Proteomics for Biomarker Discovery: Methods and Protocols [1st ed.] 978-1-4939-9163-1, 978-1-4939-9164-8

Table of contents :
Front Matter ....Pages i-xii
Pre- and Post-analytical Factors in Biomarker Discovery (Frank Klont, Peter Horvatovich, Natalia Govorukhina, Rainer Bischoff)....Pages 1-22
Pre-fractionation of Noncirculating Biological Fluids to Improve Discovery of Clinically Relevant Protein Biomarkers (Annarita Farina)....Pages 23-37
Serum Exosome Isolation by Size-Exclusion Chromatography for the Discovery and Validation of Preeclampsia-Associated Biomarkers (Rosana Navajas, Fernando J. Corrales, Alberto Paradela)....Pages 39-50
Protein Biomarker Discovery Using Human Blood Plasma Microparticles (Raghda Saad Zaghloul Taleb, Pacint Moez, Doreen Younan, Martin Eisenacher, Matthias Tenbusch, Barbara Sitek et al.)....Pages 51-64
A Standardized and Reproducible Proteomics Protocol for Bottom-Up Quantitative Analysis of Protein Samples Using SP3 and Mass Spectrometry (Christopher S. Hughes, Poul H. Sorensen, Gregg B. Morin)....Pages 65-87
Analyzing Cerebrospinal Fluid Proteomes to Characterize Central Nervous System Disorders: A Highly Automated Mass Spectrometry-Based Pipeline for Biomarker Discovery (Antonio Núñez Galindo, Charlotte Macron, Ornella Cominetti, Loïc Dayon)....Pages 89-112
Lys-C/Trypsin Tandem-Digestion Protocol for Gel-Free Proteomic Analysis of Colon Biopsies (Armin Schniers, Yvonne Pasing, Terkel Hansen)....Pages 113-122
Tube-Gel: A Fast and Effective Sample Preparation Method for High-Throughput Quantitative Proteomics (Leslie Muller, Luc Fornecker, Sarah Cianferani, Christine Carapito)....Pages 123-127
Protein Biomarker Discovery in Non-depleted Serum by Spectral Library-Based Data-Independent Acquisition Mass Spectrometry (Alexandra Kraut, Mathilde Louwagie, Christophe Bruley, Christophe Masselon, Yohann Couté, Virginie Brun et al.)....Pages 129-150
Discovering Protein Biomarkers from Clinical Peripheral Blood Mononuclear Cells Using Data-Independent Acquisition Mass Spectrometry (Xin Ku, Wei Yan)....Pages 151-161
Intact Protein Analysis by LC-MS for Characterizing Biomarkers in Cerebrospinal Fluid (Jérôme Vialaret, Sylvain Lehmann, Christophe Hirtz)....Pages 163-172
Detection of Proteoforms Using Top-Down Mass Spectrometry and Diagnostic Ions (Didia Coelho Graça, Ralf Hartmer, Wolfgang Jabs, Alexander Scherl, Lorella Clerici, Kaveh Samii et al.)....Pages 173-183
Development of a Highly Multiplexed SRM Assay for Biomarker Discovery in Formalin-Fixed Paraffin-Embedded Tissues (Carine Steiner, Pierre Lescuyer, Jean-Christophe Tille, Paul Cutler, Axel Ducret)....Pages 185-203
Development and Validation of Multiple Reaction Monitoring (MRM) Assays for Clinical Applications (Georgia Kontostathi, Manousos Makridakis, Vasiliki Bitsika, Nikolaos Tsolakos, Antonia Vlahou, Jerome Zoidakis)....Pages 205-223
Protein-Level Statistical Analysis of Quantitative Label-Free Proteomics Data with ProStaR (Samuel Wieczorek, Florence Combes, Hélène Borges, Thomas Burger)....Pages 225-246
Computation and Selection of Optimal Biomarker Combinations by Integrative ROC Analysis Using CombiROC (Mauro Bombaci, Riccardo L. Rossi)....Pages 247-259
PanelomiX for the Combination of Biomarkers (Xavier Robin)....Pages 261-273
Designing an In Silico Strategy to Select Tissue-Leakage Biomarkers Using the Galaxy Framework (Lien Nguyen, Virginie Brun, Florence Combes, Valentin Loux, Yves Vandenbrouck)....Pages 275-289
Back Matter ....Pages 291-292

Citation preview

Methods in Molecular Biology 1959

Virginie Brun Yohann Couté Editors

Proteomics for Biomarker Discovery Methods and Protocols

METHODS

IN

MOLECULAR BIOLOGY

Series Editor John M. Walker School of Life and Medical Sciences University of Hertfordshire Hatfield, Hertfordshire, AL10 9AB, UK

For further volumes: http://www.springer.com/series/7651

Proteomics for Biomarker Discovery Methods and Protocols

Edited by

Virginie Brun and Yohann Couté Université Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France

Editors Virginie Brun Universite´ Grenoble Alpes CEA, Inserm, BGE U1038 Grenoble, France

Yohann Coute´ Universite´ Grenoble Alpes CEA, Inserm, BGE U1038 Grenoble, France

ISSN 1064-3745 ISSN 1940-6029 (electronic) Methods in Molecular Biology ISBN 978-1-4939-9163-1 ISBN 978-1-4939-9164-8 (eBook) https://doi.org/10.1007/978-1-4939-9164-8 Library of Congress Control Number: 2019933724 © Springer Science+Business Media, LLC, part of Springer Nature 2019 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Humana Press imprint is published by the registered company Springer Science+Business Media, LLC, part of Springer Nature. The registered company address is: 233 Spring Street, New York, NY 10013, U.S.A.

Preface Recent progress in the fields of instrumentation, pre-analytical and analytical methods, and data treatment has significantly improved the performance of mass spectrometry (MS)based discovery proteomics analysis and the range of its applications. In biomedical research, discovery proteomics analysis is extensively used to characterize cellular, tissue, and biological fluid proteomes to extract information advancing our understanding of pathogenic mechanisms and to identify new biomarkers for diagnostic, prognostic, or therapeutic follow-up purposes. Discovery proteomics approaches to identify new biomarker candidates involve the sequential application of several steps consisting in (1) preparing biological samples to make them compatible with the analytical constraints, (2) analyzing the samples by (liquid chromatography (LC) coupled to) MS, (3) extracting the data identifying and/or quantifying proteins, and (4) performing statistical analyses to select proteins based on their relative abundance in different conditions and/or the abundance of which can be used to class patients. To obtain useful and relevant results, each of these complex steps must be carefully controlled. This volume of Methods in Molecular Biology, entitled Proteomics for Biomarker Discovery: Methods and Protocols, presents a series of modern and optimized methods by which to perform proteomics analyses aiming to discover new biomarkers in the field of human health. Chapters 1 and 2 present returns on experience, detailing critical pre- and/or postanalytical factors and providing recommendations on ways to improve the reliability of proteomics experiments for biomarker discovery in circulating (serum, plasma, cerebrospinal fluid) and noncirculating (pulmonary epithelial lining fluid, bile, etc.) biological fluids. Chapters 3 and 4 focus on the search for biomarkers present in circulating extracellular vesicles and provide up-to-date efficient protocols for the preparation of samples composed of exosomes (Chapter 3) and microparticles (Chapter 4). In response to the need to analyze large sample cohorts, Chapters 5 and 6 describe complete pre-analytical and analytical pipelines which are particularly appropriate for use when performing multiplexed quantitative analysis of cellular and tissue samples (Chapter 5) or biological fluids such as cerebrospinal fluid (Chapter 6). In Chapters 7 and 8, novel protocols to specifically optimize protein digestion (a key step in discovery proteomics workflows) are presented. Chapters 9 and 10 deal with the development of “data-independent acquisition” (DIA) mass spectrometrybased analysis methods to explore blood-based samples (plasma and peripheral blood mononuclear cells, respectively). Most discovery proteomics analyses are performed using a “bottom-up” strategy involving protein digestion followed by LC-MS/MS analysis of the resulting peptide mixture. However, for a few years now, “top-down” analyses—consisting in analyzing proteins by MS without prior digestion—have been increasingly used with the aim of identifying molecular signatures associated with disease states. We wished to include a presentation of this new bioanalytical trend. Thus, Chapters 11 and 12 in this volume describe top-down proteomics analysis workflows applied to the search for biomarkers in Alzheimer’s disease (Chapter 11) and hemoglobin disorders (Chapter 12). Exploratory proteomics analyses by LC-MS/MS, or orthogonal approaches (transcriptomics, bioinformatics, literature mining, etc.), produce lists of candidate biomarkers, the specificity of which must generally be confirmed in independent sample cohorts. Although

v

vi

Preface

this confirmation can be effected by immunological assays, the high specificity and multiplexing capacity of targeted quantitative proteomics approaches are increasingly being used in this context. Chapter 13 in this volume presents the development of a highly multiplexed targeted proteomics analysis to quantify and assess biomarker candidates in formalin-fixed paraffin-embedded (FFPE) tissues. And Chapter 14 describes the regulatory requirements and experiments that must be performed to validate a targeted proteomics method for use in a medical biology setting. The final four chapters in this volume (Chapters 15, 16, 17, and 18) reflect the considerable evolution of the field of data science applied to proteomics. Chapter 15 describes the functionalities and modalities of the use of the open-source ProStaR tool to statistically analyze datasets generated by label-free quantitative proteomics analyses. Chapters 16 and 17 present the web-based PanelomiX and CombiROC tools which, in association with a clinical question, can help select optimal biomarker combinations from proteomics (or other) data input. Finally, Chapter 18 describes the use of a workflow developed in the Galaxy environment to select candidate biomarkers from publicly accessible proteomics data. Grenoble, France

Virginie Brun Yohann Coute´

Contents Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Contributors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

v ix

1 Pre- and Post-analytical Factors in Biomarker Discovery . . . . . . . . . . . . . . . . . . . . . Frank Klont, Peter Horvatovich, Natalia Govorukhina, and Rainer Bischoff 2 Pre-fractionation of Noncirculating Biological Fluids to Improve Discovery of Clinically Relevant Protein Biomarkers . . . . . . . . . . . . . . Annarita Farina 3 Serum Exosome Isolation by Size-Exclusion Chromatography for the Discovery and Validation of Preeclampsia-Associated Biomarkers. . . . . . . Rosana Navajas, Fernando J. Corrales, and Alberto Paradela 4 Protein Biomarker Discovery Using Human Blood Plasma Microparticles . . . . . Raghda Saad Zaghloul Taleb, Pacint Moez, Doreen Younan, Martin Eisenacher, Matthias Tenbusch, Barbara Sitek, and Thilo Bracht 5 A Standardized and Reproducible Proteomics Protocol for Bottom-Up Quantitative Analysis of Protein Samples Using SP3 and Mass Spectrometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Christopher S. Hughes, Poul H. Sorensen, and Gregg B. Morin 6 Analyzing Cerebrospinal Fluid Proteomes to Characterize Central Nervous System Disorders: A Highly Automated Mass Spectrometry-Based Pipeline for Biomarker Discovery. . . . . . . . . . . . . . . . . . ˜ ez Galindo, Charlotte Macron, Ornella Cominetti, Antonio Nu´n and Loı¨c Dayon 7 Lys-C/Trypsin Tandem-Digestion Protocol for Gel-Free Proteomic Analysis of Colon Biopsies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Armin Schniers, Yvonne Pasing, and Terkel Hansen 8 Tube-Gel: A Fast and Effective Sample Preparation Method for High-Throughput Quantitative Proteomics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Leslie Muller, Luc Fornecker, Sarah Cianferani, and Christine Carapito 9 Protein Biomarker Discovery in Non-depleted Serum by Spectral Library-Based Data-Independent Acquisition Mass Spectrometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Alexandra Kraut, Mathilde Louwagie, Christophe Bruley, Christophe Masselon, Yohann Coute´, Virginie Brun, and Anne-Marie Hesse 10 Discovering Protein Biomarkers from Clinical Peripheral Blood Mononuclear Cells Using Data-Independent Acquisition Mass Spectrometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Xin Ku and Wei Yan

1

vii

23

39 51

65

89

113

123

129

151

viii

11

12

13

14

15

16

17 18

Contents

Intact Protein Analysis by LC-MS for Characterizing Biomarkers in Cerebrospinal Fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Je´roˆme Vialaret, Sylvain Lehmann, and Christophe Hirtz Detection of Proteoforms Using Top-Down Mass Spectrometry and Diagnostic Ions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Didia Coelho Grac¸a, Ralf Hartmer, Wolfgang Jabs, Alexander Scherl, Lorella Clerici, Kaveh Samii, Yury O. Tsybin, Denis Hochstrasser, and Pierre Lescuyer Development of a Highly Multiplexed SRM Assay for Biomarker Discovery in Formalin-Fixed Paraffin-Embedded Tissues . . . . . . . . . . . . . . . . . . . . Carine Steiner, Pierre Lescuyer, Jean-Christophe Tille, Paul Cutler, and Axel Ducret Development and Validation of Multiple Reaction Monitoring (MRM) Assays for Clinical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Georgia Kontostathi, Manousos Makridakis, Vasiliki Bitsika, and Nikolaos Tsolakos, Antonia Vlahou, and Jerome Zoidakis Protein-Level Statistical Analysis of Quantitative Label-Free Proteomics Data with ProStaR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Samuel Wieczorek, Florence Combes, He´le`ne Borges, and Thomas Burger Computation and Selection of Optimal Biomarker Combinations by Integrative ROC Analysis Using CombiROC . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mauro Bombaci and Riccardo L. Rossi PanelomiX for the Combination of Biomarkers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Xavier Robin Designing an In Silico Strategy to Select Tissue-Leakage Biomarkers Using the Galaxy Framework. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lien Nguyen, Virginie Brun, Florence Combes, Valentin Loux, and Yves Vandenbrouck

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

163

173

185

205

225

247 261

275

291

Contributors RAINER BISCHOFF  Department of Analytical Biochemistry, Groningen Research Institute of Pharmacy, University of Groningen, Groningen, The Netherlands VASILIKI BITSIKA  Biomedical Research Foundation Academy of Athens, Athens, Greece MAURO BOMBACI  Translational Research Unit, Protein Arrays Lab, Istituto Nazionale Genetica Molecolare, Milan, Italy HE´LE`NE BORGES  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France THILO BRACHT  Medizinisches Proteom-Center, Ruhr-Universit€ at Bochum, Bochum, Germany CHRISTOPHE BRULEY  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France VIRGINIE BRUN  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France THOMAS BURGER  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France; CNRS, BIG-BGE, Grenoble, France CHRISTINE CARAPITO  Laboratoire de Spectrome´trie de Masse Bio-Organique (LSMBO), IPHC, UMR 7178, Universite´ de Strasbourg, CNRS, Strasbourg, France SARAH CIANFERANI  Laboratoire de Spectrome´trie de Masse Bio-Organique (LSMBO), IPHC, UMR 7178, Universite´ de Strasbourg, CNRS, Strasbourg, France LORELLA CLERICI  Laboratory Medicine and Pathology, Department of Genetic, Geneva University Hospitals, Geneva, Switzerland DIDIA COELHO GRAC¸A  Clinical Proteomics and Chemistry Group, Faculty of Medicine, University of Geneva, Geneva, Switzerland; Department of Genetic, Laboratory Medicine and Pathology, Geneva University Hospitals, Geneva, Switzerland FLORENCE COMBES  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France ORNELLA COMINETTI  Proteomics, Nestle´ Institute of Health Sciences, Lausanne, Switzerland FERNANDO J. CORRALES  Functional Proteomics Facility, Centro Nacional de Biotecnologı´a (CNB-CSIC), ProteoRed-ISCIII, Madrid, Spain YOHANN COUTE´  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France PAUL CUTLER  Biomarkers, Bioinformatics and Omics, Pharmaceutical Sciences, Roche Pharma Research & Early Development (pRED), Roche Innovation Center Basel, F. Hoffmann-La Roche AG, Basel, Switzerland; Translational Biomarkers and Bioanalysis, Development Sciences, UCB Pharma, Slough, UK LOI¨C DAYON  Proteomics, Nestle´ Institute of Health Sciences, Lausanne, Switzerland AXEL DUCRET  Biomarkers, Bioinformatics and Omics, Pharmaceutical Sciences, Roche Pharma Research & Early Development (pRED), Roche Innovation Center Basel, F. Hoffmann-La Roche AG, Basel, Switzerland MARTIN EISENACHER  Medizinisches Proteom-Center, Ruhr-Universit€ a t Bochum, Bochum, Germany ANNARITA FARINA  Digestive Cancers Biomarkers Group, Faculty of Medicine, Department of Internal Medicine, University of Geneva, Geneva, Switzerland LUC FORNECKER  Laboratoire de Spectrome´trie de Masse Bio-Organique (LSMBO), IPHC, UMR 7178, Universite´ de Strasbourg, CNRS, Strasbourg, France NATALIA GOVORUKHINA  Department of Analytical Biochemistry, Groningen Research Institute of Pharmacy, University of Groningen, Groningen, The Netherlands

ix

x

Contributors

TERKEL HANSEN  Natural Products and Medicinal Chemistry Research Group, Department of Pharmacy, UiT—The Arctic University of Norway, Tromsø, Norway RALF HARTMER  Bruker Daltonics, Bremen, Germany ANNE-MARIE HESSE  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France CHRISTOPHE HIRTZ  Clinical Proteomics Platform, LBPC, IRMB, CHU Montpellier, Montpellier University, Montpellier, France DENIS HOCHSTRASSER  Clinical Proteomics and Chemistry Group, Faculty of Medicine, University of Geneva, Geneva, Switzerland; Laboratory Medicine and Pathology, Department of Genetic, Geneva University Hospitals, Geneva, Switzerland PETER HORVATOVICH  Department of Analytical Biochemistry, Groningen Research Institute of Pharmacy, University of Groningen, Groningen, The Netherlands CHRISTOPHER S. HUGHES  Canada’s Michael Smith Genome Sciences Centre, British Columbia Cancer Agency, Vancouver, BC, Canada WOLFGANG JABS  Beuth Hochschule fu¨r Technik Berlin, Berlin, Germany FRANK KLONT  Department of Analytical Biochemistry, Groningen Research Institute of Pharmacy, University of Groningen, Groningen, The Netherlands GEORGIA KONTOSTATHI  Biomedical Research Foundation Academy of Athens, Athens, Greece ALEXANDRA KRAUT  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France XIN KU  Key Laboratory of Systems Biomedicine (Ministry of Education), Shanghai Centre for Systems Biomedicine, Shanghai Jiao Tong University, Shanghai, People’s Republic of China SYLVAIN LEHMANN  Clinical Proteomics Platform, LBPC, IRMB, CHU Montpellier, Montpellier University, Montpellier, France PIERRE LESCUYER  Department of Genetic, Laboratory Medicine and Pathology, Geneva University Hospitals, Geneva, Switzerland; Clinical Proteomics and Chemistry Group, Faculty of Medicine, University of Geneva, Geneva, Switzerland MATHILDE LOUWAGIE  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France VALENTIN LOUX  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France; INRA, MAIAGE Unit, University Paris-Saclay, Jouy-en-Josas, France CHARLOTTE MACRON  Proteomics, Nestle´ Institute of Health Sciences, Lausanne, Switzerland MANOUSOS MAKRIDAKIS  Biomedical Research Foundation Academy of Athens, Athens, Greece CHRISTOPHE MASSELON  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France PACINT MOEZ  Clinical and Chemical Pathology Department, Faculty of Medicine, Alexandria University, Alexandria, Egypt GREGG B. MORIN  Canada’s Michael Smith Genome Sciences Centre, British Columbia Cancer Agency, Vancouver, BC, Canada; Department of Medical Genetics, University of British Columbia, Vancouver, BC, Canada LESLIE MULLER  Laboratoire de Spectrome´trie de Masse Bio-Organique (LSMBO), IPHC, UMR 7178, Universite´ de Strasbourg, CNRS, Strasbourg, France ANTONIO NU´N˜EZ GALINDO  Proteomics, Nestle´ Institute of Health Sciences, Nestle´ Research, Lausanne, Switzerland

Contributors

xi

ROSANA NAVAJAS  Functional Proteomics Facility, Centro Nacional de Biotecnologı´a (CNBCSIC), ProteoRed-ISCIII, Madrid, Spain LIEN NGUYEN  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France; INRA, MAIAGE Unit, University Paris-Saclay, Jouy-en-Josas, France ALBERTO PARADELA  Functional Proteomics Facility, Centro Nacional de Biotecnologı´a (CNB-CSIC), ProteoRed-ISCIII, Madrid, Spain YVONNE PASING  Tromsø Endocrine Research Group, Department of Clinical Medicine, UiT—The Arctic University of Norway, Tromsø, Norway; Division of Internal Medicine, University Hospital of North Norway, Tromsø, Norway XAVIER ROBIN  Swiss Institute of Bioinformatics, University of Basel, Biozentrum, Basel, Switzerland RICCARDO L. ROSSI  Bioinformatics, Istituto Nazionale Genetica Molecolare, Milan, Italy KAVEH SAMII  Department of Genetic, Laboratory Medicine and Pathology, Geneva University Hospitals, Geneva, Switzerland; Division of Hematology, Geneva University Hospitals, Geneva, Switzerland ALEXANDER SCHERL  Clinical Proteomics and Chemistry Group, Faculty of Medicine, University of Geneva, Geneva, Switzerland; Department of Genetic, Laboratory Medicine and Pathology, Geneva University Hospitals, Geneva, Switzerland ARMIN SCHNIERS  Natural Products and Medicinal Chemistry Research Group, Department of Pharmacy, UiT—The Arctic University of Norway, Tromsø, Norway BARBARA SITEK  Medizinisches Proteom-Center, Ruhr-Universit€ a t Bochum, Bochum, Germany POUL H. SORENSEN  Department of Molecular Oncology, British Columbia Cancer Research Centre, Vancouver, BC, Canada CARINE STEINER  Division of Laboratory Medicine, Geneva University Hospitals, Geneva, Switzerland; Biomarkers, Bioinformatics and Omics, Pharmaceutical Sciences, Roche Pharma Research & Early Development (pRED), Roche Innovation Center Basel, F. Hoffmann-La Roche AG, Basel, Switzerland; Late Stage Analytical Development, Small Molecules Technical Development, Roche Innovation Center Basel, F. Hoffmann-La Roche AG, Basel, Switzerland RAGHDA SAAD ZAGHLOUL TALEB  Medizinisches Proteom-Center, Ruhr-Universit€ a t Bochum, Bochum, Germany; Clinical and Chemical Pathology, Faculty of Medicine, Alexandria University, Alexandria, Egypt MATTHIAS TENBUSCH  Institute of Clinical and Molecular Virology, University Hospital Erlangen, Friedrich-Alexander University Erlangen-Nu¨rnberg, Erlangen, Germany JEAN-CHRISTOPHE TILLE  Division of Clinical Pathology, Geneva University Hospitals, Geneva, Switzerland NIKOLAOS TSOLAKOS  ProtATonce, NCSR Demokritos, Athens, Greece YURY O. TSYBIN  Spectroswiss, Lausanne, Switzerland YVES VANDENBROUCK  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France ´ JEROˆME VIALARET  Clinical Proteomics Platform, LBPC, IRMB, CHU Montpellier, Montpellier University, Montpellier, France ANTONIA VLAHOU  Biomedical Research Foundation Academy of Athens, Athens, Greece SAMUEL WIECZOREK  Universite´ Grenoble Alpes, CEA, Inserm, BGE U1038, Grenoble, France

xii

Contributors

WEI YAN  Key Laboratory of Systems Biomedicine (Ministry of Education), Shanghai Centre for Systems Biomedicine, Shanghai Jiao Tong University, Shanghai, People’s Republic of China DOREEN YOUNAN  Clinical and Chemical Pathology, Faculty of Medicine, Alexandria University, Alexandria, Egypt JEROME ZOIDAKIS  Biomedical Research Foundation Academy of Athens, Athens, Greece

Chapter 1 Pre- and Post-analytical Factors in Biomarker Discovery Frank Klont, Peter Horvatovich, Natalia Govorukhina, and Rainer Bischoff Abstract The translation of promising biomarkers, which were identified in biomarker discovery experiments, to clinical assays is one of the key challenges in present-day proteomics research. Many so-called “biomarker candidates” fail to progress beyond the discovery phase, and much emphasis is placed on pre- and postanalytical variability in an attempt to provide explanations for this bottleneck in the biomarker development pipeline. With respect to such variability, there is a large number of pre- and post-analytical factors which may impact the outcomes of proteomics experiments and thus necessitate tight control. This chapter highlights some of these factors and provides guidance for addressing them on the basis of examples from previously published proteomics studies. Key words Biomarker, Liquid chromatography, Mass spectrometry, Pre-analytical variability, Postanalytical variability, Proteomics, Regulated bioanalysis, Sample preparation, Stability

1

Introduction The interest toward adopting indicators of the physiological state of a human being into clinical practice did not emerge in recent times as physicians millennia ago already adopted such markers of disease. Physicians of the Hippocratic School, for example, recognized specific tumors as Karkinos, the Greek word for crab, as corresponding swellings and ulcers resembled the shape of a crab with its claw-like projections [1, 2]. These same physicians, and Galen of Pergamon in particular, furthermore practiced medicine based on the theory of “humorism” which proposes the clustering (and clinical assessment, accordingly) of illnesses in terms of excesses and deficiencies of the four bodily fluids, so-called humors (i.e., blood, yellow bile, black bile, and phlegm, which were associated with the heart, liver, spleen, and brain, respectively) [3]. Humoral medicine even continued to be one of the central principles of (Western) medical practice until the mid-nineteenth century, after which the humoral theory was abruptly abandoned

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_1, © Springer Science+Business Media, LLC, part of Springer Nature 2019

1

2

Frank Klont et al.

and medicine could start to mature in times of numerous scientific discoveries and inventions [2, 3]. Science and technology have been key drivers of medical advancements ever since the fall of humoral medicine and greatly expanded our knowledge of human (patho)physiology and the treatment options of physicians. In the past decades, “Omics” technologies (e.g., genomics, proteomics, metabolomics) made their entrance in (bio)medical sciences and raised high expectations for discovery of new biomarkers [4]. Notable successes of genomics include BRCA1/BRCA2 susceptibility testing for breast and ovarian cancer, viral load testing for diagnosing and monitoring human immunodeficiency virus (HIV) infection, genotyping and subtyping of chronic hepatitis C virus (HCV) infections based on the viral genome as well as guiding HCV treatment through (host) ITPA and IL28A genotype testing [5, 6]. Success stories of “proteomic medicine” are less frequent, though the CKD273 biomarker panel, which received a Letter of Support from the United States Food and Drug Administration (FDA) in 2016 encouraging the use of this panel in the (early) management of chronic kidney disease [7], and the OVA1 in vitro diagnostic multivariate index assay (IVDMIA), which received FDA-clearance for assessing the risk of ovarian cancer in women presenting with pelvic masses, are noteworthy examples in this respect [8, 9]. The CKD273 panel is a classifier based on 273 urinary peptides which were identified and assessed using capillary electrophoresis coupled to mass spectrometry. The OVA1 test integrates the serum levels of cancer antigen 125 (CA125), transthyretin, apolipoprotein A1, β2-microglobulin, and transferrin. Evidence to support the rationale of this combination was mainly based on protein expression profiles obtained using the Surface-Enhanced Laser Desorption/Ionization-Time Of Flight (SELDI-TOF) mass spectrometric (MS) platform [10, 11]. While the usage and applicability of both tests may (currently) be limited as compared to the genomic tests mentioned above, both tests provide good examples of successful proteomicsbased biomarker discovery and development and are testimony to the opportunities of proteomics research in this area. The combination of SELDI-TOF and ovarian cancer in the OVA1 example represents a success story of MS-based proteomics, yet this combination will also be remembered in the context of a major controversy which casted a shadow over the biomarker field of research in the first decade of this millennium [12]. Central to this controversy was a study published in The Lancet in 2002, which reported a SELDI-TOF MS-based blood test for the detection of ovarian cancer with 100% sensitivity and 95% specificity [13]. The test stirred hope for the prospect of early-stage cancer detection, though it also stirred concern about the test’s reliability [14]. Reanalysis of the original data, which were deposited in an open access repository, yielded rather dissatisfactory results leading

Pre- and Post-analytical Factors in Biomarker Discovery

3

to the conclusions that “[t]he ability to discriminate between cancer and control subjects . . . reveals the existence of a significant non-biologic experimental bias between these two groups [15]” and that “the reproducibility of the proteomic profiling approach has yet to be established [16]”. The authors of the initial publication responded that reproducibility was never claimed and indicated that the “inappropriate conclusions drawn . . . could have been avoided by communication between the producers and consumers of the data” [13, 17]. Eventually, the reported test did not reach the clinical chemist’s “toolbox” and thus was not applied for the benefit of patients. The corresponding controversy did, however, raise awareness that bias and lack of generalizability (e.g., statistical overfitting of the data) are potential threats to the validity of biomarker-based research findings and furthermore stressed the need for critically assessing pre- and post-analytical factors in biomarker discovery and development research [12, 18]. 1.1 Regulated Bioanalysis

Good quality analytical methods form the basis of the discovery potential of proteomics workflows and are furthermore a key success factor of efforts addressing the later stages of the biomarker development pipeline (i.e., qualification, verification, and validation) [19]. Research dealing with these later stages mostly comprise targeted proteomics endeavors following regulatory guidelines (e.g., FDA [20], EMA [21], CLSI [22]), which have become well rooted in corresponding practices. These guidelines have the aim to minimize inter-laboratory variance by adopting consensus criteria with respect to assay performance. In particular, recommendations are provided for addressing analytical quality attributes like accuracy, precision, sensitivity, and recovery during method validation, but also some pre-analytical factors, such as sample stability (e.g., storage, benchtop, freeze-thaw) and specific matrix effects (e.g., hemolysis, icterus, and lipemia), which, respectively, are attributed to ruptured red blood cells, bilirubin, and lipoprotein particles [23–25]. Adhering to these guidelines does not guarantee the quality or usefulness of a biomarker assay, since other, non-addressed (pre)analytical variables may have a major impact on whether the method and the experimental design are suitable for addressing the study goal. Regulatory documents are living documents that are regularly updated often based on deliberations following expert workshops and panel discussions [26]. Journals adopting requirements with respect to validation of analytical methods in an effort to raise the quality standard of published methods is a positive development with respect to the applicability and reproducibility of scientific research and the corresponding findings, which is not limited to biomarker-related work [27–29]. While strict regulation and standardization are more difficult to implement in discovery-based proteomics as compared to targeted proteomics, it is conceivable that “Good Proteomics Practice”

4

Frank Klont et al.

guidelines will emerge within the foreseeable future. One example of this type of document is the Human Proteome Project Data Interpretation Guideline, which provides guidance on how to interpret fragment-ion-based mass spectrometry data for peptide and protein identification. This guideline is now mandatory for manuscripts submitted to a number of proteomics journals [30]. Guidelines for data management and stewardship such as the Findable, Accessible, Interoperable and Reusable (FAIR) Guiding Principles may furthermore contribute to improving the quality and evaluation of proteomics studies [31]. Although recommendations and standardized procedures aiming to set quality standards for biomarker discovery research have been proposed, there is currently no consensus on their large-scale implementation [27, 32]. 1.2 The Pre-analytical Phase

Providing (consensus-based) guidance for adequately dealing with analytical variables during method development and validation is a complex task. To illustrate this, the FDA draft guidance document released in 2013 was just recently finalized replacing its predecessor from 2001 [20, 33]. Considering that these documents deal with approximately ten analytical variables which ought to be addressed during method validation, providing guidance for pre-analytical variables will be considerably more challenging as these easily outnumber the analytical ones [34]. The reader is referred to excellent reviews on this topic for further details [35–38]. We will present a few examples in Subheading 2 to outline the complexity of the pre-analytical phase and to stress the need for adequately addressing the corresponding sources of variability during biomarker discovery and development. The following subsections provide a framework for these examples.

1.2.1 Pre-sampling Factors

There are dozens of physiological and environmental factors that may affect laboratory results, which are generally condensed into terms like “biological,” “interindividual,” and “between-subject variation” or “variability” in biomarker discovery and development research [19, 23, 24, 38]. These terms explain the increased variation in biomarker levels that are observed when moving from earlystage, small-scale monocentric discovery studies to the advanced biomarker qualification level, where more heterogeneous, larger populations are studied across multiple clinical centers. Factors like age, gender, circadian rhythm, seasonal changes, altitude, menstruation, pregnancy, and lifestyle may play a role with some of them rather difficult to control [34, 39]. While groups are generally matched with respect to gender and age in biomarker studies, other factors may be equally or even more relevant. As an example, we found considerable changes in peroxiredoxin 1, uteroglobin, and serpin B3 levels in pulmonary epithelial lining fluid within 3 h after cigarette smoking on the basis of an iTRAQ-based proteomic analysis followed by validation using commercial immunoassays

Pre- and Post-analytical Factors in Biomarker Discovery

5

[40]. In one of the groups included in this study (i.e., old COPD patients), cigarette smoking led to increases in uteroglobin levels of ten and three times, as determined by the proteomics- and immunoassay-based analyses, respectively. Considering these proteins as disease biomarkers for COPD would thus necessitate tight control of the smoking history prior to sampling, which may be rather difficult in practice. 1.2.2 Sampling Factors

A group of experts in the field of clinical proteomics recommended in a perspective paper in 2010 that detailed descriptions of (appropriate and consistently applied) sampling parameters ought to be provided in publications, since the quality of samples and corresponding results may otherwise be compromised [24]. While this recommendation is rather difficult to comply with, notably for already acquired, biobanked samples, it puts the focus on potential sampling errors which may lead to spurious findings. When, for example, studying the HUPO Plasma Proteome Project specimen collection and handling recommendations published in 2005, it becomes apparent that the list of critical sampling factors, in this case related to blood-based samples, is quite extensive [23]. For each of these sampling factors, either related to venipuncture (e.g., needle gauge), phlebotomy (e.g., tourniquet technique, patient position), or collection device (e.g., tube vs. bag, glass vs. plastic, presence vs. absence of protease inhibitors), there are numerous examples of biomarkers that are affected by corresponding changes in sampling conditions [23, 34, 37, 41]. Such variables are often not controlled or standardized in proteomics research, since many projects target (long-term) stored samples for which sampling conditions were fixed prior to the design of the study and have not always been documented in the necessary detail. In this case, it is difficult to decide whether such samples are suitable for the study goal or not. Analyzing “stabilityindicating” molecules, such as proteins, that are highly susceptible to certain factors (e.g., freeze-thaw cycles) may aid in assessing the quality of the samples (see Subheading 2.1.3) [42]. But even when projects deal with samples, which have yet to be collected, thereby allowing to adopt sampling procedures that conform to prevailing best-practices, such conditions will eventually become outdated. Bias, which will become apparent as time goes by and methodologies improve, will thus inevitably become rooted in most, if not all biobanked samples. Providing detailed descriptions of how samples were obtained is thus important in view of reproducing results and defining factors that may introduce a bias. Such factors may be prioritized and studied in a Design-of-Experiment (DoE) approach as detailed in Subheading 2.1.

6

Frank Klont et al.

1.2.3 Sample Processing and Storage Factors

Pre-analytical factors related to sample processing and storage are more tangible compared to factors addressed in the previous sections. Unintentional mistakes can be made after samples have been taken (e.g., sample contamination, sample spills, improper labeling, inadequately following protocols, “forgetting” to process the samples in time, “losing” samples), and some conditions may lead to unintentionally compromised sample integrity (e.g., exposure to sunlight or moisture, the use of secondary vials with unfavorable (adsorptive) properties, temperature fluctuations in case of sample shipment, power outages, or freezer break-down and maintenance) [37]. In particular when dealing with large numbers of samples that are stored for extended time periods (e.g., in biobanks), it is important to have a tight quality control and to document events that may affect sample quality [43]. The complexity of Omics-based biomarker discovery studies is daunting and repeated failures of so-called “biomarker candidates” to translate into useful, robust clinical assays resulted in a certain skepticism and sometimes even an outright negative attitude toward performing such studies at all. While there is probably no single study that is perfect in all respects, these challenges should motivate researchers to establish and subsequently work according to standards through which the risk of bias is mitigated. Adopting a “Quality-by-Design” (QbD) concept, as originally proposed by Juran, may function as a safeguard against potential errors in biomarker discovery and development research and thus increase the success rate [44]. Lessons may be learned from the pharmaceutical industry, where stringent standards on documenting, managing, and reporting deviations are the rule. Documenting protocol deviations and violations yet also potential weaknesses in experimental design are also very helpful in biomarker discovery research, for example, for adequately interpreting (unexpected) findings. Openness to reporting such information or to sharing all experimental details should be advocated, as this will allow other scientists as well as reviewers and readers of the scientific literature to adequately draw conclusions. In case of the ovarian cancer example, more openness regarding the samples, experimental design, and data processing procedures in the initial article might have limited the extent of the ensuing controversy. In the following, we will present a number of approaches and examples to stress that pre- and post-analytical factors must be carefully controlled, if biomarker discovery and development research shall lead to translatable results that will benefit patients.

Pre- and Post-analytical Factors in Biomarker Discovery

2

7

Methods

2.1 Pre-analytical Factors

There is no doubt that pre-analytical factors can have a major influence on the results of bioanalytical measurements and most notably on the measurement of a wide range of molecules (e.g., proteins, metabolites). This is why regulatory agencies established guidelines on how to assess such factors in regulated bioanalysis, for example, related to clinical studies [35, 39, 45–47]. These guidelines are regularly updated as technology develops in a close collaboration between industrial R&D laboratories, contract research organizations (CROs), and the agencies themselves [26]. While it is comparatively easy to assess the effect of different pre-analytical factors in standard bioanalytical measurements of single compounds, this becomes quite another issue when trying to quantify multiple molecules in multiplex analyses or even in profiling entire classes of molecules in so-called “Omics” studies. To approach this challenge, it is necessary to prioritize pre-analytical factors and to define the range over which each of these factors should be studied. In the first part of this section, we will describe the combination of a Design of Experiment (DoE) approach with multivariate statistical data analysis to gain insight into the relevance of different pre-analytical factors using the example of the preparation of serum from whole blood and the analysis of the serum proteome. Subsequent sections will give examples of the effect of defined pre-analytical factors on the serum proteome (clotting time) and the cerebrospinal fluid (CSF) proteome (delay time between sampling and freezing).

2.1.1 Selecting and Prioritizing Pre-analytical Factors

Defining the pre-analytical factors to be studied and setting the levels over which their effect should be evaluated requires some insight into possible effects of a given factor. While this seems somewhat counterintuitive, since we would like to make no a priori assumptions as to which factors are most relevant, it is simply not possible to study every conceivable pre-analytical factor across a wide range of parameters, as this would lead to an excessive number of experiments, especially when interactions between the factors are included in the analysis. The situation is even more challenging for “Omics” studies, where each experiment is quite work-intensive and time-consuming. The complexity of the experimental design is dependent on whether the goal of the analysis is to determine the effect of one factor (so-called main effect) or whether it is also of interest to study interactions between factors (e.g., how the level of one factor influences the effect of other factors at various levels). Interactions can be studied between two or more factors, but it is

8

Frank Klont et al.

less likely to find interactions between multiple factors. DoE strategies help to select the most appropriate design (e.g., distribution of the levels of the various factors in the analyzed samples), to study the significance of individual factors or interactions between factors on single analytes in case of univariate statistics (e.g., targeted bioanalysis) or on multiple analytes using multivariate statistics (multiplex or “Omics” studies) with the minimal number of experiments. Main effects and two-factor interactions, which are usually of greatest interest, can be assessed with a fractional factorial design, thus further reducing the number of analyses compared to a full factorial design, which allows to study the main effects and all possible interactions between the studied levels of all factors [48, 49]. One possible starting point is to refer to factors that need to be studied to validate a regulated bioanalytical method, such as analyte stability under different conditions of storage and sample preparation. The analyte stability issue is, however, tricky when it comes to “Omics” studies, since one analyte (e.g., a protein) may be stable under one set of conditions while the opposite may be true for another analyte. One would thus need an overall readout concerning sample stability (e.g., based on comparative multivariate statistical analysis) or alternatively define a set of individual analytes that give a representative picture of sample stability. Many areas of biological and biomedical research struggle with this problem. Particularly biobanks, that often store hundred thousands of samples for decennia, need to design strategies how to ensure sample stability for “Omics” as well as for conventional bioanalyses [50–53]. On the one hand, specimen quality needs to be optimized a priori in order to minimize the risk for pre-analytical bias, while on the other hand testing of so-called “surrogate quality biomarkers” or “surrogate quality indicators” is desired to retrospectively assess and verify specimen integrity [54–56]. Controlling pre-analytical bias by design can be pursued by adhering to guidelines and recommendations dealing with biospecimen best practices, such as those from the International Society for Biological and Environmental Repositories (ISBER) and the US-based National Cancer Institute (NCI) [57, 58]. For retrospective assessment of specimen integrity, some analytes have been identified as surrogate quality biomarkers/indicators including soluble CD40 ligand (sCD40L) to assess serum exposure to high temperatures and Vascular Endothelial Growth Factor (VEGF) to assess the effect of freeze-thawing on serum samples, though more markers are likely needed to provide a more generalizable readout of biospecimen stability [54]. It is unlikely that a one-size-fits-all readout will reflect the stability of every analyte in a complex matrix such as serum, plasma, or urine, and adequate assessment of pre-analytical variables should thus ideally be performed in a fit-for-purpose manner.

Pre- and Post-analytical Factors in Biomarker Discovery

9

Much is known about sample preparation of the most widely used body fluids blood (serum, plasma) and urine. This knowledge helps to select pre-analytical factors that may affect sample composition. For example, when it comes to proteomics, protein degradation due to proteolysis must be prevented. In addition, other chemical modifications like the oxidation of sulfur-containing (Met, Cys) or aromatic amino acids (Trp) or the deamidation of Asn and Gln may be an issue. Further to this, there are operational parameters that may induce alterations in sample composition, such as freeze-thaw cycles, the type of blood collection tube, hemolysis, trypsin digestion (time and conditions), or the adsorption of peptides to vials in the autosampler [59]. Operational parameters may be controlled through Standard Operating Procedures (SOPs) while protein degradation and chemical modifications may be controlled via pH or the inclusion of protease inhibitors. Some pre-analytical factors cannot be altered for biobanked, archived samples. In that case, it is advisable to define inclusion/exclusion criteria. We introduced a novel approach to evaluate the effect of pre-analytical factors on the composition of serum in view of using samples from an existing biobank of cervical cancer patients for comparative proteomics analysis [60]. Serum is derived from whole blood by letting the blood clot, often in the presence of a so-called “clotting activator.” In a biomarker discovery study for cervical cancer, we were faced with the situation that an existing biobank containing some 30,000 serum samples had been acquired from patients and controls over the last 25–30 years. It does not come as a surprise that procedures had changed over such a long time period and that not all changes were traceable. It was thus necessary to make a selection of possible pre-analytical factors that should be studied. Based on our knowledge and consultation with clinical and bioanalytical experts we arrived at the following list of factors (see Mitra et al., 2016 for details) [60]. l

Blood collection tube: 2 types of tubes were used in the biobank (BD368430 and BD367784).

l

Hemolysis level: low and high (simulated by adding lysed red blood cells to a “low hemolysis” sample).

l

Clotting time: 2 h and 6 h (see Govorukhina et al., 2009 for details) [61].

l

Freeze-thaw cycles: 1 cycle and 3 cycles.

l

Trypsin digestion: 6% enzyme/substrate ratio and 11% enzyme/ substrate ratio.

l

Stop trypsin digestion with acid: yes and no.

l

Sample stability (in the autosampler): 0 days and 30 days.

10

Frank Klont et al.

Table 1 Fractional factorial design to study the effect of seven pre-analytical factors on the serum protein profile

Sample name

Blood collection tube

Clotting Freeze-thaw Trypsin Hemolysis time cycles digestion

Stopping trypsin

Sample stability

16090534 BD368430

Low

2h

1 cycle

6%

Yes

0 days

16090526 BD367784

Low

2h

1 cycle

11%

Yes

30 days

16090532 BD368430

High

2h

1 cycle

11%

No

0 days

16090540 BD367784

High

2h

1 cycle

6%

No

30 days

16090541 BD368430

Low

6h

1 cycle

11%

No

30 days

1609525

Low

6h

1 cycle

6%

No

0 days

16090533 BD368430

High

6h

1 cycle

6%

Yes

30 days

16090539 BD367784

High

6h

1 cycle

11%

Yes

0 days

16090531 BD368430

Low

2h

3 cycles

6%

No

30 days

16090538 BD367784

Low

2h

3 cycles

11%

No

0 days

16090536 BD368430

High

2h

3 cycles

11%

Yes

30 days

16090530 BD367784

High

2h

3 cycles

6%

Yes

0 days

16090535 BD368430

Low

6h

3 cycles

11%

Yes

0 days

16090529 BD367784

Low

6h

3 cycles

6%

Yes

30 days

16090528 BD368430

High

6h

3 cycles

6%

No

0 days

16090542 BD367784

High

6h

3 cycles

11%

No

30 days

BD367784

Each of the seven factors was varied at two levels

These factors and their levels were analyzed according to a fractional factorial design with resolution IV ( 27
3 IV —design) [48, 49] to limit the number of required experiments to 16 (Table 1). This design allowed studying the main effects of all selected factors; while two-way interactions of factors were clustered in triades (i.e., three interactions between two factors are indistinguishable in this design). The generated LC–MS data were preprocessed at the single-stage MS level using the Threshold Avoiding Proteomics Pipeline (TAPP) [62] to arrive at one common peak table, which was subsequently filtered based on fold change and statistical significance (Volcano Plot) to select the most significantly affected peptides (fold change > 2 and pvalue < 0.05), as surrogates for the corresponding proteins. The selection criteria were chosen fairly lenient in this case so as not to exclude too many proteins that may be affected by the pre-analytical factors. The selected peaks were subjected to

Pre- and Post-analytical Factors in Biomarker Discovery

11

multivariate statistical analysis by ANOVA—simultaneous component analysis (ASCA) [63] to find peptide peaks that were affected by one or multiple pre-analytical factors and to prioritize the factors according to their overall significance level. This was judged from the variance that can be explained by varying a given factor between the two studied levels, as given by the size of the principal component for that factor measured by the sum of squares (SSQ) of the elements of the matrix decomposed according to the studied main effect. We subsequently ordered the different factors according to their respective SSQ values and prioritized them according to their influence on the overall serum peptide profile. The significance of each factor was determined by permuting the levels and factors of the analyzed samples. The final list of significant factors was obtained by setting a threshold for the obtained p-value. This procedure is generic and may be applied to other “Omics” studies and in general to complex data sets that are affected by a number of pre-analytical factors to varying degrees. ASCA analysis showed that the hemolysis level had the largest overall influence on the serum profile followed by the trypsin/ protein ratio and the fact of stopping trypsin digestion with acid or not. Figure 1 gives an overview of the workflow and shows an example of a hemoglobin-derived peptide peak (upper right) that clearly discriminates a high-hemolysis from a low-hemolysis sample. It is not surprising that hemoglobin stands out as the most discriminating protein with respect to hemolysis, so its level may be used as an exclusion criterion. 2.1.2 The Effect of Clotting Time on the Serum Protein Profile

As mentioned above, serum preparation involves a clotting step during which a cascade of proteolytic enzymes and cofactors is activated. It is thus critical to evaluate the influence of this step on the protein profile and to establish the time during which blood should clot under a given set of conditions to reach a stable state. We studied this by letting blood clot at room temperature for 1, 2, 4, 6, or 8 h in glass tubes containing a separation gel and micronized silica to accelerate clotting followed by centrifugation at room temperature for 10 min at 3000 rpm. Serum samples were subsequently trypsin digested and analyzed by LC–MS [61]. This is an example of a study, where one particular pre-analytical factor was followed across a given parameter space in greater detail. The clotting time is of particular interest for biobanked serum samples, since it is often not known with certainty or has not been documented. In addition, different laboratories may use different conditions ranging from clotting at reduced temperatures without a clotting activator for prolonged time periods to clotting at room temperature in the presence of an activator, as described here. We have seen considerable differences in protein

12

Frank Klont et al.

Fig. 1 General scheme of the experimental analysis and statistical evaluation including sample preparation, LC–MS analysis, data preprocessing, and ASCA analysis. Proteins in biological samples were digested to peptides, which are in this example influenced by seven factors each at two levels. This affects the peptide profile obtained by LC–MS as shown by the extracted ion chromatogram (EIC) of one sample as an example (upper right panel). LC–MS raw data were preprocessed with the TAPP pipeline at the single-stage MS level [62]. The natural logarithm intensity values from the data matrix were then filtered using a fold change of 2 and a t-test significance of 0.05 as shown in a Volcano plot. The resulting data matrix was submitted to ASCA analysis, which identified significant factors and provided a ranked list of discriminatory peaks as shown by the extracted ion chromatogram and box plots in the upper right panel (green traces: high; blue traces: low level). Peptide identity was annotated based on LC–MS/MS data obtained from the same sample using a QTOF LC–MS/MS platform. (From Mitra et al., 2016; copyright American Chemical Society) [60]

levels between such conditions that actually led us into a dead alley in one of our biomarker discovery studies (Fig. 2). While clotting time, under our conditions, did not affect the overall protein profile significantly (Fig. 3a), it did affect proteins

Pre- and Post-analytical Factors in Biomarker Discovery

13

Fig. 2 Comparative serum analysis for patients with cervical intraepithelial lesions (CIN, a precancerous stage), early-stage cancer (EC), late-stage cancer (LC), and ovarian cancer (OvC) in comparison to serum from healthy volunteers (H). While serum from healthy volunteers (n ¼ 23) was coagulated overnight at 4  C without a clotting activator in a bleed bag unit (number not specified, provided by a commercial source), all patient sera were prepared at room temperature for 2–6 h with a clotting activator. This affected a platelet-derived protein (thrombospondin-1) and introduced a statistically significant bias for this protein

Fig. 3 (a) Principal component analysis (PCA) score plots of matched peak matrices obtained after LC–MS analysis of depleted, trypsin-digested serum after clotting for 1, 2, 4, 6, and 8 h. All analyses were performed in duplicate on different days (denominated as s1 and s2). (b) Box plots of the intensity values of a fibrinogenderived peptide peak that decreases significantly with increasing clotting time. This peptide corresponds to a fragment of the fibrinogen alpha chain (DSGEGDFLAEGGGVR). Box plots show the median as well as the lower quartile (25% of the distribution) and the higher quartile (75% of the distribution). The “whiskers” are broken lines extending from each end of the box to show the interval containing 95% of the data

14

Frank Klont et al.

and peptides that are directly involved in the clotting process, such as fibrinogen-related peptides (Fig. 3b). In fact, following the level of fibrinogen-derived peptides indicates that the clotting process continues for about 6 h, which is much longer than what is generally used in clinical practice. It is thus important to use sufficiently long clotting times to avoid variability or to control the clotting time carefully. As shown above (Fig. 2), it is also important to maintain consistent clotting conditions to avoid, for example, the release of platelet-derived proteins such as thrombospondin-1. This emphasizes the need to control pre-analytical factors to assure that no bias is introduced perturbing the final statistical data analysis. 2.1.3 The Effect of the Delay Time Between Sampling and Freezing for Cerebrospinal Fluid (CSF)

Another example from our work relates to a pre-analytical factor that is often ill-defined in clinical studies, namely the delay time between sampling and final sample storage. In routine clinical practice, it is often difficult to control the time between taking a sample from a patient and its final storage. It is thus critical to evaluate the relative importance of this factor with respect to the study goal and especially with respect to other sources of variability (e.g., interindividual variability). In a study on multiple sclerosis biomarkers in CSF, we evaluated the delay time between sampling and sample storage in relation to other pre-analytical and biological conditions [64]. While no separation according to the delay time between 0, 30, and 120 min was observed when taking the top 10,000 protein-derived isotopologue peaks into account (Fig. 4a), there was partial discrimination when selecting the 11 most discriminating features using the Nearest Shrunken Centroid (NSC) variable selection algorithm (Fig. 4b) [64, 65]. Some peptides, as surrogates of the corresponding proteins after trypsin digestion, showed a statistically significant decrease with delay time. It is noteworthy in this context that these differences were still significant even after taking interindividual differences into account and correcting for multiple testing (Bonferroni) (Fig. 4c). These examples illustrate that ill-controlled pre-analytical factors can easily bias the results of biomarker discovery studies and thus lead to biomarker candidates that will not survive the subsequent biomarker development phase let alone clinical implementation.

2.2 Post-analytical Variables

It seems to be a strange concept to refer to “post-analytical factors” but acquiring raw data of good quality is only part of the work. Identical LC–MS data from a proteomics biomarker discovery study may be processed with different quantitative workflows and different parameters leading to very different results. We addressed this with a spike-in experiment, where known peptides were added to a urine sample at three different levels [66]. We subsequently processed the data with different open-source, quantitative workflows and used a scoring algorithm (see Hoekman et al., 2013 for

Pre- and Post-analytical Factors in Biomarker Discovery

15

Fig. 4 (a) Multivariate statistical analysis (PCA) of the 10.000 most intense isotopologue peaks selected from a tryptic digest of CSF analyzed by LC–MS (n ¼ 5). No separation based on time between sampling and freezing (0, 30 or 120 min) is visible. (b) Multivariate statistical analysis by PCA based on 11 NSC-selected peaks (0 and 30 min delay time; 5 repetitive analyses of 6 human CSF samples). (c) Univariate statistical analysis of a peak at m/z 736.383 that decreased significantly after 30 and 120 min delay time. Statistical analysis was based on a 2-tailed Student t-test with Bonferroni correction of the combined data from 5 repetitive analyses of 6 human CSF samples (* p < 5  10
5; from Rosenling et al. 2011, copyright American Association for Clinical Chemistry) [64]

details [66]) to assess the ranking of true positives (features related to the spike-in peptides) vs. false positives (features not related to the spike-in peptides). Although the input data were identical, output in terms of “discovered biomarkers” (features related to spike-in peptides) vs. false positives differed significantly between different data processing approaches (Fig. 5). Errors can have many origins; however, the most commonly observed error was due to peak splitting. In this case, a non-discriminative peak is detected as two peaks in one chromatogram but not in the other, which leads to a correspondence problem. This situation results in two entries in the quantitative peak matrix that is used for statistical analysis, of

16

Frank Klont et al.

Fig. 5 (a) Comparison of the performance of different quantitative data processing workflows with ion trap LC–MS data derived from the analysis of human urine that was spiked with a range of peptides at three different concentration levels. (b) Occurrence of peak splitting in one chromatogram and matching with a corresponding peak that was not split leads to correspondence (matching) problems and may (c) result in incorrectly assigned biomarkers upon subsequent statistical analysis. (d) Visualization of the peak in all chromatograms clearly shows that it is not discriminative, which is (e) confirmed by manual integration of peaks. Quality scores were calculated as described in Hoekman et al., 2012 (copyright The American Society for Biochemistry and Molecular Biology, Inc.) [66]

which one is highly discriminative resulting in an incorrectly selected biomarker. To avoid this situation raw data and peak detection should be visualized prior to assigning biomarker status to any feature, to reveal peak splitting and other possible errors and to confirm or reject the hypothesis that a peak is differential between the two sample groups. The point is not to emphasize that certain data processing pipelines work better than others but to make the reader aware of

Pre- and Post-analytical Factors in Biomarker Discovery

17

Fig. 6 Overview over the two best performing feature selection statistical methods for data sets of different sample size and class separation (the difference between the spiking levels in different sample groups) based on the f-score. The Nearest Shrunken Centroid (NSC) algorithm shows its superior overall performance for data sets with 6 repeat analyses independent of class separation, while univariate tests (t-test and MannWhitney-test; Mw-test) perform better when sample size increases to 12 or 15 repeat analyses per class (Christine et al., 2013; copyright The American Society for Biochemistry and Molecular Biology, Inc.) [67]

the fact that data processing algorithms and parameter settings can have a significant effect on the final result and that this part of the procedure requires careful optimization as well. The same goes for the ensuing statistical analysis. Using the same processed spike-in data sets, we assessed different statistical feature selection approaches for biomarker discovery (by “biomarkers” we mean features that are related to the spike-in peptides) and further to this evaluated the relevance of repeating analyses for improving the selection of true positives. Without going into details, which can be found in Christin et al. [67], it became apparent that certain feature selection approaches are more suitable for a given kind of data than others (Fig. 6). It is thus advisable to test both the data processing and the statistical feature selection approaches with realistic spike-in samples, for which the “ground truth” is known, so that errors in the form of false positives and false negatives can be minimized. While no approach gave infallible results, it is important to know where the compromise between true positives and false positives lies for a given approach and a given set of parameters. In this way one can chose the most suitable approach and adjust parameters to favor either sensitivity or specificity or to find the best compromise

18

Frank Klont et al.

between the two depending on the goal of the study [ROC (receiver operating characteristics) curves are often used for this purpose] [67].

3

Conclusions and Perspectives There is no doubt that controlling pre-analytical factors and understanding their impact on the final result plays a crucial role in biomarker discovery studies. Changes in protein profiles due to sampling, sample handling, or sample storage may introduce significant bias and lead to false biomarker candidates that will fail during subsequent validation studies. Next to an improper design of the patient and control cohorts that serve as basis for the discovery study, this is likely the most frequent reason why biomarker candidates do not translate into clinically useful diagnostic or prognostic markers, as results cannot be reproduced in larger multicentric studies. There is also a risk of introducing bias and errors during the subsequent data processing and statistical analysis steps. This is due to the fact that signals must be discriminated from background noise and irrelevant signals based on well-defined and stringently controlled criteria and correctly adjusted parameters such as the signal-to-noise ratio threshold and mathematical criteria defining a signal (e.g., a peak) such as peak width, shape, and isotope pattern. This is far from straightforward when considering that a typical proteomics analysis may comprise more than a million signals that may or may not be related to peptide isotopologue peaks. Reliability of the subsequent statistical analysis depends critically on the quality of the input data after data preprocessing. However, statistical analysis may also introduce errors in terms of false positives and false negatives. Both data preprocessing algorithms and statistical analysis approaches make assumptions about the underlying data, such as the shape of peaks or the distribution of real signals vs. noise. These conditions may not be met for all compound-related signals, which leads to preprocessing and biomarker selection bias. Depending on the structure of the data and the goal of the study, it is advisable to try different statistical approaches and look for consistent findings. In this chapter we tried to provide guidelines how to approach the difficult problem of studying pre- and post-analytical factors in an “Omics-based” biomarker discovery experiment. Suggestions can be found in reviews and perspectives articles [24]. While the effect of some pre-analytical factors on the overall protein profile may be overrated (see our work on the clotting time for serum preparation), others may not be considered at all (e.g., the delay time between sampling and sample storage). Although some pre-analytical factors may not be known for already biobanked samples, such samples could be unique in other aspects (e.g., in

Pre- and Post-analytical Factors in Biomarker Discovery

19

terms of their longitudinal character covering decades). Using such samples for biomarker discovery requires thorough validation and possibly establishing inclusion/exclusion criteria. Stabilityindicating markers may be useful to assess the quality of biobanked samples. However, there remains a risk of biomarker candidates failing due to an initial bias. In case longitudinal samples from the same patient are available, they may serve as internal controls and trend analysis can help to support the selection of biomarkers. We showed further that, while the overall protein profile may not change, it is important to perform feature selection to look at the most discriminating signals and to exclude the corresponding analytes from future biomarker candidate lists unless this particular pre-analytical factor has been (can be) rigorously controlled. On the other hand, such molecules may serve as sensitive indicators of sample quality with respect to a given pre-analytical factor and can thus be used as inclusion/exclusion criteria. It is hard to give generic “one-size-fits-all” guidelines but common sense is a good starting point as are the guidelines of the FDA, the EMA, or the CLSI with respect to regulated, quantitative bioanalysis [20, 21, 68]. References 1. Papavramidou N, Papavramidis T, Demetriou T (2010) Ancient Greek and Greco-Roman methods in modern surgical treatment of cancer. Ann Surg Oncol 17(3):665–667. https:// doi.org/10.1245/s10434-009-0886-6 2. Weiss L (2000) Metastasis of cancer: a conceptual history from antiquity to the 1990s. Cancer Metastasis Rev 19(3–4):193–383 3. Jackson WA (2001) A short guide to humoral medicine. Trends Pharmacol Sci 22 (9):487–489. https://doi.org/10.1016/ S0165-6147(00)01804-6 4. Paulovich AG, Whiteaker JR, Hoofnagle AN et al (2008) The interface between biomarker discovery and clinical validation: the tar pit of the protein biomarker pipeline. Proteomics Clin Appl 2(10–11):1386–1402. https://doi. org/10.1002/prca.200780174 5. McCarthy JJ, McLeod HL, Ginsburg GS (2013) Genomic medicine: a decade of successes, challenges, and opportunities. Sci Transl Med 5(189):189sr184. https://doi. org/10.1126/scitranslmed.3005785 6. Calvaruso V, Craxı` A (2012) 2011 European Association of the Study of the Liver hepatitis C virus clinical practice guidelines. Liver Int 32 (Suppl 1):2–8. https://doi.org/10.1111/j. 1478-3231.2011.02703.x

7. Nkuipou-Kenfack E, Zurbig P, Mischak H (2017) The long path towards implementation of clinical proteomics: exemplified based on CKD273. Proteomics Clin Appl 11(5–6). https://doi.org/10.1002/prca.201600104 8. Toss A, De Matteis E, Rossi E et al (2013) Ovarian cancer: can proteomics give new insights for therapy and diagnosis? Int J Mol Sci 14(4):8271–8290. https://doi.org/10. 3390/ijms14048271 9. Zhang Z, Bast RC Jr, Yu Y et al (2004) Three biomarkers identified from serum proteomic analysis for the detection of early stage ovarian cancer. Cancer Res 64(16):5882–5890. https://doi.org/10.1158/0008-5472.CAN04-0746 10. Fung ET (2010) A recipe for proteomics diagnostic test development: the OVA1 test, from biomarker discovery to FDA clearance. Clin Chem 56(2):327–329. https://doi.org/10. 1373/clinchem.2009.140855 11. Li D, Chan DW (2014) Proteomic cancer biomarkers from discovery to approval: it’s worth the effort. Expert Rev Proteomics 11 (2):135–136. https://doi.org/10.1586/ 14789450.2014.897614 12. Ransohoff DF (2005) Lessons from controversy: ovarian cancer screening and serum

20

Frank Klont et al.

proteomics. J Natl Cancer Inst 97 (4):315–319. https://doi.org/10.1093/jnci/ dji054 13. Petricoin EF, Ardekani AM, Hitt BA et al (2002) Use of proteomic patterns in serum to identify ovarian cancer. Lancet 359 (9306):572–577. https://doi.org/10.1016/ S0140-6736(02)07746-2 14. Check E (2004) Proteomics and cancer: running before we can walk? Nature 429 (6991):496–497. https://doi.org/10.1038/ 429496a 15. Sorace JM, Sorace JM, Zhan M (2003) A data review and re-assessment of ovarian cancer serum proteomic profiling. BMC Bioinformatics 4:24. https://doi.org/10.1186/14712105-4-24 16. Baggerly KA, Morris JS, Edmonson SR et al (2005) Signal in noise: evaluating reported reproducibility of serum proteomic tests for ovarian cancer. J Natl Cancer Inst 97 (4):307–309. https://doi.org/10.1093/jnci/ dji008 17. Liotta LA, Lowenthal M, Mehta A et al (2005) Importance of communication between producers and consumers of publicly available experimental data. J Natl Cancer Inst 97 (4):310–314. https://doi.org/10.1093/jnci/ dji053 18. Ransohoff DF (2005) Bias as a threat to the validity of cancer molecular-marker research. Nat Rev Cancer 5(2):142–149. https://doi. org/10.1038/nrc1550 19. Rifai N, Gillette MA, Carr SA (2006) Protein biomarker discovery and validation: the long and uncertain path to clinical utility. Nat Biotechnol 24(8):971–983. https://doi.org/10. 1038/nbt1235 20. US FDA (2001) Guidance for Industry: Bioanalytical Method Validation. US Department of Health and Human Services, FDA, Center for Drug Evaluation and Research, Rockville, MD. https://www.fda.gov/downloads/ Drugs/Guidance/ucm070107.pdf 21. Guideline on Bioanalytical Method Validation (2011) European Medicines Agency (EMEA/ CHMP/EWP/192217/2009). London. http://www.ema.europa.eu/docs/en_GB/ document_library/Scientific_guideline/ 2011/08/WC500109686.pdf 22. CLSI (2014) Liquid-chromatography-mass spectrometry methods; approved guideline. CLSI document C62-A. Clinical and Laboratory Standards Institute, Wayne, PA 23. Rai AJ, Gelfand CA, Haywood BC et al (2005) HUPO Plasma Proteome Project specimen collection and handling: towards the

standardization of parameters for plasma proteome samples. Proteomics 5(13):3262–3277. https://doi.org/10.1002/pmic.200401245 24. Mischak H, Allmaier G, Apweiler R et al (2010) Recommendations for biomarker identification and qualification in clinical proteomics. Sci Transl Med 2(46):46ps42. https:// doi.org/10.1126/scitranslmed.3001249 25. Carr S, Aebersold R, Baldwin M et al (2004) The need for guidelines in publication of peptide and protein identification data. Mol Cell Proteomics 3(6):531–533. https://doi.org/ 10.1074/mcp.T400006-MCP200 26. Booth B, Arnold ME, DeSilva B et al (2015) Workshop report: crystal city V—quantitative bioanalytical method validation and implementation: the 2013 revised FDA guidance. AAPS J 17(2):277–288. https://doi.org/10.1208/ s12248-014-9696-2 27. Abbatiello S, Ackermann BL, Borchers C et al (2017) New guidelines for publication of manuscripts describing development and application of targeted mass spectrometry measurements of peptides and proteins. Mol Cell Proteomics 16(3):327–328. https://doi.org/ 10.1074/mcp.E117.067801 28. GBSI (2013) The case for standards in life science research: seizing opportunities at a time of critical need. Global Biological Standards Institute, Washington, DC. https:// www.gbsi.org/wp-content/uploads/2013/ 12/The-Case-for-Standards.pdf 29. Freedman LP, Cockburn IM, Simcoe TS (2015) The economics of reproducibility in preclinical research. PLoS Biol 13(6): e1002165. https://doi.org/10.1371/journal. pbio.1002165 30. Deutsch EW, Overall CM, Van Eyk JE et al (2016) Human proteome project mass spectrometry data interpretation guidelines 2.1. J Proteome Res 15(11):3961–3970. https:// doi.org/10.1021/acs.jproteome.6b00392 31. Wilkinson MD, Dumontier M, Aalbersberg IJ et al (2016) The FAIR Guiding Principles for scientific data management and stewardship. Sci Data 3:160018. https://doi.org/10. 1038/sdata.2016.18 32. Bruderer R, Bernhardt OM, Gandhi T et al (2017) Optimization of experimental parameters in data-independent mass spectrometry significantly increases depth and reproducibility of results. Mol Cell Proteomics 16 (12):2296–2309. https://doi.org/10.1074/ mcp.RA117.000314 33. US FDA (2013) Guidance for Industry: Bioanalytical Method Validation. US Department of Health and Human Services, Food and

Pre- and Post-analytical Factors in Biomarker Discovery Drug Administration, Center for Drug Evaluation and Research, Center for Veterinary Medicine, Rockville, MD. https://www.fda. gov/downloads/Drugs/Guidances/ ucm070107.pdf 34. Narayanan S (2000) The preanalytic phase. An important component of laboratory medicine. Am J Clin Pathol 113(3):429–452. https:// doi.org/10.1309/C0NM-Q7R0-LL2EB3UY 35. Kellogg MD, Ellervik C, Morrow D et al (2015) Preanalytical considerations in the design of clinical trials and epidemiological studies. Clin Chem 61(6):797–803. https:// doi.org/10.1373/clinchem.2014.226118 36. Salvagno GL, Danese E, Lippi G (2017) Preanalytical variables for liquid chromatographymass spectrometry (LC-MS) analysis of human blood specimens. Clin Biochem 50 (10–11):582–586. https://doi.org/10.1016/ j.clinbiochem.2017.04.012 37. Ellervik C, Vaught J (2015) Preanalytical variables affecting the integrity of human biospecimens in biobanking. Clin Chem 61 (7):914–934. https://doi.org/10.1373/ clinchem.2014.228783 38. Geyer PE, Holdt LM, Teupser D et al (2017) Revisiting biomarker discovery by plasma proteomics. Mol Syst Biol 13(9):942. https://doi. org/10.15252/msb.20156297 39. O’Bryant SE, Gupta V, Henriksen K et al (2015) Guidelines for the standardization of preanalytic variables for blood-based biomarker studies in Alzheimer’s disease research. Alzheimers Dement 11(5):549–560. https://doi. org/10.1016/j.jalz.2014.08.099 40. Franciosi L, Postma DS, van den Berge M et al (2014) Susceptibility to COPD: differential proteomic profiling after acute smoking. PLoS One 9(7):e102037. https://doi.org/10. 1371/journal.pone.0102037 41. Lippi G, Becan-McBride K, Behulova D et al (2013) Preanalytical quality improvement: in quality we trust. Clin Chem Lab Med 51 (1):229–241. https://doi.org/10.1515/ cclm-2012-0597 42. Rosenling T, Slim CL, Christin C et al (2009) The effect of pre-analytical factors on stability of the proteome and selected metabolites in cerebrospinal fluid (CSF). J Proteome Res 8 (12):5511–5522. https://doi.org/10.1021/ pr9005876 43. Tissot JD, Currat C, Sprumont D (2017) Proteomics of blood plasma/serum samples stored in biobanks: insights for clinical application. Expert Rev Proteomics 14(8):643–644.

21

https://doi.org/10.1080/14789450.2017. 1324301 44. Schweitzer M, Pohl M, Hanna-Brown M et al (2010) Implications and opportunities of applying QbD principles to analytical measurements. Pharm Technol 34(2):52–59 45. Yin P, Lehmann R, Xu G (2015) Effects of pre-analytical processes on blood samples used in metabolomics studies. Anal Bioanal Chem 407(17):4879–4892. https://doi.org/10. 1007/s00216-015-8565-x 46. Kong FS, Zhao L, Wang L et al (2017) Ensuring sample quality for blood biomarker studies in clinical trials: a multicenter international study for plasma and serum sample preparation. Transl Lung Cancer Res 6(6):625–634. https://doi.org/10.21037/tlcr.2017.09.13 47. Dakappagari N, Zhang H, Stephen L et al (2017) Recommendations for clinical biomarker specimen preservation and stability assessments. Bioanalysis 9(8):643–653. https://doi.org/10.4155/bio-2017-0009 48. Box GEP, Hunter JS, Hunter WG (2005) Statistics for experimenters: design, innovation, and discovery, 2nd edn. John Wiley & Sons, Hoboken, NJ 49. Montgomery DC (2012) Design and analysis of experiments, 8th edn. John Wiley & Sons, Hoboken, NJ 50. Mateos J, Carneiro I, Corrales F et al (2017) Multicentric study of the effect of pre-analytical variables in the quality of plasma samples stored in biobanks using different complementary proteomic methods. J Proteomics 150:109–120. https://doi.org/10.1016/j. jprot.2016.09.003 51. Malm J, Fehniger TE, Danmyr P et al (2013) Developments in biobanking workflow standardization providing sample integrity and stability. J Proteomics 95:38–45. https://doi. org/10.1016/j.jprot.2013.06.035 52. Malm J, Vegvari A, Rezeli M et al (2012) Large scale biobanking of blood - the importance of high density sample processing procedures. J Proteomics 76:116–124. https://doi.org/10. 1016/j.jprot.2012.05.003 53. Malm J, Lindberg H, Erlinge D et al (2015) Semi-automated biobank sample processing with a 384 high density sample tube robot used in cancer and cardiovascular studies. Clin Transl Med 4(27):2–8. https://doi.org/10. 1186/s40169-015-0067-0 54. Lengelle J, Panopoulos E, Betsou F (2008) Soluble CD40 ligand as a biomarker for storage-related preanalytic variations of human serum. Cytokine 44(2):275–282. https://doi.org/10.1016/j.cyto.2008.08.010

22

Frank Klont et al.

55. Betsou F, Gunter E, Clements J et al (2013) Identification of evidence-based biospecimen quality-control tools: a report of the International Society for Biological and Environmental Repositories (ISBER) Biospecimen Science Working Group. J Mol Diagn 15(1):3–16. https://doi.org/10.1016/j.jmoldx.2012.06. 008 56. Chaigneau C, Cabioch T, Beaumont K et al (2007) Serum biobank certification and the establishment of quality controls for biological fluids: examples of serum biomarker stability after temperature variation. Clin Chem Lab Med 45(10):1390–1395. https://doi.org/10. 1515/CCLM.2007.160 57. Doucet M, Becker KF, Bjorkman J et al (2017) Quality Matters: 2016 Annual Conference of the National Infrastructures for Biobanking. Biopreserv Biobank 15(3):270–276. https:// doi.org/10.1089/bio.2016.0053 58. Vaught J, Lockhart NC (2012) The evolution of biobanking best practices. Clin Chim Acta 413(19–20):1569–1575. https://doi.org/10. 1016/j.cca.2012.04.030 59. Van Midwoud PM, Rieux L, Bischoff R et al (2007) Improvement of recovery and repeatability in liquid chromatography-mass spectrometry analysis of peptides. J Proteome Res 6(2):781–791. https://doi.org/10.1021/ pr0604099 60. Mitra V, Govorukhina N, Zwanenburg G et al (2016) Identification of analytical factors affecting complex proteomics profiles acquired in a factorial design study with analysis of variance: simultaneous component analysis. Anal Chem 88(8):4229–4238. https://doi.org/10. 1021/acs.analchem.5b03483 61. Govorukhina NI, de Vries M, Reijmers TH et al (2009) Influence of clotting time on the protein composition of serum samples based on LC-MS data. J Chromatogr B Analyt Technol Biomed Life Sci 877(13):1281–1291.

https://doi.org/10.1016/j.jchromb.2008. 10.029 62. Suits F, Hoekman B, Rosenling T et al (2011) Threshold-avoiding proteomics pipeline. Anal Chem 83(20):7786–7794. https://doi.org/ 10.1021/ac201332j 63. Smilde AK, Jansen JJ, Hoefsloot HC et al (2005) ANOVA-simultaneous component analysis (ASCA): a new tool for analyzing designed metabolomics data. Bioinformatics 21(13):3043–3048. https://doi.org/10. 1093/bioinformatics/btm419 64. Rosenling T, Stoop MP, Smolinska A et al (2011) The impact of delayed storage on the measured proteome and metabolome of human cerebrospinal fluid. Clin Chem 57 (12):1703–1711. https://doi.org/10.1373/ clinchem.2011.167601 65. Tibshirani R, Hastie T, Narasimhan B et al (2002) Diagnosis of multiple cancer types by shrunken centroids of gene expression. Proc Natl Acad Sci U S A 99(10):6567–6572. https://doi.org/10.1073/pnas.082099299 66. Hoekman B, Breitling R, Suits F et al (2012) msCompare: a framework for quantitative analysis of label-free LC-MS data for comparative candidate biomarker studies. Mol Cell Proteomics 11(6):M111.015974. https://doi.org/ 10.1074/mcp.M111.015974 67. Christin C, Hoefsloot HCJ, Smilde AK et al (2013) A critical assessment of feature selection methods for biomarker discovery in clinical proteomics. Mol Cell Proteomics 12 (1):263–276. https://doi.org/10.1074/mcp. M112.022566 68. Group F-NBW (2016) BEST (Biomarkers, EndpointS, and other Tools) Resource. Silver Spring (MD), Food and Drug Administration (US), Bethesda (MD): National Institutes of Health (US)

Chapter 2 Pre-fractionation of Noncirculating Biological Fluids to Improve Discovery of Clinically Relevant Protein Biomarkers Annarita Farina Abstract Nowadays, significant difficulties remain in the diagnosis and/or prognosis of many diseases, leading to an unsatisfactory patient management and a counterproductive increase in time and costs. It is therefore crucial to bridge the gap between basic and applied research by complying with clinical requirements, notably from the design stage of the experimental workflow. In this chapter we provide key suggestions for selecting appropriate biological samples and reducing pre-analytical and analytical variabilities to improve the discovery of clinically relevant protein biomarkers. Key words Differential centrifugation, Extracellular Vesicles, Exosomes, Microvesicles, Body fluids, Differential diagnosis, Translational medicine

1

Introduction In the context of ever-increasing scientific production in the field of biomarkers, currently nearly 3000 publications per year (PubMed search for “biomarker[Title]”), the number of new biomarkers that reach the clinic is unsatisfactory. According to the FDA Premarket Approval Database, only 30 in vitro Diagnostic tests have been cleared/approved by the Pathology Advisory Committee in the last 10 years, most of which use known biomarkers to predict response to a given treatment. Less than one new medical device per year belongs, instead, to the diagnostic or prognostic category. The reasons for this inefficiency are at all stages of the translational process, but mainly in the experimental design. As other authors have already pointed out, the time has come for researchers working on biomarker discovery to fully comply with clinical requirements [1, 2]. For the sake of brevity, we will focus on the topics specifically addressed in this chapter.

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_2, © Springer Science+Business Media, LLC, part of Springer Nature 2019

23

24

Annarita Farina

1.1 Cases and Controls Inclusion

For biomarker discovery studies, we strongly suggest referring to recently published guidelines [2] with specific regard to the choice of case and control patients. In particular, where there is a diagnostic/prognostic conundrum for clinicians, the pathologies to be compared must correspond specifically to those envisaged for the application of biomarkers. For instance, using healthy controls, while in daily clinical life a differential diagnosis must be established between benign and malignant lesions, will almost certainly lead to the identification of differentially expressed but useless proteins.

1.2

Sample Selection

Due to the ease and low invasiveness of their collection, circulating fluids have always been a sample of choice. While housing biomarkers, however, they gather molecules from each physiopathological process of the body, thus shrinking the analytes of interest often below the limit of detection of available instruments. A wiser approach would consist in a shotgun screening of body fluids confined to the affected organ, to take advantage of the direct release and accumulation of pathology-related molecules [3]. Once revealed, potential biomarkers could be evaluated in circulating fluids using more sensitive targeted assays.

1.3 Sample Collection and Storage

Biological samples, as such, show interindividual variability due to intrinsic and extrinsic factors influencing molecular processes. In addition to a statistically adequate design of the cohort, subsequent sample collection and handling must ensure minimal pre-analytical variability [4, 5]. To favor a consistent resuspension of the solid components of the fluid, the patient should move (or be moved) prior to collection. The sampling procedure must produce identical samples without introducing contamination from neighboring tissues. The delay between collection and storage at 80  C should be minimized and should not differ significantly between samples. It is advisable to supply the sampling room with liquid nitrogen or dry ice to allow immediate freezing. Finally, the fluid should be directly aliquoted into low-protein binding cryotubes and not be subjected to repeated freeze and thaw cycles. If blood is used for biomarker measurement, it is mandatory to avoid typical ex-vivo artifacts. Proteolysis, hemolysis, and platelet activation, indeed, particularly affect the reliability of free- and extracellular vesicle (EV)-protein detection and are strongly influenced by negative pressure, anticoagulants, shaking, and centrifugation in the early phases of sample collection and transport. A large diameter needle (e.g., 21 G) is required to avoid in-vitro hemolysis and release of erythrocyte-derived EVs. EDTA-treated plasma is the best choice for minimizing proteolysis and EV rise due to platelet activation. For the latter, it is also recommended to centrifuge samples twice at 2000  g. Finally, since other potential sources of EVs have been detected due to shaking during sample transport, samples must be centrifuged at the collection site and

Biological Fluid Pre-Fractionation for Biomarker Discovery

25

immediately frozen. If this is impossible, a tip is given to transport samples vertically [6–8]. 1.4 Pre-fractionation Methods

Whatever the chosen fluid, the reduction of its complexity is necessary to improve the detection of biomarkers, especially in cases where it is specifically necessary to analyze the protein cargo of a given subcellular compartment (i.e., EVs). Most commonly used methods to decomplexify samples and reduce their protein dynamic range are likely to introduce large pre-analytical variability due to multiple manipulations, thus hindering downstream comparative quantitation [9, 10]. We suggest here to enrich the fluid components according to their density by differential centrifugation (Fig. 1). This method guarantees a virtually lossless protein yield and has been shown to be effective in highlighting cancer biomarkers in human bile [11–13]. An additional step of SDS-PAGE is used to trap proteins in the gel matrix for the sole purpose of improving the efficiency of tryptic digestion [14].

1.5 Volume-Based Quantitation

One of the most decisive information, generally overlooked by biomarker researchers, is that automated chemistry analyzers and nonautomated immunochemical methods (i.e., ELISA, lateral-flow devices) often used in clinical laboratories are designed to work with a given volume of biological fluids. For the introduction of a new protein biomarker in clinical settings, the quantitative discovery phase cannot avoid comparing fluid samples based on the same volume. This requires the addition of an internal standard for subsequent normalization. Experimenters can choose the spiked molecules based on their samples and their specific needs.

2

Materials

2.1 Equipment and Labware

1. Solvents and reagents of highest purity. 2. Pre-filtered solutions at 0.22 μm (for EV analysis). 3. Low-binding cryotubes. 4. EDTA-treated tubes for hematology. 5.

Ultra-low protein binding filters, 0.22 μm, for aqueous solutions.

6. Dulbecco’s Phosphate Buffered Saline (D-PBS). 7. Tubes for ultracentrifugation. 8. Lipid removal system (we classically use Cleanascite Lipid Removal Reagent and Clarification from Biotech Support Group). 9. 0.5-mL centrifugal filter units with 3-kDa MWCO cellulose membrane.

26

Annarita Farina

Fig. 1 Label-free approach. Crude samples of biological fluids collected from patients with pathologies selected for comparison (from 1 to 4) are subjected to differential centrifugation as follows: (1) 300  g for 10 min; (2) 16,000  g for 20 min; (3) 120,000  g for 70 min; (4) 200,000  g for 2 h. All the pellets and the final supernatants are then run on SDS-PAGE, up to 0.5 cm, and protein bands are cut and digested in-gel with trypsin. The resulting peptides are finally analyzed by LC-MS/MS

10. Depletion system for highly abundant serum proteins (we classically use the Proteome Purify 12 Human Serum Protein Immunodepletion Resin from R&D). 11. Refrigerated microcentrifuge. 12. Vacuum microcentrifuge.

Biological Fluid Pre-Fractionation for Biomarker Discovery

27

13. Protein Assay Kit compatible with samples containing reducing agents and detergents (we classically use the RC DC Protein Assay from Bio-Rad). 14. NuPAGE 4–12% Bis-Tris Protein Gels (ThermoFisher). 15. NuPAGE Antioxidant (ThermoFisher). 16. NuPAGE MES SDS Running Buffer (20) (ThermoFisher). 17. Ultrasonic bath or VialTweeter. 18. iTRAQ Reagents Multiplex Kit (SCIEX). 19. Microcapillary Pipet Tips. 20. Agilent 3100 OFFGEL Fractionator (Agilent Technologies). 21. 13 cm IPG strips pH 3–10. 22. Desalting devices (we classically use C18 Macro, Micro and Ultra-Micro SpinColumns from Harvard Apparatus) (see Note 1). 2.2

Solutions

2.2.1 Protein Denaturation

1. Denaturing solution: 2% SDS. 2. 1 Laemmli buffer: 62.5 mM Tris–HCl, pH 6.8, 1% SDS, 8.7% Glycerol, 1% 2-mercaptoethanol, 0.002% bromophenol blue. 3. 6 Laemmli buffer: 375 M Tris–HCl, pH 6.8, 6% SDS, 52.2% Glycerol, 6% 2-mercaptoethanol, 0.002% bromophenol blue.

2.2.2 SDS-PAGE

1. Upper chamber running buffer: 0.25% NuPAGE Antioxidant, 5% NuPAGE MES 20. 2. Lower chamber running buffer: 5% NuPAGE MES 20. 3. Staining solution: 0.1% Coomassie R-250, 50% methanol, 7.5% acetic acid. 4. Destaining solution: 30% methanol, 7.5% acetic acid; 5. Storage solution: 20% ethanol.

2.2.3 In-Gel Protein Destaining, Reduction, and Alkylation

1. Drying solution: 50 mM triethylammonium bicarbonate (TEAB), pH 8.0, 30% acetonitrile (ACN). 2. Reducing solution: 10 mM TCEP, 50 mM TEAB, pH 8.0. 3. Alkylating solution: 55 mM iodoacetamide, TEAB 50 mM, pH 8.0. 4. Digestion buffer: 50 mM TEAB, pH 8.0.

2.2.4 In-Gel Protein Deglycosylation

1. Deglycosylating solution: 0.025 U/μL PNGase F.

28

Annarita Farina

2.2.5 In-Gel Protein Digestion

1. Digestion buffer: 50 mM TEAB, pH 8.0; 2. Digestion solution: 6.25 ng/μL Trypsin, 50 mM TEAB, pH 8.0. 3. Extraction solution I: 1% FA; 4. Extraction solution II: 0.1% FA, 50% ACN.

2.2.6 Peptide Cleanup with C18 SpinColumns

1. Loading Solution: 0.1% FA, 5% ACN.

2.2.7 OFFGEL Peptide Fractionation

1. Resuspension solution: 1% SDS.

2. Elution buffer: 0.1% FA, 50% ACN.

2. Labeling buffer: 0.1 M TEAB, pH 8.0. 3. Dilution reagent: ultrapure 100% ethanol; 4. Stop solution: 5% hydroxylamine. 5. Peptide OFFGEL Stock Solution: 6% Glycerol, 0.75% IPG-Buffer, pH 3–10.

3

Methods

3.1 Best Practice for Body Fluid Collection

1. Preoperative collection: if possible, collect the fluid from a catheter already in place. Alternatively, draw with a syringe and discard the first aspirated volume (see Note 2).

3.1.1 Noncirculating Body Fluids

2. Intraoperative collection: collect the fluid directly from the body cavity (e.g., gallbladder, cyst, peritoneum) avoiding aspirating tissue debris. 3. Immediately aliquot into low-binding cryotubes and transport on dry ice. 4. Store at

3.1.2 Plasma

80  C until analysis.

1. Collect blood in EDTA-treated tubes for hematology. 2. Centrifuge twice at 2000  g for 15 min at 4  C. 3. Immediately aliquot the supernatant into low-binding cryotubes and transport on dry ice. 4. Store at

3.2 Proteome Sub-fractionation

80  C until analysis.

Dilute crude fluid samples 1:6 with pre-filtered (0.22 μm) D-PBS to reduce viscosity, and centrifuge as follows: 1. 300  g for 10 min at 4  C; transfer the supernatant into a fresh tube and centrifuge again under the same conditions. Transfer the final supernatant into a fresh tube and move to the next step. Resuspend both pellets with pre-filtered D-PBS, combine them, centrifuge again under the same conditions, and discard

Biological Fluid Pre-Fractionation for Biomarker Discovery

29

the supernatant. The resulting fraction will be regarded as the “300  g pellet” (see Note 3). 2. 16,000  g for 20 min at 4  C; transfer the supernatant into a fresh tube and centrifuge again under the same conditions. Transfer the final supernatant into an ultracentrifuge tube and move to the next step. Resuspend both pellets with pre-filtered D-PBS, combine them, centrifuge again under the same conditions, and discard the supernatant. The resulting fraction will be regarded as the “16,000  g pellet” (see Note 4). 3. 120,000  g for 70 min at 4  C. Transfer the supernatant into a fresh ultracentrifuge tube and move to the next step. Wash the pellet with pre-filtered D-PBS to remove residual supernatant and ultracentrifuge again under the same conditions. The resulting fraction will be regarded as the “120,000  g pellet” (see Note 5). 4. 200,000  g for 2 h at 4  C. Transfer the supernatant into a fresh tube. This fraction will be regarded as the “200,000  g supernatant”; Wash the pellet with pre-filtered D-PBS to remove residual supernatant, ultracentrifuge again under the same conditions, and discard the supernatant. The resulting fraction will be regarded as the “200,000  g pellet” (see Note 6). 5. Keep at 3.3 Decomplexification (Optional) and Volume Reduction of the “200,000  g Supernatant” Fraction

3.4 Protein Denaturation and Quantitation

80  C before use.

If the elimination of lipids or high-abundance serum proteins is necessary, the “200,000  g supernatant” can be treated with lipid removal (Cleanascite) or depletion (Proteome Purify 12 Human Serum Protein Immunodepletion) resins, respectively. Briefly, 1 mL of Cleanascite or Immunodepletion suspension can be added to the appropriate volume of supernatant, mixed on a rotary shaker for 30–45 min at 4  C, and the resin removed by filtration or centrifugation, according to the manufacturer’s instructions. Both the treated and untreated “200,000  g supernatant” may require a volume reduction if the protein dilution is too high. The ideal way to proceed in this case would be to reduce the volume to the original one, in order to have a direct correlation with the crude sample. For this purpose, the supernatant can be ultrafiltered by using an ultra-low protein binding filter centrifugal unit (see Note 7). 1. Apply 20 μL of Denaturing solution to the pellet fractions, incubate for 10 min at room temperature (RT), and resuspend with 100 μL of 1 Laemmli buffer.

30

Annarita Farina

2. Mix 100 μL of “200,000  g supernatant” with 20 μL of 6 Laemmli buffer. 3. Heat the tubes for 2 min at 95  C. 4. Centrifuge the tubes at 16,000  g for 5 min, transfer each supernatant in a fresh tube, and measure its volume (see Note 8). 5. Define protein concentration by using the RC DC Protein Assay, according to the manufacturer’s “Microfuge Tube Assay Protocol.” 3.5 SDS-PAGE and Gel Slicing

1. Load up to 35 μL of each sample onto a NuPAGE 4–12% Bis-Tris Protein Gel and perform electrophoretic migration in Running buffer until the migration front has reached about 0.5 cm (or more, if a second fractionation is needed) (see Note 9). 2. Incubate the gel in the Staining solution for 15 min. 3. Discard the Staining solution and incubate in the Destaining solution until the background noise has cleared. 4. Cut each gel lane in a single slice (or more, if a second fractionation is needed, see Note 10). 5. Put each slice in a microcentrifuge tube. 6. Chop each slice into smaller cubes (see Note 11). 7. If necessary, add an adequate volume of Storage solution, enough to completely cover each cut slice, and store overnight at 4  C.

3.6 In-Gel Protein Destaining, Reduction, and Alkylation

From now on, we suggest using microcapillary pipet-tips to remove solutions at the end of each step. 1. Destain and dehydrate gel pieces by incubating them in 400 μL of Drying solution for 15 min at RT. 2. Repeat the first step until all the gel pieces are completely clear. 3. Break disulfide bonds by incubating in 400 μL of Reducing solution for 35 min at 56  C. 4. Block free thiol groups by incubating in 400 μL of Alkylating solution for 30 min at RT in the dark. 5. Wash with 400 μL of Digestion buffer for 10 min at RT. 6. Dehydrate by incubating in 400 μL of Drying solution for 10 min at RT. 7. Dry by vacuum centrifugation (the duration of this step depends on the instrument). 8. Store at

80  C or go to the next section.

Biological Fluid Pre-Fractionation for Biomarker Discovery

3.7 In-Gel Protein Deglycosylation (Optional Step)

31

Deglycosylation is optional but should be used for fluids supposed to contain abundant glycosylated proteins. 1. Rehydrate with 80 μL of Deglycosylating solution for 30 min at 37  C. 2. Add 100 μL (or more, if the gel pieces are not completely covered) of Ultrapure water. 3. Incubate overnight at 37  C. 4. Centrifuge the tubes briefly to collect all possible condensate. 5. Add 800 μL of Ultrapure water. 6. Extract digested glycans by sonicating 20–30 min in a cold bath (or 0.5–1 min in a VialTweeter). 7. Discard the supernatant. 8. Repeat the extraction another 3 times. 9. Dry the gel pieces by vacuum centrifugation. 10. Store at

3.8 In-Gel Protein Digestion

80  C or go to the next section.

1. Rehydrate with 100 μL of Digestion solution for 45 min on ice. 2. Add 100 μL of Digestion buffer (or more, if the gel pieces are not completely covered). 3. Incubate overnight at 37  C. 4. Centrifuge the tubes briefly to collect all possible condensate. 5. Add 100 μL of Extraction solution I. 6. Extract tryptic peptides by sonicating 3 times 10–15 min in a cold bath (or 3 times 0.5 min in a VialTweeter). 7.

Transfer the acidic supernatant into low-protein-binding tubes.

8. Add 200 μL of Extraction solution II. 9. Extract remaining tryptic peptides by sonicating 3 times 10–15 min in a cold bath (or 3 times 0.5 min in a VialTweeter); 10. Pool the acidic supernatant with the one collected first. 11. Dry by vacuum centrifugation (the duration of this step depends on the instrument). 12. Store at 3.9 Label-Based Protein Quantitation

80  C or go to the next section.

For translational studies, we strongly suggest using a label-free approach because of reduced sample handling. If, for whatever reason, this is not possible, we defined an alternative label-based approach (Fig. 2). In this case, however, it is crucial to add an adequate exogenous spike-in standard into the samples

32

Annarita Farina

Fig. 2 Alternative label-based approach. Tryptic peptides obtained from the in-gel digestion of each centrifugal series of samples (i.e., 300  g pellets, 16,000  g pellets, 120,000  g pellets, 200,000  g pellets, and 200,000  g supernatants) (Fig. 1) are labeled with iTRAQ tags and mixed. Each peptide mixture is then fractionated by OFFGEL electrophoresis and the resulting 12 fractions collected and subjected to cleaning-up using C18 SpinColumns. Each peptide fraction is finally analyzed by LC-MS/MS

(preferentially before SDS-PAGE) in order to monitor and correct for technical variability introduced throughout the procedure. 3.9.1 iTRAQ Labeling

From now on, consider that a different set of iTRAQ reagents (composed of 2, 4, 8, or 10 tags) will be used to label each series of equivalent fractions obtained from the samples to be compared (e.g., the tags 114, 115, 116, and 117 could be used to label the “16.000  g pellet” fraction obtained from “Pathology n.1,” “Pathology n.2,” “Control n.1,” and “Control n.2,” respectively). 1. Dissolve tryptic peptides by adding 1 μL of Resuspension solution and then 24 μL of Labeling buffer. 2. Add 70 μL of Dilution reagent into each iTRAQ tag.

Biological Fluid Pre-Fractionation for Biomarker Discovery

33

3. Vortex for 1 min and centrifuge briefly. 4. Transfer each unique iTRAQ tag into a different tryptic sample of the series. 5. Incubate for 1 h at RT. 6. Interrupt the labeling reaction by adding 8 μL of Stop solution. 7. Incubate for 15 min at RT. 8. Transfer and combine all the samples in a fresh tube, together with 1–2 washes of the original tubes with 30 μL of Ultrapure water. 9. Dry by vacuum centrifugation (the duration of this step depends on the instrument). 10. Store at 3.9.2 Peptide Cleanup with C18 SpinColumns

80  C or go to the next section.

Dissolve Tryptic Peptides in 300 μL (Macro), 150 μL (Micro), or 50 μL (Ultra-Micro) (see Note 1) of Loading Solution and vortex widely. 1. Verify that the pH is 3; if necessary, adjust by adding 10% formic acid. 2. Condition dry columns by adding 150 μL (Macro), 75 μL (Ultra-Micro), or 25 μL (Ultra-Micro) of Elution Solution. 3. Wait 10 min at RT. 4. Centrifuge for 2 min at 1000  g. 5. Discard the supernatant. 6. Repeat steps 3–6 once. 7. Equilibrate by adding 150 μL (Macro), 75 μL (Ultra-Micro), or 25 μL (Ultra-Micro) of Loading Solution. 8. Centrifuge for 2 min at 1000  g. 9. Discard the supernatant. 10. Repeat step 8–10 once. 11. Load 150 μL (Macro), 75 μL (Ultra-Micro), or 25 μL (UltraMicro) of the sample. 12. Centrifuge for 2 min at 1000  g. 13. Discard the supernatant. 14. Repeat steps 12–14 as needed to load the entire sample. 15. Wash with 150 μL (Macro), 75 μL (Ultra-Micro), or 25 μL (Ultra-Micro) of Loading Solution. 16. Centrifuge for 2 min at 1000  g. 17. Discard the supernatant. 18. Repeat steps 16–18 once.

34

Annarita Farina

19. Elute by applying 150 μL (Macro), 75 μL (Ultra-Micro), or 25 μL (Ultra-Micro) of Elution Solution. 20. Centrifuge for 2 min at 1000  g. 21. Recover the eluate in a clean microcentrifuge tube. 22. Repeat steps 20–22 once. 23. Pool the recovered eluates. 24. Dry by vacuum centrifugation (the duration of this step depends on the instrument). 25. Store at 3.9.3 OFFGEL Peptide Fractionation

80  C or go to the next section.

1. Dissolve dried samples in 360 μL of Ultrapure water. 2. Add 1.44 mL of Peptide OFFGEL Stock Solution. 3. Assemble frames, electrodes and 13 cm IPG strips pH 3–10, according to the OFFGEL fractionator manual. 4. Distribute the prepared sample solution across the 12-well frame, 150 μL per well. 5. Set the parameters presented in Table 1 (corresponding to a modified version of the OFFGEL default method for peptides: OG12PE00). 6. Start the fractionation run. 7. At the end of the run, recover each fraction in a low-proteinbinding tube. 8. Measure the pH of each fraction (optional). 9. Dry by vacuum centrifugation (the duration of this step depends on the instrument). 10. Store at

3.9.4 Peptide Cleanup with C18 SpinColumns

80  C or go to the next section.

1. Refer to Subheading 3.9.2. 2. Samples are ready for analyses by nanoliquid chromatography coupled to tandem mass spectrometry.

Table 1 Parameters for the OFFGEL fractionation of peptides

Focusing Hold

Volt. hour (kVh)

Voltage (V)

Current (μA)

Power (mW)

Time (h:m)

50

8000

50

200

100:00

8000

50

200

Biological Fluid Pre-Fractionation for Biomarker Discovery

4

35

Notes 1. It is essential to adapt SpinColumn capacity to peptide quantities. The presented procedure refers to Macro, Micro, and Ultra-Micro SpinColumns (Harvard Apparatus) which allow loading 30–300 μg, 5–60 μg, and 3–30 μg of peptides, respectively. 2. Inserting needle through the body can cause sample contamination. 3. This fraction will predominantly contain whole cells and debris. If, for any reason, the sample is supposed to contain tissue contamination resulting from the collection method, we suggest removing this fraction from the analysis. 4. This fraction will preferentially contain high-density subcellular structures (e.g., mitochondria, lysosomes, peroxisomes, and large extracellular vesicles). 5. This fraction will preferentially contain low-density subcellular structures (e.g., small microvesicles and exosomes) and should be preferentially considered for the analysis of EVs. 6. This fraction will preferentially contain ribosomes and highdensity macromolecules. 7. It is strongly recommended that quantitative immunoassays be performed before and after each treatment to verify if they have introduced random changes in protein quantitation. 8. Keep in mind that this volume contains the pellet arising from the centrifugation of the whole fluid sample, while the supernatant corresponds to only a portion of the initial sample. When unfiltered and untreated supernatants are used, the original volume is approximately 16.7 μL (or else, 1/6) of the final withdrawn volume. When, instead, ultrafiltered and/or treated supernatants are used, the correspondence between the withdrawn and the original volume has to be determined on the basis of the final:crude volume ratio. 9. At this stage the investigator can choose to compare different (e.g., malignant vs. non-malignant) samples on the basis of equal protein amount (i.e., 20–25 μg) or equal volumes of original fluid. In the latter case, which is strongly recommended for translational studies, the pellet with the highest concentration should serve as a reference to determine the maximum amount of proteins to be loaded. The volume of other pellets will be adjusted accordingly. The volume of supernatants could not correspond to that of pellets. This will not affect the analysis, because the comparison has to be made between similar fractions of different samples. In this case, however, if desired, an adequate normalization procedure could be applied.

36

Annarita Farina

10. Cut larger, strongly stained protein bands distinct from thinner and faint ones. Use scalpel with removable sterile blade: one blade per sample, washed with 20% ethanol between each slice. Optionally, mark onto a photographic reproduction of the gel the borders of each slice and use the molecular weight markers, if any, as a reference for identified proteins. 11. Perform the manipulation onto a large and protein-free surface in order to be able to recollect fragments eventually “jumped” out of the tube. This could happen due to the elasticity of the gel. Also, pay attention not to produce fragments so small to be aspirated by pipet tips during the subsequent manipulations.

Acknowledgment The author sincerely thanks all the coworkers who, over the years, have substantially contributed in the acquisition of the know-how required for the development of the present protocols, notably: Dr. Yohann Coute´ (Exploring the Dynamics of Proteomes laboratory at the CEA, Grenoble, France) for the methodological advices about each phase of protein identification and quantitation; Dr. Valeria Severino (Digestive Cancers Biomarkers Group of the Medicine Faculty at the Geneva University, Geneva, Switzerland) for the improvement of protein sub-fractionation protocols; The Proteomic Core Facility (Medicine Faculty at the Geneva University, Geneva, Switzerland) for the adaptation of peptide cleanup methods; Dr. Pierre Lescuyer, (Department of Genetic, Laboratory and Pathology Medicine at the Geneva University Hospitals, Geneva, Switzerland) for crucial knowledge in clinical laboratory requirements; Dr. Jean-Marc Dumonceau (Gedyt Center, Argentina, Buenos Aires) for essential expertise in sample inclusion and collection; Prof. Jean-Louis Frossard (Gastroenterology and Hepatology Service at the Geneva University Hospitals, Geneva, Switzerland) for strong competences in digestive pathologies, constructive criticisms, and continuous help and support; all the trainee students involved in the analyses that allowed assembling and validating the protocols. References 1. Fuzery AK, Levin J, Chan MM et al (2013) Translation of proteomic biomarkers into FDA approved cancer diagnostics: issues and challenges. Clin Proteomics 10(1):13. https:// doi.org/10.1186/1559-0275-10-13 2. Pepe MS, Li CI, Feng Z (2015) Improving the quality of biomarker discovery research: the

right samples and enough of them. Cancer Epidemiol Biomark Prev 24(6):944–950. https://doi.org/10.1158/1055-9965.EPI14-1227 3. Farina A (2014) Proximal fluid proteomics for the discovery of digestive cancer biomarkers. Biochim Biophys Acta 1844(5):988–1002.

Biological Fluid Pre-Fractionation for Biomarker Discovery https://doi.org/10.1016/j.bbapap.2013.10. 011 4. Dakappagari N, Zhang H, Stephen L et al (2017) Recommendations for clinical biomarker specimen preservation and stability assessments. Bioanalysis 9(8):643–653. https://doi.org/10.4155/bio-2017-0009 5. Percy AJ, Parker CE, Borchers CH (2013) Pre-analytical and analytical variability in absolute quantitative MRM-based plasma proteomic studies. Bioanalysis 5(22):2837–2856. https://doi.org/10.4155/bio.13.245 6. Baek R, Sondergaard EKL, Varming K et al (2016) The impact of various preanalytical treatments on the phenotype of small extracellular vesicles in blood analyzed by protein microarray. J Immunol Methods 438:11–20. https://doi.org/10.1016/j.jim.2016.08.007 7. Yuana Y, Bertina RM, Osanto S (2011) Pre-analytical and analytical issues in the analysis of blood microparticles. Thromb Haemost 105(3):396–408. https://doi.org/10.1160/ Th10-09-0595 8. Jambunathan K, Galande AK (2014) Sample collection in clinical proteomics-Proteolytic activity profile of serum and plasma. Proteomics Clin Appl 8(5–6):299–307. https://doi. org/10.1002/prca.201300037 9. Chutipongtanate S, Chatchen S, Svasti J (2017) Plasma prefractionation methods for

37

proteomic analysis and perspectives in clinical applications. Proteomics Clin Appl 11(7–8). https://doi.org/10.1002/prca.201600135 10. Ivanov AR, Lazarev A (eds) (2011) Sample preparation in biological mass spectrometry. Springer, Dordrecht. https://doi.org/10. 1007/978-94-007-0828-0 11. Wisniewski JR, Wegler C, Artursson P (2016) Subcellular fractionation of human liver reveals limits in global proteomic quantification from isolated fractions. Anal Biochem 509:82–88. https://doi.org/10.1016/j.ab.2016.06.006 12. Lukic N, Visentin R, Delhaye M et al (2014) An integrated approach for comparative proteomic analysis of human bile reveals overexpressed cancer-associated proteins in malignant biliary stenosis. Biochim Biophys Acta 1844 (5):1026–1033. https://doi.org/10.1016/j. bbapap.2013.06.023 13. Farina A, Dumonceau JM, Delhaye M et al (2011) A step further in the analysis of human bile proteome. J Proteome Res 10 (4):2047–2063. https://doi.org/10.1021/ pr200011b 14. Choksawangkarn W, Edwards N, Wang Y et al (2012) Comparative study of workflows optimized for in-gel, in-solution, and on-filter proteolysis in the analysis of plasma membrane proteins. J Proteome Res 11(5):3030–3034. https://doi.org/10.1021/pr300188b

Chapter 3 Serum Exosome Isolation by Size-Exclusion Chromatography for the Discovery and Validation of Preeclampsia-Associated Biomarkers Rosana Navajas, Fernando J. Corrales, and Alberto Paradela Abstract Exosomes are extracellular nanovesicles of complex and heterogeneous composition that are released in biofluids such as blood. The interest in the characterization of exosomal biochemistry has increased over the last few years as they convey cellular proteins, lipids, and RNA that might reflect the biological or pathological condition of the source cell. In particular, association of changes of exosome proteins with specific pathogenic processes arises as a promising method to identify disease biomarkers as for the pregnancy-related preeclampsia. However, the overlapping physicochemical and structural characteristics of different types of extracellular vesicles have hindered the consolidation of universally accepted and standardized purification or enrichment protocols. Thus, it has been recently demonstrated that the exosomal protein profile resulting from in-depth proteomics analyses is highly dependent on the preparation protocol used, which will determine the particle type specificity and the presence/absence of contaminating proteins. In this chapter, an isolation method of serum exosomes based on size-exclusion chromatography (SEC) using qEV columns (Izon) is described. We show that this method is fast and reliable, as the population of exosomes isolated is homogeneous in terms of size, morphology, and protein composition. This exosome enrichment method is compatible with downstream qualitative and quantitative proteomic analysis of the samples. Key words Biomarkers, Serum/plasma, Exosomes, Size-exclusion chromatography, Shotgun proteomics, Biomedicine, Preeclampsia

1

Introduction Biomarkers are parameters, including organic molecules of varied nature (lipids, peptides, nucleic acids, proteins) whose presence or absence may be unambiguously associated to specific biological processes or disease stages, enabling their study from a quantitative and/or qualitative perspective. In biomedicine (including veterinary sciences), biomarkers are extremely valuable to characterize physiological and pathological processes and, ideally, for the

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_3, © Springer Science+Business Media, LLC, part of Springer Nature 2019

39

40

Rosana Navajas et al.

diagnosis and prognosis of human (animal) diseases. Description and characterization of early biomarkers is of special interest, since it would enable to detect pathologies at the initial stages when they are frequently asymptomatic (e.g., cancer), facilitating timely treatment and thus increasing the rate of therapeutic success [1, 2]. The best choice of biomarkers source are biofluids, since their noninvasive access might facilitate the rapid development and transfer of targeted detection methods to the clinical routine. Serum and plasma have been commonly used since, in addition to their easy accessibility and availability, they can be stored and preserved for long periods without significant negative effects in the quality of the sample [3]. It is of interest to note that blood irrigates the whole body and collects information from very diverse sources and backgrounds that makes it ideal for recapitulating biological processes and pathologies. However, from a strictly proteomic point of view, the study of serum/plasma has some disadvantages, as the overwhelming presence of a reduced subset of prevalent proteins (albumin, IgG, transferrin, etc.). These proteins encompass more than 95% of the serum/plasma proteinaceous material [4], and represent a first magnitude analytical problem, affecting significantly any in-depth proteomic analysis. There are several ways to overcome or minimize this problem, such as exhaustive sample pre-fractionation before the proteomic analysis [5, 6] or depletion of the most abundant proteins using immunoaffinity columns [7]. In addition to soluble proteins, complex macromolecular structures of varied morphology and composition, such as the microvesicles and exosomes, have been described in serum and plasma. These structures are produced inside the cell and differ in size, which typically is in the 50 nm–1 μm range, as well as in the molecular composition, that includes proteins, lipids, and nucleic acids (miRNA) [8]. It is generally considered that the molecular composition of extracellular vesicles reflects the physiological and functional state of the secreting cell and that they act as vehicles for intercellular communication. Altogether, these features imply that extracellular vesicles, and more specifically exosomes, could be an excellent source of biological material for the identification of biomarkers of different chemical composition, including proteins [9, 10]. In addition, focusing proteomics analyses in plasma/ serum-purified exosomes could overcome the aforementioned analytical problem posed by major plasma proteins. However, overlapping physicochemical and structural characteristics existing between the different types of extracellular vesicles, especially in view of their size and morphology, have hindered to date the consolidation of universally accepted and standardized purification or enrichment methods in the scientific community. Most purification protocols described so far include differential centrifugation, precipitation, and molecular exclusion, with minor modifications [11]. Yet, a series of recently published works have demonstrated

Serum Exosome Isolation by Size-Exclusion Chromatography

41

that the profile of exosomal proteins identified is highly dependent on the protocol used, with a great variability in the results obtained in relation to the morphology, presence/absence of contaminating proteins, yield, and identified proteins profile. Thus, it has been demonstrated that purification of serum/plasma exosomes by means of size-exclusion microcolumns produces preparations of high purity [12]. In addition, although the yield is typically lower than with other methods (e.g., precipitation), the presence of serum main contaminant proteins was significantly lower. In this manuscript, we describe a protocol based on such sizeexclusion principle for the purification of exosomes from human serum, in order to identify, characterize, and quantify potential biomarkers associated with preeclampsia. An overview of our strategy is summarized in Fig. 1. Preeclampsia is a pregnancy-related pathology that affects to 4–6% of the pregnancies worldwide, and is a major cause of maternal and fetal morbidity and mortality. Preeclampsia characteristic clinical symptoms are hypertension and proteinuria, which are detected around weeks 16–20 (early preeclampsia) or in later stages. In addition, other symptoms may also

Isolation of serum exosomes (SEC) Exosome characterization

TEM NTA

Protein extraction and digestion

SAMPLE PREPARATION AND CHARACTERIZATION

Biomarker Discovery Untargeted MS-based Proteomics

Biomarker Validation Targeted MS-based Proteomics Sample preparation and characterization

+

Differential proteins Protein1 Protein2 Protein3 Protein4 Protein15 Protein16

y3

Protein5 Protein6 Protein7 Protein8 Protein20 Protein22

Proteotypic peptides selection

y6 y4

Transition selection

Peptide separation

RP-LC

RP-LC

Peptide separation

Shotgun-MS SRM-MS y6

y3 y4 y5

y7

Peptide/protein identification (MS2) Peptide/protein label free quantification (MS1)

SKYLINE

Relative quantification Analysis (MS2) Statistical significance analysis of putative differential biomarkers

Fig. 1 General workflow of shotgun and targeted proteomics strategies on serum exosome samples for preeclampsia-associated biomarker discovery and validation

42

Rosana Navajas et al.

appear in different individuals, such as fuzzy vision, edema, renal dysfunction, and intrauterine growth restriction [13]. Several studies have shown that the abnormal interaction between different circulating factors and the maternal endothelium might be at the onset of the disease [14]. Some protein biomarkers, such as soluble endoglin [15] and sFlt1 [16], have been described as blood biomarkers for this disease and are used in the daily clinical practice although results are not unequivocal [17]. Thus, description of new preeclampsia-associated biomarkers would be extremely useful, especially for the diagnosis in the early stages of the disease.

2

Materials

2.1 Exosome Isolation by SizeExclusion Chromatography (SEC)

1. Prepare 1 PBS (phosphate-buffered saline) pH 7.4 solution. 2. Prepare 10% sodium azide (w/v). 3. Sterile syringe filter, pore size 0.2 μm, surfactant-free cellulose acetate. 4. qEV Size Exclusion New Zealand).

2.2 Protein Lysis, Precipitation, and Quantification

Columns

(Izon,

Christchurch,

1. RIPA lysis buffer: 150 mM NaCl, 1% (v/v) Triton X-100, 0.5% (w/v) sodium deoxycholate, 0.1% (w/v) SDS, 50 mM Tris-Cl. 2. Broad spectrum cysteine and serine proteases inhibitor cocktail: complete mini, EDTA-free (Roche, Basel, Switzerland). 3. Chloroform/methanol precipitation: reagent grade chloroform and methanol; ultrapure water (resistivity at 25  C > 18.18 MΩ cm). 4. Pierce 660 nm protein assay (Thermo Fischer Scientific, Massachusetts, USA) for determination of protein concentration.

2.3 In-Solution Digestion

1. Digestion buffer solution: 7 M urea, 2 M thiourea, 100 mM triethylammonium bicarbonate (TEAB). Store in aliquots at 20  C. 2. Reduction solution: Prepare 50 mM tris(2-carboxyethyl) phosphine (TCEP) solution from 10 stock 500 mM TCEP. 3. Alkylation reagent: 200 mM methyl methanethiosulfonate (MMTS). Store in aliquots at 20  C. 4. 0.5 M TEAB stock solution. Store in aliquots at

20  C.

5. Mass Spectrometry (MS) grade modified porcine trypsin (Pierce, Thermo Fischer Scientific). 6. Reversed phase (RP) C18 tips (Millipore, Massachusetts, USA).

Serum Exosome Isolation by Size-Exclusion Chromatography

2.4 Analysis by Liquid Chromatography Coupled to Tandem Mass Spectrometry (LC-MS/MS System)

43

1. Eksigent Technologies nanoLC Ultra 1D plus (Sciex, Massachusetts, USA). 2. LC-MS grade water, acetonitrile, and formic acid. 3. Mobile phase (loading pump): 0.1% formic acid in water. 4. Mobile phases (nano pump): (a) A: 0.1% formic acid in water. (b) B: 0.1% formic acid in acetonitrile. 5. Distal coated emitter tips (New Objective, Massachusetts, USA). 6. Shotgun analysis: (a) Analytical column: reversed phase C18 nanoAcquity BEH analytical column, 1.7 μm particle size, 100 A˚ pore size, 75 μm I.D.  15 cm (Waters, Massachusetts, USA). (b) Trapping column: C18 PepMap, 5 μm particle size, 100 μm I.D.  2 cm (Thermo Fischer Scientific, Massachusetts, USA). (c) 5600 Triple TOF Massachusetts, USA).

mass

spectrometer

(Sciex,

(d) Licensed version (v. 2.6.1) of the Mascot search engine (Matrix Science, London, UK). 7. Targeted analysis: (a) Analytical column: reversed phase, C18 BioSphere 3 μm particle size, 120 A˚ pore size, 75 μm I.D.  15 cm (Nanoseparations, Nieuwkoop, Netherlands). (b) Trapping column: C18 PepMap, 5 μm particle size, 100 A˚

pore size, 300 μm I.D.  5 cm (Thermo Fischer Scientific, Massachusetts, USA).

(c) 5500 QTRAP triple quadrupole mass spectrometer (Sciex, Massachusetts, USA). (d) Data analysis by Skyline (v. 3.7.1): Skyline is a freely available and open source Windows client application for building targeted (SRM/MRM/PRM), Data Independent Acquisition (DIA/SWATH) and targeted DDA with MS1 quantitative methods and analyzing the resulting mass spectrometer data. It can be freely downloaded from https://skyline.ms/project/home/begin.view.

3

Methods

3.1 Exosome Isolation by SizeExclusion Chromatography Using Izon qEV Columns

1. Filter 1 PBS mobile phase solution by pushing with a 10 mL syringe and passing through 0.2 μm sterile syringe filter. Degas the solution for 10 min using an ultrasonic bath (see Note 1). 2. Serum preparation: Thaw frozen serum samples on ice. Then, centrifuge 500 μL of serum sample at 1500  g and 10,000  g

44

Rosana Navajas et al.

at 4  C, for 10 and 20 min, respectively, and take carefully the supernatant (see Note 2). 3. Place the qEV column in a holder in vertical position with both bottom and end caps, and keep it at room temperature for at least 30 min to guarantee that the packing material has reached the operational temperature (15–25  C). 4. Column equilibration: (a) Carefully remove the top cap. (b) Remove the bottom luer-slip cap and rinse the column with 10–15 mL of 1 PBS (see Note 3). (c) Write down the time period required for 5 mL of PBS buffer to flow through the column (see Note 4). 5. Sample fraction collection: (a) Place the bottom cap on, remove with a pipette the remaining buffer from the top and transfer the serum sample onto the column. (b) Remove the bottom cap and let the sample to pass through the column filter before adding more PBS buffer. Add 20 mL of PBS buffer in multiple 1–2 mL aliquots. (c) Collect 0.5 mL fractions: First 3 mL (6 fractions) correspond to column void volume and are discarded. Exosomes will elute in fractions 7–9 (total volume of 1.5 mL). Most serum soluble proteins will elute from fraction 10–40 (see Note 5). (d) Save a 20 μL aliquot for exosome characterization by Transmission Electron Microscopy (TEM) and/or Nanoparticle Tracking Analysis (NTA) (see Note 6; Fig. 2). 6. Concentrate the exosomal fraction (1.5 mL) to a final volume of approximately 50 μL in the Speed-Vac at room temperature. 3.2 Protein Lysis, Precipitation, and Quantification

1. Protein lysis. (a) Add 100 μL of RIPA lysis buffer containing protease inhibitors to 50 μL of exosomal fraction, and mix thoroughly by pipetting up and down several times. (b) Incubate at 95  C for 10 min and mix again thoroughly with the pipette until complete homogenization is achieved. Incubate on ice for 5 min. (c) Sonicate in a water bath 4 for 10 s each time, at room temperature, placing the tube on ice between each sonication step. Mix thoroughly after the last step. Incubate on ice for 5 min (see Note 7).

Serum Exosome Isolation by Size-Exclusion Chromatography

45

Fig. 2 Serum exosomes were isolated by size-exclusion chromatography and characterized by Nanoparticle Tracking Analysis (NTA) and Transmission Electron Microscopy (TEM). (a) Representative NTA profile of exosome-sized vesicles, and (b) TEM image of extracellular vesicles of exosomal size and morphology

46

Rosana Navajas et al.

(d) Centrifuge at 12,000  g for 10 min at 4  C. When the centrifugation is over, save the supernatant (100–120 μL) and discard the pellet (see Note 8). 2. Protein precipitation. (a) Add the following solvents sequentially, vortexing in between: 4 volumes of methanol, 1 volume of chloroform, and 3 volumes of Milli-Q water (see Note 9). (b) Centrifuge at 15,000  g for 10 min at 4  C (see Note 10), and remove carefully the aqueous top layer. Avoid disturbing the interface. (c) Add 4 volumes of methanol to the remaining volume in the tube and vortex well (see Note 11). After a centrifugation step for 10 min at 15,000  g, a protein pellet should be visible. Remove carefully as much liquid as possible without disturbing the pellet, and dry under soft conditions, such as 20 min evaporation at room temperature. 3. Protein quantification. (a) Dissolve the exosomal protein fraction in 20 μL of 7 M urea, 2 M thiourea, 100 mM TEAB. Vortex well to ensure that solubilization is complete. (b) Determine protein amount by Pierce 660 nm Assay in a microplate format, according to manufacturer’s instructions (see Note 12). 3.3 In-Solution Digestion and Peptide Purification

1. Add tris-(2-carboxyethyl) phosphine (TCEP) reduction solution to a final concentration of 5 mM and incubate at 37  C for 1 h (see Note 13). 2. For alkylation of the reduced thiol groups, add 200 mM methylmethane-thiosulfonate (MMTS) solution to a final concentration of 20 mM, and incubate the sample at room temperature for 10 min. 3. Dilute the sample fourfold with 25 mM triethylammonium bicarbonate (TEAB) and add trypsin in a 1:20 protease-toprotein ratio (see Note 14). Incubate overnight at 37  C. After digestion, dry the sample in a speed-vacuum device. 4. Remove salts and other putative interfering compounds from peptide mixture by using RP-C18 tips, according to manufacturer’s instructions (see Note 15).

3.4 Tandem Mass Spectrometry Analysis

A bottom-up proteomic analysis strategy is followed [18]. First of all, peptide mixture is on-line desalted via trapping column to preserve the analytical column from residual salts and lipids. A sample volume of 5 μL (containing 1 μg of tryptic peptides) was injected onto the LC-MS/MS system. The effluent from the

Serum Exosome Isolation by Size-Exclusion Chromatography

47

chromatographic system was connected to distal coated emitter tips via a nano-electrospray ion source operating in positive ion mode. 1. Chromatographic conditions: (a) 100 min gradient from 2% to 40% B. (b) Flow rate: 250 nL/min. (c) Oven temperature: 50  C. 2. Mass spectrometer conditions. (a) Biomarker discovery by shotgun proteomics analysis: l

Data-dependent acquisition mode.

l

MS acquisition parameters per cycle: 10 precursor ions with an accumulation time of 250 ms in MS1 and 100 ms per precursor ion in MS2 (see Note 16).

l

Data analysis: MS and MS/MS raw data were processed using Analyst TF version 1.7, converted to mascot generic format (mgf file), and used for the search against a human sequence reference proteome database downloaded from Uniprot Knowledgebase (http:// www.uniprot.org/proteomes/UP000005640), using an in-house licensed Mascot Server v. 2.6.1. Results were filtered at the peptide level to a false discovery rate of FDR  1% (see Note 17).

(b) Biomarker verification/validation by targeted proteomics analysis:

4

l

Selective/multiple acquisition mode.

monitoring

(SRM/MRM)

l

3–4 proteotypic peptides per protein and 3–4 transitions per peptide (see Note 18).

l

Dwell times of 20–25 ms per transition to achieve MS cycles of 2.8–3.5 s.

l

Data analysis by Skyline v. 3.7.1.

Notes 1. The use of fresh filtered buffer is always recommended to avoid contamination. 2. Both centrifugation steps are necessary to remove cellular debris and large vesicles. 3. It is always essential to keep the column soaked in PBS buffer as dry columns would not work properly. A bacteriostatic agent, such as 0.04% (w/v) sodium azide, should be added to PBS solution for an appropriate medium and long-term column storage.

48

Rosana Navajas et al.

4. Elution time through the column for 5 mL of PBS is usually around 5.00  0.25 min (22–25  C). Longer elution times after reusing the column suggest that a cleaning step with 0.5 M NaOH solution is needed. qEV columns are intended to be used 5 times. 5. Exosome-containing fractions should be collected in 2 mL microtubes with labeled scale for eluate volume measurement. Nonspecific binding of proteins to the surface of the tube can be minimized by using low-binding or other high-quality material. This drawback is especially significant at low protein concentrations as is the case of exosomal serum samples. 6. Both techniques, TEM and NTA, are recommended to assess the quality and efficacy of exosomal isolation. TEM characterizes exosomes according to their size and morphology, while NTA provides size distribution and concentration of isolated vesicles/particles. 7. Harsh lysis conditions are required to disrupt exosome vesicles and to extract proteins more efficiently. 8. Centrifugation of the lysate at 12,000  g allows the removal of insoluble material such as exosome membranes. 9. One volume refers to the volume of supernatant obtained in the previous centrifugation step. 10. Proteins should be found at the interface, although at low concentrations they could not be clearly appreciated. 11. In most cases, turbidity is apparent in the solution, reflecting the presence of denatured proteins. After centrifugation, these proteins will precipitate. 12. Pierce 660 nm Assay is especially useful for protein quantification due to its compatibility with high concentrations of chaotropic agents. This will allow subsequent in-solution digestion using the same mixture buffer. Other alternative methods for protein quantification, such as micro BCA (bicinchoninic acid), Bradford or Lowry assays can also be used. 13. When urea buffer is used, incubation at higher temperatures is not recommended to prevent protein carbamylation. 14. Urea concentration has to be diluted fourfold for optimal trypsin activity. In addition, make sure that the sample pH is approximately 7–8. 15. Peptide desalting can be performed using in-house columns where reversed phase medium is packed in a 200-μL pipette tip containing a filter of appropriate pore size (10–35 μm), according to resin particle diameter. 16. Higher scan rates are usually set for high throughput protein identification by shotgun proteomics analysis, and differ from

Serum Exosome Isolation by Size-Exclusion Chromatography

49

the parameters stated in this chapter that are more suitable for label-free quantification. In label-free quantitative proteomics, it is extremely important to acquire enough data points to appropriately define elution profiles of peptide precursor ions, as quantification is carried out at MS1 level. 17. For the quality assessment of exosome isolates, a set of protein markers included in ExoCarta (http://www.exocarta.org) and EVpedia (http://www.evpedia.info) repositories, and thus considered as characteristics of exosomal origin, should be identified. 18. Theoretically, every possible peptide sequence generated after protease digestion of a given protein is amenable to detection by mass spectrometry. In practice, only a subset of peptides per protein is detected and identified by mass spectrometry. Proteotypic peptides are specific peptide sequences for a single protein that identify unambiguously that protein, and that can be consistently detected by tandem mass spectrometry. In addition to these requirements, proteotypic peptides should fulfill other essential criteria, such as length (9–20 amino acids) or that their sequences do not contain amino acids susceptible to chemical modification (Cys and Met) [19].

Acknowledgments CNB-CSIC lab is a member of Proteored, PRB2-ISCIII and is supported by grant PT13/0001, of the PE I + D + i 2013–2016, funded by ISCIII and FEDER. We thank the technical staff of the CNB-CSIC electron microscopy facility for advice and technical expertise. References 1. Gerszten RE, Wang TJ (2008) The search for new cardiovascular biomarkers. Nature 451 (7181):949–952. https://doi.org/10.1038/ nature06802 2. Hanash SM, Pitteri SJ, Faca VM (2008) Mining the plasma proteome for cancer biomarkers. Nature 452(7187):571–579. https://doi.org/10.1038/nature06916 3. Mateos J, Carneiro I, Corrales F et al (2017) Multicentric study of the effect of pre-analytical variables in the quality of plasma samples stored in biobanks using different complementary proteomic methods. J Proteome 150:109–120. https://doi.org/10.1016/j. jprot.2016.09.003

4. Anderson NL, Anderson NG (2002) The human plasma proteome: history, character, and diagnostic prospects. Mol Cell Proteomics 1(11):845–867. https://doi.org/10.1074/ mcp.R200007-MCP200 5. Cao Z, Tang HY, Wang H et al (2012) Systematic comparison of fractionation methods for in-depth analysis of plasma proteomes. J Proteome Res 11(6):3090–3100. https://doi.org/ 10.1021/pr201068b 6. Faca V, Pitteri SJ, Newcomb L et al (2007) Contribution of protein fractionation to depth of analysis of the serum and plasma proteomes. J Proteome Res 6(9):3558–3565. https://doi.org/10.1021/pr070233q

50

Rosana Navajas et al.

7. Polaskova V, Kapur A, Khan A et al (2010) High-abundance protein depletion: comparison of methods for human plasma biomarker discovery. Electrophoresis 31(3):471–482. https://doi.org/10.1002/elps.200900286 8. Mathivanan S, Ji H, Simpson RJ (2010) Exosomes: extracellular organelles important in intercellular communication. J Proteome 73 (10):1907–1920. https://doi.org/10.1016/j. jprot.2010.06.006 9. Arbelaiz A, Azkargorta M, Krawczyk M et al (2017) Serum extracellular vesicles contain protein biomarkers for primary sclerosing cholangitis and cholangiocarcinoma. Hepatology 66(4):1125–1143. https://doi.org/10.1002/ hep.29291 10. Boukouris S, Mathivanan S (2015) Exosomes in bodily fluids are a highly stable resource of disease biomarkers. Proteomics Clin Appl 9 (3-4):358–367. https://doi.org/10.1002/ prca.201400114 11. Sodar BW, Kovacs A, Visnovitz T et al (2017) Best practice of identification and proteomic analysis of extracellular vesicles in human health and disease. Expert Rev Proteomics 14 (12):1073–1090. https://doi.org/10.1080/ 14789450.2017.1392244 12. Lobb RJ, Becker M, Wen SW et al (2015) Optimized exosome isolation protocol for cell culture supernatant and human plasma. J Extracell Vesicles 4:27031. https://doi.org/ 10.3402/jev.v4.27031 13. Goulopoulou S, Davidge ST (2015) Molecular mechanisms of maternal vascular dysfunction in preeclampsia. Trends Mol Med 21 (2):88–97. https://doi.org/10.1016/j. molmed.2014.11.009

14. Powe CE, Levine RJ, Karumanchi SA (2011) Preeclampsia, a disease of the maternal endothelium: the role of antiangiogenic factors and implications for later cardiovascular disease. Circulation 123(24):2856–2869. https://doi. org/10.1161/CIRCULATIONAHA.109. 853127 15. Venkatesha S, Toporsian M, Lam C et al (2006) Soluble endoglin contributes to the pathogenesis of preeclampsia. Nat Med 12 (6):642–649. https://doi.org/10.1038/ nm1429 16. Maynard SE, Min JY, Merchan J et al (2003) Excess placental soluble fms-like tyrosine kinase 1 (sFlt1) may contribute to endothelial dysfunction, hypertension, and proteinuria in preeclampsia. J Clin Invest 111(5):649–658. https://doi.org/10.1172/JCI17189 17. Kleinrouweler CE, Wiegerinck MM, Ris-Stalpers C et al (2012) Accuracy of circulating placental growth factor, vascular endothelial growth factor, soluble fms-like tyrosine kinase 1 and soluble endoglin in the prediction of pre-eclampsia: a systematic review and metaanalysis. BJOG 119(7):778–787. https://doi. org/10.1111/j.1471-0528.2012.03311.x 18. Aebersold R, Mann M (2016) Massspectrometric exploration of proteome structure and function. Nature 537 (7620):347–355. https://doi.org/10.1038/ nature19949 19. Lange V, Picotti P, Domon B et al (2008) Selected reaction monitoring for quantitative proteomics: a tutorial. Mol Syst Biol 4:222. https://doi.org/10.1038/msb.2008.61

Chapter 4 Protein Biomarker Discovery Using Human Blood Plasma Microparticles Raghda Saad Zaghloul Taleb, Pacint Moez, Doreen Younan, Martin Eisenacher, Matthias Tenbusch, Barbara Sitek, and Thilo Bracht Abstract Cells shed into the extracellular space a population of membranous vesicles of plasma membrane origin called microparticles (MP). Given the fact that MP are abundantly present in body fluids including plasma, rich in cell-type or disease-specific proteins and formed in conditions of stress and injury, they have been extensively investigated as biomarkers in various diseases. With the advancement in the mass spectrometrybased proteome analysis, the knowledge of the protein composition of plasma MP (PMP) has been intensively expanded, which aids the discovery of novel diagnostic target proteins. However, the lack of standardized and accurate protocols for PMP isolation limits the implementation of PMP as biomarkers in clinical settings. Here, we describe in detail a robust protocol for PMP isolation from human blood plasma via ultracentrifugation followed by label-free quantitative proteome analysis of PMP. Key words Plasma microparticles, Ultracentrifugation, Label-free proteome quantification, Plasma microparticle proteome, Blood-based biomarker

1

Introduction Microparticles (MP) are a heterogeneous population of circulating membrane-enclosed vesicles that are 100–1000 nm in diameter [1]. They are shed directly from the plasma membrane into the extracellular space by budding as phospholipid vesicles (exocytosis) and retain the antigens of their cells of origin [2]. MP are released from cells under stress conditions; however, apoptosis/cell death is not the solitary prerequisite for MP formation as cellular activation also induces MP shedding [3]. The composition and cargo of MP is not only dependent on the parental cell of origin but also on the stimulus or the agonist responsible for the induction of MP formation [4]. MP are found in plasma and other biological fluids from healthy and diseased individuals and their levels reflect the underlying illness, such as cardiovascular diseases, diabetes, chronic kidney

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_4, © Springer Science+Business Media, LLC, part of Springer Nature 2019

51

52

Raghda Saad Zaghloul Taleb et al.

disease, and other medical conditions [5]. Accordingly, MPs were considered as potential circulating biomarkers of endothelial dysfunction, coagulation, inflammation, and other pathological processes. In addition to their role as circulating biomarkers, MP are suggested to exert biological effects on their own and contribute to the pathogenesis of various diseases. They represent an important mode of intercellular communication and a vehicle for transfer of biological signals between different cells [6]. Given the importance of MP as potential biomarkers and mediators of pathology in numerous diseases, many studies have examined MP isolated from biological fluids where the majority focused on plasma MP (PMP). The main challenge in studying PMP is the lack of standardization of pre-analytical and analytical issues related to MP analysis. Ideal PMP analysis should be performed with freshly prepared plasma directly after blood withdrawal [7]. Based on the physical and the antigenic properties of PMP, several approaches have been proposed for purification and characterization of PMP. Direct measurement without prior enrichment of PMP is preferred to prevent PMP loss during isolation and to preserve their characteristics [8]. For this purpose, flow cytometry and solid phase capture-based assays can be performed [8, 9]. PMP measurement with prior isolation, such as fluorescence or electron microscopy, proteome analysis and some functional PMP assays has the advantage of increased PMP concentration and less interference with plasma proteins. On the other hand, they have the disadvantage of the inevitable loss of PMP during isolation owing to the multiple washing steps [10–14]. Mass spectrometry (MS)-based proteome analysis represents a promising approach to globally examine the protein composition of PMP and the changes in different underlying pathological conditions [15]. In general, this approach requires a prior step of PMP isolation and tryptic protein digestion followed by MS analysis, protein identification, and quantification. PMP isolation by simple ultracentrifugation of plasma samples results in 1500-fold enrichment of PMP proteins [15, 16]. However, the high molecular weight and the hydrophobic nature of cell-surface antigens necessitate the use of ionic/non-ionic detergents or high pH solvents for the successful extraction of membrane proteins [16]. Then, the powerful combination of chromatographic separation and MS analysis is used to characterize and quantify PMP proteins [7]. In this chapter, we describe a simple and reproducible protocol for PMP isolation from human platelet-poor plasma (PPP) via ultracentrifugation for label-free liquid chromatography tandem mass spectrometry (LC-MS/MS)-based quantification of PMP proteins (Fig. 1).

Plasma Microparticle Isolation and Proteome Analysis

53

Fig. 1 Schematic representation of the applied workflow

2

Materials Prepare all reagents using ultrapure water (Resistivity at 25
C > 18.18 MΩ.cm). Purchase analytical grade reagents unless otherwise indicated.

2.1

Blood Sampling

1. K3 EDTA Blood collection tubes (9 mL). 2. Butterfly needles, 21 gauge assembled with adapter (see Note 1).

2.2 Plasma Separation

1. Centrifuge.

2.3 Plasma Microparticle Isolation via Ultracentrifugation

1. Ultracentrifuge with matching fixed-angle rotor.

2.4 Protein Concentration Measurement

1. Bicinchoninic acid (BCA) assay (Pierce BCA Protein Assay Kit, ThermoFisher Scientific, USA). The working reagent is prepared by mixing 50 parts of BCA reagent A with 1 part of BCA reagent B (see Note 2).

2. Polypropylene (PP) centrifugation tubes, 15 mL.

2. Thick wall polycarbonate tubes, 5–8 mL. 3. Dulbecco’s Phosphate Buffer Saline (DPBS) w/o calcium and magnesium.

2. Bovine serum albumin (BSA) standards. The following concentrations: 0.7, 0.6, 0.5, 0.4, 0.3, 0.25, 0.2, 0.15, 0.1 μg/μL are prepared from a 10 μg/μL BSA stock solution. BSA standards are stored at 4
C.

54

Raghda Saad Zaghloul Taleb et al.

3. 0.5 mL PP microcentrifuge tubes. 4. Thermoblock (specifications: temperature up to 100
C). 5. Flat-bottom 96-well plate. 6. Microplate reader with 562 nm filter. 2.5 Sample Preparation and Tryptic Protein Digestion

1. RapiGest SF powder (Waters, USA): 1 vial (1 mg) is reconstituted in 50 mM Ammonium Bicarbonate (ABC) to reach a final concentration of 0.1%. Store RapiGest SF 0.1% as aliquots at 20
C (see Note 3). 2. Reduction solution: 20 mM Dithiothreitol (DTT) in 100 mM ABC, pH 8.0. Store aliquots at 20
C. 3. Alkylation solution: 100 mM Iodoacetamide (IAA) in 100 mM ABC, pH 8.0. Store as aliquots in amber (light protection) microcentrifuge tubes at 20
C (see Note 4). 4. Trypsin (sequencing grade, modified, e.g., from Serva Electrophoresis, Germany): 1 vial (25 μg) is reconstituted in 100 μL 10 mM HCl. Stock solution is stored as 5 μL (1.25 μg) aliquots at 20
C. To prepare the working solution (0.033 μg/μL): dilute (just before use and on ice) 5 μL stock solution in 32.5 μL 100 mM ABC (see Note 5 and 6). 5. Trifluoroacetic acid (TFA) 10% and 0.1% (v/v). 6. Microcentrifuge. 7. Vacuum centrifuge.

2.6 Peptide Concentration Measurement

1. ACQUITY-Ultra-Performance Liquid Chromatography (UPLC) system with an AccQTag Ultra-UPLC column (Waters, USA). 2. AccQ Tag Ultra derivatization kit (Waters, USA). To prepare AccQ Tag Ultra Reagent: Add 1 mL of vial 2B (diluent) to vial 2A (reagent powder), vortex for 10 s and heat for 10 min at 55
C. Keep protected from light by aluminum foil. Store at room temperature. 3. 2.5 μmol/mL Pierce Amino Acid Standard (ThermoFisher Scientific, USA). To prepare a stock solution (100 pmol/μL): 80 μL is diluted with 1920 μL AccQ Tag Ultra Borate bufferVial 1 (Waters, USA). Stock solution is stored as 200 μL aliquots in glass tubes at 20
C. 4. Internal standard: 2.5 μmol/mL Norvaline. To prepare a stock solution (100 pmol/μL): 80 μL is diluted with 1920 μL AccQ Tag Ultra Borate buffer-Vial 1 (Waters, USA). Stock solution is stored as aliquots in PP microcentrifuge tubes at 20
C. To prepare the working solution (10 pmol/reaction): dilute 410 μL stock solution with 1990 μL AccQ Tag Ultra Borate buffer-Vial 1 (Waters, USA).

Plasma Microparticle Isolation and Proteome Analysis

55

5. 6 N HCl: add equal volumes of Ultrapure 12 N HCl and H2O. Store at 4
C. 6. 20 mM HCl: add 45 μL 6 N HCl to 15 mL H2O. Store at room temperature. 7. Eluent A: dilute 50 mL AccQ Tag Ultra Eluent A Concentrate (Waters, USA) with 950 mL H2O. Store at 4
C. 8. Eluent B: add 10 mL Formic acid (FA) 99–100% to 490 mL acetonitrile (ACN). Freshly prepared and used immediately. 9. Phenol crystals. 10. Thermomixer (specifications: temperature up to 100
C, mixing speed up to 500 rpm). 11. Reaction vessel and vacuum pump. 12. Argon gas. 13. Oven 14. 6  50 mm borosilicate glass tubes. 15. Glass vials 12  32 mm with blue screw neck cap and PTFE/ silicone septum (Waters, USA). 16. AccQ Tag Ultra 2.1  100 mm, 1.7 μm, 130 A˚ column (Waters, USA). 2.7 LC-MS/MS Measurement

1. LTQ Orbitrap Elite™ coupled online to an upstream connected Ultimate 3000 RSLCnano UHPLC System (ThermoFisher Scientific, USA). 2. Trap column: Acclaim PepMap100 C18 Nano-Trap Column (C18, particle size 5 mm, pore size 100 A˚, I.D. 100 mm, length 2 cm; Dionex, USA). 3. Analytical column: Acclaim PepMap RSLC, C18, 2 μm, 100 A˚, 75 μm  50 cm, nanoViper. (Dionex, USA).

4. Loading solvent: 0.1% (v/v) TFA. 5. Gradient solvent A: 0.1% (v/v) FA. 6. Gradient solvent B: 0.1% (v/v) FA (MS grade), 84% (v/v) ACN (HPLC-S gradient grade). 2.8

Data Analysis

1. Proteome Discoverer 1.4 software (ThermoFisher Scientific, USA). 2. Progenesis QI software (version 2.0.5387.52102, Nonlinear Dynamics Ltd., Newcastle upon Tyne, UK).

2.9 Statistical Analysis

1. R: A language and environment for statistical computing. (R Foundation for Statistical Computing, Vienna, Austria). 2. In-house written R scripts conducting ANOVA followed by Tukey’s “honest significant difference (HSD)” procedure for pairwise comparisons. The FDR is controlled by adjusting the p-value using the Benjamini and Hochberg method.

56

Raghda Saad Zaghloul Taleb et al.

2.10 Gene Ontology (GO) Annotation, Pathway Analysis, and Bioinformatics Analysis

3 3.1

1. Database for annotation, visualization, and integrated discovery (DAVID) bioinformatics resources v6.8 Beta, an online functional annotation tool (david.ncifcrf.gov) [17]. 2. EVpedia database. EVpedia is an integrated and comprehensive proteome, transcriptome, and lipidome database of extracellular vesicles derived from prokaryotes and eukaryotes (evpedia. info) [18].

Methods Blood Sampling

1. Blood sampling should be performed by experienced medical personal. 2. Blood samples are obtained from the antecubital vein under complete aseptic technique. 3. Puncture the antecubital vein at a 30
angle with the needle bevel. 4. Attach the blood collection tube to the adapter of the butterfly needle and allow blood to enter the tube. 5. Mix the blood immediately with the anticoagulant by gently inverting the tube five times. Do not mix vigorously to avoid hemolysis. 6. Proceed to plasma separation immediately.

3.2 Plasma Separation

1. The 9 mL collected blood are centrifuged at 1000  g, 4
C for 30 min. 2. The supernatant is quickly transferred into a 15 mL PP centrifuge tube (see Note 7). 3. To obtain platelet poor plasma (PPP), the supernatant is recentrifuged at 1000  g, 4
C for 30 min. 4. The resulting 5 mL plasma is transferred into a new 15 mL PP centrifuge and stored at 80
C until further analysis.

3.3 Plasma Microparticle Isolation via Ultracentrifugation

1. 5 mL plasma are centrifuged at 200,000  g, 4
C for 1 h (see Notes 8, 9, and 10). 2. The supernatant is carefully removed and discarded without disturbing the pellet (see Note 11). 3. The pellet is washed with 5 mL DPBS and centrifuged at 200,000  g, 4
C for 1 h. 4. The supernatant is carefully removed and discarded without disturbing the pellet. 5. The pellet is washed with 5 mL DPBS and centrifuged at 200,000  g, 4
C for 1 h.

Plasma Microparticle Isolation and Proteome Analysis

57

6. The supernatant is carefully removed and discarded without disturbing the pellet. 7. The pellet is suspended in 300 μL DPBS and stored at 80
C. 3.4 Protein Concentration Measurement

1. Pipette 10 μL of each standard, unknown sample and blank (DPBS) into PP microcentrifuge tubes (see Note 12). 2. Add 200 μL BCA working reagent to each tube and mix by vortexing. 3. Incubate the microcentrifuge tubes at 60
C for 30 min. 4. Cool to room temperature. 5. Load 200 μL of each tube in the bottom of a 96-well plate. 6. Measure the absorbance at 562 nm wavelength. Adjust the instrument to shake the plate first for 30 s with amplitude of 2.5 mm.

3.5 Sample Preparation and Tryptic Protein Digestion

1. An amount equivalent to 4 μg protein of plasma microparticle pellet is used for tryptic digestion. 2. Dry the samples using a vacuum centrifuge. 3. Suspend the dried pellets in 10 μL 0.1% RapiGest SF solution and mix by vortexing. 4. Reduce with 3.33 μL 20 mM DTT to reach a final concentration of 5 mM, incubate the samples at 60
C for 30 min, spin down, and cool to room temperature. 5. Alkylate with 2.35 μL 100 mM IAA to reach a final concentration of 15 mM, place the samples in the dark for 30 min at room temperature. 6. Add 6 μL (0.033 μg/μL) trypsin (1:20, w/w) to digest the protein overnight (max. 14 h) at 37
C (see Note 13). 7. Add 1.14 μL 10% TFA to reach a final concentration of 0.5%. This acidification ensures quenching of enzyme activity and precipitation of hydrolyzed RapiGest SF. 8. Incubate the samples at 37
C for 30 min. 9. Centrifuge the samples at 15,700  g for 10 min. 10. Transfer the supernatant to another microcentrifuge tube and dry the samples using a vacuum centrifuge. 11. Add 40 μL 0.1% TFA to the dried pellet to reach a final peptide concentration of 0.1 μg/μL (see below for the determination of peptide concentration) and store at 80
C for further usage.

3.6 Peptide Concentration Measurement

1. Place 5 μL of the peptides suspended in 0.1% TFA on the bottom of 6  50 mm borosilicate glass tubes and dry the samples using a vacuum centrifuge. Analyze the samples in duplicates.

58

Raghda Saad Zaghloul Taleb et al.

2. To perform acid gas phase hydrolysis, place 500 μL 6 N HCl together with 2 phenol crystals as antioxidant (to prevent destruction of, e.g., tyrosine) outside the sample glass tubes in a reaction vessel that can be evacuated, closed, and heated. Tightly seal the reaction vessel and connect it to the vacuum pump. Evacuate the reaction vessel alternating with flashing with argon. Repeat this for three times, the last step is the evacuation. Place the sealed vessel in the oven at 150
C for 60 min, then remove from the oven, open the vessels to allow evaporation of HCl and cooling to room temperature. Acid vapor hydrolyzes the dried samples. To ensure removal of residual HCl, connect the reaction vessel to the vacuum pump and evacuate. 3. Dissolve the hydrolyzed samples in 10 μL 20 mM HCl and vortex. Add 30 μL working solution of internal standard and vortex. To perform derivatization, add 10 μL AccQ Tag Ultra Reagent. Seal the glass tubes with parafilm, place on a thermoshaker, and incubate for 10 min at 55
C with minimal shaking. 4. For calibration, prepare the external standard in the following concentration: 5 pmol, 10 pmol, and 20 pmol. Prepare each concentration twice. Pipette from the stock solution of external standard (100 pmol/μL) the following: 2.5 μL (5 pmol), 5 μL (10 pmol), and 10 μL (20 pmol), place in 6  50 mm borosilicate glass tubes and dry for 15 min using a vacuum centrifuge. Derivatization of external standard is performed as previously described with the samples. 5. Transfer the derivatized samples and standards into 12  32 mm glass vials with blue screw neck cap and PTFE/ silicone septum. Separate and quantify the modified amino acids with an ACQUITY UPLC system. Perform the reversed phase separation on the waters AccQ Tag Ultra 2.1  100 mm, ˚ column. The derivatives are separated with a 1.7 μm, 130 A gradient from 0% to 60% Eluent B over 9.5 min at a flow rate of 700 μL/min at 55
C. The derivatives are quantified with UV detection at 260 nm. Data processing is done using Empower® software (Waters, USA). Briefly, the 17 peak areas of the amino acids in the protein sample are calculated. For quantification the ratio of the peak area of an amino acid in the sample to that of the same amino acid in the standard is calculated. The peak area data of the 17 amino acids is converted to pmol values by the external standard calibration. Each pmol value is multiplied by the molecular mass of the respective amino acid to obtain μg value of the amino acid per 1 μL injection volume. Peptide quantification is performed by summing the μg values of the contributing amino acids/1 μL injection volume. To convert μg/injection to μg/sample, the former value is multiplied by

Plasma Microparticle Isolation and Proteome Analysis

59

the dilution factor 50. To obtain the peptide concentration in μg/μL, the μg/sample is divided by the initial sample volume (5 μL). 3.7 LC-MS/MS Measurement

1. Inject 350 ng peptides dissolved in 17 μL 0.1% TFA into the HPLC for hydrophobic separation. 2. Pre-concentration of peptides is carried out on a trap column within 7 min at a flow rate of 30 μL/min with 0.1% TFA. 3. Peptides are then transferred onto an analytical column. There, peptide separation is achieved with a gradient from 5% to 40% solvent B over 98 min at 400 nL/min at 60
C. 4. MS/MS spectra are produced using data-dependent acquisition approach, which means that a fixed number of precursor ions of the first MS-scan is chosen and subjected to MS/MS analysis. 5. To obtain optimal efficiency in an experiment, one must finetune the following instrument parameters: The time period for acquiring the MS spectra, the time period for acquiring the MS/MS spectra, the number of peptides that will be selected for MS/MS analysis from an MS scan, and the method by which to select them (for example, by intensity rank or a predefined inclusion mass list) and the time period during which an m/z value will be excluded from being selected again. Full scan mass spectra in the Orbitrap analyzer are acquired in profile mode at a resolution of 60,000 at 400 m/z and within a mass range of 350–2000 m/z. For MS/MS measurements, the twenty most abundant peptide ions are fragmented by collision-induced dissociation (CID) and measured for tandem mass spectra in the linear ion trap. 6. Finally export LC-MS/MS analysis data as Thermo .raw file format.

3.8

Data Analysis

3.8.1 Protein Identification

1. Protein identification is performed with the software Proteome Discoverer 1.4 (ThermoFisher Scientific, USA). 2. The measured spectra are searched against the UniprotKB/ SwissProt database using Mascot search engine (version 2.5, Matrix Science Ltd., London, UK). Search parameters are set as follows: enzyme: trypsin; maximum missed cleavage sites: 1; taxonomy: Homo sapiens; mass tolerance: 5 ppm for precursor and 0.4 Da for fragment ions; dynamic modification: oxidation (M); static modification: carbamidomethyl (C). The percolator algorithm implemented in Proteome Discoverer is used to calculate the false discovery rate (FDR) of the identified peptides and only peptides with an FDR < 1% are considered.

60

Raghda Saad Zaghloul Taleb et al.

3.8.2 Protein Quantification

1. Ion intensity-based label-free quantification is carried out using Progenesis QI software (Nonlinear Dynamics Ltd., UK). 2. All (.raw) files resulting from LC-MS/MS measurement are imported into the software. 3. The software is adjusted to automatically assess all runs in the experiment for suitability and select the alignment reference run. The retention times of eluting peptides from all the samples within the experiment are aligned to the selected reference run, to compensate for the shifts in retention times between runs. Check the automatic alignment carefully and if it does not work optimally, manually add vectors to align the run. 4. The peptide ions are filtered to include the peptides with certain criteria: Those with a charge of þ2 to þ4 and only ions with a minimum of three isotope peaks. 5. After alignment and peptide ion filtering, normalization of raw abundances is performed to allow comparison across different samples within the experiment. The software automatically selected one of the runs that is “least different” from all the other runs in the data set to be the “Normalizing reference.” For each detected ion, a quantitative abundance ratio is calculated between the run being normalized and the reference run. The normalization should be checked and runs that exceed a threshold that has to be defined by the experimenter should be excluded from the experiment. 6. Create the experimental design, e.g., healthy vs. diseased individuals. 7. The protein identifications obtained from Proteome Discoverer are imported into Progenesis QI software in order to identify quantified peptides. Identified peptides with mass error more than 5 ppm are deleted. Only unique or nonconflicting peptides of a corresponding protein are used for relative quantification. Protein grouping, i.e., hide proteins whose peptides are a subset of another protein, is disabled. 8. As a quality control, perform a principal component analysis (PCA) to check whether the runs cluster based on the expected study groups and to determine whether there are any outliers in the data.

3.9 Statistical Analysis

1. Transform the normalized abundances using the inverse hyperbolic sine (arsinh) and perform the ANOVA test. Tukey’s “HSD” procedure for pairwise comparisons between the individual study groups is calculated in case of a significant result of the ANOVA test [19].

Plasma Microparticle Isolation and Proteome Analysis

61

Fig. 2 Exemplary results of statistical analysis, gene ontology annotation, and comparison with EVpedia database [20]. (a) Protein abundance profile of Annexin A2 (ANXA2) along three experimental groups. Boxes represent 25th and 75th percentiles; whiskers indicate the 1.5-fold standard deviation. The median is shown as a horizontal line, and the mean value as a square within the box. Significant differences between the experimental groups are indicated by an asterisk. (b) Results of the enrichment analysis of all identified proteins for the cellular component ontology (Top 10 enriched GO terms). Bars represent protein/gene numbers. Purple points represent –Log10 (pFDR value). (c) Venn diagram showing the overlap between the proteins identified and quantified in this study and the top 100 extracellular vesicle proteins (as reported by EVpedia database)

2. Calculate the absolute fold change by dividing mean normalized protein abundance (not transformed) in one study group by mean normalized protein abundance in another group. 3. Set up the filter criteria for the experiment. A significant differential abundance might be defined as an absolute fold change 2, an FDR-corrected ANOVA p-value (using Benjamini and Hochberg method) (pFDR)  0.05, and a pTukey HSD  0.05. Exemplary results [20] are displayed for the protein Annexin A2 in Fig. 2a.

62

Raghda Saad Zaghloul Taleb et al.

3.10 GO Annotation, Pathway Analysis, and Bioinformatics Analysis

1. Upload the list of Uniprot accession numbers of the identified proteins or the differentially abundant proteins into DAVID. DAVID systematically maps the large list to the associated biological annotation (e.g., GO terms), and then statistically highlights the most overrepresented (enriched) biological annotation out of thousands of linked terms and contents. An enrichment p-value (or called EASE score) based on a modified Fisher’s exact test is calculated for the most relevant (overrepresented) biological terms enriched in the given list. The enrichment p-value is corrected for multiple testing to control the false discovery rate using the Benjamini-Hochberg method (Fig. 2b). 2. To confirm the enrichment of extracellular vesicles in the preparation, compare the set of identified and quantified proteins with proteins identified in human extracellular vesicles, extracted from EVpedia (Fig. 2c). EVpedia database provides an array of useful tools such as an updated list for the top 100 extracellular vesicle proteins based on the number of publications.

4

Notes 1. For the venipuncture, needles ranging from 19 to 22 gauge are commonly used. Those large diameter needles are recommended to avoid ex vivo activation of blood platelets and possible production of erythrocyte microparticles. 2. It is possible to use other protein assay methods, e.g., the Bradford assay. 3. RapiGest SF is a reagent used to enhance the tryptic digestion of proteins by solubilizing proteins and making them more susceptible to enzymatic cleavage without significantly inhibiting the enzyme activity. 4. Iodoacetamide should be protected from the light as much as possible in order to minimize light-induced decomposition. This solution can be stored at 20
C for up to 1 year. 5. Use sequencing grade trypsin only for mass spectrometry experiments due to its stringent specificity and low contamination of alpha-chymotrypsin. Moreover, it is recommended to use modified trypsin (with methylated lysine residues) to prolong activity of the native enzyme and resist its autolysis. 6. It is advisable to divide the solution in glass vials as aliquots of 1.25 μg each (5 μL) in order to prevent multiple freeze/thaw cycles. The trypsin solution can be stored at 20
C for up to 1 year.

Plasma Microparticle Isolation and Proteome Analysis

63

7. Plasma should be cautiously collected without disturbing the platelet layer to avoid contamination of the plasma with platelets. 8. Prior to ultracentrifugation, tubes should be weighed to ensure adequate balance and equal weight distribution between opposing tubes. 9. It is recommended to mark each ultracentrifuge tube and orient the tubes in the rotor facing up and outward. The mark refers to the location of the pellet after ultracentrifugation. 10. All rotor lids and caps should also be washed with 70% ethanol. 11. For supernatant removal following ultracentrifugation, for fixed-angle rotors pour the supernatant rather than using a pipette. 12. Since the amount of samples was limited, 10 μL of unknown samples and standards were used for the test tube procedure and to maintain the sample: working reagent ratio, the working reagent used was reduced to 200 μL. 13. For protein digestion, a ratio of between 1:100 and 1:20 (w/w) of trypsin to substrate is recommended. References 1. Mause SF, Weber C (2010) Microparticles: protagonists of a novel communication network for intercellular information exchange. Circ Res 107(9):1047–1057. https://doi. org/10.1161/CIRCRESAHA.110.226456 2. Inal JM, Kosgodage U, Azam S et al (2013) Blood/plasma secretome and microvesicles. Biochim Biophys Acta 1834(11):2317–2325. https://doi.org/10.1016/j.bbapap.2013.04. 005 3. Morel O, Jesel L, Freyssinet JM et al (2011) Cellular mechanisms underlying the formation of circulating microparticles. Arterioscler Thromb Vasc Biol 31(1):15–26. https://doi. org/10.1161/ATVBAHA.109.200956 4. Lynch SF, Ludlam CA (2007) Plasma microparticles and vascular disorders. Br J Haematol 137(1):36–48. https://doi.org/10.1111/j. 1365-2141.2007.06514.x 5. Dignat-George F, Boulanger CM (2011) The many faces of endothelial microparticles. Arterioscler Thromb Vasc Biol 31(1):27–33. https://doi.org/10.1161/ATVBAHA.110. 218123 6. Burger D, Schock S, Thompson CS et al (2013) Microparticles: biomarkers and

beyond. Clin Sci (Lond) 124(7):423–441. https://doi.org/10.1042/CS20120309 7. Yuana Y, Bertina RM, Osanto S (2011) Pre-analytical and analytical issues in the analysis of blood microparticles. Thromb Haemost 105(3):396–408. https://doi.org/10.1160/ TH10-09-0595 8. Robert S, Poncelet P, Lacroix R et al (2009) Standardization of platelet-derived microparticle counting using calibrated beads and a Cytomics FC500 routine flow cytometer: a first step towards multicenter studies? J Thromb Haemost 7(1):190–197. https://doi. org/10.1111/j.1538-7836.2008.03200.x 9. Ueba T, Haze T, Sugiyama M et al (2008) Level, distribution and correlates of plateletderived microparticles in healthy individuals with special reference to the metabolic syndrome. Thromb Haemost 100(2):280–285. https://doi.org/10.1160/TH07-11-0668 10. Smalley DM, Root KE, Cho H et al (2007) Proteomic discovery of 21 proteins expressed in human plasma-derived but not plateletderived microparticles. Thromb Haemost 97 (1):67–80. https://doi.org/10.1160/TH0602-0066

64

Raghda Saad Zaghloul Taleb et al.

11. Aras O, Shet A, Bach RR et al (2004) Induction of microparticle- and cell-associated intravascular tissue factor in human endotoxemia. Blood 103(12):4545–4553. https://doi.org/ 10.1182/blood-2003-03-0713 12. Tesselaar ME, Romijn FP, Van Der Linden IK et al (2007) Microparticle-associated tissue factor activity: a link between cancer and thrombosis? J Thromb Haemost 5(3):520–527. https://doi.org/10.1111/j.1538-7836.2007. 02369.x 13. Yuana Y, Oosterkamp TH, Bahatyrova S et al (2010) Atomic force microscopy: a novel approach to the detection of nanosized blood microparticles. J Thromb Haemost 8 (2):315–323. https://doi.org/10.1111/j. 1538-7836.2009.03654.x 14. Miguet L, Pacaud K, Felden C et al (2006) Proteomic analysis of malignant lymphocyte membrane microparticles using double ionization coverage optimization. Proteomics 6 (1):153–171. https://doi.org/10.1002/ pmic.200500133 15. Jin M, Drwal G, Bourgeois T et al (2005) Distinct proteome features of plasma microparticles. Proteomics 5(7):1940–1952. https:// doi.org/10.1002/pmic.200401057

16. Josic D, Clifton JG (2007) Mammalian plasma membrane proteomics. Proteomics 7 (16):3010–3029. https://doi.org/10.1002/ pmic.200700139 17. Huang da W, Sherman BT, Lempicki RA (2009) Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc 4(1):44–57. https:// doi.org/10.1038/nprot.2008.211 18. Kim DK, Lee J, Kim SR et al (2015) EVpedia: a community web portal for extracellular vesicles research. Bioinformatics 31(6):933–939. https://doi.org/10.1093/bioinformatics/ btu741 19. Tukey JW (1949) Comparing individual means in the analysis of variance. Biometrics 5 (2):99–114. https://doi.org/10.2307/ 3001913 20. Taleb RSZ, Moez P, Younan D et al (2017) Quantitative proteome analysis of plasma microparticles for the characterization of HCV-induced hepatic cirrhosis and hepatocellular carcinoma. Proteomics Clin Appl 11 (11-12). https://doi.org/10.1002/prca. 201700014

Chapter 5 A Standardized and Reproducible Proteomics Protocol for Bottom-Up Quantitative Analysis of Protein Samples Using SP3 and Mass Spectrometry Christopher S. Hughes, Poul H. Sorensen, and Gregg B. Morin Abstract The broad utility of mass spectrometry (MS) for investigating the proteomes of a diverse array of sample types has significantly expanded the use of this technology in biological studies. This widespread use has resulted in a substantial collection of protocols and acquisition approaches designed to obtain the highestquality data for each experiment. As a result, distilling this information to develop a standard operating protocol for essential workflows, such as bottom-up quantitative shotgun whole proteome analysis, can be complex for users new to MS technology. Further complicating this matter, in-depth description of the methodological choices is seldom given in the literature. In this work, we describe a workflow for quantitative whole proteome analysis that is suitable for biomarker discovery, giving detailed consideration to important stages, including (1) cell lysis and protein cleanup using SP3 paramagnetic beads, (2) quantitative labeling, (3) offline peptide fractionation, (4) MS analysis, and (5) data analysis and interpretation. Special attention is paid to providing comprehensive details for all stages of this proteomics workflow to enhance transferability to external labs. The standardized protocol described here will provide a simplified resource to the proteomics community toward efficient adaptation of MS technology in proteomics studies. Key words SP3, Paramagnetic beads, Proteomics, Mass spectrometry, Protein cleanup, Tandem mass tagging, Protein digestion, Bottom-up, Quantification

1

Introduction Advances in sample handling and MS technology have enabled the efficient in-depth characterization of proteomes across a range of organisms and experimental scenarios [1]. To execute an optimal analysis using these advanced proteomics tools, careful optimization of a wide range of experimental steps is necessary. The most common implementation of MS in proteomics are bottom-up analyses, wherein solubilized proteins are enzymatically digested to peptides that are then examined as complex mixtures [2]. In a typical bottom-up analysis, proper execution of optimal protocols for lysis and protein solubilization, protein cleanup, quantitative

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_5, © Springer Science+Business Media, LLC, part of Springer Nature 2019

65

66

Christopher S. Hughes et al. Identification

Absorbance

Fractions

m/z

10

−log (Adjusted p-value)

contaminant removal

Intensity

Quant

Retention Time

log (Fold Change) 2

Cell Populations

Solubilized Proteins

(treated vs. untreated)

(in detergent mixture)

Protein Clean-up and Digestion

Peptide Labeling

Peptide Fractionation

Mass Spectrometry Analysis

(isobaric TMT labeling)

(high pH C18 reversed phase)

(with data interpretation)

(SP3 clean-up protocol)

Fig. 1 Workflow for quantitative whole proteome analysis using MS. Schematic depicts all steps utilized in the preparation and analysis of cellular samples with a quantitative whole proteome protocol as described in this work

labeling, peptide fractionation, MS acquisition, and data processing are all critical to the success of the experiment. Complicating experimental design is that each of these handling steps offers a broad range of suitable workflows to achieve the end result, an in-depth survey of the proteins in a sample. As a result, choosing the optimum set of workflows from the large collection of available protocols to build an appropriate sample processing pipeline can be a daunting task for both experienced and new users of MS technology in proteomics. In this work, we describe in detail a standard workflow that can be used to obtain a high-quality in-depth and quantitative survey of proteins within a complex sample that is also suitable for biomarker discovery (Fig. 1). Throughout this work, special attention will be paid to give descriptive information on each of the central workflow points in a typical proteomics analysis: (1) cell lysis and protein solubilization, (2) protein cleanup and digestion using SP3, (3) quantitative labeling, (4) peptide fractionation, (5) MS data acquisition, and (6) MS data analysis. 1.1 Cell Lysis and Protein Solubilization

The primary goal of cell lysis from a proteomics standpoint is to extract and solubilize the proteins of interest for an experiment. The diversity in protein biochemistry has resulted in a large toolbox of reagents for handling proteins. Depending on the cell, tissue, or material being analyzed, a lysis solution can include detergents, chaotropes, salts, organic solvents, and other components. To enhance recovery of proteins from varied biological materials, mechanical methods to increase cell disruption, such as bead beating, dounce homogenization, ultra-high pressure, and sonication, can be paired with chemical lysis methods. Despite the availability of this diverse set of tools, there remains no one-size-fits-all solution for global shotgun proteomics. In this work, we describe the use of a mixture of detergents and salts coupled to mechanical bead beating to maximize the extraction and solubilization of a broad range of proteins.

Standardized Protocol for Bottom-up Proteomics with MS

67

1.2 Protein Cleanup and Digestion

Bottom-up proteomics analysis relies on digestion of proteins into peptides prior to MS analysis. Problematically, many commonly used enzymes (e.g., trypsin) are inhibited by different concentrations of denaturants. In addition, detergents can disrupt both liquid chromatography (LC) and MS analysis. Depending on the lysis protocol used, a variety of cleanup methods are available to remove or dilute components prior to digestion and MS analysis [3, 4]. The optimum cleanup approach should provide a balance between reagent compatibility, contaminant removal, high quantitative recovery of proteins, simplicity and speed in application, throughput, and a low per sample cost. In this work we describe the use of the SP3 paramagnetic bead-based approach [5–7], as it encompasses all of the above traits and has been proven effective across a diverse array of sample types.

1.3 Quantitative Labeling

Quantification of peptides using signal intensities derived from MS analysis provides an effective means to determine the relative amounts of proteins within and between samples. For this purpose, a large collection of quantification methods are available, including label free and spectral counting methods, MS1-based metabolic (SILAC) labeling, chemical (reductive dimethylation) tagging, and tandem MS-based isobaric tagging approaches (tandem mass tagging—TMT, isobaric tags for relative and absolute quantification—iTRAQ) [8–10]. Each of these methods offers advantages and disadvantages in terms of multiplexing capacity, accuracy and precision, and costs. In this work, we describe application of the isobaric TMT approach as it affords the highest level of multiplexing, is compatible with virtually any sample, results in a minimum number of missing values between samples, and has been applied to achieve high depth quantitative proteome coverage in diverse contexts [6, 11–13].

1.4 Peptide Fractionation

The “in-time” nature of MS data acquisition means that a limited number of peptide ions can be sampled in a given elution window during analysis. If an elution window is too complex, peptide ions can be missed, and thus incomplete sampling of a sample will occur. In addition to sampling, the dynamic range of measurable signals can impact peptide detection. To minimize the influence of these properties, peptide pre-fractionation steps are often used to simplify samples prior to MS analysis to increase proteome coverage [14]. As the majority of MS analyses use an acidic pH reversedphase C18 separation step directly before electrospray ionization (ESI), an ideal fractionation method will be highly orthogonal with this approach [15]. In this work, we describe basic pH reversedphase C18 separation for offline fractionation of peptides. The high-pH approach with concatenation has been demonstrated to be highly orthogonal with acidic pH separation of the same type while also being simple to optimize and apply, and is routinely employed to give high proteome coverage [5–7, 12, 16].

68

Christopher S. Hughes et al.

1.5 MS Data Acquisition

The wide variety of MS instrument architectures means that an equally large set of acquisition approaches is possible for analyzing samples. In bottom-up analysis of whole proteomes, Orbitrapbased platforms (Thermo Scientific) have become the instrument of choice for a wide variety of analyses [17]. However, recent advances in quadrupole-time-of-flight (Q-ToF) instruments [18–20], especially when coupled with ion mobility, have triggered the resurgence of this platform for bottom-up protein identification. Independent of the instrument chosen for analysis, careful optimization of the scan parameters is necessary to ensure acquisition of the highest-quality data set. In this work we use an Orbitrap Fusion Tribrid MS instrument [21] for data acquisition due to the rapid scan rate and high sensitivity in ion detection afforded by this platform.

1.6 Analysis of MS Data

Similar to the wide range of platforms for MS acquisition, there is a large diversity of tools available for processing of the acquired data [22, 23]. In bottom-up experiments, a majority of tools rely on database-matching methods that use variants of the target-decoy method for controlling identification error rates. Available software packages can also offer functionalities beyond peptide identification, such as quantification, and can be open-source or developed by commercial entities. While other workflows are suitable, in this work we describe the open-source and highly flexible combination of SearchGUI [24] and PeptideShaker [25]. SearchGUI operates as a front end, interfacing and facilitating search execution from a variety of popular database matching tools. PeptideShaker provides robust processing of search results, with precise estimation and control of error rates. To perform quantification, we use the open-source and freely available RawTools tool for parsing of isobaric tag data directly from Thermo Scientific Orbitrap raw files [26]. Together, the methods used at each stage of this work interlink to provide a robust protocol capable of handling materials from a wide variety of experiments, from single proteins, to cell lines, to tissues. The described workflow can be directly applied and serves as an excellent starting point for modification toward creation of a tailored standardized operating protocol for bottom-up whole proteome analysis focused on biomarker discovery.

2

Materials Prepare all solutions using ultrapure water unless otherwise noted. Where powdered materials are used to make stock solutions, pre-made commercially obtainable solutions may be substituted. All solutions may be stored at room temperature, unless otherwise noted by the manufacturer or this protocol. Directly follow all

Standardized Protocol for Bottom-up Proteomics with MS

69

institutional regulations in relation to waste management and discarding of chemicals and solutions. Wear appropriate personal protective equipment at all times and work in a fume hood when required. 2.1 Cell Lysis, Protein Reduction, and Alkylation

1. 1.5 mL microfuge tubes. 2. Lysis buffer: 50 mM HEPES pH 8, 1% SDS, 1% Triton X100, 1% NP-40, 1% Tween 20, 1% deoxycholate, 5 mM EDTA, 50 mM NaCl, 1% glycerol, 1
protease inhibitor, 5 mM dithiothreitol. 3. FastPrep24 instrument (MP Biomedicals). This equipment is optional, but can improve lysis efficiency and protein recovery. 4. Lysing Matrix Y tubes (MP Biomedicals). Optional, only required if the FastPrep24 instrument is used. 5. 200 mM iodoacetamide stock solution. 6. 100 mM dithiothreitol stock solution 7. Benchtop centrifuge capable of holding 1.5 mL tubes, and achieving 20,000
g RCF. 8. A Thermomixer capable of holding 1.5 mL tubes (Eppendorf).

2.2 SP3 Processing and Protein Digestion

1. SeraMag carboxylate-coated beads with a hydrophilic surface (GE Healthcare). Two individual types of SeraMag carboxylate beads are commercially available and can be used interchangeably or in combination. 2. Absolute ethanol. 3. 80% ethanol stock solution prepared using absolute ethanol. 4. Magnetic rack capable of holding 1.5 mL tubes. 5. 0.2 M HEPES, pH 8 stock solution. 6. Trypsin enzyme. 7. Bath sonicator: this item is optional, but can improve protein elution from the SP3 beads.

2.3

TMT Labeling

1. HPLC-grade acetonitrile. 2. TMT reagents (Thermo Fisher). 3. 1 M glycine. 4. 10% trifluoroacetic acid (TFA in HPLC-grade water). 5. Vacuum centrifuge.

2.4 C18 Tip Peptide Cleanup

1. Mobile Phase A: 0.1% TFA in HPLC-grade acetonitrile. 2. Mobile Phase B: 0.1% TFA in HPLC-grade water. 3. Rinse solution: 0.1% formic acid (FA) in HPLC-grade water.

70

Christopher S. Hughes et al.

4. Elution solution: 0.1% FA in 80% HPLC-grade acetonitrile and 20% HPLC-grade water. 5. C18 TopTip’s, 1 mL pipette tip size (Glygen Corp.). 6. Vacuum centrifuge. 2.5 Offline High-pH C18 HPLC Peptide Fractionation

1. Mobile Phase A: 20 mM Ammonium bicarbonate, pH 8 in HPLC-grade water. 2. Mobile Phase B: HPLC-grade Acetonitrile. 3. HPLC system: there are a wide variety of systems that will be compatible with the protocol described here. Optimization for your specific system based on system volumes and elution windows will need to be carried out irrespective of what system is used. For this work, an Agilent 1100 system equipped with a diode array detector (DAD) and an analytical fraction collection module is used. 4. Fractionation column: there are a wide variety of columns that can be used for this purpose. Column selection will depend on the limits (e.g., flow rate, pressure) of your HPLC system, the desired selectivity, and sample loading requirements. For this work, a Kinetix EVO-C18 reversed phase column (250 mm
4.6 mm, 2.6 μm core-shell particles, pH 1–12 stable, Phenomenex) at a flow rate of 1 mL/min is used. 5. Precolumn: This item is optional, but will help protect your HPLC fractionation column. For this work, a KrudKatcher Ultra precolumn (Phenomenex) is used. 6. MonoSLEEVE controller unit and 15 cm HotSleeve column oven heated to 45  C (Analytical Sales and Services). This item is optional, but offers additional flexibility in separation setup (e.g., higher flow rates, smaller column particle size, variable column lengths). 7. Vacuum centrifuge.

2.6 Nano-UHPLC Column Preparation

1. Acetone, HPLC-grade. 2. 200 μm internal diameter (ID) glass microcapillaries, polyimide coated. 3. Frit kit (New Objective). This will come with a ceramic cutting device for capillaries that can also be used for “shaving” analytical column tips. 4. 1.9 μm Reprosil-Pur Basic C18 beads (Dr. Maisch). 5. Capillary Packing Bomb. Paired with a high-pressure nitrogen tank (standard purity) equipped with a gas-appropriate regulator capable of providing an output pressure of 1000 psi. 6. P-2000 Laser Puller (Sutter Instruments). This equipment is optional. Pre-made fritted nanospray capillaries purchased

Standardized Protocol for Bottom-up Proteomics with MS

71

commercially that can be packed with chromatography material may be substituted. 7. Butane torch: can be obtained from most hardware stores. 8. Nano-HPLC pump capable of flow rates of 1 μL/min using a 60% acetonitrile, 40% water mobile phase. This unit is optional. Columns can alternatively be packed using the packing bomb to the desired lengths. 9. Stainless steel union, 1/1600 fitting size, 0.25 mm bore (only required if columns will be packed using nano-HPLC unit). Tubing sleeves (1/1600 OD
0.01600 ID, Orange) will also be required for connections with this union. 2.7 Nano-UHPLC Configuration and LCMS Data Acquisition

1. MicroCross connector, PEEK, 39 nL swept volume, 0.15 mm thru hole. 2. MicroTight sleeves, green, 0.02500 OD
0.015500 ID. 3. nano-UHPLC Mobile Phase A: 0.1% FA in HPLC-grade water. 4. nano-UHPLC Mobile Phase B: 0.1% FA in HPLC-grade acetonitrile. 5. MonoSLEEVE controller unit and 20 cm AgileSLEEVE column oven heated to 50  C (Analytical Sales and Services). 6. nano-UHPLC instrument: for this work an Easy nLC-1000 system is used (Thermo Fisher). 7. 50 μm ID
65 cm nanoViper capillary—double-ended nanoViper capillary (Thermo Fisher). This part is optional. 8. Liquid junction kit: This item is optional and only required if you are using the Nanospray Flex ESI source (Thermo Fisher). 9. MS instrument: for this work an Orbitrap Fusion™ Tribid™ (Thermo Scientific) equipped with an in-house manufactured nanospray ion source (the design and parts used are based off of the University of Washington Proteomics Resource (UWPR) nano source—http://proteomicsresource. washington.edu/protocols05/nsisource.php) is used.

2.8 Analysis of Acquired MS Data

1. SearchGUI software: latest version can be downloaded from http://compomics.github.io/projects/searchgui.html [24]. 2. PeptideShaker software: latest version can be downloaded from http://compomics.github.io/projects/peptide-shaker. html [25]. 3. RawTools software: latest version and instruction guide can be downloaded from https://github.com/kevinkovalchik/ RawTools [26].

72

3

Christopher S. Hughes et al.

Methods All steps of this protocol can be carried out at room temperature (24  C), unless noted otherwise. This protocol assumes a recommended starting point of a collection of ten individual cell pellets or frozen tissue pieces (5 million cells or  5
10 μm scrolls from a 1 cm
1 cm
1 cm piece of tissue, for each sample, respectively) being processed in parallel in a single workflow. However, this protocol is also compatible with smaller input amounts (10,000 cells or  1
10 μm scroll from a 1 cm
1 cm
1 cm piece of tissue). Scale reagents appropriately if using an amount of starting material that is different from the recommendations stated above.

3.1 Cell Lysis, Protein Reduction, and Alkylation

1. Preset a Thermomixer to 60  C heating and 1000 rpm mixing. 2. If using the FastPrep instrument (or another cell disruption device, such as a sonicator), prepare and label an appropriate number of tubes for processing (see Note 1). 3. Prepare sufficient lysis buffer (see Note 2) for the number of samples to be processed. Assuming 5 million cells in 10 individual samples, a total of 500 μL of lysis buffer will be used for each (5 mL total required). 4. If the material to be processed is frozen, thaw the cells to be lysed. As cells are thawing, gently flick the bottoms of the tubes to reconstitute the cellular material pellet. Keep cells on ice during the thawing process. 5. Transfer the cell material to a FastPrep Lysing Matrix Y tube for each sample (see Note 3). 6. Add 500 μL of lysis buffer to each tube and disrupt on the FastPrep instrument using a program of speed 6 M/s, 45 s, 3 cycles, 120 s rest between cycles (see Note 4). 7. Centrifuge (spin) the tubes with lysate at 20,000
g for 1 min. 8. Recover the supernatant to fresh 1.5 mL tubes and incubate in the preheated 60  C Thermomixer for 15 min. 9. Remove the tubes from the Thermomixer and let them cool and equilibrate to room temperature in a sample rack on the bench (15-min). 10. Add 50 μL of the prepared iodoacetamide stock solution and pipette mix. Incubate in the dark at room temperature for 30 min. 11. Add 25 μL of the prepared dithiothreitol stock solution and pipette mix. Incubate at room temperature for 15 min.

Standardized Protocol for Bottom-up Proteomics with MS

73

12. If desired, measure the protein concentration using a method compatible with the applied lysis buffer. 13. Cell lysates can be frozen at 80  C indefinitely prior to further processing. 3.2 SP3 Processing and Protein Digestion

Processing from this point assumes the use of 50 μg of protein material from each of the lysates prepared above in a final working volume of 100 μL. The described workflow is compatible with a range of input amounts and should be scaled based on the requirements of the samples in question. 1. Set a Thermomixer to room temperature (24  C) heating and 1000 rpm mixing. 2. Prepare a working solution of SP3 beads based on the number of samples to be processed (see Note 5). 3. Add an appropriate amount of the SP3 bead stock to 50 μg of protein from each of the cell lysates (e.g., 500 μg). Pipette mix to homogenize the beads and lysate. 4. Add 100% ethanol to achieve a final proportion of 50% by volume (50% ethanol final concentration) (e.g., for 100 μL of lysate þ beads, add 100 μL of 100% ethanol). Pipette mix briefly to homogenize the beads, lysate, and ethanol (see Note 6). 5. Place tubes in the Thermomixer and incubate at room temperature for 5 min with shaking at 1000 rpm. 6. Place tubes in a magnetic rack and incubate for 2 min or until the beads have settled to the tube wall. 7. Remove and discard the supernatant taking care not to disrupt the beads. 8. Remove the tubes from the magnetic rack and add 200 μL of 80% ethanol. Gently pipette to reconstitute and rinse the beads. 9. Place the tubes on the magnetic rack and incubate for 2 min or until the beads have settled to the tube wall. 10. Remove and discard the supernatant. 11. Repeat steps 8–10 two further times. 12. Using a 200 μL pipette and tip, remove as much ethanol from the tubes as possible (see Note 7). 13. Add 100 μL of 50 mM HEPES pH 8 containing trypsin (1:25 enzyme:protein (μg/μg)) to each tube (see Note 8). 14. Using the tip of the pipette, gently push the beads from the side of the tube wall into the liquid. Do not try to pipette the liquid (see Note 9). 15. Sonicate the tubes in a bath for 15 s to reconstitute the beads.

74

Christopher S. Hughes et al.

16. Incubate tubes overnight at 37  C with mixing at 1000 rpm in a Thermomixer. 17. The next day, spin the tubes at 20,000
g for 1 min to pellet the beads. 18. Place on a magnetic rack for 2 min or until beads have settled to the tube wall. 19. Recover the supernatant to a fresh 1.5 mL tube taking care to not remove any magnetic beads. 20. The peptides can be frozen and stored indefinitely at this point. 3.3

TMT Labeling

Since the protocol above was described for a set of ten samples, this stage of the workflow assumes the use of a 10-plex TMT kit. However, the protocol can be adapted for 2, 6, 11, or n-plex as needed. A total of 50 μg of peptide is assumed for each sample, based on the protein input amount for SP3. 1. If starting from a new label kit, centrifuge the stock tubes of TMT label at 5000
g for 1 min. 2. Add HPLC-grade acetonitrile to reconstitute the TMT labels to a final concentration of 10 μg/μL. 3. Vortex-mix the tube for 30 s to dissolve the label reagents (see Note 10). 4. If using previously reconstituted labels, remove from the freezer and equilibrate to room temperature. 5. Add 10 μL (100 μg of label) of the appropriate TMT label to each of the sample tubes and pipette mix. 6. Incubate for 30 min at room temperature. 7. Add a further 10 μL of the appropriate TMT label to each of the sample tubes and pipette mix. 8. Incubate for 30 min at room temperature. 9. Add 10 μL of 1 M glycine to each of the tubes and pipette mix. 10. Incubate for 15 min at room temperature. 11. SpeedVac the samples to reduce the volume by approximately 50%. 12. Combine the samples into a single tube and acidify to 1% TFA (v/v) using a 10% stock TFA solution. 13. Spin the tube at 20,000
g for 2 min to pellet any formed precipitate and recover the supernatant.

3.4 C18 Tip Peptide Cleanup

Different desalting methods can be substituted at this protocol stage. TopTip’s C18 tips offer high peptide binding capacity and relative ease of use. For all steps, a flow rate of 1 drop per second eluting from the tip is desired.

Standardized Protocol for Bottom-up Proteomics with MS

75

1. Condition the tip: add 600 μL of mobile phase A and push the liquid through the tip using the manufacturer provided plastic syringe. The eluted liquid can go to waste. When adding the mobile phase initially, pipette over the inner walls of the TopTip to promote rinsing of the beads stuck at the upper portion into the rest of the slurry at the bottom. 2. Repeat step 1. 3. Add 600 μL of mobile phase A taking care to not disrupt the bead slurry and push the liquid through the tip. The eluted liquid can go to waste. 4. Repeat step 3. 5. Add 600 μL of mobile phase B taking care to not disrupt the bead slurry and push the liquid through the tip. The eluted liquid can go to waste. 6. Repeat step 5. 7. Bind sample: add sample to the tip taking care to not disrupt the bead slurry and push the liquid through the tip. The eluted liquid can go to waste. If the volume is too large, add in multiple steps. 8. Rinse bound sample: add 600 μL of Rinse solution taking care to not disrupt the bead slurry and push the liquid through the tip. The eluted liquid can go to waste. 9. Repeat step 8 two further times. 10. Elute sample: add 600 μL of Elution solution taking care to not disrupt the bead slurry and push the liquid through the tip into a 1.5 mL collection tube. 11. Repeat step 10. 12. SpeedVac the elution to a volume of approximately 80 μL or less. 13. Vortex the tube briefly and spin at 20,000
g for 2 min. 14. Recover the supernatant taking care to not remove any of the material pellet. 15. The labeled and desalted peptide mixture can be stored indefinitely at 80  C. 3.5 Offline High-pH C18 HPLC Peptide Fractionation

1. Scale the amount for injection based on the HPLC column being used for fractionation. In this work, a total amount of 500 μg of peptide material is assumed based on the starting protein quantity. This entire sample amount will be injected to the HPLC for fractionation. 2. Inject and separate the peptide mixture using a gradient of mobile phase A and B. The gradient used will depend on your HPLC system, column, and peptide labeling status. For

76

Christopher S. Hughes et al.

example, for TMT-labeled peptides a gradient that begins at a baseline of 5% B for 2 min, to 15% B in 2 min, to 22% B in 4 min, to 34% B in 12 min, to 42% B in 5 min, to 80% B in 1 min, held at 80% for 4 min, to 5% B in 1 min, and a final reconditioning at 5% for 5 min (35 min total runtime). From 6 to 28 min fractions are collected every 0.46 min across the gradient length (48 total fractions) into 4 rows of a 96 well plate and concatenated into 24 final samples. Fractions are concatenated based on a well pattern of: Fraction 1 ¼ A1 (well 1) þ C1 (well 25); Fraction 2 ¼ A2 (well 2) þ C2 (well 26); Fraction 3 ¼ A3 (well 3) þ C3 (well 27); and so on. 3. Dry fractions in a SpeedVac centrifuge and reconstitute in 20 μL of 0.1% FA, 1% DMSO in water. 4. Fractionated peptides can be stored indefinitely at 80  C. 3.6 Nano-UHPLC Column Preparation

Nano-UHPLC columns can be prepared in a wide variety of configurations using different approaches. In this work, a nano-HPLC system is used in combination with a capillary packing bomb in order to rapidly produce homogeneous columns. 1. Prepare standard fritted capillaries for analytical column reservoirs (Fig. 2a): (a) Trim 200 μm ID capillaries in 15 cm lengths. Two will be needed to prepare a single column. (b) Prepare the frit solution by combining 30 μL of Kasil solution with 10 μL of formamide (3:1 v/v ratio), pipette mix, and vortex for 15 s. (c) Dip the capillary ends into the prepared frit liquid and wipe the excess off the outside of the tubing using a lintfree tissue. (d) Lay the capillaries on a room temperature heating block and turn the temperature to “high” (95  C setting) (see Note 11). The capillaries can be secured by placing a glass bottle on top of them. (e) Wait for 30 min for frits to polymerize. 2. Prepare fritted nanospray tip capillaries for analytical columns (Fig. 2a): (a) Trim 200 μm ID capillaries in 65 cm lengths. (b) Turn on P-2000 laser puller and allow it to warm up for 15 min. (c) Remove the polyimide capillary coating for ~3 cm in the center of the 65 cm trimmed tubing lengths using a butane torch (see Note 12).

Standardized Protocol for Bottom-up Proteomics with MS

77

(a) Reservoir column 200µ m internal diameter

frit 15cm length

Analytical column

frit

200µm internal diameter 25cm length

(b) Column packing setup nano HPLC

Reservoir column frit

Analytical column

Stainless steel union

200µ m internal diameter

200µ m internal diameter

bead flow direction

(c) Easy nLC-1000 nanospray setup nano UHPLC

to W-valve

Column out 50µ m internal diameter

Analytical 200µ m internal diameter

20cm length

25cm length high-voltage connection

Fig. 2 Microcapillary configuration for packing and Easy nLC-1000 nano-UHPLC setup. (a) Schematic depicts the required capillaries and directionality of frits that are required for packing of analytical columns. (b) Schematic depicts the setup for packing analytical columns using a “reservoir capillary” and a nano-HPLC for backflushing. (c) Schematic depicts the configuration for setup of the Easy nLC-1000 nano-UHPLC instrument for MS analysis. Shaded regions in capillaries denote packed (gray) or fritted (orange) areas

(d) Pull the tip in the region where the coating was removed using the P-2000 instrument (see Note 13). (e) After laser pulling, gently stroke (or “shave”) the pulled capillary tip end in an almost parallel motion on a ceramic cutting tile to open the tip (if necessary). (f) Prepare the frit solution by combining 30 μL of Kasil solution with 10 μL of formamide (3:1 v/v ratio), pipette mix, and vortex for 15 s. (g) Dip the opened capillary tip into the 3:1 Kasil/formamide frit mix and wipe the excess from the end using a lint-free tissue soaked in 50% methanol. (h) Lay the capillaries on a room temperature heating block (secure by placing a glass bottle on top of the capillaries) and turn the temperature to “high” (95  C setting). The capillaries can be secured by placing a glass bottle on top of them. (i) Wait for 30 min for frits to polymerize. (j) After polymerization, gently stroke the capillary tip on a ceramic cutting tile to open the tip again.

78

Christopher S. Hughes et al.

3. Pack an analytical column: (a) Attach a fritted nanospray tip capillary to the nano-HPLC pump outlet at a flow rate of 1 μL/min (60% Mobile Phase B) (see Note 14). If no flow is observed out of the nanospray tip, it can be shaved further to open it. (b) Prepare the slurry of 1.9 μm C18 beads by combining 25 mg of bead with 1 mL of acetone (see Note 15). (c) Vortex the bead slurry for 15 s, and further mix by placing in a sonic water bath for 5 min. (d) Attach a 200 μm ID standard fritted capillary to the packing bomb containing a glass vial filled with acetone. (e) Flush the capillary with acetone for 15 s (at a tank output pressure of ~100 psi) (see Note 16). (f) Release the tank pressure and exchange the acetone vial with the one containing the C18 bead slurry into the packing bomb and attach the standard fritted capillary. (g) Apply pressure (~100 psi) and pack the standard fritted capillary until it is full of beads across the entire length. This is the “reservoir capillary.” (h) Gently release the pressure from the column and remove the capillary from the packing bomb (see Note 17). (i) Attach the reservoir capillary and the fritted nanospray tip capillary together using a stainless steel junction and to the nano-HPLC pump such that the beads will flow into the analytical column (the fritted end of the reservoir capillary is toward the nano-HPLC) (Fig. 2b). (j) As 200 μm capillaries are used for both the reservoir and analytical column in this work, this process will have to be repeated to reach the desired column length (see Note 18). (k) When the desired length has been reached, finish-pack the column using the nano-HPLC pump to apply an increased pressure (e.g., >350 bar) via modulation of the flow rate. (l) After packing has been completed, gently release the pressure from the prepared column. The packed column can be stored indefinitely in 50% methanol or directly connected and used on the nano-UHPLC hardware. 3.7 Nano-UHPLC Configuration and LCMS Data Acquisition

Depending on your nano-UHPLC and MS system, configuration and acquisition can be carried out in a wide variety of ways (see Note 19). This workflow is based on the use of an Easy nLC-1000 system paired with an analytical column prepared in-house (see Note 20). The protocol described here is designed to achieve an optimal

Standardized Protocol for Bottom-up Proteomics with MS

79

balance between chromatography performance, minimal betweensample injection dead time, and downtime due to repairs (see Note 21). 1. Trim the prepared 200 μm ID analytical column to a length of 25 cm. 2. Replace the 20 μm
50 cm column-out line from the S-valve on the Easy nLC-1000 with a 50 μm
20 cm version (see Note 22). 3. Plumb the nano-UHPLC system using a PEEK MicroCross connector. The S-valve column-out line and the analytical column should be in series. The waste-out W valve line should be in series with the liquid junction voltage connection. 4. With the 200 μm ID column connected (column oven set to 50  C) at a flow rate of 1.5 μL/min (95% A), the backpressure should be 300–350 bar. 5. On 200 μm ID columns, injections of approximately 2–5 μg of peptide material for analysis in 0.5–1 h runs are targeted. 6. With the 200 μm ID column setup described above, the Easy nLC-1000 is used with the following settings: (a) Sample pickup: volume of 8 μL, flow of 25 μL/min (loop size is 10 μL). (b) Sample loading: volume of 8 μL, max pressure of 400 bar (expected flow rate is approximately 2 μL/min on a 25 cm column). (c) Gradient: minute 0 (flow—1500 nL/min, %B 3), minute 2 (flow—1500 nL/min, %B 7), minute 40 (flow— 1500 nL/min, %B 24), minute 48 (flow—1500 nL/ min, %B 40), minute 49 (flow—1500 nL/min, %B 80), minute 53 (flow—1500 nL/min, %B 80). The expected backpressure at 1500 nL/min is approximately 300–350 bar. (d) Analytical column equilibration: volume 6 μL, max pressure of 400 bar (expected flow rate is approximately 2 μL/ min on a 25 cm column). (e) The time to complete the equilibration and injection cycle using this configuration is approximately 7–8 min. 7. For MS acquisition, a wide variety of method parameters are possible and should be tailored to your analysis type. For a Orbitrap MS2 analysis of high-complexity TMT samples, our method is as follows: (a) Spray voltage: 2200 V. (b) Ion Transfer Tube Temperature: 325  C.

80

Christopher S. Hughes et al.

(c) MS-OT: Orbitrap Resolution of 60,000, scan range of 400–1200 m/z, an AGC Target of 4.0e5, a max injection time of 64 ms. (d) Monoisotopic precursor selection (MIPS) filter: Monoisotopic peak determination ¼ Peptide. (e) Charge State: 2–4, include undetermined charge states. (f) Dynamic Exclusion: exclude after 1 time, duration of 15 s, tolerance of 10 ppm in both high and low, exclude isotopes enabled, dependent scan on single charge state disabled. (g) Data dependent MS2 OT HCD: Quadrupole isolation, a 1.4 m/z isolation window, HCD activation, 40% collision energy (see Note 23), auto: m/z Normal scan range, 30,000 Orbitrap resolution, 110 m/z fixed first mass, 1.2e5 AGC target, 64 ms maximum injection time, injection ions for all available time OFF (see Note 24). (h) Overall cycle time of 3 s. 3.8 Analysis of Acquired MS Data

This data analysis routine assumes the sample set was TMT 10-plex labeled, and measured using an Orbitrap MS2 method as described above. If using the GUI versions of SearchGUI and PeptideShaker, the data processing is straightforward and there are already excellent walkthroughs on the developer’s website. The analysis described here will focus on the use of the command line versions of the tools, SearchCLI and PeptideShakerCLI. Numerous other tools can be efficiently substituted here for data analysis. 1. Process data files in RawTools using an appropriate command based on your data acquisition method. For TMT 10-plex with MS2, this would be: RawTools.exe parse -f input_file. raw -quxmR -r TMT10. Processing with RawTools can be carried out directly on the raw files in Windows, OSX, and Linux. 2. Create and configure your database file. (a) Download the organism appropriate database from UniProt. (b) Add contaminant proteins to the above database. These can be downloaded from: http://www.thegpm.org/ crap/. The identifiers can be processed to match the parsing rules for the above downloaded UniProt database. (c) Process the database to generate a concatenated targetdecoy version using FastaCLI with the –decoy tag. 3. Configure SearchCLI to perform your search. (a) Use IdentificationParametersCLI parameters file.

to

generate

a

Standardized Protocol for Bottom-up Proteomics with MS

81

(b) Depending on your search engines, pay special attention to the parameters for each engine (e.g., MS-GFþ has special parameters for TMT, and instrument type). Most available parameters for general processing can be observed by invoking IdentificationParametersCLI without additional flags. (c) Invoke IdentificationParametersCLI with the –mods flag to see the available modifications for the search engines. (d) Specify to use the database generated above as part of the parameters file. 4. Start SearchCLI. (a) Run SearchCLI using the generated parameters file and database. (b) Specify the search engines to be used with the appropriate tags. (c) Specify individual outputs if processing multiple files to ease computational requirements using the –output_option 1 flag. If processing multiple fractions, these can be combined later using PeptideShaker. 5. Process the search results with PeptideShakerCLI using the same parameters and database files as with SearchCLI. Depending on the number and sizes of the files being processed, PeptideShaker can be very RAM intensive. 6. Output the results of PeptideShakerCLI using the ReportCLI and MzidCLI tools. ReportCLI will create standard text outputs of the desired data sets (e.g., peptides, proteins). MzidCLI will create publication-appropriate mzID files. 7. Combine the peptide spectral match or peptide data from ReportCLI with the quantification values output in the text reports from RawTools.

4

Notes 1. Although the use of the FastPrep instrument is described here, samples processed via other methods (e.g., sonication, standard bead beating, dounce homogenization) are equally suitable with this workflow. 2. The lysis buffer composition used here is designed as a one-size-fits-all for most proteomics applications due to the wide range of detergents present. The composition of this lysis buffer is by no means strict and can be changed to one suitable for your own analysis.

82

Christopher S. Hughes et al.

3. There are a wide variety of bead types that can be used to promote cell disruption. One pitfall of using beads in general is that recovery of the entire original sample can be challenging. To enhance recovery, one option is to poke a small hole in the bottom of the tube containing the beads using a heated needle, followed by low-speed centrifugation into a collection tube. 4. The program utilized here is a guideline and can be changed depending on your material. If a significant amount of intact chromatin or poor overall lysis after bead beating is observed, add additional cycles, extend the duration, or change the type of bead used. 5. The SeraMag carboxylate beads utilized in this protocol are provided at an approximate stock concentration of 50 mg/mL. SP3 is generally performed at a minimum of 10:1 μg/μg of beads:protein. However, it is recommended to maintain a minimum working concentration of 0.5 μg/μL for the beads in the lysate mixture to ensure efficient protein binding. Therefore, when working with small input amounts or large volumes, scale the quantity of beads to maintain this minimum bead concentration. To prepare the stock solution for general use, rinse the beads in water using the magnetic rack and reconstitute at a suitable working concentration (e.g., 50 μg/μL) in water. Scale the volume of bead stock prepared based on the number of samples and amount of protein to be processed. Prepared bead stocks can be stored at 4  C 1 month. 6. Depending on the amount of protein present, the beads may clump and become very sticky immediately after the addition of ethanol. To avoid material loss due to beads sticking to the pipette tip, minimize the amount of post-addition pipetting that is performed. 7. Try to remove as much ethanol as possible with pipetting prior to digestion. However, it is not necessary to air-dry the beads prior to the addition of the digestion solution. 8. Generally, 100 μL volumes of digestion solution provide a good balance between reconstitution and limited sample dilution. The amount of trypsin can be scaled based on preference. 9. Avoid pipette mixing the beads during reconstitution. Depending on how much material was processed with SP3, the beads may be very sticky and material can be lost on pipette tips. It is recommended to add the digestion solution and simply push the beads into it using a pipette tip. Sonication in a water bath will help break up the bead clump, which can then be pipetted. 10. The reconstituted label can be stored at 80  C. It is critical to ensure the tube lid is secure and wrapped with Parafilm M™ (Bemis) during storage. The reagent can be freeze-thawed for

Standardized Protocol for Bottom-up Proteomics with MS

83

use as required. Reseal with fresh Parafilm prior to each refreezing. 11. It is important to not preheat the heating block. Putting the capillaries containing the liquid-frit on a preheated unit can cause the frit liquid to bubble and migrate, generating uneven frits. Gradually increase the temperature during frit polymerization. 12. Do not overheat any region of the capillary tube with the butane torch while attempting to burn off the coating as doing so will cause the capillary to become brittle and prone to breakage. To help avoid overheating of the capillary, rotate the tube while heating and just singe the coating and wipe it off with a wet lint-free tissue. 13. A variety of programs are suitable to generate nanospray tips. A stepwise program of (1) Heat ¼ 320, Vel ¼ 40, Del ¼ 200; (2) Heat ¼ 310, Vel ¼ 30, Del ¼ 200; (3) Heat ¼ 300, Vel ¼ 25, Del ¼ 200; (4) Heat ¼ 290, Vel ¼ 20, Del ¼ 200 is used. 14. The pressure generated from a 1 μL/min flow rate (60% Mobile Phase B) over this capillary should be almost zero. If the pressure climbs above 1000 psi prior to packing, gently stroke the tip on the ceramic cutting tile as before to further open the tip. 15. In this work, acetone is used as the slurry solvent with the 1.9 μm C18 beads. Acetone gives a uniform suspension of the beads while also giving good flow rates during packing. Alternative solvents and combinations can be used in place of acetone. If changing solvents, be sure to check a small droplet of the bead slurry by microscopy to examine the uniformity of the suspension. 16. Performing packing steps on the capillary bomb at lower pressures is recommended. Because long columns are not being packed on the pressure bomb itself, the use of high pressure isn’t necessary. Finish packing at high pressures is performed on the nano-HPLC instrument. 17. Release the tank pressure slowly. If the pressure is released too quickly, the beads will “back-out” of the capillary. 18. If using capillaries with different internal diameters, the same procedure can be used. However, as the 200 μm reservoir capillary will have a larger ID than the desired analytical column (e.g., 75 μm ID column), the process can be completed in a single shot. 19. The protocol described in this work utilizes an Easy nLC-1000 system coupled to an Orbitrap Fusion MS and data analysis with open-source packages. The described analytical column

84

Christopher S. Hughes et al.

setup is compatible with a wide variety of alternative nanoUHPLC systems and can be easily adapted based on your own configuration or preferences. The MS analysis of the generated samples that is described here (MS2-only acquisition) can be carried out on any platform capable of resolving the TMT tags, such as the Q-Exactive series and qToF systems. If using a Fusion or Fusion Lumos system, synchronous precursor selection (SPS)-MS3 can be used (excellent method templates are available in the Fusion software). For data analysis, numerous other freely available packages can be used to efficiently process the data, such as MaxQuant, the TransProteomic Pipeline, and OpenMS. 20. This work describes the use of a one-column analytical setup. This setup is based on a wide-bore capillary column that facilitates fast equilibration and compatibility with high peptide loads. The higher ID of this column will result in reduced sensitivity compared with a smaller ID that can be compensated using increased peptide loads. If sample is limiting in quantity, a smaller ID column should be used. Alternatively, a trapping-analytical setup can also be used. The use of a trapping column as affords an additional sample cleanup step that helps reduce the amount of contaminant that is sprayed into the MS. Trapping columns are easy to prepare, inexpensive, and extend the lifetime of the analytical column (the same procedure described here to prepare the “reservoir” capillary can be used to make a trap column). In addition, with a trapping column, sample loading can be completed at much higher flow rates due to the lower backpressure generated by this short column, thus reducing between-injection dead time. Lastly, if using large quantities of peptide material during injections the trapping column can aid in protecting the separation performance of the analytical column by helping combat overloading. 21. Specific changes to the Easy nLC-1000 system can be made to maximize reliability without negatively impacting overall performance. Reduced reliability of the Easy nLC 1000 system can be observed when consistently operating at ultra-high pressures (>500 bar). The result is more MS downtime due to nano-UHPLC repairs and greater incurred expense due to lost run time and replacement part purchases. To enhance reliability of the Easy nLC-1000 system, all equilibration and loading steps are carried out at a maximum of 400 bar. 22. The standard easy nLC1000 liquid junction setup is configured with a 20 μm ID
50 cm length nanoViper capillary for the “column-out” line. This capillary is prone to clogging and generates unnecessary backpressure when working with nano and micro-flow column setups. Replacing this tube with a

Standardized Protocol for Bottom-up Proteomics with MS

85

wider bore variety (50 μm ID
20 cm length) can improve reliability and reduce system backpressure. To manufacture a liquid-junction-compatible capillary, a 50 μm ID
65 cm length double-ended nanoViper capillary can be cut in half with the protective sleeve trimmed back to generate two 20 cm length tubes. In newer generations of the Easy nLC (e.g., the 1200 series), the column-out line is equipped with an in-line filter to help reduce clogging events. Even with the filter unit installed the larger ID capillary has the benefit of reducing backpressure at the higher flowrates used during equilibration and loading which decreases lost time between sample injections. Although the 50 μm ID tube setup does increase the dead volume of the system, the overall modification is by approximately 6 s at a standard flow rate of 1500 nL/min (the increase in dead volume is approximately 150 nL). These changes are optional. 23. It is important to optimize the collision energy for each specific MS system in all scenarios (e.g., when using TMT labels). Running a standard sample across a range of collision energies can help to determine the optimum setting for a given MS. 24. It is important to balance the max ion injection time and AGC ion target values to ensure optimal operation efficiency of your MS [27]. After optimization of the collision energy, performing individual tests with a variety of AGC and maximum fill time target values can help to determine the best setting for a given sample type. With samples of all complexities it can be optimal to use high ion target values to modulate scan duration with the ion injection time parameter. For example, a high AGC target parameter will result in a majority of fragmented precursors being unable to hit to specified value, and most scans will ultimately extend for the entire allotted fill time. As a result, the fill time setting will effectively dictate the amount of time spent doing MS2 scans. With a highcomplexity sample, it can be beneficial to use a low value for the ion injection time parameter (60 ms with the OT) to maximize the number of MS2 scans performed. Conversely, when working with a low complexity sample, it can be beneficial to use a high ion injection time value (120 ms) to maximize MS2 spectral quality. Selecting this time such that the detector and fill times operate in parallel will ensure the most effective use of the MS. When using TMT, it is important to maintain a sufficient ion target value to ensure optimal ion count metrics for the reporter ions (e.g., >80 ms) [28].

86

Christopher S. Hughes et al.

Acknowledgments C.S.H. would like to acknowledge valuable discussions with Lida Radan. References 1. Larance M, Lamond AI (2015) Multidimensional proteomics for cell biology. Nat Rev Mol Cell Biol 16(5):269–280. https://doi.org/10. 1038/nrm3970 2. Gillet LC, Leitner A, Aebersold R (2016) Mass spectrometry applied to bottom-up proteomics: entering the high-throughput era for hypothesis testing. Annu Rev Anal Chem (Palo Alto, Calif) 9(1):449–472. https://doi. org/10.1146/annurev-anchem-071015041535 3. Weston LA, Bauer KM, Hummon AB (2013) Comparison of bottom-up proteomic approaches for LC-MS analysis of complex proteomes. Anal Methods 5(18):4615–4621. https://doi.org/10.1039/C3AY40853A 4. Zhang Y, Fonslow BR, Shan B et al (2013) Protein analysis by shotgun/bottom-up proteomics. Chem Rev 113(4):2343–2394. https:// doi.org/10.1021/cr3003533 5. Hughes CS, Foehr S, Garfield DA et al (2014) Ultrasensitive proteome analysis using paramagnetic bead technology. Mol Syst Biol 10:757. https://doi.org/10.15252/msb. 20145625 6. Hughes CS, McConechy MK, Cochrane DR et al (2016) Quantitative profiling of single formalin fixed tumour sections: proteomics for translational research. Sci Rep 6:34949. https://doi.org/10.1038/srep34949 7. Moggridge S, Sorensen PH, Morin GB et al (2018) Extending the compatibility of the SP3 paramagnetic bead processing approach for proteomics. J Proteome Res 17 (4):1730–1740. https://doi.org/10.1021/ acs.jproteome.7b00913 8. Bakalarski CE, Kirkpatrick DS (2016) A biologist’s field guide to multiplexed quantitative proteomics. Mol Cell Proteomics 15 (5):1489–1497. https://doi.org/10.1074/ mcp.O115.056986 9. Domon B, Aebersold R (2010) Options and considerations when selecting a quantitative proteomics strategy. Nat Biotechnol 28 (7):710–721. https://doi.org/10.1038/nbt. 1661

10. Ong SE, Mann M (2005) Mass spectrometrybased proteomics turns quantitative. Nat Chem Biol 1(5):252–262. https://doi.org/ 10.1038/nchembio736 11. Paulo JA, O’Connell JD, Gaun A et al (2015) Proteome-wide quantitative multiplexed profiling of protein expression: carbon-source dependency in Saccharomyces cerevisiae. Mol Biol Cell 26(22):4063–4074. https://doi. org/10.1091/mbc.E15-07-0499 12. Mertins P, Mani DR, Ruggles KV et al (2016) Proteogenomics connects somatic mutations to signalling in breast cancer. Nature 534 (7605):55–62. https://doi.org/10.1038/ nature18003 13. Zecha J, Meng C, Zolg DP et al (2018) Peptide level turnover measurements enable the study of proteoform dynamics. Mol Cell Proteomics 17(5):974–992. https://doi.org/10. 1074/mcp.RA118.000583 14. Kuljanin M, Dieters-Castator DZ, Hess DA et al (2017) Comparison of sample preparation techniques for large-scale proteomics. Proteomics 17(1–2). https://doi.org/10.1002/ pmic.201600337 15. Yang F, Shen Y, Camp DG et al (2012) HighpH reversed-phase chromatography with fraction concatenation for 2D proteomic analysis. Expert Rev Proteomics 9(2):129–134. https://doi.org/10.1586/epr.12.15 16. Spicer V, Ezzati P, Neustaeter H et al (2016) 3D HPLC-MS with reversed-phase separation functionality in all three dimensions for largescale bottom-up proteomics and peptide retention data collection. Anal Chem 88 (5):2847–2855. https://doi.org/10.1021/ acs.analchem.5b04567 17. Eliuk S, Makarov A (2015) Evolution of Orbitrap mass spectrometry instrumentation. Annu Rev Anal Chem (Palo Alto, Calif) 8:61–80. https://doi.org/10.1146/annurev-anchem071114-040325 18. Beck S, Michalski A, Raether O et al (2015) The impact II, a very high-resolution quadrupole time-of-flight instrument (QTOF) for deep shotgun proteomics. Mol Cell

Standardized Protocol for Bottom-up Proteomics with MS Proteomics 14(7):2014–2029. https://doi. org/10.1074/mcp.M114.047407 19. Andrews GL, Simons BL, Young JB et al (2011) Performance characteristics of a new hybrid quadrupole time-of-flight tandem mass spectrometer (TripleTOF 5600). Anal Chem 83(13):5442–5446. https://doi.org/10. 1021/ac200812d 20. Schilling B, Gibson BW, Hunter CL (2017) Generation of high-quality SWATH® acquisition data for label-free quantitative proteomics studies using TripleTOF® mass spectrometers. In: Comai L, Katz J, Mallick P (eds) Proteomics. Methods in molecular biology, vol 1550. Humana Press, New York, NY, pp 223–233. https://doi.org/10.1007/978-1-4939-67476_16 21. Senko MW, Remes PM, Canterbury JD et al (2013) Novel parallelized quadrupole/linear ion trap/Orbitrap tribrid mass spectrometer improving proteome coverage and peptide identification rates. Anal Chem 85 (24):11710–11714. https://doi.org/10. 1021/ac403115c 22. Nesvizhskii AI, Vitek O, Aebersold R (2007) Analysis and validation of proteomic data generated by tandem mass spectrometry. Nat Methods 4(10):787–797. https://doi.org/ 10.1038/nmeth1088 23. Schmidt A, Forne I, Imhof A (2014) Bioinformatic analysis of proteomics data. BMC Syst

87

Biol 8(Suppl 2):S3. https://doi.org/10. 1186/1752-0509-8-S2-S3 24. Vaudel M, Barsnes H, Berven FS et al (2011) SearchGUI: an open-source graphical user interface for simultaneous OMSSA and X!Tandem searches. Proteomics 11(5):996–999. https://doi.org/10.1002/pmic.201000595 25. Vaudel M, Burkhart JM, Zahedi RP et al (2015) PeptideShaker enables reanalysis of MS-derived proteomics data sets. Nat Biotechnol 33(1):22–24. https://doi.org/10.1038/ nbt.3109 26. Kovalchik KA, Moggridge S, Chen DDY et al (2018) Parsing and quantification of raw Orbitrap mass spectrometer data using RawTools. J Proteome Res 17(6):2237–2247. https://doi. org/10.1021/acs.jproteome.8b00072. PMID: 30462513 27. Hughes CS, Spicer V, Krokhin OV et al (2017) Investigating acquisition performance on the Orbitrap fusion when using tandem MS/MS/ MS scanning with isobaric tags. J Proteome Res 16(5):1839–1846. https://doi.org/10. 1021/acs.jproteome.7b00091 28. Hughes CS, Zhu C, Spicer V et al (2017) Evaluating the characteristics of reporter ion signal acquired in the Orbitrap analyzer for isobaric mass tag proteome quantification experiments. J Proteome Res 16(5):1831–1838. https:// doi.org/10.1021/acs.jproteome.7b00092

Chapter 6 Analyzing Cerebrospinal Fluid Proteomes to Characterize Central Nervous System Disorders: A Highly Automated Mass Spectrometry-Based Pipeline for Biomarker Discovery Antonio Nu´n˜ez Galindo, Charlotte Macron, Ornella Cominetti, and Loı¨c Dayon Abstract Over the past decade, liquid chromatography tandem mass spectrometry (LC MS/MS)-based workflows become standard for biomarker discovery in proteomics. These medium- to high-throughput (in terms of protein content) profiling approaches have been applied to clinical research. As a result, human proteomes have been characterized to a greater extent than ever before. However, proteomics in clinical research and biomarker discovery studies has generally been performed with small cohorts of subjects (or pooled samples from larger cohorts). This is problematic, as when aiming to identify novel biomarkers, small studies suffer from inherent and important limitations, as a result of the reduced biological diversity and representativity of human populations. Consequently, larger-scale proteomics will be key to delivering robust biomarker candidates and enabling translation to clinical practice. Cerebrospinal fluid (CSF) is a highly clinically relevant body fluid, and an important source of potential biomarkers for brain-associated damage, such as that induced by traumatic brain injury and stroke, and brain diseases, such as Alzheimer’s disease and Parkinson’s disease. We have developed a scalable automated proteomic pipeline (ASAP2) for biomarker discovery. This workflow is compatible with larger clinical research studies in terms of sample size, while still allowing several hundred proteins to be measured in CSF by MS. In this chapter, we describe the whole proteomic workflow to analyze human CSF. We further illustrate our protocol with some examples from an analysis of hundreds of human CSF samples, in the specific context of biomarker discovery to characterize central nervous system disorders. Key words Alzheimer, Automation, Biomarker, Brain, Clinical research, CSF, Depletion, Human, Isobaric tagging, Large scale, Mass spectrometry

1

Introduction In complement to proteomic analysis performed directly on blood, analysis of other biological fluids (e.g., cerebrospinal fluid (CSF), urine, saliva, lacrymal fluid, synovial fluid, bronchoalveolar lavage fluid, nipple aspirate, and amniotic fluid) could be relevant to a range of applications in human health and disease, in particular for

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_6, © Springer Science+Business Media, LLC, part of Springer Nature 2019

89

90

Antonio Nu´n˜ez Galindo et al.

Fig. 1 Radar plot of analytical figures-of-merit of the ASAP2 workflow compared to general proteomic workflows. For both workflow types, quantitative performances must be high, with reasonable cost and low required sample volumes. The ASAP2 workflow (in blue) provides higher throughput, but reduced proteome coverage compared to general proteomic workflows (in orange)

the discovery of biomarkers. Very deep proteome coverage of these biological fluids is now possible with the most recent mass spectrometry (MS)-based technologies; however, it can come with significant constraints in terms of analytical timelines. Indeed, time is a critical factor when performing large clinical research studies. Relevant strategies (Fig. 1) must be designed to provide adequate proteome coverage within a realistic analysis time frame for these methods to be adopted in clinical research. Neurodegenerative diseases are progressing rapidly due to aging demographics. In 2015, 46.8 million people were living with dementia, and predictions suggest that this number will almost double every 20 years from now, reaching 74.7 million in 2030 and 131.5 million in 2050 (World Alzheimer Report 2015). Alzheimer’s disease (AD) is the most common form of dementia and represents 60–80% of cases in older people [1]. AD is characterized by a long initial asymptomatic phase before the appearance of the first clinical symptoms [2]. Today, a diagnosis of AD can only be confirmed at brain autopsy; there is therefore an urgent need for other means to support this diagnosis as early as possible. The pathogenesis of AD is also still quite poorly understood. Altered proteome profiles for some tissues and body fluids have been reported in AD patients. Some circulating protein biomarkers have been reported to be able to sort AD or mild-cognitive impairment from non-dementia controls [3], but well-accepted molecular biomarkers are very rare and are generally measured in CSF [4, 5].

CSF Proteomics for Biomarker Discovery

91

The relevance of CSF as an important source of potential biomarkers for brain damage and brain diseases is mainly linked to its proximity to the brain. To some extent, the molecular composition of CSF reflects the brain’s biochemical and metabolic status and changes. In particular, CSF contains some brain-specific proteins [6, 7]. Despite a significant number of discovery studies, very few protein biomarker candidates for brain disorders have been translated to the clinic. Many results have not been replicated, which prevents their use for diagnostic, prognostic, and/or intervention monitoring purposes in the clinic. The failure to replicate results might stem from a number of causes, including disease complexity. The proteomic approach also has often a limited ability to deliver robust and usable biomarkers. Indeed, in many cases the initial sample size for the discovery studies is limited, potentially generating not only false positive but also false negative results. Under such circumstances, verification and validation of discovery findings becomes highly challenging. In this chapter, we describe a proteomic pipeline (Fig. 2a) for the analysis of human CSF samples. This pipeline is compatible with clinical research studies involving large subject cohorts, with numbers toward those used in genomics, and increasingly in metabolomics studies. The methodology considers analytical throughput as a compulsory feature and provides solutions to some of the current limitations of MS-based proteomic workflows. Automation and integration of sample processing tasks offer innovative solutions to increase reproducibility and throughput for sample handling and subsequent liquid chromatography tandem MS (LC MS/MS) analysis. Importantly, we describe how a laboratory information management system (LIMS) can be used to aid sample tracking, record analytical steps, and automatically process data. We also report on our own experience of analyzing CSF in the field of AD.

2

Materials

2.1 Sample Preparation and Depletion

1. 1.5-mL polypropylene tubes. 2. Centrifuge and speed-vac system. 3. MilliQ water (18.2 MΩ cm at 25
C). 4. 12-channel 10–100 μL multi-pipette. 5. Sterile, clear 96-well filter plate with 0.22 μm pore size PVDF membrane. 6. Vacuum manifold. 7. 96-well polypropylene plate (V-bottom). 8. Adhesive sealing mat.

92

Antonio Nu´n˜ez Galindo et al.

Fig. 2 ASAP2 workflow and representations of the liquid handling system. (a) Human CSF samples spiked with β-lactoglobulin (LACB) are depleted to eliminate 14 proteins. Using the liquid handling system, flow-through proteins are digested, labeled with TMT (6-plex), pooled and purified. Samples are analyzed by RP-LC MS/MS using a hybrid Orbitrap mass spectrometer. A 3D view (b) and view from above (c) of the liquid handling system’s deck are displayed: ❶ reagents (e.g., TCEP, IAA, TMT and endoprotease), ❷ plate heater and shaker, ❸ dark chamber, ❹ disposable reservoirs for solvents, ❺ and ❻ reservoirs fed from stock bottles of solvent, ❼ RP elution 96-DeepWell plate, and ❽ SCX elution 96-DeepWell plate

9. Multiple affinity removal system (MARS) column Human 14, Buffer A, and Buffer B (Agilent Technologies). 10. β-Lactoglobulin (LACB) from bovine milk prepared at 0.0965 mg/mL in Buffer A. 11. High performance LC (HPLC) system. 12. Fraction collection system (CTC Analytics AG). 13. 1.4- and 1.0-mL tubes, screw-caps, and 2D barcode reader (FluidX™). 14. Supports for FluidX™ tubes.

CSF Proteomics for Biomarker Discovery

93

15. Liquid nitrogen. 16. Container for sample snap freezing. 17. Low-temperature freezer (80
C). 2.2

Buffer Exchange

1. 2D barcode reader and tube capper/decapper (FluidX™). 2. Centrifuge and speed-vac system. 3. Strata-X 33u Polymeric reversed-phase (RP) (30 mg/mL) cartridges (Phenomenex). 4. Vacuum manifold and 96-well holder (Phenomenex). 5. Support for FluidX™ tubes. 6. MilliQ water (18.2 MΩ cm at 25
C). 7. Acetonitrile (CH3CN, 99.9% stock solution) and trifluoroacetic acid (TFA). 8. Conditioning buffer: 0.08% TFA in CH3CN. 9. Washing buffer: 0.1% TFA in MilliQ water. 10. Elution buffer: 70% CH3CN, 0.08% TFA in MilliQ water. 11. Vibrating platform. 12. 12-channel 50–1200 μL multi-pipette. 13. Pipette tips 50–1250 μL, two different lengths (76 and 103 mm). 14. 2.0-mL 96-DeepWell plates. 15. 96-well sealing mats. 16. Low-temperature Freezer (80
C).

2.3 Reduction, Alkylation, and Proteolytic Digestion of Proteins

1. MilliQ water (18.2 MΩ cm at 25
C). 2. 1.5-mL polypropylene tubes. 3. 100 mM triethylammonium hydrogen carbonate buffer (TEAB), pH 8.5 in MilliQ water from 1 M stock solution. 4. 2% (w/v) sodium dodecyl sulfate (SDS) in MilliQ water. 5. 20 mM tris(2-carboxyethyl) phosphine hydrochloride (TCEP) in MilliQ water. 6. 150 mM iodoacetamide (IAA) in CH3CN. 7. 0.25 μg/μL sequencing-grade modified trypsin/Lys-C (Promega) in 100 mM TEAB. 8. 4-channel Microlab Star liquid handling system (Hamilton). 9. Tube racks, temperature-controlled mixer, temperaturecontrolled dark chamber, and reservoirs mounted on the deck of the liquid handling system (Fig. 2b, c). 10. 96-well sealing mats.

94

Antonio Nu´n˜ez Galindo et al.

2.4 Tandem Mass Tag (TMT) Labeling

1. TMT reagents (Thermo Scientific). 2. MilliQ water (18.2 MΩ cm at 25
C). 3. CH3CN (99.9% stock solution) and TFA. 4. 5% (w/v) hydroxylamine in MilliQ water from a 50 wt.% hydroxylamine solution in water. 5. RP loading buffer: 5% CH3CN, 0.1% TFA in MilliQ water. 6. 1.5-mL and 0.5-mL polypropylene tubes. 7. 5-mL polypropylene tubes. 8. 4-channel Microlab STAR liquid handling system (Hamilton). 9. Tube racks and temperature-controlled mixer mounted on the deck of the liquid handling system (Fig. 2b, c). 10. Vortex mixer and centrifuge. 11. 96-well sealing mats.

2.5 Sample Purification

1. Oasis HLB 1 cc (30 mg) extraction cartridges (Waters). 2. Strata-X-C 33u Polymeric Strong Cation (SCX) (30 mg/mL) cartridges (Phenomenex). 3. 2.0-mL 96-DeepWell plates. 4. MilliQ water (18.2 MΩ cm at 25
C). 5. CH3CN (99.9% stock solution) and TFA. 6. RP conditioning buffer: 95% CH3CN, 0.1% TFA in MilliQ water. 7. RP loading buffer: 5% CH3CN, 0.1% TFA in MilliQ water. 8. RP elution buffer: 50% CH3CN, 0.1% TFA in MilliQ water. 9. SCX loading buffer: 25% CH3CN, 0.1% TFA in MilliQ water. 10. SCX elution buffer: 25% CH3CN, 400 mM ammonium acetate (NH4OAc) in MilliQ water. 11. 4-channel Microlab STAR liquid handling system (Hamilton). 12. Vacuum manifold mounted on the deck of the liquid handling system (Fig. 2b). 13. Speed-vac system.

2.6 RP-LC MS/MS Analysis

1. Nano-HPLC instrument: for this work we classically use an Ultimate 3000 RSLC nano system (Thermo Scientific). 2. Pre-column and analytical column: we use a PepMap 300 μm  5 mm (C18, 5 μm, 100 A˚) pre-column and a PepMap (C18, 2 μM, 75 μm  500 mm, nanoViper, fused silica, 1200 bars) analytical column (Thermo Scientific). 3. Stainless steel nanobore emitter (40 mm, OD 1/3200 ) mounted on a Nanospray Flex Ion Source (Thermo Scientific). 4. MilliQ water (18.2 MΩ cm at 25
C).

CSF Proteomics for Biomarker Discovery

95

5. CH3CN (99.9% stock solution) and formic acid (FA, 99% stock solution). 6. LC loading solvent: 2% CH3CN, 0.05% TFA in MilliQ water. 7. LC solvent A: 2% CH3CN 2%, 0.1% FA in MilliQ water. 8. LC solvent B: 80% CH3CN, 0.08% FA in MilliQ water. 9. LC vials. 10. Orbitrap Fusion Lumos Tribrid (Thermo Scientific) (see Note 1). 2.7

Data Analysis

1. Proteome Discoverer (version 1.4) (Thermo Scientific). 2. Mascot (version 2.4.2) (Matrix Sciences). 3. Human UniProtKB/Swiss-Prot database (see Note 2). 4. Scaffold Qþ 4.7.2 (Proteome Software) to be used with X! Tandem (CYCLONE version 2010.12.01.1). 5. Excel (Microsoft). 6. R (version 3.2.4, 2016-03-10) (http://www.r-project.org/) and mixOmics 6.2.0, ggplots2 2.2.1, dplyr 0.7.3, and reshape2 1.4.2 R packages. 7. Prism (GraphPad Software).

3 3.1

Methods General Practice

In the last few years, we have developed a scalable automated proteomic pipeline (ASAP2) (Fig. 2a) [8–10] to be deployed in clinical research studies. This method can typically measure several hundred of proteins with sufficient throughput to analyze hundreds of clinical samples [11–13]. In this chapter, we describe how ASAP2 can be specifically use to prepare human CSF samples and perform MS-based proteomic analysis [9]. The pipeline is tailored for human CSF samples; it involves evaporation, depletion of abundant proteins, buffer exchange, followed by automated reduction, alkylation, protein digestion, TMT labeling, sample combination, and purification before RP-LC MS/MS (Fig. 2a). The whole process can be considered to represent serial but independent analytical blocks, facilitating workflow flexibility. Abundant proteins are depleted by immuno-affinity using antibody-based columns and LC systems equipped with a refrigerated autosampler and fraction collector. Depleted samples are then frozen and stored ready for the next step: buffer exchange performed in a 96-well plate format. The remainder of the process is fully automated and takes place on a liquid handling workstation (Fig. 2b, c). It includes (1) sample solubilization; (2) reduction, alkylation, enzymatic digestion; (3) TMT labeling and sample pooling; (4) RP solid-phase extraction (SPE) purification; and (5) SCX

96

Antonio Nu´n˜ez Galindo et al.

SPE purification. The liquid handling platform is used to dispense predefined volumes of reagents with a pipetting arm; the mechanized arm also acts as a gripper to move plates on the deck. In addition, the system includes a block unit to mix (Fig. 2c, ❷), and/or heat 96-well plates (Fig. 2c, ❷ and ❸), and a dark chamber where alkylation reaction can be performed (Fig. 2c, ❸). A vacuum manifold is mounted on the platform for fully automated SPE of up to 96 samples at a time, with reservoirs for solvent storage (Fig. 2c, ❹, ❺, and ❻). Finally, samples are analyzed by RP-LC MS/MS. The whole process is assisted by a LIMS (Clarity from Genologics) that provides an interface to create, view, and edit sample tracking information. 3.2 Sample Preparation

1. Switch-on the speed-vac and check that both vacuum and temperature parameters are working correctly. 2. Aliquot CSF samples (i.e., 400 μL) into 1.5-mL tubes (see Note 3) and scan barcodes for tracking purposes (see Notes 4 and 5). 3. Dry samples in the speed-vac and store them at 80
C until needed (see Note 3). 4. The day the depletion is performed (see Note 6), use the LIMS to track and check the identity of the tubes containing the dried CSF samples to be depleted. 5. Dissolve the dried CSF samples in 125 μL of Buffer A containing 9.65 μg/mL LACB (Fig. 2a). 6. Transfer 120 μL of the diluted sample to a 0.22 μm filter plate, and filter using a vacuum manifold for 1 min (Fig. 2a). 7. Rapidly centrifuge the reception plate to spin down the diluted/filtered samples, cover with an adhesive mat. 8. Keep diluted/filtered CSF samples refrigerated until needed.

3.3 Depletion of Abundant Proteins

1. Check that the HPLC system (Fig. 2a) is functioning properly and refill Buffers A and B (see Note 7). Check volumes in waste bottles and empty if necessary. Mount the MARS depletion column (see Notes 8–10). Equilibrate the system with 100% Buffer A at a flow rate of 125 μL/min. Check that the LC pressure values are appropriate and stable, and monitor for leaks. 2. Place the 1.4-mL collection tubes in the autosampler to collect 1000 μL of the depleted CSF solution. Scan the barcodes of the collection tubes and register them in the LIMS (see Note 4). 3. Place 1.0-mL tubes in the autosampler and fill them with 900 μL of water for tool cleaning. 4. Place a vial filled with Buffer A for blank injections (see Note 7). 5. Prepare the sample injection sequence (see Notes 7 and 11).

CSF Proteomics for Biomarker Discovery

97

Table 1 LC chromatographic gradient used for depletion Time (min)

Buffer A (%)

Buffer B (%)

Flow rate (μL/min)

0.00

100

0

125

18.00

100

0

125

18.10

100

0

1000

20.00

100

0

1000

20.01

0

100

1000

27.00

0

100

1000

27.01

100

0

1000

36.70

100

0

1000

36.80

100

0

800

38.00

100

0

800

6. Take the diluted/filtered CSF samples, scan and record the barcode for the sample plate, and start recording the analytical step (see Note 4). 7. Place the polypropylene 96-well plate containing the diluted/ filtered CSF samples in the autosampler of the HPLC system, set temperature to 6
C. 8. Deplete samples using the HPLC method indicated in Table 1 (see Note 11) by systematically collecting the unbound fractions that elute first from the MARS column. Bound fractions contain the abundant proteins. These proteins can be eluted using Buffer B and discarded. 9. After sample depletion, collect the tubes containing the depleted CSF samples from the autosampler. Close the tubes with screw-caps. 10. Snap-freeze the samples in liquid nitrogen in an appropriate container. 11. Store the tubes at 80
C. 12. Record the analytical step and register any problems encountered in the LIMS after completing the process (see Note 4). 3.4

Buffer Exchange

1. Take 96 (or fewer) depleted CSF samples from the 80
C freezer and place them on a 96-well rack. 2. Let the samples thaw at room temperature for 60 min. If necessary, the samples can be placed in a container filled with cold water and agitated on a vibrating platform to accelerate the process.

98

Antonio Nu´n˜ez Galindo et al.

3. In the LIMS, record the sample barcodes and confirm that their identity matches the previously registered collection tubes used during the depletion process. Record the analytical step in the LIMS (see Note 4). 4. Rapidly centrifuge the tubes/96-well rack to spin down the liquid. Uncap tubes using the capper/decapper. 5. Use strata-X 33u Polymeric RP cartridges mounted on a 96-well holder and a vacuum manifold (Fig. 2a). 6. Using a 12-channel automatic pipette, condition the cartridges with 1 mL conditioning buffer (see Note 12). 7. Equilibrate the cartridges using the same 12-channel automatic pipette and 1 mL washing buffer (see Note 12). 8. Load the samples, introducing tips slowly into the cartridges using the multi-pipette fitted with long tips (see Notes 12 and 13). 9. Rinse the empty sample tubes with 300 μL washing buffer and slowly load the washing solutions onto the cartridges (see Notes 12 and 13). 10. Place a new clean 96-well plate in the vacuum manifold and elute samples with 1 mL elution buffer into a bar-coded 2.0mL 96-DeepWell plate (see Note 4). 11. At the end of elution, slowly increase the vacuum and wait for 1 min (see Note 12). 12. Recover the 96-DeepWell plate containing the bufferexchanged CSF samples and dry the samples in a speed-vac. 13. Recover the 96-DeepWell plate containing the dried samples and cover it with a polypropylene mat. 14. Store the bar-coded 96-DeepWell plate at 80
C. Confirm completion of the analytical step and register any problems encountered in the LIMS (see Note 4). 3.5 Reduction, Alkylation, and Proteolytic Digestion of Proteins

1. Place the 96-DeepWell plate containing the depleted and buffer-exchanged samples (Fig. 2a) on the robot deck (Fig. 2b, c, “Master Plate”). Check the barcode of the sample plate in the LIMS (see Note 4). Record the analytical step to be started. 2. Completely fill the 100 mM TEAB reagent reservoir (see Note 14) (see liquid handling system’s deck layout in Fig. 2b, “Solvents” and Fig. 2c, ❹). Fill the 20 mM TCEP and 2% SDS reagent tubes with 1 mL of each solution (Fig. 2b, “Reagents” and Fig. 2c, ❶) and place them, uncovered on the deck of the liquid handling system (see Note 14). 3. Start the automated run. The robot adds 5 μL of 2% SDS and 95 μL of 100 mM TEAB to each well. After shaking the

CSF Proteomics for Biomarker Discovery

99

96-DeepWell plate for 30 s, the robot adds 5.3 μL 20 mM TCEP to each well. Once all solutions have been dispensed, the robot shakes the 96-DeepWell plate for 30 s to mix before incubating for 1 h at 55
C to reduce disulfide bridges (Fig. 2b “Reduction Alkylation” and Fig. 2c, ❸) (see Note 15). 4. Just before the previous incubation finishes, prepare a fresh 150 mM IAA solution in CH3CN and transfer 1 mL to the appropriate reagent tube (Fig. 2b, “Reagents” and Fig. 2c, ❶). Place the uncovered tube on the deck of the liquid handling system (see Note 14). 5. After waiting for the samples to cool down to room temperature for 2 min, the robot adds 5.5 μL of the 150 mM IAA solution to each well, shakes the 96-DeepWell plate for 30 s and incubates the samples at room temperature for 1 h in the dark chamber (Fig. 2b, “Reduction Alkylation” and Fig. 2c, ❸) (see Note 15). 6. Freshly prepare the trypsin/Lys-C reagent at 0.25 μg/μL in 100 mM TEAB and fill the reagent tube (Fig. 2b, “Reagents” and Fig. 2c, ❶) with an appropriate volume to treat all samples (see Note 14). The robot adds 10 μL of the endoprotease mixture to each well, places a plastic cover on the 96-DeepWell plate and incubates overnight at 37
C with gentle shaking (see Note 15). 7. The next day, stop the automated program and the incubation process. 8. Rapidly centrifuge the 96-DeepWell plate containing the samples to spin down liquid. 9. Remove and discard soiled laboratory-ware. 10. Record completion of the analytical step and indicate problems encountered in the LIMS after the end of the process (see Note 4). 3.6

TMT Labeling

1. Place the 96-DeepWell sample plate on the deck of the liquid handling system (Fig. 2b, c, “Master Plate”). 2. Spin down the tubes containing the thawed/dried TMT reagents. Dissolve lyophilized TMT reagents in an adequate volume of CH3CN. Typically, 0.8 mg TMT reagent is needed for each sample to perform labeling; it should be dissolved in 41 μL of CH3CN (see Notes 16 and 17). Vortex for 30 s and spin down the liquid. 3. Transfer the dissolved TMT reagents to cleaned, appropriately labeled (i.e., 126, 127, 128, 129, 130, and 131) 500-μL tubes (see Note 17). Place the open tubes on the deck of the liquid handling system (Fig. 2b, “Reagents” and Fig. 2c, ❶).

100

Antonio Nu´n˜ez Galindo et al.

4. Start the automated run. The robot adds 41 μL of TMT solution to each sample in the 96-DeepWell plate in line with the design of the TMT experiments. Samples are incubated for 1 h at room temperature with shaking. 5. Record the analytical step in the LIMS (see Note 4). 6. Meanwhile, fill the hydroxylamine reagent tube with 1 mL of solution (Fig. 2b, “Reagents” and Fig. 2c, ❶). 7. To quench the TMT reaction and reverse occasional labeling of tyrosine, serine, and threonine residues, the robot adds 8 μL 5% hydroxylamine in water to each sample and incubates at room temperature for 15 min with shaking. 8. Next, the robot mixes/pools samples (six by six in case of 6-plex TMT) in a clean 5-mL tube (Fig. 2c, “Sample Pooling”). 96 initial samples will thus generate 16 TMT experiments at this stage. 9. Fill the dedicated reservoir with RP loading buffer (Fig. 2b, “Solvents” and Fig. 2c, ❹). 10. The robot washes each well of the 96-DeepWell plate with 120 μL of RP loading buffer and adds the washing liquid to the corresponding pooled sample. 11. Discard used and/or empty laboratory-ware. 12. Record completion of the analytical step and any problems encountered in the LIMS after the process ends (see Note 4). 13. Proceed to sample purification. 3.7 Sample Purification

1. RP purification. (a) Start recording the analytical step in the LIMS (see Note 4). (b) Fill the stock bottles containing the RP loading buffer and the RP elution buffer (see Note 18). Fill the RP conditioning buffer reservoirs (Fig. 2b, “Solvents” and Fig. 2c, ❹). (c) Place the Oasis HLB cartridges (1 cc, 30 mg) on the holder of the vacuum manifold positioned on the deck of the liquid handling system for purification of each mixed/pooled sample (Fig. 2a). Place a clean, bar-coded 96-DeepWell plate to receive purified samples (Fig. 2b, “RP & SCX SPE”). (d) Start the automated run. The liquid handler dilutes the peptide samples with 2.7 mL RP loading buffer. Next, the robot prepares the cartridges by flushing with 2  0.95 mL RP conditioning buffer and 4  0.95 mL of RP loading buffer. Samples are loaded (aspirating 5  0.86 mL from the 5-mL tubes containing the pooled

CSF Proteomics for Biomarker Discovery

101

TMT-labeled samples; Fig. 2c, “Sample Pooling”) and washed with 4  0.9 mL RP loading buffer. (e) Peptides are eluted with 2  0.8 mL RP elution buffer into the clean 96-DeepWell plate, which must be bar-coded and registered in the LIMS (Fig. 2c, ❼). (f) Discard any empty laboratory-ware. (g) Record completion of the analytical step and indicate any problems encountered during the process in the LIMS (see Note 4). 2. Proceed to SCX purification. (a) Start recording the analytical step in the LIMS (see Note 4). (b) Fill the stock bottles with SCX loading buffer and SCX elution buffer (see Note 18). (c) Place SCX cartridges on the holder of the vacuum manifold along with a clean, bar-coded 96-DeepWell plate (Fig. 2b, “RP & SCX SPE”). (d) Start the automated run. The robot equilibrates the cartridges with 4  0.93 mL of SCX loading buffer. The robot loads the samples (aspirating 2  0.95 mL from the 96-DeepWell plate containing the Oasis HLB-purified samples; Fig. 2c, ❼) and washes the samples with 4  0.9 mL SCX loading buffer. (e) Peptides are eluted with 2  0.8 mL RP elution buffer into the clean 96-DeepWell plate, which is bar-coded and has been registered in the LIMS (Fig. 2c, ❽). (f) Discard soiled laboratory-ware. (g) Lyophilize the samples in a speed-vac overnight (see Note 19). (h) After complete evaporation dissolve the samples in 1.6 mL SCX loading buffer, and lyophilize once again (see Note 19). (i) Cover the 96-DeepWell plate with a polypropylene mat and store the dried peptide samples at 80
C before RP-LC MS/MS analysis. (j) Record completion of the analytical step and indicate any problems encountered during the process in the LIMS (see Note 4). 3.8 RP-LC MS/MS Analysis

1. Start recording the analytical step in the LIMS (see Note 4). 2. Prepare the sample sequence. Wash (i.e., CH3CN) and blank (i.e., LC solvent A) samples should be inserted after every four CSF sample injections.

102

Antonio Nu´n˜ez Galindo et al.

Table 2 LC chromatographic gradient used for RP-LC MS/MS analysis Time (min)

LC solvent A (%)

LC solvent B (%)

Flow rate (μL/min)

0.00

93.7

6.3

0.300

1.00

93.7

6.3

0.300

12.00

89

11

0.300

87.00

82.5

17.5

0.300

129.00

74.5

25.5

0.300

157.00

60

40

0.300

157.10

2

98

0.300

165.00

2

98

0.300

165.10

93.7

6.3

0.300

180.00

93.7

6.3

0.300

3. Dissolve the prepared CSF samples in 200 μL of LC solvent A. Shake the 96-DeepWell plate for 5 min. Transfer 50 μL from each well into pre-labeled LC vials. Place the LC vials at 5
C in the LC autosampler (Fig. 2a). Store the remaining sample volumes at 80
C for further use. 4. Inject 5 μL of sample per analysis. 5. Run the RP-LC for 180 min using a gradient of LC solvent A and LC solvent B (as indicated in Table 2) at a flow rate of 300 nL/min. 6. Perform RP-LC MS/MS analysis (see Note 1). Using an Orbitrap Fusion Lumos Tribrid mass spectrometer, data should be acquired using a data-dependent method. Positive-ion spray voltage should be set to 1700 V, and transfer-tube temperature to 275
C. For MS survey scans in profile mode, the Orbitrap resolution is set to 120,000 at m/z ¼ 200 (automatic gain control (AGC) target of 2  105) with an m/z scan range of 300 to 1500, RF lens should be set to 30%, and maximum injection time to 100 ms. For MS/MS with higher-energy collisional dissociation (HCD) at 35% of the normalized collision energy, AGC target is set to 1  105 (isolation width of 0.7 in the quadrupole), with a resolution of 30,000 at m/z ¼ 200, first mass at m/z ¼ 100, and a maximum injection time of 105 ms with the Orbitrap acquiring in profile mode. A duty cycle duration of 3 s (top speed mode) should be used to determine the number of precursor ions to select for HCD-based MS/MS. Ions are injected for all available parallelizable time. Dynamic exclusion is set to 60 s within a  10 ppm window. A lock mass of m/z ¼ 445.1200 is used.

CSF Proteomics for Biomarker Discovery

103

7. Record completion of the analytical step and indicate any problems encountered during the process in the LIMS (see Note 4). 3.9

Data Analysis

1. Convert MS raw files to peak lists using Proteome Discoverer (see Note 20). 2. Perform database search using Mascot as the search engine. Set variable amino acid modifications to oxidized methionine, deamidated asparagine/glutamine, and 6-plex TMT-labeled peptide amino terminus (þ 229.163 Da). 6-plex TMT-labeled lysine (þ 229.163 Da) is set as a fixed modification, as is carbamidomethylation of cysteine. Trypsin is selected as the proteolytic enzyme, with a maximum of two potential missed cleavages. Peptide and fragment ion tolerance should be set to 10 ppm and 0.02 Da, respectively. 3. Automatically load Mascot result files (.dat files) into Scaffold Qþ via the LIMS for further searching with X! Tandem. Set both peptide and protein FDRs to 1%, with a requirement for two unique peptides to report protein identification. 4. Relative quantitative protein values are exported from Scaffold Qþ as Log2 of the protein fold-change ratios with respect to the reference TMT channel (see Notes 21 and 22), i.e., mean Log2 values after correction for isotopic purity but without normalization between samples and experiments (Fig. 3). 5. Some quality checks are applied to the data, for instance checking that all TMT experiments have a similar number of identified and quantified proteins and determining the standard deviation and errors in the values measured for the internal standard, LACB (see Note 2). 6. Preprocessing steps before statistical analysis may include the removal of samples for which over 70% of quantitative values are missing, or with a high level of proteins exhibiting extreme quantitative values considered as outliers. 7. Replicate values for each sample are averaged when two replicates are available. For more than two replicates, we recommend using the median value to reduce the impact of outliers (see Note 23). Otherwise, the only non-missing value between the replicates should be retained. 8. To identify outliers, determine data structure (such as by identifying clusters) and assess any potential effects specific to the analytical batch, collection center, gender, or age of patients; perform principal component analyses (PCAs) (using R package mixOmics) on the full dataset (Fig. 3) (see Note 24). 9. Normalize data if required, e.g., if for some proteins the measurement ranges vary widely from those of other proteins, or if

104

Antonio Nu´n˜ez Galindo et al.

Fig. 3 Schematic representation of initial data analysis. (a) PCA plot (second vs. first principal component) for four samples with three technical replicates each. Ellipses encompass all the replicates of the same sample. The most elongated ellipsoids are the yellow and orange; data points furthest from the other two replicates are identified as outliers. (b) Heatmap of the data. Samples have been clustered. Boxplot of the data for each batch or plate number (c) before and (d) after normalization. In this view, only one value is included per replicate, representing just one protein. Alternative views can be obtained after averaging replicates, and including all proteins

required by the model or data mining approach used. Different options available are, e.g., median normalization, quantilequantile normalization, min-max normalization, and standard deviation normalization (in Fig. 3c, d, data are shown before and after normalization). 10. Identify proteins that are present in increased or decreased amounts in one sample group compared to others (Fig. 3c, d) (see Notes 21–25). We recently applied this procedure to

CSF Proteomics for Biomarker Discovery

105

measure CSF proteomes for 120 older adults and explored their association with well-established markers of core AD pathology (see Note 26).

4

Notes 1. ASAP2 was initially developed using a hybrid linear ion trapOrbitrap (LTQ-OT) Elite (Thermo Scientific) mass spectrometer. It was recently adapted for use with an Orbitrap Fusion Lumos Tribrid mass spectrometer, with relatively unchanged RP-LC conditions. A significant increase in CSF proteome coverage was noted following this change. Other mass spectrometers can also be used such as tandem time-of-flight (TOF–TOF) and quadrupole (Q)-TOF allowing MS/MS and detection of reporter-ions in the low-mass range for isobaric labeling. MS acquisition parameters should be adapted accordingly. Once a week, the mass spectrometer should be calibrated, and its performance evaluated (e.g., by analyzing a complex protein digest sample as a reference) in line with standard operating procedures and well-defined quality checks. To verify the performance of the RP-LC MS/MS instrumentation on the complex protein digest sample and fine-tune it, we typically record the number of MS/MS scans, numbers of identified proteins and peptides, retention time for reference peptides, their elution peak widths and intensities, ion injection times, mass accuracies, and multiplier voltage values. 2. A fixed amount of an internal protein standard (e.g., LACB) is spiked into each sample at the start of the experiment; it can later be used for assessment and correct any bias occurring during experimental handling. We recommend including the LACB sequence in the most recent release of the human UniProtKB/Swiss-Prot database (or other database when studying CSF from another species). The internal protein standard should be chosen carefully: it should be absent from the samples studied, and should not generate tryptic peptides with sequences similar to peptides from the proteome studied, for instance. Correction factors can be determined and applied to correct for manipulation bias. Since the internal standard is spiked in equal amounts into individual samples, it also provides data to control the quality of the quantitative experiments (Fig. 3b). 3. Human CSF from healthy donors generally contains proteins at concentrations between 150 and 600 mg/L (U.S. National Library of Medicine; https://medlineplus.gov/), whereas the reference range for total protein concentrations in plasma is typically 60–80 g/L. To adapt such protein concentration

106

Antonio Nu´n˜ez Galindo et al.

differences to ASAP2, which was initially developed for plasma samples, we use more than 100 μL of each CSF sample. Consequently, CSF samples must be evaporated to be compatible with the workflow. We have established that an initial volume of 400 μL of human CSF provides good starting conditions to ensure qualitative and quantitative results using ASAP2 [9]. To avoid overdrying the CSF samples, it is highly recommended to optimize the drying time. In our hands, evaporation of 400 μL of CSF from a 1.5-mL tube using a standard speed-vac vacuum centrifugation system takes around 2 h. 4. To minimize the risk of errors in both the wet and dry laboratories in large-scale research studies, a customized LIMS software application was implemented for sample barcode tracking at each step of the workflow and to record each experimental procedure. This LIMS platform also supports both Mascot and Scaffold proteomic tools, and is integrated with data storage solutions, that are also used to maintain an efficient workflow. 5. To prevent or minimize the introduction of batch effects, the study and sample plate template should be carefully designed before starting. Depending on the size of the study, and on the number of CSF samples to be processed, it may be convenient to perform the drying step and the depletion step on two different days. 6. Considering the significant time required to deplete samples, we recommend processing groups of up to 24 samples per day. Samples should be filtered just before the depletion process to avoid evaporation and potential sample degradation. 7. It is important that the system be cleaned regularly to avoid any contamination or carry-over after multiple injections of CSF samples. In general, we include a blank sample (containing only Buffer A) after every four CSF samples during analysis to keep the lines and column clean, and to check for potential contamination. To do so, we record chromatograms using ultraviolet detection at 254 nm. 8. We have assessed the impact, in terms of proteome coverage, of using commercial MARS immuno-affinity LC columns to treat human CSF samples, and investigated the specificity of the depletion of 14 highly abundant human plasma proteins from human CSF samples. Typically, ~93% of the total CSF protein content is removed after depletion of the 14 specified proteins, confirming that despite the differences in protein composition in CSF and plasma, several proteins are present at high abundance in both body fluids. In addition, by comparing depleted and non-depleted CSF samples (Fig. 4a), we previously observed that proteome coverage overlapped by 63% (with 630 proteins identified) [9]. In the same study, depletion was

CSF Proteomics for Biomarker Discovery

107

Fig. 4 (a) Human CSF proteome coverage comparison including (i.e., depleted CSF) and excluding (i.e., non-depleted CSF, or total CSF) the immuno-affinity depletion step in ASAP2, as previously reported in [9]. (b) Gene ontology/biological processes enriched for the 185 genes corresponding to the proteins only identified in the depleted CSF, as previously reported [9]

shown to increase proteome coverage by 31%; using depletion, proteins relevant to cell-cell adhesion and homophilic cell adhesion (and in particular, cadherins and protocadherins, known to be involved in several neurological diseases) were detectable in CSF (Fig. 4b). 9. The quantitative performance of ASAP2 with and without the immuno-affinity step was also evaluated [9]. Inclusion of depletion results in less precise relative quantitation data (Fig. 5a), but the loss is acceptable considering the improved proteome coverage. Log2 of standard deviation values for the fold-change ratios obtained were 0.1856 and 0.1279 for depleted CSF and non-depleted CSF samples, respectively.

108

Antonio Nu´n˜ez Galindo et al.

Fig. 5 (a) Distribution of Log2 of the spectral fold-change ratios with and without immuno-affinity depletion step (depleted CSF is shown in blue; non-depleted CSF, or total CSF, in orange) as previously reported [9]. (b) Comparison of normalized distributions of Log2 of the spectral fold-change ratios for human CSF samples (in blue) and human plasma samples (in red) as previously reported [8, 9]

10. The depletion columns have proven to be stable and accurate for treatment of more than 360 samples, and we suggest replacing them after about 300 sample injections. 11. The depletion procedure is the rate-limiting step in the workflow, requiring four days to treat 96 samples using a single LC apparatus. In our laboratory, the depletion procedure allows a flexible number of samples to be processed (i.e., up to 48). We use two identical HPLC systems to allow depletion of up to 192 samples per week. Using this setup (Table 1), depletion requires ~38 min per sample. Buffer volumes should be verified before starting the depletion each day (600 mL Buffer A and 300 mL Buffer B will be needed for depletion of 24 samples). In addition, the waste bottle levels should be checked regularly to avoid overflows. 12. The buffer exchange process is currently performed manually (we usually perform buffer exchange for two 96-well plates on the same day, after completing depletion of 192 CSF samples over four days). It is important to apply the vacuum progressively and to verify that all the cartridges are free of liquid before proceeding to the next step. The plate used to recover waste liquids from the vacuum system should be emptied to avoid overflows. Elution is performed into a clean plate. 13. Longer 50–1250 μL tips (i.e., 103 mm length instead of 76 mm) are used for convenient introduction into the 1.4mL collection tubes (after the depletion process). Particular

CSF Proteomics for Biomarker Discovery

109

care should be taken to recover the whole sample and avoid the formation of bubbles. 14. It is recommended that TEAB, TCEP, IAA, and trypsin/Lys-C solutions be freshly prepared before each experiment. 2% SDS can be prepared only once a week. The reagent tubes (and reservoirs) should always contain an excess volume of at least 150–200 μL. 15. The liquid handling platform includes a shaker/heating unit (Fig. 2a, “Digestion Labeling” and Fig. 2b, ❷). The shaker speed is fixed and will be the same for all processes. The shaker/heating unit can be automatically switched on or off and is temperature-controlled. The dark chamber (Fig. 2a, “Reduction Alkylation” and Fig. 2b, ❸) is also temperaturecontrolled. 16. 6-plex TMT is convenient for use in the described procedure as 96 is a multiple of six. A 96-well plate can therefore be entirely filled with TMT experiments. When processing 96 samples on a plate, a custom order of TMT reagents is placed to obtain amounts directly compatible with labeling of 96 samples (i.e., 13 mg of each TMT reagent to be diluted in 700 μL of CH3CN). A smaller number of samples could be also processed with the liquid handling system using the ASAP2 protocol. Because TMT reagents are moisture-sensitive, reagent stocks should be equilibrated to reach room temperature before opening to avoid moisture condensation inside the vials. 17. For the TMT labeling process, reagents should be prepared as quickly and carefully as possible to prevent evaporation of the CH3CN used to dilute the TMT reagents. 18. We recommend preparing fresh stocks of solvent every three weeks. 19. To ensure optimal conditions for sample evaporation and to anticipate device mal-functioning, the speed-vac instrument and its settings should be checked and set up in advance. 20. Alternatives to Proteome Discoverer to convert raw MS files into peak lists exist. The msconvert freeware which is included in the ProteoWizard software is one solution, although it does not support vendor conversion for Linux-based operating systems. We usually select mzML data as output format and generate it using the following parameters: 32-bit precision, no zlib conversion (for compatibility with X! Tandem software), HCD activation, MS levels 1 & 2, and zero sample filter levels 1 & 2. 21. We previously validated the use of ASAP2 with human CSF samples. Using ASAP2, 96 identical CSF pool samples were prepared and the distribution of the fold-change ratios

110

Antonio Nu´n˜ez Galindo et al.

obtained was determined using TMT technology. The analytical figures-of-merit for human CSF samples appeared very similar to those obtained with human plasma samples [9] (Fig. 5b). We also showed that ASAP2 can quantify proteins at different concentrations in a calibration curve experiment using a two-proteome model where an E. coli total protein extract was spiked at different concentrations into 400 μL human CSF samples [9]. 22. To link TMT experiments between each other in a large study, a biological reference sample should be used (here, a pool of CSF samples) in each of the experiments. This biological reference serves to calculate relative protein differences across all the samples analyzed. 23. If technical replicates are available, it can be useful to identify potential outliers at replicate level. Extreme differences between replicates can be observed by plotting PCAs (Fig. 3a), where technical replicates are expected to be projected very near to each other. If only two technical replicates are available, and their data points are extensively separated in the PCA plot, it would be advisable to keep just one of the replicates, for instance, the one with the best metric regarding internal standard variability or total number of quantified proteins. If more than two technical replicates are available, then the replicate(s) that differ(s) the most from the majority can be removed (see Fig. 3, where two replicate outliers are pinpointed among the yellow and orange samples, and identified by a black star). 24. We also recommend studying principal components with lower variances than the first two or three commonly used principal components to obtain a better picture of the data and its potential effects. The data can be further visually inspected using boxplots representing proteins by analytical batch (such as the graphical representation shown in Fig. 3c), collection center, gender, and age. Plotting heatmaps of the values can also highlight samples with consistently high or low values for a large proportion of proteins (Fig. 3b). These samples should be cross-checked with the laboratory records collected in the LIMS to see if they have been flagged as problematic (e.g., blood contamination, low volume, or other). In such cases, they can be excluded from further statistical analyses. 25. Supervised machine learning techniques are well suited for biomarker identification in large clinical proteomic studies when classes are known a priori (e.g., disease vs. control, disease progression status, response to treatment/intervention). Examples of commonly used supervised machine learning tools include support vector machines (SVM), decision trees,

CSF Proteomics for Biomarker Discovery

111

random forests, rule-based classifiers, naive Bayes, and many others. While traditional tests such as t-tests are often not recommended for large datasets with few replicates, in particular for datasets with a large number of variables, they can still be useful as preliminary screening tools; however, it is essential to adjust for multiplicity testing to increase confidence in the results. Additional adjustments should be taken into account, such as confounding factors like age, gender, and any other known or suspected effects. In the field of AD, for instance, additional variables such as years of education, cognitive impairment, presence of APOE ε4 allele, levels of CSF Aβ1–42, CSF tau, and CSF phosphorylated tau have often been considered. Validation of candidate biomarkers in an independent cohort is essential to add confidence to the generalization of the protein candidates. Once the potential biomarkers are identified and validated, additional studies will be required to assess their utility and practicability as tools in the clinic. 26. The workflow described in this chapter has already been used in clinical research. In particular, we recently reported an analysis of CSF samples from 120 older community-dwelling adults with normal (n ¼ 48) or impaired (n ¼ 72) cognition [14]. A total of 790 proteins were quantified by MS in CSF samples using an LTQ-OT Elite, and associations between CSF proteins and CSF Aβ1–42, CSF tau, and CSF phosphorylated tau were made. Known and new CSF proteins related to amyloid pathology, neuronal injury and tau hyperphosphorylation were identified. We confirmed several previous results linking CSF proteins to AD, and revealed additional CSF proteome alterations involving reelin-producing cells and the myelin sheath. References 1. Prince M, Bryce R, Albanese E et al (2013) The global prevalence of dementia: a systematic review and metaanalysis. Alzheimers Dement 9(1):63–75. https://doi.org/10.1016/j.jalz. 2012.11.007 2. Musiek ES, Holtzman DM (2015) Three dimensions of the amyloid hypothesis: time, space and ‘wingmen’. Nat Neurosci 18 (6):800–806. https://doi.org/10.1038/nn. 4018 3. Lista S, Faltraco F, Prvulovic D et al (2013) Blood and plasma-based proteomic biomarker research in Alzheimer’s disease. Prog Neurobiol 101–102:1–17. https://doi.org/10. 1016/j.pneurobio.2012.06.007 4. Agrawal M, Biswas A (2015) Molecular diagnostics of neurodegenerative disorders. Front

Mol Biosci 2:54. https://doi.org/10.3389/ fmolb.2015.00054 5. Henriksen K, O’Bryant SE, Hampel H et al (2014) The future of blood-based biomarkers for Alzheimer’s disease. Alzheimers Dement 10 (1):115–131. https://doi.org/10.1016/j.jalz. 2013.01.013 6. Begcevic I, Brinc D, Drabovich AP et al (2016) Identification of brain-enriched proteins in the cerebrospinal fluid proteome by LC-MS/MS profiling and mining of the human protein atlas. Clin Proteomics 13:11. https://doi. org/10.1186/s12014-016-9111-3 7. Fang Q, Strand A, Law W et al (2009) Brainspecific proteins decline in the cerebrospinal fluid of humans with Huntington disease.

112

Antonio Nu´n˜ez Galindo et al.

Mol Cell Proteomics 8(3):451–466. https:// doi.org/10.1074/mcp.M800231-MCP200 ˜ ez Galindo A, Corthe´sy J et al 8. Dayon L, Nu´n (2014) Comprehensive and scalable highly automated MS-based proteomic workflow for clinical biomarker discovery in human plasma. J Proteome Res 13(8):3837–3845. https://doi. org/10.1021/pr500635f ˜ ez Galindo A, Kussmann M, Dayon L 9. Nu´n (2015) Proteomics of cerebrospinal fluid: throughput and robustness using a scalable automated analysis pipeline for biomarker discovery. Anal Chem 87(21):10755–10761. https://doi.org/10.1021/acs.analchem. 5b02748 ˜ ez Galindo A, Cominetti O et al 10. Dayon L, Nu´n (2017) A highly automated shotgun proteomic workflow: clinical scale and robustness for biomarker discovery in blood. In: Greening D, Simpson R (eds) Serum/plasma proteomics. Methods in molecular biology, vol 1619. Humana Press, New York, NY, pp 433–449. https://doi.org/10.1007/978-1-4939-70575_30

˜ ez Galindo A, Corthe´sy J 11. Cominetti O, Nu´n et al (2016) Proteomic biomarker discovery in 1000 human plasma samples with mass spectrometry. J Proteome Res 15(2):389–399. https://doi.org/10.1021/acs.jproteome. 5b00901 ˜ ez Galindo A et al 12. Dayon L, Wojcik J, Nu´n (2017) Plasma proteomic profiles of cerebrospinal fluid-defined Alzheimer’s disease pathology in older adults. J Alzheimers Dis 60 (4):1641–1652. https://doi.org/10.3233/ JAD-170426 ˜ ez Galindo 13. Oller Moreno S, Cominetti O, Nu´n A et al (2018) The differential plasma proteome of obese and overweight individuals undergoing a nutritional weight loss and maintenance intervention. Proteomics Clin Appl 12 (1):1600150. https://doi.org/10.1002/prca. 201600150 ˜ ez Galindo A, Wojcik J et al 14. Dayon L, Nu´n (2018) Alzheimer disease pathology and the cerebrospinal fluid proteome. Alzheimers Res Ther 10(1):66. https://doi.org/10.1186/ s13195-018-0397-4

Chapter 7 Lys-C/Trypsin Tandem-Digestion Protocol for Gel-Free Proteomic Analysis of Colon Biopsies Armin Schniers, Yvonne Pasing, and Terkel Hansen Abstract The protocol presented was specifically optimized for in-depth analysis of the human colon mucosa proteome. After cell lysis in a sodium deoxycholate/urea buffer, a tandem digestion with Lys-C and trypsin was performed. Prior to LC-MS/MS analysis, peptides were TMT-labeled and fractionated by high pH reversed-phase spin columns. This protocol is a powerful, reproducible, sample-saving, and cost-effective option when an in-depth quantitative proteome analysis is desired. Key words Colon mucosa proteome, Sodium deoxycholate, Lys-C, TMT labeling, High pH reversed-phase fractionation

1

Introduction The colon mucosa has a crucial physiological function in the absorption of salt and water [1]. It interacts with the gut flora and is a barrier against pathogens and parasites [2, 3]. However, the colon mucosa is also the site of disorders such as inflammatory bowel diseases (IBD). Both the host-symbiont-pathogen interactions [4] and the IBD pathophysiology [5] are complex and not fully understood. Shotgun proteomics, which provides quantitative data on a large number of proteins and furthermore allows for pathway and functional analyses, is a promising way to untangle these complex processes. A tandem digestion approach consisting of a predigestion with Lys-C in high urea concentrations followed by dilution and a tryptic digest in 1 M urea showed a better digestion efficiency than a simple tryptic digest for colon mucosa biopsies [6]. Besides chaotropes like urea, detergents are often used to improve protein solubilization, and sodium deoxycholate (SDC) has been found to perform best for the samples in question here as well as other difficult matrixes [7–9]. In a previous study we could show that

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_7, © Springer Science+Business Media, LLC, part of Springer Nature 2019

113

114

Armin Schniers et al.

the additional use of 5% SDC in the urea lysis buffer for the Lys-C/ trypsin tandem digestion gives highly reproducible results and brings a major improvement in peptide and protein identification numbers as compared to urea alone [9]. Like other human tissues, the colon mucosa has a wide dynamic range of protein abundances. Prefractionation techniques have been developed to reduce the sample complexity, allowing for a deeper analysis of the proteome. Prefractionation on reversed phase at high pH has shown good orthogonality and a more evenly distribution of peptides over the fractions than strong cation exchange prefractionation (SCX), as well as superiority over strong anion exchange (SAX) with regard to peptide identification numbers [10, 11]. Colon mucosa biopsies are typically small (10–15 mg wet weight), limiting the amount of protein available in some cases down to 200 μg. In order to prevent biases which could result from the prefractionation process in label-free quantifications, Tandem Mass Tag (TMT) or other isotopic labeling can be utilized. Isobaric labeling also reduces the amount of protein required from each biopsy or biological condition investigated. This is of interest, because the amount of protein needed for one experiment involving a prefractionation into eight fractions with high pH reversedphase columns can be up to 100 μg. Computational analysis returned 8703 protein identifications, which is 29% more than the previously largest dataset of the colon mucosa proteome [12]. The protocol described here is not limited to colon mucosa biopsies or biopsies in general. Given the disruptive strength of the lysis buffer and the predigestion in high urea and detergent concentration, we encourage its use especially for samples containing proteins which are hard to solubilize and digest, but also for any other samples when a deep proteome coverage is desired.

2

Material Use only ultrapure water and analytical grade reagents to prepare solutions.

2.1 Biopsy Handling and Storage

1. Cryogenic sample vials.

2.2

1. Lysis buffer: 5% Sodium deoxycholate (SDC) and 8 M urea in 100 mM triethylammonium bicarbonate buffer (TEAB) (see Note 1).

Cell Lysis

2.
70  C freezer.

2. MagNA Lyser Instrument and MagNA Lyser Green Beads (Roche Diagnostics AG, Rotkreuz, Switzerland).

Proteomics of Colon Biopsies

2.3 Lys-C/Trypsin Tandem Digest

115

1. 1.5 mL Protein LoBind microtubes (Eppendorf AG, Hamburg, Germany). 2. Pierce™ bicinchoninic acid (BCA) Protein Assay Kit (Thermo Fisher Scientific) or comparable assay. 3. 0.2 M solution dithiothreitol (DTT), 0.5 M solution iodoacetamide (IAA), and 50 mM solution CaCl2 (see Note 2). 4. 1 M triethylammonium bicarbonate buffer (TEAB). 5. Lys-C. 6. Trypsin. 7. Trifluoroacetic acid (TFA).

2.4

TMT Labeling

1. TMTsixplex™ Isobaric Label Reagent Set (Thermo Scientific). 2. 10.5% hydroxylamine hydrochloride solution.

2.5

Fractionation

1. Pierce™ High pH Reversed-Phase Peptide Fractionation Kit. 2. Low Protein Binding Collection Tubes 2.0 mL (Thermo Scientific).

2.6

LC-MS/MS

Nano-liquid chromatography system coupled to a high-resolution tandem mass spectrometer, e.g., EASY-nLC 1000 coupled to a Q Exactive (both Thermo Fisher Scientific, Bremen, Germany), consisting of: 1. Precolumn: Acclaim PepMap® 100, C18, particle size 3 μm, pore size 100 A˚, inner diameter 75 μm, length 2 cm, nanoViper, Thermo Fisher Scientific. 2. Separation column: EASY-Spray, PepMAP® RSLC, C18, particle size 2 μm, pore size 100 A˚, inner diameter 75 μm, length 50 cm, Thermo Fisher Scientific. 3. Solvents for separation: solvent A: water with 0.1% formic acid (FA), solvent B: acetonitrile with 0.1% FA. Both A and B should be purchased as ready to use solvents from the vendor of the mass spectrometer.

2.7

Data Analysis

1. MaxQuant software package for identification and quantitation of peptides and proteins [13]. 2. Perseus software package for visualization of data [14].

3

Methods

3.1 Biopsy Handling and Storage

Store biopsies at
70  C until further sample preparation.

116

3.2

Armin Schniers et al.

Cell Lysis

1. Prepare the lysis buffer fresh before use. 2. Fill 250 μL of cooled lysis buffer into each MagNA Lyser Green Bead tube and place the biopsies between the beads in the tube. Keep the tube on ice until cell lysis is completed. 3. Perform lysis in the MagNA Lyser Instrument for 35 s at 6500 rpm (see Note 3). 4. Freeze the lysates at
70  C until further sample preparation or continue immediately.

3.3

Tryptic Digest

Prepare the necessary solutions. Prepare IAA and enzyme solutions fresh and keep them in the fridge until use (see Note 4). Prepare fresh CaCl2 and DTT solutions or thaw solutions stored at
70  C immediately before use. 1. Perform a BCA assay: Dilute 5 μL of cell lysates with 95 μL water to a final concentration of 5 mM TEAB. Prepare further dilutions with 5 mM TEAB in water. Prepare bovine serum albumin (BSA) standard dilutions with final TEAB concentrations of 5 mM for the standard curve. Perform a BCA assay (such as the Pierce™ BCA Protein Assay Kit) according to the manufacturer’s protocol in order to estimate the protein concentration in the lysate (see Note 5). 2. Transfer lysate aliquots of 60 μg protein to Protein LoBind microtubes. 3. Perform reduction of disulfide bridges: Add DTT solution to a final concentration of 5 mM and incubate at 54  C for 30 min. 4. Perform alkylation: Add IAA solution to a final concentration of 15 mM and incubate for 30 min in the dark. 5. Add DTT solution corresponding to a final concentration of 5 mM to remove excess IAA. 6. Add CaCl2 solution, 0.6 μg Lys-C (enzyme-to-protein ratio 1:100, w/w), water, and 1 M TEAB to a final concentration of 1 mM calcium chloride, 6 M urea, and 100 mM TEAB. 7. Perform protein predigest under gentle agitation for 8 h at 37  C. 8. Add CaCl2 solution, 3 μg trypsin (enzyme-to-protein ratio 1:20, w/w), water, and 1 M TEAB to a final concentration of 1 mM calcium chloride, 1 M urea, and 100 mM TEAB. 9. Perform tryptic protein digest under gentle agitation for 16 h at 37  C.

3.4

TMT Labeling

1. Transfer aliquots corresponding to 25 μg peptides to new Protein LoBind microtubes. 2. Perform TMT labeling (see Note 6) according to the manufacturer’s protocol. Use 0.4 mg TMT reagent for 25 μg peptides

Proteomics of Colon Biopsies

117

from samples with starting protein concentrations 2200 μg/ mL. Increase the amount of TMT reagent for lower starting concentrations (see Note 7). 3. Mix in total 100 μg of the differently labeled peptides in equal amounts in a Protein LoBind microtube (see Notes 8 and 9). 3.5 Removal of SDC and Acetonitrile

1. Acidify protein digests with 50% FA to a final concentration of 2.5% FA and pH  2. Check the pH and add more FA if necessary to obtain pH  2 (see Note 10). 2. Centrifuge at 16,000  g for 15 min. 3. Transfer the supernatants carefully to new Protein LoBind microtubes. 4. Remove acetonitrile by evaporation in a vacuum concentrator (see Notes 11 and 12).

3.6

Fractionation

1. Add TFA to a concentration of 0.1%. 2. Fractionate the sample into eight fractions, e.g., with Pierce™ High pH Reversed-Phase Peptide Fractionation Kit, according to the manufacturer’s protocol. Load samples exceeding 300 μL stepwise by repeating the following: Add 300 μL to the column, centrifuge for 2 min at 3000  g, discard flowthrough. 3. Evaporate the fractions until dryness in a vacuum concentrator and redissolve them in 10 μL 0.1% TFA for subsequent LC-MS/MS analyses.

3.7

LC-MS/MS

1. Inject 2.0 μg peptides per sample into the nano-LC-MS/MS system, e.g., an EASY-nLC 1000 coupled to a Q Exactive mass spectrometer (see Note 13). 2. Concentrate the peptides on a reversed-phase trap column with 0.1% formic acid (FA) at a flow rate of 20 μL/min. Separate the peptides on a reversed-phase main column. Use a binary solvent gradient at 60  C column temperature and 200 nL/min flow rate. Increase the acetonitrile (ACN) proportion from 2% to 5% over 19 min, further to 30% at 180 min and to 100% at 200 min. Regenerate the column with 100% ACN for additional 10 min (see Note 14). 3. Run the Q Exactive mass spectrometer in positive mode, with the following global settings: Chromatographic peak width 15 s, default charge state 2, full MS survey scans from 400 to 2000 m/z, resolution 70,000, AGC target value 3e6, maximum injection time 100 ms for MS scans. Subject the 10 most intense peaks to MS/MS with the following settings: resolution 17,500, dynamic exclusion 10 s, underfill ratio 1%, charge states þ2, þ3, and þ4, exclude isotopes, normalized collision energy 31, isolation window 2 m/z, AGC target value 1e5 and maximum injection time 50 ms, fixed first mass 120.

118

3.8

Armin Schniers et al.

Data Evaluation

Process the raw data with a quantitative proteomics software package, e.g., MaxQuant [13]. Download the newest version of the respective fasta file, e.g., from https://www.uniprot.org/, and configure it in the MaxQuant sequence database configurations. Change the TMT reporter ion distribution in the MaxQuant PTM configuration according to the manufacturer’s Certificate of Analysis and restart MaxQuant (see Note 15). Set reporter ion MS2/TMTsixplex as quantification method. Load the raw files and specify their experiment name and fraction numbers. Search against the configured fasta file with the following parameters: Enzyme: Trypsin/P (specific), fixed modifications: Carbamidomethyl (C), variable modifications: Oxidation (M) and Acetyl (Protein N-term) (see Note 16), max. 2 missed cleavages, PSM FDR: 0.01, Protein FDR: 0.01 Site FDR: 0.01, Use Normalized Ratios For Occupancy: TRUE, Min. peptide Length: 7, Min. score for unmodified peptides: 0, Min. score for modified peptides: 40, Min. delta score for unmodified peptides: 0, Min. delta score for modified peptides: 6, Min. unique peptides: 0, Min. razor peptides: 1, Min. peptides: 1, Use only unmodified peptides and: TRUE, Modifications included in protein quantification: Oxidation (M) and Acetyl (Protein N-term), Peptides used for protein quantification: Razor, Discard unmodified counterpart peptides: TRUE, Label min. ratio count: 2, Use delta score: FALSE, iBAQ: FALSE, iBAQ log fit: FALSE, Match between runs: TRUE, Matching time window [min]: 0.7, Alignment time window [min]: 20, Find dependent peptides: FALSE, Decoy mode: revert, Include contaminants: TRUE, Advanced ratios: TRUE, Second peptides: FALSE, Calculate peak properties: FALSE, Main search max. combinations: 200, Advanced site intensities: TRUE, Max. peptide mass [Da]: 4600, Min. peptide length for unspecific search: 8, Max. peptide length for unspecific search: 25, Razor protein FDR: TRUE, Disable MD5: FALSE, Max mods in site table: 3, Match unidentified features: FALSE, MS/MS tol. (FTMS): 20 ppm, Top MS/MS peaks per Da interval. (FTMS): 12, Da interval. (FTMS): 100, MS/MS deisotoping (FTMS): TRUE, MS/MS deisotoping tolerance (FTMS): 7, MS/MS deisotoping tolerance unit (FTMS): ppm, MS/MS higher charges (FTMS): TRUE, MS/MS water loss (FTMS): TRUE, MS/MS ammonia loss (FTMS): TRUE, MS/MS dependent losses (FTMS): TRUE, MS/MS recalibration (FTMS): FALSE. Load the quantification data into Perseus. [14]. Delete identifications labeled as only identified by site, reverse hits and potential contaminants. Perform a log2 transformation. Normalize the intensities. If a standard channel was used, this can be done by subtracting its reporter ion intensities from the other channels of the same experiment and subsequent Z-score transformation (matrix access: columns). Visualize the results, e.g., as described in [15].

Proteomics of Colon Biopsies

119

The anticipated result from a single complete set of eight fractions is a dataset of approximately 7250 proteins, quantified from 59,000 peptides.

4

Notes 1. To constitute 1 mL of lysis buffer, add 100 μL 1 M TEAB buffer pH 8.5–480 mg urea and add water to dissolve the urea and fill the solution up to 1 mL. Add 1 mL of this solution to 50 mg sodium deoxycholate and dissolve the detergent by vortexing. Primary or secondary amines cannot be used in buffers or reagents, because they would react with the TMT reagents, resulting in poor labeling efficiency. 2. Make a 200 mM DTT solution by dissolving 30.85 mg DTT in 1 mL water, aliquots can be stored at
70  C for later use. Produce 500 mM IAA solution by dissolving 92.48 mg in 1 mL water. Produce fresh IAA solution right before use, it is unstable, hence it should be kept cold and protected from light. Produce 50 mM CaCl2 by dissolving 55.49 mg CaCl2 (anhydrous), 73.51 mg CaCl2·2H2O (dihydrate), or 109.54 mg CaCl2·6H2O (hexahydrate) in 10 mL water. Aliquots of 50 mM CaCl2 solution can be stored at
70  C for later use. 3. A bead mill can be an alternative, or other preferred methods of cell lysis, if no MagNA Lyser instrument is available. This should however be evaluated. 4. For optimal activity the enzyme solutions should be prepared fresh to avoid freeze-thaw cycles. 5. TEAB gives a color reaction with the BCA reagents. Hence, the TEAB concentration must be equal for all samples and standards. We recommend a target concentration of 5 mM TEAB. 6. Different TMT reagents are available, differing in the number of reporter ion channels. TMTzero is designed for method development purposes. TMTduplex, TMTsixplex, TMT10plex, and TMT11plex give two, six, ten, and eleven reporter ion channels, respectively. 10plex and 11plex demand a higher mass resolution, because they depend on the differentiation between the mass shifts from 12C to 13C and 14N to 15 N. The TMT variant should be chosen based on the performance of the available instruments and the experimental design. It should also be considered that the higher plexes can further reduce the needed sample amount as well as the running time on the LC-MS/MS, thereby reducing the total experiment costs. 7. Decreased labeling efficiencies can be observed when the amount of used TMT reagent is not corrected for increasing

120

Armin Schniers et al.

Table 1 Adjusting the amount of TMT reagents based on sample volume Sample concentration as determined by BCA assay (μg/mL)

TMT reagent for 25 μg peptides (mg)

2200

0.4

1530–2200

0.5

1173–1530

0.6

952–1173

0.7

800–952

0.8

sample volumes. We adjust the TMT amount as depicted in Table 1. 8. Before combining 100 μg for the prefractionation, we recommend to control for sample quality. To do this, combine a small amount (2 μg per sample, in total 12 μg) and desalt it after Subheading 3.5, step 4, for instance with Bond Elut OMIX pipette tips (Agilent). Analyze 0.5 μg peptides with a short gradient on the LC-MS/MS system. Control quality parameters such as the absence of contaminations, and the digestion and labeling efficiencies. The acquired data should also be used to control for over- and under-representations of single reporter ion channels. Fine-tune the composition where applicable by adjusting the volumes accordingly. 9. One TMT isotope may be dedicated to serving as a standard channel, which allows for comparisons across different LC-MS/MS experiment. To produce this standard, a mixture of peptides from several samples is labeled with one TMT isotope. This standard is added in an equal amount when the labeled peptides from the samples of interest are mixed. This is a good option when the number of samples to be compared exceeds the number of available TMT isotopes. 10. SDC precipitates under acidic conditions and forms a pellet under subsequent centrifugation. 11. Acetonitrile added with the TMT reagents needs to be removed completely to ensure complete peptide binding when loading the fractionation columns. 12. Samples or aliquots of samples can be desalted for LC-MS/MS analysis after this step, e.g., with Bond Elut OMIX pipette tips (Agilent), either to control the sample composition and quality or if no prefractionation is desired. 13. Determine the peptide concentration, e.g. with a Nanodrop 2000 (Thermo Fisher Scientific, Bremen, Germany) at a

Proteomics of Colon Biopsies

121

wavelength of 205 nm and extinction coefficient 31 mg/mL [16]. Dilute the samples with 0.1% TFA for injection into the LC-MS/MS system according to the measurements. 14. Run blanks between the samples to minimize memory effects. Inject standard samples of a known composition after approximately every tenth sample to control intensity, peak shape, and retention time. 15. Changes in the configuration only become active after a restart of MaxQuant. 16. Choose variable modifications depending on knowledge about the proteins of interest and your research focus. The search duration increases with every added variable modification, we don’t recommend to include more than three variable modifications in one search.

Acknowledgments We thank Ilona Urbarova and Jack-Ansgar Bruun for fruitful discussions, as well as Prof Jon Florholmen and Rasmus Goll for supplying colon biopsies. Our work was supported by a grant from the North Norway Regional Health Authorities. References 1. Sandle G (1998) Salt and water absorption in the human colon: a modern appraisal. Gut 43 (2):294–299. https://doi.org/10.1136/gut. 43.2.294 2. Shi N, Li N, Duan X et al (2017) Interaction between the gut microbiome and mucosal immune system. Mil Med Res 4:14. https:// doi.org/10.1186/s40779-017-0122-9 3. Kasper LH, Buzoni-Gatel D (2001) Ups and downs of mucosal cellular immunity against protozoan parasites. Infect Immun 69(1):1–8. https://doi.org/10.1128/IAI.69.1.1-8.2001 4. Curtis MM, Sperandio V (2011) A complex relationship: the interaction among symbiotic microbes, invading pathogens, and their mammalian host. Mucosal Immunol 4(2):133–138. https://doi.org/10.1038/mi.2010.89 5. Baumgart DC, Carding SR (2007) Inflammatory bowel disease: cause and immunobiology. Lancet 369(9573):1627–1640. https://doi. org/10.1016/S0140-6736(07)60750-8 6. Glatter T, Ludwig C, Ahrne´ E et al (2012) Large-scale quantitative assessment of different in-solution protein digestion protocols reveals superior cleavage efficiency of tandem Lys-C/ Trypsin proteolysis over trypsin digestion. J

Proteome Res 11(11):5145–5156. https:// doi.org/10.1021/pr300273g 7. Pasing Y, Colnoe S, Hansen T (2017) Proteomics of hydrophobic samples: Fast, robust and low-cost workflows for clinical approaches. Proteomics 17(6). https://doi.org/10.1002/ pmic.201500462 8. Pasing Y, Schniers A, Hansen T (2018) Straightforward protocol for gel-free proteomic analysis of adipose tissue. Methods Mol Biol 1788:289–296. https://doi.org/10. 1007/7651_2017_82 9. Schniers A, Anderssen E, Fenton CG et al (2017) The proteome of ulcerative colitis in colon biopsies from adults - optimized sample preparation and comparison with healthy controls. Proteomics Clin Appl 11(11–12). https://doi.org/10.1002/prca.201700053 10. Nakamura T, Kuromitsu J, Oda Y (2008) Evaluation of comprehensive multidimensional separations using reversed-phase, reversedphase liquid chromatography/mass spectrometry for shotgun proteomics. J Proteome Res 7 (3):1007–1011. https://doi.org/10.1021/ pr7005878

122

Armin Schniers et al.

11. Gilar M, Olivova P, Chakraborty AB et al (2009) Comparison of 1-D and 2-D LC MS/MS methods for proteomic analysis of human serum. Electrophoresis 30 (7):1157–1167. https://doi.org/10.1002/ elps.200800630 12. Bennike TB, Carlsen TG, Ellingsen T (2017) Proteomics dataset: the colon mucosa from inflammatory bowel disease patients, gastrointestinal asymptomic rheumatoid arthritis patients, and controls. Data Brief 15:511–516. https://doi.org/10.1016/j.dib. 2017.09.059 13. Cox J, Mann M (2008) MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteomewide protein quantification. Nat Biotechnol 26(12):1367–1372. https://doi.org/10. 1038/nbt.1511

14. Tyanova S, Temu T, Sinitcyn P et al (2016) The Perseus computational platform for comprehensive analysis of (prote)omics data. Nat Methods 13(9):731–740 15. Tyanova S, Cox J (2018) Perseus: a bioinformatics platform for integrative analysis of proteomics data in cancer research. In: von Stechow L. (eds) Cancer systems biology. Methods in molecular biology, vol 1711. Humana Press, New York, NY, p 133–148. doi:https://doi.org/10.1007/978-1-49397493-1_7 16. Scopes RK (1974) Measurement of protein by spectrophotometry at 205 nm. Anal Biochem 59(1):277–282. https://doi.org/10.1016/ 0003-2697(74)90034-7

Chapter 8 Tube-Gel: A Fast and Effective Sample Preparation Method for High-Throughput Quantitative Proteomics Leslie Muller, Luc Fornecker, Sarah Cianferani, and Christine Carapito Abstract Sample preparation is a key step in proteomics workflows. Tube-gel (TG) is a fast and repeatable sample preparation method that consists in the instantaneous trapping of the sample in a polyacrylamide gel matrix. It takes advantage of in-gel sample preparations by allowing the use of high concentrations of sodiumdodecyl sulfate but avoids the time-consuming step of electrophoresis. Therefore, TG limits the sample handling and is thus particularly suitable for high-throughput quantitative proteomics when large sample numbers have to be processed, as it is often the case in biomarker research and clinical proteomics projects. Key words Tube-gel, Sample preparation, High-throughput, Detergent-compatibility, Quantitative proteomics

1

Introduction Tube-gel (TG) principle relies on the polymerization of polyacrylamide gel directly in the sample in a Laemmli-like solution. It was first introduced in 2005 by Lu X. et al. and consequently applied and adapted by others, in particular for the analysis of membrane proteins, lipid rafts proteins, and the charge derivatization of peptides [1–8]. We have recently demonstrated its compatibility and repeatability for label-free quantitative proteomics by comparing it with a stacking gel and a standard urea-based liquid digestion protocol using well-calibrated samples [9]. TG appeared to be comparable to classical in-gel sample procedure and overperformed liquid digestion. Major advantages of TG are the considerable gain of time and the limited sample handling it allows, thus reducing the introduction of biases and the risk of sample loss. Our method described in this chapter can be easily applied to the

Leslie Muller and Luc Fornecker contributed equally to this work. Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_8, © Springer Science+Business Media, LLC, part of Springer Nature 2019

123

124

Leslie Muller et al.

quantitative analysis of large sample cohorts during biomarker discovery and verification phases.

2

Materials All solutions are prepared with ultrapure water.

2.1 Protein Extraction

1. 0.5 M Tris–HCl, pH 6.8: weigh 6 g of Tris in a glass bottle. Add 100 mL of water. Mix and adjust the pH with HCl. Store at 4
C. 2. 10% SDS solution in water. Store at room temperature. 3. Glycerol: store at room temperature. 4. Laemmli-like buffer: 62.5 mM Tris–HCl pH 6.8, 10% glycerol (w/w), 2% SDS (see Note 1). Leave one aliquot at 4
C for current use and store the remaining aliquots at 20
C (see Note 2).

2.2 Tube-Gel Preparation

1. Ammonium persulfate (APS): weigh 20 mg of ammonium persulfate and add 200 μL of water (see Note 3). 2. N, N, N, N0 -Tetramethylethylenediamine (TEMED): Store at room temperature. 3. 30% acryl/bis-acrylamide: Store at 4
C. 4. 0.5 mL microtubes (Eppendorf). 5. Material for cutting the tube-gel: scalpel and a punch cutting 2 mm sections. 6. Fixation solution: 45% (v/v) methanol, 5% acetic acid (see Note 4). Store at room temperature.

2.3 Pre-digestion Procedure

1. 25 mM ammonium bicarbonate (NH4HCO3): weigh 99 mg of ammonium bicarbonate and transfer to a plastic container. Add 50 mL of water and mix (see Note 3). 2. 10 mM dithiothreitol (DTT): weigh 15.4 mg of dithiothreitol and transfer to a plastic container. Add 10 mL of 25 mM NH4HCO3 and mix (see Note 5). 3. 55 mM iodoacetamide (IAA): weigh 102 mg of iodoacetamide and transfer to a plastic container. Add 10 mL of 25 mM NH4HCO3 and mix (see Note 5). 4. LC-grade acetonitrile (ACN). 5. 2 mL microtubes (Eppendorf).

Tube-gel Sample Preparation for High-Throughput Proteomics

3

125

Methods Carry out all procedures at room temperature unless otherwise specified.

3.1 Protein Extraction

3.2 Tube-Gel Preparation

For simple procedure, extract proteins directly in the Laemmli-like buffer (see Note 6). Extract recommended amounts of proteins between 10 and 100 μg in 30 μL of Laemmli-like buffer (final concentration 0.3–3 μg/μL). 1. Mix the 30 μL of extracted proteins in Laemmli-like buffer with 42.25 μL of ultrapure water, 25 μL of 30% acryl/bisacrylamide and 0.25 μL of TEMED in a 0.5 mL microtube. Gently vortex and centrifuge the mixture until all bubbles have disappeared (approximatively 2000  g during 10 s). Add 2.5 μL of 10% APS without introducing air bubbles and immediately briefly vortex and centrifuge the tube to remove residual bubbles (see Note 7). 2. After 1 h, the TG has to be fixed with 200 μL fixation solution for at least 1 h. 3. The TG is cut in 2 mm high sections and each section in ~2 mm2 pieces. Transfer the gel pieces into a 2 mL microtube.

3.3 Pre-digestion Procedure

1. Add 400 μL of 75% ACN and 25% 25 mM NH4HCO3 to the gel pieces in order to wash them. Briefly agitate the tube and wait approximatively 2 min before removing this solution. Repeat this step four times. 2. Dehydrate the gel pieces by adding 400 μL of ACN. Wait until the gel pieces are fully dehydrated (white color) and remove ACN. 3. Add 200 μL of 10 mM DTT in order to reduce the cysteine residues. Briefly agitate the tube and heat at 60
C for 30 min. Then place the tube at room temperature for 30 min. 4. Add 200 μL of 55 mM IAA and place the tube in the dark for 20 min. 5. Add 400 μL of ACN. Briefly agitate the tube and remove the mixture contained in the tube. 6. In order to wash the gel pieces, add 200 μL of 25 mM NH4HCO3. Briefly agitate the tube and wait 5 min before adding 200 μL of ACN. Agitate the tube and remove the mixture. Repeat this step 3 times. 7. Finally dehydrate the gel pieces with 200 μL ACN. Wait until the gel pieces are fully dehydrated (white color) and remove ACN. Repeat this step twice (see Note 8).

126

4

Leslie Muller et al.

Notes 1. To prepare this solution, mix 0.5 mL of 0.5 M Tris–HCl, pH 6.8 with 0.8 mL of 10% SDS solution, 0.4 mL of glycerol (It is also possible to weigh glycerol because it is very viscous and difficult to pipette. In this case, weigh 0.5 g of glycerol) and 2 mL of water. 2. Bring the solution at room temperature before use as SDS precipitates at 4
C. 3. Prepare this fresh each time. 4. To prepare this solution mix 450 mL of methanol with 50 mL of acetic acid and 500 mL of water in a glass bottle. 5. Prepare this solution fresh. Wear a mask when weighing dithiothreitol and iodoacetamide. Handle this solution under the laboratory fume hood. 6. Depending on the sample nature, heating, vortexing, and/or sonication steps can be added. 7. Caution should be taken with bubbles as air interferes with the polymerization process. All bubbles must be removed for an optimal polyacrylamide polymerization in order to avoid sample loss. Be careful, in these conditions the polymerization is fast and takes place in a few seconds. 8. At this step, the sample can be stored at –20
C prior to digestion. In this case, caution should be taken in order to remove all the liquid before freezing.

Acknowledgments This work was supported by the French Proteomic Infrastructure (ProFI; ANR-10-INBS-08-03). References 1. Lu X, Zhu H (2005) Tube-gel digestion: a novel proteomic approach for high throughput analysis of membrane proteins. Mol Cell Proteomics 4(12):1948–1958. https://doi.org/10.1074/ mcp.M500138-MCP200 2. An M, Dai J, Wang Q et al (2010) Efficient and clean charge derivatization of peptides for analysis by mass spectrometry. Rapid Commun Mass Spectrom 24(13):1869–1874. https://doi.org/ 10.1002/rcm.4589 3. Cao L, Clifton JG, Reutter W et al (2013) Mass spectrometry-based analysis of rat liver and hepatocellular carcinoma Morris hepatoma

7777 plasma membrane proteome. Anal Chem 85(17):8112–8120. https://doi.org/10.1021/ ac400774g 4. Cao R, He Q, Zhou J et al (2008) Highthroughput analysis of rat liver plasma membrane proteome by a nonelectrophoretic in-gel tryptic digestion coupled with mass spectrometry identification. J Proteome Res 7 (2):535–545. https://doi.org/10.1021/ pr070411f 5. Han CL, Chien CW, Chen WC et al (2008) A multiplexed quantitative strategy for membrane proteomics: opportunities for mining

Tube-gel Sample Preparation for High-Throughput Proteomics therapeutic targets for autosomal dominant polycystic kidney disease. Mol Cell Proteomics 7(10):1983–1997. https://doi.org/10.1074/ mcp.M800068-MCP200 6. Smolders K, Lombaert N, Valkenborg D et al (2015) An effective plasma membrane proteomics approach for small tissue samples. Sci Rep 5:10917. https://doi.org/10.1038/srep10917 7. Yu H, Wakim B, Li M et al (2007) Quantifying raft proteins in neonatal mouse brain by ‘tubegel’ protein digestion label-free shotgun proteomics. Proteome Sci 5:17. https://doi.org/10. 1186/1477-5956-5-17

127

8. Zhou J, Xiong J, Li J et al (2010) Gel absorption-based sample preparation for the analysis of membrane proteome by mass spectrometry. Anal Biochem 404(2):204–210. https://doi.org/10.1016/j.ab.2010.05.013 9. Muller L, Fornecker L, Van Dorsselaer A et al (2016) Benchmarking sample preparation/ digestion protocols reveals tube-gel being a fast and repeatable method for quantitative proteomics. Proteomics 16(23):2953–2961. https:// doi.org/10.1002/pmic.201600288

Chapter 9 Protein Biomarker Discovery in Non-depleted Serum by Spectral Library-Based Data-Independent Acquisition Mass Spectrometry Alexandra Kraut, Mathilde Louwagie, Christophe Bruley, Christophe Masselon, Yohann Coute´, Virginie Brun, and Anne-Marie Hesse Abstract In discovery proteomics experiments, tandem mass spectrometry and data-dependent acquisition (DDA) are classically used to identify and quantify peptides and proteins through database searching. This strategy suffers from known limitations such as under-sampling and lack of reproducibility of precursor ion selection in complex proteomics samples, leading to somewhat inconsistent analytical results across large datasets. Data-independent acquisition (DIA) based on fragmentation of all the precursors detected in predetermined isolation windows can potentially overcome this limitation. DIA promises reproducible peptide and protein quantification with deeper proteome coverage and fewer missing values than DDA strategies. This approach is particularly attractive in the field of clinical biomarker discovery, where large numbers of samples must be analyzed. Here, we describe a DIA workflow for non-depleted serum analysis including a straightforward approach through which to construct a dedicated spectral library, and indications on how to optimize chromatographic and mass spectrometry analytical methods to produce high-quality DIA data and results. Key words Mass spectrometry, Data-independent acquisition, Serum, Biomarker discovery, Labelfree quantification

1

Introduction The search for new clinical biomarkers has been steadily growing over the past two decades. Indeed, the identification and validation of diagnostic, predictive, or prognostic biomarkers could have a huge impact on patient care and clinical outcome. Classically, the biomarker development pipeline is divided into three phases: biomarker discovery, verification, and validation [1]. Quantitative mass spectrometry-based proteomics has become central to the first two steps in this pipeline as it is a powerful tool which can identify and

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_9, © Springer Science+Business Media, LLC, part of Springer Nature 2019

129

130

Alexandra Kraut et al.

quantify novel candidate biomarkers for which abundance is altered in disease-specific conditions. Every year, numerous studies report candidate protein biomarkers that have been discovered or verified by mass spectrometry (MS) [2–10]. However, a relatively small number of these candidates has been approved by the US Food and Drug Administration (FDA) and transferred for use in medical biology laboratories [11]. This low transfer rate indicates that MS-based proteomics may not be mature enough to accurately yield actual biomarkers with a low false-positive rate. As a result, biomarker discovery and evaluation pipelines require robust and high-throughput protein assays to reduce attrition and deliver valuable candidate biomarkers at earlier stages [6]. Discovery proteomics is typically performed by shotgun experiments through data-dependent acquisition (DDA). In this acquisition mode, ions are successively isolated from the survey scan for individual fragmentation by tandem mass spectrometry (MS/MS), based on a selection process which is biased toward more abundant species. The tandem mass spectra obtained are then submitted to a database search engine to identify the selected peptides and their corresponding proteins. In label-free approaches, peptide quantification is based either on spectral counting or on the extracted signal for the precursor ions [12]. Over the last decade, improvements in MS technology, and particularly in acquisition speed, have made this approach efficient for the rapid identification of high numbers of peptides and proteins in complex mixtures [13]. However, due to the stochastic nature of precursor ion selection, biased toward more abundant peptides, DDA often lacks reproducibility across samples and provides an incomplete peptide quantification matrix. Biomarker verification is typically performed by targeted proteomics approaches such as multiple reaction monitoring (MRM) or parallel reaction monitoring (PRM) [14]. Signature peptides from the selected candidate biomarkers, and their sequence-specific fragments, are used to create transition lists for selective monitoring. This approach achieves high specificity with very precise and accurate quantification when combined with isotope-dilution standards [15], but can only be used for limited numbers of target species. Furthermore, setting up and optimizing a sensitive and robust SRM protein assay is a time-consuming task that must be repeated for every new list of candidates. Thanks to recent developments in high-resolution mass spectrometry instruments, data-independent acquisition (DIA), a concept introduced more than 10 years ago [16], has become a realistic option to achieve the ultimate goal of proteomics: the comprehensive identification and quantification of proteins and peptides in complex biological matrices. DIA consists in generating series of multiplex MS/MS spectra from all the peptides contained in a defined mass-to-charge (m/z) window. To cover the full m/z range, DIA methods are designed to monitor sequential m/z

Serum Biomarker Discovery using DIA Proteomics

131

windows at regular intervals. Peptide and protein identification from the multiplex spectra produced remains challenging and requires dedicated computational solutions [17]. For example, peptide-centric approaches use existing spectral libraries to identify peptides by extracting their specific fragments from multiplex MS/MS signals [18, 19]. These solutions are derived from methods used for SRM-based targeted proteomics data analysis. Alternatively, spectrum-centric approaches attempt to deconvolute multiplexed MS/MS spectra by extracting fragments and precursor ions for which elution profiles correlate, producing pseudo singleprecursor MS/MS spectra that can be submitted to classical DDA search engines [20, 21]. This approach can identify peptides and proteins absent from public generic spectral libraries, which may have been obtained from samples differing from the samples studied. However, by using a specific spectral library produced from the samples of interest this shortcoming can be overcome. Altogether, the DIA approach mitigates the weaknesses of DDA and SRM, while combining their advantages: high-throughput, speed on the one hand, and quantification accuracy and reproducibility, and a high level of data completeness on the other hand [22]. Moreover, the resulting DIA files can be viewed as complete and permanent digital proteome maps for further analysis and reuse in a number of data-mining schemes [23]. The DIA strategy may turn out to be the long-sought robust and reproducible MS-method to discover biomarkers from large clinical cohorts, and indeed it has already been successfully employed to this end [24, 25]. Among clinical samples, plasma or serum are the most frequently studied biofluids in proteomics [6, 26] as they are routinely collected in a controlled manner and widely available in biobanks. However, due to the extremely large dynamic range of protein abundances in these fluids [27], their analysis by proteomics approaches is very challenging. Indeed, to achieve deep and extensive proteome coverage, it is very common to deplete the sample from highly abundant proteins and perform extensive prefractionation in order to reveal low-abundance proteins and thus achieve deeper proteome coverage [28]. However, such biochemical treatments have limitations and can significantly impair technical reproducibility. Firstly, plasma/serum depletion is associated with nonspecific protein loss as specifically immunodepleted protein targets can act as carriers for other, less abundant (and potentially more interesting) proteins [29, 30]. Secondly, important modifications of the plasma/serum proteome (such as the induction of acute-phase proteins or hypoalbuminemia) can modify the binding capacity of depletion devices and introduce some bias. Based on these considerations, we decided to develop a librarybased DIA workflow for the analysis of non-depleted serum samples. Although generic DIA methods perform well on fairly complex samples such as cells lysates, their performance is more limited

132

Alexandra Kraut et al.

on samples with an extreme dynamic range like non-depleted serum. Thus, we first had to perform extensive analytical optimization to achieve adequate proteome coverage and reproducible quantification in these unfavorable conditions [31, 32]. This chapter presents a detailed description of our protocol by which a high-quality spectral library dedicated to the cohort being studied can be constructed, based on a simple peptide fractionation method. Optimized analytical chromatography and mass spectrometry DIA methods are also presented. Finally, data processing and statistical analysis are described for differential analysis. We applied this strategy without protein immunodepletion or prefractionation, and—in 1 h—reproducibly quantified more than 280 proteins from serum samples from healthy subjects and patients, allowing the identification of differentially abundant proteins.

2

Materials 1. 0.5-mL and 1.5-mL polypropylene tubes, pipettes, pipette tips are needed at each step. 2. 3 μL of serum from each healthy subject and patient. 3. High-speed refrigerated micro centrifuge. 4. 0.5-mL centrifugal filter units with 10-kDa MWCO cellulose membrane (Millipore). 5. Urea buffer: 4 M urea in 25 mM ammonium bicarbonate. 6. Tris(2-carboxyethyl)phosphine (TCEP) solution: 20 mM TCEP in urea buffer. 7. Iodoacetamide (IAA) solution: 55 mM IAA in urea buffer. 8. Digestion buffer: 25 mM ammonium bicarbonate (AB). 9. Proteomics-grade Trypsin/Lys-C Mix (Promega). 10. Desalting cartridges (C18 macro SpinColumns from Harvard Apparatus, see Note 1). 11. LC-MS grade water. 12. HPLC gradient grade Acetonitrile (ACN). 13. High purity Formic acid (FA). 14. Vacuum Concentrator. 15. High pressure Nano-LC system compatible up to 700 bars, (for this work, an Ultimate 3000 nano RSLC (Thermo Fisher Scientific) was used) configured in trapping mode. 16. Trap column (for this work, an Acclaim™ PepMap™ 100 C18 300 μm
5 mm trapping column (Thermo Fisher Scientific) was used).

Serum Biomarker Discovery using DIA Proteomics

133

17. Analytical column (for this work, a Reprosil-Pur 120 C18-AQ, 1.9 μm, 75 μm
25 cm column (Dr. Maisch) was used). 18. Fused silica capillary: 25 μm internal diameter, 280 μm external diameter. 19. PTFE connection tubing 250/350 μm. 20. SilicaTip Emitters, such as distal coated, compatible with a 100–400 nL/min flow rate (e.g., 10 μm internal diameter, 360 μm external diameter). 21. High-resolution mass spectrometer (for this work, a QExactive HF equipped with Nanospray Flex Ion Source (Thermo Fisher Scientific) was used). 22. Reagent-grade Trifluoroacetic acid (TFA). 23. Loading solvent: 5% ACN, 95% water, 0.1% TFA. 24. HPLC solvent A: 2% ACN, 98% water, 0.1% FA. 25. HPLC solvent B: 80% ACN, 20% water, 0.08% FA. 26. UHPLC-compatible injection vials, inserts, and caps. 27. Strong Cation eXchange (SCX) tips (for this work, we used GL-Tips from GL-Sciences). 28. Centrifuge Adapter (GL-Sciences). 29. Ammonium acetate 7.5 M (AA). 30. Peptide mixture for quality control (e.g., Cytochrome C digest). 31. HRM Calibration kit (Biognosys): peptide mixture to calibrate retention times. 32. Personal Computer with at least 32 GB RAM and 16 cores, a 64-bit Windows operating system with Microsoft Office, a recent version of MaxQuant [33] software (http://www.cox docs.org) and a recent version of Spectronaut [34] software (https://biognosys.com) installed.

3

Methods The proposed methodology for analysis of non-depleted serum relies on DIA MS and a peptide-centric computational pipeline requiring a spectral library. Various strategies can be used to select the data sources to compile when constructing the spectral library. For instance, publicly available datasets can be mined to assemble large-scale libraries to provide comprehensive proteome coverage. Alternatively, more focused libraries can be compiled by applying DDA analyses to identify proteins contained in the same samples to be analyzed by DIA. Sample-specific spectral libraries have been shown to produce better results than large nonspecific libraries in

134

Alexandra Kraut et al.

terms of proteome coverage and quantification reproducibility [24]. Indeed, dedicated libraries improve retention time alignment between DIA acquisitions and library acquisitions, and better fragment ion pattern matching, both of which are important parameters for data processing. Below, we will first discuss sample preparation and peptide fractionation to quickly and easily build a dedicated spectral library. Then, we will describe optimization of the liquid chromatography conditions using nonlinear gradients to improve peptide separation quality [35, 36], and optimization of the DIA method using variable isolation windows for MS/MS analysis to increase proteome coverage and analysis specificity [37]. Finally, we will present DIA data processing which involves identification and quantification of peptides and proteins alongside statistical analyses to identify candidate biomarkers. 3.1 Sample Preparation

Classically, biomarker discovery studies consist in case-control comparisons performed using biofluid samples from healthy subjects compared to samples from patients with a specific disease. As a good starting point we recommend using 3 μL of serum collected from at least five different individuals from each group (healthy / disease). Ideally, a preliminary study should be performed to determine the required number of biological replicates needed to reach the intended statistical power and significance levels [38].

3.1.1 Serum Sample Digestion

Serum samples can be digested using filter-aided sample preparation (FASP [39]) with slight variations on the described protocol [40] (see Note 2). 1. For each sample, place a filter unit in a labeled collection tube (10
). 2. Dilute 3 μL of serum in 200 μL TCEP solution. Load the diluted sample onto the filter. Vortex carefully and incubate for 30 min at room temperature. Centrifuge at 12,000
g for 15 min at 4  C. Centrifugation time should be adjusted to retain around 50 μL of protein solution above the filter. Discard flow-through from the collection tube as necessary. 3. Add 200 μL of urea buffer to wash the filter. Vortex carefully and centrifuge at 12,000
g for 15 min at 4  C. 4. Add 200 μL of IAA solution onto the filter. Vortex carefully and incubate for 15 min at room temperature in the dark. Centrifuge at 12,000
g for 15 min at 4  C. 5. Add 200 μL of urea buffer to wash the filter. Vortex carefully and centrifuge at 12,000
g for 15 min at 4  C. Repeat this step once again. 6. Add 10 μL of trypsin/LysC mix (1 μg/μL in 25 mM AB). Vortex carefully and incubate for 2 h at 37  C.

Serum Biomarker Discovery using DIA Proteomics

135

7. Dilute urea by adding 250 μL digestion buffer (final urea concentration < 1 M). Vortex carefully and incubate for 5 h at 37  C (see Note 3). 8. Transfer the filter to a new collection tube and centrifuge at 12,000
g for 15 min at 4  C. 9. Add 100 μL of digestion buffer. Vortex carefully and centrifuge at 12,000
g for 15 min at 4  C. If necessary, repeat the centrifugation to pass the whole sample through the filter. 10. Remove the filter. 11. Acidify the peptide solution directly in the collection tube by adding FA until the pH is less than 3. 12. Purify the peptides on C18 SpinColumns according to published protocols [41]. After elution, store samples at 4  C until needed for qualification (see Subheading 3.1.2). 3.1.2 Sample Qualification by LC-MS/MS

After protein digestion and peptide desalting, samples must be qualified both to verify that these steps have been completed properly and to determine the final peptide amount (see Note 4). Here, we describe the LC-MS/MS analysis of the samples by a QExactive HF coupled on-line to an Ultimate 3000 nano RSLC liquid chromatography, but any other kind of LC-MS instrument can be used. 1. Prepare an LC method to separate peptides in trapping mode: After injection using μLPickUp mode on a 20-μL sample loop, peptides are loaded onto a trap column over 3 min at a flow rate of 20 μL/min in 100% solvent A. After valve switching, peptides are eluted and separated on the reversed phase column by applying a linear gradient from 4% to 50% solvent B over 1 h at a flow rate of 300 nL/min. The column is then washed with 90% solvent B for 15 min and finally re-equilibrated with 4% solvent A for 15 min. During the whole procedure, the autosampler is maintained at 5  C, and the column oven at 35  C. 2. Prepare an MS acquisition method to analyze peptides with a data-dependent top-20 strategy performed on a QExactive HF in positive mode. Acquire survey full scan MS spectra (from m/ z 400 to 1600) in the Orbitrap using a resolution of 60,000 with AGC target of 1e6 ions (max injection time: 200 ms). Isolate and fragment the 20 most intense ions (excluding z ¼ 1 and unassigned charge states) using HCD with a normalized collision energy of 30, a resolution of 15,000, and an AGC target of 1e5 ions (max injection time: 100 ms). Set dynamic exclusion to 30 s, intensity threshold to trigger MS/MS at 1e4, and default charge state to 3. The source should be operated with the following parameters: nanospray voltage, 2 kV; ion transfer capillary temperature, 270  C; and S-Lens RF amplitude, 55%. Acquire data with a lock-mass of 445.12003 Th.

136

Alexandra Kraut et al.

3. Connect trap and analytical columns to the LC instrument and follow the manufacturer’s instructions to flush and equilibrate them (see Note 5). Couple the LC column to the MS source using a silica capillary and a SilicaTip Emitter. 4. Isolate a small aliquot (equivalent to 2 μg) of each serum sample prepared as described in Subheading 3.1.1, and dry it under vacuum. Resuspend the dried peptides in loading solvent (0.2 μg/μL). 5. Inject around 200 ng of peptides into the LC-MS system and analyze them using the previously prepared LC-MS/MS methods. 6. On the resulting raw data files, verify that the elution profiles correspond to well-digested samples. Quantity can be evaluated from Total Ion Current measurement. Maximum TIC should be around 1010 counts. Considerable variations of TIC between the different samples (>20%) should be compensated for before the final data-independent acquisition (DIA) by calculating a normalization factor. 7. Prepare two pooled samples containing about 25 μg of peptides in total. For example, combine 5 μg from each healthy subject with 5 μg from each patient. Prepare a global sample pool by combining 5 μg from each sample. 8. Aliquot the remaining volume of peptides obtained following the desalting step described in Subheading 3.1.1 (for example, 5 μg /aliquot), taking the normalization factor into account to adjust quantities. Dry all samples under vacuum and store frozen at 20  C. 3.1.3 Sample Qualification by Protein Identification

In addition to direct observation of TIC intensities, it is recommended to run peptide and protein identification to verify that all digested samples are suitable for use in subsequent steps. To do so, any search engine can be used. Here, we describe data processing using MaxQuant software. The same procedure will be used at various steps later on in the protocol (gradient optimization and spectral library generation). 1. Download the latest version of MaxQuant (http://www.cox docs.org) and unzip it. 2. To obtain a human protein sequence database from the UniProtKB server (http://www.uniprot.org), download the most recent fasta file corresponding to manually annotated and reviewed sequences (Swiss-Prot). Delete all human entries from the MaxQuant contaminants.fasta file (in “..\bin\conf”). 3. Open MaxQuant. Here we describe the parameter settings for version 1.6.2.10. If not specified, use MaxQuant default settings.

Serum Biomarker Discovery using DIA Proteomics

137

4. In the “Raw data” tab, load the different raw files to be processed. Give each file a name in Experiment (the “Write template” and “Read from file” item can be used to define the ExperimentalDesignTemplate). 5. Go to the “Group-specific parameters” tab and select the protease used during the digestion step. Set the Max. missed cleavages to 3. Select the variable modifications that should be taken into account, such as methionine oxidation and protein N-terminal acetylation (see Note 6). Select the fixed modification, here cysteine carbamidomethylation, or according to the biochemical steps applied during sample preparation. 6. Go to the “Global parameters” tab and add the FASTA files previously downloaded. Set the correct identifier and description parsing rules to read them. Disable the second peptide option (see Note 7). Choose Unique peptides for quantification. 7. Set number of processors based on the number of files to process and the computer configuration. Click Start. 8. Once the search is finished, check that the number of proteins and peptides identified in the proteinGroups.txt and peptides. txt files are consistent between samples (differences should be less than 15% between replicates). 3.1.4 Fractionation of Peptide Digests

To increase proteome coverage and library size, DDA replicate runs or sample prefractionation must be performed. This is particularly important when analyzing samples such as serum or plasma which have an extreme dynamic range of protein concentrations. Recently, it was shown that in-gel or peptide fractionation produce larger library sizes than simply performing DDA replicates [42]. However, as in-gel digestion is difficult to apply to non-depleted serum or plasma samples, we chose to use a simple and rapid protocol to obtain good proteome coverage using SCX peptide fractionation performed on each sample pools (healthy/ disease). 1. Prepare the following conditioning, washing, and elution solutions (volume for two SCX tips) and mix each prepared tube thoroughly (Table 1). 2. Connect a centrifuge adaptor to a waste fluid tube and insert an SCX column (2
). 3. Add 50 μL of solution B to each tip and centrifuge at 3000
g for 2 min at room temperature. 4. Add 50 μL of solution A to each tip and centrifuge at 3000
g for 2 min at room temperature. 5. Dissolve each pooled sample in 50 μL of solution A and mix thoroughly.

138

Alexandra Kraut et al.

Table 1 Conditioning, washing, and elution solutions used during SCX fractionation Final volume (μL)

Fraction

10% TFA (μL)

ACN (μL)

350

A

3.5

250

B

2.5

200

0

47.5

150

C

7.5

45

0

97.5

150

D

15

45

0

90

150

E

30

45

0

75

150

F

45

45

0

60

150

G

45

45

2

58

150

H

60

45

2

43

150

I

0

45

10

95

17.5

AA (μL) 0

Water (μL) 329

6. Change the recovery tubes below tips and load samples into clearly labeled tips. Centrifuge at 3000
g for 5 min at room temperature. 7. Add 50 μL of solution A to each tip and centrifuge at 3000
g for 2 min at room temperature. Label the recovery tube “flowthrough” followed by the name of the sample and store at 4  C. Place a fresh recovery tube below each tip. 8. Repeat this step, 8 times with solutions B to I, labeling successive recovery tubes with letters B to I and the sample name. Store all samples at 4  C. 9. Dry all recovered fractions under vacuum and store frozen until LC-MS analysis. 3.2 Spectral Library Generation

The protocol below describes how to produce a spectral library. Basically, it consists in analyzing each sample destined for library generation using a nanoflow LC system coupled to a highresolution mass spectrometry instrument for MS/MS analysis. After nanoLC-MS/MS acquisition, data are searched against a protein database to identify peptides and proteins. Finally, a spectral library is constructed containing the full list of peptides identified with their precursor mass, their observed fragment masses and intensities, and the retention time at which they were observed in DDA runs.

3.2.1 Optimization of the Reverse-Phase Nano-LC Gradient

To improve the depth of analysis, it is recommended that LC conditions be specifically optimized for the biological matrix of samples used for biomarker discovery. Optimized nonlinear gradients make the whole run duration to be exploited and limit the number of co-eluted peptides at any given time. Depending on the

Serum Biomarker Discovery using DIA Proteomics

139

sample complexity and dynamic range, this procedure can possibly increase the numbers of peptides identified. Note that identical LC conditions should be used to construct the library and for DIA sample analysis (see Note 8). 1. Determine the dead time for your complete LC system (including the LC pump capillary volume, trap and analytical columns and any post-column capillary used to link the LC to the MS source). This time corresponds to the delay between the time when gradient instructions are sent to the pumps and when their effects can be observed in the MS file. A simple procedure consists in increasing the percentage of B solvent from 4% to 90% in less than 10 s just after the 3 min loading step. For this step, it is not necessary to use a complex peptide mixture: a simple protein digest such as Cytochrome C can be injected. 2. Dissolve the dried peptides from the global sample pool in loading solvent (0.1 μg/μL). 3. Analyze 500 ng of this sample with a 1-h linear gradient from 4% to 50% B and a conventional MS method (see Subheading 3.1.2). 4. Run MaxQuant as described in Subheading 3.1.3 to identify all the peptide sequence matches (PSM) from the raw data acquired. 5. Open the msms.txt file in Excel and use the Retention Time column after removing sequences matching PSM from the reverse database. Sort PSM according to their retention time (ascending order). Convert retention times into LC method times by simply subtracting the dead time from the observed retention time. 6. Finally, convert these LC method times into %B to know how much B is needed to elute each peptide. 7. Build the initial experimental distribution of identified PSM as a function of the percentage of B solvent and gradient time (Fig. 1a). 8. Optimize the gradient by equally distributing peptides across the gradient time: for example, divide the total number of PSM by six to determine the number of PSM to elute every 10 min. Apply this procedure to each 10-min fraction of LC time (see Note 9). Suppose that 1600 PSM are expected by 40 min in the new gradient with this calculation. On the initial distribution, 20% B is needed to obtain this number. Consequently, the optimized gradient must reach 20% B after 40 min. 9. Analyze 500 ng of the same sample with the optimized nonlinear gradient and verify that the distribution obtained is more linear (Fig. 1b).

140

Alexandra Kraut et al.

Fig. 1 Representation of the cumulative distribution of the PSM (dots) identified as a function of time. The solid line represents the percentage of B solvent in the gradient as a function of time. (a) Shows data obtained with the linear gradient. (b) Shows data obtained with the optimized gradient Table 2 Optimized 1-h gradient for serum analysis in LC-MS/MS LC time (min)

% Solvent B

0

4

3

4

5.5

11.4

38

21

54

30.4

63

50

64

90

79

90

80

4

95

4

Table 2 presents the optimized gradient for non-depleted serum samples obtained from healthy subjects and patients. 3.2.2 LC-MS/MS Analyses of Peptides Fractions

Each peptide fraction must then be analyzed by LC-MS/MS. Most of the software used to process DIA data based on matching with a spectral library require a good correlation between fragment ion intensities in the library and in DIA runs. Therefore, parameters like collision energy and MS/MS m/z range should be the same in the DIA and DDA methods. Other parameters can be the same as in a standard DDA method for shotgun proteomics. 1. Prepare HRM-IRT stock solution (10 UI/μL) as described by the manufacturer. Then prepare a diluted HRM-IRT solution

Serum Biomarker Discovery using DIA Proteomics

141

at 0.33 UI/μL: in a clean tube thoroughly mix 193 μL Loading buffer with 7 μL HRM-IRT stock solution. This volume corresponds to enough solution to add to the 16 samples obtained during the fractionation step. 2. Resuspend each SCX fraction in 40 μL of Loading buffer (around ng/μL). 3. For each sample, mix 8 μL of sample with 12 μL of the diluted HRM-IRT solution in a clean injection insert. Store the remaining samples at 20  C. The final concentration of HRM-IRT is calculated so as to inject 1 UI on the column. 4. Inject 5 μL of each sample, running the previously optimized UHPLC gradient (see Subheading 3.2.1) combined with the DDA method described in Subheading 3.1.2. Alternate serum injection with blank solvent injection to flush the fluidics and columns. 3.2.3 Library Generation

DDA raw data are searched against the protein database to identify peptides that will be used to build the spectral library. 1. Run MaxQuant, applying the same parameters described in Subheading 3.1.3. In addition to the Human fasta file, load the IRTfusion fasta file and configure the different fractions in the ExperimentalDesignTemplate. 2. Once the search is completed, start Spectronaut. Here, we present the parameter settings for version 11.0. 3. Under the “Databases” section, import the collection of fasta files used with MaxQuant (SwissProt Human, IRT fusion and Contaminants protein databases). Set the correct parsing rules. 4. Under the “Prepare” section, select Generate Spectral Library from MaxQuant. Choose the folder containing DDA runs and the combined MaxQuant subfolder created previously. Use default parameters and select the fasta files used to infer proteins (see Note 10). Name and create the library.

3.3 DataIndependent Acquisition

The protocol below describes how to discover biomarker candidates from serum samples using a spectral-library-assisted DIA strategy. Basically, the procedure consists in analyzing each prepared sample from healthy subjects and patients by nanoLCMS/MS with a high-resolution mass spectrometer. After LC-MS/ MS acquisition, data are processed with DIA-compatible software to obtain a list of identified and quantified peptides and proteins. Finally, the results can be statistically analyzed to generate a list of biomarker candidates.

3.3.1 Optimization of the DIA Method

Optimizing the width of DIA selection windows basically consists in equalizing the number of precursor ions that will be present in each window.

142

Alexandra Kraut et al.

Table 3 Correspondence between resolution, Orbitrap transient duration, and free injection time for QExactive HF Resolution

Transient (with interscan)

Free injection time

Max scan speed

15,000

32 ms (40 ms)

10 ms

25 Hz

30,000

64 ms (70 ms)

40 ms

14 Hz

60,000

128 ms (140 ms)

100 ms

7 Hz

1. Extract the MS signals of the different peptides of the HRM Calibration kit in one of the library runs acquired with the optimized nonlinear gradient. Measure the peak width of the chromatographic peaks obtained and calculate the median (on our system, this width was around 25 s). 2. Divide the median peak width by the number of data points required along each chromatographic peak to determine the cycle time needed. It is recommended to have between 10 and 15 data points to correctly sample each chromatographic peak. In our case a cycle time of 1.6 s was needed. 3. Calculate the maximum number of consecutive isolation windows that can be acquired during a single cycle. This number will depend both on the maximum ion injection time (max IT) chosen and on the resolution chosen. Table 3 indicates the correspondence between Resolution and scan duration (corresponding to transient with interscan). For each resolution, it also indicates the maximum injection time that can be set to fill the octopole cell with ions from the (N þ 1)th DIA window in parallel to scanning the Nth DIA window in the Orbitrap. If a longer injection time is set, the scan rate will be impacted. It is recommended that injection times greater than 50 ms (90 ms with interscan) be used to provide adequate sensitivity during MS/MS. Our method is designed with 18 DIA windows and 55 ms of injection time to produce a cycle time of less than 2 s. 4. From the evidence.txt file produced by the MaxQuant search used to generate the final spectral library (see Subheading 3.2.3), delete the reverse peptides and generate a nonredundant list of precursor ions (based on Sequence, Modification, and m/z columns). 5. Build the experimental distribution of the number of precursor ions as a function of m/z (Fig. 2). Choose the m/z range that will be covered by the DIA method, either full m/z range or a more limited one (see Note 11).

Serum Biomarker Discovery using DIA Proteomics

143

Fig. 2 Distribution of ions’ m/z in the serum spectral library obtained following SCX fractionation. Ninety percent of ions are identified in the 427–962 m/z range. Centers and window sizes for the DIA method were optimized for this m/z range and are represented on the graph

6. Divide the total number of precursor ions in this m/z range by the number of windows to determine the fixed number of precursors per window. Optimize and center the isolation window size to distribute the number of precursor ions equally across the different windows (see Note 12). 7. Create a DIA method on the QExactive HF with a Full MS scan and as many DIA scans as necessary to obtain the appropriate isolation window size and center. Acquire survey full scan MS spectra (from m/z 400 to 1600) in the Orbitrap with a resolution of 60,000 after accumulation of 3e6 ions (maximum injection time 200 ms). Select all DIA scans together to configure common parameters: set default charge state to 3, resolution to 30,000, AGC to 1e6 ions with a maximum injection time of 55 ms and set normalized collision energy to 30. Sequentially select each DIA scan to configure each isolation window. In “Global Lists,” add an inclusion list with the centers of each selection window ranked in ascending order. Operate the source with the following parameters: nanospray voltage, 2 kV; ion transfer capillary temperature, 270  C; S-Lens, 55%. Data are acquired with a lock-mass of 445.12003 Th. The isolation window sizes and centers optimized for serum samples with our LC-MS/MS setup are summarized in Table 4.

144

Alexandra Kraut et al.

Table 4 Optimized DIA method for serum proteome profiling Center (Th)

Window size (Da)

Center (Th)

Window size (Da)

440

26

621.5

21

465.5

25

643

22

488

20

666

24

507

18

693.5

31

524.5

17

723.5

29

542

18

757

38

560.5

19

796.5

41

580

20

845.5

57

600.5

21

918

88

3.3.2 Data-Independent Acquisition from Individual Samples

1. Prepare a diluted HRM-IRT solution at 1 UI/μL: in a clean tube, thoroughly mix 49.5 μL of Loading solvent with 5.5 μL of the HRM-IRT stock solution. Store at 4  C until required. 2. Resuspend one stored aliquot from each patient or healthy subject in 10 μL of Loading solvent (around 500 ng/μL). 3. For each sample, in a clean injection insert, mix 5 μL of sample with 5 μL of the prepared diluted HRM-IRT solution. Store the remaining samples at 20  C. The final concentration of HRM-IRT is calculated to inject 1 UI to the column. 4. Inject 2 μL of each sample, and run the previously optimized UHPLC method combined with the previously optimized DIA method. Alternate serum injections with blank solvent injections to clean the fluidics and column.

3.3.3 DIA Data Analysis

Once DIA has been performed, a dedicated software suite must be used to analyze the data. Recently, a number of software tools were benchmarked on a standard mixture. This analysis revealed good convergence of results between the different tools evaluated [17]. Slight advantages in terms of proteome coverage and quantification reproducibility were found for Spectronaut. Moreover, this software offers a user-friendly, ergonomic graphical interface, in a core facility perspective. Here, we describe the procedure we used to process DIA data; for more detailed information on each parameter, please refer to the software user’s manual. Once data have been processed, descriptive statistics must be generated and differential analysis performed between the different conditions. Any statistical tool can be used, here we describe the procedure to process a quantitative matrix using DAPAR/ProStaR packages [43] via a public server. For more technical information please

Serum Biomarker Discovery using DIA Proteomics

145

refer to the dedicated chapter in this volume of Methods in Molecular Biology. 1. After acquisition of the DIA data for the ten samples, open Spectronaut. Under the “Review” section, create a new Experiment by loading the raw files. Name the study and assign it to a spectral library. Choose the spectral library from “Prepare Perspective,” previously created in Subheading 3.2.3 (see Note 13). 2. Configure conditions by setting Condition and Replicate name. Select a reference condition (e.g., healthy) (see Note 14). 3. Define the analysis settings. Set the Library size fraction to 1, use only Protein-Group-specific peptides to quantify proteins, choose sum as the aggregation method when calculating peptide quantity from precursors and calculating protein quantity from peptides. Uncheck the top three methods to use all validated precursors and peptides. Set “by Precursor” as the Minor (Peptide) Grouping. Deselect the Cross-run normalization in Quantification and Differential Abundance Testing in Post Analysis as they will be deployed later with another tool. Select the appropriate fasta files and start the analysis (see Note 15). 4. Save the final Spectronaut Experiment.sne file and export the “Protein Quant (Pivot)” file (see Note 16). Examine the different graphs generated showing the retention time calibration or XIC Extraction Width for each DIA file to assess the quality of the alignment between DIA runs and the spectral library. 5. Prepare the protein export by replacing all “Filtered” cells by NaN. Save the file as a tab-separated text file. 6. Connect to the ProStaR server http://prostar-proteomics. org/ (see Note 17). 7. In the Dataset manager, convert the .txt file to MSnset format. Choose protein dataset rather than peptide dataset. Select “PG. ProteinAccessions” as the User ID. Select all “PG.Quantity” columns as Quantitative Data. Configure the Samples’ metadata by applying the correct “label” (at least two conditions, Healthy and Disease) and “replicate” names (here, 5 biological replicates per condition). Finally, name the study and launch the conversion. 8. Explore the Descriptive statistics panels to verify that there is good correlation between replicates of the same condition. 9. Begin Data processing by filtering data for missing values: Retain proteins with at least one complete condition (here, at least one condition with five intensity values). Filter contaminants by applying the appropriate string for the fasta file used.

146

Alexandra Kraut et al.

10. In the normalization step, choose quantile centering with median. The “overall” normalization type can be chosen if a large majority of proteins is expected not to vary between sample groups. Otherwise, if extensive changes in protein abundance are expected, choose “within conditions” normalization. 11. To be able to perform the differential analysis, impute missing values. More information about the imputation parameters can be found in Subheading 3.8 from the chapter dedicated to DAPAR/ProStaR in this volume of Methods in Molecular Biology. 12. Perform differential analysis by selecting the limma test: for each protein a fold-change (FC) and a p-value are calculated. Set the log2(FC) threshold to 1 (verify that enough proteins are selected to ensure a valid estimate of the false-discovery rate—FDR). Select the Benjamini-Hochberg p-value calibration method and select a p-value threshold to obtain a FDR below 1%. Save the result. View the Volcano plot generated (scatter plot representing –log10( p-value) as a function of log2(FC), Fig. 3a). The plot should be relatively symmetrical around 0 (corresponding to invariant proteins). A proportion of proteins should exhibit high absolute FC with very low pvalues in the upper right and left areas. These points correspond to the proteins that are statistically up- or downregulated by the disease. Export the results (dataset manager/export) as an Excel file. p-value and log2(FC) are reported in the “Feature Meta Data” sheet. The “Significant” column contains the value “1” if the protein is considered

Fig. 3 (a) Typical volcano plot obtained after differential analysis of samples from healthy subjects and patients. The red lines correspond to the p-value and fold-change filters applied. Black points correspond to proteins which are differentially expressed between the two conditions according to the analysis. (b) Example of a protein enriched in patient samples

Serum Biomarker Discovery using DIA Proteomics

147

significantly enriched/depleted in one condition compared to the other. Figure 3b shows typical results from a DIA-based biomarker discovery analysis.

4

Notes 1. Adapt the cartridge format to the peptide quantity. 2. Alternative sample preparation protocols can be used (in-solution or in-gel digestion). 3. If needed, the protocol can be stopped after digestion and samples stored frozen. 4. This step is crucial if samples are prepared by a method involving intensive biochemical prefractionation such as depletion. In this case, the final protein and peptide concentration will be difficult to derive from the initial concentration. 5. In some cases, initial column conditioning can be speeded up by injecting several high-concentration aliquots of a standard complex mixture (digest of cell lysate). 6. Adapt modifications list to reflect the sample preparation protocol if it differs from what we have suggested. 7. The second peptide option allows identification of more than one peptide in a single MS/MS spectrum. This option can be useful in shotgun analysis to increase proteome coverage but it can also lead to a larger proportion of false-positive results. In DIA, this effect can adversely affect the quality of results as it is directly linked to the quality of the spectral library. We therefore do not recommend using this option as we consider it preferable to construct a more accurate spectral library. 8. As synthetic peptides are spiked into samples to realign retention time between DDA and DIA runs, different gradient slopes and lengths may be used at each stage. However, the alignment will be better if the same LC method is used during both phases. 9. This step can be performed with Excel or with a tool like GOAT [35]. 10. Spectronaut can build libraries from Mascot, MaxQuant, Proteome Discoverer, and ProteinPilot search results, or from results generated by any other search engine using the Biognosys Generic Format. 11. Windows can be optimized either on the full m/z range (400–1600) or over a smaller m/z range. In the serum library we constructed, 90% of identified ions were in the 427–962 m/ z range. Limiting the m/z range allows DIA to be performed on smaller windows and increases specificity. Multiple DIA

148

Alexandra Kraut et al.

runs may also be performed on different m/z fractions to cover the full range, but at the cost of lower throughput and increased sample consumption. 12. This step can be performed with Excel or a tool like SwathTUNER [37]. 13. Spectronaut can use multiple libraries for a single DIA run or different libraries for each run. It depends on the experimental design chosen. For example, if fractionation is also performed on samples dedicated to DIA analysis, the corresponding fraction library can be assigned to each DIA run. 14. The experimental design can be saved and reused later for other processing via the import/export condition option. 15. Once parameter setting has been optimized, or to test different options, it is possible to create different DIA Analysis settings under the Spectronaut Settings section. 16. Different types of files can be exported, e.g., peptide export or normalization report. Once the Spectronaut Experiment file has been saved, it can be reloaded later if needed. 17. ProStaR is the web-based graphical user interface of DAPAR, a collection of tools and graphs dedicated to proteomics analysis. Both are available as R package through Bioconductor [43]. The website is a permanent link where the software can be tested before installation or used with limited datasets.

Acknowledgment This study was supported by grants from the “Investissement d’Avenir Infrastructures Nationales en Biologie et Sante´” program (ProFI project, ANR-10-INBS-08) and by the French National Research Agency in the framework of the “Investissements d’avenir” program (GRAL project, ANR-10-LABX-49-01 and LIFE project, ANR-15-IDEX-02). References 1. Parker CE, Borchers CH (2014) Mass spectrometry based biomarker discovery, verification, and validation—Quality assurance and control of protein biomarker assays. Mol Oncol 8:840–858. https://doi.org/10.1016/ j.molonc.2014.03.006 2. Liu H, Wang H, Hongmei Z et al (2018) Preliminary study of protein changes in trisomy 21 fetus by proteomics analysis in amniocyte.

Prenat Diagn 38(6):435–444. https://doi. org/10.1002/pd.5259 3. Rauniyar N, Yu X, Cantley J et al (2018) Quantification of urinary protein biomarkers of autosomal dominant polycystic kidney disease by parallel reaction monitoring. Proteomics Clin Appl 12(5):e1700157. https://doi.org/ 10.1002/prca.201700157 4. Preece RL, Han SYS, Bahn S (2018) Proteomic approaches to identify blood-based biomarkers

Serum Biomarker Discovery using DIA Proteomics for depression and bipolar disorders. Expert Rev Proteomics 15(4):325–340. https://doi. org/10.1080/14789450.2018.1444483 5. Atak A, Khurana S, Gollapalli K et al (2018) Quantitative mass spectrometry analysis reveals a panel of nine proteins as diagnostic markers for colon adenocarcinomas. Oncotarget 9:13530–13544. https://doi.org/10.18632/ oncotarget.24418 6. Geyer PE, Kulak NA, Pichler G et al (2016) Plasma proteome profiling to assess human health and disease. Cell Syst 2:185–195. https://doi.org/10.1016/j.cels.2016.02.015 7. Sandow JJ, Rainczuk A, Infusini G et al (2018) Discovery and validation of novel protein biomarkers in ovarian cancer patient urine. Proteomics Clin Appl 12(3):e1700135. https://doi. org/10.1002/prca.201700135 8. Hirao Y, Saito S, Fujinaka H et al (2018) Proteome profiling of diabetic mellitus patient urine for discovery of biomarkers by comprehensive MS-based proteomics. Proteomes 6. https://doi.org/10.3390/ proteomes6010009 9. Bostanci N, Selevsek N, Wolski W et al (2018) Targeted proteomics guided by label-free global proteome analysis in saliva reveal transition signatures from health to periodontal disease. Mol Cell Proteomics 17(7):1392–1409. https://doi.org/10.1074/mcp.RA118. 000718 10. Duriez E, Masselon CD, Mesmin C et al (2017) Large-scale SRM screen of urothelial bladder cancer candidate biomarkers in urine. J Proteome Res 16:1617–1631. https://doi. org/10.1021/acs.jproteome.6b00979 11. Anderson NL (2010) The clinical plasma proteome: a survey of clinical assays for proteins in plasma and serum. Clin Chem 56:177–185. https://doi.org/10.1373/clinchem.2009. 126706 12. Nahnsen S, Bielow C, Reinert K, Kohlbacher O (2013) Tools for label-free peptide quantification. Mol Cell Proteomics 12:549–556. https://doi.org/10.1074/mcp.R112.025163 13. Doll S, Dreßen M, Geyer PE et al (2017) Region and cell-type resolved quantitative proteomic map of the human heart. Nat Commun 8:1469. https://doi.org/10.1038/s41467017-01747-2 14. Song E, Gao Y, Wu C et al (2017) Targeted proteomic assays for quantitation of proteins identified by proteogenomic analysis of ovarian cancer. Sci Data 4:170091. https://doi.org/ 10.1038/sdata.2017.91 15. Gilquin B, Louwagie M, Jaquinod M et al (2017) Multiplex and accurate quantification

149

of acute kidney injury biomarker candidates in urine using protein standard absolute quantification (PSAQ) and targeted proteomics. Talanta 164:77–84. https://doi.org/10. 1016/j.talanta.2016.11.023 16. Venable JD, Dong M-Q, Wohlschlegel J et al (2004) Automated approach for quantitative analysis of complex peptide mixtures from tandem mass spectra. Nat Methods 1:39–45. https://doi.org/10.1038/nmeth705 17. Navarro P, Kuharev J, Gillet LC et al (2016) A multicenter study benchmarks software tools for label-free proteome quantification. Nat Biotechnol 34:1130–1136. https://doi.org/ 10.1038/nbt.3685 18. Ro¨st HL, Rosenberger G, Navarro P, et al (2014) OpenSWATH enables automated, targeted analysis of data-independent acquisition MS data. In: Nat. Biotechnol. https://www. nature.com/articles/nbt.2841. Accessed 12 Apr 2018 19. Egertson JD, MacLean B, Johnson R et al (2015) Multiplexed peptide analysis using data-independent acquisition and Skyline. Nat Protoc 10:887–903. https://doi.org/10. 1038/nprot.2015.055 20. Tsou C-C, Avtonomov D, Larsen B et al (2015) DIA-Umpire: comprehensive computational framework for data-independent acquisition proteomics. Nat Methods 12:258–264. https://doi.org/10.1038/ nmeth.3255 21. Li Y, Zhong C-Q, Xu X et al (2015) GroupDIA: analyzing multiple data-independent acquisition mass spectrometry data files. Nat Methods 12:1105–1106. https://doi.org/10. 1038/nmeth.3593 22. Bruderer R, Bernhardt OM, Gandhi T et al (2017) Optimization of experimental parameters in data-independent mass spectrometry significantly increases depth and reproducibility of results. Mol Cell Proteomics 16:2296–2309. https://doi.org/10.1074/ mcp.RA117.000314 23. Guo T, Kouvonen P, Koh CC et al (2015) Rapid mass spectrometric conversion of tissue biopsy samples into permanent quantitative digital proteome maps. Nat Med 21 (4):407–413. https://doi.org/10.1038/nm. 3807 24. Muntel J, Xuan Y, Berger ST et al (2015) Advancing urinary protein biomarker discovery by data-independent acquisition on a quadrupole-orbitrap mass spectrometer. J Proteome Res 14:4752–4762. https://doi.org/ 10.1021/acs.jproteome.5b00826

150

Alexandra Kraut et al.

25. Song Y, Zhong L, Zhou J et al (2017) Dataindependent acquisition-based quantitative proteomic analysis reveals potential biomarkers of kidney cancer. Proteomics Clin Appl 11. https://doi.org/10.1002/prca.201700066 ´ , Welinder C, Lindberg H et al 26. Ve´gva´ri A (2011) Biobank resources for future patient care: developments, principles and concepts. J Clin Bioinforma 1:24. https://doi.org/10. 1186/2043-9113-1-24 27. Anderson NL, Anderson NG (2002) The human plasma proteome: history, character, and diagnostic prospects. Mol Cell Proteomics 1:845–867 28. Keshishian H, Burgess MW, Gillette MA et al (2015) Multiplexed, quantitative workflow for sensitive biomarker discovery in plasma yields novel candidates for early myocardial injury. Mol Cell Proteomics 14:2375–2393. https:// doi.org/10.1074/mcp.M114.046813 29. Bellei E, Bergamini S, Monari E et al (2011) High-abundance proteins depletion for serum proteomic analysis: concomitant removal of non-targeted proteins. Amino Acids 40:145–156. https://doi.org/10.1007/ s00726-010-0628-x 30. Tu C, Rudnick PA, Martinez MY et al (2010) Depletion of abundant plasma proteins and limitations of plasma proteomics. J Proteome Res 9:4982–4991. https://doi.org/10.1021/ pr100646w 31. Lin L, Zheng J, Yu Q et al (2018) High throughput and accurate serum proteome profiling by integrated sample preparation technology and single-run data independent mass spectrometry analysis. J Proteomics 174:9–16. https://doi.org/10.1016/j.jprot. 2017.12.014 32. Nigjeh EN, Chen R, Brand RE et al (2017) Quantitative proteomics based on optimized data-independent acquisition in plasma analysis. J Proteome Res 16:665–676. https://doi. org/10.1021/acs.jproteome.6b00727 33. Cox J, Hein MY, Luber CA et al (2014) Accurate proteome-wide label-free quantification by delayed normalization and maximal peptide ratio extraction, termed MaxLFQ. Mol Cell Proteomics 13:2513–2526. https://doi.org/ 10.1074/mcp.M113.031591 34. Bruderer R, Bernhardt OM, Gandhi T et al (2015) Extending the limits of quantitative

proteome profiling with data-independent acquisition and application to acetaminophen treated 3D liver microtissues. Mol Cell Proteomics 14(5):1400–1410. https://doi.org/10. 1074/mcp.M114.044305 35. Trudgian DC, Fischer R, Guo X et al (2014) GOAT—a simple LC-MS/MS gradient optimization tool. Proteomics 14:1467–1471. https://doi.org/10.1002/pmic.201300524 36. Moruz L, Pichler P, Stranzl T et al (2013) Optimized nonlinear gradients for reversedphase liquid chromatography in shotgun proteomics. Anal Chem 85:7777–7785. https:// doi.org/10.1021/ac401145q 37. Zhang Y, Bilbao A, Bruderer T et al (2015) The use of variable Q1 isolation windows improves selectivity in LC-SWATH-MS acquisition. J Proteome Res 14:4359–4371. https://doi. org/10.1021/acs.jproteome.5b00543 38. Forshed J (2017) Experimental design in clinical ’Omics Biomarker Discovery. J Proteome Res 16:3954–3960. https://doi.org/10. 1021/acs.jproteome.7b00418 39. Wis´niewski JR, Zougman A, Nagaraj N, Mann M (2009) Universal sample preparation method for proteome analysis. Nat Methods 6:359–362. https://doi.org/10.1038/ nmeth.1322 40. Lebert D, Louwagie M, Goetze S et al (2015) DIGESTIF: a universal quality standard for the control of bottom-up proteomics experiments. J Proteome Res 14:787–803. https://doi.org/ 10.1021/pr500834z 41. Gundry RL, White MY, Murray CI et al (2009) Preparation of proteins and peptides for mass spectrometry analysis in a bottom-up proteomics workflow. Curr Protoc Mol Biol Chapter 10:Unit10.25. https://doi.org/10. 1002/0471142727.mb1025s88 42. Govaert E, Van Steendam K, Willems S et al (2017) Comparison of fractionation proteomics for local SWATH library building. Proteomics 17. https://doi.org/10.1002/pmic. 201700052 43. Wieczorek S, Combes F, Lazar C et al (2017) DAPAR & ProStaR: software to perform statistical analyses in quantitative discovery proteomics. Bioinformatics 33:135–136. https:// doi.org/10.1093/bioinformatics/btw580

Chapter 10 Discovering Protein Biomarkers from Clinical Peripheral Blood Mononuclear Cells Using Data-Independent Acquisition Mass Spectrometry Xin Ku and Wei Yan Abstract Global proteomics analyses are traditionally performed in data-dependent acquisition (DDA) mode, which results in inadequate reproducibility across large sample cohorts due to the under-sampling inherent to shotgun proteomics. Recently, data-independent acquisition (DIA) strategies were introduced to allow reproducible detection and quantification of thousands of proteins with consistent sensitivity across samples. Here, we present an approach to analyze changes to the protein network in human peripheral blood mononuclear cells (PBMCs) from clinical blood samples, using DIA as a unique platform for biomarker discovery. We describe how to generate spectral PBMC proteome libraries by applying peptide fractionation followed by DDA analysis, and then how to apply DIA methods to PBMC samples from individual patients using a high-resolution Orbitrap Fusion mass spectrometer. Key words Mass spectrometry, Peripheral blood mononuclear cell (PBMC), Data-independent acquisition, Label-free quantification, Biomarker

1

Introduction Mass spectrometry (MS)-based proteomics technology has emerged as an extremely valuable tool to address biological and medical questions raised by both basic and translational research [1–5]. As it can identify and quantify thousands of proteins and their associated networks in complex biological samples, this holistic approach is often used to detect protein changes between different physiological states and thus allow disease progression to be monitored [6, 7] with the aim of developing clinical applications in diagnosis, prognosis, therapeutic evaluation, etc. Owing to its powerful and incomparable capacity to profile proteins from complex clinical samples including numerous body fluids and tissues, MS has become the method of choice for biomarker development. The power of this technique facilitates the discovery and validation

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_10, © Springer Science+Business Media, LLC, part of Springer Nature 2019

151

152

Xin Ku and Wei Yan

of biomarkers that differentiate between healthy and disease states [8, 9]. Traditionally, MS-based proteomics mainly involves datadependent acquisition (DDA), as featured in the shotgun approach [10, 11]. In this approach, the mass spectrometer scans the peptide precursor ions (MS1) and the ion intensity determines which peptide precursors are selected for fragmentation to generate a tandem mass spectrum (MS2) pattern from which the selected peptides can be identified. The DDA-based shotgun approach has been successfully applied in many discovery-driven studies to identify and quantify a vast number of proteins for screening purposes [3, 5, 7, 9]. However, it has shown limited success in biomarker development studies, where highly reproducible results are required when working with complex clinical samples such as plasma/serum [12, 13]. Due to its relatively stochastic selection of precursor peptides, DDA inherently suffers from under-sampling and thus proteins may not be reproducibly detected across multiple samples. Given that the peptides selected for fragmentation are typically chosen based on the peptide ion intensities from high to low, the DDA approach is also biased against low-abundance proteins, which are often biologically important and thus constitute attractive biomarker candidates [14, 15]. To circumvent these disadvantages, targeted proteomics approaches—represented by select reaction monitoring (SRM) or multiple reaction monitoring (MRM)—have been developed. These methods focus on a predetermined list of proteins to obtain reproducible MS measurements [16, 17] and have become popular in biomarker studies over the past few years [18, 19]. However, the targeted analysis involved in SRM/MRM usually requires prior selection/optimization of the target peptides and their corresponding transitions, often making the development process laborious. Recently, data-independent acquisition (DIA) emerged as a novel and promising alternative to DDA methods and SRM/MRM [20, 21]. In DIA, continuous broad sequential acquisition/isolation mass-to-charge (m/z) windows (between 2 and 200 m/z) are defined to cover the whole mass range for all potential candidate peptides in the sample. All peptides falling into the predefined m/z window are fragmented and the resulting fragmentation spectra (MS2) are recorded by high-resolution mass spectrometers. Since recording of the MS2 spectra for each peptide identified is independent of its MS1 intensity, DIA allows unbiased measurement of all detectable analytes in a given sample. However, due to the enormous complexity of the MS2 spectra within the large m/z window and the loss of a clear link between the precursor mass and its fragment patterns, protein identification in DIA can be challenging. To overcome this difficulty, a spectral library strategy has been introduced whereby a library of spectral information including retention time, precursor m/z, and MS2 spectra for each detectable

Biomarker Discovery from PBMCs using DIA

153

peptide is generated from a comprehensive DDA analysis of similar (or the same) samples (e.g., a pool of multiple samples) [21]. The spectral library can then be used to match the chromatographic areas of a set of fragment ions acquired by DIA analysis of each individual sample, allowing identification and quantification of the peptides contained in the spectral library. Fractionation and/or enrichment strategies can be applied during the DDA analysis to produce a spectral library with deeper proteome coverage; the subsequent DIA analysis of individual samples will benefit from the increased protein coverage, and allow consistent identification of a significant number of proteins. Here, we present a method that we have applied to blood biomarker development for cancer diagnosis using a DIA approach. In addition to analyzing the plasma proteome of clinical samples, we specifically investigated the proteome of the PBMC from each clinical blood sample. We believe critical disease-correlated information can be obtained by profiling PBMC proteomes in clinical samples as these cells may reveal cross talk with disease progression, and thus be of benefit to cancer biomarker development. Cancer immune-therapy has rapidly progressed in the past few years and disease progression is well known to correlate with compromise of the human immune system [22, 23]. In addition to cancer, other diseases that could perturb the surveillance network, such as infections [24] and inflammation [25], can also be studied and monitored by examining changes to the PBMC protein network. Furthermore, this DIA-based proteomics method can be adapted and expanded to study cells from a range of sources, including surgical tissues or biopsy samples.

2

Materials All solutions were prepared with purified deionized water or HPLC-grade reagents. All solutions were stored at room temperature unless indicated otherwise. 1. EDTA K2 blood collection vacutainer (purple top) for blood collection. 2. 15-mL human lymphocyte separation tube. 3. Red blood cell (RBC) lysis buffer: 139.6 mM NH4Cl in 16.96 mM Tris-buffer (pH 7.3). 4. Lysis buffer (see Note 1): 50 mM HEPES buffer (pH ¼ 8) with 1 protease inhibitor cocktail (e.g., Roche; according to the manufacturer’s instruction) and 0.2% acid-labile surfactant (w/w) such as RapiGest (Waters Co.). 5. Ammonium bicarbonate (AMBIC) solution (freshly prepared): 60 mg AMBIC per 15 mL HPLC-grade water; final concentration 50 mM.

154

Xin Ku and Wei Yan

6. Urea solution (freshly prepared, see Note 2): weigh 8.1 g urea, and add HPLC-grade water to 15 mL to reach a final concentration of 9 M. 7. Tris(2-carboxyethyl)phosphine (TCEP) stock solution: weigh 100 mg TCEP, and add HPLC-grade water to 1 mL to reach a final concentration of 400 mM; store at 4  C. 8. Iodoacetamide (IAA) solution (freshly prepared in the dark): weigh 92.5 mg IAA, and add HPLC-grade water to 1 mL to reach a final concentration of 500 mM. 9. Sequence modified trypsin (0.5 μg/μL): store in 80  C freezer. 10. (Optional) iRT peptide standard kit (10): to calibrate retention times. Store at 4  C. 11. 96-well plate desalting cartridge or SPE columns (Waters Co.) 12. Equilibration buffer: 0.1% trifluoroacetic acid (TFA) in water. 13. Elution buffer: 50% acetonitrile in equilibration buffer. 14. High-pH fractionation solvents: Basic equilibration buffer: 5 mM ammonium hydroxide; Basic elution buffer: 90% acetonitrile, 5 mM ammonium hydroxide. 15. High-pH-compatible C18 reversed-phase column: 4.6 mm ID  250 mm. 16. High-performance liquid chromatography (HPLC) system with UV detector, pressure sensor, and fraction collector is recommended. 17. NanoLC solvents: A: 0.1% formic acid in water; B: acetonitrile in 0.1% formic acid in water. 18. Self-packed C18 analytical column: 3 μM particle size, 150 mm  75 μm. 19. Software for data processing: Proteome Discoverer or Maxquant (for spectral library construction) [26], and Skyline [27] or Spectronaut (for protein identification and quantification from DIA files) [28]. 20. NanoLC system compatible with pressure up to 300 bars, linked to MS. 21. High-resolution Orbitrap mass spectrometer capable of MS analysis, such as Q-Exactive or Fusion (Thermo Fisher). 22. 1.5-mL microtubes, sample vials, 50-mL centrifuge tubes, pipettes, pipette tips, and other disposable plastic labware 23. Personal protective equipment: gloves, gown, and safety goggles.

Biomarker Discovery from PBMCs using DIA

3

155

Methods The schematic workflow is illustrated in Fig. 1. 1. Collect blood in EDTA K2 blood collection vacutainer (purple top). Blood should be used within 4 h of collection to ensure maximum cell yield. The blood in the vacutainer should never be stored in the fridge. 2. To isolate PBMC, spin the human lymphocyte separation tube at 2000  g for 1 min at 20  C. Transfer 3–5 mL whole blood from the vacutainer to a 15-mL human lymphocyte separation tube and spin at 800  g for 15 min at 20  C. Adjust acceleration and braking profile to curve number 3 (curve number 1 is the slowest and curve number 9 the fastest). 3. Collect the PBMC layer from the plasma/separation liquid interface using a transfer pipette (aspirate as little plasma and separation liquid as possible) and transfer to a 15-mL

Fig. 1 Schematic workflow of the DIA protocol

156

Xin Ku and Wei Yan

centrifuge tube. Isolate plasma for separate storage. Add 10 mL PBS to PBMCs, spin cells at 200  g for 10 min at 20  C, and decant supernatant. 4. To specifically lyse red blood cells, add 10 mL of RBC lysis buffer, resuspend cells, and incubate with mixing by inversion for 5 min at RT. Spin cells once again at 200  g for 10 min at 4  C. Decant supernatant, resuspend cells in 1.5 mL PBS, and transfer to a 1.5-mL microfuge tube (take 20 μL to count cells, see Note 3). Finally, spin cells at 320  g for 10 min at 4  C, decant supernatant and store dry pellet at 80  C. 5. To extract proteins from PBMCs, add 0.5 mL lysis buffer to 106 PBMCs (see Note 3) and incubate cells on ice for 30 min. Centrifuge lysates at 20,000  g for 30 min at 4  C. Transfer supernatant to a clean tube and determine protein concentrations of the supernatant by standard BCA assay. 6. Denature the proteins at room temperature for 1 h with a final concentration of 6 M urea. Then, reduce the proteins with TCEP (5 mM, dilute 80 times from stock solution) at room temperature for 30 min. Add IAA (6.25 mM final concentration; dilute 80 times from freshly prepared solution) to alkylate proteins, incubate for 30 min at room temperature in the dark. Dilute the mixture with six volumes of 50 mM AMBIC buffer before digestion with sequence modified trypsin (1:100 w/w) for 12 h at 37  C. 7. Quench digestion by adding 1 μL phosphoric acid, and adjusting pH to 2–3. Activate the 96-well cartridge by rinsing with methanol three times, and equilibrate it by flushing three times with equilibration buffer (200 μL each time). Load the acidified peptide solution onto the cartridge (loading should be performed very slowly to achieve optimal interaction), and wash three times with equilibration buffer. Elute desalted peptides with elution buffer, and dry under vacuum (Speedvac). 8. To fractionate the peptide digest using high-pH reverse phase (see Note 4), add high-pH fractionation basic equilibration buffer to the dried peptides (the sample volume will depend on the column’s capacity). The final volume should be kept below the maximum volume of the injection loop. After injection of the basic peptide mixture, run a 120-min gradient (duration can be adjusted), varying the proportion of basic elution buffer stepwise between 0% and 100%. Collect fractions every 30 or 60 s. If a reduced number of fractions is desired for MS analysis, different parts of the gradient can be mixed so that the combined fractions remain orthogonal to downstream LC/MS analysis at low pH. Dry combined fractions under vacuum (Speedvac).

Biomarker Discovery from PBMCs using DIA

157

9. To perform liquid chromatography separation, dissolve the dried peptide fractions in nanoLC solvent A (0.5 μg/μL). To calibrate retention time, iRT peptides can be added prior to LC-MS/MS analysis, according to manufacturer’s instructions. 10. LC-ESI-MS/MS is performed by coupling a nanoLC to an Orbitrap Fusion mass spectrometer (also applicable for QE-plus and QE-HF). For each analysis, 1 μg of dissolved peptides is delivered to an analytical column and separated by applying an 80-min gradient from 7 to 35% of nanoLC solvent B at a flow-rate of 300 nL/min. 11. To generate the spectral library, perform tandem mass spectrometry (LC-MS/MS) analyses on an Orbitrap Fusion mass spectrometer operating in DDA mode and automatically switching between MS and MS/MS. Acquire full-scan MS spectra (400–1000 m/z) in the Orbitrap at 60,000 resolution (at m/z 400) after accumulation of precursor ions to a target value of 3  106 for a maximum of 20 ms. Apply internal lock mass calibration using the signal for the (Si(CH3)2O)6 H+ ion at m/z 445.120025; this ion is present in the air in the laboratory. Tandem mass spectra should be recorded for a maximum of 3 s by high-energy collision-induced dissociation (HCD, target value of 1  105, max 35 ms accumulation time) at a normalized collision energy of 30% in the Orbitrap. To maximize the number of precursors targeted for analysis, enable dynamic exclusion with one repeat count and 60 s exclusion time. 12. DIA analyses can also be performed using the Orbitrap Fusion (see Note 5). Choose tMS2 as scan type and set the isolation window to 21 m/z (1 m/z overlap between neighboring windows, see Note 6). The mass spectra are also acquired by HCD (target value of 1  105, max 60 ms accumulation time) at a normalized collision energy of 30% in the Orbitrap at 30,000 resolution (at 400 m/z), with the scan range set to 400–1000 m/z. Then import the mass list table where each mass number should correspond to the median of the corresponding isolation window (see Table 1 for a sample target mass list between 400 and 1000). This information can be easily calculated using Skyline (see Skyline manual for details). 13. To identify proteins from DDA files, generate peak lists from raw MS data files and use Proteome Discoverer software (Thermo Fisher; see Note 7) to search for them against a modified Uniprot Human Protein Database (latest version available on www.uniprot.org), to which the iRT peptide sequence (its fasta file can be downloaded from the vendor’s website) has been added if necessary. Perform searches with

158

Xin Ku and Wei Yan

Table 1 Example of calculated target mass values for equal-width (20 m/z) isolation windows Window number

Target mass (m/z)

Window number

Target mass (m/z)

Window number

Target mass (m/z)

1

411.4369

11

611.5279

21

811.6188

2

431.446

12

631.5369

22

831.6279

3

451.4551

13

651.546

23

851.637

4

471.4642

14

671.5551

24

871.6461

5

491.4733

15

691.5642

25

891.6552

6

511.4824

16

711.5733

26

911.6643

7

531.4915

17

731.5824

27

931.6734

8

551.5006

18

751.5915

28

951.6825

9

571.5097

19

771.6006

29

971.6916

10

591.5188

20

791.6097

30

991.7007

carbamidomethylation on cysteine residues defined as fixed modification and methionine oxidation as a variable modification. Specify trypsin as the proteolytic enzyme, and allow up to two missed cleavages. Set mass tolerance to 10 ppm for the precursor ions and 0.02 Da for the fragments. Filter all peptides at a high confidence level (or 1% FDR for other software), as calculated by the software. Export the MSF file (*.msf) and the protein annotation file (txt format, finishing with psms) from Proteome Discoverer. 14. To generate the spectral library, open Spectronaut software (Biognosys), choose “Prepare” tab and then click on “Generate Spectral Library from Proteome Discoverer” (bottom left), choose the exported *.msf file, assign the corresponding raw (*.raw) and protein annotation files. Adjust the parameters in the setup procedure if required (e.g., specify digest enzyme and proteome database), finally click “load” to generate a spectral library which will be available in “spectral libraries” when finished. 15. To identify and quantify proteins from DIA files, first convert each DIA raw file into a HTRMS file using HTRMS converter software which can be downloaded together with Spectronaut. After all files have been converted, open Spectronaut and go to “settings” tab, choose “protein database” tab, and go to lower left and click “import” to import the fasta file of the previously modified human protein database (with iRT sequence if appropriate). Then go to “review” tab and choose “Load Raw from file” to select all the DIA files for analysis. Return to the

Biomarker Discovery from PBMCs using DIA

159

“experiment setup” section. In this step, we first assign the previously constructed spectral library from “prepare perspective” and then specify the experimental conditions by clicking on “configure conditions.” In the condition editor, the number of sample-groups to be created must be specified, e.g., disease and control, and certain groups of samples must be identified as reference samples for the analysis. Fraction information should also be indicated if several fractions are derived from the same sample. In the lower panel of the experiment setup interface, go through the options from “data extraction” to “post analysis,” and choose the desired processing workflow and parameters (for a more detailed description of each parameter/workflow, see the software instruction manual which can be downloaded from the vendor’s website). Finally, click “start” to begin the analysis. 16. After the analysis is completed, go to the “post analysis” tab where numerous results can be graphically visualized for data interpretation. Quantification data can be exported from the “report” tab, by selecting the desired columns and applying the appropriate filters.

4

Notes 1. Many lysis buffers can be used; however, MS-incompatible detergents such as SDS would require additional cleanup steps prior to LC/MS analysis. 2. Urea can be difficult to dissolve at these high concentrations. A warm water bath (37  C) can be used to speed up the dissolution. However, hot water is not recommended as urea tends to generate cyanate at high temperatures, leading to protein/ peptide carbamylation. 3. This protocol should yield 106 PBMCs from 2–3 mL whole human blood. 4. High-pH fractionation of samples before LC-MS/MS analysis can be considered for spectral library generation, since it can provide higher proteome coverage in the library. 5. The number of acquired data points is a very important parameter to increase quantitative reproducibility of an elution peak. For an average peak elution time of 30 s, 9–10 data points are considered acceptable, and will limit cycle time to no more than 3 s. The cycle time (the length of time it takes to cycle through the entire target list) is highly influenced by the number of targets (isolation windows), the scan rate of the mass spectrometer, and the resolving power used. The maximum number of isolation windows can be calculated based on these parameters.

160

Xin Ku and Wei Yan

6. The isolation window can be of either equal or variable width. The variable width is usually calculated from peptide identification results from the DDA library, to generate equal numbers of peptide identifications in every window. Thus, the isolation window width in this case will be expanded/shortened, to account for regions where the peptides are expected to be less/more abundant in a certain m/z range. This approach will generate more windows with a smaller isolation width in the middle of the whole m/z range, which will reduce the complexity of the spectra and potentially allow easier interpretation. 7. Free software is also available to generate spectral libraries and extract ions from DIA files, such as Maxquant (http://www. maxquant.org), TransProteomicPipeline (http://tools.pro teomecenter.org/software.php), and Skyline (https://skyline. ms/project/home/begin.view). Instructions can be found on each developer’s website.

Acknowledgments This work was supported by the Chinese National Key Research and Development Programme (No. 2016YFC0902100), the Chinese National Natural Science Foundation (No. 21708024), Shanghai Sailing Programme (No. 16YF1406400), and Shanghai Jiao Tong University’s Medical Engineering Cross Fund (No. YG2016MS8014). References 1. Yates JR 3rd, Gilchrist A, Howell KE et al (2005) Proteomics of organelles and large cellular structures. Nat Rev Mol Cell Biol 6 (9):702–714. https://doi.org/10.1038/ nrm1711 2. Karr TL (2008) Application of proteomics to ecology and population biology. Heredity (Edinb) 100(2):200–206. https://doi.org/ 10.1038/sj.hdy.6801008 3. Liu Y, Huttenhain R, Collins B et al (2013) Mass spectrometric protein maps for biomarker discovery and clinical research. Expert Rev Mol Diagn 13(8):811–825. https://doi.org/10. 1586/14737159.2013.845089 4. Aebersold R, Mann M (2016) Massspectrometric exploration of proteome structure and function. Nature 537 (7620):347–355. https://doi.org/10.1038/ nature19949 5. Nomura DK, Dix MM, Cravatt BF (2010) Activity-based protein profiling for

biochemical pathway discovery in cancer. Nat Rev Cancer 10(9):630–638. https://doi.org/ 10.1038/nrc2901 6. Borrebaeck CA (2017) Precision diagnostics: moving towards protein biomarker signatures of clinical utility in cancer. Nat Rev Cancer 17 (3):199–204. https://doi.org/10.1038/nrc. 2016.153 7. Wulfkuhle JD, Liotta LA, Petricoin EF (2003) Proteomic applications for the early detection of cancer. Nat Rev Cancer 3(4):267–275. https://doi.org/10.1038/nrc1043 8. Pernemalm M, Lehtio J (2014) Mass spectrometry-based plasma proteomics: state of the art and future outlook. Expert Rev Proteomics 11(4):431–448. https://doi.org/10. 1586/14789450.2014.901157 9. Parker CE, Borchers CH (2014) Mass spectrometry based biomarker discovery, verification, and validation—quality assurance and control of protein biomarker assays. Mol

Biomarker Discovery from PBMCs using DIA Oncol 8(4):840–858. https://doi.org/10. 1016/j.molonc.2014.03.006 10. Zhang Y, Fonslow BR, Shan B et al (2013) Protein analysis by shotgun/bottom-up proteomics. Chem Rev 113(4):2343–2394. https:// doi.org/10.1021/cr3003533 11. Rauniyar N, Yates JR 3rd (2014) Isobaric labeling-based relative quantification in shotgun proteomics. J Proteome Res 13 (12):5293–5309. https://doi.org/10.1021/ pr500880b 12. Anderson NL, Anderson NG (2002) The human plasma proteome: history, character, and diagnostic prospects. Mol Cell Proteomics 1(11):845–867. https://doi.org/10.1074/ mcp.R200007-MCP200 13. Santini AC, Giovane G, Auletta A et al (2016) Translational research and plasma proteomic in cancer. J Cell Biochem 117(4):828–835. https://doi.org/10.1002/jcb.25413 14. Faria SS, Morris CF, Silva AR et al (2017) A timely shift from shotgun to targeted proteomics and how it can be groundbreaking for cancer research. Front Oncol 7:13. https:// doi.org/10.3389/fonc.2017.00013 15. Domon B, Aebersold R (2010) Options and considerations when selecting a quantitative proteomics strategy. Nat Biotechnol 28 (7):710–721. https://doi.org/10.1038/nbt. 1661 16. Ebhardt HA, Root A, Sander C et al (2015) Applications of targeted proteomics in systems biology and translational medicine. Proteomics 15(18):3193–3208. https://doi.org/10. 1002/pmic.201500004 17. Picotti P, Aebersold R (2012) Selected reaction monitoring-based proteomics: workflows, potential, pitfalls and future directions. Nat Methods 9(6):555–566. https://doi.org/10. 1038/nmeth.2015 18. Lassman ME, Fernandez-Metzler C (2014) Applications of low-flow LC-SRM for the analysis of large molecules in pharmaceutical R&D. Bioanalysis 6(13):1859–1867. https://doi. org/10.4155/bio.14.141 19. Surinova S, Schiess R, Huttenhain R et al (2011) On the development of plasma protein biomarkers. J Proteome Res 10(1):5–16. https://doi.org/10.1021/pr1008515 20. Anjo SI, Santa C, Manadas B (2017) SWATHMS as a tool for biomarker discovery: from

161

basic research to clinical applications. Proteomics 17(3–4). https://doi.org/10.1002/ pmic.201600278 21. Gillet LC, Navarro P, Tate S et al (2012) Targeted data extraction of the MS/MS spectra generated by data-independent acquisition: a new concept for consistent and accurate proteome analysis. Mol Cell Proteomics 11(6): O111 016717. https://doi.org/10.1074/ mcp.O111.016717 22. Ingersoll SB, Ahmad S, McGann HC et al (2015) Cellular therapy in combination with cytokines improves survival in a xenograft mouse model of ovarian cancer. Mol Cell Biochem 407(1–2):281–287. https://doi.org/ 10.1007/s11010-015-2475-2 23. Zelig U, Barlev E, Bar O et al (2015) Early detection of breast cancer using total biochemical analysis of peripheral blood components: a preliminary study. BMC Cancer 15:408. https://doi.org/10.1186/s12885-015-14147 24. Gupta P, Liu B, Wu JQ et al (2014) Genomewide mRNA and miRNA analysis of peripheral blood mononuclear cells (PBMC) reveals different miRNAs regulating HIV/HCV co-infection. Virology 450–451:336–349. https://doi.org/10.1016/j.virol.2013.12. 026 25. Lenna S, Assassi S, Farina GA et al (2015) The HLA-B*35 allele modulates ER stress, inflammation and proliferation in PBMCs from limited cutaneous systemic sclerosis patients. Arthritis Res Ther 17:363. https://doi.org/ 10.1186/s13075-015-0881-1 26. Tyanova S, Temu T, Cox J (2016) The MaxQuant computational platform for mass spectrometry-based shotgun proteomics. Nat Protoc 11(12):2301–2319. https://doi.org/ 10.1038/nprot.2016.136 27. Egertson JD, MacLean B, Johnson R et al (2015) Multiplexed peptide analysis using data-independent acquisition and Skyline. Nat Protoc 10(6):887–903. https://doi.org/10. 1038/nprot.2015.055 28. Bruderer R, Bernhardt OM, Gandhi T et al (2016) High-precision iRT prediction in the targeted analysis of data-independent acquisition and its impact on identification and quantitation. Proteomics 16(15–16):2246–2256. https://doi.org/10.1002/pmic.201500488

Chapter 11 Intact Protein Analysis by LC-MS for Characterizing Biomarkers in Cerebrospinal Fluid Je´roˆme Vialaret, Sylvain Lehmann, and Christophe Hirtz Abstract In the field of proteomics, the emerging “top-down” MS-based proteomics approach can be used to obtain a bird’s eye view of all intact proteoforms present in a sample. This alternative to the “bottom-up” approach based on tryptic protein digestion processes has some unique advantages for assessing PTMs and sequence variations. However, it requires some dedicated tools for sample preparation and LC-MS analysis, which makes it more complex to handle than the bottom-up approach. In this study, a simple methodology is presented for characterizing intact proteins in biological fluid. This method yields quantitative information using an MS1 profiling approach and makes it possible to identify the proteins regulated under various clinical conditions. Key words Biomarker, Top-down, LC-MS, Cerebrospinal fluid, Protein precipitation, SPE

1

Introduction Cerebrospinal fluid (CSF) is a colorless biological fluid which is directly in contact with the brain. Thanks to this specificity, CSF is a potential indicator of abnormal states of the central nervous system such as inflammation, infection, neurodegeneration, and tumor growth [1, 2]. CSF protein monitoring is not only useful for diagnostic and prognostic purposes, but it also provides valuable information about the pathophysiology of these conditions. In previous clinical Mass Spectrometry Proteomics (cMSP) based studies [3, 4], two main strategies, the “bottom-up” and “top-down” strategies, were presented. To date, the bottom-up approach has been the most frequently used strategy in the field of cMSP, where it serves to identify thousands of proteins in a single experiment [5]. With the bottom-up approach, proteins from the sample of interest are digested into peptides, mostly using trypsin, and analyzed by Liquid Chromatography-Mass Spectrometry (LC-MS). Although the bottom-up strategy is a powerful tool which has been widely used

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_11, © Springer Science+Business Media, LLC, part of Springer Nature 2019

163

164

Je´roˆme Vialaret et al.

in medical test laboratories, it has some major shortcomings. The proteins produced from a single gene can vary considerably in their molecular structure, in terms of genetic variations, splice variants, RNA edits, and posttranslational modifications (PTMs), giving rise to proteins of many forms, known as proteoforms [6]. Depending on the rates of protein sequence coverage by identified peptides, some ambiguity always persists as to the origin of the redundant peptide sequences detected. It has therefore become necessary to find means of identifying the proteoforms present in biological fluids in order to understand the pathological processes involved, since proteoforms often play dramatically different functions [7]. As an alternative to the bottom-up approach, the top-down approach (involving no proteolytic digestion step) now makes it possible to identify and characterize intact proteoforms using highresolution MS instruments [8]. In this study, a simple and efficient method for analyzing proteoforms from CSF was developed. The sample preparation protocol includes an acidic protein precipitation step followed by a protein cleanup step prior to the injection of the sample into an LC-MS system. For this top-down approach, a LC-MS system was specially tuned and an appropriate software program was used for protein identification to obtain an efficient and robust cMSP workflow. The potential of this top-down approach was previously demonstrated using CSF samples from 30 patients [8]. In the latter study, patients’ samples were subdivided into three different clinical groups involving neurodegenerative disorders. This robust and quantitative top-down cMSP method was used to identify several proteoforms known to be specifically present in Alzheimer’s disease (AD). These proteoforms, which originate mainly from three canonical protein families (i.e., clusterin, secretogranin-2, and chromogranin-A) have been widely used as biomarkers of AD and other neurodegenerative disorders [8].

2

Materials All solutions used in this study were prepared with ultrapure water with a resistivity of 18 MΩ.cm at 25
C.

2.1

Samples

1. Control CSF samples were obtained from patients who had undergone a lumbar puncture for headaches or memory complaints and whose etiological findings turned out to be negative. The samples were collected in accordance with protocols approved by the relevant Ethics Committee, and informed consent forms were signed by all patients, in line with the Helsinki Declaration. In all these samples, cytology was normal as well as protein, glucose, amyloid peptide 1–42, tau, and phospho-tau protein contents.

Towards Top-Down Approaches for Biomakers Discovery in CSF

165

2. 10-mL polypropylene tubes (Tube 10 mL, 92/15,3, PP, round (Sarstedt)) (see Note 1). 3. 1.5-mL low-adsorption polypropylene tubes (LoBind, colorless, Eppendorf). 2.2 Protein Precipitation

2.3 Solid Phase Extraction

1. 70% perchloric acid. 2. 1.5-mL low-adsorption polypropylene tubes (LoBind, colorless, Eppendorf). This step must be performed under a fume hood because of the organic and acidic vapors release. 1. Collection plate: a Deepwell Plate 96/2000 μL with clear wells, 2000 μL, PCR clean, white. 2. Elution plate: an Eppendorf twin.tec PCR Plate 96 LoBind, skirted, PCR clean, colorless (Eppendorf) (see Note 2). 3. Solvent A: 100% methanol ULC-MS grade. 4. Solvent B: 0.1% trifluoroacetic acid (TFA). To obtain solvent B, mix 9 mL of ultrapure water and 1 mL of 10% TFA. 5. 99.0% TFA LC-MS grade (see Note 3). 6. Dilute TFA giving a 10% solution by mixing 9 mL of ultrapure water and 1 mL of TFA 99.0% (see Note 4). 7. Solvent C: 35% acetonitrile/65% water with 0.1% TFA. Acetonitrile solvent is ULC-MS grade. To obtain solvent C, 3.5 mL of acetonitrile is mixed with 6.5 mL of 0.1% TFA. 8. Oasis HLB 96-well μElution Plate, 2 mg Sorbent per Well, 30 μm Particle Size (Waters).

2.4

Sample Drying

2.5 Sample Preparation for Liquid Chromatography

Equilibration plate: an Eppendorf twin.tec PCR Plate 96 LoBind, skirted, PCR clean, colorless (see Note 5). 1. Phase A: ultrapure water with 1% Formic acid, both of ULC-MS grade. To obtain 1 L of phase A, mix 10 mL pure formic acid to 990 mL water. 2. LC-MS vials (crimp/snap top polypropylene, 250 μL vial, Agilent).

2.6

LC-MS Analysis

All the chemicals mentioned in this section must be HPLC-MS grade. 1. QC HSA: dilute 100 amol/μL of digested HSA to 1 fmol/μL with BSA dilution solution. 2. Digest of BSA (Bovine Serum Albumin) is provided by Agilent. Resuspend the dry digest at 1 pmol/μL with 500 μL of 15% acetonitrile/85% water containing 0.1% formic acid. Dilute this solution 1/100 with the same buffer (15% acetonitrile/85%

166

Je´roˆme Vialaret et al.

water containing 0.1% formic acid) to obtain 10 fmol/μL of “BSA dilution solution.” 3. Digest of HSA (Human Serum Albumin) is provided by Agilent. Resuspend dry peptides with 50 μL of BSA dilution solution to obtain HSA at a concentration of 10 pmol/μL. Make a serial dilution of HSA solution with “BSA dilution solution” to obtain a final sample of 100 amol/μL of HSA dilute in 1 fmol/μL of BSA digest. 4. Phase A: ultrapure water with 1% formic acid. Prepare 1 L by adding 10 mL of pure formic acid and 990 mL of water. 5. Phase B: acetonitrile with 0.8% formic acid. Prepare 1 L by adding 8 mL of pure formic acid and 992 mL of acetonitrile. 6. The liquid chromatography system used consists of a pre-column (Acclaim PepMap300, 5 μm, 300 A˚ Wide Pore, 300 μm  5 mm C4 cartridge, Thermo Fischer Scientific) and a column (Proswift RP-4H 50 cm  100 μm monolithic column, Thermo Fisher Scientific). 7. Lockmass calibrant: Hexakis (1H, 4H-hexafluorobutyloxy)phosphazine (Agilent). 2.7

Data Analysis

1H,

1. Compass DataAnalysis 4.4 software for raw data processing (Bruker Daltonics). 2. Profile Analysis 2.1™ for statistical analyses (Bruker Daltonics). 3. Skyline software v3.3 (Skyline Targeted Mass Spec Environment, MacCoss Lab, Seattle, WA, USA). 4. Byonic™ (ProteinMetrics). 5. BioTools 3.2 (Bruker Daltonics).

2.8 Equipment and Supplies

1. A refrigerated centrifuge for 10 mL tubes. 2. A vortex mixer. 3. A refrigerated centrifuge for 1.5 mL tubes. 4. A fume hood. 5. An extraction plate manifold connected to a vacuum pump. 6. Acid-resistant CentriVap Vacuum concentrators. 7. A vibrating platform shaker—Titramax 1000 (Heidolph Instruments, Schwabach, Germany). 8. An Eppendorf 5804 centrifuge for plates. 9. An LC-MS system: nanoElute LC system (Bruker Daltonics) combined to Impact II benchtop UHR-Q-TOF mass spectrometer (Bruker Daltonics).

Towards Top-Down Approaches for Biomakers Discovery in CSF

3 3.1

167

Method Samples

1. Lumbar puncture (LP) is performed under standardized conditions at the end of the morning. LP is carried out at the L3/L4 or L4/L5 interspace after ruling out the occurrence of any potential contraindications. 2. Collect CSF samples directly in polypropylene tubes to prevent variations in the adsorption of biomarkers to the container surface (see Note 1). 3. Transfer the sample tubes on ice to the biochemical laboratory within 4 h. 4. Centrifuge the CSF samples at 1000  g for 10 min at 4
C. 5. Transfer the supernatant and aliquot into 1.5 mL LoBind polypropylene tubes. 6. Store in biobank at 80
C.

3.2 Protein Precipitation

1. Unfreeze 500 μL of CSF sample on ice. 2. Acidify the sample by adding 25 μL of 70% perchloric acid. 3. Vortex for 10 s at maximum speed. 4. Wait 15 min on ice (see Note 6). 5. Centrifuge at 1600  g for 15 min at 4
C. 6. Carefully collect the supernatant and transfer on a new 1.5 mL polypropylene tubes (see Note 7).

3.3 Solid Phase Extraction

This step must be performed under a fume hood because of the toxic organic and acidic vapor release. SPE steps are performed using an extraction plate manifold connected to a vacuum pump. To aspirate the liquid, the depression is gradually increased to 55 kPa (see Note 8). The liquid is collected below the SPE plate in the collection plate, except during the elution step. 1. Wash the phase with 300 μL of solvent A. Aspirate. 2. Condition the phase with 500 μL of solvent B. Aspirate. 3. Load the sample by transferring all the sample volume into the well. Wait 2 min before and aspirate gently for 30 s until no liquid passes through (see Note 9). 4. Wash the sample with 500 μL of solvent B. Aspirate. 5. Repeat step 4. 6. Change the collection plate by the elution plate below the SPE plate. 7. Elute the cleaned sample with 100 μL of solvent C. Wait 2 min before aspirating it gently for 30 s until all the liquid passes through (see Note 9).

168

3.4

Je´roˆme Vialaret et al.

Sample Drying

1. Switch on the concentrator system. 2. Put the Elution plate in the centrifugal unit without covering it. Equilibrate with an equivalent plate. 3. Select the centrifugal program corresponding to 30
C for 1 h (see Notes 10 and 11). 4. Switch on the centrifugal unit and open the depression valve.

3.5 Sample Preparation for Liquid Chromatography

1. Dried sample is resuspended by adding 25 μL of phase A. 2. Cover the plate and stir for 15 min at 800 revolutions per minute (rpm) on a vibrating platform shaker. 3. Transfer the liquid into a LC-MS vial (see Note 12).

3.6

LC-MS Analysis

1. Remove gas from solvents A and B. Purge the pump. Run two blank runs to equilibrate all the fluidic parts of the LC system. 2. LC-MS performances are checked with the QC HSA which is injected prior to sample analysis. 3. Inject 2 μL of the sample into the chromatographic system in a 180 min LC run (Table 1).

Table 1 LC and autosampler parameters Autosampler

Tray temperature Injection mode Volume Bottom sense

4
C μL Pickup 2 μL Yes

Trapping system

Pre-column Equilibration on trap column

300 μm  5 mm C4 cartridge 10 column volumes

Chromatographic system

Column

Proswift RP-4H 50 cm  100 μm 4 column volumes

Separation column equilibration Column temperature Flow rate Gradient

40
C 1 μL/min Time (min) 0 5 15 125 133 142 145 158 159 180

Phase B (%) 5 5 9 35 40 60 95 95 5 5

Towards Top-Down Approaches for Biomakers Discovery in CSF

169

Table 2 ESI source and MS parameters ESI source

Source NanoBooster NanoBooster gas pressure Drying gas Capillary

CaptiveSpray Acetonitrile ULCMS grade 0.2 Bar 5 L/min at 180
C 1200 V

MS parameters

Mode MS scan rate Mass range Lockmass Transfer

Positive 1 Hz 300–4000 m/z 1221.99 m/z isCID Hexapole RF Collision RF Retention time tolerance Exclude MS/MS rate for a precursor of 15,000 cts

Collision cell MS/MS mode: SPL

40 eV 600 Vpp 3500 Vpp 2 min After 3 spectra 0.33 Hz 1 Hz

4. Mass spectrometer operates in MS1 scan mode (Table 2) (see Note 13). 5. After MS1 profiling comparisons, regulated MS peaks are fragmented in additional LC-MS injections operating in the Scheduled Precursor List (SPL) mode (Table 2) (see Note 14). 3.7

Data Analysis

1. To process raw data, proceed as follows: (a) Start raw data processing with DataAnalysis software, starting with MS data re-calibration with the lockmass before signal extraction step . (b) Evaluate LC-MS performances using processed QC HSA with a dedicated skyline document. Check the peak profile at m/z ¼ 575.31 in terms of peak shape, peak intensity at MS and MS/MS level, retention time, and MS precision. 2. Process the MS1 profile: (a) Process top-down raw data with a dedicated automatized method provided by Bruker Daltonics. After MS re-calibration, the protein signal is extracted based on the chromatographic elution profile with the Dissect™ algorithm in order to also remove MS background noise. (b) Once the data had been monoisotopically resolved, deconvolute the data with SNAP™. When monoisotopic resolution is not performed, the data can be deconvoluted with MaxEnt. The deconvoluted data are exported, including the monoisotopic masses with the corresponding retention times and intensities.

170

Je´roˆme Vialaret et al.

Table 3 Byonic™ search parameters MS tolerance

10 ppm

MS/MS tolerance 30 ppm Protein database

Full human proteome Full human proteome with known proteolytic products

Variable modifications

Pyro-Glu N-terminus Deamidated asparagine Phosphorylations on S/T/Y O-linked glycosylation (HexNAc, HexNAc(2), HexNAc-Hex, HexNAc-HexNeuAc, HexNAc-Hex-NeuAc(2), HexNAc-Hex-NeuAc(3))

3. Statistical analyses (using Profile Analysis software) are performed by first generating “bucket” tables based on the retention times and intensities of the deconvoluted masses. Similar compounds correspond to the same mass (tolerance of 2 ppm) and retention times (tolerance of 2 min) in all analyses. Values for each compound have to be present in at least 60% of each group analyzed. Missing values are replaced by the average value of the bucket in the relevant class. Intensity values are normalized with a quantile normalization algorithm. Discriminating groups of compounds are detected by performing with a final Student’s t-test ( p-value precursor to 6 ions, always include N-terminal to proline ions, exclude product ions within 20 m/z window around m/z of precursor ion. At this stage, you might want to apply stricter rules to peptides which have already been observed experimentally in order to reduce the number of transitions to monitor (and instrument time). In our example, we had experimental MS/MS data (Subheading 3.5.1) for 1351 of the 1901 synthetic peptides. For these peptides, we selected only the precursors (charge 2+ or 3+ or both) which had been experimentally observed previously, thereby drastically reducing the number of transitions to screen. For the peptides which had not been observed experimentally (n ¼ 523), we screened both the 2+ and the 3+ precursors. We tolerated a maximum of 200 transitions per screening method and kept all peptides belonging to a given protein in the same method. Export the

Multiplexed SRM for Biomarker Discovery in FFPE Tissues

193

transition lists from Skyline and import them into the mass spectrometer software (in this case Xcalibur). In our example, this led to 124 unscheduled instrument methods to screen the >19,000 transitions corresponding to the 1901 heavy-labeled peptides in the synthetic peptide mix we ordered. Dissolve an aliquot of the synthetic peptides in mobile phase C and dilute to reach a final concentration of 20 fmol/μL per peptide. Analyze repeatedly using each of the screening methods. In this example, we injected 5 μL of the peptide solution on column (corresponding to approximately 100 fmol of each synthetic peptide) for SRM analysis. The analytical separation was run using a 2 h gradient as follows: 2–30% mobile phase D in 90 min, 30–60% mobile phase D in 6 min, and 60–80% mobile phase D in 2 min at a flow rate of 250 nL/min. The spray voltage was set to 2600 V, ion transfer capillary temperature was set to 240  C, resolution was set to 0.7 Th for Q1 and Q3 (FWHM), and the collision gas pressure was set at 1.5 mTorr. In order to ensure sufficient quality of the chromatographic peaks, cycle time was set to 2.0 s and a maximum of 200 concomitant transitions was allowed, thereby ensuring a minimum dwell time of 10 ms per transition. Collision energy was set individually for each transition using the following linear equation embedded in Skyline: CE ¼ slope
(precursor m/z) + intercept. This equation is instrument specific and each charge state of the precursor, namely 2+ and 3+, is allowed to have a different equation [26]. Then, perform data review to remove precursor ions with poor signal. As mentioned previously, we aimed to have three peptides per protein and three transitions per peptide in the final SRM assay for optimal specificity [27] (see Note 11). After SRM analysis, upload the generated raw files in Skyline and start by restricting the number of transitions to the five highest ranks (based on intensity) for each precursor. Then, visually inspect each peptide and use the dotp score to perform a first selection of precursor ions based on the quality of the SRM signal (see Note 12). An optimal SRM signal is characterized by co-elution of all monitored transitions and by a dotp score close to one (the dotp score represents the closeness of the match between the SRM data and the peptide library) (Fig. 1a). Precursor ions were removed in the following cases: low intensity signal (for one or both precursor ions) (Fig. 1b), data was acquired for both precursor ions (2+ and 3+) of a peptide and one showed a significantly lower intensity (Fig. 1c), irregular peak shape (even though the intensity might be higher than that of the second precursor ion) (Fig. 1d). In rarer cases, we also removed any precursor where inconsistencies were observed in the fragment profile (see Note 13) as well as any precursor which had less than three valid transitions as this would compromise specificity.

194

A

Carine Steiner et al.

y8 - 904.5238+ (heavy) y5 - 549.3131+ (heavy)

y7 - 775.4813+ (heavy) y9 - 516.7948++ (heavy)

y6 - 662.3972+ (heavy)

400

C

y11 - 1245.7917+ (heavy) y7 - 863.5337+ (heavy)

y10 - 1174.7546+ (heavy) y6 - 762.4860+ (heavy)

Precursor charge: +++

y9 - 1061.6705+ (heavy) y11 - 623.3995++ (heavy)

y9 - 1061.6705+ (heavy)

6000

Precursor charge: ++

y5 - 649.4019+ (heavy)

Precursor charge: +++ 120

57.5

5000

63.8

y6 - 762.4860+ (heavy) y9 - 531.3389++ (heavy)

140

57.5

DISCARD

300

KEEP

200

Intensity (10^3)

Intensity

Intensity (10^3)

100 4000

3000

80

60

2000 40

100 1000

0 62.5

63.0

63.5

64.0

64.5

65.0

20

0

65.5

57.0

Retention Time

B

y9 - 1212.6036+ (heavy) y6 - 875.4285+ (heavy)

57.5

58.0

58.5

y8 - 1125.5715+ (heavy) y12 - 779.3898++ (heavy)

y7 - 1012.4875+ (heavy)

D

57.0

57.5

y9 - 1077.6079+ (heavy) y5 - 663.3812+ (heavy)

y8 - 930.5395+ (heavy) y8 - 465.7734++ (heavy)

y7 - 833.4867+ (heavy)

58.5

y8 - 930.5395+ (heavy) y9 - 539.3076+ (heavy)

y7 - 833.4867+ (heavy) y8 - 465.7734++ (heavy)

y5 - 663.3812+ (heavy)

1.8

Precursor charge: ++

Precursor charge: +++ 1.6

102.0

74.9 (dotp 0.68)

75.0

40

1400

58.0

Retention Time

50

Precursor charge: +++

1.4

DISCARD 1200 1000 800 600 400

Intensity (10^6)

1.2

Intensity (10^3)

Intensity

0

Retention Time

1800 1600

KEEP

30

20

KEEP

DISCARD 1.0 0.8 0.6 0.4

10

200

0.2

0 101

102

Retention Time

103

0

0.0 73.5

74.0

74.5

75.5

Retention Time

75.5

76.0

73.5

74.0

74.5

75.0

75.5

76.0

Retention Time

Fig. 1 Selection of precursor ions. (a) Precursor ion with optimal signal. (b) Precursor was discarded due to a low intensity signal. (c) Precursor ion with lowest signal was discarded due to a significantly lower signal than second precursor ion. (d) Precursor ion with highest signal was discarded due to an unstable signal

3.5.3 Select Peptides with Low Analytical Interference When Spiked in a Biological Matrix

In this experiment, spike the heavy-labeled peptides in a relevant biological matrix and screen both the heavy and the light transitions in order to further select peptides with low analytical interference. For this, first prepare a relevant biological background for screening experiments. We developed this SRM assay to ultimately study breast cancer; however, for ethical reasons, we could not justify using clinical samples collected for this study as a background to screen peptides. We therefore used cell lines, which we suspended and fixed in 5% buffered formalin solution at pH 7.4 for 24 h in order to be as close as possible to FFPE clinical samples. After washing the pellets with TrisCl 20 mM, pH 8.5 buffer, the peptides were extracted using the same procedure as described previously [10]. The extracts were reconstituted in 50 μL of mobile

Multiplexed SRM for Biomarker Discovery in FFPE Tissues

195

phase C and an estimation of peptide amount was performed using the Pierce BCA Protein Assay Kit (Thermo Fischer Scientific). Depending on which type of samples you wish to investigate with your SRM assay, choose a suitable biological system to be used for screening purposes. In this example, we used five breast cancer cell lines. Then, generate scheduled instrument methods for a second screening of peptides. The second screening is meant to remove transitions which are subject to interferences or signal suppression by the sample matrix. As a starting point for this experiment, use the Skyline document generated in previous step. Add the corresponding light transitions to the already listed heavy transitions. At this stage, we also started using the indexed retention time (iRT) option in Skyline [28] (see Note 14). Using the retention times measured in the first screening experiment, we generated scheduled methods for the second screening, which reduces instrument time. Export transition lists from Skyline using the scheduling option with 10 min acquisition window and a maximum of 100 concurrent transitions in order to maximize data quality. Import transition lists in the mass spectrometer software to generate scheduled instrument methods (in our case, this led to 35 methods for the second screening). Based on peptide amount estimation, mix the peptide extracts from cell lines in equal proportions and spike with synthetic peptides in order to reach a concentration of approximately 160 ng/μL of background peptides and 20 fmol/μL of each synthetic peptide. Analyze repeatedly by injecting 5 μL (corresponding to 800 ng of background peptides and 100 fmol of each synthetic peptide on column) and using the SRM conditions described above. Perform data review to select peptides/transitions least affected by biological matrix. For this, upload the SRM raw files in Skyline and review data with the aim to select the peptides performing best in the presence of the sample matrix. Visually inspect each peptide and select candidates which display distinct and consistent signals (Fig. 2). For this step, we selected the four peptides per protein with the most consistent signal by considering the following criteria (in order of importance): 1. Similarity of fragment pattern with data from first screening. 2. Presence of a signal for the endogenous peptide which co-elutes with the heavy peptide and shows the same fragment ion pattern. 3. High intensity of heavy peptide. 4. If possible choose peptides which are distributed across the entire chromatographic run (see Note 15).

196

Carine Steiner et al.

A

First screening

Second screening

(in absence of background peptides)

(in presence of background peptides)

TNBC_screening_heavy_01

TNBC_screening_02

y8 - 700.3612+ (heavy) y5 - 473.2342+ (heavy) y10 - 437.7163++ (heavy)

y6 - 530.2557+ (heavy) y4 - 416.2127+ (heavy)

y8 - 700.3612+ (heavy) y5 - 473.2342+ (heavy) y10 - 437.7163++ (heavy)

y6 - 530.2557+ (heavy) y4 - 416.2127+ (heavy)

250 10

54.3 56.2

200

B 150

Intensity

Intensity (10^3)

8

6

Second screening (in presence of background peptides)

TNBC_screening_heavy_01

100

4

First screening (in absence of background peptides)

TNBC_screening_2

y9 - 1129.5009+ (heavy) y6 - 854.3739+ (heavy) y12 - 730.3128++ (heavy)

6000

y9 - 1129.5009+ (heavy) y6 - 854.3739+ (heavy) y12 - 730.3128++ (heavy)

y7 - 1001.4423+ (heavy) y5 - 717.3150+ (heavy)

1200

y7 - 1001.4423+ (heavy) y5 - 717.3150+ (heavy)

DISCARD 49.3 (dotp 0.77)

5000

50

2

1000

800

4000 0 57.0

53.8

54.0

54.2

54.4

54.6

54.8

Intensity

56.5

Retention Time

Retention Time

TNBC_screening_heavy_01

TNBC_screening_02

y8 - 700.3612+ (heavy) y5 - 473.2342+ (heavy) y10 - 437.7163++ (heavy)

y8 - 700.3612+ (heavy) y5 - 473.2342+ (heavy) y10 - 437.7163++ (heavy)

y6 - 530.2557+ (heavy) y4 - 416.2127+ (heavy)

y6 - 530.2557+ (heavy) y4 - 416.2127+ (heavy)

250 10

Intensity

0 56.0

55.5

3000

53.0 (dotp 0.83)

600

2000

400

1000

200

54.3 56.2

200

DISCARD

8 0 48.0

48.5

49.0

49.5 Retention Time

50.0

50.5

42

44

46

48

50

52

54

56

Retention Time

Intensity

Intensity (10^3)

0 150 6

100 4

50

2

0

0 50

52

54

56

58

60

62

64

Retention Time

50

52

54

56

58

60

62

64

Retention Time

Fig. 2 Selection of peptides in the presence of sample background. (a) Example of a peptide where the fragment pattern is different between both screening experiments, indicating that the assigned peak is incorrect. No other peak shows a better match in the unzoomed window, indicating that the original signal is probably suppressed in the presence of background peptides. (b) Example of a peptide with a low signal in the first screening experiment and no distinguishable signal in the second screening experiment

At this stage, keep only one precursor ion for each peptide and reduce the number of transitions to the three most intense based on the signal of the heavy-labeled peptide. 3.5.4 Selecting Peptides with Best Linear Response

Another option to select peptides is based on the linearity of their response. Using the Skyline document created in previous section, export transition lists to create new scheduling methods. In this case, we created five scheduled SRM methods with 7 min acquisition windows and a maximum of 200 concurrent transitions. Based on the peptide amount estimation, mix the cell extract background peptides in equal proportions and spike with synthetic peptides using sequential (1:1) dilutions. In this example, the 14 dilutions ranged from 20 fmol/μL (100 fmol on column) per synthetic peptide to approximately 2.4 amol/μL (12 amol on column). The amount of background peptides was 200 ng/μL (1 μg on column) for all dilutions. The injection volume was 5 μL and

Multiplexed SRM for Biomarker Discovery in FFPE Tissues

C

A

Y axis: peak area

197

KEEP

B

KEEP

D

KEEP

DISCARD

X axis: amount heavy peptide on column (light peptide is constant)

Fig. 3 Selection of peptides based on linearity performance. (a) Peptide with optimal linear response. (b) Peptide with intermediate linear response. (c) Peptide with intermediate linear response. (d) Peptide with insufficient linear response

SRM analysis was performed as described above. Start the experiment by analyzing the lowest dilution in order to minimize potential carry-over effects. Import the generated data into Skyline and visually check that all peaks are correctly assigned (see Note 16). After review, export the peak area corresponding to the sum of the three fragments to excel and import in a data visualization software such as Spotfire for further processing (see Note 17). In order to help the decision process, we represented both the heavy and the light peptide on the same graph (see Note 18). We selected the three best performing peptides for each protein based on the following criteria (in order of importance) (Fig. 3): 1. Length of linear domain for the heavy peptide. 2. Intensity of signal for the endogenous (light) peptide. 3. Reproducibility of signal for the endogenous peptide signal. 3.5.5 Final Steps

Using the scheduling tool in Skyline set the width of the scheduling window for the final SRM assay according to the maximum of tolerated concurrent transitions (Fig. 4). The latter is constrained by the dwell time (10 ms in our case) required for sufficient sensitivity and by the need to have at least 8 measurement points across the chromatographic peak to ensure reliable peak integration. In this example, setting the scheduling window to a narrow 2 min still led to having more than 200 concurrent transitions at some points during analysis. We considered 2 min acquisition windows too tight

198

Carine Steiner et al.

Fig. 4 Number of concurrent transitions at any given point in the method as a function of the size of the scheduling window. In this example, setting the scheduling window to 2 min still led to more than 200 concurrent transitions at several points during the analysis

due to the complex nature of the human samples analyzed and the small shifts in retention time which often occur. We therefore set the scheduling window to a more comfortable 3 min and distributed the transitions across two assays using the “multiple methods” option in Skyline. We further used the iRT peptides in the final methods and additionally used the intelligent SRM (iSRM) feature available on the TSQ Vantage instrument [29], which allows to perform dynamic scheduling (compensates for retention time shifts by automatically adjusting acquisition windows for all peptides based on the retention time of the iRT peptides) (see Note 19). In summary, the proper assay development process started with 224 proteins (corresponding to >1800 peptides and >38,000 transitions), which represented approximately 10 days of instrument time in unscheduled mode. After refinement, the assay still contained 200 proteins (corresponding to 480 peptides and 2880 transitions), which were separated in two distinct methods of 2 h each. We were thereby able to retain most of the biological information (number of proteins measured) but still drastically reduce instrument measurement time (see Note 20).

Multiplexed SRM for Biomarker Discovery in FFPE Tissues

4

199

Notes 1. The Skyline website (https://skyline.ms/project/home/soft ware/Skyline/begin.view) contains many useful tutorials to learn how to best use the software. We particularly recommend the “Skyline Targeted Method Editing,” the “Skyline Targeted Method Refinement,” and the “iRT Retention Time Prediction” tutorials. 2. We recommend the approach of creating your own peptide library specific to FFPE material as it is known that peptides generated from fresh frozen material can be different from those generated from formalin-fixed material [30, 31]. 3. The NIST library should be used to evaluate whether a protein is generally present in sufficient abundance to be experimentally detected. However, it should not be used to select suitable peptides for SRM measurement because the experimental data used to generate this library has not been specifically obtained from formalin-fixed material. 4. Ideally, we would have liked to measure all proteins of the target list; however, this was not realistically feasible in our example due to instrument capability, budget for synthetic heavy-labeled peptides, and measurement time (we aimed to use the developed assay in a cohort of approximately 100 samples). If you have the capacity of including all proteins of interest in your final SRM assay, this step can be skipped. 5. Considering these additional proteins is worthwhile in our opinion, because the untargeted MS/MS experiment described only allows to measure a very limited number of samples (four in this case), which is too low to be really representative of the variability encountered in a cohort of several dozen clinical samples. 6. We chose the option of restricting the number of proteins and ordering more peptides per protein because there are often many theoretical peptides for a given protein but predicting the best performing ones is difficult. We expected this to be even more true for FFPE material. The option chosen here also depends on the abundance of the proteins in the samples to measure with the final assay. In our case, we expected many of the proteins of interest to be low-abundant, which made the selection of peptides with the best analytical response even more important. 7. We recommend performing this step in a separate Skyline document. 8. We chose to have the peptides premixed in equimolar proportions (“SpikeMix”) and aliquoted so as to contain 1 nmol of

200

Carine Steiner et al.

each peptide per aliquot (1901
1 nmol). The peptides are delivered crude; therefore, the indicated concentration is approximate but is sufficient for relative quantification purposes. 9. Before ordering synthetic peptides, check that your list does not contain any C-terminal peptides because these generally do not have an arginine or lysine at the C-terminus. As a consequence, they will not be synthesized properly unless the specific C-terminal amino acid is specified (this option will probably be more expensive). 10. Although spiking the peptides in a biological system would be more representative of real samples, we used buffer as a background for the first screening. This allowed us to perform a first peak selection without having to deal with too many interferences in the signal traces, which would happen if we had chosen a background originating from tissues or cell lines for example. Also, there were too many transitions to include the light peptides in addition to the heavy peptides at this stage (instrument time). Therefore, the presence of endogenous (light) peptides was not mandatory at this point. 11. Three peptides per protein and three transitions per peptide is the best case scenario but is not always feasible. Reasons for this include proteins with very few proteotypic peptides or peptides which cannot be used for measurement due to poor analytical performance for example. 12. During this visual inspection step, we only removed precursor ions, not individual transitions. 13. By inconsistencies, we mean when a fragment ion is present in the MS/MS spectra of the peptide library but the same fragment ion does not generate a peak in the co-eluting SRM trace. Such examples typically have low dotp scores (approximately 0.5). 14. The iRT function is described in detail in a specific “iRT Retention Time Prediction” tutorial on the Skyline Website. We recommend investing time to use an iRT calibrator, especially if you plan to use scheduling methods with narrow time windows (3–4 min). The iRT function allows to easily predict small retention time shifts (for example, when installing a new column or new eluents) without having to measure again all transitions in unscheduled mode. As iRT calibrators, we selected 12 peptides from our pool of synthetic peptides, which were equally distributed throughout the chromatographic gradient. 15. The majority of peptides elute near the center of the gradient. Focusing on these peptides will lead to a higher number of

Multiplexed SRM for Biomarker Discovery in FFPE Tissues

201

transitions to monitor in this area and will require to set narrower scheduling windows in order to compensate for this. 16. In some cases, we observed a shift in retention time and the peak was partially outside of the acquisition window. In those cases, we set an arbitrary threshold of approximately 60%. If the peak apex was clearly visible, we kept and used the data, otherwise the data was discarded. This way of proceeding does not represent a problem because in the final assay, the peak area of the light peptide will be normalized using the heavy-labeled peptide, which has the same elution profile and is equally truncated. Moreover, several peaks were ambiguous in that the transition pattern was not consistent with previous documents or several peaks were visible in the acquisition window. These peptides were flagged in Skyline as “correct peak could not be identified” and were not used for data analysis. 17. Alternatively, this step can also be performed in excel; however, it is more time-consuming because individual plots need to be created for each peptide. 18. Plotting the light peptide (which is constant throughout all dilutions) helps getting a sense of the level of endogenous peptide in the sample material (usually the higher, the better) as well as of the reproducibility of the signal (the more reproducible, the better). 19. We recommend using the iSRM option, especially if you plan to set very small scheduling windows of 2–3 min. In our example, we observed some peaks drifting out of the acquisition window even though we used the iRT and the iSRM features. For iSRM, open the exported transition list from Skyline and manually change the acquisition windows of the iRT peptides to 10 min and set them as reference peptides (set as “2” in the corresponding column). 20. 24 proteins were lost during assay development for the following reasons: (a) 3 had only one peptide which was at the C-terminal position and was not terminated with R or K. Therefore it was not synthesized correctly as a heavy analog in the pool of synthetic peptides. (b) 14 were lost during the first screening due to unsatisfying performance. (c) 2 were lost during second screening due to unsatisfying performance. (d) 5 were lost during linearity testing.

202

Carine Steiner et al.

Acknowledgments We are grateful to Tom Dunkley and Arno Friedlein for fruitful discussions about SRM assay development and to Thomas McKee for valuable input on breast cancer biomarkers. Furthermore, we are indebted to Gaby Walker for growing the breast cancer cell lines, to Paola Antinori for her help with the isoelectric focusing, to Alex Scherl for running the Perl script on the peptides, and to Marco Berrera for retrieving the isoelectric point for all peptides. References 1. Szasz AM, Gyorffy B, Marko-Varga G (2017) Cancer heterogeneity determined by functional proteomics. Semin Cell Dev Biol 64:132–142. https://doi.org/10.1016/j. semcdb.2016.08.026 2. Matboli M, El-Nakeep S, Hossam N et al (2016) Exploring the role of molecular biomarkers as a potential weapon against gastric cancer: a review of the literature. World J Gastroenterol 22(26):5896–5908. https://doi. org/10.3748/wjg.v22.i26.5896 3. Barbieri CE, Chinnaiyan AM, Lerner SP et al (2017) The emergence of precision urologic oncology: a collaborative review on biomarker-driven therapeutics. Eur Urol 71 (2):237–246. https://doi.org/10.1016/j. eururo.2016.08.024 4. Hinestrosa MC, Dickersin K, Klein P et al (2007) Shaping the future of biomarker research in breast cancer to ensure clinical relevance. Nat Rev Cancer 7(4):309–315. https:// doi.org/10.1038/nrc2113 5. Perez-Gracia JL, Sanmamed MF, Bosch A et al (2017) Strategies to design clinical studies to identify predictive biomarkers in cancer research. Cancer Treat Rev 53:79–97. https://doi.org/10.1016/j.ctrv.2016.12.005 6. Ikeda K, Monden T, Kanoh T et al (1998) Extraction and analysis of diagnostically useful proteins from formalin-fixed, paraffin-embedded tissue sections. J Histochem Cytochem 46 (3):397–403. https://doi.org/10.1177/ 002215549804600314 7. Hood BL, Conrads TP, Veenstra TD (2006) Mass spectrometric analysis of formalin-fixed paraffin-embedded tissue: unlocking the proteome within. Proteomics 6(14):4106–4114. https://doi.org/10.1002/pmic.200600016 8. Vincenti DC, Murray GI (2013) The proteomics of formalin-fixed wax-embedded tissue. Clin Biochem 46(6):546–551. https://doi. org/10.1016/j.clinbiochem.2012.10.002

9. Giusti L, Lucacchini A (2013) Proteomic studies of formalin-fixed paraffin-embedded tissues. Expert Rev Proteomics 10(2):165–177. https://doi.org/10.1586/epr.13.3 10. Steiner C, Tille JC, Lamerz J et al (2015) Quantification of HER2 by targeted mass spectrometry in formalin-fixed paraffin-embedded (FFPE) breast cancer tissues. Mol Cell Proteomics 14(10):2786–2799. https://doi.org/10. 1074/mcp.O115.049049 11. Hembrough T, Thyparambil S, Liao WL et al (2013) Application of selected reaction monitoring for multiplex quantification of clinically validated biomarkers in formalin-fixed, paraffin-embedded tumor tissue. J Mol Diagn 15 (4):454–465. https://doi.org/10.1016/j. jmoldx.2013.03.002 12. Catenacci DV, Liao WL, Thyparambil S et al (2014) Absolute quantitation of Met using mass spectrometry for clinical application: assay precision, stability, and correlation with MET gene amplification in FFPE tumor tissue. PLoS One 9(7):e100586. https://doi.org/10. 1371/journal.pone.0100586 13. Nuciforo P, Thyparambil S, Aura C et al (2016) High HER2 protein levels correlate with increased survival in breast cancer patients treated with anti-HER2 therapy. Mol Oncol 10 (1):138–147. https://doi.org/10.1016/j. molonc.2015.09.002 14. Carr SA, Abbatiello SE, Ackermann BL et al (2014) Targeted peptide measurements in biology and medicine: best practices for mass spectrometry-based assay development using a fit-for-purpose approach. Mol Cell Proteomics 13(3):907–917. https://doi.org/10.1074/ mcp.M113.036095 15. MacLean B, Tomazela DM, Shulman N et al (2010) Skyline: an open source document editor for creating and analyzing targeted proteomics experiments. Bioinformatics 26 (7):966–968. https://doi.org/10.1093/bioin formatics/btq054

Multiplexed SRM for Biomarker Discovery in FFPE Tissues 16. Perou CM, Sorlie T, Eisen MB et al (2000) Molecular portraits of human breast tumours. Nature 406(6797):747–752. https://doi.org/ 10.1038/35021093 17. Liu NQ, Stingl C, Look MP et al (2014) Comparative proteome analysis revealing an 11-protein signature for aggressive triplenegative breast cancer. J Natl Cancer Inst 106 (2):djt376. https://doi.org/10.1093/jnci/ djt376 18. Sorlie T, Perou CM, Tibshirani R et al (2001) Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci U S A 98 (19):10869–10874. https://doi.org/10. 1073/pnas.191367098 19. Parker JS, Mullins M, Cheang MC et al (2009) Supervised risk predictor of breast cancer based on intrinsic subtypes. J Clin Oncol 27 (8):1160–1167. https://doi.org/10.1200/ JCO.2008.18.1370 20. Smid M, Wang Y, Zhang Y et al (2008) Subtypes of breast cancer show preferential site of relapse. Cancer Res 68(9):3108–3114. https://doi. org/10.1158/0008-5472.CAN-07-5644 21. Rody A, Karn T, Liedtke C et al (2011) A clinically relevant gene signature in triple negative and basal-like breast cancer. Breast Cancer Res 13(5):R97. https://doi.org/10.1186/ bcr3035 22. O’Toole SA, Beith JM, Millar EK et al (2013) Therapeutic targets in triple negative breast cancer. J Clin Pathol 66(6):530–542. https:// doi.org/10.1136/jclinpath-2012-201361 23. Hubner NC, Ren S, Mann M (2008) Peptide separation with immobilized pI strips is an attractive alternative to in-gel protein digestion for proteome analysis. Proteomics 8 (23–24):4862–4872. https://doi.org/10. 1002/pmic.200800351 24. Kuster B, Schirle M, Mallick P et al (2005) Scoring proteomes with proteotypic peptide

203

probes. Nat Rev Mol Cell Biol 6(7):577–583. https://doi.org/10.1038/nrm1683 25. Scherl A, Shaffer SA, Taylor GK et al (2008) Genome-specific gas-phase fractionation strategy for improved shotgun proteomic profiling of proteotypic peptides. Anal Chem 80 (4):1182–1191. https://doi.org/10.1021/ ac701680f 26. Maclean B, Tomazela DM, Abbatiello SE et al (2010) Effect of collision energy optimization on the measurement of peptides by selected reaction monitoring (SRM) mass spectrometry. Anal Chem 82(24):10116–10124. https:// doi.org/10.1021/ac102179j 27. Lange V, Picotti P, Domon B et al (2008) Selected reaction monitoring for quantitative proteomics: a tutorial. Mol Syst Biol 4:222. https://doi.org/10.1038/msb.2008.61 28. Escher C, Reiter L, MacLean B et al (2012) Using iRT, a normalized retention time for more targeted measurement of peptides. Proteomics 12(8):1111–1121. https://doi.org/ 10.1002/pmic.201100463 29. Gallien S, Peterman S, Kiyonami R et al (2012) Highly multiplexed targeted proteomics using precise control of peptide retention time. Proteomics 12(8):1122–1133. https://doi.org/ 10.1002/pmic.201100533 30. Sprung RW Jr, Brock JW, Tanksley JP et al (2009) Equivalence of protein inventories obtained from formalin-fixed paraffin-embedded and frozen tissue in multidimensional liquid chromatography-tandem mass spectrometry shotgun proteomic analysis. Mol Cell Proteomics 8(8):1988–1998. https://doi. org/10.1074/mcp.M800518-MCP200 31. Palmer-Toy DE, Krastins B, Sarracino DA et al (2005) Efficient method for the proteomic analysis of fixed and embedded tissues. J Proteome Res 4(6):2404–2411. https://doi.org/ 10.1021/pr050208p

Chapter 14 Development and Validation of Multiple Reaction Monitoring (MRM) Assays for Clinical Applications Georgia Kontostathi, Manousos Makridakis, Vasiliki Bitsika, Nikolaos Tsolakos, Antonia Vlahou, and Jerome Zoidakis Abstract Selected/multiple reaction monitoring-mass spectrometry (SRM/MRM) is an analytical method that is frequently combined to the use of stable isotope-labeled standard (SIS) peptides for absolute protein quantification. The application of SRM/MRM is a relatively recent development in the proteomics field for analysis of biological samples (plasma, urine, cell/tissue lysates) targeting to a large extent biomarker validation. Although MRM generally by-passes the use of antibodies (being linked to sub-optimal assay specificity in many cases), bioanalytical validation of MRM protocols has not been robustly applied because of sensitivity issues, in contrary to antibody-based methods. In this chapter, we will discuss the points that should be addressed for MRM method development in clinical proteomics applications. Key words MRM, Proteomics, Plasma, Bioanalytical validation, Quantification

1

Introduction Selected/multiple reaction monitoring-mass spectrometry (SRM/MRM) is a high-throughput “-omics” technology that has emerged as a powerful tool for protein detection and absolute quantification in several biological samples such as plasma [1–13], urine [14–17] and cell/tissue lysates [18]. It has been applied in multiple applications of clinical proteomics, including quantification of individual diagnostic biomarkers (e.g., for cardiovascular disease in plasma [1]) or biomarker panels (e.g., differentiating benign from malignant pulmonary nodules [19, 20] or evaluating response of patients with hepatocellular carcinoma to sorafenib treatment [21]). SRM/MRM is useful for achieving “targeted proteomics analysis,” in which the mass spectrometer is programmed to analyze a preselected group of proteins, in contrast to “shotgun proteomics,” in which the spectra generated from all detectable proteins in a sample are interpreted by database searching [22]. Targeted proteomic analysis via MRM/SRM relies on the

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_14, © Springer Science+Business Media, LLC, part of Springer Nature 2019

205

206

Georgia Kontostathi et al.

detection of proteotypic peptides whose sequence is present only in the protein of interest. The specificity and accuracy of MRM quantification relies on the use of stable isotope-labeled standard (SIS) variants of these proteotypic peptides. Specific 13C and/or 15N isotopes are incorporated at the C-terminal lysine or arginine resulting in a mass difference of 8 or 10 Da, respectively, from the endogenous (or natural-NAT) peptide of interest. SIS peptides are spiked in the sample of interest at known amounts and measured in parallel to the respective endogenous peptides. In this way, absolute quantification of the latter can be achieved. In most cases, LC-MRM-MS analysis is performed with triple quadrupole instruments, or lately on hybrid systems, combining quadrupole, timeof-flight, and Orbitrap mass analyzers for improved mass resolution and sensitivity (Parallel Reaction Monitoring analysis) [23]. Precursor ions are selected in the first stage (Q1 quadrupole), fragmented in the second stage (Q2 quadrupole) via collision-induced dissociation (CID), and selected and/or measured in the third stage (Q3 quadrupole). The pair of m/z values corresponding to the precursor and fragment ions is called a transition. Transitions are recorded for both the SIS reference peptide and the NAT peptide and used for absolute quantification [16, 18]. Immuno-assay based methods such as enzyme linkedimmunosorbent assay (ELISA) have been extensively applied in the clinical setting according to Food and Drug Administration (FDA) guidelines [24]. Their application in the clinical practice has been preferred due to the high sensitivity (pmol quantification levels) and accuracy in protein quantification. However, they present some pronounced disadvantages such as high cost, limited multiplexing capability due to antibody cross reactivity and questionable specificity that varies based on the epitope of the antibody and the biological sample analyzed [8, 16]. In fact, the disappointing analytical performance of several ELISA tests for seven urinary biomarkers in terms of standard curve analysis, assay reproducibility, linearity, and spiking experiments performed in our lab [25] prompted us to start evaluating MRM assays, initially in plasma samples. Despite the promising results that MRM has provided as a potential biomarker validation/clinical implementation assay, there is a lack of thoroughly reported protocols to be uniformly followed across different labs. FDA guidelines for bioanalytical method validation are not specific for MRM, since they were published in 2001, when MRM methodology was not widely applied for peptide quantification. In this chapter, we will describe a detailed protocol incorporating FDA regulations (related to selectivity/interference screening, calibration/standard curves accuracy/precision/recovery, stability) [26] for rigorous bioanalytical validation of MRM assays for clinical applications.

MRM Method Development and Validation for Clinical Assays

207

The points to be discussed are the following: (1) Sample preparation (using plasma as an example) for LC-MRM-MS, describing thoroughly the required steps for sample treatment prior to MRM mass spectrometric analysis (2) Method development/selection of appropriate peptides for LC-MRM-MS, describing the selection process of proteotypic peptides (3) Selectivity/interference screening, in order to eliminate the effect of false positive results (4) Calibration/Standard Curve, defining the appropriate concentration of SIS peptides for subsequent analysis and quantification (5) Quantification of Absolute Peptide Abundance via Skyline, describing the bioinformatics analysis of spectra for quantification (6) Accuracy/ Precision/Recovery, regarding the analytical performance of the MRM methodology (7) Stability, determining the stability of the: (a) NAT peptides within the sample matrix and (b) SIS peptides, and finally (8) Application of method to routine analysis, describing a detailed protocol for high-throughput quantification of many samples. A workflow encompassing the aforementioned steps is summarized in Fig. 1.

1) Sample preparation (using plasma as an example) for LC-MRM-MS

2) Method development/selection of appropriate peptides for LC-MRM-MS Protein

Peptides

Proteotypic Peptide 3) Selectivity/interference screening

1) Concentration balanced SIS in clinical samples 2) Quantification via a) standard curve equation b) NATarea/SISarea ratio

Interferencefree peptides

Time (min)

4) Calibration/Standard Curve CVs Prostar()

3.2

Data Loading

1. To upload data from tabular file (see Notes 7 and 8) (i.e., stored in a file with one of the following extensions: .txt, .csv, .tsv, .xls, or .xlsx) click on the upper menu “Data manager,” then chose “Convert data.” 2. Go to the “Select File” tab. 3. Click on the “Browse. . .” button and select the tabular file of interest (see Note 9). 4. Once the upload is complete, indicate that it is a protein-level dataset (i.e., each line of the data table should correspond to a single protein, see Note 10). 5. Indicate if the data are already log-transformed or not. If not they will be automatically log-transformed (see Note 11). 6. If the quantification software uses “0” in places of missing values, tick the last option “Replace all 0 and NaN by NA” (as in ProStaR, 0 is considered a value, not a missing value). 7. Move on to the “Data Id” tab.

Protein-Level Statistical Analysis with ProStaR

229

Fig. 1 ProStaR home screen

Fig. 2 After loading a dataset (here thanks to the demo mode, see Note 7), the menu appears contextually. Table 1 provides a view of the menu contents

230

Samuel Wieczorek et al.

Table 1 Synoptic view of the ProStaR menu (once enabled by data loading, see Figs. 1 and 2) Prostar

Data manager

Data processing

Data mining

Help

Home

Open MSnset file

Filter data

Descriptive statistics

Useful links

Global settings

Convert data

Normalization

GO analysis

Release notes

Demo mode

Miss value imputation

Differential analysis

FAQ

Export dataset

Hypothesis testing

Session log

Check for updates Bug reports

8. If the dataset already contains an ID columns (a column where each cell has a unique content, which can serve as an ID for the proteins), choose “User ID” and select the appropriate column name. In any case, it is possible to use the “Auto ID” option, which creates an artificial index. 9. Move on to the “Exp. and feat. data” tab. 10. Select the columns which contain the protein abundances (one column for each sample of each condition). To select several column names in a row, click-on on the first one, and click-off on the last one. Alternatively, to select several names which are not continuously displayed, use the “Ctrl” key to maintain the selection. 11. If, for each sample, a column of the dataset provides information on the identification method (e.g., by direct MS/MS evidence, or by mapping) check the corresponding tick box. Then, for each sample, select the corresponding column. If none of these pieces of information is given, or, on the contrary, if all of them are specified with a different column name, a green logo appears, indicating it is possible to proceed (however, the content of the specified columns is not checked, so that it is the user’s responsibility to select the correct ones). Otherwise (i.e., the identification method is given only for a subset of samples, or a same identification method is referenced for two different samples), a red mark appears, indicating some corrections are mandatory. 12. Move on to the “Sample metadata” tab. This tab guides the user through the definition of the experimental design. 13. Fill the empty columns with as different names as biological conditions to compare (minimum two conditions and two samples per conditions) and click on “Check conditions.” If necessary, correct until the conditions are valid. When achieved, a green logo appears and the sample is reordered according to the conditions.

Protein-Level Statistical Analysis with ProStaR

231

Fig. 3 Definition and validation of the experimental design

14. Choose the number of levels in the experimental design (either 1, 2, or 3), and fill the additional column(s) of the table (see Note 12). 15. Once the design is valid (a green check logo appears), move on to the “Convert” tab (see Fig. 3). 16. Provide a name to the dataset to be created and click on the “Convert” button. 17. As a result, a new MSnset structure is created and automatically loaded. This can be checked with the name of the file appearing in the upper right hand side of the screen, as a title to a new drop-down menu. So far, it only contains “Original—protein,” but other versions of the dataset will be added along the course of the processing. 3.3

Data Export

As importing a new dataset from a tabular file is a tedious procedure, we advise to save the dataset as an MSnset binary file right after the conversion. This makes it possible to restart the statistical analysis from scratch if a problem occurs without having to convert the data another time. To do so: 1. Click on “Export” in the “Data manager” menu. 2. Choose MSnset as file format and provide a name to the object (see Note 13).

232

Samuel Wieczorek et al.

3. Optionally, it is possible to select a subset of the column metadata to make the file smaller. 4. Click on “Download.” 5. Once the downloading is over, store the file in the appropriate directory. 6. To reload any dataset stored as an MSnset structure, refer to Note 3. 3.4 Descriptive Statistics

1. By clicking on “Descriptive statistics” in the “Data mining” menu, it is possible to access several tabs generating various plots (see Note 14) that provides a comprehensive and quick overview of the dataset (see Note 15). 2. On the first tab (named “overview”), a brief summary of the quantitative data size is provided. It roughly amounts to the data summary that is displayed along with each dataset during the loading step of the demo mode. 3. On the second tab (named “miss. values”), barplots depicts the distribution of missing values: the left hand side barplot represents the number of missing values in each sample. The different colors correspond to the different conditions (or groups, or labels). The second barplot (in the middle) displays the distribution of missing values; the red bar represents the empty protein count (i.e., the number of lines in the quantitative data that are only made of missing values). The last barplot represents the same information as the previous one, yet, condition-wise. Let us note that protein with no missing values are represented in the last barplot while not on the second one (to avoid a too large Y-scale). 4. The third tab is the data explorer (see Fig. 4): it makes it possible to view the content of the MSnSet structure. It is made of four tables, which can be displayed one at a time thanks to the radio button on the left menu. The first one, named “Quantitative data” contains quantitative values. The missing values are represented by empty cells. The second one is referred to “Protein metadata.” It contains all the column dataset that are not the quantitative data. The third tab, “Replicate metadata,” summarizes the experimental design, as defined at the import step (see Subheading 3.2, step 13). Finally, the last tab, “Dataset history,” contains the logs (see Note 16) of the previous processing. 5. In the fourth tab (“Corr. matrix”), it is possible to visualize to what extent the replicate samples correlates or not. The contrast of the correlation matrix can be tuned thanks to the color scale on the left hand side menu.

Protein-Level Statistical Analysis with ProStaR

233

Fig. 4 Screenshot of the Data Explorer

6. A heatmap as well as the associated dendrogram is depicted on the fifth tab. The colors represent the intensities: red for high intensities and green for low intensities. White color corresponds to missing values. The dendrogram shows a hierarchical classification of the samples, so as to check that samples are related according to the experimental design. It is possible to tune the clustering algorithm (see Note 17) that produces the dendrogram by adjusting the “distance” and “linkage” parameters, as described in the hclust R function [16]. 7. Tabs 6, 7, and 8 represent in a slightly different way the same information, that is the distribution of intensity values by replicates and conditions: respectively, boxplots, violin-plots and smoothed histograms (a.k.a. kernel density plots) are used. Depending on the needs, it is possible to shift from one representation to any of the others. 8. Finally, the last tabs display a density plot of the variance (within each condition) conditionally to the log-intensities (see Note 18). 3.5

Filtering

This stage aims at filtering out proteins according to their number of missing values, as well as according to some information stored in the protein (or feature) metadata. 1. Click on “Filter data” in the “Data processing” menu.

234

Samuel Wieczorek et al.

Fig. 5 Missing value filter

2. On the first tab (called “Missing values,” see Fig. 5), select among the various options which proteins should be filtered out or not. The options are the following: (a) None: No filtering, the quantitative data is left unchanged (see Note 19). (b) Whole Matrix: proteins that contain in the quantitative dataset (across all conditions) fewer non-missing values than a user-defined threshold are deleted. (c) For every condition: proteins that contain fewer non-missing values in each condition than a user-defined threshold are removed. (d) At least one condition: proteins that contain fewer non-missing values in at least one condition than a userdefined threshold are suppressed. 3. Visualize the effect of the filtering options without changing the current dataset by clicking on “Perform filtering.” If the filtering does not produce the expected effect, test another one. To do so, simply choose another method in the list and click again on “Perform filtering.” The plots are automatically updated. This action does not modify the dataset but offers a preview of the filtered data. Iterate this step as long as necessary. 4. Move on to the second tab (called “String based filtering”), where it is possible to filter out proteins according to information stored in the metadata. 5. Among the columns constituting the protein metadata listed in the drop-down menu, select the one containing the information (see Note 20) of interest (for instance, “Contaminant” or

Protein-Level Statistical Analysis with ProStaR

235

“Reverse,” see Note 21). Then, specify in each case the prefix chain of characters that identifies the proteins to filter (see Note 22). 6. Click on “Perform” to remove the corresponding proteins. A new line appears in the table listing all the filters that have been applied. 7. If another filter must be applied, go back to step 4. 8. Once all the filters have been applied, move on to the last tab (called “Visualize and Validate”) to check the set of filtered out proteins. This visualization tools works similarly as the Data explorer (see Subheading 3.4, step 4). 9. Finally, click on “Save filtered dataset.” The information related to the type of filtering as well as to the chosen options appears in the Session log tab (“Session logs” tab from the “Session logs” option in the “Data manager” menu). 3.6 Navigating Through the Dataset Versions

Once the filters have been applied and the results saved, a new dataset is created. It is referred to as “Filtered—protein,” and its name appears right below “Original—protein” in the upper right drop-down menu, beside the dataset name (see upper right corner of Fig. 2, as well as Subheading 3.2, step 17). Unless modified, the newest created dataset is always the current dataset, i.e., the dataset on which further processing will be applied. As soon as the current dataset is modified, all the plots and tables in ProStaR are automatically updated. Thus, as soon as a new dataset is created, we suggest to go back to the descriptive statistics menu (see Subheading 3.4) to check the influence of the latest processing on the data. It is possible to have a dynamic view of the processing steps by navigating back and forth in the dataset versions, so as to see the graphic evolutions (see Note 23).

3.7

The next processing step proposed by ProStaR is data normalization (see Fig. 6). Its objective is to reduce the biases introduced at any preliminary stage (such as for instance batch effects).

Normalization

1. Choose the normalization method among the following ones (see Note 24). (a) None: No normalization is applied. (b) Global quantile alignment: The Quantile of the intensity distributions of all the samples are equated, as described in [17]. (c) Column sums: The total intensity values of all the samples are equated. The rationale behind is to normalize according to the total amount of biological material within each sample.

236

Samuel Wieczorek et al.

Fig. 6 Normalization tab. The density plot, the distortion plot and the box plot makes it possible to visualize the influence of each normalization method

(d) Quantile Centering: A given quantile of the intensity distribution is used as reference (see Note 25). (e) Mean Centering: sample intensity distributions are aligned on their mean intensity values (and optionally, the variance distributions are equated to one). (f) Variance Stabilizing Normalization: A wrapper to the method described in [18]. (g) Loess normalization: The intensity values are normalized by means of a local regression model [19] of the difference of intensities as function of the mean intensity value (see [20] for implementation details). 2. Then, for each normalization method, the interface is automatically updated to display the method parameters that must be tuned. Notably, for most of the methods, it is necessary to indicate whether the method should apply to the entire dataset at once, or whether each condition should be normalized independently of the others. For other parameters, which are specific to each method, the reader is referred to ProStaR user manual, available through the “Help” section of the main menu (see Note 26). 3. Click on “Perform normalization.” 4. Observe the influence of the normalization method on the graphs of the right hand side panel. Optionally, click on “Show plot options,” so as to tune the graphics for a better visualization.

Protein-Level Statistical Analysis with ProStaR

237

5. If the result of the normalization does not correspond to the expectations, change the normalization method or change its tuning. 6. Once the normalization normalization.”

is effective,

click on

“Save

7. Check that a new version appears in the dataset version dropdown menu, referred to as “Normalized - Protein”. 8. Remember that at any time, it is possible to return to the menu “Descriptive statistics” to have a refined view of each step of the processing. 3.8

Imputation

In protein-level datasets, ProStaR makes it possible to have separate processing for two different types of missing values: POV (standing for Partially Observed Value) and MEC (standing for Missing in the Entire Condition). All the missing values for a given protein in a given condition are considered POV if and only if there is at least one observed value for this protein in this condition. Alternatively, if all the intensity values are missing for this protein in this condition, the missing values are considered MEC (see Note 27). 1. On the first tab (see Fig. 7), select the algorithm to impute POV missing values. According to our expertise, we advise to select the SLSA algorithm (Giai Gianetto Q et al., submitted) but other methods are also of interest in specific situations. 2. Tune the parameters of the chosen imputation method. 3. Click on “Perform Imputation.” It will enable the next tab, on which the result of the imputation is shown. 4. Move on to the second tab and decide how to deal with MEC. As a matter of fact, it is always dangerous to impute them, as in the absence of any value to rely on, the imputation is arbitrary and risks to spoil the dataset with maladjusted values. As an

Fig. 7 Imputation of the Partially Observed Values

238

Samuel Wieczorek et al.

alternative, it is possible to (1) keep the MEC as is in the dataset, yet, it may possibly impede further processing, (2) discard them at the filter step (see Subheading 3.5, step 2) so as to process them separately. However, this will make it impossible to include these proteins (and their processing) in the final statistics, such as for instance FDR. 5. If MEC are not going to be imputed, select the imputation method referred to as “None.” Otherwise, select the appropriate method. Based on our experience, we advise to use detQuantile (see Note 28). 6. Tune the parameters of the MEC imputation method (see Note 29). 7. Click on “Perform Imputation” and move on to the next tab (“Validate & save”). 8. Observe the influence of the chosen imputation methods on the graphs of the right hand side panel. If the result of the imputation does not correspond to the expectations, change the imputation methods or their tuning (by going back to step 1 of this section). 9. Once the imputation is effective, click on “Save imputation.” 10. Check that a new version appears in the dataset version dropdown menu, referred to as “Imputed - Protein” (see Notes 30 and 31). 3.9 Hypothesis Testing

For datasets that do not contain any missing values, or for those where these missing values have been imputed, it is possible to test whether each protein is significantly differentially abundant between the conditions. To do so, click on “Hypothesis testing” in the “Data processing” menu (see Fig. 8). 1. Choose the test contrasts. In case of two conditions to compare, there is only one possible contrast. However, in case of

Fig. 8 Tuning of the null hypothesis significance testing (to prepare the differential analysis)

Protein-Level Statistical Analysis with ProStaR

239

N > 2 conditions, several pairwise contrasts are possible. Notably, it is possible to perform N tests of the “1vsAll” type, or N (N
1)/2 tests of the “1vs1” type. 2. Then, choose the type of statistical test, between limma [20] or t-test (either Welch or Student). This makes appear a density plot representing fold-change (FC) (as many density curves on the plot as contrasts). 3. Thanks to the FC density plot, tune the FC threshold (see Note 32). 4. Run the tests and save the dataset to preserve the results (i.e., all the computed p-values). Then, this new dataset, containing the p-values and FC cutoff for the desired contrasts, can be explored in the “Differential analysis” tabs available in the “Data mining” menu. 3.10 Differential Analysis

If one clicks on “Differential analysis” in the “Data mining” menu, it is possible to analyze the results of all statistical tests (see Subheading 3.9). To do so: 1. Select a pairwise comparison of interest. The corresponding volcano plot is displayed. 2. Possibly, swap the FC axis with the corresponding tick box, depending on layout preferences. 3. Some proteins may have an excellent p-value, while a too great proportion of their intensity values (within the two conditions of interest in this comparison) are in fact imputed values, so that they are not trustworthy. To avoid such proteins become false discoveries, it is possible to discard them (by forcing their p-value to 1). To do so, fill in the last parameters of the right hand side menu, which are similar to the MV filtering options described in Subheading 3.5. 4. Click on “Perform p-value push” and move on to Tab 3, “pvalue calibration.” 5. Tune the calibration method, as indicated in [21] as well as in ProStaR user manual. 6. Move on to the next tab and adjust the FDR threshold (see Fig. 9). 7. Save any plot/table of interest and move on to the next tab (“Summary”) to have a comprehensive overview of the differential analysis parameters. 8. Possibly, go back to step 1 to process another pairwise comparison. Alternatively, it is possible to continue with the current protein list so as to explore the functional profiles of the proteins found statistically differentially abundant between the compared conditions (as explained in the next section).

240

Samuel Wieczorek et al.

Fig. 9 Volcano plot to visualize the differentially abundant proteins according to a user-specified FDR

3.11

GO Analysis

1. To Perform a Gene Ontology (GO) enrichment analysis, click on the corresponding option in the “Data mining” menu. (We assume the reader is familiar with the interest and use of the GO terms [22].) 2. Go to the first tab (“GO setup”) and tune the “Source of protein ID” by ticking the “Select a column in dataset” radio button (see Note 33). 3. Select the dataset column that contains the protein ID. 4. Indicate the type of protein ID (GeneID, Uniprot). 5. Select the organism of interest among the proposed list. 6. Select the ontology of interest (either Molecular Function, Biological Process, or Cellular Component). 7. Click on “Map protein ID.” 8. Check on the right panel the list of proteins that could not be mapped in the ontology (see Note 34). 9. Move on to the next tab “GO classification,” or depending on the needs, skip it to directly move to “GO Enrichment.” 10. GO classification: choose the GO level(s) of interest and click on “Perform GO grouping” (see Note 35). Once the result appears, right-click on the plots to save them. Then, depending on the needs, go to the next tab “GO Enrichment,” or directly move to the “Summary” tab.

Protein-Level Statistical Analysis with ProStaR

241

11. GO Enrichment: choose the set of proteins that should be used as reference to perform the enrichment test (“Entire organism,” “Entire dataset,” “Custom,” see Note 36) and tune the FDR threshold before clicking on the “Perform enrichment analysis” button. As with “GO classification,” save the plots of interest. 12. Move on to the “Summary” tab. It displays a table summarizing the parameters of the GO analysis.

4

Notes 1. ProStaR versions which are posterior to 1.14 may slightly differ from what is described in this protocol. However, the general spirit of the graphical user interface remains unchanged allowing any user accustomed to an earlier version to easily adapt to a newer version. 2. With only two replicates per conditions, the computations are tractable. It does not mean that statistical validity is guaranteed. Classically, three replicates per conditions are considered a minimum in case of a controlled experiment with a small variability, mainly issuing from technical or analytical repetitions. Analysis of complex proteomes between conditions with a large biological variability requires more replicates per conditions (5–10). 3. To reload a dataset that has previously been stored as an MSnset file, go to “Open MSnset file” in the “Data manager” menu and simply browse the file system to find the desired file. 4. Before installing the software, any user can have a quick overview by testing its demo mode (see Note 7) on the following URL: http://www.prostar-proteomics.org. Any user can also test the website version on his own data, yet we do not recommend it since the server has limited computational capabilities that are shared between all the users connected at the same moment. Overloading is therefore possible, which would lead to data loss. Moreover, we are currently working on a portable version of ProStaR that could directly be downloaded for the above URL, without requiring any installation. This feature will likely be available in future versions of ProStaR. 5. We advise to use the latest version and to make regular updates, so as to guarantee the compatibility with the latest ProStaR developments. 6. If a package is still missing, it means that some problem occurred during its install, or during the install of another package it depends on. In such a case, we first advise to try again ProStaR install, by executing the commands indicated in

242

Samuel Wieczorek et al.

Subheading 2.7. If it still does not work, then it is necessary to manually install the corresponding packages. Depending on whether these are CRAN packages or BioConductor packages, as well as on the possible use of an IDE, the procedure will change, and we refer the reader to the corresponding documentation. 7. Before uploading a real dataset, any user can test ProStaR thanks to the demo mode. This mode provides a direct access to the DAPARdata packages where some toy datasets are available, either in tabular or MSnset formats. Concretely, the demo mode is accessible in the “Data manager” menu: on the corresponding tab, a drop-down menu lists the datasets that are available through DAPARdata. After selecting a proteinlevel datasets (see Note 10), click on “Load demo dataset.” 8. The DAPARdata package also contains tabular versions (in txt format) of the datasets available in the demo mode. Thus, it is also possible to test the import/export/save/load functions of ProStaR with these toy datasets. Concretely, one simply has to import them from the folder where the R packages are installed, in the following sub-folder: . . .\R\R-3.4.0\library \DAPARdata\extdata. Note that each dataset is also available in the MSnset format, but these datasets should not be considered to test conversion functions from/to tabular formats. 9. If the user chooses an Excel file, a drop-down menu appears and the user is asked to select the spreadsheet containing the data. 10. If the dataset under consideration is not a protein dataset (each line of the quantitative table does not represent a protein, but for instance a peptide), do not apply the present protocol. 11. ProStaR cannot process non-log-transformed data. Thus, do not cheat the software by indicating data on their original scale are log-transformed. 12. In case of difficulty, either to choose the adapted design hierarchy or to fill the table design, it is possible to click on the interrogation mark beside the sentence “Choose the type of experimental design and complete it accordingly.” Except for flat design, which are automatically defined, it displays an example of the corresponding design. It is possible to rely on this example to precisely fill the design table. 13. Alternatively, it is possible to export data as excel spreadsheets or as a zip containing text files. This has no interest in case of a preliminary export; however, it may be useful to share a dataset once the statistical analysis is completed. 14. The user can download the plots showed in ProStaR by rightclicking on the plot. A contextual menu appears and let the user

Protein-Level Statistical Analysis with ProStaR

243

choose either “Save image as” or “Copy image.” In the latter case, the user has to paste the image in appropriate software. 15. It is essential to regularly go back to these tabs, so as to check that each processing step has produced the expected results. 16. Contrarily to the “Session logs” panel (see Subheading 3.5, step 9), the information here does not relate to the session: it is saved from a session to the next one. 17. Computing the heatmap and the dendrogram may be very computationally demanding depending on the dataset. 18. As is, the plot is often difficult to read, as the high variances are concentrated on the lower intensity values. However, it is possible to interactively zoom in on any part of the plot by clicking and dragging. 19. This is the default option, however, we warn about such an absence of filtering: if too many missing values remain, the statistical analysis will be spurious. Moreover, we recommend at least filtering out proteins with only missing values in all the conditions, as such proteins do not carry any trustworthy quantitative information. Finally, depending on the imputation choices with regard to MEC (see Subheading 3.8 as well as Note 27), it may be wise to filter out proteins whose intensity values are completely missing in at least one condition. 20. To work properly, the selected column must contain information encoded as a string of characters. For each protein, the beginning of the corresponding string is compared to a given prefix. If the prefix matches, the protein is filtered out. Otherwise, it is conserved in the protein list. Note that the filter only operates a prefix search (at the beginning of the string), not a general tag match search (anywhere in the string). Similarly, filters based on regular expressions are not implemented. 21. In datasets resulting from a MaxQuant [23] search, metadata indicates under a binary form which proteins are reversed sequences (resulting from a target-decoy approach) and which are potential contaminants. Both of them are indicated by a “+” in the corresponding column (the other proteins having a NA instead). It is thus possible to filter both reversed and contaminants out by indicating “+” as the prefix filter. However, if adequately encoded, filtering on other type of information is possible. 22. If one has no idea of the prefixes, it is possible to switch to the “Data Explorer” in the “Descriptive Statistics” menu (see Subheading 3.4), so as to visualize the corresponding metadata. 23. It is not possible to keep in parallel several datasets or multiple versions of a dataset at a particular level (for instance, a dataset filtered using various rules). Thus, if one goes on with the next

244

Samuel Wieczorek et al.

processing steps on an older dataset, or if one goes back to a previous step and restart it, the new results will overwrite the previously saved ones at this same step, without updating other downstream processing, leading to possible inconsistencies. 24. This list corresponds to the normalization methods available in ProStaR version 1.12; future versions will possibly propose slightly different methods. 25. This normalization method should not be confused with Global quantile alignment. 26. It should be noted that the choice of a normalization method and its tuning is highly data dependent, so that a single protocol cannot be proposed. The data analyst should gather expertise on the normalization methods, so as to be able to choose soundly. Thus, we advise to refer to ProStaR user manual, as well as to the literature describing the normalization methods (to be found in the “Help” section of ProStaR). 27. As a result, all the missing values are either POV or MEC. Moreover, for a given protein across several conditions, the missing values can be both POV and MEC, even though within a same condition they are all of the same type. 28. The detQuantile method proposes to impute each missing value within a given sample by a deterministic value, usually a low value. The rationale behind is that MEC values corresponds to proteins that are below the quantification limit in one condition, so that they should not be imputed according to observed values in the other conditions. Although the use of a deterministic value slightly disturb the intensity distribution, it makes the MEC values easy to spot, as they all correspond to a known numerical value. 29. If detQuantile is used, we advise to use a small quantile of the intensity distribution to define the imputation value, for instance, 1–2.5%, depending on the stringency you want to apply on proteins quantified only in one condition of a pairwise comparison. In case of a dataset with too few proteins, the lower quantile may amount to instable values. In such a case, we advise to use a larger quantile value (for instance 10% or 20%) but to use a smaller multiplying factor (for instance 0.2 or smaller) so as to keep the imputation value reasonably small with respect to the detection limit. In any case, when using detQuantile, the list of imputation values for each sample appears above the graphics on the right panel. 30. When exporting the data in the Microsoft Excel format, imputed values are displayed in a colored cell so that they can easily be distinguished.

Protein-Level Statistical Analysis with ProStaR

245

31. Recall that at any moment, it is possible to go back to the “Descriptive statistics” menu to have a refined view of each step of the processing. 32. As explained in [24], we advise to tune the FC conservatively by avoiding discarding to many proteins with it. Moreover, it is important tune the FC to a small enough value, so as to avoid discarding too many proteins. In fact, it is important to keep enough remaining proteins for the next coming FDR computation step (see Subheading 3.10), as (1) FDR estimation is more reliable with many proteins, (2) FDR, which relates to a percentage, does not make sense on too few proteins. 33. The “Choose a file” radio button should only be used to perform a GO analysis on a protein list that does directly come from the quantitative dataset. 34. Proteins that could not be matched will no longer be considered for GO analysis. Thus, it is the user’s responsibility to determine whether this will significantly affect the results. If so, it may be necessary to go back to the original data and fix any issues with protein accession names. 35. It may take a while, so be patient. 36. In the latter case, select the file that was preliminary prepared on purpose (see the ProStaR user manual [14] for details).

Acknowledgment ProStaR software development was supported by grants from the “Investissement d’Avenir Infrastructures Nationales en Biologie et Sante´” program (ProFI project, ANR-10-INBS-08) and by the French National Research Agency (GRAL project, ANR-10LABX-49-01). References 1. Zhang Y, Fonslow BR, Shan B et al (2013) Protein analysis by shotgun/bottom-up proteomics. Chem Rev 113(4):2343–2394. https:// doi.org/10.1021/cr3003533 2. Ong SE, Foster LJ, Mann M (2003) Mass spectrometric-based approaches in quantitative proteomics. Methods 29(2):124–130. https:// doi.org/10.1016/S1046-2023(02)00303-1 3. Schwanh€ausser B, Busse D, Li N et al (2011) Global quantification of mammalian gene expression control. Nature 473 (7347):337–342. https://doi.org/10.1038/ nature10098

4. Tyanova S, Temu T, Sinitcyn P et al (2016) The Perseus computational platform for comprehensive analysis of (prote) omics data. Nat Methods 13(9):731–740. https://doi.org/10. 1038/nmeth.3901 5. Choi M, Chang CY, Clough T et al (2014) MSstats: an R package for statistical analysis of quantitative mass spectrometry-based proteomic experiments. Bioinformatics 30 (17):2524–2526. https://doi.org/10.1093/ bioinformatics/btu305 6. MacLean B, Tomazela DM, Shulman N et al (2010) Skyline: an open source document

246

Samuel Wieczorek et al.

editor for creating and analyzing targeted proteomics experiments. Bioinformatics 26 (7):966–968. https://doi.org/10.1093/bioin formatics/btq054 7. Zhang X, Smits AH, van Tilburg GB et al (2018) Proteome-wide identification of ubiquitin interactions using UbIA-MS. Nat Protoc 13(3):530–550. https://doi.org/10. 1038/nprot.2017.147 8. Contrino B, Miele E, Tomlinson R et al (2017) DOSCHEDA: a web application for interactive chemoproteomics data analysis. PeerJ Comput Sci 3:e129. https://doi.org/10.7717/peerjcs.129 9. Singh S, Hein MY, Stewart AF (2016) msVolcano: a flexible web application for visualizing quantitative proteomics data. Proteomics 16 (18):2491–2494. https://doi.org/10.1002/ pmic.201600167 10. Efstathiou G, Antonakis AN, Pavlopoulos GA et al (2017) ProteoSign: an end-user online differential proteomics statistical analysis platform. Nucleic Acids Res 45(W1): W300–W306. https://doi.org/10.1093/ nar/gkx444 11. Goeminne LJ, Argentini A, Martens L et al (2015) Summarization vs peptide-based models in label-free quantitative proteomics: performance, pitfalls, and data analysis guidelines. J Proteome Res 14(6):2457–2465. https:// doi.org/10.1021/pr501223t 12. Wieczorek S, Combes F, Lazar C et al (2017) DAPAR & ProStaR: software to perform statistical analyses in quantitative discovery proteomics. Bioinformatics 33(1):135–136. https://doi.org/10.1093/bioinformatics/ btw580 13. Gatto L, Lilley K (2012) MSnbase-an R/bioconductor package for isobaric tagged mass spectrometry data visualization, processing and quantitation. Bioinformatics 28 (2):288–289. https://doi.org/10.1093/bioin formatics/btr645 14. Wieczorek S, Combes F, Burger T (2018) DAPAR and ProStaR user manual. Bioconductor. https://www.bioconductor.org/packages/ release/bioc/vignettes/Prostar/inst/doc/ Prostar_UserManual.pdf?attredirects¼0

15. RStudio Team (2015) RStudio: integrated development for R. RStudio, Inc., Boston, MA. http://www.rstudio.com/ 16. http://stat.ethz.ch/R-manual/R-devel/ library/stats/html/hclust.html 17. Bolstad B (2018) preprocessCore: a collection of pre-processing functions. R package version 1.42.0. https://github.com/bmbolstad/ preprocessCore 18. Huber W, von Heydebreck A, Sueltmann H et al (2002) Variance stabilization applied to microarray data calibration and to the quantification of differential expression. Bioinformatics 18(Suppl 1):S96–S104. https://doi.org/10. 1093/bioinformatics/18.suppl_1.S96 19. Cleveland WS (1979) Robust locally weighted regression and smoothing scatterplots. J Am Statist Assoc 74(368):829–836. https://doi. org/10.1080/01621459.1979.10481038 20. Smyth GK (2005) Limma: linear models for microarray data. In: Gentleman R, Carey VJ, Huber W, Irizarry RA, Dudoit S (eds) Bioinformatics and computational biology solutions using R and bioconductor. Statistics for biology and health. Springer, New York, NY, pp 397–420. https://doi.org/10.1007/0-38729362-0_23 21. Giai Gianetto Q, Combes F, Ramus C et al (2016) Calibration plot for proteomics: a graphical tool to visually check the assumptions underlying FDR control in quantitative experiments. Proteomics 16(1):29–32. https://doi. org/10.1002/pmic.201500189 22. Ashburner M, Ball CA, Blake JA et al (2000) Gene ontology: tool for the unification of biology. Nat Genet 25(1):25–29. https://doi.org/ 10.1038/75556 23. Cox J, Mann M (2008) MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteomewide protein quantification. Nat Biotechnol 26(12):1367–1372. https://doi.org/10. 1038/nbt.1511 24. Giai Gianetto Q, Coute´ Y, Bruley C et al (2016) Uses and misuses of the fudge factor in quantitative discovery proteomics. Proteomics 16(14):1955–1960. https://doi.org/10. 1002/pmic.201600132

Chapter 16 Computation and Selection of Optimal Biomarker Combinations by Integrative ROC Analysis Using CombiROC Mauro Bombaci and Riccardo L. Rossi Abstract The diagnostic accuracy of biomarker-based approaches can be considerably improved by combining multiple markers. A biomarker’s capacity to identify specific subjects is usually assessed using receiver operating characteristic (ROC) curves. Multimarker signatures are complicated to select as data signatures must be integrated using sophisticated statistical methods. CombiROC, developed as a user-friendly web tool, helps researchers to accurately determine optimal combinations of markers identified by a range of omics methods. With CombiROC, data of different types, such as proteomics and transcriptomics, can be analyzed using Sensitivity/Specificity filters: the number of candidate marker panels arising from combinatorial analysis is easily optimized bypassing limitations imposed by the nature of different experimental approaches. Users have full control over initial selection stringency, then CombiROC computes sensitivity and specificity for all marker combinations, determines performance for the best combinations, and produces ROC curves for automatic comparisons. All steps can be visualized in a graphic interface. CombiROC is designed without hard-coded thresholds, to allow customized fitting of each specific dataset: this approach dramatically reduces computational burden and false-negative rates compared to fixed thresholds. CombiROC can be accessed at www.combiroc.eu. Key words Biomarker, Protein, miRNA, ROC curve, Statistical analysis, Combinatorial analysis, Multimarker signatures

1

Introduction In the area of diagnostic medicine, biomarkers have emerged as important tools in diagnostic, clinical, and research settings. Accurate markers are powerful tools to help clinicians to choose the most appropriate treatment, ultimately improving personalized patient management. In recent years, the identification of diseaseassociated signatures has been accelerated through the use of high-throughput omics techniques [1–3]. Unfortunately, most of the signatures identified would be too expensive to implement with the specificity (SP) and/or sensitivity (SE) required for diagnostics and routine clinical practices. The literature provides various statistical modeling strategies by which to combine biomarkers.

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_16, © Springer Science+Business Media, LLC, part of Springer Nature 2019

247

248

Mauro Bombaci and Riccardo L. Rossi

Threshold-based [4, 5], logistic regression [6, 7], and tree-based [8] methods are probably the most frequently used, whereas techniques such as Support Vector Machines [9] can be very useful in many multidimensional problems. Even if quite a number of statistical methods exist to combine biomarkers, their application in the clinical setting remains the domain of analytically skilled researchers, mainly due to the difficulty of extrapolating simplified, standardized, and interpretable results from these complex statistical strategies [10–12]. CombiROC is an easy-to-use tool implemented as a web application. It was developed to accurately determine optimal combinations of markers from diverse complex omics data. CombiROC takes advantage of combinatorial analysis and ROC curves to deliver immediate visual feedback, based on SP and SE graphs alongside ROC curves. CombiROC can automatically open data files, perform analyses appropriate for user-selected applications, create result files that can be opened as a spreadsheet, and create a summary of results for download and visualization as a PDF file or in tabular format for further processing with additional software.

2

Materials

2.1 Operating System and Hardware Requirements

CombiROC is a web tool developed in R and running on a Shiny server. As such, it does not need to be installed locally on the user’s machine. The CombiROC app can be executed from inside any recent web browser by accessing www.combiroc.eu. A working internet connection is needed. Computational and memory limitations are imposed by the R-server on which the CombiROC application is deployed. No fixed limitations have been set on the number of concurrent biomarkers that can be analyzed, but with the current web server, a maximum of ten markers is recommended (see Note 1). Combinations of larger numbers of biomarkers will be possible with a local R-based version of the application, but computational capacity and speed will depend on the resources of the user’s local machine.

2.2

The user provides the input dataset which should contain detection values for a maximum of ten biomarkers for two groups of samples (e.g., healthy donors and patients, or wild-type and mutant cells). Since the goal of CombiROC is to rank and select the best combinations of a limited number of biomarkers, these biomarkers should have been selected following an external high-throughput screening analysis performed by the user (see Note 2). Input detection values can be of various natures depending on the technology used to produce them. For example, detection values may be geneexpression values from microarrays, read-counts from RNA

Input Datasets

Diagnostic Accuracy by Multiple Marker Combinations

249

sequencing analyses, protein-expression values from mass spectrometry-based quantitative proteomics, fluorescence signals from protein arrays, or other types of data. Input datasets must be pre-processed before they can be uploaded to CombiROC. File format requirements are detailed below. 2.3

File Formats

CombiROC accepts two types of input files in which data is organized in columnar format: Tab-delimited text files (.txt), comma- or semicolon-separated values (.csv). Input data files can be prepared in any spreadsheet programme: each row (except the first one which contains the header) identifies a sample, and each column from the third on identifies a marker. The first two columns must be called “Sample_ID” and “Class,” subsequent columns contain the signal readout for featuredetection (Fig. 1). Text used as column headers must be identical to those listed in Table 1, the system is case-sensitive. Make sure you are using the English locale for decimal separators (use the dot “.” to separate decimals in numerical fields, and the comma “,” as grouping delimiter). Save the file as a plain text (. txt or .csv) file. CombiROC can then open and read the resulting file and analyze the data it contains.

Fig. 1 Sample illustration of input file formatting. Input files must be organized as outlined in the text and depicted here. The header (first row) is mandatory and must contain the first two columns as shown. The first column contains sample names and the second column the two classes the samples belong to (indicated by “A” or “B” here). Marker detection data (detection values) start in the third column

250

Mauro Bombaci and Riccardo L. Rossi

Table 1 Description of input data fields Column header

Description

Sample_ID

The name associated with the sample as defined in the “Sample_ID” field in your experiments (e.g., patient number)

Class

The name associated with the sample’s category as defined in the “class” field of your experiments (e.g., healthy and disease; tagged “A” and “B”)

Marker#

Data values (e.g., detection level FI, OD) associated with the features in that row of the data matrix. “#” is a progressive integer numeral

3

Methods

3.1 CombiROC User Interface

CombiROC is organized as a multistep workflow to select optimal combination(s) of markers through a simple analytical method based on the introduction of a double filter scoring method. The general workflow consists of a combinatorial analysis (Phase I) and an evaluation of sensitivity and specificity to obtain results (Phase II) (Fig. 2). These two phases are organized as four sequential steps, as described in Table 2. The main menu in the web application reflects this workflow in the first two menu items (Data; Analysis). Two additional items (Download; Accessories) will be detailed below (Fig. 2). In the following sections we describe the individual steps of the workflow based on the main menu structure in the graphic user interface.

3.2

Once the input data have been appropriately formatted, they can be uploaded to the application for analysis. Each page of the CombiROC application contains widgets (delimited areas) dedicated to specific functions or output display. The “Data upload” page shows the user three widgets (Table 3). In the “Enter data” widget, users can view the preloaded demonstration dataset by choosing the “Load demo data” option. Any uploaded dataset can be copied or downloaded as a csv or pdf by clicking on the “Copy,” “CSV,” or “PDF” buttons, respectively (in the upper right corner of the widget). Only entries displayed on-screen will be downloaded or printed, so if the user wants the entire file to be downloaded, all entries must be selected by clicking the “Show ALL entries” toggle on the left. If the input dataset is not automatically recognized, it may be necessary to specify the presence or absence of the header and the type of separator used (comma, semicolon or tab). If an error is detected, a warning will be displayed inside the widget itself.

Data

3.2.1 Upload

Diagnostic Accuracy by Multiple Marker Combinations

251

Fig. 2 Sequential flowchart and main menu of the CombiROC web interface. The flowchart is divided into two main phases. Phase I, multiple marker profiling data are uploaded and plotted; users can define the stringency of their test. Combinatorial analysis is performed and a list of markers and combinations of markers (Combo list) is obtained. In Phase II, various SE and SP thresholds can be visually explored and interactively adjusted to observe how many marker combinations survive the different cutoffs; the application then computes the outputs and automatically selects the best combinations (displayed in the Gold table) and pre-generates the corresponding ROC curves. The right panel schematically represents the main elements on an output page

3.2.2 Plots

After the data have been correctly uploaded to the application they can be viewed using a few plotting options. Box plots of the two data classes are displayed by default, but other options are available through the “Display” and “Options” sub-items on the Plot menu (Table 4). Both the box plot and marker plot displays are presented alongside tables summarizing the main statistics. The information includes distribution parameters which can be useful when determining the best cutoff values for subsequent steps (combinatorial analysis). To obtain box plot panels of the two classes plotted over the same range on the Y-axis, the “Display Options/Adjust Y-axis

252

Mauro Bombaci and Riccardo L. Rossi

Table 2 Overview of the main workflow steps Step

Description

1. Data upload and optional pre-processing

Open CombiROC and identify the data using the pull-down menu in the upper-left of the user interface. Use the Upload button to open single data files in CombiROC, users are offered a choice of simple data viewing with plotting and optional data processing methods

2. Set parameters and start combinatorial analysis

Use the Analyse button to analyse the data. Users can define the stringency of their test to identify features scoring as a hit in the assay. Results can be saved in a range of file formats (txt, csv, png, etc.)

3. Select best combinations

Select the Gold Combination to obtain the best combinations. CombiROC’s analytical approach is based on SE and SP filters, interpreted in terms of recognition frequency, optimizing the number of potentially interesting panels emerging from a previous combinatorial analysis step

4. Rank combinations using ROC curves

Use the ROC analysis button to visualize ROC curves which are automatically generated based on the previously-set parameters. Users can review results and download combinatorial analysis results and ROC curves in a number of file formats (txt, csv, png, etc.)

Table 3 Detailed procedure for data upload and pre-processing Widget

Description/function

Enter data

Here, users can either select preloaded demonstration datasets or upload their own formatted input dataset by selecting the Upload file option and the file from their local machine

Table

Immediately after loading, data will be visible in a tabular form in the table widget on the right; if the data are correctly formatted the header and data will be displayed as they appear in the original file. By default, only the first ten entries (rows) are displayed, but this number can be adjusted by clicking in the upperleft section of the Table widget

Details of uploaded dataset

The “Details of uploaded dataset” widget summarizes some details of the uploaded data: number of samples, markers and categories, data value type; it also indicates whether any values are missing

range” field can be used to type in the desired range (in the “lower value—comma—upper value” format, e.g., “0,1000”). Beware that when the Y-axis is adjusted, extremes values may no longer be visible. 3.2.3 Pre-processing (Optional)

Uploaded input data do not necessarily need further processing and the application can prompt the user through the “Pre-processing option” widget that data are ready to go. If desired, data can be

Diagnostic Accuracy by Multiple Marker Combinations

253

Table 4 Details of plotting options Plot menu sub-items

Description/functions

Display Box plot

Box plot is the default option. The appearance of box plots can be customized using the “Options” sub-menu Marker plots display the signal intensity for each individual sample. Individual markers must be selected using the “plot graph” button below the graph

Marker plot Options Change data option Change display option Change plot labels

(available for boxplots only) Used to toggle between plot canvas options to change boxplot styles (Tukey, Spear, Altman) and to display means and numbers of data-points (available for boxplots only) Used to change colors, plot orientation and to adjust height and width (available for boxplots only) Used to customize labels and label fonts

further processed to facilitate data analysis and model interpretation. Available processing options are: log transformation and two scaling methods (unit variance or Pareto scaling). Unit variance scaling, also known as auto-scaling, is widely applied and uses the standard deviation of the dataset as the scaling factor; Pareto scaling uses the square root of the standard deviation as the scaling factor. If the application detects problems with the processing method, it displays an error message directly inside the widget (see Note 3). On the right side of the page, the transformed and/or scaled data are displayed in tabular form. As for the other tables visualized in the application, by default only the first ten entries (rows) are displayed but this number can be adjusted by clicking in the upper-left section in the Table widget. 3.3

Analysis

3.3.1 Combinatorial Analysis

The first step of the actual analysis is to set the detection parameters. In this step, the user must set the parameters based on the experiment and the method used to generate the input data. These parameters (signal cutoff and minimum features) will de facto discriminate between noise and real signal (see Note 4). Parameters can be defined through the Graphic widget on the combinatorial analysis page (Table 5). As an example, upon loading the “Demo data (proteomics)” provided with the application, the user will see the pre-set cutoff value of “450” (in the specific context of the demo dataset, this value corresponds to the Fluorescence Intensity calculated from the mean value obtained for the buffer-control class plus three times the standard deviation). For these demonstration data, the minimum number of positive features is set to “1” (i.e., minimum stringency) which means that at least one marker must have an intensity of 450 to be considered a true signal. Most marker combinations from the proteomics demonstration data have a SE of

254

Mauro Bombaci and Riccardo L. Rossi

Table 5 Detailed description of the combinatorial analysis procedure Widget

Description/function

Graphics

Here, the user sets the test signal cutoff, based on the type of experiment used to generate the input data; as a rule, data-points with values above this threshold are considered positively detected (i.e., not noise). In addition, the minimum number of positive features (minimum features) reaching the previously-set cutoff must be defined. Upon clicking the “Distribution” button, a double histogram is displayed showing the distributions for SE and SP values satisfying the cutoffs set. Examination of these graphs helps to select the best SP&SE for use in further analysis

Mathematical details

Displays the formula used to calculate the number of countable combinations

Combo list

Lists all the possible combinations and their SE/SP values for both sample classes in tabular format

Fig. 3 Sensitivity and specificity evaluation chart. SE (blue bars) is defined as the true-positive rate, expressed as a percentage of your sample size. SP (black bars) is defined as the true-negative rate, expressed as a percentage of the control class. The X-axis corresponds to the number of each positive feature (left-wise blue bars for SE, right-wise black bars for SP). The Y-axis shows the distribution of SE and SP in percent intervals. Bars can be hovered over to display values

more than 40%, with a peak of 12 combinations in the 81–90% SE range; for the SP distributions, all combinations have a SP exceeding 50%, and a substantial number of them are above 80% (Fig. 3). Any evaluation and choice at this point strictly depends on these distributions and thus on the specific nature of the experiment that generated the data, and on the user’s ultimate goal. The

Diagnostic Accuracy by Multiple Marker Combinations

255

Table 6 Procedure to select the best marker combination Widget

Description/function

Explore SE and SP values

This widget displays two sliders for selecting SE and SP values. Users can move the sliders to obtain a reasonable trade-off between SE and SP

Gold combination bubble plot

In this widget, SE (y-axis) and SP (x-axis) are visually displayed on a bubble chart. All marker combinations are automatically plotted; the size of the bubbles is proportional to the number of markers in the combination (the bigger the bubble, the more markers)

Gold table

Produced once the SE/SP cutoffs have been set, this widget generates the corresponding list of marker combinations. Tables listing all Gold combinations are displayed

demonstration proteomics data SE and SP values (>40% and >80%, respectively) are good enough to produce a usable tradeoff (see Note 5). 3.3.2 Gold Combination

The array of marker combinations, “Combos,” obtained can be quite long. The best ones (the “Gold combinations”) can be selected by the user, taking the distributions displayed by the histogram graph in the previous step into account. For clarity, SE ¼ 40 and SP ¼ 80 were chosen for the proteomics demonstration dataset. The widgets in the Gold combinations page allow these cutoffs to be dynamically and visually set in the context of all possible marker combinations (Table 6). Marker combinations below the SE and SP thresholds are depicted as blue bubbles; yellow bubbles correspond to combinations above the SE/SP cutoffs. Initially, the user can move the sensitivity and specificity sliders and observe how many bubbles (¼marker combos) remain yellow at higher SE and SP values (Fig. 4). Yellow bubbles represent the Gold marker combinations with SE and SP values above the chosen thresholds; these combinations will subsequently be used to calculate the ROC curve (see Note 6).

3.3.3 ROC Analysis

Receiver operating characteristic (ROC) curves are used in medicine to determine a cutoff value for clinical tests. When creating a diagnostic test, a ROC curve helps to visualize and understand the trade-off between high SE and high SP when discriminating between clinically “normal” and clinically “abnormal” laboratory values. ROC curves can be calculated for combinations of selected biomarkers on the “ROC Analysis” page of the application. This page presents two main widgets, the “Select Combos” and the “Results” widgets (Table 7). Selecting the “Check to plot multiple curves” box in the Select Combos widget makes it possible to

256

Mauro Bombaci and Riccardo L. Rossi

Fig. 4 Sensitivity and specificity selection chart. Interactive bubble chart showing the SE (Y-axis) and SP (X-axis) of all the feature (marker) combinations and their distribution across the SE/SP landscape; the size of bubbles is proportional to the number of markers in the combination

Table 7 ROC analysis steps Widget

Description/function

Select Combo Allows selection between three results output options: (1) selection of single markers or specific marker combinations (default option) (2) selection of multiple features or combinations to view ROC comparisons (Check to plot multiple curves) (3) provide an unbiased estimate of the panel’s performance in a cross-validation procedure (Check to perform permutation test (single curves only)) Results:

ROC curves for individual markers and combinations selected in the previous step are calculated and graphically displayed. A table for each single marker or combination selected from the marker dialogue is also displayed, listing the values for Area Under the Curve, SE, SP and Optimal cutoff

l

Predictions

l

Permutation (Available for single curves only, not for overlays of ROC curves) Used to perform test multiple rounds of cross-validation applying different test sets to give a reliable estimation of model performance: a family of ROC curves is generated and performance results are averaged over the rounds. A tenfold CV strategy is used to compare the different models generated

Used to visualize a violin plot in the predictions section showing the “probability density” of the data at different values (Prediction probabilities are plotted for both classes based on the previously determined optimal cutoff)

Diagnostic Accuracy by Multiple Marker Combinations

257

visually compare multiple ROC curves for individual markers or combinations of the markers listed in the Gold table. The parameters table is automatically updated when each additional curve is overlaid in the multi-comparison, showing Area Under the Curve (AUC), SE, SP and optimal cutoff of the ROC curves displayed. Performance analysis returns the same ROC curve for the selected combination, overlaid with the corresponding crossvalidation to assess its performance. A table reports parameters for the whole cohort and the permutation models: accuracy (ACC) and error rate, as well as the corresponding SE, SP, and AUC. This analysis is available for single curves only, not for overlays of ROC curves. Finally, for each single ROC curve (single marker or combination) the four possible categories (false-negative, false-positive, true-negative (TN), and true-positive (TP)) can be visualized in a simple pie chart. This plot helps to determine the proportion of samples falling into the four possible quadrants, especially in those in the TN and TP prediction categories, and can be used to assess the power of the underlying markers. 3.4 Download and Accessories

4

The Download and Accessories sections provide ancillary functions, not necessarily connected to the analysis. From here, users can download datasets for the “demo data” in table format, and a printable pdf file of the tutorial. In addition, the logs of the application versioning, contacts and frequently asked questions are accessible.

Notes 1. It is not advisable to upload to CombiROC signatures containing data for more than ten biomarkers. Although, a signature can be composed of tens of biomarkers (genes, proteins, antigens, or other features), from a diagnostics point of view (clinical feasibility) efficient signatures tend to be composed of a low number of biomarkers. Thus, it is good practice to select only the top 5 or top 10 markers (based on the original screening process) and study combinations of those candidates. Moreover, ten markers can produce more than one thousand possible combinations, which makes the computational handling challenging. The combinatorial increase of combinations is exponential, and to deal with more than 210 combinations, higher processing power would be required. 2. CombiROC is a tool to rank existing signatures or groups of candidate biomarkers; it is not a screening tool. Screening, which can be performed by any high-throughput omics method, should be completed before the combinatorial

258

Mauro Bombaci and Riccardo L. Rossi

analysis; it is highly dependent on the omics method used (i.e., proteomics, transcriptomics, or other). 3. The choice of data pre-processing method depends on the biological/clinical data available. A variable/biomarker could contain one or a few extreme measurements which may dominate over the other measurements, indicating that this variable/marker departs from normality. One way to achieve a more symmetric data distribution is by transformation, such as “log transformation.” Data scaling is equivalent to giving all the variables equal importance; hence, this procedure could be particularly useful in “omics” datasets where variables are affected by large-scale effects. Data can be scaled in a number of ways, but the most common techniques are “unit variance” or “pareto scaling.” 4. The test signal cutoffs and minimum features apply to the original dataset values and NOT to the SE and SP values that are subsequently calculated. The user must decide on the cutoffs based on the DETECTION threshold for the system used. 5. Once the cutoffs have been set, the SE/SP distribution can be assessed by viewing the histogram plot to choose the best interval (e.g., where SE and SP are maximized) for use in subsequent steps. 6. Bubble plots can be used to rank and select the combination (s) performing best. This step is not mandatory, indeed all combinations can be retained, taking the total number of combinations available and their performance into account. This step can be done iteratively (it is probably the most interactive, trial-and-error step in the entire analysis), and this is where the power of CombiROC expresses itself. References 1. Pfaffl MW (2013) Transcriptional biomarkers. Methods 59(1):1–2. https://doi.org/10. 1016/j.ymeth.2012.12.011 2. Janvilisri T, Suzuki H, Scaria J et al (2015) High-throughput screening for biomarker discovery. Dis Markers 2015:108064. https:// doi.org/10.1155/2015/108064 3. Sotiriou C, Piccart MJ (2007) Taking geneexpression profiling to the clinic: when will molecular signatures become relevant to patient care? Nat Rev Cancer 7(7):545–553. https://doi.org/10.1038/nrc2173 4. Hainard A, Tiberti N, Robin X et al (2009) A combined CXCL10, CXCL8 and H-FABP panel for the staging of human African trypanosomiasis patients. PLoS Negl Trop Dis 3

(6):e459. https://doi.org/10.1371/journal. pntd.0000459 5. Turck N, Vutskits L, Sanchez-Pena P et al (2010) A multiparameter panel method for outcome prediction following aneurysmal subarachnoid hemorrhage. Intensive Care Med 36 (1):107–115. https://doi.org/10.1007/ s00134-009-1641-y 6. Fung KYC, Tabor B, Buckley MJ et al (2015) Blood-based protein biomarker panel for the detection of colorectal cancer. PLoS One 10 (3):e0120425. https://doi.org/10.1371/jour nal.pone.0120425 7. Li J, Zhang Z, Rosenzweig J et al (2002) Proteomics and bioinformatics approaches for identification of serum biomarkers to detect breast cancer. Clin Chem 48(8):1296–1304

Diagnostic Accuracy by Multiple Marker Combinations 8. Bombois S, Duhamel A, Salleron J et al (2013) A new decision tree combining Abeta 1-42 and p-Tau levels in Alzheimer’s diagnosis. Curr Alzheimer Res 10(4):357–364. https://doi. org/10.2174/1567205011310040002 9. Zhang F, Deng Y, Drabier R (2013) Multiple biomarker panels for early detection of breast cancer in peripheral blood. Biomed Res Int 2013:781618. https://doi.org/10.1155/ 2013/781618 10. Buyse M, Michiels S, Sargent DJ et al (2011) Integrating biomarkers in clinical trials. Expert

259

Rev Mol Diagn 11(2):171–182. https://doi. org/10.1586/erm.10.120 11. de Gramont A, Watson S, Ellis LM et al (2015) Pragmatic issues in biomarker evaluation for targeted therapies in cancer. Nat Rev Clin Oncol 12(4):197–212. https://doi.org/10. 1038/nrclinonc.2014.202 12. Kramar A, Faraggi D, Fortune´ A et al (2001) mROC: a computer program for combining tumour markers in predicting disease states. Comput Methods Prog Biomed 66 (2–3):199–207. https://doi.org/10.1016/ S0169-2607(00)00129-2

Chapter 17 PanelomiX for the Combination of Biomarkers Xavier Robin Abstract Proteomics has allowed the discovery and validation of a massive number of biomarkers. However most of them suffer from insufficient specificity and sensitivity and therefore didn’t translate to clinical practice. Combining biomarkers with different properties into panels can be an efficient way to bypass these limitations and facilitate the translation of biomarkers into clinical practice. Key words Biomarkers, Panel, Combination of biomarkers, Machine learning, Clinical study

1

Introduction The number of biomarkers discovered by proteomics methods keeps increasing over the years. However only very few of them have been translated into clinical practice. An explanation for this translation gap is that most biomarkers lack the required specificity that would be required to make useful decisions. To overcome these limitations, several biomarkers can be combined into a panel of biomarkers, and a decision is made based on a linear or nonlinear expression of the individual marker measurements, often with increased specificity and/or sensitivity. PanelomiX is a web-based tool to compute panels of biomarkers based on thresholds [1]. It searches for the combination of biomarkers and thresholds that optimizes a user-defined criteria that can be customized for a specific clinical question. It provides options to speed up the search when the search space becomes too large. The results are a set of biomarkers with clear and easily interpretable thresholds, so that the classification of a patient can be understood in terms of the contribution of each individual biomarker. In this chapter, we show how to generate a panel of biomarkers with PanelomiX. To serve as an example, we provide a dataset of 113 patients with aneurysmal subarachnoid hemorrhage (aSAH) and investigate the outcome of these patients 6 months after aSAH.

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8_17, © Springer Science+Business Media, LLC, part of Springer Nature 2019

261

262

Xavier Robin

One clinical measurement (WFNS) and two biological markers (S100B and NDKA) are reported [2, 3]. Finally, we show how to analyze the results with the R statistical environment and perform a ROC analysis with pROC.

2

Materials

2.1 System Requirements for Panelomix

PanelomiX is available as a web-based tool and can be used with any computer with an Internet connection, a modern web browser such as Mozilla Firefox, Google Chrome, Apple Safari or Microsoft Internet Explorer or Edge. A screen resolution of 1280x720 or greater is recommended.

2.2 System Requirements for pROC

For the ROC analysis with pROC [1], version 3.0 or greater of the R statistical environment is needed [4]. It is available for Microsoft Windows, MacOS, and Linux and can be downloaded for free from the CRAN repository at cran.r-project.org.

2.3

Most common formats are recognized, in particular Comma Separated Values (CSV) and Microsoft Excel (XLS and XLSX). If you are preparing your data in a spreadsheet editor such as Microsoft Excel, keep the following requirements in mind:

Input Data

1. In the first row, enter the title of the column. Do not use more than one row for titles and set a title on each column. Do not merge cells. 2. The following rows contain the patient data, one patient per line and one line per patient (wide format). 3. Missing values must be encoded with NA or a blank (empty) cell. No other text (even a dot) are permitted. 4. Cell formatting is allowed as long as it doesn’t change the table structure. For example, bold, colors (text and background), fonts, or alignment is OK, but merging cells or leaving blank cells is not. 5. Only the first sheet is considered in multi-sheets documents. The selection status (active sheet) is not taken into account.

3

Methods

3.1 User Creation and Login

1. Start by opening the PanelomiX home page in your browser. Type www.panelomix.net in the address bar. 2. If it is the first time you are using PanelomiX, you need to register your account. In the bottom right panel, under the Register title, enter your desired username and your password

PanelomiX for the Combination of Biomarkers

263

twice. The optional email address will be used to send notifications when panels complete, and to reset a forgotten password. 3. Alternatively if you don’t want to register but only give PanelomiX a try, you can use the bottom left panel under guest to login with the guest account (see Note 1). 3.2 Creating and Managing Panels

After registering and logging in, you will be redirected to the main page that lists existing panels (once you created some) and allows to create new panels and manage existing panels. The main elements of this page are described in Fig. 1. 1. The list of panels on the left menu (Fig. 1A) allows navigating between panels from any page in PanelomiX. 2. Click on the Create a new panel button (Fig. 1B) to create a new panel. A new field will appear at the top of the list below (Fig. 1C) and ask for a name to identify the panel. Enter a descriptive name for the panel and click on Create to create it. 3. The Edit the panel list button (Fig. 1D) allows to copy, rename, and delete existing panels. Click on it to make the corresponding buttons appear (Fig. 1E). 4. The Help page is available throughout PanelomiX (Fig. 1C) and gives some additional information on the methods used and how to use PanelomiX. To get started with PanelomiX, click on the Create a new panel button (Fig. 1B) to start a new panel. Enter a name and click Create (Fig. 1C). A new panel is created and you are redirected to the Data Entry page.

Fig. 1 Panel list page. (A) Persistent navigation menu. (B) Button to create a new panel. (C) When pressed, a new field appears to choose a name. (D) Button to edit panels. (E) When pressed, new buttons appear next to the panel list, which allows to edit the name, copy and delete a panel, respectively. (F) Additional help and algorithm description is available on most pages

264

3.3

Xavier Robin

Data Entry

1. The first step to create a panel is to upload the data to PanelomiX. Most common formats are recognized, in particular Comma Separated Values (CSV) and Microsoft Excel (XLS and XLSX). If you are preparing your data in a spreadsheet editor such as Microsoft Excel, please refer to subheading 2.3 to format the data appropriately. 2. Data can be provided as direct upload of a saved file (the file format is detected automatically from the file extension), by copying and pasting directly from the spreadsheet editor (using the Copy from spreadsheet button), or copied directly from an other panel. 3. Finally a sample dataset featuring 113 patients with an aneurysmal subarachnoid hemorrhage is available. 6-month outcome is measured and reported as good or poor. One clinical measurement (WFNS) and two biological markers (S100B and NDKA) are reported [2, 3]. We will use it as an example in the next sections of this chapter. 1. Click on the “Aneurysmal subarachnoid hemorrhage example data” button. 2. Click on the Submit data button to continue.

3.4 Selection of Grouping Variables and Biomarkers

The details tab (Fig. 2) allows the selection of the grouping variable, as well as the list of biomarkers to be included in the panel.

3.4.1 Select Fixed and Test Markers

There are two ways to introduce biomarkers in a panel: test or fixed. In addition, the bottom of the page provides a quick summary of the data uploaded in the previous step (Fig. 2A). Test markers (Fig. 2B) will be tested for inclusion in the panel. They will be included only if they improve the classification. By contrast, Fixed markers (Fig. 2C) will always be included in the panel, even if they worsen the classification. Select biomarkers in the Fixed markers list if you already know for sure they will improve the classification or if you want to include them anyway. Otherwise select them as Test markers (see Note 2).

3.4.2 Choose Grouping Variables

The grouping variable (Fig. 2D) determines the two groups that will be discriminated by the markers. A typical example of grouping variable is the diagnosis variable. Typically, a grouping variable can take only a few possible values which you will then need to select as control (negative) or case (positive). If you are following the aneurysmal subarachnoid hemorrhage example, make sure that 113 patients and 6 variables were detected by checking the numbers at the bottom of the page (Fig. 1A).

PanelomiX for the Combination of Biomarkers

265

Fig. 2 (A) The panel variable selection page displays a summary of the data uploaded in the previous step. (B) Fixed and (C) test markers must be selected, as well as (D) the grouping variable and classes

1. Select wfns as a fixed marker that will be included in all panels. 2. Select the s100b and ndka variables for inclusion as test markers which will be added only if they improve the classification. 3. Finally select outcome as grouping variable, with Good prognosis patients as controls and Poor prognosis as case observations. 4. Click on the Submit details button to proceed. 3.5

Options

The options page enables the selection of important options that will have an impact on what combinations are searched and how they will be validated. 1. At the top of the page, the Estimated time icon (Fig. 3A) gives a rough estimate of the time that will be needed to perform all the panel computations, including cross-validation and pre-filtering if those are selected. This time is subject to many variations such as server load, or some characteristics of the data, but should give a good idea of the order of magnitude of computation time (second, minutes, hours, days, or years). Together with the markers selection in the previous tabs, options on this page will significantly influence the running time of the panel. For testing purposes it is recommended to target run times of a few minutes or less, while it is not uncommon to run the final panel for hours or days. It should be noted

266

Xavier Robin

Fig. 3 The panel options page displays a summary of the data uploaded and markers selected in the previous steps (bottom of the page, not shown). (A) The estimated time to run the panel is influenced by options in this page, and updated automatically (allow a few seconds to refresh the data). (B) Maximum panel size, (C) optimization rule, and whether to (D) perform cross-validation and (E) pre-filtering, as well as (F) the desired output format can be changed

that panels running for more than a few days may not complete. 2. The Maximum panel size option (Fig. 3B) selects the maximum size of the panels. For instance, to search for panels containing 1, 2, or 3 biomarkers, it must be set to 3 (see Note 3). 3. The next step is to select the optimization rule for the panel (Fig. 3C). By default, the Accuracy is optimized, which maximizes the sum of sensitivity and specificity. It is also possible to optimize sensitivity (with a Minimum specificity) or specificity

PanelomiX for the Combination of Biomarkers

267

(with a Minimum sensitivity). In those two cases, the minimal specificity or sensitivity is given in the adjacent text field as a fraction (between 0 and 1). Panels that doesn’t reach this performance will be ignored. If you are following the aneurysmal subarachnoid hemorrhage example, select a maximum of 3 panels, and a minimum specificity of 0.95. 4. Cross-validation. Cross-validation (CV) is used to assess the stability of panels, and evaluate the expected performance they would have on a test set. It is done by randomly splitting the data set into training and testing sets, then repeating the training process with each test set removed, one by one. The panel that was trained on the reduced training set can then be tested on the test set, and its performance is expected to be similar to the performance it would have on an independent test set. When cross-validation is enabled by clicking on the yes button (Fig. 3D), the Cross-validation folds field determines in how many sets the data will be divided into, often termed k for k-fold cross-validation (see Note 4). Higher numbers will result in a slower cross-validation as there are more splits to train, but have the advantage that the training sample of each fold is more representative of the data set. This is especially significant with small data sets, where removing 10 or 20% can significantly change the performance of the training procedure. The Cross-validation repeats field indicates how many time to repeat the whole cross-validation. Higher numbers give better estimates of the stability and performance, but also longer computation time. Choose yes to enable cross-validation and leave the two fields Cross-validation folds and Cross-validation repeats to their default values of 10 and 5 respectively. 5. Pre-filtering. Pre-filtering is sometimes required to speed up the computation. Once enabled by clicking on the yes button (Fig. 3E), a Random forest classifier is trained with R’s randomForest[5, 6] package and the following filters can be selected: (a) Number of markers: only this number of markers will be used in the computations. If more markers are selected in the details page, only those being selected most often as split points of the decision trees in the random forest will be retained. (b) Number of thresholds: several thresholds are tested for each markers. By default, all the thresholds corresponding to local maxima of the ROC curve are used. If an integer N is defined here, only the N thresholds being selected most often as split points of the decision trees in the random forest for a given marker will be retained.

268

Xavier Robin

Leave the Perform randomForest filtering field with the no option selected. 6. Output. The output options define how the results will be presented. The following options are available (Fig. 3F): (a) HTML only produces a simple HTML output. This is the fastest option, best used during early screening steps. (b) PDF report generates a PDF file summarizing the results (including cross-validation if this option was selected above). A ROC analysis is performed without computing confidence intervals. (c) PDF report with confidence intervals is similar to the previous one, but ROC curves come with additional confidence intervals for (p)AUC, sensitivity and specificity. This is most complete but also the slowest option. Check the Skip cross-validation of individual biomarker box to perform the cross-validation only on the panel, and not on each separate biomarker (see Note 5). To finish with the aneurysmal subarachnoid hemorrhage example, leave all fields with their default values and click on Submit details. 3.6 Running the Panel

3.7

Results

The Run page summarizes the data provided, marker selection and panel options. It shows the estimated time that will be needed to complete the panel. If the time is longer than a few days, go back to the details or options page and select fewer markers, more aggressive Random Forest filtering options or fewer cross-validation repeats. Click on Run the panel. A page showing the progress appears and is automatically reloaded every 30 seconds. Wait until the results page appears. 1. Once the panel is completed the results page appears (Fig. 4). The page displays a quick summary of the data that was used to build the panel (Fig. 4A) and results (Fig. 4B), which can also be viewed in text format by clicking on the View Simple Text Report (Fig. 4C) button. 2. If a PDF option was selected (Subheading 3.5, step 6, Fig. 3F), a report in PDF format can be viewed by clicking on the View PDF Report button (Fig. 4D). It shows data about the performance of the panel, comparison with the best biomarker, stability of the cross-validation and a comparison with other combination methods. A detailed explanation of these results and their interpretation is available in our original paper [3]. 3. The third available format is a machine-readable file that can be read into the R statistical environment (Fig. 4E). An example of how to use this file for further analysis of the results is shown in the next section of this chapter.

PanelomiX for the Combination of Biomarkers

269

Fig. 4 The panel results page displays the data that was used to build (A) the panel and (B) results. The data can be downloaded in (C) plain text, (D) as a detailed PDF with cross-validation and (E) in R Data format for the R statistical environment

3.8 Further Analysis with pROC

In this final section we will download the results in a format that can be read by the R statistical environment [4] and show a possible analysis of the results with pROC [1]. The possibilities are virtually unlimited but we will show a representative analysis of the panel and cross-validation, with the analysis of the partial area under the ROC curve (pAUC). 1. On the results page, click on the R icon (Fig. 4E) to download the results as an R data object and save it on your computer. 2. Open an R session and set the working directory to the folder containing the downloaded file. Use the setwd function or the File, Change dir. . . menu if you are using the GUI on

270

Xavier Robin

Windows. If you are using Rstudio, you can either use the Session menu, and Set Working Directory, or create a new project with in the folder where you downloaded the file with the File, New Project, and select Existing directory and navigate to the folder containing the downloaded file. 3. Load the results file. If you used the default name, this is done with the following command: > load("PanelomiX-result.RData")

This loads two objects in memory: panels which stores the panels that were found on the whole dataset, and panels.cv that store cross-validation information. 4. We can investigate how many panels were found: > length(panels) [1] 1

A single panel was found by PanelomiX with the parameters given. 5. We can then concentrate on this panel and check what thresholds were tested: > panel str(panel$possible.thresholds) List of 3 $ s100b: num [1:18] 0.065 0.075 0.085 0.095 0.105 0.115 0.135 0.155 0.205 0.245 ... $ ndka : num [1:26] 3.44 5.11 7.24 8.16 8.72 ... $ wfns : num [1:4] 1.5 2.5 3.5 4.5

6. Next we can calculate the value of the panel for each patient. This value, or score, will be required later to plot a ROC curve with pROC. We define a predict.panel function, which loops over all the marker included in the panel and calculates the PanelomiX score for every patient. > predict.panel library(pROC) > # Plot the ROC curve > plot(roc(panel$train.data[[panel$response]], +

patient.scores,

+

levels = panel$levels,

+

percent = TRUE,

+

partial.auc = c(90, 100)),

+

max.auc.polygon = TRUE,

+

print.auc = TRUE,

+

print.thres=TRUE)

8. Cross-validation Cross-validation results are saved in the panels.cv variable and organized by cross-validation repeat and fold (see Note 6). First we need to update the prediction function to use the test data:

272

Xavier Robin

> predict.cv 10 TPM) as input file. Additional input parameters: “Does file contain header?”: leave “Yes” (default value) as there is a header in your dataset; “Column number on which apply the comparison”—this will be column No. 1 (named “Gene”), where the gene identifiers (i.e., Ensembl gene or ENSG ID) are listed (“c1” is the default value); for the “Enter the name of this list” option: enter the type of data: “Heart (RNAseq).” Now the protein profiles based on IHC experiments can be added to the “List to compare” menu item below. In the “Enter your list” option: click the file browser and select “Retrieve tissue-specific expression data” that corresponds to the IHC-based protein expression information created at the first step. For the additional parameters: “Does your input file contain header?”: Leave button set to “Yes” as there is a header in this file; “Column number on which apply the comparison”: leave “c1” (default value), as column No. 1 contains the ENSG ID; “Enter the name of this list”: enter an informative name, e.g., Heart (IHC). Then run “Jvenn.” Two outputs are created in the history panel: a textual output and a Venn diagram. Click the “eye” icon to view each one in the central panel. The textual output lists each element and indicates whether it is shared between the two datasets or not (three columns). Thus, for example, the proteins identified in both datasets are reported in column No. 3. The “Venn diagram” history item can be used to visualize genes/proteins that are in common between the two data sources (RNAseq and IHC), in this case there are 931 elements in common (Venn diagram shown in Fig. 1). 4. The next step is to refine the selection to biomarkers that are highly specific to the heart using additional expression data

In Silico Strategy for Biomarker Selection

285

from HPA (e.g., RNA tissue category, RNA tissue specificity abundance). In the tool panel, select the “Add expression data (RNAseq or Immuno-assays)[Human Protein Atlas]” (Proteore -> “HUMAN ANNOTATION” subsection). Under “Select your file,” select the output previously created (“Venn diagram text output”) as input file; “Column IDs (e.g : Enter c1 if ENSG ID are in column n 1)”: enter “c3” in which the ENSG ID identified in both datasets are reported. In the “RNAseq/Ab-based expression data” menu, select the following features: “Gene name,” “Gene description,” “RNA tissue category,” and “RNA tissue-specificity abundance.” Run the tool by clicking the “Execute” button. We now wish to focus on transcripts that have been classed as “tissue enriched” (expression in one tissue at least fivefold higher than all other tissues), “group enriched” (fivefold higher average TPM in a group of two to seven tissues compared to all other tissues) and “tissue enhanced” (fivefold higher average TPM in one or more tissues/cell lines compared to the mean TPM for all tissues) according to the HPA definitions [12]. This information will be listed in column No. 6 “RNA tissue category” of the newly created file. Select the “Filter by keywords and/or numerical value” in the tools panel. In the central panel, the submission form is displayed. It should be completed as follows: as input file, keep the “Add expression data on data 7” generated in the last step (normally proposed by default); click the “+ Insert Filter by keywords” box below and enter the following parameters: “Column number on which you want to apply the filter” set to “c6,” corresponding to the “RNA tissue category” column in the previously created file. In the “Filter by keyword” box, select the “Enter keywords (copy/ paste)” mode, then enter the following keywords “enriched enhanced” (present in column No. 6 and that cover the three RNA tissue categories of interest), then click the “Execute” button. Click the “eye” icon to view the file containing the filtered lines (the last one created in the history panel, the columns of interest are Nos. 3–7, which bear the following headings: “Heart (RNAseq)_Heart (IHC),” “Gene,” “Gene description,” “RNA tissue category,” “RNA TS TPM”). Note that 115 candidates are considered to have significantly higher expression in heart muscle according to HPA criteria. 5. The next step in the selection process is to retrieve protein features from the neXtProt database to select biomarker candidates without transmembrane domains and which are reported to be localized in the cytoplasm. Since HPA only considers ENSG identifiers (i.e., related to the gene) and neXtProt uses Uniprot identifiers (i.e., related to the protein), we first need to map the Ensembl identifiers contained in our list of candidates

286

Lien Nguyen et al.

to their corresponding Uniprot accession number. To do so, select the “ID_Converter” tool from the tool panel (ProteoRE -> “DATA MANIPULATION” subsection). Choose the previously generated file as input. Set the “ID_converter” parameters as follows: the “column number of IDs to map” corresponds to column No. 3, where the Initial ENSG ID are listed, enter “c3” (ENSG ID shared between IHC and RNAseq data); “Species” is set to “Human (homo sapiens)” by default; in the “Type/source of IDs” menu, scroll down and choose “Ensembl gene ID” as the source-identifier listed. “Target ID type” corresponds to the type of ID you want to map to; in this case, we selected “Uniprot accession number” and “Uniprot ID.” Click “Execute,” and in the history panel, click on the “eye” icon to view the output created and note that two further columns have been added (column No. 8 and No. 9). To collate protein features, select the “Add protein features [neXtProt]” tool from the tool panel (ProteoRE -> “HUMAN ANNOTATION” subsection). Choose the file previously generated as input file (“ID Converter on data” is proposed by default). Choose the following “Add protein features” parameters: “Column IDs (e.g : Enter c1 for column n 1)”: Enter “c8,” where the Uniprot accession numbers are listed in your input file; “Type of IDs”: leave “Uniprot accession number”; in the “Select features” menu, select “Number of transmembrane domains,” and “Subcellular Location” by clicking on the appropriate checkboxes. Run the tool by clicking the “Execute” button. Click the “eye” icon to view the file containing the latest output created in the history panel, note that three columns have been added to your list containing the following information: “TMDomain” (number of transmembrane domains) (column No. 10), “SubcellLocations” (column No. 11), and “Diseases” (column No. 12 added by default). From this file, you can identify proteins reported as localized in the cytoplasm (or cytosol) and having no transmembrane domain by re-running the “Filter by keywords and/or numerical value” from the tools panel. Fill in the submission form as follows: keep the output previously generated as input file (“Add information from neXtProt” is proposed by default); click the “Filter by keywords” box below and enter the following parameters: “Select an operator to combine your filters (if more than one)” set to “AND” (“OR” by default), this allows both filters to be combined for this file; click the “Insert Filter by keywords” box below, select “Enter keywords (copy/paste)” mode, then type keywords such as “cytoplasm cytosol” (which are present in column No. 11) (keywords must be separated by a space or a tab character); “Column number on which to apply the filter” set to “c11,” which corresponds to the “SubcellLocations” column in the

In Silico Strategy for Biomarker Selection

287

file created by the “Add protein features [neXtProt]” tool. Then, to apply the second filter (i.e., the number of transmembrane domain), click the “Insert Filter by numerical value” box below and enter the following parameters: “Column number on which to apply the filter” set to “c10,” which corresponds to the “TMDomain” column. In the “Filter by value” box: set “Filter by value” to the “¼” symbol and for the “Value”: enter “0” using the number pad, then run the program using the “Execute” button. Click the “eye” icon to view the file containing the filtered lines (the last one created in the history panel). The resulting output contains 44 proteins which are highly enriched in heart tissue, have a cytoplasmic location, and no transmembrane domains. 6. The final step consists in checking whether these candidate biomarkers have previously been observed in the human plasma and heart proteome builds contained in Peptide Atlas. The underlying idea is to inform the end-user as to whether a protein has been observed in human plasma, to allow them to select the most appropriate surrogate peptides representing a target protein in a targeted experiment (LC/MRM–MS analysis). From the tools panel, select the “Get MS/MS observations in tissue/fluid [Peptide Atlas]” (ProteoRE -> “HUMAN ANNOTATION”). Select the 44-protein file previously generated (e.g. “Filter by keywords and/or numerical value on Add_information_from_neXtProt - Filtered lines”) as input in the file selector (“Enter your IDs (UniProt Accession number only)”), enter “c8” for the “Column of IDs”, check the “Human plasma 24/08/2018” and “Human heart 24/08/ 2018” boxes in the “Proteomics dataset (biological sample)” menu. Run the tool. View the result: the columns “Human_Heart_24-08-2018” (No. 13) and “Human_Plasma_24-082018” (No. 14) report the number of peptide spectrum matches (PSM) observed by LC-MS/MS in human plasma and heart samples. Proteins for which no peptides were detected in plasma (i.e., with a value of “NA”) can now be removed using the “Filter by keywords and/or numerical value” tool with the following parameters: keep the output previously generated as input file; click the “Insert Filter by keywords” box below and enter the following parameters: “Column number on which to apply the filter” set to “c13” corresponding to the “Nb of times peptide Observed_Human_Plasma” column; in the “Filter by keyword” box, select the “Enter keywords (copy/paste)” mode, type the keyword “NA,” then click the “Execute” button. The final list contains 19 candidate biomarkers known to be detectable by MS in human plasma (Table 1).

288

4

Lien Nguyen et al.

Notes 1. For people who are not familiar with the Galaxy framework, we recommend that they read the introductory tutorial: http:// galaxyproject.github.io/training-material/topics//introduc tion/. 2. For help and support specific to ProteoRE, please e-mail [email protected]. ProteoRE can currently be used in two ways: (a) Unregistered user: you can start a session by uploading your data (e.g., proteomics identification results in tabular format). Analysis can be performed using the available tools. Note that the allocated storage space for unregistered users is limited to 20 Mb and data cannot be conserved between sessions. For frequent use and/or numerous analyses, we highly recommend that you create an account (see below). (b) Registered user (password-protected access): creating an account is the best way to keep track of and record your analyses and your workflows. In addition, additional storage space is available to registered users (up to 20 Gb). To create an account, click on “Account registration or login” on the main menu bar (upper part of the ProteoRE central panel), click on “registration” and fill out the form. Once the form is submitted, your account is created. This Galaxy platform is constantly evolving and being improved, as a result, some minor interface changes may occur, and the online version may not strictly comply with what is described in this chapter, which presents the state at the time of writing. In addition, results may differ slightly due to updates of the public databases searched.

Acknowledgments This work was partly supported by grants from the “Investissement d’Avenir Infrastructures Nationales en Biologie et Sante´” program (ProFI project, ANR-10-INBS-08 and French Bioinformatics Infrastructure grant ANR-11-INBS-0013) and by the French National Research Agency (GRAL project, ANR-10-LABX-4901). We would like to thank the following for their contributions to the design and beta-testing of these tools: Benoit Gilquin, Maud Lacombe, Lisa Perus, Marianne Tardif. We are very grateful to Christophe Caron and Christophe Bruley for their contribution at the earliest stages of the ProteoRE project, and for their advice and continuous support.

In Silico Strategy for Biomarker Selection

289

References 1. Anderson NL (2010) The clinical plasma proteome: a survey of clinical assays for proteins in plasma and serum. Clin Chem 56(2):177–185. https://doi.org/10.1373/clinchem.2009. 126706 2. Mcdermott JE, Wang J, Mitchell H et al (2013) Challenges in biomarker discovery: combining expert insights with statistical analysis of complex omics data. Expert Opin Med Diagn 7(1):37–51. https://doi.org/10.1517/ 17530059.2012.718329 3. Greene CS, Troyanskaya OG (2010) Integrative systems biology for data driven knowledge discovery. Semin Nephrol 30(5):443–454. https://doi.org/10.1016/j.semnephrol. 2010.07.002 4. Foster KR, Koprowski R, Skufca JD (2014) Machine learning, medical diagnosis, and biomedical engineering research—commentary. Biomed Eng Online 13:94. https://doi.org/ 10.1186/1475-925X-13-94 5. Kulasingam V, Pavlou MP, Diamandis EP (2010) Integrating high-throughput technologies in the quest for effective biomarkers for ovarian cancer. Nat Rev Cancer 10(5):371–378. https://doi. org/10.1038/nrc2831 6. Robinson WH (2013) Mechanistic biomarkers for clinical decision making in rheumatic diseases. Nat Rev Rheumatol 9(5):267–276. https://doi.org/10.1038/nrrheum.2013.14 7. Maes P, Donadio-Andre´i S, Louwagie M et al (2017) Introducing plasma/serum glycodepletion for the targeted proteomics analysis of cytolysis biomarkers. Talanta 170:473–480. https://doi.org/10.1016/j.talanta.2017.04. 042 8. Prassas I, Chrystoja CC, Makawita S et al (2012) Bioinformatic identification of proteins with tissue-specific expression for biomarker discovery. BMC Med 10:39. https://doi.org/ 10.1186/1741-7015-10-39 9. Martı´nez-Morillo E, Garcı´a Herna´ndez P, Begcevic I et al (2014) Identification of novel

biomarkers of brain damage in patients with hemorrhagic stroke by integrating bioinformatics and mass spectrometry-based proteomics. J Proteome Res 13(2):969–981. https://doi.org/10.1021/pr401111h 10. Carr SA, Abbatiello SE, Ackermann BL et al (2014) Targeted peptide measurements in biology and medicine: best practices for mass spectrometry-based assay development using a fit-for-purpose approach. Mol Cell Proteomics 13(3):907–917. https://doi.org/10.1074/ mcp.M113.036095 11. Goecks J, Nekrutenko A, Taylor J et al (2010) Galaxy: a comprehensive approach for supporting accessible, reproducible, and transparent computational research in the life sciences. Genome Biol 11(8):R86. https://doi.org/10. 1186/gb-2010-11-8-r86 12. Uhlen M, Hallstro¨m BM, Lindskog C et al (2016) Transcriptomics resources of human tissues and organs. Mol Syst Biol 12(4):862. https://doi.org/10.15252/msb.20155865 13. Lane L, Argoud-Puy G, Britan A et al (2012) neXtProt: a knowledge platform for human proteins. Nucleic Acids Res 40(Database issue):D76–D83. https://doi.org/10.1093/ nar/gkr1179 14. Farrah T, Deutsch EW, Hoopmann MR et al (2013) The state of the human proteome in 2012 as viewed through PeptideAtlas. J Proteome Res 12(1):162–171. https://doi.org/10. 1021/pr301012j 15. Pelsers MMAL, Hermens WT, Glatz JFC (2005) Fatty acid-binding proteins as plasma markers of tissue injury. Clin Chim Acta 352 (1–2):15–35. https://doi.org/10.1016/j. cccn.2004.09.001 16. Lin S, Yokoyama H, Rac VE et al (2012) Novel biomarkers in diagnosing cardiac ischemia in the emergency department: a systematic review. Resuscitation 83(6):684–691. https:// doi.org/10.1016/j.resuscitation.2011.12. 015

INDEX A Alkylation.......................... 26, 30, 42, 46, 54, 69, 72–73, 92, 95, 96, 98–99, 109, 116, 219 Alzheimer’s disease (AD)............v, 90, 91, 105, 111, 164 Analyte stability ................................................................. 8 Automation ...................................................... 25, 89–111

B Bioanalysis ......................................................3–4, 7, 8, 19 Bioinformatics ....................v, 56, 61, 207, 220, 225, 277 Biomarker discovery ................................. v, 1–19, 23–36, 39–49, 51–63, 89–111, 124, 129–147, 151–160, 185–201, 226, 277 evaluation..................................................12, 130, 205 panel.......................................... 2, 205, 261, 264, 268 Biopsy ................................................................... 153, 185 Blood ............................... v, 1–3, 5, 7–10, 24, 28, 40, 42, 51–63, 89, 110, 151–160, 174, 175, 178, 276 Body fluids.................................... 9, 24, 28, 90, 106, 151 Brain............................................1, 90, 91, 163, 276, 278 Buffer exchange...................................92, 95, 97–98, 108

C Cell lysis ............................. 66, 69, 72–73, 114, 116, 119 Centrifugation ........................... 8, 24–26, 29–31, 33–35, 40, 46–48, 53, 82, 106, 120, 134, 135, 175 Cerebrospinal fluid (CSF) ............... v, 7, 12, 15, 89–111, 163–172, 276 Cleanup peptides......................................28, 33–34, 69, 74–75 proteins ....................................................... 65–67, 164 Clinical research ........................................ 90, 91, 95, 111 Colon mucosa ...................................................... 113, 114 Combinatorial analysis ........................248, 250–254, 257

D Data-independent acquisition (DIA)....... v, 43, 129–147, 151–160 Data processing .......................6, 15–17, 58, 66, 79, 132, 134, 136, 145, 154, 166, 169, 230, 233, 238, 252 Deglycosylation .........................................................26, 31 Depletion of abundant proteins ........................ 40, 95–97 Detergents ..........27, 52, 66, 67, 81, 113, 114, 119, 159

Diagnosis ............................... v, 2, 15, 23, 24, 39, 42, 90, 91, 114, 129, 151, 153, 163, 173–182, 185, 186, 188, 205, 247, 254, 257, 264, 278 Differential analysis ........................... 132, 144, 146, 230, 238, 239 Differential centrifugation ................................. 25, 26, 40 Discovery mass spectrometry .............. 89–111, 129–146, 151–160

E Electron transfer dissociation (ETD).................. 174, 182 Exosomes.........................................................v, 35, 39–49 Extracellular vesicles (EV) ......................v, 24, 25, 35, 40, 45, 56, 61

F Formalin-fixed paraffin-embedded (FFPE) tissues ........vi, 185–201

H Hemoglobin ............................................ 9, 174, 177, 180

I Isobaric tagging iTRAQ labelling ........................................................ 67 TMT labelling .................................................. 67, 114

L Label-free quantifications ................................49, 60, 114 Large-scale analysis........................................................ 133 Liquid chromatography (LC)..................... v, 8, 9, 12, 13, 15, 16, 26, 43, 46, 52, 55, 58, 60, 67, 71, 77–80, 91, 92, 94–97, 101–104, 106, 108, 115, 117, 119–121, 124, 132, 134–136, 138–141, 145, 156, 157, 159, 163–173, 188, 189, 192, 206–214, 216, 219, 225, 277, 287

M Machine learning........................................................... 110 Medicine .....................................1, 2, 104, 185, 247, 254 Method validation....................................... 3, 4, 206, 216 Microparticles........................................................ v, 51–63 Microvesicles .............................................................35, 40

Virginie Brun and Yohann Coute´ (eds.), Proteomics for Biomarker Discovery: Methods and Protocols, Methods in Molecular Biology, vol. 1959, https://doi.org/10.1007/978-1-4939-9164-8, © Springer Science+Business Media, LLC, part of Springer Nature 2019

291

PROTEOMICS

292 Index

FOR

BIOMARKER DISCOVERY: METHODS

Multiple reaction monitoring (MRM) ................ 47, 130, 152, 205–221, 287 Multiplexed assay ................................................. 185–201

O OFFGEL fractionation ............................... 27, 28, 32, 34

P Paramagnetic beads......................................................... 67 Peptide fractionation high-pH reversed-phase........................ 114, 115, 117 strong cation exchange (SCX).................... 92, 94, 96, 100, 101, 114, 133, 137, 138, 141, 143 Peptide library ............................189, 190, 193, 199, 200 Peripheral blood mononuclear cells (PBMCs)................ v, 151–160 Plasma ....................... v, 5, 8, 24, 28, 40, 41, 51–63, 104, 106, 108, 110, 131, 137, 152, 153, 155, 175, 180, 205, 207–210, 212–214, 216–218, 220, 275–277, 281, 287 Plasma microparticles................................................51–63 Post-analytical variability ..........................................12–18 Pre-analytical variability ............................................24, 25 Preeclampsia ..............................................................39–49 Protein desalting................................................................... 175 extraction ........................................... 31, 66, 124, 125 in-gel digestion.........................................32, 137, 145 in-solution digestion .................................... 42, 46, 48 precipitation................. 41, 42, 44–46, 164, 165, 167

Q Quantification absolute.................. 67, 205–207, 209, 212–215, 226 relative............. 60, 67, 102, 107, 186, 187, 200, 225

R Receiver operating characteristic (ROC) curves ........248, 251, 252, 254, 256, 257, 267–272 Reduction ......................... 25, 26, 29, 30, 42, 46, 54, 69, 72–73, 92, 95, 98–99, 109, 116, 283 Reverse-phase chromatography.................................... 157

S Sample collection ..........................24, 35, 44, 56, 66, 67, 134 preparation ............................ 8, 9, 12, 54, 57, 91–93, 96, 115, 116, 123–126, 137, 145, 164, 165, 168, 181, 207, 208, 210 Sample fractionation ..................115, 134, 137, 138, 141 Sample stability........................................................3, 8–10 SDS-PAGE ......................................................... 25, 26, 30

AND

PROTOCOLS

Selected reaction monitoring (SRM)................... 43, 130, 131, 152, 185–202, 205, 276 Serum........................ v, 2, 7–15, 26, 29, 39–49, 53, 116, 129–147, 152, 165, 166, 278 Size-exclusion chromatography (SEC) ....................39–49 Software analyst ...................................................................... 208 BioTools ................................................ 166, 170, 172 Byonic .................................................... 166, 170, 172 Compass DataAnalysis ............................................ 166 Mascot ...............................43, 58, 95, 102, 106, 172, 188, 189, 192 MaxQuant ............................. 84, 115, 121, 133, 136, 139, 141, 142, 145, 154, 243 PeptideShaker............................................... 68, 79, 81 Perseus ............................................................ 115, 118 Prism .......................................................................... 95 Profile Analysis ............................................... 166, 170 Progenesis QI ......................................................55, 60 ProStaR............................ vi, 144–146, 148, 225–245 Proteome Discoverer ........................... 55, 58, 60, 95, 102, 109, 145, 154, 157, 158 R.......................................................95, 148, 191, 227 RawTools ............................................... 68, 71, 79, 81 Scaffold .....................................................95, 102, 106 SearchGUI.................................................... 68, 71, 79 Skyline............................................... 43, 47, 166, 169, 187, 199 Spectronaut .........................133, 141, 144, 145, 148, 154, 158 Spotfire ........................................................... 188, 197 X! Tandem ................................................95, 102, 109 Spectral library.................................... 129–146, 152–154, 157–159, 189, 191, 192 Statistics ............................... v, 3, 7–9, 12, 13, 15–17, 24, 55, 60, 61, 102, 110, 132, 134, 141, 144–146, 166, 170, 220, 225–245, 247, 248, 251, 262, 266, 269 Synthetic peptides .....................145, 187, 190, 192, 193, 195, 200, 201

T Targeted mass spectrometry ................................ 186, 205 Tissue-leakage biomarker .................................... 275–288 Top-down proteomics ............................. v, 163, 173–182 Tube-gel (TG)...................................................... 123–126

W Web-based tools CombiRoc ................................................. vi, 250–256 Galaxy ........................................................ vi, 275–288 PanelomiX ................................................. vi, 261–271 ProteoRE ................................................278, 281–288