic control of directions and funding of research, the less the scientists are free to optimize their progress. The greater the control, the more they avoid innovation, and vote with their feet. Universities become filled with temporary teaching staff, and neither teachers nor students have much chance of a great career. The change was very noticeable in the ANU after 1990, and especially so when the Institute of Advanced Studies was closed down in favour of more teaching of more students. This is nothing new. We can read about it in Gibbon’s The History of the Decline and Fall of the Roman Empire (1983; original 1788), after the Praetorian Guards sold it to Didius Julianus in March, ad 193. The Spanish have a phrase for it, ‘leyenda negra’, the meaning of which is ‘authoritarian and backward’. All durable empires have had to solve this problem by regularly changing

the leadership. Professors, however, cannot be changed because they take a long time to be educated and mature. How can we manage their special requirement?

Social and Political Milieu of German Bee Research 1914–1945 Almost all of the research on insect vision, up to the end of World War II, was done in Germany and published in German, therefore little read elsewhere. After a decade of post-war reconstruction, this tradition continued after 1956, but publication shifted into English. Most of this huge effort, especially histology and anatomy, was superb, but I note that those topics were descriptive, and did not require an understanding of the logic of science or experimental design, nor was that taught effectively to researchers. The Napoleonic system of education put misplaced trust in human intuition as the best way to judge scientific theories, until proved wrong. However, to rely on science just because it has not been proved wrong is a convenient excuse to stick with the status quo. In the 20th century several substantial research efforts led by professors, went badly astray. Factors that made the decay possible were: (i) the freedom to use the funding in any way they wished to control their own research and that of their assistants; (ii) the isolation of the tradition within one language group; (iii) the power of professors over all appointments; (iv) the political situation that encouraged obedience to a leader; (v) their mutual dependence that was based on the appointments system; and (v) their inability to admit that they were wrong. To understand how serious scientific errors could be ignored for decades, and the truth be blatantly denied, we have to understand the complex and sometimes terrible situation in German science in the 20th century. In the last years of the 19th century, science and research flourished relatively freely and was state supported, in the Austro-­ Hungarian Empire based on Vienna, from where it spread into Bavaria, and further into

224

Chapter 11

Germany. Professors had a high status (Coen, 2002). The early industrial revolution had created wealth; governments favoured academic freedom, and a great deal of excellent science was published, particularly biology: for example, studies of marine invertebrates at the Naples Laboratory. The first catastrophe was the destruction of wealth by the war of 1914–1918, and subsequent division of the Empire, accompanied by the rise of Communism in Russia, lack of employment, collapse of the economy in Germany and Austria in the 1920s, gross inflation, and widespread severe poverty. The reaction was a political shift to the right, and eventual rise of National Socialism, beginning in Münich. The second catastrophe was the use of Darwinian argument that the unfit should not survive, and the spread of eugenics from the USA, which was applied first to the insane, gypsies and eventually to the Jews. Anti-Semitism was already latent in Eastern Europe, partly as a result of the dominance of Jews in financial and academic positions, music and the arts. The third catastrophe was German acceptance of authoritarian culture ruled by those at the top, particularly officials, professors and doctors, that developed as a result of the failure of the post-war democratic state. The fourth catastrophe was the terrible epidemic of Spanish flu, which killed off another phalanx at random and shook faith in science and medicine. The fifth catastrophe was the rise of Hitler, and his taking absolute power as Chancellor in 1933, which eventually led to the destruction of the old Europe.

The power of the professors In Germany, after graduation, students could apply to take a PhD, then a further oral examination and dissertation to become Dozent (with pay as an assistant) or Privatdozent (without pay). Professors might have up to five or so assistants, depending on their influence and their teaching load. This system offered professors great power over the choice

and promotion of their students, their research and publications. The conclusions of the professor became the accepted scientific orthodoxy. There was nothing for students who did not support their professor. As an unfortunate accident of history, positions in universities or research institutes (branches of the Kaiser Wilhelm Institute in Berlin) were all funded from the Deutsch Forschungsrat, which became the Max Planck Organization after 1945. Each appointment was approved by state authorities, and tenured with a pension, but all state employees were expected to keep in line with national policies. Furthermore, appointments and promotions of professors were in the hands of committees that included state representatives and several senior professors in similar topics. Effectively, these small national committees decided which candidate would go to each university vacancy. Even recently, I have known of examples where the successful candidate was not the best available, and where the cabal played a more important part than research ability. In the case of von Frisch, the successful but hot-headed young man was first sent to Rostock in the far north, no doubt because he had created a ruckus and offended the Professor of Ophthalmology in Münich. When von Hess died in 1923, von Frisch managed to get back in 1925 before the worst of the social catastrophe began, no doubt with the help of his former friends and relatives. Over the course of the 19th century, professors with Jewish family names (as distinct from Jewish by faith) had everywhere become more common. Von Frisch, via his grandmother was one-quarter Jewish; he belonged to one such family group from Vienna that included his uncles Karl and Sigmund Exner and the Hertwig brothers, all relevant to my topic. The Exner family had produced ten professors. Another example was Mathilde Hertz, in Berlin, although her family had been Lutheran Christians for a couple of generations. Neither that, nor her father’s discovery of Hertzian (radio) waves, was considered when she lost her job as an assistant to Richard Goldschmidt, who also lost his job as Director of



What Was Not Mentioned 225

the Kaiser Wilhelm Institute. By 1933, it was the blood, not the intellect, family or even religion, that mattered. Once appointed, a professor was supposed to continue actively along one research line with his students, and not encroach on the topics of others. A new professor might start a new line of research, but there was no mechanism to encourage innovation. In 1962, I published a paper showing that a collection of central neurons in the hearing system of the locust would enable discrimination of pitch, then was surprised to receive a letter from a German professor. He said that I encroached upon his own belief that Orthoptera are able to discriminate (behaviourally) only the intensity modulation of the sound waves, not the pitch. He complained because I had destroyed his reputation, but did not mention my discovery. Whereas in St Andrews and Canberra I encouraged my students to make discoveries, then consolidate and publish them as single authors, assistants in Germany had to wait until promoted before they could follow their own ideas. German biologists had to maintain a reputation for being painstaking, thorough and correct. It was a mistake to disprove, or even improve upon, the work of others, in case the funding authority might think that standards had slipped as a result, referees were little used in Germany until after 1945. All conclusions of professors were supported by this strong tradition. Even in my time, in my opinion, almost all German referees rejected novel thoughts and seemed to enjoy nit picking. They demanded that their own papers be referenced even if they were in error. Two of the worst offending professors founded their own journals and published as they pleased; with corrosive effect on validity. It is not surprising that, after 1936, there is evidence of editorial control, cover-ups and suppression of the discussion of anomalous data. The work of von Hess on the inability of bees to learn yellow was suppressed, and the anomalies in von Frisch’s paper were never mentioned. All reviews were selective, restricted to the orthodox opinion, and they never mentioned anomalous findings or their authors. Even today, referees following the

orthodox view will not look at any revision of their outdated paradigms. Novelty was rejected (Blackford, 2018). The book on animal vision published by von Hess in 1912 stimulated the research by von Frisch, but not much that was relevant had been discovered before. The century of bee research divides neatly into 1914–1940 and 1956–2016, when bee research was resumed after World War II. It was an extraordinary period in European history, from 1933 to 1945, disgraced by the Nazi persecutions. The early scientific leadership of interest herein was almost confined to one generation. Richard Goldsmith (1878–1958) began as Assistant to Professor Richard Hertwig in the Zoology Department in Münich, where von Frisch was later appointed as an assistant. He made a famous study of the nervous system of a nematode worm and later initiated the science of genetics of the fruit fly Drosophila. He was appointed Director of the Kaiser Wilhelm Research Institute in Berlin in 1914 and effectively became the most influential scientist in Germany. One of his assistants was Mathilde Hertz, the best early researcher on the honeybee visual system. Karl von Frisch was from the same Austrian academic stock as Professors Oscar and Richard Hertwig (Professor of Zoology in Münich), Sigmund Exner and many others, spreading into Germany, all with Jewish family names if not actually practising Jews. Von Frisch was born in 1886; Hitler, also an Austrian, in 1889. There was an unfortunate convergence of disasters; their academic world was destroyed. Cleansing the German blood The rejection of the unfit is far from being a modern phenomenon. Nomadic people are forced to leave behind the infirm, aged and seriously sick. Triage was usual after a battle. Abnormal children were neglected. In outback Indonesia, I have seen small children throw stones to drive away one who was obviously weak and lame. Racial discrimination is universal. ‘Otherness’ is rejected. To combat this, strong ethics are essential.

226

Chapter 11

The attitudes that led to the racial cleansing were formed in Britain, the USA, Germany and some other countries, long before the sudden crystallization of opinion at the formation of the National Socialist Party in Germany in the 1920s. In these countries, Darwinism taught that the fittest thrived and survived, which led to the formation of eugenics societies in the UK and the USA. In Germany, founders like Woltmann and Ploetz produced books and journals of social theory linked to race, but with good intentions. Slowly, anthropologists and psychiatrists became more involved because jobs were available for them. In 1913, a report of research done by E. Fischer on racial crosses in German South West Africa was a popular publication. In 1918, starvation to death of half the number of patients in German mental hospitals was blamed on the war. By 1925, in Münich, the centre of the active bee research, there was also Dr Fr. Lenz, Professor of Rassenhygiene. The Lehmanns Verlag in Münich was a principal publisher of journals and books on eugenics. It must have been a very ominous and difficult place for von Frisch and his assistants. In 1923, Adolf Hitler was imprisoned in Landsberg gaol after the failure of an armed revolt by his men, the Sturmabteilung, the brown shirts, at the Münich Beer Garden Putsch. In prison, he read Grundriß der menschlichen Erblichkeitslehre und Rassenhygiene by E. Bauer, E. Fischer and F. Lenz, and never forgot its contents (Fig. 11.1). By 1931, attitudes towards mental patients, gypsies and Jews were more explicit in Germany. Here is a translation from the third edition of the book by Lenz, Bauer and Fischer: ‘We must of course deplore the one-sided anti-Semitism of National Socialism. Unfortunately, it seems that the masses need such “anti” feelings…. We cannot doubt that National Socialism is striving for a healthier race’ (Bauer et  al., 1927). That was a convenient story that served the interests of the National Socialists. Times change. In 2002, the Institute of Ethics and Practice in Medicine of the University of Göttingen published a call for this frightfiul book to be completely revised (Fangerau and Müller, 2002). More recently Germany

has accepted a million ethnically diverse refugees. On 30 January 1933, Hitler became Chancellor of the German Reich, and on 7 April 1933 proclaimed a ‘law for the reclamation of the civil service’, so that all Jews and half-Jews employed by the state, universities and research institutes were dismissed with no compensation. The reaction was weak. A few complaints about the loss of valuable Jews were soon squashed and the sackings were carried out. Mathilde Hertz and even Director Goldschmidt lost their positions at the Kaiser Wilhelm Institute in Berlin. Many scientists, engineers and artists migrated to the USA, where they greatly improved the quality of academic life and technology. The net was spread very wide. In February 1936, all patients in mental hospitals were documented, and 75,000 of them were condemned to death. In the spring of 1937, all German coloured children were sterilized. Doctors and psychiatrists were recruited to examine every suspect, and to fill in forms that became death warrants. Following the invasion of Poland in September 1939, German laws were applied in Poland. About 10,000 patients in mental hospitals there were shot. By September 1941, about 70,000 mental patients in Germany had been gassed with hydrogen cyanide. Following the invasion of Russia, the murder of gypsies, mental hospital patients and Jews began there. On 17 March 1943, it was decreed that one-quarter Jews should not be classed as German until racial ‘experts’ classified them as not Jewish. Von Frisch had a Jewish grandmother but he was Austrian and by that time had disappeared, to live in isolation near the family’s alpine chateau. On 3 August 1943, 2897 gypsies were the last to be gassed at Auschwitz. In February 1945, all incriminating documents were destroyed. The war ended on 8 May 1945. Many of the officials involved in the killing committed suicide (data from Müller-Hill, 1984/1988). The frightful situation in Germany did not end with the war. Thousands of refugees arrived from the East, notably from the Ukraine. My subsequent post-doc, Randolph Menzel,



What Was Not Mentioned 227

Fig. 11.1.  This is an advertisement for the book by Bauer, Fischer and Lenz, in a major scientific journal, the SitzungBerichte Gesellschaft für Morphologie und Physiologie, volume 36 (1925). This is the book that Hitler read while in prison.

was one of them, coming from the Sudetenland with his family and all their stuff packed on a cart pulled by a horse. Quite a number of people I funded were from families

with similar experience. People in crowded areas were fed at soup kitchens funded by the Hoover Foundation. European transport, rail stations, airports, bridges, factories and

228

Chapter 11

whole cities had been smashed. Von Frisch returned to his chair in Münich and survived through all this. In my opinion, the probability that all that, or something similar or worse, will happen elsewhere, is quite high. The movement of peoples is already upon us, as refugees flee from wars about land ownership, water use, mineral wealth or religious differences. The movement of peoples, as guest workers, economic migrants, students and academics, is continuous. It inevitably causes shifts in the ownership of wealth and resources, but worse, it puts power in the hands of ruthless irresponsible elites. It is essential to find ways of easing these pressures in an equitable way, and not increase them by continually adding to world population.

The response of the biologists The logic of Social Darwinism appealed to the German psyche. Ernst Haeckel, as a student at Würzburg and later professor at Jena, and a spokesman for Darwin, insisted that the theory of evolution and the implied human struggle for existence provided answers to all social, intellectual and political problems. Alfred Ploetz, who founded the Archiv für Rassen- und Gesellschaftsbiologie and a ‘Society for Racial Hygiene’ in 1904, stressed the fear that medical progress would preserve the feeble and feeble-minded. The necessity to preserve the purity of a superior Aryan race and the perceived competition from Jewish financiers, businesses and intellectuals added up to powerful populist intuitive beliefs that Hitler was later able to act upon. In the early 20th century, the study of racial physical anthropology was popular among students, and research was even supported by the Rockefeller Foundation of the USA. After 1933, undistinguished German doctors and professors eagerly filled the suddenly vacant desirable university positions. Study of animal and human behaviour moved away from verifiable science and became holistic. Orthodox Nazi medicine was based on vague ideas of purity of the blood, to

be preserved with herbs, exercise and service to the state. Departments of racial science were established in many German universities. The philosophical support for this racial and political activity was entirely intuitive, and deeply held beliefs in racial purity were quite at odds with scientific studies of hybrid vigour. Among biologists, Karl von Frisch was Austrian and not subject to the new German laws, and being one-quarter Jewish, was not a racist, but he was threatened. He was saved several times by his knowledge of pollination of crops, bee diseases, an outbreak of the very destructive nosema disease, and he appears to have kept a low profile. My own view is that von Frisch was at first wildly enthusiastic about his tests for colour vision, but he noted the findings of von Hess and never published again on that topic. So far as I can discover, his paper of 1914 was never published in English and translation appears to have been deliberately avoided. Clearly, he relied on intuition rather than detailed analysis, and ignored his errors. As the years passed and students discovered anomalies, he used his authority and position as editor to suppress conclusions other than his own. Certainly, several men who disagreed with him had to leave and find jobs in the USA, notably Harald Esch and Rudolf Jander, who both went to the University of Notre Dame, Indiana. On the other hand, he assisted Jewish scientists, and was himself accused from time to time. He supported female students with Jewish family names, and gave them a chance to get a qualification, and leave Germany. None of them returned to work on bees, and their contributions were misinterpreted or ignored. National Socialist Biology went completely off the rails. In 1936, Himmler had set up a new organization, the Ahnenerbe, to research the history, culture and biology of the Aryan race. Himmler seems to have believed that he was a reincarnation of the first German king, and that the Aryan race had been independently created, not evolved from monkeys. Several establishments were set up to revive Aryan culture: for example, the brewing of mead. Traditional herbs for use



What Was Not Mentioned 229

by the army were cultivated in 200 acres tendered by at least 1000 prisoners at the Dachau concentration camp. Rabbits with long hair were bred to make underwear for soldiers on the Russian front. The pseudoscience of the Rassenkunde was also included. For example, Himmler believed that a Grecian nose indicated Aryan ancestry, so he had a special Schutzstaffel (SS) unit with this feature so he could separately record their performance. Matters were far more serious for the wretched gypsies, patients from mental hospitals and Jews who were experimented on in Auschwitz, in efforts to discover racial differences when infected by lethal germs. It is easy to see that scientific validity was thrown out of the window. In 1940, a new monthly journal was published by Lehmanns Verlag, Berlin: Der Biologie; Monatsschrift … der Lebens und Rassenkunde. A prominent contributor and member of the editorial board was Konrad Lorenz, who called for euthanasia for the unfit. In one article, he compared the effects of civilized life with the softening of animals when domesticated. Also Austrian, he had joined the Nazi Party in 1938, when his country was overrun by the German army. In 1940, he was a professor at Königsberg, and was appointed as official psychologist to the provinces of conquered western Poland, where he studied children of mixed German and Polish descent. This period was not well documented. For a time, Lorenz was suspected of being a war criminal by the Austrian Government but after the war hurried to clear himself of any crimes. In 1973, with von Frisch and Tinbergen, he was awarded the Nobel Prize for Medicine for their joint work in founding the Science of Ethology, supposedly the explanation of behaviour. After the war ended, many doctors, anthropologists and psychiatrists who had been involved committed suicide. Others, like Lorenz, made great efforts to get rehabilitated. Social Darwinism was swept under the carpet. Ethology for a time became a respectable science but was based on the study of performance of whole animals followed by intuitive guesswork about what happened inside. However, as said so

many times, cognition is just a word that explains nothing. Performance is insufficient to explain anything except the purpose of the behaviour, and even that is often guesswork. It was no more than popular natural history, but still permeates all behavioural biology of bees. A biting critique of the new science of ethology soon showed that the ethological concepts of inherited maturation of behaviour during development were based on guesswork, ideas about innateness had no foundation, and intuitive concepts about control of behaviour were not necessarily related to concepts about the action of the nervous and hormonal systems. Any theory that is based on concepts of cognition or innateness of behaviour avoids the essential investigation of what actually happens during development as totally naïve hatchlings learn to perform in an unfamiliar environment (Lehrman, 1953). It assumes the very point at issue, that the experimenter was supposed to test. The alternative to the innateness of behaviour is the observation that newly hatched animals, even Protozoa, all learn by tedious trial and error (Jennings, 1905). Piaget developed similar concepts for development of human behaviour. Later it was discovered that something as apparently innate as control of posture of insect leg (Horridge, 1962) or the optomotor response of the fly (Heisenberg and Wolf, 1984) involves continual learning and flexible adjustment by trial and error. When they first enter the outside environment, honeybees must learn every detail of their foraging behaviour by trial and error, and even the use of their eyes, although the neural circuitry is already in place. Simple explanations using the concept of innateness in ethology were based on artificial definitions of cognition and a predisposition to find a cause, any cause. As a result of such criticism, ethology became neuroethology, and then changed again to behavioural ecology. Today it is absolutely clear that perpetual vigilance is essential to prevent populism or bureaucratic drag, and neither Social Darwinism nor any other philosophy or religion is required to initiate creeping or sudden ethnic cleansing, but there is hope

230

Chapter 11

that remedies are in sight. The world has been forcibly made aware. Population has been controlled in China by a ruthless onechild policy that was later softened. Elsewhere, the education of women has been most effective in controlling the worst excesses, but has a long way to go yet. Huge personal footprints and waste of assets are obvious in every rich society, but again education will help.

The Bottom Line Despite the terrible situation in Germany, the study of bee behaviour persisted until 1940, and recovered in Frankfurt, Freiburg and Münich after the war, but the orthodox view did not change because the very doggone persistence of those in charge did not change. Spectral sensitivities of the receptors were unknown until 1964, and there was no alternative theory. Moreover, the culture that protected status and reputations has not changed to this day, and even

grows among us, with signs that the huge research effort of all prosperous nations is slowed down by ever more conservative bureaucratic power dedicated to control of the funds and objectives. Trust in science is easily thrown out of the window. I live in comfortable retirement in Australia, a remarkably free country, multicultural and peaceful, uncluttered by historical drag, and free to research on anything that does not damage others. At least four of my students who became professors had parents from the camps of displaced persons. I have supported post-docs and students from many of the countries worst affected by war and atrocities in Europe. My first assistant technician was a young German captured by the Americans, my last assistant is Croatian, my first post-doc, Josef Hamori, became Dean of Science in Budapest, Hungary, my last student a Sikh in Australia. My 55 years of research has taught me the duplicity of those in power, and how easily the misuse of science, humanity and religion persists, and how pernicious is the dishonesty and its consequences.

References Bauer, E., Fischer E. and Lenz F. (1921) Grundriß der menschlichen Erblichkeitslehre und Rassenhygiene. J.F. Lehmanns Verlag, Munich, Germany. Bauer, E., Fischer E. and Lenz F. (1927) Grundriß der menschlichen Erblichkeitslehre und Rassenhygiene, 3rd edn. J.F. Lehmanns Verlag, Munich, Germany. Blackford, R. (2018) The Tyranny of Opinion: Conformity and the Future of Liberalism. Bloomsbury, London and New York. Coen, D.R. (2002) Vienna in the Age of Uncertainty: Science, Liberalism and Private Life. University of Chicago Press, Chicago, Illinois. Fangerau. H. and Müller, I. (2002) Das Standardwerk der Rassenhygiene von Erwin Baur, Eugen Fischer und Fritz Lenz im Urteil der Psychiatrie und Neurologie 1921–1940. Nervenarzt 73, 1039–1046. DOI: 10.1007/s00115-002-1421-1 Gibbon, E. (1983; original 1788) The History of the Decline and Fall of the Roman Empire. Eight volumes. Folio Society, London. Heisenberg, M. and Wolf, R. (1984) Vision in Drosophila: Genetics of Microbehavior. Springer, Berlin. Horridge G.A. (1962) Learning of leg position by the ventral nerve cord in headless insects. Proceedings of the Royal Society of London 157, 33–52. Jennings, H.S. (1905) The Behaviour of the Lower Organisms. Columbia University Press, New York. Lehrman, D.S. (1953) A critique of Konrad Lorenz’s theory of instinctive behaviour. Quarterly Review of Biology 28, 337–363. Müller-Hill, B. (1984) Tödliche Wissenschaft. Reinbeck, Hamburg, Germany. Translated by Fraser, G.F. (1988) Murderous Science: Elimination by Scientific Selection of Jews, Gypsies, and Others, Germany 1933–1945. Oxford University Press, New York. Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222.

Chapter 12 What We Learned

They all exclaimed at the perfection in colour and pattern of the Emperor’s dress. For none of them wished to be thought stupid. ‘But he hasn’t got anything on’ cried a little child. (Hans Christian Anderson; translated by Robyn N. Sheahan, 1998)

One might say this book is a summary of one personal scientific endeavour on the vision of the bee, but that is only a small part of the story. From 1961, when I returned to the Gatty Marine Laboratory, St Andrews, Scotland, until 1992, when I retired from ANU, insect vision offered many varied topics for training 51 PhD students in difficult techniques of mechanistic analysis of nervous systems of medium complexity, and how to design experiments with a logical outcome. Some analysed the optics, others probed into receptor cells or optic lobe neurons with microelectrodes, some used data analysis online, or microanatomy of neurons labelled with dyes. Some tried to unravel the connections of the neurons of the optic lobes. The mix of related topics, specialists and background education was an essential recipe for success. Eye structure and function in many insects yielded many new principles, for example the function of ocelli, and how the insect retina is limited by the physics of light, and how receptor weight is limited by the need to fly. They discovered a great deal about optics,

receptors, neurons and behaviour of whole insects, especially locusts, flies, dragonflies, butterflies and bees. One result was that 37 of them became professors in 14 different countries. Doing science created scientific leaders and is the only way of making ­progress. It was very strange that, despite a general belief that we were studying insect ­vision, we never discovered what insects actually detected with their eyes that was of interest to them. The finer the detail of the mechanistic analysis, the further we were from explaining the vital mechanisms of visual behaviour. When I retired in 1992, a large part of our understanding of the bees’ visual world required another 20 years of testing trained bees to sort out the feature detectors, cues, and vision of pattern and colour. The hardest lesson was to change direction, away from the detail of neurons, synapses and stimulus–response relations, and known unknowns, and turn attention to the critical unknown unknowns that governed behaviour. The task became the analysis of what trained bees had in memory when they recognized a target, or distinguished two targets. Finally, for lack of alternative insects that could readily be trained, bees have become the leading example of vision for all arthropods. Based on a few examples, we can now boldly affirm the principle that insect natural visual systems rely on rates of

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)

231

232

Chapter 12

change of the optical inputs, and the power and total energy of the green modulation signal, as well as intensity of blue. As a result, out of the window go all the clever correlations between flower colour, shapes and behaviour. Bees are not interested in them. They need only blue and the coincidences of a few cues that identify the place, and they use the same cues at decision points along the route (Chapter 7, this volume). The use of blue perhaps goes right back to the common marine ancestors that had only blue available under water.

What Do Bees Detect With Their Eyes? So, let us look at what they are known to detect using a variety of our symbolic representations of this science. We have formally deduced interactions (Figs 12.1, 12.5, 12.6), probable neuron interactions (Figs 12.2, 12.3, 12.4), pictures of the cues as humans might see them (Fig. 12.7), and the visual world of the bee (Fig. 12.8). We are obliged to use words with established meanings. Therefore they have a kind of colour vision, but there is no reason to support, and plenty of reasons against, the earlier conclusion that they have something like human colour vision of hues and tints. Bees certainly distinguish a palette of colours, but in a very novel, extremely simple way. They distinguish different flowers as more blue or less blue than the background of foliage or earth. The whole visual system for foraging depends on only three inputs in parallel (Fig. 12.1). First, in order of preference, they readily locate and learn to go to blue, and distinguish all colours by a measure of their blue emission and blue contrast relative to background. Because they take a total over a local region (Figs 12.2, 12.3A), they confound blue content with aspects of pattern structure. The blue detector has a large field that is useless for distinguishing pattern in the usual sense of the word. They also measure the height of blue (Fig. 12.3A). Reception of blue is tonic (i.e. persistent in a coloured area), and the single blue receptor cell (R8) in each ommatidium sends an axon through the lamina to end in the medulla. Therefore, there must be other

connections in the medulla that signal blue modulation (Fig. 12.4; from blue receptor cell R7 to blue phasic neuron). Green receptor axons terminate at the lamina at synapses that rapidly adapt (Figs 12.2, 12.4). The signal is transformed with a time constant of about 0.1 s into a measure of green modulation, with the result that bees locate and distinguish edges everywhere by green modulation. The green modulation detectors are very small, only one receptor-­field wide, with a narrow inhibitory surround (Fig. 12.5B). In the medulla, green modulation is summed over a group of edges of various lengths, so green modulation is confounded with edge length and aspects of the pattern structure. The summation is mainly along vertical rows of ommatidia, and the total is located in the horizontal direction (Fig. 12.3B). Likewise, blue contrast is detected with similar resolution, measured and learned, but the location of the adapting synapse is not known. There is no evidence that they measure a ratio of blue contrast to green contrast to give a palette of edge contrasts, and much against that idea because green contrast inhibits blue contrast at the same place. That is not quite the end of story for colour vision, because some insects and other arthropods have an additional red- or yellow-sensitive receptor in each ommatidium, probably for specific recognition of food plants, mates or prey. Of the three inputs, bees least prefer blue contrast. Scanning in flight excites modulation of the blue receptors that measure the averaged local intensity of retinotopic blue receptor modulation. All modulation detectors are very small, only three facets wide, and there are no blue orientation detectors. The single UV receptor in each ommatidium sends a long axon to the medulla, but there is no firm evidence that it is useful for recognition of colour, although it may be useful to detect the direction of the scattered UV in the sky in the escape response of disturbed insects. The six green-sensitive receptors in each ommatidium make synapses with several lamina neurons. There is no evidence that green tonic colour is transmitted via these connections. Bees detect, learn and remember a number of cues via green modulation



What We Learned

(A)

(B)

233

(C)

OUTSIDE BEE Blue colour summed

Blue contrast scanned

Green contrast scanned

Tonic signal

Phasic signal

Phasic signal INSIDE BEE

INSIDE BEE Blue content detector response

Scalar strength

Blue modulation Green modulation detector detector response response

Measure of change

Measure of change

Receptor level

Feature detectors

Quantitative measure

Three independent variables ±S

dB/dt

dG/dt Search for coincidences

Coincidences of ratios and angles of separation on the eye

Cue detectors

Learned preferred coincidences

Fig. 12.1.  The three types of visual input from each ommatidium of the bee in the foraging behaviour. Solid evidence of a UV input to vision of colour is lacking. A blue tonic signal does not rapidly fade away, but phasic signals adapt rapidly, so they signal each change in intensity inside as the eye scans across contrasts outside. Tonic and phasic responses are located on the right in Fig. 12.4. S is the stimulus to the blue receptors. It is less or more than background. dB/dt is a differential operator, a rate of change in blue over time t; dG/dt is the equivalent for green.

r­ esponses summed over local regions. The totals play an important part in recognition but pattern is lost. Asymmetry of the distribution of green modulation and the positions

of ­concentrations of modulation are also detected and remembered. Small groups of green channels from six adjacent ommatidia form arrays of orientation

234

Chapter 12

One omatidium (facet)

UV Blue receptors peak in green

I°

LAMINA Lamina ganglion cells MEDULLA

TO DEEP OPTIC LOBE AND BRAIN

Fig. 12.2.  Three ommatidia (facets) showing receptors and principal neurons of the eye and superficial optic lobe, typical of all insects. The lamina ganglion cells rapidly adapt, and transmit the rates of change of the input, which is the modulation signal serving many functions. Responses of blue receptors are summed to provide a measure of blue in local regions, and distinguish colours relative to background. Connections of UV cells are not shown. Eac hv o tras f facets ertical t an lin s pos d det ums gr e ects ition een i by l ine ts horiz iden o tity ntal

con

Each horizontal line of facets sums blue intensity and detects its vertical position by line identity (A)

(B)

Tangential fibres in the medulla

Orthogonal tangential fibres in the medulla

Fig. 12.3.  Arrangement of axons of blue receptors, and second-order axons of the green receptor channel, that sum upon horizontal fibres of the optic medulla. (A) Horizontal lines of blue detectors identify the vertical position of blue by line identity. (B) Vertical lines of green modulation detectors identify its horizontal position by line identity.

detectors. Responses of orientation detectors are summed over local regions in such a way that equal lengths of edges at right angles cancel the orientation (Figs 6.9, 6.10). Isotropic

orientations in a texture would be detected only as a region of modulation, while an area of vertical plant stems would show up as orientation as well as modulation. The positions



What We Learned

235

One ommatidium 30 μm diameter = 0.03 mm Retina detects intensity

Six green receptors

One blue receptor

One UV receptor

R7

R1–6

R8 Tonic responses

Stimulus

Lamina cells have small fields

Lamina detects change L3?

Medulla has lateral interactions

Blu

ep

ha

sic

Green phasic

UV tonic

L2

UV phasic ?

L1

Blue tonic

L4

Receptor response is phasic and tonic

Medulla

Dm8

Phasic responses

Tm?

Tm?

Tm? Summation from a horizontal line of facets detects vertical position of blue

Tonic responses Stimulus

Neuron response profiles

Stimulus Summation from a vertical line of facets detects horizontal position of green contrast

Fig. 12.4.  Montage of known peripheral neuron connections. Receptors are well known for many insects, including bee. Lamina neurons are best known in fly and locust, and medulla neurons in fly and (anatomically) in Drosophila. Dm8 and Tm are neurons, L1–4 are nerve cells in the lamina. Compare with Fig. 12.1.

and orientations of responding feature detectors are lost in the summation, but measures of total modulation and average orientation can be located and learned. Responses of edge detectors are not strung together to make continuous lines. The green channel also feeds into a distributed network that detects radially or tangentially arranged edges in relation to a hub. Bees detect whether a pattern is predominantly radial or tangential but they cannot distinguish between patterns of the same type, except by a modulation difference. The position of the hub is remembered irrespective of the type of pattern, and a hub can be

r­ eplaced by a blue or black spot (Fig. 6.21C). So far, tests have been unable to demonstrate recognition of more complex aspects of pattern beyond this limited menu of feature detectors and coincidences of their responses. The green receptor channel is also the input to feature detection for motion of a contrast from one ommatidium to the next, so that motion perception is colour blind in green. This is not a significant factor in the natural situation because every green leaf and brown stone is normally edged by a shadow. An interesting detail is that responses to an edge are independent of the direction of scanning, which implies that the detectors

236

Chapter 12

of modulation have circular symmetry and are unaware of the scan direction. The detectors of edge orientation have bilateral symmetry about lines that run in three different directions. As a result bees cannot distinguish a black-white edge from a white-black one. This system of separate summation of blue and of edge detectors makes it impossible to reassemble a pattern, although it is clear from the scanning behaviour and ordered columnar structure of the optic lobes that the gross distribution of stimuli across the eye is somehow partially recoverable.

A Simple Neuron Interaction Yields a Useful Cue A mechanism for the measurement of the height of blue by bees can now be suggested by referral to known neuron anatomy in the fly Drosophila. Blue receptor axons originating in a horizontal line of 10–16 facets of the eye are summed upon a horizontal fibre deep in the medulla (Fig. 12.3A). The line label (identity) of the horizontal fibre identifies the position of a small region of blue in the vertical direction. Similarly, green modulation is located horizontally by the position of the terminals of a group of inputs originating in the green channels of a vertical row of ommatidia (Fig. 12.3B). Coincidences between feature detectors commonly form cues. The angle subtended between two prominent vertical edges (width) can be measured and remembered (Fig. 5.15). Furthermore, detectors of edges with green contrast have small retinotopic fields and are separate from detection of blue content of areas. Bees especially look for a coincidence between a patch of blue and a landmark with green contrast, detect the horizontal polarity of the combination, and measure the included angle (Fig. 12.5M). This is an important algorithm (way of processing data) in the mechanisms of discrimination and learning (Chapter 7, this volume). Such a system distinguishes patterns by the different sets of coincidences in the responses of detectors, and has extreme astigmatism, so in no sense could it see as humans do.

Behaviour Can Be Correlated With Neuron Types Ground-breaking work of Osorio, Srinivasan, Meinertzhagen, Shaw, Laughlin, Hardie, my assistant Ljerka Marcelja, and many others from the 1960s to 1990s, revealed the actions of receptors, neurons of the lamina, and a few neurons of the medulla. They agree very well with the more recent descriptions of the feature detectors and cues. Deeper down in the medulla, correlations between neurons and functions are rare. The long lost circuit for fly motion perception is beginning to appear in the exact column anatomy of Drosophila medulla, but the simple repeated connections for orientation detection (Fig. 12.5C, D, E) are not yet located. Representations of the feature detectors and cues show that they are mostly simple spatial sums of modulation, and cues are coincidences of individual bits of modulation. No wonder that recordings from optic lobe neurons yielded no information about the function of each neuron type, because the many responses in parallel all look like modulation, and they cannot be separately related to behaviour. The geometry of the individual cues suggests that many neurons will be found with antagonistic green-green inputs that generate the cancellation of orientation at orthogonal edges (Figs 6.9, 6.10; 12.5F, G, H, J) but this stimulus has never been tried during recording. All cues presumably are generated by specific neuron connections that detect coincidences (Fig. 12.5K, L, M, O). Picturing the cues in this way is the natural conclusion of mechanistic analysis that combines data from neurophysiology and results of testing trained bees. Hopefully it will lead to better design of the analysis of real neuron activity.

Symbolic Representation of Cue Structure There is plenty of hard data on spatial resolution showing that modulation is detected by individual lamina neurons with a small



What We Learned

Green sensitive receptors 1°

237

Three edge orientation detectors –1

(A)

(C)

(D) –1 –1

(B)

–1 –1

+6

(F)

Vertical

–1

Hub

3° –1

–1 –1 +2

Hub

+1

Hub

Orientation cancelled

(J)

orientation cue

(L)

(P)

+1

–1

orientation cue

Oblique

One tangential cue

–1 (E)

(G)

(H)

(K)

+2

–1

Feature detectors

+1 +2 –1

–1

Modulation detector

–1 –1

+1

+1

–1

+1

–1

Orientation cancelled (M)

(N)

(O)

Hub One radial cue (Q)

Left/right polarity (R)

Height and total blue (S)

Asymmetry along horizontal Modulation gradients detected in scanning Fig. 12.5.  Visualization of feature detectors and cues as organized symbols. (A) Receptors. (B) Modulation detector of the lamina, with resolution near 2°. (C–E) Orientation detectors, with resolution near 3°. (F–J) Summation of orientation by separate detectors upon a collector neuron. (K, L) Circles and spokes detected by coincidences of local orientations. (M, N) Polarities with blue and a landmark. (O) Height and summation of blue emission. (P, Q) Polarity of modulation irrespective of pattern. (R, S) Polarities of modulation ­distributions.

238

Chapter 12

inhibitory surround that sharpens the field a little. Modulation detectors are individually separate, and summed in groups that form known cues. This can be visualized with radial (Fig. 12.5L) or tangential symmetrical units (Fig. 12.5K). Orientation detectors of the medulla must be formed by groups of seven columns that contribute an input from each axis and form bilaterally symmetrical detectors of orientation. There must be at least three types of these orientation detectors with axes at 120° to each other (Fig. 12.5C, D, E). Next, groups of responses of orientation detectors are summed upon fibres that run horizontally in different directions in the medulla, and detect local areas of average orientation, with much cancellation of directions that are less strongly represented. This must happen in the medulla because that is where the grouping can occur, but the output terminals may be in the lobula. In the medulla, also, are detectors of coincidences between blue and green modulation (Fig. 12.5M, N), detectors of height and amount of blue (Fig. 12.5O), detectors of asymmetrical outlines of modulation (Fig. 12.5P, Q) and detectors of gradients of modulation (Fig. 12.5R, S). These diagrams may allow the corresponding neurons to be labelled in a circuit diagram. The neurons are identical from one individual worker bee to another, which is a great help, so that all can be given index numbers, and connections between them can be slowly visualized. The formal model (Fig. 12.5) then becomes a model of neuron function (Fig. 12.4). It turns out to be surprisingly peripheral in the lamina and medulla. You may have noticed that these cues are not like human diagnostics. They work by taking totals, indicating relative amounts, relative positions, go left or right, above or below, increasing or decreasing in a scan, making themselves independent of range, light intensity, or pattern. Quantitative relations between the new input variables and cue preferences have yet to be measured. The evolution of the flower colours will have to be reconsidered in the light of the detectors that the bees had previously evolved when they behaved more like ants that were presumably also colour-blind in blue. Into the bucket go the models of

c­ olour space, the city-grid and other methods of calculating colour separation, how bees detect radial pattern, the relation to background, out-dated theories of how bees detect green, black and white, how bees detect the removal of black, explanations of responses of optic lobe neurons, location of memory, what it is like to be a bee, and much more. Even more difficult to digest, researchers will have to remodel their own mental models, improve experimental design, and their teaching. Perhaps they should use less intuition, and do more simple experiments leading to firm deductions.

Visualizing the Formal Interactions Within When we put together everything we know or infer about the bee optic lobe, and the way that three distinguishable parts relate to each other, we can make an imaginary picture of the interactions during the visual processes. Really, this is no more than interactions that have been progressively discovered or inferred over at least two decades of experimental analysis (Fig. 12.6).

Visualizing the cues In addition to the above, the cues can be visualized as humans see them (Fig. 12.7). We actually see them, whereas bees only detect feature detector responses that are lost as soon as used. This type of system, with special adaptations to other habits and habitats, will probably be the rule for all arthropods, and we might expect related types of vision of colour and pattern in lower vertebrates and invertebrates.

Defining the visual world of the bee, and other arthropods Having legs and more critically, wings, bees function well in our three-dimensional (3D) world but they certainly do not image it as



What We Learned

239

?

Green receptors

Blue receptors

UV receptors

? Modulation of green-sensitive monopolar cells

Other lamina cells?

Retinal

on

coordinates

all

Modulation resolution 2.5°

Local orientation detectors resolution 3.5°

Measures blue content of field of each eye

180°

90°

Field sizes

5°–90°

Measures angle between green modulation and blue position Measures angle between two positions of green modulation Measures angle between two positions of blue modulation Coincidences of two inputs

Lamina detects temporal modulation

channels

Medulla Measures blue and perhaps UV modulation

Retina detects intensity

detects Nondirectional motion detection

local features Directional motion detection resolution 2°

Lobula sums features and detects cues

Local retinotopic orientation Range Global average orientation

Total is odometer

Tangential vectors Radial vectors Radial hubs Tangential hubs

Piloting

2.5°

External

yaw, roll, climb, land

Cues from summed edge orientations Protocerebrum

Coincidences of cues, recognition of landmarks

Directs fixation

Flight control

Sequence of retinotopic labels moved to memory

Coincidences of labels recognition of place Internal

Fig. 12.6.  A map of the formal interactions between the different processing channels in the honeybee optic lobe. At the top is a single ommatidium above a cartridge of lamina neurons. Below that is a column of the medulla, where most of the features are detected. Non-directional motion has been replaced by totals of green modulation. Detection of asymmetry is not shown. Responses of feature detectors combine to form cues, and coincidences of cues enable recognition of places and landmarks. Approximate field sizes are shown on the left. A resemblance to the optic lobe of the bee is not accidental. Compare Figs 4.14 and 4.15.

240

Chapter 12

Action of blue receptor input

Action of green receptor input

Modulation

Active when green is not available

First preferred input from all targets

Black area covering blue

Tonic blue input summed, measured and located

Not sensitive to areas or colours

Feature

Weak

Strong

Width

Height

Blue

Blue

Polarity

Blue

Blue

Retinotopic

Measures and locates angular width to 90°

Measures blue content and height

Inactive

Locates and measures angular width from tonic blue to green modulation and polarity in horizontal direction

Angle Radial

Not sensitive

Active

Bilateral symmetry

Not sensitive

Active

Orientation

Not sensitive

Active

Tangential

Not sensitive

Active

Fig. 12.7.  Diagrams that approximate to cues detected by bees, and that humans recognize. The illustration is only to assist the reader visualize. The cues are each detected by a group of feature detectors leading into an activated neuron circuit within the bee. Black implies an area of no stimulus and missing blue. Edges of black signal maximum contrast.

we do. The panorama is unavoidable, all around, but objects are detected and recognized by totals of local feature detectors and angles between them, and outstanding edges, not as separate things. Route signs are detected by their horizontal asymmetry, landmarks by a coincidence of cues, and places by landmarks and blue content. The 3D world of the bee is rich with odours, as judged by the number and variety of receptors and the huge numbers of

neurons of the olfactory lobes and tracts of the brain. For bees within the hive, and for many burrowing, deep aquatic or nocturnal arthropods, the main sensory environment is one of odour and taste. Several tracts of neurons run from the olfactory input to the deep optic lobes, so perhaps we should consider perception at a distance by vision as a useful addition for exploration of the environment by the olfactory system. The idea makes a great deal of sense as a bee searches



What We Learned

for food. When a bee pauses at a flower, perhaps the main cue is odour, not to see the flower. Stationary insects sitting on a leaf see only blue content unless there is relative movement. In the new paradigm, the 3D world of flying insects is a panorama of range to the nearest edges in each direction (Fig. 12.8), all derived from the optic flow. There are also moving landmarks, and polarity of anything with horizontal asymmetry. It is easy for us to imagine because this is the world of a ship or aircraft in fog when surrounding objects are detected by radar that gives range but no image or identification. This is the world of the electric fishes that swim in muddy water, and of bats relying on sonar, with a panorama of range, but no way to categorize separate objects. By having huge visual processing centres, higher vertebrates are the odd ones out in the world of animal vision. With reference to colour, we have a new paradigm in the bee, monochromatic in blue in one scalar dimension where blue content is measured, and rates of change and modulation power are measured in blue and green

241

channels independently, and where the signals are measures that confound pattern structure with contrast. Varieties of this new paradigm are probably found in all natural visual systems. Certainly in primates most of the neurons connecting the eye to the visual cortex are phasic (i.e. detect change, not tonic intensity) and what is known about vision in annelids, molluscs, fishes, amphibia and birds reveals phasic signals, often with mutually antagonistic inputs. Human vision also first separates the edges in the image, and fills in the colours of areas later with separate channels. Efforts to make robot vision often start with a map of the responses to edges, then look for hubs. The limits of our knowledge We have real bees that learn to distinguish patterns or signposts, and we have no idea how they distinguish or learn. They are not computers; they make mistakes when shown unfamiliar displays that contain the cues they have in memory. That is often called ‘generalization’, but, in fact, the bees have

Measure of modulation

Measure of range

Yellow flower lack of blue Modulation Modulation Range

Range

Asymmetry of modulation Blue relative to landmark

Circular pattern Radial pattern Average orientation

Fig. 12.8.  The visual world of the bee is an extremely wide visual field, with resolution of motion and modulation down to near 2°. Modulation (dotted line) and range (dashed line) can be measured in every direction by the relative motion of the eye. Some cues display polarity; others do not.

242

Chapter 12

insufficient in memory to distinguish many alternative patterns. We have no idea what we mean by the phrase ‘in memory’. Where and what is the place of memory? Bees detect cues and then make a decision. They usually choose by avoiding the unrewarded display. We have no idea what we mean by the phrase ‘they choose’. Hopefully, the process is a balance of preferences ‘for’ and ‘against’ that is signalled in neurons that will be identified. This must be somewhere near the locus of memory. There, neurons concerned with the visual input feed into far fewer neurons that pass on ‘Yes’ or ‘No’ decisions to the place where motor options are stored; but this is pure guesswork. Judging by rapid advances in understanding the motor system, these neurons will be identified, but we will have no idea how or when bees ‘think’. Why was progress so slow? Just to ask this question implies that innovation should arrive on time to oblige, and that a way forward exists, which was not true for bee vision. Scientific limitations were severe. In the early 20th century, there was confusion between phototaxis (attraction to light), the optomotor response (passive visual stabilization on surroundings), recognition of places and food, and other visual responses. Nobody imagined they differed in inputs and processing. Next, there was no information about the variety and properties of the receptor types, and complete ignorance of neural processing and what it could achieve. Spectral sensitivities were unknown, and the first microelectrode recordings of them were in the early 1960s. Ideas about coding, edge detectors and simple cues came from wartime work on electronics and radar, and machine vision, which spread slowly into biology. Third, for the whole century, performance of bees was studied, not mechanism. As I have so frequently pointed out, however, you have to see the spark and pistons to understand a motor car. About a century was wasted in training bees without exhaustively testing them to see what they learned.

Innovation failed to appear because it was so obvious that bees distinguished flower colours and shapes, and nobody could imagine Ronacher (1979) an alternative paradigm. ­ demonstrated that. Five professors asked to advise him could not offer an explanation of his data (Figs 3.10–3.11). We can rule out National Socialism as a factor simply because research on bee behaviour is so cheap, and has no political ­impact. Anti-Semitism was not a factor because von Frisch protected assistants that had Jewish family names, and their research prospered. Most managed to escape overseas when qualified. Also, they certainly were not lazy or slow, although there was a good deal of copying reviews and textbooks from old work without question. Over time, post-war work on bees was all governed by male professors with orthodox convictions, and little changed. In my laboratory between 1963 and 1992, there were several German students and staff. Together, we all believed in the orthodox trichromatic colour vision of bees without question. At first, I also looked only at performance in training. We learned about equiluminant colours because Srini had worked in a laboratory for human psychophysics at Yale. The Y-maze choice chamber was invented because we had to define the range for making measurements of spatial acuity, not for pattern vision. Once we put the logical experimental design together, with very many tests of trained bees and used logical deduction with no guesswork to reach a conclusion, the whole upset of the orthodox view was unstoppable. After 2006, as other researchers on bee vision closed down their projects one by one, it is certain there was a conspiracy of silence. I am sure that the reputation of family, loss of esteem and honour of country were somewhat to blame, especially for full colour vision, the defence of Exner’s image, and the optics of the eye of the flour moth, Ephestia (Chapter 4, this volume). Isolation in small groups, lack of training to be critical, and pressure to perform and to be approved, were all large factors. ‘Research findings may often be simply accurate measures of the prevailing bias’ (Ioannidis, 2005).



What We Learned

There was also ancestor respect, pride, and because, having exposed themselves and gone into print loaded with error, ‘none of them wished to be thought stupid’.

Goodbye to Intuition Based on Performance Over the years of searching for the correct thought, we learned a lot about intuition. A powerful cause of the errors of thought was the human habit of looking at, comparing and measuring the performance (i.e. what the animal does; and then making guesses and judgements based on our own experience). Humans have the strong tendency to observe performance and infer an intuitive conclusion based on outwardly similar systems, or human performance. Bees distinguish some colours – therefore they have colour vision. Bees (sort of) distinguish a colour from all shades of grey – therefore they have colour vision like us. Bees distinguish between many pairs of shapes, and recognize single shapes, but that does not show that they use the shape in the task; actually they use small cues. Bee vision is hopelessly anti-intuitive and not at all like our vision. Finally, humans evolved in the same world as insects but have almost no ability to imagine the unknown mechanisms of insect vision that must perform superficially like ours. Performance tells little about mechanism. You learn little about a camera mechanism from the picture or looking down the viewfinder. To understand the flight of a plane, you must understand the direction and velocity of the invisible airflow around the wings. The bare ability of bees to distinguish patterns, shapes and colours was of little help, in fact, extremely misleading. Retraction of hasty intuitive conclusions was painful, and a new paradigm was urgent. Consider the following. Bees learn asymmetry, not symmetry; they learn modulation, not contrast; they distinguish colours by using blue content relative to background; each ommatidium has good resolution but the information is summed, they distinguish ­

243

patterns by structure of modulation, not by shape. To discover visual mechanisms in the bee required extensive new test sets that were carefully designed to produce firm data and deduce what cues in the training patterns the bees had really detected. It was a waste of time to train animals without thoroughly testing them. The Napoleonic education system in Europe taught that the human mind was adapted or created to make correct rapid decisions based on experience. It was called ‘rational mind’ or ‘intuition’. In the 1920s, it was the basis of Popper’s idea that reasonable explanations should be accepted as truth until proved false, but this was a fragile foundation for the search for scientific truth. In the time of Galileo, it was already pre-empted by demonstrable experiments with logical conclusions: nullius in verba; ‘words are worth nothing’. A shift away from intuition that accompanied the student riots in France in the late 1960s marked the end of postmodernism. Scientists demanded that, to be valid, conclusions should be based on logical deductions from firm, repeatedly observed data, not temporary guesswork, dogmas or belief in the words of ancestors. The student riots reached the USA, but in Berlin and elsewhere in Germany they soon fizzled out as the professors reclaimed their authority. They boasted about their victory. Even in France, much of the reliance on intuition remained. They are now overtaken by the revolutions of molecular biology and digital technology that demand complete material and logical precision. In the case of bee vision, refusal to investigate anomalies was the most obvious delaying factor, and one that I exploited wherever it was found. The major efforts by von Frisch, Menzel and Giurfa, on bee vision of colour, and those of Wehner, Srinivasan and Zhang, on patterns and shapes, abounded in minor contradictions. New difficulties of interpretation appeared with every new batch of experiments. A few single-­­shot authors with very awkward findings, like Carricaburu, Esch, von Hess, Jander, Kriston, Kuiper, Ronacher and Ibbotson, were not allowed a platform (Blackford, 2018).

244

Chapter 12

I believe by tribal allegiance to him, and lack of an alternative, von Frisch delayed research by 100 years. Clearly, everybody except von Hess accepted the undigested field notes of an enthusiastic youth. Effectively, von Frisch had jumped to a ‘reasonable conclusion’ after accumulating a good deal of data. Since Aristotle, almost everyone who altered the course of science had done just that, hoping to be correct until proved wrong. However, von Frisch actually was wrong, and probably he had realized and suppressed that sad fact. In this topic, however, nature demanded anti-intuitive conclusions.

Innovation Innovation is a term for anything between applications of the mobile phone that assist banking, to fundamental discoveries about the nature of the photon or interactions between galaxies. Practical innovations that create convenience for everyone and fortunes for their owners are their own reward and must be governed by new laws.

Fundamental discoveries The rare opportunities when significant fundamental discoveries appear are the result of careful planning, but that alone does not ensure success. Four or five conditions must be synchronized. The most important is to have available someone with long experience and extensive knowledge of the literature, who would know the boundaries of the relevant sciences. Knowledge alone merely stifles innovation, but someone who can map out the known unknowns, and perhaps suggest a few unknown unknowns, can save a lot of useless exploration. Secondly, the magic ingredient is the confident experimentation and unrolling of ideas by enthusiastic youth, or youthful oldies. For some innovations, the above may be sufficient. With a group in a research laboratory, management must be generous, so that every youthful avenue is explored, but discipline

must be slack to give them room to think. Youth performs best when left to explore, and also when a goal is in sight. Most innovations require much more – for example: (i)  an abundance of cash for studentships; (ii) expensive equipment; (iii) user time on big machines; (iv) expendable supplies; (v) animal and plant house; (vi) technicians to teach techniques and how to use equipment; and (vii) travel to conferences. Sometimes research is pushed by an urgent need to solve a problem. There is much to be said for attempting something that is ripe for discovery at the right place, at a pregnant moment. All these improve the chances of individual success in a well-founded research institute. We can seek innovation, and occasionally fall upon it by informed ­accident, but for consistent success, a great deal of background reading and deep thought is required, and a lot of hard work flowing from each crucial new thought. Application and marketing of fundamental discoveries requires a totally different and separate larger effort by an independent commercial company with long experience. It is essential to have a well-policed system of patents to allow them time and opportunity to profit by the findings. University researchers rarely worry about patents and do not have immediate funds to make their discoveries secure. In fact, to gain a reputation and get a grant, academic researchers may reveal all their secrets at the next conference on their calendar and publish as fast as possible. This creates a drag on the application of research because industrial companies want exactly the opposite, to keep results confidential and make use of them. The Swiss method, of linking finance for PhD students and university research grants serves for practical topics, but fundamental innovation has to be separately funded generously by taxpayers. The greatest drag on the advance of fundamental university research is the power exerted by bureaucrats who decide the topic, select the winners, decide and restrict the budget, and how it is to be used. The official system has many faults and is easily corrupted by back-room deals. Worse is a requirement to state at the beginning



What We Learned

what is intended in great detail, then apply for every bit of funding to support all the research activity, maybe even including rent for the space, then submit regular progress reports that are judged by the very competitors of the enterprise. When the success rate for applications is less than 20%, university staff regard the situation as unfair, and abandon research to become teachers of students. The very best researchers search for and may find positions in research institutes where staff are fully funded by a wealthy university, large company or government organization. Another way is to work on a self-­funded project, following ­Lavoisier, Darwin and a few modern entrepreneurs. Power Truly, fear of being out of line and annoying the boss, or professor, has strongly influenced all human societies, but those in charge have to be worthy. Being mistaken can easily be forgiven, because most learning is by trial and error, and most fields of knowledge have been ploughed and replanted many times. Also, the boss is an essential part of all societies, and cannot be dispensed with entirely. Finding the leader for the occasion is critical. Following Confucius, or the Pope, or Marx, will not suffice for coping with climate change and overpopulation. Continual new leadership is obviously essential to cope with the one planet we share. Science has become the heart of our modern civilization and is central to our intention to survive and prosper. Science is the only path of self-improvement and survival that depends upon validation by reference to the real world, empirical tests and logic. Science, however, needs new ideas, continually changes, and hopefully advances. Applauding the existing Emperor for the beauty of his clothes saves trouble, but will creep over us if we don’t take care to teach, learn and practise transparent honesty based on scientific investigation in every walk of life and have a style of government that will encourage and adopt new discoveries (Nurse, 2015).

245

Trust In research, there is a lot to learn about trust. The only liberal international example of knowledge that is based upon critical validation of evidence is the best of science, including mathematics, scientific economics and medicine, as practised in the prosperous democracies. This is the only natural system that endures. As they matured and lost their youthful innocence, transparency and enthusiasm, other liberal worlds of politics, trade, finance, banking, even religions, and certainly huge and wealthy international companies, and every other pursuit with a controlling bureaucracy, have forfeited the trust of thoughtful and critical educated, liberal, democratic populations. Wealthy permanent managers exploit all the resources, obscure their private transactions, clip all tickets in sight, follow the money, blame everyone else, pay little tax, create oceans of waste, control from the top by fiat, then respond to reaction with increased surveillance, and passing of more restrictive laws. By a natural process of bureaucratic growth, scientific journals became straightjackets; legal systems, strata-title rules, textbooks, school curricula, speaking and writing correctly, dog shows, even children’s games were pushed to the limit of complexity and had to conform to rules. Those in charge won the games of life, and gave competitors a hard time. God knows, world history and events closer to home, demonstrate what autocratic power can do, and the frightful consequences for the lives of millions and their children. This book on bee vision is but an allegory to illustrate the dangers. We are presently like Dante in Purgatory staggering across a miasma between stinky pools in sticky mud. So, science, with its tests of validity by experiment, demanding and reliance on strict impartial logic and advance wherever the ground is solid, excludes murky dealings and cabals in management. With its essential ingredients of honesty, accuracy and publicity, science seems to be the last hope on this planet, the only path to ensure the survival of general human welfare and freedoms. With similar but woolly slogans,

246

Chapter 12

many, like the French Revolution, soon failed. Christians and other religions have an excellent philosophy of hope to offer, but clearly do not meet the requirement for empirical tests of validity, self-monitoring, equality, democratic justice, a moderate work ethic, recycling, no war, a no-waste technology, and care for the planet for ever. Why bother if God is in charge. Government by authoritarian rules on scientific lines made by elite technocrats is also a dangerous road to loss of liberty. The only known path for survival of freedom is to be able to vote out of office those who show signs of behaving in any way not in the interests of our living planet. All the other philosophies and political systems grow the seeds of their own ultimate decay by maximizing populations, depleting resources, seeking perpetual growth, eventually descending into war and decay, causing ruination and universal hardship. Professors and scientists, therefore, with some of the necessary knowledge and training, have to face a moral dilemma when exploitation of science is at the expense of the future. If this reasoning is true, as dead civilisations clearly show it is, then our first priority is preservation of trust in science. ‘Trust in Science’ was the title of the Anniversary Address by Sir Paul Nurse, President of the Royal Society, December 2014. By the way, my moral dilemma with the superposition image and the colour vision of bees was quite small, but they were contested none the less. Trust appears in every aspect of the operations of a scientific institute. All appointments are made expecting that the staff will be totally trustworthy in the way they do their science, report it openly, and give credit to their students and reference sources. The second condition is the capability to do the job at the highest standard, to have read the literature and learned the techniques. Ethics and capability are insufficient alone, however, because the whole effort must be in a direction that is likely to yield the required result, and perhaps a surprising result. It is also essential to keep up with recent discoveries and new techniques. Keeping a research group on track requires someone with long

experience in the subject, and with great imagination and a questioning mind, rarely combined in the same person. However, trust is not sufficient. Power corrupts, and rides over trust. I learned a lot about power in a long career. Search for autonomy, freedom in research, assured funding and funds for others, took me to Australia. In education and research, unusually, power must be benign, and should be brief; the greater the power, the shorter the tenure. The secret of good power in academia is great generosity and respect received and given, which is easy to say but difficult to achieve. Public comment in the media must be tolerated, but not slander. Journalists, researchers with awkward demands, and activists, should be treated like precious prized bulls, essential for the future survival, but given limits. Lower management is best achieved by alternating technocrats with bureaucrats to operate the system, with councils, committees and boards of management democratically elected with limited tenure. Professors present a special problem, hence the German experience where growth in size and power of universities retained a structure with many assistants dependent on a single God-professor. Professors require a long tenure to achieve the necessary knowledge of the literature and expertise to reach the frontiers of knowledge. For this, they need privilege. Some systems provide the opportunity of all the academics to rise to the position of professor, based on their achievement. As everyone knows, there are professors and Professors, certainly a strange and interesting lot. When I was a PhD student, Vincent B. Wigglesworth was Quick Professor of Biology in Cambridge. He established and published The Principles of Insect Physiology (1965). Not many professors deserve the words in his Biographical Memoir: Many of us also regret the passing of what he stood for in science because we fear what has taken its place. He solved naturally posed problems in carefully crafted papers that he had written and illustrated. … Now we live in an age when research is managed, and ‘science’ is unpublished or even hidden, with untested



What We Learned

information reserved for industrial gain, business profit, (etc.) …. V.B.W.’s life shows that science for the public good comes from giving talented individuals the opportunity to solve problems with free dissemination of the results. [my emphasis] The world is richer because he had that opportunity. (Locke, 1996)

To achieve fundamental discoveries in science, this is the atmosphere and scene to be recovered. Brilliant youth will not tolerate anything else; they go elsewhere.

Management for discoveries by research students Bright young graduates in a hurry may take a year to change their lifestyle from competitive to collaborative. They learn from each other and from technicians who are initially more skilled. They discover that doing science requires foresight to look ahead towards a career, with new manual skills, and operating advanced equipment with patience and care. For some, learning to experiment successfully took a long time, and so they were left to learn at their own pace. My lot had accepted a studentship for 3 years to become competent in ways to analyse a relatively simple nervous system. For some, it was a good entry into a later career in hospital research. Often a topic and an experiment were suggested, which they took, and ran with. As undergraduate students, they had written a different essay every week. In research they started a long apprenticeship in several skills that may take years to perfect, and to write a thesis and acceptable papers. It is the practice of science that makes scientists. In general, my students kept their own data and published it themselves; that was a huge incentive. The current system of using the work of students and temporary staff to win the next round of grants is not the way to encourage innovation or a commitment to science. My research habit was to collaborate with paid as­ sistants who were expert in electron microscopy and electrophysiology. Students and staff grumbled that they had to do their own

247

routine work, but continuity and high standards were maintained for decades. Science by itself does not generate discovery, but bright well-trained youth can sometimes achieve it. My most important service was to provide background knowledge acquired in writing the book with Bullock (Bullock and Horridge, 1965), and provision of plenty of first-class equipment at the cutting edge of technology, technical help in learning to use it, and security in a well-found laboratory. Secondly, PhD students must feel secure and good about themselves, their project and their social set-up. They must not be lonely, or they droop, pine, lose heart and need a holiday. My students knew that a job awaited them, because the topic of single cell recording with microelectrodes filled with dye, accompanying microscopy, and data processing with a computer online, ancillary knowledge of the optics, and neuron circuitry, was the great new leap forward in the decades from 1950 to 1980. Many believed that an intense study of single neuron activity and neuronal organization would lead to explanations of behaviour, so every university had to appoint a graduate to teach neurobiology. They were overly hopeful, of course, as this book illustrates. This book is intended to show that standing on the shoulders of giants is a precarious enterprise, because they invented errors, and stuck with them. It also shows how enthusiastic study of bees over a century did not generate the required innovation. It also shows that returning to the same problems with new thoughts, better techniques and a persistent effort over decades, can prevail. At St Andrews and again at ANU, we had the advantage that the research group was relatively self-contained. We had plenty of cash, excellent library and reprint collections. Skilled technical assistants could be found because in St Andrews the only occupation for skilled local people was to be a golf caddy or university technician, and in Canberra we employed scientifically trained graduates. It was easy to form a lab-centred community where often several keen ones were experimenting all weekend. My advice to them was to learn from each other and

248

Chapter 12

from the technical staff, and do an experiment every day. To compensate for this dedication, provision of equipment was very generous, and discipline was very slack, so they generated their own work ethic, a responsible independence, and to respect ideas and possessions of others. However, with support from the group, innovation always came from individuals. It paid off, as demonstrated by the 37 who became professors. The best that governments can do is to appoint brilliant dedicated young leaders, and fund them well for 10 years. Nowadays, the very brightest go to the few elite establishments that are fully funded. Referees The story of referees is different for each of the sciences. Bee vision had little to do with referees in the early 20th century. The high esteem of German science did not derive from referees, for they hardly existed in German science until about 1960; the editor ran the show. In the 1970s, I joined von Frisch and Autrum as an editor of the principal journal in the topic, the Journal of Comparative Physiology, when it shifted into English. I sent a few papers that I did not understand to suitable referees, and checked the rest myself. Referees became a problem for me only when I introduced feature detectors and cues. From 1989 to about 1995, our publications on bee visual discriminations were mostly rubbish, as I soon discovered, but the referees had failed to notice. As soon as I started using extensive testing for cues in 1995, referees were bitterly opposed to the conclusions because they differed from the orthodox story inferred from spontaneous preference or learning performance only. It was OK to make advances but not if existing texts and tradition was questioned. The problem disappeared when I switched to an American journal with a physiological tradition. What came to be the orthodox view about bee behaviour was supported by von Frisch from 1923 after the death of von Hess; then after 1982 by his pupils and their pupils. Referees became significant only after

1995, when the orthodox view was seriously threatened. To find suitable referees, editors would commonly use the names in the reference list. As a result, a group of orthodox authors always recommended each other as referees and would never cite any author likely to be critical. For bee vision, this was a serious problem from 1990 to 2006. However, referees did not slow down innovation because an author can shift to another journal. For bee vision, there were no better ideas or methods available, and no alternative was published until the new paradigm in 2014–2016. At the heart of the complaints about referees are two intractable problems. First they have negative apprehension bias when the idea is to promote innovation (Fölster, 1995), especially when the new offering proves that they themselves were in error. Secondly, to be really expert, referees must be active in the topic, therefore having a conflict of interest. As a result, they never validate the findings of others, so the system of refereeing is now the very opposite of ‘nullius in verba’. Worse; current referees select future referees by deciding who publishes. The outstanding referee of bee visual behaviour was Miriam Lehrer, who checked every statement for its agreement with the views of von Frisch. She could not break out of the orthodox view, but made discoveries that did not conflict. Apart from grammar, I never saw a referee’s effort that resulted in a higher standard, and I doubt that the science of Newton, or Darwin, suffered for lack of referees. Judging by modern journal content, huge volumes of referee comments have done nothing to stem the flow of dull, irrelevant and (frankly useless) inconsequential papers. Journals nowadays seem to select papers by the place or department of origin, and by the market for the topic. Errors continually pass the bureaucratic gates. Editors should bear in mind that the requirement of science is to validate the observations and conclusions, for which a referee is usually useless. So, what alternatives are there in the current crowded fund-driven scene? None; however, the problem of referees has been sidelined by the use of prepublication



What We Learned

­ebsites, invited lectures, chapters in w books and by whole books, following the 19th-­century tradition. The need to validate the experiments and the conclusions is still there. Maybe granting bodies should continue only those projects that are validated by data from another source. In 2006, my experiments stopped until 2013, by which time other bee researchers had closed down their projects. A new series of experiments on colour vision generated the series of publications in 2014–2017. It was essential to get the discoveries out to establish precedence. I gave two plenary lectures to the International Society of Invertebrate Neurobiology at Tihany, in Hungary, and the Hungarian Academy published them quickly. There were five referees for the paper on polarity in PLoS One, none agreeing with another. Referees for this book looked in vain for scientific errors, and reluctantly considered it might be a good idea to have some new ideas. This is the historical way to deal with a serious upset in science, as repeatedly happened in the 19th century. I quote: ‘Discovery entails the abandonment of previously respected authorities. But when such an authority has the status of a national icon, matters proceed more slowly’, in case ‘the rhetoric of revolution threatened science’s claim to stability [my emphasis]’ (Clarke, 2014). From 1945 to 1980, refereeing in neurobiology was a benign occupation. When funds became short, however, and there was danger of losing a grant, it was not advisable to be disproved by a competitor. Everyone strived for more and more publications in the best journals, which put a double onus on the referee. Referees at a loss to know how to defend the orthodox usually aimed ad hominem, (i.e. they attacked the character of the author, not the content of the work). An editor will ignore this if it is drawn to their attention. In the case of bees, referees always quoted the very same stuff that the new data disproved. In that regime, this author responded fiercely to every detail of every point made by the referee, and carried on regardless.

249

Opportunity Now that we have a formal model of the inputs and interactions in bee vision, a large amount of new work beckons. None of the analysis of vision of colour was quantitative, none of the time constants, resolution of cues, persistence of memory, and all the rest of the detail, has been catalogued. There is a lot of follow-up work to do. Secondly, an opportunity has opened up to discover how neuron activity is correlated with the inferred pathways illustrated at the beginning of this chapter. For electrophysiology, large tropical bees might be more convenient than the honeybee. Some of the field observations will have to be repeated when the blue content of different flowers and their contrast against background have been measured. As happened with monkey brain, trained bees will be used for electrophysiology. The neuronal basis of the visual world of the bee will slowly become available. Reverse engineering offers a far greater opportunity. In 1962, I had found that large insects without a head learn a comfortable position and walking action independently for each leg. In the 1980s, Rodney Brooks and others at Massachusetts Institute of Technology copied insects to make walking vehicles with independent feedback from the stress on the feet. They carried a load over uneven ground where wheels could not pass. ‘Fundamentalist AI [artificial intelligence] … emphasizes ongoing physical interaction with the environment as the primary source of constraint on the design of intelligent systems’ (Brooks, 1990). In this effort, they discovered Moravec’s paradox, which says, contrary to common assumption, complex reasoning requires very little computer power, but low-level sensorimotor skills require enormous computation. Computers can play chess or calculate the future position of the sun, but cannot control a doll that walks, jumps and dances like a 2 year old. In my view, Moravec’s error was to think in terms of classical software and programming. If you could copy a bee in every detail, you would get an artificial bee; therefore it is a hardware problem. Animal ­performance

250

Chapter 12

is done with its own special hardware, heavily parallel, thick with feedback loops, instantly adapting to the environment and task in hand, and very messy to analyse and comprehend. Computers depend on absolute accuracy of coding and software, irrespective of hardware; animals depend on the accuracy and learning ability of the hardware, irrespective of the messiness and intensity of the signals. This is where insects can show us how to see, swim, jump or fly; as we analyse their control systems, we describe the anatomy exactly and the line-labelling of neurons; not much has been learned from the sequence of nerve impulses.

Natural Vision for Engineers Our familiar technical society has little between stupid automatically opening doors and human vision, which is too subtle to understand, except the rapidly developing machine vision for self-driving cars, which attempts to start with the camera and copy human visual performance, but also makes use of the optical flow field generated by a moving vehicle. Apparent angular velocity measured by an optical or radar device is inversely proportional to the range. To steer and prevent a crash, range is what we need to monitor, but optic flow is useless for identification of the obstacle, which must be done by a different discrimination system. A few tasks that are well defined within simple limits, for example reading car number plates or recognizing faces or finger prints, can be done by specialized scanners that use features that were previously detected by humans, speeded up and refined. An example is the use of the total length of edge in a fixed area, exactly as used by the bee, as a first pass when identifying fingerprints. Superimposed layers of detectors connected together can be arranged with feedback, so that deeper layers control the inputs, but this is just a way of collapsing a pattern to simplify identification of the input. Nothing is added and it needs a final assessor, which in natural vision was the process of natural selection.

Bee vision appears to be designed for the rapid recognition of the place where food was previously found. Understanding bee vision may be a short cut to using tricks and avoiding flaws in ultra-lightweight machine vision of medium complexity in similar tasks. Examples are already in use for measurements of crop growth and ripening, detection of ­illegal crops, or even swarms of locusts, by satellite. There are immense applications of similar systems in industrial preparation of food and drinks, detection of wilt in crops or significant detail in robot surgery. The real problem emerges when we try to make use of the detail and relations between parts of a camera image that records detail but sees nothing. The meaning in the relations between every pixel and every other pixel must be realized, but the task is impossible because the number of possible combinations defeats the computer. Natural vision starts with simple primitive features, which are processed in various ways in parallel. Some are merely summed in groups, others are antagonistic to each other, some are sorted by colour, and others lie along edges. Then we look for commonly occurring coincidences. The next step is to detect standard features and components, such as corners, hubs of closed shapes or radial spokes, eyes, outside dimensions, heads on bodies, cars, motorbikes and pedestrians, and so on, depending on the task, but it soon fills memory, increases search time in memory, and it is all ad hoc. A great improvement is to let an automatic learning machine discover by trial and error what can be detected that might be relevant in each task. This is fine if important features occur repeatedly, but what happens if the very significant detail occurs only once, like a key word in a telephone conversation. The listening computer must be preloaded with all possible significant words, and then be able to look in immediate memory to see what came before. This would not defeat a modern computer but looking for combinations of words or understanding meaning in rapid speech would soon slow the detection process. We do not understand how this so-called combinatorial explosion in perception is so



What We Learned

easily processed by the human brain, and in higher mammals and birds. My own view is that we learn a great many preformed images and sentences at our mother’s knee, and store them ready to be recalled when triggered by an exact combination that we see or hear. Some people can see letters and words flow along as they stare into a black hole after reading a lot. To me that huge store explains the long learning period in animals that appear intelligent or can distinguish a lot of images. Yet, wait a moment, if education is just about stuffing in more memories and skills, it explains why innovation is so difficult. Innovation requires a long period of learning, as well as a great deal of thought, and what is thought if not a search through existing ­memory. Bees have a mechanism of vision that evolved over 400 million years. It enables them to fly around in a safe, sensible way without crashing, without seeing the panorama, without identifying objects, things, colours or patterns, and without a huge brain, but conversely, a bee cannot play chess or read text. In a similar world to ours, with similar primary visual inputs (Fig. 12.1), bees have a totally different visual world. They measure total inputs in each input channel, they detect contrast multiplied by edge length to measure modulation. They total the optic flow to measure distance flown and scan modulation to detect asymmetry. They detect horizontal position of blue relative to a green contrast, and its polarity. The cues are partially independent of range, but a far cry from the vision of higher vertebrates. The explanation of how bees make decisions that are fast, decisive, accurate and predictive still eludes us. This is the essence of the problem of explaining behaviour. There are many other visual systems to investigate, in octopus, fish, frog, lizard,

251

turtle and many others. Our present understanding of these animals is at square one (Lasareva et al., 2012).

Adieu It would be wise to remember von Frisch, not as a disgrace, but as an enthusiastic lover of nature, and popularizer, and that he cared for his staff and students, as long as they agreed with him. At the time, there was no synthesis of honeybee vision. He nurtured a high standard of scientific publications and studies of bee behaviour by others through extraordinary difficult times, and survived. Ariel: Was’t well done? Prospero: Bravely, my diligence. Thou shalt be free. Alonso: This is as strange a maze as ere men trod, And there is more in this business than ever nature was conduct of; Some oracle must rectify our knowledge. … Prospero: Do not infest your mind with beating on the strangeness of this business. At picked leisure, which shall be shortly, Single, I’ll resolve you … of every of these happened accidents; till then be cheerful, and think of each thing well. Come hither spirit; untie the spell. Bees detect and measure coincidences of responses of simple feature detectors. Ariel: Where the bee sucks, there suck I; In the cowslip’s bell I lie: … After summer, merrily: Merrily, merrily shall I live now, Under the blossom that hangs on the bough. (William Shakespeare, The Tempest; Final Scene)

References Andersen, H.C. (1998) The Emperor’s New Clothes. Retelling in English by Robyn M. Sheahan. Jam Roll Press, Nundah, Queensland, Australia. Blackford, R. (2018) The Tyranny of Opinion: Conformity and the Future of Liberalism. Bloomsbury, London and New York.

252

Chapter 12

Brooks, R.A. (1990) Elephants don’t play chess. Robotics and Autonomous Systems 6, 3–15. Bullock, T.H. and Horridge, G.A. (1965) Structure and Function in the Nervous Systems of Invertebrates. Freeman, San Francisco. Clarke, I. (2014) How to manage a revolution: Isaac Newton in the early twentieth century. Notes and R ­ ecords: the Royal Society Journal of the History of Science 68, 323–337. Fölster, S. (1995) The perils of peer review in economics and other sciences. Journal of Evolutionary Economics 5, 43–57. Ioannidis, J.P.A. (2005) Why most published research findings are false. PLoS Med 2(8), e124. Lasareva, O.F., Shimizu, T. and Wasserman, E.A. (2012) How Animals See The World. Oxford University Press, New York. Locke, M. (1996) Sir Vincent Brian Wigglesworth, C.B.E. Biographical Memoirs of Fellows of the Royal Society 42, 539–553. Nurse, P. (2015) Address of the President, Sir Paul Nurse, given at the Anniversary Meeting on 1 December 2014. Notes and Records of the Royal Society of London 69, 217–222. Ronacher, B. (1979) Äquivalenz zwischen Größen- und Helligkeitsunterschieden im Rahmen der visuellen Wahrnehmung der Honigbiene. Biological Cybernetics 32, 63–75. Shakespeare, W. (1923) The Complete Works of William Shakespeare. Collins Clear Type Press, London. Wigglesworth, V.B. (1965) The Principles of Insect Physiology. Revised 6th edn. Methuen & Co., London.

Appendix Training and Testing Bees

In the past, bees have usually been trained by a method that in detail was unique for each investigator, some of whom used quite different arrangements and sizes of the display. The standard method of encouraging the bees to search and look at the display, invented by Lubbock, Turner, and again by several others, was to move the display repeatedly as the bees learned and also while they were tested. This eliminated the cues relating to place, but other cues remained constant, and it slowed down the process of learning. In the early days, bees were scored as successful when they landed on the rewarded display (Figs 1.1, 1.4 and 2.1–2.5), but this made it impossible to determine the angle subtended by the target at the moment of choice. When targets are very large and fixed in position relative to the nearest landmarks, bees learn the position of blue and of green contrast relative to a landmark in one visit, and if the display is moved or changed, they fail, relearn or simply go away. With black/white patterns in the mid-20th century, this led to the false conclusion that the whole pattern was projected into memory, although only one or two cues had been learned. In some cases, patterns were very small (about 1 cm), so bees made a decision when near (Fig. 2.1); in other examples, each pattern subtended 130° at the eye of the bee at the moment of decision (Figs 6.8

and 8.7). Four methods are illustrated here (Fig. A.1). Hertz (1939) knew well that the size of the target at the moment of choice influences what bees detect and learn. With small targets and any target with fine detail of edges or lines, bees measured the total modulation of green receptors with targets presented on a horizontal plane (Fig. 6.5). With targets presented vertically I found the stimulus to be vertical edges crossed at each single scan in the horizontal plane, so tree trunks make good landmarks. The orientation of the display was a major factor for some cues. In his long review, Wehner (1981) believed that before his effort, displays were horizontal, but von Frisch (1914) used vertical display for patterns and horizontal for colours. Baumgärtner (1928; Fig. 2.1) also used a vertical display. For almost all her publications, Hertz used horizontal displays. Zerrahn (1933) and Hertz (1933) showed that bees distinguished three types of pattern irrespective of detail: (i) circular (tangential); (ii) star-like (radial); and (iii) disrupted irregular (neither tangential nor radial). The fourth type, (iv) orientation of single straight edges or lines, or a group of them parallel, was not discovered until a vertical display was used (Wiechert, 1938; ­ Jander, et al., 1970, on a wasp; Srinivasan and Lehrer, 1988; van Hateren et al., 1990). However, the bees used an unsuspected input, the

© A. Horridge 2019. The Discovery of a Visual System: the Honeybee (A. Horridge)

253

254 Appendix

(C)

(A) 1914

Elevation 1968

130° 20 cm

Bees land on the target

Reward box

Plan view

11

cm

Targets

Transparent

(B)

(D) Partitions define range

Target

27

10 cm 1996

Air

cm

27 cm

27

10 cm

screen

Reward hole

1928

No reward

cm

Choice chamber

Transparent baffle Bees fly in here

Fig. A.1.  Various earlier apparatus for training bees. (A) Von Frisch boxes. (B) Baumgartner. (C) Wehner. (D) Horridge. Bee training box with complete control of the visual input.

measure of ­modulation induced by a scan, which differs for each inclination of an edge, and had to be eliminated before the orientation sense was demonstrated (Fig. 6.5). More recently, new ideas generated a variety of ways of presenting displays, usually with the idea of eliminating the bees’ use of odours. For example, a completely different piece of apparatus was used for the training and testing, often with clean new targets for every test. All were unsatisfactory because the criterion of success was the landing on the target, and no tests were done to discover exactly what cues the bees were using. In fact, any method of training is satisfactory if successful, and the whole idea of a simple cue led to the use of repeated test displays at a fixed distance (Figs 3.2 and 6.1). The surprising point is that every result obtained by this great variety of training

methods has for a century been accepted as valid. Bees can now be trained on the original patterns and then tested properly to discover the cues, and most of the old data can be fitted into current ideas because only a very few cues are available in bee vision, and all training and testing methods ultimately reveal the same ones.

A New Apparatus for Measurement of Resolution In earlier experiments, the bees learned the cues at an unknown range while they prepared to land, and made their choice with the external panorama around them, except in Wehner’s (1967) experiments (Fig. 6.8A, B). To measure the resolution of bee vision required a new apparatus in which flying



Training and Testing Bees

255

Fig. A.2.  Y-shaped choice chamber as in Fig. 3.2 with yellow and blue striped patterns on the target.

bees chose between two targets at a known distance (Srinivasan and Lehrer, 1988). The Y-choice apparatus (Figs 3.2, 6.1 and 9.5) has been used for a great variety of experiments with carefully controlled conditions. The base and side are of wood, the top of thin transparent polycarbonate that absorbs ultraviolet (UV). The whole is light enough to carry easily and is sheltered under a polycarbonate roof. In experiments after 1994, a thin transparent Perspex film was added across the opening of each choice-channel. This baffle stopped the bees in flight and forced them to pause and look. Hovering behind the baffles, bees detected patterns at a fixed range and fixed angular size, so calculations of resolution were improved. By chance, 1 cm on the target subtended 2° at the eye of the bee. To change the apparent size of patterns, some researchers adjusted the lengths of the arms for the whole experiment, but changes of range between training and testing should be avoided because bees can measure range and absolute size if the task demands it (Horridge et al., 1992). At first, it was feared that this new apparatus might restrict the visual angle of the bees like blinkers, but they scan as they fly

and they pause at the baffles as if they scan the targets. Because the targets are shuffled between the two arms, they learn to search and look, and to ignore features that are the same on each target. Also they tolerate test patterns that differ from the training patterns. This change in their behaviour slows learning but improves identification and characterization of cues. They also learn to ignore odours because it is the pattern that is rewarded, not the odour or the place. Controls are essential: for example, a check that there is no side-preference with two identical displays, and a switch to a clean new display that does not carry an odour. The apparatus was placed under a shelter so the bees could not use the compass of the blue sky when inside, and was lined with clean white paper, which was originally intended to help the bees find the patterns, but leaves plenty of vertical and horizontal edges which the bees can use to stabilize themselves in flight and orient themselves with reference to the patterns. The bees must look for the expected positions of the parameters by use of a frame of reference within the box, usually the internal edges and the hole at the centre of each target. I did not realize the significance of these changes in the task until, about 1998, the retinotopic position of a vertical edge, or

256 Appendix

Fig. A.3.  Y-shaped choice chamber like that shown in the diagram for Fig. 3.2 with reward box in foreground.

the hub in radial patterns, was recognized as a parameter. The reward is a sugar solution that is adjusted between 3% and 9% w/w so that trained bees keep coming back, but naïve bees are not recruited. It is essential to have an additional feeder at a distance of 5–10 m, filled with weaker sugar solution as a lure for bees that are unwanted in the vicinity. Initially, bees are brought into the apparatus by use of a blue feeder that is carried by hand in 1 m steps from the hive, right into the box. After 1995, a stream of air was drawn out at each side so the bees made their choice in clean air without interference from new odours. However, air extraction made no difference to the results because if there is an odour that might favour one display it is dispersed when the patterns are changed and when displays change sides, so the bees are trained all the time to ignore odours and look at displays. During training,

the two sides were interchanged every 5 min (10 min before 1995) so the bees could not learn which side to go, and so must look at the targets, which can both be seen from the choice chamber. The bees learn the geometry of the inside of the apparatus and learn to prefer the rewarded display, not the side where they were last rewarded. Usually 2 h of training is sufficient. Training takes time because they have to learn to look at the display and ignore everything else. Naïve bees usually head first towards blue emissions and green contrasts before they reach the baffles. If they choose correctly, they have learned nothing because they find the reward without effort. If they fail to find the reward, they are forced to exit that side and go to the other. At this point, they learn by trial and error to avoid the display that gives them nothing. In contrast to the popular belief that they learn when rewarded, bees usually learn the unrewarded target first, because learning is by trial and error, and they recognize their error when not rewarded. Learning at first is not Freudian association of the decision with the reward; it may become that after days of learning, but initially it is trial and error. To investigate what the bees learned, they are given a variety of different tests with unfamiliar patterns on the targets. Several different tests are intercalated so that they see a given test only once or twice a day. When they arrive, they look at one test pattern and then at the other if the first is not recognized. By watching them in the choice chamber, one can see whether they decide quickly, or whether they spend a long time looking. In each experiment the aim is to see whether the bees can do the task or not after a reasonable period of training. They were allowed only one test visit with a test pattern on each side, then training resumed for 20 min before a different test was displayed. It is preferable to use many tests in a sequence so that the bees cannot learn any one of them. Usually a small group of individually marked bees was trained on a Monday morning and each experiment lasted a week. The success of the experiments depended very much on first identifying a set



Training and Testing Bees

of effective training patterns. By use of training patterns equiluminant for either blue or the green receptors, it was possible to train bees excluding green or blue receptor pathways. Bees could be trained to measure the modulation input by training on two gratings of similar colour but differing period. Bees ignored cues that were the same on both targets because they learned on one and unlearned on the other. Successful discriminations of colours provided no information about the mechanism. It was essential to give a succession of different tests, so that they could not learn any one test. As a result, it was convenient to give a variety of different tests that supported the final conclusion. Care was necessary when bees passed a test because they may have learned a cue that the experimenter did not notice or intend. Care was also necessary when bees failed a test because there may have been mutually antagonistic cues that cancelled. Failures by the bees (not significantly different from 50% of choices) provided the most useful information, showing that they detected nothing in the test that they had learned in the training, or that inputs cancelled out. The main effort was to determine whether trained bees passed or failed each test. It is important to understand that the result of each test is a unique piece of data for that pair of test patterns, and unrelated to the other tests in that experiment. The scores were not comparable because each is a forced choice between two unfamiliar targets, so in an ideal world, the bees would be 50% or 100% correct (i.e. fail or pass every time). Therefore, in each test, only a significant pass or fail was required, so we need to know whether each test score was theoretically different from 50%. Every effort was made to design training displays and tests that gave clear yes/no answers. From the result of all the tests in each experiment, it was possible to deduce by simple logic exactly what the bees had detected, learned and later recognized from that pair of training patterns. With continued training and other tests intervening, each test was continued until 100–200 counts

257

had been made. Statistics are not necessary because the result soon becomes obvious if the training and test patterns were cleverly designed. Validation by use of different patterns is usually easy, and if necessary, counting can be continued until obviously significant. Fancy statistics are rarely necessary. The data are frequencies, so if statistics are required, standard deviations are calculated from the formula s.d. = √[p.(1 − p)/n] where p is the measure of probability of a correct response, and n is the number of observations. This formula is valid when the choices of the bees are independent and the scores have no trend. Scores of more than 0.57 (57%) for n = 200, or a score of 0.60 (60%) for n = 100, is more than two standard deviations greater than chance (p < 0.05); which is acceptable.

Calibration of the Photon Emission of the Coloured Papers in Sunlight Eyes catch the light with photopigments that absorb it as photons. One photon at a time is absorbed by individual rhodopsin molecules. The correct measure of the intensity is the photon flux, which can be measured as the number of photons at each wavelength arriving per second per square centimetre on the surface, or the number absorbed per second per receptor cell. Receptor types differ in absolute sensitivity, so the best that can be done is to make measurements in the constant conditions used in the experiments, and calculate values of the stimulus to each receptor type from each paper, relative to white paper. Photon flux can be measured by a commercial calibrated photon spectrometer that gives a digital output and draws a graph of the emission spectrum of the paper in sunlight. This data must then be multiplied, at each wavelength by the spectral sensitivity of each receptor type and the whole added up to give a number of photons per second across the whole spectrum for that receptor from that paper, relative to white paper. All display papers were calibrated for blue and green receptors (Table 5.1).

258 Appendix

In the visual region, humans see each wavelength as a different colour but there are colours such as grey, brown and black, which are not found in the spectrum. You hallucinate them and black; in fact, you hallucinate all colours with your large brain. Bees cannot do that, so they detect black by its edges and measure its width; for the rest, they measure blue content. Colours are usually considered to be different ratios of photon flux at different wavelengths. Bees have a constant order

of preference in the learning process depending on blue content, and the green receptor response is a modulation of the signal, not photon flux. So, with UV excluded, my experiments tested whether bees used either blue photon flux or modulation caused by scanning edges with green or blue receptors or both. There were only three types of input for all recognition (Fig. 12.1), and the angles between them (Figs 3.11B, 5.15, 7.6 and 12.5M, N).

References Baumgärtner, H. (1928) Der Formensinn und der Sehschärfe der Bienen. Zeitschrift für vergleichende Physiologie 7, 56–143. Hertz, M. (1933) Über figurale Intensität und Qualitäten in der optische Wahrnehmung der Biene. Biologische Zentralblatte 53, 10–40. Hertz, M. (1939) New experiments on colour vision in bees. Journal of Experimental Biology 16, 1–8. Horridge, G.A., Zhang, S.W. and Lehrer, M. (1992) Bees can combine range and visual angle to estimate ­absolute size. Philosophical Transactions of the Royal Society of London B 337, 49–57. Jander, R., Fabritius, M. and Fabritius, M. (1970) Die Bedeutung von Gliederung und Kantenrichtung für die visuelle Formunterscheidung der Wespe Dolichovespula saxonica am Flugloch. Zeitschrift für Tierpsychologie 27, 881–893. Srinivasan, M.V. and Lehrer, M. (1988) Spatial acuity of honeybee vision, and its spectral properties. Journal of Comparative Physiology A 162, 159–172. van Hateren, J.H., Srinivasan, M.V. and Wait, P.B. (1990) Pattern recognition in bees, orientation discrimination. Journal of Comparative Physiology A 167, 649–654. von Frisch, K. (1914) Der Farbensinn und Formensinn der Bienen. Zoologische Jahrbücher. Abteilung für allgemeine Zoologie und Physiologie der Tiere 35, 1–188. Wehner, R. (1967) Pattern recognition in bees. Nature, London 215, 1244–1248. Wehner, R. (1981) Spatial vision in arthropods. In: Autrum, H. (ed.) Handbook of Sensory Physiology, Volume VII/ Part 6C: Vision in Invertebrates. Springer, Berlin, pp. 287–616. Wiechert, E. (1938) Zur Frage der Koordinaten des subjectiven Sehraumes der Biene. Zeitschrift für vergleichende Physiologie 25, 455–493. Zerrahn, G. (1933) Formdressur und Formunterscheidung bei der Honigbiene. Zeitschrift für vergleichende Physiologie 20, 117–150.

Author Index

Note: bold page numbers indicate figures. Anderson, Hans Christian  231 Arikawa, K.  7 Aristotle  198, 209, 243 Autrum, Hans  24, 25, 28, 39, 57, 62, 66, 248 Avarguès-Weber, A.  34

Dittrich, M.  40 Dohrn, Anton  3 D¢ving, K.B.  65 Duft, U.  151 Dyer, A.G.  7

Backhaus, W.  24 Baerends, G.P.  202, 218, 219 Barlow, H.B.  177 Bauer, E.  226, 227 Baumgärtner, H.  9, 13–14, 24, 54, 253, 254 Benard, J.  159 Benndorf, Hans  213 Bertholf, L.M.  19 Bethe, Albrecht  3, 12, 24, 198–199, 202 Bischof, S.  31 Bogdany, Franz  23, 25 Brooks, Rodney  249 Bullock, T.H.  63, 247

Efler, D.  151 Esch, Harald  9, 184, 212, 214, 216, 228, 243 Exner family  3, 9, 224 Exner, Karl  4, 61, 62, 224 Exner, Sigmund  4, 44, 57, 59, 61–64, 66, 72–74, 85, 224, 225, 242

Cajal, S.R.  57 Campan, R.  123 Carricaburu, P.  63, 64, 243 Chittka, L.  160 Cicero 216 Clarke, I.  249 Cleary, P.  66 Collett, Tom S.  151, 217

Darwin, Charles  1, 11, 198, 248 Daumer, K.  9, 19, 20, 215

Fabre, J.H.  198 Fischer, E.  226, 227 Flecker, James Elroy  90 Forel, Auguste  2, 8, 12, 24, 61, 199 Friedlaender, Marianne  9, 15, 19, 20, 25, 54, 91

Gibson, J.J.  194 Giger, A.  36, 54 Giurfa, Martin  14, 32–33, 54, 159, 243 Goldschmidt, Richard  3, 12, 17, 224–225, 226 Goldsmith, Richard  225 Gould, James L.  7, 212, 214, 218 Greggers, U.  2 Grenacher, Georg Hermann  45, 61, 72 Grindley, G.C.  194 Guiraud, M.  160–161

259

260

Author Index

Haeckel, Ernst  3, 228 Hardie, Roger  40, 53, 90, 236 Hassenstein, B.  19, 175 Hecht, Selig  17, 110, 175 Heisenberg, Martin  177, 179, 182 Helmholz, Hermann von  182, 194 Hempel di Ibarra, N.  33–34 Hertwig family  3, 9, 44, 225 Hertz, Mathilde  9, 12, 16, 17–20, 18, 24, 35, 40, 106, 110, 118, 119, 130, 151, 224, 225, 226, 253 Heyes, Tony  192 Hooke, R.  44 Horridge, G.A.  63, 135, 159, 160, 189, 254

Ioannidis, J.P.A.  242

Jander, R.  9, 115, 116, 155, 228, 243

Kennedy, John  175, 182, 189 Kirchhoffer, O.  62–63 Kirschfeld, Kuno  46, 47 Köhler, Otto  15, 16 Koltermann, R.  22, 23, 24, 25 Komissar, Alexander  35 Kriston, Irmgard  22, 25, 90, 243 Kühn, Alfred  12, 15 Kuiper, J.W.  62, 243 Kunze, Peter  62, 63–64, 65–66

Labhart, Tom  214 Land, M.F.  54, 67–70, 85 Laughlin, S.B.  40, 90, 236 Lehrer, Miriam  30, 31–32, 54, 90, 111, 118, 123, 183, 192, 203, 248 Lenz, F.  226, 227 Levick, W.R.  177 Lindauer, M.  22, 89, 160, 214, 216 Locke, M.  247 Lopatina, N.G.  23 Lorenz, Konrad  211, 229 Lotmar, Ruth  9, 15–16, 20, 25, 52 Lovell, J.H.  3, 4 Lubbock, John  1–2, 24, 35, 216, 253

Maeterlinck, M.  209 Marcelja, Ljerka  236 Meinertzhagen, I.A.  236 Menzel, Randolph  2, 22, 24, 25, 89, 160, 215–216, 218, 226–227, 243 Mill, J.S.  121 Miller, S.  30 Miller, W.H.  65

Nagle, Martin  192, 193 Neumeyer, Christa  21 Niggebrügge, C.  33–34 Ninham, Barry  47 Nurse, Paul  1, 25, 28, 44, 89, 106, 135, 150, 173, 197, 222, 246

Osorio, D.  236

Piaget, J.  229 Pièron, H.  199 Plateau, Felix  2, 3, 12, 24 Ploetz, Alfred  228 Popper, K.  243

Rabaud, E.  202 Reichardt, Werner  176–177, 179 Riehle, Alex  24 Romanes, George  198, 216 Ronacher, Bernhard  38–39, 40, 136, 151, 242, 243 Rossel, Sam  214

Sanchez, S.D.  57 Sandeman, David  179, 209 Sander, W.  19 Santschi, Felix  200, 209, 213 Schnetter, B.  115 Schultze, Max Sigmund  44–45, 72 Seitz, G.  63 Shakespeare, William  251 Siedl, R.  54 Snyder, Allan  47–48 Sobey, Peter  192, 193, 195 Srinivasan, M.V.  18, 30, 36, 54, 90, 111, 112, 156, 169, 182–183, 184, 192, 194, 195, 212, 214, 216, 236, 242, 243 Stach, S.  159, 160

Tautz, Jürgen  209, 215, 216 Thorpe, Bill  10, 19 Turner, Charles  3, 4, 207, 253 Tyndall, J.  212

van Hateren, J.H.  111 von Buttel-Reepen, H.  199 von Frisch, Karl  2, 3–9, 4, 7, 10–11, 12, 14, 19, 20, 24, 25, 28, 31, 32, 34, 35, 37, 44, 50, 89, 93, 118, 160, 184, 202, 209, 211–212, 213, 214, 215, 222, 224, 225, 228, 229, 242, 243, 244, 248, 251, 253, 254



Author Index

von Gavel, Lotte  175 von Helversen, Otto  9, 20–21 von Hess, Carl  2, 3–4, 4, 7, 8–9, 10–11, 12, 15, 24, 25, 34, 90, 222, 224, 225, 243, 244 von Zwehl, Vera  28, 39 Vorobyev, M.  32–33

Wells, H./Wells, P.H.  35 Wenner, Adrian  211–212, 216 Wiechert, E.  9, 20, 25 Wigglesworth, Vicent B.  246–247 Wolf, Ernst  9, 17, 24, 110, 175, 177, 179, 182

Yonge, C.D.  216 Wakakuwa, M.  40 Wehner, Rudiger  19, 111–112, 115–117, 151, 155, 167–168, 182, 209, 214, 243, 253, 254, 254 Weichert, Elsbeth  16–17

Zeil, J.  151 Zerrahn, Gertrud  9, 15, 17, 20, 25, 118, 253 Zhang, S.W.  158, 159, 192, 204, 243

261

Subject Index

Note: bold page numbers indicate figures; italic page numbers indicate tables. absolute conditioning  32–34, 32 acceptance angle  77 achromatic vision  7, 25, 31, 33, 34, 38–39, 39 acone eyes  44–45, 59, 60 acuity, visual  17, 33, 49, 75, 242 adaptation  62, 64, 64, 65, 67 to colour change  21 rapid  17, 49 see also dark-adapted eyes afocal lens/optics  61, 63, 64 Agrotis spp.  66–67 AI (artificial intelligence)  249–250 see also robots alder fly (Raphidia) 77 Ammophila 202 angular field/span  13 angular sensitivity  33, 50, 70, 71, 84, 114–115 by day/night  75–76 angular size  31, 32, 33, 132, 151, 156, 169, 200, 207, 255 angular velocity  49, 175, 176, 177, 182, 183–187, 250 and landing  192 and military/industrial applications  194 and moving/stationary wall experiment  183–184, 184 and optic flow  184–185, 186 and range  185, 185, 186 and route finding  200, 211, 214, 215 anomalies in colour constancy  21 in colour vision  2 in contrast  23 in discrimination of black spots  38–39, 39

in Giurfa group  32–35 after training to grey  37 in von Frisch students  20 Anoplognathus  61, 66–67, 71, 75, 76, 77 anthropomorphism  21, 53, 89, 243 anti-Semitism  224, 226, 242 ants  1, 46, 50, 179, 199–200, 200, 204, 216 desert (Cataglyphis)  185, 200, 208, 209, 215 ANU see Australian National University aperture  45, 46, 49, 53, 60, 72, 78 of night-adapted eyes  174 and ray tracing  74 and resolution–sensitivity trade-off 74, 79, 82, 84 applications of visual systems research  244, 250–251 aids for the blind  192–193 drones 194 fingerprint readers/facial/number plate recognition  250 machine vision  250 military/industrial 193–195 mobile robots  193–195 silicon chip technology  192–193, 195 and trial and error  250 World Wide Web  46–48 see also radar; reverse engineering; robots apposition eyes  46–47, 60 aquatic insects optomotor response in  175 see also giant water bug; water beetle arc lamps  12, 29, 69 Archichauliodes 64, 64

263

264

Subject Index

areas  10, 19 discrimination of colour of  13, 14 and edges, differences in detection of  18–19 arthropods  45, 52, 177–179, 197, 200, 231–232 red-/yellow-sensitive receptors in  232 visualizing 3D world of  238–241 artificial flowers, experiments with  2 artificial intelligence see AI astigmatism  13, 54, 91, 236 Atelophlebia  77, 78–82, 83, 84, 85 Australia  199, 230 Australian National University (ANU, Canberra)  177, 183–184, 195, 215, 231 funding/research team at  183, 192, 193, 195, 223, 225, 247–248 impact of Nazi Germany in  223, 226–227 research on colour in  22, 24, 25, 28, 30, 31, 90 research on compound eyes in  47, 61, 66 research on context in  24 research on feature detectors in  106 research on pattern/shape in  18, 117, 118

bars, thin  16, 101, 106, 112–114, 115–116, 167, 168 Baumgärtner apparatus  254 bee wars  211–212, 215–216 beekeepers 3 beelines  199, 200, 200, 211, 218 beetles  45, 46, 60, 62 Bell Laboratories  48 Berlin 243 see also Kaiser Wilhelm Institute binocular vision  174 black centre of  146, 146 and green/blue contrast  126 lack of receptors for  28, 31 and modulation  111–112, 113 versus blue  93, 94 black, position of  7, 25, 109, 112, 116, 120, 130, 154, 155, 165, 201 peripheral 111–112, 113 black/white patterns  20, 25, 31, 31, 33, 37, 40, 106–107, 118, 126, 253 and symmetry/asymmetry  138, 146–147, 146, 147 Blattoidea 60 blind/visually impaired people  192–193 blowflies  70, 182 blue  7–8, 16, 22, 34, 36–37, 107, 232 at edges  6, 7, 9 preference for  1, 2, 4, 7, 32 and recognition at hive entrance  35–36, 35 and spots  36–37, 37 versus white  90

blue content  9, 16, 37, 37, 97, 106, 107, 205, 241, 243 and absence of green contrast  102, 102 and blue modulation  101, 102–103, 103 in grey  4, 6, 7, 12, 21, 91 and size of spots  40, 41 total  21, 34, 37, 40, 41, 96–97, 96, 151, 153, 155 of white  7, 14, 15, 25, 40, 98, 106 and width  103–104 blue content relative to green contrast  6, 13, 32, 32, 34, 36, 51–52, 53, 251 absence of green contrast  102, 102 and discrimination of yellow  14, 93 and landmarks  33, 143, 146, 253 measurement of  92 blue contrast  7, 40, 97, 106, 111, 232 absence of  37, 39, 41 and green contrast  98 blue, height of  51, 98–99, 128, 161, 232, 236 blue modulation  31, 40, 96, 97, 97, 98, 111 and absence of green contrast  102, 102 and blue content  101, 102–103, 103 and green modulation  40, 99, 100–102, 101 and width  103–104 blue paper, spectral emission of  30, 89 blue, position of  36, 37, 97, 126, 166, 239, 253 average  97, 98, 142, 156 below reward hole  14 direction of polarity of  13 horizontal 251 relative to green/blue contrast  126, 251 and symmetry/asymmetry  139, 141–143, 144, 146, 147 vertical  16, 51, 98–99, 128, 143, 144, 147, 151, 154, 156, 161, 232, 234, 235, 236 blue receptor channel  30, 31, 35, 110, 111, 126 blue receptors  21, 28, 36, 98, 138, 232, 234 and changes in brightness  49–50 and polarized light  51–52 in pseudocone eyes  60 blue-green 5, 6, 7 blue/yellow squares  13, 13, 90 brain 57, 57 see also medulla; mushroom bodies; neurons; optic lobe brightness, discrimination of  1, 2, 9, 15, 34, 49–50 in human vision  17 and motion perception  17 see also adaptation Britain (UK)  57, 182, 215, 226, 231 see also St Andrews buff  33, 36, 37, 37, 91, 92–93, 97, 102 versus blue  91–96, 152–153



Subject Index

buff paper, spectral emission of  30, 89 butterflies  45, 46, 53, 85, 185, 213, 231 see also skipper butterflies

cabbage white butterfly (Pieris) 45 Campan, R.  123 Canberra (Australia) see Australian National University Canson papers  30, 89, 91–92, 138 cantharid beetles  60, 77, 81, 85 Carcinus  179, 186–187 Cataglyphis  185, 200, 208, 209, 215 CCD (charged couple device) camera systems 192–193 centering response  184 cephalopods  2, 3 cerebrum 57 Chauliognathus 77, 81 Chernobyl disaster (1987)  193 chlorophyll 5 chromatic contrast  13, 21, 22, 32–34 simultaneous 21–22 see also green contrast Chrysopa 76–77, 80 Cicindelinae 62 circular patterns  18, 18, 19, 31, 54, 118, 119, 120, 135 clamped head, insects’ vision with  104, 176, 178 clear zone eyes  60–82 endopterygote 64–82 and Exner’s contribution  61–64 lens cylinders in  61–62, 64, 74 superposition in see superposition eyes Cloeon 77–78, 82 cockroaches  46, 197, 213 cognition  34, 42, 160, 229 colour constancy  21 colour intensity, measurement of  90 colour opponency  12 colour preferences  1, 2, 4, 7, 21, 22, 32, 152–154, 232 and brightness  1, 2 right/left see right/left discrimination colour triangle  19, 20, 32, 41 colour vision  232, 243, 246, 258 and changes in light intensity  1 Frisch paradigm of  11, 20 new paradigm for  241 and radial/tangential cues  120, 123 colour-blindness  2, 4, 5, 9, 10, 53, 203, 235, 238 in fish  3 in humans  1, 8 coloured light, responses to  21, 22, 25 compound eye  44–85, 45 acone 44–45, 59, 60 apposition  46–47, 60

265

array of optical axes  44, 45, 46 bee eye performance  49–50 clear zone see clear zone eyes cone 45–46, 45, 47 corneal cone see cones, corneal/cone cells dorsal rim organ  50–51, 50, 174, 213–214 eucone 46 and F value  17, 49, 79 in flies/bugs  60 light guide optics in  46–48, 48 microvilli 47, 49 and motion perception  174, 179 ommatidia see ommatidia optic lobe see optic lobe polarized light sensitivity of see polarized light, sensitivity to pseudocone  46, 60 resolution of  13, 14, 53–54, 53, 54 retina see retina rhabdoms/rhabdomeres  45, 46, 47, 53 and UV  50, 50, 51, 52–53, 60, 80, 81, 82, 85, 174, 213 very small  82–84 concentric circles  118, 119, 124, 126 conditioning, Freudian  136 cones, corneal/cone cells  44–46, 58 in clear zone eyes  60–61, 62, 63, 65, 71 crystalline 65, 70, 71, 74, 77, 78, 80 context 24 contrast  17, 31, 126 and modulation  99, 100, 110 see also chromatic contrast; edge contrast contrast frequency  177, 182, 184, 185, 186 and velocity parallax  187–189 cornea  44, 45, 46, 48, 58, 59 in clear zone eyes  61, 62, 65 curvature of  47, 71 refractive index in see refractive index in cornea see also cones, corneal/cone cells corpora pedunculata see mushroom bodies counting ability of bees  154 crabs  40, 177, 179, 197 Carcinus  179, 186–187, 201 crickets  50, 197, 214 crosses (test pattern)  112–114, 113, 155–156, 157, 164–166, 165, 166 crystalline cones  65, 70, 71, 74, 77, 78, 80 crystalline tracts  59, 61, 62, 64, 64, 65, 71, 74, 75, 76, 79, 81, 83, 84, 85 of Anoplognathus  75 of Archichauliodes 64, 64 of Chauliognathus  81 of Chrysopa 76 of Ditiscus 74, 79

266

Subject Index

crystalline tracts (continued ) of firefly  61, 62–63 of Hemiptera  59 of Macrogyrus 75 of mayflies  83, 85 of moths  65, 84 of skipper butterflies  71 cues 130–133, 131, 150, 151, 169, 171, 253 and avoidance of cue absent from training  132 different states of  22, 23, 107 and feature detector resolution  132 learning of several  132–133 linkage of coincidental see context and modulation  232–233, 236 order of  130–131 salience versus retinotopic  132 structure, symbolic representation of  236–238, 237 switching 34, 95, 142, 199, 211 visualized as humans see them  238, 240

dances of bees  23, 187, 189, 198, 202, 209–212, 210, 213, 214, 215, 216 and bee wars  211–212 dark-adapted bees  25, 49, 60 dark-adapted eyes  49, 60, 62, 62, 63, 65, 66–67, 74, 75–76 of Chauliognathus  81 of Chrysopa 76, 80 of dung beetles/Agrotis spp.  66–67 of Dytiscus  78 of Ephestia 65, 65, 66, 68 of fireflies  62, 62, 63 of Lethocerus 60 of Macrogyrus 75–76, 80 of skipper butterflies  71, 71, 74 DARPA (Defence Advanced Research Projects Agency) 194 dead reckoning  186–187, 198, 199–201, 209 deep optic lobe  23, 40–41, 114 democracy 246 depolarization 46, 55, 81 dichroic molecules  213 differential conditioning  34 direction of motion  51 directional response, righting response as  2 discs 31 blue 14, 14, 217 size of  38 distance travelled, measurement of  51, 184, 185–186, 188, 212, 214–216 and contrast frequency/velocity parallax 187–189 and opening/closing parallax  187 dorsal rim organ  50–51, 50, 174, 213–214

dragonflies  46, 52, 53, 58, 174, 185, 191, 197, 213, 231 drone aircraft  194 Drosophila  60, 175, 177, 179, 180, 182, 185, 189, 192, 200, 225 neuron anatomy of  234, 236 dung beetle (Anoplognathus)  61, 66–67, 71, 75, 76, 77 Dytiscus  45, 61, 72–74, 78, 79

earwigs  45, 60 edge contrast  8, 9, 16, 17, 21, 91, 232 and recognition of hive entrance  35–36 edge detectors  114, 115, 123, 124–125, 129, 180, 187, 236 edge length  9, 10, 17–18, 19, 20, 32, 96, 99, 110, 251 and fingerprint readers  250 ignored by authors  30 and orientation detectors  111, 112, 114, 115 and spot size  40, 41 edge, moving  17, 18, 40, 51, 91, 111 edge orientation  18, 22, 25, 30, 36, 36, 91, 108, 112–115, 115, 117, 131, 235–236 at all angles  44, 107 at different angles  44, 107, 114–115 at right angles  114, 121–122, 122, 155–157, 157 average 40, 56, 163 differences of, in separate eye regions 159–161 and green receptor channel  235 edges  4, 5, 10, 16, 18–19, 151, 255 and blue  6, 7, 9, 12 fuzzy 114 and modulation/movement  17, 18, 40, 51, 91, 111 and route finding  201 of shadows  51, 106, 179 tangential see tangential edges/edge detectors editors, duties of  248 eidetic image  111, 115, 121, 133 electrophysiology  23, 23–24, 24, 40, 75, 85, 186, 247, 249 emission spectra  28–30, 29, 30, 33, 89 Endopterygota insects  64–82, 90 Ephemeroptera  52, 60 Ephestia  64, 65–67, 65, 68, 69, 242 diagram of eye of  69 eyeshine in  65–66 ray tracing in  66, 67, 68 equiluminant colours  28–31, 90, 91, 93, 242 erect image  61, 64, 65, 74 escape response  2 ethics  225, 246



Subject Index

ethology 229 eucone eyes  46 eugenics 224, 227 Exner line  67, 68, 74 experiment design  231, 238 faults in  158, 159–160 see also mazes eyeshine 65–66, 70, 71, 72 apparatus for measuring  69

F number  17, 49, 54, 79 Failure to recognise colour  91, 94 rewarded pattern  143, 140, 145, 148 shape  161, 162, 163, 164, 165, 166, 167, 168, 169, 170 symmetry  137, 149, 148 failures in tests, usefulness of  10–11 feature detectors  25, 28, 41, 42, 107, 133, 138, 236, 237 and field size  54 graphic definition of  125–126, 128 and medulla  56 and optic lobe  107 resolution by  132 and shape/pattern discrimination  159, 160, 169, 170–171 feedback loops  176–177, 176 fibre optics  48 figural quality  18, 18, 19 filters, neutral/grey  1 fireflies (Lampyridae)  44–45, 46, 59, 60–61, 62, 63–64, 64, 85 firefly (Lampyris)  44–45, 61, 62, 63–64 firefly (Photuris) 61, 62, 63, 63 fish  3, 4, 8, 241, 251 flies  40, 46, 53, 58, 60, 180–181, 185, 191–192, 231; see also specific fly species flight, visual control of  173–195 and angular velocity see angular velocity and binocular vision  174 and contrast frequency see contrast frequency and distance travelled see distance travelled, measurement of and height  189, 190 and hovering  191 and landing  191–192 and measurement of range  181, 182–183, 183 and military/industrial applications  193–195 and motion perception see motion perception/detection and navigation see piloting and optic flow see optic flow and optomotor response see optomotor response

267

and responses to light  173–179 righting response  2, 51 and saccades see saccades and speed of flight  190–191 and visual aid application  192–193 floral guides  14, 104, 174 odours as  2, 22, 23, 182, 212, 217 flour moth see Ephestia flower-like patterns, experiments with  118–130 and coloured spots  119–120, 121 and defining number of cues  120–121, 122 and discrimination with changing states/ cues 127–129, 129, 130 and discrimination of squares/circles  119 and hub position  120, 124–125, 125, 126, 127, 129 and number of cues  120–121 and radial patterns  123–125, 126 shortcomings of  123, 133 and symmetry  115, 118, 119, 128, 131 and tangential/radial bars  121–122, 123 and visualization of feature detectors  125–126, 133 see also shape/pattern discrimination flowers, evolution of  118, 153, 238 foraging  3, 9, 12, 197, 201 finding route home from see route finding and measurement of distance travelled see distance travelled, measurement of and symmetry/asymmetry  147–149 fovea  58, 191 France  199, 243 Frankfurt (Germany)  20, 22, 214, 216, 230 Freiburg Zoological Congress (1914)  7–8, 9 Fujitsu Computer Company  193, 194 funding of research  183, 192, 193, 195, 222, 223, 248 and bad effects of grant applications  247 drags on  244–245 self- 245 fused rhabdoms/rhabdomeres  45, 46, 47, 62, 76

ganglia/ganglion cells  23, 40, 55 Gatty Marine Laboratory see St Andrews (Scotland) geometrical patterns see pattern perception; shape/pattern discrimination Germany, research in  199, 223–230, 243, 248 in 19th century  223–224 in First World War/period following  224 Nazi period see Nazi Germany power of professors in  222–223, 224–225, 246 subversion of scientific research in  223 Gestalt theory  19 giant water bug (Lethocerus) 60

268

Subject Index

Goldschmidt, Richard  3, 12, 17, 224–225, 226 grant applications see funding of research grasshoppers  46, 181, 181 grating test patterns  17, 20, 30, 31, 40, 91 and memory  129–130 minimum resolution period for  49, 51 and modulation  30–31, 31, 93, 99, 100–102, 101, 102, 110–111, 110 green 5, 6, 14, 21, 22 preference for  32 receptor channel for  30, 35 versus black  95, 96 versus grey/white  93–96, 94, 95 versus grey/yellow  94 versus sequence of black/grey/white  95, 96 green contrast  13, 13, 22, 32–33, 34, 100–102, 161, 253 absence of, and blue content/modulation  102, 102 at edges  4, 6, 16, 21, 23, 25, 33, 33, 36, 53, 96, 106, 159 and blue content see blue content and green contrast and blue contrast  98 detectors of  13, 18 minimum measure for detection of  104 and motion perception  36, 117, 177, 182, 183, 184 and orientation  111 and recognition of hive entrance  35–36 relative to blue see blue content and green contrast; blue content relative to green contrast and spot colour/size  36–37, 37, 40 and symmetry/asymmetry  138, 139, 140, 143, 144, 145 green modulation  7, 22, 32, 33, 37, 40, 54, 92, 126, 151, 153, 232–233 at edges  25, 40, 90, 106–107, 162, 232 and blue modulation  40, 99, 100–102, 101 discrimination with no difference in  91–92 and receptor channel adaptation  49 and symmetry/asymmetry  138, 139, 140, 144, 145, 146, 147, 233 green paper, emission spectrum of  89 green receptor channel  30, 35, 90, 91, 91, 107, 110, 124, 126, 232–235, 234 and edges in relation to hub  235 green receptors  19, 28, 34, 36, 37, 40, 90–91, 111, 138, 232 and changes in brightness  49–50 and edge orientation see edge orientation and polarized light  51 in pseudocone eyes  60 grey  1, 2–3, 4–7, 4, 6, 28, 34 and inhibitory effect of UV  19

and level of blue content  4, 6, 7, 12, 21, 31, 35, 37, 91, 96–97 set out in single array  7 and spot colour/size  37, 37, 38–39, 39 Gröningen University (Netherlands)  63

Haldane principle  222 half-discs 25 head movement  174, 175, 176–179, 180, 181 prevention of, in experiments  104, 177, 178, 179 Heidelberg (Germany)  20 Hemiptera  59, 60 hemipterans 45 Hering series  5, 6 Hesperiidae see skipper butterflies heterochromatic flicker  24, 30 Himmler, Heinrich  228, 229 Hitler, Adolf  224, 225, 226, 227, 228 hive entrance, recognition of  35–36, 35 honey, attraction to  1, 5 horizontal scanning  13, 17, 25, 33, 46, 51, 124, 138, 181 horizontal surfaces, and modulation  18 horizontal/vertical rotation see rotation, detection of Horridge apparatus  254 housefly (Musca)  53, 176–177, 179, 180 hoverflies  45, 191 hub position  120, 124–125, 125, 126, 127, 129, 131, 151, 235, 256 and robot vision  241 human vision  2, 4, 49, 241, 258 and anthropomorphism  21, 53, 89, 90, 243 and colour-blindness  1 and contrast  99 and cues  238, 240 and estimation of range  181 Hydrophilus 61

ignoble prizes  vii Imperial College London (UK)  182 innateness concept  229 innovation  viii, xiii, 244–245, 248, 251 insects  5, 24, 118, 231–232 applications of research on see applications of visual systems research aquatic see aquatic insects compound eye of see compound eye motion perception in see motion perception/detection nocturnal see nocturnal insects optomotor response in see optomotor response primitive  63, 64, 197



Subject Index

and reverse engineering  249–250 visualizing 3D world of  240–241 see also specific insects interneurons 184 interommatidial angle  13, 54, 54, 175, 177, 178 intuition 243

Japan  193, 194 journals  177, 225, 245 German  9, 11, 12 and National Socialism  226, 227, 229 and referees  118, 248–249

Kaiser Wilhelm Institute (Berlin)  17, 224, 225, 226 Kiev (Ukraine)  35

lacewing (Chrysopa) 76–77, 80 lamina  40, 41, 51, 55, 56, 57, 118, 232, 236 Lampyris  44–45, 61, 62, 63–64 landmarks  2, 3, 24, 115, 129, 197, 200, 200, 201–203, 207, 211, 215, 216, 217, 218 and modulation  110 and recognition of hive entrance  35–36 retinotopic learning of  177, 201 learning in bees  3, 8, 10, 107, 199, 236 and landmarks  2, 3 and retinotopic vision  19 and switching to new colours  22–23, 23 see also mazes; training/testing bees left/right discrimination see right/left discrimination Lethocerus 60 light guides/light guide optics  46–48, 48, 49, 54 applications of  47–48 and clear zone eyes  60, 61, 62, 63, 63 light intensity see brightness, discrimination of lobula  55, 56, 57 locusts  47, 57, 70, 174, 177, 179, 182, 185, 225, 231 Lovell, J.H.  3, 4

Macrogyrus 75–76, 80 magnetic field  198, 203, 216–217 mantids  46, 174, 181, 182, 188–189 maps, internal/cognitive  217–219 Max Planck Institute (Germany)  12, 47, 176–177, 223, 224 Maxwell colour triangle  19, 20, 32, 41 mayflies (Ephemeroptera)  60, 85, 182 mayfly (Atelophlebia)  77, 78–82, 83, 84, 85 mayfly (Cloeon) 77–78, 82 mazes  24, 204–207, 206, 216, 217, 218, 219 see also Y-choice maze

269

mechanoreceptors  173, 185, 191 medulla 51, 55, 56, 57, 57, 60, 82, 118, 156, 161, 177, 232, 234, 239 evolution of  118 and neural circuits  179, 235 neurons of  147, 236 orientation detectors of  238 serial sectioning of  177 memory  17, 57, 106, 129–130, 151, 205, 207, 231, 251 destruction of (retroactive interference)  22–24 limits to understanding of  241–242 optokinetic 183–185 retinotopic see retinotopic vision/memory and route finding see under route finding and time of day  24 microelectrode recordings  52, 54, 60, 67, 71, 82, 84, 231, 242, 247 microvilli 47, 49 mirror images  15, 15, 36, 36, 131 and polarity  137–138, 139, 140 of spots  143–144, 144, 145, 154 symmetrical/asymmetrical  140, 141, 141, 146–147, 146, 147 triangles with vertical edge  162–163, 163 misplaced concreteness  24 modulation  7, 16, 17–18, 33, 90–91, 96, 104, 110–118, 124, 243 and angular sensitivity see angular sensitivity at boundaries  9, 12, 40 and contrast  99, 100, 111 and deep optic lobe  40–41 detectors 110, 128, 232, 233, 234, 237, 238 different states of  107 and grating patterns see under grating test patterns irrespective of pattern  111 and orientation  30–31, 31, 111, 112–114, 112 and peripheral position of black  111–112, 113, 116, 117 and polarization of light  51 receptor 20 symmetry/asymmetry  140, 141, 142, 143, 144–147, 146, 147 visualization of  241, 241 see also blue modulation; green modulation Moiré effect  175 monochromatic vision  23 Moravec’s paradox  249–250 mosquitoes  175, 182 moths  60, 61, 64, 182 see also Ephestia; Phalaenoides glycine motion detectors  114, 175, 177, 179, 180

270

Subject Index

motion perception/detection  17, 33, 36, 173, 235 and contrast frequency  184 and direction  51 and green contrast  36, 117, 177, 182, 183 neural circuit for  177, 179 reverse engineering of  179 see also angular velocity movement perception  19 Münich (Germany)  20, 28, 44, 57, 224, 226, 230 Münich University  2, 3, 8, 9, 12, 28, 44 Musca  53, 176–177, 179, 180 mushroom bodies (corpora pedunculata)  55, 57, 57, 160, 161, 197

naïve bees  98, 167, 203, 208, 229 and order of preference  22, 151–152 and symmetry  119, 135, 136, 143 and trial and error learning  152, 256 Nasonov gland  8, 203 National Socialism see Nazi Germany navigation  12–13, 173, 197–219 Nazi Germany  223, 224, 225–229, 242 destruction of science/academia in  225, 226–228, 229 eugenics/Social Darwinism in  226, 227, 228, 229 pseudoscience in  228–229 and von Frisch  211, 226, 228 neuron fields  23, 24, 116 neurons  23–24, 41, 57, 89, 125–126, 187, 197, 198 analyser 214 and behaviour  236, 247 collector  237 disparity 174 interactions of  232, 234, 235, 236 lamina 232, 235, 236–238, 239 large-field  24, 40, 56 line-labelling of  250 olfactory 240 and optic flow  186 and optomotor response  177, 179 orientation detector  114, 116 peripheral connections of  235 phasic 241 and pitch  225 research approaches for  231, 247, 249 second-order 40, 56, 58, 90 voltage oscillations in  107 neutral filter  1 newspaper controversy  215–216 nocturnal insects  62, 83–84, 85, 104 see also dark-adapted eyes noise-limited theory  33

object recognition  35, 184 ocelli  55, 57, 174, 231 octopus  2, 251 odometer see distance, measurement of odours  2, 8, 22, 23, 161, 182, 185, 217, 240 and route finding see under route finding spatial arrangements of  212 and training apparatus  255, 256 ommatidia  28, 46, 48, 49, 58, 232, 234 apparatus for mapping axes of  47 in clear zone eyes  61, 62, 62, 64, 65, 70, 75, 80, 81, 82, 83 and optomotor response  175–176 and orientation detectors  114 and polarized light  50, 51, 213, 214 and resolution  54 three types of visual input from  233 see also retinula cells optic flow  4, 17, 173, 179, 182, 241, 250, 251 and angular velocity  184–185 applications of  192–194 and drone aircraft  194 and flight height  189 and flight speed  191 neurons coding for  186 and route finding  207, 209, 210–211, 215, 216 optic lobe  24, 41, 51, 56, 57–60, 231, 238 and clear zone eyes  61 and feature detectors  107, 126 and lifestyle  57, 58 serial sectioning of  177, 179 see also deep optic lobe optics, light guide see light guides/light guide optics optokinetic memory  183–185 optomotor response  173, 174–179, 181, 184, 192, 229, 242 at different light intensities  175 and feedback loops  176–177, 176 and fixed head experiments  177, 178 and learned muscle control  179 motion stimulus not required in  177–179 and motion of sun/moon  179 and ommatidia  175–176 reversal of  178 slowness of  175, 177, 180 orange 22 orientation 54, 108, 169, 203, 233–235 edge see edge orientation effect of range/proximity on  157–159, 157, 158 as least preferred cue  107 and modulation  30–31, 54, 111, 112–114, 112 not learned/recognised  16, 22, 31, 36, 102, 108, 109 in training of bees  253–254



Subject Index

orientation detectors  18, 111, 112–114, 112, 128, 160, 233–234, 238 and angular sensitivity see angular sensitivity arrays of  18 length of  114, 116, 126 misunderstandings with  115–118 overpopulation  230, 245 owl flies  52

Papaver  15, 53 paper, coloured  90, 183 calibration of photon emission of  257–258 emission spectra of  28–30, 29, 30, 33, 89 paradigms 11 new, introduction of  20, 21, 28, 241, 242, 243, 248 orthodox  12, 21, 34, 225 pattern perception see shape/pattern discrimination pattern types  18, 18, 19 Pavlov Institute (Leningrad)  23 performance, study of  10, 13, 25, 89, 107, 118, 123, 173–174, 199, 229, 242 and intuition  243 and mazes  204, 205, 207 and navigation  207, 208, 209, 212, 219 petal numbers  14 Phalaenoides  66, 67–70, 70, 78 PhD students  215, 224, 231, 244 management of  247–248 photon flux  13, 23, 33, 41, 90, 257, 258 photoreceptor, three types of see trichromatic vision phototaxis  1, 2, 17, 35, 242 and colour blindness  9 and foraging  3, 9 and UV  10 Photuris 61, 62, 63, 63 Pieris 45 piloting, and green contrast  173 polarity, detection of  147–149 left/right 13, 36, 128, 147–149, 203 yellow/blue 13, 13 polarized light, sensitivity to  50–53, 212–213, 217 and blue-sensitive receptor cell  51–52 and green receptors  51 and midday  50–51 and UV-sensitive receptor cell  52–53 and yellow/red receptor  53 pollen guides  14 poppy (Papaver)  15, 53 population control  230, 245 potato beetle  175–176 primates  170, 174, 241 pseudocone eyes  46, 60

271

queens, recognition of hive entrance by see hive entrance, recognition of

racism/racial cleansing  225–226, 229 radar  202, 216, 218, 251 radial bars  120–122, 122, 123 radial edges  123, 125, 128, 131, 159, 160 radial patterns  15, 18, 19, 22, 31, 36, 119, 120, 123–125, 126 innate preference for  18, 32, 118 see also spoked pattern randomized testing  110, 113 range, measurement of  157–159, 157, 158, 182–183, 183, 191, 255 as aid for visually impaired people  192–193 and angular velocity  185, 185, 186, 251 and route finding  203, 205 visualization of  241, 241 Raphidia 77 rates of change of optical input  231–232, 234, 241 ray tracing  59, 62, 63, 65, 66, 67, 74, 77 receptor axons  40, 41, 45, 59 in pseudocone eyes  60 receptor channels  24, 30, 31, 32, 110 adaptation to dark of  49 spectral sensitivity of  17 receptor modulation  20 receptors  47, 231 sensitivity/resolution of  54–57, 54, 55, 61 in vertebrates  53–54 red receptor  53 referees  25, 34, 118, 119 little used in Germany  8, 9, 225 unreliability of  248–249 refractive index in cornea  47, 60, 61, 62, 63, 64, 65, 66, 67, 68, 71, 72–73, 74, 79, 85 refugees 226–228 relative motion  192–193, 194, 208, 241 and brain  57 and range measurement  182, 183, 187, 203 research  222, 242–250 applications of see applications of visual systems research and experiment design see experiment design failures in tests, usefulness of  10–11 and fundamental discoveries  244–245 in Germany see Germany, research in grants/funding see funding of research and innovation  244, 248, 251 and management of PhD students  247–248 and opportunity  249–250 and power  222–223, 224–225, 244, 245 and referees  248–249

272

Subject Index

research (continued) and scientific orthodoxy see scientific orthodoxy and trust  245–247 resolution  13, 14, 53–54, 53, 54, 158 apparatus for measuring  254–257 resolution–sensitivity trade-off  54, 79, 82, 83–84 retina  13, 40, 44–47, 45, 56, 57, 65, 72–74, 79, 118 four types of  45–46 retinotopic vision/memory  15, 19, 34, 91, 107, 108–110, 109, 133, 141, 151, 177, 255 and measurement of turning  186–187 retinula cells  45, 48, 62, 64, 65, 65, 71, 74, 75, 80–81, 83, 214 and changes in intensity  23, 40 distal 75, 78, 79 extensions 60, 71, 78, 80, 82 as light guides  77–78, 81 recordings from  84 and UV  213, 214 retroactive interference  22–24, 23 reverse engineering  249 reward holes  14, 16, 16, 30, 31, 91 direct/indirect flight to  14 rhabdoms/rhabdomeres  45, 46, 47, 53, 60, 61, 62, 62, 65, 65, 71, 72, 74, 80, 81–82, 85, 213 in clear zone eyes  60, 61, 62, 62, 65, 65, 71, 72, 74, 80, 81–82, 85 fused  45, 46, 47, 62, 76 off-axis 77 in pseudocone/acone eyes  60 rhodopsin 47, 49, 90, 213 right/left discrimination  15, 118, 130 and detection of polarity  13, 36, 128, 147–149, 203 righting response  2, 51 robots  193, 194, 195, 241 and Moravec’s paradox  249–250 Ronacher’s enigma  38–39, 39, 40, 242 Rostock (Germany)  9, 45, 209, 224 rotation, detection of  16–17 route finding  197–219 and angular velocity  200, 211, 214, 215 and bee wars  211–212 and beelines  199, 200, 200, 211, 218 by trial and error  197, 201, 204–205, 218–219 and cognitive maps  217–219 and colour  201, 205, 217 and dances  198, 202, 209–211, 210, 213, 214, 215, 216 and dead reckoning  186–187, 198, 199–201, 209 and distance flown see distance travelled, measurement of

early observations on  198–199 and earth’s magnetic field  198, 203, 216–217 and first observational flights  207 and landmarks see landmarks and light polarization/UV  51, 198, 212–214, 217 in mazes  204–207, 206, 216, 217, 218, 219 and memory  179, 186–187, 197, 201, 202, 203, 218, 219 and motivational state  201 and odour  197, 198, 199, 200, 201, 202, 203, 209, 211, 212, 216, 217, 218 with one eye  208 and optic flow  207, 209, 210–211, 215, 216 and position of sun  198, 200, 202, 211 and radar technology  202, 216, 218 and range  203, 205 and signposts see signposts on routes and sky compass  20, 51, 200, 202, 209, 212, 214, 216 and stages of journey  203–204, 206 and symmetry/asymmetry  147–149, 203, 204, 239 and turn back and look (TBL) flights  207, 208, 217 Royal Society  vii, viii, ix, 25, 28, 48, 195, 222, 246

saccades  56, 176, 177, 179–182, 217 and edge detectors  180 and odour plumes  182 and peering/scanning before jump  181, 181 and scanning in flight  182 in wind tunnel experiment  180, 180 St Andrews (Scotland)  57, 225, 231, 248 scanning before a jump  181 and detection of modulation  102, 128, 235–236, 237 in flight  19, 110, 181, 182, 232 horizontal  13, 17, 25, 33, 46, 51, 124, 138, 186 and symmetrical detector  91 vertical 185 science 245–246 and bureaucracy  223, 230, 245 and democracy  246 and religion  246 and trust  245–247 and validation  119, 245, 248, 249, 257 see also research scientific orthodoxy  7, 12, 16, 20, 21, 32, 34–35, 66, 74, 216, 222–223 and hierarchical authority  12, 25 reasons for persistence of  242–243 and referees  248 see also paradigms; trichromatic theory



Subject Index

shadows  24, 51, 106, 179, 207, 235 edges of  51, 106, 179 shape/pattern discrimination  3, 10, 14, 18–20, 18, 42, 123, 124, 135, 150–171, 195, 243 diamond versus square  161–162, 162 and differences in edge orientation  159–161, 160, 169 disc versus triangle  163–164, 164 faults in experiments on  158, 159–160 and feature detectors  159, 160, 169, 170–171 global/local  159, 170–171 and green modulation  232, 235 lack of ability for  36, 61, 150–151, 154, 168, 170 and mirror images see mirror images misleading appearance of  150–151, 154, 161, 161, 169–170 O versus S  167, 168 and orientation  157–159 and randomized cue positions  169–170 and regions of eye  152 ring versus cross  164–165, 165 ring versus disc  156, 166–167 ring/cross versus plain white target  165–166, 166 and rotation of sectors  167–168, 169 shapes 3, 10, 14, 42 see also figural quality signposts on routes  2, 107, 138, 145, 147–149, 171, 185, 201, 203, 212–213 silicon chip technology  192–193, 195 simultaneous contrast  21–22 skipper butterflies (Hesperiidae)  60, 61, 63, 66, 70–71, 71, 72, 73, 74, 79, 85 afocal optics in  64 eyeshine in  72 superposition eyes of  60, 61, 63, 66, 70–71, 71, 85 sky compass  20, 51, 200, 202, 209, 212, 214, 216 Social Darwinism  226, 227, 228, 229 soldier beetle (Chauliognathus) 77, 81 Spanish flu epidemic  224 spectral bands, training in  12 spectral sensitivities  17, 28, 29, 52, 90 see also polarized light, sensitivity to spectrum of bees’ vision, and UV  15–16 spiders  174, 181, 197 spiral patterns  11, 111, 113, 120, 135 spoked pattern  18, 20, 36, 36, 54, 118, 119, 124, 125, 135 hubs of see hub position spot colour  36–38, 37, 38, 152–154, 152, 153 spot size  31–32, 32, 38–40, 39, 41, 54, 106 spots  13, 19, 22–23, 119–120, 121, 124 arrangements of  154, 155 detection of  13, 19, 22–23 versus rings  166–167, 167

273

subtense, minimum angular  14, 32–33 successes xiv Süddeutschen Zeitung 215–216 sun, determination of position of  50–51, 50 sunlight  28, 50 superposition eyes  60–61, 63–64, 85, 246 evolution of  63 and eyeshine  64, 65–66 in moths  64 Switzerland  199, 212 see also Zürich symmetry/asymmetry  42, 115, 118, 119, 128, 131, 135–149, 151, 243, 251 and avoidance of unrewarded patterns  135, 136–137, 140–141, 142 bilateral 118, 119, 120, 129, 132, 135–137, 137 and blue/fixed landmarks  138–146, 139 and centre of black  146, 146 and chevron pattern  135, 136 and earlier results  146–147, 146, 147 and green contrast/modulation  138, 139, 140, 144, 145, 146, 147 not recognised  141, 142, 147, 148, 149 and position of blue  139, 141–143, 144, 145, 146, 147 and recognition of polarity  137–138, 138, 139, 147–149, 203 and route finding  147–149, 203, 204, 239 and width as cue  139–140, 141 Syrphidae 45

tangential bars  120–122, 122, 123 tangential edges/edge detectors  118, 124, 125, 127, 131, 135, 151, 159, 160 and saccades  180 tangential patterns  18, 18 TBL (turn back and look) flights  207, 208, 217 Tempest, The (Shakespeare)  251 Tenodera 182 tiger beetles (Cicindelinae)  62 time, measurement  24 trading curve  38–39, 39 training/testing bees  1–2, 4, 7, 8, 14, 16, 183, 253–258 calibration of photon emission of papers  257–258 controls in  255 and cues  253, 254, 257 early apparatus for  254 and grey preferences  5 and landmarks  3 learning time for  256 and measurement of resolution  254–257 and odours  255, 256 and orientation of display  253–254

274

Subject Index

training/testing bees (continued ) scoring pass/fail/statistical analysis in  257 and size of target  253 and trial and error learning  256 see also mazes trial and error  7, 8, 15, 34, 92, 137, 197, 201, 204–205, 218–219, 229, 250, 256 trichromatic theory  7, 12, 24–25, 28, 32, 35, 89, 242 anomalies in  5, 6–7, 6, 9–10, 11, 14, 15, 15, 20, 22, 25, 28, 35, 36, 37 and intensity discrimination  17 and peak sensitivities of receptors  20–21 and spectral sensitivities  28 trust 245–247 turn back and look (TBL) flights  207, 208, 217

ultramarine 37, 38 ultraviolet (UV)  2, 7, 8, 10, 12, 22, 80, 104 absence of  31, 40 and brightness discrimination  15–16 inhibitory effect of  19, 20, 40, 52–53 innate preference for  32 and recognition of hive entrance  35 and sky compass  51, 213, 214 United States (USA)  199, 211, 212, 214, 218, 243 eugenics/Social Darwinism in  224, 226, 228 Jewish scientists in  226 unrewarded target  34, 136, 137, 140–143, 140 USAF (US Air Force)  194 UV (ultraviolet)  2, 7, 8, 10, 12, 22, 80, 104, 232 absence of  31, 40 and brightness discrimination  15–16 inhibitory effect of  19, 20, 40, 52–53 innate preference for  32 and recognition of hive entrance  35 and sky compass  51, 213, 214 UV receptors  28, 41, 60 and polarized light  51, 52–53

vertebrate vision  53–54, 133, 174, 213, 238, 241, 251 vertical presentation  16, 16 vertical/horizontal rotation see rotation, detection of Vespa germanica  117 Vienna (Austria)  3, 4, 8, 44, 61, 223, 224 violet  15, 21, 31, 32 vision in animals  2 visual acuity  17 visualization of visual interactions  238–243 and cues as humans see them  238, 240 and memory/decision-making  241–242

and persistence of scientific orthodoxy 242–243 and research history  242–243 wide visual field of  241, 2412 von Frisch boxes  254

waggle dance  209–211, 214, 215, 216 wasps  1, 46, 53, 58, 117, 117, 158, 253 digger (Ammophila) 202 route finding in  197, 207, 216, 217, 218, 219 water beetle  85, 175 Dytiscus  45, 61, 72–74 Hydrophilus 61 whirligig (Macrogyrus) 75–76, 80 Wehner apparatus  254 whirligig water beetle (Macrogyrus)  75–76, 80 white  12, 25 blue content of  7, 14, 15, 40, 98, 106 lack of receptors for  28, 31 preference for  90 and recognition of hive entrance  35 UV in  19, 20 versus blue  93, 94 width as cue  32, 34, 51, 90, 101, 103–104, 103, 110, 111, 129, 136 and overlap of visual fields of two eyes  13, 13 and spot size  40, 41 and symmetry/asymmetry  139–140, 140 women’s education  230 worker bee  24, 45, 46, 47, 53, 197, 238 spectral sensitivities in  28, 29, 52 World Wide Web  47–48 Würzburg (Germany)  2, 182, 215, 216, 228

Y-choice maze  30, 32, 107, 107, 111, 137, 151, 183, 203, 217, 219, 242, 254, 255, 255, 256 yellow  12, 20, 32, 32, 225 and black  8, 16, 161, 162 blue content of  7, 14, 143–144 discrimination of  92–93, 93 preference for  90 receptor 53 seen as black  8, 16 versus blue  8, 13, 15, 15 versus grey  6, 8 yellow paper, spectral emission of  30, 89

Zürich (Switzerland)  2, 30, 182, 183, 195, 209, 214

CABI – who we are and what we do This book is published by CABI, an international not-for-profit organisation that improves people’s lives worldwide by providing information and applying scientific expertise to solve problems in agriculture and the environment. CABI is also a global publisher producing key scientific publications, including world renowned databases, as well as compendia, books, ebooks and full text electronic resources. We publish content in a wide range of subject areas including: agriculture and crop science / animal and veterinary sciences / ecology and conservation / environmental science / horticulture and plant sciences / human health, food science and nutrition / international development / leisure and tourism. The profits from CABI’s publishing activities enable us to work with farming communities around the world, supporting them as they battle with poor soil, invasive species and pests and diseases, to improve their livelihoods and help provide food for an ever growing population. CABI is an international intergovernmental organisation, and we gratefully acknowledge the core financial support from our member countries (and lead agencies) including: Ministry of Agriculture People’s Republic of China

Discover more To read more about CABI’s work, please visit: www.cabi.org Browse our books at: www.cabi.org/bookshop, or explore our online products at: www.cabi.org/publishing-products Interested in writing for CABI? Find our author guidelines here: www.cabi.org/publishing-products/information-for-authors/