Parasitoid Population Biology 9780691230894

Citation preview

PARASITOID POPULATION BIOLOGY

PARASITOID POPULATION BIOLOGY

Edited by Michael E. Hochberg and Anthony R. Ives

PRINCETON

UNIVERSITY

PRESS

PRINCETON

AND

OXFORD

Copyright © 2000 by Princeton University Press Published by Princeton University Press, 41 William Street, Princeton, New Jersey 08540 In the United Kingdom: Princeton University Press, 3 Market Place, Woodstock, Oxfordshire 0X20 1SY All Rights Reserved Library of Congress Cataloging-in-Publication Data Parasitoid population biology / edited by Michael E. Hochberg and Anthony R. Ives. p. cm. Includes bibliographical references (p. ). ISBN 0-691-04981-5 (cloth : alk. paper) — ISBN 0-691-04982-3 (pbk. : alk. paper) 1. Parasitoids. 2. Insect populations. I. Hochberg, Michael E. II. Ives, Anthony R., 1961QL496.12.P37 2000 595.717'857—dc21 00-020748 This book has been composed in Times Roman The paper used in this publication meets the minimum requirements of ANSI/NISO Z39.48-1992 (R1997) (Permanence of Paper) www.pup.princeton.edu Printed in the United States of America 10

9 8 7 6 5 4 3 2 1

10 9 8 7 6 5 4 3 2 1 (Pbk.)

To Julien, Kevin, Joelle, and Mary Ann

Contents Preface List of Contributors One Introduction Michael E. Hochberg and Anthony R. Ives

xi xiii 3

PART ONE: POPULATION DYNAMICS

15

Two Host Location and Selection in the Field

17

Jerome Casas Three Effects of Parasitoid Clutch Size on Host-Parasitoid Population Dynamics George E. Heimpel Four Host-Parasitoid Models: The Story of a Successful Failure

27

41

Carlos Bernstein Five A Field Guide to Studying Spatial Pattern Formation in Host-Parasitoid Systems Susan Harrison

58

Six Parasitoid Spread: Lessons for and from Invasion Biology Alan Hastings

70

Seven Landscape Ecology of Parasitism

83

Jens Roland PART TWO: POPULATION DIVERSITY

101

Eight The Evolution of Parasitoid Egg Load Minus van Baalen

103

Viii

CONTENTS

Nine Host Resistance, Parasitoid Virulence, and Population Dynamics

121

H. C. J. Godfray Ten Developmental Traits and Life-History Evolution in Parasitoids Michael R. Strand Eleven Host Specificity and Trophic Relationships of Hyperparasitoids

139

163

Jacques Brodeur Twelve Comparing Parasitoid-Dominated Food Webs with Other Food Webs: Problems and Future Promises

184

Marcel Holyoak Thirteen Species Coexistence in Parasitoid Communities: Does Competition Matter? Bradford A. Hawkins

198

PART THREE: POPULATION APPLICATIONS

215

Fourteen Biological Control: The Need for Realistic Models and Experimental Approaches to Parasitoid Introductions Nick Mills Fifteen Parasitoid Populations in the Agricultural Landscape

217

235

Teja Tscharntke Sixteen Threats, Flies, and Protocol Gaps: Can Evolutionary Ecology Save Biological Control?

254

Bernard D. Roitberg Seventeen "What, Conserve Parasitoids?" Michael E. Hochberg

266

CONTENTS

IX

Eighteen Conclusions: Debating Parasitoid Population Biology over the Next Twenty Years Anthony R. Ives and Michael E. Hochberg

278

References

305

Index

359

Preface HosT-parasitoid systems are being employed at an ever-increasing rate as models of evolution, population and community dynamics, species diversity, and biological control. Thus, we hope that this volume is timely, both to give a constructively critical review of past work and to propose new and interesting venues for future research. We decided that an exciting way to achieve this was to ask each contributor to discuss a subject about which he or she has strong opinions, particularly if the subject is poorly represented in the current literature. We were looking for fresh perspectives on a growing field. With our invitation to contribute to this book, we gave contributors the following rules: Each chapter could have only a single author; this ensures that the ideas expressed in the chapters are not open to veto by coauthors. Each chapter should begin by presenting a brief review of empirical and theoretical work. Based on this review, the author should identify what he or she feels is the most exciting element for future research. This could be a question that remains unanswered, one that has been answered unsatisfactorily, or one unanswered from another field and that could be better pursued using parasitoids as model systems. The author should discuss in detail why the question is interesting and important, and make some headway toward proposing solutions. This could include a literature review, a comparative analysis, an analysis or re-analysis of data, and/or the development of mathematical models. The chapter should close with a short, critical discussion of what we should hope to achieve over the next twenty years or so. We leave it to you to judge, based on this sample of perspectives (chapters 2 through 17) and our own perspectives (chapters 1 and 18), the state of parasitoid population biology research today and where it is going tomorrow. We express our sincere thanks to the people of Princeton University Press for their encouragement and efficient help in the production of this book. Michael E. Hochberg Anthony R. Ives April 1999

List of Contributors Carlos Bernstein. Biometrie et Biologie Evolutive, University Claude Bernard-Lyon I, 43 Bd. du 11 Novembre 1918, 69622 Villeurbanne, France Jacques Brodeur. Centre de Recherche en Horticulture, Departement de Phytologie, Universite Laval, Sainte-Foy, Quebec, G1K 7P4, Canada Jerome Casas. Institut de Recherche sur la Biologie de l'lnsecte, IRBICNRS ESA 6035, Universite de Tours, 37200 Tours, France H. C. J. Godfray. Department of Biology and NERC Centre for Population Biology, Imperial College at Silwood Park, Ascot, Berkshire SL5 7PY, UK Susan Harrison. Department of Environmental Science and Policy, University of California, Davis, CA 95616, USA Alan Hastings. Department of Environmental Science and Policy, University of California, Davis, CA 95616, USA Bradford A. Hawkins. Department of Ecology and Evolutionary Biology, University of California, Irvine, CA 92697, USA George E. Heimpel. Department of Entomology, University of Minnesota, St. Paul, MN 55108, USA Michael E. Hochberg. Institut des Sciences de l'Evolution, Universite Montpellier 2, Place Eugene Bataillon, 34095 Montpellier Cedex 05, France Marcel Holyoak. Department of Entomology, University of California, Davis, CA 95616-8584, USA Anthony R. Ives. Department of Zoology, University of Wisconsin, Madison, WI 53706, USA Nick Mills. Department of Insect Biology, University of California, Berkeley, CA 94720-3112, USA Bernard D. Roitberg. Department of Biological Sciences, Simon Fraser University, Burnaby, BC, V5A 1S6, Canada Jens Roland. Department of Biological Sciences, University of Alberta, Edmonton, Alberta T6G 2E9, Canada Michael R. Strand. Department of Entomology, University of Wisconsin, Madison, WI 53706, USA

LIST OF C O N T R I B U T O R S

Teja Tscharntke. Agroecology, University of Gottingen, Waldweg 26, D-37073 Gottingen, Germany Minus van Baalen. Institut d'Ecologie, Universite Pierre et Marie Curie, 7 quai St. Bernard, 75252 Paris Cedex 05, France

PARASITOID POPULATION BIOLOGY

One Introduction MICHAEL E. HOCHBERG AND ANTHONY R. IVES

communicating "where we are" and "where do we go from here" is difficult in any discipline. This is particularly true for parasitoid population biology, because it is a relatively new and still diffuse field. The term "parasitoid" was coined less than one hundred years ago, and only in the past two decades has the term been widely accepted by biologists. Furthermore, population biology as we know it today is essentially a product of the last three or four decades. With relatively little history, it is difficult to predict how the field will develop in the future. It is not our intention to give an extensive historical recast of parasitoid population biology; there are numerous books that already do this, and interested readers are referred to Hassell (1978), Price (1980), Waage and Greathead (1986), LaSalle and Gauld (1993), Godfray (1994), Hawkins (1994), Hawkins and Sheehan (1994), Jervis and Kidd (1996), Quicke (1997), Barbosa (1998), and Hawkins and Cornell (1999). Rather, we believe that it is now time to dedicate a volume to asking how satisfactory our current understanding is and discussing the prospects for research into the twenty-first century. This book is composed of three parts: population dynamics, population diversity, and population applications. These rubrics reflect three main endeavors involving parasitoids. Population dynamics concerns how abundance changes in space and time, and what factors may be responsible for these changes. Parasitoid-based models are the most often used for understanding the population dynamics of exploiter-victim systems. Population diversity is the number and range of biologies that one finds within populations, assemblages of parasitoid species, and entire ecological communities. Here, parasitoids have established themselves as the prima facie model for diversity amongst the metazoa. Finally, population applications are the means by which parasitoids have found their way into the human world, usually in a positive fashion (e.g., as biological control agents), but sometimes in a negative one (e.g., as biological control agents that inadvertently attack nontarget species). Although parasitoids are the most employed agents for the biological control of insect pests, many still regard their application as much of an art form as a science. ACCURATELY

4

CHAPTER ONE

Population Dynamics For many population biologists, mention of the "Nicholson-Bailey model" conjures thoughts of instability and irrelevance. This has often led to its rejection as a useful descriptor of population dynamics. Nevertheless, we would argue that the vast majority of theoretical developments on parasitoid population biology unavoidably have antecedents in the work of A. J. Nicholson and V. A. Bailey (Nicholson and Bailey 1935; Bailey et al. 1962)1. Nicholson and Bailey demonstrated that the most basic property of hostparasitoid systems—persistence—is not an easy thing to explain, thus generating a legacy of attempts to explain persistent natural systems. Although a range of models are often presented as "Nicholson-Bailey" equations, the form accepted by most population ecologists is Nt+1 =

FNte"aP'

Pt+1 =

cN t (l - e- a P l ).

N and P are densities of host and parasitoid, respectively, which change from generation t to t + 7 if the system is not exactly at equilibrium. F is known as the "finite" or "net" rate of increase of the host population (i.e., host growth rate when the parasitoid is absent from the system), a is the "area of search" of the parasitoid, and c is the number of female adult parasitoids produced per parasitized host. Hassell (1978) and, more recently, Mills and Getz (1996) discuss the development of the model. The Nicholson-Bailey model poses an interesting puzzle: This, the simplest dynamic model for a tightly coupled host-parasitoid interaction, predicts growing population oscillations until the parasitoid becomes extinct due to sheer low densities, and the host either becomes extinct due to the action of the parasitoid, or exhibits unchecked growth if the parasitoid should become extinct first. Since it is obviously the case that some parasitoid populations exist, there must be one or more essential elements that are missing from the Nicholson-Bailey model. Thus, the problem boils down to knowing the range of biological, ecological, and environmental complexities needed for interacting parasitoids and hosts to persist. The challenge to population biologists has been to identify and evaluate the various possibilities. 1

It is somewhat unfair to cast the problem as one developed only by Nicholson and Bailey, since it should be acknowledged that there were diverse earlier works, that of Thompson (1924) being the most relevant to the problem of parasitoids (see Mills and Getz 1996 for historical discussion). Recent years have also seen a recognition that models of host age-structure and of overlapping host-age classes can produce different dynamics to those cast in discrete, nonoverlapping frameworks (i.e., Nicholson-Bailey models and their descendants) (see Briggs et al. 1999).

INTRODUCTION

5

What the overwhelming majority of models indicate is there must be density dependence, but not just of any form. For example, density-dependent reductions in parasitoid attack rates leading to a type II functional response tends to make the Nicholson-Bailey model even less stable. In contrast, density dependence increases stability when the parasitoid population experiences contest competition (Taylor 1988b), when the parasitoid inflicts mortality on the host in a heterogeneous fashion (Chesson and Murdoch 1986; Ives 1992b), or when the host population is regulated by other density dependent forces (Hochberg and Lawton 1990a). Spatial mechanisms of density dependence dominate the literature on host-parasitoid interactions. One of the much-studied ways to create spatial density dependence is through behavioral aggregation of adult parasitoids. At an individual behavioral level, Casas (chapter 2) begins with the premise that much of what we consider as "real" in adult parasitoid foraging behavior may be the artifacts of laboratory experimental designs. Employing real examples, he points to four ways in which a field approach may provide an essential understanding of host foraging behavior by parasitoids: (1) identification of new and important behavioral processes; (2) verification of processes widely accepted as important; (3) testing of theories developed in the laboratory or through the use of models; and (4) production of a priority list of processes requiring study. He believes that new technologies may make many once-impossible field studies on parasitoids a reality, but that there will remain insurmountable difficulties associated with parasitoid fieldwork that may impede progress. Once a host is located by a female parasitoid, a variety of behaviors may ensue. Studies on parasitoid population dynamics often overlook the decisions that must be made by adult females during oviposition, such as sex ratio and clutch size. The few studies investigating these have discovered diverse effects on population dynamics. According to Heimpel (chapter 3), what is missing from previous models is an interdependence of life-history parameters. Diverse biological parameters may be phenotypically associated, and because each on its own may affect parasitoid-host interactions, inclusion of more than one via positive or negative trade-offs may have complex effects on population dynamics. The possibilities are immense, and Heimpel limits analysis to tradeoffs between clutch size (the number of eggs deposited per host) and adult parasitoid attack rate mediated by egg limitation. He finds that egg limitation weakens the negative relationship between clutch size and host density, and he calls for future models to incorporate intraspecific variability in clutch size. The classic approach of varying a single parameter at a time and observing its effects on dynamics can be quite misleading, both for within- and betweenpopulation comparisons. His approach is all the more pertinent because the life-history and behavioral parameters that determine the population dynamics predicted by his models are known to be highly variable not only among species, but also among individuals within the same species.

6

CHAPTER ONE

Casas's empirical approach and Heimpel's more theoretical one point to how easily we may be misled by contrived experiments and simplistic models. Bernstein (chapter 4) addresses this directly by evaluating the utility of mathematical models in resolving the fundamental question of parasitoid population biology: What stabilizes host-parasitoid population dynamics? He traces models from Thompson's work in the 1920s to the present and highlights three apparent answers to this question: (1) sigmoidal functional responses; (2) mutual interference between adult parasitoids; and (3) spatial heterogeneity in parasitism. Bernstein's historical approach lucidly illustrates the scientific process: The sigmoidal functional response was promoted as a potential stabilizing mechanism for about fifteen years, mutual interference has had its ups and downs over the past thirty years, and although spatial heterogeneity received particular attention in the mid-1980s to early 1990s, it now appears to be receiving less attention. The latter two mechanisms are still generally accepted as potential candidates as the magic missing piece needed to stabilize host-parasitoid models, thereby resolving the contrast between instability in models and persistence in nature. Dovetailing nicely with Casas's and Heimpel's chapters, Bernstein prescribes that the way forward is to look at the behavioral nuts and bolts behind these nebulous stabilizing mechanisms. He does not dismiss the modeling approach, but rather says that for the future, it should have more behavioral realism and be testable in the field. Harrison and her colleagues have conducted the first field test of one of the classic phenomena predicted by mathematical theory: diffusion-driven instability. Harrison (chapter 5) discusses how a large body of theoretical work points to fairly distinct predictions of how predator-prey interactions may generate fixed spatial patterns in abundance. Models that include host and parasitoid movement as diffusion processes predict that when parasitoids disperse at greater rates than their hosts, it is possible for areas of high host abundance to arise, surrounded and constricted by a ring of high parasitism caused by dispersing parasitoids. Harrison and colleagues conducted manipulative field studies on the western tussock moth and members of its parasitoid assemblage to test this prediction. Their findings are strikingly consistent with the model predictions. Nonetheless, the models and reality differ in key ways, and more system-specific models need to be made to test whether the experiments do in fact show all of the ingredients needed to produce diffusion-driven spatial patterns. Harrison's chapter gives a clear example in which a striking natural pattern was first observed (the persistence of a spatially restricted tussock moth outbreak), theory provided a potential explanation (the more rapid dispersal of parasitoids than hosts from the center of the outbreak), and experiments confirmed many of the theoretical requirements for diffusion-driven spatial patterns. At a larger spatial scale than that considered by Bernstein and Harrison,

INTRODUCTION

7

Hastings (chapter 6) addresses the problem of how parasitoid populations spread through time. Specifically, he presents a biologically simple model for parasitoid diffusion over a landscape and is able to arrive at a simple formula for the rate of parasitoid spread, employing information only about the per capita rate of population increase and rate of movement. He presents five case studies for which data are sufficient to say something about parasitoid spread. Interestingly, in three of the four studies where it could be determined, parasitoids spread tens of kilometers per year! In the remaining study, slow rates of spread may be a consequence of the parasitoid and host having been introduced into the habitat simultaneously; thus, Hastings concludes that the spread of the parasitoid is limited by the spread of its host. Hastings's results are interesting because they provide indirect methods for assessing how individual movement rates translate into the overall rate of spread of the population. He suggests that movement studies combined with life-table analyses are a sensible way forward toward estimating the distance between the site of parasitoid emergence and where the parasitoids lay their eggs. In the future, Hastings's approach could prove important both for understanding regional and geographical patterns in population dynamics and community structure, and for the employment of parasitoids in biological control. How landscape patterns may affect parasitoid movement is discussed in detail by Roland (chapter 7). Over the past few years, Roland has designed an impressive series of experiments to uncover patterns in percent parasitism of the forest tent caterpillar over different spatial scales. Roland takes a landscape ecology perspective, explicitly focusing on the distribution of different types of habitat, both habitats containing forest tent caterpillars and habitats without. Roland points to three main features of landscapes: (1) composition, which is the characteristics of a given local habitat; (2) context, or the distribution of habitat types surrounding a local habitat; and (3) connectivity, which is determined by the global distribution of habitat types and how this distribution affects the ability of different parasitoids to navigate through the landscape. Roland argues that different features of landscapes affect different parasitoids in different ways, with each species experiencing landscape patterns at different spatial scales. Like Hastings, Roland calls for an integration of life-table studies with data on movement, but he concedes that this will be a difficult task to achieve.

Population Diversity Parasitoids are distinguished from taxa employing other resource exploitation strategies by their astronomical diversity among just a handful of families. It goes without saying that phenomena hors norm attract the curious,

8

CHAPTER

ONE

and evolutionary biological approaches have been some of the most fruitful in explaining the diverse world of parasitoids (e.g., Godfray 1994; Quicke 1997). Although a large body of research is accumulating on the evolutionary biology and community structure of parasitoids, we are still far from the level of synthesis that is now possible in parasitoid population dynamics (see previous section). Much work on parasitoid diversity has focused on how individual adult and larval parasitoid behavioral decisions percolate to determine structure at the population and community levels. Parasitoid behavior has many facets (Godfray 1994), conveniently divisible into adult search for hosts; oviposition on, in, or near hosts; and larval development on or within hosts. When behaviors and the population and community patterns to which they contribute are integrated over long time periods and/or over large spatial expanses, important patterns emerge (Hawkins and Sheehan 1994; Quicke 1997), such as host preferences and host range, niche partitioning with competing natural enemies, life-history strategies, and local adaptation and speciation. At the behavioral level, there has been a renaissance over the past twentyfive years regarding how animals forage for resources that vary in space and time. Casas outlined how our field knowledge of foraging biology is still rudimentary, and Heimpel showed how the interdependence of search parameters could greatly affect population dynamics. Van Baalen's premise (chapter 8) complements these two views on parasitoid behavior by considering how one should expect environmental conditions to determine fitness. Is the main determinant of parasitoid fitness the time the parasitoid has to search for hosts, or is it the number of eggs a parasitoid carries or can develop over its lifetime? Van Baalen suggests that a parasitoid's biology should be molded through what amounts to gambling: In the face of an uncertain environment, a parasitoid must trade off investments in searching ability against investments in egg-laying potential, selecting the optimal trade-off to give the greatest chances of a big fitness payoff. Van Baalen employs a simple model to derive an explicit representation of parasitoid fitness. Analyzing this model, he finds that the optimal number of eggs a female parasitoid should carry depends both on the population densities of hosts and parasitoids and on the nature of their interaction, concluding that the optimal (evolutionarily stable) strategy is probably something intermediate between being limited by egg production and by the time available to search for hosts. Considered in a evolutionarily dynamic setting, populations may show cycles in the prevalence of time and egg limitation, indicating that a diversity of strategies may be maintained within a population. Van Baalen closes by highlighting the need for future studies on the evolution of gregariousness (as a mechanism for swamping the host's immune system) and on host egg size (as a means of provisioning parasitoid juveniles in competitive environments).

INTRODUCTION

9

An intriguing dichotomy in parasitoid life-history strategies is between koinobionts and idiobionts (Strand, chapter 10). Koinobionts generally attack mobile and growing host stages, thereafter usually developing endoparasitically over a prolonged period within the live host. In contrast, adult female idiobionts usually paralyze or kill immobile host stages (i.e., eggs and pupae), their larvae subsequently exploiting the host as an ectoparasitoid. Somewhat metaphorically, therefore, koinobionts are more akin to true parasites than to the more predatory strategy of idiobionts. Given that parasitoids exert strong selection pressures on their hosts to escape death, hosts should be expected to have a certain degree of latitude to combat each type of parasitoid according to its life history. For koinobionts, this may mean the evolution of physiological responses, the most studied being the primitive immune response called encapsulation. Godfray (chapter 9) gives a lucid presentation of the interest and importance of understanding the biology of parasitoid virulence (i.e., the ability of the parasitoid juvenile to develop to maturity) and host resistance (the ability to encapsulate parasitoid eggs). He discusses the theoretical advances in understanding host-parasitoid coevolution, most of this work occurring only in the last several years. Coevolutionary interactions can be usefully classified as "graded" or "matching." Graded interactions are those in which on an absolute scale, some genotypes are better at either encapsulating eggs (host) or overcoming encapsulation (parasitoid). Matching interactions are those in which, to overcome host encapsulation defenses, the genotype of the parasitoid must match that of the host; no particular genotype for host or parasitoid is better, because the performance of the genotypes depends solely on the genotype of the opponent. Godfray suggests that graded interactions are probably more relevant to the process of encapsulation and provides empirical support for the prevalence of graded interactions. He calls for future work to examine in detail cellular dynamics, similar to that which has been done for vertebrate immune systems. It is quite possible that insect immune systems will shed unexpected light on those of vertebrates. Godfray focuses on how reciprocal selection pressures could generate diversity in adaptations of parasitoids to exploit hosts and hosts to escape parasitoids. This is a micro-evolutionary problem. When considering crossspecies comparisons of the evolution of life-history traits, one enters into the realm of macro-evolution. Strand's thesis (chapter 10) is that developmental traits regulating parasitoid growth have been largely neglected as forces shaping parasitoid life histories. He begins by reviewing the two classic ways of understanding cross-species life-history variation: phylogeny and mode of host exploitation (exo- or endoparasitoid / idio- or koinobiont). He believes that failure to understand the developmental processes regulating life-history traits can lead to spurious conclusions about phylogenetic patterns. Strand presents a wealth of empirical studies showing that embryonic

10

CHAPTER ONE

traits and the mode of parasitoid larval development have affected the direction of life-history evolution. He beckons for future studies to distinguish ecology and macro-evolution in shaping developmental traits. A most fruitful way to integrate community structure has been to represent species ensembles in terms of their feeding relationships in the form of food webs. This perspective was pioneered in the 1960s, but for some reason, parasitoids seem to have missed out on food web models until only recently. Given the recurrent observation that many species, especially predators, feed at more than one trophic level, it is interesting that organisms as specialized as parasitoids are able to do so as well. Brodeur (chapter 11) presents a diversity of examples of parasitoids feeding in complex ways in food webs. He begins by investigating taxonomic patterns in the occurrence of hyperparasitoids, noting that (1) no parasitoid family consists uniquely of hyperparasitoids and (2) there are diverse lifestyles of hyperparasitism within any given clade, meaning that lower taxonomic scales are needed to understand the macro-evolution of this phenomenon. Brodeur then tackles the problems of hyperparasitoid host range and multitrophic interactions, concluding that even if some patterns are noticeable, much remains to be done. He calls for a better appreciation of the genesis of hyperparasitism, both as a biological strategy and as a macro-evolutionary process. As does Brodeur, Holyoak (chapter 12) addresses issues involving parasitoid food chains; he tests the hypothesis that as parasitoids increasingly dominate communities, community food-chain length should decrease. This hypothesis is based on the following premises: (1) tightly coupled one predator-one prey interactions are more likely to produce unstable dynamics than diffuse many predator-many prey interactions; and (2) parasitoids have narrower prey ranges than predators. Thus, the greater instability of parasitoidhost food chains should limit their length relative to the more stable generalist predator-prey food chains. To investigate this prediction, Holyoak analyzes the patterns that appear when combining all fifty-eight available published food webs. He finds that there is a small but significant positive correlation between the proportion of enemies that are parasitoids and the number of species in the food webs—exactly the opposite of what was hypothesized. Although there is no relationship between the proportion of parasitoids and food-chain length, a very interesting pattern is revealed in terms of the distribution of species over trophic levels: In food chains with relatively more parasitoids than other natural enemies, fewer species occur at intermediate levels (3, 4, and 5) in the chain and more at the top levels (6 and 7). Holyoak points to the value of experimental manipulation in more powerfully resolving certain food web questions. He concludes, however, that we must ultimately rely on cross-web comparisons to assess the generality of pattern and process. Holyoak's premise is based on the cumulative destabilizing influence of

INTRODUCTION

11

tightly woven food chains. Another destabilizing force is interspecific competition within a trophic level. A central question for population biologists for almost a century has concerned the existence, nature, and strength of competitive interactions in ecological communities. One would think that if parasitoid assemblages are so speciose (with a mean of about five to six species per insect host species, but sometimes reaching more than one hundred), then they should be appropriate models for developing theories and conducting experiments and comparative analyses on competition. The problem is that competition may be idiosyncratically expressed in different macro-evolutionary periods, different communities, and different places. Therefore, detecting competition and extrapolating its importance should be done with caution. Hawkins (chapter 13) is aware of this. Using parasitoid assemblages, he searches for patterns that may reveal the impact of competition on community structure. There are many examples of strong competition between parasitoid species, particularly coming from biological control. Even though competition may be demonstrable within a community, however, the existence of competition does not necessarily imply that it affects the number and type of species in a community. By comparing local and regional parasitoid species richness, Hawkins finds scant evidence for the importance of competition to community structure. He explains this discrepancy using previous theoretical and empirical developments, which show that intense competition may be consistent with idiosyncratic patterns in vacant niches and with a lack of community saturation from local to regional levels. Hawkins suggests that although we can learn something about competition by following the intentional or serendipitous introduction of parasitoids into new geographical regions, we should be cautious about overinterpreting the results and applying them to natural, established communities.

Population Applications Parasitoids have played a central role in the application of biological control (Hawkins and Cornell 1999). Indeed, many of the hundreds of studies modeling parasitoid population dynamics and diversity (see previous sections) have explicitly discussed applications to biological control (Mills and Getz 1996; Hochberg and Holt 1999). For the most part, these studies have focused on what ecological characteristics of host-parasitoid interactions contribute to the depression and stability of host (i.e., pest) populations. The major finding here is that greater stability of the interaction may sometimes come at a cost of less mean impact—and, therefore, less economic benefit— from biological control (e.g., Murdoch 1990). A population biological perspective permits a scientific (i.e., repeatable and testable) means of understanding the successes and failures of natural

12

CHAPTER ONE

enemies in biological control. Conversely, a better understanding of how biological control functions is not only desirable for basic research, but has also been championed as the most sensible way to conduct natural manipulative experiments toward a theoretical end (Hawkins and Cornell 1999). These manipulative experiments have two important caveats reducing their pertinence (at least for classical biological control, where an exotic parasitoid is imported for the control of an invading pest). First, exotic natural enemies do not necessarily have a long evolutionary history with the target host and the environment into which they are released. Therefore, they might not represent natural examples of parasitoid-host interactions. Second, biological control generally takes place in simplified ecosystems, such as agricultural crops or managed forests. Therefore, the environment in which parasitoid-host interactions take place may not be natural. Taken together, these two caveats indicate that classical biological control introductions constitute unnatural testing grounds for population biological forces, and with less experimental control than would be the case in the more contrived cage experiments. It is our view, however, that if there is a science behind population biology, then it should be applicable to any kind of system, including the artificial systems of biological control. Classical biological control can be broken down into a series of sequential steps, and each step can be analyzed using a population biological framework. The two principal steps Mills (chapter 14) distinguishes are establishment and impact. Establishment involves the following problems: How many parasitoids must be released? Where should they be released in the pest population? What sex ratio should be used? Should genetic variation be considered? Mills highlights the population dynamic phenomena of "Allee effects," which require the initial density of released parasitoids to exceed a threshold for establishment to occur. He presents empirical evidence contrasting the effects on the establishment of host taxonomy, host refuges, parasitoid mating, landscape characteristics, climatic mismatch, and parasitoid founder effects. Turning to established parasitoid impact, Mills highlights the lasting depression of the pest population as the ideal. Three of the most important population approaches toward understanding impact in biological control are then discussed: spatial heterogeneity, parasitoid coexistence, and tritrophic interactions. Mills finds some evidence for each of these three factors to be related to impact in biological control, but it is far too early to draw any firm conclusions. He concludes that equilibrium-oriented theoretical approaches are not appropriate for understanding short-term dynamics. He calls for new approaches explicitly dealing with short-term dynamics, and for more dialogue between scientists and practitioners so that field testing becomes routine. Tscharntke (chapter 15) considers agricultural landscapes as an unexpected habitat for conserving biological diversity. His applied view is sur-

INTRODUCTION

13

prisingly similar to Roland's more fundamental one that spatial scale over landscapes sets the stage for population biological patterns. Tscharntke begins by drawing attention to how harsh agro-ecosystems really are for the component species; tillage, pesticides, fertilizers, and lack of vegetation diversification may all have adverse effects on parasitoids. In championing landscape aspects of agro-ecosystems, he focuses on vegetation adjacent to crop fields as parasitoid reservoirs. The idea is that cultivation perturbs parasitoid populations, and set-aside areas are necessary to create temporal and spatial bridges for the influx of parasitoids into the economically relevant arena. He presents substantial empirical evidence that the species composition and diversification of set-asides are important for parasitoids. In particular, he presents results that structural diversity of landscapes and the spatial distributions themselves are important to parasitism rates. Tscharntke predicts that landscape perspectives will be increasingly necessary in biological control, and they may even interact with other environmental perturbations, such as habitat destruction. He believes that generalizations will be impossible due to the unappreciated complexity of agricultural systems, but that the only way to build a scientific approach to understanding landscape effects on parasitoids is to start amassing well-conducted studies. Although there are clear benefits to biological control, there are also risks. Effects of biological control agents on nontarget species is one of these. Even though candidate control agents are screened for possible nontarget effects, Roitberg (chapter 16) believes that type II errors (i.e., incorrectly accepting a null hypothesis, which, in this case, is no effect on nontargets) may permeate these screening programs. He argues that the complex reality of parasitoid behavior and the simplified laboratory methods used for assessing these behaviors are at the root of uncertainty in screening programs; this is very similar to Casas's argument, but with an applied bent. Parasitoid effects on nontarget organisms can occur either if the parasitoids are not completely screened for all possible nontarget species, or if parasitoids evolve a broader host range following introduction. Roitberg argues that existing screening processes cannot assess the full host range likely to be adopted by parasitoids following release for biological control. He believes that evolutionary ecology provides a way to anticipate potential nontarget dangers based on life-history theory. Together with his colleagues, he has developed a series of dynamic life-history models to predict how life-history characteristics and environmental conditions influence the propensity of parasitoids to reduce host fidelity. Roitberg argues that host fidelity amounts to a complex surface of reaction norms (i.e., it is expressed differently in different environments), and therefore, evolutionary trajectories will be somewhat uncertain. It is the shape of host-fidelity reaction norms (steep, flat, linear, nonlinear) that provides an approach to guide practitioners in screening agents. Roitberg closes with a call for evolutionary biologists to

14

CHAPTER ONE

"take the bull by the horns" and evaluate whether the ecological sciences have something to contribute to the predictability of nontarget effects. Nontarget effects in biological control bring the specter of parasitoids threatening nontarget species with extinction. Hochberg (chapter 17) considers the converse problem—factors that threaten parasitoids with extinction. Given their immense diversity, it is reasonable to say that in terms of the number of species becoming extinct per unit time, we are losing more species of parasitoids than of any other type of insect. It is therefore perhaps somewhat puzzling that parasitoids have rarely been selected for conservation measures. Hochberg presents a number of "values," in addition to biological control, that make parasitoids worthy of conservation measures. He also discusses how parasitoids may become endangered in the first place and prescribes rules of thumb that practitioners may use to develop conservation measures. Hochberg closes by presenting a case example of a detailed modeling approach to conserving rare parasitoids. The case involves an ichneumonid parasitoid, Ichneumon eumerus, which, although not appearing in the World Conservation Union's IUCN Red Book of endangered species, should, since its unique host, the Lycaenid butterfly Maculinea rebeli, does. This is the most speculative chapter of the volume, but also appropriately serves as a final clarion call for more attention to noneconomic motives for the conservation of parasitoid biodiversity.

Conclusion We echo the view of Hawkins and Sheehan (1994) that these are indeed "exciting times" for parasitoid ecologists, and more generally for parasitoid population biologists. However, we are also of the firm opinion that because recognition of the utility of parasitoids as biological models has really come to the fore only in the latter half of the 1900s, growth has been somewhat slow in terms of the number and diversity of research programs worldwide. One of our main objectives in this book is to communicate to interested young researchers that many questions, and even fields, are wide open to discovery using parasitoids as model systems.

Two Host Location and Selection in the Field JERGME

CASAS

of what is known about the behavior and ecology of parasitoids has been discovered in the laboratory (Godfray 1994; Quicke 1997), and behavioral field studies of parasitoid species are rare (Waage 1983; Thompson 1986; Casas 1989; Janssen 1989; Driessen and Hemerik 1992; Connor and Cargain 1994; Visser 1994; Volkl 1994; Volkl and Kranz 1995; Heimpel et al. 1996, 1997; Volkl and Kraus 1996; Ellers et al. 1998; Henneman 1998). The lack of knowledge about host searching and host location in the field leads to two legitimate questions about (1) the importance, in the field, of the mechanisms studied in the laboratory and (2) the rationale in the choice of parameters in individual based models of host-parasitoid interactions (see Bernstein, chapter 4). Foraging behavior in the field can be inferred indirectly from capturerecapture data and sampling of host and parasitoid populations. The information available by using this approach is on a time scale ranging from one hour to a generation. The processes of host finding and host selection occur on a much shorter time scale, however, typically of the order of minutes, and requires the direct observation of the foraging behavior of females. In this chapter, I will identify three foraging parameters whose importance have been identified by conducting direct observations of foraging females in the wild. These parameters have been neglected in laboratory and theoretical studies so far. They are (1) the abundance of hosts as perceived by the parasitoid, (2) imperfect foraging cues, and (3) the time available for foraging. My arguments are developed by exploring in detail the few case studies in which parasitoids have been tracked continuously and their behavior recorded. Information about searching behavior in the field as observed in other, less studied, host-parasitoid systems is included when possible. It is my opinion that a deeper understanding of the foraging behavior of parasitic wasps will emerge through a comparative analysis of detailed case studies. MOST

Sampling Rules and Host Abundance A scientist's sampling rules designed to obtain unbiased estimates of host density may be quite different from those used by foraging parasitoids. We

18

CHAPTER TWO

know surprisingly little, except for the two examples described below, about the sampling rules used by parasitoids in the field, and how abundant hosts really are from the point of view of a parasitoid. In the first example, the parasitoid seems to have adopted a sampling strategy very well suited to the distribution of its host. In the second example, the low frequency of encounters with hosts leads to the acceptance of suboptimal hosts. Hence, both examples can be interpreted to show that perceived host abundance and distribution act as strong selection pressures on parasitoid traits related to host searching and host selection in the field. A third example shows how new knowledge about host density and host distribution in the field has changed our understanding of the patch leaving mechanisms of a parasitoid. The moth Greya subalba (Lepidoptera: Incurvariidae) feeds within immature seeds of Lomatium dissectum (Umbelliferae) (Thompson 1987). The flowers are grouped into umbellets, commonly with five to fifteen flowers; these umbellets are, in turn, grouped into compound umbels of fifty to two hundred flowers. Until they mature, seeds are held together tightly in pairs, or a "schizocarps." G. subalba females lay one—or less frequently, two— eggs per schizocarp, and the larva feeds within the immature schizocarp. Females tend to distribute their eggs broadly among umbellets, so that most umbellets have some larvae and the great majority of plants are attacked to some degree (25%-40% of seeds per plant are attacked). The distribution of attacked schizocarps among umbellets is well fitted by a truncated geometric distribution. The geometric distribution is the discrete analogue of the exponential, and also possesses the Markovian property. This property implies for the parasitoid that finding an attacked schizocarp does not change the likelihood of finding another one. Hence, the moth is distributing its progeny in a way that minimizes the information available to the parasitoid. Given these circumstances, how should its parasitoid, Agathis sp. (Hymenoptera: Braconidae), search for hosts? Searching females seem unable to distinguish seeds with larvae from those without larvae, as the following behaviors show (Thompson 1986). First, the distributions of time needed to check empty schizocarps and to oviposit are similar. Second, the parasitoids did not preferentially probe schizocarps with many hosts. Third, they preferentially probed large schizocarps, but large schizocarps were not more likely to contain larvae. Finally, females did not probe more schizocarps on umbellets in which many of the schizocarps had larvae. In conclusion, Agathis has to probe to detect host presence. The distribution of schizocarps probed by Agathis is also a truncated geometric distribution, with almost the same mean. This is the only case study in which the sampling strategy of the parasitoid has been studied in relation to its host distribution, and the similarity between the two distributions is striking. However, it remains unclear whether other sampling rules would be better and how the parasitoid per-

HOST LOCATION AND SELECTION

19

ceives host abundance (abundance being defined here as the successful proportion of probes). In contrast to the previous example, host abundance is best approximated for Drosophila parasitoids as the frequency of encounters with hosts per unit time. The density of Drosophila on fermenting fruits and sap fluxes in temperate woodlands is low (A. Janssen, pers. comm.). The rate of host finding by the parasitoids Asobara tabida Nees (Hymenoptera: Braconidae) and Leptopilina heterotoma Thompson (Hymenoptera: Eucoliidae) is normally between one and five hosts per hour, with a maximum of ten hosts per hour. Oviposition does not take much time (about one minute) and the majority of these parasitoids seldom run out of eggs (Driessen and Hemerik 1992; Ellers et al. 1998). Observed females searched most of the time (A. Janssen, pers. comm.), so resting did not affect potential foraging time. Once they find a host, A. tabida females accept it readily, almost irrespective of the survival chances of their offspring (Janssen 1989). Hence, the near-total acceptance of hosts can be explained only by the fact that the rate of encounters is so low that there is a marginal fitness gain from an oviposition in a suboptimal host. These two studies strongly suggest that host abundance, whether perceived or real, exerts a strong selection pressure on traits related to host finding and host selection. The third example shows how knowledge about field situations can make the difference between alternative theories based on laboratory experiments. Dissection of wild fruits containing the moth Ectomyelois ceratoniae (Lepidoptera: Pyralidae), a host of the Venturia caenescens (Hymenoptera: Ichneumonidae), revealed that fallen fruits harbor only one, and seldom two, hosts (Driessen et al. 1995; Driessen and Bernstein 1999). A large portion of fruits have no host. Under such conditions, Venturia would be best served by increasing its tendency to leave after each oviposition—a "decremental" rule. This result is in contradiction with the incremental model developed for the same species by Waage (1979). This model, which found its way into many textbooks (Krebs and Davies 1984; Bell 1991; Godfray 1994; Begon et al. 1996a), was based on highly unnatural petri dish experiments with host densities several times higher than those encountered in the known field situations.

Imperfect Foraging Cues Parasitoids use a set of cues associated with their hosts. Examples are semiochemicals from host feces, attacked plants, and visual cues such as galls or mines (Godfray 1994). These cues have been widely studied, the assumption

20

CHAPTER TWO

being that a parasitoid using them is at a reproductive advantage compared to a parasitoid searching at random. This assumption is best met when host density is low; metaphorically, any clue is welcome when it leads to a needle in a haystack. High host density may drastically reduce the efficiency of the searching parasitoid due to the distraction caused by unsuitable hosts. Also, unsuitable hosts often produce or trigger the same cues that were so effective at low host densities. However, searching for metallic objects is not really helpful in a haystack full of screws and nails. While the reliability and detectability of different cues have been the focus of much recent work (see Vet et al. 1995 for a review), the constancy in time and space of these cues in the field remains largely unstudied. The following two field studies identify the declining reliability of cues as a handicap for the foraging parasitoid. In the first case, the decline is due to parasitism itself, introducing a negative feedback effect. In the second case, the necessity to use easily detectable cues for finding hosts at low density actually inhibits host-finding at high density, because the parasitoid is constantly arrested by low-quality, unsuitable hosts. Tentiform leaf miners (Phyllonorycter sp., Lepidoptera: Gracillariidae) make very conspicuous spotted mines, which are searched for visually by eulophid parasitoids attacking later larval and pupal stages (Casas 1989; Connor and Cargain 1994). A field observational study showed that Sympiesis sericeicornis (Hymenoptera: Eulophidae) is unable to assess the quality of the inhabitant from a distance. It is only after the parasitoid has landed that the time spent and the sequence of behaviors on the mine become functions of the content of the mine. Unsuitable mines may be empty, or contain dead or already parasitized hosts. The percentage of unsuitable hosts increases over time; this is a real problem for those Sympiesis that continue to find unsuitable hosts. The problem is exacerbated because the probability of being found is not uniformly distributed over the host population. Hence, while some hosts escape parasitism, others are frequently rediscovered. In the light of the problems faced by Sympiesis foraging under those conditions, it is not so surprising that it is able to recognize and reject a dead host in about four seconds, 10% of the time needed for oviposition. Broadly, the same pattern seems to emerge for Cameria hamadryadella (Lepidoptera: Gracillariidae), a leaf miner on oak in North America, and its parasitoid Closerocerus tricintus (Hymenoptera: Eulophidae) (Connor and Cargain 1994; E. Connor pers. comm.). Aphytis melinus (Hymenoptera: Aphelinidae), the highly successful biological agent of the California red scale, Aonidiella aurantii (Murdoch 1994), forages in two strikingly different habitats within a given tree (pers. obs.). In the outer canopy, host density is very low and the analogy of the needle in a haystack seems appropriate. While foraging on bark, A. melinus is literally surrounded by hosts: There are enough suitable hosts within a few

HOST LOCATION AND SELECTION

21

square centimeters for A. melinus to lay all its daily egg complement in a fraction of the time available for foraging (one oviposition lasts around six minutes; pers. obs.)- Given these considerations, Aphytis seems surprisingly inefficient, having a mean oviposition rate of 0.6 eggs per hour (Casas et al. 2000). Part of the explanation of this low realized parasitism rate is that suitable hosts make up a tiny fraction of the total scale population; as much as 90% of the scale are, in fact, dead! Adding the amount of time spent dealing with dead hosts to the time spent searching increases the fraction of the total time spent in searching by only 20%, from 40% to 60%. Thus, handling dead scale is not a major factor determining the oviposition rate. The most likely hypothesis for the inefficiency of Aphytis is that the presence of so much dead scale and debris on the bark makes it difficult for Aphytis to discover or recognize desirable hosts. The description is valid not only for this species, but also for Aphytis aonidiae, a parasitoid of San Jose scale on almond trees. About 21% of scale examined by the parasitoid for more than sixty seconds were dead, and the percentage was much higher for shorter encounters (Heimpel et al. 1996; G. E. Heimpel, pers. comm.). Laboratory studies on Aphytis have demonstrated host-size discrimination in the context of host selection, sex allocation, superparasitism and host feeding (Luck and Podoler 1985; Opp and Luck 1986; Walde et al. 1989; van Lenteren 1994; Collier 1995; Morgan and Hare 1997; Morgan and Hare 1998). All of these finetuned behaviors become blunted in the field under circumstances such as those described, where finding a suitable host is simply difficult. However, it is a fact that many parasitoid species do make fine-tuned choices in specific situations. Since these choices have consequences in terms of fitness, appropriate behaviors in these situations, even if they are rarely encountered in the field, do matter. The unique contribution of behavioral studies in the field is an estimation of the frequency with which parasitoids encounter these situations. The relationship between field and lab studies will be dealt with again at the end of this chapter.

Other Sources of Foraging Variability Both the industry of near perfect petri dish experiments and decade-long bivariate host-population dynamics have left the misleading impression that host-parasitoid systems are tightly coupled pairwise interactions occurring in a vacuum. Experience in the field tells quite another story. Parasitoids encounter a range of situations unlike those in laboratory studies: for example, incredibly complex spatial structures of the foraging environment, and microclimatic conditions varying severalfold over very short distances. It is a truism that the foraging behavior of parasitoids is under the influence of

22

CHAPTER TWO

many factors. More discomforting is the fact that parasitoid behavior may be sometimes totally unrelated to host and parasitoid population densities, as the next preliminary results show. In order to explore the relationship between aggregation of searching parasitoids, host density, and parasitism rate in the field ("spatial aggregation of parasitism"; for a review, see Hassell and Wilson 1997, and Bernstein, chapter 4), twenty-three leaves at the interior of a single grapefruit tree were tagged and the number of Aphytis spotted on the leaves counted every hour during one day from the onset to the end of activity. There was no relationship between the total number of visits per day on a leaf and the number of unparasitized hosts, the number of live scale of all stages and the total number of scale (live, parasitized, and dead scale). Only five Aphytis eggs were recovered. Three were laid on the same leaf, which was not among the highly visited ones. Given the lack of relationship between Aphytis behavior (visits and ovipositions) and host population size on the different leaves, one may tentatively conclude that leaves were chosen at random. However, a statistical analysis showed that visits to the different leaves were not at random, but slightly aggregated (data from the first day CV = 1.28, mean = 2.4, S.E. = 0.64, n = 23). An extension of the experiment over two days for six of the original leaves rejected the hypothesis of random visits. The ranking of visits between leaves over the two days was very similar (table 2.1). The sum, over the six leaves, of the absolute differences between the daily visits is a good measure of constancy of attractiveness of leaves over two days. The smallest possible sum of differences is four, as we observed ten visits the first day, and fourteen visits on the second day. The observed sum of differences is six. There is a single permutation able to produce a sum of differences of four (permuting the visits of leaves four and five on the second day). The other 718 permutations of the number of visits on the second day produce larger sums. The probability of observing a sum of differences smaller than or equal to six by randomly assigning visits to leaves is 0.003. While a lack of dependence of parasitoid aggregation and parasitism rate on host density is rather frequent ("host-density-independent heterogeneity"; see Hassell and Wilson 1997 for a review, and Smith and Maezler 1986 for similar results on the same system), these field observations show that paraTABLE 2.1 Number of Visits by Aphytis melinus to Six Leaves over Two Consecutive Days Leaf Number

1

2

3

4

5

6

Total

Day 1 Day 2

6 7

1 1

0 0

2 1

0 4

1 1

10 14

HOST LOCATION AND SELECTION

23

sitoids visit specific host patches preferentially and repeatedly, for reasons apparently independent of host density.

Time Available for Foraging Climatic conditions, such as temperature, wind, and rain, strongly determine the foraging window available to parasitoids. Daily activity of parasitoids in the field has hardly been studied (but see Volkl and Kranz 1995), despite its obvious importance for behavioral ecology and population dynamics of hostparasitoid systems. The mean and variance in the extent of the foraging window determine the rate of oviposition and, thereby, the parasitism rate. This was demonstrated in a recent study by Weisser and colleagues (1997). They first studied the impact of climatic conditions on the length of foraging activity of Aphidius rosae (Hymenoptera: Aphidiidae), an aphid parasitoid, in the laboratory. Using weather data from Bavaria and a set of assumptions, they then estimated the realized fecundity of this species in the field. Although some individuals could reach their full potential, which is more than nine hundred hosts parasitized, most were predicted to perform poorly. The average could be as low as eighty to one hundred if unfavorable weather is included in the model. Given that the quantification of the daily foraging window of parasitoids based on behavioral observations in the field is exceedingly rare, the following preliminary results are worth presenting. In a study on Aphytis melinus attacking red scale, I scanned the bark of the lower portion of a tree visually for Aphytis for six minutes every hour from the onset of activity on two separate days. Aphytis forages at the interior of trees (bark and a few twigs) for only a few hours per day (figure 2.1). Somewhat more time is spent in the outer canopy, but the exact extent of the increase is unknown. The reasons for such a restricted use of total available time to a few hours are unclear, but could be related to lower light intensities in interior portions of trees and to lower temperatures. When females forage for six hours, models of egg load dynamics based on field experiments predict that between onethird and one-half of the population of Aphytis run out of eggs at least once during the foraging period (Casas et al. 2000; see also van Baalen, chapter 8). Restricting the foraging window to four hours reduces the egg-limited proportion of the population to one-quarter. In contrast, more than half of the population experiences egg limitation when the foraging period is extended to eight hours (pers. obs.). Given the current debate about the dichotomy of time and egg limitation strategies (Collier 1995; Getz and Mills 1996; Rosenheim 1996; Heimpel et al. 1996; Heimpel et al. 1998; Mangel and Heimpel 1998; Ellers et al. 1998; Rosenheim 1999; Sevenster et al. 1998;

24

CHAPTER TWO

1

1 —

1

•g

/

1

/'

0.5-

Q.

\

- 1

y / /-

V' 1

\

\l \

1

7

0-

o o o

1

o o

\

r CO

O O

ia

time of day

Figure 2.1. Percentage of foraging Aphytis melinus on bark on each of two sampling days (solid line, October 23, 1994; dashed line, October 30, 1994). see also van Baalen, chapter 8), hard data such as these are urgently needed to keep theoretical arguments rooted in reality.

Field versus Laboratory Experiments In this chapter, I have argued that field observations provide a unique understanding of host searching and host selection in four ways: (1) by identifying new and important processes and parameters we ought to study in the laboratory and include in our theoretical models; (2) by confirming and reconsidering the role and importance of widely established processes; (3) by distinguishing between alternative theories developed in the laboratory or through theory; and (4) by producing a priority list of all the parameters and processes that require our attention.

HOST LOCATION AND SELECTION

25

Given the obvious benefits of field behavioral studies, one may wonder how laboratory experiments fare compared to them. In the context of behavioral studies conducted in an evolutionary mindset, laboratory experiments are ideally suited to explore how effective parasitoids are at solving particular problems and the way in which they solve them. Care has to be taken in the interpretation phase, in particular when referring to "the natural conditions" under which a particular trait may have evolved. I regard a combination of field and laboratory experiments as the most promising approach. It is fair to recognize that field studies have their own set of limitations. Small sample size is an obvious one. The insidious consequence is that one requires situations (e.g., patches, time of the year) characterized by high densities of hosts and parasitoids in order to obtain a reasonable sampling size. This, in turn, may lead to a biased view of the conditions usually encountered by a foraging parasitoid. Studying rare species and species occurring at low densities is a daunting task, even though the great majority of host-parasitoid associations probably are of these types.

Conclusions and Future Directions The future for field studies is bright because technology continues to provide better experimental apparatuses, and because there is ample room for originality in how this technology is employed. Among the battery of new technologies available for field studies, I see miniature chemical and physical sensors able to characterize the environment in which parasitoids forage in real time, I see ever-smaller tracking devices, and I see long-distance microscopes enabling the observation of a foraging parasitoid at a distance of several meters. Some of these tools do already exist: insects as small as tachinid flies (yes, parasitoids) can be tracked using harmonic radar (Roland et al. 1996), a miniature accelerometer as light as 0.08 grams can be placed on large leaves without perturbing the field of vibrations, the body temperature of a parasitic wasp can be measured at a distance of three meters (by combining an infrared CCD to a Questar long-distance microscope resolving 12 microns at three meters), and portable electroantennograms, which are about one hundred times more sensitive than gas chromatograph measurements (Metcalf 1998) are now being commercialized (van der Pers and Minks 1998). At an even more challenging level, there is plenty of opportunity for original contributions linking field studies and laboratory work. After a long, unfinished, but necessary period of observational studies in the field, it is encouraging to see manipulative studies now being conducted (Waage's 1983 pioneer study was partly manipulative). The reverse can (and should)

26

CHAPTER TWO

also be done: catching wild foraging females and conducting pseudo-laboratory experiments on the spot, employing well-defined, controlled protocols. Despite these exciting perspectives, I do not see the number of field observational/manipulative studies increasing dramatically over the next several years. The highly unbalanced ratio of field studies to laboratory and theoretical studies will remain constant. The main reason for this is that field studies are particularly difficult and time-consuming. The optimistic conclusion is that studies on the behavioral ecology of parasitic wasps in the wild will continue to be a rewarding field of investigation for the scientist inspired by challenging tasks; indeed, we need much more fieldwork to make our understanding of host-parasitoid interactions approach reality. Acknowledgments. I thank very much the following scientists who shared their intimate understanding of the systems they studied over many years: Edwards Connor, George Heimpel, Arne Janssen, Bob Luck, Bill Murdoch, Jay Rosenheim, and Sue Swarbrick. Carlos Bernstein, George Heimpel, Tony Ives, and Ana Rivero made very useful comments on previous versions of this chapter.

Three Effects of Parasitoid Clutch Size on Host-Parasitoid Population Dynamics GEORGE E. HEIMPEL

of host-parasitoid population dynamics have incorporated the effects of numerous parasitoid behaviors (Hassell 1978; Walde and Murdoch 1988; Kidd and Jervis 1996; Bernstein, chapter 4). One behavior that has largely escaped analysis, however, is parasitoid clutch size. This has occurred despite the fact that clutch size is known to vary greatly within and between species, and that a large body of theory has been developed to explore the evolutionary significance of clutch size variation in parasitoids (Godfray 1994). Three classes of variation in parasitoid clutch size have been identified. First, parasitoids can develop singly within hosts ("solitary" parasitoids) or as part of a brood of siblings ("gregarious" parasitoids). These classifications are somewhat imprecise, in that females of solitary species may lay more than a single egg per host (Rosenheim and Hongkham 1996), and females of some gregarious species may facultatively lay eggs singly (e.g., Luck et al. 1982; Olson and Andow 1997). Still, they underscore the important distinction between solitary species, which are constrained to developing singly within hosts due to physical or physiological suppression among siblings, and gregarious species, which are free from this constraint (e.g., le Masurier 1987; Visser and Rosenheim 1998; Strand, chapter 10). Second, there is interspecific variation in the mean clutch size produced by parasitoids. For instance, mean clutch size in twenty-two species of parasitoids in the family Bethylidae range from two to more than one hundred (Griffiths and Godfray 1988), a range that is not uncommon for the parasitoid Hymenoptera as a whole (Mayhew and Hardy 1998). Third, intraspecific variation in clutch size has been widely reported (Godfray 1994; Hardy et al. 1998). This variation has been attributed to host size (Vet et al. 1993), the host encounter rate (Podoler et al. 1978; Rosenheim and Rosen 1991; Rosenheim and Hongkham 1996), and the physiological state of the parasitoid (Rosenheim and Rosen 1991). MODELS

In this chapter, I will introduce a series of discrete-time models of hostparasitoid population dynamics based on the Nicholson-Bailey framework that incorporate effects of parasitoid clutch size. For simplicity, clutch size is

28

CHAPTER THREE

treated as a fixed parameter in these models so that the analyses are most germane to comparisons between species that have different mean clutch sizes or between individuals that produce consistently different clutch sizes. A general form of the model that I use as a starting point can be expressed as follows (Hassell 1978): Ht + 1 = AH t e"°"

(3.1a)

P t + 1 = cH t (l - e-*).

(3.1b)

Here, H and P are the densities of hosts and female parasitoids, respectively; X is the host reproductive rate in the absence of parasitoids; c is parasitoid clutch size; and the subscript t stands for time (in generations). The term a is the rate at which parasitoids attack hosts which can depend on parasitoid and host densities, and exp( — a) is the proportion of hosts escaping parasitism under the assumption that parasitism is distributed randomly among the host population. Several authors have allowed c to affect parasitoid recruitment, leading to a negative relationship between clutch size and equilibrium host density (Hassell et al. 1983; Taylor 1988b; Murdoch et al. 1996a). The analyses that I will describe below extend these models by introducing a trade-off between clutch size and the attack rate that is mediated by the potential for egg limitation.

The Trade-Off between Clutch Size and the Attack Rate The specific attack rate that I use here is a slight modification of a function introduced by Getz and Mills (1996) that incorporates parasitoid clutch size while responding to both host and egg limitation: a = a((3/c)P / (((3/c) + aH)

(3.2)

Here, a is the parasitoid search efficiency constant, which can be described as the per-parasitoid probability of host encounter (Hassell 1978), and (3 is the maximum parasitoid fecundity. The ratio (J/c therefore represents the number of clutches laid per parasitoid. Equation 3.2 describes a trade-off between clutch size and the attack rate that is mediated by parasitoid searching efficiency, a, and results from the increased effects of egg limitation at higher clutch sizes (figure 3.1). As clutch size increases for a given level of 3, the number of clutches laid (and, correspondingly, the number of hosts attacked) decreases. The trade-off is strongest when parasitoid searching efficiency is high (figure 3.1) and when parasitoid egg supply is most likely to

EFFECTS

OF

PARASITOID

CLUTCH

SIZE

29

Clutch size Figure 3.1. Effects of clutch size on the attack rate, a, at three levels of the parasitoid searching efficiency constant, a, using equation 3.2. Other parameters are: P = 1, H = 10, p = 100. limit fecundity (e.g., Driessen and Hemerik 1992; Rosenheim 1996; Mangel and Heimpel 1998; van Baalen, chapter 8). As has been discussed by Getz and Mills (1996), equation 3.2 is equivalent to a type II functional response (Holling 1959a) in which the attack rate is limited by fecundity rather than by handling time constraints. Thus, for equation 3.2, the maximum attack rate is set by the ratio (3/c (figure 3.2). While increases in clutch size may lead to diminished attack rates for a given level of fecundity, greater clutch sizes also produce more offspring for a given number of hosts attacked. Within the simple framework described by equation 3.1 in a single generation, the number of parasitoid offspring produced is the product of clutch size and the number of hosts killed (figure 3.3). In summary, clutch size has two important effects when maximum fecundity, p, is held constant. First, the attack rate decreases with increasing clutch size when parasitoid searching efficiency is high enough to lead to egg limitation at higher clutch sizes. This leads to a type II functional response in which the asymptote is set by the number of clutches laid in the lifetime of the parasitoid. Second (and by definition), higher clutch sizes lead to more progeny produced per parasitized host. In the remainder of

30

CHAPTER THREE

"8 !

en

1 3

20

40

60

80

100

Host density Figure 3.2. Effects of host density on the number of hosts killed, H(\ — exp[ —a]), for various clutch sizes, c, using equation 3.2. Other parameters are: P = 1, (3 = 100, a = 0.5. this chapter, I will explore the consequences of these processes for hostparasitoid population dynamics. The analyses I present are most relevant for two classes of comparisons. First, they can be used to compare parasitoid species or strains that lay different average clutch sizes but have the same maximum fecundity. One potential problem with interspecific comparisons is that fecundity will often vary with clutch size. Allowing fecundity to increase with clutch size in the model presented here leads to an attenuation of the trade-off between clutch size and the attack rate, as I will discuss at the end of the chapter with two examples from the literature. Differences in host use patterns or encounter rates can obviously confound interspecific comparisons as well. Second, the models can be used for intraspecific comparisons between populations or individuals that consistently deposit different clutch sizes by attacking different host species or host size classes sizes (e.g., Klomp and Teerink 1967; Luck et al. 1982; Taylor 1988a; Rosenheim and Rosen 1991; Vet et al. 1993). In this case, maximum fecundity is presumably the same, but differences in the host encounter rate may exist (e.g. Nell et al. 1976; van Lenteren 1994). For simplicity, I incorporate neither adaptive clutch size adjustment (Parker and Courtney 1984; Charnov and Skinner 1984; Skinner 1985; Ives 1989; Rosenheim and Rosen 1991; Mangel et al. 1994; Mangel

EFFECTS OF PARASITOID

31

CLUTCH SIZE

100

80 -

5

"GO

60 -

o •8

40 -

2

. '

Parasitoid offspring

O

20 -

Hosts killed

20

40

60

80

100

Clutch size Figure 3.3. Effects of clutch size on the number of parasitoid progeny produced (Pt+, using equation 3.1) and the number of hosts killed {Ht{\ — exp[ — a]) using equation 3.1) in a single generation. Other parameters: Ho = 100, Po = 1, X = 2, f$ = 100, a = 0.5. and Heimpel 1998) nor the effects of clutch size on offspring fitness (e.g., Hardy et al. 1992; Vet et al. 1994; Ode et al. 1996).

Coexistence and Clutch Size Getz and Mills (1996) concluded that the dynamics described by equation 3.1 were inherently unstable, and including the clutch size dynamics discussed above does not change this result. Following Getz and Mills (1996), I introduced stability into the model by replacing e x p ( - a t ) in equation 3.1 with the negative binomial form of the escape function

f(a) = (1 + a/k) - k

(3.3)

in which the parameter k is inversely proportional to the extent to which parasitism is aggregated among the host population (May 1978; Getz and Mills 1996). I evaluated the effect of the various parameters in the model on host and parasitoid coexistence numerically by determining whether either or both populations persisted for at least one hundred generations. Starting population sizes for the simulations were 100,000 and one for hosts and parasitoids, respectively.

32

CHAPTER

THREE

100

Coexistence at all c, a

c
0.2, host and parasitoid coexistence depends on X, c, and a. Coexistence is dependent on c and a over a relatively narrow range of intermediate levels of k and X, the dynamics become unstable at all c and a when k exceeds a threshold level and X falls below a threshold level (figure 3.4). The exact threshold levels change with initial host and parasitoid levels and the arbitrary level assigned to extinction, but the qualitative trends are robust with respect to these parameters. This analysis therefore suggests that host-parasitoid coexistence is dependent on clutch size only under relatively narrow conditions of spatial heterogeneity and host population growth. Under these specified conditions, clutch size (and parasitoid searching efficiency) affect coexistence in two ways. First, coexistence itself is favored by smaller clutch sizes, and second, extinction of both host and parasitoids (as opposed to only parasitoids) is favored at the largest clutch sizes (figure 3.5). Thus, it appears that higher clutch sizes can be a destabilizing force in host-parasitoid interactions. Why would higher parasitoid clutch sizes destabilize host-parasitoid interactions? In principle, increasing the clutch size could destabilize dynamics

EFFECTS

OF PARASITOID

CLUTCH

33

SIZE

1.0

Hosts and parasitoids go extinct

0.8 -

Parasitoids go extinct

0.6 0.4

0.2 -

itoids \_ Hosts and parasitoids coexist 0.0

20

40

60

80

100

Figure 3.5. Values of c and a at which hosts and parasitoids coexist, parasitoids only because extinct, or both hosts and parasitoids become extinct. Other parameters: Ho = 100,000, Po = 1, p = 100, \ = 2, k = 0.6. Lines appear as step functions because simulations were run over values of a separated by 0.1. by increasing the number of parasitoid offspring available to put pressure on the host population in the next generation. However, this effect should be canceled out by the lower attack rate that is associated with higher clutch sizes (see figures. 3.1-3.3). The explanation comes from the effect that hostand egg-limitation have on the trade-off between clutch size and the attack rate. Consider first a parasitoid with a clutch size that equals its maximum fecundity (c = (3 = 100 in the current analysis). While the individual attack rate is very low for this parasitoid, the maximum fecundity (i.e., egg limitation) can be achieved at very low host densities and for low values of the parasitoid searching efficiency constant, a. Consider next a parasitoid with the same maximum fecundity, but that produces a low clutch size. While the attack rate per parasitoid can, in principle, be much higher in this scenario, the realized fecundity and attack rate are dependent on high host density and/or high parasitoid searching efficiency. Parasitoids with lower clutch sizes are therefore more vulnerable to decreases in fecundity at low host densities. The effect of this increased level of vulnerability is to decrease the extent to which low clutch sizes are canceled out by high attack rates. This decreases pressure on the host population from parasitoids with low clutch sizes and allows for host-parasitoid coexistence over a broader set

34

CHAPTER THREE

of ecological conditions. A related effect of high parasitoid clutch size is to increase the amplitude of population fluctuations (see below). This can lead to scenarios in which parasitoids become extinct while host populations continue to expand (figure 3.5; see also Hochberg and Lawton 1990a and Hochberg and Holt 1995).

Population Dynamics and Mean Densities Examples of coupled host and parasitoid population dynamics over one hundred generations are shown for representative parameter values in figure 3.6.

T3

Hosts

'3

120

C

c= \ a = 0.5

100 80 60 40 20 n

20

40

60

80

100

o

Generations Figure 3.6. Population dynamics of hosts (solid lines) and parasitoids (dotted lines) over 100 generations under four combinations of parasitoid clutch size, c, and the parasitoid searching efficiency constant, a. Other parameters: Ho = 100, Po = 10, \ = 2, |3 = 100, k = 0.75.

EFFECTS OF PARASITOID CLUTCH SIZE

35

The general pattern is damped oscillations with hosts more abundant than parasitoids and a clutch size of five (figure 3.6A, C). When the clutch size is increased, parasitoid abundance can greatly exceed host abundance, and host densities are decreased (figure 3.6B, D). To explore these relationships further, I calculated mean host and parasitoid densities as well as the proportion of hosts parasitized. Mean host and parasitoid densities over one hundred generations are shown in figure 3.7 and can be best understood as an interaction among clutch size, egg limitation, and host limitation (see also van Baalen, chapter 8). When parasitoid searching efficiency, a, is low, eggs are not limiting, and the attack rate, a, is unrelated to clutch size (see figure 3.1). Fewer parasitoids are produced per host at low clutch sizes regardless of a, however, and this leads to less host suppression and higher mean host densities at lower clutch sizes (figure 3.7A). When parasitoid searching efficiency is high, eggs become increasingly limiting as clutch size increases, leading to proportionately lower attack rates at high clutch sizes, as was discussed above (see figure 3.1). Thus, the decrease in the number of offspring produced per host with lower clutch sizes can be offset by a higher attack rate because there is less egg limitation (figure 3.7A). Egg limitation therefore serves to weaken the negative relationship between clutch size and host suppression by parasitoids. Mean parasitoid densities are equal to the product c X mean host densityX mean proportion parasitism, leading to a monotonic increase in parasitoid density with clutch size (figure 3.7B). For the particular example illustrated in figure 3.7, the mean level of parasitism ranges between 0.4 and 0.5 and decreases slightly with clutch size and the parameter a (figure 3.7C).

Discussion and Future Directions These analyses suggest that parasitoid clutch size can interact with parasitoid searching efficiency and the risk of host versus egg limitation in parasitoids to influence the dynamics of host-parasitoid interactions. In the absence of egg limitation, the attack rate is unrelated to clutch size and higher clutch sizes lead to lower average host densities because more parasitoids are produced per host attacked, as has been found in other models (e.g., Taylor 1988b; Murdoch et al. 1996a). However, conditions of potential egg limitation drive a trade-off between clutch size and the attack rate, which weakens the negative relationship between clutch size and host density. These relationships follow from the form of the attack function, which produces a positive but decelerating relationship between the number of hosts killed and host density. This form of the attack function is analogous to a type II functional response in which fecundity sets the maximum attack rate (Getz and

36

CHAPTER THREE 30 25 -

20

40

60

0.40 20

40

60

80

80

100

100

Clutch Size Figure 3.7. Effect of clutch size on mean host density (A), mean parasitoid density (B), and mean proportion parasitism ( Q over 100 generations for three levels of a. Other parameters: Ho = 100, Po = 10, X = 2, p = 100, k = 0.75.

EFFECTS OF PARASITOID CLUTCH SIZE

37

Mills 1996). Since parasitoids experience varying levels of egg and host (or time) limitation as a function of both reproductive strategies and environmental conditions (Driessen and Hemerik 1992; Rosenheim 1996, 1999a, 1999b; Weisser et al. 1997; Ellers et al. 1998; Heimpel and Rosenheim 1998; Heimpel et al. 1998; Sevenster et al. 1998; van Baalen, chapter 8), this function is particularly appropriate for these analyses that require both egg and host limitation to influence the attack rate. An important caveat accompanying these results is that maximum fecundity, p, was kept constant while clutch size was varied. Although such an approach may be valid for intraspecific comparisons in which individual parasitoids consistently lay different clutch sizes (e.g., by attacking different species or stages of hosts; Vet et al. 1993), it is not clear that it would apply to interspecific comparisons, where species that lay larger clutches may have higher fecundity. If this were the case, (3 would increase with c, and depending on the specific relationship between these parameters, such a correlation could weaken or eliminate the trade-off between clutch size and the attack rate that is central to the analyses presented here. For example, the two braconid parasitoids Cotesia rubecula and C. glomerata are closely related and have overlapping host ranges, but C. rubecula is solitary while C. glomerata is gregarious, laying 20 to 30 eggs per host when attacking Pieris rapae larvae (Nealis 1990; le Masurier 1991). The potential fecundity of C. glomerata exceeds that of the solitary C. rubecula by a factor approximating its clutch size: the maximum egg complements of C. rubecula and C. glomerata are approximately 100 and 2,000, respectively (Laing and Levin 1982; Nealis 1990). Thus, the attack rates of these two species as defined by equation 3.2 are identical. However, the higher clutch sizes and greater fecundity produced by a species like C. glomerata can, in principle, result in higher parasitoid densities and enhanced host suppression (figure 3.8A). This difference is greatest at low levels of searching efficiency, when fecundity is disproportionately limited at the lower clutch size. This comparison is not designed to predict actual population numbers for Cotesia sp. or P. rapae; rather, it illustrates potential differences between parasitoids with the same attack rate but differing clutch size and potential fecundity. Other factors that could affect the comparison include differences in juvenile mortality associated with solitary versus gregarious development (Brodeur and Vet 1995; Brodeur et al. 1997, 1998) and differences in host encounter rates between females of the two species. A similar qualitative comparison can be made for solitary versus gregarious strains of the pteromalid Muscidifurax raptorellus (Legner 1989; Antolin et al. 1996). Females from a gregarious strain lay between one and five eggs per host, while females from a solitary strain of M. raptorellus lay a single egg per host (unless they are mated to a male of the gregarious strain, in which case two eggs may be laid; Legner 1987). Legner (1967, 1987,

38

CHAPTER THREE

40

35 30 25 20 15 ' " • . . C. rubecula (solitary)

10 5a

o

a

^glomerata

(gregarious)

0 0.2

0.4

0.6

0.8

1.0

40

35 30 25 20 -

' • • . M. raptorellus solitary strain

M. raptorellus gregarious strain 0.2

0.4

0.6

0.8

1.0

Figure 3.8. Mean host density (over 100 generations and as a function of the parasitoid searching efficiency constant, a) from model runs in which clutch size, c, and potential fecundity, fJ, values are associated with two species of Cotesia (A) and two strains of Muscidifurax raptorellus (B). In A, c = 1, p = 100 for C. rubecula and c eq , (3 = 2,000 for C. glomerata. In B, c = 1 and (3 = 105 for the solitary strain and c = 3 and (3 = 177 for the gregarious strain. See text for literature sources of c and p. Other parameters for both panels: Ho = 100, Po = 10, r = 2, k = 0.75.

EFFECTS

OF P A R A S I T O I D

CLUTCH

SIZE

39

1988, 1989) compared the clutch size and attack rate of these two strains in the laboratory and found that, as in the Cotesia example, mean fecundity of the gregarious strain was higher than that of the solitary strain. However, the difference did not compensate for clutch size, and the attack rate of the gregarious strain was consistently lower than that of the solitary strain. While average clutch size of the gregarious strain exceeded that of the solitary strain by a factor of approximately 3, fecundity was only 1.7 times higher in the gregarious strain. The effect of the incorporation of these differences on mean host density using values from Legner (1988) is shown for a range of parasitoid searching efficiency in figure 3.8B. As in the previous comparison, the gregarious strain suppressed the host population to a greater extent than the solitary strain, and the difference between the two scenarios was greatest at low levels of parasitoid searching efficiency. An important omission from the models analyzed here is intraspecific (or intrastrain) variability in clutch size. As I have explained above, the models presented here are relevant mainly for comparisons among parasitoid species. Within species or strains, however, clutch sizes are predicted to decrease at higher host encounter rates in anticipation of potential egg limitation (Mangel 1987; Mangel et al. 1994; Mangel and Heimpel 1998), as has been observed in a number of laboratory systems (Podoler et al. 1978; Rosenheim and Rosen 1991; Rosenheim and Hongkham 1996). Also, Taylor (1988b, 1997) analyzed a series of models in which the cumulative clutch size of multiple females attacking single hosts was dependent on both host and parasitoid population levels. Unlike the models presented here, c did not enter into the host equations in Taylor's models, and the functional response was unaffected by egg limitation. The combined effects of (1) downward adjustment of clutch size with increasing host availability and (2) upward adjustment of cumulative clutch size through superparasitism have not been explored. Regardless of whether clutch size differences come from variability in individual behavior (as predicted by optimality theory) or from superparasitism (as in Taylor's models), these clutch size adjustments result in parasitoid offspring with differing potential fecundity (Godfray 1994). A realistic clutch size model that incorporates intraspecific variability in clutch size should incorporate the effects of all of these processes on population dynamics, as well as the effects of the attack rate variability that I have explored here. Perhaps the most general outcome of these analyses is the observation that parasitoids with lower clutch sizes are more vulnerable to host limitation than are parasitoids with higher clutch sizes. Conversely, parasitoids that lay higher clutch sizes are more vulnerable to egg limitation (assuming constant fecundity), which is synonymous with attaining maximum fecundity within the modeling framework used here. The realized fecundity of parasitoids with higher clutch sizes will therefore more closely approximate potential

40

CHAPTER THREE

fecundity under a broader range of ecological conditions. This dynamic leads to the results of the models presented here, which include (1) more hostparasitoid coexistence at low clutch sizes, (2) more host suppression at high clutch sizes, and (3) a higher ratio of parasitoids to hosts at high clutch sizes. How might these predictions be tested empirically or comparatively? One approach would be to test the underlying processes by comparing attack rates or functional responses of closely related parasitoids that produce different clutch sizes. As explained above, Legner (1967, 1987, 1988, 1989) adopted this approach with M. raptorellus and showed that attack rates were lower in a strain that produced larger clutches and that this led to lower asymptotes in functional response curves for the gregarious strain (Legner 1967). Another approach would be to test the model outcomes listed above with field data or with multigenerational laboratory studies. A further implicit prediction of these analyses is that parasitoids with higher clutch sizes should outcompete parasitoids with lower clutch sizes over many generations, assuming all else is equal. This prediction could also form the basis of empirical or comparative tests. In all cases, potential interactions among clutch size, egg limitation, and host limitation should be considered. The prediction that parasitoids with higher clutch sizes can produce greater host suppression than parasitoids with lower clutch sizes suggests important implications for biological control. Are parasitoids with higher clutch sizes superior biological control agents? A general comparative treatment of this question remains to be conducted, but Murdoch and colleagues (1996a) have shown that the ability of Aphytis melinus to use smaller hosts than A. lingnanensis to produce clutches of two eggs (rather than one) contributed to its superiority as a biological control agent. In a more general treatment of the effect of parasitoid life histories on biological control, Murdoch and Briggs (1996) argued that superior biological control agents are species that use a minimum of host individuals to replace themselves in the next generation. Increasing parasitoid clutch size does just this by increasing parasitoid recruitment from single hosts. Acknowledgments. I thank Tim Collier, Charles Godfray, Jeff Harvey, Michael Hochberg, Tony Ives, Marc Mangel, Roger Moon, Paul Ode, Jay Rosenheim, and two anonymous reviewers for helpful comments on this chapter.

Four Host-Parasitoid Models: The Story of a Successful Failure CARLOS BERNSTEIN

ecology seems to have been born from the combination of simple observations about nature (the world is green, the same species are found in the same areas and with relatively the same abundance, and so on), and a program of physics (reduce processes to their bare fundamentals, reconstruct these processes mathematically, draw predictions from these models, and test them experimentally). In the realm of host-parasitoid interactions, the pioneering work of Thompson (1924) and Nicholson and Bailey (1935) was developed under the general assumption that natural populations are stable. The aim was to identify these stabilizing processes. Both attempts failed to identify any, however. In what can be seen either as a truly scientific endeavor or as a hopeless quest for the Holy Grail, theoretical ecologists ever since have striven to identify regulatory processes and to incorporate them into their models. In this chapter, I will retrace the fate of three of the first stabilizing mechanisms proposed: sigmoid functional responses, mutual interference, and spatial heterogeneity. Rather than give a full account of the theory of hostparasitoid interactions or of its history, my aim is to attempt to answer these questions: What were we trying to do? Was it worth the effort? and What have we learned, if anything? My choice of the bibliography is clearly biased: My intention is to emphasize the polemic aspects of the story. THEORETICAL

The Story of a Successful Failure In 1924, Thompson published a model intended to explain and to guide decisions on biological control of insect pests by parasitoids. He centered his model on the probability of a given host's escaping parasitism; his basic assumptions were that parasitoids distribute their attacks at random and that they are unable to avoid superparasitism. In contradiction with his expectation that "insect parasitoids form a harmonious system in which the specific activities of the beings involved result in a relatively stable equilibrium,"

42

CHAPTER FOUR

Thompson's (1924) model always leads to the extinction of both populations. As he had set his model in a biological control scenario, he did not seem to be troubled by the contradiction (see Mills, chapter 14). To build his model, Thompson (1924) made the restrictive assumption that parasitoid reproduction was independent of the number of hosts encountered. Nicholson and Bailey (1935) overcame this limitation by assuming that parasitized hosts become parasitoids in the next generation, and that each host attacked produces a single adult parasitoid (see Heimpel, chapter 3). They also assumed that each healthy host produces a constant number of offspring (k). Their model is well known:

Nt+1 =

XNtf{NpP}

P,+i = N, (1 - f{N,P}) where Nt and Pt are the numbers of hosts and parasitoids at generation t, and f{Nt, P} is the proportion of hosts escaping parasitism. Under the assumption of a random distribution of attacks, the proportion of hosts escaping parasitism is given by the zero term of the Poisson distribution, f{Nt, Pt} = e~a, where a is the mean number of attacks per host. Nicholson and Bailey (1935) assumed a linear (infinitely rising) relationship between the number of attacks produced per parasitoid and the number of hosts (the "functional response"): Na/Pt = (a/A) T Nt, where Na is the number of hosts attacked, a is the area searched for hosts by a parasitoid in a unit of time, A is the total area of the environment, and T is the generation time, normally taken as unity. The mean number of attacks per host is a = Na/Nt = (a/A) TPt, and f{Nt, Pt} = e x p ( - a ' Pt), where the searching efficiency is a' = (a/A) T. In clear contradiction to the empirical fact that parasitoids and hosts may coexist, this model predicts that both populations should oscillate with ever increasing amplitude. To explain this contradiction, the authors proposed mechanisms that could check the oscillations: (1) a spatially extended framework in which the asynchrony between local populations would lead to the overall persistence of the system; (2) a behavioral change in the parasitoids that would lead to the more efficient exploitation of host-rich patches; and (3) the influence of other density-dependent processes. Nicholson and Bailey seldom get any credit for these far-seeing predictions that anticipated nearly sixty-five years of subsequent work. Years later, Bailey, Nicholson, and Williams (1962) acknowledged the divergence between the Nicholson-Bailey model and the observation of stability in the field and set the modern program of looking for phenomena that would dampen oscillations.

HOST-PARASITOID MODELS

43

Type III Functional Responses Twenty-five years after Nicholson and Bailey, C. S. Holling (1959a) stated that the linear (type I) functional response assumed by the Nicholson-Bailey model does not describe the facts and explored what would be more realistic relationships between the number of hosts (or prey) attacked per parasitoid (or predator) and the number of victims available. Holling showed that different factors (mainly the time elapsed in handling each prey item) would set an upper limit to the number attacked and lead to a decelerating relationship (a type II functional response). He did not, however, inquire what the consequences of this would be for the dynamics of predator-prey systems. It turned out that a type II functional response further contributes to instability (Hassell and May 1973). Holling (1959b) suggested that a sigmoid (type III) functional response would lead to the stability lacking in the Nicholson-Bailey model. Years later, Hassell and colleagues (1977) claimed that type III functional responses are not atypical of invertebrate predators and parasitoids. They suggested that this type of functional response is likely to arise whenever there is a threshold host density below which the efficiency of search by parasitoids over hosts decreases—for instance, when hosts are difficult to locate. We had then, for the first time, a realistic candidate to stabilize the Nichoson-Bailey model. The possible influence of the functional response on host-parasitoid and predator-prey systems stimulated much field and experimental work, and many empirically fitted functional responses were published (most of them of type II). Nevertheless, this first candidate to stabilize host-parasitoid systems was not long-lived: In 1978, Hassell and Comins showed that, because of the one-generation time delay between changes in parasitoid population density and the level of host mortality, no functional response (whether type III or otherwise) in which the area of discovery is exclusively a function of the number of hosts (i.e., of the form a = g{N}) can stabilize the NicholsonBailey model. In a continuous-time framework, a type III functional response can stabilize a host-parasitoid system, but only under restrictive conditions (Oaten and Murdoch 1975). In conjunction with these theoretical demonstrations, few data sets were found in which the functional response for parasitoids was unambiguously of type III (two examples of type III functional responses in parasitoids are given in Hassell et al. 1977). The fact that functional responses per se cannot stabilize a NicholsonBailey model did not end interest in them. Functional responses are an essential constituent of any predator-prey and host-parasitoid system, especially when attempting to model them from elementary behavioral processes (Casas et al. 1993).

44

CHAPTER FOUR

Mutual Interference Hassell and Varley (1969) identified another weakness of the NicholsonBailey model: the assumption of a constant searching efficiency by parasitoids. They showed that in many data sets, this value decreases with the number of parasitoids. In a log-log transformation, the relationship was approximately linear: log(a') = log(g) — m log (P) or a' = QPtm.

(4.1)

They called Q the "quest constant" and m the "coefficient of interference" (for a detailed definition of the different forms interference and a discussion of the processes leading to them, see Visser et al. 1999). Hassell and Varley (1969) introduced interference into the Nicholson-Bailey model, as in equation 4.1, and showed by simulation that for m > 0, the oscillations damped to a stable equilibrium. The suggestion that mutual interference would stabilize host-parasitoid systems was formalized by Hassell and May in 1973. They showed that, provided the values of X are not very large, even relatively low values for m (on the order of m > 0.25) will contribute markedly to stability and may even give complete stability. These authors include a table of values of m found in the literature. In it, m was always higher than 0.25, and they suggested that mutual interference could help account for the observed stability of host-parasitoid systems. These results led to a small cottage industry of measuring interference coefficients. Lists of such coefficients were tabulated in works such as Hassell (1976, 1978). The purely empirical approach used by Hassell and Varley (1969) was criticized both by Hassell and Rogers (1972) and Beddington (1975), who proposed models based on assumptions about the individual level consequences of encounters between parasitoids. Beddington (1975) assumed that upon each encounter, a parasitoid would waste tw units of time. Each parasitoid could encounter P — 1 conspecifics at a rate b/A, where b is area perceived (i.e., the area in which it can discover conspecifics) by a predator per unit of time. Under these assumptions and after some elementary algebra the searching efficiency becomes

a' = (a/A)/[I + (b/A)tw(Pt-l)]. Free and colleagues (1977) showed that for the Beddington (1975) model: m* = btw inX /(aT) where m* is the value of the coefficient of interference at the equilibrium density, and T is the generation time. Hence, because the constants a and b

HOST-PARASITOID MODELS

45

are essentially similar (both depend on parasitoid behavior), m* scales approximately as tJT. Thus, m* is determined by the ratio of the time a parasitoid wastes through interference at a single encounter to the duration of its adult life, a ratio that is much lower than the theoretically used values for m*. Free and coworkers (1977) concluded that, in general, parasitoid mutual interference is too weak to have any real influence on host-parasitoid stability. This result was perceived as a mortal blow to mutual interference as an important stabilizing process. Note that the idea of the stabilizing influence of interference was virtually abandoned just because a single theoretical study suggested that this process was unimportant. Biologists, or at least theoreticians, seemed prepared to accept that a model can be invalidated by another model, even in the absence of any data. In fact, developments fifteen years later suggested that the epitaph to mutual interference might have been pronounced too prematurely. Arditi and Akcakaya (1990) pointed out that the classical way in which a' is calculated assumes a type I functional response. They showed that both for predators and for indiscriminant parasitoids, if handling time is not negligible (i.e., type II functional responses), the regression of log a' as a function of log P will lead to severe underestimations of m. They proposed a new method to calculate this coefficient and applied it to different data sets in the literature. For both predators and parasitoids, they obtained large values of m, most of them not significantly different from unity; with one exception out of seven parasitoids, they obtained values of m > 0.6. Ruxton and colleagues (1992), employing a behavior-based model, aimed to overcome some of the limitations of Beddington (1975). For instance, Beddington (1975) assumed that a predator can interact with any conspecific in the population, independently of its current behavior (searching, handling prey, or already engaged in another encounter). Ruxton and coworkers (1992) considered that, at any given time, a predator population (denoted P) is divided into three subpopulations: individuals searching for prey ("searchers," S), those handling prey ("handlers," H), and those engaged in aggressive encounters with conspecifics ("wasters," W). The population states are mutually exclusive, so that S + H + W = P. They also assumed that a searcher becomes a handler by finding a prey item, and that two searchers become wasters by meeting each other. Handlers and wasters decay into searchers at rates equal to the inverse of mean handling time (th), and wasting time (tw), respectively. The authors adopted a chemical kinetics approach and derived a differential equation for each subpopulation. By assuming that aggressive interactions and prey captures take place on a much shorter time scale than predator birth and death processes (P constant),

46

they deduced the in prey searching response, written the environment)

CHAPTER FOUR

equilibrium proportion of the predator population engaged (denoted S*/P). The intake rate per predator (the functional I{F,P}, with F being the number of prey items present in is calculated as I{F,P} = Vs (S*/P)F

where Vs is the volume searched by a predator in a unit of time (a threedimensional equivalent of the area of discovery). Ruxton and colleagues (1992) introduced this functional response into two predator-prey models. The first is an unstructured, two-differential equation model. The second is an age-structured, delay-differential equation model (a continuous-time equivalent of the Nicholson-Bailey difference equation model). Continuous, age-structured models have the interesting property of presenting two types of cyclic behavior: damped or undamped multigeneration oscillations or single-generation cycles (Gurney and Nisbet 1985; Godfray and Hassell 1989). Ruxton and coworkers (1992) showed that even weak interference can stabilize either the unstructured model or the age-structured version. In fact, the influence of interference is stronger in the latter model. This applies to multigeneration cycles. The effectiveness of mutual interference against single-generation cycles is much weaker, and there are substantial regions of parameter space in which single-generation cycles are impossible to eliminate. While entomologists lost interest in interference, ecologists and behavioral ecologists working mainly on birds found clear evidence of the influence of this phenomenon in natural populations (see, for instance, Ens and GossCustard 1984; Goss-Custard and Durrell 1987). Theoretical developments in behavioral ecology also renewed the interest in interference. Fretwell and Lucas (1970) introduced the notion of the ideal free distribution (IFD): the frequency-dependent equilibrium distribution of consumers of a heterogeneously distributed resource. At the IFD, all consumers get the same rate of gain and no animal can increase its share by moving elsewhere. The IFD predicts that if there is no interference between consumers (m = 0), then all animals should concentrate on depleting the richest patch. When food availability in the richest patch reaches the value of the second richest one, the consumers should distribute equally between them, until they are depleted to the level of the third richest patch, and so on. Such a process has never been observed under natural conditions. In fact, animals tend to aggregate predominantly in the richest patches, leaving few poor patches unexploited. This is what is predicted by the IFD when m > 0 (Sutherland 1983). Ignoring interference is not the only factor that may lead to discrepancies between experimental data and the IFD. As in many other optimality models, this

HOST-PA RASITOID

MODELS

47

theory assumes that animals have full knowledge of their environment (they are "omniscient"). Bernstein and colleagues (1988, 1991) have explored the influence of constraints in information acquisition on the distribution of consumers of a heterogeneously distributed resource. They showed that the influence of deviations from full knowledge would be to lead to less aggregated distributions (see also Abrahams 1986; Gray and Kennedy 1994). A behavioral-based approach to modeling interference similar to that of Beddington (1975) was used by van der Meer and Ens (1997) and by Stillman and colleagues (1997). Van der Meer and Ens (1997) compared different expressions for mutual interference from the literature and analyzed these in terms of their effects on predator distribution at the ideal free distribution. They showed that the Hassell and Varley (1969) and the Ruxton and coworkers (1992) models lead to very similar aggregation responses. Stillman and colleagues (1997) demonstrated that differences in dominance (i.e., the probability of winning a fight) between the animals, and the use of optimality rules to decide whether to fight or retreat, can have a strong influence on the distribution of predators in a patchy environment. To sum up, the question of the real influence of mutual interference on the dynamics of host-parasitoid systems is still an open one. There is, nevertheless, a renewed interest in the subject because of the possible influence of interference on consumer distribution over a heterogeneously distributed resource. Recent years have seen an increasing interest in population models based on individual behavior (Sutherland 1996). One of the reasons for this endeavor is that it is hoped that these models would better predict how key population parameters that depend on behavior change under different situations. In this regard, interference can be an interesting starting point to build host-parasitoid models. Because of the high level of realism that models based on individual behavior can achieve, and because of the detailed experimental work they stimulate, we might expect that the problem of the real influence of mutual interference will be solved in the near future. Better still, it is likely that new surprises and new unsolved problems will arise that will enrich our general understanding of host-parasitoid systems.

Spatial Heterogeneity Early Models The interest in spatial heterogeneity as a factor potentially contributing to the stability of host-parasitoid systems started in earnest with the classical experiments by Huffaker (1958) and Pimentel and coworkers (1963), both of whom showed that spatially extended environments lead to more persistent

48

CHAPTER FOUR

relationships than simpler systems. Pimentel and coworkers' system consisted of a series of boxes connected by pipes. They studied the relationship between houseflies and the parasitoid Nasonia vitripenis and compared the persistence of the system in environments of single boxes with environments of sixteen and thirty interconnected boxes. While the system never persisted for more than sixteen weeks in single boxes, in an environment of sixteen boxes it persisted for more than thirty weeks, and in thirty boxes for at least eighty weeks. Hassell and May (1973) presented a theoretical model that is not too different from Pimentel's experimental design. They considered an environment formed by q cells. Host and parasitoids are distributed among the cells at the beginning of each generation and in each cell parasitoids search randomly (Poisson) for hosts. If pt parasitoids search for hosts in a given patch with ht hosts, at the end of the generation the number of surviving hosts will be hte~a pt. If all surviving hosts disperse in the environment to lay eggs, the total number of hosts at generation t +1 will be

Nt+l = X Z, hjexp(-a'

p^

where A, is the host's finite rate of increase. If a, and (3, are, respectively, the proportions of hosts and parasitoids in patch i and if we assume that Pj = c of

(4.2)

then q

Nl+1 = X N, 2 a,- exp(-a' c a? Pt). The equation for the number of parasitoids at generation t + 7, Pt+j, is derived in a similar way. The aggregation coefficient, |x controls how parasitoids distribute as a function of the number of hosts in each patch. On the basis of optimality considerations, it was expected that parasitoids would aggregate in the richest patches, that is, that |x would be positive and take high values. Hassell and May (1973) showed that if the environment is sufficiently heterogeneous, and (x and q are high ((JL > 1 for relatively large environments), the system would be stable. This was a very interesting result, because it pointed to a putative stabilizing principle that is well rooted in the knowledge and expectations of animal behavior. This stimulated the study of the distribution of parasitism under

HOST-PARASITOID MODELS

49

field conditions. Two reviews of this work (Stiling 1987; Walde and Murdoch 1988) list more than one hundred studies. Only one-quarter of these studies showed the kind of spatial density dependence that was expected. However, Bernstein and colleagues (1991) have shown that sampling scale has a strong influence on these results. For instance, nearly 100% of the results in Stiling 1987 that took the plant as the sampling unit showed density-dependent aggregation. Sutherland (1983) added one of the most elegant chapters of this saga. Without making any assumption about the form of parasitoid distribution as a function of host distribution, he showed that if interference is modeled as in equation 4.1, at the IFD, parasitoids will aggregate following equation 4.2 with |x = Mm. This result stresses how tight the association is between interference and the distribution of parasitoids in a heterogeneous environment. To study the influence of spatial heterogeneity on population dynamics, May (1978) used an approach that differed from that of Hassell and May (1973). He considered that the overall consequence of spatial heterogeneity would be that some hosts would be more susceptible than others to parasitoid attacks. In this case, at the whole population level, the attacks will not follow a Poisson distribution but as another, unspecified, distribution. He proposed to approximate this distribution by the negative binomial. May (1978) quoted different work showing that the negative binomial describes the distribution of attacks in many field populations well. He also provided a more formal argument by showing that, if the within-patch distribution of attacks is Poisson and the distribution of parasitoids among patches is independent of that of hosts and is gamma distributed, then the compounded distribution of attacks will be negative binomial. In this case, the probability that a given host will escape parasitism is

f{Nt, P}= [1 + a'PAl -k where k is the parameter that controls the degree of aggregation. The smaller the value of k, the more aggregated the distribution. At the limit k—>°°, the Poisson distribution is recovered. May (1978) showed that k < 1 is the local stability condition for a Nicholson-Bailey model where the probability of escaping parasitism is given by the zero term of the negative binomial. He also showed that there is a tight connection between the distribution of attacks and that of the parasitoids among patches. Indeed, k = (l/CV)2 where CV is the coefficient of variation of the among-patch distribution of parasitoids. The stability condition is then CV2 > 1.

50

CHAPTER FOUR

This result led to a series of very interesting developments (see below) and is known in the literature as the "CV2 rule" (Hassell and Pacala 1990). An interesting twist to the story is that in 1962, Bailey and coworkers had already proposed the use of the negative binomial using the same theoretical argument, but a somewhat more elegant formalism. Bailey and coworkers (1962) proposed the negative binomial as a particular case of a more general problem. While they found that the conditions for stability under a negative binomial distribution are improbable in natural settings, based on some experimental data, May (1978) concluded the opposite. A General Rule? In a series of papers Hassell, Pacala, and coworkers (Hassell and Pacala 1990; Godfray and Hassell 1990; Hassell et al. 1991b; Pacala et al. 1990; Pacala and Hassell 1991) suggested that the CV2 rule had a more general scope. They considered five different situations: (1) The original May (1978) model, in which the distribution of parasitoids is host-density-independent (HDI, or "pure error model," Chesson and Murdoch 1986). In this case the CV2 rule gives an exact prediction of the behavior of the system. (2) A model in which it is assumed that the distribution of parasitoids is hostdensity-dependent (HDD or "pure regression model," Chesson and Murdoch 1986) and follows equation 4.2, and that hosts among patches are gamma distributed. The authors show that in this case the CV2 rule gives an approximate, but acceptable, prediction of the regulation of the system (especially if the rate of increase of the prey is not too high). The main difference from the original CV2 rule is that now, the coefficient of variation refers to the density of parasitoids per patch weighted by the number of hosts in that patch. (3) A model that combines both HDI and HDD components, and for which the same result holds. (4) In a somewhat different context, Godfray and Hassell (1991) analyzed a model that considers the immunological response of hosts to parasitoids and in which the probability that a given parasitoid egg will be encapsulated (i.e., rejected by its host) is a random variable (see Godfray, chapter 9). Godfray and Hassell (1991) showed that also in this case the CV2 rule gives an adequate prediction of the behavior of the model. (5) Finally, Hassell and May (1988) studied a model in which it is assumed that only a fraction of the hosts and parasitoids that emerge in a given patch abandon it to lay elsewhere. Also in this case, the CV2 rule gives a good prediction of the stability of the model. These results were extremely attractive, not only because the CV2 rule was so compact and seemed so general but also because Pacala and coworkers showed how the CV2 could be estimated from field data (Pacala and Hassell 1991). They found that the CV2 could be approximated as

HOST-PARASITOID MODELS

51

2

CV « CjCD - 1 where C7 represents the component of heterogeneity that is independent of host density and CD represents the component that depends on host density. Measures of the distribution of searching parasitoids are rarely available for natural populations. In contrast, many authors have measured the relationship between percent parasitism and host density per patch. Hassell and Pacala (1990) and Pacala and Hassell (1991) proposed a maximum likelihood procedure to estimate Q and CD from these data sets. They analyzed sixty-five such sets and found that only in eighteen cases, CV2 > 1. Furthermore, in fourteen of these cases, CD «* 1 and C7 > 2, which means that Q alone suffices to make CV2 > 1. In summary, contrary to conventional wisdom and what has been proposed by some of the authors of this work, spatial heterogeneity did not appear to be a strong stabilizing principle. When it did, it seemed to be mainly as a consequence of variation that cannot be explained by the behavioral response of parasitoids to host densities within patches. Of course, this conclusion depends on the validity of the models, which deserves close scrutiny. Two key (and unanswered) questions are: How robust are these models? and How well do they capture what actually happens in the field? Taylor (1993) shows that there are variety of factors that can interact or combine with aggregation to alter the accuracy of the CV2 stability criterion or entirely eliminate the effect of aggregation. On the other hand, as is often the case with these theoretical developments, the interest in the rule faded away before any real experimental test could be performed.

Controversy While the results of Hassell, Pacala, and coworkers were in development, the full idea of the stabilizing influence of spatial heterogeneity experienced another vigorous assault. Murdoch and Stewart-Oaten (1989) pointed out that most of the models studied so far considered that there was a single episode in the distribution of hosts and parasitoids, at the beginning of each generation. They suggested that the results obtained depended critically on this assumption. To overcome this limitation, Murdoch and Stewart-Oaten (1989) adopted a continuous time framework in the form of the classical Lotka-Volterra model: dh/dt =

ah — g{h}p

dp/dt =

cg{h}p — dp

52

CHAPTER FOUR

where h and p are the average number of hosts and parasitoids per patch; specifically, h = E(H) and p = E(P), where H and P are the numbers of host and parasitoids in a randomly chosen patch. g{h} is the functional response. Murdoch and Stewart-Oaten (1989) assumed that the functional response is linear (type I), and that the distribution of parasitoids is dependent on the distribution of hosts. The instantaneous death rate of hosts in a patch is bHP, and the average death rate over all patches is

bE(HP) = b(hp + cov(H.P)) where the covariance describes how the parasitoid distribution depends on local host density. If H and P are independently distributed, cov(//,P) = 0, so E(HP) = hp and the classical, neutrally stable, Lotka-Volterra model is recovered; this is independent of how variable the distribution of parasitoids is among patches. This suggests that HDI aggregation has no influence on the stability of host-parasitoid systems. If the parasitoids aggregate in the richest patches, co\(H,P) > 0. This could lead to system stability if the functional response is accelerating (type III). Murdoch and Stewart-Oaten assumed that the proportion of parasitoids in a patch increases linearly with the proportion of hosts in that patch. In this case, the form of the functional response depends critically on the variance in the number of hosts among patches. The authors suggested that under most natural conditions, the distribution of hosts in the environment is such that it will lead to type II functional responses. On this basis, they argued that HDD aggregation would promote the instability of host-parasitoid systems. They then analyzed some more general models and reach a similar conclusion. Murdoch and Stewart-Oaten (1989) argued that the contrasting results obtained both in the Nicholson-Bailey framework and in their own stem from an unrealistic assumption of the first: the absence of within-generation dynamics. In this sense, the influence of spatial heterogeneity on stability would be an artifact. By not allowing the parasitoids to redistribute, the Nicholson-Bailey model would force parasitoids to become increasingly inefficient as their numbers increase (the patches attacked become more rapidly depleted), thus leading to "pseudo-interference" (as found in Free et al. 1997), a process that would have the same stabilizing effect as real interference. Real animals are not constrained by handy mathematical assumptions, and they would simply move to more profitable patches. Many people have risen to meet the challenge set by Murdoch and Stewart-Oaten (1989). Godfray and Pacala (1992) pointed out several weaknesses in this work. They showed that Murdoch and Stewart-Oaten's (1989) results for both HDI and HDD can be obtained from a Nicholson-Bailey framework by assuming infinitely fast dispersal, and suggested that this contradicts the observation that, in many cases, hosts are immobile (lepidoptera eggs) or are

HOST-PARASITOID MODELS

53

relatively sedentary (larvae). Moreover, Godfray and Pacala (1992) argued that Murdoch and Stewart-Oaten's (1989) conclusion that density-independent aggregation does not affect stability is structurally unstable to the introduction of finite dispersal rates. Rohani and colleagues (1994) made a most valuable contribution to this discussion. They analyzed a model that is an extension of the classical one by Hassell and May (1973) that we have already discussed. As in the earlier model, at the beginning of each season, female hosts and parasitoids disperse in the environment to lay eggs in the different patches. The initial parasitoid distribution can be either HDI or HDD. In the latter case, the proportion of parasitoids in each patch is given by equation 4.2. The model also has within-season dynamics. During the season, hosts mature and pass through different stages (eggs, larvae, pupae, and adults). Only host larvae are attacked by parasitoids. Parasitoids migrate between patches in either an HDI or an HDD manner. A key parameter of this model is M, the within-season parasitoid dispersal rate. For the HDD situation, when M = 0 the original Hassell and May (1973) model is recovered and aggregation is highly stabilizing. The movement of parasitoids between patches markedly reduces the stabilizing influence of parasitoid aggregation. This is so because the redistribution of parasitoids allows them to move away from areas already depleted, decreasing pseudo-interference. Parasitoid migration might also make some contribution to stability, as shown by the model when the initial distribution of parasitoids is density-independent. In contrast with these results, Rohani and colleagues (1994) found that the stabilizing influence of initial HDI aggregation was not influenced by (HDI) parasitoid movement. Because parasitoid movement is not affected by host density, the net movements into and out of patches are equal and preserve the aggregated distribution of parasitoids across patches. One of the things that make these results so interesting is the fact that they coincide with the field data analysis of Pacala and Hassell (1991) and Hassell and Pacala (1990), which suggest that HDI aggregation has a more prevalent stabilizing influence than HDD. This does not mean that the puzzle was solved, however, as we will see below.

The Influence of Time Scales Ives (1992a) explored the possible influence of time scales and the timing of dispersal on the stability of host-parasitoid systems. To do this, he developed three models that differed in which developmental stage of the host is susceptible to parasitoid attacks, in the duration of the host susceptible stage, and with respect to the parasitoid's dispersal rate. The first model does not consider different developmental stages, dispersal of both hosts and parasitoids is continuous, and dispersal rates are of the

54

CHAPTER FOUR

same order of magnitude as reproduction, death, and parasitism rates. Parasitoid dispersal is assumed to be host-density dependent. Ives (1992a) showed that, in general, spatial heterogeneity and parasitoid aggregation have a stabilizing influence, but there are conditions where strong aggregation might have a destabilizing influence. Ives's (1992a) second model divides the host population into a sedentary susceptible stage, and a dispersing adult stage. It is assumed that the parasitoid's emigration rate depends on the number of hosts in a patch relative to the average number of hosts across all patches. Parasitoid dispersal is assumed to be very rapid. In this model, parasitoid aggregation has a stabilizing influence. In contrast with the previous model, the influence of spatial heterogeneity depends on the strength of parasitoid aggregation: If parasitoid aggregation is strong, increasing heterogeneity is stabilizing; if parasitoid aggregation is weak, increasing heterogeneity is destabilizing. Ives's third model considers, as does the previous one, that the host population is divided into two stages, a mobile adult stage and a sedentary juvenile one that is attacked by the parasitoids. This model also assumes that the susceptible stage is very short and that parasitoids do not distinguish between healthy and parasitized hosts, and that parasitoids move very rapidly among patches. This model is always stable but in it, both spatial heterogeneity and aggregation can promote either stability or instability (measured by the tendency of the dynamics to oscillate), depending on the parameter values. These results suggest that the effects of spatial heterogeneity and parasitoid aggregation on population dynamics depend both on the model and on the parameter values. This shows the difficulties in making generalizations about the stability of host-parasitoid systems and calls for the inclusion of more realistic descriptions of the spatial and temporal structure of populations and of animal behavior. This has stimulated work in two directions: the study of optimal and realistic migration decisions in host-parasitoid models (van Baalen and Sabelis 1993; Holt 1985; Krivan 1997; Bernstein et al. 1999) and the experimental study of these emigration decisions by parasitoids (for a review, see Driessen and Bernstein 1999), but this is another chapter of this continuing story that I will not tell here.

Discussion and Future Directions In the previous section, I discussed three processes that were proposed as stabilizing host-parasitoid systems. One of them, type III functional responses, seems unimportant both in terms of its occurrence in nature and in terms of what models predict its real influence would be. The discussion of

HOST-PARASITOID MODELS

55

the real influence of the other two processes, mutual interference and parasitoid dispersal in a heterogeneous environment, is still open. The three main questions are those we should ask about any model: How well do these models capture the basic ingredients of the problems? How robust are the predictions to the structure of the models? How well do the models predict the behavior of real systems? From the point of view of ingredients and assumptions, the processes included in the models have made important strides toward realism. This is, in fact, one of the domains in which the theory of host-parasitoid dynamics has been most successful: It has prompted biologists to study empirically based phenomena (more on this below). The problem of model robustness has increasingly come to the fore. Different authors tackle the same basic theoretical problem using different mathematical formalisms, different population structures, and different representations of the processes under study. The proliferation of models, perceived by many experimentalists as demonstrating that theory is a futile enterprise, will help reveal whether model predictions are robust, provided we find the way to put some order in the different, sometimes contradictory, answers we get. A powerful approach to contribute to the first two questions is to begin at the individual level. Behavioral ecology leads progressively to a clearer understanding of animal behavior with a strong evolutionary emphasis. Optimal strategies are a concise way to include animal behavior into host-parasitoid and predator-prey models to explore the consequences for population dynamics, regardless of whether animals converge to such strategies (see Roitberg, chapter 16, for applications). Two methodological developments are of great help. On the one hand, both delay-differential equations (Gurney et al. 1983, Murdoch et al. 1987, Gordon et al. 1991) and aggregation techniques (Auger and Poggiale 1996, Michalski et al. 1997, Bernstein et al. 1999) allow the description of realistic situations that combine different time scales and demographical (age structure, aging, reproduction) and behavioral (prey capture, migration) processes without sacrificing mathematical tractability. On the other hand, fast and inexpensive computers allow the development of individual-based models that can help explore the consequences of different behavioral strategies. Nevertheless, one should keep in mind that this is essentially a force brute approach, often leading to results that are difficult to interpret or generalize in the absence of a clear theoretical framework. This individual-centered approach can help explore the influence of different aspects of behavior in isolation or in small groups, but not their relative influence on whole systems. This is why Casas (chapter 2) calls for a more field-based choice of the aspects of behavior we study. The weakest point of all in this endeavor is the almost complete absence

56

CHAPTER FOUR

of quantitative comparisons between model and real-system behaviors. This is needed to evaluate how accurately models predict reality and to detect missing ingredients. There are two reasons for the rarity of quantitative data. One is well known: We lack a sufficient number of long-term field studies of host-parasitoid systems under natural conditions. The fact that repeating this is becoming a sort of rite in reviews and discussions like the present one does not make it less true. I cannot offer any substitutes for the comparison between data and models, and I do not think there are any. The second reason is somewhat more subtle: because of the intrinsic conflict among generality, realism, and precision in ecological models (Levins 1966), most theoretical models (that aim to reach some degree of generality) cannot be expected simultaneously to simulate and to predict the dynamics of any particular host-parasitoid system. One of the main consequences of these shortcomings is that biologists are sometimes constrained to compare models with models, and not with data. Case models (empirically based, mechanistic formulations tailored to particular systems) play a very important role. They can include a high degree of biological realism and make it possible to evaluate assumptions and validate predictions. These models also allow one to compare the relative importance of different processes and their interactions. For instance, Hochberg and coworkers (1996) explored the relationship between individual searching behavior in the parasitoid Ichneumon eumerus and refuges from parasitism for its host, Maculinea rebeli, on the dynamics of the system. The story I have just told can be read in very different ways. If the aim of models was to accurately predict the behavior of natural populations and to explain the persistence of host-parasitoids systems, this would be the story of a clear failure: Of the first three candidates to stabilize host-parasitoid systems, one is ineffective, and of the remaining two, the best that can be said is that the evidence is contradictory. A more bold suggestion would be that host-parasitoid systems are not stable at the level studied by these models, and that such theoretical modeling is a futile exercise. Proponents of this point of view are not lacking. Besides some old-fashioned biologists who resent the introduction of mathematics into ecology, some authors have called for a resignation of pursuing what is perceived as an unreachable goal (Judson 1994) and, instead, building detail-rich models in which the fate of each individual is followed through time (individual-oriented and individualbased models, Uchmanski and Grimm 1996). This approach leads to models that can only be solved numerically and which, in some cases, are strongly dependent on the characteristics of the particular system in mind. The danger here is a progressive abandoning of the aim of building a general conceptual framework that could help understand how nature works. If the aim of models is to generate new hypotheses, to understand what might be the influence of different processes, and so to build a general con-

HOST-PARASITOID MODELS

57

ceptual framework and to guide experimental work, then the verdict may be quite different. Theoretical and experimental work have stimulated and fertilized one another. Theoretical work has shown the possible influence of different biological factors. Experimental work has studied the workings of such factors and corrected some of the assumptions introduced into the models. Even if theoretical work has not yet given us the right answers, it has done a lot to help us ask the right questions. Maybe the two most frustrating aspects of the full endeavor are linked to the time it takes to make progress. On one hand, as we have seen, interesting theoretical developments and concepts are sometimes completely forgotten soon after they are published, and it may take decades for them to be rediscovered. On the other hand, the different investments in terms of both resources and time required by theoretical and experimental work makes it difficult for theory and experiments to make progress hand in hand. Both these problems result in a waste of time and effort. Up to now, attempts to identify unambiguously the processes that stabilize host-parasitoid systems have been a failure in my view, but a successful failure that has resulted in a progressive understanding of host-parasitoid systems. Other domains of ecology have not reached the same level of development. Probably what they lacked is models that "failed" as clearly as the Nicholson-Bailey model. Acknowledgments. I would like to thank Michael Hochberg, Tony Ives, Jerome Casas, and Bernard Roitberg for all their suggestions and comments that helped me improve this text. Yoram Ayal, Manu Desouhant, Christian Gautier, and Thierry Spataro commented on earlier versions and spotted different weaknesses. I have done my best to correct them.

Five A Field Guide to Studying Spatial Pattern Formation in Host-Parasitoid Systems SUSAN HARRISON

INSECT herbivores often show patchy spatial distributions, which are not always successfully explained by variation in environmental factors such as host-plant quality or predator abundance. For more than twenty years (e.g., Segel and Jackson 1972; Levin 1976), a branch of mathematical theory has predicted that patchiness in population distributions may arise intrinsically when mobile and specialized natural enemies interact with their less-mobile victims (reviewed in Kareiva 1990; Deutschmann et al. 1993; McLaughlin and Roughgarden 1993; Holmes et al. 1994). Despite decades of theoretical work on this subject, and despite the preeminence of insect parasitoids as mobile and specialized natural enemies, virtually no empirical work has tested the idea of intrinsic spatial pattern formation in host-parasitoid systems. This is perhaps because the theory is quite abstract, and remains inaccessible or simply unknown to many empiricists. Another obvious reason is that parasitoid population dynamics and dispersal are far harder to study in the field than in theory.

The goals of this chapter are to explain the theory of spatial pattern formation in empiricists' terms, and to suggest appropriate systems and techniques for testing it in the field (for landscape perspectives, see Roland, chapter 7; Tscharntke, chapter 15). I will then describe one study suggesting that hostparasitoid spatial dynamics may be an important mechanism of regulation of the host population.

Theoretical Background Early ecological work on spatial pattern formation was motivated by questions such as "Why are plankton patchy even though the ocean is continuous?" (Levin and Segel 1976). Theoreticians used the reaction-diffusion model first introduced by Turing (1952) to portray a predator and prey "reacting" according to Lotka-Volterra or similar continuous-time dynamics, and "diffusing" by moving randomly through continuous space (e.g., Segel and Jackson 1972; Levin 1976; Mimura and Murray 1978; Okubo 1980;

STUDYING SPATIAL PATTERN FORMATION

59

Conway 1984). In the majority of ecological reaction-diffusion models, reviewed by Kareiva (1990) and Holmes and colleagues (1994), two features lead to the formation of fixed spatial patterns in predator and prey population densities. First, the predator diffuses faster than the prey. Second, prey population growth is autocatalytic, i.e., inversely density-dependent at low densities (however, alternative assumptions about prey growth are possible as well; see below). In biological terms, what happens in these models is that the prey population begins to grow in a localized area, leading to an incipient patch of prey. This local increase in prey causes the predator population to grow as well. Because it disperses more quickly, the predator spills over the edges of the newly forming patch of prey, which creates a peripheral zone in which predator-prey ratios are elevated. In turn, this "halo" or "zone of death" around the prey patch locks the patch into place, because prey dispersing out the patch suffer a very high risk of being eaten. But within the patch, the prey are saved from annihilation by the constant diffusive loss of part of the predator population. The end result is the formation of standing waves in the densities of both species through space, with a wavelength determined by the predator's dispersal abilities. Models of this kind have been applied in other areas of biology—for example, to explain pattern formation during embryonic development (Murray 1989). In their ecological manifestation, reaction-diffusion models suffer from a number of questionable assumptions. Perhaps the most obvious is that both the predator and the prey are passive, dispersing independently of their own and each others' densities. Also, while inverse density dependence in the prey may be contrived in a number of ways, such as by giving the predator a limited appetite and a population ceiling, these assumptions may not always be met in nature. Finally, the model applies only to systems in which the predator and prey strongly control one another's abundance. Some of these limitations have been relaxed in further elaborations of the theory. One significant development was the recasting of the original continuoustime models into discrete-time models based on integro-difference equations (Kot 1992; Neubert et al. 1995; Kot et al. 1996). Space is still continuous in integro-difference models, and dispersal is governed by probability distributions known as "contact distributions" or "dispersal kernels." Besides being more biologically appropriate for annual species than the original continuoustime models, discrete-time formulations reveal a broader set of conditions under which spatial patterns can form. Inversely density-dependent prey growth is not needed. Also, a wide range of empirically derived shapes for the dispersal function can be shown to yield pattern (e.g., Gaussian, exponential, or leptokurtic). Related integro-difference models yield traveling, rather than static, waves in population density, and have been used to study invasions (Kot 1992; Kot et al. 1996). Still, this class of models remains

60

CHAPTER FIVE

somewhat unrealistic in that dispersal is independent of species densities. Predators, in other words, are assumed not to move preferentially toward their prey. More flexible, and possibly realistic, movement rules have been examined in a few analytic models, mostly in continuous space and time. For example, Turchin (1989) and Lewis (1994) modeled the population consequences of aggregative behavior. Holmes and colleagues (1994) suggest that, although strong aggregation of predators toward their prey will annihilate spatial pattern, a weak (undercompensating) tendency for predators to aggregate to prey could actually promote pattern formation. In fact, such a spatial response could substitute to some extent for a numerical response by predators. Pattern formation in predator-prey systems in discrete space, as well as discrete time, has seldom been examined analytically. Instead, it has been the province of individually based models, cellular automata, and coupled map lattices (e.g., de Roos et al. 1991; Hassell et al. 1991a; Molofsky 1994; Ruxton and Rohani 1996). The focus in much of this work has been on whether explicit space can stabilize the temporal dynamics of predator-prey interactions. Such models have also been used to ask whether spatial dynamics can promote coexistence in competing parasitoids. In general, compared with continuous-space models, discrete-space simulations make it even easier to generate spatial patterns in population density within a uniform environment. However, adding random variability in host or parasitoid demography may sometimes break down these self-generated patterns (Ruxton and Rohani 1996; Wilson and Hassell 1997). Discrete-space simulations are potentially useful for examining the long-lasting transient behavior of spatially distributed populations, as well as for incorporating the effects of complex habitat geometry. Very few models have examined whether predator-prey interactions might reinforce underlying spatial variation in the environment, instead of creating population patchiness in a completely uniform world. However, McLaughlin and Roughgarden (1993) analyzed a continuous-space analogue of the LotkaVolterra model with a spatial gradient in the prey's carrying capacity, as well as predators that disperse faster than their prey. They found that predator abundance simply tracked the spatial pattern in prey carrying capacity, but with respect to prey abundance, the interaction amplified the variation in carrying capacity. The strength of this amplification increased with decreasing amounts of prey dispersal.

Testing for Pattern Formation in the Field This brief overview of theory on spatial pattern formation underlines two essential conditions. First, the natural enemy must be capable of reducing its

STUDYING SPATIAL PATTERN FORMATION

61

victim's abundance and of increasing its own abundance in response to that of the victim. In other words, the interaction must be a relatively strong one. This condition seems especially likely to be met by parasitoids, since they tend to be more specialized than predators. Second, the natural enemy must disperse substantially faster than its victim. Very few studies have quantified movement by parasitoids, let alone compared it to that of their hosts (but see Roland, chapter 7). In one rare exception, Jones and coworkers (1996) found that four coexisting species of parasitoid wasps all dispersed more than their shared host, a tephritid fly. Overall, the handful of available evidence suggests good dispersal ability in parasitoids, with the exception of a small number of flightless species (Godfray 1994; Hastings, chapter 6). In contrast, there is greater variation in dispersal ability among herbivorous insects; a substantial minority have flightless females—for example, many geometrid and lymantriid moths. Across taxa, insect flightlessness is associated with continuous and stable habitats, such as forests (Barbosa et al. 1989; Roff 1990; Wagner and Liebherr 1992). While it remains unclear whether parasitoids will, in general, disperse better than their host insects (see Hastings, chapter 6), winged parasitoids that happen to attack poorly dispersing hosts would seem ideal subjects for testing the theory. Of course, it is hardly worth testing a theory on spatial pattern formation unless there are natural observations that it may explain. Then the questions become: How frequently is the abundance of herbivorous insects strikingly patchy within the distribution of their host plants? In what kinds of species or systems is this most prevalent? How satisfactory are the more traditional extrinsic explanations for such patchiness, such as variation in host-plant quality, microclimate, or the abundance of generalist natural enemies? A survey of the forest insect literature reveals that it is very common for insect defoliators to remain at high densities for several years in some stands of their host trees, while being much sparser in adjacent stands (Furniss and Knopf 1971; Mason and Luck 1978; Wickman and Beckwith 1978; Berryman 1987; Harrison 1987; Mason and McManus 1981; Watt and Leather 1990; Nair 1990; Hunter et al. 1991). Cases in which patchy insect herbivory has been studied experimentally are reviewed by Mopper and Simberloff (1995). Although variation in herbivore densities may sometimes be related to variations in plant genotype (e.g., Karban 1987), environmental conditions (e.g., Harrison 1987), or plant phenology (e.g., Mopper and Simberloff 1995), such explanations fail in other cases (e.g., Hunter et al. 1991; Harrison 1997). At present, it is unknown whether a tendency toward patchy herbivory is associated with flightlessness or other life-history or ecological traits. Given an appropriate system, what experiments could be used to test the hypothesis that patchy herbivory arises from interactions between herbivores

62

CHAPTER FIVE

and their parasitoids, in the manner portrayed by models of spatial pattern formation? Perhaps the strongest prediction made by theory is that, if the dispersal of parasitoids away from patches of host insects could somehow be blocked, the patches of hosts would spread. But this would have to be accomplished in a manner that did not block the dispersal of the host insect. Neither cutting the wings off the parasitoids, nor erecting barriers permeable to host insects but not parasitoids, seems entirely feasible. Alternatively, it might be possible to eliminate the parasitoid population entirely in a tightly controlled situation, but not, for example, at the edge of a natural gypsy moth outbreak. Nor would control or replication be achieved easily in a natural system, unless a large number of independent outbreaks were available. One test almost as strong as the above, but more feasible in the field, is to experimentally measure parasitoid abundance, rates of parasitism, and host population growth as functions of the distance from naturally occurring patches of host insects. Theory provides the counterintuitive prediction that, even though parasitoids will be most abundant within the patch of hosts, rates of parasitism will peak strongly just beyond the patch edge and then will decline again at distances greater than those to which the parasitoid typically disperses. In turn, population growth by the host insect will be positive within the patch, negative just outside it, and positive again at some distance beyond the parasitoid-imposed "halo of death."

A Case Study One striking case of spatial variation in population density is the western tussock moth (Orgyia vetusta Bdv., Lymantriidae) feeding on perennial lupines (Lupinus arboreus Sims and L. chamissonis Eschs.) in dunes and grasslands on the California coast. Adult females of this moth are flightless. The moth may achieve densities at which larvae completely defoliate lupine bushes, and may maintain such densities at a single site for more than ten years. In several extensive searches of the California coast, four dense populations of the tussock moth were found, each inhabiting tens to hundreds of bushes within lupine stands often numbering in the tens of thousands (Harrison 1997). The tussock moth thus appears to use a tiny fraction of its available resources extremely heavily, while leaving vast amounts almost unexploited. Experiments at one site, the Bodega Marine Reserve (BMR; University of California, Sonoma County, California) revealed some of the reasons for the striking temporal stability of tussock moth outbreaks. A density manipulation experiment revealed that the outbreak population was strongly foodlimited (Harrison 1994). Nonetheless, defoliation seldom killed the host

STUDYING SPATIAL PATTERN FORMATION

63

plants, and heavily attacked plants recovered their biomass and seed production fully within one growing season (Harrison and Maron 1995). Experiments also showed that outbreaking tussock moths on lupines did not suffer delayed density dependence through long-term reductions in food quality (induced defenses) or larval quality (maternal effects), in contrast to some other outbreaking insects (Harrison 1995). The spatial stability of tussock moth outbreaks appears to be related, at least in part, to the extremely limited dispersal ability of the moth. In two experimentally created outbreaks at BMR, the median displacement of moths from their release point after one complete generation was only two meters (Harrison 1994). Variable host-plant quality did not appear to play a role in restricting the outbreaks; moths reared from hatching to pupation on bushes inside and outside the outbreak areas at four sites showed equal survival, pupal weight, and development time (Harrison 1994, 1997). Abundances of generalist predators, such as ants and spiders, also did not differ between outbreak and nonoutbreak areas (Harrison and Wilcox 1995; Harrison 1997). Parasitoids of the tussock moth at BMR are all locally specialized on the moth. They include the scelionid wasp Telenomus californicus Ashmead, which parasitizes eggs, and three tachinid flies that attack larvae and emerge from pupae: Tachinomyia similis Williston; Patelloa pluriseriata or fuscimacula Aldrich and Webber, which inserts its eggs into the larva; and Protodejeania echinata Townsend, which oviposits on host plant foliage. An ichneumonid wasp has emerged from a small number (< 5%) of tussock moth pupae. In 1994 and 1995, Brodmann and colleagues (1996) placed groups of tussock moth eggs or larvae on bushes within the outbreak area, and along a transect emanating 500 m from its edge through suitable but unoccupied habitat. Parasitism by all the above species was significantly higher outside than inside the outbreak, e.g., 6% inside versus 14% outside for T. californicus, and 25-35% inside versus 40-60% outside for T. similis. Parasitism by most species followed the predicted parabolic relationship with distance from the outbreak, peaking at 100 m to 200 m from its edge, although parasitism by T. similis increased monotonically with distance from the outbreak (Brodmann et al. 1996). These results are consistent with the idea that the tussock moth and parasitoid interaction leads to the spatial restriction of the outbreak. Left unclear, however, is whether this observed distance-dependent parasitism is actually strong and consistent enough to control the spread of the tussock moth population. To answer this question, Maron and Harrison (1997) created small experimental outbreaks by placing one thousand newly hatched moth larvae on forty-four lupine bushes at distances of 0 to 700 m from the edge of the 1996 outbreak. To examine the possible role of ground-dwelling predators, these bushes were surrounded by either 30-cm fences topped with Tangle-

64

CHAPTER FIVE

foot®, or control fences with four 10-cm-wide gates. Half the experimental bushes were located in sand dunes and half in grassland habitat. Survival of moths to the pupal stage was substantially higher farther from the natural outbreak, and the numbers of parasitized larval corpses found beneath the bushes showed the opposite trend (figure 5.1). Numbers of egg masses produced by surviving experimental moths also increased strongly with distance from the edge of the outbreak (figure 5.2). The experimental populations were completely suppressed at distances up to 200 m (dunes) or 300 m (grasslands) from the natural outbreak, while at greater distances, net population growth was sometimes (grasslands) or nearly always (dunes) strongly positive. In contrast, even though predator exclusion affected the survival of first- and second-instar larvae, it did not significantly affect the numbers of egg masses per bush, showing that distance-dependent parasitism overwhelmed the initially strong effect of predation (Maron and Harrison 1997). These experiments on the tussock moth population were the first to show that parasitoids emanating from a population of host insects could suppress the growth of incipient host populations nearby, while allowing ones farther away to increase. Recent observations have also corroborated the role of parasitism in the spatial dynamics of tussock moth outbreaks. The original outbreak studied at BMR is now at its lowest size in the last six years. In 1996, a small incipient outbreak appeared approximately 1,000 m from the old one; by 1997, the new outbreak had grown tremendously, increasing more than tenfold in area. In 1997, the new outbreak had much lower rates of parasitism than the original one, although predation rates and egg mass sizes (an indicator of plant quality) did not differ (Maron and Harrison, submitted). In contrast, satellite outbreaks had generally failed to become established in previous years, when the original outbreak was much larger and presumably supported a heavy population of parasitoids. In 1998, the new outbreak collapsed because of a massive die-off of the lupine population. System-specific models are now needed to explore more fully the dynamic properties of the tussock moth and parasitoid interaction and to generate further testable predictions. One relatively simple model of this system has successfully predicted that host densities should be highest near the edge of the host patch (Hastings et al. 1997). Wilson and colleagues (1999) have shown that an individual-based model reflecting the detailed features of the tussock moth system gives qualitatively similar results to a simple discretetime analytic model of pattern formation. McCann and coworkers (in press) have shown that spatial pattern formation can arise in an analytic model with discrete space as well as discrete time, motivated by the tussock moth system. Many questions remain about this system, which can best be addressed through a combination of modeling and empirical work. One is how the

65

STUDYING SPATIAL PATTERN FORMATION

50 -,

Q.

A

GRASSLANDS

30

Q.

b 20 ^

i_

n

| 10 —I 0 0

(O

120

perl

140 100

2. 3

I 200

I 300

I 400

B

I 500

I I 600 700

DUNES

80

Q.

60

o

-2 -

1 -4

1

*

i

continuous fragmented stand type

Figure 7.1. Box plots of parasitism (log10 odds-ratio) by the braconid wasp Aleiodes malacosomatus attacking the forest tent caterpillar in continuous forest and in forest fragments at North Bay, Ontario, Canada. Parasitism is estimated for 8 continuous sites and 8 fragmented sites (50-100 larvae at each). Median values are indicated by the horizontal line; the box around each is the interquartile range; whiskers beyond the ends of the box include all observations within 1.5 times the inter-quartile range, and the asterisks indicate observed values beyond 1.5 times the interquartile range (Systat® 6.0 for Windows, Graphics, 1996).

forest tent caterpillar, is in small fragments surrounded by agricultural meadow, compared to when it is surrounded by more aspen forest (figure 7.1). Although the composition within the stands is very similar, the context of each is very different. A. malacosomatus is presumably able to move across clearings to isolated patches of aspen (and hosts) as evidenced by its presence at all sites. Once wasps colonize small patches, they attack proportionately more hosts. This pattern could result if the forest edge constrains search within the isolated patches, resulting in repeated encounter with host colonies; in continuous forest (absence of edge), the search would not be so constrained, resulting in lower attack rates. Parasitism is not only higher in isolated patches, but is also more consistently so.

LANDSCAPE ECOLOGY OF PARASITISM

89

Habitat Connectivity for Arachnidomyia Aldrichi (Sarcophagidae) Extreme fragmentation into isolated small patches may, alternatively, result in lowered rates of parasitism because parasitoids cannot disperse to these patches. This appears to be the case for the tachinidfly,Patelloa pachypyga, and the sarcophagidfly,Arachnidomyia ( = Sarcophaga) aldrichi, attacking the forest tent caterpillar (Roland and Taylor 1995, 1997). A possible consequence of reduced connectivity in a fragmented habitat (clearings acting as barriers) is the inability to respond to spatial variation in host density. In fact, A. aldrichi produces density-dependent parasitism of its hosts within large blocks of continuous forests (figures 7.2A and 7.3A), but parasitism is independent of host density in fragmented forests (figures 7.2B and 7.3B; F(interaction) = 11.58, P = 0.001; and Roland and Taylor 1997). Therefore, loss of connectivity because of forest fragmentation could obscure or preclude density-dependent parasitism by thisfly.The overall parasitism rate is slightly higher but less variable in the contiguous forest (log-odds = 0.905 or 89% parasitism, CV = 0.73) compared to that in the fragmented forest (log-odds = 0.776 or 85% parasitism, CV = 0.89). It is interesting, but perhaps premature, to speculate that the generally lower, more variable, and density-independent parasitism in stands fragmented at this scale is associated with longer outbreaks of the host seen in areas of large-scale fragmentation (Roland 1993). Landscape, in this case, may help explain why searching for hosts by parasitoids often results in density-vague host mortality (Strong 1986). Whether such patterns can be generalized to other parasitoids remains to be seen; clearly, the pattern is opposite to that seen for the braconid wasp, A. malacosomatus, described above.

Habitat Graininess for Patelloa Pachypyga and Leschenaultia Exul (Tachinidae) and Arachnidomyia Aldrichi (Sarcophagidae) Although parasitism by these flies is affected by landscape, the scale at which parasitoids perceive and respond to landscape elements likely differs among species. For example, the forest within 100 m of a point may be continuous but be situated in an area that is otherwise highly fragmented; on a local scale, the forest is not fragmented, but at a larger scale it is. Conversely, a small area of clearings would be locally fragmented, but might be embedded in a large, otherwise continuous, forest. Determining how landscape affects parasitism requires that habitat structure be measured at a variety of spatial scales. Measuring landscape at several scales serves two purposes in understanding spatial patterns of parasitism: (1) we do not know a priori at what scale parasitoids might perceive or

90

CHAPTER SEVEN

B

Figure 7.2. Sample grids with (A) contiguous forest and (B) fragmented forest, used to estimate rates of parasitism by several species of parasitoid at Ministik Hills, Alberta, Canada. Sample sites are separated by approximately 50 m. Shaded areas are aspen forest; white areas are open meadow.

respond to landscape structure, and (2) the processes that ultimately determine rates of parasitism may operate at more than one spatial scale. As an example of the latter, there may be large-scale (km2) effects of the amount of forest on host abundance, but fine-scale effects of forest edge (barriers, microclimate) on parasitoid foraging behavior. In this case, we would expect additional effects of large-scale forest fragmentation on parasitism (through host abundance) beyond those explained by fine-scale fragmentation (through

LANDSCAPE ECOLOGY OF PARASITISM

91

behavior). One of the strengths of tools like GIS are their ability to reestimate landscape variables at multiple spatial scales. Parasitism by several species of parasitoid attacking the forest tent caterpillar illustrate a few of these points (Roland and Taylor 1997). The collapse of tent caterpillar outbreaks is attributed, in part, to insect parasitoids, in particular the tachinid flies P. pachypyga and Leschenaultia exul, and the sarcophagid fly A. aldrichi. These flies use both visual and olfactory cues to locate hosts (Hodson 1939; Mondor and Roland 1998). Parasitism is generally lower in fragmented forests compared to that in continuous forest (Roland and Taylor 1997), although the mechanism by which fragmentation reduces parasitism is not entirely known. One hypothesis is that movement (searching) by parasitoids is reduced in fragmented stands (Roland and Taylor 1995, 1997), but it is unclear at what scale the fragmentation has its greatest effect. For this reason, the rates of parasitism by each were compared to the level of fragmentation measured at a variety of scales around each population sample point. Parasitism by the three fly species was estimated at each of 127 sample points across a 20 X 20 km grid in 1995 (approximately 1.8 km between sample points, figure in Roland and Taylor 1997). Parasitism by L. exul was estimated from fifty larvae collected at each site, and parasitism by P. pachypyga and A. aldrichi were estimated from fifty pupae collected two weeks later at each site. Forest structure around each sample site was estimated from a classified photomosaic of the entire area, on which each 5.3 m pixel was classified as being forested or unforested using SPANS GIS software. The proportion of forested area around each sample point was used as the estimate of continuity of forest (1 = continuous forest and 0 = no forest). Forest continuity was estimated within a 53 m square centered on the site, and was also estimated for increasingly larger squares around the same points: 106, 212, 425, 850, and 1700 m on a side. Comparing parasitism levels to forest continuity at several scales allowed for the data to indicate at which scale forest structure has its greatest effect. For some species, the local structure may have little effect on parasitism and only larger-scale structure is relevant. Conversely, some fly species may be strongly affected by the local structure and show little or no response to larger-scale structure. To determine the scale at which parasitism is affected most by forest structure, we first fit a generalized linear model (McCullagh and Nelder 1989) of parasitism for a given species as a function of host abundance in the current year, host abundance in the previous year, and location on the large grid (to control for historical patterns of the outbreak). Second, we added the landscape term (forest structure) measured at the 53 m scale to the model, and assessed its effect on parasitism by the change in deviance of the model (McCullagh and Nelder 1989), and by the magnitude of the coefficient of the relation between landscape and parasitism. The inclusion of a

92

CHAPTER SEVEN

Contiguous forest

3r

CO

CO

•a •a

o o> o

P = 0.001 r2 = 0.11

-1

0.0

0.5

1.0

J L 1.5

2.0

2.5

log host density Figure 7.3. Parasitism (log10 odds-ratio) by the sarcophagidfly,Amchnidomyia aldrichi, as a function of density of its host, the forest tent caterpillar. Data are from: (A) a 32-ha grid with contiguous aspen forest and (B) a 32-ha fragmented aspen forest, respectively, near Edmonton, Alberta, Canada. Sample points within each grid

landscape term was refitted for each of the six scales at which it was measured; therefore, six models were fitted for each parasitoid species. The landscape term that caused the greatest change in deviance (and which had a significant coefficient) for its respective model was identified as the scale at which forest structure most strongly affects that parasitoid species. This process was repeated for each of the fly species (figure 7.4). For P. pachypyga, forest structure measured within an area of 212 m of the sample point caused the greatest deviance in the model (figure 7.4A). Parasitism was higher in continuous forest and lower in fragmented forest (positive coefficient). The implication is that P. pachypyga in some way assess or respond to forest structure at this scale, either because this is the normal scale over which females move during their adult stage, or this may be the scale over which all resources it needs is distributed. For A. aldrichi,

93

LANDSCAPE ECOLOGY OF PARASITISM

B

Fragmented forest

3r

o

o

CO

S

xy

o

P = 0.37 r2 = 0.01

-1

0.0

0.5

1.0

1.5

2.0

2.5

log host density

(figure 7.3 continued) are separated by 50-m intervals. Lines are the linearfitto the logit estimates of parasitism: (A) continuous forest b = 0.62, P = 0.001, r 2 = 0.11; (B) fragmented forest b = —0.2l,P = 0.37, r2 = 0.01. Log-odds of parasitism is used in this analysis because it is the linear form of the logistic (logit) response variable. forest measured at 425 m has a greater effect on parasitism (figure 7.4B), and for L. exul, forest has its strongest effect on parasitism at 850 m (figure 7.4C). Two general patterns emerge from this analysis. First, despite the fact that the three fly species search the same landscape and encounter the same distribution of hosts, they respond to forest structure at very different spatial scales. Therefore, the landscape effects on any one species is not simply an artifact of an indirect effect of landscape on host abundance imposing a pattern on parasitism. Second, the smallest fly species, P. pachypyga (41 mg pupal mass) responds to forest at the finest scale, and the larger species, A. aldrichi (58 mg) and L. exul (68 mg), respond to forests at correspondingly larger scales of 425 m and 850 m. That large flies respond to large-scale structure argues that movement links the landscape pattern to the pattern of parasitism. Interestingly, a fourth, even smaller species, Carcelia malaco-

94

CHAPTER SEVEN

50

Patelloa pachypyga

~\5

40

/

\ deviance

30 c

CO

20

o

o-

CD'

10

100

i

i

i

i

i i ii

1000

scale at which forest is measured (m)

B

50

100

1000

scale at which forest is measured (m)

95

LANDSCAPE ECOLOGY OF PARASITISM 50

Leschenaultia exul

40

I

30

3

1

§3. o

CO

20

10

100

1000

scale at which forest is measured (m)

Figure 7.4. Deviance accounted for (left y-axis) and coefficient for the regression (right y-axis) from each of several regressions of parasitism versus landscape, where landscape was estimated separately for each of several spatial scales (x-axis). The scale for which landscape estimates provide the greatest change in deviance (peak estimates in the figure) and for which the coefficients are steepest, is considered the scale at which forest structure affects parasitism the most for that parasitoid species. somatus (34 mg), is positively affected by fragmentation, but to fragmentation at the very local (53 m) scale (Roland and Taylor 1997). The three larger fly species described here cause the highest rates of parasitism in continuous forests. Reduced parasitism in fragmented forests is a pattern consistent with longer outbreaks of the host in fragmented forests in Ontario (Roland 1993).

Inferred Scale of Movement for P. Pachypyga and A. Aldrichi One of the analytical tools used by landscape ecologists is the "correlogram" (or "semivariogram"), which assesses the distance over which estimates are similar, by virtue of their proximity to each other. Plots of semivariance increase as more variation is accounted for over increasing distance, and then flatten at distances beyond which there is no further increase in vari-

Patelloa pachypyga

CD O

.2

I CD

B

Arachnicomyia aldrichi

§ CO

CD CO

86

172

259

345

432

lag distance (m) Figure 7.5. Semivariograms indicating pattern of spatial autocorrelation of parasitism caused by: (A) Patelloa pachypyga (Tachinidae) and (B) Arachnidomyia aldrichi (Sarcophagidae) attacking the forest tent caterpillar. The data come from 109 sample points across a single 32-ha grid (figure 7.2A), with 50-m spacing between sample points. The range is indicated by the dotted line. Semivariograms are fitted using GS + geostatistical software (Gamma Design Software, Plainwell, Michigan).

LANDSCAPE ECOLOGY

OF PARASITISM

97

ance. This type of analysis includes an estimate of the spatial scale within which a variable is autocorrelated (i.e., the "range," in the vocabulary of geostatisticians) and hence the scale at which a process that causes the autocorrelation may operate. For example, the seed set of plants pollinated by a flightless insect species might be autocorrelated at only short distances, whereas plants pollinated by more mobile insects would be expected to show patterns of seed set autocorrelated at larger spatial scales. Similarly, parasitism by poorly dispersing parasitoids might be expected to be narrowly autocorrelated, compared to that for a vagile parasitoid species. Although the scale of movement of each parasitoid species is difficult to estimate directly, the pattern of parasitism on a common host could indicate their relative scale of search. The four species of fly parasitoid attacking the forest tent caterpillar produce patterns of parasitism that are autocorrelated at very different distances (Roland and Taylor 1997). Among 109 sites within a 32 ha plot (figure 7.2A), parasitism by the tachinid fly P. pachypyga is autocorrelated to a distance of only 53 m (figure 7.5A and Roland and Taylor 1997), whereas parasitism by the larger tachinid fly L. exul (Roland and Taylor 1997) and by the large sarcophagidfly,A. aldrichi, is autocorrelated at distances of more than 400 m (figure 7.5B and Roland and Taylor 1997). Parasitism by each species of parasitoid is autocorrelated over different spatial scales despite their searching the same landscape and same distribution of hosts. Interestingly, the short range of autocorrelation for P. pachypyga (53 m) is consistent with its response to forest structure at a relatively fine scale (212 m, figure 7.4A), whereas the larger range of autocorrelation for parasitism by A. aldrichi (421 m) is consistent with its response to forest structure at a larger spatial scale (425 m, figure 7.4B). Again, the strength of this comparison among parasitoid species comes from the fact that all four parasitoid species are searching across, and responding to, the same landscape and host distribution.

Discussion and Future Directions How does a landscape perspective improve our understanding of the interaction of parasitoids and their hosts? Fine-scale studies of habitat structure and invertebrate predator foraging behavior (Kareiva 1987; Huffaker 1958) suggest that the impact of landscape on large-scale dynamics could be profound, with respect both to herbivore abundance and to the stability of their numbers. Whether the direct effect of landscape is on the natural enemy or on its host may not matter as much as the potential to decouple the two. The effect of spatial scale of measurement on the detection of density-dependent parasitism has been clearly and repeatedly made (e.g., Heads and Lawton 1983; Rothman and Darling 1991), but because parasitoids are not searching within

98

CHAPTER SEVEN

a uniform arena, interpretation of spatial patterns of parasitism in largerscale studies will require that the landscape effects also be assessed—and be assessed at multiple scales. Although landscape ecology implicitly deals with spatial patterns, population dynamics are, of course, more directly concerned with temporal pattern. Often, the link between spatial and temporal patterns is not made explicitly and the link remains unclear. The real challenge for landscape-level studies of host-parasitoid interactions, I think, will be to determine whether landscape effects on natural enemies result in a meaningful impact on host dynamics (Bernstein, chapter 4). Although landscape may dramatically reshuffle or constrain parasitoid distribution and movement relative to that of the hosts, and hence alter the spatial patterns of parasitism relative to what they might be in a homogeneous environment, it remains to be seen whether dynamics are subsequently altered on a large scale. Host dispersal and other sources of host mortality could relegate the effects of landscape to a minor role. For this reason, demonstrating a landscape effect on temporal dynamics will require that the landscape structure be manifest over a sufficiently large area to prevent the swamping out of the landscape "treatments." The size of such studies should be significantly greater than the usual dispersal distance of both host and parasitoid, and, if possible, larger still. Although landscape ecology does not have to be large-scale ecology, detecting effects on population dynamics will require that such studies be done at the largest scale possible (May 1994), particularly for highly mobile hosts and parasitoids. The true test of the importance of landscape effects on parasitism will come from studies combining large-scale life-table studies of population dynamics across different landscapes, with concurrent studies of mechanisms by which landscape features modify the host-parasitoid interaction. It is important that replicate monitoring sites be located within the normal dispersal distance of the host and parasitoid to permit sample sites to be linked. The requirement for linking of sites is, of course, at odds with the call by statisticians to ensure that replicate samples be independent. For this reason, future work needs to be done on statistical techniques for nonindependent data (e.g., Lele et al. 1998). Large-scale life-table data will permit correlation between population parameters such as rate of population change, Rt (Royama 1994), and landscape features. The degree to which any one ecological process, including parasitism, can explain the patterns of population change could be used as a guide to the importance of that factor. Spatial characteristics of population change, for example the scale of autocorrelation for Rt versus the scale of autocorrelation for individual factors, including parasitism, could similarly be used as a guide for assessing the impact of factors driving dynamics. If we accept that parasitoids play a significant role in the regulation of herbivore numbers, then landscape management could be used as a pest

LANDSCAPE ECOLOGY OF PARASITISM

99

management alternative (Marino and Landis 1996; Hochberg, chapter 17). Landscape structure could facilitate, or at least conserve, the impact of natural enemies (Letourneau 1998). Indeed, the management of spatial structure of host (pest) insect populations may be very difficult to do directly, especially with the loss of many insecticides, but as history shows, we have been particularly (and alarmingly) good at altering landscapes. Acknowledgments. Data presented in this chapter come from studies of forest tent caterpillar populations funded by the Network of Centres of Excellence for Sustainable Forest Management, Natural Sciences and Engineering Research Council, and the Canadian Forest Service Green Plan. Discussions with Barry Cooke, Phil Taylor, and Brian van Hezewijk helped clarify my ideas. The Spatial Ecology Group at the University of Helsinki provided quality time and space to complete the task of writing.

Eight The Evolution of Parasitoid Egg Load MINUS VAN BAALEN

INSECT parasitoids lay their eggs in, on, or close to their insect hosts. These eggs hatch into larvae that use a single host to complete their development. If the risk of parasitoid attack is substantial, the antagonistic interaction between the parasitoids and their hosts may lead to intense coevolution of attack and counterattack strategies. The hosts try to escape detection, and those that are nevertheless found and attacked may try to neutralize the parasitoid's eggs—for example, by encapsulation. The parasitoids, in turn, will try to increase their searching efficiency and may develop countermeasures to overcome their hosts' defensive tactics (Carton and Nappi 1991; Kraaijeveld and van Alphen 1994; Kraaijeveld and Godfray 1997; Hochberg 1997; Godfray, chapter 9). From the viewpoint of the hosts, what is at stake is simple: their lives. The risk of attack determines how well a host should be prepared; this will set an optimum balance of costs (allocation of resources to defense that could have been used otherwise) and benefits (probability of surviving parasitoid attack). For the parasitoids, however, what is at stake is less clear. If eggs were cheap and fast to produce, and if the time required to attack a host were minimal, a parasitoid may lose nothing by parasitizing any host it encounters. If the hosts evolve defense strategies, however, this is not likely to be the case. Expensive eggs that require substantial investment of resources (or hosts that are dangerous to attack) could imply a direct cost of parasitizing an encountered host. Thus, it is necessary to include the potential costs of parasitoid eggs (and/or attack) to understand host-parasitoid coevolution. Traditionally, parasitoids are subdivided into "egg-limited" and "timelimited" species. Time-limited parasitoids have an ample supply of eggs or they can quickly mature new ones if their supply has become depleted, so their fitness is proportional to the number of hosts they are able to attack during their lifetimes. This type of parasitoid has been the paradigm of optimal foraging theory (Charnov and Stephens 1988; Stephens and Krebs 1986; see also Roitberg, chapter 16). In contrast, egg-limited parasitoids have a finite egg supply ("egg load"), which sets an upper limit to their lifetime reproductive success. For such parasitoids, eggs must be costly, and indeed, theory and observation suggest that such parasitoids are more choosy when

104

CHAPTER EIGHT

presented with hosts of varying qualities (Iwasa et al. 1984; Roitberg and Mangel 1989; Mangel 1992). This dichotomous view has been giving way to the more integrative view that time limitation and egg limitation are the extremes of a continuous spectrum, and that, given the stochasticity inherent in their way of life, parasitoids are to some extent subject to both (Minkenberg et al. 1992; Rosenheim 1996; Heimpel and Rosenheim 1998). Even when most individuals die without having depleted their egg loads, some may have to spend their last hours or days (or a proportion of their time, in the case of synovigenic parasitoids) in vain because they have no more eggs to deposit. The evolutionary questions are where on the spectrum we expect parasitoids to be and how this depends on the ecological characteristics of hostparasitoid interaction. On one hand, a female parasitoid's investment in her capacity to go on searching for hosts is wasted once she has laid her last egg. On the other hand, preparing more eggs than the maximum number of oviposition opportunities a parasitoid can expect is also wasteful. As Rosenheim (1996) has put it, parasitoids should strike the "optimal balance" between being egg-limited (because it wastes foraging opportunities) and being timelimited (because this wastes eggs). This gambling aspect of oviposition strategies has received recent attention by Rosenheim (1996) and Sevenster and colleagues (1998). In these studies, the stochastic component of the parasitoid's life history takes the form of stochastic survival: A parasitoid may die before she has depleted all her eggs. The unpredictability of the rate of host encounter has received less attention, but the effect is analogous. If a parasitoid is living in a completely predictable environment, she should never store more eggs than the number of hosts she is going to encounter. In unpredictable environments, however, it is unlikely that the best strategy is to prepare eggs just for the expected number of hosts encountered. Even if the probability is low that a parasitoid will be lucky and encounter many hosts, the potential payoff may still be sufficiently high to be equipped for it (Godfray 1994). Models that address optimum strategies for egg-limited parasitoids are generally based on a life-history framework (Iwasa et al. 1984; Mangel 1992; Roitberg and Mangel 1989; Minkenberg et al. 1992; Rosenheim 1996; Sevenster et al. 1998; Heimpel and Rosenheim 1998). This approach assumes that the environment is constant: Under such conditions, the parasitoid should try to maximize the number of successful ovipositions, given an unvarying mortality rate. From such studies, it appears that it does not pay to economize on eggs if mortality rates are high; such parasitoids should therefore tend to be more time-limited. Dynamic programming models that keep track of the state of the egg complement of parasitoids throughout their life show that optimum decisions depend on the age of the parasitoid. (Iwasa et al. 1984; Mangel 1992; Roitberg and Mangel 1989) Young parasitoids

EVOLUTION OF PARASITOID EGG LOAD

105

should not waste eggs (behaving like egg-limited parasitoids) whereas parasitoids near the end of their lives should be less choosy (behaving like timelimited parasitoids). Models that have addressed the ecological consequences of egg limitation usually draw the analogy with handling time in predator foraging (Thompson 1924, cited in Getz and Mills 1996; Hassell and May 1973; Hassell 1978; Hochberg 1997; Heimpel, chapter 3). If host density is low, parasitoid fitness is likely to be constrained by encounters with hosts, whereas "saturation" will occur with higher host densities; ergo, egg limitation ensues. Because there is no explicit relationship between egg load and fitness, this approach is less satisfactory for the dynamics of egg limitation per se than an approach explicitly modeling the condition of parasitoid females. Shea and colleagues (1996) analyzed a population-dynamic model that includes explicit oviposition dynamics. In contrast to what could be expected on the basis of the handling time analogy, they found no effect of egg limitation on population stability. This result is difficult to interpret, because their model was based on a continuous-time Lotka-Volterra system of differential equations, whose dynamic behavior is very different from that of the discretetime Nicholson-Bailey model that is usually taken to represent hostparasitoid dynamics (May 1974b; Hassell 1978). The evolution of egg limitation depends on the resource cost of an egg, which is an individual-level characteristic, as well as on the expected number of encounters with hosts, which is a population-level characteristic. This calls for an integrative approach that links individual-level optimization and population-level consequences. There have been some approaches to integrating ecological and evolutionary models involving parasitoids (e.g. Driessen and Hemerik 1992; Hochberg and Holt 1995; Getz and Mills 1996; Shea et al. 1996; Hochberg 1997), but these are limited for the dynamics of egg limitation by the lack of a fitness measure for egg-limited parasitoids that is consistent with the Nicholson-Bailey model. The fitness concept underlying most life-history models can be integrated into a population-dynamic setting, but typically it leads to a mathematical nightmare, characterized by a profusion of parameters, additional assumptions, and so on. Consequently, it is difficult to reassess the model results in terms of the "standard" Nicholson-Bailey model. What I will do here is derive, from first principles, the per capita fitness of parasitoids with a finite egg supply in a classical Nicholson-Bailey setting. In contrast to the standard derivation, which is based on the consideration of what can happen to individual hosts (escape from attack, being parasitized, and so on), I shall derive the model from the point of view of individual female parasitoids (e.g., whether she encounters more or fewer hosts than her egg load, or whether there is competition with conspecifics). For simplicity, the model is tailored to solitary parasitoids that lay only a single egg per host attacked.

106

CHAPTER EIGHT

As I will show, the model is essentially equivalent to the Nicholson-Bailey model. The merit of this approach is not that the Nicholson-Bailey model is realistic (it surely is not), but that it is the theoretical benchmark, and I hope that the modification that I will present will lead to hypotheses that can be tested using more detailed models or experimental observations. When the model derived from individual behavior of the parasitoids is extended with an equation that describes the dynamics of the host population, the long-term consequences and possible evolutionary feedbacks that govern the selection of egg load can be investigated. I will discuss only a very simple example to illustrate the principles involved, leaving more complex cases for future analysis.

Parasitoid Fitness in the Absence of Competition I assume that the parasitoids have an "area of discovery" of size a, and that they can detect and attack every host that is found within this area. Since this area is finite, and since hosts come in discrete units (e.g., eggs or larvae), there will be stochastic variation in the number of hosts encountered. I will begin by assuming an unspecified probability distribution ())„ for the number of hosts encountered, n. (The underline is to indicate that n is a stochastic variable.) The model assumes also that parasitoids are proovigenic; that is, they have a fixed egg load of E eggs. If, during her lifetime, a parasitoid encounters fewer hosts than her egg load E, she can parasitize them all, but if she encounters more, some opportunities will go unused. Therefore, the number of hosts attacked by a parasitoid (which may be called its "oviposition success" or, more loosely, its "gain," G) with E eggs that encounters n hosts is, in the absence of competitors, G(n) =

[n

in < E)

[E (n>E)

(8.1)

The parasitoid's expected gain, EG, is EG(n) =

2

n=0

4>n G(n)

(8.2)

This expression can be rewritten as EG(n) =

2

«=0

„«-

2

n-E+l

(f>n{n - E)

(8.3a)

107

EVOLUTION OF PARASITOID EGG LOAD

= En - 2 n ~

(8.3b)

n=0

where the first term gives the parasitoid's potential oviposition success (the expected number of hosts encountered) and the second term represents the opportunity cost of egg limitation (the opportunities that are unused for oviposition). An equivalent way of representing the parasitoid's expected gain is

EG(n) = E - 2 4>n (E - ri) n=0

(8.4)

which shows that she can never attack more than E hosts (figure 8.1). The expected gain depends not only on the mean of the distribution of host encounters, but also on the variance. This effect is demonstrated in figure 8.2, which shows the expected gain as a function of egg load, E, for different values of the clumping parameter k of the negative binomial distribution (with the same mean host density). Such a clumped distribution of

0

10

15 20

25 30

35 40

45 50

host density

Figure 8.1. Parasitoid gain (lifetime number of hosts attacked) in absence of competition with other parasitoids, as a function of host density and egg load. The hosts are randomly encountered (i.e., according to a Poisson distribution).

108

CHAPTER EIGHT

10

20 30 egg load

40

50

Figure 8.2. Expected parasitoid gain (lifetime number of hosts attacked) as a function of egg load E, when hosts are encountered according to negative binomial distributions with the same mean (aN = 10) but with different values of the clumping parameter k. The solid line shows expected gain when the variance in the number of hosts encountered is zero. encounters can have multiple causes in nature. If the parasitoids have sophisticated foraging strategies, clumped encounter rates will ensue (Chesson and Murdoch 1986). Nonetheless, here I will assume that parasitoids search randomly (which is the assumption underlying the Nicholson-Bailey model), in which case clumpedness is a direct consequence of the host's spatial distribution. The more clumped the hosts (i.e., the smaller the k) the greater the variance in the number of hosts encountered. The greater the variance, the greater the relative probability for a parasitoid to encounter more hosts than her egg load allows her to parasitize. Fitness, therefore, keeps increasing with egg load long past the mean host density, when the hosts are very clumped. In contrast, when the hosts are very homogeneously distributed, it does not pay to carry more than the mean host density. One obvious result is that it will never pay to carry more eggs than the maximum ever to be encountered. However, the Poisson and negative binomial distributions have no maximum, so for these distributions there is always an opportunity cost, because there is always a chance that a parasitoid will encounter more hosts that her egg load allows her to parasitize.

EVOLUTION OF PARASITOID EGG LOAD

109

Overcoming Host Defenses If the hosts have defenses that allow them to encapsulate parasitoid eggs, the parasitoids are forced to take countermeasures (see Godfray, chapter 9). This may simply mean larger eggs, but also may mean eggs requiring more resources to equip them with anti-encapsulation properties (Kraaijeveld and van Alphen 1995). In any case, such countermeasures are likely to be costly. Suppose that parasitoids have allocated a fixed amount of resources to reproduction (/?), but that they can subdivide this into larger or smaller units, so that size s is given by s = R/E.

(8.5)

For simplicity, I will refer to s as egg sized, though size does not necessarily refer to the egg's physical size, but could equally well apply to the amount of an "anti-encapsulation agent" (for example, polydnavirus, Fleming 1992; Godfray 1994) that is injected together with the egg. The crucial assumption is that a parasitoid's total supply of this agent is fixed, so that more eggs means less agent per egg. If the probability of successful development of a parasitoid egg is a function of its size, c(s), then we have that expected fitness (EF) is the product of survival probability and expected gain: EF = c(R/E) EG.

(8.6)

To calculate the numerical examples, I took the arbitrary function c(x) = 1 — exp( — x). Here I assume that the parasitoid does not self-superparasitize. If superparasitism occurs, however, the assumption that only one egg will survive to maturity produces the same equation. As can be seen in figure 8.3, the optimum egg load depends on the expected host density. Since this particular model assumes a random (i.e., Poisson) host distribution, the variance increases with the mean, which implies that it pays to be prepared for encounters with large numbers of hosts when the mean increases. Since such preparedness requires increased egg loads, a fraction of each of these eggs' survival chances has to be sacrificed.

Within-Host Competition If parasitoid density is sufficiently high, a focal parasitoid will encounter hosts that are already parasitized by other females, and hosts parasitized by this focal parasitoid may in turn be found by other females. We therefore have to incorporate competition for hosts into the model. Again, let n be the

110

A

CHAPTER EIGHT

survival 1.0.80.6 0.40.2-

10

30

20

40

egg load

5n

aN = 40

432-

.

| S

"'*-.

• ••

•

" • •. 20

'*

"'•t#

10-

5

10

20

egg load

30

40

Figure 8.3. (A) Offspring survival and (B) expected fitness as a function of egg load E, when egg size is inversely proportional to egg load, .s = R/E with R = 5, and offspring survival is a function of egg size, taken arbitrarily to be c(s) = l-exp( — s). Expected fitness is shown for a range of mean host densities, indicated in the plot.

EVOLUTION OF PARASITOID EGG LOAD

111

number of hosts in the patch, and let there be p competing resident parasitoids (each with an egg complement of E* eggs). Then, there are E + pE* parasitoid eggs to be distributed over the n hosts. Assuming that the parasitoids can distinguish between parasitized and unparasitized hosts, every parasitoid can oviposit her entire egg load if n = E + pE*, but if n < E + pE*, there is competition for hosts. I assume that superparasitism will then occur, and that the ensuing survival probability of an egg is inversely proportional to the total number of eggs deposited in its host. Assuming no differences between parasitoids other than their egg load, the probability that an egg of the focal parasitoid will hatch, on average, is E/(E + pE*). Putting everything together, the expected gain of the focal parasitoid becomes E

G(n,p) =

n< E + pE*

(8.7)

E + pE* E «> E + pE*

The focal parasitoid's overall expected oviposition success is therefore

EF = 2

2 4>n,pG(n,p)

(8.8)

p = 0n= 0

where §nj> gives the joint probability of finding n hosts and p parasitoid competitors in the patch. As before, this can be rewritten as cc

EG = E - X

P=o

E+pE*

2

«= o

>

, y

\

faJE-n

E + pE* j

.

(8.9)

This expression requires the evaluation of an infinite sum that cannot be solved for finite egg complements E and E*. However, as a check, it can be shown that if (1) the focal parasitoid has the same egg load as the resident parasitoids (E = £*), (2) E approaches infinity, and (3) hosts and parasitoids are distributed over the patches according to a joint Poisson distribution with means of aN hosts and aP parasitoids, then the expected gain of the focal parasitoid will be

EG = ^1

J)

(8,0)

which is the expected gain of a parasitoid in the Nicholson-Bailey model. This is no surprise, of course, because the underlying assumptions are the same as those of the Nicholson-Bailey model.

112

CHAPTER EIGHT

10

20 30 mean host density

40

50

Figure 8.4. Expected fitness of an egg-limited (E = 30) parasitoid as a function of mean host density (aN), for various values of resident (E* = 30) parasitoid density as indicated in the plot. Encounters with parasitized and unparasitized hosts are assumed to be randomly distributed. The thin lines indicate per capita fitness in the absence of egg limitation (aN (1 — e~aP)laP). I have not been able to find a closed expression for the expected fitness of egg-limited parasitoids (equation 8.9). Numerical exploration reveals that the effect of egg limitation decreases if the mean density of resident parasitoids increases (figure 8.4). This is no surprise, as competition for hosts effectively decreases the number of available hosts. Figure 8.4 does not show an optimum; this is because I did not include a cost of larger egg complements. If such a cost is included (for example, because larger egg complements are associated with smaller eggs that have smaller survival chances, as in figure 8.3A), the curves would eventually decline for high enough parasitoid densities. To arrive at this figure, I assumed that hosts and parasitoids are randomly and independently distributed. It may well be that the results would be different if the parasitoids aggregate independently of host density, because this increases the variance in the number of hosts encountered by a parasitoid. Figure 8.4 shows the expected fitness of a parasitoid with egg complement E*; it shows that the satiating effect of egg limitation decreases with mean parasitoid density. This effect is not taken into account in the "handling

EVOLUTION OF PARASITOID EGG LOAD

113

time" analogy used in previous models (Thompson 1924, cited in Getz and Mills 1996; Hassell and May 1973; Hassell 1978; Hochberg 1997; Heimpel, chapter 3), because handling time depends only on host density. To calculate the force of selection on egg number, we have to compare this with the fitness of parasitoids having different egg loads. An example is shown in figure 8.5. Here, it becomes apparent that the benefit of an additional egg depends on the combination of host and parasitoid densities (figure 8.5A). If the mean host density (aN) is larger than the egg complement, the benefit of an additional egg decreases with parasitoid density (figure 8.5B). This is because competition with other parasitoids brings the number of available hosts within the range where the focal parasitoid does not "feel" the consequences of egg limitation (and as a consequence, an additional egg is of little value). In contrast, if the mean host density is substantially less than the parasitoid's egg complement, the value of an additional egg may actually increase with parasitoid density (figure 8.5B). Under these conditions, parasitoids always encounter fewer hosts than their egg complement allows them to parasitize, but an increase in egg complement gives them an advantage over other parasitoids when superparasitization is common (Ives 1989). Since it is assumed that host survival is not affected by the number of times the hosts have been parasitized, an additional egg represents an extra ticket in the superparasitism lottery. Note that the fitness benefit, depicted in figure 8.5, does not yet take into account the costs associated with increased egg loads. To work out whether it is selectively advantageous to increase the egg complement, the decreased survival probabilities of every egg should be included. In summary, analysis of the individual fitness of a focal parasitoid reveals that optimum egg load depends on the details of the interactions between host individuals and the parasitoids, and also on details of the interactions among the parasitoid larvae themselves in the case of superparasitism. However, the optimum egg load also depends on the densities and distributions of hosts and parasitoids. The latter are not fixed constants, but instead are set by die dynamics of the resident host-parasitoid system. The resulting population-dynamic feedbacks have figured importantly in the chapters by Heimpel (chapter 3), Bernstein (chapter 4) and Godfray (chapter 9). In the context of egg limitation, population dynamics mediates a feedback in the evolution of oviposition strategies of parasitoids. The next section aims to investigate this feedback in more detail.

Population Dynamical Feedback The environment that is faced by a focal parasitoid is determined by the interaction of the other parasitoids and the hosts. Because it leads to satura-

CHAPTER EIGHT

10

20 30 mean host density

40

50

B

aN =

0

4 6 mean parasitoid density

8

10

Figure 8.5. Fitness benefit A = EG(E* + 1) - EG(E*) associated with an increased egg complement, as a function of (A) mean host density and (B) mean resident (E* = 30) parasitoid density.

EVOLUTION OF PARASITOID EGG LOAD

115

tion of the parasitoid functional response, one would expect egg limitation to be a destabilizing factor analogous to the effect of a type II functional response in Lotka-Volterra systems. This leads to the following speculative evolutionary scenario. Suppose the host population is at a stable equilibrium. This implies that, to the parasitoids, the availability of hosts is predictable, which favors increased egg limitation because parasitoids are selected to carry no more eggs than necessary. Because egg limitation increases instability, the system may start to cycle. These cycles will decrease predictability: At some times, the parasitoids will encounter large numbers of hosts, whereas at other times, hosts will be rare. If the cycles become too violent, selection pressure will favor less egg-limited parasitoids, decreasing the tendency of populations to cycle. Thus, ecology and evolution may interact to regulate egg limitation. To explore this hypothesis more quantitatively, I will present some simulations of a simple host-parasitoid model that is based on the Nicholson-Bailey framework. The model is chosen for its simplicity, rather than its realism. Because its underlying assumptions are identical to the behavioral model presented in the last section, the fitness concept I have employed applies to population-dynamic models as well. A problem with the Nicholson-Bailey model is that its equilibrium is always unstable and, since no limit cycles occur, populations are nonpersistent. To render the resident system persistent, I assume that a proportion a of the hosts is in a refuge and cannot be attacked by the parasitoids; in these refuges, hosts can survive but not reproduce (for a discussion of this model, see Hochberg and Holt 1995). For a model without egg limitation, this leads to the following pair of recurrence equations: Nl+1 = Pt+1 =

(1 - a)XN^aPt

+ aN, aP

(1 - a)cNt(l - e- ')

(8.11a) (8.11b)

where X represents the host's finite rate of increase in absence of parasitism, and c is the probability of successful development of a parasitoid egg (note that any attacked host is killed, irrespective of parasitoid survival). To introduce egg limitation in the model, we need to use expressions (8.6) and (8.9) for a parasitoid's expected fitness, where Nt+l =

X,[(l - a)Nt - PtEG] + aNt

(8.12a)

Pt+1 =

Ptc(R/E)EG

(8.12b)

Note that the host equation (8.12a) results from the assumption that every host that is not attacked reproduces. Numerical exploration of system 8.12 shows that depending on the egg

116

CHAPTER EIGHT

aN,aP 30 -i E*=2 20-

E*=6 10-

E*=4

= E* 5

= E* 6

-0.2 J t Figure 8.6. Population dynamics of hosts (thin lines) and parasitoid (thick lines) (A) and the resulting selection differential on egg load (B). Parameters: X = 2, a = 0.5, R = 5. Displayed are the results over 100 generations, after the transients have disappeared.

EVOLUTION OF PARASITOID EGG LOAD

117

load of the resident host population, the system cycles with varying mean and amplitude. As expected, the amplitude is largest with the most egglimited parasitoids (figure 8.6A). Note also that the mean density of the parasitoids decreases for larger egg loads, while that of the host increases; this is a direct consequence of the survival cost of small eggs. For sufficiently large egg loads (resulting in small eggs with low survival chances), the resident parasitoid population may not be able to persist at all. For this simulation, parameters were chosen deliberately to emphasize the mechanism of egg limitation affecting population dynamics rather than to mimic some real system. Using equation 8.9 to calculate the fitness of a focal parasitoid with an egg complement E* + 1, we can get insight into the selective pressure on egg load. Figure 8.6B indicates that for low egg complements the selection differential is always positive (increasing egg load), whereas for high egg loads it is always negative, which indicates that lower egg loads are favored. Thus, some intermediate egg complement will be optimal. It is not possible to deduce the evolutionarily stable strategy (ESS) exactly from figure 8.6B; for this, the geometric mean fitness differential should be calculated over the cycle (Holt and McPeek 1996; van Baalen and Sabelis 1999). I refrain from analyzing this model in excessive detail because of the limited value of the underlying population-dynamic model. Indeed, more than one hundred studies have been dedicated to the question of host-parasitoid persistence, and many variants of the basic Nicholson-Bailey model have been considered (see Hochberg and Holt 1999; Bernstein, chapter 4). The most promising approach is probably a model taking into account demographic stochasticity (such as that analyzed by Wilson and Hassell 1997), because this would give insight into how the population distributions (here embodied in the parameters (|)n o) depend on individual characteristics.

Discussion and Future Directions Most model studies of host-parasitoid interactions have been based on variants of the discrete-time Nicholson-Bailey model (Hassell 1978; but see Murdoch and Stewart-Oaten 1989; Shea et al. 1996; Hochberg and Holt 1999). There are good biological reasons for this. The nature of many hostparasitoid interactions implies that there will be a time delay between foraging and fitness return for the parasitoids: Hosts parasitized in one generation will give rise to parasitoids in the next generation. This time delay turns out to be a strongly destabilizing mechanism. This has been the incentive to study the possible mechanisms that render the interaction persistent (Hassell and May 1973; Hassell 1978; van Baalen and Sabelis 1993; Bernstein, chapter 4).

118

CHAPTER EIGHT

The Nicholson-Bailey model and most of its variants are based on populationlevel bookkeeping. For the parasitoids, this usually leads to an equation of the form Pt+1 = cNt»[proportion of hosts attacked]. That is, the NicholsonBailey framework does not provide a natural per-capita fitness concept for an individual parasitoid of the form Pt+l = Pr»[per capita fitness]. From the Nicholson-Bailey equations, one can derive aNI{\ — e~aP) as the per capita fitness of the parasitoids, but here I have shown that this is correct only for time-limited parasitoids. For egg-limited parasitoids, I could not derive a simple expression for per capita fitness. In this respect, it is perhaps ironic that parasitoids have been popular model organisms in more individual-based evolutionary ecology studies because of the clarity of the fitness concept. Indeed, for time-limited parasitoids, there is a consistent relationship between foraging behavior and fitness (Stephens and Krebs 1986; Charnov and Stephens 1988). Parasitoids must find their hosts in order to parasitize them, and given the vagaries of finding hosts, foraging optimization is not a trivial problem. Thus, much research has focused on aspects related to parasitoid searching (spatial distributions in population-dynamic studies, foraging strategies in evolutionary ecology, see Stephens and Krebs 1986; Charnov and Stephens 1988; van Alphen and Visser 1990; Driessen and Visser 1993; Hochberg et al. 1996). This work is based on what may be an appropriate assumption that once a parasitoid has located and attacked a host, everything is determined. In contrast, a growing number of studies into the defense mechanisms that are available to the hosts (see Godfray, chapter 9) show that there may be more going on in some host-parasitoid interactions. In particular, many hosts can encapsulate—and thereby neutralize—parasitoid eggs. There is evidence for spatial variation in the outcomes of the parasitoid attack/host defense outcome (Carton and Nappi 1991; Kraaijeveld and van Alphen 1995; Kraaijeveld and Godfray 1997). My point is that if we want to understand the evolutionary outcome of parasitoid attack and host defense, we need to have insight into the costs of an egg to a parasitoid. To this end, we need to be able to calculate the per capita fitness of parasitoids. In this chapter, I have indicated how such fitness measures can be derived from first principles, and I have tried to illustrate some of the basic aspects that are involved. As noted by Rosenheim (1996), egg limitation is an expected evolutionary outcome if there is a constraint relating egg survival and egg complement, and there is stochasticity in the number of reproductive opportunities that parasitoids have. Rosenheim (1996) and Sevenster and coworkers (1998) assume that the stochasticity arises from random mortality of the parasitoids. Here I assumed that the stochasticity arises from variability in the distribution of the hosts. In reality, of course, both mechanisms may be important. The ecological model that I analyzed is a caricature: A constant propor-

EVOLUTION OF PARASITOID EGG LOAD

119

tion of hosts remains inert in refuges, and hosts and parasitoids are distributed randomly across space with encounter probabilities given by a joint Poisson distribution. In reality, many environmental aspects may be important, and moreover, the spatial distribution of the populations, as well as the host's use of refuges, are likely to be affected by all sorts of behaviors that are subject to evolution. Active searching by the parasitoids will lead to population distributions that deviate from the Poisson because they are likely to end up in locations of high host density. For this reason, however, hosts may be selected to avoid areas with high densities of conspecifics (van Baalen and Sabelis 1993, 1999). Depending on how behaviorally flexible hosts and parasitoids are, this may lead to spatiotemporal variations in host and parasitoid densities. I have argued that egg limitation may therefore be the result of a complex interplay of all these factors. To what extent the general hypothesis proposed in this chapter (i.e., increased egg limitation increases instability, which in turn favors decreased egg limitation and more stability) holds in these more realistic settings remains open for investigation. How the evolution of egg limitation is correlated with the evolution of other parasitoid traits is as yet an open question. It is conceivable, for example, that parasitoid populations may diverge into populations with different foraging strategy/egg-load combinations, if different foraging strategies lead to different variances in the number of hosts encountered (see Godfray, chapter 9). To what extent would egg limitation affect the legitimacy of optimal foraging models based on gain rate maximization? Recently, Sevenster and colleagues (1998) have argued that because egg limitation occurs only infrequently, it is probably not important. However, this ignores the "jackpot effect": If ninety patches contain ten hosts but ten patches each contain one hundred hosts, then a parasitoid with an egg complement of ten is egglimited in only 10% of the cases, but it nevertheless forgoes almost half of its opportunities for reproduction (its expected gain is ten, whereas it could have been 0.9 • 10 + 0.1 »100 = 19). From the model that I analyzed here, it is clear that parasite fitness always increases with foraging efficiency, but at decreasing rates. It does not pay for a parasitoid to search so efficiently that she will encounter more hosts than she can deal with. But at the same time, an increase in a will affect the optimum egg load. The only way to understand how egg limitation will affect rate maximization is, therefore, to consider joint models. This will require insight into possible trade-offs between foraging efficiency and other aspects of parasitoid life-history parameters. An example of such an analysis is given in Heimpel, chapter 3. Last, I want to briefly point out some interesting points of overlap of my analysis with other aspects of parasitoid biology. From the start, I assumed that the parasitoids are solitary parasitoids, which lay only one egg in each host they encounter. Many parasitoids are "gregarious," depositing a clutch

120

CHAPTER EIGHT

of eggs in every host. I am not aware of any studies that address this situation, but this might be a case where the parasitoid "swamps" its host's defensive abilities (see Strand, chapter 10). Whether or not this is the case, however, gregariousness gives an extra dimension to the distinction between egg-limited and time-limited parasitoids (Heimpel, chapter 3). Larger eggs may also be the consequence of within-host competition among the offspring of different parasitoid females, if superparasitism occurs. Then it may well be that larger eggs give rise to offspring that have greater competitive abilities. To understand this effect, we can no longer focus on the interaction between single hosts and single parasitoids, but we have to take into account the interaction among the different parasitoid individuals within a host. This involves a rather complicated bookkeeping of all singly and multiply parasitized hosts. Though an interesting extension, although it may yield new perspectives on the conditions favoring avoidance of superparasitim (Parker and Courtney 1984; Bakker et al. 1985; van Alphen and Visser 1990; Nagelkerke et al. 1996). Acknowledgments. I thank George Heimpel, Mike Hochberg, Tony Ives, Lex Kraaijeveld, and an anonymous referee for their comments on this chapter.

Nine Host Resistance, Parasitoid Virulence, and Population Dynamics H. C. J.

GODFRAY

A MAJOR division in parasitoid life-history strategies is between koinobionts and idiobionts1: The former allow their hosts to continue to feed and grow in size after parasitism, while the latter either kill or permanently paralyze their hosts at oviposition, or attack a nongrowing stage (see Strand, chapter 10). The distinction is important, especially to the parasitoid, as the first type must be able to withstand the defensive responses of a still-active host, whereas the second feeds on a dead or inert unit of resource. Parasitoids are often said to occupy a position intermediate between parasites and predators—the first type of parasitoid life history is nearer the parasite end of the spectrum as, like a true parasite, the insect has to coexist at least temporarily with the host; the second type is closer to that of a predator, as successful host capture ends any chance of host recovery. In this chapter, I will explore the consequences for population dynamics of the coevolutionary interaction between host resistance and parasitoid virulence in koinobiont parasitoids. By "resistance," I mean the ability of the host to destroy an internal parasitoid egg or larvae after oviposition, and by "virulence," the ability of the parasitoid to overcome host defenses (note that virulence has different meanings in related fields). I will begin by sketching the physiological basis of host resistance and parasitoid virulence and then describe the growing number of studies of its genetic and evolutionary basis. I shall then review existing models that have incorporated resistance and 1 The terms "koinobionts" and "idiobionts" were introduced by Askew and Shaw (1986), although they were based on Haeselbarth's (1979) distinction between koinophytants and idiophytants (Askew and Shaw understandably preferred to avoid the botanical associations of these earlier names). The terms derive from the Greek words koinos, common or shared; idios, own or private; and bios, life—koinobionts live the common or shared life with a living host, whereas idiobionts live a private life with a dead host. Curiously, the combination of koinos + bios has given us a second English word, coenobite: a monk living in a religious community (although in ascetic Christianity, the opposite of a coenobite is an eremite—a solitary and, hence, hermit—rather than an "idiobite"), and this word has also found its way into biology. A coenobium, a religious community of coenobites each living in his or her own cell, is a term used to describe a colony of unicellular organisms.

122

CHAPTER NINE

virulence in a population-dynamic framework, and finish by speculating about how the subject might progress. In doing this, I also want to argue that resistance and virulence in host-parasitoid interactions may be a useful model system for studying the genetics and population dynamics of coevolving traits in higher eukaryotes.

Background Physiology Insects fight pathogen and parasite attack using a mixture of cellular and humoral defense strategies (Brehelin 1986; Lackie 1988; Ratcliffe 1993; Strand and Pech 1995a; Gillespie et al. 1997; Strand, chapter 10). Against parasitoids, the cellular defense system is the most important. Certain cell types circulating in the hemocoel (i.e., classes of hemocyte) recognize a parasitoid as foreign, and then adhere to it and rupture. This triggers other cells to aggregate to the parasitoid and form a multilayer capsule, a process known as encapsulation. Although there is some variation across host species, what normally happens is that the cells in the capsule break down and the whole structure melanizes and hardens. This results in the death of the parasitoid, either by asphyxiation or through the release of toxic compounds by the capsule. Unlike the vertebrate immune response, that in invertebrates has no memory: Stronger responses do not occur to a second challenge by the same class of pathogen. Also in contrast to our knowledge of the vertebrate system, we do not yet have a good understanding of the molecular basis of immunity, although this is a field that is currently advancing rapidly. There is evidence that some hosts use different types of defense against parasitoids (e.g., Henter and Via 1995), although encapsulation is undoubtedly the most common. To counteract host defenses, parasitoids have adopted both passive and active defenses (Salt 1970; Fleming 1992; Beckage et al. 1993; Rizki and Rizki 1994; Strand and Pech 1995a; Carton and Nappi 1997). The main types of passive defense are mimicry of the host and concealment from hemocytes. Many parasitoid eggs and larvae are not recognized by hemocytes, at least in their favored hosts. It appears that the outer integument of the parasitoid mimics the host basement membrane. That mimicry is skin deep is illustrated by the strong immune response of hosts to parasitoids that have been wounded, for example by fighting among first instar larvae within a host (Salt 1970). The protective covering of some parasitoid eggs can be removed experimentally; this results in their encapsulation when injected into the host (Lewis and Vinson 1968). A number of parasitoids lay their eggs precisely in certain host tissues—for example, the brain or ganglia— where they are not encountered by circulating hemocytes (Salt 1970). The eggs of some parasitoids adhere to the fat body and other tissue, and the

RESISTANCE AND VIRULENCE

123

strength of this adhesion is directly proportional to the eggs' ability to avoid encapsulation (Kraaijeveld and van Alphen 1994). Active defense takes several forms. First, at oviposition, the parasitoid may inject substances into the host that disable its immune system. Second, the wasp may inject viruses carrying genes that are expressed in the host and again destroy the immune system. The best studied of these viruses are the polydnaviruses that occur in a variety of braconid and ichneumonid wasps, especially those attacking Lepidoptera (Fleming 1992). The viral genome is stably integrated into the parasitoid genome, but in the calyx cells of the female reproductive system, the virus is replicated and assembled prior to injection. The precise mode of action of the viral products in the host is not fully understood, but infected hemocytes appear to commit apoptosis (Strand and Pech 1995b). No viral replication occurs in the host. A number of different viruses have been associated with other groups of parasitoids, although their actions are still poorly characterized. Curiously, several parasitoids inject viruslike particles—protein coats that do not contain DNA—into thenhosts, which are essential for parasitoid survival (Schmidt and SchuchmannFedersen 1989; Rizki and Rizki 1990). Third, the parasitoid larvae themselves may secrete compounds into the host body cavity that attack its immune system. Finally, the cells of the parasitoid embryonic membranes may disassociate and persist in the host hemocoel, typically swelling in size and developing a complex cytosecretory apparatus. These cells, called teratocytes, probably have several functions, an important one being counteracting host immunity (Pennacchio et al. 1994). Different parasitoids use different combinations (or none) of these active defense strategies. Often, different strategies are used sequentially; thus, substances injected by the parasitoid may confer immediate protection, while injected viral products protect the egg and young larva until it is old enough to look after itself.

When Does the Parasitoid Influence Population Dynamics? If all hosts have the same probability of defending themselves against parasitoid attack, and equally, if all parasitoids have the same probability of overcoming host defenses, then as far as a population dynamicist is concerned, a detailed knowledge of resistance and virulence is of little importance in understanding host-parasitoid interactions. All he or she requires is to know the outcome of a parasitoid attack: whether the host or parasitoid survives (or neither, as sometimes happens), and if the host survives whether its fitness is lower than hosts that have not been attacked (Carton and David 1983; Fellowes et al. 1999a). Conceivably, the probability of parasitoid survival may be dosage-dependent; that is, it may be influenced by the number of times a host is attacked, or by the clutch size of a gregarious species, but

124

CHAPTER NINE

such fixed functions can be estimated and included in population-dynamic models (Taylor 1997 has provided an appropriate framework for studying these issues; see also Heimpel, chapter 3). There is considerable evidence, however, that host resistance changes with age (Salt 1970): Younger hosts tend to be less able to resist parasitoid attack than older individuals. Such age-dependent parasitism can have interesting dynamic consequences, as has been well illustrated by the Santa Barbara group over the last few years (see Murdoch et al. 1997 and included references). The group's work has concentrated on the opposite pattern to that just described: Older, but not younger, individuals are suitable hosts for developing (female) parasitoids. Younger hosts either are used for food because they are of too poor quality for oviposition, or are used to produce male parasitoids, rather than female ones. A related phenomenon is variable clutch size in gregarious species, with small clutches laid on smaller hosts. All these patterns are predicted by behavioral ecological models (Godfray 1994) and share the feature that a parasitoid attack on young hosts contributes fewer female recruits to the next generation than does an attack on old hosts. Moreover, if young hosts are killed, they are clearly not available for more profitable (from the population point of view) parasitism later on. As the value of young hosts is reduced, the effect is initially to contribute to the stabilization of the dynamics because a large parasitoid population, which decimates the stock of young hosts, has less opportunity to propagate itself on the remaining old hosts. If the value of young hosts is further reduced, however, this density dependence becomes overcompensating and the result is cyclic population dynamics with a period of one to two generations. Similar, though possibly subtly different, effects will occur if only young hosts produce parasitoid recruits, a subject that could be studied using the same modeling approach. Things become more interesting if there is variation among individuals in their resistance or virulence. Such variation might be purely environmental, or may have a genetic component. First, consider purely environmental variation in encapsulation ability. Two factors that are known to influence the outcome of host defense are temperature and stress. The relationship between temperature and the outcome of parasitism differs among species, although the most common pattern is for host defense to be more efficient under warmer conditions (Salt 1970). Change in temperature is likely to affect other components of the host-parasitoid interaction as well, and its composite influence on population dynamics will be difficult to predict. A number of studies have found stressed hosts to have lower encapsulation abilities (Bouletreau 1986). Mounting a successful defense against parasitism requires the maintenance of immune machinery and the redirection of resources after a parasitoid attack. An unparasitized stressed host may have to allocate resources to more immediate needs than maintaining its defensive

RESISTANCE AND VIRULENCE

125

capabilities, and when parasitized, it may be in too poor a condition to respond. Stress may result from a number of causes, some biotic and some abiotic. From a population-dynamic viewpoint, stress that results from reduced food intake caused by high population densities is particularly interesting, as it is density-dependent. In such circumstances, increased parasitoid attack may be part of the proximate mechanisms through which densitydependent resource limitation operates. A related observation is that the costs associated with selection for increased resistance in Drosophila become manifest only under conditions of high resource competition (Kraaijeveld and Godfray 1997; Fellowes et al. 1998). What is the evidence for genetic variation in resistance or virulence? There are two main ways in which this question can be investigated: by classical quantitative genetic techniques, or through selection studies—either observations of natural selection in the wild or artificial selection in the laboratory (for a fuller review of this area, see Kraaijeveld et al. 1998). The only large-scale quantitative genetic study is the important work of Henter and Via (1995) and Henter (1995) on the pea aphid, Acyrthosiphon pisum, and its braconid parasitoid, Aphidius ervi. Pea aphids are parthenogenetic throughout the summer, so variation in resistance can be measured directly by assessing the ability of different clones to escape parasitism. The genetic variation measured in such a comparison includes not only the additive genetic variation on which natural selection operates, but also components of dominance and epistatic variation, so it cannot be used directly to predict the response of the population to selection. Henter and Via (1995) found substantial between-clone variability in parasitoid resistance (coefficient of clonal variation 50-77%). The way their parasitism assay was designed meant that part of the resistance might occur through differences in behavior among clones, but they concluded that the majority of the effect reflected differences in the noncellular immune mechanism of the aphid. The magnitude of the genetic variation measured, and the rates of parasitism observed in the field, led Henter and Via to predict that the frequency of resistant clones should increase over the summer. This did not occur, however, and they suggested that highly resistant strains might suffer trade-offs reducing other components of fitness. Measuring genetic variation for virulence in the parasitoid was harder, as A. ervi is sexual and a half-sib analysis has to be performed. Henter (1995) found substantial genetic variation in wasp virulence (the coefficient of additive genetic variation was 26%) against one particular aphid clone. Unfortunately, the amount of work involved precluded seeing if certain wasp genotypes were specialized on certain aphid clones. A commonly used method for assessing genetic variation is the production and study of isofemale lines: lines derived from a single female. This technique has been used extensively by French workers to show widespread

126

CHAPTER NINE

genetic variation in parasitoid resistance by Drosophila species (e.g., Carton and Nappi 1991, 1992). Virulence has been studied less frequently, but genetic variation in this trait has been demonstrated in at least one parasitoid of Drosophila (Carton and Nappi 1991). As with aphid clones, the genetic variation revealed by isofemale lines contains additive, dominance, and epistatic components, so predictions about population responses to selection have to made be with care. This is especially true here as the inbreeding required to produce isofemale lines serves to unmask the nonadditive components. One might expect that introductions of parasitoids as part of biological control programs would offer excellent opportunities to study the natural selection of resistance and virulence. Unfortunately, this has not proved to be so, probably because most biological control programs, by their very nature, are done in a hurry with little opportunity for associated research. There are cases in which resistance or virulence has been thought to change after parasitoid release, but these data are at best suggestive and, in some cases, in error (reviewed by Godfray 1994 and Holt and Hochberg 1997). Artificial selection is a powerful means of assessing genetic variation in a trait; there is a long history of its use in studying Drosophila resistance to parasitoids (Schlegel-Oprecht 1953; Hadorn and Walker 1960; Walker 1961—but see Carton and Kitano 1981 for a critique of the early work— Bouletreau 1986; Hughes and Sokolowski 1996; Kraaijeveld and Godfray 1997; Fellowes et al. 1998). For example, individuals from a large outbred population of Drosophila melanogaster established from flies collected in the wild encapsulated eggs of two parasitoids, Asobara tabida and Leptopilina boulardi, at rates of 5% and 0.5%, respectively. By breeding only from those flies that had encapsulated parasitoids, Kraaijeveld and Godfray (1997) and Fellowes and colleagues (1998) were able to select artificially for increased resistance to the two species of parasitoid. In both cases, the rate of encapsulation rose over a period of five to eight generations to a plateau of about 60%. Interestingly, selection for increased virulence seems to be much harder: Walker (1962) failed with the Drosophila parasitoid Leptopilina, and our efforts with A. tabida (Kraaijeveld and Godfray, unpublished) have so far met with no success. Of course, the lack or low level of genetic variation in resistance or virulence does not indicate that such variation has been unimportant in the past. There are clear geographical patterns in the strength of resistance and virulence in D. melanogaster and its parasitoids when sampled across Europe (Kraaijeveld and van Alphen 1994, 1995; Kraaijeveld and Godfray 1999), which presumably reflect past and possible present selection pressures. Similarly, there are potentially interesting cross-species patterns in Drosophila resistance to parasitoids (Carton and Kitano 1981). The concentration of studies involving Drosophila reflects both its ease of culture in the laboratory (similarly, its parasitoids) but also the hope that the

RESISTANCE AND VIRULENCE

127

formidable armory of Drosophila genetics might be applied to this problem. There is increasing activity in this field, although it is still in its infancy. Single genes with a major effect on resistance and virulence have been identified by groups working with isofemale lines (Carton et al. 1992; Carton and Nappi 1997; Benassi et al. 1998; Dupas et al. 1998), but such genes may be recessive deleterious mutants of genes that make little contribution to withinand between-population additive genetic variation. However, Orr and Irving (1998) studied the genetic basis of the cline in increasing host resistance in D. melanogaster from the north to the south of Europe. They concluded that the cline had a genetically simple, possibly monogenic, basis, which they localized to the centromere region of chromosome 2. In parasitoids, isofemale line studies of Drosophila parasitoids have revealed single loci that influence virulence (Dupas et al. 1998), although as with similar studies of their hosts, whether these loci are variable in the field is unknown. To conclude this section, there is ample evidence of host variability in resistance, some of which may be purely environmental, but other parts of which are certainly genetic. There is also some evidence for variation in parasitoid virulence, but the limited evidence available so far indicates less variation in virulence than in resistance. Finally, of uncertain concern, the vast majority of our information comes just from Drosophila and its parasitoids.

Population Dynamic Models I will now turn to more formal attempts to investigate the populationdynamic consequences of variation in resistance and virulence. I will begin with studies that have assumed that variation in these traits is static and unaffected by population dynamics; in other words, that the population and genetic dynamics of the system are uncoupled. I will then move on to the technically much harder problem of modeling joint population and genetic dynamics. Uncoupled Genetics Let us suppose that there is constant variation in resistance or virulence. The constancy might occur because variation in the traits is purely environmental and determined by processes that remain approximately constant over time and space. A different justification is to assume that as variation is lost through selection of favorable traits, it is generated by migration or mutation. This is the approach to quantitative genetics pioneered by Lande in the 1970s. The problem with its application to resistance and virulence in the

128

CHAPTER NINE

present context is that when variation in resistance and virulence is significant enough to affect population dynamics, selection on the two traits is necessarily strong, and hence will influence the standing genetic variation. The Nicholson-Bailey is the Ur-model of parasitoid population dynamics and has been modified in innumerable ways to explore the dynamic consequences of different aspects of host-parasitoid biology (see Bernstein, chapter 4). One of the earliest studies of this type is that by Bailey, Nicholson, and Williams (1962). The Nicholson-Bailey model itself assumes that all hosts have the same probability of attack, and that attacks lead to successful parasitoid emergence with constant probability, so that the probability of surviving parasitism is the zero term of a Poisson distribution. Bailey and colleagues (1962) first asked what the consequences were of assuming that hosts differed in their probability of being located by parasitoids, and found that this could lead to stable coexistence. The primary biological motivation behind their work seems to have been differences in host discoverability, but in their abstract (although not in the body of their paper) they also talk about some hosts having "superior intrinsic defensive properties," which suggests that they were also thinking about differences in resistance. The first explicit consideration of host resistance was by Hassell and Anderson (1984). Their approach was to simulate a population of parasitoids attacking a population of hosts. At the beginning of each simulation, all hosts were assigned an "immunity," /, from a rectangular distribution on the unit interval. When a parasitoid discovered a host, a second random number, M, was chosen from the same distribution. Successful parasitism was assumed to occur if M > /. The simulation was run for many values of host and parasitoid population densities and standard models fitted to the resulting functional responses. When these functions were substituted into the Nicholson-Bailey model, the model predicted stable population dynamics. The Nicholson-Bailey model is unstable because parasitoids tend to overexploit their hosts, which drop to low population densities. This drop is followed by a decrease in parasitoid densities to such a level that hosts suffer very little mortality and their populations begin to rocket, followed by an even bigger increase in parasitoid densities. These successive rounds of host overexploitation and host, then parasitoid, recovery end with the extinction of the parasitoid or of both host and parasitoid. The presence of a proportion of hosts immune to parasitism stops the host population from being decimated by high parasitoid densities and allows coexistence. Another way of viewing this is in terms of density dependence acting on the parasitoid population. At high parasitoid densities, the per capita rate of successful parasitism declines as more and more parasitoids concentrate on a relatively small pool of susceptible hosts. Godfray and Hassell (1991) also took Nicholson-Bailey as their basic model, but incorporated variation in resistance in a different way. In their

RESISTANCE AND VIRULENCE

129

first model, they assumed that a certain fraction of hosts were invariably able to encapsulate parasitoids, irrespective of the number of times they were attacked (all-or-none encapsulation). They noted that this assumption was exactly equivalent to assuming that a constant fraction of hosts were in a physical refuge, a situation first modeled by Hassell and May (1973). As the proportion of hosts resistant to encapsulation increases, the behavior of the system moves from instability to stability to limit cycles, and finally to host exponential growth when too few hosts are exposed to parasitism to allow parasitoids to ever rein in host populations. The transition from instability to stability requires a larger refuge (i.e., more resistance) as fecundity increases. In recent years, a number of models of the dynamic consequences of fixed refuges have been studied (e.g., Holt and Hassell 1993, Holt et al. 1999); these models can also be interpreted in an all-or-none encapsulation context. In their second model, they assumed probabilistic encapsulation2 where the probability of surviving attack by x parasitoids is a constant raised to the jrth power. The effect of this is to scale the parasitoid searching efficiency (a, which becomes a\\ — x\), which has no influence on the dynamics of the system (i.e., divergent oscillations still occur). However, introducing individual variation in encapsulation does make a difference, and is capable of stabilizing the dynamics. Exact solutions are difficult, but a second-order approximation showed that the system was stable when the variance in probability of surviving parasitoid attack (x) exceeded a certain threshold, and as long as fecundity was not too great. Limited data suggest that encapsulation in Drosophila is probabilistic rather than all-or-none. As in the Hassell and Anderson model, resistance is stabilizing when it allows a certain fraction of hosts to escape parasitism when parasitoid densities are high. Straight probabilistic parasitism is not stabilizing because all hosts suffer identical parasitoid pressure. More generally, as all these authors have recognized, variation in resistance is one of a larger class of phenomena that can stabilize the diverging oscillations of the Nicholson-Bailey equation by providing means by which host populations avoid destruction by abundant parasitoids. These mechanisms include physical refuges, temporal asynchrony of hosts and parasitoids, and certain types of parasitoid aggregation to hosts distributed in patches across the environment. In all these cases, there is heterogeneity in the risk of parasitism experienced by hosts (Chesson and Murdoch 1986; Walde and Murdoch 1988). An advantage of thinking about heterogeneity of risk is the remarkable result obtained by Pacala and colleagues (1990, 1991; Hassell et al. 1991b) that for a wide variety of models based on the 2 We actually called probabilistic encapsulation "dosage-dependent encapsulation," but in retrospect this seems misleading, as the critical distinction is between certain survival of a fraction of hosts and a fixed probability of survival by everyone. Also, when writing the paper, we had not appreciated the similarities of our work to that of Bailey and colleagues (1962).

130

CHAPTER NINE

Nicholson-Bailey framework, stability occurs whenever the coefficient of variation of risk exceeds one. Heterogeneity of risk implies that the per capita efficiency of the parasitoid declines as parasitoid densities increase, because more and more parasitoids are chasing a diminished pool of potential hosts. Adding the Genetics The final words of Hassell and Anderson's (1984) paper are: "The extension of host-parasitoid population models to include such genetical components would be an interesting direction for further work." In host-parasitoid studies, this has really occurred only within the last five years, although there is a much longer tradition of jointly modeling population and genetic dynamics in other resource-consumer systems, particularly those involving hosts and parasites. Seger (1992) provides a useful two-way categorization of resourceconsumer coevolutionary models. The major distinctions concern the assumptions about the underlying genetics and the nature of the interaction between host and parasitoid genotypes. The first distinction is between (1) models that assume that the outcome of the host-parasitoid interaction can be described by the mean of host and parasitoid traits and (2) models that specify the interaction between specific host and parasitoid genotypes. The second distinction is between models that assume (a) graded or (b) matching interactions between genotypes. A graded interaction is something like running speed in a predator-prey interaction: The resource and consumer genotypes can be ranked along a single axis of increasing efficacy. An example of a matching interaction is the classical gene-for-gene interaction that occurs in certain plants and their pathogens. Here, the survival of host or parasitoid depends on the degree of matching of the two genotypes. A more formal way of describing the two types of interactions is to call the level of resistance x and the level of virulence y, in which case in a graded interaction, host survival is a function of x — y and in a matching interaction a function of \x — y\. Consider models of category (1) where the outcome depends on mean trait values. Such models normally consist of a pair of population-dynamic equations for resource and consumer density, and a pair of population genetic equations for changes in the mean values of host and parasitoid traits (see Dieckmann et al. 1995 for a different, but related, approach). The population genetics is usually based on Lande's (1976) formulation of change in a quantitative genetic character where the trait is incremented in each generation by the product of the additive genetic variation of the trait and the selection gradient. The additive genetic variation is normally assumed to be constant,

RESISTANCE AND VIRULENCE

131

losses due to selection in effect being constantly replenished by mutation (although it is possible to add a third pair of equations for changes in genetic variance). Lande's formulation does not properly apply to frequency-dependent selection such as is occurring here, but is approximately correct when genetic variances are small (Abrams et al. 1993). These considerations suggest that this type of model is most appropriate for modeling long-term evolutionary changes, rather than rapid genetic changes in ecological time. A number of workers have examined predator-prey coevolution using this type of model (Rosenzweig 1973; Rosenzweig et al. 1987; Abrams 1992; Saloniemi 1993; Matsuda and Abrams 1994b; and see especially Abrams and Matsuda 1997) and have shown that predator-prey coevolution can lead to predator extinction, predator-prey cycles, or stable interactions. Hochberg and Holt (1995) modeled the evolution of a refuge using a category (lb) model. The population-dynamic framework assumed was Nicholson-Bailey, with (1) nonrandom parasitism described by May's (1978) negative binomial model; (2) a type II functional response with a parameter, T), representing the maximum number of hosts that can be attacked by a female parasitoid; (3) a proportional refuge—a fraction a of hosts are protected from parasitoid attack; and (4) direct host density dependence acting prior to parasitism. Including all these features together makes for a model that can only be fully investigated numerically. One of the most interesting of its features is that for weak direct density dependence, the host can be regulated either at a relatively high population density by self-limitation (such as competition for food), or at a lower density by the parasitoid (Beddington et al. 1978). Many factors act together to determine the precise dynamics of the model, but one of the most important is the size of the proportional refuge, and this can be thought of as the fraction of the population exhibiting all-or-none encapsulation. Hochberg and Holt (1995) assumed that the size of the refuge was a function of the difference between a host and a parasitoid trait (though they also constrained the parasitoid trait never to exceed the host trait). In both cases, an increase in trait value is costly: to the host in terms of a linear decline in fecundity, and to the parasitoid through a linear decline in maximum attack rate, T| (this means that costs are host-density dependent because they are only experienced when the parasitoid is host-limited). The results of simulating this model are complex and not easy to summarize. However, coevolution could change the size of the refuge and move a population from cyclic/chaotic dynamics to a stable point equilibrium (and vice versa), and move the system from one dominated by parasitoid regulation to one dominated by host density dependence. For parasitoid regulation to occur, certain ecological criteria must be met, and to this Hochberg and Holt (1995) added an evolutionary criterion concerning the maximum growth rate of the parasitoid relative to that of the host when both have

132

CHAPTER NINE

invested at equilibrium levels in the two coevolving traits. They suggested that their model supports the argument that most host-parasitoid systems are bottom-up regulated. However, the model incorporates many specific assumptions, and extrapolating to such a general statement must be done with caution. Turn now to category (2) models, in which interactions between specific host and parasitoid genotypes are modeled. There has been a long history of models of matched interactions of this type, initially stimulated by the discovery of gene-for-gene interactions in plants, and more lately by immune specificity in mammalian pathogens (e.g., Anderson and May 1982; May and Anderson 1983; Frank 1993, 1994a, 1996; Ebert and Hamilton 1996; and reviewed by Seger 1992). Graded interactions have received less attention (Seger 1992; Frank 1994b), perhaps because of the perception that they change more slowly and are more appropriately modeled by category (1) models. I know of three category (2) host-parasitoid models, one of which assumes a matching interaction (Doebeli 1997) and the other two graded interactions (Hochberg 1997; Sasaki and Godfray 1999). In addition, some other more general resource-consumer models have been applied to parasitoids. Like that of Hochberg and Holt, Doebeli's (1997) model does not explicitly consider the coevolution of resistance and virulence, but his model can be interpreted in this light. He starts with an unadorned, and hence unstable, Nicholson-Bailey equation and assumes that the searching efficiency or attack rate is a function of the genotypes of both the host and the parasitoid. As probabilistic encapsulation can be interpreted as scaling searching efficiency, his model can be thought as describing this type of probabilistic host defense (although the parallel is not exact, as the result of attack by parasitoids of more than one genotype is not specified). The underlying genetics is of a multilocus trait, where each allele takes the value 0 or 1. The trait (x in the host) is the sum across loci and thus lies in the range 0 to c if c/2 loci are involved. The parasitoid trait (v) has exactly similar genetics and the attack efficiency is assumed to be greatest when x = y, with the efficiency dropping symmetrically as the difference in trait values increases. Finally, diploid sexual genetics are assumed and the trait distribution in the next generation is calculated on the basis that all loci are unlinked. With constant searching efficiency, the Nicholson-Bailey model is unstable. Variation in searching efficiency allows stability (Bailey et al. 1962; Godfray and Hassell 1991), but where this variation is adaptive we would expect it to be destroyed by natural selection. Doebeli's model provides a mechanism by which sufficient variation can be maintained to allow persistence. In the one-dimensional trait space, the host can evolve into a relatively safe area, although in evolutionary time it is chased by the parasitoid. The host then jumps in trait space to an area where most current parasitoids

RESISTANCE AND VIRULENCE

133

have poor searching efficiency, although after a lag, the parasitoid follows the host. This evolutionary dance allows sufficient variability to be maintained to prevent the overexploitation of the host population that brings about extinction in the Nicholson-Bailey model. Only if there is insufficient trait space (too few loci or searching efficiency insufficiently affected by the difference in trait values) does the population become monomorphic, and hence doomed to extinction. The basic picture is unaltered by assuming that increasing trait values are costly to the host and parasitoid, although more complex dynamics result as the trait space is no longer symmetrical. Doebeli (1997) argues that the explicit inclusion of sexual reproduction is essential for population persistence, which bears on the argument about the evolutionary advantages of sex. Hochberg (1997) studied a model with two types of graded defense. The first represented host concealment; the second, host encapsulation ability. The parasitoid had two equivalent types of graded countermeasures—one against concealment, the other against probabilistic encapsulation. All defenses and countermeasures incur costs (to host fecundity and parasitoid juvenile survival) and the dynamics are studied within a Nicholson-Bailey framework, including host density dependence and a type II functional response. Clonal dynamics are assumed; the model simulates the interaction between two arrays of host and parasitoid genotypes, varying in the two dimensions of defense and countermeasure. Hochberg focuses on the question of whether concealment or encapsulation is more likely to evolve. This is a complex model; Hochberg attempts only a partial analysis, focusing on the question of whether concealment or encapsulation is more likely to evolve. From the point of view of the host, and other things being equal, concealment and encapsulation have the same effect: They both allow the host to survive. If handling time is negligible, the two categories of host defense are equivalent from the parasitoid's of view: They both act to reduce effective searching efficiency. But a difference appears when handling time is non-negligible: Parasitoids cannot recognize hosts with high encapsulation ability; hence, ovipositions on this type of host waste time (or, equivalently, eggs). As a result, parasitoids tend to evolve strong countermeasures against encapsulation, and it then becomes worthwhile for the host to concentrate investment in concealment. Only if countermeasures are cheap do parasitoids invest against both forms of host defense, and in these circumstances, hosts tend to evolve either high encapsulation ability or high levels of concealment, the two defenses being equivalent. Sasaki and Godfray (1999) also worked with graded defenses, although their model has more in common with Doebeli's. They considered an explicit model of host resistance and parasitoid virulence, again based on the original Nicholson-Bailey or May's (1978) negative-binomial modification. Like Doebeli, they assumed that individual hosts and parasitoids have trait

134

CHAPTER NINE

values that determine the outcome of parasitism, and that escape is probabilistic, rather than all-or-none. However, their model is asexual and assumes an array of host and parasitoid genotypes investing to different degrees in resistance and virulence, with the outcome of parasitism determined by a function of the difference in investment in the two traits: The host is most likely to survive if it has invested heavily in resistance and if it is attacked by a parasitoid that has invested to a lesser degree in virulence. Background mutation assures that no genotype becomes extinct. There are costs to both resistance and virulence: Higher investment in the former reduces host fecundity and, in the latter, immature parasitoid survival. In analyzing the model, Sasaki and Godfray assumed that the host-parasitoid interaction was stabilized by some heterogeneity in parasitoid attack (May 1978). Given this, the joint population and genetic dynamics of the model were potentially complex, although two main outcomes were found for what we believed were biologically realistic parameter values. First, hosts could evolve to invest nothing in resistance, while parasitoids invested an intermediate amount in virulence. In this case, the host in effect elects to take its chances, trading off the costs of defense against the risk of parasitoid attack. This "no-resistance ESS" was favored when (1) resistance was costly; (2) fecundity was high, so that at population-dynamic equilibrium the risks of parasitism were great; and (3) costs of virulence were low, which resulted in virulent parasitoids that were expensive to defend against. Second, persistent cycles in resistance and virulence were observed. What happens is that the host and parasitoid engage in an arms race, with spiraling levels of resistance and virulence and their concomitant costs. However, a point is reached at which resistance is so costly that the host is selected to cease investment and again gamble on not being found by the parasitoid. Once hosts cease to invest in resistance, the parasitoid comes under selection to reduce investment in virulence, and this then resets the conditions that allow the cycle to begin anew. Interestingly, in a model of graded host-parasite evolution with very different population dynamic assumptions, Frank (1994b) found regions of parameter space in which hosts did not invest in defense, as well as regions of persistent cycles that we suspect share a common biological mechanism with ours. Hochberg and van Baalen (1999) have taken a related model to Frank's with Lotka-Volterra population dynamics, and explored how investment in resistance and virulence changes over a productivity gradient. Their main conclusion is that greater investment in both traits occurs in regions of high productivity. This appears to conflict with Sasaki and Godfray's (1999) prediction that a no-resistance ESS is more likely when host fecundity is high. Hochberg and van Baalen tentatively suggest that the geographical patterns in resistance and virulence found in D. melanogaster and its parasitoids (Kraaijeveld and van Alphen 1994, 1995) may support their predictions, al-

RESISTANCE AND VIRULENCE

135

though these patterns can also be explained by geographical differences in host-parasitoid community structure (Kraaijeveld and van Alphen 1994, 1995, Kraaijeveld and Godfray 1999).

Discussion and Future Directions Some of the founding fathers of modern population dynamics worked on host-parasitoid populations, but it is only in the last five or so years that ecological and evolutionary dynamics have begun to be considered together.3 In this chapter, I have concentrated on a limited field where ecology and evolution interact the coevolution of resistance and virulence between hosts and koinobiont endoparasitoids. I have not discussed general issues of enemy-free space and niche coevolution (Jeffries and Lawton 1984; Lawton 1986; Godfray 1994; see also Hawkins, chapter 13), nor specific models of the coevolution of host and parasitoid patch selection strategies (van Baalen and Sabelis 1993) or diapause (Ringel et al. 1998). The evolution of egg load and the interplay of population dynamic and evolutionary factors is discussed by van Baalen (chapter 8). I have also done little more than mention Holt and Hochberg's (1997) and Holt and coworkers' (1999) interesting discussion of why pests evolve resistance to insecticides so much more readily than they evolve resistance to parasitoids (and other natural enemies). What is the appropriate model framework with which to study the coevolution of resistance and virulence? What follows is a personal and shamelessly parti pris view. I think models need to specify interactions between particular host and parasitoid genotype, rather than assuming that the interaction can be summarized by the means of the two traits in the host and parasitoid. The latter quantitative genetic approach is more appropriate for studying the gradual evolution of traits under weak selection, rather than the intense interactions that are more likely to characterize the evolution of resistance and virulence. I have described Hochberg and Holt's (1995) quantitative genetic model in some detail, but I remind readers that they set out to study the evolution of a host refuge and it was I rather than they who, for illustrative purposes, reinterpreted it as a resistance-virulence model. Of those models that do specify how genotypes interact, there is a major division between those that assume matching interactions and those that assume graded interactions. I believe that the available evidence points to resistance and virulence being graded traits (see review of some apparently conflicting evidence in Kraaijeveld et al. 1998), although I readily acknowl3 The early twentieth-century giants of population genetics were not altogether sympathetic to the embryonic field of host-parasitoid population dynamics. Cambridge University Press sent Sir Ronald Fisher a book outline of A. J. Nicholson's on population dynamics, which he advised the press not to publish.

136

CHAPTER NINE

edge that the experimental basis of this conclusion is slender and that it ought to be regarded as provisional. I thus think, ungraciously but predictably, that Sasaki and Godfray's (1999) approach is a better one than that of Doebeli (1997). While both models predict cycles in resistance and virulence, their biological bases are very different. In Doebeli's model, parasitoids chase hosts through trait space in evolutionary time and, because the trait space is linear, the host has to "jump over" the parasitoid to locate enemy-free space (extension to multidimensional trait spaces, as has been done in host-parasite studies, would be interesting). In our model, the cycles arise as the host goes through episodes of escalation and surrender. Our model also predicts that "no-resistance ESSs" should occur. All models to date are simplistic; hence, relating them to data must be done with caution. Yet I think the no-resistance ESS of Sasaki and Godfray (1999) may explain the common observation that some hosts offer no resistance to their specialist parasitoids. A good example of this is Drosophila subobscura, a northern European species that invariably succumbs to its main parasitoid Asobara tabida (Kraaijeveld and van der Wei 1994). Clearly, these cases may arise because no resistance mechanisms are possible for the host, but this seems unlikely, as closely related species show effective resistance. At a no-resistance ESS, parasitoids are predicted to have a fixed level of investment in resistance; this may explain why Lex Kraaijeveld and I (unpublished) have been unable to select for higher resistance in A. tabida. In other species, however, additive genetic variation in resistance appears to be common, and there is at least one case of additive genetic variation in virulence (Henter 1995). Such variation may be a reflection of the cycles in resistance and virulence predicted by Doebeli (1997) and Sasaki and Godfray (1999). In one of the best studied system {Drosophila melanogaster and its parasitoids), however, we suspect that other processes are at work that have yet to be included in formal models (Kraaijeveld et al. 1998; Kraaijeveld and Godfray 1999; Fellowes et al. 1999b), specifically interactions between multiple species of parasitoid or host, and density dependence in the costs of resistance. In the pea aphid system, a suggestion by Henter and Via (1995) that there may be a trade-off between resistance against parasitoids and resistance against fungal pathogens, and a demonstration by Losey and colleagues (1997) with the same insect of a trade-off between susceptibility to parasitoids and predators, both suggest that understanding host-parasitoid coevolution may require a consideration of other natural enemies. Much of the initial interest in the coevolution of resistance and virulence coevolution was prompted by questions of whether such effects could influence population stability. All-or-none resistance acts as a refuge and is stabilizing, but probabilistic resistance—the more realistic type—is not stabilizing unless that there is variance in resistance across populations. Doebeli (1997) showed that the genetic cycles in his model could produce sufficient

RESISTANCE AND VIRULENCE

137

heterogeneity to stabilize the Nicholson-Bailey model. The cycles in Sasaki and Godfray (1999) contributed to population stability, but could not of themselves stabilize the Nicholson-Bailey model. My guess is that the difference between these results probably reflects the different genetic structure of the models, rather than the different assumptions about matching or graded interactions (but it would be nice to know this for certain). Whereas Sasaki and Godfray worked with a clonal model to retain some analytical tractability, Doebeli, using some very elegant modeling, incorporated diploid genetics, sex, and recombination. I think Doebeli is correct that sex is essential to generate sufficient genetic variation to have a major stabilizing effect on population dynamics. These are exciting times to be working with resistance and virulence in host-parasitoid systems. The molecular basis of invertebrate immunity is likely to be established in the next decade, while molecular tools are shifting the focus in population genetics from the study of common deleterious mutations to rare favorable mutations. As Orr and Irving (1998) have recently pointed out, parasitoid resistance in insect hosts (and specifically in Drosophila) is a clear example of an adaptive trait, and an excellent model system for investigating the genetic basis of adaptation. The armory of classical Drosophila genetics supplemented with modern weapons, such as dense microsatellite maps, is already available for probing these questions. But Drosophila parasitoids are as easy to maintain in the laboratory as their hosts, and this, coupled with the increasing ease of constructing genetic maps in previously unstudied species, offers the prospect of studying the genetic basis of reciprocal adaptation and coevolution in host-parasitoid systems. Parasitoids have provided some of the best model systems for testing hypotheses in behavioral and evolutionary ecology; I predict that they will be as useful in the new ecological genetics. I finish by flagging one aspect of host-parasitoid biology that may benefit from a new, explicitly dynamic approach. Above, I briefly described the physiological basis of resistance and virulence: how circulating hemocytes recognize a parasitoid egg or larva as foreign and cause other hemocytes to aggregate and form a capsule; and how the parasitoid can disable the immune system by producing toxins, or by injecting viruses that infect the hemocytes and cause them to apoptose. In some cases, either the host or parasitoid is the invariable winner, but in other cases, the interaction can go either way—hence probabilistic encapsulation. But the processes determining the outcome are essentially dynamic. The host produces haemocytes of different classes at different rates, possibly accelerating production after parasitism. The dynamics of hemocyte recognition and aggregation determine the speed with which a capsule is formed, while the aliquot of toxin or virus injected by the ovipositing parasitoid, or the rate at which toxins are secreted by the larvae, determine the rate at which haemocytes are de-

138

CHAPTER NINE

stroyed. Modeling the dynamics of the interaction of the parasitoid with the host immune system might provide a useful framework for organizing our increasing knowledge of the physiological processes involved. Simple backof-the-envelope calculations suggest the existence of thresholds in quantities such as hemocyte and virus/toxin density on one side of which the system evolves (in the mathematical sense) to parasitoid encapsulation, and on the other side to host immune failure. Perhaps the models will just quantify the obvious, but the great success of dynamic models in helping explain the workings of the vertebrate immune system, and the diseases such as HIV that attack it, strongly suggest that this is an avenue worth exploring. Acknowledgments. Many thanks to Mark Fellowes, Mike Hassell, Lex Kraaijeveld, and Akira Sasaki for their valuable discussion and arguments.

Ten Developmental Traits and Life-History Evolution in Parasitoids MICHAEL R. STRAND

are insects that develop as parasites of other arthropods during their immature stages but are free-living as adults. Hosts are generally located by the adult female, who lays her eggs in, on, or in close proximity to the host's body. Parasitoids almost always complete their immature development by feeding on one host, and hosts almost always die as a consequence of successful parasitoid development. While most parasitoids are well described by these characteristics, anyone who has worked with these insects recognizes that parasitoids exhibit incredible levels of species richness, accompanied by an equally high level of diversity in biological habits. Clearly, among the greatest challenges facing parasitoid biologists is understanding the origins of this diversity and how different life-history strategies have evolved. Among the most important issues in this regard are to (1) determine whether particular traits correlate with one another, (2) characterize trait variation among and within species, and (3) assess how specific traits are influenced by phylogenetic history versus ecological factors. PARASITOIDS

The optimality approach (Maynard-Smith 1978) has been especially favored by parasitoid researchers as a means for studying the evolution of lifehistory traits. As with many animals (Calder 1984; Sibley and Calow 1986; Harvey et al. 1989), among- and within-species comparisons of parasitoids suggest that size is a primary target for selection, and that adult parasitoids allocate offspring to hosts in ways that maximize this parameter (Charnov and Skinner 1985; Waage and Godfray 1985; King 1989; Mackauer and Sequiera 1993). In contrast, other developmental traits have primarily been relegated to the status of constraints on life-history evolution. In this chapter, I will suggest that developmental traits regulating parasitoid growth should, in their own right, be studied more thoroughly, and that a failure to consider such traits can lead to misinterpretations about the factors that have shaped the parasitoid lifestyle (see Brodeur, chapter 11, for related discussion of hyperparasitoids). Too few studies of developmental traits have been conducted to perform phylogenetically based comparative analyses (see Felsenstein 1985; Ridley 1989; Harvey and Pagel 1991). However, there are

140

CHAPTER TEN

enough data to provide examples on how developmental studies can enhance our understanding of parasitoid life-history evolution. I will begin this chapter by reviewing the literature and highlighting those factors traditionally considered most important in shaping parasitoid life histories. I will then compare the early development of parasitoids to other insects, and provide examples for how changes in embryogenesis have potentially influenced parasitoids at the population level. In the last section, I will examine how traits regulating larval growth and development will differentially affect idiobiotic and koinobiotic parasitoids, and how this variation has potentially influenced progeny allocation strategies. My discussion focuses on parasitoids in the order Hymenoptera, although suggested trends are potentially applicable to parasitoids in other orders.

Parasitoid Life-History Variation: Traditional Considerations Most studies on parasitoid life-history evolution can be divided into two broad categories. The first category includes studies that have sought patterns in parasitoid life histories in relation to phylogeny. Modern treatments of hymenopteran phylogeny recognize that the traditional suborder Symphyta (sawflies and woodwasps, e.g., Xyeloidea, Tenthredinoidea, Megalodontoidea, Cephoidea, and Siricoidea) is paraphyletic, whereas the suborder Apocrita, which contains most species of parasitoids, is monophyletic (Rasnitsyn 1988; Dowton and Austin 1994; Hanson and Gauld 1995; Whitfield 1998). The Apocrita suborder is divided into five main lineages (Stephanoidea, Ichneumonoidea, Aculeata, Proctotrupomorpha, Evaniomorpha), and the parasitoid lifestyle appears to have evolved once in the common ancestor of the Orussoidea and Apocrita (figure 10.1). This ancestor almost certainly developed during its immature stages as an ectoparasite on concealed hosts, since many of the basal clades within the Orussoidea, Stephanoidea, Ichneumonoidea and Aculeata share this lifestyle. Thereafter, however, free-living (phytophagous-gall forming, pollen-nectar feeding, or predatory), ectoparasitic, and endoparasitic species have arisen independently many times within and/or between each lineage. For example, most aculeates are free-living, yet endoparasitism has arisen in a few groups, such as the crysidids. In contrast, the sister group to the aculeates is the Ichneumonoidea, which is composed primarily of ecto- or endoparasitoids. However, a few ichneumonoid genera contain seed-feeding or secondarily phytophagous species (Hanson and Gauld 1995). Phylogenetic analysis of comparatively well-studied families like the braconids, also suggests that phylogeny correlates with host usage patterns in some taxa, but not in others. For example, alysiine braconids appear to attack cyclorrhaphan Diptera across a variety of habitats, whereas other subfamilies, like the Rogadinae,

TRAITS AND LIFE-HISTORY

141

EVOLUTION -Xyeloidea* Tenthredinoidea* Megalodontoidea* Cephoidea* Siricoidea* Orussoidea^ • Stephanoideat

Ichneumonoidea*^ a

Chrysidoidea1'*

Vespoit = Aculeata

Apoidea+A Cynipoidea* n Proctotrupoidea* Chalddoideat:t a

= Proctotrupomorpha

Platygastroidea* Trigonalyoidea* Megalyroidea' Ceraphronoideat*

= Evaniomorpha

Evanioidea^*

Figure 10.1. Consensus phylogeny and larval feeding habits for the Hymenoptera. The figure is based on the discussions and analyses of several authors (Gauld and Bolton 1988; Rasnitsyn 1991; Dowton and Austin 1994; Whitfield 1998). Symbols are defined as follows: *, phytophagous; t> ectoparasitic; $, endoparasitic; A, pollenfeeding or predatory; n, gall-forming.

142

CHAPTER TEN

attack hosts from different taxa that co-occur in similar habitats (Gauld 1988; Gauld and Bolton 1988; Wharton 1993; Whitfield 1998). The important point for this chapter is that replicate shifts in life history have occurred within the Apocrita, and that phylogeny is an essential consideration when evaluating the directions and factors that have influenced these changes. The second approach to classifying parasitoid life histories has been in terms of how parasitoids interact with their hosts, or how factors related to host usage influence key fitness traits such as size, fecundity, or longevity. It has long been recognized that most parasitoids are either ectoparasitic, with progeny feeding on the surface of hosts, or endoparasitic (Price 1973; Force 1975). Askew and Shaw (1986) have suggested that a more useful approach to classifying parasitoid life histories is to categorize species in relation to how they interact with their hosts (Haeselbarth 1979). By this scheme, idiobionts are parasitoids that develop on hosts whose body mass does not increase in size after parasitism, whereas koinobionts develop on hosts that remain mobile and continue to grow. Idiobionts include most ectoparasitoids that paralyze their hosts, as well as endoparasitoids that attack sessile host stages such as eggs and pupae. Koinobionts include primarily endoparasitoids of larvae and adults. Askew and Shaw (1986) emphasized that the intimate relationship that exists between most koinobionts and their hosts will favor greater specializaton and narrower host ranges than seen among idiobionts. Comparative ecological studies of selected taxa support this suggestion (summarized by Askew and Shaw 1986; Sheehan and Hawkins 1991), as do phylogenetic studies that indicate that taxa composed of koinobionts tend to have narrower host ranges than assemblages composed of idiobionts (Gauld and Bolton 1988; Whitfield 1998). The large literature on the causal mechanisms by which parasitoids affect hosts or hosts affect parasitoids also indicate that levels of specialization among koinobionts far exceed those of idiobionts (Vinson and Iwantsch 1980; Thompson 1993; Lawrence and Lanzrein 1993; Strand and Pech 1995a; Strand and Obrycki 1996). Ecto-/endoparasitism and idio-/koinobiosis have been correlated with several different traits. One of these is egg size. Flanders (1942) noted that parasitoids lay either anhydropic (lecithal) eggs that possess rigid chorions and large quantities of yolk, or hydropic (alecithal) eggs that have thin chorions, are yolk-deficient, and increase in volume after oviposition. All ectoparasitoids lay anhydropic eggs, whereas some endoparasitoids lay anhydropic eggs or eggs intermediate between anhydropic and hydropic (le Ralec 1995). In contrast, hydropic eggs are laid exclusively by endoparasitoids. Many parasitoid species host-feed to obtain energy for foraging or oogenesis (Heimpel and Collier 1996). Jervis and Kidd (1986) noted that host-feeding species usually lay anhydropic eggs and mature eggs over their lifetime (i.e., are synovigenic), and Iwata (1960) found that the number of mature eggs parasitoids carry is inversely related to egg size and ovariole number. The

TRAITS AND LIFE-HISTORY

EVOLUTION

143

inference from these studies is that anhydropic eggs are energetically more expensive to produce than hydropic eggs, and that potential fecundity for species producing hydropic eggs should be, on average, higher than for species that produce anhydropic eggs (see Heimpel, chapter 3). Price extended these observations by suggesting that potential fecundity (the number of eggs a parasitoid can mature) will reflect average realized fecundities (the number of eggs a parasitoid actually lays), and that realized fecundity must balance immature mortality rates (Price 1973, 1974, 1980). Since immature parasitoids are likely to experience similar mortality schedules as those of their hosts, Price's "balanced mortality hypothesis" predicts that parasitoid fecundity should increase as opportunities to find hosts rises and as the probability of offspring surviving to adulthood declines. Since earlier host stages, like early instar larvae, are more abundant than late stages, like pupae, parasitoids attacking younger stages are predicted to have greater fecundities than parasitoids attacking older stages. By similar logic, parasitoids attacking concealed hosts are predicted to have lower fecundities that parasitoids attacking exposed hosts, because concealed hosts will usually experience lower levels of extrinsic mortality. Godfray (1994) noted that increased variation in realized fecundities and the population dynamics of the parasitoid-host system will likely effect specific outcomes. Nonetheless, empirical support for these predictions has been found in several studies, including Price's own comparative work on ichneumonids and tachinids (Price 1973, 1974, 1980). Last, general life-history models predict that development time of organisms (from egg to the adult) will increase with body size and adult longevity (Calder 1984; Sibley and Calow 1986). However, the comparative analysis of Blackburn (1991a) revealed no such relationships among parasitoids, possibly because the development of many parasitoids is physiologically tied to the development of their hosts (Beckage 1985; Strand 1986). Blackburn did find, however, that idiobionts had, on average, shorter immature development times than koinobionts. Mackauer and coworkers (Mackauer and Sequiera 1993; Mackauer 1996) came to a similar conclusion and suggested that the reason for this is best explained in terms of the quality of the hosts that idio- and koinobionts encounter at oviposition. Host quality can be affected by many factors, but size is usually considered to be especially important, because it determines the amount of resources available to a developing parasitoid larva. In turn, parasitoid size has been shown many times, within species, to be positively correlated with other fitness traits, such as fecundity and longevity (Charnov and Skinner 1985; Waage and Godfray 1985; King 1989). Thus, if parasitoid size is the primary target of selection, then development times of idiobionts will, on average, be shorter and less variable than those of koinobionts, because idiobionts attack host resources of fixed size, whereas koinobionts attack hosts whose size can later increase. Experimental

t ge

}

asitoid I its |

Size

Small

Intermediate

Large

r

r Large

t

Late Larva-Pu

i Medium

High

Mid Larva

Small

Medium

Small

Young Larva

Adult size Low

Intermediate

Egg

Searching ability Large

Large

Fecundity Small/hydropic

Ectop

t

Egg size/type Endoparasitic

t

Larval development

10.2. Generalized trends for host traits (size, survival) and parasitoid traits (size, searching, fecundity, egg type, and la in relation to host developmental stage (egg, young larva, mid-larva, and late larva-pupa). Host size usually increases w mortality rates decline. Parasitoid adult and egg size tend to increase with host stage, but fecundity tends to decline. The rasitism also tends to increase with host stage.

TRAITS AND LIFE-HISTORY EVOLUTION

145

studies of selected idiobionts and koinobionts lend qualitative support for these predictions (Salt 1940; Corbet 1968; Beckage and Templeton 1985; Sequeira and Mackauer 1992; Harvey et al. 1994). To conclude, the parasitoid literature reveals several important linkages between life-history traits and host usage patterns. These include negative correlations between parasitoid egg sizes and fecundity, positive correlations between fecundity and mortality rates, and the tendency for idiobionts to have shorter development times than koinobionts (figure 10.2). Since idiobionts parasitize hosts of a static size, selection will favor oviposition on larger, late-stage hosts, which suffer lower mortality rates and thus select for lower parasitoid fecundities and larger egg sizes. In contrast, endoparasitic koinobionts can attack younger stage hosts that, on average, experience higher mortality rates. This will select for higher parasitoid fecundities and smaller egg sizes. Clearly, there is value in thinking about parasitoid lifehistory evolution in these broad terms, as long as one does not overgeneralize. For example, the scheme illustrated in figure 10.2 does not apply to endoparasitic idiobionts like egg parasitoids, and selection pressure for any one of these traits will vary with the particular circumstances that individual species confront. In addition, failure to consider the developmental processes regulating these traits can lead to misinterpretations about why certain traits have arisen in some taxa but not others. Below, I will consider two aspects of parasitoid development largely ignored in the life-history literature that nonetheless may be key to understanding patterns in the traits exhibited by parasitoids.

Embryonic Traits Affect Life-History Evolution Despite the attention paid to the relationship between egg size and fecundity, relatively little effort has been directed toward understanding how the early development of parasitoids might affect other life-history traits. As noted by Quicke (1997), the number of studies on parasitoid egg types and embryology is small, highly descriptive in nature, and dominated by papers published before 1970. Yet, if reducing egg size is the means by which parasitoids increase fecundity, we need to think about whether reductions in egg size will affect other developmental traits. The importance of this is twofold. First, studies of holometabolous insects overall conclude that insect eggs are morphologically very similar across the major orders and that the processes regulating early development are strongly conserved (summarized by Anderson 1972; Sander 1983; Bunning 1994; Tautz and Sommer 1995). However, the descriptive literature on parasitoids appears at odds with this generalization since, as noted above, dramatic differences exist in the size, morphology, and development of parasitoid eggs (Clausen 1940; Ivanova-Kasas

146

CHAPTER TEN

1972; Strand and Grbic 1997). Familiarizing ourselves, therefore, with the processes regulating embryogenesis puts us in a stronger position to distinguish between traits that are strongly conserved among insects and phenotypically plastic traits that exhibit reaction norms. Second, by understanding how (developmentally speaking) parasitoids reduce egg size, we improve our ability to predict which taxa or parasitoid lifestyles are likely to show the greatest plasticity in egg size or fecundity. Below, I will first summarize our current understanding of how insect eggs develop. I will then present evidence that early development of parasitoid eggs varies dramatically among species, and that this variation correlates with the host environment in which embryogenesis proceeds. I will then conclude this section with one example of how a specific embryonic trait has influenced the direction of life-history evolution.

Early Development of Insects Most insects lay yolky eggs surrounded by a rigid chorion, and initiate development by undergoing syncytial cleavage, a process characterized by several rounds of nuclear division without cytokinesis (Schwalm 1988). Outward migration of some nuclei in late cleavage results in the formation of a syncytial blastoderm that then cellularizes (Turner and Mahowald 1976; Schwalm 1988). The cellular blastoderm becomes subdivided into the germ anlage (embryonic rudiment) and extraembryonic blastoderm. Insect embryos can be divided into two basic groups on the basis of how the body pattern is formed at the cellular blastoderm stage (reviewed by Sander 1983). In long-germband species, the germ anlage gives rise to the entire metameric (i.e., segmental) pattern, whereas in short-germband species, the germ anlage consists only of the future procephalon (head region). All segments posterior to the head region are formed sequentially in an anteroposterior progression during post-blastodermal development. Long-germband development occurs almost exclusively in holometabolous species, whereas short-germband development occurs more commonly among hemimetabolous species. At the completion of germband extension, all insect embryos look very similar. This stage is known as the phylotypic stage for arthropods, and is important to the current discussion because developmental changes occur more commonly during early embryonic development than during later stages of embryongenesis (Weisblat et al. 1994; Davidson et al. 1995). Germband extension is followed by germband retraction and segmentation to yield the conserved metameric pattern characteristic of all insect species. Our understanding of how these events are regulated derives primarily from the molecular analysis of the long germband dipteran, Drosophila melanogaster. In Drosophila, the patterning process is initiated by maternal factors

TRAITS AND LIFE-HISTORY EVOLUTION

147

localized during oogenesis that trigger transcription of gap and pair-rule segmentation genes whose zygotic products produce gradients of positional information (St. Johnston and Nusslein-Volhard 1992). By the time the blastoderm cellularizes, these factors have programmed the cells in different regions of the embryo to express segment polarity and homeotic genes that define segment-specific and regional identities (Ingham 1988; Stanojevic et al. 1991). The most important question from a life-history perspective is whether these events are representative of insects generally. Developmental biologists have generally assumed the answer to be yes. Some differences in patterning have been documented in primitive, short-germband species, such as grasshoppers (Patel et al. 1992; Dawes et al. 1994), but most studies with holometabolous insects agree well with the paradigms established in Drosophila (Tautz and Sommer 1995). For example, analysis of the patterning cascade in long-germband, advanced insects like the housefly (Musca domestica) and honeybee (Apis mellifera) indicate that expression of gap, pair-rule, and segment-polarity genes are highly conserved with Drosophila (Flieg 1990; Flieg et al. 1992; Sommer and Tautz 1991, 1993). Even among intermediate and short-germband insects, such as beetles (Coleoptera) and moths (Lepidoptera), homologues of the Drosophila gap and homeotic genes are expressed in patterns broadly similar to those of flies (Sommer and Tautz 1993; Kraft and Jackie 1994; Patel et al. 1994; Brown et al. 1994; Nagy and Carroll 1994; Wolff et al. 1995). In summary, the embryological features shared by most insects include a rigid chorion, a prepackaged yolk source, and a phylotypic stage that coincides with the end of germband extension. Molecular analyses of selected species in diverse orders indicate that development always begins in a syncytial environment and invariably includes a pair-rule phase, followed by expression of segment polarity and homeotic genes. Differences between more primitive insects, like grasshoppers, and other insects suggest that some of the patterning mechanisms identified in Drosophila may have arisen in association with the evolution of the Holometabola. Development of holometabolous insects as a group, however, agrees well with the paradigms established in Drosophila. Most importantly, syncytial cleavage is assumed to be essential for establishment of the insect body plan, because the transcription factors regulating axis formation are thought to require an acellular environment to function. Early Development of Parasitoids Varies with Host Environment The results discussed above argue that insect embryogenesis is phylogenetically conserved, and that parasitoid eggs should develop similarly to

148

CHAPTER TEN

other holometabolous insects. On the other hand, the eggs laid by endoparasitoids develop in an environment (the host) quite different from the terrestrial conditions in which most insect eggs develop. As noted above, the only hymenopteran whose embryogenesis had been studied at the molecular level is the aculeate, Apis mellifera. Thus, we recently compared two parasitoids with opposite life histories in the sister group to the aculeates, the Ichneumonoidea (Grbic and Strand 1998). Bracon ( = Habrobracon) hebetor is an ectoparasitic idiobiont that lays anhydropic eggs on larvae of certain pyralids (Lepidoptera), whereas Aphidius ervi is an endoparasitic koinobiont that lays hydropic eggs that develop in the hemocoel of aphids. Both parasitoids are in the family Braconidae. We found that early embryogenesis of B. hebetor proceeds in a syncytium similar to other canonical long-germband insects, like the honeybee (figure 10.3). In contrast, A. ervi eggs undergo holoblastic (complete) cleavage that results in formation of blastomeres of two different sizes. Large blastomeres give rise to a serosal membrane that envelops the smaller blastomeres, which go on to develop into a morula stage embryo (figure 10.3). The morula then ruptures from the chorion and, thereafter, undergoes morphogenesis in a manner consistent with short-germband development. At eclosion, the first stadium larva is released into the host's hemocoel. Molecular analysis of representative pairrule, segment polarity, and homeotic genes reveals that expression patterns in B. hebetor are almost identical to those seen in Drosophila and other long-germband insects. A. ervi exhibits major differences in early patterning events, but late patterning events are similar to those described for all insects. This suggests that late development of B. hebetor and A. ervi is conserved irrespective of how development begins. Although B. hebetor and A. ervi reside in the same monophyletic family of parasitoids, the differences in their early development are greater than any described previously for insects in the developmental biology literature. Clearly, the alterations in the early development of A. ervi do not support the suggestion that embryogenesis of apocritans is phylogenetically conserved. However, is there any evidence to suggest that the alterations seen in A. ervi occur among endoparasitoids in other taxa? Few studies have been conducted, but the small literature indicates the answer is emphatically yes. The only other molecular data available are for Copidosoma floridanum (Proctotrupomorpha: Chalcidoidea: Encyrtidae), a polyembryonic koinobiont that also lays hydropic eggs. Like A. ervi, C. floridanum eggs undergo holoblastic cleavage, and embryos exhibit several alterations in early gene expression (Grbic et al. 1996a, b, 1998; Strand and Grbic 1997). The descriptive literature, however, provides additional examples of parasitoids from diverse taxa that undergo holoblastic cleavage (Parker 1931; Tremblay and Calvert 1972; Ivanova-Kasas 1972; Tremblay and Caltagirone 1973; Koscielski et al. 1978; Koscielski and Koscielska 1985; Strand 1986). In each case, altered cleavage

TRAITS AND LIFE-HISTORY

EVOLUTION

149

Figure 10.3. Embryogenesis of B. hebetor and A. ervi. Confocal, fluorescent, and Nomarski images of embryonic development. (A) After oviposition, the B. hebetor egg has a clear polarity corresponding to the dorsal-ventral and anterior-posterior embryonic axes. Embryonic nuclei {arrows) divide without cytokinesis. (B) During the first few syncytial cleavages, nuclei remain in the yolk. After the tenth cleavage, nuclei migrate to the periphery of the egg, where they undergo two additional division cycles in the syncytium before finally forming a cellular blastoderm. (C) Following germband formation, the germband undergoes retraction and segmentation (anterior and posterior limits of embryo marked by arrows). (D) After oviposition, the A. ervi egg is lemon-shaped and does not exhibit any axial polarity (nucleus marked by an arrow, chorion by an arrowhead). (£) At the sixteen-cell stage, blastomeres of

150

CHAPTER TEN

correlates with oviposition of hydropic eggs and development as an endoparasitoid. In contrast, every ecto- and endoparasitoid that lays anhydropic eggs undergoes syncytial cleavage (Flanders 1942; Bronskill 1959; 1964; Abe and Koyama 1991). Comparison of these trends with the consensus phylogeny in figure 10.1 reveals three important lessons for parasitoid life-history evolution. First, production of anhydropic eggs correlates with syncytial cleavage. This form of embryogenesis also represents the ancestral state for the Hymenoptera, given that the parasitoid lifestyle arose from ectoparasitic ancestors, who almost certainly laid eggs with a prepackaged yolk source. However, production of hydropic eggs and elimination of a prepackaged yolk source has arisen independently multiple times in several major parasitoid lineages (table 10.1). The shift from anhydropy to hydropy correlates with the shift to endoparasitism rather than koinobiosis, because examples of hydropy and total cleavage are known for both endoparasitic idiobionts like egg parasitoids (Scelionidae, Mymaridae) and koinobionts like A. ervi. What remains unclear is why many endoparasitoids lay anhydropic eggs or eggs intermediate between anhydropic and hydropic. One possibility is that the transition from ecto- to endoparasitism reflects a stepwise process, whereby traits required for a terrestrial existence (desiccation-resistant chorion, prepackaged source of nutrition) are progressively altered or lost as a consequence of developing in the nutrient-rich, aquatic environment of the host. Anhydropy in endoparasitoids, therefore, may reflect how recently the shift from ecto- to endoparasitism has occurred in a given taxon. Once the loss of yolk and the evolution of hydropy have occurred, however, hydropy may act as a constraint that would prevent an endoparasitoid from reverting back to an ectoparasitic lifestyle. On the other hand, anhydropy may persist in some groups of endoparasitoids, because it positively affects fitness or is linked to other critical traits. For example, heteronomous aphelinids probably lay anhydropic eggs because, even though females develop as endoparasitoids, males often develop as ectoparasitoids (Walter 1983). A second lesson is that among-species comparisons of egg size are more likely to reflect where parasitoids develop than trade-offs, from a parasi-

(figure 10.3 continued) the A. ervi egg are autonomous, as evidenced by injection of 3 kDa methylrhodamine-conjugated dextran (phalloidin staining demarcates the cell cortex underlying the cell membranes [green], dextran-injected cell [red]). (F) Later in development, the extraembryonic membrane (arrow) surrounding the morula stage embryo ruptures from the chorion (arrowhead). (G) The embryo undergoes germband extension by posterior growth followed by condensation and segmentation (extraembryonic membrane removed to facilitate viewing). Scale bars: (A-C, G), 80 |xm; (D, F), 7 (Jim; (E), 10 (xm. (adapted from Grbic and Strand 1998).

TRAITS AND LIFE-HISTORY

EVOLUTION

151

toid's perspective, between egg size and fecundity. That is, within-species or -taxon comparisons are likely to demonstrate significant trends between egg size and fecundity, whereas among-species or -taxon comparisons are likely to yield ambiguous results if egg types and cleavage patterns are unknown. For species that lay anhydropic eggs, reductions in egg size can be achieved only by proportionately reducing embryo size to match a reduction in quantity of yolk. In contrast, far fewer resources are allocated to hydropic eggs, which has likely paved the way for the marked increases in potential fecundity documented among endoparasitoids generally, and koinobionts in particular (Blackburn 1991a, b). Thus, thinking about egg size solely as a constraint on life-history evolution ignores the selective forces that favor production of anhydropic or hydropic eggs by parasitoids in the first place. Hydropic eggs clearly overcome a constraint on fecundity, but the factors that led to the formation of hydropic eggs probably had nothing to do with fecundity. A third lesson of significance to life-history evolution is that the alterations in early development associated with hydropic eggs have also been essential preadaptations for the evolution of other traits. Among the most dramatic of these is polyembryony, which is defined as the formation of multiple embryos from a single egg (table 10.1). In insects, polyembryony is known only from endoparasitoids in four families of Hymenoptera (Braconidae, Platygasteridae, Dryinidae, and Encyrtidae) and the Strepsiptera. Detailed studies of encyrtids like Copidosoma floridanum and descriptions of polyembryony in other wasp families reveal remarkable similarities in early development of all polyembryonic species. These include oviposition of hydropic eggs, complete cleavage, and formation of a trophamnion of polar body origin (reviewed by Strand and Grbic 1997). In each polyembryonic taxon, multiple embryos arise from the simultaneous proliferation of blastomeres and partitioning of these cells by ingrowth of a trophamnion. Strand and Grbic (1997) concluded that syncytial cleavage and the constraints on volume inherent in the architecture of typical insect eggs would prevent polyembryony from ever evolving in most insect groups. However, since the transition from syncytial to complete cleavage arose in monoembryonic endoparasitoids, polyembryony has evolved at least four times. Among the genetic regulatory changes required for polyembryony is the uncoupling of pattern formation processes from early cleavage events in the egg. Analysis of C. floridanum indicates that this has occurred since blastomeres remain undifferentiated, and no patterning genes are expressed during embryo proliferation (Grbic et al. 1996a, b; 1998). Freed from the constraint of early specification of cell fate, embryo proliferation can then proceed by partitioning of blastomeres over the course of the life cycle of the host. Among the ecological transitions favoring polyembryony would be host shifts toward increased size or conditions in which risks of immature mortality are high. Not surprisingly, all

TABLE 10.1 Taxa Containing Endoparasitoids in which Egg Cleavage Patterns or Polyembryony Have Been Reported Taxon Ichneumonoidea Braconidae Microgastrinae Cheloninae Cardiochilinae

Host Usage

Egg Type/Cleavage Pattern

Polyembryony

Selected References

Larval endoparasitoids of Lepidoptera Koinobionts

intermediate/unclear

Dahlman (1990) Gauld and Bolton (1988) Quicke (1997)

Larval endoparasitoids Koinobionts

intermediate/unclear

Whitfield (1998)

Aphidiinae

Endoparasitoids of aphids Koinobionts

hydropic/total

Tremblay and Calvert (1972)

Macrocentrinae

Larval endoparasitoids Koinobionts

hydropic/total

Euphorinae Meteorinae Helconinae

yes, in selected species

Parker (1931)

Crysidoidea Dryinidae

Chalcidoidea Encyrtidae

Mymaridae Platygastroidea Scelionidae Platygasteridae

Endoparasitoids of Homoptera Koinobionts

hydropic/total

yes, in selected species

Kornhouser (1919)

Diverse biologies Polyembyronic species are Egg-larval endoparasitoids of Lepidoptera Koinobionts

anhydropic or hydropic/total, intermediate, and syncytial

yes, in selected genera

Ivanova-Kasas (1972) Strand and Grbic (1997)

Egg parasitoids Idiobionts

hydropic/total total

no

Sahad (1984)

Egg parasitoids Idiobionts

hydropic/total total

Egg-larval parasitoids Koinobionts

hydropic/total

Strand (1986) yes, in selected species

Leiby and Hill (1924)

154

CHAPTER TEN

polyembryonic wasps are egg-larval parasitoids or larval parasitoids that oviposit in young hosts. However, rather than increasing maternal fecundity (Broodryk 1969; Price 1974) (figure 10.2), potential fecundities of adult polyembryonic parasitoids are relatively small (about 30 to 150 eggs) (Leiby 1922; Parker 1931; Ode and Strand 1995; Heimpel, pers. comm.).

The Influence of Larval Development on Parasitoid Life-History Another area where developmental considerations can contribute to our understanding of parasitoid life-history evolution is in relation to host usage and progeny allocation strategies. As noted above, hosts represent a finite resource for offspring, and as such maternal fitness is determined largely by the rate at which a female finds hosts, the quality of hosts she parasitizes, and the number (and sex) of the eggs she lays (Charnov and Skinner 1985; Iwasa et al. 1984; Parker and Begon 1986; Godfray 1987a; Heimpel, chapter 3). Parasitoids are either gregarious, with more than one offspring developing per host, or solitary, with only one offspring completing development per host. Solitary species sometimes lay more than one egg per host (Rosenheim and Hongkham 1996), but supernumerary progeny are eliminated by physical combat or physiological suppression (Vinson and Iwantsch 1980; Godfray 1987b). Among-species comparisons indicate that solitary development is much more common than gregariousness, and that gregarious parasitoids almost certainly evolved from solitary ancestors (Rosenheim 1993; Mayhew 1998). Godfray (1987b) suggested that larval fighting has been a key factor in clutch size evolution, because the optimal clutch size for offspring could be smaller than it is for parents. Larval fighting, in effect, reinforces the solitary lifestyle and prevents the evolution of gregariousness unless the per-capita fitness of individuals developing together exceeds the fitness of individuals developing alone. Nonetheless, the stringent conditions proposed by Godfray for the evolution of gregariousness have apparently been relaxed many times, given that gregariousness has arisen independently on multiple occasions (Rosenheim 1993; Mayhew 1998). As to why gregariousness has evolved in some parasitoid taxa but not in others, most explanations have focused on changes in either adult or larval behavior. For example, oviposition of multiple-egg clutches by some solitary parasitoids may arise because they increase the probability that one offspring will survive in the host. Once multiple-egg clutches are produced, they would also serve as an important preadaptation favorable to the evolution of gregariousness (Rosenheim 1993). Oviposition of female-biased or singlesex clutches by haplodiploid (i.e., arrhenotokous) species would also reduce genetic asymmetries among offspring and thereby favor altruistic, nonfight-

TRAITS AND LIFE-HISTORY EVOLUTION

155

ing behavior among siblings (Ode and Rosenheim 1998). In the absence of larval fighting, therefore, clutch sizes are predicted to vary continuously such that both small and large gregarious broods could evolve. In the presence of fighting, however, clutch sizes are predicted to be discontinuous such that species will either be solitary or produce large, gregarious broods. The empirical literature offers some support for both predictions (le Masurier 1987; Mayhew and Hardy 1998). Here, I suggest that the traits regulating larval development have also played a key role in shaping host usage and progeny allocation strategies. Below, I will briefly review the literature and suggest that the factors regulating metamorphosis are fundamentally conserved among insects. I will then argue that because of this conservation, the host environments confronted by idio- and koinobionts present very different risks for offspring survival. Idiobionts are more likely to minimize these risks via plasticity in maternal oviposition behavior. In contrast, koinobionts are more likely to evolve host usage strategies that increase the predictability of how much in the way of host resources will be available to offspring. Some strategies adopted by koinobionts severely constrain any opportunity for gregariousness to evolve, whereas others are potentially more favorable to host switching and the evolution of gregariousness.

Traits Regulating Larval Development Are Conserved Molting and metamorphosis of all insects are regulated by conserved endocrine molecules. Depending on environmental conditions, most holometabolous insects show some plasticity in the number of larval molts they undergo or in the duration of individual instars (Riddiford 1985; Nijhout 1994). However, the timing of metamorphosis of all insects studied to date is linked to an absolute, species-specific threshold size (Nijhout 1994). If an individual is below this threshold, it molts to another larval instar or remains in that instar until attaining the threshold size. In contrast, if the individual molts to an instar that is above the threshold size, changes in endocrine state occur that result in the next molt being to a pupa. Once above the threshold size, the larva continues to feed because of the amount of time required for the appropriate endocrine signals to induce the metamorphic molt. During this period, insect larvae commonly increase their weight twofold or more. Moreover, since insect larval growth is exponential, no matter how many instars an insect goes through, most of the increase in an insect's mass occurs during the last larval instar. Thus, the threshold size represents the minimum larval mass that will yield an adult, whereas the maximum possible mass depends on how far above the threshold mass the insect was when it molted to the final instar. The precise timing of when larvae cease to feed and initiate

156

CHAPTER TEN

metamorphosis depends on environmental factors and sensory inputs that "inform" the insect of its size, as well as of the status of its food source. For example, fruit- and flesh-feeding Diptera that achieve their size threshold cease feeding a few hours later and then wander from the feeding site to pupate. This behavior is essential because larvae in contact with water do not pupate (Denlinger and Zdarek 1994). This occurs because sensory receptors that perceive external moisture inhibit metamorphosis by preventing the release of prothoracicotropic hormone (PTTH) or ecdysteroids (Ohtaki 1966; Zdarek and Denlinger 1991). Thus, neural inputs that monitor size provide positive stimuli for the initiation of metamorphosis, while inputs that monitor the feeding environment provide negative stimuli that suppress metamorphosis. Parasitoids differ from other insects only in the sense that they must contend with the specialized environments of their hosts. Host resources are essentially of fixed value for idiobionts. In contrast, the hosts of koinobionts change after oviposition in a bewilderingly large variety of ways. Hosts attacked by so-called conformer species develop very similarly to nonparasitized hosts, and the parasitoid larva synchronizes its own immature development with that of the host (Lawrence 1986). At the other extreme, regulator species induce substantial changes in host growth by injecting factors like polydnaviruses and venoms (summarized by Vinson and Iwantsch 1980; Beckage 1985; Lawrence and Lanzrein 1993). While these factors clearly increase the range of possible interactions that can occur between parasitoids and hosts, none change the fundamental factors that regulate metamorphosis—namely, the host is a finite resource and parasitoids develop into adults by consuming sufficient resources to attain a critical larval size. All parasitoids must consume some minimum amount of host resources in order to pupate. Above this minimum, all parasitoids will also exhibit some plasticity in final size up to a species-specific maximum. All parasitoids will also exhibit a species-specific size maximum beyond which additional host resources cannot be consumed.

Conserved Size Thresholds Pose Different Risks to Ecto- and Endoparasitoids Current evidence suggests that, within species, insect size thresholds are conserved and change very slowly over evolutionary time (Nijhout 1994). Mayhew and Hardy (1998) report a similar trend among species in the family Bethylidae, which suggests that parasitoids may also have a limited capacity to change average body size. If so, ecto- and endoparasitoid offspring face the same mortality risks from developing on hosts too small to attain a critical size, but face very different risks if host resources exceed an off-

TRAITS AND LIFE-HISTORY EVOLUTION

157

spring's capacity to consume them. An ectoparasitoid larva can simply feed up to its size maximum, wander, and pupate. In contrast, excess resources will often be fatal to endoparasitoids because parasitoid larvae, like the dipterans mentioned above, are seemingly incapable of pupating in the presence of moisture (i.e., unconsumed host resources) (Flanders 1935; Jackson 1958; Strand and Vinson 1985; Beckage and Templeton 1985; Ode and Strand 1995; Harvey 1996). As a result, offspring fitness (as measured by size) for ectoparasitoids tends to increase monotonically with host size to a maximum, and then asymptotes. In contrast, endoparasitoids often exhibit domed fitness functions. Offspring fitness initially increases with host size, but as resources increase beyond the capacity of the parasitoid larva to consume them, fitness rapidly declines (see Charnov and Skinner 1985, and Waage and Godfray 1985 for examples). All of the above information underscores the fact that imprecision in egg allocation to hosts may have fitness costs (Godfray and Ives 1988). The questions are how do parasitoids achieve accuracy, and are certain host usage strategies more compatible with host switching or the evolution of gregariousness than others? Consider first idiobionts that attack nongrowing host stages. The most direct way of accurately allocating progeny to available resources is for the ovipositing female to assess host size directly. A female can then either attack hosts of variable size and adjust the number of eggs laid accordingly, or it can preferentially attack hosts of a narrow size range that corresponds to the maximum size of offspring (see Casas, chapter 2). Both options are clearly available to idiobionts, and many empirical studies conclusively demonstrate that idiobionts accurately assess host size before ovipositing (Wylie 1967; Strand and Vinson 1983; Schmidt and Smith 1985; Takagi 1986; Hardy et al. 1992). Provided that no trade-offs exist in attacking larger host species, the offspring of ectoparasitoids should face the fewest risks in switching from smaller to larger hosts, because host consumption is not a prerequisite for pupation. The morphology and lack of mobility of ectoparasitoid larvae also limits the possibility for larval fighting, which, as discussed above, should also favor the evolution of gregariousness. For endoparasitic idiobionts, like egg parasitoids, excess resources are potentially fatal and offspring often possess fighting mandibles. However, the ability to accurately allocate offspring provides a mechanism to females for adjusting the number and average relatedness of eggs laid into hosts. This in turn provides conditions favorable to the evolution of gregariousness (Gauld and Bolton 1988; Whitfield 1998). Consider now the situation faced by koinobionts. Precise oviposition is more difficult for these parasitoids because hosts continue to grow after oviposition. Larval endoparasitoids also face significant mortality risks from attacking aggressive hosts, such that the costs of increased examination time might outweigh any benefits from more accurately assessing host size. Given

158

CHAPTER TEN

these difficulties, I suggest that koinobionts largely do not rely on plasticity in maternal behavior to achieve accuracy in progeny allocation. Instead, I would argue that koinobionts have evolved strategies that either make host resources more predictable or that reduce the costs of imprecisely allocating offspring to hosts. Larval endoparasitoids in the families Ichneumonidae and Braconidae have been studied more often than all other koinobionts, and provide good examples for how host usage patterns can differ along these lines. The strategy of increasing resource predictability is typified by ichneumonids in the subfamily Campopleginae. These wasps parasitize larval Lepidoptera, produce fighting larvae, and complete their immature development by consuming the host and pupating within the remaining host cuticle. Most campoplegines are able to parasitize hosts that vary greatly in initial size because (1) females inject factors that reduce host growth rates and final size, and (2) immature (larval) development is often synchronized with the time at which the host molts to its final instar. As a result, development times for campoplegine larvae often vary with which host instar is parasitized, but final offspring size is relatively constant (Corbet 1968; Smilowitz and Iwantsch 1973; Beckage and Templeton 1985; Dover et al. 1987; Harvey 1996; Harvey et al. 1998). In essence, campoplegines like Venturia canescens maximize offspring size at the potential expense of increased development time (figure 10.4). Because larvae pupate within the remnant cuticle of their host, however, offspring remain highly constrained at the upper end of the host size continuum. In the case of V. canescens, offspring develop successfully in several species of pyralid larvae that attain final weights of less than 40 mg (see figure 10.4), but do not survive in larger host species because the parasitoid larva is unable to consume all host tissues (Harvey et al. 1994; Harvey and Vet 1997). The strategy of reducing the risk associated with imprecise allocation offspring to hosts is exhibited by many endoparasitic braconids. Most, if not all, braconids of the microgastroid complex are hemolymph feeders that emerge from their hosts to pupate (Gauld and Bolton 1988; Wharton 1993; Wharton, pers. comm.; Shaw, pers. comm.). Parasitoids in this group also inject factors into hosts that reduce growth, and larvae are capable of fighting (Beckage 1985; Lawrence and Lanzrein 1993; Strand and Pech 1995a). Unlike campoplegines, however, development time and offspring size of microgastrines are often very uniform (Beckage and Riddiford 1978; Strand et al. 1988; Harvey et al. 1999). In the case of Microplitis demolitor, development time and offspring size vary minimally across a range of host sizes (figure 10.4). Emerging from a host and pupating after a set developmental period potentially reduces the maximum size offspring could attain by consuming additional host resources. On the other hand, M. demolitor is able to develop

TRAITS AND LIFE-HISTORY

159

EVOLUTION

A

C Venturia canescens

40 -

Microplitis demolitor

15-

11 _

5

10

20

40

5

10

30

60

2nd

3K1

4th

5th

1st

2nd

3id

4th

D 3.5

2.5

2.5

5 2nd

10

20

40

5

10

30

60

3K1

4th

5th

1st

2nd

3id

4th

Anagasta kuehniella

Pseudoplusia includens

Host size (mg) Host instar

Figure 10.4. Relationship between host size and the development time and size of two solitary endoparasitoids. The tissue feeder Venturia canescens (Ichneumonidae: Campopleginae) parasitizes the larval stage of Anagasta kuehniella (Lepidoptera: Pyralidae) (A, B), whereas the hemolymph feeder Microplitis demolitor (Braconidae: Microgastrinae) parasitizes the larval stage of Pseudoplusia includens (Lepidoptera: Noctuidae) (C, D). (A, Q Development time of V. canescens decreases significantly with host size (mg), but development time of M. demolitor is unaffected by host size. (B, D) The size of V. canescens adults (as measured by hind tibia length) remains almost constant as host size at oviposition increases, but M. demolitor size (as measured by wet mass) increases with host size at oviposition. Figures are adapted from Harvey and Vet (1997) and Harvey and colleagues (2000).

160

CHAPTER TEN

successfully in hosts much larger than itself, because it does not have to consume all host resources to pupate. Campoplegine ichneumonids and microgastrine braconids have very similar host ranges (mainly larval Lepidoptera), are broadly similar in size, and both produce first instar larvae with fighting mandibles. Why, then, have the microgastrines overcome the constraint of siblicidal behavior and, on several occasions, become gregarious (le Masurier 1987; Strand et al. 1988; Harvey et al. 1999), while campoplegines have simultaneously remained exclusively solitary? I would suggest that the answer is best explained by the fundamental differences in how their offspring develop. Given that the solitary habit is ancestral in parasitoids, the evolution of fighting mandibles and traits that suppress host growth or that synchronize parasitoid development likely arose early in the evolution of koinobiosis itself. The "campoplegine" pattern of synchronizing growth and altering host size so that it corresponds to maximum offspring size clearly is essential for survival under conditions where offspring pupate inside their host. However, this strategy also severely limits the amount of host resource available for consumption and, in conjunction with larval fighting behavior, reinforces the solitary habit. In contrast, hemolymph feeding and external pupation greatly reduce the risks offspring face when developing in hosts larger than themselves. Unlike internal pupation, this developmental strategy also would be more compatible with host range shifts toward larger species, because a single parasitoid larva can still survive without consuming the host. With additional host resources available, the potential fitness gains from gregariousness could then be favored over solitary development and overcome the constraint imposed by larval fighting. As alluded to earlier, solitary microgastrines often lay multiple egg clutches and parasitize hosts of sufficient size to easily support development of more than one offspring (see Rosenheim 1993). Larval fighting usually prevents this, but in the case of M. demolitor, approximately 3% of parasitized hosts yield two adult offspring (Strand, unpublished). Thus, the evolution of gregariousness in well-known microgastrine genera, like Cotesia, in my opinion most likely arose in conjunction with host range expansion, because of the preexisting traits of hemolymph feeding and external pupation.

Discussion and Future Directions Optimality approaches have been enormously helpful in generating hypotheses and testing different ideas regarding the evolution of parasitoid traits. However, the consequences of mutation for phenotypic change are conditioned by molecular, cellular, and physiological characteristics that promote the evolvability of some traits and constrain the evolution of others. As such,

TRAITS AND LIFE-HISTORY EVOLUTION

161

I argue here that an understanding of developmental processes is essential when considering the forces that have shaped parasitoid life-history evolution. By ignoring the cellular and molecular events that regulate parasitoid development, we compromise our ability to recognize patterns in trait evolution among and within taxa, and at the population level we potentially make erroneous assumptions about the factors that control traits of long-standing interest, such as size and fecundity. Three main themes were emphasized in this chapter. First, the transition from ecto- to endoparasitism correlates with pronounced alterations in early development of parasitoids relative to almost all other insects. The shift from ecto- to endoparasitism is also likely responsible for eliminating a key constraint on the size of insect eggs—namely, the need for a prepackaged yolk source. Loss of yolk, in turn, has had important population-level effects on fecundity and host usage patterns by koinobionts. A second theme of this chapter is that parasitoid larval development is governed by the same factors that regulate molting and size thresholds of all insects. As such, the risks of mismatching host resources to the capacity of parasitoid larvae to consume them will be especially high for koinobionts. However, developmental traits such as external pupation relax host size constraints and have potentially facilitated the transition from solitary to gregarious development by koinobionts. Last, I have argued that greater attention to the comparative development of parasitoids is essential if we are to understand how the transition from ectoparasitism and idiobiosis to endoparasitism and koinobiosis occurs. Important areas for future study include the need to distinguish between the importance of ecological variables and phylogenetic history in shaping developmental traits (see also Brodeur, chapter 11). In this regard, quantitative analysis of developmental traits in relation to established phylogenies should be a primary goal. This will require more rigorous comparison of embryonic and larval development of species from appropriate taxa to determine how evolutionarily labile specific traits might be. For example, have species that lay hydropic eggs lost the capacity to produce yolk, or have they retained functional genes for vitellogenin production? If the former is true, hydropic eggs would represent a major constraint that would prevent endoparasitoids from reverting back to ectoparasitism. Such study will also require additional phylogenetic information. For certain parasitoid groups, like the braconids, current phylogenies are reasonably strong, but for many other taxa this clearly is not the case. Other areas of study would be to determine whether larval feeding and pupation strategies are uniformly linked and conserved within taxa. It is also important to determine the degree of plasticity in parasitoid size thresholds. The general literature suggests that size thresholds will be conserved and that size ranges of individual species will likely be narrow. If so, shifts in host range to hosts of larger size are more likely to lead to gregariousness

162

CHAPTER TEN

than increases in parasitoid size. Parasitoid taxa containing smaller species may show a greater tendency toward gregariousness than taxa composed of large species that already approach the average size of preferred hosts. Last, much more comparative information is needed about the specific genes and pathways regulating parasitoid growth. A view held by many developmental and evolutionary biologists is that core processes (metabolism, signaling pathways, genetic regulatory circuits) are highly conserved and act as constraints to evolution (see Raff 1996). Although they are conserved, others argue that these core processes actually reduce constraint by conferring extreme flexibility on other processes that regulate the novel functions and morphologies that have arisen among metazoans (Kirschner and Gerhart 1999). Parasitoids arguably exhibit greater developmental variation than any other arthropodan group (Strand and Grbic 1997). As such, comparative study of these insects offers a real opportunity for distinguishing between these views and enhancing our understanding of insect evolution. Acknowledgments. I wish to thank M. Grbic and J. Harvey for their helpful discussions and important contributions to some of the data discussed in this chapter. I also wish to thank M. Shaw and R. Wharton for providing helpful information on feeding and pupation behavior of ichneumonoids, and A. Ives and G. Heimpel for valuable comments on the manuscript. Some of the work presented was supported by grants from the National Science Foundation, National Institutes of Health, and U.S. Department of Agriculture to MRS.

Eleven Host Specificity and Trophic Relationships of Hyperparasitoids JACQUES BRODEUR

the last few decades, insect parasitoids have become fruitful models for studying questions in host-parasite physiological interactions, behavior, population biology, and evolutionary ecology (Godfray 1994; Hawkins 1994; chapters in this volume). Parasitoids exhibit very diverse, and often peculiar, life histories which allow them to exploit a wide range of niches. For various reasons, most of the literature has been devoted to primary parasitoids of saprophagous, herbivorous, and predatory arthropods. Despite their ecological importance and ubiquity, we have comparatively ignored aspects of the biology of hyperparasitoids. In their excellent books on the biology and evolution of parasitoids, Godfray (1994) and Quicke (1997) treat virtually every attribute of parasitic Hymenoptera in detail, yet dedicate only a few pages to secondary parasitoids, reflecting the scarcity of information and discussion on the topic of hyperparasitism. Conspicuously, hyperparasitoids are usually excluded from discussions on topics such as foraging behavior, competition, community structure, trophic interactions, host-parasitoid coevolutionary processes, and, to a lesser extent, population regulation. OVER

The first convincing attempt to synthesize our knowledge on hyperparasitoids and suggest stimulating perspectives for future research resulted from a symposium organized by Rosen in 1976 (Rosen 1981). A strong motivating factor behind this meeting was the role of hyperparasitoids in biological control, as they were traditionally thought to impede the actions of beneficial parasitoids (see DeBach 1964a; Luck et al. 1981; Mackauer and Volkl 1993; Rosenheim 1998; Mills, chapter 14). Predictably, in the years that followed, a number of increasingly refined hypotheses and models about the ways in which primary parasitoids interact with hyperparasitoids (Holler et al. 1994; Weisser et al. 1994; Volkl et al. 1995; Boenisch et al. 1997), how hyperparasitoids respond to spatial variation in host density (Weseloh 1986; Schooler et al. 1996; Muller and Godfray 1998), and how they may regulate herbivore populations by stabilizing host-parasitoid interactions (Beddington and Hammond 1977; Luck et al. 1981; May and Hassell 1981; Hassell and Waage 1984; Briggs 1993) were developed. Nevertheless, the debate on the

164

CHAPTER ELEVEN

ecological significance of hyperparasitism in arthropod communities and biological control remains open, in part because most current models neglect aspects of parasitoid foraging behavior (Mackauer and Vdlkl 1993). In a recent review, Rosenheim (1998) concluded that a limited amount of experimental evidence supports the idea that hyperparasitism significantly disrupts the "short-term regulation" of herbivorous host populations by parasitoids, but critical multigeneration studies have yet to be conducted to assess the long-term effects. The 1976 symposium also provided a timely synthesis of information on the taxonomic occurrence, host specificity, and sex determination of hyperparasitoids. Some of the observations revealed distinctive developmental and behavioral characteristics of hyperparasitoids. Although the degree of similarity between primary and secondary parasitoids is obvious because of their common evolutionary origins and life-history strategies, hyperparasitoids are likely to possess specific biological attributes enabling them to exploit resources from the third trophic level (see Strand, chapter 10, for related discussion). In this chapter, my aim is to show that host spectra of hyperparasitic wasps are complex and often transcend trophic levels. I will first review the rather scant literature on the taxonomic distribution and the host associations of hyperparasitoids. Next, I will examine whether available field records of host-hyperparasitoid associations fit the existing theory of parasitoid-host specificity. I will then look at different scenarios that illustrate how host utilization by hyperparasitoids allows them to play different roles in food web structure. I will focus on hymenopteran hyperparasitoids, as there are too few references to dipteran and coleopteran hyperparasitoids to be discussed meaningfully in a broad context.

Taxonomic Affiliation and Host Associations Hyperparasitoids are parasitoids that attack other species of parasitoids. The great majority of hyperparasitoids are members of the order Hymenoptera, and a few species belong to the Diptera and the Coleoptera (Gordh 1981). As with primary parasitoids, the dichotomies of endo- versus ectoparasitoid and idio- versus koinobiont parasitoids pertain to the developmental strategies of hyperparasitoids and thus remain useful to characterize life-history strategies (Strand, chapter 10). Hyperparasitism may be obligatory or facultative (i.e., species developing as either primary or secondary parasitoids). Facultative hyperparasitism may originate from competition between parasitoids for the host resource, and is considered as one potential evolutionary pathway to obligate hyperparasitism (Godfray 1994). In most cases, facultative hyperparasitism is interspecific, but van Baaren and colleagues (1995)

HYPERPARASITOIDS

165

recently described the example of a mymarid egg parasitoid that may develop as a facultative intraspecific hyperparasitoid. A few examples of facultative intra- or interspecific tertiary and quaternary parasitism have been reported (Sullivan 1987). However, Gordh (1981) hypothesized that tertiary hyperparasitism is too precarious to evolve as an obligate developmental strategy. Furthermore, there are fitness costs associated with such a mode of development, as shown by Kfir and Rosen (1981), who observed a significant decrease in the size of tertiary and quaternary parasitoids following the depletion of host resources. Another unique lifestyle of hyperparasitoids is heteronomy. In the family Aphelinidae, heteronomous species produce females as primary parasitoids of homopteran hosts and males as hyperparasitoids (Viggiani 1981; Walter 1983; Williams and Polaszek 1996). Last, Shaw and Askew (1976) introduced the term "pseudohyperparasitoids" to describe species that attack the cocooned primary parasitoid after it has destroyed the host, as opposed to true hyperparasitoids, which attack the primary parasitoid in its still-living host. At present, it is impossible to provide an exhaustive analysis of the taxonomic affiliation of hyperparasitoids, because their biology is poorly known and their host relationships not established for sufficiently diverse families, genera, and species of parasitoids. Nevertheless, using data gleaned primarily from Gordh (1981) (the first taxonomical analysis of hyperparasitoids), Gauld and Bolton (1988), and Goulet and Huber (1993), I have identified families of parasitic wasps that include obligate or facultative hyperparasitoids, estimated the occurrence of hyperparasitism within families, and compiled general information on host associations. Among Hymenoptera, hyperparasitism is only known in the suborder Apocrita "Parasitica," where it occurs in seven of eleven superfamilies (table 11.1). Hyperparasitism is common in sixteen families, mainly in the Ceraphronoidea, Chalcidoidea, Ichneumonoidea, and Trigonalyoidea; rare in five families; and not known in the twenty-four families of hymenopteran parasitoids. Despite the paucity of information for several groups, some inferences can be drawn from table 11.1. First, hyperparasitism has a wide taxonomic distribution, suggesting diverse evolutionary origins (Gordh 1981). Gauld and Bolton (1988) described four possible evolutionary pathways leading to hyperparasitism based on host-natural enemy relationships. Hyperparasitoids may have evolved in association with the primary host (type A), parasitoids of the primary host (type B), predators of the primary host (type C), or without any association with the primary host or its natural enemies (type D). There are very few studies that have examined the mechanisms by which the different modes of hyperparasitism may have evolved. Therefore, without detailed phylogenies of hyperparasitic species, it is impossible to reach conclusions about their relative occurrence and evolutionary significance within a group (Godfray

166

CHAPTER ELEVEN

TABLE 11.1 Families of Hymenopteran Parasitoids with Hyperparasitoid Members Hyperparasitism Facultative (F) / Obligatory (O)

Ectophagous (Ec) / Endophagous (En)

F/?

Ec

?/0

Ec

F/O

Ec/En

Chalcididae

F/O

Ec/En

Elasmidae

7

Ec

?/O

En

F/O F/?

Ec Ec

F/O

Ec/?

F F/O F/O

En 7 Ec

F/O

Ec/?

Taxon

Occurrence

Host Associations

SYMPHYTA Orussoidea Orussidae APOCRITA 'Aculeata' Chrysidoidea Dryinidae Chrysididae Vespoidea Tiphiidae APOCRITA 'Parasitica' Ceraphronoidea (2/2) Ceraphronidae

Megaspilidae Chalcidoidea (14/19) Aphelinidae

Encyrtidae

VV

Eucharitidae Eulophidae Eupelmidae Eurytomidae Leucospidae Mymaridae2 Ormyridae Perilampidae Pteromalidae

VV

Bethyhdae, Dryinidae, Braconidae, Ichneumonidae Chalcididae, Cynipidae Braconidae Aphelinidae, Eulophidae, Encyrtidae Ichneumonidae, Braconidae, Tachinidae Braconidae, Ichneumonidae Several Hymenoptera, mainly Chalcidoidea Hymenoptera, Diptera Chalcidoidea, Platygastroidea Diverse Hymenoptera Mymaridae Chalcidoidea Mainly Tachinidae and Ichneumonidae Diverse host associations

167

HYPERPARASITOIDS TABLE 11.1 (cont.) Hyperparasitism Occurrence

Facultative (F) / Obligatory (O)

1Zctophagous (Ec) / Endophagous (En)

Rotoitidae Signiphoridae

VVV

?/O

En

Chalcididae, Encyrtidae, Tachinidae

Tanaostigmatidae Tetracampidae Torymidae

VV

F/O

Ec/?

Trichogrammatidae2

Mainly Ichneumonidae, Pteromalidae, Tachinidae

"V

F

VVV

O

En

Aphelinidae, Encyrtidae, Braconidae

V

0

En

VV

F/O

Ec/En

Ichneumonidae, Tachinidae, Phoridae Ichneumonidae, Braconidae, Tachinidae

V

F

?

Taxon

Cynipoidea (1/5) Charipidae

Host Associations

Eucoilidae Figitidae Ibaliidae Liopteridae Evanioidea (0/3) Aulacidae Evaniidae Gasteruptiidae Ichneumonoidea (2/2) Braconidae Ichneumonidae Megalyroidea (0/1) Megalyridae Mymarommatoidea (0/1) Mymarommatidae Platygastroidea (1/2) Platygastridae Scelionidae1 Proctotrupoidea (1/9) Austroniidae

7

168

CHAPTER ELEVEN

TABLE 11.1 (cont.) Hyperpamsitism Taxon Diapriidae

Occurrence

Facultative (F) / Obligatory (O)

Ectophagous (Ec) / Endophagous (En)

vv

F/O

En

Eulophidae, Dryinidae, Braconidae, Tachinidae

VVV

F/O

En

Ichneumonidae, Tachinidae

Heloridae Monomachidae Pelecinidae Peradeniidae Proctotrupidae Roproniidae Vanhorniidae

Host

Associations

Stephnoidea (0/1) Stephanidae Trigonalyoidea (1/1) Trigonalyidae

'V = Rare; VV = common; VVV = very common. Only one species has been shown to develop as a facultative hyperparasitoid. Note: For each superfamily of Apocrita, the ratio of families including hyperparasitoid species are given in parentheses. Data are retrieved primarily from Gordh (1981), Gauld and Bolton (1988), and Goulet and Huber (1993). 2

1994). I will show in the next section that these evolutionary pathways are not necessarily exclusive. Second, as a consequence of the varied lifestyles in most families of Hymenoptera, none of the parasitoid families consist exclusively of hyperparasitoids. For example, species of the Pteromalidae may be ectophagous or endophagous, idiobionts or koinobionts, solitary or gregarious. Some are primary or secondary parasitoids, while others are predators. Within the Eulophidae, although most species are primary parasitoids, obligate and facultative hyperparasitism have been frequently observed in three of the four subfamilies. At a lower taxonomic level, however, all species within a subfamily or genus may be obligate hyperparasitoids, suggesting that, within these groups, hyperparasitism follows phylogenetic lines. The most important subfamilies of obligate hyperparasitoids are the Alloxystinae (Charipidae), the Mesochorinae (Ichneumonidae), and the Eucerotinae (Ichneumonidae). Notably, all members of these three lineages are koinobiont

HYPER PA RASITOIDS

169

endoparasites, although the status of the Eucerotinae is questionable (Gauld 1988). Third, hyperparasitism is a rare phenomenon in three important groups of parasitic wasps. In the Braconidae, one of the largest families in the Hymenoptera, hyperparasitoid species occur only in the subfamily Alysiinae (Wharton 1984). This is not surprising, since very few braconids parasitize Hymenoptera (Gauld 1988). Furthermore, idiobiont braconids do not exploit cocooned hosts, which are common hosts of facultative hyperparasitoids, notably of several lineages of ichneumonids (Gauld 1988). In the Proctotrupoidea, a superfamily characterized by its morphological and host spectrum diversity, hyperparasitism is, as far as we know, restricted to the Diapriidae. In addition, there is limited evidence of facultative hyperparasitism within families of egg parasitoids. Nevertheless, hyperparasitism may be more common than presented in table 11.1. As noted by Gordh (1981), "the nature of hyperparasitism often makes its detection difficult," and misidentification of a secondary parasitoid as a primary parasitoid may be frequent. Finally, host-hyperparasitoid associations are quite diverse. Members belonging to either a holophyletic (a group of taxa that have a single ancestral species and include all the descendants of the ancestor) or a paraphyletic (a taxon that includes some, but not all, descendants of a common ancestor) assemblage may parasitize primary parasitoids from several unrelated taxa in the Hymenoptera and Diptera, reflecting the great variety of species and lifestyles within each family of hyperparasitoid. An analysis of host-use patterns and their underlying mechanisms is, therefore, required at a lower taxonomic level to better understand patterns of host use by Hyperparasitoids.

Hyperparasitoid Host Range According to Askew (1994), host range is the most variable biological trait of parasitoids. Very few parasitoids are 100% host-specific. Some are restricted to a single genus or host family, whereas others are polyphagous and exploit hosts from different orders sharing a particular habitat. In recent years, extensive efforts have been made to characterize the host spectrum of several groups of parasitic wasps through analysis of parameters from the habitat, the host plant, the life cycle, physiological suitability, feeding biology, and abundance of the host, and the foraging behavior and competitive capacity of the parasitoid (Gross and Price 1988; Hawkins et al. 1990; Sheehan 1991; Sheehan and Hawkins 1991; Mills 1992; Askew 1994; Whitfield 1994; Brodeur and Vet 1995; Brodeur et al. 1997; Geervliet 1997; Miiller et al. 1999). Although host associations of the vast majority of species remain unexplored, patterns have emerged indicative of a general model

170

CHAPTER ELEVEN

of host range in parasitoids (Askew and Shaw 1986; Gauld 1988; Shaw 1994). Mode of development, phylogeny, and shared ecology have been identified as the three most important determinants for the evolution of parasitoid host range (summarized by Strand and Obrycki 1996; see also Strand, chapter 10) and are key concepts for making inferences about host-hyperparasitoid associations. Askew and Shaw (1986) made a significant advance in the study of parasitoid host specificity when they developed the concept of idiobiont versus koinobiont mode of development (sensu Haeselbarth 1979). They predicted that koinobiont parasitoids, whose developing larvae are constrained by the physiological suitability and nutritional value of the host, would be more host-specific than idiobiont parasitoids. This hypothesis was subsequently supported by studies of parasitoid assemblages (Sheehan and Hawkins 1991; Shaw 1994). The correlation between phylogeny and parasitoid host use is self-evident, but not fully explored, partly due to the patchy reliability of existing hymenopteran phylogenies (Gauld 1986). Although a few studies have shown that parasitoid specificity is strongly determined by the phylogenetic relationship of the hosts (Griffiths 1964; Wharton 1993), others have not (Gauld 1986; Hoffmeister 1992). In general, phylogenetic history is more reliable in predicting which taxa would likely serve as hosts for a given parasitoid species than for predicting the absolute number of host species a given parasitoid actually uses (Gauld 1986). In contrast, evidence of the importance of shared ecology for the host-specificity of parasitoids is well established. Variations in parasitoid host range are likely to be related to "the ecological opportunities offered by the current environment" (Shaw 1994). For example, host plant species, phenology, and architecture largely determine patterns of host use by leaf miner parasitoids (Askew 1994). Accordingly, recent theory has emphasized the role of female foraging behavior in shaping parasitoid host range (Vet and Dicke 1992). A commonly held assertion in the literature is that hyperparasitoids have a wider host range than primary parasitoids (Muesebeck and Dohanian 1927; Hagen and van den Bosch 1968; Krombein et al. 1979; Gordh 1981). However, there are very few detailed studies, if any, that seriously document this. Furthermore, the observed patterns may be misleading due to uncertain taxonomic status (synonymy) and unreliable host-hyperparasitoid records, since the identity of the primary parasitoid is often unknown or assessed by association with the most commonly reared species. Existing records of hosthyperparasitoid associations are not likely to settle this question. Nevertheless, they tend to confirm the prediction of Askew and Shaw (1986) that idiobiont ectophagous parasitoids have the potential to exploit a broad spectrum of hosts. To illustrate this point and merge the influence of shared ecology, I will examine the host specificity of aphid hyperparasitoids, as these are certainly the best known models in terms of taxonomy, host asso-

HYPER PA RASITOIDS

171

ciation, mode of development, behavior, and impact on primary parasitoid populations (Sullivan 1987, 1988; Mackauer and Volkl 1993). Secondary parasitoids of aphids are found in four families of Hymenoptera: the Megaspilidae, Pteromalidae, Alloxystidae, and Encyrtidae. They are solitary and attack both the aphidiine and aphelinine parasitoids of aphids. Hyperparasitoids of the first two families are ectophagous, idiobiont parasitoids, with females ovipositing on the prepupa or pupa of their host after the primary parasitoids have killed and mummified the aphid. Alloxystinae are endophagous, koinobiont hyperparasitoids, where females oviposit in the primary parasitoid larva within the living aphid. The Alloxysta larva hatches from the egg and feeds internally on the parasitoid prepupa only after aphid mummification. The developmental strategy of encyrtid aphid hyperparasitoids is more complex and not well understood. The females appear to have the capacity to attack both living and mummified aphids. They lay their eggs inside the primary parasitoids, where the larvae first develop as endophagous parasites, but feed ectophagously in later larval stages (Kanuck and Sullivan 1993). In the seventies, extensive field surveys of aphid mummies and laboratory experiments on host selection suggested that alloxystine hyperparasitoids exhibit a high degree of host specificity (Gutierrez 1970; Sullivan and van den Bosch 1971; Evenhuis 1976; Andrews 1978). However, there are a number of potential sources of error in the determination of host associations in aphid hyperparasitoids, mainly the alloxystines. One way to circumvent such problems is to analyze hyperparasitoid specificity at the generic level of primary parasitoids, the most common genera of aphid parasitoids being distinguishable through differences in structure, form, and coloration of the mummies (Stary 1970). Although not primarily designed to examine the host range of aphid hyperparasitoids, a study by Holler and colleagues (1993) has yielded fresh insight into this issue. Data collected from cereal fields in northern Germany showed that the mean number of aphid parasitoid genera attacked by hyperparasitoid species was three for Megaspilidae, four for Pteromalidae, and two for Alloxystidae (table 11.2). These results indicate that koinobiont Alloxystidae have a narrower host range than idiobiont Megaspilidae and Pteromalidae. Except for Dendrocerus aphidum, all idiobiont species have been reared from a minimum of three different primary parasitoid genera. Notably, the two Asaphes species not only had the capacity to parasitize the four genera of the Aphidiinae, but were also the only ones recovered from the aphelinid. Alloxysta species were restricted to one or two genera of the aphidiine. One remarkable exception to the general pattern observed between idiobiont and koinobiont species was Phaenoglyphis villosa, an alloxystine that developed in all four genera of the Aphidiinae. The factors that have produced what appears to be an atypical pattern of host range use are unknown.

172

CHAPTER ELEVEN

TABLE 11.2 Host Ranges of Aphid Hyperparasitoids Collected from Cereal Fields in Northern Germany Genus of primary aphid parasitoid Hyperparasitoid

Aphidius

Ephedrus

Praon

Trioxys

Aphelinus

Total

Megaspilidae Dendrocerus carpenteri Dendrocerus aphidum Dendrocerus laticeps Dendrocerus dubiosus

4 2 3 3

Pteromalidae Asaphes vulgaris Asaphes suspensus Pachyneuron solitarium Coruna clavata

5 5 3 3

Alloxystidae Alloxysta victrix Alloxysta leunisii Alloxysta macrophadna Phaenoglyphis villosa

1 1 2 4

Source: Modified from Holler et al. (1993).

More recently, Miiller and coworkers (1999) described the primary and secondary parasitoid complexes of aphid hosts in an abandoned field in southern England. Their findings suggest that hyperparasitoid host specificity was similar to that observed by Holler and colleagues (1993), with the mean number of aphid parasitoid genera attacked per hyperparasitoid species at 2.75 for Megaspilidae, 3.75 for Pteromalidae, and 2.3 for Alloxystidae. Of significance in both studies is the fact that within idiobiont species, pteromalids appear to have broader host ranges than megaspilids. An extensive survey of aphid hyperparasitoids from Australia provide complementary information, at the species level, about the host specificity of Alloxystidae (Carver 1992). The cosmopolitan A. fuscicornis has been reared from five aphidiine species belonging to two genera, whereas A. australiae, known only in eastern Australia, is host-specific to Aphidius colemani (table 11.3). Similarly, A. darci, also an Australian species, specializes on aphelinids. As observed by Holler and coworkers (1993) in Germany, P. villosa has been recovered from all four genera of the Aphidiinae in Australia. My analysis is essentially qualitative, as the relative distribution and abundance of primary and secondary parasitoids, as well as the foraging behavior and temporal pattern of activities of hyperparasitoids, are not taken into ac-

HYPERPARASITOIDS

173

TABLE 11.3 Host Ranges of Alloxystine Aphid Hyperparasitoids in Australia

=§

= hyperparasitoid, Q > = tertiary parasitoid, \ / = predator, / \ = plant. The shaded symbols refer to a single species that can occupy different roles but is a hyperparasitoid at least under some conditions. © and 0 indicate the beneficial or detrimental impact that each trophic protagonist may have on the host plant in a biological control context.

HYPER PA RASITOIDS

179

hyperparasitoid could therefore play a beneficial role if it attacks the primary parasitoid of the coccinellid, but a detrimental one to biological control if it parasitizes the primary parasitoid of the mealybug. The same pattern has been observed for P. aegyptiacus in Africa, where it has been reared from primary parasitoid species of both the cassava mealybug and its coccinelid predators (Neuenschwander et al. 1987). The fourth scenario reveals the complexity of relationships hyperparasitoids may have with their hosts (figure 11.2D). Gauld and Bolton (1988) reported that the chalcidoid Eurytoma brunniventris (Eurytomidae), which inhabits cynipid galls, has the capacity to feed on gall tissues (Askew 1961), parasitize cynipid gall makers, or parasitize primary chalcid parasitoids of gall makers. This facultative idiobiont ectoparasitoid thus has the potential to exploit any one of three trophic levels within the same microhabitat. To my knowledge, this type of scenario has been observed only for a few species from the genera Eurytoma and Sycophila (Eurytomidae). The fifth scenario describes the fascinating host associations of Pachyneuron concolor (Foerster), a pteromalid species that develops as a pupal ectoparasite within cocoons or mummies of its host. Following detailed field observations and laboratory rearings, Kfir and Rosen (1981) and Rosen and Kfir (1983) established that P. concolor is a polyphagous obligate hyperparasitoid of mealybugs and soft scales, but a primary parasitoid of predators within a distinct aphid system (figure 11.2E). Pachyneuron concolor has been reared from Microterys flavus, an encyrtid parasitoid of the brown soft scale, Coccus hesperidum (L.), and from various species of the encyrtid primary parasitoid of Chilocorus bipustulatus (L.), a coccinelid predator of soft scales and mealybugs. Pachyneuron concolor may also develop as facultative intra- or interspecific tertiary hyperparasitoids. Furthermore, P. concolor has the capacity to develop as a primary parasitoid of Leucopis (Diptera: Chamaemyiidae), an aphidophagous fly predator. Establishing the status of P. concolor in biological control is difficult, given the multiplicity of trophic connections (figure 11.2E). However, Rosen and Kfir (1983) concluded that its overall effect is more likely to be detrimental because of the "benevolent, 'Dr. Jekyll' aspect" of P. concolor, which corresponds to relatively rare events of secondary and tertiary parasitism, is overwhelmed by the more frequent "malevolent, 'Mr. Hyde' aspect." The competition for, and partitioning of, the host resource that occur among primary and secondary parasitoid species could select for the acceptance of alternate hosts from various trophic levels. Undoubtedly, several hyperparasitoid species have the capacity to adapt rapidly to the presence of alternative hosts and use new food resources, suggesting that they possess behavioral plasticity with respect to host searching and host acceptance. The scenarios of figure 11.2 indicate that the common representation of hyperparasitoids set on top of food-web diagrams in textbooks of ecology

180

CHAPTER ELEVEN

does not necessarily reflect reality in nature. Most organisms occupy a single trophic level in resource-consumer interactions: Scavengers develop on decaying materials, fungivorous arthropods feed and reproduce on fungal hyphae, herbivores exploit plant materials, predators eat prey, and so forth. However, various types of natural enemies may exploit resources across different trophic levels (Polis and Winemiller 1996). Facultative hyperparasitism typically involves complex trophic relationships as one species may occupy two, three, and even four different levels. This has inevitable dynamic consequences for constituent species of a community (see next section). The study of parasitoid food webs is in early stages (Memmott and Godfray 1994; see also Holyoak, chapter 12), and identifying the role of hyperparasitoids may be a Pandora's box for ecologists. Careful observations and field studies would definitely produce additional complex and fascinating scenarios.

Discussion and Future Directions This short review indicates that our knowledge of insect hyperparasitism is based on a paucity of information. It also pleads for more recognition of the opportunities provided by hyperparasitoids for studying the evolution and ecology of parasitoid host-range and trophic relationships (see Holyoak, chapter 12). I believe that studies of the life history of hyperparasitoids, the evolution of facultative hyperparasitism, and the consequences of hyperparasitism for community ecology should be especially rewarding to ecologists. Life History The evolutionary transitions from primary parasitism to facultative hyperparasitism, and from facultative hyperparasitism to obligate hyperparasitism, are likely to induce changes in life-history characteristics in response to the different selective forces associated with each type of parasitism. Unfortunately, the life histories of only a few species of hyperparasitoids have been studied in detail, making a discussion and search for patterns rather tenuous. Major gaps exist in our knowledge of the mode of development, life-table characteristics, searching behavior, and competitive capacity of hyperparasitoids. This is due, in part, to the obvious experimental difficulties encountered where, in the most simple case, plants, herbivores, primary parasitoids, and hyperparasitoids have to be maintained simultaneously in the laboratory. Given their diversity and the inherent difficulties in studying aspects of their biology and ecology, we are unlikely to ever have a complete

HYPERPARASITOIDS

181

understanding of the natural history of hyperparasitoids. However, accurate knowledge of the natural history of some important groups of hyperparasitoids is a prerequisite for improving our understanding of their origin, distinctive biological attributes, and role in community structure. Furthermore, our efforts should be directed toward obtaining a better understanding of the distribution of hyperparasitism among evolutionarily diverse groups of primary parasitoids. As previously mentioned, a few taxa within certain families of parasitic Hymenoptera are exclusively hyperparasitoids, suggesting phylogenetic origins. Thus, a comparison of hyperparasitic taxa with the most closely related taxa composed exclusively of primary parasitoids could be valuable and might suggest patterns in the evolution of hyperparasitism. For example, this could be achieved within the Ichneumonidae, since their phylogenies are fairly reliable (Quicke 1997). By contrast, very few, if any, inferences could be made about the evolution of hyperparasitism within polyphyletic groups, such as the Pteromalidae or the Proctotrupoidea. Data from table 11.1 are preliminary and generally come from studies of host-parasitoid systems with particular relevance to human activities. More attention must be devoted to poorly known, small taxa. Just as importantly, studies of the distribution of hyperparasitism among dipteran and, perhaps, coleopteran parasitoids could be helpful to identify general conditions under which hyperparasitism may have evolved.

Facultative Hyperparasitism The question of facultative hyperparasitism is key to understanding parasitoid trophic shifts. Competition for hosts is likely to be the most significant factor that has shaped the evolution of facultative hyperparasitism (Godfray 1994). Such a life-history strategy not only provides an additional host source, but also may reduce interspecific competition; facultative hyperparasitoids may benefit from parasitism on primary parasitoids by excluding them from the habitat. How accurately can we predict the nature and outcome of complex interactions among primary and secondary parasitoids? Presumably, a familiarity with the developing literature on intraguild predation (Polis and Holt 1992; Polis and Winemiller 1996) would lead to novel and fruitful approaches to the study of these ecological questions. As with intraguild predation, several factors may determine the magnitude, symmetry, and outcome of intraguild parasitism, the two most important factors probably being host specificity and extraguild host density. For example, Lucas and colleagues (1998) recently suggested various scenarios to characterize the relationship between extraguild prey density and intraguild predation. Data from different aphid predator combinations indicated that, in most cases, intraguild predation decreases with increased aphid den-

182

CHAPTER ELEVEN

sity. No study has quantified the dynamics of host use by facultative hyperparasitoids. Do they have the capacity to modulate their host selection behavior in response to variations in host abundance and quality? We should expect facultative hyperparasitoids to behave as secondary parasitoids when primary hosts are rare or unavailable. The latter prediction suggests that facultative hyperparasitoids would be better competitors than related primary parasitoids under conditions of low host density. Alternatively, there might be fitness costs associated with facultative hyperparasitism. The benefit of parasitizing a parasitoid may be lower than the benefit of developing in a primary (herbivorous) host for a facultative hyperparasitoid. In the absence of relevant studies, we have no evidence to support or reject this hypothesis. However, if there are trade-offs in facultative hyperparasitoid performance between parasitized and nonparasitized hosts, the probability that a given host will be accepted should be expected to vary accordingly. Kfir and colleagues (1993) showed that the facultative hyperparasitoid Tetrastichus howardi (Eulophidae) prefers to attack stem borers over two species of dipteran and hymenopteran primary parasitoids. In the future, it will be important to learn the proximate and ultimate causes of facultative hyperparasitism. Community Structure An important component of future studies in community ecology should be the role of hyperparasitoids in agricultural and natural ecosystems (Tscharntke, chapter 15). Hyperparasitoids can attack a wide range of hosts; some are capable of transferring from one trophic level to another as competition increases or host condition deteriorates. What are the ecological consequences of these complex host associations of hyperparasitoids for the dynamics of arthropod communities? As discussed above, most models and reviews have dealt with hyperparasitoids within the context of biological control (Luck et al. 1981; Hassell and Waage 1984; Mackauer and Volkl 1993; Rosenheim 1998). To my knowledge, there has been no attempt to analyze the role of transitions by hyperparasitoids on the constituent species of a community. Because hyperparasitoids are ubiquitous in many systems, and because they increase the complexity of trophic connections among species, we may have underestimated their role in influencing community stability. Presumably, facultative trophic shifts would promote the persistence of interacting species. In addition, the possibility that secondary parasitoids significantly constrain primary parasitoid richness in certain systems needs to be examined. Hyperparasitism represents a unique challenge for population ecologists. Any attempt to broaden current parasitoid population models by adding hy-

HYPERPARASITOIDS

183

perparasitoids with different life histories will contribute to the development of a more complete theory of parasitoid food webs (Beddington and Hammond 1977; Hassell 1979; May and Hassell 1981). Nevertheless, theoretical studies remain ahead of experimental evidence. Although direct observations and correlations of complex trophic patterns may be difficult to achieve in nature, efforts must be made, because this information is crucial for understanding how communities function. Acknowledgments. I sincerely acknowledge Y. Carriere, B. Hawkins, M. E. Hochberg, A. R. Ives, J. N. McNeil, J. Rosenheim, W. Volkl, and an anonymous reviewer for critical and helpful comments on an earlier version of the manuscript. I thank A. Bouchard for assistance with creating the figures.

Twelve Comparing Parasitoid-Dominated Food Webs with Other Food Webs: Problems and Future Promises MARCEL HOLYOAK

webs depict patterns in the flow of energy or materials among living organisms, and sometimes include flows to the abiotic and dead biotic environments. Food webs are potentially important for solving environmental problems and making practical predictions about the response of species assemblages to hunting, pollution, and pest management. Theory about food webs has also led to some very general ecological predictions; for example, in stable food webs, the product of the total number of species and the frequency of connections between species is expected to be a constant value (May 1973; Pimm 1980, 1982). Empirical observations can also reveal food web patterns that are attractive because of their apparent generality; for example, Cohen (1978) and Briand and Cohen (1984) found that in a wide range of published food webs, the ratio of predator to prey species was usually close to 4:3. Whether patterns such as these are real or not has been the subject of much debate (e.g., Paine 1988; Martinez 1991, 1993). Food webs are studied with many different aims in mind. The net result is that it has been difficult to standardize the factors measured, how they are measured, and the methods used to analyze food web patterns. Paine (1988) commented that food webs "are what naturalists have observed and assembled to provide a convenient, qualitative guide or road map to those relationships that fascinated them." He even went as far as to point out that these qualitative descriptions were never intended to be data for testing theoretical predictions (Paine 1988). Paine's arguments are not without substance; for example, he points out that the 4:3 predator-prey ratio is not influenced by lumping groups of organisms together, suggesting that it is so general that it is not measuring anything useful. Notwithstanding this criticism, there have been numerous attempts to analyze published food webs for ecological patterns, and a large body of ecologists believes that these analyses serve a useful purpose (e.g., Martinez 1991; Cohen et al. 1993). Here, I will report on one such analysis, where I will compare food webs that are dominated by parasitoids to those that are dominated by other insect predators. While the results of this analysis are tantalizing, they also

FOOD

PARASITOID-DOMINATED FOOD WEBS

185

reveal some of the problems that are inherent in present-day comparative analyses of food webs. Below, I will first introduce the hypotheses that I tested, and then describe some of the problems with measuring and comparing food-web parameters. After describing the analyses, I will end by discussing their interpretation and by suggesting how we might improve comparative analyses of food webs that involve parasitoids.

Hypotheses Koinobiont parasitoids (and endoparasitoids) are in more intimate contact with their hosts for longer time periods than are idiobiont parasitoids (and ectoparasitoids) and most predators. This is likely to necessitate a greater degree of trophic specialization among koinobiont parasitoids than among idiobionts or predators (Askew and Shaw 1986; Hawkins 1994). Trophic specialization may also have repercussions for food web structure. Modeling work by Pimm and Lawton (1977) suggests that population dynamics could limit the length of food chains, so that we might expect species that exhibit more variable population dynamics to be found in shorter food chains because it is harder for a predator species to survive on a prey population that fluctuates widely in abundance. MacArthur (1955) proposed that trophic generalists should have less variable abundances through time than trophic specialists because they are less likely to crash to low abundances (where extinction is likely) during unfavorable time periods, since it is unlikely that all of their food sources will crash simultaneously. Conversely, Watt (1964) proposed that generalists should actually be more variable, because they can exploit a wider range of food sources during favorable periods. For parasitoids, Redfearn and Pimm (1992) showed that, as predicted by MacArthur (1955), those that attacked a greater number of hosts had lower temporal population variability than species with a smaller number of hosts. Recent modeling work also shows that a special form of generalism, omnivory (feeding at more than one trophic level), is stabilizing as long as interactions are not too strong (McCann and Hastings 1997). This effect of omnivory is opposite to that found by Pimm and Lawton (1978); the two models can be reconciled by noting that Pimm and Lawton used random interaction strengths that were occasionally very strong, and that were also destabilizing in the model of McCann and Hastings (e.g., McCann et al. 1998). Overall, we might expect food chains to be longer in webs that contain more generalists, but this has not been tested in webs that are dominated by parasitoids. I tested two hypotheses related to the prevalence of parasitoids in food webs consisting mainly of terrestrial arthropods: 1. As the proportion of enemies that are parasitoids increases, food chains should become shorter. This is expected because parasitoids are

186

CHAPTER TWELVE

frequently koinobionts or endoparasitoids that are expected to be more closely adapted to the physiology and/or behavior of their host than are idiobiont parasitoids, ectoparasitoids, and most predators (Askew and Shaw 1986; Hawkins 1994). A close adaptation to host traits is something that we expect to see most frequently in trophic specialists. Webs with many trophic specialists are expected to have shorter food chains if food chain length is limited by population fluctuations being larger for specialists than generalists (e.g., Redfearn and Pimm 1992), or because trophic specialization rules out the possibility of omnivory that may stabilize food webs (McCann and Hastings 1997). 2. Webs in which more of the natural enemies are parasitoids should be less speciose than webs in which the enemies are predators. This follows if webs with more parasitoids have shorter food chains (hypothesis 1), providing that there are not more species at lower trophic levels in webs with fewer trophic levels. Food webs to test these hypotheses came from a published survey by Schoenly and colleagues (1991), supplemented with some additional published webs that I found during a survey of the literature. The emphasis of their work was to compare both how the numbers of insects in food webs differed among habitats and whether the proportion of species that were insects influenced food web patterns, such as web shape, food chain length, predator-prey ratios, and links per species. Source webs are those that originate with a single host or prey (herbivore or detritivore) species. Eleven community webs (e.g., Pimm et al. 1991) that include all taxa in a single habitat without regard for trophic relationships were split to separate source webs for each host or prey species. Hawkins, Martinez, and Gilbert (1997) showed that, compared to community webs, source webs gave biased impressions of numbers of species; fractions of basal, intermediate, and top species; connectance; and linkage density. It is therefore not advisable to compare source and community webs directly. In source webs I included all plant or basal resource species for each prey. Sink webs, which trace links down from one top predator, were not considered in this analysis. In total, webs came from thirteen independent sites and included fifty-eight semiindependent source webs. Statistical analyses allowed for the fact that data points (source webs) from the same study are not independent. The selected webs were collected with many different aims in mind, including studies of decomposition, island biogeography, and food web theory. Consequently, there might be many reasons why the webs are not comparable. For the present study, some of the more important problems are (from Paine 1988 and Cohen et al. 1993): (1) Authors differ in the extent to which species are resolved taxonomically; this is as a consequence of the authors' interests and knowledge, and the level of difficulty of identifying the taxa

PARASITOID-DOMINATED FOOD WEBS

187

involved. Frequently, species are lumped into "trophic species," which share the same prey and predator species (e.g., Cohen et al. 1993). For this study, lumping species in this manner would bias assessment of the numbers of the species and the numbers of specialists and generalists (see also Schoenly et al. 1991). Many webs include entries at the genus level without indicating if these represent a single species that was not identified to the species level, or whether these entries represent several species that were not (or could not be) separated. Webs also vary in whether life stages with different diets are separated and whether cannibalism is included. (2) Authors vary in what they consider to represent a trophic link. The problem of identifying links encompasses variation in the number of times a feeding interaction needs to be observed, whether feeding is observed directly or is inferred from indirect evidence (e.g., feeding marks and literature references), and interaction strength, which measures the per capita impact of one species on densities of another species. It should be noted that the problems caused by lumping species into trophic species are much worse for studies that consider quantitative interaction strengths, because different species will rarely have similar interaction strengths. (3) Sampling effort is usually not reported, and in some cases authors do not detail sampling methods, the spatial and temporal scale of sampling, or whether the web is cumulative or based on samples at a single point in time and space. Cumulating species through time or across space might be misleading to studies of population dynamics if temporal or spatial heterogeneity is present. Estimates of trophic specialization are likely to be especially dependent on sampling effort. Another major concern in the comparison of existing food webs is that observed patterns are correlative. Most importantly, correlations do not identify cause and effect: There may be many potential causes for the patterns observed. It is wise to regard the results of such studies as being indicative of patterns that require further study, in the form of further correlative studies to test whether the patterns are general and experimental studies to test mechanisms. Although there are many potential pitfalls to conducting food web studies, here I will primarily discuss the potential problems of sampling artifacts, because the hypotheses I address may be particularly sensitive to these problems.

Methods Table 12.1 presents a list of webs used during this study. In selecting webs: 1. I allowed webs to contain up to 5% of taxa that consisted of more than one similar species (within the same family and, ideally, the same genus), and rejected all webs that had either prey or enemy species

188

CHAPTER TWELVE

lumped into larger categories, such as "parasitic hymenoptera" or "aphids." 2. Webs had to come from a single habitat. 3. I required that the web was either illustrated or that all trophic links were tabulated—in cases of doubt I excluded the web (Schoenly et al. 1991). 4. Species were sometimes present in published webs more than once if there was a change of diet across life-history stages; the "number of species" might therefore be more correctly interpreted as the number of trophic species. 5. No attempt was made to scale different webs for variation in sampling effort. While an increase in sampling effort would have added species to the webs, Schoenly and coworkers (1991) showed that the problem was not too severe. They used rarefaction of the number of observed links to show that, in general, there was a high return of species from the reduced samples compared to the total reported number of species. Therefore, continued sampling would have added relatively few species to the webs used in this 1991 study. It is difficult to find all hosts of a particular parasitoid, but relatively easy to find all parasitoids of a particular host (Shaw 1994). This might lead to more complete records of predatory links for parasitoids than for other types of predator. Consequently, patterns of specialism in parasitoid-dominated webs are likely to depend on the variety of host species examined, and whether the predatory species are parasitoids or predators. In predators, analyses of gut contents may reveal large numbers of hosts, and such analyses are not possible in parasitoids. 6. Where there were uncertain links within webs, I arbitrarily chose to take the more complete version of the web. This was shown by Schoenly and colleagues (1991) not to be a large form of bias for webs similar to those used here. In the case of parasitoids, a further problem was how to score cases of multiple parasitism (where several parasitoid species attack the same host individual). If parasitoids also consume other parasitoid species within the host, then this should probably be scored as intraguild predation (where two predators consume the same prey and one predator species can also consume the other), and another trophic level should be added to the food chain. The incidence and consequences of multiple parasitism were rarely reported in food web studies, but are likely to be particularly important for scoring species as specialists or generalists and counting trophic levels. Therefore, where authors reported multiple parasitism as intraguild predation, an additional trophic level was added (figure 12.1). For each web, I recorded total number of species, predators, and parasitoids (table 12.1); numbers of species in each trophic level; modal food

PARASITOID-DOMINATED

189

FOOD WEBS

Trophic level: 6

1 Figure 12.1. An example of a source web (Hopkins 1984). The web has been redrawn from Hopkins to illustrate treatment of intraguild predation and assignment of trophic levels. Rumex is a plant, Apion is a weevil, and all other species are parasitoids; 100% of the enemies (trophic level three or higher) are parasitoids. The web contains ten chains (R-A-Tr-M-Te; R-A-Te; R-A-Eu-M-Te; R-A-En-C-Eu-M-Te; R-A-Eu-C; R-A-EnM-Te; R-A-En-Eu-M-Te; R-A-En-Eu-C; R-A-C-Eu-M-Te; and R-A-H, where each species is represented by its initial one or two letters). The modal chain length is four species. A tenth species recorded by Hopkins (1984) was excluded from the web because it was known to be a hyperparasitoid, but it was not clear which species was the host.

chain length (figure 12.1); and habitat category (detailed in table 12.1). Preliminary analyses showed that the differences in habitat categories of webs with different proportions of enemies that were parasitoids did not interfere with the patterns reported here, so I did not consider habitat further. As a consequence of considering several source webs from a site (table 12.1), some of which came from the same community web, the sample of source webs is not independent. Statistics should therefore use independent subsets of source webs. Studies were regarded as independent if they had no recorded species in common; altogether, there were thirteen independent sets of studies, with between one and fifteen source webs from each study (table 12.1). Both hypotheses I tested involve correlations between two variables. To test for a significant correlation between two variables (xt and yy-, i = 1

190

CHAPTER TWELVE

TABLE 12.1 Source Food Webs Used in Analyses Source

Identification

Habitat*

Species

Rejmanek and Stary 1979 Rejmanek and Stary 1979 Rejmanek and Stary 1979 Force 1974 Whittaker 1984 Hopkins 1984 Fagan 1997 Fagan 1997 Fagan 1997 Fagan 1997 Fagan 1997 Fagan 1997 Fagan 1997 Fagan 1997 Fagan 1997 Fagan 1997 Fagan 1997 Fagan 1997 Mayse and Price 1978 Mayse and Price 1978 Mayse and Price 1978

Betula pendula canopy Pinus sylvestris canopy Quercus robur canopy Baccharis sp. Phoradendron tomentosum Rumex sp. Malezonotus Sciarid Tetrix Empoasca Nysius Cuerna Aceratogallia Macrosiphum Philaenus Lygus Ligyrocoris Bromius Empoasca fabae wk3 Sericothrips variabilis wk3 Other herbivorous thrips wk3 Black alate aphid wk 3 Empoasca fabae wkl2 Sericothrips variabilis wkl2 Other herbivorous thrips wkl2 Colias eurytheme Andricus curvator Andricus kollari Andricus ostreus Andricus o. furunculus Andricus quadrilineatus Cynips divisor Cynips longiventris Cynips quercus-folii Neuroterus albipes Neuroterus aprilinus Neuroterus numismalis Neuroterus quercus-baccarum

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2

13 12 13 8 22 9 6 4 4 6 7 5 5 5 4 4 4 4 5 4

6 7 8 6 3 7 0 0 0 0 0 0 0 0 0 0 0 0 0 0

0 0 0 0 9 0 4 2 2 4 5 2 2 2 2 2 2 2 3 2

2 2 2

4 5 8

0 0 0

2 3 6

2

8

0

6

2 2 3 3 3 3 3 3 3 3 3 3 3

7 6 16 14 7 3 8 14 12 11 9 3 6

0 0 14 12 5 1 6 12 10 9 7 1 4

5 4 0 0 0 0 0 0 0 0 0 0 0

3

11

9

0

Mayse and Price 1978 Mayse and Price 1978 Mayse and Price 1978 Mayse and Price 1978 Mayse and Price 1978 Askew 1975 Askew 1961 Askew 1961 Askew 1961 Askew 1961 Askew 1961 Askew 1961 Askew 1961 Askew 1961 Askew 1961 Askew 1961 Askew 1961

Parasitoids

Predators

191

PARASITOID-DOMINATED FOOD WEBS

TABLE 12.1 (cont.) Source

Identification

Habitat*

Species

Parasitoids

Predators

Askew 1961 Askew 1961 Hawkins and Goeden 1984 Hawkins and Goeden 1984 Hawkins and Goeden 1984 Hawkins and Goeden 1984 Hawkins and Goeden 1984

Andricus curvator Callirhytis glandium Leaf cluster bud Club stem A. polycarpa nodular stem A. polycarpa wooly stem A. polycarpa woody + smooth stem A. canescens blister leaf + nipple bud A. canescens, tumor stem A. canescens, oval twig A. canescens, oval bud/ flower A. canescens fuzzy bud/ flower Asphondylia borrichiae Phyllonorycter Ground cherry Horse nettle Tree flux Haematobia irritans Sarcophaga sp. Paregle cinerella

3 3 3 3 3 3

10 3 3 9 9 24

7 1 1 6 6 13

0 0 0 1 1 3

3

9

4

2

3 3 3

13 19 12

8 9 9

1 3 1

3

12

8

1

3 3 4 4 4 5 6 6 6

12 6 11 8 15 9 6 5 6

7 4 9 6 13 0 3 1 1

1 0 0 0 0 4 2 2 3

Hawkins and Goeden 1984 Hawkins and Goeden 1984 Hawkins and Goeden 1984 Hawkins and Goeden 1984 Hawkins and Goeden 1984 Stiling and Rossi 1994 Askew 1975 Gross and Price 1988 Gross and Price 1988 Robinson 1953 Mohr 1943 Mohr 1943 Mohr 1943

* Habitats: 1 = plant-herbivore-predator (18 cases); 2 = Agricultural crops (8); 3 = Gallers on trees (25); 4 = Leaf miners (3); 5 = Decaying wood (1); 6 = Dung (3). Note: Each web contained only one resource. "Identification" lists either the host plant, host/prey, or other feature that identifies this web within the original paper.

. . . N, j = 1 . . . N, where N is the number of samples), I used Pearson's correlation coefficients (r). Let robserved and rnuu be values of r from observed and null distributions, respectively. I randomly selected one web per study and calculated robserved, and repeated this for all 172,800 possible combinations of thirteen webs to obtain a mean value of robserved. One value of r nuii P e r combination of thirteen independent webs was calculated using randomly paired JC, and yj values. The proportion of rnuU values that had greater absolute values than the mean value of robserved is P in a two-tailed test (Manly 1991). As a test that is more sensitive to small differences in the correlation coefficient, I also tested whether the proportion of values of rObserved a n d rnuU that were positive differed using a chi-square test.

192

CHAPTER TWELVE

Results The fifty-eight source webs contained between three and twenty-four species. Contrary to expectations, there was no evidence that webs where a greater proportion of the natural enemies were parasitoids contained a smaller number of species. The mean correlation between the proportion of enemies that were parasitoids and the number of species was actually positive (figure 12.2; mean robserved = 0.107), but did not differ from the null value in a randomization test (P = 0.394). However, 87.6% of the observed values of this correlation were positive, compared to only 50.0% of values in the null distribution (P < 0.0001 in a chi-square test). This suggests that there is a small, but significant, positive correlation between numbers of species and the proportion of enemies that are parasitoids. There was no evidence that the proportion of enemies that are parasitoids was correlated with modal food chain length (figure 12.3); the mean correlation was only 0.03 (P = 0.92 in a randomization test) and there were similar numbers of positive and negative correlations in the observed and null distributions (49.999% of observed values were positive, compared to 50.0% of null values, P > 0.99 in a chi-square test). As the proportion of enemies that were parasitoids changed, there were some interesting changes in the shape of food webs. Mean correlations between numbers of species within a trophic level and proportion of enemies

24

a

20

SPECIE

CO

16 a

12 8 4

a •

a

4

•

•

•

0.2

0.4

»

« \

i

:•

r>

0

i

0.6

0.8

1 case 2 cases 3 cases 4 cases 5 cases 8 cases

PROPORTION PARASITOIDS Figure 12.2. The relationship between the proportion of enemies that were parasitoids and the number of species in source webs. See text for statistics.

PARASITOID-DOMINATED

o.z CD

z LJJ Z X

o DAL .FOO

Q

193

FOOD WEBS

0

o

o o

•

4.8 4.4 4

•

A

•>•

•

3.6

Ol/M

3.2 O Q

•

0

o

e

o

0.2 0.4 0.6 0.8 PROPORTION PARASITOIDS

^

1

1 case 2 cases 4 cases 10 cases 12 cases 13 cases

Figure 12.3. The relationship between modal food chain length (see figure 12.1 and Schoenly et al. 1991 for a definition) and the proportion of natural enemies that were parasitoids within source webs. that were parasitoids were not significant at P < 0.1 in randomization tests for any trophic level. However, more sensitive randomization tests, which looked at differences in the proportion of correlations that were positive in null and observed distributions of correlations, showed some complex changes: As the proportion of enemies that were parasitoids increased, there were lower numbers of species at trophic levels three, four, and five, but greater numbers of species at trophic levels six and seven (P < 0.0001 in chi-square tests in all cases; mean correlations were —0.05, —0.19, —0.15, + 0.28, and + 0.34, respectively). Thus, as the proportion of parasitoids in source webs increased, species were lost at trophic levels three to five and gained at trophic levels six and seven. These effects might be caused by competition or other mechanisms (see Hawkins, chapter 13). The changes might be expected to produce an increase in the modal food chain length in webs where more of the enemies were parasitoids; however, no such increase was seen, presumably because modal food chain length and numbers of species within trophic levels are measuring different things.

Discussion and Future Directions Webs with many parasitoids are expected to have shorter food chains if food chain length is limited by population fluctuations being larger for specialists

194

CHAPTER TWELVE

than generalists (e.g., Redfearn and Pimm 1992), or because trophic specialization rules out the possibility of omnivory that may stabilize food webs (McCann and Hastings 1997). On the contrary, I found weak statistical support for both the number of trophic levels and total number of species increasing with the proportion of enemies that were parasitoids; in tests of mean correlations there were no statistical differences, but there were changes in the proportions of positive and negative correlations that suggest weak statistical effects of the proportion of parasitoids on these factors. Because of the correlational nature of the observed patterns and the difficulties inherent in using data from multiple sources, the results need to be interpreted cautiously. The observed patterns might be explained by considering the way that feeding links are typically recorded for parasitoids and predators. For parasitoids, researchers typically dissect hosts or rear out parasitoids, so that it is easy to find all the parasitoids that feed on a particular host but much harder to find all the hosts for a particular parasitoid species (Shaw 1994). For predators, prey species are noted either by directly observing feeding events or through indirect methods, such as gut content analyses or immunological, isotopic, or chemical assays (van Dinther and Mensink 1971; Fry 1991). These techniques make it more likely that all prey species are identified for a particular predator than all predator species are found for a particular prey species. The webs used here either recorded predators by direct observation or inferred predation by the presence of both predators and prey in the same rotting wood, dung, or galls. These differences in sampling methods have several consequences: Predators are more likely to be recorded as generalists and parasitoids as specialists, and predators are more likely to be omitted from webs than are parasitoids. Sampling differences should lead to webs with more parasitoids having longer food chains, which is the opposite of hypothesis 1, but concurs with the observation that there were more species in higher trophic levels in webs with greater proportions of parasitoids. Inclusion of more parasitoids than predators in webs should also cause webs with more parasitoids to be more speciose. This is the opposite of hypothesis 2, but was found in the statistical analyses. The observed increases in total numbers of species in source webs and lack of a decline in the length of food chains in webs with greater proportions of parasitoids could be a result of the ways in which parasitoids and predators are sampled, or it might be because the webs with the most parasitoids are also dominated by herbivores that are gallers. Askew (1971) noted that gall chains frequently include serial parasitism and consumption of previous occupants by later arrivals (e.g., eupelmid, pteromalid, torymid, and chalcid wasps). Hawkins (chapter 13) also examines evidence that competition is important to the diversity of parasitoids that coexist. The present analyses found complex changes in the numbers of species at different tro-

PARASITOID-DOMINATED FOOD WEBS

195

phic levels, such that in food webs dominated by specialists or parasitoids, trophic levels three to five contained relatively fewer species and trophic levels six and seven contained relatively more species. The present analyses reveal a major gap in food web data. The differences in sampling methods of predators and parasitoids make it difficult to sensibly compare webs with insect predators versus parasitoids. The only way that such a comparison could be made with confidence would be to use an identical method for sampling both predators and parasitoids, such as direct observation of both feeding by predators and oviposition by parasitoids. It is also important that food web studies report a yield-effort curve that gives the cumulative number of species (or trophic links) on the ordinate and sampling effort on the abscissa (Cohen et al. 1993). A flattening of the curve at higher sampling efforts indicates that most of the species have been included, whereas a curve that continues to rise indicates that species have probably been missed out of the web. Multiple parasitism also presents problems for drawing food webs. Events within hosts are important for distinguishing hyperparasitism from competition among parasitoid larvae, making it necessary to either dissect hosts or rear out parasitoids. These questions are addressed in more detail by Brodeur (chapter 11). For predators, a further problem is distinguishing cases of incidental predation of parasitoid larvae, which would elevate the predator to a higher trophic level. There is no simple solution to the dilemma of distinguishing accurately what is occurring, versus recording predation and parasitism in comparable ways. Perhaps the best compromise would be to use host dissections and rearing to identify the kinds of interactions that occur within hosts and whether species are hyperparasitoids or competitors, but to include links and species in webs only when they are observed directly. Comparison of food webs and the search for patterns in food webs are valuable endeavors that have the potential to find biological features that cannot be found in any other way. Food web studies are appealing because of their ability to find patterns that are both very general and are either invariant of scale (total number of species in the web), or vary predictably with scale (e.g., Martinez 1993). Such analyses might reveal constraints and similarities among food webs that occur either in all ecosystems (Cohen 1978; Pimm 1982; Briand and Cohen 1984) or certain types of ecosystems (Martinez 1991). The connectance of food webs has potential links with food web stability (May 1973; Pimm 1980, 1982). More modern theory about food web stability (McCann et al. 1998) has shifted the emphasis from the qualitative proportion of possible links to the strengths of interactions, but this is still closely related to connectance. Connectance might also reflect biological details; for example, changes in connectivity as species are gained or lost at a trophic level might answer questions about shifts in trophic dynamics that result from factors like optimal foraging, diet shifts, and omni-

196

CHAPTER TWELVE

vory (for an applied perspective, see Roitberg, chapter 16). Despite its potential usefulness, connectivity has been sidestepped by many ecologists because of the difficulties of measuring it in food webs that were measured using small sample sizes. However, this is not to say that more accurate food webs would not find interesting patterns resulting from changes in connectivity (e.g., Martinez 1991). Certain other food web patterns, including patterns in food web architecture and dynamics with latitude or climate, are also unlikely to be testable using experiments. Cohen and colleagues (1993) call for more explicitness and more exhaustiveness in studies of food webs. Detail should be sufficient to enable any field study to be repeated. In the introduction, I reported the following requirements for field studies to be used to compare different food webs: (1) Taxa must be resolved down to a level where all food web entries have the same prey and predators (Paine 1988; Cohen et al. 1993). (2) What constitutes a trophic link must be clear (Cohen et al. 1993). (3) Sampling effort, sampling methods, the time of sampling, the spatial and temporal scale of sampling, or whether the web is cumulative or based on samples at a single point in time and space must all be clearly reported (Cohen et al. 1993). Furthermore, it is sensible to (4) clearly report any taxonomic groups that are knowingly excluded—e.g., birds and mammals from insect webs, or microorganisms (Cohen et al. 1993); (5) clearly report the setting (latitude, longitude, how boundaries were set, physical volume if relevant, whether boundaries shifted through space and time) and surroundings (continuous habitat or isolated fragment) (Cohen et al. 1993); (6) include a matrix or table with columns as consumers and rows as prey species, or when the total number of species is large, sublists of prey species for each predator, to facilitate unambiguous interpretation of findings (Cohen et al. 1993); and (7) search for and explicitly report omnivory, cannibalism, and resources crossing habitat boundaries (Polis and Strong 1996). Comparative analyses of food webs are not the only approach to addressing questions about food webs. The vast majority of questions can be addressed by careful experimental manipulation combined with the use of the same statistics that are used in comparisons of food webs. There is a rich array of such manipulations in the literature, but few have been conducted in parasitoid-dominated systems. Fagan (1997) demonstrated how knowledge of a small food web could be combined with experimental manipulation of omnivory to investigate the effects of omnivory on food web stability. Wootton (1994) showed that path analysis and experimental manipulation are powerful tools for predicting and testing how changes in the abundance of a key species in a food web are likely to influence the abundance of other food web members. Another approach that is likely to be particularly useful for investigating how food web architecture influences the dynamics of food webs is to select food web modules for experimentation and modeling, such

PARASITOID-DOMINATED FOOD WEBS

197

as intraguild predation, apparent competition, or linear food chains (e.g., Holt 1993; McCann et al. 1998). Ultimately however, to test the generality of experiments conducted using small numbers of species and in more controlled settings, we will need either larger-scale experiments or comparative analyses of food webs of the kind discussed here. Acknowledgements. I wish to thank Brad Hawkins, Tony Ives, Sharon Lawler, and an anonymous referee for comments. This work was supported by NSF DEB 96-29876 to MH and Susan Harrison.

Thirteen Species Coexistence in Parasitoid Communities: Does Competition Matter? BRADFORD A. HAWKINS

is generated over evolutionary time, but its maintenance depends on processes operating in ecological time. Understanding these processes represents the core of community ecology. For several decades, ecologists concentrated on the influence of localized biotic processes (e.g., competition and predation) on communities, but recent developments have re-emphasized the need for incorporating historical and large-scale processes into community ecology, particularly with respect to understanding local patterns of diversity (Ricklefs 1987; Cornell and Lawton 1992; Ricklefs and Schluter 1993). A common form of this argument stresses the need to quantify the balance between local and regional processes, and it is widely appreciated that identifying the relative roles of local and regional processes in determining the size and structure of communities is of fundamental importance for development of the field (Cornell 1985a, 1993; Compton et al. 1989; Lewinsohn 1991; Schluter and Ricklefs 1993; Underwood and Petraitis 1993; Brown 1995; Roland, chapter 7). After this is accomplished, a reasonable expectation is the conceptual unification of biogeography and community ecology. In the 1960s and 1970s, the problem of species coexistence was largely addressed in terms of local interspecific competition. But as most ecologists are aware, the competition paradigm has been seriously undermined in the past twenty years for animals, and particularly for phytophagous insect communities (Lawton and Strong 1981; Strong et al. 1984a; but see Denno and colleagues 1995 for a more recent review of competition among herbivores). On the other hand, the extent of competition in parasitoid communities remains unresolved (Askew and Shaw 1986; Godfray 1994). This is not because parasitoid workers are slower to accept challenges to established ideas than ecologists working on other groups. Rather, the belief that competition may be widespread exists because parasitoids have evolved a range of physiological and behavioral attributes that suggest that it is common (Salt 1961), the host-stage specificity of many species ensures that they will encounter each other in host individuals (Mills 1994c), and laboratory experiments for

BIODIVERSITY

DOES COMPETITION MATTER?

199

competition in parasitoid complexes usually find it (Force 1970, 1974; Leveque et al. 1993; van Alebeek et al. 1993; Wen et al. 1994; Wen and Brower 1995). Interest also persists because of the implications for biological control, and an extensive literature deals with the potential of interspecific competition to disrupt the regulation of pest populations (e.g., Waage and Mills 1992; Mills 1994c; Ehler 1994). Biological control practice has also produced several cases of competitive exclusion among parasitoids, the best understood being the case of Aphytis species attacking California red scale (DeBach and Sundby 1963; Luck and Podoler 1985; Murdoch et al. 1996a). Thus, there is little doubt that the potential for strong competition exists and parasitoid introductions can influence species coexistence. My goal in this chapter is to evaluate the role of interspecific competition in structuring parasitoid communities, with an emphasis on species coexistence. Following Wiens (1989b), I will distinguish the intensity of competition (its proximate effects on individuals and populations) and its importance (the ultimate consequences on community composition). The evidence that parasitoids sometimes compete fiercely for hosts is quite strong; whether or not this competition limits species coexistence is another matter. My discussion is focused on the effects of either interference or exploitative competition arising from the mortality that parasitoid larvae inflict on hosts. By interference competition, I mean competition due to encounters among larvae sharing a host individual; by exploitative competition, I mean either preemptive competition when one parasitoid attacks and/or kills a host individual at an earlier stage than another parasitoid, or situations where earlyattacking parasitoids reduce host densities low enough to limit the ability of later-attacking parasitoids to find hosts. Adult parasitoids may also compete via mutual interference among females searching for hosts. However, most of this work is theoretical or in laboratory experiments, and there is some question as to its applicability to field situations (Godfray 1994). There are also doubts that such behavioral mechanisms would be expected to strongly influence parasitoid coexistence, so I will not consider them here.

Evidence of Competition It has long been realized that parasitoids make excellent subjects for studies of competition. Whereas documenting prey usage by predators can be difficult and labor-intensive, parasitoid biologies couple them tightly to their host resources, and it is relatively easy to document interspecific interactions occurring within or on individual hosts (when at least one of the species is an idiobiont; e.g., Hawkins and Goeden 1984). Thus, it is often not necessary to conduct detailed manipulative experiments to document levels of competition in the field. Indeed, when the host is sessile and leaves telltale evidence

200

CHAPTER THIRTEEN

of its presence, as do scale insects, leaf miners, and gall formers, it is often necessary to only pop a scale cover or dissect the gall to determine directly the outcome of interactions between parasitoid larvae. It is also often possible to determine the intrinsic competitive abilities of parasitoids a priori, simply based on their biologies. For example, koinobionts (which permit a prolonged period of host growth and development after parasitization) are almost always competitively inferior to idiobionts (which kill or paralyze the host immediately) when the latter act as facultative hyperparasitoids (Varley 1947; Parnell 1964; Askew 1975; Hawkins and Goeden 1984; Tscharntke 1992; Shaw 1994; Stiling and Rossi 1994), and facultative hyperparasitoids provide some of the clearest evidence that individual parasitoids compete for hosts, in the guise of intraguild predation (i.e., facultative hyperparasitoids, by definition, eat their competitors). Rosenheim and colleagues (1995) recently reviewed examples of intraguild predation among parasitoids, highlighting its potential effects on biological control. Because idiobionts are generally strongly represented in the parasitoid assemblages associated with endophytic hosts, it is not surprising that much of the evidence for interference competition among parasitoids comes from these systems. For example, in his classic study of the population dynamics of the knapweed gallfly, Varley (1947) found that approximately 25% of the larvae of the koinobiont Eurytoma curta were subsequently killed by several idiobiont species. Further, because few of these facultative hyperparasitoids distinguish between healthy and previously parasitized host larvae, there was also substantial competition among them; for example, the idiobiont Macroneura vesciularis suffered 68% mortality due to interactions with other idiobiont species. There is little doubt that interspecific competition is intense in this system. Although generally harder to document, interference competition among koinobionts also occurs. For example, Eichhorn (1996) released gypsy moth larvae in an Austrian forest supporting a low density of moths, which he subsequently recovered over the course of the summer to measure parasitism. Based on larval dissections, he found that average multiparasitism varied from 2.2% to 25%, and in one sample reached 44% of larvae. Eichhorn also cited work in Germany on an outbreaking gypsy moth population, in which multiparasitism by the tachinids Parasetigena silvestris and Blepharipa pratensis reached 92%. A somewhat unusual example of multiparasitism was reported by Kaneko (1995), who found not only that multiparasitism of the scale Nipponaclerda biwakoensis was common (at least 24%), but that two parasitoid species, and rarely three, successfully emerged from multiparasitized hosts. All five of the parasitoid species involved were gregarious, and the number of adults emerging from hosts was lower when a species co-occurred with another species than when they occurred alone. Although this study demonstrates

DOES COMPETITION MATTER?

201

that competition does not necessarily always generate a winner and a loser, it does provide another clear example in which parasitoids commonly compete in the field. An alternative approach to quantifying competition has compared parasitism rates by individual species against rates when they co-occur with other species. Ehler (1979) used this method for the parasitoids of the gall midge Rhopalomyia californica and found that in these multilocular galls, parasitism rates by the strictly primary idiobiont Torymus koebelei fell from 54% in galls when it occurred alone to 45% when it occurred with one other species, and 33% when two other species were present. Similarly, parasitism by the primary koinobiont Platygaster sp. fell from 32% when alone in galls to 10% when co-occurring with two other species. Only the idiobiont Zatropis capitis was not influenced by the presence of other species, which Ehler presumed was because it is a facultative hyperparasitoid. Despite the examples outlined above that certainly indicate that parasitoids compete for hosts in the field, not all studies that have looked for competition have found it. Dean and Ricklefs (1979) analyzed parasitoid records from the Canadian Forest Insect Survey and concluded that parasitism rates among parasitoid species were independent of those by other species, concluding that the parasitoids of forest Lepidoptera do not compete for hosts. This study attracted strong criticism from Force (1980) and Bouton and colleagues (1980), and a defensive reply by Dean and Ricklefs (1980). Given the questionable appropriateness of the data set to address the question, however, little can be concluded from this study. Even so, Dean and Ricklefs (1980) made one comment that requires mention. They argued that competition may be present when host utilization rates are high but not when they are low, as was the case for most of the host species in their data set. (Force 1989 provides an example of this straightforward idea in the parasitoid complex of the seed-feeding moth Mesepiola specca in California.) However, this would be true only if all host individuals are equally accessible to parasitoids. If a large proportion of the host population is immune from attack, parasitoids could compete strongly for the relatively few vulnerable hosts, even though overall parasitism rates are low. Variability in susceptibility among potential host individuals is probably extremely common in nature. In endophytic parasitoid-host systems, this can arise from host individuals residing in structural refuges, and in exophytic systems, it can arise from large numbers of hosts being too small or in an invulnerable developmental stage (see Luck 1990 for a discussion of this from theoretical and behavioral perspectives), or from variability in levels of chemical defenses among host individuals. Whereas very high parasitism rates probably indicate that competition is occurring, low parasitism rates are not prima facie evidence that it is not. It also follows that the depression of host populations is not necessary for parasitoids to compete. If most of the

202

CHAPTER THIRTEEN

potential hosts are inaccessible to attack, parasitoids must share the few available hosts and could suffer intense competition with minimal impact on the host's population. An additional twist to evaluating competition among parasitoids is that they not only may have to compete among themselves, but may compete with other types of natural enemies as well. Although long recognized, this has attracted concerted study only relatively recently (Hochberg and Lawton 1990b; Hochberg et al. 1990), and evidence is rapidly accumulating that competition among parasitoids, predators, and pathogens is widespread (e.g., Rosenheim et al. 1995; Tscharntke 1997). Obviously, this complicates the issue somewhat, but only reinforces the potential importance of competition in the third trophic level, since it expands the potential number of competitive interactions with which parasitoids must contend. All the above examples concern interference competition, where the mechanism involves direct contact among competing larvae. Strong exploitative competition among parasitoids has also been documented. Luck and Podoler (1985) demonstrated that the competitive exclusion of Aphytis lingnanensis by A. melinus is due to the ability of A. melinus to utilize smaller scale individuals than can A. lingnanensis, thereby severely limiting the number of hosts that will grow large enough to be available to the latter species. However, this example is entirely synthetic (the plant, host, and parasitoids are all exotic), and I am aware of no examples where levels of exploitative competition have been documented in more natural communities, whether for a parasitoid attacking earlier host stages or for a generally more efficient species that reduces host densities to a level below which a less-efficient species can be maintained. Thus, although it seems sensible to assume that a parasitoid that decimates a host population could have severe impacts on the success of other parasitoids searching for very rare hosts (see Price 1975 for a perspective of how temporal exploitative competition influences parasitoid fecundities), there is very limited direct evidence of this in natural systems. This does not seriously compromise the otherwise good evidence that parasitoids often must compete for hosts, and the handful of examples outlined above is illustrative rather than exhaustive. It is likely that many workers who study multispecies parasitoid assemblages in the field can provide additional examples showing that interspecific competition occurs at some level.

Competition and Species Coexistence Given that interspecific competition does commonly occur, what is its contribution to the problem of species coexistence? Here, the evidence is more equivocal. As already mentioned, there is excellent evidence that competi-

DOES COMPETITION MATTER?

203

tive exclusion has occurred as a consequence of parasitoid introductions for biological control (Luck and Podoler 1985 and references therein). Even so, the ecological significance of these examples is something about which reasonable people can disagree. Traditionally, both biological control specialists and ecologists have assumed that classical biological control (the introduction of exotic enemies to control exotic pests) represents ecological experiments on a grand scale, and that the lessons learned from such introductions tell us about how parasitoid-host systems work in more natural situations (e.g., DeBach 1964a; Strong et al. 1984a; Hawkins 1994). On the other hand, a recent analysis of insect life tables and the biological control record (Hawkins et al. 1999) suggests that classical biological control overestimates the importance of parasitoids to host population control, for two reasons. First, the biological control record indicates that it is much more successful when conducted in simplified habitats based on exotic plants (as in most crop fields and orchards) than when attempted against pests feeding on native plants in diverse habitats (most forests), suggesting that the simplification of both food chains and the connections to nontarget habitats facilitate top-down control by parasitoids. Second, life tables identify parasitoids as key mortality factors in biological control systems, whereas in natural systems, generalist predators are key. Although it is beyond the scope of this chapter to address this issue in detail, these results suggest that classical biological control does not in fact reflect the nature and strength of parasitoid-host interactions in many natural systems. If this is true, then competitive exclusion by biological control agents, albeit a real phenomenon, may provide only limited insight into the role of competition in the coexistence of parasitoids. Until this issue is resolved, we should be cautious about using the biological control record to evaluate the factors that influence the size and structure of parasitoid communities. If we exclude the "experimental" evidence provided by biological control practice, we are left with less compelling types of evidence. This is necessarily so, since few people would advocate introducing parasitoids into natural systems where the host is not a serious pest. It is also logistically difficult to experimentally exclude particular parasitoids from native systems to see what impact they have on the dynamics of the system. Consequently, the only studies of which I am aware are either correlative or comparative. Given the limitations of such data, what do they tell us about the role of competition in species coexistence? Hawkins and Goeden (1984), in a study of the parasitoid complexes associated with gall-forming Asphondylia spp. (Diptera: Cecidomyiidae) in southern California, found the eulophid Paragaleopsomyia cecidobroter ( = Tetrastichus cecidobroter) acting as a keystone species in two of the eight gall systems studied. For example, the abundances of P. cecidobroter over one year were negatively correlated with the number of parasitoid spe-

204

CHAPTER THIRTEEN CO

• Palm Desert (rs = -0.736, PO.001) o Valle Vista (rs == -0.517, P>0.05)

CD O CD Q.

o

•

CO

O

•

•4—»

o

•

CO

CD Q.

CD

0*

O

•o

O

o

•

•» 8 o

•

•

o

o o

o

E 20

40

60

80

100

Percent "parasitism" by Paragaleopsomyia Figure 13.1. Association between "parasitism" rates by Paragaleopsomyia cecidobroter (proportion of galler larval cavities attacked) and the number of co-occurring galler parasitoid species in samples of Asphondylia atriplicis galls on Atriplex canescens, at two southern California sites. rs = Spearman rank correlation coefficient. cies at a desert site in galls of Asphondylia atriplicis, although the effect was apparent only at very high P. cecidobroter attack rates (figure 13.1, Palm Desert). This suggests that competitive exclusion may be occurring at least some of the time in this local community. Even so, the relationship was not significant at a chaparral site, although the correlation coefficient was also negative (figure 13.1, Valle Vista). There are at least three relevant points about these data. First, being correlative, the causal relationship between the abundance of P. cecidobroter and the presence of other species cannot be unambiguously established. Second, even if competitive exclusion is occurring, it appears to be very local and perhaps transient. Third, the keystone species is not a parasitoid. Paragaleopsomyia is a phytophagous inquiline that forms galls around the larval chambers of the Asphondylia (Hawkins and Goeden 1982). During formation of this gall, the midge larval cavity is obliterated, physically crushing the midge and any associated parasitoid larvae. Thus, this is not a classic case of competition, since P. cecidobroter and the parasitoids feed on different trophic levels. On the other hand, dynamically, it is indistinguishable from con-

DOES COMPETITION MATTER?

205

ventional competition. The message of this study is that there is some evidence that the diversity of the parasitoid community is constrained by the actions of a co-occurring species, but it is not clear how widespread such strong interactions are, even within this system. Other studies to test whether or not competition among parasitoids is sufficiently strong to limit the number of species that can coexist have used a comparative approach. As noted in the introduction, it is now widely realized that constraints on species coexistence may not depend on local processes, such as competition, but instead reflect the outcome of forces acting at the regional level (see Ricklefs and Schluter 1993 for a discussion of this point of view and numerous examples of its application). At one logical extreme, local communities may be strongly interactive, and local biotic processes place hard limits on the number of species that can coexist. Alternatively, local biotic interactions may be feeble, and community richness depends solely on the size of the regional pool of species that are available to colonize local habitats. A technique to test these extremes regresses the species richness of local communities against the species richness of regional pools to determine if the relationship is linear or asymptotic (Cornell and Lawton 1992). A linear relationship indicates that regional processes dominate and local interactions are insufficiently strong to limit species coexistence (i.e., communities are not saturated with species), whereas an asymptotic relationship is consistent with hard limits on community size due to local competition (i.e., communities may be saturated; but see caveats below). This technique has been most extensively used by insect ecologists to examine phytophagous insect communities, and in all cases, workers have concluded that local communities are unsaturated (Cornell 1985a,b; Zwolfer 1987; Lewinsohn 1991; Compton and Hawkins 1992; Lawton et al. 1993). Care must be taken when generating and interpreting local-regional richness regressions (see Srivastava 1999). There are several methodological pitfalls, but the solutions to these problems are relatively straightforward (Wiens 1989a; Hawkins and Compton 1992; Cresswell et al. 1995). More important, the test is asymmetrical. A linear relationship between local and regional richness represents reasonable evidence that local interactions are insufficiently intense to cause competitive exclusion, as long as the resources on which the communities are based can potentially support equal numbers of species. On the other hand, curvilinear relationships can arise for reasons unrelated to saturation, such as stochastic equilibria that balance colonization and extinction, or habitat heterogeneity (Cornell 1993). Thus, when asymptotic relationships are found, additional evidence is required to judge whether this actually reflects local saturation. A final comment on the technique is that it should be clear that it is most convincing when it is resourcebased, as opposed to being habitat-based. Because the underlying mechanism is presumed to be competition for one or more specific resources, esti-

206

CHAPTER THIRTEEN

mating local community richness in terms of those resources (e.g., in the case of parasitoids, the number of parasitoid species per host species) requires fewer assumptions about how limiting those resources might be than would estimating richness in terms of some habitat characteristic or type in which resource levels are not quantified. To date, local-regional richness regressions have been done in only two host-parasitoid systems. Hawkins and Compton (1992) analyzed the parasitoids associated with the fruit-inhabiting chalcidoids of fifteen species of figs (Ficus spp.) in southern Africa, and found little compelling evidence that the communities are saturated. However, this study suffered from the problem mentioned above. Because of the scope of the project, individual rearings of hosts and parasitoids were not possible, so parasitoid richness was measured as the number of species per fig (the habitat) rather than the number per galler species (the host). To partially overcome this problem, we adjusted the richness estimates by the number of potential hosts in a gall. The resulting relationship between local and regional richness (the number of parasitoid species per galler species per fig species) was nonasymptotic (figure 13.2A; see the original paper for the statistical methodology used to test for nonlinearities), suggesting that the richness of local parasitoid complexes is due more to the number of potential coexisting species than to local constraints. Because of the uncertainties inherent to the method we used, we also examined variation in local parasitoid richness on individual trees and in individual fruits. At both spatial scales, most fruits and trees support far fewer parasitoids than they are capable of supporting (e.g., many fruits and trees supported no parasitoids at all, even though potential hosts were present). Based on the local-regional regressions and the low and variable levels of parasitoid representation in local sites, we concluded that interspecific competition to the point of competitive exclusion is highly unlikely in this system. The second study compared local and regional species richness in the parasitoids associated with grass-infesting chalcidoids in Great Britain. Using an extremely well-sampled system of fifteen Tetramesa spp. (Eurytomidae) on ten grass species, Dawah and coworkers (1995) also found a very strong linear relationship between local and regional parasitoid species richness, expressed as the number of parasitoid species per host species (figure 13.2B). Moreover, the relationship differed in one respect when compared to other examples of local-regional regressions. In all cases studied to date, local communities have represented subsets of regional species pools, a process of local community assembly referred to as "proportional sampling" (Cornell and Lawton 1992). In 7e?rame.ra-parasitoid communities, however, local communities comprise every parasitoid species occurring in Britain, with a few exceptions due to undersampling (Dawah et al. 1995). The general conclusion of this study was that variation in the number of locally

207

DOES COMPETITION MATTER?

CO CD C

y = 0.291 + 0.397x R2 = 0.735

1 2 3 Regional species richness

2 4 6 8 Regional species richness

10

Figure 13.2. Relationships between regional and mean ( ± 1 SE) local parasitoid species richness for (A) agaonid hosts in 15 species of southern African figs, and (B) 14 Tetramesa host species in ten British grasses. Both regressions are highly significant (P < 0.001).

208

CHAPTER THIRTEEN

coexisting parasitoid species depends exclusively on the number of species composing the regional pool, and there was no evidence that competition or other local processes restrict the local recruitment of parasitoids. This leaves open the question of what restricts regional pool size, but competition is not the cause. The conclusion that competition does not constrain the richness of Tetramesa-parasitoid communities in Britain has also been strengthened by an analysis of similar systems in Germany. T. Tscharntke and I (unpublished data) compared the species richness and host utilization patterns of Tetramesa parasitoids sampled near Karlsruhe and found that most British Tetramesa apparently support far fewer parasitoids than they are capable of supporting (table 13.1). For example, the mean number ( ± 1 SE) of parasitoid species per Tetramesa species was 4.80 ± 0.587 in Britain and 9.76 ± TABLE 13.1 Number of Parasitoids Known from Tetramesa Species in Great Britain and Germany In Germany

In Britain

Tetramesa sp.

Total

Mono

Poly

Total

Mono

Poly

fulvicollis longula longicornis eximia calamagmstidis linearis hyalipennis new sp. sp. 3 agrostidis schmidti sp. 1 phragmitis cornuta brevicornis brevicollis angustipennis petiolata airae albomaculata pratense

13 17 9 17 17 9 3 8 5 11 4 6 8

1 1 2 9 4 1 2 1 4 5 1 2 7

6 10 7 8 13 8 1 7 1 6 3 4 1

4 3 6 9 8 6 4

4 2 5 7 5 5 3

0 1 1 2 3 1 1

2 5 6 7 5 4 1 2

2 4 4 5 5 3 1 2

0 1 2 2 0 1 0 0

Note: The total parasitoid complexes have also been separated into their monophagous (mono = restricted to one known host species) and polyphagous (poly = more than one known host species) components.

DOES COMPETITION MATTER?

209

1.369 in Germany. The difference was even greater among host species occurring in both places; 5.71 ± 0.837 in Britain and 12.14 ± 2.040 in Germany. Most of the difference between the areas was due to a loss of more generalized species in Britain; across all Tetramesa species there were 3.80 ± 0.416 monophagous species in Britain versus 4.00 ± 0.768 in Germany, whereas there were 1.00 ± 0.239 polyphagous species in Britain versus 5.77 ± 1.02 in Germany. The densities of hosts were very similar in both areas, so interregional variation in host availability cannot explain the large differences in species richness of the parasitoids. These data indicate that most of the parasitoid communities in Britain are impoverished relative to continental communities due to a loss of more generalized species (possibly due to a species-area effect). Of concern here is that Tetramesa are clearly capable of supporting richer parasitoid complexes than they do in Britain, making it unlikely that competitive exclusion is operating within these systems. We do not have sufficient data to examine the possibility that the richer German systems, on the other hand, are limited by competition. Before moving on to another approach that has been used to test for saturation of parasitoid communities, two comments are needed regarding the results outlined above. First, two cases are insufficient to judge the extent of saturation in the large number of parasitoid communities that actually exist. Second, both tests used "atypical" systems. Fig-wasp parasitoid communities are inordinately depauparate compared to those normally associated with tree-galling hosts and, despite some taxonomic uncertainties, are probably dominated by specialist parasitoids (Hawkins and Compton 1992). Similarly, the parasitoids of grass-infesting Tetramesa are mostly specialists, at least in Britain. Specialist communities are expected to be less structured by competition than are communities of generalists (see, for example, Price 1984; Holyoak, chapter 12), so before any general conclusions can be reached, it is necessary to examine communities that are both highly diverse and dominated by generalists. An a priori expectation that generalist-dominated parasitoid communities should be more strongly structured by competition arises because generalists are often idiobionts and facultatively hyperparasitic, and levels of hyperparasitism can be very high (Askew 1961; Ehler 1992; Stiling and Rossi 1994; Brodeur, chapter 11). This form of intraguild predation represents an extreme form of competition, in which the hyperparasitoids are unambiguously superior to their nonhyperparasitic primary-parasitoid hosts. Therefore, such analyses for other types of parasitoid communities are sorely needed before the competitive exclusion hypothesis can be summarily rejected. A more general attempt to evaluate whether or not competition may limit species coexistence has been done by classifying parasitoids into guilds. Variability in guild representation between and within host species provides a tool to quantify the extent that resources are underused and to determine if

210

CHAPTER THIRTEEN

community richness is not constrained by interspecific interactions (Lawton and Price 1979; Lawton 1982; Lawton et al. 1993). Hawkins and Mills (1996) conducted a literature-based comparative analysis using the parasitoid guilds associated with 381 host species and with 389 populations of 66 host species to identify the extent and distribution of "vacant niches," both between closely related hosts and between local populations within host species. We found that a minimum of one-third of operationally defined host niches are not being used by parasitoids, and concluded that parasitoid communities are not likely to be saturated in many cases. We also found that the representation of parasitoid guilds among different host taxa was idiosyncratic and depended more on the taxonomic identity of hosts than on their biologies. This suggests that the functional structure of parasitoid communities is unpredictable in any general sense and reflects largely the outcomes of chance historical events and patterns of parasitoid diversification. This conclusion is very similar to that reached in a study of the distribution of herbivore guilds on British trees (Cornell and Kahn 1989; Cornell 1993). Thus, the local-regional species richness regressions and the analyses of guild structure done so far for both herbivore and parasitoid communities are consistent with the notion that competitive exclusion is uncommon or absent in insect communities at either the second or the third trophic level. In sum, there is little—if any—evidence that competition plays a widespread role in the coexistence of parasitoids in natural systems, although biological control practice clearly demonstrates that it is possible. On the other hand, the data are limited, so it is premature to claim that local processes have no role in species coexistence in parasitoid communities. Although few workers claim to have too much data, that additional studies are still needed, particularly for speciose complexes containing large numbers of facultative hyperparasitoids (see Brodeur, chapter 11). If competition matters, it is in these systems where it should be most evident.

If Not Competition, Then What? If we take the existing data at face value, we must conclude that competition is not a major constraint on the species richness of many host-parasitoid systems. If this is true, then what does permit coexistence among organisms that, due to their biologies, must compete for individual hosts? An obvious alternative hypothesis is spatial and temporal heterogeneity in host resources. As is apparent from other chapters in this book (Bernstein, chapter 4; Mills, chapter 14), the role of heterogeneity in host-parasitoid interactions has received a great deal of attention, although much of it is on how spatial heterogeneity stabilizes host-parasitoid population dynamics (see also Ives 1995); its role in species coexistence is less often discussed. Hochberg and Hawkins

DOES COMPETITION

MATTER?

211

(1992, 1993, 1994) used a population-dynamic model to investigate the factors that permit coexistence in multispecies parasitoid complexes. The main focus of the model was on proportional refuges provided to hosts, but we also found that spatial heterogeneity in parasitoid attack rates could contribute to the maintenance of species-rich assemblages by reducing competitive exclusion. Of course, refuges themselves represent a source of heterogeneity in the susceptibility of potential hosts to parasitoid attack, and the richest parasitoid complexes are possible when both types of heterogeneity are operating in concert. Within the constructs of the model the lesson is clear: Variability in parasitoid-host encounter rates is essential for maintaining model parasitoid communities at levels of richness comparable to those found in nature. Using real parasitoids rather than digital ones, Tscharntke (1992) argued that spatial and temporal heterogeneity accounts for the coexistence of fourteen species of parasitoids associated with the gall midge Giraudiella inclusa on the common reed, Phragmites australis, in Germany. Although an expansive monoculture of reed may appear homogeneous to the human eye, Tscharntke found that the various species of specialist parasitoids concentrate their attacks in different microhabitats, based on the size and location of the reed-shoots supporting galls, the size of the gall cluster (local host density), and different host generations. Because of this resource partitioning, each parasitoid species has to commonly compete with, at most, one other species, presumably reducing the probability of competitive exclusion. If heterogeneity in resource use can operate in such structurally simple habitats, it is even more likely to be important in more diverse and complex habitats. As in all such studies that examine resource partitioning among coevolved parasitoid complexes (see also Price 1970 and Hawkins and Goeden 1984), it is possible that strong competition was the driving force behind the patterns identified by Tscharntke (1992), the famous "ghost of competition past" (Connell 1980). Because of this, such studies cannot provide critical tests of the role of competition in community structure. However, they do show that parasitoids are able to respond to heterogeneity in the environment and provide evidence that it contributes to the maintenance of rich parasitoid complexes.

Discussion and Future Directions In closing, the answer to the question posed in the title—does competition matter?—has to be: probably not. However, it is premature to completely dismiss the potential importance of competition among parasitoids. I see the way forward encompassing several approaches. First, comparisons of re-

212

CHAPTER THIRTEEN

gional versus local patterns of diversity represent a relatively simple methodology for identifying systems in which local processes may be playing a role. When such regressions are asymptotic, more detailed work is warranted to determine what sets the maximum number of species. Conversely, strong linear relationships such as those found so far provide little rationale for investing the time and money needed to conduct manipulative experiments. An important limitation of the comparative method is that it requires information on a large number of closely related systems, which may not be possible when hosts belong to unusual or small taxonomic groups. Irrespective of this, ultimately determining whether or not competitive exclusion is occurring requires experiments in which particular species are excluded from local systems, to see how other species respond. Unfortunately, given the extreme mobility of most parasitoids and similar (small) body sizes, such selective exclusions are difficult to execute. Thus, the success of such experiments depends largely on the creativity of the experimenter. An alternative method, introducing a parasitoid into a system where it does not normally occur, is difficult to justify since any changes made to the system are likely to be permanent and irreversible. Perhaps the least objectionable form of this approach occurs when different combinations of parasitoid species coexist in different local host populations. In these cases, parasitoids can be moved between local populations without severely damaging the community as a whole. This might be most appropriate when host populations have become fragmented and isolated due to changes in land use, and the loss of parasitoid species in local patches due to the fragmentation process can, in a sense, be reversed. If the scale of the "experiment" is large enough, this may allow us to generate unique combinations of parasitoid species on local populations of a geographically widespread host that never coexisted previously. Although this idea is simple in theory, I do not know if it is that simple in practice. Finally, serendipity can provide valuable data. The introduction of parasitoids for biological control will continue for the foreseeable future, and despite our best efforts, an introduced agent may move onto local nontargets that support their own parasitoid complexes. If we are sensitive to this possibility and the invasion process can be discovered early, any changes in the invaded system can be documented. If we find that such introductions do result in the loss of species in the nontarget system, it will not bode well for the continued practice of classical biological control, but it will tell us much about parasitoid community structure. I do not mean to advocate using this approach, but it would be a shame not to use such data if the opportunity arises. Although it is not particularly fashionable at the moment to discuss competitive exclusion in community ecology, this does not make the problem any less interesting or less important. It seems to me that until we can an-

DOES COMPETITION MATTER?

213

swer this question with some degree of confidence, we are going to be constrained about what we can really say concerning how multiple parasitoid species sharing a host resource interact in the field. Whether interspecific competition has a major or trivial role in the diversity of parasitoids is yet to be determined, but I, for one, would like to know which it is.

Fourteen Biological Control: The Need for Realistic Models and Experimental Approaches to Parasitoid Introductions NICK MILLS

biological control, or the introduction of exotic natural enemies for the control of invading pests, has captured the attention of ecologists throughout the century. The apparent simplicity with which an exotic pest population can be reduced dramatically by the introduction of an exotic natural enemy is intriguing, and has challenged ecologists to provide an equally simple explanation of how biological control works. The early successes in classical biological control (henceforth referred to simply as biological control) encouraged ecologists to study population phenomena, and, in particular, the influence of trophic interactions on the equilibrium density and persistence of insect populations. Such studies have generated an impressive body of theory on predator-prey interactions, but a simple explanation for the phenomenon of biological control has proved to be a more elusive goal. Although biological control is concerned with introductions of both predators and parasitoids for the control of insect pests, the vast majority of introductions and successes have been restricted to insect parasitoids (75%), due primarily to their greater level of specificity to the target pest (Greathead and Greathead 1992). As parasitoids require only a single host to complete thendevelopment, this greatly simplifies the relationship between hosts attacked and parasitoid progeny produced, and provides an ideal model system for both theoretical and experimental analysis of biological control. As a result, biological control has been conceptualized as a simple host-parasitoid system that is relatively isolated and closed from the rest of the environment, as both pest and control agent are exotic species with few linkages to the surrounding natural community. The historical record of biological control contains numerous examples of the successful control of insect pests through introductions of one of more parasitoids. These examples provide us with some of the best evidence in ecology that top-down trophic interactions can reduce the average abundance of herbivore populations, albeit from managed rather than from natural ecosystems. However, the historical record also provides us with ample eviCLASSICAL

218

CHAPTER FOURTEEN

dence that the introduction of parasitoids does not always result in the successful control of the target pest. So why is it that parasitoid introductions can, on occasion, be spectacularly successful in reducing the abundance of a pest, but more generally have had little impact on pests? Can we improve the success of biological control by identifying the most susceptible target hosts or by selecting the most effective parasitoid species? Biological control provides an ideal opportunity to combine theory and practice in the advancement of population ecology and the development of environmentally benign solutions for the management of insect pests that threaten our natural resources. In this chapter, I will review the more pertinent ecological theory on host-parasitoid interactions (see Bernstein, chapter 4, for a more detailed discussion), examine the empirical patterns of parasitoid success in biological control, and highlight some promising directions for future research. Before proceeding further, however, it is important to distinguish between two facets of biological control that will be used to structure the rest of the chapter: establishment—the colonization of a new environment by an introduced parasitoid; and impact—the reduction of pest population abundance by the action of an established parasitoid.

Parasitoid Establishment Theoretical Framework There has been considerable recent interest in ecological invasions (Kareiva 1996a). The invasion and colonization of novel environments has both demographic and genetic components in population biology. From a demographic viewpoint, colonization is dependent on population growth. The simplest description of population growth in a continuously reproducing population is the familiar logistic model: dN/dt = rN(l - N/K)

(14.1)

where N is the abundance of the population, K is the carrying capacity of the environment that sets the equilibrium abundance of the population, and r is the instantaneous per capita net rate of increase of the population. Consequently, the success of population establishment in a novel environment is dependent on the availability of resources (K) and the per capita growth rate (r). From a biological control perspective, there is generally no shortage of hosts as a resource for an introduced parasitoid and, thus, reproduction and survival determine population growth and the success of establishment. However, this very simple deterministic model of population growth ignores a critical feature of ecological invasions—namely, the viability of small founder populations. The success of a small founder population may

BIOLOGICAL CONTROL

219

be limited by an "Allee effect," stochastic variation, or a genetic bottleneck. An Allee effect is a negative population growth rate at low population abundance due, for example, to the difficulty of finding mates. The importance of Allee effects has been widely discussed (Dennis 1989; Viet and Lewis 1996; McCarthy 1997), and its relevance to biological control has been considered by Hopper and Roush (1993). Following Philip (1957), the success of mating, as an Allee effect, can be represented as a process of random search for mates, such that the probability of a female escaping a mating, given by e~sN, with s the mating efficiency, can be incorporated into the logistic model as: dN/dt = rN(l - N/K) - bNe~sN

(14.2)

where b is the instantaneous per capita birth rate. This modified logistic model poses two constraints to successful colonization: a minimum carrying capacity Kmin (the positive root of ln(bsK/r) = sK — 1), and a minimum initial population size Nmin = [ln(b/r)]/s as K -» °°. Thus, the success of colonization increases with initial population size, per capita growth rate (r), and mating success (5). Incorporating the same expression for mating success into a reaction-diffusion model of parasitoid population growth and dispersal, Hopper and Roush (1993) showed that the success of parasitoid colonization not only depends on initial population size and population growth rate, but is also increased by a greater mate detection distance and proximity between individuals. It is interesting to note from this latter model that haplodiploidy, by which virgin female parasitoids (Hymenoptera only) can still produce male offspring rather than none at all, permits successful establishment in parasitoid populations that are 30% smaller in initial population size than corresponding diplodiploid populations. Stochastic variation could also limit the ability of a small founder population to colonize a novel environment. Shaffer (1981) discussed three types of stochastic variation: demographic stochasticity (the random variation between individuals in their birth and death rates), environmental stochasticity (the random variation in birth and death rates that affects all individuals simultaneously), and catastrophes (large random reductions in population size). In the context of biological control, both demographic and environmental stochasticity will affect small founder parasitoid populations, whereas random catastrophes are probably too rare to be of importance in colonization events. Using a discrete logistic population model, Ludwig (1996) demonstrated that the success of colonization increases with initial population size, population growth rate, and a reduction in the variance of environmental stochasticity. From their analysis of a host-parasitoid metapopulation model, Wilson and Hassell (1997) have also shown that demographic stochasticity increases the probability of extinction of small parasitoid populations, and that under such circumstances, greater dispersal rates are re-

220

CHAPTER FOURTEEN

quired to ensure the persistence of the parasitoid population. From a review of the dynamics of dispersal in models of invading populations, Hastings (1996) has noted that rates of dispersal are proportional to net rates of population increase, once again highlighting the importance of population growth rate to the success of parasitoid colonization. Almost by definition, small founder populations must exhibit reduced genetic variation in relation to their source populations, but it is not clear whether such a decrease would limit the potential of a parasitoid population to colonize a novel environment. Although genetic bottlenecks have been considered a potential reason for failures in the establishment of natural enemies (Messenger et al. 1976; Hagvar 1991), rapid population growth following colonization reduces the severity of genetic loss (Nei et al. 1975). It is also more generally believed that genetic revolutions or mutation, drift, and selection acting on founder populations could enhance adaptation to novel environments (Barton and Charlesworth 1984; Carson 1990; Hopper et al. 1993). In fact, the successful colonization of a novel environment can be viewed as the balance of two processes: one demographic, by which population growth rates of less than unity lead to the extinction of founder populations, and the other evolutionary, by which natural selection increases population growth rates to rescue founder populations from extinction. Using coupled models of population growth and natural selection, Gomulkiewicz and Holt (1995) were able to examine the balance of demography and evolution in greater detail. Although simplistic, these models provide the valuable insight that a founder population is likely to avoid extinction only when sufficiently large and sufficiently pre-adapted to the novel environment.

Empirical Evidence The overall rate of establishment of parasitoids in biological control introductions has been estimated to be 34-38% (Hall and Ehler 1979; Stiling 1990; Mills 1994a). Despite the importance of population growth rate as a factor in theoretical models, Stiling (1990) was unable to find a clear relationship between the rate of establishment of parasitoids and their fecundity, voltinism, or offspring per host. This is somewhat surprising, as both Crawley (1986), in the context of biological weed control, and Lawton and Brown (1986), in the context of accidental introductions of insects into the British Isles, found the probability of establishment to increase with characteristics associated with population growth rate (small body size, high fecundity, and short generation time). Similarly, Veltman and colleagues (1996) found population growth rate to be associated with the success of establishment of exotic birds in New Zealand.

BIOLOGICAL

I

221

CONTROL

0.8

0.6'

I 0.4 i CD

0.2-

03

(0 T3

CD (0 2

(DO) (0 (0 TJ "O

Q.

.2

(0

T3 C

£

>.

(0

tB

1

S 2

Q

8 2 o

Q.

ffl

.=

(0

2

T3 3

CO

o

o c (U

a> cu

a)

(0

11

CL

0.6

B

0.4'

CD

cu

i—

cu

0) (0

O

O

CU

(0 T3 !|

(0 >-

1

o

iaspidi

2

f. See also agricultural landscapes

364 Nicholson, A. J., 4, 41, 42, 43 Nicholson-Bailey model: egg limitation and, 105-6, 115, 117-18; on functional responses, 43; on host-parasitoid interaction, 42; instability of, 132, 133; local stability condition for, 49; parasitoid interference introduced into, 44-47; parasitoid population dynamics and, 4, 27, 128. See also host-parasitoid interactions Nipponaclerda biwakoensis, 200 nontarget organism evolutionary potential, 260-64. See also biological control Okubo, A., 70, 72 Olesicampe geniculate spread, 78/—79, 81 Ophiomyia phaseoli, 233 Orgyia vetusta (tussock moth) outbreaks, 6 2 68

Ormia depleta spread, 77-78, 81 Orr, H. A., 127, 137 Oscinella frit, 239 Pacala, S. W., 50, 51, 53, 129 Pachyneuron californicum, 174 Pachyneuron concolor, 179 Paragaleopsomyia cecidobroter, 203-4/ Parasetigena silvestris, 200 parasitoid clutch size: classes of variation in, 27; coexistence and, 31-34, 32/ 33/; conclusions on attack rate and, 35-40; impact on mean host density, 36/ 38/ impact on parasitoid progeny, 31; larval fighting and, 154-55; model parameters of, 27-28; trade-off between attack rate and, 28—31, 29/ See also parasitoid egg load parasitoid community ecology, 295-99 parasitoid conservation: case study of Ichneumon eumerus and Maculinea rebeli, 275-76; efficiency used in, 270-72; implications of studies on, 276-77; maximum growth rate (MGR) used in, 269-72; of parasitoid ensembles, 274; reasons to support, 267-68; of spatially extended populations, 272-73 parasitoid defense strategies, 122-23 parasitoid-dominated food webs: community ecology on, 296-97; description of, 18485; future direction of studies on, 193-97; hypotheses on, 185-87; methods used in study of, 187-91; relationship between enemy and source web species, 192/ rela-

INDEX tionship between food chain and enemies within, 193/ results of study on, 192-93; source food webs used in study of, 190«911; source web example, 189/ See also parasitoid food chains parasitoid egg load: absence of competition and, 106-8/ 107/ implications of studies on, 117-20; life-history evolution and embryonic traits of, 145-54; NicholsonBailey model on, 105-6, 115, 117-18; overcoming host defenses and, 109, 118; overview of, 103—6; population dynamical feedback and, 113-17; taxa containing endoparasitoids and cleavage in, 152/-53/ within-host competition and fitness of, 109, 110/ 111-13, 112/ See also parasitoid clutch size parasitoid embryonic traits, 145-54 parasitoid establishment: biological control and rate of, 220-21/ biological control success and, 230; problems involved in, 12 parasitoid food chains: hypothesis testing on, 10, 185-87; trophic pathway of hyperparasitoid, 177, 179-80. See also parasitoid-dominated food webs parasitoid foraging behavior: field vs. laboratory experiments on, 24-25; future research directions on, 25-26; host abundance sampling rules and, 17-19; imperfect cues for, 19-21; parameters identified for, 17; sources of variability in, 2 1 23; time available for, 23-24; time-limited parasitoids and theory of, 103 parasitoid-host interactions. See hostparasitoid interactions parasitoid life-history: body size/longevity and, 143-44; development of traits during, 144/-45; dynamic life-history models on, 257-60; embryonic traits and evolution of, 145-54; future studies on hyperparasitoid, 180-81; host-parasitoid interaction and, 139; Hymenoptera phylogeny and, 1404 1 / implications of studies on, 160-62; larval development and, 154—56; optimality approach to, 139-40, 160-61; risks of conserved size thresholds and, 156-60; traditional approaches to, 140, 142-43, 145 parasitoid population biology: biocontrol and, 291-95; described, 3; impact on dynamic patterns by, 4-7, 281-82; perspective and

365

INDEX applications of, 11-14; population diversity and, 7-11 parasitoid population dynamics: community ecology and, 295-99; diversity of, 7—11; environmental stochasticity in, 279—80, 281, 288; evolution and, 288-91; GodfrayHassell model on, 128-29; Hassell-Anderson model on, 129, 130; HochbergHawkins model on, 210-11; host resistance/parasitoid virulence and, 123-27; implications of studies on, 135-38; as indicators in applied areas, 282; landscape ecology and, 299-301; linking behavior and, 285—88; maximum growth rate (MGR) and, 269-72; models of genetical components on, 130-35; molecular techniques used in, 301-3; Nicholson-Bailey model on, 4, 27, 128; overview of, 4-7, 278-82; parasitoid egg load and, 113-17; population biology and, 4-7, 281-82; Sasaki-Godfray model on, 133—34, 136, 137; spatial heterogeneity and, 282-85. See also biological control; hostparasitoid interactions parasitoid resistance: between-clone variability in, 125; implications of studies on, 135-38; isofemale lines and, 125—26, 127; uncoupled genetics models on, 127-30 parasitoids: defense strategies of, 122-23; defining, 139; egg-limited and time-limited subdivisions of, 103—5; ensemble levels of, 274; foraging field research on, 17; generalized trends in traits of, 144/-45; host abundance and, 17-19; imperfect foraging cues and, 19-21; as insect system regulators, 284; landscape ecology of, 8 3 99; mechanisms of extinction for, 266-67. See also hyperparasitoids; idiobionts; koinobionts parasitoid spatial spread: of Bracon phylacteophagus, 79; of Eurytoma serratulae, 80, 81; integro-difference models on, 73-76; landscape approach to, 84; models of, 72; of Olesicampe geniculate, 78/-79, 81; of Ormia depleta, 77-78, 81; reaction-diffusion models on, 72-76; semivariograms on autocorrelation pattern in, 96/; studies on, 70-72, 80-82; of Torymus beneficus, 76; of Torymus sinensis, 16-11/, 81. See also landscape ecology

parasitoid virulence: coevolution of resistance and, 136-37; Godfray-Hassell model on, 128-29; models of genetical components on, 130-35; Nicholson-Bailey model on, 128; parasitoid population dynamics and, 123-27 pasture weevil control, 224 Patelloa pachypyga, 91, 92, 93, 94/, 95, 96/, 97 Patelloa pluriseriata, 63 pea aphid {Acyrthosiphon pisum), 125 pest management, 98-99. See also biological control; landscape ecology Phacelia tanacetifolia, 247 Phaenoglyphis villosa, 171, 172 Phenacoccus manihoti (cassava mealybug), 233 Phillips, C. B., 224 Phragmites australis, 211 Phylacteophaga froggatti, 79 phylogeny: as host range determinant, 170; Hymenoptera, 140-41/ Pieris rapae, 37 Pimentel, D., 47, 48 Pimm, S. L., 185 Planococcus citri, 111 Podoler, H., 202 Poisson distribution, 42, 49, 111, 119, 128 population biological theory, 268-69, 27071 Praon pequodorum, 280 predator-prey system: grass varieties and ratio in, 240/; pattern formation in, 60 predators: biological control programs and, 222-23; impact of landscape changes on, 236; relationship between food chain and food web, 193/; relationship between source web species and, 192/; sampling methods to identify, 194 Price, P. W., 3, 143 Prochiloneurus aegyptiacus, 177, 179 Prochiloneurus spp., 177 Protodejeania echinata, 63 Pseudaletia unipunctata (armyworm), 251 pseudohyperparasitoids, 165 Quednau, F. W., 79 Quicke, D.L.J., 3, 145, 163 random amplified polymorphic DNA (RAPD), 302

366 rape pollen beetle (Meligethes aeneus), 244/45, 250/-51, 252 reaction-diffusion models: described, 58-60; on parasitoid spatial spread, 7 2 76; on tussock moth population, 64, 6 5 66 Redfearn, A., 185 resource-consumer revolutionary models, 130 Rhopalomyia californica, 201 Ribautodelphax albostriatus, 239 Ricklefs, R. E., 201 Rogers, D. J., 44 Rohani, P., 53 Roitberg, B. D., 13, 263, 277 Roland, J., 7, 13, 300 Rosen, D., 163, 165, 179 Rosenheim, J. A., 40, 104, 118, 200, 242 Roughgarden, J., 60 Roush, R. T., 219, 223 Ruxton, G. D., 45, 46, 47 Sasaki, A., 133-34, 136, 138 Sasaki-Godfray model, 133-34, 136, 137 Schoenly, K., 186, 188 Seger, L. A., 130 set-asides of crop fields, 247-48/ See also agricultural landscapes Settle, W. H., 227 Sevenster, J. G., 104, 118, 119 Shaffer, M. L., 219 Shaw, M. R., 142, 165, 170, 174 Shea, K., 105 Sheehan, W., 3, 14 Simberloff, D., 61, 255, 264 Skellam, J. G., 70, 71, 72 SLOSS (single large or several small) debate, 243 "Space—the final frontier in theoretical ecology" (Kareiva), 283 spatial heterogeneity: models of, 47-50; population dynamics and, 282-85 species coexistence: competition paradigm of, 198-99; contribution of competition to, 202-10 Stewart-Oaten, A., 51, 52, 53 Stiling, P., 220, 222, 224, 229, 232, 255, 264 Stillman, R. A., 47 Strand, M. R., 9-10 Sutherland, W. J., 49 Sympiesis sericeicornis, 20

INDEX Tachinomyia similis, 63 Taylor, A. D., 39, 51 Telenomus californicus Ashmead, 63 tent caterpillar (Malacosoma disstria), 87, 91 7e/rame.sa-parasitoid communities (Britain/ Germany), 206-9, 207/ 208f Telramesa sp., 239 Tetrastichus howardi, 182 Thompson, W. R., 41-42 time-limited parasitoids: egg loads of, 103-4; foraging behavior and fitness of, 118 time scales, 53-54 Torymus beneficus spread, 76 Torymus koebelei, 201 Torymus sinensis spread, 76-77/ 81 transgenic crop development, 292-93 Tscharntke, T., 12-13,208,211,242,243,298 Turing, A. M., 72 tussock moth (Orgyia vetusta) outbreaks, 62-68 type I functional responses, 45 type II errors (B), 254-56 type II functional responses, 52 type III functional responses, 43, 52, 54 Unruh, T. R., 225 Urophora cardui L., 80 Urtica dioica (nettle stands), 243-46/ See also agricultural landscapes van Baalen, M , 8, 134, 164, 262, 272 van den Bosch, R, 73 Van der Meer, J., 47 Varley, G. C , 44, 47, 200 Vaugh, T. T., 302 Veltman, C. J., 220 Venturia caenescens, 19, 158 Via, S., 125 Vikberg, V, 173 Volkl, W., 183 Waage, J. K., 3, 19 Walker, T. J., 126 Watt, K.E.F., 185 Weinberger, H. K, 73 Wharton, R., 162 Williams, E. J., 42 Williamson, M., 222 Wilson, H. B., 64, 219 Wootton, J. T., 196 World Conservation Union, 14 Zatropis capitis, 201