Stock Identification Methods: Applications in Fishery Science

Stock Identiﬁcation Methods Applications in Fishery Science This Page Intentionally Left Blank Stock Identiﬁcation ...

Author: Steven X. Cadrin | Kevin D. Friedland | John R. Waldman

140 downloads 1130 Views 8MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Stock Identiﬁcation Methods Applications in Fishery Science

This Page Intentionally Left Blank

Stock Identiﬁcation Methods Applications in Fishery Science Edited by

Steven X. Cadrin Kevin D. Friedland John R. Waldman

AMSTERDAM • BOSTON • HEIDELBERG • LONDON NEW YORK • OXFORD • PARIS • SAN DIEGO SAN FRANCISCO • SINGAPORE • SYDNEY • TOKYO Academic Press is an imprint of Elsevier

Editorial Coordinator: Kelly Sonnack Publishing Services Manager: Andre Cuello Project Manager: Justin Palmeiro Cover Design: Eric DeCicco Composition: Best-Set Text Printer: Maple Press Cover Printer: Phoenix Color Elsevier Academic Press 200 Wheeler Road, Burlington, MA 01803, USA 525 B Street, Suite 1900, San Diego, California 92101-4495, USA 84 Theobald’s Road, London WC1X 8RR, UK This book is printed on acid-free paper. Copyright © 2005, Elsevier Inc. All rights reserved. Except Chapter 21; Copyright © 2005, Mark B. Bain No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Permissions may be sought directly from Elsevier’s Science & Technology Rights Department in Oxford, UK: phone: (+44) 1865 843830, fax: (+44) 1865 853333, e-mail: [email protected]. You may also complete your request on-line via the Elsevier homepage (http://elsevier.com), by selecting “Customer Support” and then “Obtaining Permissions.” Library of Congress: Application submitted. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN: 0-12-154351-X For all information on all Academic Press publications visit our Web site at www.books.elsevier.com Printed in the United States of America 04 05 06 07 08 09 9 8 7 6 5 4 3 2 1

ANNOTATED TABLE OF CONTENTS

Foreword by Michael Sissenwine

xi

Contributors

xiii

PART

I

Introduction 1. Stock Identiﬁcation Methods: An Overview, by Steven X. Cadrin, Kevin D. Friedland, and John R. Waldman A brief introduction to stock identiﬁcation, including the historical development of the ICES Working Group. 2. Deﬁnition of Stocks: An Evolving Concept, by John R. Waldman A more detailed introduction to the ﬁeld, including technical deﬁnitions. 3. Fish Migration and the Unit Stock: Three Formative Debates, by D. H. Secor A review of ecological and historical issues related to stock connectivity and metapopulations. 4. Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters, by Douglas P. Swain, Jeffrey A. Hutchings, and Chris J. Foote An overview on the major categories of stock identiﬁcation approaches and their relative strengths for identifying stocks.

3

7

17

45

v

vi

Annotated Table of Contents

PART

II

Life History Traits 5. The Use of Early Life Stages in Stock Identiﬁcation Studies, by Jonathan A. Hare A description of methods that examine geographic range and distribution during early life history, including planktonic stages as well as juvenile and adult stages. 6. Life History Parameters, by Gavin A. Begg A review and critique of approaches that use differences in ontogenetic rates to distinguish stocks.

89

119

PART

III

Natural Marks—Morphological Analyses 7. Morphometric Landmarks, by Steven X. Cadrin A description of techniques used to measure and analyze general morphometry, including traditional multivariate morphometrics and more advanced geometric analyses. 8. Morphometric Outlines, by Steven X. Cadrin and Kevin D. Friedland A review and critique of methods that describe shape of outlines for structures such as scales and otoliths, and how methods are used to distinguish individuals with differently shaped features. 9. Analyses of Calciﬁed Structures: Texture and Spacing Patterns, by Kevin D. Friedland and Steven X. Cadrin A review of methods used to analyze spacing patterns of circuli on scales, otoliths, and vertebrae through image analysis, including digital photomicrograph examples. 10. Meristics, by John R. Waldman An evaluation of using the number of discrete morphological elements (e.g., number of vertebrae, ﬁn rays) for identifying stocks, with illustrative examples.

153

173

185

197

vii

Annotated Table of Contents

PART

IV

Natural Marks—Environmental Signals 11. Parasites as Biological Tags, by K. MacKenzie and P. Abaunza A summary of how parasitological analysis has been used to discriminate stocks.

211

12. Otolith Elemental Composition as a Natural Marker of Fish Stocks, by Steven E. Campana A review of rapidly developing techniques that use chemical composition of secreted hard parts to identify environmental differences and individuals that inhabited different habitats throughout their life history.

227

13. Fatty Acid Proﬁles as Natural Marks for Stock Identiﬁcation, by O. Grahl-Nielsen A description of a relatively new method for determining different populations according to fatty acids in tissues, with demonstrations on ﬁnﬁsh and marine mammals.

247

PART

V

Natural Marks—Genetic Analyses 14. Chromosome Morphology, by Ruth B. Phillips A review and critique of techniques for detecting differences among stocks through inspection of chromosome form, including several example photomicrographs.

273

15. Genetic Analysis: Allozymes, by M.-L. Koljonen and R. Wilmot A description of traditional electrophoretic methods, with many examples of stock identiﬁcation applications.

295

16. Mitochondrial DNA, by Antonios Magoulas A comprehensive review and protocol for detecting mitochondrial genetic characters and analyzing stock differences.

311

viii

Annotated Table of Contents

17. Use of Nuclear DNA in Stock Identiﬁcation: Single-Copy and Repetitive Sequence Markers, by Isaac Wirgin and John R. Waldman A review and critique of methods using single-copy, coding and noncoding, repetitive nuclear DNA for stock identiﬁcation.

18. Random Ampliﬁed Polymorphic DNA (RAPD), by P. J. Smith A review of polymerase chain reaction and RAPD techniques, which have had a rapidly increased application for stock identiﬁcation in recent years.

19. Ampliﬁed Fragment Length Polymorphism (AFLP), by Zhanjiang (John) Liu A description of a relatively new technique with great potential for stock identiﬁcation, including a comparative review with other genetic approaches.

331

371

389

PART

VI

Applied Marks 20. Internal and External Tags, by J. A. Jacobsen and L. P. Hansen A summary and critique of conventional tagging methods and their application for identifying stocks.

415

21. Electronic Tags, by Mark B. Bain A description of rapidly developing techniques involving telemetry and archival tags.

435

22. Otolith Thermal Marking, by Eric C. Volk, Steven L. Schroder, and Jeffrey J. Grimm A description of relatively new methods involving thermal signatures on ﬁsh otoliths, with photomicrographs illustrating their application for stock identiﬁcation.

447

ix

Annotated Table of Contents

PART

VII

Stock Identiﬁcation Data Analysis 23. Experimental Design and Sampling Strategies for Mixed-Stock Analysis, by Mary C. Fabrizio A protocol for sampling and a description of how sampling issues affect precision and accuracy of stock composition analysis.

467

24. An Introduction to Statistical Algorithms Useful in Stock Composition Analysis, by Michael H. Prager and Kyle W. Shertzer An evaluation of methods used to determine the contributions of different stocks in mixed-stock samples.

499

25. Classical Discriminant Analysis, Classiﬁcation of Individuals, and Source Population Composition of Mixtures, by Jerome Pella and Michele Masuda A description of linear discriminant analysis, with focus on stock identiﬁcation applications.

517

26. Neural Networks Used in Classiﬁcation with Emphasis on Biological Populations, by Saul B. Saila An introduction to a relatively new method of data analysis, with illustrative examples for identifying stocks. 27. Maximum Likelihood Estimation of Stock Composition, by Jon Brodziak A review of stock identiﬁcation applications using maximum likelihood to estimate contributions of different stocks in mixed-stock samples. 28. Estimation of Movement from Tagging Data, by Carl James Schwarz A description of analytical methods used to examine mark-recapture data for identifying stocks and quantifying interchange rates among stocks.

553

571

591

x

Annotated Table of Contents

PART

VIII

Application of Stock Identiﬁcation Data in Resource Management 29. Stock Identiﬁcation for Conservation of Threatened or Endangered Species, by Michael A. Banks A summary of how information on stock structure is used in resource management decisions. 30. The Role of Stock Identiﬁcation in Formulating Fishery Management Advice, by Cornelius Hammer and Christopher Zimmermann A description of how information on stock structure is considered in advice on stock status and management alternatives. 31. Identifying Fish Farm Escapees, by Peder Fiske, Roar A. Lund, and Lars. P. Hansen A review of the issue of escaped ﬁsh from aquaculture operations and a protocol for monitoring methods.

Index

609

631

659

681

FOREWORD

Fisheries scientists and managers use the term stock so frequently that you would think we know all there is to know about it. We speak of stock assessments, stock structure, spawning stock biomass, stock-recruitment relationships, stock complexes, stock production models, and so forth. I think it is fair to say that most scientists and managers take “stock” for granted as if it were sufﬁcient to rely on the old adage that “we’ll know one when we see it.” The truth is that its impossible to know how many scientiﬁc conclusions or ﬁshery management decisions may have been led astray by assuming we were seeing a stock that wasn’t! Information on the stocks is needed to meet objectives of ﬁsheries management to achieve sustainable yield, avoid recruitment failures, rebuild overﬁshed stocks, as well as to conserve threatened and endangered species. With growing acceptance of the need to conserve biodiversity (including genetic diversity), apply a precautionary approach, and operationalize the ecosystem approach (which places greater emphasis on spatial distributions and place-based management, such as MPAs), know what is and is not, a stock has never been more important. For more than a decade, the International Council for Exploration of the Sea, Stock Identiﬁcation Methods Working Group has been promoting standard protocols for sampling, data processing, and analytical methods, for data being generated by both traditional methods (e.g., meristics and morphometrics, traditional tags, parasites as natural tags) and new technologies (otolith chemistry, molecular genetics, electronic tags). The result is this comprehensive volume prepared by an outstanding team of international scientists. It addresses the stock concepts, historical development, applications to ﬁsheries science and management, use of natural marks (some traditional and some recently developed techniques), genetics, recent advances in tagging technology, and analytical methods. Although this volume will be a valuable reference for years to come, I think that we should all be excited by the prospect of innovative advances in the near future that surely will render some of the conclusions in the book out of date. xi

xii

Foreword

The scientists responsible for the volume, and ICES as the sponsor of the Working Group, do not want to rest on their laurels. Advances in biochemistry, analytical chemistry, and electronics (including microtechnology with nanotechnology on the horizon) foretell a very productive era unfolding when it comes to stock information in support of better science and better resource management. Speaking as both the President of ICES and the Chief Science Advisor for the U.S. National Marine Fisheries Service, I look forward to important and exciting discoveries in the future. Michael Sissenwine

CONTRIBUTORS

P. ABAUNZA, Instituto Espanol de Oceangraﬁa, Santander, Spain MARK B. BAIN, Center for the Environment, Cornell University, Ithaca, New York, USA MICHAEL A. BANKS, Coastal Oregon Marine Experiment Station, Hatﬁeld Marine Science Center, Department of Fisheries and Wildlife, Oregon State University, Newport, Oregon, USA GAVIN A. BEGG, CRC Reef Research Centre, James Cook University, Townsville, Queensland, Australia JON BRODZIAK, National Marine Fisheries Service, Woods Hole, Massachusetts, USA STEVEN X. CADRIN, National Marine Fisheries Service, Woods Hole, Massachusetts, USA STEVEN E. CAMPANA, Marine Fish Division, Bedford Institute of Oceanography, Dartmouth, Nova Scotia, Canada MARY C. FABRIZIO, National Marine Fisheries Service, Highlands, New Jersey, USA PEDER FISKE, Norwegian Institute for Nature Research, Trondheim, Norway CHRIS J. FOOTE, Department of Fisheries and Aquaculture, Malaspina University-College, Nanaimo, British Columbia, Canada O. GRAHL-NIELSEN, Department of Chemistry, University of Bergen, Bergen, Norway JEFFREY J. GRIMM, Washington Department of Fish and Wildlife, Olympia, Washington, USA xiii

xiv

Contributors

EVIN D. FRIEDLAND, NOAA Cooperative Marine Education and Research Program, University of Massachusetts, Amherst, Massachusetts, USA CORNELIUS HAMMER, Federal Research Centre for Fisheries, Institute for Baltic Sea Fisheries, Rostock, Germany M.-L. KOLJONEN, Finnish Game and Fisheries Research Institute, Helsinki, Finland L. P. HANSEN, Norwegian Institute for Nature Research, Oslo, Norway JONATHAN A. HARE, NOAA National Ocean Service, Center for Coastal Fisheries and Habitat Research, Beaufort, North Carolina, USA JEFFREY A. HUTCHINGS, Department of Biology, Dalhousie University, Halifax, Nova Scotia, Canada J. A. JACOBSEN, Faroese Fisheries Laboratory, Tórshavn, Faroe Islands ZHANJIANG (JOHN) LIU, The Fish Molecular Genetics and Biotechnology Laboratory, Department of Fisheries and Allied Aquacultures and Program of Cell and Molecular Biosciences, Aquatic Genomics Unit, Auburn University, Auburn, Alabama, USA ROAR A. LUND, Norwegian Institute for Nature Research, Trondheim, Norway K. MACKENZIE, School of Biological Sciences, Department of Zoology, The University of Aberdeen, Aberdeen, Scotland, United Kingdom ANTONIOS MAGOULAS, Hellenic Centre for Marine Research, Institute of Marine Biology and Genetics, Heraklion, Crete, Greece MICHELE MASUDA, National Marine Fisheries Service, Auke Bay, Alaska, USA JEROME PELLA, National Marine Fisheries Service, Auke Bay, Alaska, USA RUTH B. PHILLIPS, Washington State University, Vancouver, Washington, USA MICHAEL H. PRAGER, National Marine Fisheries Service, Beaufort, North Carolina, USA SAUL B. SAILA, University of Rhode Island, Graduate School of Oceanography, Narragansett, Rhode Island, USA STEVEN L. SCHRODER, Washington Department of Fish and Wildlife, Olympia, Washington, USA CARL JAMES SCHWARZ, Statistics and Actuarial Science, Simon Fraser University, Burnaby, British Columbia, Canada

Contributors

xv

D. H. SECOR, Chesapeake Biological Laboratory, University of Maryland Center for Environmental Science, Solomons, Maryland, USA KYLE W. SHERTZER, National Marine Fisheries Service, Beaufort, North Carolina, USA P. J. SMITH, National Institute of Water and Atmospheric Research Ltd., Wellington, New Zealand DOUGLAS P. SWAIN, Department of Fisheries and Oceans, Gulf Fisheries Centre, Moncton, New Brunswick, Canada ERIC C. VOLK, Washington Department of Fish and Wildlife, Olympia, Washington, USA JOHN WALDMAN, Hudson River Foundation for Science and Environmental. Research, New York, York; currently, Biology Department., Queens College, The City University of New York, New York, New York, USA R. WILMOT, National Marine Fisheries Service, Juneau, Alaska, USA ISAAC WIRGIN, Department of Environmental Medicine, New York University School of Medicine, Tuxedo, New York, USA CHRISTOPHER ZIMMERMAN, Institute for Sea Fisheries, Hamburg, Germany

PART

Introduction

I

This Page Intentionally Left Blank

CHAPTER

1

Stock Identiﬁcation Methods— An Overview STEVEN X. CADRIN,* KEVIN D. FRIEDLAND,† AND JOHN R. WALDMAN‡ *National Marine Fisheries Service, Woods Hole Massachusetts, USA †University of Massachusetts, Amherst, Massachusetts, USA ‡Queens College, The City University of New York, New York, New York, USA

References

Stock identiﬁcation is an interdisciplinary ﬁeld that involves the recognition of self-sustaining components within natural populations and is a central theme in ﬁsheries science and management. The obvious role of stock identiﬁcation is as a prerequisite for the tasks of stock assessment and population dynamics, because most population models assume that the group of individuals has homogeneous vital rates (e.g., growth, maturity, and mortality) and a closed life cycle in which young ﬁsh in the group were produced by previous generations in the same group. Because stock structure and delineation are uncertain, the reliability of stock assessments, and therefore the effectiveness of ﬁshery management, is severely limited for many ﬁshery resources. There are also roles for stock identiﬁcation in ﬁshery science that may be equally important but less obvious. Any study that wishes to represent a living resource through ﬁeld sampling, or even laboratory studies, should consider the species’ population structure in the sampling and analytical design. Whether the research concerns general life history, growth, physiology, or diet, the population of inference and its stock components should be identiﬁed. Therefore, stock identiﬁcation can be viewed as a prerequisite for any ﬁshery analysis, just as population structure is considered a basic element of conservation biology (Thorpe et al., 1995). Despite its importance, stock identiﬁcation remains one of the most confusing subjects in ﬁsheries science for nonspecialists, with a wide variety of approaches and conﬂicting terminologies and interpretations. There have been some excellent reviews of stock identiﬁcation research, including concise overviews (Simon and Larkin, 1972; Templeman, 1982; Pawson and Jennings, 1996) and conference proceedings that include various case studies (Ihssen et al., Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

3

4

Steven X. Cadrin et al.

1981; Kumpf et al., 1987; Begg et al., 1999). However, a synthetic overview of stock identiﬁcation was not previously available, and a focus on application of stock identiﬁcation results to ﬁshery science and management was generally lacking. Many of the case studies on stock identiﬁcation are result-oriented and narrowly focused, and overview perspectives lack the detail needed to guide researchers. Furthermore, in recent years signiﬁcant advances have been made in many approaches to stock identiﬁcation. In 1992, the International Council for the Exploration of the Sea (ICES) established a “Study Group on Stock Identiﬁcation Protocols for Finﬁsh and Shellﬁsh Stocks,” chaired by Kevin Friedland, to review methodologies of stock identiﬁcation and develop a protocol for the application of stock identiﬁcation results. The Study Group was organized in an open format to invite a wide participation of experts on stock identiﬁcation to summarize the various approaches. Over the next decade, the group expanded to the “Stock Identiﬁcation Methods Working Group,” currently co-chaired by Kevin Friedland, John Waldman, and Steve Cadrin, and developed a volume of contributions that is aimed at synthesizing the many disciplines involved in stock identiﬁcation. An outline of major stock identiﬁcation approaches and applications was formed and authors were solicited to draft chapters that review each speciﬁc method, with emphasis on recent advances, review of benchmark case studies, critique of strengths and weaknesses, and guidance for effective protocol. The support of this work by the ICES community reﬂects the continuing leadership of ICES in oceanographic and ﬁshery research. After all, it was an early ICES committee that ﬁrst promoted stock identiﬁcation as an important consideration for ﬁshery science in the late 1800s (Smith, 1994). This book introduces a wide variety of methods and provides guidance and example applications. As described in the introductory chapters, the deﬁnition of the term stock is somewhat conditional on methodology, and different methods offer complementary perspectives on population structure. Accordingly, the “stock concept” evolved as methodological approaches evolved, from early morphological methods through the development of genetic techniques and the increased power to detect environmental signals in tissues. The compendium of methodological reviews (sections II to VI) was designed to serve as a resource for researchers interested in comparative studies in stock identiﬁcation as well as a general introduction for all scientists and managers of natural resources. Methodological chapters are not necessarily comprehensive reviews but focus more on historical development, benchmark case studies, critique of current issues, and prescriptions for the most effective protocols for stock identiﬁcation. Section VII, Stock Identiﬁcation Data Analysis, offers insights into a variety of statistical procedures and provides guidance on their proper application. Considerations for proper sampling, data treatments, and interpretations are dis-

Stock Identiﬁcation Methods—An Overview

5

cussed. The ﬁnal section describes how stock identiﬁcation information is used in resource management, illustrating the difﬁculties and limitations, and reviewing case studies in which stock identiﬁcation played a central role. One theme that emerged throughout the development of this volume is the strength of interdisciplinary analyses. Over the history of stock identiﬁcation, new methods were developed and promoted as better ways to approach stock identiﬁcation, often leading to equivocal information from competing methodological camps. However, when results from each approach are viewed in the context of what precise aspect of stock structure they reveal (deﬁned in this volume), a more holistic view with multiple perspectives is possible, providing more reliable information for resource management. As new methods continue to emerge, their results should be considered, along with those from traditional approaches, in improving our ability to study stock structure. Finally, despite its importance in the development of ﬁshery advice and management, stock identiﬁcation continues to be an afterthought. The ﬁshery science community has a habit of building assessments from back to front, often giving only cursory treatment to stock identiﬁcation, and in the name of being expeditious, population vital rates are estimated without regard to lingering questions about stock structure. We hope this volume not only will provide source material to improve the quality of stock identiﬁcation research, but also will stimulate new research on stocks being assessed without the beneﬁt of reliable stock identiﬁcation.

ACKNOWLEDGMENTS We thank the ICES community, particularly David Grifﬁth and Mette Bertelson, for their support throughout this endeavor. We are also grateful to Dave Cella, Kelly Sonnack, and Chuck Crumly for their assistance in the publication.

REFERENCES Begg, G., Friedland, K. D., and Pearce, J. B. 1999. Stock identiﬁcation—its role in stock assessment and ﬁsheries management. Fisheries Research 43: 1–8. Ihssen, P. E., Bodre, H. F., Casselman, J. M., McGlade, J. M., Payne, N. R., and Utter, F. 1981. Stock identiﬁcation: materials and methods. Canadian Journal of Fisheries and Aquatic Sciences 38: 1838–1855. Kumpf, H. E., Vaught, R. N., Grimes, C. B., Johnston, A. G., and Nakamura, E. L. 1987. Proceedings of the Stock Identiﬁcation Workshop. NOAA Tech. Mem. NMFS-SEFC-199. Pawson, M. G. and Jennings, S. 1996. A critique of methods for stock identiﬁcation in marine capture ﬁsheries. Fisheries Research 25: 203–217. Simon, R. C. and Larkin P. A. (eds.). 1972. The stock concept in Paciﬁc salmon. H. R. MacMillan Lectures in Fisheries. University of British Columbia.

6

Steven X. Cadrin et al.

Smith, T. D. 1994. Scaling Fisheries: A Science Driven by Economics and Politics 1855–1955. Cambridge University Press, Cambridge, UK. Templeman, W. 1982. Stock discrimination in marine ﬁshes. NAFO SCR Doc. 82/IX/79. Thorpe, J., Gall, G., Lannan, J., and Nash, C. 1995. Conservation of Fish and Shellﬁsh Resources: Managing Diversity. Academic Press, San Diego, CA.

CHAPTER

2

Deﬁnition of Stocks: An Evolving Concept JOHN R. WALDMAN Hudson River Foundation for Science and Environmental Research, New York, New York Currently, Biology Department, Queens College, The City University of New York, New York, New York, USA

I. Introduction II. Deﬁning the “Stock”: An Evolving Concept A. Early Viewpoints B. Late Twentieth-Century Deﬁnitions C. Stocks and Conservation Biology III. Some Fundamental Issues in Deﬁning Stocks A. Genotypic vs. Alternative Approaches B. Negative Results C. Congruence IV. Conclusions References

I. INTRODUCTION Fisheries science is based on the notion of an idealized “unit stock,” a discrete entity with its own origin, demographics, and fate. Fish stock assessment is the science of estimating certain key population dynamics parameters to better manage the unit stock. If the rates of growth, natural mortality, reproduction, and present ﬁshing mortality can be estimated, their effects may be combined in order to evaluate the effects of changes in the ﬁshing mortality rate. An understanding of stock structure is necessary for designing appropriate management regulations in ﬁsheries where multiple stocks are differentially exploited (Ricker, 1981). Stock identiﬁcation is an essential partner to stock assessment—unit stocks cannot be assessed unless they are circumscribed, that is, their boundaries deﬁned in relation to other units of the same species. At its core, stock identiﬁcation is the process that seeks to identify coherent units of individuals that have complete-to-partial discreteness in space or time Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

7

8

John R. Waldman

from their congeners. These units have certain characteristics that render them fundamental to management: (1) they occupy their own physical life history circuit, including spawning grounds that are geographically or temporally unique; (2) they experience their own natural demographic inﬂuences, such as mortality suffered from a particular suite of predators; (3) their complete-to-partial isolation allows ﬁne tuning of their morphological and genetic characteristics to their particular environmental circumstances; and (4) their abundances and life history characteristics respond to their own sets of unnatural inﬂuences, such as ﬁsheries and contamination of their habitats. Before identifying these units, ﬁsheries science must grapple with deﬁning them, a task that remains confusing and problematic. Deﬁnitions of both parts of the term stock identiﬁcation are muddled, that is, “stock” and “identiﬁcation,” pattern and process. But of the two, process is far more easily speciﬁed. Concerning process, although the term stock identiﬁcation is often used to represent an entire realm of investigation, in a narrower sense it may be taken to mean only the initial identiﬁcation of units within a species, whereas stock discrimination means the process of classifying individuals or collections of individuals to those units. The term stock composition analysis is an extension of stock discrimination where the proportions of unit stocks to a mixed-stock ﬁshery are estimated. It is common that stock identiﬁcation and stock discrimination are used interchangeably. Stock composition analysis is also referred to as mixed-stock analysis and relative contribution analysis. Concerning pattern or structure in ﬁshery resources, however, deﬁnitions abound. Royce (1972) believed there were “. . . a bewildering array of semantic problems because there is little agreement on the meaning of the words used to deﬁne groups in the hierarchy with the rank of subspecies and below. . . .” Reaching a consensus in this area remains immensely difﬁcult. Older categories beneath subspecies that have largely fallen out of favor among ﬁshery biologists include subpopulation, race, strain, breed, and variety. A more current hierarchy below subspecies is illustrated by Secor (this volume; Fig. 3-1) that includes metapopulation, population, contingent, year-class, school, brood, and individual. A fundamental problem in deﬁning stocks (and other biological categories such as species) is assuming the optimal balance between precision and generality: too much precision and the deﬁnition will not be robust enough to serve all situations, too general and the deﬁnition will have little utility. This conundrum runs throughout the history of stock deﬁnition. Kutkuhn (1981) observed that “. . . the literature is replete with observations and opinions on the concept, character, and implications of something we commonly refer to as the (ﬁsh) ‘stock.’ But nowhere, insofar as I can determine, has any body of authority, as is often established to perform such functions, acclaimed a deﬁnition of it.” Nonetheless, many deﬁnitions have been proposed.

Deﬁnition of Stocks: An Evolving Concept

9

II. DEFINING THE “STOCK”: AN EVOLVING CONCEPT

A. EARLY VIEWPOINTS Secor (this volume, Chapter 3) reviews notions of unit stock and population thinking, noting how the unit stock arose as a means of practically deﬁning vital rates and renewal rates pertinent to geographic areas where ﬁsheries were undertaken. Dahl (1909) viewed stock as a source of ﬁsh, that is, a speciﬁc portion of a population that is inﬂuenced by an anthropogenic activity that affects population productivity. This deﬁnition is purely operational in which the portion of ﬁsh exposed to the ﬁshery deﬁnes the stock. This category might be referred to as a “harvest stock.” The baseline for establishment of a harvest stock is not nature’s knife (i.e., isolation and the process of differentiation) and its natural subdivisions below that of species. Instead, a harvest stock is deﬁned by human interaction with the species, cutting across and grouping whichever natural subdivisions (lineages) happen to encompass a particular ﬁshery.

B. LATE TWENTIETH-CENTURY DEFINITIONS Despite its beginnings as an operationally deﬁned ﬁsheries term, the term stock began to drift from the practical toward more theoretical deﬁnitions that recognized the microevolutionary subdivisions occurring below the species level, regardless of whether ﬁsheries are promulgated. A transitory deﬁnition was provided by Larkin (1972), who wrote that a stock is “. . . a population of organisms which, sharing a common gene pool, is sufﬁciently discrete to warrant consideration as a self-perpetuating system which can be managed.” Both a common gene pool and management are explicit in this deﬁnition. However, deﬁnition of the stock by the ﬁshery alone continued in the Magnuson Fishery Conservation and Management Act of 1976 in which it meant a species, subspecies, geographic grouping, or other category of ﬁsh capable of management as a unit (Fox and Nammack, 1995). The decades-long conundrum over the deﬁnition of stock brought about an ambitious symposium held in 1980 in Ontario, Canada, to consider that subject. Indeed, the papers originating from the Stock Concept International Symposium (Canadian Journal of Fisheries and Aquatic Sciences, 1981, vol. 33, no. 12) provide a ﬁne summary of the state of our perceptions of stock at that time. Several observations can be made. One is that the issue was far from settled at, or even after, the gathering—the resultant papers reﬂect a wide diversity of unreconciled opinion. Another was a movement from utilitarian deﬁnitions to ones that seek to outline naturally occurring units. A third was a sometimes expressed desire to base the deﬁnition on genetics, a response, in part, to the growing wealth

10

John R. Waldman

of genotypic knowledge gained through the application and maturation by 1980 of protein electrophoretic analysis. Among these papers, Booke (1981) tackled the issue head on, wishing ambitiously to propose a working deﬁnition of stock that would be useful for all ﬁsh species. He offered one that he recognized was highly general, that is, “a species group, or population, of ﬁsh that maintains and sustains itself over time in a deﬁnable area.” He then suggested more precisely that a genotypic stock is a population of ﬁsh maintaining and sustaining Castle–Hardy–Weinberg equilibrium. And that if his preferred genotypic stock characterization is not possible, a phenotypic stock “. . . has to be recognized as a group, or population of ﬁsh maintaining characteristics which are expressed in one or more ways depending on the type of environment or domicile.” Bailey and Smith (1981) echoed this genotypic and phenotypic dichotomy, deﬁning stocks as local populations that maintain recognizable genetic differentiation by separation of their spawning place or time, but also stating that “. . . a (phenotypic) stock has to be recognized as a group, or population, of ﬁsh maintaining characteristics which are expressed in one or more ways depending on the type of environment or domicile.” However, their genotypic criterion contrasted with Booke’s in that it was based on differentiation among units, whereas Booke mentioned “group” and “population” but did not deﬁne them as differentiated (but he did uniquely invoke Castle–Hardy–Weinberg equilibrium, a withinpopulation measure). Another attempt was made by Casselman et al. (1981), who stated that a stock is “. . . a population of ﬁsh that behaves as a cohesive unit whose members exhibit common responses to environmental conditions within its geographic boundaries.” They believed this includes but is not restricted to population units that are reproductively isolated, which is less restrictive than many other deﬁnitions and which conceivably could embrace other concepts such as a regional stock or metapopulation. The deﬁnition of stock made at this symposium, which I believe was most robust, yet sufﬁciently speciﬁc to be useful, was offered by Ihssen et al. (1981), who proposed that a stock is “. . . an intraspeciﬁc group of randomly mating individuals with temporal or spatial integrity.” This deﬁnition places an upper boundary at the species level yet allows for a kind of operational deﬁnition anywhere below that boundary at whatever level temporal or spatial integrity is displayed; that is, by this deﬁnition various units of interest expressed by Royce (1972) as groups in the hierarchy with the rank of subspecies and below (such as metapopulation, population, or contingent) would qualify, depending on the pattern expressed and the degree of technical stringency applied to the problem. A newer and less restrictive deﬁnition was offered by Hilborn and Walters (1992) in which stocks are considered as arbitrary groups of ﬁsh large enough

Deﬁnition of Stocks: An Evolving Concept

11

to be essentially self-reproducing, with members of each group having similar life history characteristics.

C. STOCKS

AND

CONSERVATION BIOLOGY

Beyond its operational deﬁnitions linked to ﬁsheries, stocks may be considered to be units below species that are naturally occurring and of interest to managers and scientists beyond reasons concerning harvest. With its emphasis on preserving biodiversity, the relatively new science of conservation biology is intimately involved with identifying these below-species units. Many of the technical approaches to resolving stocks for ﬁsheries’ purposes are also applicable to the goals of conservation biology. However, new terms have emerged that border on, yet differ from, more traditional ﬁsheries’ stock concepts. Primary among these today is the “evolutionarily signiﬁcant unit” (ESU) of the biological species, a concept reﬁned for Paciﬁc salmon under the general demands of the U.S. Endangered Species Act to conserve genetic diversity within and between species (Waples, 1991, 1995). The ESU as deﬁned by Waples (1991) is a population that (1) is substantially reproductively isolated from other conspeciﬁc reproductive units, and (2) represents an important component in the evolutionary legacy of the species. The ﬁrst criterion is essentially the deﬁnition of a stock. But the second criterion attempts to answer a question with respect to a population’s evolutionary legacy, that is, if the population became extinct, would this represent a signiﬁcant loss to the ecological-genetic diversity of the species? Thus, the ESU goes beyond the stock concept in that its designation includes a qualitative and often difﬁcult judgment (King and Ludke, 1995) about each unit’s importance. Other deﬁnitions of ESU beyond Waples (1991) were reviewed by Fraser and Bernatchez (2001). They argued that no single approach works best in all situations, but that each has its strengths and weaknesses under different circumstances. Thus, they propose “adaptive evolutionary conservation” in which alternative ESU criteria are ﬁt to situational circumstances.

III. SOME FUNDAMENTAL ISSUES IN DEFINING STOCKS

A. GENOTYPIC

VS.

ALTERNATIVE APPROACHES

Some workers have referred to the genotypic approach to deﬁning stocks as if it is the avatar of stock identiﬁcation (e.g., Booke, 1981). Genotypic approaches do offer many advantages to alternative techniques, including permanence across an individual’s life cycle, freedom from environmental modulation, often large

12

John R. Waldman

substrates of variation, and usually, nonfatal sampling (Wirgin and Waldman, 1994). But it is important not to be so swayed by these advantages as to dismiss the fact that any consistently detectable differences among stocks are both valid and useful; that is, a stock is a stock if a marker discriminates among them, even if these differences are partly or completely environmental in origin. Indeed, there are situations in which it is conceivable that nongenetic approaches might be preferable. One such situation is when there are simple, qualitative characteristics, such as observations of different timings of runs of an anadromous species in a river; another might be an obvious phenotypic marker available, such as a color difference. Another is when newly formed populations through restocking or natural colonization do not yet show genetic differences despite their discreteness. Yet another is where stock identiﬁcation information can emerge from mark-recapture work conducted for stock assessment purposes. (However, it would help the stock identiﬁcation ﬁeld if more controlled “common garden” rearing experiments in which the environment was held constant were conducted in order to gain a better understanding of genetic and nongenetic control of stock differences.) Today’s recognition of the complementarity of genotypic and phenotypic approaches contrasts with the apparently short-lived view espoused by Marr (1957) who believed that “subpopulations” are characterized by genetic selfsustainment, whereas stocks are populations or portions of populations of which all members are characterized by similarities that are not heritable, but are induced by the environment. The scientist interested in identifying ﬁsh stocks has a large number of tools available in the kit. Genotypic approaches may be highly prominent among them, but they have not supplanted the remainder.

B. NEGATIVE RESULTS Kutkuhn (1981) summarized the scientiﬁc process of stock identiﬁcation. First, one needs indication of the existence of stocks, from which a hypothesis is mounted (vs. the null hypothesis of a single stock). Second, a well-conceived survey must be undertaken. Finally, an efﬁcient discrimination technique must be applied to the collections. Each of these stages may be nulliﬁed by ﬂaws that result in a failure to falsify the null hypothesis, including defective study logic, faulty survey techniques, and questionable classiﬁcation procedures (Kutkuhn, 1981). Notwithstanding the ﬁrst two problems, it is possible in a study that a competent classiﬁcation procedure was applied but that it lacked the sensitivity to detect evidence for the stocks at a statistically signiﬁcant level; that is, a careful and earnest effort using one approach yields negative results (Type 1 error). Should analysis stop there and the absence of discrete stocks be accepted?

Deﬁnition of Stocks: An Evolving Concept

13

My answer is—only with caveats. Nonexistence cannot be proved and, in a formal sense, what has occurred is that the null hypothesis has not been falsiﬁed. It is entirely possible that a different stock identiﬁcation approach would falsify the null hypothesis. Unfortunately, there is no hard-and-fast rule to know how far to proceed with additional analyses. Most often, resource limitations preclude any, or at least, much more analysis. But it is important to make a distinction in such cases where negative results have been obtained: The existence of stocks has not been disproved, only the null hypothesis of a single stock has not been falsiﬁed. Also, although failure to falsify the null hypothesis may be less compelling to the investigator, such negative results should nonetheless be published inasmuch as they still provide valuable information about the nature of the species studied and the sensitivities of the stock identiﬁcation approaches employed.

C. CONGRUENCE Falsiﬁcation of the null hypothesis of a single stock implies that multiple stocks exist, and much of what we know about the existence of particular stocks is derived from the successful application of single stock identiﬁcation approaches. But why might one approach be successful in detecting multiple stocks, whereas another is unsuccessful? That is, why do different stock identiﬁcation approaches have different sensitivities to a particular stock problem? One factor is the “depth” of the stock division being addressed. As anyone who has examined branching diagrams of relationships among ﬁsh stocks has observed, stocks show subdivisions with different levels of relatedness, sometimes within a metapopulation structure (Stephenson, 1999). For example, several anadromous ﬁsh stocks in rivers that drain into a common bay or sound may group as a regional stock, yet there may be tributary stocks within one of these single rivers. This depth is a product of space and time, that is, the greater the degree and length of isolation, the greater the depth of the stock division. But differentiation of stocks is estimated by examining a wide variety of phenotypic and genotypic features, and each of these features may undergo microevolution at their own rate (Waldman, 1999). Fish have literally been “taken apart” down to the level of DNA in the search for useful stock discrimination characters. Some of these features respond rapidly to differentiating forces (e.g., ﬁn ray counts, nuclear DNA microsatellites), whereas others are more conservative (e.g., vertebral counts, the coding region of any nuclear DNA gene). Thus, a particular stock identiﬁcation problem involves choosing among approaches that address features of the ﬁsh that compose these stocks, which are passing through differing microevolutionary trajectories—some of which may be applicable to the problem (offering a high signal-to-noise ratio) and others which may be less, or not at all applicable.

14

John R. Waldman

Application of additional stock discrimination techniques (a holistic approach) has several advantages to single approach studies (Begg and Waldman, 1999). One advantage is that one of these techniques may detect stock structure where others fail to do so. Another is that greater conﬁdence is gained when contrasting approaches provide congruent results. Also, additional levels of stock structure may be observed with approaches that offer different sensitivities. Finally, multiple applications of discrete approaches to individual stock identiﬁcation problems provide more empirical information on the relative merits of alternative approaches to this still developing ﬁeld.

IV. CONCLUSIONS Since the advent of modern ﬁsheries science, the concept of the unit stock has remained the same while the practice of identifying ﬁsh stocks has undergone signiﬁcant change. The unit stock is still an idealized fundamental unit which, when correctly identiﬁed, allows estimation of important population parameters without the biases that result from incorporation of unknown components of additional unit stocks. But the stock searched for in the stock identiﬁcation process may not always be a unit stock. Under an early, and presently minority, point of view, this does not matter; that is, a harvest stock is simply “what is caught” and so the ﬁshery itself deﬁnes the stock. Secor (1999) noted that the increased technical ability to discriminate ﬁsh populations has led to synonomy between population and stock. He argues, however, that this classical deﬁnition still has validity in the sense that accessibility is implicit in stock deﬁnitions and that accessibility is mediated largely by within-population ecological attributes such as migratory modalities. How consideration of these within-population differences will affect future notions of stock is unclear, but it is likely that much new information will emerge as powerful new tools are employed, such as microelemental analysis of otoliths (e.g., Secor et al., 2001) and stable isotope analysis (e.g., Adams et al., 2003). However, the numerous modern deﬁnitions of stock reﬂect an evolution toward naturally deﬁned divisions or lineages within species that has occurred for at least two reasons, one philosophical and one practical. The philosophical rationale is that lineage-based divisions correspond more closely with the notion of the idealized unit stock. Although deﬁning such stocks involves necessary steps beyond letting the ﬁshery itself delimit the stock, and these steps may be costly and technically challenging (sometimes producing unclear results), they nonetheless represent a search for fundamental biological units, not units of convenience. The practical reason is the century-long trend toward improved stock identiﬁcation capabilities; that is, the tool kit has grown considerably larger and far more technically keen, as evidenced by this volume.

Deﬁnition of Stocks: An Evolving Concept

15

Earlier ﬁsheries scientists had no choice but to consider harvest stocks inasmuch as stock discrimination techniques did not exist and so synonomy between stock and population was not even possible. Today a wealth of techniques can be applied to discriminate among stocks, and in many instances these results can be applied to estimate their proportions in mixed stocks. Thus, the possibly “mixed” harvest stock may be quantitatively decomposed to its constituent unit stocks. Begg and Waldman (1999) underscored the analogy between the surprisingly difﬁcult problems in deﬁning species (e.g., Mayden and Wood, 1995) and in deﬁning stocks. Deﬁning the more recently conceptualized ESU presents similar problems (Fraser and Bernatchez, 2001). All of these notions have shifted and diversiﬁed over time partly in response to changes in thinking about evolution (albeit, at different levels) and to technological advances. Yet, despite the challenges in deﬁning them, on the operational level, all are usually discriminated without much debate. If a single stock deﬁnition is needed, I favor that of Ihssen et al. (1981), but others may be more appropriate in particular cases. Deﬁnitions of stock will likely continue to evolve as management requirements change and technologies advance.

ACKNOWLEDGMENTS Thanks to Steven Cadrin, Kevin Friedland, and David Secor for their insightful comments on the manuscript.

REFERENCES Adams, C., Fraser, D., McCarthy, I., Shields, S., Waldron, S., and Alexander, G. 2003. Stable isotope analysis demonstrates ecological segregation in a bimodal size polymorphism in Arctic charr from Loch Tay, Scotland. Journal of Fish Biology 62: 474–481. Bailey, R. M. and Smith, G. R. 1981. Origin and geography of the ﬁsh fauna of the Laurentian Great Lakes basin. Canadian Journal of Fisheries and Aquatic Sciences 38: 1539–1561. Begg, G. A. and Waldman, J. R. 1999. An holistic approach to ﬁsh stock identiﬁcation. Fisheries Research 43: 35–44. Booke, H. E. 1981. The conundrum of the stock concept—are nature and nurture deﬁnable in ﬁshery science? Canadian Journal of Fisheries and Aquatic Sciences 38: 1479–1480. Casselman, J. M., Collins, J. J., Crossman, E. J., Ihssen, P. E., and Spangler, G. R. 1981. Lake whiteﬁsh (Coregonus clupeaformis) in the Great Lakes region. Canadian Journal of Fisheries and Aquatic Sciences 38: 1772–1789. Dahl, K. 1909. The problem of sea ﬁsh hatching. Special Part B, No. 5. In J. Hjort (ed.). Rappurt sur les Travaux de Commission A, Dans La Période 1902–1907, Conseil Permanent International pour l’Exploration de La Mer. Rapports et Procès-Verbaux, Vol. X. Fox, W. W., Jr. and Nammack, M. F. 1995. Conservation guidelines on signiﬁcant population units: responsibilities of the National Marine Fisheries Service. American Fisheries Society Symposium 17: 419-422.

16

John R. Waldman

Fraser, D. J. and Bernatchez, L. 2001. Adaptive evolutionary conservation: towards a uniﬁed concept for deﬁning conservation units. Molecular Ecology 10: 2741–2752. Hilborn, R. and Walters, C. J. 1992. Quantitative Fisheries Stock Assessment: Choice, Dynamics and Uncertainty. Chapman and Hall, New York. 570 pp. Ihssen, P. E., Booke, H. E., Casselman, J. M., McGlade, J. M., Payne, N. R., and Utter, F. M. 1981. Stock identiﬁcation: materials and methods. Canadian Journal of Fisheries and Aquatic Sciences 38: 1838–1855. King, T. L. and Ludke, J. L. 1995. A national biological service perspective on deﬁning unique units in population conservation. American Fisheries Society Symposium 17: 425–429. Kutkuhn, J. H. 1981. Stock deﬁnition as a necessary basis for cooperative management of Great Lakes ﬁsh resources. Canadian Journal of Fisheries and Aquatic Sciences 38: 1476–1478. Larkin, P. A. 1992. The stock concept and management of Paciﬁc salmon. In R. C. Simon and P. A. Larkin (eds.), The Stock Concept in Paciﬁc Salmon. H. R. MacMillan Lectures in Fisheries. University of British Columbia, Vancouver, BC, pp. 11–15. Marr, J. C. 1957. Contributions to the study of sub-populations of ﬁshes. U.S. Fish and Wildlife Service Special Scientiﬁc Report—Fisheries 208: 1–6. Mayden, R. L. and Wood, R. M. 1995. Systematics, species concepts, and the evolutionary signiﬁcant unit in biodiversity and conservation biology. American Fisheries Society Symposium 17: 58–113. Ricker, W. E. 1981. Changes in the average size and average age of Paciﬁc salmon. Canadian Journal of Fisheries and Aquatic Sciences 38: 1636–1656. Royce, W. F. 1972. Introduction to the Fishery Sciences. Academic Press, New York. 351 pp. Secor, D. H. 1999. Specifying divergent migrations in the concept of stock: the contingent hypothesis. Fisheries Research 43: 13–34. Secor, D. H., Rooker, J. R., Zlokovitz, E., and Zdanowicz, V. S. 2001. Identiﬁcation of riverine, estuarine, and coastal contingents of Hudson River striped bass based upon otolith elemental ﬁngerprints. Marine Ecology Progress Series 211: 245–253. Stephenson, R. L. 1999. Stock complexity in ﬁsheries management: a perspective of emerging issues related to population sub-units. Fisheries Research 43: 247–249. Waldman, J. R. 1999. The importance of comparative studies in stock analysis. Fisheries Research 43: 237–246. Waples, R. S. 1991. Paciﬁc salmon, Oncorhynchus spp. and the deﬁnition of “species” under the endangered species act. Marine Fisheries Review 53: 11–22. Waples, R. S. 1995. Evolutionarily signiﬁcant units and the conservation of biological diversity under the Endangered Species Act. American Fisheries Society Symposium 17: 8–27. Wirgin, I. I. and Waldman, J. R. 1994. What DNA can do for you. Fisheries 19(7): 16–27.

CHAPTER

3

Fish Migration and the Unit Stock: Three Formative Debates D. H. SECOR Chesapeake Biological Laboratory, University of Maryland Center for Environmental Science, Solomons, Maryland, USA

I. The Unit Stock and Population Thinking II. Debate 1. Early Cod Hatchery Enhancement: Local or Global Effects on Fisheries? III. Debate 2. The Parent Stream Theory: Directed Migrations or Local Wanderings? A. Natural Tags B. Marking Experiments C. Transplant Experiments D. The Year-Class Phenomenon (Again) E. Ocean Studies F. The Parent Stream Theory and Population Thinking IV. Debate 3. The Eel Problem: Who Contributes to Reproduction? A. Mediterranean vs. North Atlantic Origin of Eels B. North Atlantic Eels: Speciation or Population Structure? C. Failed Adults D. “Freshwater Eels” and Contingent Thinking E. More Eel Problems and Population Thinking V. Summary References

I. THE UNIT STOCK AND POPULATION THINKING Modern ﬁsheries science emerged through the early efforts of the International Council for the Exploration of the Sea (ICES) to understand factors that caused Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

17

18

D. H. Secor

ﬂuctuations in important coastal ﬁsheries (Smith, 1994; Smed and Ramster, 2002). A major research goal early on for ICES scientists was to establish relationships among ﬁsh distribution, behavior, and environmental factors (Nakken, 2002). Late nineteenth-century investigations on Atlantic herring by Heincke (1898) supported a widely held view that many local races structured the distribution of herring and their availability to ﬁsheries. Local ﬁsheries might then have local effects. This so-called modern migration theory displaced an earlier view of panmixia in cod and other coastal ﬁshes. In contrast to these earlier ideas, ICES scientists discovered that ﬁsh became accessible to ﬁsheries according to population renewal processes structured over large but distinct geographic regions (Hjort, 1914). The unit stock arose as a means of practically deﬁning vital rates and renewal rates pertinent to geographic regions where ﬁsheries were undertaken. Early use of the term was by Dahl (1909), who deﬁned stock as the source of ﬁsh. Later, Russell (1931), in his catch equation, deﬁned stock as the exploitable portion of a population. Stock as an operational deﬁnition could be tailored for regions where vital rates and recruitment were deemed homogenous (Gulland, 1983; Cushing, 1995). Thus, while population thinking underlies the way we deﬁne and manage stocks, stocks themselves are not ecological entities per se; they are operationally deﬁned by the geographic extent of anthropogenic effects and other practical considerations. Levels of biological structure relevant to stock issues can range from species to brood (Fig. 3-1). In some instances, lineage (species to population levels) may be critical in deﬁning management units, as is the case when large reductions occur to abundance and threaten maintenance of the underlying gene pool. This often entails an evolutionary perspective that can extend millions of years and across ocean basins. In other cases, ecological time and spatial scales may be of greatest relevance. For instance, a manager may need to know how important a given habitat is as a source for recruits. Here, stocks are deﬁned according to regions of production, and biological levels such as contingents, cohorts, and shoals may be important (e.g., Beck et al., 2001). Despite the operational deﬁnition of stocks, there is little doubt that population structure is fundamental to how we assess and manage living resources (Sinclair, 1988; Sinclair and Smith, 2002). Population thinking arose during the early twentieth century as scientists contended with Hjort’s discovery of population cycles common over large coastal regions (Solemdal and Sinclair, 1989). Issues remained as to which geographic scales were relevant to populations (scale and entity), which behaviors contributed to population structure (migration, life cycle closure, and philopatry), and the consequences of population structure (or alternatively, the consequence of deviations from life cycle closure). Here, I introduce three formative debates, which relate to pattern, process, and consequence. These include: (1) the effectiveness of cod hatcheries in fjords of southern

19

Fish Migration and the Unit Stock: Three Formative Debates Region of Coherence

Years of Coherence

Stock Entity

107 Species

106 Subspecies

105 Metapopulation

104 103 102 10

Population Contingent Year-class School Brood

Individual

FIGURE 3-1. Temporal and spatial domains, and levels of biological organization relevant to the unit stock. Here, the unit stock is operationally deﬁned by the anthropogenic effects of interest (Gulland, 1983).

Norway; (2) evaluation of the parent stream theory for Paciﬁc salmon; and (3) the eel problem: Who contributes to population renewal processes?

II. DEBATE 1. EARLY COD HATCHERY ENHANCEMENT: LOCAL OR GLOBAL EFFECTS ON FISHERIES? Johan Hjort’s pioneering observation of population response to periodically strong year-classes remains a principal paradigm that guides ﬁsheries science and management. To make this revelation, Hjort and his team had to turn away from prevailing theory, which stipulated local races and local effects (Sinclair and Solemdal, 1988), and discover a new way of thinking about the renewal process. Several scientists and historians have argued that Hjort’s “population thinking” may have originated from practical considerations on the effectiveness of hatchery releases of larval cod into Norwegian fjords over a century ago (Solemdal et al., 1984; Sinclair and Solemdal, 1988; Smith, 1994; Schwach, 1998; Secor, 2002; Sinclair and Smith, 2002). Did these releases produce local effects in a fjord? Or were released cod swamped by natural juvenile production within the fjord or, perhaps, from outside the fjord?

20

D. H. Secor

By the mid-1890s the Flødevigen hatchery under the leadership of G. M. Dannevig was releasing millions of cod larvae into southern Norway fjords. Early development of egg fertilization and hatching led to the belief that “. . . cod fry artiﬁcially hatched have the power and energy to live, grow, and develop when set at liberty and left to care for themselves in their natural element, the sea” (Rognerud, 1887). From 1890 to 1906, 23.3 billion yolk-sac larvae were released into small fjords, sounds, and bays in southern Norway (Solemdal et al., 1984). Hjort questioned the effectiveness of these releases and a study was commissioned in 1903 by the Norwegian Parliament (Smith et al., 2002) that required the participation by both Dannevig and Hjort. Hjort assigned Norwegian scientist K. Dahl to work with Dannevig. During spring 1904 and 1905, millions of cod larvae were released into two fjords, and subsequent summertime juvenile abundances were estimated through beach seine sampling. These abundances were compared to summertime abundances in 1903, when no releases were made (Table 3-1). Dannevig reported increased juvenile abundances associated with the releases for 1904–1905 in comparison to 1903, the control year. Dahl, on the other hand, investigated other fjord systems that did not receive hatchery larvae in 1904 and 1905. In 1904, large numbers of juveniles were encountered in the fjords Dahl surveyed regardless of whether releases were made. But in 1905, the abundance of juveniles was relatively scarce in all fjords, again independent of whether larvae were released or not (Table 3-1). Dahl concluded (1909) that, “. . . the formation of the ﬁsh stock (the ﬁsh supply) in these fjords is not really much dependent on the spawning and hatching taking place in the fjord itself, it is more dependent on the quantities of fry brought by currents (the circulation of fry).” Thus, the same study resulted in conﬂicting conclusions. Parliament

TABLE 3-1. Number of Released Larvae (“Fry”) into Different Fjords as Part of G. M. Dannevig’s Cod Hatching Programa Søndeled Fjord

Year

Larvae released

1903 0 1904 33.5 ¥ 106 1905 33 ¥ 106

Helle Fjord

Sandnes Fjord

Støle Fjord

Cod per haul

Larvae released

Cod per haul

Larvae released

Cod per haul

Larvae released

6.3 33.7 11.4

0 0 10 ¥ 106

5.4 10.9 1.5

— 0 0

— 49 4.1

— 0 0

Cod per haul

Christiania Fjord Larvae released

Cod per haul

— — — 112 20 ¥ 106 10.8 2.7 20 ¥ 106 1.9

a The study on the effectiveness used “cod per haul” (number of juveniles collected using beach seines) to index juvenile abundance among years for months of July and August. Dannevig emphasized interannual comparisons in Søndeled and Helle Fjords, 1903–1905, using 1903 as a control year. K. Dahl emphasized between Fjord contrasts in 1904 and 1905 using Sandnes, Støle, and Christiania Fjords as controls. Data from Dahl, 1909 (p. 31).

Fish Migration and the Unit Stock: Three Formative Debates

21

assigned a committee to evaluate the opposing conclusions, which determined that the uncertainty in the ﬁndings dictated continued study. Hatchery releases of cod and juvenile monitoring continued into modern times, and the issue of the effectiveness of hatchery releases of cod into fjords remains a point of contention (e.g., Smith et al., 2002). The diverging views on the effectiveness of hatchery releases, which persist today (e.g., Secor et al., 2000, 2002), center on scale. Dahl maintained that ﬂuctuations in juvenile cod abundances had to be thought about at a spatial scale larger than a fjord. Dannevig subscribed to the modern migration theory, which prescribed restricted movements inshore and offshore by numerous local races (Heincke, 1898; Smith, 1994). Dahl (1909) speciﬁed the theoretical underpinnings of Dannevig’s program: “. . . the importance of the size of the waters was reduced by accepting the doctrine that each area of the sea, even the smallest, possessed its own tribe of ﬁsh. These tribes were supposed to be highly local during the whole life of the individuals. They were easily injured by overﬁshing and had to be replaced by the aid of man.” Based on his studies of differing stages of Norwegian coastal cod, Dahl (1909) thought it “just to consider the ﬁsh stock of a considerable stretch of coast as belonging to the whole of the area.” How could relevant spatial domains be applied to issues of stock renewal? The answer came through migration studies. Because adult cod do not reside the year round in spawning or nursery habitats, ontogenetic and seasonal migration patterns were needed to link stage-speciﬁc distribution patterns. Through early tagging studies, Hjort (1909) conﬁrmed seasonal migrations between Lofoten Island spawning grounds and far-off Barents Sea feeding habitats. For Atlantic herring, Hjort and his colleagues used “certiﬁcates of origin”—unique optical patterns of scale annuli—to chart the seasonal migrations to and from spawning grounds. Based on herring studies, Hjort argued that different stages or sizes of herring located in different regions were in fact members of the same population (Secor, 2002). Such studies, informed by the hatchery investigation, provided Hjort with the critical spatial domain needed to combine samples across relevant scales and observe decadal cycles in abundance of Norwegian herring (Hjort and Lea, 1914). Thus, Hjort used complex life cycles to reveal the spatial scale relevant to stock renewal processes. But how then were complex life cycles maintained? Resolution of this issue came with early studies of the parent stream theory in Paciﬁc salmon.

III. DEBATE 2. THE PARENT STREAM THEORY: DIRECTED MIGRATIONS OR LOCAL WANDERINGS? Prior to the twentieth century, naturalists, commercial ﬁshermen, and anglers long wondered: Where do spawning Paciﬁc and Atlantic salmon originate? Does

22

D. H. Secor

a stream’s spawning run represent a random mixture of mature ﬁsh migrating in from the coast? Or, does each spawning run represent the return of a generation of salmon to their parent’s stream? The speciﬁc migration and spawning behaviors associated with salmon reproduction would seem to represent adaptations that only lineage could guarantee. Thus, the parent stream theory became dogma well in advance of scientiﬁc evidence. Two central issues required resolution: (1) Do salmon home to natal streams? (2) Where do spawning runs of maturing adults originate—from distant oceanic waters or from regional coastal waters? For Paciﬁc salmon, scientiﬁc investigations and debate related to these issues were concentrated during the ﬁrst half of the twentieth century starting with C. H. Gilbert’s important work on natural tags (circuli patterns in scales) and homing for sockeye salmon, and culminating with work on ocean migrations by the International North Paciﬁc Fisheries Commission (Neave, 1964; Royce et al., 1968). Nevertheless, much research remains to be done on mechanisms of migration, which remain controversial (Hansen and Quinn, 1998). Is homing to natal streams obligatory or facultative in Paciﬁc and Atlantic salmon? With 100 years of hindsight, this may seem like a ridiculous question, but as the famous ichthyologist Jordan (1887) pointed out, “It seems more probable than parent stream theory that the young salmon hatched in any river mostly remain in the ocean, within a radius of 20, 30, or 40 miles of its mouth.” This idea was in keeping with the modern migration theory, prevalent in Europe during the late nineteenth century, which viewed seasonal disappearances of diadromous and coastal ﬁshes as restricted migrations to deeper near-shore coastal regions rather than the result of extensive ocean migrations. In this view, homing is not due to directed behaviors ensuring philopatry, but is due only to proximity.

A. NATURAL TAGS Gilbert provided early evidence for the parent stream theory in a series of reports on Fraser River sockeye salmon (1914–1919) with an early application of natural tags. Like his contemporary Hjort (1909, 1914), Gilbert developed a demographic approach based on annuli in scales to explain cycles of abundance. Further, like Hjort, he used scale circuli patterns during the ﬁrst year of life as certiﬁcates of origin. Distinct frequencies of circuli that were associated with the early freshwater juvenile period occurred among tributaries and, on occasion, even within tributaries (Fig. 3-2; Gilbert, 1919). Comparisons between juveniles and adults captured at Fraser River spawning localities showed the same number of circuli corresponding to the freshwater juvenile period, leading Gilbert (1915) to conclude, “Examination of the scales had removed any possible doubt that the progeny of the Fraser River ﬁsh return to the Fraser at their maturity, and that this is true also of the ﬁsh of each of the large river basins.”

23

Number of First Growth Season Circuli

Fish Migration and the Unit Stock: Three Formative Debates

A

30

A D

B C

E

20

10

0

Birkenhead Harrison Hatchery Pitt Lake Hatchery FishCanyon Morris Creek Skookumchuck

Spawning Locality FIGURE 3-2. Box-whisker plots of circuli counts of Fraser River juvenile sockeye salmon, collected in 1918 (Table III; Gilbert, 1919). Not shown due to low sample size (n = 12) is the Hanceville locality. Sample sizes for other sites ranged from 40 to 153. Letters represent similar and discrete statistical groupings (ANOVA; Tukey post hoc test; a = 0.05).

Were these natural tags the result of environment or lineage? Gilbert argued for both causes. He believed that heredity ﬁxed racial differences existed among tributaries, but also stated that “. . . during their life in freshwater salmon are subjected to obviously diverse external conditions. These are most marked perhaps, between salmon inhabiting different watersheds, but undoubtedly exist, if in less degree, between those that live for a year or more in different lakes belonging to the same river basin” (Gilbert, 1915). Although speculative, reconciliation of these two apparently diverging views may have come from Gilbert’s appreciation for trait variance within species and populations: “. . . the peculiarities of each race in the matter of habitat do not lie outside but within the total range of variation as found in other river basins. Nothing new to the species, then, is found in a rare phenomenon, within the range of variation exhibited elsewhere” (Gilbert, 1914). Thus, Gilbert may have recognized well in advance of population genetics (e.g., Fisher, 1930) that heredity leads to modalities in trait expression rather than invariant race-speciﬁc traits. To refute Jordan’s criticism of the parent stream theory, it was insufﬁcient to only recognize trait differences between tributaries—this could be due to local

24

D. H. Secor

ambits rather than philopatry. Initial evidence for philopatry came from observations of low straying rates between adjacent spawning tributaries. For instance, from many thousands of sockeye salmon examined from Rivers Inlet Race, only 24 were observed to exhibit different scale patterns, which were attributed to the adjacent Fraser River race. Gilbert (1917) concluded, “. . . the spawning runs in these streams and tributaries enjoy practical isolation, each from every other. . . .” The question of hatcheries underlay the early work on Paciﬁc salmon, as it did with Norwegian cod. As previously described, Dahl (1909) and Hjort (1909) used the issue of scale to argue against local beneﬁts due to local releases of cod. In contrast, Gilbert’s early work suggested that hatcheries could have local beneﬁts. Gilbert (1915) advised that for hatcheries to be effective, “every stream must receive its own quota of fry.” Using natural tags to evaluate philopatry entails a troubling conundrum. What if the salmon of two tributaries exhibit the same certiﬁcate of origin? Gilbert (1919) noted two possible causes: Racial segregation has not occurred, or segregation has occurred, but has not resulted in divergence in the natural tag. Gilbert and others (e.g., Neave, 1964; Harden Jones, 1968; see also Waldman, 1999) have used multiple phenotypic traits (e.g., growth, age at maturity, length of life in freshwater, color, quality of ﬂesh) to reduce the chance that common certiﬁcates of origin for different races will be misinterpreted. Still, the parent stream theory entails lineage and natural tags can only provide circumstantial evidence. Later, Huntsman (1937a), a chief antagonist of the parent stream theory, would argue strongly against the use of natural tags in support of philopatry: “The characters that have been used to distinguish ‘races’ in species of marine ﬁshes, such as herring and cod, are being demonstrated to result from the action of the environment on the individual during its lifetime, so that it seems doubtful whether there are heritable differences between the populations of different districts.”

B. MARKING EXPERIMENTS Tagging should provide advantages in investigating homing because origins (spawning tributaries and localities) can be identiﬁed without the reliance of distinguishing natural tags. A central problem in early marking experiments for Paciﬁc and Atlantic salmon was low rates of returning adults in comparison to tagged and released juveniles. Return rates of adults from a marked sample of juveniles ranged from 0% to 10% in early studies. What then happened to the majority of tagged juveniles? Were these lost due to straying to other systems, tagging and release mortality, mortality at sea, ﬁshing mortality, or were they

Fish Migration and the Unit Stock: Three Formative Debates

25

somehow otherwise “lost at sea,” as Huntsman and others believed? Interestingly, this debate remains important in investigations on homing mechanisms in salmon. Geolocation homing mechanisms entail high overall ocean mortality but efﬁcient (>90%) homing by adult survivors (Quinn and Groot, 1984; Quinn and Dittman, 1990). Random walk models imply low ocean losses prior to spawning migrations, but less efﬁcient (<10%) homing by adult survivors (Saila and Shappy, 1963; Jamon, 1990). Presumably in such models those adults that do not home, wander and are lost at sea. In his compendium of migration studies, Harden Jones (1968) reviewed decades of early marking experiments, but highlighted the Cultus Lake experiment on Fraser River sockeye salmon conducted by Foerster (1936) as being instrumental in providing evidence for homing, and experiments conducted by A. L. Pritchard as evidence against substantial straying to adjacent tributaries. In successive years, Foerster’s group marked (adipose ﬁn clip) ~105,000 and 365,000 sockeye juveniles as they migrated seaward through a restricted outlet from Cultus Lake. For the larger release, a total of 12,803 returning salmon were recovered over a 3-year period. Of these, 71% were recovered from salmon canneries, which obtained salmon from commercial ﬁsheries in coastal areas approaching the Fraser River. The remaining 2,856 ﬁsh returned to Cultus Lake. Thus, those salmon not captured in coastal ﬁsheries homed to Cultus Lake. Pritchard marked a portion (<5%) of pink salmon during their seaward migration from McClinton Creek in northern British Columbia (Queen Charlotte Islands) during the 1930s. In 1940, he examined returns in McClinton Creek and immediately adjacent streams for ﬁn clips. In McClinton Creek, 781 of 35,521 (2.2 ± 0.1%) ﬁsh were marked. In four adjacent streams, only two marked ﬁsh were observed of 4,075 ﬁsh examined (0.04%). Early marking experiments showed evidence for directed spawning migrations from coastal areas (Foerster, 1936) and lack of straying to adjacent streams. Still, these studies can show strong biases due to low recapture rates, unknown tagging effects, ocean and ﬁshing mortality, and recapture reporting rates. Further, we still cannot conﬁdently distinguish the role of proximity vs. philopatry. For instance, might the propensity of Cultus Lake sockeye salmon to home be due to the sampling domain, which did not include the ocean or coastal approaches to rivers other than the Fraser River itself? Further, subsequent tagging studies showed higher incidence of straying than indicated by the Pritchard study, leading Huntsman (1937b) to later question the prevalence of straying: “In quite a number of instances salmon marked or tagged in one river have been recaptured in another. . . .” These issues would motivate a more manipulative approach to test straying rates by transplanting salmon across distant watersheds.

26

D. H. Secor

C. TRANSPLANT EXPERIMENTS Many early transplant studies addressed practical issues such as restoring depleted runs or enhancing ﬁsheries, but Snyder (1931) deliberately tested the parent stream theory by transplanting marked Chinook salmon juveniles of Sacramento River progeny into the Klamath River, and vice versa. Of 25,000 marked Sacramento juveniles released into the Klamath River, 10 were recovered at sea and 50 were recovered in the Klamath River. None of the 15,000 Klamath River juveniles released into the Sacramento River were recovered outside of coastal areas. This result was evidence that in one instance, spawning adults return to the stream in which they experienced their early growth rather than the stream where their parents resided. If the parent stream theory implied that progeny return to the stream of parents regardless of early rearing environments, then Snyder’s experiment and subsequent ones showing similar results refuted the popular theory. From a practical viewpoint, transplant and marking experiments did support the view that anthropogenic effects on spawning and rearing habitats of Paciﬁc salmon would directly affect returns of subsequent generations.

D. THE YEAR-CLASS PHENOMENON (AGAIN) Much like Norwegian and North Sea herring, region-speciﬁc recruitment cycles in Paciﬁc salmon supported the view that populations were spatially structured. Contemporary to Hjort’s (1914) important observation of the impact of the 1904 year-class on Norwegian herring ﬁsheries, Gilbert (1914) noted large interannual variation in Fraser River sockeye spawning runs with a 4-year dominant yearclass cycle. Based on scales, differences in age structure were noted between the Fraser Rivers Inlet races. Gilbert attributed these cyclical patterns to differential reproductive success among years, which were tributary-speciﬁc. For the upper Fraser River, this cycle was interrupted by construction of a train corridor through Hell’s Canyon, which resulted in a large rock fall into the river (see Smith, 1994, for a review of this case study). Gilbert lamented that the 1917 spawning run “. . . was deprived to a large extent by the progeny of up-river spawners. . . . Four years before, on their ascent of the river, these had been arrested at the canyon and died in the lower river without laying their eggs.” Here, local anthropogenic inﬂuences provided strong circumstantial evidence for the signiﬁcance of homing on population dynamics. To varying degrees the localized inﬂuences of dams, hatcheries, pollution, and ﬁsheries in spawning tributaries have a larger impact on anadromous salmon than other temperate coastal ﬁshes due to the precise nature of their homing (e.g., Schaller et al., 1999).

Fish Migration and the Unit Stock: Three Formative Debates

27

While recruitment cycles and local effects showed that population structure was associated with homing within river systems, they did not provide support for the parent stream theory per se. We have yet to address the second issue prompted by the theory: Where do spawning runs of maturing ﬁsh originate?

E. OCEAN STUDIES In 1937, University of Toronto Professor A. G. Huntsman, in two letters to the journal Science, established himself as a principal skeptic of the parent stream theory. In particular, he was critical of the view that the return spawning migrations begin in distant ocean feeding grounds. “I maintain that the return of salmon to their native rivers from distant feeding grounds in the ocean is not only merely a theory, but one with a frail basis” (Huntsman, 1939). Huntsman’s view was similar to that of Jordan (see previously) and similar in concept to the modern migration theory of the late nineteenth century. Atlantic salmon (and presumably Paciﬁc salmon) moved into coastal waters, proximate to the spawning river’s “zone of inﬂuence.” Once outside of this zone, salmon became lost and may have moved to other rivers or traveled out to sea. Very little was known about ocean distributions of salmon. Why should we expect salmon to occur outside of coastal waters where they were principally harvested? Understanding the oceanic occurrences of salmon would require a decade of concerted ocean studies, which came about due to the rapid emergence of ocean purse-seine and long-line ﬁsheries for Paciﬁc salmon. Harden Jones (1968) cast Huntsman as a stalwart skeptic who presented important challenges and uncertainties to the widely held, but poorly supported parent stream theory. Indeed, Huntsman stated, “For questioning homing, I have suffered the fate of a heretic.” Still, Huntsman held a pretty high standard for this theory throughout his career. Initially, he argued against homing based on natural tag studies, suggesting that there was little evidence for directed homing migrations within tributaries (Huntsman, 1937a, b); then when presented with such evidence, he focused on uncertainties in oceanic migrations and the mechanism of homing: “Homing is a very deﬁnite feature in salmon migration, but it is the end of wandering rather than a directive factor” (Huntsman, 1950). And when multiple mark-recapture data showed that when an individual tagged as a juvenile occurred in distant ocean waters and then was recaptured as an adult in the natal river, Huntsman (1952) argued that “Recognition of home water does not direct the ﬁsh home, but may stop the wandering. Very foreign water may actually be avoided by the ﬁsh.” Huntsman held the view that while some salmon could home from distant waters, most did not. His contemporaries also thought that many salmon must be “lost at sea” (e.g., Saila and Shappy, 1963; Harden

28

D. H. Secor

Jones, 1968). Such views were in part reﬂective of a lack of directed studies on oceanic distributions of Atlantic salmon, a situation that would soon be remedied for Paciﬁc salmon by the International North Paciﬁc Fisheries Commission. In 1955, a convention by ﬁshing nations on North Paciﬁc salmon (Canada, Japan, and the United States) formed the International North Paciﬁc Fisheries Commission and initiated migration and distribution studies of salmon in the North Paciﬁc Ocean and Bering Sea. Data and analyses from the resulting studies conducted in the late 1950s and early 1960s remain fundamental to knowledge and understanding of Paciﬁc salmon oceanic migrations (Pearcy, 1992; Groot and Margolis, 1998). Further, these studies yielded a new dynamic understanding about salmon migration and oceanography. For Atlantic salmon, ﬁsherydependent mark-recapture studies on Atlantic salmon have occurred over the past several decades, but due to the limited scale of these investigations, patterns of oceanic distribution and oceanic migration circuits remain poorly understood (Hansen and Quinn, 1998). Neave (1964) and Royce et al. (1968) summarized much of the Commission’s research ﬁndings for pink, sockeye, and ocean-run steelhead, which included trawling directed at intercepting the migrations of juveniles, extensive mark recapture, and ocean ﬁshery (purse-seine and long-line) data. These sources of evidence left little doubt as to the oceanic dependency of local races. Royce et al. (1968) described entire oceanic migration circuits, the underlying oceanography, and possible mechanisms leading to homing. These oceanic migration circuits occurred as single annual cycles for pink salmon, but were repeatedly undertaken as annual cycles by longer-lived salmons (Fig. 3-3). “Downstream” (conﬂuent with major ocean current) migration circuits were described for British Columbia pink salmon in the Alaska Gyre; for eastern Kamchatka pink salmon in the current system centered on the Bering Sea and North Paciﬁc Ocean; and for Bristol Bay pink salmon in the Bering Sea and North Paciﬁc Ocean (Fig. 33). Neave (1964) focused on migration trajectories from oceanic aggregations to spawning grounds. Based on Japanese high sea ﬁsheries data, he observed “. . . well ordered direction and timing of the movements towards appropriate spawning grounds . . . ,” and concluded that “. . . the evacuation of the high seas is not a random affair. . . .” Thus, the weight of evidence from directed North Paciﬁc studies countered the idea that salmon were lost at sea: “. . . we believe that the salmons’ migrations could not be performed if they migrated or drifted at random . . .” (Royce et al., 1968).

F. THE PARENT STREAM THEORY and POPULATION THINKING The parent stream theory or philopatry is a ﬁrst principle in ﬁsheries science. There is no more important assumption that underlies the management of ﬁshery

Fish Migration and the Unit Stock: Three Formative Debates

29

FIGURE 3-3. Circuits of oceanic migrations of Bristol Bay sockeye salmon based on landings and tagging data. From Figure 17, Royce et al. (1968).

stocks. As Huntsman and Harden Jones emphasized, the parent stream theory not only requires homing, but also requires a complex life cycle. A ﬁsh must leave natal areas to juvenile and adult habitats and then return. The spatial ambit and temporal frequency of this life cycle was codiﬁed in the migration triangle, conceived by Harden Jones (1968) and later modiﬁed by D. H. Cushing (Secor, 2002). For marine ﬁshes, Cushing (1975) elaborated the migration triangle to include the concept of hydrographic containment, that life stages and migrations were constrained within oceanographic systems that favored denatant migrations of juveniles away from spawning areas and countranatant migrations by adults back to spawning areas (Fig. 3-4), a view favored by Harden Jones (1968) based on the many case studies he examined. According to the views of philopatry and hydrographic containment, salmon were not lost or randomly cast about the oceans, but were distributed in time and space according to a complex life cycle designed to guarantee high levels of philopatry. Once such patterns were elucidated, recruitment cycles could be well matched with the inﬂuence of ﬁsheries that were arrayed across the migration circuit. Similarly, other natural and anthropogenic inﬂuences that varied in time and space could be allocated in proportion to their effect on a given spawning population. The studies on Paciﬁc salmon were taken as demonstrative of a general pattern for temperate and boreal marine ﬁshes: “In temperate seas ﬁsh usually return to the same spawning ground each year at the same season, much as do Paciﬁc salmon” (Cushing, 1995, p. 86).

30

D. H. Secor

Adult Stock

t en

Denatant

itm

A

Spawning area

cru Re

De na tan Co t ntr an ata nt

C

B

Nursery area

FIGURE 3-4. Migration triangle as depicted by D. H. Cushing (1982). Cushing substantially modiﬁed F. W. Harden Jones’s (1968) migration triangle, which depicted seasonal migrations of adults rather than an ontogenetic circuit (Secor, 2002).

Despite remaining issues related to the assumption of the parent stream theory and population substructure (see last section), marine ﬁsheries research and management during the last half of the twentieth century moved to a typological view of stock structure and ﬁsh migrations. Migration studies and stock structure studies were directed at determining for each population a single spawning area, a single nursery area, and a single migration triangle (Stephenson, 2002). This approach has been substantially aided with the advent of biochemical markers, some of which represent population lineage and are thus more directly applicable to the issue of parental natality. An interesting late twentieth-century theory related to philopatry was the “member-vagrant” hypothesis (Iles and Sinclair, 1982; Sinclair, 1988), which countered the idea of denatant dispersal of juveniles, but rather emphasized the importance of larval habitats with physical retention features. Those larvae/juveniles that failed to arrive in prescribed nursery areas were “vagrants”—lost at sea. In this view, populations could persist only if spawning behaviors guaranteed that a large fraction of larvae arrived at nursery areas where retention would occur. Interestingly, Harden Jones doubted that single population circuits would survive generations. First, in consideration of homing migrations, “Homing could be a disadvantage when ﬁsh persist in returning to spawn in an area . . . where conditions have become unfavorable. . . . The only biological insurance against this is a satisfactory level of straying and a multiplicity of spawning grounds. . . .” Further, in consideration of population substructure, “. . . the capacity to meet change lies, not in the ﬂexibility of each unit, but the multiplicity of units.” These observations were remarkably prescient of metapopulation and contingent concepts, which have only emerged during the past decade to meet some of the shortcomings of the parent stream theory.

Fish Migration and the Unit Stock: Three Formative Debates

31

IV. DEBATE 3. THE EEL PROBLEM: WHO CONTRIBUTES TO REPRODUCTION? Population thinking in ﬁsheries science largely developed outside ideas related to the role of races in natural selection (e.g., Darwin, 1859). Rather than evolutionary groups, races (populations) were conceived as ecological entities—groups of individuals that had common natality and migration patterns and were therefore affected similarly by exploitation, climate, and environment. Later, the widespread use of neutral lineage markers would depend on theoretical developments of microevolution. Still, stock structure studies were not explicitly related to evolution or speciation. Sinclair (1988) provided an initial theoretical basis for the role of evolution in stock structure by suggesting that population richness was itself a selected attribute, deﬁned by oceanographic retention areas that favored larval survival. In addition, recent emphasis on conservation and biodiversity of aquatic resources has emphasized the relationship between population structure and endangerment of exploited species (e.g., Waples, 1995). Still, there remains little cross-fertilization between ﬁsheries scientists and evolutionary biologists who work on ﬁsh (Conover, 2000). As recent evidence, consider that a series of papers devoted to the species concept in ﬁshes, published in the journal Fish and Fisheries [Vol. 3(3); Stauffer et al., 2002], failed to address the role of population thinking in the species concept. Indeed, the journal issue did not contain a single reference to population or stock structure in ﬁshes. Within this dearth of studies on the species concept and population thinking in ﬁsheries, there exists the so-called eel problem. During the past century, the linked problems of natality and migrations of Atlantic eels have required simultaneous thought about population structure and speciation. Indeed, debate on the role of population structure and speciation remains active for Anguilla species and important for their management (e.g., Tsukamoto and Aoyama, 1998; Tsukamoto et al., 2002). Recurring as part of the resolution to the eel problem is the view that adult eels in major portions of their range do not reproduce (Tucker, 1959; Harden Jones, 1968; Tsukamoto et al., 1998), or that some eels disproportionately contribute to reproduction dependent on the habitats to which they recruit (Limburg et al., 2003). That some eels are destined never to reproduce seems a peculiar idea, but such proposals have been advanced for salmon due to ocean wanderings (see previously) and for vagrant ﬁshes that range too far from their migration triangle (Sinclair, 1988). Might some juvenile/adult habitats and associated stocks require constant subsidy from reproduction from adults that grew elsewhere? Managing the effects of exploitation would engender risk if reproducers were selectively harvested or otherwise affected (Castonguay et al., 1994), an issue now attracting consideration in the design of exploitation refuges and marine protected areas.

32

D. H. Secor

A. MEDITERRANEAN vs. NORTH ATLANTIC ORIGIN

OF

EELS

Well after Aristotle’s musings on spontaneous generation of eels in mud, evidence for the marine origins of eels began with the Italian naturalist G. B. Grassi, who recognized that Leptocephalus brevirostrum was in fact the larval form (termed leptocephalus) of Anguilla (Grassi and Calandruccio, 1897). Based on local collections of leptocephali, Grassi assumed that Anguilla spawned in the Straits of Messina (narrow straights north of Sicily). Proximity between inland freshwater ﬁsheries for eels and coastal spawning areas (the modern migration theory) implied local populations of eels. “If Grassi’s hypothesis was correct, there were likely to be problems of over-ﬁshing if too many eels were trapped on their way to the sea, the implication being that each river was supplied by elvers derived from a local spawning ground” (Harden Jones, 1968). Schmidt, through decades of directed ocean collections of the pelagic leptocephali, discovered that eels in the Mediterranean and North Atlantic originated in the Sargasso Sea (Schmidt, 1922). The Danish scientist’s efforts in tracking ever-decreasing sizes of leptocephali across the Atlantic represented important pioneering research in ﬁsheries oceanography (Jakobsson, 2002). Schmidt argued for a single spawning region in the Sargasso Sea for eel stocks throughout Northern and Southern Europe. Evidence against a separate Mediterranean spawning location included: (1) similar myomere counts for leptocephali captured in the Mediterranean with those captured in the Northeast Atlantic; (2) increased leptocephalus size inside the Mediterranean in comparison to outside the Mediterranean, implying older ages and longer periods of dispersal from the Sargasso Sea; and (3) earlier seasonal occurrences of leptocephali outside of the Strait of Gibraltar in comparison to those captured off Italy.

B. NORTH ATLANTIC EELS: SPECIATION POPULATION STRUCTURE?

OR

That the European eel life cycle must include distant oceanic spawning in the Sargasso Sea was a surprising discovery; one not well accommodated by the emphasis of racial differences and population structure emerging from the ICES community (Hjort, 1909; 1914). If all eels, from the Baltic to the Mediterranean, were the result of spawning in the Sargasso and years of larval drift along the Gulf Stream, then all eel ﬁsheries were dependent on one panmictic population: species and population were one and the same. Further, the Sargasso Sea not only included all European eel A. anguilla spawning, but also proved to be the only spawning region for the congeneric American eel A. rostrata. Indeed, in many of Schmidt’s samples both species (distinguished by myomere counts) co-occurred as leptocephali. This ﬁnding presented a particularly difﬁcult conundrum: Over

Fish Migration and the Unit Stock: Three Formative Debates

33

generations of presumed co-occurrence of spawning in the Sargasso Sea what led to maintenance of distinct eel species? Schmidt’s hypothesis was that leptocephali of European eels were entrained into the Gulf Stream and drifted for 2 to 3 years until they reached regions of Northern Europe (earlier) and the Mediterranean (later), whereupon they metamorphosed into glass eels and entered the estuaries and rivers of Europe. Juvenile habitats in North America were substantially more proximate to the Sargasso Sea, and leptocephali could be expected to reach most of these habitats within one year. Thus, if the leptocephalus development rates and stage durations were substantially different between American and European eels, then speciation might be maintained over generations. Evidence for this came from (1) drift bottle studies, which suggested leptocephalus drift must be substantially greater for eels arriving in European as opposed to North American estuaries; (2) European leptocephali that were over 33% longer at metamorphosis than American eels; (3) slower growth and development rates of European eels vs. American eels based on an assumed overlapping spawning season during spring (Harden Jones, 1968) (for instance, when European eels passed Bermuda, they were not competent to ingress into freshwater); and (4) sample sizes of A. rostrata leptocephali collected by Schmidt that were substantially less than A. anguilla leptocephali, corresponding to differences in ﬁshery yields between the two species. (While early twentieth-century harvest records are probably insufﬁcient to make this claim, production of European eels is likely higher. From 1980 to 2000, ﬁshery yields were an order of magnitude higher for A. anguilla in comparison to A. rostrata.) Still, without direct observations of spawning eels, these lines of evidence only provided circumstantial support for speciation in North Atlantic eels.

C. FAILED ADULTS In an inﬂuential Nature article, Tucker (1959) played the lone role of skeptic by suggesting an alternative set of explanations for regulation of eel stocks produced by Sargasso Sea spawning. Eel species were not distinct, but represented different phenotypes due to environmentally mediated meristic characters. He proposed that temperature differences across the large region over which small leptocephali had been observed in the Sargasso Sea could generate meristic differences in vertebral (myomere) counts. Different origins within the Sargasso Sea not only resulted in different dispersal fates (North America or Europe), but also associated vertebral counts. But this proposal was not the most controversial part of Tucker’s thesis. He forcefully argued that the distance for adults to migrate from European waters to the Sargasso Sea was too vast. “European eels . . . do not succeed in returning to the ancestral spawning ground, but perish in their own

34

D. H. Secor

continental waters.” Tucker believed that stocks of so-called European eels were sustained by American eel parentage. The idea of failed reproduction was similar to the view that much Paciﬁc salmon production was lost at sea (see previously), but here stock structure of eels had consequence: If stocks failed in North America, so too would they fail in Europe. A scientiﬁc debate on Tucker’s proposal quickly emerged. Chief criticisms were against temperature affecting stable differences in myomere counts in American and European eels and Tucker’s view that “. . . Europe and North Africa are regularly colonized by eels of American ancestry doomed to perish in a fruitless suicide-migration” (Tucker, 1959). D’Ancona (1959) argued that Tucker’s hypothesis of temperature gradient differences would produce continuous variation in myomere counts rather than the observed bimodal pattern unless there was a threshold or critical temperature. Tesch (1977) and Harden Jones (1968) argued that expected temperature effects on vertebral counts from salmonid studies would lead to a difference of less than one vertebral count rather than seven, which would be required to explain the observed difference between European and American eels. In recent decades, karyotyping and other genetic studies (Passakas, 1981; Lintas et al., 1998) have determined the separation of species on grounds other than meristics. The issue of whether European eels can successfully return to the Sargasso Sea has been more difﬁcult to resolve and remains an active area of speculation. Tucker remarked that it was surprising that no emigrating silver (mature) eels had ever been observed in or near the Strait of Gibraltar. Others stated that the reason for an apparent absence of silver eels there and elsewhere was lack of ﬁsheries and appropriate gear to observe returning eels (D’Ancona, 1959; Deelder, 1960). From an energetics perspective Tucker articulated this colorful metaphor: “. . . the advanced modiﬁcations of the European eel appear to be ill-adapted to . . . a return journey. We may indeed know that a motorist is in the best of health but, as we watch him driving down a mountain road, intoxicated and without brakes, we can still legitimately say that his celebration was premature that he is likely to reach his destination.” Did European eels have sufﬁcient reserves to undertake a 3,000- to 4,000-km migration to the Sargasso Sea? Based on telemetry studies, Tesch (1977) argued that with a fat content of >20%, and using advantageous currents, an adult eel could accomplish a 6,000-km migration in 5 months. More recent studies have failed to deﬁnitively resolve whether energetic stores are sufﬁcient to undertake spawning migrations from waters as distant as the Baltic Sea. Svedäng and Wickström (1997) observed that many eels occurring in the Baltic Sea had limited fat reserves (<10%) and proposed that most of these eels may delay maturation and migration to the Sargasso Sea. Further, they believed that transformation to the adult silver eel form and related migration behaviors may be reversible. Van Ginneken and

Fish Migration and the Unit Stock: Three Formative Debates

35

vanden Thillart (2000) conducted swimming metabolism studies on large eels and predicted that only 40% of the total fat reserves of a 2-kg eel with a 20% fat content would be utilized for a 6,000-km spawning migration. Still, a 2-kg eel would represent an extremely large eel. For a more typical size of 500 g, approximately 125% of fat reserves would be required to accomplish a 6,000-km spawning migration. Limburg et al. (2003) estimated a range of possible costs for eels of different sizes and origins, of which only a subset would be able to accomplish a 6,000-km spawning run based on energy reserves. While the issue of speciation in eels has been resolved through genetic studies, there remains the question of whether more distant yellow eel stocks such as those in the Baltic Sea substantially contribute to the pool of spawners.

D. “FRESHWATER EELS”

AND

CONTINGENT THINKING

Anguilla eels are commonly referred to as freshwater eels because it is assumed that their longest juvenile period, the yellow eel-phase, is dependent on freshwater habitats. Still, important ﬁsheries are centered in estuaries (ICES, 2001), and scientists have suggested that brackish water regions may represent better growth habitats than freshwater systems (Helfman et al., 1987). In a recent letter to Nature, Tsukamoto et al. (1998) made the noteworthy discovery that some A. anguilla complete their life cycles without ever moving into freshwater. Further, based on a small sample of silver eels collected from the North Sea and using otolith microconstituent analysis, they concluded that freshwater eels did not contribute to the spawning population. Tsukamoto’s speculation on failed reproduction by freshwater eels was controversial, and subsequent analysis of a larger sample indicated that some Baltic Sea silver eels did in fact originate from freshwater eels (Limburg et al., 2003). Still, estuarine and marine eels made up the majority of these silver eels. Tsukamoto et al.’s observation of divergent life cycles within eel stocks has been conﬁrmed for several Anguilla species worldwide based on otolith microconstituent analysis (Arai et al., 2000; Tzeng et al., 2000; Tsukamoto and Arai, 2001; Limburg et al., 2003; Morrison et al., 2003). Thus, the issue of how freshwater vs. brackish and marine habitats contribute to reproduction is important for eel management, particularly because freshwater and estuarine systems are differentially affected by exploitation and habitat degradation (ICES, 2001). The observed modalities of individuals, within the same population, structured by discrete migration is well epitomized by eels, but in fact has been observed frequently for other marine species. Secor (1999) termed groups of individuals structured by lifetime migration patterns as contingents [the term contingent was based on early usage by Hjort (1914) and Gilbert (1914)] and proposed that

36

D. H. Secor

contingents represented important components of population substructure, particularly in strategies of spatially explicit ﬁsheries and habitat management. This is exempliﬁed by a recent proposal to establish exploitation refuges in freshwater eel habitats where little exploitation now occurs but could develop in the future (ICES, 2001). If freshwater contingents produce relatively few reproducing adults, then such a management strategy would prove ineffective (Morrison and Secor, 2003).

E. MORE EEL PROBLEMS

AND

POPULATION THINKING

The issue of speciation mechanisms for Atlantic eels has resurfaced in recent research. Wang and Tzeng (2000) observed hatch dates, which were fairly discrete for each species. Although differing substantially in season, direct observations of small leptocephali by McCleave and Kleckner (1987) also supported the view of allopatric spawning by A. anguilla and A. rostrata. While the assumption of daily increment formation in otoliths has not been substantiated, analyses indicate that dispersal is of much shorter duration for European eels (9–18 months) than suggested by Schmidt and other early investigators (LecomteFiniger and Yahyaoui, 1989; Wang and Tzeng, 2000). This shorter duration has called into question dispersal mechanisms for European eel leptocephali that now have an insufﬁcient larval duration to allow transport by the Gulf Stream. McCleave et al. (1998) have proposed an alternative route by prevailing northeast ﬂow from northeast regions of the Sargasso Sea. In contrast to recent oceanographic and otolith studies, genetics studies have further muddied the issue of population structure in Atlantic eels. While A. rostrata remains consistently different from A. anguilla when genetic markers are used (e.g., Tsukamoto and Aoyama, 1998; Lehmann et al., 2000), Wirth and Bernatchez (2001) interpreted microsatellite DNA as evidence for population structure within A. anguilla. They observed a latitudinal gradient in genetic differentiation and suggested that population structure was maintained through differential seasonality in silver eel return migrations. In a second independent study, further evidence was given for nonrandom genetic structure of A. anguilla stocks (Daemen et al., 2001). Still, a more conservative genetic marker (mitochondrial DNA) showed genetic homogeneity of A. anguilla (Lintas et al., 1998), and the conclusion of population structure in Atlantic eels remains controversial, in part because it is difﬁcult to conceive of a likely mechanism to retain population structure. Flaming controversy, an Italian scientist has sought to redeem history (Casellato, 2002): “[Recent genetic studies] could lead to support of . . . Grassi, who was convinced that European eels could reproduce in the Mediterranean Sea.” The mystery of speciation and population structure in Atlantic eels will endure until spawning eels are captured (Tsukamoto et al., 2003). Issues of how eel

Fish Migration and the Unit Stock: Three Formative Debates

37

stocks contribute to overall reproduction remain very active and important, particularly as both species have experienced large declines in ﬁshery yields during the past 20 years (ICES, 2001). The question of who reproduces, epitomized by eel investigations over the past century, has become critical as we become more reliant on management strategies of regional ﬁshing allocation, exploitation refuges, and habitat protection. The eel example also demonstrates that such strategies may sometimes necessitate alternatives to “population thinking” (e.g., contingent and metapopulation thinking).

V. SUMMARY These three examples of population structure in marine ﬁshes support the strongly held views of Hjort and Gilbert that dynamics of living resources occurred within spatial domains deﬁned by population migration circuits. There can be no doubt after decades of genetic study that salmon and most other marine ﬁshes show varying degrees of homing. Fish commonly return to restricted spawning regions, and migration trajectories are seasonal and connect important reproductive, nursery, feeding, and winter habitats. Still, Hjort and Gilbert also noted an important source of biodiversity within the migration circuits. Nordland “fat” herring undertook a different migration circuit than the other Norwegian herring; a contingent of anadromous sockeye salmon did not partake in oceanic migrations. Later, Tsukamoto observed that some freshwater eels in fact persisted in marine waters throughout their life history. These deviations have been ignored during the ﬁrst century of ﬁsheries science, in part due to the view that populations are each deﬁned by a single migration triangle (Secor, 2002). Biodiversity in spatial behaviors within populations can have signiﬁcant consequences on recruitment rates to regional ﬁsheries, efﬁciency of hatchery enhancement programs, anthrophogenic effects due to degraded or lost habitats, exposure of ﬁsh to contaminants, and overall resiliency of a population or metapopulation to longer term climate or anthropogenic change. Here, Booke’s (1981) call to operationally deﬁne stock based on the issue at hand is prudent. Regardless of the complexity of population structure, stock can be deﬁned at scales appropriate to the question at hand. Still, an increased focus on resiliency and conservation by policy makers and managers will mean that stock deﬁnition will encompass a broader perspective in space and time (Fig. 3-1). Further, many of the assumptions underlying population thinking remain unveriﬁed or inadequate. Consider the following: 1. In several presumably large subocean basin populations, it is increasingly evident that local renewal processes may be important to population and metapopulation resiliency. Large temperate ﬁsheries for Norwegian spring-spawning herring, Japanese sardine, and Scotian Shelf Atlantic

38

D. H. Secor

cod have all experienced severe range contraction into local coastal or shelf regions during nadirs of population abundance (Hutchings, 1996; Watanabe and Wada, 1997; Holst et al., 2002). Thus, during decadal phases of collapsed abundance, local subpopulations may be critical as sources for the next population expansion. Also, through use of oceanographic models and otolith microconstituent analysis, retentive life cycles (“self-recruitment”) have been observed in pelagic spawning species despite the large potential for advection, drift, and straying (Jones et al., 1999; Thorrold et al., 2001; Swearer et al., 2002; Sponaugle et al., 2002). Thus, local processes may be important in sustaining large opensystem populations. 2. Knowledge of the underlying mechanism of homing is incomplete for salmon and unstudied for the vast majority of ﬁshes managed as unit stocks. For salmon, Great Lake studies support the role of olfaction as a local/regional means for identifying natal stream (Hasler and Scholz, 1983), but mechanisms of oceanic migration remain unknown and speculative (Hansen and Quinn, 1998). Still, rapid progress is being made in the use of electronic tags in oceanic migration studies, which may provide critical clues about possible homing mechanisms (Friedland et al., 2001). Indeed, the lack of a certain homing mechanism in salmon has kept Huntsman’s views of oceanic wanderers alive (Jaman, 1990; but see Quinn and Groot, 1984, for evidence against this view). Thus, in salmon and other marine ﬁshes there remains no complete explanation for the most important process underlying the unit stock—philopatry. 3. The emphasis on natural and lineage markers in deﬁning the unit stock has not always provided useful resolution in stock structure issues. The confusion in using different types of markers should highlight that stock is not always clearly deﬁned (Begg and Waldman, 1999). Also, markers result from processes that range in scale from ecological to generational to evolutionary (Waldman, 1999). Thus, deﬁning stock in an operational manner should dictate which markers are used (phenotypic vs. lineage), rather than vice versa. Further, we should not take the term stock to be synonymous with population, as has been done in the past (e.g., Booke, 1981; Pitcher and Hart, 1982). Stock (Schmidt, 1909: “source of ﬁsh”) is a speciﬁc portion of a population that is inﬂuenced by an anthropogenic activity that affects population productivity (Russell, 1931; Ricker, 1975; Secor, 1999). Similarly, tests of the parent stream theory using natural or genetic markers do not test the same assumptions related to natality— natural tags focus on source habitat; genetic markers focus on source lineage. It should be noted that the use of natural tags and electronic tags in recent years has seen rapid progress in better tests of homing in marine ﬁshes and in more precisely quantifying anthropogenic

Fish Migration and the Unit Stock: Three Formative Debates

39

inﬂuences on ﬁsh populations (e.g., Block et al., 1999; Zlokovitz and Secor, 1999). 4. Alternative migratory pathways remain poorly accommodated in ﬁsheries and habitat management. Unusual occurrences that were previously viewed as anomalies, strays, or vagrants were ignored because they did not match expectations for a given migration circuit. The “multiplicity of units” within a population (Harden Jones, 1968) or contingent structure was recognized early but forgotten for many decades, which favored research and management directed in support of a more monolithic view of migration—the migration triangle. The role of contingents and strays in ﬁshery problems has again emerged with the advent of changed management goals and new stock identiﬁcation methods (Secor, 1999; Secor et al., 2001; Limburg et al., 2001; Tzeng et al., 2000).

ACKNOWLEDGMENTS Support for preparation of this manuscript came from the National Science Foundation (OCE0324850). I thank J. Waldman and A. Folkvord for comments on an earlier draft. This is Contribution No. 3774 of the University of Maryland Center for Environmental Science.

REFERENCES Arai, T., Otake, T., and Tsukamoto, K. 2000. Timing of metamorphosis and larval segregation of the Atlantic eels Anguilla rostrata and A. anguilla, as revealed by otolith microstructure and microchemistry. Marine Biology 137: 39–45. Beck, M. W. and 12 others. 2001. The identiﬁcation, conservation and management of estuarine and marine nurseries for ﬁsh and invertebrates. Bioscience 51: 633–641. Begg, G. A. and Waldman, J. R. 1999. A holistic approach to ﬁsh stock identiﬁcation. Fisheries Research 43: 37–46. Block, B. A., and 10 others. 1999. Migratory movements, depth preferences, and thermal biology of Atlantic blueﬁn tuna. Science 293: 1310–1314. Booke, H. E. 1981. The conundrum of the stock concept—are nature and nurture deﬁnable in ﬁshery science? Canadian Journal of Fisheries and Aquatic Sciences 38: 1479–1480. Casellato, S. 2002. European eel: a history which must be rewritten. Italian Journal of Zoology 69: 321–324. Castonguay, M., Hodson, P. V., Couillard, C. M., Eckersley, M. J., Dutil, J.-D., and Verreault, G. 1994. Why is recruitment of the American eel, Anguilla rostrata, declining in the St. Lawrence River and Gulf? Canadian Journal of Fisheries and Aquatic Sciences 51: 479–488. Conover, D. O. 2000. Darwinian ﬁshery science. Marine Ecology Progress Series 208: 303–307. Cushing, D. H. 1975. Marine Ecology and Fisheries. Cambridge University Press, Cambridge, UK. 277 pp. Cushing, D. H. 1982. Fisheries Biology: A Study in Population Dynamics. 2nd ed. University of Wisconsin Press, Madison. 295 pp. Cushing, D. 1995. Population Production and Regulation in the Sea: A Fisheries Perspective. Cambridge University Press, Cambridge. UK. 354 pp.

40

D. H. Secor

Daemen, E., Cross, T., Ollevier, F., and Volckaert, F. A. M. 2001. Analysis of the genetic structure of European eel (Anguilla anguilla) using mitochondrial DNA and mtDNA markers. Marine Biology 139: 755–764. Dahl, K. 1909. The problem of sea ﬁsh hatching. Special Part B, No. 5. In J. Hjort (ed.), Rapport sur les Travaux de Commision A, Dans La Période 1902–1907. Conseil Permanent International pour l’Exploration de La Mer. Rapports et Procès-Verbaux, Vol. X. D’Ancona, U. 1959. Old and new solutions to the eel problem. Nature 4672: 1405–1406. Darwin, C. R. 1859. The Origin of the Species. John Murray, London. Deelder, C. L. 1960. The Atlantic eel problem. Nature 4713: 589–591. Fisher, R. A. 1930. The Genetical Theory of Natural Selection. Oxford University Press, London, UK. Foerster, R. E. 1936. The return from the sea of sockeye salmon (Onchorhynchus nerka) with special reference to percentage survival, sex proportions and progress of migration. Journal of the Biological Board of Canada 3: 26–42. Friedland, K. D., Walker, R. V., Davis, N. D., Myers, K. W., Boehlert, G. W., Urawa, S., and Ueno, Y. 2001. Open-ocean orientation and return migration routes of chum salmon based on temperature data from data storage tags. Marine Ecology Progress Series 216: 235–252. Gilbert, C. H. 1914. Contributions to the Life History of the Sockeye Salmon. Report to British Columbia Fisheries Department, Vancouver, BC. 26 pp. Gilbert, C. H. 1915. Contributions to the Life History of the Sockeye Salmon. Report to British Columbia Fisheries Department, Vancouver, BC. 64 pp. Gilbert, C. H. 1917. Contributions to the Life History of the Sockeye Salmon. Report to British Columbia Fisheries Department, Vancouver, BC. 80 pp. Gilbert, C. H. 1918. Contributions to the Life History of the Sockeye Salmon. Report to British Columbia Fisheries Department, Vancouver, BC. 52 pp. van Ginneken, V. J. T. and van den Thillart, G. E. E. J. M. 2000. Eel fat stores are enough to reach the Sargasso. Science 403: 156–157. Groot, C. and Margolis, L. (eds). 1998. Paciﬁc Salmon Life Histories. UBC Press Vancouver, BC. 564 pp. Gulland, J. A. 1983. Fish Stock Assessment. Wiley, New York. 223 pp. Hansen, L. P. and Quinn, T. P. 1998. The marine phase of the Atlantic salmon (Salmo salar) life cycle, with comparisons to Paciﬁc salmon. Canadian Journal of Fisheries and Aquatic Sciences 55 (suppl. 1): 104–118. Harden Jones, F. R. 1968. Fish Migration. St. Martin’s, New York. 325 pp. Hasler, A. D. and Scholz, A. T. 1983. Olfactory Imprinting and Homing in Salmon: Investigations into the Mechanism of the Imprinting Process. Springer-Verlag, Berlin. 134 pp. Heincke, F. 1898. Naturgeschicte des Herings I. Die Lokalformen und die Wanderungen des Herings in den europäischen Meeren. Abhandlung des Deutschen Seeﬁscherei-Vereins. 2,1. 136 pp. Helfman, G. S., Facey, D. E., Stanton Hales, J. L., Earl, J., and Bozeman, L. 1987. Reproductive ecology of the American eel. American Fisheries Society Symposium 1: 42–56. Hjort, J. 1909. A. General Part. II. Summary of the Results of the Investigations. In J. Hjort (ed.), Rapport sur les Travaux de Commision A, Dans La Période 1902–1907, pp. 20–159. Conseil Permanent International pour l’Exploration de La Mer. Rapports et Procès-Verbaux, Vol. X. Hjort, J. 1914. Fluctuations in the great ﬁsheries of Northern Europe. Conseil Permanent International pour l’exploration de la Mer, 20: 1–228. Hjort, J. and Lea, E. 1914. The age of herring. Nature 94: 60–61. Holst, J. C., Dragesun, O., Hamre, J., Misund, O. A., and Østvedt, O. J. 2002. Fifty years of herring migrations in the Norwegian sea. ICES Marine Science Symposium 215: 352–360. Huntsman, A. G. 1937a. “Migration” and “homing” of salmon. Science NY 85: 313–314.

Fish Migration and the Unit Stock: Three Formative Debates

41

Huntsman, A. G. 1937b. “Races” and “homing” of salmon. Science NY 85: 582–583. Huntsman, A. G. 1939. Races and homing instinct. Salmon and Trout Magazine 96: 233–237. Huntsman, A. G. 1950. Factors which may affect migration. Salmon and Trout Magazine 130: 227–239. Huntsman, A. G. 1952. Wandering versus homing in salmon. Salmon and Trout Magazine 136: 185–191. Hutchings, J. A. 1996. Spatial and temporal variation in the density of northern cod and a review of hypotheses for the stock’s collapse. Canadian Journal of Fisheries and Aquatic Sciences 53: 943–962. ICES (International Council for the Exploration of the Sea). 2001. Report of the EIFAC/ICES Working Group on Eels. ICES CM 2001/ACFM:03, Copenhagen, Denmark. (http://www.ices.dk/reports/ acfm/2000/wgeel/wgeel00.pdf) Iles, T. D. and Sinclair, M. 1982. Atlantic herring—stock discreteness and abundance. Science 215: 627–633. Jakobsson, J. 2002. Ocean travelers. ICES Marine Science Symposium 215: 330–342. Jamon, M. 1990. A reassessment of the random hypothesis in the ocean migration of Paciﬁc salmon. Journal of Theoretical Biology 143: 197–213. Jordan, D. S. 1887. Bulletin of the U.S. Fisheries Commission. Jones, G. P., Milicich, M. J., Emslie, M. J., and Lunow, C. 1999. Self-recruitment in a coral reef ﬁsh population. Nature 6763: 802–804. Lecomte-Finiger, R. and Yahyaoui, A. 1989. Otolith microstructure in the study of the early life history of the European eel Anguilla anguilla. Comptes Rendus de l’Academie des Sciences, Serie III. Sciences de la Vie/Life Sciences 308: 1–7. Lehmann, D., Hettwer, H., and Taraschewsk, H. 2000. RAPD-PCR investigations of systematic relationships among four species of eels (Teleostei: Anguillidae), particularly Anguilla anguilla and A. rostrata. Marine Biology 137: 195–204. Limburg, K. E., Landergren, P., Westin, L., Elfman, M., and Kritsiansson, P. 2001. Flexible modes of anadromy in Baltic sea trout: making the most of marginal streams. Journal of Fish Biology 59: 682–695. Limburg, K. E., Wickström, H., Svedäng, H., Elfman, M., and Kristiansson, P. 2003. Do stocked freshwater eels migrate? Evidence from the Baltic suggests “yes.” American Fisheries Society Symposium 33: 275–284. Lintas, C., Hirano, J., and Archer, S. 1998. Genetic variation of the European eel (Anguilla anguilla). Molecular Marine Biology and Biotechnology 7: 263–269. McCleave, J. D. and Kleckner, R. C. 1987. Distribution of leptocephali of the catadromous Anguilla species in the western Sargasso Sea in relation to water circulation and migration. Bulletin of Marine Science 41: 789–806. McCleave, J. D., Brickley, P. J., O’Brien, K. M., Kistner, D. A., Wong, M. W., Gallagher, M., and Watson, W. M. 1998. Do leptocephali of the European eel swim to reach continental waters? Status of the question. Journal of the Marine Biological Association UK 78: 285–306. Morrison, W. E. and Secor, D. H. 2003. Demographic attributes of yellow-phase American eels in the Hudson River Estuary. Canadian Journal of Fisheries and Aquatic Sciences 60: 1487–1501. Morrison, W. E., Secor, D. H., and Piccoli, P. M. 2003. Estuarine habitat use by Hudson River American eels as determined by otolith strontium:calcium ratios. American Fisheries Society Symposium 33: 87–99. Nakken, O. 2002. Understanding environmental controls on the ﬁsh stocks and progress towards their inclusion in ﬁsh stock assessment. ICES Marine Science Symposium 215: 247–255. Neave, F. 1964. Ocean migrations of Paciﬁc salmon. Journal of the Fisheries Research Board Canada 21: 1227–1244.

42

D. H. Secor

Passakas, T. 1981. Comparative studies on the chromosomes of the European eel (Anguilla anguilla L.) and the American eel (Anguilla rostrata Le Sueur). Folia Biologica 29: 41–58. Pearcy, W. G. 1992. Ocean Ecology of North Paciﬁc Salmonids. Washington Sea Grant Program, Seattle. 179 pp. Pitcher, T. J. and Hart, P. J. B. 1982. Fisheries Ecology. Chapman and Hall, New York. 414 pp. Pritchard, A. L. 1939. Homing tendency and age at maturity of pink salmon (Oncorhyncus gorbuscha) in British Columbia. Journal of Fisheries Research Board of Canada 4: 233–251. Quinn, T. P. and Groot, C. 1984. Paciﬁc salmon (Oncorhynchus) migrations: orientation versus random movement. Canadian Journal of Fisheries and Aquatic Sciences 41: 1319–1324. Ricker, W. E. 1975. Computation and Interpretation of Biological Statistics of Fish Populations. Canada Fisheries and Oceans, Ottawa, Ontario. 382 pp. Rognerud, C. 1887. Hatching cod in Norway. Bulletin of the U.S. Fisheries Commission 7(8): 113– 119. Royce, W. F., Smith, L. S., and Hartt, A. C. 1968. Models of oceanic migrations of Paciﬁc salmon and comments on guidance mechanisms. U.S. Fisheries Bulletin 66: 441–462. Russell, E. S. 1931. Some theoretical considerations on the “overﬁshing” problem. Conseil Permanent International pour l’Exploration de la Mer 6: 3–20. Saila, S. A. and Shappy, R. A. 1963. Random movement and orientation in salmon migration. Journal Conseil Permanent International pour l’Exploration de la Mer 28: 153–166. Schaller, H. A., Petrosky, C. E., and Langness, O. P. 1999. Contrasting patterns of productivity and survival rates for stream-type Chinook salmon (Oncorhynchus tshawytscha) populations of the Snake and Columbia Rivers. Canadian Journal of Fisheries and Aquatic Sciences 56: 1031–1045. Schmidt, J. 1909. The distribution of the pelagic fry and spawning regions of the Gadoids in the North Atlantic from Iceland to Spain. In J. Hjort (ed.), Rapport sur les Travaux de Commision A, Dans La Période 1902–1907, Special Part B, No. 4. Conseil Permanent International pour l’Expl oration de La Mer. Rapports et Procès-Verbaux, Vol. X. Schmidt, J. 1922. The breeding places of the eel. Philosophical Transactions of the Royal Society London, B 211: 179–208. Schwach, V. 1998. The impact of artiﬁcial hatching on cod on marine research. In Sonderdruck aus Historich-Meereskundliches Jahrbuch, Band 5, pp. 27–47. Deutches Museum für Meereskunde and Fischerei. Stralsund, Germany. Secor, D. H. 1999. Specifying divergent migrations in the concept of stock: the contingent hypothesis. Fisheries Research 43: 13–34. Secor, D. H. 2002. Historical roots of the migration triangle. ICES Journal of Marine Science 215: 329–335. Secor, D. H., Arevyev, V., Nikolaev, A., and Sharov, A. 2000. Restoration of sturgeons: lessons from the Caspian Sea sturgeon ranching program. Fish and Fisheries. 1: 215–230. Secor, D. H., Rooker, J. R., Zlokovitz, E., and Zdanowicz, V. S. 2001. Identiﬁcation of riverine, estuarine, and coastal contingents of Hudson River striped bass based on otolith elemental ﬁngerprints. Marine Ecology Progress Series 211: 245–253. Secor, D. H., Hines, A. H., and Place, A. R. 2002. Japanese Hatchery-Based Stock Enhancement: Lessons for the Chesapeake Bay Blue Crab. MD Sea Grant Report UM-SG-TS-2002-02. College Park, MD. 44 pp. Sinclair, M. 1988. Marine Populations: An Essay on Population Regulation and Speciation. Washington Sea Grant Program, Seattle, WA. 252 pp. Sinclair, M. and Solemdal, P. 1988. The development of “population thinking” in ﬁsheries biology between 1878 and 1930. Aquatic Living Resources, 1: 189–213. Sinclair, M. M. and Smith, T. D. 2002. The notion that ﬁsh species form stocks. ICES Marine Science Symposium 215: 297–304.

Fish Migration and the Unit Stock: Three Formative Debates

43

Smed, J. and Ramster, J. 2002. Overﬁshing, science and politics: the background in the 1890s to the foundation of the International Council for the Exploration of the Sea. ICES Marine Science Symposium 215: 13–21. Smith, T. D. 1994. Scaling Fisheries, the Science of Measuring the Effects of Fishing, 1855–1955. Cambridge University Press, Cambridge, UK. 392 pp. Smith, T. D., Gjøsæter, J., Christian Stenseth, N., Kittilsen, M. O., Danielssen, D. S., and Tveite, S. 2002. A century of manipulating recruitment in coastal cod populations: the Flødevigen experience. ICES Marine Science Symposium 215: 402–415. Snyder, J. O. 1931. Salmon of Klamath River, California. Bulletin Department of Fish and Game California. 129 pp. Solemdal, P. and Sinclair, M. 1989. Johan Hjort—founder of modern Norwegian ﬁshery research and pioneer in recruitment thinking. Rapport Permanent International pour l’Exploration de la Mer 191: 339–344. Solemdal, P., Dahl, E., Danielssen, D. S., and Moksness, E. 1984. The cod hatchery in Flødevigen— background and realities. In E. Dahl, D. S. Danielssen, E. Moksness, and P. Solemdal (eds.). The Propagation of Cod, pp. 17–45. Institute of Marine Research, Arendal, Norway. 895 pp. Sponaugle, S. and 10 others. 2002. Predicting self-recruitment in marine populations: Biophysical correlates and mechanisms. Bulletin of Marine Science 70 (suppl.): 341–375. Stauffer, J. R., Kocovsky, P. M., and Rufﬂing, R. A. (eds.). 2002. Species concepts and speciation of ﬁshes. Special issue. Fish and Fisheries 3: 143–232. Stephenson, R. L. 2002. Stock structure management and structure: an ongoing challenge for ICES. ICES Marine Science Symposium 215: 305–314. Svedäng, H. and Wickström, H. 1997. Low fat contents in female silver eels: indications of insufﬁcient energetic stores for migration and gonadal development. Journal of Fish Biology 50: 475–486. Swearer, S. E. and 9 others. 2002. Evidence of self-recruitment in demersal marine populations. Bulletin of Marine Science 70 (suppl.): 251–271. Tesch, F. W. 1977. The Eel: Biology and Management of Anguillid Eels. Chapman and Hall, New York. Thorrold, S. R., Latkoczy, C., Swart, P. K., and Jones, C. M. 2001. Natal homing in a marine ﬁsh metapopulation. Science 291 (5502): 297–299. Tsukamoto, K. and Aoyama, J. 1998. Evolution of freshwater eels of the genus Anguilla: a probable scenario. Environmental Biology of Fishes 52: 139–148. Tsukamoto, K., Nakai, I., and Tesch, F. W. 1998. Do all freshwater eels migrate? Nature 396: 635– 636. Tsukamoto, K., Aoyama, J., and Miller, M. J. 2002. Migration, speciation, and the evolution of diadromy in Anguillid eels. Canadian Journal of Fisheries and Aquatic Sciences 59: 1989– 1998. Tsukamoto, K. and 11 others. 2003. Seamounts, new moon and eel spawning: the search for the spawning site of the Japanese eel. Environmental Biology of Fishes 66: 221–229. Tsukamoto, K. and Arai, T. 2001. Facultative catadromy of the eel Anguilla japonica between freshwater and seawater habitats. Marine Ecology Progress Series 220: 265–276. Tucker, D. W. 1959. A new solution to the Atlantic eel problem. Nature 4660: 495–501. Tzeng, W. N., Wang, C. H., Wickström, H., and Reizenstein, M. 2000. Occurrence of the semicatadromous European eel Anguilla anguilla in the Baltic Sea. Marine Biology 137: 93–98. Waldman, J. R. 1999. The importance of comparative studies in stock analysis. Fisheries Research 43 (1–3): 237–246 Wang, C. H. and Tzeng, W. N. 2000. The timing of metamorphosis and growth rates of American and European eel leptocephali: a mechanism of larval segregative migration. Fisheries Research 46: 191–205. Waples, R. S. 1995. Evolutionarily signiﬁcant units and the conservation of biological diversity under the Endangered Species Act. American Fisheries Society Symposium 17: 7–27.

44

D. H. Secor

Watanabe, Y. and Wada, T. 1997. Expansion and contraction of migration range of the Japanese sardine in the northwestern Paciﬁc. Memoirs of the Faculty of Fisheries, Hokkaido University 44: 49–52. Wirth, T. and Bertatchez, L. 2001. Genetic evidence against panmixia in the European eel. Nature 409: 1037–1040. Zlokovitz, E. R. and Secor, D. H. 1999. Effect of habitat use on PCB body burden in Hudson River striped bass (Morone saxatilis). Canadian Journal of Fisheries and Aquatic Sciences 56 (suppl.1): 86–93.

CHAPTER

4

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters DOUGLAS P. SWAIN,* JEFFREY A. HUTCHINGS,† AND CHRIS J. FOOTE‡ *Department of Fisheries and Oceans, Gulf Fisheries Centre, Moncton, New Brunswick, Canada, † Department of Biology, Dalhousie University, Halifax, Nova Scotia, Canada, ‡ Department of Fisheries and Aquaculture, Malaspina University-College, Nanaimo, British Columbia, Canada

I. Introduction II. Factors Underlying Intraspeciﬁc Patterns of Phenotypic and Genetic Diversity III. Meristic Characters A. Environmental Inﬂuences B. Genetic Inﬂuences C. Variation Among Populations IV. Morphometric Characters A. Environmental Inﬂuences B. Genetic Inﬂuences C. Variation Among Populations V. Life History Characters A. Environmental Inﬂuences B. Genetic Inﬂuences C. Variation Among Populations VI. Conclusions References

I. INTRODUCTION Characters used to identify ﬁsh stocks can be divided into three groups: those that are purely genetic, those that are purely environmental, and those that may reﬂect both genetic and environmental variation. Early studies characterized Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

45

46

Douglas P. Swain et al.

stocks on the basis of phenotypic variation in life-history, meristic, morphometric, and life history traits. These characters are quantitative genetic traits, typically controlled by many genes and affected by the environment in which those genes are expressed (Falconer, 1981; Hard, 1995). They also are generally related to ﬁtness and thus, molded by natural and sexual selection, they reﬂect local adaptation (e.g., Carvalho, 1993; Hard, 1995; Conover, 1998). Over the past 40 years, severa lmolecular genetic techniques have been developed to directly examine genetic variation within and between groups (e.g., protein allozymes, mitochondrial DNA, microsatellite DNA). These genetic markers usually are assumed to be neutral or nearly neutral to selection (Carvalho and Hauser, 1994; McKay and Latta, 2002), although this is not always the case (e.g., Karl and Avise, 1992; Pogson et al., 1995; Streelman et al., 1998; Merila and Crnokrak, 2001). Finally, environmental markers, such as the elemental composition of otoliths, have also been used to delineate stock structure in recent decades (e.g., Thresher, 1999; Campana et al., 2000). These markers, signatures of the habitats or areas occupied at each life history stage, are usually thought to reﬂect purely environmental differences between groups of ﬁsh, although it has been suggested that genetic effects also may contribute to differences between individuals, stocks, or species (Kalish, 1989; Thresher et al., 1994). In this chapter, we focus on a comparison between the “traditional” approaches to stock identiﬁcation using phenotypic characters, in particular meristic, morphometric, and life history traits, and the newer approaches using molecular genetic markers. The strengths and weaknesses of the different approaches depend on the working deﬁnition of the term stock. A wide range of deﬁnitions has been used, spanning a continuum from the “ﬁshery stock,” a group of ﬁsh exploited in a speciﬁc area, to the “genetic stock,” a reproductively isolated unit, genetically different from other such units (Carvalho and Hauser, 1994). Most commonly, a stock is considered equivalent to a population, at least partly reproductively isolated from other populations, and genetically different from them as a result of adaptation to its local environment (e.g., MacLean and Evans, 1981). However, the importance of delineating “phenotypic” stocks (Booke, 1981), groups of ﬁsh characterized by phenotypic differences that may be entirely environmentally induced, is being increasingly emphasized (e.g., Shepherd, 1991; Haddon and Willis, 1995; Jerry and Cairns, 1998; Lowe et al., 1998; Cadrin, 2000). For example, Cadrin and Friedland (1999) argue that intraspeciﬁc groups with persistent phenotypic differences in life history traits need to be recognized in stock assessment and ﬁsheries management, even if these differences do not reﬂect genetic differentiation. Meristic, morphometric, and life history characters are clearly appropriate for delineating phenotypic stocks. In this chapter, we focus on the advantages and disadvantages of these types of characters for delineating genetically distinct, locally adapted stocks.

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

47

II. FACTORS UNDERLYING INTRASPECIFIC PATTERNS OF PHENOTYPIC AND GENETIC DIVERSITY Genetic divergence between intraspeciﬁc groups depends on the level of gene ﬂow between the groups, genetic drift, and selection (Endler, 1986). Genetic drift refers to the random ﬂuctuation in allele frequencies that arises from the sampling of gametes in ﬁnite populations. Gene ﬂow is the change in allele frequency that results from the movement of gametes or individuals between groups. Gene ﬂow reduces or prevents differentiation, while drift promotes it. Selection can do either, depending on whether the same or different genotypes are favored in the different groups. On evolutionary time scales, neutral molecular evolution between reproductively isolated groups appears to proceed at a steady pace, with the extent of neutral genetic differentiation between groups being a measure of the time since they became isolated (Wilson et al., 1977; Clayton, 1981). However, gene ﬂow is a potent force opposing neutral genetic differentiation between groups that are not reproductively isolated. The exchange of a single migrant per generation will prevent different neutral alleles from being nearly ﬁxed in different populations (Wright, 1931), although a higher exchange rate is required to maintain the same allele frequencies between populations (Allendorf and Phelps, 1981; Adkison 1995). Given the sample sizes usually employed in stock identiﬁcation studies, Carvalho and Hauser (1994) argue that even a small number of migrants per generation is sufﬁcient to prevent detectable heterogeneity in neutral genetic markers. While neutral genetic markers provide an important tool for identifying the extent of reproductive isolation between groups and for determining phylogenetic relationships among isolated groups, they may not accurately reﬂect quantitative genetic divergence between groups in adaptive traits (Hard, 1995; Volpe and Ferguson, 1996; Conover, 1998). Genetic divergence in response to natural selection can be rapid (e.g., Reznick et al., 1997; Conover and Munch, 2002; Koskinen et al., 2002) and can occur despite regular gene ﬂow that would prevent the accumulation of neutral genetic differences (Carvalho, 1993; Allendorf, 1995). Reed and Frankham (2001) reported only a weak correlation between molecular and quantitative measures of genetic variation and concluded that molecular markers do not accurately measure differentiation between populations as a result of natural selection. Karhu et al. (1996) found that Finnish Scots pine populations showed a high degree of genetic differentiation in adaptive quantitative traits but little differentiation in molecular markers. Similarly, genetic divergence in quantitative traits tended to be much stronger than molecular genetic (microsatellite) differentiation in recently evolved grayling populations (Koskinen et al., 2002). Finally, there is strong evidence of extensive adaptive genetic divergence among four morphs of sympatric Arctic charr, Salvelinus alpinus, in developmental, life history, behavioral, and morphological traits (Skulason et al., 1989,

48

Douglas P. Swain et al.

1993, 1996, 1999; Snorrason et al., 1994), yet there is relatively little variation in neutral molecular traits (Magnusson and Ferguson, 1987; Danzmann et al., 1991; Volpe and Ferguson, 1996). In some cases, natural selection may also inﬂuence the divergence between populations in molecular markers. In cases where greater divergence is indicated by microsatellite DNA than by allozymes, it has been suggested that the allozyme loci are subject to stabilizing selection acting to maintain similar genotypic frequencies in different populations (e.g., Karl and Avise, 1992; Pogson et al., 1995). In other cases, rapid divergence in allozyme loci has been attributed to contrasting selection pressures between populations (e.g., Vuorinen et al., 1991). However, in most cases, molecular genetic markers, in particular noncoding microsatellite DNA variants, are thought to be largely neutral to selection (Carvalho and Hauser, 1994). Local adaptation and differentiation into genetically distinct populations or stocks are generally thought to be less signiﬁcant in marine ﬁshes than in freshwater or anadromous ﬁshes (e.g., Blankenship and Leber, 1995; Pawson and Jennings, 1996). This view stems from the lack of obvious physical barriers to migration in the sea and the high vagility of most marine ﬁshes, either as migratory adults or as planktonic eggs or larvae. Consistent with this view, molecular genetic studies suggest lower genetic diversity in marine ﬁshes than in anadromous and freshwater ﬁshes (Gyllensten, 1985; Ward et al., 1994). However, gene ﬂow may be limited despite the absence of physical barriers (e.g., Tallman and Healey, 1994; Wood and Foote, 1996; Skúlason et al., 1999), and genetic differentiation in adaptive traits may occur despite gene ﬂow (e.g., Wood and Foote, 1996; Foote et al., 1997). Genetic differences in phenotypic traits related to ﬁtness have been identiﬁed between populations of marine ﬁshes by the few studies that have tested for such differences (e.g., Conover and Present, 1990; Svåsand et al., 1996; Schultz et al., 1996; Puvanendran and Brown, 1998; Purchase and Brown, 2000). Studies of the Atlantic silverside, Menidia menidia, provide an example of extensive evidence for local adaptation in a marine ﬁsh (Conover, 1998). This species is widely distributed along the east coast of North America, with few physical barriers to gene ﬂow, a neustonic larval stage, and migratory adults. Genetic differentiation in life history and morphological traits has been demonstrated at ﬁne geographic scales even though molecular genetic studies provide little evidence of population subdivision from Florida to Maine (Conover, 1998). The main advantage in using life history and morphological traits in studies of population structure is that these traits are often related to ﬁtness and respond to selection, and thus may reveal genetic differentiation not evident in neutral genetic traits. Their main disadvantage results from phenotypic plasticity, the ability of a genotype to produce different phenotypes across an environmental gradient (Bradshaw, 1965; Stearns, 1989; Thompson, 1991; Schlichting and Pigliucci, 1998; Debat and David, 2001). Two types of plasticity have been

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

49

distinguished (Schmalhausen, 1949). One type involves an environmentally cued switch in developmental programs, producing one phenotype in one set of environmental conditions and a different phenotype in a second set of conditions (Levins, 1963; Smith-Gill, 1983). Usually assumed to be adaptive, examples of this type of plasticity include the predator-induced shell dimorphism of some barnacles (Lively, 1986) and the female “morphotypes” found in some rotifers (Gilbert, 1980). The second type involves an effect of the environment on the rate or degree of expression of a given developmental program, rather than a switch between programs, and has been termed phenotypic modulation (SmithGill, 1983). Plasticity in life history and morphological traits of ﬁshes appears to be mostly of this type. Although often considered nonadaptive, reﬂecting a failure to buffer development against environmental perturbations (Smith-Gill, 1983), recent research has focused on the possibility that this type of plasticity is also an adaptation to environmental heterogeneity in the phenotypes favored by selection (e.g., Thompson, 1991; Via et al., 1995; Schlichting and Pigliucci, 1998). Phenotypic plasticity can be described as a norm of reaction (Schmalhausen, 1949; Schlichting and Pigliucci, 1998), a function that expresses how the mean phenotypic value of a given genotype changes with the environment (Fig. 4-1). Reaction norms provide information on the magnitude of plasticity in a trait, the presence of genotype ¥ environment (G ¥ E) interactions in the phenotypic expression of a trait, and the extent to which the additive genetic variance or

FIGURE 4-1. Hypothetical norms of reaction of two different genotypes to an environmental variable. The reaction norms specify the mean phenotype produced at each level of the environmental variable. Variation about the mean phenotype produced by a particular genotype in a particular environment generally occurs as a result of developmental noise (Waddington, 1957). Genotype ¥ environment interaction is indicated by the intersection of the reaction norms. Common garden experiments conducted only in the vicinty of points A and B would fail to identify the difference between the two genotypes. Reprinted from Swain and Foote (1999) with permission from Elsevier.

50

Douglas P. Swain et al.

heritability changes with the environment. Crossing reaction norms, indicative of G ¥ E interaction, suggest the presence of genetic variation in plasticity. If that genetic variation is additive (i.e., heritable), then selection can produce changes to the shapes of reaction norms, resulting in population differences in responses to environmental change. Phenotypic differences between groups in the wild may reﬂect genetic differentiation, environmental differences, or a combination of the two (Thompson, 1991). Differences that are consistent with expected or known differences in selection pressure are sometimes taken as evidence for genetic differentiation (e.g., Fleming and Gross, 1989 ). However, these patterns may be entirely environmentally induced (e.g., Swain et al., 1991), reﬂecting adaptive phenotypic plasticity (e.g., Bradshaw, 1965; Meyer, 1987; Schlichting and Pigliucci, 1998; Robinson and Parsons, 2002) rather than genetic divergence. An experimental approach is needed to disentangle the genetic and environmental components of phenotypic variation (Thompson, 1991; Conover and Schultz, 1995). Two approaches have been used: “common garden” experiments and reciprocal transplants. In common garden experiments, individuals from the different areas or populations are reared in the same controlled environments (e.g., Tallman, 1986; Kinnison et al., 2001; Koskinen et al., 2002). Ideally, a series of environments, spanning the range of conditions experienced in the wild, are used to delineate reaction norms and test for genetic differences in plasticity between groups. The use of a number of levels for the controlled environmental factors also reduces the chances that a genetic difference is overlooked because experiments are conducted at the intersection of reaction norms (e.g., points A and B in Fig. 4-1). In reciprocal transplant experiments, individuals from each population are placed together in each of their natural habitats (e.g., Robinson and Wilson, 1996). This approach has the advantage that genotypes are tested in each of the environments experienced in the wild, but has the disadvantage that care must be taken to avoid the escape of transplanted genotypes. Maternal effects and other preexperimental environmental inﬂuences pose a difﬁculty for both approaches. Differences due to these effects can be avoided by using offspring of parents that have themselves been reared in common environments (Conover and Schultz, 1995), though this will not be possible for many ﬁshes. Alternatively, reciprocal hybrids (e.g., Hatﬁeld, 1997; Craig and Foote, 2001) or quantitative genetic analyses (e.g., Koskinen et al., 2002) can be used to distinguish between genetic and maternal effects. The possibility that phenotypic similarity reﬂects genetic differentiation is rarely considered in stock identiﬁcation studies. However, this situation is likely to exist when phenotypic expression depends on environmental conditions and stabilizing selection favors the same phenotype in groups developing in different environments (e.g., Bervan et al., 1979; Bervan and Gill, 1983; Conover and Present, 1990; Tallman and Healey, 1991; Billerbeck et al., 2001). In this

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

51

situation, selection will favor genetic differentiation between the groups to counteract the differing environmental inﬂuences. This geographic pattern, in which differences between areas in the genetic inﬂuences on a trait oppose the differences in environmental inﬂuences, has been termed countergradient variation (Levins, 1968; Conover and Shultz, 1995). The alternative pattern, in which genetic and environmental differences reinforce each other, has been termed cogradient variation. Numerous examples of cogradient have been identiﬁed (see below), perhaps because the phenotypic differences produced by cogradient variation have tended to attract the attention of ecologists and evolutionary biologists. Even though countergradient variation may often be overlooked because of the phenotypic similarity that it produces, a growing number of cases have been demonstrated in marine, anadromous, and freshwater ﬁshes (Conover and Present, 1990; Wood and Foote, 1990; Tallman and Healey, 1991; Present and Conover, 1992; Nicieza et al., 1994; Schultz et al., 1996; Schultz and Conover, 1997; Conover et al., 1997; Arendt and Wilson, 1999; Purchase and Brown, 2000; Lankford et al., 2001; Craig and Foote, 2001).

III. MERISTIC CHARACTERS

A. ENVIRONMENTAL INFLUENCES Meristic characters are the numbers of discrete, serially repeated, countable characters such as vertebrae, gill rakers, and ﬁn rays (e.g., Waldman, this volume). Phenotypic plasticity of meristic characters has been studied extensively (reviewed by Lindsey, 1988). The developmental environment can have a great effect on the number of parts formed in ﬁsh. However, early in ontogeny, the number of parts is ﬁxed and remains unchanged regardless of subsequent changes in the environment. Some meristic characters (such as the number of vertebrae) are ﬁxed relatively early in development, usually well before hatching, while others (such as median ﬁn rays counts) may remain labile to environmental inﬂuences even well after hatching. The number of parts may be ﬁxed well before the ﬁnal number is visible and countable (cf. Lindsey et al., 1984). Meristic characters are inﬂuenced by a wide variety of environmental factors including salinity, light, and dissolved oxygen (see review by Lindsey, 1988), but the inﬂuence of temperature has been the most widely studied (e.g., Heuts, 1949; Tåning, 1952; Lindsey, 1962; Lindsey and Harrington, 1972; Ali and Lindsey, 1974). Norms of reaction to temperature are most commonly negative (i.e., more parts produced at colder temperatures) for both vertebrae and ﬁn rays, though U-shaped responses (i.e., minimum number at an intermediate temperature) are also common for vertebrae while positive or arched responses are frequent for ﬁn rays (Lindsey, 1988).

52

Douglas P. Swain et al.

Attempts have been made to delimit the developmental period when meristic characters are labile to environmental inﬂuences by transferring embryos between temperatures at various developmental stages (e.g., Tåning, 1944; Ali and Lindsey, 1974). This work has led to the deﬁnition of “sensitive” periods before and after which meristic counts are not labile, and “supersensitive” periods when a sudden temperature change produces a “shock effect” on meristic counts (e.g., Tåning, 1952). However, subsequent work has cast doubt on the ideas of shock effects and of sensitive periods before which meristic count is unaffected by environmental inﬂuences. Instead, meristic number appears to be continuously labile to environmental inﬂuences from fertilization to the ﬁnal ﬁxation of count, with no special periods of supersensitivity (Lindsey, 1988). Meristic characters can also be affected by environmental inﬂuences acting on gametes before fertilization. The temperatures experienced by parents prior to fertilization affected the numbers of vertebrae and ﬁn rays formed in offspring in the zebraﬁsh Brachydanio rerio (Dentry and Lindsey, 1978) and in the cyprinodont ﬁsh Rivulus marmoratus (Swain and Lindsey, 1986a). Likewise, the number of parts formed in R. marmoratus differed between offspring produced soon or long after the onset of oviposition in parents (Swain and Lindsey, 1986b). These prefertilization inﬂuences need to be considered when interpreting the results of common garden experiments. Meristic differences between groups of ﬁsh reared in the same environment may not indicate genetic differences, but instead reﬂect different prefertilization environmental inﬂuences acting on genetically similar parents.

B. GENETIC INFLUENCES Despite the strong environmental inﬂuences on meristic traits, there appears to be a strong genetic component to meristic variation within populations. High heritabilities (0.4–0.9) have been reported for a variety of meristic characters, including the numbers of vertebrae, ﬁn rays, spines, and gill rakers (e.g., Hagen, 1973; Kirpichnikov, 1981; Hagen and Blouw, 1983; Tave, 1984; Leary et al., 1985; Beacham, 1990; Foote et al., 1999). Although polygenically controlled, usually by many genes (Kirpichnikov, 1981; Hatﬁeld, 1997), these characters may in some cases be inﬂuenced by relatively few loci with major effects (Hagen and Blouw, 1983; Leary et al., 1984). Hagen (1973), Leary et al. (1985), and Foote et al. (1999) found no evidence for maternal or sex-linked effects on meristic traits. A number of common environment experiments have also revealed genetic variation in phenotypic plasticity of meristic characters (e.g., Ali and Lindsey, 1974). The high heritabilities reported for meristic characters indicate a large store of additive genetic variation within populations available for response to selection. Because natural selection is expected to deplete additive genetic variation, high

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

53

heritability in a trait is sometimes assumed to indicate that the trait is not closely related to ﬁtness. However, a link with ﬁtness has been demonstrated for many meristic characters (e.g., Hagen and Gilbertson, 1973; Blouw and Hagen, 1984b; Swain, 1992b; Day and McPhail, 1996). Additive genetic variance in meristic characters appears to be maintained within populations by spatial, temporal, and ontogenetic heterogeneity in selection pressures (e.g., Reimchen, 1980; Swain, 1992b; Foote et al., 1999; Reimchen and Nosil, 2002).

C. VARIATION AMONG POPULATIONS Widespread patterns in meristic variation occur among populations of related ﬁshes. For example, vertebral number tends to be higher in populations occurring at higher latitudes or in colder waters (Jordan’s rule; Jordan, 1892) as well as in those with greater maximum body lengths (pleomerism; Lindsey, 1975). Gill raker number tends to be higher in limnetic forms than in benthic forms (e.g., Lindsey, 1981; McPhail, 1984, 1992; Snorrason et al., 1994; Hatﬁeld, 1997), with the difference associated with prey size and the efﬁciency with which prey are handled (Lavin and McPhail, 1986; Schluter, 1993; Day and McPhail, 1996). Differences between areas or populations in spine and lateral plate numbers of sticklebacks appear to be related to differences in the intensity or type of predation (e.g., Hagen and Gilbertson, 1972; Reimchen, 1980; Blouw and Hagen, 1984a). These widespread trends suggest that meristic traits may be subject to natural selection. Because of their link to ﬁtness through effects on survival under predation (e.g., Hagen and Gilbertson, 1973; Blouw and Hagen, 1984b), meristic differences between populations in traits like spine and lateral plate numbers are usually expected to have a genetic component, and this has been conﬁrmed by common environment experiments (McPhail, 1992). Gill raker number also has clear adaptive signiﬁcance through effects on feeding efﬁciency, and controlled rearing experiments have generally conﬁrmed a genetic component to differences between populations or morphs (Svärdson, 1979; McPhail, 1984, 1992; Snorrason et al., 1994; Hatﬁeld, 1997; Foote et al., 1999). In contrast to gill raker number and meristic traits related to the body armature of sticklebacks, the adaptive signiﬁcance of many meristic traits such as the numbers of vertebrae and ﬁn rays has been obscure. For example, Fowler (1970) argued that, within narrow limits, the precise vertebral count is without selective signiﬁcance, and that, within these limits, it may be more adaptive to let the number of vertebrae vary in response to environmental changes than to ﬁx the number genetically. Concordant with this view, geographic variation in meristic traits is often attributed to environmental differences rather than to genetic differentiation (e.g., Brander, 1979; Templeman, 1981; Beacham, 1984; Shepherd,

54

Douglas P. Swain et al.

1991), particularly when molecular genetic markers suggest genetic homogeneity between areas (Pepin and Carr, 1993). In contrast, Barlow (1961) argued that genetic differences may often underlie the meristic dissimilarities between populations or races of ﬁsh. He suggested that these differences may result from selection acting on correlated physiological processes, arguing that it is unlikely that “the addition or subtraction of a few elements would materially affect the probability of the survival of a ﬁsh.” However, effects of the precise vertebral phenotype on swimming performance and survival under predation have been demonstrated in larval ﬁshes and suggest that the trends observed among populations in this meristic trait reﬂect local adaptation to contrasting selection pressures acting directly on the number of vertebrae (Swain, 1988, 1992a,b). Common environment experiments have indicated a strong genetic component to geographic variation in the numbers of vertebrae and ﬁn rays in the few cases examined (references in Lindsey, 1988; Taylor and Foote, 1991; Billerbeck et al., 1997). For example, Billerbeck et al. (1997) revealed a genetic basis to latitudinal variation in vertebral number in the Atlantic silverside using common garden experiments. Maternal effects and other prefertilization inﬂuences were ruled out by using offspring of parents that had themselves been reared in a common environment. In this case, genetic and environmental inﬂuences were cogradient, since there was also a negative relationship between vertebral number and rearing temperature. However, this environmental effect was slight compared to the strong genetic differences between populations. These genetic differences occurred despite little differentiation in molecular genetic traits (Conover, 1998).

IV. MORPHOMETRIC CHARACTERS

A. ENVIRONMENTAL INFLUENCES Morphometric characters describe aspects of body shape. In contrast to meristic characters, they are continuous variables and depend on body size. Thus, a key step in measuring morphometric characters is disentangling shape from size (e.g., Rohlf and Bookstein, 1987). Morphometric characters also typically show ontogenetic changes associated with allometric growth (Gould, 1966). These ontogenetic changes in body shape may be particularly rapid at key life history stages, such as metamorphosis from larval to juvenile body forms, smoltiﬁcation in salmon, and sexual maturation. Unlike meristic characters, which are ﬁxed early in life, morphometric characters may be labile to environmental inﬂuences throughout life (e.g., Wainwright et al., 1991). Although environmental inﬂuences on morphometric characters have not been as well studied as those on meristic characters, a number of inﬂuential factors have been identiﬁed. Body shape in ﬁshes can be modiﬁed by

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

55

rearing temperature (Martin, 1949; Beacham, 1990), water velocity (Imre et al., 2002), quantity of food (Currens et al., 1989) and type of food or feeding mode (Meyer, 1987, 1990; Witte et al., 1990; Wimberger 1991, 1992; Wainwright et al., 1991; Day et al., 1994; Robinson and Wilson, 1995; Day and McPhail, 1996). Structures made of bone remodel and change shape depending on the stresses imposed on them (Lanyon, 1984; Lanyon and Rubin, 1985). These changes are usually considered to be adaptive (Lanyon and Rubin, 1985). Plasticity in trophic morphology induced by diet or feeding mode is usually assumed to result from bone remodeling in response to differences in loading regime. Other environmental inﬂuences may involve heterochrony, changes in the relative timing of developmental events (Meyer, 1987) such as switches between growth stanzas (Martin, 1949). Phenotypic plasticity in morphometric traits may often be adaptive (Robinson and Parsons, 2002). Predator-induced changes in body shape of crucian carp Carassius carassius provide a striking example of adaptive plasticity in morphology. In the presence of northern pike Esox lucius, a predator of carp, crucian carp develop a deeper body (Brönmark and Miner, 1992). This change in body form occurs in response to chemical cues released by piscivores fed a diet of ﬁsh (Brönmark and Pettersson, 1994) and reduces the vulnerability of carp to predation by pike, a gape-limited predator (Nilsson et al., 1995). Adaptive plasticity may also contribute to the morphological differences between the benthic and limnetic forms observed in a variety of ﬁsh taxa (Robinson and Wilson, 1994; Robinson and Parsons, 2002). Coexisting limnetic and benthic forms of stickleback (Gasterosteus sp.) in Enos and Paxton Lakes, British Columbia, provide an intensively studied example. The limnetic forms have more and longer gill rakers; shallower bodies and heads; longer heads, snouts, and upper jaws; and larger eyes than the benthic forms (McPhail, 1984, 1992). Foraging success is superior for the limnetic forms in open-water habitats feeding on zooplankton and for the benthic forms in benthic habitats feeding on benthic prey (Bentzen and McPhail, 1984; Schluter, 1993). Common environment experiments indicate that the differences in morphology are inherited (McPhail, 1984, 1992; Hatﬁeld, 1997), but the two forms also exhibit morphological plasticity in the adaptive direction (Day et al., 1994). Each species more closely resembled the other when raised on the latter’s diet. Diet-induced plasticity resulted in improved foraging efﬁciency (Day and McPhail, 1996) and reduced the morphological gap between species by 30% to 60% (Day et al., 1994). Similar diet-induced plasticity has been demonstrated in the cichlids Geophagus brasiliensis and G. steindachneri (Wimberger, 1991, 1992). Fish fed brine shrimp nauplii (a planktonic prey) developed longer snouts, larger eyes, longer and shallower heads, longer paired ﬁns, and shallower bodies and tails than those fed chironomid larvae (a benthic prey). Similarly, changes in feeding orientation induced morphological differences in guppies Poecilia reticulata (Robinson and Wilson, 1995). Guppies feeding on ﬂoating food developed

56

Douglas P. Swain et al.

a more fusiform shape than those feeding on food attached to a plate on the bottom. Morphological differentiation associated with trophic specialization has also been extensively studied in Arctic char Salvelinus alpinus and involves both genetic differences and phenotypic plasticity (e.g., Skúlason et al., 1989, 1993, 1996; Snorrason et al., 1994). Finally, Meyer (1987) reported extensive phenotypic plasticity in trophic morphology resulting from differences in diet and feeding mode in the cichlid Cichlasoma managuense. This plasticity paralleled the differences in morphology used to divide cichlids into functional groups. Adaptive phenotypic differences between groups of ﬁsh may thus reﬂect plasticity instead of indicating genetic differentiation between the groups.

B. GENETIC INFLUENCES Like meristic characters, morphometric traits are quantitative genetic characters, generally thought to be inﬂuenced by many genes of small individual effect, though some adaptive morphometric differences may be explained by relatively few genes of large effect (Hatﬁeld, 1997). Estimates of the heritability of morphometric characters range between low and moderate values. Riddell et al. (1981) reported heritabilities less than 0.1 for morphometric traits in Atlantic salmon. The average heritability reported for morphometric characters of chum salmon ranged between 0.3 and 0.6, depending on rearing temperature (Beacham, 1990). Lavin and McPhail (1987) reported heritabilities between 0.19 and 0.84 for morphometric characters in the threespine stickleback G. aculeatus. Grudzien and Turner (1984) reported a heritability of 0.44 for the mouth width polymorphism in Ilyodon. Genetic variation for morphological plasticity has also been demonstrated whenever it has been tested in ﬁshes (Robinson and Parsons, 2002).

C. VARIATION AMONG POPULATIONS Body shape in ﬁshes is generally thought to reﬂect adaptation to their ecological niches. Associations between body shape and ecological variables are commonly observed among populations of related ﬁshes. Correlations between body shape and the extent of stream residence in Paciﬁc salmon provide a well-studied example. Body shape and the extent of stream residence are correlated among the species of Oncorhynchus (see Scott and Crossman, 1973, p. 145), among life history types of Chinook salmon O. tshawytscha (Carl and Healey, 1984), and between ecological types (stream- vs. lake-rearing) of coho salmon O. kisutch (Swain and Holtby, 1989). Forms with longer stream residence have deeper bodies with larger, more brightly colored median ﬁns. This pattern presumably

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

57

reﬂects contrasting selection pressures between stream and open-water habitats. In streams, both as territorial juveniles and spawning adults, selection may be predominantly for burst swimming and agonistic performance (e.g., Fleming and Gross, 1989). A deep body with large median ﬁns is advantageous for burst swimming (Webb, 1978, 1984; Taylor and McPhail, 1985b) and during agonistic interactions (Holtby et al., 1993). In open waters (and during migrations to and from the sea), selection may be predominantly for sustained swimming performance, favoring a fusiform or streamlined body shape. Differences in body shape between interior and coastal populations of coho salmon are also consistent with these predictions. Interior coho, which must undertake longer migrations to and from the sea, are more streamlined, with shallower bodies and smaller median ﬁns (Taylor and McPhail, 1985a), and have superior sustained swimming performance and inferior burst swimming performance in comparisons with coastal populations (Taylor and McPhail, 1985b). Similar differences in body shape and swimming performance have been demonstrated between sockeye salmon and kokanee, the anadromous and nonanadromous forms of O. nerka (Taylor and Foote, 1991). Body shape also differs between wild and hatchery populations of coho salmon. Both as juveniles (Taylor, 1986; Swain et al., 1991) and as adults (Fleming and Gross, 1989), wild salmon have deeper bodies and larger median ﬁns than do hatchery salmon. These differences are in the direction predicted from the expected differences in selection between wild and hatchery salmon (Fleming and Gross, 1989; Swain et al., 1991). Numerous other patterns in body shape have been identiﬁed between populations of related ﬁshes. As previously noted, predictable differences in trophic morphology and general body shape occur between pelagic and benthic forms of related ﬁshes (e.g., McPhail, 1984, 1992; Robinson and Wilson, 1994). Other examples include associations between body shape and stream velocity in Atlantic salmon Salmo salar (Riddell and Leggett, 1981), spawning stream size in chum O. keta, pink O. gorbuscha, and sockeye salmon O. nerka (Beacham, 1984; Beacham et al., 1988: Hamon et al., 2000; Quinn et al., 2001), and predation intensity in Galaxias platei (Milano et al., 2002). Such differences among stocks can result in differential exploitation rates, with increased risk of stock extinction, in cases where morphological variation is related to catchability by a ﬁshery (Hamon et al., 2000). Because morphometric differences often result from changes in the timing of developmental events, they can be an indication of life history differences between groups, differences that are central to the assessment and management of exploited ﬁsh stocks (Cadrin, 2000). Differences in body shape between areas, populations, or morphs could reﬂect genetic differences, phenotypic plasticity, or a combination of both factors. A genetic component to these differences has generally been demonstrated when it has been tested for by controlled rearing in common environments (Riddell et al., 1981; Todd et al., 1981; McPhail, 1984; Taylor and McPhail, 1985a; Taylor

58

Douglas P. Swain et al.

and Foote, 1991; McPhail, 1992; Robinson and Wilson, 1996; Hatﬁeld, 1997). An environmental component has also been found in most cases examined. Genetic differences and phenotypic plasticity both contribute to trophic polymorphism in pumpkinseed sunﬁsh Lepomis gibbosus (Robinson and Wilson, 1996) and in stickleback species pairs (Day et al., 1994). In each case, environmental and genetic inﬂuences are cogradient, both operating in the adaptive direction. However, morphological differences between groups that appear to be adaptive given known or expected differences in natural selection between the groups are no guarantee that there is a genetic component to the differences. For example, differences in body shape of juvenile coho salmon between hatchery and wild populations, differences that appear to be adaptive given the expected differences in selection between the two types of salmon, are entirely environmentally induced (Swain et al., 1991). Much of the genetic variation in body shape between ﬁsh populations may be overlooked because of the tendency to focus research on the phenotypic differences observed between areas or groups of ﬁsh (Tallman and Healey, 1991; Conover and Schultz, 1995). Body shape is similar between early- and latespawning stocks of O. keta in the wild (Tallman and Healey, 1991) but differs distinctly between the stocks when reared in common laboratory environments (Tallman, 1986). Tallman and Healey (1991) suggested that stabilizing selection favored the same body shape in the two populations, resulting in genetic differences between the populations to compensate for the differences in their developmental environments. Given the demonstrated inﬂuences of the environment on body shape in ﬁshes (see above), phenotypic similarity between groups of ﬁshes developing in different environments may be more likely to reﬂect genetic differentiation than genetic homogeneity. A danger in using multivariate morphometric analyses to discriminate stocks of ﬁshes is that slight differences between groups in individual characters, possibly just sampling artifacts, can result in statistically signiﬁcant but biologically insigniﬁcant differences between groups in multivariate analyses with large sample sizes and many characters (Cadrin, 2000; e.g., Bowering, 1988; Bowering et al., 1998). For example, Bowering et al. (1998) found statistically signiﬁcant morphometric differences between all 18 samples of American plaice Hippoglossoides platessoides that they examined. Sample sizes were large, averaging 241 ﬁsh per sample. Bowering et al. (1998) concluded that a large body of additional biological evidence did not support the ﬁne-scale stock separation suggested by the morphometric differences. Morphological differences with clear adaptive signiﬁcance or ontogenetic causes that are stable over time, persisting between repeat samples taken over a number of years, provide a more meaningful basis for stock separation than differences occurring among a single collection of samples and lacking any evident ecological signiﬁcance (Cadrin, 2000).

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

59

V. LIFE HISTORY CHARACTERS

A. ENVIRONMENTAL INFLUENCES Life history traits reﬂect the ways in which individuals vary their stage- or age-speciﬁc expenditures of reproductive effort in response to intrinsic and extrinsic factors that inﬂuence survival and fecundity. As such, life histories reﬂect the expression of traits most closely related to ﬁtness, such as age and size at maturity, number and size of offspring, and longevity, and the timing of the expression of those traits throughout an individual’s life. Within an evolutionary context, natural selection is predicted to favor those individuals whose age-speciﬁc rates of survival and fecundity generate the highest rate of genotypic increase, commonly expressed as either r or R0 (Stearns, 1992; Roff, 2002). When expressed at the individual level, r and R0 represent ﬁtness. When expressed at the population or stock level, they represent rates of population growth, a parameter that reﬂects a broad range of characteristics of interest to ﬁsheries scientists, including sustainable rates of exploitation, resilience following population collapse, and probability of extinction (Hutchings, 2002). Despite this rather clear and fundamental link between individual life histories and population growth rate, life history traits have been underrepresented as means of distinguishing putative stocks. Given that the harvest rates that a stock can sustain are ultimately a function of that stock’s maximum rate of increase, it would seem logical to distinguish putative stocks on the basis of their life histories. However, the observation that the environment can affect life history traits appears to have contributed unduly to the relatively infrequent application of life history traits as metrics of stock identity. In ﬁshes, as with most indeterminately growing organisms, the inﬂuence of the environment on life history traits is realized primarily through factors that affect body size and the rate at which body size changes throughout an individual’s life. Thus, an environmental factor, such as density, food supply, or temperature, that directly inﬂuences size can be expected to have some concomitant impact on several life history traits. A key question is whether these concomitant changes reﬂect phenotypic or genetic correlations among traits; in many cases, they almost certainly represent both. Body size has a positive inﬂuence on many life history traits. Perhaps most notable among these associations is the observation that larger females produce more eggs than smaller females (Wootton, 1998; Roff, 2002). Larger individuals also tend to produce larger eggs, a correlation that holds for a number of species, such as Atlantic cod (Chambers and Waiwood, 1996), Atlantic herring, Clupea harengus (Hempel and Blaxter, 1967), capelin, Mallotus villosus (Chambers et al., 1989), and striped bass, Morone saxatilis (Zastrow et al., 1989). Increased

60

Douglas P. Swain et al.

body size is also associated with lower mortality in several ﬁshes (Hutchings, 1994; Schluter, 1995; Schultz et al., 1998; Post and Parkinson, 2001). Reﬂecting both individual size at age and the rate at which that size is attained, few parameters have greater inﬂuence on the life history traits of ﬁshes than growth rate. One of the best-documented associations between growth rate and life history is the observation that fast-growing individuals mature earlier in life than slower growing individuals (Alm, 1959; Hutchings, 1993a; Fox, 1994; Trippel et al., 1995; Godø and Haug, 1999). Independent of its effect on body size, growth rate can also potentially affect fecundity. Scott (1962), for example, reported that rainbow trout (Oncorhynchus mykiss) fed ad libitum produced more eggs per unit body mass than individuals fed a restricted diet. Regarding offspring size, there is evidence that growth rate in early life can both negatively (Jonsson et al., 1996) and positively (Morita et al., 1999) inﬂuence egg size in Atlantic salmon and white-spotted char (Salvelinus leucomaenis), respectively. Although the life history consequences of phenotypic changes to growth rate are reasonably well understood, it is not clear how growth rate affects survival independent of its inﬂuence on body size. To some extent, this depends on the scale at which the association is examined. One of the classic life history invariants in ﬁsh is that denoted by K/M. Representing the ratio of a metric of growth (from the von Bertalanffy growth equation) to the instantaneous rate of mortality, its invariance across taxa implies that species or populations characterized by rapid individual growth are also characterized by high mortality (Beverton and Holt, 1959; Charnov, 1993). At the proximate level, negative consequences to survival may be effected by physiological, metabolic, and developmental costs associated with compensatory growth, deﬁned as accelerated growth following a period of retarded growth (Metcalfe and Monaghan, 2001). For example, rapid growth during larval development is associated with delays in cranial ossiﬁcation in pumpkinseed sunﬁsh (Lepomis macrochirus), leading to reduced survival early in life (Arendt and Wilson, 2000). From an ecological perspective, faster growing individuals may place themselves at greater risk of predation if faster growth can only be achieved by riskier foraging behavior (Holtby and Healey, 1990; Werner and Anholt, 1993). Thus, environmental inﬂuences on life history traits are realized primarily through correlated responses to life history generated by environmentallyinduced variation in body size and individual growth rate. Indeed, given their heritable basis, their intimate links with ﬁtness, and their manifestation above the level of the individual as population growth rate (r), the utility of using life history characters as metrics of stock identiﬁcation merits reexamination.

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

61

B. GENETIC INFLUENCES Life history traits are quantitative traits that vary continuously and are controlled by the expression of many genes, each having a small effect (Roff, 2002). Across taxa, one analysis of 341 wild, outbred animal populations revealed that the heritability of life history traits frequently exceeds 0.30 (Mousseau and Roff, 1987). In ﬁshes, studies have consistently revealed substantive genetic variation for life history traits, particularly in salmonids (Gjedrem, 1983; Gjerde, 1984, 1986; Robison and Luempert, 1984; Roff, 2002). Moderately high heritabilities are also suggested by comparatively rapid changes in life history characters produced by intense selection pressures. In the laboratory, these selection responses are most evident in aquaculture breeding programs as a consequence of the intentional selecting and breeding of only those individuals possessing characteristics desired by the aquaculture industry (e.g., faster growth rate, delayed maturity). For example, artiﬁcial selection has produced changes in growth rate of 10% per generation in Atlantic salmon, Salmo salar (Gjoen and Bentsen, 1997) and coho salmon, Oncorhynchus kisutch (Hershberger et al., 1990). Experimentally, after subjecting Atlantic silverside (Menidia menidia) to intensive size-selective harvesting, Conover and Munch (2002) documented signiﬁcant changes in size-atage after only four generations, caused by differential selection of genotypes with slower or faster rates of growth. Genetic changes resulting from selection on one life history trait can produce genetic changes in other traits (Sinervo and Svensson, 2002). These changes, a consequence of correlated selection, result when traits are genetically correlated with one another. Genetic correlations express the extent to which two characters are determined by the same genes. Estimates of genetic correlations between life history traits in ﬁshes have been limited primarily to the aquaculture literature (e.g., Su et al., 2002). Genetic differences in life history traits can also be generated by genotypically variable thresholds that appear to determine age and size at maturity in some ﬁshes, notably those that adopt alternative reproductive strategies (Taborsky, 1994). There is good evidence, for example, that alternative mating strategies in salmonids have both a genetic and an environmental component. Comparing the incidence of jacking (early maturity) among male progeny sired by jacks and by “hooknose” (late-maturing) Chinook salmon (Oncorhynchus tshawytscha), Heath et al. (1994) estimated the heritability of jacking to be about 0.4. Breeding experiments also suggest that the incidence of parr maturity is higher among the progeny of mature male parr than it is among those of anadromous males (Thorpe et al., 1983; Gjerde, 1984; Glebe and Saunders, 1986). Whatever the genetic basis to alternative life histories, it is clear that environmental factors are also of signiﬁcant importance. In particular, there is often a signiﬁcant correlation between growth rate and/or condition and the likelihood of adopting the

62

Douglas P. Swain et al.

small-male strategy (Myers et al., 1986; Thorpe, 1986; Bohlin et al., 1990; Metcalfe, 1998). Several authors have suggested that adoption of alternative strategies in male salmonids depends on whether an individual’s growth rate in early life exceeds that speciﬁed by a growth-rate threshold, that is, that the strategies are conditional upon an individual’s state (Leonardsson and Lundberg, 1986; Thorpe, 1986; Hazel et al., 1990; Hutchings and Myers, 1994). In the quantitative genetic sense, threshold traits describe characters that are determined by alleles at multiple loci and that can be assigned to one of two or more distinct classes (Roff, 1998). The loci affecting threshold traits are assumed to each have some small effect on a trait that varies continuously. For alternative strategies in salmonids, the continuously varying trait may be the concentration of a hormone, the amount of lipid deposition, or a metabolic efﬁciency (Thorpe, 1986; Metcalfe, 1998). Genotypes expressing less than the threshold value of this underlying trait will express one phenotype, while those exceeding the threshold will express the alternative phenotype. Growth-rate thresholds can be modeled as norms of reaction for age at maturity. The existence of substantive differences in the incidence in parr maturity among families reared in a common environment (Glebe and Saunders, 1986; Herbinger, 1987) suggests that differences in reaction norms for the probability of parr maturity exist among individuals in the same population.

C. VARIATION AMONG POPULATIONS Across species, life history traits vary tremendously in ﬁshes. For example, age at maturity ranges from weeks in annual cyprinodonts (Simpson, 1979) to decades in dogﬁsh sharks (Saunders and McFarlane, 1993). Size at maturity varies from less than 1 cm in some gobies (Winterbottom and Emery, 1981) to several meters in whale sharks (Helfman et al., 1997). Fecundity ranges between 2 in some elamobranchs to millions in many broadcast-spawning marine ﬁsh. And egg size differs 100-fold, varying from 0.3 mm in the surfperch, Cymatogaster aggregata (Kamler, 1992) up to 30 mm in mouth-brooding catﬁshes (Tyler and Sumpter, 1996). Signiﬁcant differences in life history traits are not limited to among-species comparisons, and it is this within-species variability upon which stock differentiation could be undertaken (Begg et al., 1999; Begg, this volume, Chapter 6). Among populations of Atlantic salmon (Salmo salar), for example, egg size can vary threefold, age at maturity by one order of magnitude, size at maturity can differ 14-fold, and fecundity can vary almost 500-fold (Hutchings and Jones, 1998). Among marine ﬁshes, the Atlantic cod (Gadus morhua) can also exhibit wide-ranging differences in age at maturity (2–7 years; Brander, 1994; Trippel et al., 1997), length at maturity (35–85 cm; Patriquin, 1967; Morris and Green,

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

63

2002), and size-speciﬁc fecundity (Marteinsdottir and Begg, 2002; McIntyre and Hutchings, 2003). There is considerable evidence to suggest that population variation in life history reﬂects adaptation by ﬁshes to local environments (see reviews by Hindar et al., 1991; Taylor, 1991; Carvalho, 1993; Conover and Schultz, 1997). Smallscale life history variation among populations of brook trout, Salvelinus fontinalis, on Cape Race, Newfoundland, serves to provide one such example (Ferguson et al., 1991; Hutchings, 1991, 1993a,b, 1994, 1996, 1997). Brook trout inhabiting these rivers are unexploited, do not interbreed, do not differ in density, and are uninﬂuenced by interspeciﬁc competition or predation (Ferguson et al., 1991; Hutchings, 1993a). Comparing the most divergent populations, females in Freshwater River mature, on average, more than a full year earlier than those in Cripple Cove River at a ﬁvefold smaller weight. Regarding reproductive effort, relative to Cripple Cove River females, the smaller Freshwater River females allocate more than twice as much body tissue to gonads, produce signiﬁcantly more eggs, and produce 40% larger eggs (all comparisons corrected for body size; Hutchings, 1991, 1993a, 1996). These population differences in brook trout life history can be attributed to the consequences of food supply, and possibly habitat, to age-speciﬁc rates of survival and fecundity (Hutchings, 1991, 1993a,b, 1994, 1996, 1997), assertions supported by recent genetic evidence of directional selection on body size in response to sizeselective overwinter mortality and individual differences in growth (Wilson et al., 2003). Adaptive population variation in life history is the result of natural selection, acting within populations, favoring those genotypes whose age-speciﬁc rates of survival and fecundity generate the highest per capita rate of increase, or ﬁtness. One of the fundamental premises of life history theory is that natural selection acts on age-speciﬁc expectations of producing future offspring (Fisher, 1930) in response to environmental and genetic inﬂuences on age-speciﬁc survival and fecundity. Thus, if population variation in life history is adaptive, one can assume that this is a consequence of differential selection responses to environments that have different effects on survival and/or fecundity. The potential for ﬁshing to effect signiﬁcant evolutionary change within a population is no different from that of any other form of predator-induced mortality that differentially affects the survival of individuals of different ages and sizes. The question is not whether ﬁshing represents a primary selective pressure effecting genetic change in exploited ﬁsh populations—clearly it must. As Rijnsdorp (1993) put it, ﬁsheries are large-scale experiments on life history evolution. Irrespective of the causal factors, life history responses to selection are manifested primarily by population differences in (a) age and size at maturity, (b) reproductive effort, and (c) phenotypic responses to environmental change, that is, plasticity.

64

Douglas P. Swain et al.

1. Age at Maturity Age at maturity reﬂects an evolutionary compromise between the costs and beneﬁts to ﬁtness of reproducing comparatively early or late in life (Hutchings, 2002; Roff, 2002). Beneﬁts associated with early maturity include increased probability of surviving to reproduce and an increased rate of gene input into the population, resulting in reduced generation time. However, early maturity can also result in reduced fecundity and/or postreproductive survival because of the smaller body size typically associated with earlier maturity within a population. By contrast, the primary cost of delaying one’s initial spawning is the increased risk of death prior to reproduction. The primary ﬁtness advantage to delaying maturity in ﬁshes is the larger initial body size attained by individuals when they ﬁrst reproduce. Selection is predicted to favor earlier maturity with reductions in either the ratio of adult to juvenile survival (i.e., survival from birth until age at ﬁrst reproduction) or with increases in the variance in adult survival relative to the variance in juvenile survival (Cole, 1954; Gadgil and Bossert, 1970; Schaffer, 1974a,b). Intuitively these predictions make sense. As external mortality, that is, that unassociated with reproduction, at potentially reproductive ages increases, selection is expected to favor those individuals (genotypes) that reproduce prior to those ages, thus increasing their probability of contributing genes to future generations. A similar argument can be made for environmental perturbations that increase the variance in survival at potentially reproductive ages, increased variance in survival being associated with increased uncertainty in an individual’s (genotype’s) persistence. These predictions are generally presumed to hold true for ﬁshes. Leggett and Carscadden (1978) examined population differences in age at maturity among ﬁve populations of American shad, Alosa sapidissima, from Florida, United States, north to New Brunswick, Canada. They found that males and females in the northern populations, for which they presumed juvenile mortality to be more variable than that in southern populations, matured as much as 11 and 14% older, respectively, than their southernmost counterparts. Similarly, Hutchings and Jones (1998) reported a negative correlation between temporal variance in adult survival and age at maturity in anadromous Atlantic salmon. Reznick et al. (1990), documenting selection responses to predator-induced changes to mortality in guppies, Poecilia reticulata, found age at maturity among males and females to be 17 and 7% higher, respectively, in the high juvenile mortality environment relative to that in the environment characterized by comparatively low juvenile mortality. Fox and Keast (1991), comparing life histories of pumpkinseed sunﬁsh, Lepomis gibbosus, populations subjected to either high or low overwinter mortality, documented 1- to 2-yr reductions in age at maturity among males and females in the high mortality environments. And age at maturity in

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

65

bluegill sunﬁsh, Lepomis macrochirus, populations exposed to high juvenile predation is reported greater than that in populations that experience comparatively low juvenile mortality (Belk, 1995). The predominant changes to life history associated with ﬁshing are reduced age and size at maturity, the latter often being a simple consequence of the former, although increases in both characters might occur under certain circumstances (Heino, 1998; Rochet, 1998). The rapidity with which many of these changes occur within stocks is consistent with the hypothesis of a phenotypically plastic response to exploitation. In theory, reductions in density effected by ﬁshing should lead to reduced competition for resources, resulting in an increase in individual growth rate and possibly body condition. Given the widely documented negative association between individual growth rate and age at maturity in ﬁshes (e.g., Alm, 1959; Roff, 1992; Hutchings, 1993a), a comparatively rapid decline in age at maturity can be explained as a plastic response to increases in individual growth. However, while some short-term changes in age at maturity appear to be linked to increases in individual growth rate and/or condition and can potentially be explained as plastic responses to ﬁshing, others are not. One example is that of the northern stock of Atlantic cod extending from southeastern Labrador to the northern half of Newfoundland’s Grand Bank. Between the mid-1980s and the mid-1990s, female median age at maturity declined by more than 1 year, a reduction of approximately 17% (Lilly et al., 2001). However, these changes were not associated with either faster individual growth rate or improved condition (Lilly et al., 2001). Hutchings (1999) suggested that the most parsimonious explanation for these changes was an extremely rapid differential reduction of late-maturing genotypes, by severe overﬁshing, relative to that experienced by early-maturing genotypes. The high exploitation rates experienced by other Northwest Atlantic cod were also sufﬁciently high to effect genetic selection responses to life history traits, possibly favoring earlier maturity and also slower growth (Sinclair et al., 2002) as a consequence. Thus, as a life history metric of stock identiﬁcation, age at maturity would be a useful trait to consider, given its heritable basis, its responsiveness to selection, and its close correspondence with individual ﬁtness and population growth rate. 2. Reproductive Effort, Fecundity, and Egg Size Metrics of reproductive effort may also be useful as tools for distinguishing stocks, particularly if used in conjunction with other life history traits. Reproductive effort can be deﬁned as the proportion of total energy devoted to the physiological and behavioral aspects of reproduction, measured across a biologically meaningful time period (Hirshﬁeld and Tinkle, 1975) such as gonad development, movement/migration to spawning grounds, reduction of feeding prior to and

66

Douglas P. Swain et al.

concomitant with reproduction, mate competition, nest construction, or parental care. A common surrogate of reproductive effort is the gonadosomatic index, or GSI, that is, the weight of an individual’s gonads relative to that individual’s body weight. Among species, GSI ranges from as little as 0.2% in male Tilapia spp. (Helfman et al., 1997) and 0.5% in male white sticklebacks, Gasterosteus sp. (C-A. Smith and J. A. Hutchings, unpublished data) to as much as 47% in female European eels, Anguilla (Kamler, 1992). Among populations, GSI in females has been reported to differ 1.5 times in brook trout (Hutchings, 1993a) and twofold among Northwest stocks of Atlantic cod (McIntyre and Hutchings, 2003). Population differences in GSI will be reﬂected by differences in size-speciﬁc fecundity (number of eggs per unit of body mass), differences in absolute or size-speciﬁc egg size, or both. Regarding the former, there are surprisingly few comparisons among populations, although one recent study of Atlantic cod reported that size-speciﬁc fecundity can differ several-fold among stocks, depending on maternal size (Marteinsdottir and Begg, 2002). Population differences in average egg size have often been considered a proxy for adaptive variation. But, as noted elsewhere (Hutchings, 1991; Reznick and Yang, 1993), the relationship between offspring size and offspring survival must differ among environments, or among populations, for environment- or population-speciﬁc egg size optima to exist. Various hypotheses have been proposed to explain population variation in egg size and fecundity in ﬁshes. Many of these center on proposed selection responses to species- and age-speciﬁc differences in the quality of parental care (Sargent et al., 1987) and to seasonal (Rijnsdorp and Vingerhoed, 1994; Trippel, 1998), population (Kaplan and Cooper, 1984; Hutchings, 1991, 1997), and individual (Jonsson et al., 1996) differences in access to food resources. Quinn et al. (1995) suggested that among-population variation in sockeye salmon, Oncorhynchus nerka, egg size can be explained as adaptive responses to differences in the size composition of incubation gravel, arguing that the positive association between egg size and substrate size may be related to the latter’s inﬂuence on dissolved oxygen supplies relative to the surface-to-volume ratio constraints of eggs. Although somewhat more problematic to compare among populations, it has recently been shown that within-female variability in egg size can differ signiﬁcantly among populations, and that there is good reason to believe that these differences are adaptive. In a study of 10 brook trout populations, Koops et al. (2003) found that there is less egg size variability, both within and among females, when environments are more predictable, and hypothesised that females use variability in egg size to offset the cost of imperfect information about their environment when producing smaller eggs. Population differences in reproductive effort are thought to be associated with adaptive responses to age at maturity. For example, a decline in survival during

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

67

potentially reproductive ages (adult survival), relative to that during prereproductive ages (juvenile survival), is predicted to favor genotypes that increase reproductive effort, in addition to maturing earlier in life (Gadgil and Bossert, 1970; Law, 1979). Evidence of such a direct response to selection has been forthcoming from Reznick et al.’s (1990) work on Trinidad guppies. From 30 to 60 generations after a shift in predator-induced mortality from adults to juveniles, guppies responded to the presumed increase in the ratio of adult to juvenile survival by reducing reproductive allotment and by increasing age at maturity. Life history comparisons among populations have also supported these predictions concerning effort and maturity. Comparing pumpkinseed sunﬁsh from ﬁve populations that experienced either high or low levels of overwinter mortality, Fox and Keast (1991) found males and females inhabiting high-mortality environments to mature earlier and to have signiﬁcantly higher GSIs than those inhabiting low-mortality environments. Among brook trout populations in Newfoundland, declines in the ratio of adult to juvenile survival are associated with earlier maturity and increase in reproductive effort, as approximated by GSI (Hutchings, 1993a). A negative association between age at maturity and reproductive effort (GSI and size-speciﬁc fecundity) has also been reported for yellow perch, Perca ﬂavescens (Jansen, 1996). Reductions in adult survival attributable to ﬁshing should also, in theory, be predicted to increase reproductive effort. Although there has been considerably less attention directed to such changes, there is evidence that temporal changes in size-speciﬁc fecundity in the orange roughy, Hoplostethus atlanticus, may reﬂect a life history response in reproductive effort to ﬁshing. Between 1987 and 1992, when the roughy stock off east Tasmania was reduced by 50%, individual fecundity increased 20% on average (Koslow et al., 1995). Law (1979) reported a 60% increase in the fecundity of 3-year-old northern pike, Esox lucius, 12 years after an experimental harvest in Lake Windermere, U.K. 3. Phenotypic Plasticity and Norms of Reaction When life history trait optima differ among environments inhabited largely at random with respect to genotype within and among generations, selection can be expected to act on the way in which a genotype alters its life history in response to environmental change; that is, selection will act on a genotype’s norm of reaction (Schmalhausen, 1949; Via and Lande, 1985; Schlichting and Pigliucci, 1998). Such adaptive phenotypic plasticity may underlie many life history responses by ﬁsh to environmental change, notably to nongenetic variation in individual growth rate, but also to differences in temperature, habitat quality, and food supply. Experimental studies have revealed how reaction norms for juvenile growth rate (a proxy for environmental change) in brown trout (Salmo trutta) (Einum

68

Douglas P. Swain et al.

and Fleming, 1999) and juvenile survival in brook trout (Hutchings, 1991) can be inﬂuenced by egg size, and there is corresponding evidence that growth rate in early life can negatively (Jonsson et al., 1996; Morita, et al., 1999) and positively (Morita et al., 1999) inﬂuence egg size as an adult in Atlantic salmon and white-spotted char (Salvelinus leucomaenis), respectively. By modeling reaction norms as threshold traits, researchers have been able to account for the inﬂuence of both environmental and genetic inﬂuences on age and size at maturity when explaining the maintenance of conditional alternative mating strategies in Atlantic salmon (Salmo salar) (Hutchings and Myers, 1994) and coho salmon (Oncorhynchus kisutch) (Hazel et al., 1990). Comparatively few studies have examined population variation in the plasticity of life history traits, that is, population differences in the ways in which individuals alter their life history in response to environmental change. Using optimality theory, Hutchings (1996) predicted the ﬁtness consequences associated with different ages at maturity and different rates of juvenile growth (a proxy for environmental change), permitting the construction of norms of reaction for age at maturity for different populations of brook trout in southeastern Newfoundland. In one population (Freshwater River), age at maturity was invariant with growth rate (individuals being favored to mature as early in life as possible), while in the other populations age at maturity was inversely related to growth. The adaptive basis for these population differences in reaction norms underscores the point that the ﬁtness advantages of delaying maturity (increased fecundity and higher overwinter survival because of larger body size) are inevitably balanced by the probability of realizing those beneﬁts. Thus, Freshwater River females appear to be favored to mature early in life, regardless of growth rate, because the survival costs of delaying maturity are too high relative to the apparently marginal beneﬁts of increased fecundity. The existence of genetic variation in the shapes of reaction norms in some taxa (Schlichting and Pigliucci, 1998; Pigliucci, 2001) raises the possibility that plastic responses by individuals to environmental change can be modiﬁed by selection. Furthermore, if natural selection can act on reaction norms, it is possible that selection induced by anthropogenic activities, such as ﬁshing, also may be important. The adaptive signiﬁcance of interpopulation differences in plasticity, and the possibility that selection is responsible for these differences, has recently been examined for ﬁve Norwegian populations of grayling (Haugen, 2000a,b,c; Haugen and Vøllestad, 2000, 2001). Since their introduction from Lesjaskogsvatn into Hårrtjønn and Øvre Mærrabottvatn in 1910, grayling have since dispersed among several other lakes in south-central Norway, including Aursjøen and Osbumagasinet. Over a period of time ranging from 9 to 22 grayling generations, there has been considerable divergence in life history among these

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

69

populations, notably with respect to age at maturity, size at maturity, and fecundity (Haugen, 2000a,b,c). Population differences in life history have manifested themselves as differences in the shapes of reaction norms for age and size at maturity (Haugen, 2000b). Delayed age at maturity is associated with smaller size at maturity for grayling in four populations, with the population-level reaction norms crossing in state space. Interestingly, the reaction norm for grayling in Øvre Mærrabottvatn expresses an invariance in age at maturity with changes to size at maturity, all individuals maturing at age 3. Consistent with the explanation discussed previously for population differences in reaction norms for age at maturity in Newfoundland brook trout, there appears to be a direct association between the steepness of the grayling reaction norms and average adult mortality (Haugen, 2000b). For individuals aged 4 through 8 years, the instantaneous rate of mortality, Z, was highest (0.77) for Øvre Mærrabottvatn grayling, those apparently favored to reproduce as early in life as possible. By contrast, the reaction norms with the shallowest slopes, encompassing the greatest ranges of ages, are those for Hårrtjønn and Aursjøen grayling, which have the lowest rates of mortality (Z = 0.36 for both). To test the hypothesis that population differences in reaction norms are a result of selection, acting over a comparatively short period of time (9–22 generations), Haugen and Vøllestad (2000) undertook a common garden experiment in which they reared grayling from three different populations under the same experimental conditions in the laboratory. Speciﬁcally, they measured survival and speciﬁc growth rate during the ﬁrst 180 degree-days of exogenous feeding at three different temperatures. These temperatures corresponded to the average temperatures experienced by grayling in each of the three populations during this stage of life in the wild. The common garden experiments revealed signiﬁcant genetic differences in reaction norms for survival and growth rate among populations. Survival declined with temperature for the “cold” population (Aursjøen), peaked at the intermediate temperature for the “medium” population (Lesjaskogsvatn), and increased with temperature for the “warm” population (Hårrtjønn). Thus, survival was highest at the temperatures typically experienced by these grayling during this phase of their lives in the wild. There were also clear differences in plasticity for growth rates among population, although the fastest growth rates were not always achieved at the temperatures experienced by grayling in the wild (given the existence of life history trade-offs, this need not be unexpected). Thus, life history research on grayling has revealed genetic differences in reaction norms for life history traits among populations, providing data consistent with the hypothesis that the shapes of at least some reaction norms [temperature-dependent survival in early life) are adaptive and are the product of natural selection (see also the work by Hendry et al. (1998) on sockeye salmon, O. nerka].

70

Douglas P. Swain et al.

The results are also in agreement with research indicating that natural selection can take place over relatively short (10–20 generations) periods of time (Hendry and Kinnison, 1999). The question of whether ﬁshing can change the shapes of reaction norms by selection has received comparatively little attention. Reznick (1993) hypothesized that the primary effect may be to change the elevation of the reaction norms, assuming that ﬁshing would select against individuals genetically predisposed to mature at large body sizes. Considering how ﬁshing might affect the slopes of reaction norms, Hutchings (1993b, 1997, 2002) used age-speciﬁc survival and fecundity data on brook trout populations to predict how reaction norms for age, size, and effort at maturity might change in response to increases in adult mortality. As ﬁshing mortality increased, selection was predicted to favor a ﬂattening of reaction norms, notably for age and effort at maturity, such that individuals would be favored to reproduce as early in life as possible and to expend the maximum amount of effort at that age, irrespective of growth rate. Flattening of the reaction norm for age at maturity would be expected as the probability of realizing the ﬁtness beneﬁts of delayed maturity declines with increases in mortality due to ﬁshing. Similarly, as longevity declines with increased ﬁshing pressure, selection should favor increases in reproductive effort. It seems reasonable to conclude that ﬁshing can result in selective changes to reaction norms in heavily exploited populations (Hutchings, 2002). These changes may involve changes to both the slopes and the elevations of reaction norms for several life history traits. Detecting such changes, however, will be exceedingly difﬁcult, given the near absence of research on phenotypic plasticity and reaction norms on commercially exploited ﬁshes. Perhaps the primary reason that life history traits have not formed the primary basis for distinguishing stocks lies in the presumption that life history variability is predominantly environmental in origin. If so, then within the context of establishing management strategies for the purpose of providing protection for genetically distinguishable units, the use of life history variation thus becomes problematic. On the other hand, given their clear links to individual ﬁtness and population rates of increase, one could argue that it would be prudent to assume that life history variation among putative stocks has a genetic basis until demonstrated otherwise.

VI. CONCLUSIONS Molecular genetic markers and quantitative phenotypic characters both have advantages and disadvantages for delineating ﬁsh stocks. The main advantage to molecular markers is that they are direct measures of genetic differences, unaffected by any environmental differences between groups. Thus, a molecular

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

71

genetic difference between groups is an unambiguous indication of genetic differentiation. These markers are also usually assumed to be neutral to selection, and are thus useful for identifying reproductively isolated groups and for determining the phylogenetic relationships among groups. However, this selective neutrality is the main drawback of molecular markers for delineating ﬁsh stocks. Selection can result in rapid genetic divergence, much more rapid than expected as a result of mutation and genetic drift. Selection can also generate genetic differentiation between groups in the face of gene ﬂow. Thus, adaptive genetic differences between recently diverged groups or between incompletely isolated groups are not likely to be reﬂected by neutral molecular genetic differences. The main advantage to quantitative phenotypic characters such as life history and morphological traits is that these traits are generally related to ﬁtness and respond to natural selection. Thus, local adaptation, rapid adaptive divergence between recently separated groups, and genetic differences maintained by selection in the face of gene ﬂow may all be reﬂected in these traits. The main disadvantage to phenotypic characters is that they are subject to environmental as well as genetic inﬂuences. Thus, differences between groups in these characters, even differences that appear to be adaptive, may reﬂect phenotypic plasticity rather than genetic differentiation. An experimental approach, in which individuals from different areas or groups are reared in common environments, is needed to disentangle these two sources of phenotypic variation. Common environment experiments to identify the genetic and environmental components of phenotypic variation are not feasible for many ﬁsh species, particularly marine ﬁshes. In these cases, a precautionary approach would be to tentatively treat groups characterized by persistent phenotypic differences as separate stocks, recognizing that these differences may be environmentally induced rather than genetically based. This is particularly true when differences are in life history traits affecting productivity and responses to exploitation. A difﬁculty with this approach is that when countergradient variation occurs, genetic variation among groups will be cryptic, manifest by phenotypic similarity rather than by differences. When common environment experiments cannot be conducted to disentangle genetic and environmental inﬂuences and uncover hidden countergradient variation, the best recourse may be to adopt a ‘holistic’ approach, employing a broad spectrum of complementary techniques (Begg and Waldman, 1999).

REFERENCES Adkison, M. D. 1995. Population differentiation in Paciﬁc salmon: local adaptation, genetic drift, or the environment? Canadian Journal of Fisheries and Aquatic Sciences 52: 2762–2777. Ali, M. Y. and Lindsey, C. C. 1974. Heritable and temperature-induced meristic variation in the medaka, Oryzias latipes. Canadian Journal of Zoology 52: 959–976.

72

Douglas P. Swain et al.

Allendorf, F. W. 1995. Genetics: deﬁning the units of conservation. American Fisheries Society Symposium 17: 247–248. Allendorf, F. W. and Phelps, S. R. 1981. Use of allelic frequencies to describe population structure. Canadian Journal of Fisheries and Aquatic Sciences 38: 1507–1514. Alm, G. 1959. Connection between maturity, size, and age in ﬁshes. Report of the Institute for Freshwater Research, Drottningholm 40: 1–145. Arendt, J. D. and Wilson, D. S. 1999. Countergradient selection for rapid growth in pumpkinseed sunﬁsh: disentangling ecological and evolutionary effects. Ecology 80: 2793– 2798. Arendt, J. D. and Wilson, D. S. 2000. Population differences in the onset of cranial ossiﬁcation in pumpkinseed (Lepomis gibbosus), a potential cost of rapid growth. Canadian Journal of Fisheries and Aquatic Sciences 57: 351–356. Barlow, G. W. 1961. Causes and signiﬁcance of morphological variation in ﬁshes. Systematic Zoology 10: 105–117. Beacham, T. D. 1984. Age and morphology of chum salmon in southern British Columbia. Transactions of the American Fisheries Society 113: 727–736. Beacham, T. D. 1990. A genetic analysis of meristic and morphometric variation in chum salmon (Oncorhynchus keta) at three different temperatures. Canadian Journal of Zoology 68: 225–229. Beacham, T. D., Withler, R. E., Murray, C. B., and Barner, L. W. 1988. Variation in body size, morphology, egg size, and biochemical genetics of pink salmon in British Columbia. Transactions of the American Fisheries Society 117: 109–126. Begg, G. A., this volume, Chapter 6. Begg, G. A. and Waldman, J. R. 1999. An holistic approach to ﬁsh stock identiﬁcation. Fisheries Research 43: 35–44. Begg, G. A., Hare, J. A., and Sheehan, D. D. 1999. The role of life history parameters as indicators of stock structure. Fisheries Research 43: 141–163. Belk, M. C. 1995. Variation in growth and age at maturity in bluegill sunﬁsh—genetic or environmental effects. Journal of Fish Biology 47: 237–247. Bentzen, P. and McPhail, J. D. 1984. Ecology and evolution of sympatric sticklebacks (Gasterosteus): specialization for alternative trophic niches in the Enos Lake species pair. Canadian Journal of Zoology 62: 2280–2286. Bervan, K. A. and Gill, D. E. 1983. Interpreting geographic variation in life-history traits. American Zoologist 23: 85–97. Bervan, K. A., Gill, D. E., and Smith-Gill, S. J. 1979. Countergradient selection in the green frog, Rana clamitans. Evolution 33: 609–623. Beverton, R. J. H. and Holt, S. J. 1959. A review of the lifespans and mortality rates of ﬁsh in nature, and their relation to growth and other physiological characteristics. CIBA Foundation Coloquia on Ageing 54: 142–180. Billerbeck, J. M., Ortí, G., and Conover, D. O. 1997. Latitudinal variation in vertebral number has a genetic basis in the Atlantic silverside, Menidia menidia. Canadian Journal of Fisheries and Aquatic Sciences 54: 1796–1801. Billerbeck, J. M., Lankford, Jr., T. E., and Conover, D. O. 2001. Evolution of intrinsic growth and energy acquisition rates: I. Trade-offs with swimming performance in Menidia menidia. Evolution 55: 1863–1872. Blankenship, H. L. and Leber, K. M. 1995. A responsible approach to marine stock assessment. American Fisheries Society Symposium 15: 167–175. Blouw, D. M. and Hagen, D. W. 1984a. The adaptive signiﬁcance of dorsal spine variation in the fourspine stickleback, Apeltes quadracus. II. Phenotype— environment correlations. Canadian Journal of Zoology 62: 1340–1350.

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

73

Blouw, D. M. and Hagen, D. W. 1984b. The adaptive signiﬁcance of dorsal spine variation in the fourspine stickleback, Apeltes quadracus. III. Correlated traits and experimental evidence on predation. Heredity 53: 371–382. Bohlin, T., Dellefors, C., and Faremo, U. 1990. Large or small at maturity—theories on the choice of alternative male strategies in anadromous salmonids. Annales Zoologici Fennici 27: 139–147. Booke, H. E. 1981. The conundrum of the stock concept—are nature and nurture deﬁnable in ﬁsheries science? Canadian Journal of Fisheries and Aquatic Sciences 38: 1479–1480. Bowering, W. R. 1988. An analysis of morphometric characters of Greenland halibut (Reinhardtius hippoglossoides) in the Northwest Atlantic using multivariate analysis of covariance. Canadian Journal of Fisheries and Aquatic Sciences 45: 580–585. Bowering, W. R., Misra, R. K., and Brodie, W. B. 1998. Application of a newly developed statistical procedure to morphometric data from American plaice (Hippoglossoides platessoides) in the Canadian Northwest Atlantic. Fisheries Research 34: 191–203. Bradshaw, A. D. 1965. Evolutionary signiﬁcance of phenotypic plasticity in plants. Advances in Genetics 13: 115–155. Brander, K. 1979. The relationship between vertebral number and water temperature in cod. Journal du Conseil International pour l’Exploration de la Mer 38: 286–292. Brander, K. M. 1994. Patterns of distribution, spawning, and growth in North Atlantic cod: the utility of inter-regional comparisons. ICES Marine Science Symposium 198: 406–413. Brönmark, C. and Miner, J. G. 1992. Predator-induced phenotypical change in body morphology in crucian carp. Science 258: 1348–1350. Brönmark, C. and Pettersson, L. B. 1994. Chemical cues from piscivores induce a change in morphology in crucian carp. Oikos 70: 396–402. Cadrin, S. X. 2000. Advances in morphometric identiﬁcation of ﬁshery stocks. Reviews in Fish Biology and Fisheries 10: 91–112. Cadrin, S. X. and Friedland, K. D. 1999. The utility of image processing techniques for morphometric analysis and stock identiﬁcation. Fisheries Research 43: 129–139. Campana, S. E., Chouinard, G. A., Hanson, J. M., Fréchet, A., and Brattey, J. 2000. Otolith elemental ﬁngerprints as biological tracers of ﬁsh stocks. Fisheries Research 46: 343–357. Carl, L. M. and Healey, M. C. 1984. Differences in enzyme frequency and body morphology among three juvenile life history types of chinook salmon (Oncorhynchus tshawytscha) in the Nanaimo River, British Columbia. Canadian Journal of Fisheries and Aquatic Sciences 41: 1070–1077. Carvalho, G. R. 1993. Evolutionary aspects of ﬁsh distribution: genetic variability and adaptation. Journal of Fish Biology 43 (suppl. A): 53–73. Carvalho, G. R. and Hauser, L. 1994. Molecular genetics and the stock concept in ﬁsheries. Reviews in Fish Biology and Fisheries 4: 326–350. Chambers, R. C. and Waiwood, K. G. 1996. Maternal and seasonal differences in egg sizes and spawning characteristics of captive Atlantic cod, Gadus morhua. Canadian Journal of Fisheries and Aquatic Sciences 53: 1986–2003. Chambers, R. C., Leggett, W. C., and Brown, J. A. 1989. Egg size, female effects, and the correlations between early life history traits of capelin (Mallotus villosus): an appraisal at the individual level. Fishery Bulletin 87: 515–523. Charnov, E. L. 1993. Life History Invariants: Some Explorations of Symmetry in Evolutionary Ecology. Oxford University Press, Oxford, UK. Clayton, J. W. 1981. The stock concept and the uncoupling of organismal and molecular evolution. Canadian Journal of Fisheries and Aquatic Sciences 38: 1515–1522. Cole, L. C. 1954. The population consequences of life history phenomena. Quarterly Review of Biology 29: 103–137. Conover, D. O. 1998. Local adaptation in marine ﬁshes: evidence and implications for stock enhancement. Bulletin of Marine Science 62: 477–493.

74

Douglas P. Swain et al.

Conover, D. O., Brown, J. J., and Ehtisham, A. 1997. Countergradient variation in growth of young striped bass (Morone saxatilis) from different latitudes. Canadian Journal of Fisheries and Aquatic Sciences 54: 2401–2409. Conover, D. O. and Munch, S. B. 2002. Sustaining ﬁsheries yields over evolutionary time scales. Science 297: 94–96. Conover, D. O. and Present, T. M. C. 1990. Countergradient variation in growth rate: compensation for length of the growing season among Atlantic silversides from different latitudes. Oecologia 83: 316–324. Conover, D. O. and Schultz, E. T. 1995. Phenotypic similarity and the evolutionary signiﬁcance of countergradient variation. Trends in Ecology and Evolution 10: 248–252. Conover, D. O. and Schultz, E. T. 1997. Natural selection and the evolution of growth rate in the early life history: what are the trade-offs? In R. C. Chambers and E. A. Trippel (eds.), Early Life History and Recruitment in Fish Populations. Chapman and Hall, New York, pp. 305–332. Craig, J. K. and Foote, C. J. 2001. Countergradient variation and secondary sexual color: phenotypic convergence promotes genetic divergence in carotenoid use between sympatric anadromous and nonanadromous morphs of sockeye salmon (Oncorhynchus nerka). Evolution 55: 380–391. Currens, K. P., Sharpe, C. S., Hjort, R., Schreck, C. B., and Li, H. W. 1989. Effects of different feeding regimes on the morphometrics of chinook salmon (Oncorhynchus tshawytscha) and rainbow trout (O. mykiss). Copeia 1989: 689–695. Danzmann, R. G., Ferguson, M. M., Skúlason, S., Snorrason, S. S., and Noakes, D. L. G. 1991. Mitochondrial DNA diversity among four sympatric morphs of Arctic charr, Salvelinus alpinus L, from Thingvallavatn, Iceland. Journal of Fish Biology 39: 649–659. Day, T. and McPhail, J. D. 1996. The effect of behavioural and morphological plasticity on foraging efﬁciency in the threespine stickleback (Gasterosteus sp.). Oecologia 108: 380–388. Day, T., Pritchard, J., and Schluter, D. 1994. A comparison of two sticklebacks. Evolution 48: 1723–1734. Debat, V. and David, P. 2001. Mapping phenotypes: canalization, plasticity and developmental stability. Trends in Ecology and Evolution 16: 555–561. Dentry, W. and Lindsey, C. C. 1978. Vertebral variation in zebraﬁsh (Brachydanio rerio) related to the prefertilization temperature history of their parents. Canadian Journal of Zoology 56: 280–283. Einum, S. and Fleming, I. A. 1999. Maternal effects of egg size in brown trout (Salmo trutta): norms of reaction to environmental quality. Proceedings of the Royal Society of London Series B 266: 2095–2100. Endler, J. A. 1986. Natural Selection in the Wild. Princeton University Press, Princeton, NJ. Falconer, D. S. 1981. Introduction to Quantitative Genetics. 2nd ed. Longman, London. 340 pp. Ferguson, M. M., Danzman, R. G., and Hutchings, J. A. 1991. Incongruent estimates of population differentiation among brook charr Salvelinus fontinalis from Cape Race, Newfoundland, Canada, based upon allozyme and mitochondrial DNA variation. Journal of Fish Biology 35 (suppl. A): 79–85. Fisher, R. A. 1930. The Genetical Theory of Natural Selection. Oxford University Press, Oxford, UK. Fleming, I. A. and Gross, M. R. 1989. Evolution of adult female life history and morphology in a Paciﬁc salmon (coho: Oncorhynchus kisutch). Evolution 43: 141–157. Foote, C. J., Brown, G. S., and Wood, C. C. 1997. Spawning success of males using alternative mating tactics in sockeye salmon, Oncorhynchus nerka. Canadian Journal of Fisheries and Aquatic Sciences 54: 1785–1795. Foote, C. J., Moore, K., Stenberg, K., Craig, K. J., Wenburg, J. K., and Wood, C. C. 1999. Genetic differentiation in gill raker number and length in sympatric anadromous and nonanadromous morphs of sockeye salmon, Oncorhynchus nerka. Environmental Biology of Fishes 54: 263–274.

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

75

Fowler, J. A. 1970. Control of vertebral number in teleosts—an embryological problem. The Quarterly Review of Biology 45: 148–167. Fox, M. G. 1994. Growth, density, and interspeciﬁc inﬂuences on pumpkinseed sunﬁsh life-histories. Ecology 75: 1157–1171. Fox, M. G. and Keast, A. 1991. Effect of overwinter mortality on reproductive life history characteristics of pumpkinseed (Lepomis gibbosus) populations. Canadian Journal of Fisheries and Aquatic Sciences 48: 1791–1799. Gadgil, M. and Bossert, W. 1970. Life historical consequences of natural selection. American Naturalist 104: 1–24. Gilbert, J. J. 1980. Female polymorphism and sexual reproduction in the rotifer Asplanchna: evolution of their relationship and control by dietary tocopherol. American Naturalist 116: 409–431. Gjedrem, T. 1983. Genetic variation in quantitative traits and selective breeding in ﬁsh and shellﬁsh. Aquaculture 33: 51–72. Gjerde, B. 1984. Response to individual selection for age at maturity in Atlantic salmon. Aquaculture 38: 229–240. Gjerde, B. 1986. Growth and reproduction in ﬁsh and shellﬁsh. Aquaculture 57: 37–55. Gjoen, H. M. and Bentsen, H. B. 1997. Past, present, and future of genetic improvement in salmon aquaculture. ICES Journal of Marine Science 54: 1009–1014. Glebe, B. D. and Saunders, R. L. 1986. Genetic factors in sexual maturity of cultured Atlantic salmon (Salmo salar) parr and adults reared in sea cages. In D. J. Meerburg (ed.), Salmonid Age at Maturity. Canadian Special Publication of Fisheries and Aquatic Sciences 89, pp. 24–29. Godø, O. R. and Haug, T. 1999. Growth rate and sexual maturity in cod (Gadus morhua) and Atlantic halibut (Hippolglossus hippoglossus). Journal of Northwest Atlantic Fishery Science 25: 115–123. Gould, S. J. 1966. Allometry and size in ontogeny and phylogeny. Biological Review 41: 587–640. Grudzien, T. A. and Turner, B. J. 1984. Direct evidence that the Ilyodon morphs are a single biological species. Evolution 38: 402–407. Gyllensten, U. 1985. The genetic structure of ﬁsh: differences in the intraspeciﬁc distribution of biochemical genetic variation between marine, anadromous, and freshwater species. Journal of Fish Biology 26: 691–699. Haddon, M. and Willis, T. J. 1995. Morphometric and meristic comparison of orange roughy (Hoplostethus atlanticus: Trachichthyidae) from the Puysegur Bank and Lord Howe Rise, New Zealand, and its implications for stock structure. Marine Biology 123: 19–27. Hagen, D. W. 1973. Inheritance of numbers of lateral plates and gill rakers in Gasterosteus aculeatus. Heredity 30: 303–312. Hagen, D. W. and Blouw, D. M. 1983. Heritability of dorsal spines in the fourspine stickleback (Apeltes quadracus). Heredity 50: 275–281. Hagen, D. W. and Gilbertson, L. G. 1972. Geographic variation and environmental selection in Gasterosteus aculeatus L. in the Paciﬁc Northwest, America. Evolution 26: 32–51. Hagen, D. W. and Gilbertson, L. G. 1973. Selective predation and the intensity of selection acting upon the lateral plates of threespine sticklebacks. Heredity 30: 273–287. Hamon, T. R., Foote, C. J., Hilborn, R., and Rogers, D. 2000. Selection on morphology of spawning wild sockeye salmon by a gill-net ﬁshery. Transactions American Fisheries Society 129: 1300–1315. Hard, J. J. 1995. A quantitative genetic perspective on the conservation of intraspeciﬁc diversity. American Fisheries Society Symposium 17: 304–326. Hatﬁeld, T. 1997. Genetic divergence in adaptive characters between sympatric species of sticklebacks. American Naturalist 149: 1009–1029. Haugen, T. O. 2000a. Early survival and growth in populations of grayling with recent common ancestors—ﬁeld experiments. Journal of Fish Biology 56: 1173–1191.

76

Douglas P. Swain et al.

Haugen, T. O. 2000b. Growth and survival effects on maturation pattern in populations of grayling with recent common ancestors. Oikos 90: 107–118. Haugen, T. O. 2000c. Life-history evolution in grayling: evidence for adaptive phenotypic divergence during 8–28 generations. Ph.D. Thesis, University of Oslo, Oslo, Norway. 146 pp. Haugen, T. O. and Vøllestad, L. A. 2000. Population differences in early life-history traits in grayling. Journal of Evolutionary Biology 13: 897–905. Haugen, T. O. and Vøllestad, L. A. 2001. A century of life-history evolution in grayling. Genetica 112–113: 475–491. Hazel, W. N., Smock, R., and Johnson, M. D. 1990. A polygenic model for the evolution of and maintenance of conditional strategies. Proceedings of the Royal Society of London B 242: 181–187. Heath, D. D., Devlin, R. H., Heath, J. W., and Iwama, G. K. 1994. Genetic, environmental and interaction effects on the incidence of jacking in Oncorhynchus tshawytchsha (chinook salmon). Heredity 72: 146–154. Heino, M. 1998. Management of evolving ﬁsh stocks. Canadian Journal of Fisheries and Aquatic Sciences 55: 1971–1982. Helfman, G. S., Collette, B. B., and Facey, D. E. 1997. The Diversity of Fishes. Blackwell Science, Oxford, UK. Hempel, G. and Blaxter, G. H. S. 1967. Egg weight in Atlantic herring (Clupea harengus L.). Journal du Conseil, Conseil International pour l’Exploration de la Mer 31: 170–195. Hendry, A. P., Hensleigh, J. E., and Reisenbichler, R. R. 1998. Incubation temperature, developmental biology, and the divergence of sockeye salmon (Oncorhynchus nerka) within Lake Washington. Canadian Journal of Fisheries and Aquatic Sciences 55: 1387–1394. Hendry, A. P. and Kinnison, M. T. 1999. The pace of modern life: measuring rates of contemporary microevolution. Evolution 53: 1637–1653. Herbinger, C. M. 1987. A study of Atlantic salmon (Salmo salar) maturation using individually identiﬁed ﬁsh. Ph.D. Thesis, Dalhousie University, Halifax, Nova Scotia. Hershberger, W. K., Myers, J. M., Iwamoto, R. N., McCauley, W. E., and Saxton, A. M. 1990. Genetic changes in the growth of coho salmon (Oncorhynchus kisutch) in marine net-pens, produced by ten years of selection. Aquaculture 85: 187–197. Heuts, M. J. 1949. Racial divergence in ﬁn ray variation patterns in Gasterosteus aculeatus. Journal of Genetics 49: 183–191. Hindar, K., Ryman, N., and Utter, F. 1991. Genetic effects of cultured ﬁsh on natural ﬁsh populations. Canadian Journal of Fisheries and Aquatic Sciences 48: 945–957. Hirshﬁeld, M. F. and Tinkle, D. W. 1975. Natural selection and the evolution of reproductive effort. Proceedings of the National Academy of Sciences of the U.S.A 72: 2227–2231. Holtby, L. B. and Healey, M. C. 1990. Sex-speciﬁc life history tactics and risk-taking in coho salmon. Ecology 71: 678–690. Holtby, L. B., Swain, D. P., and Allan, G. M. 1993. Mirror-elicited agonistic behaviour and body morphology as predictors of dominance status in juvenile coho salmon (Oncorhynchus kisutch). Canadian Journal of Fisheries and Aquatic Sciences 50: 676–684. Hutchings, J. A. 1991. Fitness consequences of variation in egg size and food abundance in brook trout, Salvelinus fontinalis. Evolution 45: 1162–1168. Hutchings, J. A. 1993a. Adaptive life histories effected by age-speciﬁc survival and growth rate. Ecology 74: 673–684. Hutchings, J. A. 1993b. Reaction norms for reproductive traits in brook trout and their inﬂuence on life history evolution effected by size-selective harvesting. In T. K. Stokes, J. M. McGlade, and R. Law (eds.), The Exploitation of Evolving Resources. Springer-Verlag, Berlin, pp. 107–125. Hutchings, J. A. 1994. Age- and size-speciﬁc costs of reproduction within populations of brook trout, Salvelinus fontinalis. Oikos 70: 12–20.

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

77

Hutchings, J. A. 1996. Adaptive phenotypic plasticity in brook trout, Salvelinus fontinalis, life histories. Ecoscience, 3: 25–32. Hutchings, J. A. 1997. Life history responses to environmental variability in early life. In R. C. Chambers and E. A. Trippel (eds.), Early Life History and Recruitment in Fish Populations. Chapman & Hall, London, pp. 139–168. Hutchings, J. A. 1999. Inﬂuence of growth and survival costs of reproduction on Atlantic cod, Gadus morhua, population growth rate. Canadian Journal of Fisheries and Aquatic Sciences 56: 1612–1623. Hutchings, J. A. 2002. Life histories of ﬁsh. In P. J. B. Hart and J. D. Reynolds (eds.), Handbook of Fish and Fisheries. Vol. 1. Fish Biology. Blackwell, Oxford, UK, pp. 149–174. Hutchings, J. A. and Myers, R. A. 1994. The evolution of alternative mating strategies in variable environments. Evolutionary Ecology 8: 256–268. Hutchings, J. A. and Jones, M. E. B. 1998. Life history variation and growth rate thresholds for maturity in Atlantic salmon, Salmo salar. Canadian Journal of Fisheries and Aquatic Sciences 55 (suppl. 1): 22–47. Imre, I., McLaughlin, R. L., and Noakes, D. L. G. 2002. Phenotypic plasticity in brook charr: changes in caudal ﬁn induced by water ﬂow. Journal of Fish Biology 61: 1171–1181. Jansen, W. A. 1996. Plasticity in maturity and fecundity of yellow perch, Perca ﬂavescens (Mitchill): comparisons of stunted and normal-growing populations. Annales Zoologici Fennici 33: 403–415. Jerry, D. R. and Cairns, S. C. 1998. Morphological variation in the catadromous Australian bass, from seven geographically distinct riverine drainages. Journal of Fish Biology 52: 829–843. Jonsson, N., Jonsson, B., and Fleming, I. A. 1996. Does early growth cause a phenotypically plastic response in egg production of Atlantic salmon? Functional Ecology 10: 89–96. Jordan, D. S. 1892. Relations of temperature to vertebrae among ﬁshes. Proceedings of the U.S. National Museum 14: 107–120. Kalish, J. M. 1989. Otolith microchemistry: validation of the effects of physiology, age and environment on otolith composition. Journal of Experimental Marine Biology and Ecology 132: 151–178. Kamler, E. 1992. Early Life History of Fish: An Energetics Approach. Chapman & Hall, London. Kaplan, R. H. and Cooper, W. S. 1984. The evolution of developmental plasticity in reproductive characteristics: an application of the “adaptive coin ﬂipping” principle. American Naturalist 123: 393–410. Karhu, A., Hurme, P., Karjalainen, M., Karvonen, P., Kärkkäinen, K., Neale, D., and Savolainen, O. 1996. Do molecular markers reﬂect patterns of differentiation in adaptive traits of conifers? Theoretical and Applied Genetics 93: 215–221. Karl, S. A. and Avise, J. C. 1992. Balancing selection at allozyme loci in oysters: implications from nuclear RFLPs. Science 256: 100–102. Kinnison, M. T., Unwin, M. J., Hendry, A. P., and Quinn, T. P. 2001. Migratory costs and the evolution of egg size and number in introduced and indigenous salmon populations. Evolution 55: 1656–1667. Kirpichnikov, V. S. 1981. Genetic Bases of Fish Selection. Springer-Verlag, Berlin. Koops, M. A., Hutchings, J. A., and Adams, B. K. 2003. Environmental predictability and the cost of imperfect information: inﬂuences on offspring size variability. Evolutionary Ecology Research 5: 29–42. Koskinen, M. T., Haugen, T. O., and Primmer, C. R. 2002. Contemporary ﬁsherian life-history evolution in small salmonid populations. Nature 419: 826–830. Koslow, J. A., Bell, J., Virtue, P., and Smith, D. C. 1995. Fecundity and its variability in orange roughy: effects of population density, condition, egg size, and senescence. Journal of Fish Biology 47: 1063–1080.

78

Douglas P. Swain et al.

Lankford, T. E., Jr., Billerbeck, J. M., and Conover, D. O. 2001. Evolution of intrinsic growth and energy acquisition rates. II. Trade-offs with vulnerability to predation in Menidia menidia. Evolution 55: 1873–1881. Lanyon, L. E. 1984. Functional strain as a determinant for bone remodeling. Calciﬁed Tissue International 36: S56–S61. Lanyon, L. E. and Rubin, C. T. 1985. Functional adaptation in skeletal structures. In M. Hildebrand, D. M. Bramble, K. F. Liem, and D. B. Wake (eds.), Functional Vertebrate Morphology. Belknap Press, Cambridge, MA, pp. 1–25. Lavin, P. A. and McPhail, J. D. 1986. Adaptive divergence of trophic phenotype among freshwater populations of threespine stickleback (Gasterosteus aculeatus). Canadian Journal of Fisheries and Aquatic Sciences 43: 2455–2463. Lavin, P. A. and McPhail, J. D. 1987. Morphological divergence and the organization of trophic characters among lacustrine populations of the threespine stickleback (Gasterosteus aculeatus). Canadian Journal of Fisheries and Aquatic Sciences 44: 1820–1829. Law, R. 1979. Optimal life histories under age-speciﬁc predation. American Naturalist 114: 399–417. Leary, R. F., Allendorf, F. W., and Knudsen, K. L. 1984. Major morphological effects of a regulatory gene: Pgm1-t in rainbow trout. Molecular Biology and Evolution 1: 183–194. Leary, R. F., Allendorf, F. W., and Knudsen, K. L. 1985. Inheritance of meristic variation and the evolution of developmental stability in rainbow trout. Evolution 39: 308–314. Leggett, W. C. and Carscadden, J. E. 1978. Latitudinal variation in reproductive characteristics of American shad (Alosa sapidissima): evidence for population speciﬁc life history strategies in ﬁsh. Journal of the Fisheries Research Board of Canada 35: 1469–1478. Leonardsson, K. and Lundberg, P. 1986. The choice of reproductive tactics as mixed evolutionarily stable strategy: the case of male Atlantic salmon (Salmo salar L.) Reports of the Institute of Freshwater Research Drottningholm 63: 69–76. Levins, R. 1963. Theory of ﬁtness in a heterogeneous environment. II. Developmental ﬂexibility and niche selection. American Naturalist 98: 75–90. Levins, R. 1968. Evolution in Changing Environments. Princeton University Press, Princeton, NJ. Lilly, G. R., Shelton, P. A., Brattey, J., Cadigan, N. G., Healey, B. P., Murphy, E. F., and Stansbury, D. E. 2001. An assessment of the cod stock in NAFO Divisions 2J+3KL. Department of Fisheries and Oceans Canadian Stock Assessment Secretariat Research Document 2001/044. 148 pp. Lindsey, C. C. 1962. Experimental study of meristic variation in a population of threespine sticklebacks, Gasterosteus aculeatus. Canadian Journal of Zoology 40: 271–312. Lindsey, C. C. 1975. Pleomerism, the widespread tendency among related ﬁsh species for vertebral number to be correlated with maximum body length. Journal of the Fisheries Research Board of Canada 32: 2453–2469. Lindsey, C. C. 1981. Stocks are chameleons: plasticity in gill rakers of coregonid ﬁshes. Canadian Journal of Fisheries and Aquatic Sciences 38: 1497–1506. Lindsey, C. C. 1988. Factors controlling meristic variation. In W. S. Hoar and D. J. Randall (eds.), Fish Physiology. Academic Press, New York, pp. 197–274. Lindsey, C. C., Brett, A. M., Swain, D. P., and Arnason, A. N. 1984. Responses of vertebral numbers in rainbow trout to temperature changes during development. Canadian Journal of Zoology 62: 391–396. Lindsey, C. C. and Harrington, R. W., Jr. 1972. Extreme vertebral variation induced by temperature in a homozygous clone of the self-fertilizing cyprinodontid ﬁsh Rivulus marmoratus. Canadian Journal of Zoology 50: 733–744. Lively, C. M. 1986. Predator-induced shell dimorphism in the acorn barnacle Chthamalus anisopoma. Evolution 40: 232–242.

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

79

Lowe, S. A., Van Doornik, D. M., and Winans, G. A. 1998. Geographic variation in genetic and growth pattern of Atka mackerel, Pleurogrammus monopterygius (Hexagrammidae), in the Aleutian archipelago. Fishery Bulletin 96: 502–515. MacLean, J. A. and Evans, D. O. 1981. The stock concept, discreteness of ﬁsh stocks, and ﬁsheries management. Canadian Journal of Fisheries and Aquatic Sciences 38: 1889–1898. Magnusson, K. P. and Ferguson, M. M. 1987. Genetic analyses of four sympatric morphs of Arctic charr, Salvelinus alpinus, from Thingvallavatn, Iceland. Environmental Biology of Fishes 20: 67–73. Martin, W. R. 1949. The mechanics of environmental control of body form in ﬁshes. University of Toronto Studies, Biological Series, No. 58; Publications of the Ontario Fisheries Research Laboratory, No. 78. 91 pp. Marteinsdottir, G. and Begg, G. A. 2002. Essential relationships incorporating the inﬂuence of age, size and condition on variables required for estimation of reproductive potential in Atlantic cod, Gadus morhua. Marine Ecology Progress Series 235: 235–256. McIntyre, T. M. and Hutchings, J. A. 2003. Small-scale temporal and spatial variation in Atlantic cod (Gadus morhua) life history. Canadian Journal of Fisheries and Aquatic Sciences 60: 1111–1121. McKay, J. K. and Latta, R. G. 2002. Adaptive population divergence markers, QTLs and traits. Trends in Ecology and Evolution 17: 285–291. McPhail, J. D. 1984. Ecology and evolution of sympatric sticklebacks (Gasterosteus): morphological and genetic evidence for a species pair in Enos Lake, British Columbia. Canadian Journal of Zoology 62: 1402–1408. McPhail, J. D. 1992. Ecology and evolution of sympatric sticklebacks (Gasterosteus): evidence for a species-pair in Paxton Lake, Texada Island, British Columbia. Canadian Journal of Zoology 70: 361–369. Merila, J. and Crnokrak, P. 2001. Comparison of genetic differentiation at marker loci and quantitative traits. Journal of Evolutionary Biology 14: 892–903. Metcalfe, N. B. 1998. The interaction between behaviour and physiology in determining life history patterns in Atlantic salmon (Salmo salar). Canadian Journal of Fisheries and Aquatic Sciences 55 (suppl. 1): 93–103. Metcalfe, N. B. and Monaghan, P. 2001. Compensation for a bad start: grow now, pay later? Trends in Ecology and Evolution 16: 254–260. Meyer, A. 1987. Phenotypic plasticity and heterochrony in Cichlasoma managuense (Pisces, Cichlidae) and their implications for speciation in cichlid ﬁshes. Evolution 41: 1357– 1369. Meyer, A. 1990. Ecological and evolutionary consequences of trophic polymorphism in Cichlasoma citrinellum (Pisces: Cichlidae). Biological Journal of the Linnean Society 39: 279–299. Milano, D., Cussac, V. E., Macchi, P. J., Ruzzante, D. E., Alonso, M. F., Vigliano, P. H., and Denegri, M. A. 2002. Predator associated morphology in Galaxias platei in Patagonian lakes. Journal of Fish Biology 61: 138–156. Morita, K., Yamamoto, S., Takashima, Y., Matsuishi, T., Kanno, Y., and Nishimura, K. 1999. Effect of maternal growth history on egg number and size in wild white-spotted char (Salvelinus leucomaenis). Canadian Journal of Fisheries and Aquatic Sciences 56: 1585–1589. Morris, C. J. and Green, J. M. 2002. Biological characteristics of a resident population of Atlantic cod (Gadus morhua L.) in southern Labrador. ICES Journal of Marine Science 59: 666–678. Mousseau, T. A. and Roff, D. A. 1987. Natural selection and the heritability of ﬁtness components. Heredity 59: 181–198. Myers, R. A., Hutchings, J. A., and Gibson, R. J. 1986. Variation in male parr maturation within and among populations of Atlantic salmon, Salmo salar. Canadian Journal of Fisheries and Aquatic Sciences 43: 1242–1248.

80

Douglas P. Swain et al.

Nicieza, A. G., Reiriz, L., and Braña, F. 1994. Variation in digestive performance between geographically disjunct populations of Atlantic salmon: countergradient in passage time and digestive rate. Oecologia 99: 243–251. Nilsson, P. A., Brönmark, C., and Pettersson, L. B. 1995. Beneﬁts of a predator-induced morphology in crucian carp. Oecologia 104: 291–296. Patriquin, D. G. 1967. Biology of Gadus morhua in Ogac Lake, a landlocked ﬁord on Bafﬁn Island. Journal of the Fisheries Research Board of Canada 24: 2573–2594. Pawson, M. G. and Jennings, S. 1996. A critique of methods for stock identiﬁcation in marine capture ﬁsheries. Fisheries Research 25: 203–217. Pepin, P. and Carr, S. M. 1993. Morphological, meristic and genetic analysis of stock structure in juvenile Atlantic cod (Gadus morhua) from the Newfoundland shelf. Canadian Journal of Fisheries and Aquatic Sciences 50: 1924–1933. Pigliucci, M. 2001. Phenotypic Plasticity: Beyond Nature and Nurture. Johns Hopkins University Press, Baltimore, MD. Pogson, G. H., Mesa, K. A., and Boutilier, R. G. 1995. Genetic population structure and gene ﬂow in the Atlantic cod Gadus morhua: a comparison of allozyme and nuclear RFLP loci. Genetics 139: 375–385. Post, J. R. and Parkinson, E. A. 2001. Energy allocation strategy in young ﬁsh: allometry and survival. Ecology 82: 1040–1051. Present, T. M. C. and Conover, D. O. 1992. Physiological basis of latitudinal growth differences in Menidia menidia: variation in consumption or efﬁciency? Functional Ecology 6: 23– 31. Purchase, C. F. and Brown, J. A. 2000. Interpopulation differences in growth rates and food conversion efﬁciencies of young Grand Banks and Gulf of Maine Atlantic cod (Gadus morhua). Canadian Journal of Fisheries and Aquatic Sciences 57: 2223–2229. Puvanendran, V. and Brown, J. A. 1998. Effect of light intensity on the foraging and growth of Atlantic cod larvae: interpopulation difference? Marine Ecology Progress Series 167: 207–214. Quinn, T. P., Hendry, A. P., and Wetzel, L. A. 1995. The inﬂuence of life history trade-offs and the size of incubation gravels on egg size variation in sockeye salmon (Oncorhynchus nerka). Oikos 74: 425–438. Quinn, T. P., Wetzel, L., Bishop, S., Overberg, K., and Rogers, D. E. 2001. Inﬂuence of breeding habitat on bear predation and age at maturity and sexual dimorphism of sockeye salmon populations. Canadian Journal of Zoology 79: 1782–1793. Reed, D. H. and Frankham, R. 2001. How closely correlated are molecular and quantitative measures of genetic variation? A meta-analysis. Evolution 55: 1095–1103. Reimchen, T. E. 1980. Spine deﬁciency and polymorphism in a population of Gasterosteus aculeatus: an adaptation to predators? Canadian Journal of Zoology 58: 1232–1244. Reimchen, T. E. and Nosil, P. 2002. Temporal variation in divergent selection on spine number in threespine stickleback. Evolution 56: 2472–2483. Reznick, D. N. 1993. Norms of reaction in ﬁshes. In T. K. Stokes, J. M. McGlade, and R. Law (eds.), The Exploitation of Evolving Resources. Springer-Verlag, Berlin, pp. 72–90. Reznick, D. N., Bryga, H., and Endler, J. A. 1990. Experimentally induced life history evolution in a natural population. Nature 346: 357–359. Reznick, D. N., Shaw, F. H., Rodd, F. H., and Shaw, R. G. 1997. Evaluation of the rate of evolution in natural populations of guppies (Poecilia reticulata). Science 275: 1934–1937. Reznick, D. N. and Yang, T. 1993. The inﬂuence of ﬂuctuating resources on life history: patterns of allocation and plasticity in female guppies. Ecology 74: 1011–1029. Riddell, B. E. and Leggett, W. C. 1981. Evidence of an adaptive basis for geographic variation in body morphology and time of downstream migration of juvenile Atlantic salmon (Salmo salar). Canadian Journal of Fisheries and Aquatic Sciences 38: 308–320.

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

81

Riddell, B. E., Leggett, W. C., and Saunders, R. L. 1981. Evidence of adaptive polygenic variation between two populations of Atlantic salmon (Salmo salar) native to tributaries of the S. W. Miramichi River, N.B. Canadian Journal of Fisheries and Aquatic Sciences 38: 321–333. Rijnsdorp, A. D. 1993. Fisheries as a large-scale experiment on life-history evolution: disentangling phenotypic and genetic effects in changes in maturation and reproduction of North Sea plaice, Pleuronectes platessa L. Oecologia 96: 391–401. Rijnsdorp, A. D. and Vingerhoed, B. 1994. The ecological signiﬁcance of geographical and seasonal differences in egg size in sole, Solea solea (L). Netherlands Journal of Sea Research 32: 255–270. Robinson, B. W. and Parsons, K. J. 2002. Changing times, spaces, and faces: tests and implications of adaptive morphological plasticity in the ﬁshes of northern postglacial lakes. Canadian Journal of Fisheries and Aquatic Sciences 59: 1819–1833. Robinson, B. W. and Wilson, D. S. 1994. Character release and displacement in ﬁshes: a neglected literature. American Naturalist 144: 596–627. Robinson, B. W. and Wilson, D. S. 1995. Experimentally induced morphological diversity in Trinidadian guppies (Poecilia reticulata). Copeia 1995: 294–305. Robinson, B. W. and Wilson, D. S. 1996. Genetic variation and phenotypic plasticity in a trophically polymorphic population of pumpkinseed sunﬁsh (Lepomis gibbosus). Evolutionary Ecology 10: 631–652. Robison, O. W. and Luempert, L. G. 1984. Genetic variation in weight and survival of brook trout (Salvelinus fontinalis). Aquaculture 38: 155–170. Rochet, M.-J. 1998. Short-term effects of ﬁshing on life history traits of ﬁshes. ICES Journal of Marine Science 55: 371–391. Roff, D. A. 1998. The maintenance of phenotypic and genetic variation in threshold traits by frequency dependent selection. Journal of Evolutionary Biology 11: 513–529. Roff, D. A. 2002. Life History Evolution. Sinauer, Sunderland, MA. Rohlf, F. J. and Bookstein, F. L. 1987. A comment on shearing as a method for “size correction.” Systematic Zoology 36: 356–367. Sargent, R. C., Taylor, P. D., and Gross, M. R. 1987. Parental care and the evolution of egg size in ﬁshes. American Naturalist 129: 32–46. Saunders, M. W. and McFarlane, G. A. 1993. Age and length at maturity of the female spiny dogﬁsh, Squalus acanthias, in the Strait of Georgia, British Columbia, Canada. Environmental Biology of Fishes 38: 49–58. Schaffer, W. M. 1974a. Selection for optimal life histories: the effects of age structure. Ecology 55: 291–303. Schaffer, W. M. 1974b. Optimal reproductive effort in ﬂuctuating environments. American Naturalist 108: 783–790. Schlichting, C. D. and Pigliucci, M. 1998. Phenotypic Evolution: A Reaction Norm Perspective. Sinauer Associates, Sunderland MA. 387 pp. Schluter, D. 1993. Adaptive radiation in sticklebacks: size, shape and habitat use efﬁciency. Ecology 74: 699–709. Schluter, D. 1995. Adaptive radiation in sticklebacks: trade-offs in feeding performance and growth. Ecology 76: 82–90. Schmalhausen, I. I. 1949. Factors of Evolution: The Theory of Stabilizing Selection. Blakiston, Philadelphia, PA. Schultz, E. T. and Conover, D. O. 1997. Latitudinal differences in somatic energy storage: adaptive responses to seasonality in an estuarine ﬁsh (Atherinidae: Menidia menidia). Oecologia 109: 516–529. Schultz, E. T., Conover, D. O., and Ehtisham, A. 1998. The dead of winter: size-dependent variation and genetic differences in seasonal mortality among Atlantic silverside (Atherinidae: Menidia

82

Douglas P. Swain et al.

menidia) from different latitudes. Canadian Journal of Fisheries and Aquatic Sciences 55: 1149–1157. Schultz, E. T., Reynolds, K. E., and Conover, D. O. 1996. Countergradient variation in growth among newly hatched Fundulus heteroclitus: geographic differences revealed by common-environment experiments. Functional Ecology 10: 366–374. Scott, D. P. 1962. Effect of food quantity on fecundity of rainbow trout, Salmo gairdneri. Journal of the Fisheries Research Board of Canada 19: 715–731. Scott, W. B. and Crossman, E. J. 1973. Freshwater ﬁshes of Canada. Fisheries Research Board of Canada Bulletin 184. Shepherd, G. 1991. Meristic and morphometric variation in black sea bass north of Cape Hatteras, North Carolina. North American Journal of Fisheries Management 11: 139–148. Simpson, B. R. C. 1979. The phenology of annual killiﬁshes. Symposium of the Zoological Society of London 44: 243–261. Sinclair, A. F., Swain, D. P., and Hanson, J. M. 2002. Measuring changes in the direction and magnitude of size-selective mortality in a commercial ﬁsh population. Canadian Journal of Fisheries and Aquatic Sciences 59: 361–371. Sinervo, B. and Svensson, E. 2002. Correlational selection and the evolution of genomic architecture. Heredity 89: 329–338. Skúlason, S., Noakes, D. L. G., and Snorrason, S. S. 1989. Ontogeny of trophic morphology in four sympatric morphs of arctic charr Salvelinus alpinus in Thingvallavatn, Iceland. Biological Journal of the Linnean Society 38: 281–301. Skúlason, S., Snorrason, S. S., and Jónsson, B. 1999. Sympatric morphs, populations and speciations in freshwater ﬁsh with emphasis on Artic charr. In A. E. Magurran and R. M. Roberts (eds.), Evolution of Biological Diversity. Oxford University Press, Oxford, UK, pp. 71–92. Skúlason, S., Snorrason, S. S., Noakes, D. L. G., and Ferguson, M. M. 1996. Genetic basis of life history variations among sympatric morphs of arctic char, Salvelinus alpinus. Canadian Journal of Fisheries and Aquatic Sciences 53: 1807–1813. Skúlason, S., Snorrason, S. S., Ota, D., and Noakes, D. L. G. 1993. Genetically based differences in foraging behaviour among sympatric morphs of arctic charr (Pisces: Salmonidae). Animal Behaviour 45: 1179–1192. Smith-Gill, S. J. 1983. Developmental plasticity: developmental conversion versus phenotypic modulation. American Zoologist 23: 47–55. Snorrason, S. S., Skúlason, S., Jonsson, B., Malmquist, H. J., Jónasson, P. M., Sandlund, O. T., and Lindem, T. 1994. Trophic specialization in arctic charr Salvelinus alpinus (Pisces: Salmonidae): morphological divergence and ontogenetic niche shifts. Biological Journal of the Linnean Society 52: 1–18. Stearns, S. C. 1989. The evolutionary signiﬁcance of phenotypic plasticity. BioScience 39: 436–445. Stearns, S. C. 1992. The Evolution of Life Histories. Oxford University Press, Oxford, UK. Streelman, J. T., Zardoya, R., Meyer, A., and Karl, S. A. 1998. Multilocus phylogeny of cichlid ﬁshes (Pisces: Perciformes): evolutionary comparison of microsatellite and single-copy nuclear loci. Molecular Biology and Evolution, 15: 798–808. Su, G. S., Liljedahl, L. E., and Gall, G. A. E. 2002. Genetic correlations between body weight at different ages and with reproductive traits in rainbow trout. Aquaculture 213: 85–94. Svärdson, G. 1979. Speciation of Scandnavian Coregonus. Reports of Institute of Freshwater Research Drottingholm 57: 1–95. Svåsand, T., Jørstad, K. E., Otterå, H., and Kjesbu, O. S. 1996. Differences in growth performance between Arcto-Norwegian and Norwegian coastal cod reared under identical conditions. Journal of Fish Biology 49: 108–119. Swain, D. P. 1988. Evidence of selection for vertebral number of fry in peamouth, Mylocheilus caurinus. Canadian Journal of Fisheries and Aquatic Sciences, 45: 1279–1290.

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

83

Swain, D. P. 1992a. The functional basis of natural selection for vertebral traits of larvae in the stickleback Gasterosteus aculeatus. Evolution 46: 987–997. Swain, D. P. 1992b. Selective predation for vertebral phenotype in Gasterosteus aculeatus: reversal in the direction of selection at different larval sizes. Evolution 46: 998–1013. Swain, D. P. and Foote, C. J. 1999. Stocks and chameleons: the use of phenotypic variation in stock identiﬁcation. Fisheries Research 43: 113–128. Swain, D. P. and Holtby, L. B. 1989. Differences in morphology and behavior between juvenile coho salmon (Oncorhynchus kisutch) rearing in a lake or in its tributary stream. Canadian Journal of Fisheries and Aquatic Sciences 46: 1406–1414. Swain, D. P. and Lindsey, C. C. 1986a. Inﬂuence of reproductive history of parents on meristic variation in offspring in the cyprinodont ﬁsh Rivulus marmoratus. Canadian Journal of Zoology 64: 1456–1459. Swain, D. P. and Lindsey, C. C. 1986b. Meristic variation in a clone of the cyprinodont ﬁsh Rivulus marmoratus related to temperature history of the parents and of the embryos. Canadian Journal of Zoology 64: 1444–1455. Swain, D. P., Riddell, B. E., and Murray, C. B. 1991. Morphological differences between hatchery and wild populations of coho salmon (Oncorhynchus kisutch): environmental versus genetic origin. Canadian Journal of Fisheries and Aquatic Sciences 48: 1783–1791. Taborsky, M. 1994. Sneakers, satellites, and helpers: parasitic and cooperative behavior in ﬁsh reproduction. Advances in the Study of Behavior 23: 1–100. Tallman, R. F. 1986. Genetic differentiation among seasonally distinct spawning populations of chum salmon, Oncorhynchus keta. Aquaculture 57: 211–217. Tallman, R. F. and Healey, M. C. 1991. Phenotypic differentiation in seasonal ecotypes of chum salmon, Oncorhynchus keta. Canadian Journal of Fisheries and Aquatic Sciences 48: 661–671. Tallman, R. F. and Healey, M. C. 1994. Homing, straying and gene ﬂow among seasonally separated populations of chum salmon (Oncorhynchus keta). Canadian Journal of Fisheries and Aquatic Science 51: 577–588. Tåning, A. 1944. Experiments on meristic and other characters in ﬁshes. 1. On the inﬂuence of temperature on some meristic characters in sea-trout and the ﬁxation period of these characters. Meddelelser fra Danmarks Fiskeri og Havundersøgelser. Serie Fiskerier 11: 1–66. Tåning, A. V. 1952. Experimental study of meristic characters in ﬁshes. Biological Reviews of the Cambridge Philosophical Society 27: 169–193. Tave, D. 1984. Quantitative genetics of vertebrae number and position of dorsal ﬁn spines in the velvet belly shark, Etmopterus spinax. Copeia 1984: 794–797. Taylor, E. B. 1986. Differences in morphology between wild and hatchery populations of juvenile coho salmon. Progressive Fish-Culturalist 48: 171–176. Taylor, E. B. 1991. A review of local adaptation in salmonidae, with particular reference to Paciﬁc and Atlantic salmon. Aquaculture 98: 185–208. Taylor, E. B. and Foote, C. J. 1991. Critical swimming velocities of juvenile sockeye salmon and kokanee, the anadromous and non-anadromous forms of Oncorhynchus nerka (Walbaum). Journal of Fish Biology 38: 407–419. Taylor, E. B. and McPhail, J. D. 1985a. Variation in body morphology among British Columbia populations of coho salmon, Oncorhynchus kisutch. Canadian Journal of Fisheries and Aquatic Sciences 42: 2020–2028. Taylor, E. B. and McPhail, J. D. 1985b. Variation in burst and prolonged swimming performance among British Columbia populations of coho salmon, Oncorhynchus kisutch. Canadian Journal of Fisheries and Aquatic Sciences 42: 2029–2033. Templeman, W. T. 1981. Vertebral numbers in Atlantic cod, Gadus morhua, of the Newfoundland and adjacent areas, 1947–1971, and their use in delineating cod stocks. Journal of Northwest Atlantic Fisheries Science 2: 21–45.

84

Douglas P. Swain et al.

Thompson, J. D. 1991. Phenotypic plasticity as a component of evolutionary change. Trends in Ecology and Evolution 6: 246–249. Thorpe, J. E. 1986. Age at ﬁrst maturity in Atlantic salmon, Salmo salar: freshwater period inﬂuences and conﬂicts with smolting. In D. J. Meerburg (ed.), Salmonid Age at Maturity. Canadian Special Publication of Fisheries and Aquatic Sciences 89, pp. 7–14. Thorpe, J. E., Morgan, R. I. G., Talbot, C., and Miles, M. S. 1983. Inheritance of developmental rates in Atlantic salmon, Salmo salar L. Aquaculture 33: 119–128. Thresher, R. E. 1999. Elemental composition of otoliths as a stock delineator in ﬁshes. Fisheries Research 43: 165–204. Thresher, R. E., Proctor, C. H., Gunn, J. S., and Harrowﬁeld, I. R. 1994. An evaluation of electronprobe microanalysis of otoliths for stock delineation and identiﬁcation of nursery areas in a southern temperate groundﬁsh, Nemadactylus macropterus (Cheilodactylidae). Fishery Bulletin 92: 817–840. Todd, T. N., Smith, G. R., and Cable, L. E. 1981. Environmental and genetic contributions to morphological differentiation in ciscoes (Coregoninae) of the Great Lakes. Canadian Journal of Fisheries and Aquatic Sciences 38: 59–67. Trippel, E. A. 1998. Egg size and viability and seasonal offspring production of young Atlantic cod. Transactions of the American Fisheries Society 127: 339–359. Trippel, E. A., Buzeta, M. L., and Hunt, J. J. 1995. Evaluation of the cost of reproduction in Georges Bank Atlantic cod (Gadus morhua): utility of otolith back-calculation. In D. H. Secor, J. M. Dean, and S. E. Campana, (eds.), Recent Developments in Fish Otolith Research. University of South Carolina Press, Columbia, SC, pp. 599–610. Trippel, E. A., Morgan, M. J., Fréchet, A., Rollet, C., Sinclair, A., Annand, C., Beanlands, D., and Brown, L. 1997. Changes in age and length at sexual maturity of northwest Atlantic cod, haddock and pollock stocks, 1972–1995. Canadian Technical Report of Fisheries and Aquatic Sciences No. 2157. Tyler, C. R. and Sumpter, J. P. 1996. Oocyte growth and development in teleosts. Reviews in Fish Biology and Fisheries 6: 287–318. Via, S. and Lande, R. 1985. Genotype-environment interaction and the evolution of phenotypic plasticity. Evolution 39: 505–522. Via, S., Gomulkiewicz, R., De Jong, G., Scheiner, S. M., Schlichting, C. D., and Van Tienderen, P. H. 1995. Adaptive phenotypic plasticity: consensus and controversy. Trends in Ecology and Evolution 10: 212–217. Volpe, J. P. and Ferguson, M. M. 1996. Molecular genetic examination of the polymorphic Arctic charr Salvelinus alpinus of Thingvallavatn, Iceland. Molecular Ecology 5: 763–772. Vuorinen, J., Næsje, T. F., and Sandlund, O. T. 1991. Genetic changes in a vendace Coregonus albula (L.) population, 92 years after introduction. Journal of Fish Biology 39 (suppl. A): 193–201. Waddington, C. H. 1957. The Strategy of the Genes. Allen and Unwin, London, UK, 262 pp. Waldman, J. R. this volume, Chapter 10. Wainwright, P. C., Osenberg, C. W., and Mittelbach, G. G. 1991. Trophic polymorphism in the pumpkinseed sunﬁsh (Lepomis gibbosus Linnaeus): effects of environment on ontogeny. Functional Ecology 5: 40–55. Ward, R. D., Woodwark, M., and Skibinski, D. O. F. 1994. A comparison of genetic diversity levels in marine, freshwater, and anadromous ﬁshes. Journal of Fish Biology 44: 213–232. Webb, P. W. 1978. Fast-start performance and body form in seven species of teleost ﬁsh. Journal of Experimental Biology 74: 211–226. Webb, P. W. 1984. Form and function in ﬁsh swimming. Scientiﬁc American 251: 72–82. Werner, E. E. and Anholt, B. R. 1993. Ecological consequences of the trade-off between growth and mortality rates mediated by foraging activity. American Naturalist 142: 242–272.

Environmental and Genetic Inﬂuences on Stock Identiﬁcation Characters

85

Wilson, A. C., Carlson, S. S., and White, T. J. 1977. Biochemical evolution. Annual Review of Biochemistry 46: 573–639. Wilson, A. J., Hutchings, J. A., and Ferguson, M. M. 2003. Selective and genetic constraints on the evolution of body size in a stream-dwelling ﬁsh. Journal of Evolutionary Biology 16: 584–594. Wimberger, P. H. 1991. Plasticity of jaw and skull morphology in the neotropical cichlids Geophagus brasiliensis and G. steindachneri. Evolution 45: 1545–1563. Wimberger, P. H. 1992. Plasticity of ﬁsh body shape. The effects of diet, development, family and age in two species of Geophagus (Pisces: Cichlidae). Biological Journal of the Linnean Society 45: 197–218. Winterbottom, R. and Emery, A. R. 1981. A new genus and two new species of gobiid ﬁshes (Perciformes) from the Chagos Archipelago, central Indian Ocean. Environmental Biology of Fishes 6: 139–149. Witte, F., Barel, C. D. N., and Hoogerhoud, R. J. C. 1990. Phenotypic plasticity of anatomical structures and its ecomorphological signiﬁcance. Netherlands Journal of Zoology 40: 278–298. Wood, C. C. and Foote, C. J. 1990. Genetic differences in the early development and growth of sympatric sockeye salmon and kokanee (Oncorhynchus nerka), and their hybrids. Canadian Journal of Fisheries and Aquatic Sciences 47: 2250–2260. Wood, C. C. and Foote, C. J. 1996. Evidence for sympatric genetic divergence of anadromous and nonanadromous morphs of sockeye salmon (Oncorhynchus nerka). Evolution 50: 1265–1279. Wootton, R. J. 1998. Ecology of Teleost Fishes. Chapman & Hall, London, UK. Wright, S. 1931. Evolution in Mendelian populations. Genetics 16: 97–159. Zastrow, C. E., Houde, E. D., and Saunders, E. H. 1989. Quality of striped bass (Morone saxatilis) eggs in relation to river source and female weight. Rapports et Procès-Verbaux des Réunions, Conseil International pour l’Exploration de la Mer 191: 34–42.

This Page Intentionally Left Blank

PART

Life History Traits

II

This Page Intentionally Left Blank

CHAPTER

5

The Use of Early Life Stages in Stock Identiﬁcation Studies JONATHAN A. HARE NOAA National Ocean Service, Center for Coastal Fisheries and Habitat Research, Beaufort, North Carolina, USA

I. The Stock Concept, Again II. Role of ELS Information in the Stock Concept A. ELSs and Discreteness of Spawning B. ELSs and Genetic Population Structure C. ELSs and Phenotypic Traits III. Examples of ELS Information in the Deﬁnition of Stocks A. Atlantic Herring in the Northwest Atlantic Ocean B. Japanese Eel in the Western Paciﬁc Ocean C. American Lobster in the Northwest Atlantic Ocean IV. Conclusions and Future Directions A. Increased Use of Geostatistics B. An Ecosystem Framework for the Inclusion of ELSs in Stock Identiﬁcation C. Modeling and Measuring Actual Planktonic Transport D. Spatial Population Models and the Stock Concept Acknowledgments References

I. THE STOCK CONCEPT, AGAIN A stock is a group of individuals for which population parameters can be meaningfully estimated for speciﬁc management applications, typically ﬁshery stock assessments (Brown et al., 1987; Carvalho and Nigmatullin, 1998; Begg and Waldman, 1999). The purpose of a ﬁshery stock assessment is to use various statistical and mathematical calculations to make quantitative predictions about Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

89

90

Jonathan A. Hare

the status of exploited ﬁshes and their response to alternative management options (Hilborn and Walters, 1992). The parameters used in ﬁshery stock assessments include, but are not limited to, fecundity, reproductive rate, recruitment, immigration/emigration rate, age structure, size structure, growth rate, and natural mortality rate, as well as ﬁshing mortality rate, catchability, and accessibility. As such, stock has no strict biological deﬁnition (Brown et al., 1987), and the concept of stock lies at the intersection of biological organization and human activities (Secor, 1999). A main purpose of stock deﬁnition is to design management strategies that prevent the depletion of “weaker” stocks that might be harvested at an optimal rate appropriate for “stronger” stocks (Begg et al., 1999a). Misleading management strategies can originate from stock assessment models if several stocks or only a portion of a stock are the components actually modeled. Thus, stocks must be identiﬁed and then managed as discrete units (Kutkuhn, 1981; Brown et al., 1987; Begg et al., 1999a). There are two general methods by which stocks are deﬁned: by genotype and by phenotype (see Booke, 1981). A “genotypic” stock consists of randomly interbreeding members of a species whose genetic integrity persists whether they remain spatially and temporally isolated as a group or segregate for breeding and otherwise mix freely with other “genotypic” stocks (Kutkuhn, 1981). Thus, a “genotypic” stock is synonymous with an ecological population, which is deﬁned as “a group of conspeciﬁc organisms that occupy a more or less well-deﬁned geographic region and exhibit reproductive continuity from generation to generation” (Futuyma, 1986). In contrast, “phenotypic” stocks are deﬁned as intraspeciﬁc groups that differ in the expression of certain characters owing to environmental or genetic effects. The use of phenotypic characters in stock identiﬁcation diminished with the improvement in genetic techniques (Kutkuhn, 1981; Ihssen et al., 1981). However, the identiﬁcation of “phenotypic” stocks still has value, particularly when the characters used to distinguish stocks are the same parameters used in ﬁsheries stock assessments (e.g., growth, mortality, and maturity). The combined approach of multiple characters examining differing aspects of the spatial relationship of groups of individuals through time has led to the increasing use of a “holistic” approach, combining both “genotypic” and “phenotypic” concepts to identify stocks (Begg and Waldman, 1999; Smith et al., 2002). Several variations on the “genotypic” and “phenotypic” stock concepts have been proposed. One that warrants discussion, particularly in light of the “holistic” approach, is Clark’s (1968) and Secor’s (1999) “contingent” stock: a group of individuals that maintain spatial and temporal integrity by engaging in a distinct pattern of migration not shared by individuals of other contingents. Fish following a distinct pattern of movement through space and time will experience the same environment and interbreed, thereby potentially satisfying the conditions of “genotypic” and “phenotypic” stocks. Fish that are lost from a “contingent” potentially provide gene ﬂow to other “contingents,” sensu “vagrants” of Sinclair’s

The Use of Early Life Stages in Stock Identiﬁcation Studies

91

(1988) member-vagrant hypothesis. Whether contingents are ultimately used to deﬁne stocks is unknown, but the “contingent” concept is attractive because it combines ideas regarding distribution and dispersion/migration throughout the life cycle, which are important elements of several unifying hypotheses of ﬁsh population structure: Harden Jones’s (1968) migration triangle, Sinclair’s (1988) member-vagrant hypothesis, MacCall’s (1990) basin model, and Tsukamoto and Aoyama’s (1998) and Tsukamoto’s et al. (2002) migration loop hypothesis. Early life stages (ELSs) can be useful in identifying stocks in context of the “genotypic,” “phenotypic,” and “contingent” stock concepts. The purpose of this chapter is to review the use of planktonic ELSs, in the deﬁnition of ﬁsh and invertebrate stocks. Planktonic ELSs include eggs, larvae, and, in some species, juveniles. The emphasis will be on marine systems, however, the material should be relevant to freshwater systems as well. This chapter is not meant to provide a deﬁnitive review of all instances where ELS information has been used in stock delineation, nor is it a review of the methodology for collecting ELS data; any interested readers are referred to Heath (1992) and Wiebe and Benﬁeld (2003) for reviews of the latter topic. The intent is to review the role of ELS information in the stock concept, provide some examples of the application of ELS information to stock delineation, and conclude with a discussion of the future of ELS studies in stock identiﬁcation.

II. ROLE OF ELS INFORMATION IN THE STOCK CONCEPT All three stock deﬁnitions discussed above are in one form or another based on spatially and temporally discrete groups of individuals at some point during their life cycle. “Genotypic” stocks imply isolation, at least during spawning. “Phenotypic” stocks imply different phenotypes resulting from genotypic or environmental differences during the time when the phenotypic characters develop. The “contingent” concept bridges the “genotypic” and “phenotypic” deﬁnitions and implies discrete distributions during the life cycle (Fig. 5-1). Most marine ﬁshes and invertebrates have planktonic stages that last from days to months (Fig. 5-2). These planktonic eggs and larvae disperse over large areas owing to transport by ocean currents (Scheltema, 1986), and the traditional view is that larvae with longer planktonic durations are dispersed over larger areas. This traditional view carries to the stock concept: greater larval dispersal leads to more mixing of ELSs among spawning locations, which results in less stock structure. This simple view of larval dispersal and stock structure is partially supported by population genetic studies of marine invertebrates and ﬁsh. Invertebrate species with direct development (i.e., crawling juveniles developing directly from benthic egg masses) exhibit greater spatial genetic structure compared to related

92

Jonathan A. Hare Eggs & Preadult Spawning Larvae Juvenile & Adult Spawning

“Genotypic”

“Phenotypic”

Latitude

“Contingent”

ng Lo

itu

de

Time

FIGURE 5-1. Conceptualization of three stock concepts. The “genotypic” stock concept emphasizes discrete spawning from generation to generation. ELSs are used to infer discrete spawning locations. The “phenotypic” stock concept emphasizes the phenotypic expression of characters, typically during the preadult and adult stages, but also larval and juvenile characters. The “contingent” stock concept emphasizes spatially discrete distributions integrated over the entire life cycle. “Genotypic” stocks are often explained in terms of larval transport mechanisms that maintain discrete larval distributions. Here, preadult is deﬁned as juvenile ﬁsh that use adult habitats rather than spatially discrete juvenile nursery habitats (presentation modiﬁed from Secor, 1999).

species with planktonic larvae (Yamada, 1989; Parsons, 1996). Similarly, ﬁsh with shorter planktonic durations have more genetic differences among locations than ﬁsh with longer planktonic durations owing to limited mixing across space (Doherty et al., 1995; Riginos and Victor, 2001). However, the inverse relation between planktonic duration and the amount of spatial population structure is not universal. Shulman and Bermingham (1996) found no evidence for a relationship between population structure and planktonic duration in a large-scale

Number of species examined

The Use of Early Life Stages in Stock Identiﬁcation Studies

90 80 70 60 50 40 30 20 10 0

wk <1

mo –2 k 1w

93

mo >2

Larval duration FIGURE 5-2. Distribution of larval durations in demersal marine ﬁsh and invertebrate species (modiﬁed from Bradbury and Snelgrove, 2001, with permission). Fish are displayed as solid bars and invertebrates as empty bars.

study of several ﬁsh species in the Caribbean (see also Lacson, 1992). One resolution to this apparent dichotomy was readily pointed out by Bradbury and Snelgrove (2001): planktonic duration is only a proxy for actual planktonic transport. Actual planktonic transport, which includes the successful arrival to juvenile habitat, is a function of multiple interacting factors: spawning, advection, dispersion, egg and larval survival, egg and larval behavior, stage durations and the distribution of juvenile habitat. Spawning time and location initiates the transport process (e.g., Iles and Sinclair, 1982; Checkley et al., 1999). Eggs and larvae are then advected and dispersed (actual physical dispersion) in the threedimensional ﬂow ﬁeld (e.g., Werner et al., 1997; Epifanio and Garvine, 2001). The vertical distribution of eggs and larvae modify the outcome of transport in vertically structured ﬂows (e.g., Cowen et al., 1993, 2000; Hare et al., 1999), and some species are able to affect their horizontal distributions through active swimming (e.g., Stobutzki and Bellwood, 1997; Leis and Carson-Ewart, 1997; Kingsford et al., 2002). Egg and larval survival are an important component of larval transport (e.g., Helbig and Pepin, 1998a,b; Cowen et al., 2000; Hare et al., 2002) and are affected by the abiotic environment, feeding conditions, and predation (e.g., Bailey and Houde, 1989; Dower et al., 1997, 2002; Pepin et al., 2002). Stage durations determine the time over which transport occurs and can be ﬁxed or variable (e.g., Cowen, 1991; Bradbury and Snelgrove, 2001). The distribution of ELSs must then intersect with appropriate juvenile habitat to make the transition from larvae to juvenile (e.g., Etherington and Eggleston, 2000; Steves et al., 2000; Chant et al., 2000). Thus, actual planktonic transport—the

94

Jonathan A. Hare

survival of individuals from spawning to arrival at suitable juvenile habitat—is distinct from potential larval dispersal. Actual planktonic transport is an important element in the population dynamics of marine organisms, as well as in the formation of “genotypic,” “phenotypic,” and “contingent” stocks. Many overriding theories of marine population biology involve ELSs: member vs. vagrant, open vs. closed populations, long-distance transport vs. local retention, recruitment limitation vs. density-dependence (Sinclair, 1988; Chesson, 1998; Armsworth, 2002; Strathmann et al., 2002). Local larval retention contributes to membership in a group, closed populations, and potentially density-dependent dynamics. Long-distance transport contributes to vagrancy, open populations, and potentially recruitment-limited dynamics. These contrasts are a matter of scale (Sinclair, 1988), but deﬁne the importance of ELSs to the structure of marine populations. Broadly speaking, “genotypic” stocks form where exchange of individuals among stocks is limited—actual planktonic transport is only one form of exchange that occurs during the life cycle. The primary use of ELSs in stock identiﬁcation pertains to “genotypic” stocks. Egg and larval distributions are used to deﬁne discrete spawning areas and evaluate the potential for mixing among spawning areas during the planktonic stages. A second use of ELSs involves quantifying actual larval transport to test hypotheses resulting from genetic studies of stock structure—many genetic studies use generalizations of actual larval transport to support their ﬁndings. A third use of ELSs in stock identiﬁcation involves characters from ELSs to deﬁne “phenotypic” stocks. A more detailed discussion of these three applications of ELSs to stock identiﬁcations follows.

A. ELSS

AND

DISCRETENESS

OF

SPAWNING

The distribution of eggs and early larvae indicate the spatial and temporal distribution of spawning. When the distributions of these stages are spatially discrete at one time, there is an indication of distinct spawning areas; when distributions are spatially continuous at one time, there is an indication of continuous spawning or mixing of eggs and larvae following distinct spawning. This application of ELSs to stock deﬁnition is the most common and is usually used in the context of deﬁning “genotypic” stocks (Neilson et al., 1988). As an example, Gaughan et al. (2002) found a near continuous spatial occurrence of sardine (Sardinops sagax) eggs along southwest Australia. In addition, Gaughan et al. (2001) used larval ages and measures of eastward ﬂow along South Australia and concluded there was potential for long-range larval transport from west to east along the coast. Both studies provided evidence of a single stock. In contrast, earlier studies found spatial differences in allozymes and otolith chemistry, indicating separate stocks (Edmonds and Fletcher, 1997, cited in Gaughan et al., 2001). These apparently conﬂicting results can be explained. For example,

The Use of Early Life Stages in Stock Identiﬁcation Studies

95

a panmictic population with freely mixing ELSs can experience localized selection that results in different allozyme frequencies among areas, or can experience local differences in the environment that result in spatial differences in otolith chemistry. Recognizing these conﬂicting results and the attributes of the different stock identiﬁcation techniques, Gaughan et al. (2002) concluded that management must give consideration to smaller-scale reproductive units, within a single breeding stock. The results and conclusions of Gaughan et al. (2001, 2002) emphasize the general limitation of egg and larval distribution studies in identifying “genotypic” stocks. “Genotypic” stocks require that most individuals spawned in an area either remain in that area to spawn or return to that area to spawn. Egg and larval distributions can be used to determine if spawning is spatially and temporally distinct. Movement of eggs and larvae after spawning can be inferred from age or stage distributions, model results, or actual observations (e.g., Gaughan et al., 2001), but it is impossible through ELS studies alone to determine if individuals remain in or return to their spawning location. The separation of ELSs in space and time further confounds conclusions regarding stock identiﬁcation. For example, based on larval distributions, Kendall and Walford (1979) concluded that blueﬁsh (Pomatomus saltatrix) spawn in the spring on the southeast United States shelf and in the summer on the northeast United States shelf. Two stocks could be inferred from the distinct distribution of ELSs, but Chiarella and Conover (1991) found spring-spawned blueﬁsh spawning in the summer, thereby refuting the possibility of “genotypic” stocks. A real problem when spawning is separated in space and time is determining the individual patterns in reproduction and migration (Hare and Cowen, 1993). The blueﬁsh example demonstrates important limitations of ELSs applied to stock identiﬁcation. For the “genotypic” stock deﬁnition, spawning distribution and mixing after spawning can be inferred, but the patterns of individual spawning and homing are not addressed. For the “contingent” stock deﬁnition, separation in time and space during ELSs provides only a partial picture of the temporal and spatial distribution of individuals throughout the life cycle (see Fig. 5-1).

B. ELSS

AND

GENETIC POPULATION STRUCTURE

The results of genetic stock identiﬁcation studies are often discussed in the context of dispersal during ELSs. Typically when genetic differences are found among locations, retention mechanisms for larvae within locations are identiﬁed. When genetic differences are not found among locations, mechanisms of larval transport among locations are discussed. As an example, Ruzzante et al. (1999) reviewed a series of studies that showed signiﬁcant differences in genetic composition among Atlantic cod (Gadus morhua) at a variety of geographic scales in the northwest Atlantic Ocean. They concluded that topographically inﬂuenced

96

Jonathan A. Hare

gyre-like circulation on cod spawning banks retain cod eggs, larvae, and juveniles, and facilitate late-juvenile/adult spawning site ﬁdelity, which in turn contributes to the observed genetic differences. Similarly, Stepien (1999) found signiﬁcant genetic divergences in Dover sole (Microstomus paciﬁcus) from Alaska to southern California and hypothesized that the observed differences result from barriers to larval dispersal or local retention of larvae in currents. As a ﬁnal example, Planes (1993) explained the genetic structure of convict surgeonﬁsh (Acanthurus triostegus) in the context of prevailing currents in French Polynesia and hypothesized that the existence of gyres around islands could cause local recruitment of larvae. The examination of ELS distributions and, more explicitly, the tracking of ELSs through time provides a method by which these hypotheses can be tested (see Drouin et al., 2002).

C. ELSS

AND

PHENOTYPIC TRAITS

Phenotypic stocks are most often deﬁned using characters from juvenile and adult ﬁsh collected at different locations. The range of characters used is large and includes body and otolith morphometrics (e.g., Swain and Foote, 1999; Cadrin and Friedland, 1999) and life history traits (e.g., Begg et al., 1999b). Analyses then test whether there are spatial differences in the traits in question. Phenotypic traits of eggs and larvae are rarely used in the deﬁnition of “phenotypic” stocks. Jamieson and Phillips (1993) provided one example. Speciﬁcally, the vertical distribution and size-at-settlement of Dungeness crab (Cancer magister) larvae was different between locations on the shelf and in the Straits of Georgia. These phenotypic differences would limit the mixing of larvae between locations, potentially contributing to genotypic separation. Larval morphometrics also have been used in stock identiﬁcation similar to adult morphometrics. Harding et al. (1993) used seven morphometric measurements of American lobster (Homarus americanus) larvae from the northwestern Atlantic Ocean and identiﬁed three spatial groups, inferring three “phenotypic” stocks. However, Harding et al. (1993) also found a signiﬁcant effect of temperature on the morphometric variables, and once this effect was removed, two spatial groups were identiﬁed. By removing the effects of a dominant environmental variable, Harding et al. (1993) extended their phenotypic analysis closer to “genotypic” stock identiﬁcation, but they point out that maternal effects, food resources, or other unknown environmental factors could cause the spatial differences in temperature-corrected morphometrics. In addition to behavior and morphometrics, other phenotypic and life history traits exhibited by larvae have been used in stock identiﬁcation studies. Burke et al. (2000) examined a suite of larval and early juvenile characters of summer ﬂounder (Paralichthys dentatus), including ingress patterns, size and develop-

The Use of Early Life Stages in Stock Identiﬁcation Studies

97

mental stage-at-ingress, meristics, and laboratory growth and survival. They concluded that observed differences support the hypothesis of two “phenotypic” stocks separated at Cape Hatteras, North Carolina. As another example, Castello and Castello (2003) argued that larval ﬁsh growth is controlled by the environment, and differences in growth can be used as a proxy for differences in the environment experienced. They found variable larval growth in anchovy (Engraulis anchoita), providing evidence for distinct stocks. Although ELS phenotypic traits are not commonly used in stock identiﬁcation, this approach may become more important because of recent studies demonstrating the importance of maternal and ELS traits to subsequent juvenile and adult traits and survival (Meekan and Fortier, 1996; Hare and Cowen, 1997; Searcy and Sponaugle, 2000, 2001).

III. EXAMPLES OF ELS INFORMATION IN THE DEFINITION OF STOCKS There are a number of examples of ELS information used in stock identiﬁcation. Some have already been discussed brieﬂy as they pertain to the value of ELSs in stock identiﬁcation. In this section, three case studies are presented where ELSs made a signiﬁcant contribution to the deﬁnition of stocks. The one overriding theme is that ELSs do not provide a comprehensive approach to stock identiﬁcation. Rather, ELSs contribute to a holistic approach, one using multiple characters pertaining to different facets of the stock concept (Begg and Waldman, 1999).

A. ATLANTIC HERRING

IN THE

NORTHWEST ATLANTIC OCEAN

Atlantic herring (Clupea harengus) provides the quintessential example of ELSs used in stock identiﬁcation. Atlantic herring are widely distributed on both sides of the North Atlantic. Since the late 1800s, “stocks” of herring have been identiﬁed based on speciﬁc spawning areas and times (see Sinclair, 1988). Primary spawning locations off the northeast United States include the Maine coast, Jeffreys Ledge, Nantucket Shoals, and Georges Bank (Fig. 5-3A). Iles and Sinclair (1982), Sinclair and Iles (1985), and Sinclair (1988) proposed that discrete spawning occurs in locations that promote larval retention and permit the persistence of larval distributions for a few months after hatching (Fig. 5-3B). Following the larval stage, juveniles and prespawning adults mix, but adults then home to their natal spawning areas. Thus, these discrete spawning areas represent distinct “genotypic” stocks. Just as larval distributions were used to support the existence of Atlantic herring stocks based on spawning areas, larval distributions also were used to refute aspects of the hypothesis. Fortier and Gagné (1990) found that Atlantic

A

B

C

< 2 wks

5-8 wks

2-5 wks

> 8 wks

FIGURE 5-3. (A) Spawning and (B) larval distributions of Atlantic herring (Clupea harengus) in the Gulf of Maine region of the northwestern Atlantic Ocean (from Sinclair, 1988, with permission). These data were used to support stock identiﬁcation based on spawning areas. (C) Composite larval Atlantic herring distributions by approximate age from 1971 to 1975 (from Smith and Morse, 1993). There are six gradations in the grayscale representing 0 (white, 1–2, 2–5, 5–10, 10–25, >25 (black) larvae 10m-2. Distributions indicate mixing of larvae after approximately 8 weeks between the Georges Bank and Nantucket Shoals spawning groups.

The Use of Early Life Stages in Stock Identiﬁcation Studies

99

herring larvae spawned in the St. Lawrence River during the spring were retained near spawning grounds, whereas larvae spawned during the fall were advected away from spawning grounds. Other studies have found that Atlantic herring larvae were advected away from spawning areas (e.g., Townsend et al., 1986; Chenoweth et al., 1989), bringing into question the retention of Atlantic herring larvae near spawning grounds. Based on nearly 20 years of ichthyoplankton data from the northeast United States continental shelf, Smith and Morse (1993) found evidence of discrete distributions of larvae <5 weeks old on Georges Bank and Nantucket Shoals, but also found evidence for mixing between the groups of larvae between 5 and 8 weeks old (Fig. 5-3C). The questions become: What is the actual transport of Atlantic herring larvae from discrete spawning areas and do adults home to natal spawning areas? Genetic stock identiﬁcation studies do not support the hypothesis that discrete spawning areas of Atlantic herring form “genotypic” stocks. Allozyme and restriction fragment length polymorphism data found little or no genetic difference between Atlantic herring from separate spawning locations (see Smith and Jamieson, 1986; Kornﬁeld and Bogdanowicz, 1987). Similarly, McPherson et al. (2001) found limited genetic differences between Georges Bank and coastal Nova Scotia Atlantic herring using microsatellites, but did ﬁnd differences between these two groups and Atlantic herring from the Cape Breton Lakes region of northern Nova Scotia. These genetic studies support the hypothesis of one, or at least only a few, genotypic stocks of Atlantic herring in the northwest Atlantic Ocean. McQuinn (1997) attempted to reconcile the two points of view—multiple stocks deﬁned by multiple discrete spawning areas vs. multiple discrete spawning areas but limited number of stocks. McQuinn (1997) framed his reconciliation in the metapopulation concept and pointed out that an important issue is the fate of individuals that “stray” from their natal spawning location. Thus, strays or migrants between local populations (i.e., spawning locations) are integral to the structure and dynamics of the larger metapopulation. Stocks in a metapopulation concept are reviewed elsewhere (Secor, 2004), and management of Atlantic herring does not currently use a metapopulation model, but managers are aware of the spatial complexity of Atlantic herring population structure (Stephenson et al., 2001). In the future, studies of actual planktonic transport between spawning locations will play a role in building the “metapopulation” stock concept for Atlantic herring in the northwest Atlantic Ocean.

B. JAPANESE EEL

IN THE

WESTERN PACIFIC OCEAN

The Japanese eel (Anguilla japonica) is distributed from south of Taiwan, throughout Japan, and along the coasts of China and Korea (see Fig. 5-4). Matsui (1957)

100

Jonathan A. Hare

FIGURE 5-4. Ages of Japanese eel (Anguilla japonica) elvers by latitude. Ages increase with latitude from 20°–25°N to 30–36°N, suggesting larvae could be from the same source, with age increasing with distance along the larval transport route. Age data from 20°–25°N were from elvers collected in Taiwan (Tzeng and Tsai, 1992). Age data from 30°–36°N were from elvers collected in Japan (Tsukamoto, 1990).

(cited in Tzeng and Tsai, 1992) proposed that Japanese eel spawn south of Okinawa and east of Taiwan, implying that there is one stock in the western Paciﬁc. Kuo (1971) and Chen (1975) (cited in Tzeng and Tsai, 1992) hypothesized a second spawning area southwest of Taiwan, based on the observation that elvers (i.e., larvae transforming into juveniles) were more abundant along the northern and western coasts of Taiwan (i.e., the opposite side of Taiwan from the presumed spawning location). Two aspects of the distribution of Japanese eel ELSs support the one stock hypothesis. First, ichthyoplankton surveys in the western Paciﬁc found progressively smaller and younger leptocephali (i.e., the larval stages of eel) eastward from Taiwan (Kajihara, 1988; Ozawa et al., 1989; Tsukamoto et al., 1989; Tsukamoto, 1992; Kimura et al., 1994). If there were two spawning locations, young larvae would be expected closer to Taiwan (Tsukamoto, 1992; Kimura et al., 1994). Second, Tzeng and Tsia (1992) hypothesized that if there were two stocks, then there would be older elvers on the southern and eastern Taiwanese coast, which result from spawning east of the Philippines, and younger elvers on the northern and western Taiwanese coasts, which result from more localized spawning southwest of Taiwan. Tzeng and Tsia (1992) found, however, that there were not signiﬁcant age differences among Taiwanese estuaries. In a similar study along the southern coast of Japan, Tsukamoto (1990) found similar ages among

The Use of Early Life Stages in Stock Identiﬁcation Studies

101

locations. Combining the age data from Japan and Taiwan indicates that the age of elvers entering estuaries increases as distance along the presumed larval transport route increases (Fig. 5-4) (see Kimura et al., 1994 for more detail on larval transport). Cheng and Tzeng (1996) also found a cline in ages at estuarine ingress, supporting the previous conclusions of Tsukamoto (1990) and Tzeng and Tsia (1992). A subsequent genetic study found no evidence for spatial or temporal genetic heterogeneity in glass eels (i.e., juvenile stage) from Taiwan, China, and Japan, providing further support for the one stock hypothesis (Ishikawa et al., 2001). Two explanations can be provided for the pattern of increasing age of eels from east to west. Either larvae are spawned in the east and transported west or spawning occurs over time from west to east, along the transport route. These two explanations for the same data illustrate the problem of analyzing distribution data, recognized by Andrewartha and Birch (1954): mapping the distribution at one time provides a static description confounding the effects of spawning, advection, dispersion, and survival. Knowing the source of larvae and mapping their distribution repeatedly over time provides an explicit deﬁnition of spawning, advection, dispersion, and survival. However, in the case of Japanese eel, Cheng and Tzeng (1996) found that speed of the Kuroshio Current coupled with the distances between sampling sites agreed with the difference in ages of estuarine ingress, supporting the hypothesis of one spawning site with subsequent larval transport along the transport route. The ability to interpret larval distributions and larval ages in the context of physical oceanography strengthens the inferences regarding actual planktonic transport, whether samples are collected at one time or at multiple times (see Hare and Cowen, 1991; Cowen et al., 1993).

C. AMERICAN LOBSTER

IN THE

NORTHWEST ATLANTIC OCEAN

American lobster (Homarus americanus) is distributed in the northwest Atlantic Ocean from Labrador to North Carolina, from the coast to the shelf edge. Lobster were traditionally harvested by an inshore trap ﬁshery, but an offshore trawl ﬁshery started in the 1950s and an offshore trap ﬁshery developed in the 1970s (Lockhart and Estrella, 1997). Concern grew that the heavily ﬁshed inshore lobsters depended on larval supply from offshore lobster, and the spread of ﬁshing into offshore areas could be a signiﬁcant threat to the inshore ﬁshery (Lockhart and Estrella, 1997). Fogarty (1998) demonstrated that even limited supply of lobster from offshore could explain the continued resilience of inshore lobster to very high ﬁshing mortality rates. The issue is one of stock structure: are inshore and offshore lobster components of the same stock? A number of studies have examined potential mixing of lobster between inshore and offshore areas and have reached different conclusions. Tagging

102

Jonathan A. Hare

studies found movement of lobsters from offshore to inshore (e.g., Cooper and Uzmann, 1971). Morphometric studies have found differences (Cadrin, 1995), and no differences (Harding et al., 1993; Cadrin, 1995), between inshore and offshore samples. However, morphometric differences were found between Gulf of St. Lawrence lobsters and individuals from more southern locations (Harding et al., 1993). Finally, genetic studies indicated little difference between inshore and offshore samples, but found some evidence for differences between Gulf of St. Lawrence samples and more southern samples (e.g., Tracey et al., 1975; Harding et al., 1997). ELSs of lobster also were used to examine the issue of stock delineation between inshore and offshore lobsters. Harding and Trites (1988) used larval surveys and drift card releases to infer that larvae spawned offshore (over Browns Bank) would be transported inshore (into the Bay of Fundy and along the coast of Maine). Katz et al. (1994) described an offshore to inshore gradient in the ontogenetic stages of larval lobster (Fig. 5-5A), suggesting offshore to inshore transport (or an inshore to offshore pattern of spawning with little transport). They developed a two-dimensional circulation model that indicated offshore to inshore transport of larvae could occur through a combination of advection, windinduced surface ﬂow, and directional swimming by late-stage larvae (Fig. 5-5B). Incze and Naimie (2000) examined lobster larval transport in the Gulf of Maine region using a three-dimensional numerical circulation model. Results showed examples of long-distance transport and local retention and indicated the potential importance of diurnal sea breeze in the cross-isobath transport of larvae. Incze and Naime (2000) did not expressly examine the cross-shelf transport of lobster larvae, but their model indicated the potential for mixing of larvae between locations. Data derived from ELSs contribute to the conclusion that inshore and offshore groups of lobster do not function independently and should be managed as one unit stock. Although there are data that support “phenotypic” stock separation (e.g., Cadrin, 1995), the precautionary approach and analyses of Fogarty (1998) indicate that larval supply from offshore needs to be considered in the management of the heavily exploited inshore lobster ﬁshery. One spatial difference that appeared in most comparisons was lobsters from the Gulf of St. Lawrence were distinct from lobsters further south. Harding et al. (1997) hypothesized that genotypic and phenotypic differences might occur as a result of “one-way” larval transport from the Gulf of St. Lawrence, forming a partial hydrographic barrier to larval transport. ELS studies could be used to test this stock hypothesis through broader-scale surveys of berried females and larvae, coupled with a larger spatial domain three-dimensional circulation model to examine actual planktonic transport.

A

B

FIGURE 5-5. (A) Cross-shelf distribution of American lobster (Homarus americanus) larvae by stage. The data show an increase in stage from offshore to inshore, suggesting onshore transport. PL: postlarvae. (B) Simulated transport of American lobster larvae as simulated by average currents, winddriven ﬂow, and diurnal horizontal swimming behavior (from Katz et al., 1994, with permission). 103

104

Jonathan A. Hare

IV. CONCLUSIONS AND FUTURE DIRECTIONS The strength of ELSs to stock identiﬁcation is their part in a holistic approach— one tool to examine the mechanisms responsible for forming “genotypic” stocks or to provide a set of characters for examining “phenotypic” stock structure. Further, understanding actual larval transport provides one piece of the contingent puzzle (see Fig. 5-1). Speciﬁcally, ELSs provide a variety of information applicable to stock identiﬁcation. Egg and larval distributions deﬁne spawning areas (Iles and Sinclair, 1982; Gaughan et al., 2002). Size, age, and stage of larvae used in combination with physical oceanographic data provide a measure of actual planktonic transport and assist in the interpretation of distributions (Smith and Morse, 1993; Katz et al., 1994; Gaughan et al., 2001). Genetic studies on larvae provide another facet to examining genetic population structure (Ruzzante et al., 1999) and for testing hypotheses regarding the role of planktonic transport in maintaining genetic population structure (Drouin et al., 2002). Larval traits contribute to the identiﬁcation of phenotypic stocks (Harding et al., 1993; Burke et al., 2000). In the future, improvements in modeling and quantifying exchange between local groups will enhance our ability to identify stocks and, ultimately, manage marine populations. In particular, research and development in four areas will greatly improve the application of ELSs to stock identiﬁcation: increased use of geostatistics in the analysis of ELS distributions, a system-level framework for the application of ELSs to stock identiﬁcation, bio-physical modeling of actual planktonic transport, and the development of a “metapopulation” stock concept.

A. INCREASED USE

OF

GEOSTATISTICS

Future stock identiﬁcation studies using egg and larval distributions will beneﬁt from increased use of geostatistics. Begg et al. (1999b) qualitatively examined egg, larval, juvenile, and adult distributions and combined this information with quantitative measures of other life history parameters to evaluate stock structure of Atlantic cod, haddock (Melanogrammus aegleﬁnus), and yellowtail ﬂounder (Limanda ferruginea). The qualitative examination of ELS distribution could have been improved by the use of quantitative analysis of the distribution data. The application of geostatistics to egg and larval distribution data is relatively new. Geostatistics are used to examine and quantify the spatial patterns of a given variable, such as abundance. Several studies have applied geostatistical techniques to quantify the distribution of ELSs. Fletcher and Sumner (1999) determined the spatial scale of different early stages of sardine (Sardinops sagax) using two-dimensional spatial correlograms (see also Bez and Rivoirard, 2001). Grioche et al. (2001) used variograms to deﬁne the links between larval distributions and

The Use of Early Life Stages in Stock Identiﬁcation Studies

105

environmental variables. Although these studies did not address stock identiﬁcation questions, the approaches could be used to quantitatively examine the link between egg, larval, and juvenile distributions. Such use would strengthen the inferences regarding the discreteness or lack of discreteness of ELSs spawned in particular areas and thereby contribute to stock identiﬁcation studies.

B. AN ECOSYSTEM FRAMEWORK IN STOCK IDENTIFICATION

FOR THE INCLUSION OF

ELSS

Most applications of ELSs to stock identiﬁcation involve a single species. However, in a given ecosystem, the ELSs of different species that are released or hatch in the same location at the same time will be subjected to the same physical processes of advection and dispersion. Subsets of these ELSs may interact differently with the physical environment owing to differences in vertical or horizontal behavior, survival, planktonic duration, or transition to juvenile habitats, and as a result their planktonic distributions will change accordingly. These spatially and temporally co-occurring groups of eggs and larvae are termed assemblages (Cowen et al., 1993), and by comparing distributions among and within assemblages in terms of life history similarities, with concomitant measures of the physical environment, insights can be gained into the processes that affect ELS distribution, transport, and, ultimately, the spatial structure of marine populations (Cowen et al., 1993; Hare et al., 2001). The use of ELSs in stock identiﬁcation will beneﬁt from an assemblage/ecosystem approach. For example, the stock structure of species on the northeast United States shelf has been studied extensively: Atlantic cod (e.g., Begg et al., 1999b; Ruzzante et al., 1999), haddock (e.g., Begg, 1998; Begg et al., 1999b), Atlantic herring (e.g., Kornﬁeld and Bogdanowicz, 1987; Sinclair, 1988), American lobster (e.g., Harding and Trites, 1988; Harding et al., 1997), yellowtail ﬂounder (e.g., Neilson et al., 1988; Begg et al., 1999b), American plaice (e.g., Neilson et al., 1988), and witch ﬂounder (e.g., Neilson et al., 1988). Many of these studies refer to ELSs and stock identiﬁcation. Doyle et al. (1993) deﬁned larval assemblages on the northeast United States shelf, and the occurrence of managed species in discrete larval assemblages (Table 5-1) indicates that the processes contributing to the formation of larval distributions of some of these species are similar (e.g., Atlantic cod, haddock, and American plaice). To the extent that ELSs affect stock structure, there should be similarities in the stock structure of species that are part of the same larval assemblage (e.g., Atlantic cod, haddock, and American plaice). Identifying these similarities and developing and testing hypotheses regarding the processes that result in similarities and/or dissimilarities in stock structure among species will strengthen the basis for understanding stock structure in marine systems. Through such comparisons of ELSs and stock structure

106

Jonathan A. Hare

TABLE 5-1. Larval Assemblage Membership of Species of Management Concern on the Northeast United States Continental Shelf as Deﬁned by Doyle et al. (1993)a Species Atlantic herring Clupea harengus Atlantic cod Gadus morhua Haddock Melanogrammus aegleﬁnus American plaice Hippoglossoides platessoides Atlantic mackerel Scomber scombrus Witch ﬂounder Glyptocephalus cynoglossus Yellowtail ﬂounder Limanda ferruginea Gooseﬁsh Lophius americanus Summer ﬂounder Paralichthys americanus Atlantic menhaden Brevoortia tyrannus Atlantic croaker Micropogonias undulatus

Spring

Summer

A1

Fall

Winter

A5

A7

A5

A1 A1 A2 A2

A4

A2

A4

A3

A4 A6 A6 A6

a

Species shown here are a subset of the total number of species examined by Doyle et al. (1993), and the assemblages indicated here have been renumbered from Doyle et al. (1993) for clarity. Analysis based on ichthyoplankton sampling conducted on the northeast continental shelf from 1977 to 1987. The same number indicates membership in the same larval assemblage.

of a suite of managed species occurring in an ecosystem, stock deﬁnition in individual species will improve, and management in general will move from a single species emphasis toward an ecosystem approach (National Marine Fisheries Service, 1999).

C. MODELING AND MEASURING ACTUAL PLANKTONIC TRANSPORT The next great improvement in the application of ELSs to stock identiﬁcation will come from quantifying actual planktonic transport through both modeling and observation. Modeling of larval transport is already well developed in the form of individual based bio-physical models (see review by Werner et al., 2001).

The Use of Early Life Stages in Stock Identiﬁcation Studies

107

Individual-based models, as applied to marine ﬁsh populations, were developed initially to examine individual feeding and the population consequences of feeding and growth (e.g., Beyer and Laurence, 1980; Laurence, 1985). Roughly during the same period, sophisticated and realistic circulation models were developed (e.g., Blumberg and Mellor, 1987), which initially predicted paths of passive or ﬁxed-depth particles, but then incorporated more realistic vertical and horizontal behaviors (Werner et al., 1993; Hare et al., 1999). The two approaches— trophodynamic and transport—were recently combined (Hermann et al., 1996; Hinckley et al., 2001), and as these models become more integrated and more complex, they will capture more of the processes involved in determining actual planktonic transport. The next step is applying the coupled, bio-physical individual-based models to stock identiﬁcation questions. As an example, Skogen et al. (1999) examined stock separation in northeast Atlantic blue whiting (Micromesistius poutassou) using a three-dimensional circulation model. They released ﬁxed-depth particles in the model domain from the southwest England shelf, along the shelf west of Ireland, to the northern shelf of Scotland, and then quantiﬁed the direction of transport. They found that southward drift occurred for particles released in the southern portion of the domain and northward drift occurred for particles in the northern portion of the domain, demonstrating potential stock separation on the basis of discrete transport of planktonic stages. However, over the 20 years examined, the “line” separating the two groups varied by 200 km, indicating a zone of mixing between the two potential stocks. In the end, the results were equivocal, but provide some evidence for differences in larval transport based on spawning area. In the future, coupling the stock identiﬁcation focus of Skogen et al. (1999) with more realistic models of actual planktonic transport will provide a powerful approach for addressing stock identiﬁcation questions based on ELSs. Another hurdle facing the use of ELSs in stock identiﬁcation is observing and quantifying actual larval transport from known source locations. The traditional view of long-distance transport and large, open marine populations is giving way to the view of spatially complex marine populations that are created in part by a continuum of long-distance transport and local retention (e.g., Cowen et al., 2000). However, without being able to follow larvae spawned at known locations, it is impossible to quantify actual larval transport and, as a result, impossible to quantify planktonic connectivity between spawning groups. A number of methods exist for marking larvae in the ﬁeld (see review by Thorrold et al., 2002). Markers can be either artiﬁcial or natural. A marker not discussed by Thorrold et al. (2002) is larvae that are transported beyond their spawning range. Larvae expatriated from the southeast to northeast United States shelf illustrate longdistance and cross-shelf transport mechanisms (Cowen et al., 1993; Hare et al., 2002), and larvae that spawn on the shelf and enter estuarine nursery habitats illustrate mechanisms of estuarine ingress and retention (Churchill et al., 1999;

108

Jonathan A. Hare

Forward et al., 1999). Although there are major logistical obstacles to actually following larvae in the ﬁeld (Thorrold et al., 2002), the application of marked eggs and larvae in the ﬁeld to study actual planktonic transport will greatly enhance the application of ELSs to stock identiﬁcation.

D. SPATIAL POPULATION MODELS

AND THE

STOCK CONCEPT

The growing awareness of the spatially complex structure of marine populations led to the reintroduction of the “contingent” stock hypothesis (Secor, 1999) and the call for a metapopulation approach to stock identiﬁcation (McQuinn, 1997; Thorrold et al., 2001; Smedbol and Wroblewski, 2002; but see Smedbol et al., 2002 for a critique of the term metapopulation as applied to marine ﬁshes). A critical element of these stock concepts is quantifying the exchange of individuals among separate groups. Even limited exchange will result in genetic homogeneity between the groups (Whitlock and McCauley, 1999), but from the point of view of population dynamics and stock identiﬁcation, different groups may exhibit different dynamics, even with exchange. In the future, stock identiﬁcation based on the “contigent” or “metapopulation” concept depends on quantifying exchange rates during all stages of the life history. A relevant example comes from weakﬁsh (Cynoscion regalis) along the east coast of the United States. Currently, ﬁshery management assumes one stock along the east coast of the United States, based on genetic homogeneity of allozymes, mtDNA, microsatellites, and intron markers (Crawford et al., 1989; Graves et al., 1992; Cordes and Graves, 2003). However, Thorrold et al. (2001) demonstrated signiﬁcant homing to estuarine nursery areas by adults using otolith microchemistry, and indicated that management would be improved by the application of a spatially structured population model to weakﬁsh along the east coast of the United States (Fig. 5-6). Rates of exchange between juvenile and adult stages can be quantiﬁed from data in Thorrold et al. (2001). However, rates of exchange between spawning and the juvenile stage are unknown. Rowe and Epifanio (1994a,b) found that weakﬁsh larvae are retained within estuaries owing to interactions between larval vertical distributions and estuarine ﬂow ﬁelds. However, weakﬁsh eggs and larvae have been collected on the inner shelf (e.g., Cowen et al., 1993; Berrien and Sibunka, 1999), indicating potential exchange of ELSs between estuarine systems. To fully apply spatially structured concepts to stocks and ﬁshery management, rates of exchange need to be quantiﬁed throughout the life cycle. In the case of weakﬁsh, this would require examining the transport and survival of larvae between estuaries along the east coast of the United States (Fig. 5-6). The “metapopulation” concept of Atlantic herring (Clupea harengus) offered by McQuinn (1997) presents similar challenges: the need to estimate the rate of

The Use of Early Life Stages in Stock Identiﬁcation Studies

109

First Year Second Year S E L J JO S A AO S Thorrold et al. (2001) ? ? Peconic Bay

?

?

?

?

?

?

?

?

Delaware Bay

Chesapeake Bay

Pamlico Sound

Coastal Georgia FIGURE 5-6. Schematic of life history of weakﬁsh (Cynoscion regalis) along the east coast of the United States. Spawning (S) occurs in or near estuaries (Daniels and Graves, 1994; Luczkovich et al., 1999). Mixing of eggs (E) and larvae (L) between estuarine spawning areas is unknown. Larvae do exhibit selective tidal stream transport that would result in larval retention within estuaries (Rowe and Epifanio, 1994a,b), but larvae have also been collected on the continental shelf (Cowen et al., 1993; Berrien and Sibunka, 1999), which could lead to transport between estuaries. Juveniles (J) use estuarine nursery habitats through the summer and into the fall. Juveniles then leave estuaries and move southward and offshore to juvenile overwintering grounds (JO) (Wilk, 1974), where ﬁsh from different spawning estuaries mix. Adults (A) also mix during winter (AO), but return to their juvenile estuaries with 65% to 80% ﬁdelity (Thorrold et al., 2001). There is apparently enough exchange to limit genetic differentiation between spawning sites (Cores and Graves, 2001), but the rate of natal homing indicates substantial population structure (Thorrold et al., 2001). To develop a spatially explicit population model, information regarding exchange of eggs and larvae between estuaries is needed.

exchange among local populations within a metapopulation. Here the modeling approaches previously described would be quite useful. For example, James et al. (2002) used a regional-scale hydrodynamic model to examine larval transport and metapopulation structure of reef ﬁsh in the Great Barrier Reef Marine

110

Jonathan A. Hare

Park. Future integration of biophysical models with spatial population structure concepts will truly advance the use of ELSs in stock identiﬁcation. The concepts of member-vagrant, open vs. closed populations, long-distance transport vs. local retention, and recruitment limitation vs. density dependence are all related (Sinclair, 1988; Armsworth, 2002; Strathmann et al., 2002) and relevant to spatially structured populations and the deﬁnition of stocks, particularly the “contingent” concept. With regard to ﬁsheries species, however, the easy trap is to focus on these concepts from the point of view of ELSs and ignore connections during other parts of the life cycle. The assumption that the planktonic larvae stage is the only stage of exchange between spatially separated groups is valid for sessile species. However, homing in weakﬁsh (Thorrold et al., 2001) and Atlantic herring (McQuinn, 1997) serves as an important reminder that most ﬁshery species move during all life stages, and that these movements are important to the spatial structure of the population. This reminder of the implementation of ELSs to stock identiﬁcation also serves to reiterate that the distribution and traits of ELSs are only one piece of the life cycle, and that the whole life cycle must be considered to examine spatial structure of ﬁsh stocks. ELSs can only be used to infer “genotypic” or “phenotypic” stocks; these stages do not provide direct evidence of generational consistency in spawning and rarely do they exhibit traits directly relevant to ﬁsheries management. Early life history stages should only be used in combination with other approaches (Begg and Waldman, 1999).

ACKNOWLEDGMENTS First, I thank Gavin Begg, Steve Cadrin, Kevin Friedland, and John Waldman for spurring me to think about early life stages and stock identiﬁcation. Emma Jugovich (Center for Coastal Fisheries and Habitat Research) provided assistance accumulating literature. Allyn Powell (Center for Coastal Fisheries and Habitat Research), Frank Hernandez (Center for Coastal Fisheries and Habitat Research), and Dariusz Fey (Morski Instytut Rybacki) reviewed earlier drafts of the manuscript. Katsumi Tsukamoto (Ocean Research Institute, The University of Tokyo), Bill Overholtz (Northeast Fisheries Science Center), and Mike Fogarty (Northeast Fisheries Science Center) reviewed and improved the sections on Japanese eel, Atlantic herring, and American lobster, respectively. Finally, Dennis Hedgecock (University of Southern California) contributed to my thoughts and to ﬁnding literature regarding exchange and genetic structure in marine populations.

REFERENCES Andrewartha, H. G. and Birch, L. C. 1954. The Distribution and Abundance of Animals. University of Chicago Press, Chicago, Illinois. 782 pp. Armsworth, P. R. 2002. Recruitment limitation, population regulation, and larval connectivity in reef ﬁsh metapopulations. Ecology 83: 1092–1104.

The Use of Early Life Stages in Stock Identiﬁcation Studies

111

Bailey, K. M. and Houde, E. D. 1989. Predation on eggs and larvae of marine ﬁshes and the recruitment problem. Advances in Marine Biology 25: 1–67. Begg, G. A. 1998. A review of stock identiﬁcation of haddock, Melanogrammus aegleﬁnus, in the Northwest Atlantic Ocean. Marine Fisheries Review 60(4): 1–15. Begg, G. A. and Waldman, J. R. 1999. An holistic approach to ﬁsh stock identiﬁcation. Fisheries Research 43: 35–44. Begg, G. A., Friedland, K. D., and Pearce, J. B. 1999a. Stock identiﬁcation and its role in stock assessment and ﬁsheries management: an overview. Fisheries Research 43: 1–8. Begg, G. A., Hare, J. A., and Sheehan, D. D. 1999b. The role of life history parameters as indicators of stock structure. Fisheries Research 43: 141–163. Berrien, P. and Sibunka, J. 1999. Distribution patterns of ﬁsh eggs in the U.S. northeast continental shelf ecosystem, 1977–1987. NOAA Technical Report NMFS 145. Beyer, J. E. and Laurence, G. C. 1980. A stochastic model of larval growth. Ecological Modeling 8: 109–132. Bez, N. and Rivoirard, J. 2001. Transitive geostatistics to characterize spatial aggegrations with diffuse limits: an application on mackerel ichthyoplankton. Fisheries Research 50: 41–58. Blumberg, A. F. and Mellor, G. L. 1987. A Description of a Three-Dimensional Coastal Ocean Circulation Model. Three-Dimensional Coastal Ocean Models. American Geophysical Union, Washington, D.C. 208 pp. Booke, H. E. 1981. The conundrum of the stock concept—are nature and nurture deﬁnable in ﬁshery science? Canadian Journal of Fisheries and Aquatic Sciences 38: 1479–1480. Bradbury, I. R. and Snelgrove, P. V. 2001. Contrasting larval transport in demersal ﬁsh and benthic invertebrates: the roles of behavior and advective processes in determining spatial pattern. Canadian Journal of Fisheries and Aquatic Sciences 58: 811–823. Brown, B. E., Darcy, G. H., and Overholtz, W. 1987. Stock assessment/stock identiﬁcation: an interactive process. NOAA Technical Memorandum NMFS-SEFSC 199: 1–19. Burke, J. S., Monaghan, J. P., and Yokoyama, S. 2000. Efforts to understand stock structure of summer ﬂounder (Paralichthys dentatus) in North Carolina, USA. Journal of Sea Research 44: 111–112. Cadrin, S. X. 1995. Discrimination of American lobster (Homarus americanus) stocks off southern New England on the basis of secondary sex character allometry. Canadian Journal of Fisheries and Aquatic Sciences 52: 2712–2723. Cadrin, S. X. and Friedland, K. D. 1999. The utility of image processing techniques for morphometric analysis and stock identiﬁcation. Fisheries Research 43: 129–139. Carvalho, G. R. and Nigmatullin, C. M. 1998. Stock structure analysis and species identiﬁcation. FAO Fisheries Technical Papers 376: 199–232. Castello, L. and Castello, J. P. 2003. Anchovy stocks (Engraulis anchoita) and larval growth in the SW Atlantic. Fisheries Science 59: 409–421. Chant, R. J., Curran, M. C., Able, K. W., and Glenn, S. M. 2000. Delivery of winter ﬂounder (Pseudopleuronectes americanus) larvae to settlement habitats in coves near tidal inlets. Estuarine, Coastal and Shelf Science 51: 529–541. Checkley, D. M., Jr., Ortner, P. B., Werner, F. E., Settle, L. R., and Cummings, S. R. 1999. Spawning habitat of the Atlantic menhaden in Onslow Bay, North Carolina. Fisheries Oceanography 8 (suppl. 2): 22–36. Chen, T. S. 1975. Preliminary study of the elver of Anguilla japonica in Taiwan. China Fisheries Monthly 268: 11–19. Cheng, P. W. and Tzeng, W. N. 1996. Timing of metamorphosis and estuarine arrival across the dispersal range of the Japanese eel Anguilla japonica. Marine Ecology Progress Series 131: 87–96. Chenoweth, S. B., Libby, D. A., Stephenson, R. L., and Power, M. J. 1989. Origin and dispersion of larval herring (Clupea harengus) in coastal waters of eastern Maine and southwestern New Brunswick. Canadian Journal of Fisheries and Aquatic Sciences 46: 624–632.

112

Jonathan A. Hare

Chesson, P. 1998. Recruitment limitation: a theoretical perspective. Australian Journal of Ecology 23: 234–240. Chiarella, L. A. and Conover, D. O. 1991. Spawning season and ﬁrst-year growth of adult blueﬁsh from the New York Bight. Transactions of the American Fisheries Society 119: 455–462. Churchill, J. H., Forward, R. B., Luettich, R. A., Hensch, J. L., Hettler, W. F., Crowder, L. B., and Blanton, J. O. 1999. Circulation and larval transport within a tidally dominated estuary. Fisheries Oceanography 8 (suppl. 2): 173–189. Clark, J. 1968. Seasonal movements of striped bass contingents of Long Island Sound and the New York Bight. Transactions of the American Fisheries Society 97: 320–343. Cooper, R. A. and Uzmann, J. R. 1971. Migration and growth of deep-sea lobsters, Homarus americanus. Science 171: 288–290. Cordes, J. F. and Graves, J. E. 2003. Investigation of congeneric hybridization in and stock structure of weakﬁsh (Cynoscion regalis) inferred from analyses of nuclear and mitochondrial DNA loci. Fishery Bulletin 101: 443–450. Cowen, R. K. 1991. Variation in the planktonic larval duration of the temperate wrasse Semicossyphus pulcher. Marine Ecology Progress Series 69: 9–15. Cowen, R. K., Hare, J. A., and Fahay, M. P. 1993. Beyond hydrography: can physical processes explain larval ﬁsh assemblages within the Middle Atlantic Bight? Bulletin of Marine Science 53: 567–587. Cowen, R. K., Lwiza, K. M. M., Sponaugle, S., Paris, C. B., and Olson, D. B. 2000. Connectivity of marine populations: open or closed? Science 287: 857–859. Crawford, M. K., Grimes, C. B., and Buroker, N. E. 1989. Stock identiﬁcation of weakﬁsh, Cynoscion regalis, in the Middle Atlantic region. Fishery Bulletin 87: 205–211. Daniels, L. B. and Graves, J. E. 1994. Morphometric and genetic identiﬁcation of eggs of springspawning sciaenids in lower Chesapeake Bay. Fishery Bulletin 92: 254–261. Doherty, P. J., Planes, S., and Mather, P. 1995. Gene ﬂow and larval duration in seven species of ﬁsh from the Great Barrier Reef. Ecology 76: 2373–2391. Dower, J. F., Miller, T. J., and Leggett, W. C. 1997. The role of microscale turbulence in the feeding ecology of larval ﬁsh. Advances in Marine Biology 13: 169–220. Dower, J. F., Pepin, P., and Leggett, W. C. 2002. Using patch studies to link mesoscale patterns of feeding and growth in larval ﬁsh to environmental variability. Fisheries Oceanography 11: 219–232. Doyle, M. J., Morse, W. W., and Kendall, A. W. 1993. A comparison of larval ﬁsh assemblages in the temperate zone of the northeast Paciﬁc and northwest Atlantic oceans. Bulletin of Marine Sciences 53: 588–644. Drouin, C.-A., Bourget, E., and Tremblay, R. 2002. Larval transport processes of barnacle larvae in the vicinity of the interface between two genetically different populations of Semibalanus balanoides. Marine Ecology Progress Series 229: 165–172. Edmonds, J. S. and Fletcher W. J. 1997. Stock discrimination of pilchards, Sardinops sagax, by stable isotope ratio analysis of otolith carbonate. Marine Ecology Progress Series 152: 241–247. Epifanio, C. E. and Garvine, R. W. 2001. Larval transport on the Atlantic continental shelf of North America: a review. Estuarine, Coastal and Shelf Science 52: 51–77. Etherington, L. L. and Eggleston, D. B. 2000. Large-scale blue crab recruitment: linking postlarval transport, post-settlement planktonic dispersal, and multiple nursery habitats. Marine Ecology Progress Series 204: 179–198. Fletcher, W. J. and Sumner, N. R. 1999. Spatial distribution of sardine (Sardinops sagax) eggs and larvae: an application of geostatistics and resampling to survey data. Canadian Journal of Fisheries and Aquatic Sciences 56: 907–914. Fogarty, M. J. 1998. Implications of migration and larval interchange in American lobster (Homarus americanus) stocks: spatial structure and resilience. Canadian Special Publication of Fisheries and Aquatic Sciences 125: 273–283.

The Use of Early Life Stages in Stock Identiﬁcation Studies

113

Fortier, L. and Gagné, J. A. 1990. Larval herring (Clupea harengus) dispersion, growth and survival in the St. Lawrence estuary: match/mismatch or membership/vagrancy? Canadian Journal of Fisheries and Aquatic Sciences 47: 1898–1912. Forward, R. B., Reinsel, K. A., Peters, D., Tankersley, R. A., Churchill, J. H., Crowder, L. B., Hettler, W. F., Warlen, S. M., and Green, M. 1999. Transport of ﬁsh larvae through a tidal inlet. Fisheries Oceanography 8 (suppl. 2): 153–172. Futuyma, D. J. 1986. Evolutionary Ecology. Sinauer, Sunderland, MA. 600 pp. Gaughan, D. J., Fletcher, W. J., and McKinlay, J. P. 2002. Functionally distinct adult assemblages within a single breeding stock of the sardine, Sardinops sagax: management units within a management unit. Fisheries Research 59: 217–231. Gaughan, D. J., White, K. V., and Fletcher, W. J. 2001. The links between functionally distinct adult assemblages of Sardinops sagax: larval advection across management boundaries. ICES Journal of Marine Science 58: 597–606. Graves, J. E., McDowell, J. R., and Jones, M. L. 1992. A genetic analysis of weakﬁsh Cynoscion regalis stock structure along the mid-Atlantic coast. Fishery Bulletin 90: 469–475. Grioche, A., Koubbi, P., Harlay, X., and Sautour, B. 2001. Sole larval distribution (Solea solea) in the eastern English Channel and Southern Bight of the North Sea. Journal of the Marine Biological Association of the United Kingdom 81: 673–678. Harden Jones, F. R. 1968. Fish Migration. St. Martin’s, New York. 325 pp. Harding, G. C. and Trites, R. W. 1988. Dispersal of Homarus americanus larvae in the Gulf of Maine from Brown Bank. Canadian Journal of Fisheries and Aquatic Sciences 45: 416–425. Harding, G. C., Kenchington, E. L., and Zheng, Z. 1993. Morphometrics of American lobster (Homarus americanus) larvae in relation to stock determinations in the Maritimes, Canada. Canadian Journal of Fisheries and Aquatic Sciences 50: 43–52. Harding, G. C., Kenchington, E. L., Bird, C. J., Pezzack, D. S., and Landry, D. C. 1997. Genetic relationships among subpopulations of the American lobster (Homarus americanus) as revealed by random ampliﬁed polymorphic DNA. Canadian Journal of Fisheries and Aquatic Sciences 54: 1762–1771. Hare, J. A. and Cowen, R. K. 1993. Ecological and evolutionary implications of the larval transport and reproductive strategy of blueﬁsh Pomatomus saltatrix. Marine Ecology Progress Series 98: 1–16. Hare, J. A. and Cowen, R. K. 1997. Size, growth, development and survival of the planktonic larvae of Pomatomus saltatrix (Pisces: Pomatomidae). Ecology 78: 2415–2431. Hare, J. A., Churchill, J. H., Cowen, R. K., Berger, T., Cornillon, P., Dragos, P., Glenn, S., Govoni, J. J., and Lee, T. N. 2002. Routes and rates of larval ﬁsh transport from the southeastern to the mid-Atlantic North American continental shelf. Limnology and Oceanography 47: 1774–1789. Hare, J. A., Fahay, M. P., and Cowen, R. K. 2001. Springtime ichthyoplankton of the slope region off the north-eastern United States of America: larval assemblages, relation to hydrography and implications for larval transport. Fisheries Oceanography 10: 164–192. Hare, J. A., Quinlan, J. A., Werner, F. E., Blanton, B. O., Govoni, J. J., Forward, R. B., Settle, L. R., and Hoss, D. E. 1999. Larval transport during winter in the SABRE study area: results of a coupled vertical larval behavior–three-dimensional circulation model. Fisheries Oceanography 8 (suppl. 2): 57–76. Heath, M. R. 1992. Field investigations of the ELS of marine ﬁsh. Advances in Marine Biology 28: 2–174. Helbig, J. A. and Pepin, P. 1998a. Partitioning the inﬂuence of physical processes on the estimation of ichthyoplankton mortality rates. 2. Application to simulated and ﬁeld data. Canadian Journal of Fisheries and Aquatic Sciences 55: 2206–2220.

114

Jonathan A. Hare

Helbig, J. A. and Pepin, P. 1998b. Partitioning the inﬂuence of physical processes on the estimation of ichthyoplankton mortality rates. 1. Theory. Canadian Journal of Fisheries and Aquatic Sciences 55: 2189–2205. Hermann, A. J., Hinckley, S., Megrey, B. A., and Stabeno, P. J. 1996. Interannual variability of the early life history of walleye Pollock near Shelikof Strait as inferred from a spatially explicit, individual based model. Fisheries Oceanography 5 (suppl. 1): 39–57. Hilborn, R. and Walters, C. J. 1992. Quantitative Fisheries Stock Assessment: Choice, Dynamics and Uncertainty. Chapman and Hall, New York. 570 pp. Hinckley, S., Hermann, A. J., Mier, K. L., and Megrey, B. A. 2001. Importance of spawning location and timing to successful transport to nursery areas: a simulation study of Gulf of Alaska walleye Pollock. ICES Journal of Marine Science 58: 1042–1052. Ihssen, P. E., Booke, H. E., Casselman, J. M., McGlade, J. M., Payne, N. R., and Utter, F. M. 1981. Stock identiﬁcation: materials and methods. Canadian Journal of Fisheries and Aquatic Sciences 38: 1838–1855. Iles, T. D. and Sinclair, M. 1982. Atlantic herring: stock discreteness and abundance. Science 215: 627–633. Incze, L. S. and Naimie, C. E. 2000. Modelling the transport of lobster (Homarus americanus) larvae and postlarvae in the Gulf of Maine. Fisheries Oceanography 9: 99–113. Ishikawa, S., Aoyama, J., Tsukamoto, K., and Nishida, M. 2001. Population structure of Japanese eel Anguilla japonica as examined by mitochondrial DNA sequencing. Fisheries Science 67: 246–253. James, M. K., Armsworth, P. R., Mason, L. B., and Bode, L. 2002. The structure of reef ﬁsh metapopulations: modeling larval dispersal and retention patterns. Proceedings of the Royal Society of London B 269: 2079–2086. Jamieson, G. S. and Phillips, A. 1993. Megalopal spatial distribution and stock separation in Dungeness crab (Cancer magister). Canadian Journal of Fisheries and Aquatic Sciences 50: 416–429. Kajihara, T. 1988. Distribution of Angullia japonica leptocephali in western Paciﬁc during September 1986. Nippon Suisan Gakkaishi 54: 929–933. Katz, C. H., Cobb J. S., and Spaulding, M. 1994. Larval behavior, hydrodynamic transport, and potential offshore-to-inshore recruitment in the American lobster Homarus americanus. Marine Ecology Progress Series 103: 265–273. Kendall, A. W. and Walford, L. A. 1979. Sources and distributions of blueﬁsh, Pomatomus saltatrix, larvae and juveniles off the east coast of the United States. Fishery Bulletin 77: 213–227. Kimura, S., Tsukamoto, K., and Sugimoto, T. 1994. A model for the larval migration of the Japanese eel: roles of the trade winds and salinity front. Marine Biology 119: 185–190. Kingsford, M. J., Leis, J. M., Shanks, A., Lindeman, K. C., Morgan, S. G., and Pineda, J. 2002. Sensory environments, larval abilities and local self-recruitment. Bulletin of Marine Science 70: 309–340. Kornﬁeld, I. and Bogdanowicz, S. M. 1987. Differentiation of mitochondrial DNA in Atlantic herring, Clupea harengus. Fishery Bulletin 85: 561–568. Kuo, H. 1971. Approach of eel elvers to the land in Taiwan. Aquaculture (Yoshoku) 1: 52–56. Kutkuhn, J. H. 1981. Stock deﬁnition as a necessary basis for cooperative management of Great Lakes ﬁsh resources. Canadian Journal of Fisheries and Aquatic Sciences 38: 1476–1478. Lacson, J. M. 1992. Minimal genetic variation among samples of six species of coral reef ﬁshes collected at La Parguera, Puerto Rico, and Discovery Bay, Jamaica. Marine Biology 112: 327–331. Laurence, G. C. 1985. A report on the development of stochastic models of food-limited growth and survival of cod and haddock larvae in Georges Bank. NOAA Technical Memorandum NMFS-F/NEC 36: 83–150. Leis, J. M. and Carson-Ewart, B. M. 1997. In situ swimming speeds of the late pelagic larvae of some Indo-Paciﬁc coral-reef ﬁshes. Marine Ecology Progress Series 159: 165–174.

The Use of Early Life Stages in Stock Identiﬁcation Studies

115

Lockhart, F. and Estrella, B. 1997. Amendent 3 to the interstate ﬁshery management plan for American lobster. Fishery Management Report No. 29, Atlantic States Marine Fisheries Commission, Washington, D.C. 44 pp. Luczkovich, J. J., Sprague, M. W., Johnson, S. E., and Pullinger, R. C. 1999. Delimiting spawning areas of weakﬁsh Cynoscion regalis (family Sciaenidae) in Pamlico Sound, North Carolina using passive hydroacoustic surveys. Bioacoustics 10: 143–160. MacCall, A. D. 1990. Dynamic geography of marine ﬁsh populations. Washington Sea Grant, Seattle, WA. 155 pp. Matsui, I. 1957. On the records of a leptocephalus and catadromous eels of Angullia japonica in the waters around Japan with a presumption of their spawning places. Journal of Simonoseki College of Fisheries 7: 151–167. McPherson, A. A., Stephenson, R. L., O’Reilly, P. T. O., Jones, M. W., and Taggart, C. T. 2001. Genetic diversity of coastal Northwest Atlantic herring populations: implications for management. Journal of Fish Biology 59 (suppl. A): 356–370. McQuinn, I. H. 1997. Metapopulations and the Atlantic herring. Reviews in Fish Biology and Fisheries 7: 297–329. Meekan, M. G. and Fortier, L. 1996. Selection for fast growth during the larval life of Atlantic cod Gadus morhua on the Scotian Shelf. Marine Ecology Progress Series 137: 25–37. National Marine Fisheries Service (NMFS). 1999. Ecosystem-based ﬁshery management, a report to Congress by the Ecosystems Principles Advisory Panel. U.S. Department of Commerce. Washington, D.C. 54 pp. Neilson, J. D., DeBlois, E. M., and Hurley, P. C. F. 1988. Stock structure of Scotia shelf ﬂatﬁsh as inferred from ichthyoplankton survey data and the geographic distribution of mature females. Canadian Journal of Fisheries and Aquatic Sciences 45: 1674–1685. Ozawa, T., Tabeta, O., and Mochioka, N. 1989. Anguillid leptocephali from the western north Paciﬁc east of Luzon, in 1988. Nippon Suisan Gakkaishi 55: 627–632. Parsons, K. E. 1996. The genetic effects of larval dispersal depend on spatial scale and habitat characteristics. Marine Biology 126: 403–414. Pepin, P., Dower, J. F., Helbig, J. A., and Leggett, W. C. 2002. Estimating the relative roles of dispersion and predation in generating regional differences in mortality rates of larval radiated shanny (Ulvaria subbifuracta). Canadian Journal of Fisheries and Aquatic Sciences 59: 105–114. Planes, S. 1993. Genetic differentiation in relation to restricted larval dispersal of the convict surgeonﬁsh Acanthurus triostegus in French Polynesia. Marine Ecology Progress Series 98: 237–246. Riginos, C. and Victor, B. C. 2001. Larval spatial distributions and other life-history characteristics predict genetic differentiation in eastern Paciﬁc blennioid ﬁshes. Proceedings of the Royal Society of London B 268: 1931–1936. Rowe, P. M. and Epifanio, C. E. 1994a. Tidal stream transport of weakﬁsh larvae in Delaware Bay, USA. Marine Ecology Progress Series 110: 105–114. Rowe, P. M. and Epifanio, C. E. 1994b. Flux and transport of larval weakﬁsh in Delaware Bay, USA. Marine Ecology Progress Series 110: 115–120. Ruzzante, D. E., Taggart, C. T., and Cook, D. 1999. A review of evidence for genetic structure of cod (Gadus morhua) populations in the NW Atlantic and population afﬁnities of larval cod off Newfoundland and the Gulf of St. Lawrence. Fisheries Research 43: 79–97. Scheltema, R. S. 1986. On dispersal and planktonic larvae of benthic invertebrates: An eclectic overview and summary of problems. Bulletin of Marine Science 39: 290–322. Searcy, S. and Sponaugle, S. 2000. Variable larval growth in a coral reef ﬁsh. Marine Ecology Progress Series 206: 213–226. Searcy, S. and Sponaugle, S. 2001. Selective mortality during the larval-juvenile transition in two coral reef ﬁshes. Ecology 82: 2452–2470.

116

Jonathan A. Hare

Secor, D. H. 1999. Specifying divergent migrations in the concept of stock: the contingent hypothesis. Fisheries Research 43: 13–34. Secor, D. H. 2004. Fish migration and the unit stock: three formative debates. In S. X. Cadrin, J. Waldman, and K. Friedland (eds.), Stock Identiﬁcation Methods. Academic Press, Burlington, MA, pp. 15–41. Shulman, M. J. and Bermingham, E. 1995. Early life histories, ocean currents, and the population genetics of Caribbean reef ﬁshes. Evolution 49: 897–910. Sinclair, M. 1988. Marine populations: an essay on population regulation and speciation. Washington Sea Grant Program, Seattle, Washington. 252 pp. Sinclair, M. and Iles, T. D. 1985. Atlantic herring (Clupea harengus) distributions in the Gulf of Maine–Scotian Shelf area in relation to oceanographic features. Canadian Journal of Fisheries and Aquatic Sciences 42: 880–887. Skogen, M. D., Monstad, T., and Svendsen, E. 1999. A possible separation between a northern and a southern stock of the northeast Atlantic blue whiting. Fisheries Research 41: 119–131. Smedbol, R. K. and Wroblewski, J. S. 2002. Metapopulation theory and northern cod population structure: interdependency of subpopulations in recovery of a groundﬁsh population. Fisheries Research 55: 161–174. Smedbol, R. K., McPherson, A., Hansen, M. M., and Kenchington, E. 2002. Myths and moderation in marine “metapopulations?” Fish and Fisheries 3: 20–35. Smith, P. J. and Jamieson, A. 1986. Stock discreteness in herrings: a conceptual revolution. Fisheries Research 4: 223–234. Smith, P. J., Robertson, S. G., Horn, P. L., Bull, B., Anderson O. F., Stanton B. R., and Oke, C. S. 2002. Multiple techniques for determining stock relationships between orange roughy, Hoplostethus atlanticus, ﬁsheries in the eastern Tasman Sea. Fisheries Research 58: 119–140. Smith, W. G. and Morse, W. W. 1993. Larval distribution patterns: early signals for the collapse/recovery of Atlantic herring Clupea harengus in the Georges Bank area. Fishery Bulletin 91: 338–347. Stephenson, R. L., Clark, K. J., Power, M. J., Fife, F. J., and Melvin, G. D. 2001. Herring stock structure, stock discreteness, and biodiversity. In Lowell Wakeﬁeld Fisheries Symposium Series No. 18. Alaska Sea Grant College Program, Fairbanks, Alaska, pp. 559–571. Stepien, C. A. 1999. Phylogeographical structure of the Dover sole Microstomus paciﬁcus: the larval retention hypothesis and genetic divergence along the deep continental slope of the northeastern Paciﬁc Ocean. Molecular Ecology 6: 923–939. Steves, B. P., Cowen, R. K., and Malchoff, M. H. 2000. Settlement and nursery habitats for demersal ﬁshes on the continental shelf of the New York Bight. Fishery Bulletin 98: 167–188. Stobutzki, I. and Bellwood, D. R. 1997. Sustained swimming abilities of the late pelagic stages of coral reef ﬁshes. Marine Ecology Progress Series 149: 35–41. Strathmann, R. R., Hughes, T. P., Kuris, A. M., Lindeman, K. C., Morgan, S. G., Pandolﬁ, J. M., and Warner, R. R. 2002. Evolution of local recruitment and its consequences for marine populations. Bulletin of Marine Science 70: 377–396. Swain, D. P., and Foote, C. J. 1999. Stocks and chameleons: the use of phenotypic variation in stock identiﬁcation. Fisheries Research 43: 113–128. Thorrold, S. R., Jones, G. P., Hellberg, M. E., Burton, R. S., Swearer, S. E., Neigel, J. E., Morgan, S. G., and Warner, R. R. 2002. Quantifying larval retention and connectivity in marine populations with artiﬁcial and natural markers. Bulletin of Marine Science 70: 291–308. Thorrold, S. R., Latkoczy, C., Swart, P. K., and Jones, C. M. 2001. Natal homing in a marine ﬁsh metapopulation. Science 291: 297–299. Townsend, D. W., Graham, J. J., and Stevenson, D. K. 1986. Dynamics of larval herring (Clupea harengus L.) production in tidally mixed waters of the eastern coastal Gulf of Maine. In M. J. Bowman, C. Yentsch, and W. T. Peterson (eds.), Tidal Mixing and Population Dynamics, Springer-Verlag, Berlin, pp. 253–277.

The Use of Early Life Stages in Stock Identiﬁcation Studies

117

Tracey, M. L., Nelson, K., Hedgecock, D., Shleser, R. A., and Pressick, M. L. 1975. Biochemical genetics of lobsters: genetic variation and the structure of American lobster (Homarus americanus). Journal of the Fisheries Research Board of Canada 32: 2091–2101. Tsukamoto, K. 1990. Recruitment mechanism of the eel, Anguilla japonica, to the Japanese coast. Journal of Fish Biology 36: 659–671. Tsukamoto, K. 1992. Discovery of the spawning area for Japanese eel. Nature 356: 789–791. Tsukamoto, K. and Aoyama, J. 1998. Evolution of freshwater eels of the genus Anguilla: a probable scenario. Environmental Biology of Fishes 52: 139–148. Tsukamoto, K., Aoyama, J., and Miller, M. J. 2002. Migration, speciation, and the evolution of diadromy in anguillid eels. Canadian Journal of Fisheries and Aquatic Sciences 59: 1989–1998. Tsukamoto, K., Umezawa, A., Tabeta, O., Mochioka, N., and Kajihara, T. 1989. Age and birth date of Angullia japonica leptocephali collected in western North Paciﬁc in September 1986. Nippon Suisan Gakkaishi 55: 1023–1028. Tzeng, W.-N. and Tsai, Y.-C. 1992. Otolith microstructure and daily age of Anguilla japonica, Temminck and Schlegel elvers from the estuaries of Taiwan with reference to unit stock and larval migration. Journal of Fish Biology 40: 845–857. Werner, F. E., Page, F. H., Lynch, D. R., Loder, J. W., Lough, R. G., Perry, R. I., Greenberg, D. A., and Sinclair, M. M. 1993. Inﬂuence of mean 3-D advection and simple behavior on the distribution of cod and haddock early life stages on Georges Bank. Fisheries Oceanography 2: 43–64. Werner, F. E., Quinlan, J. A., Blanton, B. O., and Luettich, R. A., Jr. 1997. The role of hydrodynamics in explaining variability in ﬁsh populations. Journal of Sea Research 37: 195–212. Werner, F. E., Quinlan, J. A., Lough, G., and Lynch, D. R. 2001. Spatially explicit individual based modeling of marine populations: a review of the advances in the 1990s. Sarsia 86: 411–421. Whitlock, M. C. and McCauley, D. E. 1999. Indirect measures of gene ﬂow and migration: FST1/(4Nm+1). Heredity 82: 117–125. Wiebe, P. H., and Benﬁeld, M. C. 2003. From Hensen net toward four-dimensional biological oceanography. Progress in Oceanography 56: 7–136. Wilk, S. J. 1974. Biological and ﬁsheries data on weakﬁsh, Cynoscion regalis (Bloch and Schneider). NOAA Technical Series Report/NMFS Northeast Fisheries Science Center. Yamada, S. B. 1989. Are direct developers more locally adapted than planktonic developers? Marine Biology 103: 403–411.

This Page Intentionally Left Blank

CHAPTER

6

Life History Parameters GAVIN A. BEGG CRC Reef Research Centre, James Cook University, Townsville, Queensland, Australia

I. Introduction II. Life History Parameters A. Distribution and Abundance B. Age, Growth, and Mortality C. Reproduction, Spawning, and Maturity D. Recruitment III. Temporal Stability IV. Spatial Stability V. Conclusions References

I. INTRODUCTION Identiﬁcation of ﬁsh stocks is necessary to ﬁsheries management for allocation of catch between competing ﬁsheries, recognition and protection of nursery and spawning areas, and for development of optimal harvest and monitoring strategies (Kutkuhn, 1981; Grimes et al., 1987; Smith et al., 1990; Begg et al., 1999a). A ﬁsh “stock” can be considered to be a group of randomly mating, reproductively isolated individuals with temporal or spatial integrity (Ihssen et al., 1981), which may exhibit differences in vital population or life history parameters compared to other stocks of the same species (Begg and Waldman, 1999). Differences in these life history parameters among groups of ﬁsh have long been used as a basis for the identiﬁcation of ﬁsh stocks because of the relative ease of assessing these parameters and their dual functionality as input into ﬁsheries stock assessment and management strategies. In this chapter, I review and critique the range of life history parameters that have typically been used in stock identiﬁcation studies and provide guidance for the effective protocol of the use of life history parameters in ﬁsh stock identiﬁcation. Life history parameters are the consequences of life history strategies to which ﬁsh stocks have evolved, reﬂecting the underlying dynamics of a ﬁsh stock, and Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

119

120

Gavin A. Begg

include intrinsic vital population parameters such as abundance, growth, survival, reproduction, maturity, and recruitment (Ihssen et al., 1981; Pawson and Jennings, 1996; Begg et al., 1999b). Estimates of these parameters are thought to be representative of individual ﬁsh within a putative stock, and can be used to distinguish among discrete stocks of ﬁsh because these parameters are phenotypic expressions of the interaction between genotypic and environmental inﬂuences. As a result, differences in life history parameters between groups of ﬁsh are assumed to be evidence that populations of ﬁsh are geographically and/or reproductively isolated (although mixing can occur seasonally), and therefore are discrete stock units for management purposes (Ihssen et al., 1981). Differences in life history parameters have long been used to identify separate stocks or as proxies by which ﬁsh stocks are initially recognized (e.g., Boyar, 1968; Casselman et al., 1981; Beacham, 1982; Rulifson and Dadswell, 1995; Grifﬁths, 1997; Begg and Sellin, 1998), and have provided a basis for population differentiation when genetic methodologies have failed (Utter, 1991; Pawson and Jennings, 1996). Life history parameters generally provide no deﬁnitive information about the genetic composition of a stock (Ihssen et al., 1981) and tend to characterize the environment occupied by a stock because of their sensitivity to extrinsic variables (Beacham, 1982). As a result, the long-term utility of these parameters has been questioned owing to their assumed plasticity in response to short-term environmental variation (Pawson and Jennings, 1996), although few studies have actually examined the temporal stability of life history parameters, nor the mechanisms and spatial stability by which differences in these parameters between stocks are maintained. However, if members of a stock respond in a similar manner to environmental perturbations, such a response itself can be viewed as a stock attribute (Casselman et al., 1981; Ihssen et al., 1981). In this chapter, I detail life history parameters that have typically been used as indicators of stock structure, including abundance, growth, mortality, reproduction, maturity, and recruitment, with an emphasis on recent advances. I review benchmark case studies and critique strengths and weaknesses of using each parameter for stock identiﬁcation. I discuss the temporal and spatial stability of these parameters to determine if differences between stocks persist over generations and review the mechanisms by which individual stock structures and their corresponding life history parameters are maintained. Illustrative examples for estimating and comparing life history parameters among putative stocks of ﬁsh are provided and used as templates for future use of life history parameters in ﬁsh stock identiﬁcation.

II. LIFE HISTORY PARAMETERS Typically, data on vital population or life history parameters are collected as part of routine surveys to provide baseline information on population dynamics and

121

Life History Parameters

productivity rates needed for stock assessments, with no direct purpose for stock identiﬁcation. However, depending on the spatial resolution provided by the surveys, often these data can provide an initial indication of stock boundaries.

A. DISTRIBUTION

AND

ABUNDANCE

Distribution and abundance data are often, and where available should be, the ﬁrst point of reference for identifying ﬁsh stocks. These life history parameters provide fundamental information that enables delineation of geographic regions that may be representative of individual stocks (Pawson and Jennings, 1996; Begg and Waldman, 1999; Begg and Marteinsdottir, 2002a). Fisheries-dependent and -independent landings and catch-per-unit-effort (CPUE) data are often routinely collected to assess the status of ﬁsheries, and can be used to estimate relative indices of abundance that may assist in the initial recognition and delineation of stock boundaries (Pawson and Jennings, 1996). Abundance indices can be estimated for all life history stages including eggs, larvae, juveniles, and spawning adults, detailing important natal spawning and nursery grounds that may indicate geographic gaps in population distribution, reﬂective of reproductive isolation and stock separation (Fig. 6-1). Recently, spatial statistics (i.e., kriging), Generalized Additive Models (GAMs), and Geographic Information Systems (GIS) have all been used to examine relationships between ﬁsheries distribution and abundance data (e.g., Swartzman et al., 1992; Petitgas, 1993; Maravelias and Reid, 1997; Fletcher and Sumner, 1999; Eastwood et al., 2001; Stoner et al., 2001; Begg and Marteinsdottir, 2002a; Overholtz, 2002), albeit with no direct purpose for stock identiﬁcation (Fig. 6-2). However, these contemporary approaches to georeferenced data and spatial statistics are highly applicable and innovative techniques that should be examined in any stock identiﬁcation study where these types of data are available. Although spatial distribution of catch abundance or simple presence/absence ratios are standard data collected from research surveys or commercial ﬁsheries and provide the most basic guide to stock boundaries, the use of distribution data is not discussed in depth in this chapter, but instead I refer the reader to the preceding chapter (this volume). Fisheries-dependent (i.e., commercial) landings, catch and effort-based abundance data provide an indication of distribution and movement patterns that most likely could not be obtained on such an extensive basis by any other means (Pawson and Jennings, 1996). Often, seasonal progression of landings from a commercial ﬁshery have been used to follow the migration routes of ﬁsh, providing an initial indication of home range and stock boundaries, albeit at a rather coarse level (Pawson and Jennings, 1996; Begg et al., 1997) (Fig. 6-3). Similarly, numerous studies have examined the degree of synchrony in ﬁsheries-dependent

Egg distributions

Larval distributions

Adult and juvenile distributions

FIGURE 6-1. Distribution and abundance patterns of haddock (Melanogrammus aegleﬁnus) eggs, larvae, juveniles, and adults in the northwest Atlantic Ocean from spring (March to May) research survey data (1977–1988), demonstrating potential spawning and nursery grounds that indicate geographic junctions in population distribution and stock separation (Begg et al., 1999b). Eggs and larvae data were obtained from the Marine Resources Monitoring, Assessment and Prediction Program (MARMAP) (Morse et al., 1987), and juvenile and adult data from the Northeast Fisheries Science Center (NEFSC) spring bottom trawl surveys (Azarovitz, 1981). 122

0.0

63

-3.0

64

-2.0

65

-1.0

66

67

1985-1999

-26

-24

-22

-20

-18

-14

-16

-12 -10

-20

-25

-10

-15

-2.0 -4.0

-1.0 -2.0

100

0

200

300

400

500

0

20

40

60

80

100

120

140

Total length (cm)

-2.0

-2.0

0.0

0.0

2.0

2.0

Bottom depth (m)

-4.0

-6.0

-4.0

Effect on spawning cod distribution

0.0

0.0

1.0

Longitude W

-2

0

2

4

6

8

0

10

10

5

o

15

Age (years)

-0.6

-4.0

-0.2

-2.0

0.2

0.0

0.6

Bottom temperature ( C)

63

64

65

Latitude N

66

67

1986

1988

1990

1992

1994

1996

1998

Year

FIGURE 6-2. Generalized Additive Model (GAM) results that show spatial relationships with ﬁsheries distribution data which can be used to identify putative stock boundaries. The GAM results demonstrate the effects of various covariates (bottom depth, temperature, latitude, longitude, length, age, and year) on the spatial distribution (presence/absence) of spawning Atlantic cod (Gadus morhua) in Icelandic waters (1985–1999). The y-axes are scaled to zero and reﬂect the relative importance of the respective covariate; rugplot on the x-axes represent the number of observations; dashed lines are the 95% conﬁdence intervals on the smooths generated from bootstrap resampling of the original data; and (o) represent locations of spawning cod (Begg and Marteinsdottir, 2002a).

123

Cairns (n=14) 100 80

Townsville (n=86) 16 12

60

8

40

4

20

100 80

Frequency (%)

Rockhampton (n=360) 12 8 4

20

4

20

Moreton Bay (n=163) 12 8 4

20

8

40

J F MAM J JASON D

16

40

12

60

Hervey Bay (n=1559)

60

16

100 80

J F MAM J JASON D

100 80

4

20 J F MAM J JASON D

16

40

8

40

Mackay (n=191)

60

12

60

J F MAM J JASON D

100 80

16

Harvest (t)

A

100 80

16 12

60

8

40

4

20

J F MAM J JASON D

J F MAM J JASON D

Month

Month

B

FIGURE 6-3. Seasonal progression of spotted mackerel (Scomberomorus munroi) commercial landings and tag-recapture data to discern migration route of putative stock, which provides an indication of home range and stock boundaries in Queensland, Australian waters. (A) Monthly distribution of numbers tagged (hatched bars) and mean commercial harvest (open squares) by region; and (B) movements (>100 km) from tag-recapture data (Begg et al., 1997). 124

Life History Parameters

125

FIGURE 6-4. Fisheries-dependent annual commercial catch-per-unit-effort (CPUE) and mean catch of silver kob (Argyrosomus inodorus) for line ﬁshers operating along the South African coastline. CPUE and catch data were used to indicate three modal groups in abundance that were indicative of separate stocks (Grifﬁths, 1997).

catch statistics to provide information on the association and geographic continuity of adjacent groups of ﬁsh (Casselman et al., 1981; Ihssen et al., 1981; Campbell and Mohn, 1983; Grifﬁths, 1996, 1997; Begg, 1998a). For example, Grifﬁths (1997) used CPUE data to indicate that there were three modal groups in abundance of silver kob (Argyrosomus inodorus) along the South African coastline that were indicative of separate stocks (Fig. 6-4). Moreover, covariability between stock abundances at geographically dispersed locations is a valuable source of information regarding the dynamic structure and physical forcing of spatially distributed populations (Botsford and Paulsen, 2000). Although there are obvious advantages with using ﬁsheries-dependent data compared to expensive and often logistically unfeasible ﬁsheries-independent research survey data, there are also potentially severe limitations with using such data that need to be considered. One of the main limitations with using ﬁsheries-dependent data is the inherent potential for the samples to be biased or unrepresentative of the stock owing to the selective nature of the ﬁshing gear, handling and discard practices, and/or market demands (Hilborn and Walters, 1992). In addition, stocks can be hyperstable, migrate on a temporal and spatial basis, exhibit density-dependent effects

126

Gavin A. Begg

such as changing distribution with abundance, and form age-, size-, or sexsegregated groups (Swain and Wade, 1993; Jennings et al., 2001). Distribution and abundance parameters derived from ﬁsheries-dependent data are innately biased because these typically reﬂect only where high-density ﬁshable aggregations exist. In contrast, ﬁsheries-independent research surveys are designed to avoid these biases by sampling the approximate range of a species distribution. Abundance parameters derived from such data are generally estimated via a timeor area-speciﬁc CPUE measure, which is assumed to be proportional to stock size. Several methods have been established for collecting ﬁsheries-independent research survey data to estimate distribution and abundance parameters, although the most common are based on stratiﬁed-random or ﬁxed survey sampling designs (Azarovitz, 1981; Clark, 1981; Pálsson et al., 1989; Anonymous, 1992). The effects of these biases, however, may not be so pronounced or as important when the objective is to use distribution and abundance data for stock identiﬁcation. Stock identiﬁcation is rarely the main aim of distribution and abundance studies, and many research surveys and commercial ﬁsheries collections are too irregular or geographically imprecise to provide good evidence of stock separation ( Jennings et al., 2001). However, ﬁsheries-dependent data are particularly important in data-sparse ﬁsheries where there is often little ﬁscal or logistic support for extensive ﬁsheries-independent research surveys and in many situations may be the only data available. Distribution and abundance data, where available, should be the ﬁrst data examined in any stock identiﬁcation study to assist in reﬁning more speciﬁc questions concerning spatial detail of stock structure. Following on from such initial investigations and collation of baseline data, similarly readily available data for estimation of other life history parameters such as age, growth, and reproduction should be examined to corroborate and reﬁne preliminary interpretations of stock structure derived from distribution and abundance data.

B. AGE, GROWTH,

AND

MORTALITY

Age, growth, and mortality characteristics are the most frequently used life history parameters to identify putative ﬁsh stocks. Strong geographic differences in age or size composition, if not reﬂective of ﬁshing gear differences and other factors (i.e., sampling biases), suggest independence of recruitment or other biological or ﬁshery factors as a basis for assuming discrete stocks (Begg and Waldman, 1999). Like other life history parameters, age- and size-based parameters are strongly inﬂuenced by environmental factors, although differentiating the effects of these factors from exploitation is inherently difﬁcult (Serchuk et al., 1994). Several derivatives of age and size can be used to describe the dynamics of a stock, which in turn can be used as a means for stock identiﬁcation, including

Life History Parameters

127

age and size distributions (Boyar, 1968; Casselman et al., 1981; Hanchet, 1999). maximum age, length, and weight (or a percentile thereof) (Begg et al., 1999b; Fromentin and Fonteneau, 2001), modal size and age class (Hanchet, 1999); natural or instantaneous mortality (Begg et al., 1999b; Horn and Hurst, 1999; Williams et al., 2003), length–weight relationships ( Japp, 1990; Lowe et al., 1998), and commonly used growth curves and their associated parameters (Grifﬁths, 1996; DeVries and Grimes, 1997). Age- and size-based life history parameters have received the greatest attention in stock identiﬁcation studies because they are also the vital parameters used in calculations of yield and productivity, and, as such, provide the basis for stock assessment and management (Casselman et al., 1981). Furthermore, the resultant yield dynamics of individual stocks may differ with respect to their biological productivity as a function of age- and size-based life history parameters and the relevant impacts of exploitation (Pawson and Jennings, 1996), thereby providing an additional basis for stock differentiation, while satisfying a fundamental criterion of ﬁsheries management. Individual life history parameters of longevity, senescence, mortality, growth, and year-class strength can be described parametrically and provide a basis for stock differentiation, although their plasticity in response to short-term environmental variation through density-dependent control mechanisms and selective effects of ﬁshing have been suggested to reduce their value (Dizon et al., 1992; Pawson and Jennings, 1996). In phenotypic-based studies, it is essential to consider any confounding variation or sampling biases that may be present owing to selective effects or differences in ﬁshing gears and/or differences between samples in age group, year-class, or sex ratio, so as to not mistake stock differences for sample differences (Castonguay et al., 1991; Begg and Brown, 2000; Bolles and Begg, 2000). Parameter estimates will be biased if samples used in any comparison of stock dynamics are unrepresentative of the stock in question or if results cannot be corrected to account for sampling bias (Jennings et al., 2001). Failure to account for such extraneous inﬂuences may result in falsely attributing differences between stocks to a stock effect, whereas differences may in fact simply be reﬂective of sample variation (Begg and Waldman, 1999; Begg and Brown, 2000). Differential sex-speciﬁc growth is a common characteristic among ﬁsh stocks and is a typical factor that needs to be accounted for in any interstock comparisons (DeVries and Grimes, 1997; Begg and Sellin, 1998; Hanchet, 1999). Likewise, aggregating data across sampling years because of insufﬁcient sample numbers may bias the resulting parameter estimates, although some may question the general applicability of the ﬁndings if sampling is restricted to a single year (Simard et al., 1992). Prior to analysis, therefore, samples should initially be stratiﬁed according to sampling year, age group, sex, and so forth, or standardized with respect to ﬁsh length or age, depending on the particular variable under scrutiny, to minimize such biases and potential errors. Appropriate statistical analyses can then be conducted to determine if there is a need to account for any

128

Gavin A. Begg

sampling biases before conclusions regarding stock status are made. In addition to these analytical and sampling errors or biases that need to be considered, there are also the inherent process errors associated with measurement and age interpretation. Although development and validation of reliable aging techniques is an evolving process designed to reduce these errors and permit more effective interpretation of age and growth information (Pawson and Jennings, 1996), I do not discuss this further, but direct readers to several excellent sources that review these issues in more detail (see Beamish and McFarlane, 1983; Summerfelt and Hall, 1987; Campana, 2001). Comparison of growth curves between putative ﬁsh stocks is the most commonly used approach in life history–based stock identiﬁcation studies. A plethora of studies have examined differences in population or individual growth trajectories, at both the daily and annual level, as a basis for differentiating stocks (e.g., Japp, 1990; Simard et al., 1992; Grifﬁths, 1996; DeVries and Grimes, 1997; Begg and Sellin, 1998). Several growth models have been ﬁtted to observed and backcalculated length-at-age data including Schnute, Richards, Gompertz, and logistic (Schnute, 1981; Haddon, 2001), although the von Bertalanffy (1938) growth model is the most commonly used in ﬁsheries science since its ﬁrst application by Beverton and Holt (1957): L t = L • [1 - e - K ( t -t 0 ) ] where

Lt L• K t0

= = = =

length at age t; mean asymptotic length; rate at which L• is approached; and age at which ﬁsh have a theoretical length of zero.

The von Bertalanffy growth model is often ﬁtted to length-at-age data using nonlinear least squares methods and frequently due to ﬁshing selectivity and size limits, when there are few data available for younger and older ﬁsh (Fig. 6-5). This typical lack of available data at the extremes of the distributions invariably results in the estimation of L• and t0 by means of extrapolation, where care must be taken in the interpretation of the resultant growth parameters and comparisons with other putative stocks (Haddon, 2001). Back-calculation methods have attempted to minimize these sampling biases by estimating lengths at ages of younger ﬁsh that are rarely observed in ﬁshery-dependent samples (Campana, 2001). Nonetheless, the nonlinearity that depicts most growth curves has led to major difﬁculties in comparing these curves, which, in turn, has generated an expanse of literature detailing how best to address these difﬁculties (e.g., Kimura, 1980; Misra, 1980; Bernard, 1981). Likelihood ratio, t-, univariate (Chi)2-, and Hotelling’s T2-tests have all been used to compare von Bertalanffy–derived growth parameters among stocks, with the accuracy of each dependent on the functionality of the growth model, sample size, and the degree of heterogeneity in the

A 900

LCF (mm)

800 700 600 500

Northern Central Southern

400 300 0

1

2

3

4

6

5

7

8

9

10

Age class (years) B 490

Mackay

L• (mm)

480 470 460 450

Townsville

Storm Cay

440 0.34

0.39

0.44

0.49

0.54

C

75

Mean fork length (cm)

K (years-1)

60 45 Div. 4TVW Div. 4X Div. 5Y Div. 5Z

30 15

0

2

4

6 10 8 Age (years)

12

14

FIGURE 6-5. Comparison of age and growth data as a basis for differentiating putative ﬁsh stocks: (A) von Bertalanffy growth curves ﬁtted to length to caudal fork (LCF) at age data for female school mackerel (Scomberomorus queenslandicus) sampled in three regions throughout Queensland, Australian waters (Begg and Sellin, 1998); (B) 95% conﬁdence ellipsoids for the von Bertalanffy growth model parameters, K and L•, for a large coral reef ﬁsh (Lethrinus miniatus) sampled from three regions of the Great Barrier Reef, Australia (Williams et al., 2003); and (C) mean lengths at age among four haddock (Melanogrammus aegleﬁnus) stocks in the northwest Atlantic Ocean (Begg, 1999a). 129

130

Gavin A. Begg

error variances (Cerrato, 1990). In an empirical comparison of these tests, Cerrato (1990) found that the likelihood ratio test was the most accurate and advocated that it be the test of choice for growth curve comparisons. Kimura (1980) ﬁrst introduced the likelihood ratio test as a means of comparing von Bertalanffy growth curves among populations and also demonstrated practical approaches for constructing conﬁdence ellipsoids around parameter estimates that could be used to identify differences in stock dynamics (Fig. 6-5). Although a common occurrence, growth models ﬁtted to ﬁshery-dependent data can have many biases and the resultant growth parameters may not be representative of the actual stock in question due to size-selective ﬁshing where larger, faster-growing individuals are differentially removed from the stock (Kimura, 1980; Haddon, 2001). Parameters of growth models that do not adequately ﬁt observed data are also likely to produce erroneous results when used in subsequent analyses and should be used with caution (Grifﬁths, 1996). In addition, von Bertalanffy and other growth model parameters are correlated, making univariate tests inappropriate for comparing differences between like parameters from putative stocks (Bernard, 1981; DeVries and Grimes, 1997). An alternative approach to avoid these biases, and more importantly, the difﬁculties associated with comparing nonlinear growth curves and extrapolating parameter estimates beyond the range of the data is to statistically compare mean lengths at age among putative stocks across a common age range. Instead of applying growth models to observed length at age data, age-length keys can be derived where mean lengths at a given age or the proportional distribution of numbers at different sizes for given ages are estimated with an associated variance that can be compared in an analysis of variance or Generalized Linear Model (GLM) framework (Haddon, 2001) (Fig. 6-5). Gear selectivity problems, differences in sampling times, and the general lack of homogeneity in samples, however, may still confound results and need to be considered in any stock comparisons (Ihssen et al., 1981; Begg and Sellin, 1998). Consistent differences in age- and size-based life history parameters such as growth and mortality rates, therefore, have frequently been used to separate stocks (Begg and Waldman, 1999). The degree of asynchrony in population statistics of these parameters provides useful information on the disparity and phenotypic separation of adjacent stocks that is beneﬁcial to ﬁsheries management (Casselman et al., 1981). Signiﬁcant differences in population statistics between ﬁsh stocks can be accepted as evidence that different environments, and hence, different territories are occupied throughout the life history of the ﬁsh (Ihssen et al., 1981), but cannot be used to indicate whether stocks of these ﬁsh are genetically discrete (Pawson and Jennings, 1996). Discreteness determined in this manner may not be complete and, although often assumed, reproductive isolation is not necessarily manifest for stocks so identiﬁed (Ihssen et al., 1981). More deﬁnitive life history parameters related to spawning, however, may provide

131

Life History Parameters

a greater foundation for the assumption of reproductive isolation and, hence, genotypic separation of ﬁsh stocks.

C. REPRODUCTION, SPAWNING,

AND

MATURITY

Reproductive life history parameters provide fundamental information to assist in understanding biological processes that may be responsible for maintaining the underlying stock structure of a species (Begg, 1998b). The underpinning of the biological deﬁnition of a ﬁsh stock is that they are self-reproducing or reproductively isolated units, with members of each putative stock exhibiting similar life history characteristics (Hjort, 1914; Ihssen et al., 1981; Hilborn and Walters, 1992). This deﬁnition depends directly on our knowledge of spawning behavior and other reproductive parameters that are necessary for the formation and maintenance of stock structure. Individual stocks can develop phenotypic and genotypic differences in these parameters over time due to reproductive isolation (Waldman et al., 1988), which arise from diverse environmental conditions, differential selection pressure, and evolutionary divergence through drift and local adaptation (Dizon et al., 1992; Adkinson, 1995; Waldman, 1999). Numerous reproductive life history parameters have been used to describe the dynamics of a ﬁsh stock and provide the basis for stock differentiation, including timing, duration, and location of spawning (Finucane et al., 1986; Hutchings et al., 1993; Begg, 1998b); egg and larval distributions (O’Boyle et al., 1984; Begg et al., 1999b; Bruce et al., 2001); median or mean age, length, and weight at maturity (Beacham, 1982; O’Brien, 1990; Grifﬁths, 1997; Trippel et al., 1997); egg weight, size, viability, and fecundity relationships (Bradford and Stephenson, 1992; Marteinsdottir et al., 2000a); proportion of recruit and repeat spawners (Rochet, 2000); and, more recently, maternal effects and reproductive potential (Trippel, 1999; Marteinsdottir and Begg, 2002). Knowledge of the temporal and spatial extent of spawning can provide information on intraspeciﬁc variation in life history parameters that can be used to discriminate separate stocks (Schaefer, 1987). Differences in timing and location of spawning provide a particularly valuable criterion because they can result in reproductive isolation among stocks by restricting gene ﬂow to a level that effectively isolates stock units (Iles and Sinclair, 1982; Dizon et al., 1992; Bailey et al., 1999). Reproductive isolation among stocks is necessary for the formation and maintenance of stock integrity which can be derived from concurrent spawning of stocks in geographically dispersed locations (Horrall, 1981). Several genetically distinct stocks of herring (Clupea harengus) in the northwest Atlantic Ocean, for example, have been determined by the number of geographically stable spawning and larval retention areas, where the stocks spawn in relatively discrete geographic locations (Iles and Sinclair, 1982; Stephenson, 1991).

132

Gavin A. Begg

Likewise, separate stocks of king mackerel (Scomberomorus cavalla) in the United States have been postulated because of disjunct spawning in locations along the Atlantic seaboard and in the Gulf of Mexico (Finucane et al., 1986). Numerous other studies have also implied stock discreteness for a range of species because of differential spawning times and locations of adult spawning ﬁsh (e.g., Hutchings et al., 1993; Sinclair and Tremblay, 1984; Page and Frank, 1989). Differences in spawning and hatch-day distributions of eggs and larvae have also been used to demonstrate differential stock or spawning components (Begg and Marteinsdottir, 2000, 2002b; Marteinsdottir et al., 2000b; Bruce et al., 2001; Gaughan et al., 2001) (Fig. 6-6). Discrete larval distributions linked to particular geographic regions or hydrological features provide a mechanism for stock structure, imprinting and spawning site ﬁdelity (O’Boyle et al., 1984; Stephenson, 1991; Begg, 1998a). Egg and larval surveys frequently provide information which assists with stock identiﬁcation because stock integrity depends on spawning ﬁsh from different stocks being separated in space or time, even if they mix at other stages of their life history (Pawson and Jennings, 1996). Eggs identiﬁed from such surveys provide a direct or immediate response to spawning, while larvae provide an indication of movement to nursery grounds. Prediction of larval movements using oceanographic models can further indicate the extent to which progeny from different spawning stocks are dispersed and separated, and, in turn, assist the identiﬁcation of stock-speciﬁc spawning locations which provide recruitment to speciﬁc nursery grounds (Pawson and Jennings, 1996). Collection and analysis of life history data on spawning adult ﬁsh and their progeny should be a priority of any stock identiﬁcation study because of the direct relationship to reproductive isolation and stock discreteness. Individual ﬁsh should be sampled from putative stocks during their respective spawning season to maximize stock discreteness which may otherwise be obscured by spatial overlap and stock mixing during other times of the year (Casselman et al., 1981). Spawning (i.e., ripe and running) ﬁsh collected from assumed spawning locations would overcome the potential problem of stock mixing (Stephenson, 1991) and reﬁne spawning periodicity. Likewise, given the imminent release of eggs once hydration occurs, the collection location of spawning females with hydrated eggs should closely approximate spawning locations (Hutchings et al., 1993). As discussed previously, spatial distribution plots and georeferenced statistical techniques could be applied quite readily to presence/absence ratios or abundance indices of mature or spawning individuals, eggs, and/or larvae to deﬁne stock boundaries. In addition, data collected on individual ﬁsh during the spawning season could be used to estimate age and length at maturity and other reproductive parameters that have proved useful as indicators of stock status (Fig. 67). Similar sampling and analytical issues, as discussed previously, will be encountered with the use of reproductive-based life history parameters and need to be examined accordingly depending on the parameter in question. For

A 140 300

1

5

120

400

Frequency

2

6

25

250

3

600

7

200

4

50

22 Mar

19 Feb

100

3 May

10 Apr

200

150 30 May

8

50

19 Jul

22 Mar

19 Feb

100

3 May

10 Apr

Spawning day (DOY)

150 30 May

200 19 Jul

Spawning day (DOY)

B 68°

3

4

0.87

0.83

2

Latitude N

8

5 0.77

0.97 *

66° 0.99 0.74

6

1

64°

0.63

0.99

7

62° 36°

32°

28°

24°

20°

16°

12°

8°

Longitude W

FIGURE 6-6. Spawning day distributions of pelagic juvenile (0-group) Atlantic cod (Gadus morhua) used to demonstrate differential stock components in Icelandic waters: A) spawning day distributions of 0-group cod sampled in eight offshore regions (1970–1998 data combined). Dashed vertical lines: mean spawning day and 99th and 95th percentiles for main spawning grounds (Region 1); and (B) proportion of 0-group cod in each region predicted to have originated from main spawning grounds. Proportions are based on number of cod spawned at a later DOY than that estimated for the 99th percentile. Results demonstrate that large numbers of the surviving juvenile population may originate from other stocks besides that of the main spawning stock in the south (Begg and Marteinsdottir, 2000b).

133

1.0 0.8 0.4

0.6

1989 South n=1058 a=-5.6699 b= 0.0837 L50=67.7

0.0

0.2

North n=3077 a=-8.2215 b= 0.1014 L50=81.1

0.8

40

60

80

100

120 140

1994 South n=1200 a=-6.3391 b= 0.0988 L50=64.2

0.4

0.6

20

0.2

North n=1959 a=-7.2835 b= 0.1046 L50=69.6

0.0

Proportion mature

1.0

0

1.0

0

20

40

60

80

100

120

140

0.8

1999

0.4

0.6

South n=1414 a=-7.0447 b= 0.1100 L50=64.0

0.0

0.2

North n=2684 a=-7.4881 b= 0.1018 L50=73.6

0

20

40

60

80

100

120 140

Total length (cm) FIGURE 6-7. Logistic regression models ﬁtted to length at maturity data for Atlantic cod (Gadus morhua) collected in waters of northern and southern Iceland from spring (March) Marine Research Institute groundﬁsh surveys (1989–1999). The models were used to estimate the length at 50% maturity (L50), which has proved useful as an indicator of stock status. n, sample size; a, intercept term; and b, regression coefﬁcient (Begg and Marteinsdottir, 2002a). 134

Life History Parameters

135

example, estimates of age and length at maturity need to account for sampling biases, as discussed for growth and mortality, while analytical issues associated with georeferenced data will need to be considered for distribution and abundance studies, irrespective of the life history stage investigated. Reproductive life history parameters are extremely useful in discerning the underlying biological characteristics that shape stock structure. These parameters provide insight into the isolating mechanisms that are responsible for the maintenance of stock integrity and are elementary inputs for deﬁning the productivity and discreteness of a ﬁsh stock. Inexplicably linked with reproductive life history parameters are those related to recruitment and critical early life history stages, which are postulated to be the principal determinants of year-class strength (Cushing, 1969; Campana et al., 1989; Mertz and Myers, 1994).

D. RECRUITMENT Relative indices of recruitment (i.e., the number of ﬁsh that have attained the age at which they are vulnerable to ﬁshing) and abundance or biomass of early life history stages in putative ﬁsh stocks can provide information on year-class strength and stock resilience, as well as stock relatedness. Information on the origins of early life history stages is needed to understand stock structure and the mechanisms responsible for recruitment variability (Marteinsdottir et al., 2000a). Similar to other life history parameters, intraspeciﬁc geographic differences in recruitment provide an indirect basis for stock separation (Begg and Waldman, 1999). Recruitment and growth are the major contributors to the productivity and long-term sustainability of a stock (Haddon, 2001). The productivity of individual stocks and their relative contribution to a ﬁshery shift with time due to stockspeciﬁc differences in recruitment and exploitation rates (Waldman and Fabrizio, 1994). Recruitment success is most likely dependent on favorable environmental and stock effects that inﬂuence the dispersal and survival of early life history stages (Begg and Marteinsdottir, 2002b). Recruitment to ﬁsh stocks is naturally highly variable, and the mechanisms responsible are often poorly known. However, differentiating environmental and stock effects on early life history stages and the relationship between spawning stock biomass and recruitment are important steps in understanding the factors that govern recruitment dynamics and stock structure (Ricker, 1954; Beverton and Holt, 1957; Myers et al., 1995). Similar to other life history parameters, asynchronous ﬂuctuations in recruitment over time and among geographic regions provide circumstantial evidence for the existence of separate stocks (Waldman et al., 1988). Conversely, interrelationships may exist between stocks when the same year-classes demonstrate similar patterns in recruitment strength (Clark et al., 1982; Koslow et al., 1987;

136

Gavin A. Begg A

B

FIGURE 6-8. Time series of recruitment indices lagged back to the year of spawning for: (A) ﬁve Atlantic cod (Gadus morhua) and (B) four haddock (Melanogrammus aegleﬁnus) stocks in the northwest Atlantic Ocean. Asynchronous ﬂuctuations in recruitment over time and among geographic regions provide circumstantial evidence for the existence of separate stocks (Koslow et al., 1987).

Thompson and Page, 1989) (Fig. 6-8). The occurrence of similar year-class strength in walleye pollock (Theragra chalcogramma) throughout different regions of the Bering Sea is one example where uniformity in recruitment patterns was cited as evidence of stock panmixia (Dawson, 1994). Several studies

Life History Parameters

137

have suggested that large-scale physical and biological forcing are partly responsible for synchrony in recruitment and year-class strength among other gadoid stocks in the northwest Atlantic Ocean (Koslow, 1984; Koslow et al., 1987), although other studies have indicated that local-scale processes are more inﬂuential (Cohen et al., 1991). This disparity highlights one of the major issues in ﬁsheries science, that being the difﬁculty in deciphering the mechanisms responsible for year-class strength and recruitment variability. However, if recruitment is determined primarily by biological interactions, stronger correlations should be found in recruitment among stocks co-occurring within the same geographic region than among stocks with disjunct distributions (Koslow, 1984). Putative ﬁsh stocks have usually been identiﬁed via coarse observations that reveal a hiatus in some aspect of their life history such as large intraannual differences in recruitment and year-class strength (Waldman, 1999). Evidence of stock structure has been provided by spawning and recruitment patterns in relation to localized oceanographic conditions within each region where members of a stock reside (Begg, 1998a). Recruitment indices, as with other life history parameters, are useful indicators of phenotypic stock differences, but typically do not enable classiﬁcation of individual ﬁsh to a speciﬁc stock owing to the wide variability that naturally occurs within these parameters between individual ﬁsh (Waldman et al., 1988). Life history parameters also vary temporally within stocks as well as spatially between stocks, bringing into question the long-term stability of these parameters as indicators of stock structure (Ihssen et al., 1981; Pawson and Jennings, 1996; Begg et al., 1999b). In addition, some parameters may be less invariant over time, further questioning their utility as long-term indicators of stock status, although ﬁsheries management is innately short-term and as such, information derived from life history parameters may be more useful, depending on the management objective to be addressed.

III. TEMPORAL STABILITY Fish stocks are dynamic, interactive groups characterized by spatial boundaries and associated vital life history parameters that vary with time (Cowen et al., 1993). Numerous studies have demonstrated temporal variation in a range of life history parameters that have occurred in response to changing environmental conditions and/or exploitation patterns (e.g., Beacham, 1982; O’Brien, 1990; DeVries and Grimes, 1997; Trippel et al., 1997; Overholtz, 2002). Since the 1960s, when peak landings occurred for a suite of groundﬁsh and pelagic ﬁsh stocks in the northwest Atlantic Ocean, for example, there have been consistent and deﬁnitive changes in their life history parameters. Generally throughout this period, stocks of Atlantic cod (Gadus morhua), haddock (Melanogrammus aegleﬁnus), silver hake (Merluccius bilinearis), and yellowtail ﬂounder (Limanda

138

Gavin A. Begg

ferruginea) have grown to lower maximum lengths, reached lower ages and lengths at sexual maturity, and suffered higher mortality rates (e.g., O’Brien et al., 1993; Sinclair and Murawski, 1997; Begg et al., 1999b) (Fig. 6-9). This interdependence of different life history parameters is implicit in compensatory responses of exploited stocks to ﬁshing pressure (Beverton, 1963; O’Brien et al., 1993), where these responses occur between parameters to most likely maintain evolutionary ﬁtness (Jennings and Beverton, 1991). Atlantic herring (Clupea harengus) stocks, and others, have also experienced signiﬁcant expansions and contractions in their distributions relative to changing abundance levels (Overholtz, 2002) (Fig. 6-10). Associated with these changes in life history parameters have been large-scale ﬂuctuations in atmospheric circulation patterns (Drinkwater, 1996; Drinkwater and Mountain, 1997) and signiﬁcant increases in ﬁshing pressure (Sinclair and Murawski, 1997). However, differentiating the inﬂuence of environmental conditions from exploitation patterns is inherently difﬁcult, particularly given the magnitude of ﬁshery-induced changes that have occurred on these stocks (Serchuk et al., 1994). The existence of temporal variability in life history parameters does not negate their utility for the purpose of stock identiﬁcation, although it does highlight the need to examine these parameters among stocks over consistent time frames. If only a particular time frame was analyzed among stocks, or worse still, dissimilar time frames, erroneous results concerning the stock structure of a species could be derived because of the plasticity of vital life history parameters to the effects of changing environmental conditions and ﬁshing pressures (Begg et al., 1999b). The inﬂuence of temporal variability in life history parameters for stock identiﬁcation has largely been ignored because, typically, the assumption is made that temporal variation within stocks is not confounded with geographic variation among stocks or that within-stock variation is insigniﬁcant relative to among-stock variation (Blouw et al., 1988; Stephenson, 1991). Failure to account for such temporal variation, however, as mentioned previously, may result in falsely attributing differences between stocks to a stock effect, whereas these differences may in fact be reﬂective of differences in sampling times (Begg and Waldman, 1999). The use of inconsistent or a single time frame by which to interpret life history parameters for stock identiﬁcation can be misleading and stresses the need to investigate the temporal variation in these parameters (Begg et al., 1999b). Studies that fail to account for such within-stock variability are subject to error, which will depend on the magnitude of within-stock variation relative to among-stock variation, the proportion of life history parameters that demonstrate temporal variation, and the degree to which ﬁsh stocks vary synchronously for the parameters in question (Blouw et al., 1988). Temporal variation should be minimized in any attempt to identify stock structure by collecting or analyzing data when temporal effects are least pronounced (Ihssen et al., 1981). Synchronous

139

Life History Parameters

A

Females

1.28

CC GB

0.96

K (± s.e.)

0.96

K (± s.e.)

Males

1.28

CC GB

0.64

0.64 0.32

0.32

1970 1975 1980 1985 1990 1995

1970 1975 1980 1985 1990 1995

Year 5

Year 5

CC GB

4

A 50 (years)

A 50 (years)

4 3 2 1

3 2 1

1970 1975 1980 1985 1990 1995

1970 1975 1980 1985 1990 1995

Year

Year

40

40

CC GB

CC GB

30

L50 (cm)

30

L50 (cm)

CC GB

20 10

20 10

1970 1975 1980 1985 1990 1995

1970 1975 1980 1985 1990 1995

Year

Year

FIGURE 6-9. Temporal variation in life history parameters. (A) Yellowtail ﬂounder (Limanda ferruginea) sex-speciﬁc mean von Bertalanffy growth rate coefﬁcient (K), and age (A50) and length (L50) at 50% maturity for putative Cape Cod (CC) and Georges Bank (GB) stocks in the northwest Atlantic Ocean (1970–1997) (Begg et al., 1999b); and (B) Atlantic cod (Gadus morhua) sex-speciﬁc mean age and length at 50% maturity in waters of northern and southern Iceland (Marteinsdottir and Begg, 2002).

sampling across putative stocks is also recommended, as too is sampling during periods when stocks are likely to be homogeneous and optimally separated, such as during spawning times for adult ﬁsh (Casselman et al., 1981; Ihssen et al., 1981).

140

Gavin A. Begg

B

FIGURE 6-9. Continued

Stock identiﬁcation should be an evolving process where, because of their plasticity, life history parameters are reexamined periodically in response to changing environmental and resource conditions. Importantly, the existence of temporal variation in life history parameters does not negate their use, particularly if members of a stock respond in a similar manner over time to environmental and anthropogenic perturbations, as such a response in itself can be viewed as a stock attribute (Casselman et al., 1981; Ihssen et al., 1981).

IV. SPATIAL STABILITY Maintenance of temporal and spatial integrity among ﬁsh stocks is important for population structuring, particularly where spawning is adapted to the physical dispersive properties of a geographic location (Heath, 1992). Local hydrological conditions can minimize the mixing of eggs and larvae between neighboring stocks, effectively maintaining genetic discreteness and reproductive isolation (Iles and Sinclair, 1982; Palumbi, 1994). Alternatively, mixing between stocks may occur during egg, larval, or juvenile stages, with subsequent resegregation

Life History Parameters

141

FIGURE 6-10. Temporal variation in the relative center of abundance of Atlantic herring (Clupea harengus) collected from Northeast Fisheries Science Center (NEFSC) spring bottom trawl surveys (1968–1998) (Overholtz, 2002).

later in life (Swain et al., 1980), leading to potential differences among stocks in a range of life history parameters. In such circumstances, the level of stock integrity largely depends on the degree of larval mixing and postsettlement movement (Bruce et al., 2001). Understanding these structuring mechanisms, both biological and physical, can yield insights on factors affecting the stability of individual ﬁsh stocks and their associated life history parameters. Mechanisms by which putative ﬁsh stocks and their respective vital life history parameters are maintained are undoubtedly a combination of the biological processes and the physical environment in which they reside. Reproduction plays a fundamental role in the dispersal or retention of progeny through their initial spatial and temporal placement (Hare and Cowen, 1993) and assists in deﬁning the stock structure of a species, as genetic isolation must involve the spawning life history stage to restrict gene ﬂow to the level that effectively isolates stock

142

Gavin A. Begg

units (Iles and Sinclair, 1982). Spatial distributions and spawning times of a ﬁsh stock may represent evolutionary adaptations to circulation patterns (Parrish et al., 1981; Sinclair, 1988), which assists in maintaining reproductive isolation and stock integrity. Physical oceanographic processes also inﬂuence the distribution of early life history stages on a variety of spatial scales that can delineate and maintain stock boundaries (Bruce et al., 2001). Spawning times and locations of demersal gadoid stocks are usually associated with well-deﬁned circulation or hydrological features that enhance the retention of eggs and larvae (Hutchings et al., 1993; Page et al., 1999). Likewise, spawning locations of marine pelagic ﬁsh species and their stocks have been hypothesized to be dependent on the geographic extent of oceanographic larval retention areas (Iles and Sinclair, 1982; O’Boyle et al., 1984). Several studies have supported this theory, particularly for Atlantic herring (Clupea harengus) (Sinclair, 1988; Stephenson, 1991), but also for a number of other species (e.g., Johnson et al., 1994; Bruce et al., 2001; Gaughan et al., 2001). In contrast, coral reef ﬁsh species generally exist as metapopulations of sedentary adult stocks linked by pelagic larval dispersal (Sale, 1998), resulting in differences in life history parameters at a range of spatial scales dependent on their position within the respective reef connectivity matrix and the degree of selfseeding (Kritzer, 2002; Williams et al., 2003). Studies of biophysical processes and early life history dynamics are critical in understanding the mechanisms responsible for the maintenance of stock structure and for the determination of spatial scales over which potentially connected stocks operate (Gaughan et al., 2001). Maintenance of stock integrity requires that progeny recruit to their source population and that mixing between stocks is restricted (Bruce et al., 2001). However, despite some degree of stock mixing, which often occurs during early life history stages of marine species, persistent differences in life history parameters are generally maintained between putative ﬁsh stocks (e.g., Begg et al., 1999b). Several reasons may explain these persistent differences in the possibility of stock mixing, including the following: (1) mixing, is variable between years and depends on the speciﬁc spawning locations, circulation patterns, and survival of progeny during a given year; (2) mixing occurs during the larval stages, with segregation occurring at some later point in the life history; and, more pertinently, (3) some stocks are not genotypic groups of ﬁsh, but simply reﬂect differences in phenotypic life history parameters in response to environmental variation and ﬁshing pressure (Begg et al., 1999b).

V. CONCLUSIONS Life history parameters provide fundamental biological information for ﬁsheries management and have the dual functionality in being useful descriptors of puta-

Life History Parameters

143

tive ﬁsh stocks. Distribution, abundance, growth, mortality, reproduction, recruitment, and other life history parameters provide baseline information that typically assists with the initial recognition and delineation of geographic regions that are representative of individual stocks (Pawson and Jennings, 1996) and is an almost essential prerequisite for successful stock identiﬁcation (Grifﬁths, 1997). The use of such parameters is an efﬁcient and cost-effective means for stock identiﬁcation, as these data are routinely collected for assessment and management purposes (Ihssen et al., 1981; Pawson and Jennings, 1996) and can often be derived from archived databases. Although the utility of these parameters for stock identiﬁcation appears to decrease with stock complexity, their applicability increases with the number and diversity of parameters examined (Ihssen et al., 1981; Begg and Waldman, 1999). Life history parameters are useful indicators of phenotypic stock differences, but typically do not enable classiﬁcation of individual ﬁsh to a speciﬁc stock because of the wide variability that naturally occurs within these parameters between individual ﬁsh (Waldman et al., 1988; Begg et al., 1999b). Phenotypic differences in life history parameters also do not provide direct evidence of genetic isolation between stocks, but can indicate the prolonged separation of postlarval ﬁsh subject to different environmental conditions and/or ﬁshing pressures (Campana et al., 1995; Begg et al., 1999a). Moreover, phenotypic differences in life history parameters among putative stocks, irrespective of genetic differences, provide a ﬁrm basis for separate management units and should be modeled separately for stock assessment purposes because of inherently related productivity differences (Cadrin and Friedland, 1999). Failure to recognize such differences and the related stock structure of an exploited species, as is typical of most stock assessment methods which model the dynamics of closed populations and assume homogeneous life history parameters, can lead to overﬁshing and depletion of less productive stocks with unknown ecological consequences (Ricker, 1954; Stephenson, 1999). Anthropogenic, biophysical, and ecological processes affecting life history parameters operate at a range of temporal and spatial scales that govern the formation and maintenance of stock structure. Indeed, life history parameters have been used successfully for stock identiﬁcation at a diversity of scales, although which parameters to use will ultimately depend on the management objective and relevant scale for assessment. Nonetheless, it is critical to estimate life history parameters across a range of scales to infer which are of greatest importance for assessment and management (Stephenson, 1991; Sale, 1998). Furthermore, because of the diversity of these scales and complexity of the processes operating on life history parameters (Kritzer, 2002; Williams et al., 2003), stock identiﬁcation studies should examine a multitude of parameters to maximize the likelihood of correctly deﬁning stock structures. Investigation of any single parameter will not necessarily reveal stock differences even when true stock

144

Gavin A. Begg

differences exist (in statistical hypothesis testing this is referred to as “Type 1 error”), while the strongest inferences on stock structure are drawn from a suite of complementary parameters that cover multiple aspects of the biology of a species (Begg and Waldman, 1999). Future studies should adopt a holistic approach to stock identiﬁcation by integrating information in a multivariate framework that captures the salient aspects of a range of life history parameters (e.g., Begg and Waldman, 1999; Rochet, 2000; Fromentin and Fonteneau, 2001). Such an approach would optimize available information and provide a comprehensive understanding of a species life history, which is an essential requirement for successful stock identiﬁcation (Pawson and Jennings, 1996). Life history parameters, therefore, should be the ﬁrst data examined in any stock identiﬁcation study, as these are typically available for exploited species in archived databases of most ﬁsheries and management agencies or can be readily collected in the ﬁeld and analyzed accordingly in a cost-effective manner. Information derived from these parameters can be used to describe stock boundaries at a range of spatial scales that may assist in directing future studies to reﬁne stock structures using more sophisticated techniques. Temporal and spatial variability in life history parameters, and potential biases in sampling, should also be examined in an effort to understand the mechanisms responsible for structuring putative stocks and to ensure that conclusions reached are reﬂective of true stock differences and not sampling anomalies.

ACKNOWLEDGMENTS I would like to thank Steven Cadrin for the invitation to contribute this chapter and his continued perseverance and encouragement to ensure its completion, and Jon Hare for his collaboration on an earlier manuscript which formed the basis of this work.

REFERENCES Adkinson, M. 1995. Population differentiation in Paciﬁc salmon: local adaptation, genetic drift, or the environment? Canadian Journal of Fisheries and Aquatic Sciences 52: 2762–2777. Anonymous. 1992. Report of the ICES workshop on the analysis of trawl survey data. ICES CM 1992/D: 6. Azarovitz, T. R. 1981. A brief historical review of the Woods Hole Laboratory trawl survey time series. In W. G. Doubleday and D. Rivard (eds.), Bottom Trawl Surveys. Canadian Special Publication of Fisheries and Aquatic Sciences 58: 62–67. Bailey, K. M., Quinn, T. J., II, Bentzen, P., and Grant, W. S. 1999. Population structure and dynamics of walleye pollock, Theragra chalcogramma. Advances in Marine Biology 37: 179–255. Beacham, T. D. 1982. Some aspects of growth, Canadian exploitation, and stock identiﬁcation of Atlantic cod (Gadus morhua) on the Scotian Shelf and Georges Bank in the northwest Atlantic Ocean. Canadian Technical Report of Fisheries and Aquatic Sciences 1069: 43 pp.

Life History Parameters

145

Beamish, R. J. and McFarlane, G. A. 1983. The forgotten requirement for age validation in ﬁsheries biology. Transactions of the American Fisheries Society 112: 735–743. Begg, G. A. 1998a. A review of stock identiﬁcation of haddock, Melanogrammus aegleﬁnus, in the northwest Atlantic Ocean. Marine Fisheries Review 60: 1–15. Begg, G. A. 1998b. Reproductive biology of school mackerel (Scomberomorus queenslandicus) and spotted mackerel (S. munroi) in Queensland east-coast waters. Marine and Freshwater Research 49: 261–270. Begg, G. A. and Brown, R. W. 2000. Stock identiﬁcation of haddock Melanogrammus aegleﬁnus on Georges Bank based on otolith shape analysis. Transactions of the American Fisheries Society 129: 935–945. Begg, G. A., Cameron, D. S., and Sawynok, W. 1997. Movements and stock structure of school mackerel (Scomberomorus queenslandicus) and spotted mackerel (S. munroi) in Australian east-coast waters. Marine and Freshwater Research 48: 295–301. Begg, G. A., Friedland, K. D., and Pearce, J. B. 1999a. Stock identiﬁcation and its role in stock assessment and ﬁsheries management: an overview. Fisheries Research 43: 1–8. Begg, G. A., Hare, J. A., and Sheehan, D. D. 1999b. The role of life history parameters as indicators of stock structure. Fisheries Research 43: 141–163. Begg, G. A. and Marteinsdottir, G. 2000. Spawning origins of pelagic juvenile cod Gadus morhua inferred from spatially explicit age distributions: potential inﬂuences on year-class strength and recruitment. Marine Ecology Progress Series 202: 193–217. Begg, G. A. and Marteinsdottir, G. 2002a. Environmental and stock effects on spatial distribution and abundance of mature cod Gadus morhua. Marine Ecology Progress Series 229: 245–262. Begg, G. A. and Marteinsdottir, G. 2002b. Environmental and stock effects on spawning origins and recruitment of cod Gadus morhua. Marine Ecology Progress Series 229: 263–277. Begg, G. A. and Sellin, M. J. 1998. Age and growth of school mackerel (Scomberomorus queenslandicus) and spotted mackerel (S. munroi) in Queensland east-coast waters with implications for stock structure. Marine and Freshwater Research 49: 109–120. Begg, G. A. and Waldman, J. R. 1999. An holistic approach to ﬁsh stock identiﬁcation. Fisheries Research 43: 35–44. Bernard, D. R. 1981. Multivariate analysis as a means of comparing growth in ﬁsh. Canadian Journal of Fisheries and Aquatic Sciences 38: 233–236. Beverton, R. J. H. 1963. Maturation, growth and mortality of clupeid and engraulid stocks in relation to ﬁshing. Rapports et Procés-Verbaux des Réunions, Conseil International pour l’Exploration de la Mer 154: 44–67. Beverton, R. J. H., and Holt, S. J. 1957. On the dynamics of exploited ﬁsh populations. Fisheries Investigation Series II London 19: 553 pp. Blouw, D. M., Saxon, S. D., and Chadwick, E. M. P. 1988. Temporal variation of meristic traits within an Atlantic salmon (Salmo salar) stock, and implications for stock identiﬁcation. Canadian Journal of Fisheries and Aquatic Sciences 45: 1330–1339. Bolles, K. L. and Begg, G. A. 2000. Distinction between silver hake (Merluccius bilinearis) stocks in U.S. waters of the northwest Atlantic based on whole otolith morphometrics. Fishery Bulletin 98: 451–462. Boyar, H. C. 1968. Age, length, and gonadal stages of herring from Georges Bank and the Gulf of Maine. Northwest Atlantic Fisheries Research Bulletin 5: 49–61. Botsford, L. W. and Paulsen, C. M. 2000. Assessing covariability among populations in the presence of intraspeciﬁc correlation: Columbia River spring-summer chinook salmon (Oncorhynchus tshawytscha) stocks. Canadian Journal of Fisheries and Aquatic Sciences 57: 616–627. Bradford, R. G. and Stephenson, R. L. 1992. Egg weight, fecundity, and gonad weight variability among northwest Atlantic herring (Clupea harengus) populations. Canadian Journal of Fisheries and Aquatic Sciences 49: 2045–2054.

146

Gavin A. Begg

Bruce, B. D., Evans, K., Sutton, C. A., Young, J. W., and Furlani, D. M. 2001. Inﬂuence of mesoscale oceanographic processes on larval distribution and stock structure in jackass morwong (Nemadactylus macropterus: Cheilodactylidae). ICES Journal of Marine Science 58: 1072–1080. Cadrin, S. X. and Friedland, K. D. 1999. The utility of image processing techniques for morphometric analysis and stock identiﬁcation. Fisheries Research 43: 129–139. Campana, S. E. 2001. Accuracy, precision and quality control in age determination, including a review of the use and abuse of age validation methods. Journal of Fish Biology 59: 197–242. Campana, S. E., Frank, K. T., Hurley, P. C. F., Koeller, P. A., Page, F. H., and Smith, P. C. 1989. Survival and abundance of young Atlantic cod (Gadus morhua) and haddock (Melanogrammus aegleﬁnus) as indicators of year class strength. Canadian Journal of Fisheries and Aquatic Sciences 46 (suppl. 1): 171–182. Campana, S. E., Gagné, J. A., and McLaren, J. W. 1995. Elemental ﬁngerprinting of ﬁsh otoliths using ID-ICPMS. Marine Ecology Progress Series 122: 115–120. Campbell, A. and Mohn, R. K. 1983. Deﬁnition of American lobster stocks for the Canadian maritimes by analysis of ﬁshery-landing trends. Transactions of the American Fisheries Society 112: 744–759. Casselman, J. M., Collins, J. J., Crossman, E. J., Ihssen, P. E., and Spangler, G. R. 1981. Lake whiteﬁsh (Coregonus clupeaformis) stocks of the Ontario waters of Lake Huron. Canadian Journal of Fisheries and Aquatic Sciences 38: 1772–1789. Castonguay, M., Simard, P., and Gagnon, P. 1991. Usefulness of Fourier analysis of otolith shape for Atlantic mackerel (Scomber scombrus) stock discrimination. Canadian Journal of Fisheries and Aquatic Sciences 48: 296–302. Cerrato, R. M. 1990. Interpretable statistical tests for growth comparisons using parameters in the von Bertalanffy equation. Canadian Journal of Fisheries and Aquatic Sciences 47: 1416– 1426. Clark, S. 1981. Use of trawl survey data in assessments. In W. G. Doubleday and D. Rivard (eds.), Bottom Trawl Surveys. Canadian Special Publication of Fisheries and Aquatic Sciences 58: 82–92. Clark, S. H., Overholtz, W. H., and Hennemuth, R. C. 1982. Review and assessment of the Georges Bank and Gulf of Maine haddock ﬁshery. Journal of the Northwest Atlantic Fishery Science 3: 1–27. Cohen, E. B., Mountain, D. G., and O’Boyle, R. 1991. Local-scale versus large-scale factors affecting recruitment. Canadian Journal of Fisheries and Aquatic Sciences 48: 1003–1006. Cowen, R. K., Hare, J. A., and Fahay, M. P. 1993. Beyond hydrography: can physical processes explain larval ﬁsh assemblages within the Middle Atlantic Bight? Bulletin of Marine Science 53: 567–587. Cushing, D. H. 1969. The regularity of the spawning season of some ﬁshes. Journal Conseil International pour l’Exploration de la Mer 33: 81–97. Dawson, P. 1994. The stock structure of Bering Sea walleye pollock (Theragra chalcogramma). Master of Science thesis. University of Washington, Seattle, WA. 220 pp. DeVries, D. A. and Grimes, C. B. 1997. Spatial and temporal variation in age and growth of king mackerel, Scomberomorus cavalla, 1977–1992. Fishery Bulletin 95: 694–708. Dizon, A. E., Lockyer, C., Perrin, W. F., Demaster, D. P., and Sisson, J. 1992. Rethinking the stock concept: a phylogeographic approach. Conservation Biology 6: 24–36. Drinkwater, K. F. 1996. Atmospheric and oceanic variability in the northwest Atlantic during the 1980s and early 1990s. Journal of the Northwest Atlantic Fishery Science 18: 77–97. Drinkwater, K. F. and Mountain, D. G. 1997. Climate and oceanography. In J. Boreman, B. S. Nakashima, J. A. Wilson, and R. L. Kendall (eds.), Northwest Atlantic Groundﬁsh: Perspectives on a Fishery Collapse. American Fisheries Society, Maryland, pp. 3–25. Eastwood, P. D., Meaden, G. J., and Grioche, A. 2001. Modelling spatial variations in spawning habitat suitability for the sole Solea solea using regression quantiles and GIS procedures. Marine Ecology Progress Series 224: 251–266.

Life History Parameters

147

Finucane, J. H., Collins, L. A., Brusher, H. A., and Saloman, C. H. 1986. Reproductive biology of king mackerel, Scomberomorus cavalla, from the southeastern United States. Fishery Bulletin 84: 841–850. Fletcher, W. J. and Sumner, N. R. 1999. Spatial distribution of sardine (Sardinops sagax) eggs and larvae: an application of geostatistics and resampling to survey data. Canadian Journal of Fisheries and Aquatic Sciences 56: 907–914. Fromentin, J.-M. and Fonteneau, A. 2001. Fishing effects and life history traits: a case study comparing tropical versus temperate tunas. Fisheries Research 53: 133–150. Gaughan, D. J., White, K. V., and Fletcher, W. J. 2001. The links between functionally distinct adult assemblages of Sardinops sagax: larval advection across management boundaries. ICES Journal of Marine Science 58: 597–606. Grifﬁths, M. H. 1996. Age and growth of South African silver kob Argyrosomus inodorus (Sciaenidae), with evidence for separate stocks. South African Journal of Marine Science 17: 37–48. Grifﬁths, M. H. 1997. The life history and stock separation of silver kob, Argyrosomus inodorus, in South African waters. Fishery Bulletin 95: 47–67. Grimes, C. B., Johnson, A. G., and Fable, W. A., Jr., eds. 1987. Delineation of king mackerel (Scomberomorus cavalla) stocks along the U.S. east coast and in the Gulf of Mexico. In H. E. Kumpf, R. N. Vaught, C. B. Grimes, A. G. Johnson, and E. L. Nakamura (eds.), Proceedings of the Stock Identiﬁcation Workshop 5–7 November 1985, Panama City Beach, Florida. NOAA Technical Memorandum NMFS-SEFC-199. United States Government Printing Ofﬁce, Panama City, pp. 186–187. Haddon, M. 2001. Modelling and Quantitative Methods in Fisheries. Chapman and Hall, Boca Raton, FL. 406 pp. Hanchet, S. 1999. Stock structure of southern blue whiting (Micromesistius australis) in New Zealand waters. New Zealand Journal of Marine and Freshwater Research 33: 599–609. Hare, J. A. 2004. The use of early life history stages in stock identiﬁcation studies. In S. X. Cadrin, K. D. Friedland, and J. R. Waldman (eds.), Stock Identiﬁcation Methods. Academic Press, Burlington, MA, pp. 85–113. Hare, J. A. and Cowen, R. K. 1993. Ecological and evolutionary implications of the larval transport and reproductive strategy of blueﬁsh Pomatomus saltatrix. Marine Ecology Progress Series 98: 1–16. Heath, M. R. 1992. Field investigations of the early life stages of marine ﬁsh. Advances in Marine Biology 28: 1–174. Hilborn, R. and Walters, C. J. 1992. Quantitative Fisheries Stock Assessment. Choice, Dynamics and Uncertainty. Chapman & Hall, London, UK. 570 pp. Hjort, J. 1914. Fluctuations in the great ﬁsheries of northern Europe viewed in the light of biological research. Rapports et Procés-Verbaux des Réunions, Conseil International pour l’Exploration de la Mer 20: 1–228. Horn, P. L. and Hurst, R. J. 1999. Age and stock structure of gemﬁsh (Rexea solandri) in New Zealand waters. Marine and Freshwater Research 50: 103–115. Horrall, R. M. 1981. Behavioural stock isolating mechanisms in Great Lake ﬁshes with special reference to homing and imprinting. Canadian Journal of Fisheries and Aquatic Sciences 38: 1481–1496. Hutchings, J. A., Myers, R. A., and Lilly, G. R. 1993. Geographic variation in the spawning of Atlantic cod, Gadus morhua, in the northwest Atlantic. Canadian Journal of Fisheries and Aquatic Sciences 50: 2457–2467. Ihssen, P. E., Booke, H. E., Casselman, J. M., McGlade, J. M., Payne, N. R., and Utter, F. M. 1981. Stock identiﬁcation: materials and methods. Canadian Journal of Fisheries and Aquatic Sciences 38: 1838–1855.

148

Gavin A. Begg

Iles, T. D. and Sinclair, M. 1982. Atlantic herring: stock discreteness and abundance. Science 215: 627–633. Japp, D. W. 1990. A new study on age and growth of kingklip Genypterus capensis off the south and west coasts of South Africa, with comments on its use for stock identiﬁcation. South African Journal of Marine Science 9: 223–237. Jennings, S. and Beverton, R. J. H. 1991. Intraspeciﬁc variation in life history tactics of Atlantic herring (Clupea harengus L.) stocks. ICES Journal of Marine Science 48: 117–125. Jennings, S., Kaiser, M. J., and Reynolds, J. D. 2001. Marine Fisheries Ecology. Blackwell Science, Oxford, UK. 417 pp. Johnson, A. G., Fable, W. A., Jr., Grimes, C. B., and Trent, L. 1994. Evidence for distinct stocks of king mackerel, Scomberomorus cavalla, in the Gulf of Mexico. Fishery Bulletin 92: 91–101. Kimura, D. K. 1980. Likelihood methods for the von Bertalanffy growth curve. Fishery Bulletin 77: 765–776. Koslow, J. A. 1984. Recruitment patterns in northwest Atlantic ﬁsh stocks. Canadian Journal of Fisheries and Aquatic Sciences 41: 1722–1729. Koslow, J. A., Thompson, K. R., and Silvert, W. 1987. Recruitment of northwest Atlantic cod (Gadus morhua) and haddock (Melanogrammus aegleﬁnus) stocks: inﬂuence of stock size and climate. Canadian Journal of Fisheries and Aquatic Sciences 44: 26–39. Kritzer, J. P. 2002. Variation in the population biology of stripey bass Lutjanus carponotatus within and between two island groups on the Great Barrier Reef. Marine Ecology Progress Series 243: 191–207. Kutkuhn, J. H. 1981. Stock deﬁnition as a necessary basis for cooperative management of Great Lakes ﬁsh resources. Canadian Journal of Fisheries and Aquatic Sciences 38: 1476–1478. Lowe, S. A., Van Doornik, D. M., and Winans, G. A. 1998. Geographic variation in genetic and growth patterns of atka mackerel, Pleurogrammus monopterygius (Hexagrammidae), in the Aleutian archipelago. Fishery Bulletin 96: 502–515. Maravelias, C. D. and Reid, D. G. 1997. Identifying the effects of oceanographic features and zooplankton on prespawning herring abundance using generalized additive models. Marine Ecology Progress Series 147: 1–9. Marteinsdottir, G. and Begg, G. A. 2002. Essential relationships incorporating the inﬂuence of age, size and condition on variables required for estimation of reproductive potential in Atlantic cod Gadus morhua stocks. Marine Ecology Progress Series 235: 235–256. Marteinsdottir, G., Gudmundsdottir, A., Thorsteinsson, V., and Stefánsson, G. 2000a. Spatial variation in abundance, size composition and viable egg production of spawning cod (Gadus morhua L.) in Icelandic waters. ICES Journal of Marine Science 57: 824–830. Marteinsdottir, G., Gunnarsson, B., and Suthers, I. M. 2000b. Spatial variation in hatch date distributions and origin of pelagic juvenile cod in Icelandic waters. ICES Journal of Marine Science 57: 1182–1195. Mertz, G. and Myers, R. A. 1994. Match/mismatch predictions of spawning duration versus recruitment variability. Fisheries Oceanography 3: 236–245. Misra, R. K. 1980. Statistical comparisons of several growth curves of the von Bertalanffy type. Canadian Journal of Fisheries and Aquatic Sciences 37: 920–926. Morse, W. W., Fahay, M. P., and Smith, W. G. 1987. MARMAP surveys of the continental shelf from Cape Hatteras, North Carolina, to Cape Sable, Nove Scotia (1977–1984). Atlas no. 2. Annual distribution patterns of ﬁsh larvae. NOAA Technical Memorandum NMFS-S/NEC, 47: 215 pp. Myers, R. A., Barrowman, N. J., Hutchings, J. A., and Rosenberg, A. A. 1995. Population dynamics of exploited ﬁsh stocks at low population levels. Science 269: 1106–1108. O’Boyle, R. N., Sinclair, M., Conover, R. J., Mann, K. H., and Kohler, A. C. 1984. Temporal and spatial distribution of ichthyoplankton communities of the Scotian Shelf in relation to biological,

Life History Parameters

149

hydrological, and physiographic features. Rapports et Procés-Verbaux des Réunions, Conseil International pour l’Exploration de la Mer 183: 27–40. O’Brien, L. 1990. Effects of ﬂuctuations in stock abundance upon life history parameters of Atlantic cod, Gadus morhua, for the 1970–1987 year classes from Georges Bank and the Gulf of Maine. Master of Science thesis. University of Washington. Seattle, WA. 95 pp. O’Brien, L., Burnett, J., and Mayo, R. K. 1993. Maturation of nineteen species of ﬁnﬁsh off the northeast coast of the United States, 1985–1990. NOAA Technical Report National Marine Fisheries Service 113: 66 pp. Overholtz, W. J. 2002. The Gulf of Maine–Georges Bank Atlantic herring (Clupea harengus): spatial pattern analysis of the collapse and recovery of a large marine ﬁsh complex. Fisheries Research 57: 237–254. Page, F. H. and Frank, K. T. 1989. Spawning time and egg stage duration in northwest Atlantic haddock (Melanogrammus aegleﬁnus) stocks with emphasis on Georges and Browns Bank. Canadian Journal of Fisheries and Aquatic Sciences 46 (suppl. 1): 68–81. Page, F. H., Sinclair, M., Naimie, C. E., Loder, J. W., Losier, R. J., Berrien, P. L., and Lough, R. G. 1999. Cod and haddock spawning on Georges Bank in relation to water residence times. Fisheries Oceanography 8: 212–226. Pálsson, O. K., Jonsson, E., Schopka, S. A., Stefánsson, G., and Steinarsson, B. A. 1989. Icelandic groundﬁsh survey data used to improve precision in stock assessments. Journal of Northwest Atlantic Fishery Science 9: 53–72. Palumbi, S. R. 1994. Genetic divergence, reproductive isolation, and marine speciation. Annual Review of Ecology and Systematics 25: 547–572. Parrish, R. H., Nelson, C. S., and Bakiun, A. 1981. Transport mechanisms and reproductive success of ﬁshes in the California current. Biological Oceanography 1: 175–203. Pawson, M. G. and Jennings, S. 1996. A critique of methods for stock identiﬁcation in marine capture ﬁsheries. Fisheries Research 25: 203–217. Petitgas, P. 1993. Geostatistics for ﬁsh stock assessments: a review and an acoustic application. ICES Journal of Marine Science 50: 285–298. Ricker, W. E. 1954. Stock and recruitment. Journal of the Fisheries Research Board of Canada 11: 559–623. Rochet, M.-J. 2000. May life history traits be used as indices of population viability? Journal of Sea Research 44: 145–157. Rulifson, R. A. and Dadswell, M. J. 1995. Life history and population characteristics of striped bass in Atlantic Canada. Transactions of the American Fisheries Society 124: 477–507. Sale, P. F. 1998. Appropriate spatial scales for studies of reef-ﬁsh ecology. Australian Journal of Ecology 23: 202–208. Schaefer, K. M. 1987. Reproductive biology of black skipjack, Euthynnus lineatus, an eastern Paciﬁc tuna. Inter-American Tropical Tuna Commission Bulletin 19: 260 pp. Schnute, J. 1981. A versatile growth model with statistically stable parameters. Canadian Journal of Fisheries and Aquatic Sciences 38: 1128–1140. Serchuk, F. M., Grosslein, M. D., Lough, R. G., Mountain, D. G., and O’Brien, L. 1994. Fishery and environmental factors affecting trends and ﬂuctuations in the Georges Bank and Gulf of Maine Atlantic cod stocks: an overview. ICES Marine Science Symposium 198: 77– 109. Simard, P., Castonguay, M., D’Armours, D., and Magnan, P. 1992. Growth comparison between juvenile Atlantic mackerel (Scomber scombrus) from the two spawning groups of the northwest Atlantic. Canadian Journal of Fisheries and Aquatic Sciences 49: 2242–2248. Sinclair, A. F. and Murawski, S. A. 1997. Why have groundﬁsh stocks declined? In J. Boreman, B. S. Nakashima, J. A. Wilson, and R. L. Kendall (eds.), Northwest Atlantic Groundﬁsh: Perspectives on a Fishery Collapse. American Fisheries Society, Bethesda, MD, pp. 71–93.

150

Gavin A. Begg

Sinclair, M. 1988. Marine Populations: An Essay on Population Regulation and Speciation. University of Washington Press, Seattle, WA. 252 pp. Sinclair, M. and Tremblay, M. J. 1984. Timing of spawning of Atlantic herring (Clupea harengus) populations and the match-mismatch theory. Canadian Journal of Fisheries and Aquatic Sciences 41: 1055–1065. Smith, P. J., Jamieson, A., and Birley, A. J. 1990. Electrophoretic studies and the stock concept in marine teleosts. Journal du Conseil International pour l’Exploration de la Mer 47: 231–245. Stephenson, R. L. 1991. Stock discreteness in Atlantic herring: a review of arguments for and against. In V. Wespestad, J. Collie, and E. Collie (eds.), Proceedings of the International Herring Symposium, Anchorage, Alaska, October 23–25, 1990 (9th Lowell Wakeﬁeld Fisheries Symposium). University of Alaska, Fairbanks, pp. 659–666. Stephenson, R. L. 1999. Stock complexity in ﬁsheries management: a perspective of emerging issues related to population sub-units. Fisheries Research 43: 247–249. Stoner, A. W., Manderson, J. P., and Pessutti, J. P. 2001. Spatially explicit analysis of estuarine habitat for juvenile winter ﬂounder: combining generalized additive models and geographic information systems. Marine Ecology Progress Series 213: 253–271. Summerfelt, R. C. and Hall, G. E. (eds.). 1987. Age and Growth of Fish. Iowa State University Press, Ames, IA. 544 pp. Swain, A., Parrish, B. B., and Horsted, S. A. 1980. Tagging of salmon smolts in European rivers with special references to recaptures off West Greenland in 1972 and earlier years. Rapports et ProcésVerbaux des Réunions, Conseil International pour l’Exploration de la Mer 176: 93–113. Swain, D. P. and Wade, E. J. 1993. Density-dependent geographic distribution of Atlantic cod (Gadus morhua) in the southern Gulf of St. Lawrence. Canadian Journal of Fisheries and Aquatic Sciences 50: 725–733. Swartzman, G., Huang, C., and Kaluzny, S. 1992. Spatial analysis of Bering Sea groundﬁsh survey data using generalized additive models. Canadian Journal of Fisheries and Aquatic Sciences 49: 943–962. Thompson, K. R. and Page, F. H. 1989. Detecting synchrony of recruitment using short, autocorrelated time series. Canadian Journal of Fisheries and Aquatic Sciences 46: 1831–1838. Trippel, E. A. 1999. Estimation of stock reproductive potential: history and challenges for Canadian Atlantic gadoid stock assessments. Journal of the Northwest Atlantic Fishery Science 25: 61–81. Trippel, E. A., Morgan, M. J., Fréchet, A., Rollet, C., Sinclair, A., Annand, C., Beanlands, D., and Brown, L. 1997. Changes in age and length at sexual maturity of northwest Atlantic cod, haddock and pollock stocks, 1972–1995. Canadian Technical Report of Fisheries and Aquatic Sciences 2157: 120 pp. Utter, F. M. 1991. Biochemical genetics and ﬁshery management: an historical perspective. Journal of Fish Biology 39(A): 1–20. Von Bertalanffy, L. 1938. A quantitative theory of organic growth. Human Biology 10: 181–213. Waldman, J. R. 1999. The importance of comparative studies in stock analysis. Fisheries Research 43: 237–246. Waldman, J. R. and Fabrizio, M. C. 1994. Problems of stock deﬁnition in estimating relative contributions of Atlantic striped bass to the coastal ﬁshery. Transactions of the American Fisheries Society 123: 766–778. Waldman, J. R., Grossﬁeld, J., and Wirgin, I. 1988. Review of stock discrimination techniques for striped bass. North American Journal of Fisheries Management 8: 410–425. Williams, A. J., Davies, C. R., Mapstone, B. D., and Russ, G. R. 2003. Scales of spatial variation in demography of a large coral-reef ﬁsh—an exception to the typical model? Fishery Bulletin 101: 673–683.

PART

Natural Marks— Morphological Analyses

III

This Page Intentionally Left Blank

CHAPTER

7

Morphometric Landmarks STEVEN X. CADRIN National Marine Fisheries Service,1 Woods Hole, Massachusetts, USA

I. Introduction II. Methodological Protocols A. Sampling B. Choice of Characters C. Statistical Analysis III. Interpretation of Morphometric Differences A. Salmon Case Studies IV. Discussion References

I. INTRODUCTION Patterns of morphometric variation in ﬁshes indicate differences in growth and maturation rates because body form is a product of ontogeny. Morphometric landmark methods comprise one of the two major categories of morphometric analysis, the other being outline methods (see this volume, Chapter 8). The distinction between the two is that landmark methods analyze data derived from discrete morphometric points, linear distances between points, and geometric relationships among points, whereas outline methods deal with perimeter shapes. Beginning with Huxley and Teissier’s pioneering work on bivariate allometry of crustaceans and ﬁnﬁsh in the 1920s (reviewed by Huxley, 1932 and Teissier, 1960), stock identiﬁcation studies have played a central role in the development of traditional landmark methods (see Cadrin, 2000 for a brief history). Royce (1957) reviewed methods of multivariate morphometrics for studying subpopulations of ﬁshes more than a decade before general texts on morphometrics were published (e.g., Blackith and Reyment, 1971; Pimentel, 1979), and many early applications of multivariate morphometrics were for stock identiﬁcation. For 1 Much of this work was completed while under the employment of the Massachusetts Division of Marine Fisheries.

Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

153

154

Steven X. Cadrin 70

7000 morphometrics fisheries

6000

50

5000

40

4000

30

3000

20

2000

10

1000

0

Fisheries Abstracts per Year

Morphometric Abstracts per Year

60

0 1970-1974

1975-1979

1980-1984

1985-1989

1990-1994

1995-2000

FIGURE 7-1. Frequency of published case studies on morphometric stock identiﬁcation by 5-yr period, with ﬁshery references for comparison (data source: Cambridge Scientiﬁc Abstracts, Aquatic Sciences and Fisheries, Biological Sciences and Living Resources, http://www.csa2.com).

example, Saila and Flowers’ (1969) morphometric discrimination of American lobster stocks is considered a benchmark study in the ﬁeld (see reviews by Gould and Johnston, 1972; Thorpe, 1976). The proliferation of morphometric applications for stock identiﬁcation during the 1970s and 1980s is illustrated by the case studies reviewed by Lee (1971), Winans (1987), and Saila and Martin (1987), as well as the subsequent increase in published case studies (Fig. 7-1). The development of digital imaging systems and advances in analytical methods revolutionized the study of morphometric variation, and have increased the power of morphometric analysis for stock identiﬁcation (Cadrin and Friedland, 1999). However, the application of advanced geometric methods as applied to stock identiﬁcation lags behind applications to other biological ﬁelds such as taxonomy and biomedical research (Cadrin, 2000). In the context of interdisciplinary stock identiﬁcation, or the use of information from various approaches, morphometric analysis provides information on phenotypic stocks, groups of individuals with similar growth, mortality, and reproductive rates (Booke, 1981). The phenotypic stock deﬁnition is less conservative than the genetic stock deﬁnition because it allows for some mixing among stocks, but partial isolation is enough that geographic differences persist. Despite dependence on the environment, ontogenetic rates inﬂuence many population attributes (e.g., reproduction, fecundity, longevity, size structure) that are intimately related to population dynamics (intrinsic rate of increase, carrying capacity, productivity, resilience, and so on; Cole, 1954) and determine how each stock responds to exploitation (Garrod and Horwood, 1984). Therefore, for the

Morphometric Landmarks

155

purpose of ﬁshery stock assessment, groups with different growth or reproductive dynamics should be modeled and managed separately, regardless of genetic homogeneity. The objective of this chapter is to review common protocols for sampling, analyzing, and interpreting variation associated with morphometric landmarks for stock identiﬁcation applications.

II. METHODOLOGICAL PROTOCOLS

A. SAMPLING Analysis of morphometric data has become more powerful for stock identiﬁcation through more rigorous sampling designs and more efﬁcient data acquisition. Optimal sampling designs vary as a function of precise objectives (Cadrin, 2000). For species that have little information about stock structure, morphometric analyses may be exploratory in nature. The optimal sampling design for exploratory studies involves obtaining representative samples from the entire geographic range as well as all seasons to investigate patterns of variation and potential for mixing. A second tier of inquiry is for species with information on the location of spawning grounds and ﬁdelity to them, for which morphometric analysis can be used to test hypotheses about putative stocks. The optimal design for stock discrimination is sufﬁcient sample sizes from locations at the center of spawning grounds, during spawning seasons, when mixing among putative stocks is minimal. The most reﬁned tier of investigation is either delineation stocks (for species with little mixing) or stock composition analysis (for species that mix seasonally). Ideally, sampling for stock delineation or composition analysis has comprehensive representation, geographically and seasonally, to deﬁne discrete boundaries of morphometric variation, clines in morphometry, or the proportion of each morphometric variant in mixed samples. Optimal sample sizes are a function of the degree of morphometric variation within groups and the magnitude of difference among stocks that is desired to detect. Reliable estimates of bivariate correlation require at least 50 observations (Tabachnick and Fidell, 1989), multivariate analyses require more samples (n) than variables (p), and the stability of multivariate ordination is related to the ratio n : p. Saila and Martin (1987) advocated an ad hoc rule that sample sizes should be three times the number of variables. One complicating factor in sampling morphometric features for stock discrimination is that morphometry changes during the spawning process. Ripe gonads can contribute up to 40% of female body mass, greatly altering body proportions. Analysis of exploratory sampling conducted before, during, and after spawning can describe the degree of morphometric change, identify the time or stage that should be sampled to minimize the confusion between geographic

156

Steven X. Cadrin

variation and spawning-induced variation, and deﬁne which morphometric characters are inﬂuenced by spawning (e.g., Armstrong and Cadrin, 2001). If geographic variation is based on different spawning stages, subsequent stock delineations or composition analyses will be biased toward classifying spawning ﬁsh to the area that had the most spawners in the discrimination samples. A similar complication may be associated with feeding. Ideally, comparisons should be made among specimens with comparable stomach fullness and spawning stage. In addition to stomach and gonad observations, other individual attributes such as age, gender, and color may be associated with morphometric variation, and thus should be recorded for each specimen. Location information such as depth, salinity, temperature, and sea ﬂoor substrate should also be recorded. Such data serve as useful covariates and possibly causal factors for exploring patterns of morphometric variation. Although traditional measurement systems such as calipers and measuring boards are commonly used in morphometric studies, digital imaging with calibration provides superior data format, accuracy, design ﬂexibility, and potential for substantially increasing sample size. Coordinates of digital images can be calibrated if specimens are placed on a plane with a grid of known distances for a standard view (usually lateral or dorsal). Images can be archived from digital camera, digital video, or frame-grabbing software for videotape. Images can be calibrated to unit distances and corrected for tilt as a function of known distance in each corner of the image (Fig. 7-2). One major advantage of deriving morphometric data from digital images is the ability to store the image and the potential for reprocessing each individual to conﬁrm anomalous measurements or drive alternative character sets. Storage of images allows detailed inspection of extreme variants or outliers, as well as more ﬂexible character selection (Cadrin and Friedland, 1999). Imaging software also allows enhancement of images to accentuate subtle features. Another advantage of image processing is that data are stored as coordinates, facilitating geometric methods of analysis.

B. CHOICE

OF

CHARACTERS

Similar to other phenotypic approaches, there are an inﬁnite number of morphometric dimensions that can be measured to study patterns of variation. We can study general morphometry (i.e., shape of the entire organism) or shape of individual features (e.g., scales, otoliths, vertebrate, chelae). Perhaps the best criterion for choosing morphometric features is their use in interpretation. Morphometric stock identiﬁcation can be designed to focus on features that exhibit changes associated with development or maturation because geographic variation in ontogenetic rates is the basis of effective phenotypic stock identiﬁcation (i.e., identifying groups with different biological rates). Information from life history

157

Morphometric Landmarks

studies can guide a researcher’s choice of morphometric features. Such a priori choice of characters that are sensitive to ontogenetic processes may lead to clearer interpretation of intrapopulation patterns of morphometric variation. For example, differences in locomotion may be associated with general body form and ﬁn size of ﬁshes (e.g., Riddell and Leggett, 1981; Winans, 1984; Taylor and McPhail, 1985a,b; Swain and Holtby, 1989; Taylor, 1991), variation in mouth shape may be correlated to differences in diet (e.g., Skulason et al., 1989, 1996; Albertson and Kocher 2001), and differences in size of male secondary sex characters can indicate differences in size at maturity (Holtby et al., 1993; Cadrin, 1995, 2000). Landmarks should be homologous, representing the same developmental feature among specimens, and should be easily located (Winans, 1987; Bookstein, 1990). The most effective landmarks are those deﬁned by the intersection of different tissues, such as insertion points of ﬁns and anal pores. Extreme points, such as the posterior edge of the caudal ﬁn or tip of the snout or a spine, are also effective, but may not be strictly homologous. Using the same yellowtail ﬂounder example, landmarks range from homologous and clearly deﬁned, to nonhomologous and arbitrary (Fig. 7-2). Some landmarks are homologous, such

5

2

10

1

6

7

3

12

4

11 8 9

FIGURE 7-2. A set of 12 morphometric landmarks for yellowtail ﬂounder (numbered) and 12 calibration points.

158

Steven X. Cadrin

as the center of eyes (landmarks 2 and 3) and ﬁn insertion points (5, 6, 8, 10, 11); some are nearly so, like the tips of bones (1 and 4) and ﬁn tips (6, 7, 12). Landmark 5 (insertion of dorsal ﬁn in line with landmarks 6 and 8) is particularly nonhomologous. However, there are no homologous features along the dorsal ﬁn because the number of dorsal ﬁn rays varies among individuals. Note that 12 landmarks were also located on the measured grid to calibrate the image. Linear morphometric distances can be measured directly with calipers or a ruler or calculated from digital landmark coordinates. For example, 90 distances can be calculated between the 10 nonarticulating landmarks (i.e., those that do not move position relative to other landmarks) illustrated in Figure 7-2, but many distances would measure the same general feature. Bookstein et al. (1985) refer to this set of distances as the “globally redundant network.” However, redundancy should be minimized because minimum sample sizes required for multivariate analysis increase as a function of the number of variables, and the likelihood of spurious signiﬁcant differences increases with the number of characters used in analysis (Misra and Easton, 1999). Strauss and Bookstein (1982) developed the box-truss network in which trapezoidal cells with two crossing diagonals are formed between four adjacent landmarks. A box-truss network of 22 distances (plus two distances for pelvic and pectoral ﬁn lengths) is illustrated in Figure 73. Note that two of the landmarks (7 and 9) are the tips of articulating ﬁns, and the only meaningful distances derived from them are to their associated insertion points (6 and 8, respectively). An even less redundant set of linear distances is the triangle-truss network (Bookstein et al., 1985) in which triangular cells are formed between three adjacent landmarks with no crossing lines [e.g., a set of 10 triangle-truss networks plus the two paired ﬁn lengths were used for analysis of the same yellowtail ﬂounder landmarks by Cadrin and Silva (2004)]. The choice and number of morphometric distances for traditional multivariate analysis is a trade-off between a comprehensive measure of shape and sample size limitations. As the number of variables increases, the required number of specimens also increases by at least a factor of three (Saila and Martin, 1987). One subjective criterion for the selection of distances and networks is the visual resemblance of the network to the specimen from which it was derived. For example, the network in Figure 7-3 resembles the specimen somewhat, especially in the head region, but body depth and posterior tail shape is not represented as accurately. As described for landmark 5, few homologous landmarks can be found along the dorsal or anal ﬁns, and measuring yellowtail ﬂounder shape is a compromise between homology and comprehensive depiction. In addition to linear distances between landmarks, geometric deformations from one set of landmarks to another can also be considered as morphometric characters. Traditional morphometric analyses use linear distances as correlated variables for multivariate analyses (e.g., principal components analysis, cluster analysis, discriminant function analysis). However, the geometry of linear dis-

Morphometric Landmarks

159

FIGURE 7-3. A box-truss network of linear distances among morphometric landmarks for yellowtail ﬂounder.

tances (i.e., the orientation and proximity of line segments) is ignored by traditional analyses. Geometric methods were developed to quantify shape variation as a distribution of deformations from an average shape (Rohlf and Marcus, 1993). For example, sexual dimorphism of yellowtail ﬂounder can be viewed as a deformation from an average male to an average female (Fig. 7-4). Deformations can be measured using thin-plate spline analysis (Bookstein, 1991; Rohlf, 1998), and the resulting partial warps, composite measures of shape variation, can be used as morphometric characters for multivariate analysis. Note that landmarks on articulated structures, such as ﬁns and jaws, cannot be considered in thin-plate spline analysis because geometric positions relative to other landmarks are not ﬁxed, but methods have been developed to consider the geometry of articulating landmarks (Adams, 1999). Although traditional morphometric analyses are common for stock identiﬁcation (Fig. 7-1), few studies have used geometric characters, such as partial warps, to investigate geographic variation in morphometry (Corti and Crosetti, 1996; Walker, 1996, 1997; Sheehan et al., 2004; Cadrin and Silva, 2004). Perhaps geometric methods have not been widely

160

Steven X. Cadrin

Female

Male

FIGURE 7-4. Sexual dimorphism of yellowtail ﬂounder, viewed as a thin-plate spline deformation. For comparison to traditional morphometric analysis, linear distances that loaded strongly positive in discriminant function analysis are indicated with solid bold lines, and those that loaded negatively are indicated with dashed bold lines.

applied because biological interpretation of partial warps is difﬁcult (Rohlf, 1998). Therefore, Rohlf and Marcus (1993) suggest that partial warps be used to locate dimensions that vary among groups for selection of characters for traditional analysis of linear distances. For example, if the depth of the head is the principal deformation between two putative stocks, linear head depth should be included in a traditional multivariate analysis.

C. STATISTICAL ANALYSIS Morphometric characters are continuous variables with meaningful correlations and are therefore appropriate for conventional multivariate analysis (Blackith and Reyment, 1971; Pimentel, 1979; Reyment et al., 1984; Marcus, 1990; Klingenberg, 1996). As with all proper statistical analysis, routine descriptive diagnostics should be conducted to identify statistical outliers, assess normality (or lognormality), and inspect linearity (or log-linearity) of correlations. Principal components analysis (PCA) is a valuable diagnostic and exploratory tool. On the basis of multidimensional growth, correlation among log-transformed distances and resulting principal components can be interpreted in terms of isometric size variation and allometric shape variation (Teissier, 1960; Jolicoeur, 1963; Reyment, 1990; Klingenberg, 1996). If all characters are positively correlated and load nearly equally on the ﬁrst principal component (i.e., are all simi-

Morphometric Landmarks

161

larly correlated with the PC1 score), PC1 can be interpreted as isometric size and scales the relative size of specimens. Differences in size distribution among putative stocks may result from growth or mortality differences among areas and need to be considered in discriminations so that classiﬁcation is based on shape differences rather than size differences. For example, a recent study that discriminated sturgeon species was found to be invalid because it incorrectly classiﬁed individuals to species based on size (Rinçon, 2000). Several methods of multivariate size correction have been developed, but Burnaby’s (1966) method, which involves the removal of within-group multivariate size, appears to be the most appropriate (Rohlf and Bookstein, 1987; Klingenberg, 1996). The second principal component accounts for the maximum amount of variation remaining after isometric size variance is removed by PC1, and therefore measures shape variation. Within-group PCA is an effective method for detecting statistical outliers from processing errors or abnormal morphometric development. Morphometric measurements should be repeated for all specimens that are considered to be statistical outliers according to their shape component scores (e.g., PC2 scores) to eliminate processing errors (thus emphasizing the advantages of archiving images). If reprocessed measurements conﬁrm that the outlier specimens have a signiﬁcantly different shape, the nature of the difference should be determined. For example, if a specimen is somehow mutilated, it can be removed from the analysis without biasing results. However, if the specimen represents an extreme of regular natural variation, alternative data transformations should be considered to normalize the distribution and retain the observation to represent natural variation. Pooled-group PCA is a powerful exploratory tool for examining patterns of morphometric variation and choosing character sets that may efﬁciently discriminate groups. Individual attributes and location information collected during sampling can be used as categories for labeling observations in PC score plots to illustrate group membership. Attributes that cluster together can guide subsequent analytical designs. For example, if observations cluster by sex, the signiﬁcance of sexual dimorphism should be tested, and sex should be considered as a covariate for detecting geographic differences or stock discriminations should be separate for each sex. Group differences can be easily interpreted from PC loadings. Characters that load strongly positive or strongly negative on PC2 have large inﬂuence and reveal shape contrasts. In the simplistic example of “boxﬁsh” illustrated in Figure 7-5, PC1 accounts for size (i.e., both body length and body depth load positively on PC1) with large boxﬁsh having high PC1 scores. In the boxﬁsh example, PC2 contrasts length and width (i.e., body depth loads strongly positive and length strongly negative), and PC2 scores distinguish long-narrow boxﬁsh from short-wide ones (Fig. 7-5). Cluster analysis is also useful for exploring patterns of shape variation in size-adjusted data, but interpreting group differences is more difﬁcult than from PCA results.

162

Steven X. Cadrin

3

2

PC2 Score (shape)

1

0

-1

-2

-3 -3

-2

-1

0

1

2

3

PC1 Score (size)

FIGURE 7-5. Schematic principal components analysis of a sample of “boxﬁsh,” illustrating the interpretation of PC1 as an index of size and PC2 as an index of shape.

Signiﬁcance of morphometric differences among putative stocks is commonly tested using multivariate analysis of variance or discriminant function analysis. However, multivariate tests with a large number of morphometric characters and many observations are extremely sensitive, and statistical signiﬁcance may be spurious (Misra and Easton, 1999). A more meaningful criterion for detecting differences is the ability of a discriminant function to classify extrinsic specimens to the correct stock with greater accuracy than random classiﬁcation (Solow, 1990). Interpretation of discriminant function results is more difﬁcult than PCA. Pooled within-groups correlations between variables and discriminant scores can be used to interpret canonical variates, similar to the way PCA loadings are interpreted. Plotting truss networks or thin-plate spline deformations as canonical

163

Morphometric Landmarks 5

G ra nd B an k

4

A

S t.P ierre B an k N ortheas t U .S .

3

CV 2 S core

2 1 0 -1 -2 -3 -4 -5

-4

-3

-2

-1

0

1

2

3

4

5

CV 1 S core

5

B

4 3

CV2 Score

2 1 0 -1 -2 -3 -4 -5

-4

-3

-2

-1

0

1

2

3

4

5

CV1 Score

FIGURE 7-6. Geographic variation of female yellowtail ﬂounder as illustrated by canonical variates scores of size-adjusted morphometric data (A), and thin-plate splines for four extreme examples. (B) dashed lines indicate features that load negatively; solid lines indicate features that load positively.

variate scores can also help to interpret discriminant functions (Sheehan et al., 2004; Cadrin and Silva, 2004). For example, plotting the partial warps for yellowtail ﬂounder with extreme CV1 scores shows that Newfoundland females have relatively longer snouts and shorter bodies than those off the northeastern United States (Fig. 7-6).

164

Steven X. Cadrin

III. INTERPRETATION OF MORPHOMETRIC DIFFERENCES Although image analysis techniques and geometric analyses enhance descriptions of morphometric variation, results remain equivocal with respect to genetic or environmental bases of morphometric differences. However, if morphometry is locally adaptive, morphometric analyses can suggest the existence of discrete genetic stocks if supplemented with functional tests and rearing experiments. “Common garden experiments” can be used to partition morphometric variance into environmental and genetic variation. Adaptation involves natural selection of characters that improve survival and reproduction of individuals. Local environments vary, selecting for different characters in different areas. Differences in selected characters are maintained through reproductive isolation among groups. If morphometric variation among stocks is adaptive to different environments, it is more likely to indicate genetic stocks. However, adaptive interpretations require explicit functional hypotheses and tests (Homberger 1988), and common environment rearing experiments are needed to determine heritability (Swain and Foote, 1999, Chapter 4, this volume).

A. SALMON CASE STUDIES Several studies of morphometric variation among salmon stocks illustrate how hypotheses about the functional utility of phenotypic variation can be tested. Such case studies complement morphometric stock identiﬁcation and serve as models for the advancement of morphometric analysis. Anadromous salmon inhabit a wide range of aquatic environments, from relatively calm spawning beds, to river rapids, to pelagic waters of the open ocean. Accordingly, they exhibit a range of swimming behavior including prolonged swimming (“subcarangiform periodic axial undulation”) and burst acceleration (“transient axial undulation,” Webb and Blake, 1985). In Webb’s (1984) continuum of functional morphology for swimming form and function, salmon are intermediates between generalists and speed specialists. As will be demonstrated, species do not occupy a single point on the morphology continuum but demonstrate a variety of behavior (i.e., a range of intraspeciﬁc points in the continuum), with some individuals more specialized for speed and others more specialized for acceleration. 1. Atlantic Salmon Riddell and Leggett (1981) found that body morphology of Atlantic salmon varied between two tributaries of the Miramichi River, New Brunswick. Parr from Rocky Brook had more fusiform bodies (i.e., less robust and longer head length) and larger pelvic and pectoral ﬁns than those from the Sabbies River. They hypothe-

Morphometric Landmarks

165

sized that morphometric differences were adaptive because salmon from Rocky Brook had higher energetic costs from overwintering in the stream as compared to those from the Sabbies River, which leave the river in spring. General body shape was measured with “ﬁneness ratio” (body length: maximum body diameter), so that a greater ratio indicates a more fusiform body. The difference in ﬁneness ratio between tributaries was only 5%, but was speculated to result in substantial differences in energetic costs associated with feeding and maintaining position. Salmon parr maintain position on river beds by extending their pelvic and pectoral ﬁns, which act as hydrofoils, creating greater water pressure over the ﬁns and forcing the body downward against the substrate. The larger paired ﬁn size of Rocky Brook salmon is capable of generating more negative lift because lift is proportional to surface area of ﬁns. Riddell and Leggett (1981) tested their hypothesis by comparing observed morphometrics to ﬂow-predicted morphology of Atlantic salmon from two other tributaries in the Miramichi River system as well as two tributaries of the Big Salmon River, based on their moderate and fast ﬂows. As expected, specimens from the Big Salmon River were signiﬁcantly more streamlined and had signiﬁcantly larger ﬁns than those from tributaries with more moderate ﬂow. The accuracy of morphometric predictions from ﬂow information strongly suggests that morphometric differences are adaptive, especially because they were tested on two very different river systems (the Miramichi ﬂows into the Gulf of Saint Lawrence and the Big Salmon River ﬂows into the Bay of Fundy). Riddell et al. (1981) conducted breeding experiments to determine if the different morphologies from different tributaries had a genetic basis. They sampled eggs from each tributary and reared them over two years in controlled environments. They found that ﬁsh from Rocky Brook retained more fusiform bodies and larger ﬁns than those from the Sabbies River. Therefore, they concluded that morphometric characters were heritable and differences among populations were adaptive. Similar morphometric patterns were also found in other comparisons among Atlantic salmon groups. Atlantic salmon from the Machias River also had more fusiform bodies and deeper caudal peduncles than those from nearby rivers with lower ﬂow and elevations (“machias” is a native American word meaning “bad little falls”), and morphometric differences were maintained in common environment experiments (Sheehan et al., 2004). 2. Coho Salmon Taylor and McPhail (1985a) compared morphometrics of juvenile coho salmon from interior portions of the Fraser River to those from coastal streams and found that interior ﬁsh were more streamlined, having longer heads, shallower bodies, and narrower peduncles. They hypothesized that the more fusiform bodies of

166

Steven X. Cadrin

interior ﬁsh were adapted for prolonged swimming performance, and the more robust bodies of coastal ﬁsh were adapted for high burst performance. They collected eggs from each stream and cultured them in controlled environments in rearing troughs to ﬁnd that morphometric differences were retained in the laboratory. Taylor and McPhail (1985b) used two separate tank experiments to test their functional hypotheses. Burst performance was tested in a tank equipped with a high-speed video camera and electrical stimulus, and comparisons were made between similarly sized ﬁsh from coastal and interior brood stock. Prolonged swimming performance was tested in an oval tank with a ﬁxed-velocity tube and downstream electrical ﬁeld; swimming was timed from the onset of ﬂow to time to impingement on a downstream screen to measure time to fatigue. Results showed that coastal juveniles (with robust bodies) attained greater burst velocities, and interior juveniles (with streamlined bodies) had greater swimming stamina at several ﬂow rates. These patterns conﬁrm the general morphologies associated with fast-start bursts and prolonged fast swimming (Webb, 1984; Webb and Blake, 1985): the large girth of robust ﬁsh adds inertia and minimizes the energy lost to lateral recoil produced from burst acceleration (i.e., C-start transient oscillation), and narrow peduncles of streamlined ﬁsh reduce surface area in the most posterior portion of the body where drag is greatest during prolonged, periodic oscillation. Taylor and McPhail (1985b) then used reared ﬁsh in the same tank experiments that were conducted on wild ﬁsh (described above) and found the same patterns: Coastal ﬁsh had greater burst velocities, and interior ﬁsh had greater swimming stamina. Therefore, they once again concluded that the morphometric differences were heritable and adapted to local environments. Streamlined bodies appear to be adaptive for long migrations to inland streams, and robust bodies appear to be adaptive for avoiding predators, which are much more abundant in coastal streams than interior streams. Hale (1999) also found that C-start burst performance is maximal for small coho salmon, which are the most vulnerable to predation. In a similar case study, Swain and Holtby (1989) found that juvenile coho salmon reared in lakes had more posteriorly placed pectoral ﬁns, more streamlined bodies, and smaller dorsal and anal ﬁns than those reared in streams, and differences persisted in common environment experiments. They concluded that differences were adaptive, providing greater prolonged swimming performance, citing the functional conclusions of Taylor and McPhail (1985b). 3. Other Salmon Most salmon species have generally similar life histories, similar morphology, and presumably similar locally selective forces. For example, Hale (1999) found

Morphometric Landmarks

167

similar swimming kinematics for juvenile coho salmon, Chinook salmon, and brown trout. Riddell and Leggett (1981) reported that the patterns of body form and ﬁn size with respect to ﬂow rate were also found in brown trout. Winans (1984) speculated that morphometric differences observed in chinook salmon may be adaptive to migration differences. Taylor (1991) found that pink and chum salmon demonstrate similar patterns to those previously described for Atlantic and coho salmon, that is, salmon from higher ﬂow environments were more fusiform. Beacham (1985) found that pink salmon from large rivers in British Columbia and Puget Sound generally had larger heads, peduncles, and ﬁns than those from small rivers and speculated that differences were adapted to greater water velocity in large rivers. Beacham et al. (1988b) expanded the sampling to more rivers and analyzed allozymes to conﬁrm the morphometric differences between small and large rivers. They concluded that morphology was locally adaptive. In a similar series of ﬁeld studies, Beacham (1984) found that chum salmon from large rivers in British Columbia also had longer ﬁns, larger heads, and thicker peduncles than those from small rivers. He concluded that morphometric differences were adapted to differences in water ﬂow because larger ﬁns provide greater lift and wider peduncles provide greater propulsion for migrating long distances. Furthermore, he proposed that such adaptive morphology explains homing of salmonids. Beacham and Murray (1987) sampled chum salmon from more rivers in British Columbia and conﬁrmed the morphometric differences among ﬁsh from large and small rivers. They also repeated the speculation that migrating long distances required more muscular peduncles. The speculations about robust bodies being more adapted to long migrations from case studies on pink and chum salmon are contrary to the ﬁndings of Taylor and McPhail (1985b), who found that more streamlined bodies provide greater swimming endurance. The distinction between the two case studies is that the functional hypotheses for Atlantic and coho salmon were tested, whereas the hypotheses for pink and chum salmon were not (Beacham, 1984, 1985; Beacham and Murray, 1987; Beacham et al., 1988b). Interestingly, Beacham et al. (1988a) compared morphometry of chum salmon from British Columbia and from the Yukon, an extremely large river system, and found that Yukon ﬁsh were more streamlined than British Columbia ﬁsh. Although Beacham et al. (1988b) speculated that wide peduncles of pink salmon were adapted to greater distances of upstream migration, the same authors (Beacham et al., 1988a) speculated that the narrow peduncles observed for Yukon ﬁsh were adapted for long-distance migrations. Clearly, adaptive hypotheses can easily be proposed for any morphometric pattern. Although adaptive explanations are attractive, they do not support morphometric stock identiﬁcation unless they are rigorously tested. Taylor (1991) reviewed the issue of local adaptation in salmon stocks and discussed implications of these case studies for many aspects of ﬁsheries science.

168

Steven X. Cadrin

For conservation biology, functional interpretations facilitate the identiﬁcation of all genetic components to maintain minimum viable populations. For ﬁshery management, self-sustaining resources can be delineated for setting separate catch limits, seasons, and so on for each stock. For population enhancement, managers can ensure that streams are stocked with spawners that are from the same genetic stock, or at least they can stock one that has morphometric attributes necessary to survive and reproduce in the target environment. For population restoration (i.e., reestablishing salmon runs in extirpated rivers) morphology can be used for “population matching” so that a parental stock is chosen that is most ﬁt for the physical conditions of the extirpated river. The case studies just reviewed illustrate how practical aspects of functional morphology (e.g., working hypotheses and tests) and rearing experiments complement morphometric stock identiﬁcation. Functional hypotheses and associated tests lend interpretability to morphometric variation with respect to local adaptation and temporally stable differences and thereby increase the value of morphometric analysis for stock identiﬁcation in an interdisciplinary approach. Such integration of functional morphology, quantitative genetics, and morphometric analysis is surely applicable to many other ﬁshery resources. For example, any populations that are adapted to different ﬂow regimes should exhibit similar morphometric differences. Also, many other functional hypotheses such as variation in feeding morphometry or secondary sex characters may enhance morphometric stock identiﬁcation. For example, local groups of arctic charr (Salvelinus alpinus) that have different feeding habits have different mouth shapes (Skulason et al., 1989), and different morphs are genetically different (Skulason et al., 1996). Morphometric research is more biologically meaningful if coupled with functional hypotheses regarding the adaptive signiﬁcance of differences in body shape. Such synthesis is needed to further develop the advancement of morphometric research. However, the more formalized methodology described by Homberger (1988), involving the construction of structural and functional models, is necessary to ensure accurate interpretations. As stated by Taylor and McPhail (1985b), most case studies of morphometric stock identiﬁcation merely report differences among groups, with no functional interpretation. Such case studies have limited value for determining separate stocks because differences may not persist over time or may not be biologically signiﬁcant. Many other morphometric case studies pose functional hypotheses or speculate on the adaptive value of observed differences, but do not test their hypotheses (e.g., Beacham, 1984, 1985; Beacham and Murray, 1987; Beacham et al., 1988a,b). Unfortunately, applications of functional morphology for morphometric stock identiﬁcation (as illustrated by the case studies on Atlantic and coho salmon) are rare. The ﬁeld validation of functional predictions (e.g., Riddell and Leggett, 1981) and laboratory experiments (e.g., Taylor and McPhail, 1985b) for salmon stocks should serve as models for

Morphometric Landmarks

169

the integration of functional morphology and morphometric stock identiﬁcation. Thus, the powerful advances in morphometric analysis can have greater success in achieving the goals of stock identiﬁcation.

IV. DISCUSSION As a potential indicator of phenotypic stocks, analysis of morphometric landmarks is a valuable tool that complements other stock identiﬁcation methods. The identiﬁcation, discrimination, and delineation of phenotypic stocks are essential for population modeling, which generally assumes homogeneous ontogenetic rates within a stock. Recent reviews agree that the most comprehensive and effective strategy for stock identiﬁcation is to integrate results from disparate methods and disciplines to form conclusions about population structure that are consistent with the various approaches (Hohn, 1997; Coyle, 1998; Begg and Waldman, 1999). Carvalho and Hauser (1994) concluded that simultaneous collection of molecular genetic data and phenotypic information is crucial for stock structure analysis. Although the use of morphometric landmark characters to identify phenotypic stocks is more than a century old, imaging technology and analytical techniques have increased the power of morphometric analysis for stock discrimination and stock composition analysis. The development of affordable digital cameras and storage capacity offers the potential for collecting many images of specimens during routine ﬁshery and research sampling. The technological advances and subsequent methodological advances that occurred in the past decade are expected to continue, enhancing the discriminatory and interpretive power of landmark morphometrics for stock identiﬁcation.

ACKNOWLEDGMENTS I thank Kevin Friedland, Robert Edgar, Michael Armstrong, and Timothy Sheehan for our collaborations, and I appreciate their inﬂuences on my perspectives toward morphometric analysis. Cheryl Wilga was instructive in my understanding of functional morphology. Steve Murawski provided helpful comments on the draft manuscript. I also thank Kevin Friedland for the invitation to contribute to the work of the ICES Stock Identiﬁcation Methods Working Group.

REFERENCES Adams, D. C. 1999. Methods for shape analysis of landmark data from articulated structures. Evol. Ecol. Res. 1: 959–970. Albertson, R. C. and Kocher, T. D. 2001. Assessing morphological differences in an adaptive trait: a landmark-based morphometric approach. J. Exp. Zool. 289: 385–403.

170

Steven X. Cadrin

Armstrong, M. P. and Cadrin, S. X. 2001. Morphometric patterns within and among spawning aggregations of Atlantic herring (Clupea harengus) off the northeast United States. In F. Funk, J. Blackburn, D. Hay, A. J. Paul, R. Stephenson, R. Toressen, and D. Witherell (eds.), Herring 2000. Alaska Sea Grant Report AK-SG-2000-01, pp. 575–590. Beacham, T. D. 1984. Age and morphology of chum salmon in southern British Columbia. Trans. Am. Fish. Soc. 113: 727–736. Beacham, T. D. 1985. Meristic and morphometric variation in pink salmon (Oncorhynchus gorbuscha) in southern British Columbia and Puget Sound. Can. J. Zool. 63: 366–372. Beacham, T. D. and Murray, C. B. 1987. Adaptive variation in body size, age, morphology, egg size, and developmental biology of chum salmon (Oncorhynchus keta) in British Columbia. Can. J. Fish. Aquat. Sci. 44: 244–261. Beacham, T. D., Murray, C. B., and Withler, R. E. 1988a. Age, morphology, developmental biology, and biochemical genetic variation of Yukon river fall chum salmon, Oncorhynchus keta, and comparisons with British Columbia populations. Fish. Bull. 86: 663–674. Beacham, T. D., Withler, R. E., Murray, C. B., and Barner, A. L. 1988b. Variation in body size, morphology, egg size, and biochemical genetics of pink salmon in British Columbia. Trans. Am. Fish. Soc. 117: 109–126. Begg, G. A. and Waldman, J. R. 1999. An holistic approach to ﬁsh stock identiﬁcation. Fish. Res. 43: 35–44. Blackith, R. E. and Reyment, R. A. 1971. Multivariate Morphometrics. Academic Press, London, UK. 71 pp. Booke, H. E. 1981. The conundrum of the stock concept—are nature and nurture deﬁnable in ﬁshery science? Can. J. Fish. Aquat. Sci. 38: 1479–1480. Bookstein, F. L. 1990. Introduction to methods for landmark data. In F. J. Rohlf and F. L. Bookstein (eds.), Proceedings of the Michigan Morphometrics Workshop. University of Michigan Museum of Zoology Special Publication 2, pp. 215–226. Bookstein, F. L. 1991. Morphometric Tools for Landmark Data. Cambridge University Press, Cambridge, UK. 435 pp. Bookstein, F. L., Chernoff, B., Elder, R. L., Humphries, J. M., Smith, G. R., and Strauss, R. E. 1985. Morphometrics in evolutionary biology, the geometry of size and shape change with examples from ﬁshes. Acad. Natl. Sci. Philadelphia Spec. Pub. 15, 277 pp. Burnaby, T. P. 1966. Growth-invariant discriminant functions and generalized distances. Biometrics 22: 96–110. Cadrin, S. X. 1995. Discrimination of American lobster stocks off southern New England based on secondary sex character allometry. Can J. Fish. Aquat. Sci. 52(12): 2712–2723. Cadrin, S. X. 2000. Advances in morphometric analysis of ﬁsh stock structure. Rev. Fish Biol. Fish. 10: 91–112. Cadrin, S. X. and Friedland, K. D. 1999. The utility of image processing techniques for morphometric analysis and stock identiﬁcation. Fish. Res. 43: 129–139. Cadrin, S. X. and Silva, V. M. 2004. Morphometric variation of yellowtail ﬂounder. J. Northwest Atl. Fish. Sci. (in review). Carvalho, G. R. and Hauser, L. 1994. Molecular genetics and the stock concept in ﬁsheries. Rev. Fish. Bio. Fish. 4: 326–350. Corti, M. and Crosetti, D. 1996. Geographic variation in the grey mullet Mugil cephalus (Pisces: Mugilidae): a geometric morphometric analysis using partial warp scores. J. Fish Biol. 48: 255–269. Cole, L. C. 1954. The population consequences of life history phenomena. Quart. Rev. Biol. 29: 103–137. Coyle, T. 1998. Stock identiﬁcation and ﬁsheries management: the importance of using several methods in a stock identiﬁcation study. In D. A. Hancock (ed.), Taking Stock: Deﬁning and Managing Shared Resources. Australian Society for Fishery Biology, Sydney, pp. 173–182.

Morphometric Landmarks

171

Garrod, J. D. and Horwood, J. W. 1984. Reproductive strategies and the response to exploitation. In G. W. Potts and R. J. Wootton (eds.), Fish Reproduction. Academic Press, New York, pp. 367–384. Gould, S. J. and Johnston, R. F. 1972. Geographic variation. Ann. Rev. Ecol. Syst. 3: 457–498. Hale, M. 1999. Locomotor mechanics during early life history: effects of size and ontogeny on faststart performance of salmonid ﬁshes. J. Exp. Biol. 202: 1465–1479. Hohn, A. A. 1997. Design for a multiple-method approach to determine stock structure of bottlenose dolphins in the mid-Atlantic. NOAA Tech. Mem. NMFS-SEFSC 401. Holtby, L. B., Swain, D. P., and Allan, G. M. 1993. Mirror-elicited agonistic behaviour and body morphology as predictors of dominance status in juvenile coho salmon (Oncorhynchus kisutch). Can. J. Fish. Aquat. Sci. 50: 676–684. Homberger, D. G. 1988. Models and tests in morphology: the signiﬁcance of description and integration. Am. Zool. 28: 217–229. Huxley, J. S. 1932. Problems of Relative Growth. Dial Press, New York. 276 pp. Jolicoeur, P. J. 1963. The multivariate generalization of the allometric equation. Biometrics 19: 497–499. Klingenberg, C. P. 1996. Multivariate allometry. In L. F. Marcus, M. Corti, A. Loy, G. J. P. Naylor, and D. E. Slice (eds.), Advances in Morphometrics. NATO ASI Series A: Life Sci. 284: 23–49. Lee, P. J. 1971. Multivariate analysis for the ﬁsheries biology. Fish. Res. Bd. Can. Tech. Rep. No. 244: 1–182. Marcus, L. F. 1990. Traditional morphometrics. In F. J. Rohlf and F. L. Bookstein (eds.), Proceedings of the Michigan Morphometrics Workshop. University of Michigan Museum of Zoology Special Publication. 2, pp. 77–122. Misra, R. K. and Easton, M. D. L. 1999. A note on the number of morphometric characters used in ﬁsh stock delineation studies employing a MANOVA. Fish. Res. 42: 191–194. Pimentel, R. A. 1979. Morphometrics, the Multivariate Analysis of Biological Data. Kendall Hunt, Dubuque, Iowa. 276 pp. Reyment, R. 1990. Reiﬁcation of classical multivariate analyses in morphometry. In F. J. Rohlf and F. L. Bookstein (eds.), Proceedings of the Michigan Morphometrics Workshop. University of Michigan Museum of Zoology Special Publication. 2, pp. 123–144. Reyment, R., Blackith, R. E., and Campbell, N. A. 1984. Multivariate Morphometrics, 2nd ed. Academic Press, London, UK. 232 pp. Riddell, B. E. and Leggett, W. C. 1981. Evidence of an adaptive basis for geographic variation in body morphology and time of downstream migration of juvenile Atlantic salmon (Salmo salar). Can. J. Fish. Aquat. Sci. 38, 308–320. Riddell, B. E., Leggett, W. C., and Sanders, R. R. 1981. Evidence of an adaptive polygenic variation between two populations of Atlantic salmon (Salmo salar). Can. J. Fish. Aquat. Sci. 38, 321– 333. Rinçon, P. A. 2000. Big ﬁsh, small ﬁsh: still the same species. Lack of morphometric evidence of the existence of two sturgeon species in Guadalquivir River. Mar. Biol. 136: 715–723. Rohlf, F. L. 1998. On applications of geometric morphometrics to studies on ontogeny and phylogeny. Syst. Biol. 47: 147–158. Rohlf, F. J. and Bookstein, F. L. 1987. A comment on shearing as a method for “size correction.” Syst. Zool. 36: 356–367. Rohlf, F. L. and Marcus, L. F. 1993. A revolution in morphometrics. Trends Ecol. Evol. 8: 129–132. Royce, W. F. 1957. Statistical comparison of morphological data. In J. C. Marr (ed.), Contributions to the Study of Subpopulations of Fishes. U.S. Fish and Wildlife Serv. Spec. Sci. Rep. Fisheries 208, pp. 7–28. Saila, S. B. and Flowers, J. M. 1969. Geographic morphometric variation in American lobster. Syst. Zool. 18, 330–338.

172

Steven X. Cadrin

Saila, S. B. and Martin, B. K. 1987. A brief review and guide to some multivariate methods for stock identiﬁcation. In H. E. Kumpf, R. N. Vaught, C. B. Grimes, A. G. Johnson, and E. L. Nakamura (eds.), Proceedings of the Stock Identiﬁcation Workshop. NOAA Tech. Mem. NMFS-SEFC 199, pp. 149–175. Sheehan, T. F., Kocik, J. F., Atkinson, E., Cadrin, S. X., Legault, C., Atkinson, E., and Bengston, D. 2004. Marine growth and morphometrics for three populations of Atlantic salmon from eastern Maine, USA. Trans. Am. Fish. Soc. (in review). Skulason, S., Noakes, D. L., and Snorranson, S. S. 1989. Ontogeny of trophic morphology in four sympatric morphs of Arctic charr Salvelinus alpinus in Thingvallavatn, Iceland. Biol. J. Linn. Soc. 38: 281–301. Skulason, S., Snorranson, S. S., Noakes, D. L. G., and Ferguson, M. M. 1996. Genetic variation of life history variations among sympatric morphs of Arctic charr Salvelinus alpinus. Can. J. Fish. Aquat. Sci. 53: 1807–1813. Solow, A. 1990. A randomization test for misclassiﬁcation probability in discriminant analysis. Ecology 71: 2379–2382. Strauss, R. E., and Bookstein, F. L. 1982. The truss: body form reconstructions in morphometrics. Syst. Zool. 31: 113–135. Swain, D. P. and Foote, C. J. 1999. Stocks and chameleons: the use of phenotypic variation in stock identiﬁcation. Fish. Res. 43: 113–128. Swain, D. P. and Holtby, L. B. 1989. Differences in morphology and behavior between juvenile coho salmon (Oncorhynchus kisutch) rearing in a lake or in its tributary stream. Can. J. Fish. Aquat. Sci. 46: 1406–1414. Tabachnick, B. G. and Fidell, L. S. 1989. Using Multivariate Statistics. Harper Row & Collins, New York. 746 pp. Taylor, E. B. 1991. A review of local adaptation in Salmonidae, with particular reference to Paciﬁc and Atlantic salmon. Aquaculture 98: 185–207. Taylor, E. B. and McPhail, J. D. 1985a. Variation in body morphology among British Columbia populations of coho salmon, Oncorhynchus kisutch. Can. J. Fish. Aquat. Sci. 42: 2020–2028. Taylor, E. B. and McPhail, J. D. 1985b. Variation in burst and prolonged swimming performance among British Columbia populations of coho salmon, Oncorhynchus kisutch. Can. J. Fish. Aquat. Sci. 42, 2029–2033. Teissier, G. 1960. Relative growth. In T. H. Waterman (ed.), The Physiology of Crustacea. Academic Press, New York, pp. 537–560. Thorpe, R. S. 1976. Biometric analysis of geographical variation and racial afﬁnities, Biol, Rev. 51: 407–452. Thorpe, R. S. 1988. Multiple group principal components analysis and population differentiation. J. Zool. Lond. 216: 37–40. Walker, J. A. 1996. Principal components of body shape variation within an endemic radiation of threespine stickleback. In L. F. Marcus, M. Corti, A. Loy, G. J. P. Naylor, and D. E. Slice (eds.), Advances in Morphometrics. NATO ASI Series A: Life Sciences 284, pp 321–334. Walker, J. A. 1997. Ecological morphology of lacustrine threespine stickleback Gasterosteus aculeatus L. body shape. Biol. J. Linn. Soc. 61: 3–50. Webb, P. W. 1984. Form and function in ﬁsh swimming. Sci. Am. 251: 72–82. Webb, P. W. and Blake, R. W. 1985. Swimming. In M. Hildebrand, D. M. Bramble, K. F. Liem, and D. B. Wake (eds.), Functional Vertebrate Morphology. Harvard University Press, Cambridge, MA, pp. 110–128. Winans, G. A. 1984. Multivariate morphometric variability in Paciﬁc salmon: technical demonstration. Can. J. Fish. Aquat. Sci. 41: 1150–1159. Winans, G. A. 1987. Using morphometric and meristic characters for identifying stocks of ﬁsh. In H. E. Kumpf, R. N. Vaught, C. B. Grimes, A. G. Johnson, and E. L. Nakamura (eds.), Proceedings of the Stock Identiﬁcation Workshop. NOAA Tech. Mem. NMFS-SEFC 199, pp. 135–146.

CHAPTER

8

Morphometric Outlines STEVEN X. CADRIN* AND KEVIN D. FRIEDLAND† *National Marine Fisheries Service, Woods Hole, Massachusetts, USA, †UMass/NOAA Cooperative Marine Education and Research Program, University of Massachusetts, Amherst, Massachusetts, USA

I. Introduction II. Methods A. Image Processing B. Statistical Model Fitting C. Multivariate Analysis III. Interpretation IV. Case Studies in Stock Identiﬁcation V. Discussion References

I. INTRODUCTION The study of shape variation has advanced from measuring simple linear distances to deriving geometric variables. The shift from traditional morphometrics to more complex geometric functions was facilitated by the development of image processing tools (Cadrin and Friedland, 1999). Recently developed geometric approaches to morphometric analysis are generally categorized as either “landmark methods” (this volume, Chapter 7) or “outline methods” (Bookstein et al., 1985; Rohlf and Bookstein, 1990; Rohlf and Marcus, 1993; Marcus et al., 1996). Geometric outline methods quantify boundary shapes so that patterns of shape variation within and among groups can be evaluated. Patterns of variation in the shape of ﬁsh scales and otoliths as well as mollusc valves have been used to identify phenotypic stocks. Prior to the development of outline methods, researchers noted that otolith shape varied among stocks. For example, Messieh (1972) described two different “types” of otoliths between spring and autumn spawning herring. However, efﬁcient classiﬁcation of ﬁsh to the correct stock requires quantiﬁcation of shape and its variation. The most common outline method involves ﬁtting a Fourier series to the point coordinates along the perimeter of a morphometric feature (Jarvis et al., 1978). Fourier Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

173

174

Steven X. Cadrin and Kevin D. Friedland

coefﬁcients are commonly used as multivariate observations for discriminant analysis, and several studies have successfully used Fourier transforms of scale or otolith shape for group discrimination of ﬁnﬁsh stocks.

II. METHODS The general approach of geometric outline methods is (A) capturing the outline of a structure, (B) ﬁtting a geometric model to concisely describe the outline, and (C) multivariate analysis of model parameter estimates to detect patterns of variance, discriminate groups, and classify individuals to groups.

A. IMAGE PROCESSING The development of hardware and software for image analysis enhanced the capability of archiving outline shapes. Digital images of hard structures (e.g., scales, otoliths, valves) are usually recorded with video cameras, and more recently digital cameras, mounted on microscopes. Image processing tools are used to enhance contrast of the image, transforming gray-scale images to blackand-white silhouettes of the structure. Image analysis software includes search algorithms to trace the outline, deriving Cartesian (x,y) coordinates of outline shape. The spatially calibrated boundary shape is now easily sampled for morphometric measurements such as area and perimeter, providing useful shape indices and the outline itself, which is usually analyzed to produce derived discriminating variables. Although most stock identiﬁcation studies have utilized scales and otoliths, other hard parts (e.g., vertebrae) can also exhibit shape variation and are potential stock discriminators. Campana and Casselman (1993) found that among the types of otoliths, saggita shape variables performed best for discriminating cod stocks. Little variation has been found between right and left sagitta within individual ﬁsh (Bird et al., 1986; Castonguay et al., 1991; Campana and Casselman, 1993).

B. STATISTICAL MODEL FITTING The goal of mathematically modeling an outline is to describe the form accurately (i.e., achieving a good statistical ﬁt) with the fewest number of model parameters. Efﬁcient description of shape is critical for stock identiﬁcation because each parameter is treated as a variable for conventional multivariate analysis, and the number of samples required to detect differences increases as a function of the number of variables (Saila and Martin, 1987).

Morphometric Outlines

175

Homology, the developmental similarity of a structure, is one aspect of morphology that should be considered for describing outline shape (Rohlf and Bookstein, 1990). Most geometric outline methods require that boundary traces begin at a standard, homologous point on the outline. For example, Begg and Brown (2000) used the distal edge of otolith rostrum to start the outline trace of haddock otoliths. Some researchers also use a homologous point near the center of the outline (e.g., the focus of the scale or otolith) to derive radius functions. However, Campana and Casselman (1993) found no noticeable advantage in discriminating cod stocks by using the focus (a homologous feature) over the geometric centroid (a geometrically derived feature that may not be homologous). Most outline models require the outline to be transformed to a series of radial distances from a central point to locations on its perimeter. Each radius of the series is spaced by an equal step angle of rotation from the previous radius (Fig. 8-1). The unrolled series of radii is then treated as a truncated time series to be ﬁt by the statistical model of choice. The shape can also be expressed as tangent angle functions rather than radius functions. Several commonly used methods for ﬁtting outlines will be described. However, only Fourier analysis has been routinely used for stock identiﬁcation. For a more detailed explanation of statistically ﬁtting outlines, see Rohlf (1990).

FIGURE 8-1. Derivation of a radius function from outline coordinates.

176

Steven X. Cadrin and Kevin D. Friedland

1. Polynomials One empirical approach to representing any curvilinear function is to ﬁt a polynomial equation with enough terms to adequately describe the shape: Y = b 0 + b1 X + b 2 X 2 + . . . + b n X n A ﬁrst-degree polynomial (Y = b0 + b1X) is a linear equation, second degree is quadratic (Y = b0 + b1X + b2X2), etc. The model is ﬁt to the data using least squares, with stepwise addition of successively greater degrees from a linear equation or stepwise elimination of degrees from a multinomial equation. Polynomial splines are also used to describe outlines. These include linear splines (discrete points connected with line segments), quadratic splines (points connected with tangents), or cubic splines (points connected with ﬁrst and second derivatives of the curve). 2. Fourier Analysis Outline shapes are more commonly described by trigonometric functions of angles around a central point. Periodic functions generally use polar coordinates (r, q), rather than Cartesian coordinates, where x = r cos q, y = r sin q. Fourier functions involve periods (complete cycles), amplitudes (maximum radii), and harmonics (orthogonal components of waveform): k

q = a 0 + Â [a i cos (iq) + b i sin(iq)] i= 1

where q is the angle (in radians, ranging from 0 to 2p), and a and b are the Fourier coefﬁcients of the ith harmonic. The greater number of harmonics included in the model, the closer the ﬁt to the original shape. Lower-order harmonics are associated with general circularity and elongatedness, whereas higher-order harmonics are associated with details of the shape (Fig. 8-2). Images must be oriented similarly to compare coefﬁcients among different shapes (i.e., the angle of the starting radius, at the homologous point, must be equal among specimens). Alternatively, the equation can be rearranged to account for amplitude and phase-lags: k

q = A 0 + Â [ A i cos (iq - f i )] i= 1

where A represents amplitude and f represents phase-lag (Rayner, 1971). The solution sets are commonly calculated by an implementation of the algorithm of Wallace and Wintz (1980). An example of Fourier analysis is provided by

177

Morphometric Outlines

125

Magnitude

100

75

50

25

0

0

5

10

15

20

Harmonic FIGURE 8-2. Estimated amplitudes of each Fourier harmonic for the otolith shape illustrated in Figure 8-1.

Campana and Casselman (1993), who explained 97% to 99% of shape variance in cod otoliths using the ﬁrst ten harmonics and 99.9% of variance using the ﬁrst 20 harmonics (Fig. 8-3). 3. Elliptical Fourier Analysis An alternative method to describe outline shape is elliptical Fourier analysis, which involves the decomposition of ﬁrst differences of x,y coordinates k

x t = a 0 + Â [a i cos ( it k ) + b i sin(it k -1 )] i= 1 k

y t = c 0 + Â [c i cos ( it k ) + d i sin(it k -1 )] i= 1

where t is a distance, ranging from 0 to 2p radians. Other outline methods include median axis analysis (Straney, 1990) and eigenshape analysis (Lohmann and Schweitzer, 1990), but neither has been used for ﬁsh stock identiﬁcation.

178

Steven X. Cadrin and Kevin D. Friedland

FIGURE 8-3. Fourier reconstruction of a cod otolith. The number represents the number of harmonics used to derive the shape. The actual digitized shape is presented last [from Campana and Casselman (1993), with permission].

C. MULTIVARIATE ANALYSIS A rich selection of variables is produced from the outline shapes, including derived variables and morphometric variables from physical measurements. The variables are suited to parametric methods such as discriminant analysis as well as nonparametric classiﬁcation techniques such as decision trees. Two of the morphometric variables derived from physical measurements are rectangularity, deﬁned as the otolith area divided by the area of its minimum enclosing rectangle, and circularity, deﬁned as the perimeter squared divided by the area. The derived variables take the form of polynomial coefﬁcients, Fourier amplitudes, phase angles, or combinations of data types. As with all multivariate analyses, standard statistical diagnostics should be explored to test assumptions and detect outliers. Because it is possible to derive a large number of shape parameters, cross-validation of classiﬁcations from discriminant analysis (i.e., classiﬁcation of extrinsic specimens with known group membership) is essential for evaluating model performance (Campana and Casselman, 1993). There are limited returns in model performance from using many characters (i.e., higherorder harmonics or polynomial terms); though they tend to increase intrinsic classiﬁcation accuracy, they decrease extrinsic accuracy. An alternative to conventional multivariate analysis of Fourier coefﬁcients is contingency tables of amplitude frequency distributions using chi-square tests (Bird et al., 1986).

Morphometric Outlines

179

Similar to traditional morphometric analysis, size adjustment is a consideration for stock discrimination because classiﬁcation should be based on shape of the structure, not its size. For example, a size effect confounded stock discrimination of red snapper (Smith, 1992). Two common approaches to removing size from outline data are removing a size covariate or standardizing to a mean radius. Campana and Casselman (1993) evaluated both methods. They removed the common within-group slope of otolith length on all variables, which should remove all otolith/ﬁsh size effects. They also standardized data to the mean radius, which still had some relationships with size. However, they found no obvious differences in classiﬁcation accuracy between the two methods for removing size variance from the data. Another method of removing size is normalizing Fourier amplitudes by setting the zero harmonic component to zero and dividing the amplitude coefﬁcients by the value for the ﬁrst component, respectively (Wallace and Wintz, 1980). Other covariates that should be considered for stock identiﬁcation are age and sex. Signiﬁcant age effects have been found in otoliths of several species (Bird et al., 1986; Campana and Casselman, 1993). Age effects on otolith shape confounded discrimination of Atlantic mackerel from the Northeast and Northwest Atlantic (Castonguay et al., 1991). A signiﬁcant sex effect was found in cod (Campana and Casselman, 1993), but not in herring (Bird et al., 1986).

III. INTERPRETATION Similar to other morphometric approaches, patterns of outline shape can be used to infer phenotypic stocks. However, otolith or scale shapes are less subject to short-term variability than body shape caused by changes in feeding or spawning condition. Furthermore, otolith shape has been correlated with individual growth rate (Campana and Casselman, 1993). For example, Atlantic mackerel groups with different growth rates had the largest differences in otolith shape (Castonguay et al., 1991). Biological interpretations of Fourier coefﬁcients have been criticized because coefﬁcients may not be strictly homologous among specimens (Bookstein et al., 1982; Bookstein, 1990). However, others maintain that Fourier harmonics can be oriented by homologous landmarks, and coefﬁcients for lower-order harmonics are interpretable (e.g., the second harmonic coefﬁcient is a measure of “elongatedness”; Ehrlich et al., 1983). Biological signiﬁcance should not be interpreted directly from eigenshape functions either and can only be suggested indirectly by correlations between eigenshapes and ancillary information on group membership (Lohmann and Schweitzer, 1990).

180

Steven X. Cadrin and Kevin D. Friedland

IV. CASE STUDIES IN STOCK IDENTIFICATION A brief review of some case studies in stock identiﬁcation that use outline morphometrics illustrates the rapid advancement of both data collection and analytical systems as well as general strengths and weaknesses of the approach. Among the ﬁrst applications of outline morphometrics for stock identiﬁcation, Jarvis et al. (1978) discriminated walleye from two locations in Lake Erie according to scale shape, adopting Fourier analysis, a method previously used to identify fossil ostracodes according to their marginal shape (Kaesler and Waters, 1972). Scales were projected onto a digitizer and manually traced to derive digital coordinates of scale outlines. The ﬁrst 10 harmonics of scale outlines accounted for 95% of variance in radial lengths. The discriminant function correctly classiﬁed 80% of specimens to Seneca Shoals or Cattaraugus Falls. With the ﬁrst 20 harmonics, intrinsic classiﬁcation accuracy increased to 100%. Riley and Carline (1982) extended the method to discriminate ﬁve putative stocks of walleye, but classiﬁcation accuracy using 20 harmonics was low (57% of intrinsic samples and 25% of extrinsic samples). When stocks were grouped to western Lake Erie and Lake St. Clair, accuracy increased (82% intrinsic and 65% extrinsic), but they concluded that differences in scale shape were not sufﬁcient to discriminate stocks. Ferson et al. (1985) used the ﬁrst 10 harmonics from elliptical Fourier analysis of common mussel shells to discriminate between two electrophoretically different populations. The right valve of each mussel specimen was digitally scanned, and the resulting gray-scale images were transformed to a silhouette of black and white pixels. Outlines were derived using a line-tracing algorithm to search the transition from white to black pixels. Intrinsic classiﬁcation accuracy was 97%, and cross-validated accuracy was 74%. Scale outline features were used to discriminate groups of striped bass (Ross and Pickard, 1990; Richards and Esteves, 1997a,b). Images of scales were captured using a frame-grabber system for video that was recorded from a camera mounted on a microscope. Outlines were derived from image processing and a boundary-tracking algorithm. Ross and Pickard (1990) used the ﬁrst 8 harmonics from Fourier analysis and classiﬁed striped bass to wild or hatchery origins with 75% accuracy. Richards and Esteves (1997a and 1997b) found signiﬁcant differences in the ﬁrst 12 Fourier harmonics among two wild stocks of striped bass. Both studies (Ross and Pickard, 1990; Richards and Esteves, 1997a,b) then used Fourier harmonics in combination with other scale pattern features to classify stocks. A multidisciplinary analysis of stock identiﬁcation techniques for striped bass compared classiﬁcation accuracies from various information, including mitochondrial DNA characters, immunoassay data, meristics, landmark morphometrics, and scale outline features (Waldman et al., 1997). Their results indicate that morphometric features (including Fourier amplitudes from scale outlines) produced the most accurate classiﬁcations.

Morphometric Outlines

181

More recently, otolith shape was used to challenge the current assumptions used to manage the king mackerel ﬁshery (DeVries et al., 2002). Otolith outlines were archived with an image analysis system, and a discriminant function was derived from the standardized amplitudes of the ﬁrst 20 Fourier harmonics as well as perimeter and area. Extrinsic classiﬁcation accuracy was 71% to 78%. The discriminant function was used for stock composition analysis to determine the origin of ﬁsh caught in the 1996–1997 winter ﬁshery. Although the management plan assumed that all ﬁsh in the winter ﬁshery were from the Gulf of Mexico stock, the discriminant function indicated that 99.8% of ﬁsh sampled from the winter ﬁshery were from the Atlantic stock.

V. DISCUSSION The extent to which otolith shape differences are genetically or environmentally induced is not clear. Otolith morphology is commonly used as a species-level character in ﬁsh systematics; thus, the pattern of phenotypic expression is considered to be similar to that of other calciﬁed structures (Lagler et al., 1977). However, recent work has demonstrated that otolith morphology can vary in response to differences in growth regimes for a range of species (Reznick et al., 1989; Secor and Dean, 1989; Smith, 1992). A limited comparison made by Friedland and Reddin (1994) suggests that the greater inﬂuence on otolith shape may be genetic. In comparisons to other stock identiﬁcation methods, scale and otolith outline shape generally performed well for discriminating stocks (Casselman et al., 1981; Waldman et al., 1997). However, size and age effects commonly confound stock discrimination and require appropriate statistical treatment and interpretation (Castonguay et al., 1991; Smith, 1992). Therefore, outline morphometrics is a promising approach for stock identiﬁcation, but interpreting patterns of variance can be difﬁcult.

ACKNOWLEDGMENTS We thank Robert Edgar and Richard Strauss for their instruction on statistical methods for ﬁtting outlines. We are grateful to Steve Campana for permission to use Figure 8-3, and to Fred Serchuk for reviewing the manuscript.

REFERENCES Begg, G. A. and Brown, R. W. 2000. Stock identiﬁcation of haddock Melanogrammus aegleﬁnus on Georges Bank based on otolith shape analysis. Trans. Am. Fish. Soc. 129: 935–945.

182

Steven X. Cadrin and Kevin D. Friedland

Bird, J. L., Eppler, D. T., and Checkley, D. M., Jr., 1986. Comparisons of herring otoliths using Fourier series shape analysis. Can. J. Fish. Aquat. Sci. 43: 1228–1234. Bookstein, F. L. 1990. Introduction to methods for landmark data. In F. J. Rohlf and F. L. Bookstein (eds.), Proceedings of the Michigan Morphometrics Workshop. University of Michigan Museum of Zoology Special Publication 2, pp. 215–226. Bookstein, F. L., Strauss, R. E., Humphries, J. M., Chernoff, B., Elder, R. L., and Smith, G. R. 1982. A comment on the use of Fourier methods in systematics. Syst. Zool. 31: 85–92. Bookstein, F. L., Chernoff, B., Elder, R. L., Humphries, J. M., Smith, G. R., and Strauss, R. E. 1985. Morphometrics in Evolutionary Biology, the Geometry of Size and Shape Change with Examples from Fishes. Academy of Natural Sciences, Philadelphia, Special Publication 15. 277 pp. Cadrin, S. X. and Friedland, K. D. 1999. The utility of image processing techniques for morphometric analysis and stock identiﬁcation. Fish. Res. 43: 129–139. Campana, S. E. and Casselman, J. M. 1993. Stock discrimination using otolith shape analysis. Can. J. Fish. Aquat. Sci. 50: 162–1083. Casselman, J. M., Collins, J. J., Crossman, E. J., Ihssen, P. E., and Spangler, G. R. 1981. Lake whiteﬁsh (Coregonus clupeaformis) stocks of the Ontario water of Lake Huron. Can. J. Fish. Aquat. Sci. 38: 1772–1789. Castonguay, M., Simard, P., and Gagnon, P. 1991. Usefulness of Fourier analysis of otolith shape for Atlantic mackerel (Scomber scombrus) stock discrimination. Can. J. Fish. Aquat. Sci. 48: 296–302. DeVries, D. A., Grimes, C. B., and Prager, M. H. 2002. Using otolith shape analysis to distinguish eastern Gulf of Mexico and Atlantic Ocean stocks of king mackerel. Fish. Res. 57: 51–62. Ehrlich, R., Baxter Pharr, R., Jr., and Healy-Williams, N. 1983. Comments on the validity of Fourier descriptors in systematics: a reply to Bookstein et al. Syst. Zool. 32, 202–206. Ferson, S., Rohlf F. J., and Koehn, R. K. 1985. Measuring shape variation of two-dimensional outlines. Syst. Zoo. 34: 59–68. Friedland, K. D. and Reddin, D. G. 1994. The use of otolith morphology in stock discriminations of Atlantic salmon (Salmo salar L.). Can. J. Fish. Aquat. Sci. 51: 91–98. Jarvis, R. S., Klodowski, H. F., and Sheldon, S. P. 1978. New method of quantifying scale shape and an application to stock identiﬁcation in Walleye (Stizostedion vitreum vitreum). Trans. Am. Fish. Soc. 107: 528–434. Kaesler, R. L. and Waters, J. A. 1972. Fourier analysis of the ostracode margin. Geol. Soc. Am. Bull. 83: 1169–1178. Lagler, K. F., Bardach, J. E., Miller, R. R., and Passino, D. R. M. 1977. Ichthyology. John Wiley and Sons, New York. Lohmann, G. P. and Schweitzer, P. N. 1990. On eigenshape analysis. In F. J. Rohlf and F. L. Bookstein (eds.), Proceedings of the Michigan Morphometrics Workshop. University of Michigan Museum of Zoology Special Publication 2, pp. 147–166. Marcus, L. F., Corti, M., Loy, A., Naylor, G. J. P., and Slice, D. E. 1996. Advances in Morphometrics. NATO ASI Series A: Life Sciences 284. 587 pp. Messieh, S. N. 1972. Use of otoliths in identifying herring stocks in southern Gulf of St. Lawrence and adjacent waters. J. Fish. Res. Board Can. 29: 1113–1118. Rayner, J. N. 1971. An Introduction to Spectral Analysis. Pion, London, UK. 174 pp. Reznick, D., Linbeck, E., and Bryga, H. 1989. Slower growth results in larger otoliths: an experimental test with guppies (Poecilia reticulata). Can. J. Fish. Aquat. Sci. 46: 108–112. Richards, R. A. and Esteves, C. 1997a. Stock-speciﬁc variation in scale morphology of Atlantic striped bass. Trans. Am. Fish. Soc. 126: 908–918. Richards, R. A. and Esteves, C. 1997b. Use of scale morphology for discriminating wild stocks of Atlantic striped bass. Trans. Am. Fish. Soc. 126: 919–925. Riley, L. M. and Carline, R. F. 1982. Evaluation of scale shape for the identiﬁcation of walleye stocks. Western Lake Erie. Trans. Am. Fish. Soc. 111: 736–741.

Morphometric Outlines

183

Rohlf, F. J. 1990. Fitting curves to outlines. In F. J. Rohlf and F. L. Bookstein (eds.), Proceedings of the Michigan Morphometrics. University of Michigan Museum of Zoology Special Publication 2, pp. 167–177. Rohlf, F. J. and Bookstein, F. L. (eds.). 1990. Proceedings of the Michigan Morphometrics Workshop. University of Michigan Museum of Zoology Special Publication 2. Rohlf, F. L. and Marcus, L. F. 1993. A revolution in morphometrics. Trends in Ecology and Evolution 8: 129–132. Ross, W. R. and Pickard A. 1990. Use of scale pattern and shape as discriminators of wild and hatchery striped bass stocks in California. In N. C. Parker, A. E. Giorgi, R. C. Heidinger, D. B. Jester, E. D. Prince, and G. A. Winans (eds.), Fish-Marking Techniques. American Fisheries Society Symposium 7. American Fisheries Society, Bethesda, MD, pp. 71–77. Saila, S. B. and Martin, B. K. 1987. A brief review and guide to some multivariate methods for stock identiﬁcation. In H. E. Kumpf, R. N. Vaught, C. B. Grimes, A. G. Johnson, and E. L. Nakamura (eds.), Proceedings of the Stock Identiﬁcation Workshop. NOAA Tech. Mem. NMFS-SEFC 199, pp. 149–175. Secor, D. H., and Dean, J. M. 1989. Somatic growth effects on the otolith—ﬁsh size relationship in young pond-reared striped bass (Morone saxatilis). Can. J. Fish. Aquat. Sci. 46: 113–121. Smith, M. K. 1992. Regional differences in otolith morphology of the deep slope red snapper (Etelis carbunculus). Can. J. Fish. Aquat. Sci. 49: 795–804. Straney, D. O. 1990. Median axis methods in morphometrics. In F. J. Rohlf and F. L. Bookstein (eds.), Proceedings of the Michigan Morphometrics Workshop. University of Michigan Museum of Zoology Special Publication 2, pp. 179–200. Waldman, J. R., Richards, R. A., Schill, W. B., Wirgin, I., and Fabrizio, M. C. 1997. An empirical comparison of stock identiﬁcation techniques applied to striped bass. Trans. Am. Fish. Soc. 126: 369–385. Wallace, T. P. and Wintz, P. A. 1980. An efﬁcient three dimensional aircraft recognition algorithm using normalized Fourier descriptors. Comp. Graph. Image Processing 13: 99–126.

This Page Intentionally Left Blank

CHAPTER

9

Analyses of Calciﬁed Structures-Texture and Spacing Patterns KEVIN D. FRIEDLAND* AND STEVEN X. CADRIN† *NOAA Cooperative Marine Education and Research Program, University of Massachusetts, Amherst, Massachusetts, USA, †National Marine Fisheries Service, Woods Hole, Massachusetts, USA

I. Introduction II. Methodology A. Spacing and Texture from Optical Density Proﬁles B. Selection of Characters III. Selected Case Studies A. Salmon B. Striped Bass C. Aquaculture Releases and Escapees IV. Discussion References

I. INTRODUCTION Analysis of texture and spacing patterns found in hard body parts is a wellestablished technique of stock separation. Harden Jones (1968) described the usefulness of scale and otolith patterns for determining origin of ﬁsh and the method’s historical development. Hjort’s (1914) investigation of herring scale patterns was among the earliest identiﬁcations of distinct intraspeciﬁc groups of ﬁshes. The approach developed utilizing traditional laboratory techniques (Clutter and Whitesel, 1956; Henry, 1961; Mosher, 1963; Anas and Murai, 1969; Lear and Sandeman, 1980; Antere and Ikonen, 1983; Reddin et al., 1988; Lund and Hansen, 1991). However, newer imaging methods have greatly enhanced the approach and increased its power for stock discrimination (Cook, 1982; Barlow and Gregg, 1991; Schwartzberg and Fryer, 1989; Ross and Pickard, 1990; Friedland et al., 1994; Cadrin and Friedland, 1999). Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

185

186

Kevin D. Friedland and Steven X. Cadrin

The interpretation of texture patterns is partly dependent on the correlation between the growth of the animal and the calciﬁed structure to be analyzed. For example, ﬁsh scale growth and the rate of circuli deposition are related (Doyle et al., 1987; Barber and Walker, 1988; Fisher and Pearcy, 1990). The variation of these features throughout the species’ range, as a product of both genetic and environmental inﬂuences, allows discrimination of groups with different growth patterns. Therefore, texture patterns of calciﬁed structures relate to phenotypic stocks (Booke, 1981), which need to be deﬁned to model population dynamics. Spacing patterns can be measured with conventional measurement techniques or with enhanced technologies such as image processing. Simple features, such as the distance from one life history transition zone to another, can be recognized visually and measured manually. For example, the distance from an otolith focus to a check marking the transition from juvenile to adult life stages could serve as a stock separation character. Spacing patterns may also be a more complex set of measurements requiring the use of image processing techniques to achieve measurement accuracy and to process a large volume of data. For example, the spacing or distance from a landmark to a large number of structures can be extracted. When this approach is applied to ﬁsh scales or otoliths, the location of upward of 50 circuli may be measured, thus generating 49 spacing variables per individual. Image processing allows quick and precise measurement of many distances. For example, Cook and Guthrie (1987) found that 64,000 luminescence measures could be measured on a sockeye salmon scale in 6 sec, and with updated equipment, image handling is now even more powerful. The texture of calciﬁed body parts can be measured with physical probes or by the analysis of optical density proﬁles. The idea of developing a two- or threedimensional map of a scale or otolith using a microstylus has been considered and appears technically feasible. However, it does not appear that the approach has been attempted in stock identiﬁcation research. On the other hand, using optical density proﬁles to represent texture has been applied to stock identiﬁcation. The method was ﬁrst proposed by Major et al. (1972) and simply involves the use of transmitted light from a specimen as a representation of surface features. For example, the circuli on a ﬁsh scale appear dark because they are surface features, which are optically denser than intercirculi spaces. The periodicity of circuli can be analyzed, and spacing can be calculated from modes of luminescence. Analysis of calciﬁed structures can effectively reveal patterns in circuli spacing that correspond to the specimen’s growth history. Therefore, phenotypic stocks that have signiﬁcantly different growth rates can be discriminated using texture analysis.

Analyses of Calciﬁed Structures-Texture and Spacing Patterns

187

II. METHODOLOGY

A. SPACING

AND

TEXTURE

FROM

OPTICAL DENSITY PROFILES

Specimens should be prepared and examined with an image processor equipped with an optics subsystem appropriate for the specimen. For small specimens, this may entail the use of a compound microscope, whereas for larger specimens, the use of a dissecting microscope may be sufﬁcient. A transect from the focus to a homologous point on the margin is identiﬁed for data extraction (Fig. 9-1). The data path is selected to represent a growth phase or life history feature comparable to other specimens. From this line, a luminescence proﬁle of transmitted light is extracted which can be processed further. Manual methods or automated algorithms can be used to identify features along the data path (e.g., the location of circuli or annuli relative to the focus). The raw measurements can be saved for use as spacing indices or landmark distances. Alternatively, the luminescence proﬁle can be treated as a frequency domain series. A Fourier transform of the luminescence pattern can be calculated, and various statistics from the transform may serve as classiﬁcation model variables. For example, textural features of a scale can be expressed as the magnitudes of the Fourier transform. Other frequency domain statistics may also be used.

B. SELECTION

OF

CHARACTERS

Imaging techniques offer the investigator the ability to produce vast quantities of data. A number of investigators have begun to explore ways of reducing the number of variables generated by these sorts of analyses. One such approach has been to employ variable averaging or expressing spacing or circuli distance measures as means of pairs or quadruplets of adjacent variables. In studies using circuli spacing data, Barlow and Gregg (1991) reported that model efﬁciency was similar or only slightly higher for averaged data. In a similar study, Friedland et al. (1994) reported similar or slightly lower model classiﬁcation efﬁciency. The appeal of averaged data lies in the anticipated robustness of models to the potential problem that information content may be dispersed over a number of adjacent variables. However, there is no clear advantage to treating the data in this manner, and there may be a loss of information content when averaging is performed. Friedland et al. (1994) suggested that there may be important information in individual imaging derived variables (spacing or magnitude values) and that preprocessing of variables, like averaging, should be applied on a case-bycase basis only where it improves classiﬁcation efﬁciency. One potential source of bias from image processing techniques to extracting circuli spacing data was identiﬁed by Friedland et al. (1994). When automated

188

Kevin D. Friedland and Steven X. Cadrin

A

B

FIGURE 9-1. Plot of the luminescence proﬁle from a sockeye salmon scale (A) and (B) the same proﬁle with the circuli lactations marked based on local negative minima of luminescence. From Cook and Guthrie (1987).

procedures are used to mark a circulus and images are visually inspected for manual correction, unmarked circuli are infrequent. However, circuli are often marked more than once, and the double mark could escape manual correction. This has been observed to occur with circuli that are very wide or are of complex

Analyses of Calciﬁed Structures-Texture and Spacing Patterns

189

morphology. This error will tend to add to the total number of circuli for a specimen and decrease the circuli spacing for the adjacent circuli pairs. The direction of the bias would be to reduce circuli spacing for the specimen and miscode subsequent circuli so that comparisons with other specimens would mislead interpretations about growth differences. Texture and spacing data are continuous variables with strong correlations and are usually analyzed using conventional multivariate morphometrics [e.g., principal components analysis, cluster analysis, and discriminant analysis (Tabachnick and Fidell, 1989)]. Accordingly, exploratory diagnostics should be performed to identify outliers and test multivariate assumptions such as multivariate normality, equal variance among groups, and equal covariance among groups.

III. SELECTED CASE STUDIES

A. SALMON Several researchers have used growth patterns on scales and otoliths to discriminate groups of sockeye salmon, and the series of studies demonstrated the methodological progression from manual measurements to image analysis. In the early 1900s, Charles Gilbert used the frequency of circuli in the ﬁrst year to classify sockeye from the Fraser River to spawning location (reviewed by Secor in Chapter 3, this volume). Gilbert’s methods were reﬁned and applied widely over the next century. High-seas samples of sockeye salmon were classiﬁed to their river of origin with 72% to 95% accuracy based on the number of scale circuli in seasonal growth zones and the width of growth zones (Cook and Lord, 1977). Manual scale measurements also were used to discriminate sockeye salmon to region of origin (Bristol Bay, Gulf of Alaska, or Kamchatka Peninsula), with approximately 70% accuracy (Cook, 1982). Image processing was used by Cook (1987) to discriminate 14 races of sockeye salmon in the Fraser River, but classiﬁcation accuracy varied by year, from 35% to 100%. Waltemyer et al. (1996) also found that scale pattern variables (number of circuli, width of seasonal growth zones, and circuli distances; derived from the projection of images onto a digitizing pad) varied by year and had signiﬁcant overlap among sockeye stocks in the Upper Cook Inlet. Otolith banding patterns formed during incubation (derived from Fourier analysis of luminescence proﬁles) were used by Finn et al. (1997) to discriminate sockeye from the wild and from hatcheries, with 83% accuracy. Quinn et al. (1999) also compared otolith microstructure to estimate straying rates of sockeye from spawning sites within a river system. Much of the work being done on Paciﬁc salmonids remains in the gray literature, where it is being directly applied to management issues.

190

Kevin D. Friedland and Steven X. Cadrin

A similar series of studies demonstrate the utility of scale spacing patterns for discriminating groups of Atlantic salmon (reviewed in detail by Reddin and Friedland, 1999). Lear and Sandeman (1980) were the ﬁrst to use scale characters to discriminate the continent of origin of Atlantic salmon in the West Greenland ﬁshery. A reﬁned discriminant analysis based on the number of circuli in the ﬁrst marine summer and winter growth zones was developed by Reddin (1986) to classify salmon to North American or European origins. Reddin et al. (1988) used the discriminant function to estimate the portion of salmon caught off West Greenland that were spawned in North America. Reddin et al. (1990) expanded the discriminant function to include electrophoretic data with circuli numbers. Circuli spacing, expressed as the ﬁrst 50 Fourier coefﬁcients of luminescence proﬁles, classiﬁed farmed and wild salmon with 90% accuracy (Friedland et al., 1994). Salmon researchers also have derived great utility in applying these stock identiﬁcation techniques to identify the contrasts in ecological variation in stock structure. Circuli spacing was used to compare post-smolt growth because circuli spacings are wider for 1-seawinter than 2-seawinter salmon of the same cohort (Friedland and Haas, 1996). Differences in post-smolt growth among spawning populations were detected using intercirculi distances, with salmon from the Penobscot River having wider spacing (i.e., faster growth) than those from the Connecticut River (Friedland et al., 1996). Annual variability in circuli spacing led to inferences about the residence time in estuarine and coastal nursery habitats (Friedland et al., 1999). Circuli spacing patterns have been used to investigate recruitment synchrony among Atlantic salmon stocks (Friedland and Reddin, 2000). Scale circuli patterns currently are being used to discriminate Atlantic salmon from geographic regions of the Narraguagus river system (Haas-Castro et al., 2004). Similar patterns of growth and circuli patterns have been used to discriminate interspeciﬁc groups of other salmonid species. For example, Fisher and Pearcy (1990) used scale circuli spacing to compare growth rates among groups of coho salmon, and Marcogliese and Casselman (1998) discriminated rainbow trout from the wild and from hatcheries, with 90% accuracy.

B. STRIPED BASS Scale circuli spacing has been used to discriminate groups of striped bass. Ross and Pickard (1990) classiﬁed Paciﬁc striped bass to wild or hatchery origin with approximately 90% accuracy and to hatchery with 66% to 88% accuracy using luminescence proﬁles with a weighted moving average ﬁlter. Methods for using width and spacing of the ﬁrst 10 scale circuli were developed by Richards and Esteves (1997a) to discriminate groups of Atlantic striped bass (Fig. 9-2), and

Analyses of Calciﬁed Structures-Texture and Spacing Patterns

191

FIGURE 9-2. Transect of a striped bass scale used to measure circulus patterns. From Richards and Esteves (1997), with permission.

classify specimens to three wild stocks (Hudson River, Chesapeake Bay, and Roanoke River), with 57% to 84% accuracy (Richards and Esteves, 1997b). These example case studies demonstrate that circuli patterns can be used to discriminate groups with different growth patterns.

C. AQUACULTURE RELEASES

AND

ESCAPEES

With the proliferation of ﬁnﬁsh aquaculture, managers are faced with the contrasting issues of assessing intentional releases of aquaculture ﬁsh in stock enhancement programs and the unintentional escape of cultured ﬁsh that potentially pose a threat to wild stocks. The stock identiﬁcation of hatchery products can be simple if the ﬁsh are marked prior to release, but many species are cultured at a size that precludes effective marking. Growers will not assume the cost of tagging without mandated tagging requirements. Analysis of scale features analyses, among other characters, has proved to be effective to meet this data requirement. Silva and Bumguardner (1998) attempted to develop a classiﬁcation function to identify the hatchery contribution of red drum to Texas ﬁsheries. These researchers relied on circuli spacing patterns of juvenile stages to identify adults. Similar approaches were used to assess releases of barramundi in Australia (Barlow and Gregg, 1991) and silver perch, also in Australia (Willett, 1993). Marcogliese and Casselman (1998) discriminated rainbow trout from the wild and from hatcheries with 90% accuracy. Friedland et al. (1994) looked at the issue of identifying Atlantic salmon ﬁsh farm escapees using scale characteristics. Circuli spacing and texture, expressed as the ﬁrst 50 Fourier coefﬁcients of

192

Kevin D. Friedland and Steven X. Cadrin

Farm Ranch Wild

Magnitude

2500

2000

1500

1000

0

10

20

30

40

50

Harmonic FIGURE 9-3. Magnitude of the ﬁrst 50 harmonics differentiating three groups of Atlantic salmon: farm escapees, ranched ﬁsh, and wild ﬁsh. From Friedland et al. (1994).

luminescence proﬁles, classiﬁed farmed and wild salmon with 90% accuracy (Fig. 9-3).

IV. DISCUSSION Analysis of texture and spacing patterns offers a promising method for stock identiﬁcation. A major strength of the approach is that characters have meaningful interpretations. Differences in circuli distance and spacing patterns indicates groups that grow differently, and relative position of major checks indicate the relative timing of important life history events (e.g., the movement from riverine to marine environments). Detecting signiﬁcant differences in these characters is essential for phenotypic stock identiﬁcation. One advantage of deriving spacing and texture data from image processing is that it is quantitative and objective in that measurements are automated and without the problems associated with manual scale readings (Lund et al., 1989). Douglas, Minckley, and Tyus (1989) suggested that qualitative characters are

Analyses of Calciﬁed Structures-Texture and Spacing Patterns

193

excellent features upon which to base group separations. They presented extensive data that demonstrated even untrained observers show a high degree of feature interpretation. However, the automated approach does offer improvements since it addresses sources of procedural inaccuracy, such as those associated with reader fatigue, and removes any doubt of reader subjectivity from potentially sensitive management data. Manual scale reading and image processing techniques use essentially the same features of the scale or otolith to form information databases to classify the samples to origin. With image processing, the classiﬁcation algorithm can be deﬁned explicitly for review, and there is a complete quantitative audit trail for each decision. Spacing and texture data are undoubtedly inﬂuenced by the environment; therefore, it is essential to test the robustness of the variables based on these data for annual or long-term sources of variability. For example, annual variations in climate and food resources are known to affect circuli deposition in Atlantic salmon (Reddin et al., 1988). This has necessitated setting up new parameters for classiﬁcation models with that species (Reddin et al., 1990). It is probably wise with any scale-based discrimination procedure to maintain reference collections so that classiﬁcation models can be updated. Annual or longer-term sources of variability will be irrelevant if the proper reference samples are collected and applied. Precision with image processing techniques is very high, which is in contrast to approaches dependent on a scale reader. It is well known that fatigue, pattern of prior observations, and long term familiarity can affect the precision of scale readers (Lund et al., 1989). Feature extraction with the image processor is identical regardless of stage of the analysis, and it is also insensitive to problems created by changes in project personnel. Scales, otoliths, vertebrae, statoliths, and other hard parts are routinely collected from many species for age determinations. A large number of samples have generally been collected to represent the population or the catch, and sampling has been maintained for many years. Therefore, a wealth of archived hard parts is available for many populations, allowing the analysis of spacing patterns and exploration of growth differences and stock structure. Despite the continued promise associated with this family of techniques, they have not enjoyed widespread testing and application as predicted when hardware and software capabilities to extract texture and spacing features were ﬁrst introduced. Because negative ﬁndings do not usually make their way into the literature, it is difﬁcult to assess whether poor results or low usage of the methods account for their poor dissemination. A systematic evaluation of the discriminating power of this class of features would be useful.

194

Kevin D. Friedland and Steven X. Cadrin

REFERENCES Anas, R. E. and Myrai, S. 1969. Use of scale characteristics and a discriminant function for classifying sockeye salmon Oncorhychus nerka by continent of origin. Int. Pac. Salmon Fish. Com. Bull. 26. Antere, I. and Ikonen, E. 1983. A method of distinguishing wild salmon from those originating from ﬁsh farms on the basis of scale structure. ICES CM 1983/M: 26. 6 pp. Barber, W. E. and Walker, R. J. 1988. Circuli spacing and annulus formation: is there more than meets the eye? The case for sockeye salmon, Oncorhynchus nerka. J. Fish Biology 32: 237–245. Barlow, C. G. and Gregg, B. A. 1991. Use of circuli spacing on scales to discriminate hatchery and wild barramundi, Lates calcarifer (Bloch). Aquaculture Fish. Management 22: 491–498. Brooke, H. E. 1981. The conundrum of the stock concept—are nature and nurture deﬁnable in ﬁshery science? Canadian Journal of Fisheries and Aquatic Sciences 38: 1479–1480. Cadrin, S. X. and Friedland, K. D. 1999. The utility of image processing techniques for morphometric analysis and stock identiﬁcation. Fisheries Research 43: 129–139. Clutter, R. and Whitesel, L. 1956. Collection and interpretation of sockeye salmon scales. Int. Pac. Salmon Fish. Com. Bull. 9. Cook, R. C. 1982. Stock identiﬁcation if sockeye salmon (Oncorhychus nerka) with scale pattern recognition. Can. J. Fish. Aquat. Sci. 39: 611–617. Cook, R. C. 1987. Optical pattern recognition for stock identiﬁcation: past, present and future. NOAA Tech. Mem. NMFS-SEFC-199: 182–183. Cook, R. C. and Guthrie, I. 1987. In-season stock identiﬁcation of sockeye salmon (Oncorhynchus nerka) using scale pattern recognition. In H. D. Smith, L. Margolis, and C. C. Wood (eds.), Sockeye Salmon (Oncorhynchus nerka) Population Biology and Future Management. Can. Spec. Publ. Fish. Aquat. Sci., no. 96, pp. 327–334. Cook. R. C. and Lord, G. E. 1977. Stock identiﬁcation of sockeye salmon (Oncorhychus nerka) with scale pattern recognition. Fish. Bull. 76: 415–423. Douglas M. E., Minckley, W. L., and Tyus, H. M. 1989. Qualitative characters, identiﬁcation of Colorado River chubs (Cyprinidae: Genus Gila) and the “art of seeing well.” Copeia 89: 653–662. Doyle, R. W., Talbot, A. J., and Nicholas, R. R. 1987. Statistical interrelation of length, growth, and scale circulus spacing: appraisal of growth rate estimator for ﬁsh. Can. J. Fish. Aquat. Sci. 44: 1520–1528. Finn, J. E., Burger, C. V., and Holland-Bartels, L. 1997. Discrimination among populations of sockeye salmon frye with Fourier analysis of otolith banding patterns formed during incubation. Trans. Am. Fish. Soc. 126: 559–578. Fisher, J. P. and Pearcy, W. G. 1990. Spacing of scale circuli versus growth rate in young coho salmon. Fish. Bull. 88: 637–643. Friedland, K. D., Dutil, J.-D., and Sadusky, T. 1999. Growth patterns in postsmolts and the nature of the marine juvenile nursery for Atlantic Salmon, Salmo salar. Fish. Bull. 97: 472–481. Friedland, K. D., Esteves, C., Hansen, L. P., and Lund, R. A. 1994. Discrimination of Norwegian farm, ranch, and wild origin Atlantic salmon by image processing. Fish. Management Ecol. 1: 117–128. Friedland, K. D. and Haas, R. E. 1996. Marine post-smolt growth and age at maturity of Atlantic salmon. J. Fish Biol. 48: 1–15. Friedland, K. D., Haas, R. E., and Sheehan, T. F. 1996. Post-smolt growth, maturation, and survival of two stocks of Atlantic salmon. Fish. Bull. 94: 654–663. Friedland, K. D. and Reddin, D. G. 2000. Growth patterns of Labrador Sea Atlantic salmon postsmolts and the temporal scale of recruitment synchrony for North American salmon stocks. Can. J. Fish. Aquat. Sci. 57: 1181–1189. Harden Jones, F. R. 1968. Fish Migration. St. Martin’s, New York. 325 pp. Hass-Castro, R. E., Sheehan, T. F., and Cadrin, S. X. 2004. Scale pattern analysis discriminates Atlantic salmon by river-reach rearing. Can. J. Fish. Aquat. Sci. (in press).

Analyses of Calciﬁed Structures-Texture and Spacing Patterns

195

Henry, K. A. 1961. Racial identiﬁcation of Fraser River sockeye salmon by means of scales and its application to salmon management. Int. Pac. Salmon Fish. Com. Bull. 12. Hjort, J. 1914. Fluctuations in the great ﬁsheries of northern Europe. Rapp. P.-v. Reun. Cons. Inst Explor. Mer 20. Lear, W. H. and Sandeman, E. J. 1980. Use of scale characteristics and discriminant functions for identifying continental origin of Atlantic salmon. Rapp. P.-v. Reun. Cons. int. Explor. Mer. 176: 68–75. Lund, R. A. and Hansen, L. P. 1991. Identiﬁcation of wild and reared Atlantic salmon, Salmo salar L., using scale characters. Aquaculture Fish. Management 22: 499–508. Lund, R. A., Hansen, L. P., and Järvi, T. 1989. Identiﬁcation of reared and wild salmon by external morphology, size of ﬁns and scale characteristics. NINA Forskningsrapport 1: 1–54. Major, R. L., Mosher, K. H., and Manson, J. E. 1972. Identiﬁcation of stocks of Paciﬁc salmon by means of scale features. In R. C. Simon and P. A. Larkin (eds.), The Stock Concept in Paciﬁc Salmon, H. R. MacMillan Lectures in Fisheries, University of British Columbia, Vancouver, BC, pp. 209–223. Marcogliese, L. A. and Casselman, J. M. 1998. Scale methods for discriminating between Great Lakes stocks of wild and hatchery rainbow trout, with a measure of natural recruitment in Lake Ontario. North Am. J. Fish. Management 18: 253–268. Mosher, K. H. 1963. Racial analysis of red salmon by means of scales. Int. Pac. Salmon Fish. Com. Bull. 11. Quinn, T. P., Volk, E. C., and Hendry, A. P. 1999. Natural otolith microstructure patterns reveal precise homing to natal incubation sites by sockeye salmon (Oncorhynchus nerka). Can. J. Zool. 77: 766–775. Reddin, D. G. 1986. Discrimination between Atlantic salmon (Salmo salar L.) of North American and European origin. J. Cons. Int. Explor. Mer 43: 50–58. Reddin, D. G. and Friedland, K. D. 1999. A history of identiﬁcation of continent of origin of Atlantic salmon (Salmo salar L.) at West Greenland, 1969–97, Fisheries Research 43: 221–235. Reddin, D. G., Stansbury, D. E., and Short, P. B. 1988. Continent of origin of Atlantic salmon (Salmo salar L.) at West Greenland. J. Cons. Int. Explor. Mer 44: 180–188. Reddin, D. G., Verspoor, E., and Downton, P. R. 1990. An integrated phenotypic and genotypic approach to stock discrimination of Atlantic salmon. J. Cons. Int. Explor. Mer 47: 83–88. Richards, R. A. and Esteves, C. 1997a. Stock-speciﬁc variation in scale morphology of Atlantic striped bass. Trans. Am. Fish. Soc. 126: 908–918. Richards, R. A. and Esteves, C. 1997b. Use of scale morphology for discriminating wild stocks of Atlantic striped bass. Trans. Am. Fish. Soc. 126: 919–925. Ross, W. R. and Pickard, A. 1990. Use of scale pattern and shape as discriminators of wild and hatchery striped bass stocks in California. In N. C. Parker, A. E. Giorgi, R. C. Heidinger, D. B. Jester, E. D. Prince, and G. A. Winans (eds.), Fish-Marking Techniques. American Fisheries Society Symposium 7. American Fisheries Society, Bethesda, MD, pp. 71–77. Schwartzberg, M. and Fryer, J. 1989. Experiments in identifying hatchery and naturally spawning stocks of Columbia Basin spring chinook salmon using scale pattern analysis. Technical Report 89-3, Columbia River Inter-Tribal Fish Commission. 25 pp. Silva, P. S. and Bumguardner, B. W. 1998. Use of scale circuli pattern analysis to differentiate between hatchery and wild red drum. Trans. Am. Fish. Soc. 60: 38–43. Tabachnick, B. G. and Fidell, L. S. 1989. Using Multivariate Statistics. Harper Row & Collins. 746 pp. Waltemyer, D. L., Bue, B. G., and Tarbox, K. E. 1996. Evaluation of scale pattern analysis for Upper Cook Inlet sockeye salmon stocks. Alaska Fish. Res. Bull. 3: 69–80. Willett, D. J. 1993. Discrimination between hatchery stocks of silver perch, Bidyanus bidyanus (Mitchell), using scale growth pattern. Aquaculture Fish. Management 24: 347–354.

This Page Intentionally Left Blank

CHAPTER

10

Meristics JOHN R. WALDMAN Hudson River Foundation for Science and Environmental Research, New York, New York, USA Currently, Biology Department, Queens College, The City University of New York, New York, New York, USA

I. II. III. IV.

Introduction Use of Meristics for Stock Identiﬁcation A Case History: Striped Bass Conclusions References

I. INTRODUCTION Meristic characters are enumerable morphological features of ﬁshes. Strictly, meristic features are those that corresponded evolutionarily with body segmentation; however, today, the term is used more broadly (Strauss and Bond, 1990). The most commonly enumerated features have been external (Fig. 10-1), including ﬁn spines and ﬁn rays, gill rakers, and scales (along several possible vectors). Internal meristic features that have been analyzed include pterygiophores, vertebrae, branchiostegal rays, and pyloric caeca. Values attained for meristic features are the products of interactions between genetics and environment (Marr, 1957; Swain and Foote, 1999). Environmental factors inﬂuencing meristic counts include temperature, salinity, oxygen, pH, food availability, and growth rate, among others (Tåning, 1952; Barlow, 1961; Lindsey, 1988). The ﬁnal counts of most meristic features in ﬁshes are set prior to transformation from the larval to juvenile stages; for some meristic features, the sensitive period governing the actual number of elements attained may be ﬁxed prior to or after hatching, depending on the taxon (Barlow, 1961). On a global level, a phenomenon (termed Jordan’s rule) has often been observed in which the number of vertebrae tends to be higher in ﬁsh populations from colder waters. Other meristic features also seem to adhere to this pattern (Lindsey, 1988). An important aspect of meristic analysis is that the data obtained are discrete, not continuous, as results from analysis of body dimensions, that is, morphometrics. Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

197

198

John R. Waldman

FIGURE 10-1. (a) Common external meristic characters, and (b) additional meristic (and morphometric) characters. From Cailliet et al. (1986).

There is a long history of stock identiﬁcation of ﬁshes through meristic analysis; most ﬁsh species that occur as multiple stocks and that have been the subject of ﬁshery management also have received at least some meristic analysis. Examples of stock-related meristic studies conducted on ﬁsh by family include clupeidae (Rounsefell, 1930; Carscadden and Leggett, 1975; Schweigert, 1990), coregonidae (Lindsey, 1981; Henault and Fortin, 1989), engraulidae (McHugh, 1951), gadidae (Clark and Vladykov, 1960; Templeman, 1981), myxinidae

Meristics

199

(Martini et al., 1998), pleuronectidae (Lux, 1963; Bowering and Misra, 1982), sciaenidae (Perlmutter et al., 1956), sebastinae (Ni, 1982), serranidae (Shepherd, 1991), salmonidae (Winter et al., 1980; MacCrimmon and Claytor, 1985), scombridae (Schaefer, 1991), and trachichthyidae (Haddon and Willis, 1995).

II. USE OF MERISTICS FOR STOCK IDENTIFICATION Standardization of technique is imperative in meristic analysis so that both repeatability and comparison among studies are possible. Conventions exist for enumerating features that may take ambiguous forms, such as branched ﬁn rays and pyloric caeca, and for which gill arch should be used in making gill raker counts. Hubbs and Lagler (1958) provide widely accepted criteria for making meristic counts. Care should be taken to ensure that counts are accurate. Good lighting is essential; ﬁn spines and rays are often most readily viewed when backlit. Scales may be more easily counted if ﬁrst blotted to remove moisture, which reveals their margins. One approach to assure accuracy is to incorporate some level of quality control. In a study of coho salmon, Oncorhynchus kisutch, Hjort and Schreck (1982) recounted the meristic characters of two ﬁsh from each sample. If an error was found, meristic characters of all ﬁsh from that sample were recounted. Many meristic features, such as ﬁn rays and gill rakers, are elaborated in serial fashion. Fin rays normally reach their full complement by the early juvenile stage, but gill raker counts may increase for a much longer period, for example, for alosids (Smith, 1985). Thus, when comparing meristic counts among stocks composed of different aged or sized ﬁshes, it is important to consider whether a full complement of these features has been achieved. Meristic analysis also may be used for larval ﬁshes; however, characters examined of postlarval individuals may not yet be formed or fully elaborated in larvae. In these instances, it may be useful to examine internal features that are correlated with external features. For example, ﬁn spines and ﬁn rays form only after their internal bases, that is, pterygiophores are formed. These may be viewed through clearing and staining (Dingerkus and Uhler, 1977) or radiography (Tucker and Larouche, 1984). Another meristic-based character that has not received attention at the stock level is the interdigital relationship between vertebral spines and pterygiophores or supraneurals. However, Waldman and Andreyko (1993) found evidence of stock differences among both striped bass, Morone saxatilis, and white perch, Morone americana, across much of their respective ranges and also among tributaries of Chesapeake Bay. Conventions exist for coding these features (Olney et al., 1983). Prior to stock discrimination analysis, it is necessary to determine whether statistically signiﬁcant differences exist among reference stocks. Histograms and Dice

200

John R. Waldman

FIGURE 10-2. Graphic analysis of gill-raker counts from Spanish mackerel (see Table 10-1). Upper graph is of Dice diagrams showing sample range (horizontal line), mean (vertical line), one standard deviation on either side of mean (open bar), and t standard errors (hatched bar) on either side of mean. Thus, nonoverlapping hatched bars indicate that there is only a 5% chance or less of the two means being samples of the same population. Lower graph is a frequency histogram (hatched vertical bars) of gill-raker counts describing a sample from one population. Areas under the curve estimate the probabilities of counting the included numbers of rakers. From Cailliet et al. (1986).

diagrams (Fig. 10-2) may be useful in visualizing the raw data which usually are displayed in frequency tables (Table 10-1). Simple, descriptive, univariate statistical analysis may be conducted so as to forecast which characters will be effective in stock identiﬁcation and to provide another error check by scanning for anomalous data. For example, in their analysis of meristic variation among world hatchery stocks of rainbow trout, Oncorhynchus gairdneri, MacGregor and

201

Meristics

TABLE 10-1. Frequency Tabulation of Gill-Raker Counts Making Up Three Collections from Presumably Different Biological Populations in Two Species of Spanish Mackerel (Scomberomorus)a Total gill rakers on ﬁrst arch Species and geographic population Scomberomorus sierra (E. Paciﬁc) Mexico Central and South America S. maculatus (W. Atlantic)

11

2

12

13

14

15

16

17

Sample size

4

10 1 16

10 5 6

8 14 2

1 12 1

2

33 34 34

7

a

Adapted from Cailliet et al. (1986).

MacCrimmon (1977) calculated the mean, standard error, range, mean intrastock coefﬁcient of variation (CV), and overall CV for each meristic character. Traditionally, stock differences for individual characters have been tested either with analysis-of-variance (t-tests for paired comparisons) or c2 tests. In the latter case, it is common that numerous rare categories are encountered, and until recently, these rare categories needed to be grouped so as not to violate the statistical assumptions of c2. However, such grouping diminishes the information content contained in the rare phenotypes. An alternative approach based on c2 is now available that was developed for genetic frequency data (Roff and Bentzen, 1989). This Monte Carlo-based approach allows for virtually unlimited numbers of rare categories; software is available as part of the REAP package (McElroy et al., 1992). Waldman and Andreyko (1993) applied this approach successfully to an analysis of osteological interdigitation patterns (as many as 32 phenotypes per character) among stocks of striped bass and white perch. Although establishment of the existence of statistically signiﬁcant differences is essential toward the use of meristic data, Royce (1953) concluded that even with samples from closely related stocks, signiﬁcant statistical differences could always be found by increasing the size of the sample or by considering enough characters. Royce (1957) believed that statistically signiﬁcant differences among samples is a necessary but trivial preliminary in stock studies, and that once a predetermined level of signiﬁcance is reached, larger samples and further sampling merely reiterates the conclusion. Consequently, additional interest should be focused on the quantity and direction of the differences because these differences largely will determine the efﬁcacy of stock composition analysis of mixed stocks. Royce (1957) provided a number of ways in which overlap in character values among stocks can be assessed. For multivariate analyses, multivariate statistical signiﬁcance is important, but the utility of the multivariate model is best assessed in terms of its classiﬁcation

202

John R. Waldman

performance with cross-validation techniques, including jackkniﬁng and bootstrapping methods (Lachenbruch, 1975; James and McCulloch, 1990). Its utility may further be evaluated by removing the effects of correct classiﬁcation by chance by use of the kappa statistic (Titus et al., 1984). Within the univariate context, characters need not necessarily remain as counts of single meristic features. Prior to when routine multivariate analysis became possible with computers, Raney and de Sylva (1953) obtained improved univariate separation among striped bass stocks through creation of a compound character, which they labeled their “character index.” The index was the sum of the counts of the soft rays of the second dorsal and anal ﬁns, and the left and right pectoral ﬁn rays. This compound character was later used as one of ﬁve morphological variables in a stock composition study (discriminant analysis) of mid-Atlantic striped bass (Berggren and Lieberman, 1978). Although compound characters represent an interesting historical innovation, they usually have peculiar statistical properties and are rarely used anymore. Meristic data are, by deﬁnition, discontinuously distributed, and such distributions have different statistical properties than continuously distributed data. As such, fewer complications and violations of normality are likely to arise if statistical methods appropriate to discontinuously distributed data are used (e.g., c2 or logistic regression) rather than their continuous counterparts (ANOVA or discriminant analysis). Most multivariate stock composition analyses of meristic data have been conducted using discriminant analysis, which may yield satisfactory, but not necessarily optimal results. Alternative multivariate stock composition approaches that are receiving increased attention are maximum-likelihood analysis and logistic regression. In analyses of the same data, Van Winkle et al. (1988) found little difference between discriminant analysis and maximum-likelihood analysis in their estimated and true error rates in a relative contribution study of striped bass stocks. In contrast, Prager and Fabrizio (1990) analyzed a set of meristic data obtained from American shad, Alosa sapidissima, using both discriminant analysis and logistic regression; they found that logistic regression provided both lower allocation and classiﬁcation errors. With meristic data, measurement means and variances are often positively correlated (Winans, 1987). This effect may be reduced by transforming the data. Sokal and Rohlf (1981) recommended transforming meristic data to square roots. Meristic data also may be transformed to better ﬁt the requirements of a particular statistical model; such tailoring may result in better resolution among stocks. However, the necessity of data transformation is less clear for multivariate analyses. Van Winkle et al. (1988) chose not to transform raw meristic and morphometric data because they could not determine statistically whether transformations provided a better or poorer ﬁt to a multivariate normal distribution; such a distribution is assumed for both the discriminant function and maximumlikelihood methods they used.

Meristics

203

In multivariate analyses, meristic data may be used with other discrete or continuous character data, and these other data may be morphological or nonmorphological in origin. For example, Grove et al. (1976) combined protein electrophoresis data with both meristic and morphometric data in a discriminant function analysis to discriminate among three mid-Atlantic striped bass stocks. Fournier et al. (1984) presented a robust maximum-likelihood stock composition analysis model useful for discrete data, continuous data, or any combination of the two. Stock identiﬁcation using meristics (and other features) also may be complicated by stocks that are made up of substocks (Stephenson, 1999). In a simulation study involving stock composition estimates of the Hudson River and Chesapeake Bay striped bass stocks to a mixed ﬁshery, Waldman and Fabrizio (1994) used meristic and morphometric data of striped bass from 4 of the 11 or so tributary-speciﬁc populations that make up the Chesapeake Bay striped bass stock. Although the inclusion of one, two, three, or four substocks in the discriminant analysis did not greatly affect the overall correct-classiﬁcation rates, the speciﬁc combination of substocks signiﬁcantly affected the relative contribution estimates derived from the mixed stock sample.

III. A CASE HISTORY: STRIPED BASS Striped bass provide an excellent example in which meristic analysis helped provide fundamental information on their stock structure. As recently as the mid-1930s (Vladykov and Wallace, 1938), there was a lack of understanding of whether striped bass were a migratory species and if those that occurred along the mid-Atlantic were composed of multiple stocks. To answer these questions, E. C. Raney of Cornell University and his students and colleagues began systematic analyses of the meristic features of striped bass over their entire range. Raney and de Sylva (1953) compared ﬁn spine and ﬁn ray counts between striped bass from the Chesapeake Bay and the Hudson River; ﬁn ray counts showed substantial differences between ﬁsh from these locations. Raney and Woolcott (1955) found differences in counts of lateral line scales and ﬁn rays among striped bass from rivers of the southeastern United States. Lewis (1957) used gill raker counts and Raney (1957) used lateral line and ﬁn ray counts to distinguish three Chesapeake Bay subpopulations. Murawski (1958) compared lateral line scale counts of striped bass from Canada to Florida and found evidence of distinct populations throughout this range. He also identiﬁed three Chesapeake Bay subpopulations, consistent with the ﬁndings of Lewis (1957) and Raney (1957). Barkuloo (1970) analyzed lateral line scale counts of striped bass from the St. John’s River, a Florida tributary to the Atlantic Ocean, with those of other Atlantic and Gulf coast stocks and concluded that the St. John’s population is endemic.

204

John R. Waldman

Throughout these studies, the authors found year-class effects, that is, environmental inﬂuences on meristic counts that strongly affected the degree of overlap in character values among stocks. However, the stock divisions indicated by these studies largely have been conﬁrmed by subsequent studies that used more sensitive biochemical approaches (Waldman et al., 1988).

IV. CONCLUSIONS Meristic analysis remains a technically simple, inexpensive alternative for the stock identiﬁcation of ﬁshes. Drawbacks include that in almost all instances, specimens must be sacriﬁced—it is difﬁcult (although not impossible) to make accurate counts of meristic features on living ﬁsh. Also, variation among year-classes as a consequence of environmental effects must be considered. However, there may be instances in which ecophenotypic variation of meristic traits is more effective toward stock identiﬁcation than are genetic approaches. One instance is where genetic differences are limited because some or all populations to be analyzed are very recent in origin due to natural recolonization or stocking. Another instance is where there is sufﬁcient gene ﬂow among populations to preclude or erode genetic differentiation.

REFERENCES Barkuloo, J. M. 1970. Taxonomic status and reproduction of striped bass (Morone saxatilis) in Florida. U.S. Bureau of Sportﬁsheries and Wildlife Technical Paper 44. Barlow, G. W. 1961. Causes and signiﬁcance of morphological variation in ﬁshes. Systematic Zoology 10: 105–117. Berggren, T. J. and Lieberman, J. T. 1978. Relative contribution of Hudson, Chesapeake, and Roanoke striped bass, Morone saxatilis, stocks to the Atlantic coast ﬁshery. U.S. National Marine Fisheries Service Fishery Bulletin 76: 335–345. Bowering, W. R. and Misra, R. K. 1982. Comparisons of witch ﬂounder (Glyptocephalus cynoglossus) stocks of the Newfoundland-Labrador area, based upon a new multivariate analysis method for meristic characters. Canadian Journal of Fisheries and Aquatic Sciences 39: 564–570. Cailliet, G. M., Love, M. S., and Ebeling, A. W. 1986. Fishes: A Field and Laboratory Manual on Their Structure, Identiﬁcation, and Natural History. Wadsworth, Belmont, California. Carscadden, J. E. and Leggett, W. C. 1975. Meristic differences in spawning populations of American shad, Alosa sapidissima: evidence for homing to tributaries in the St. John River, New Brunswick. Journal of the Fisheries Research Board of Canada 32: 653–660. Clark, J. R. and Vladykov, V. D. 1960. Deﬁnition of haddock stocks of the northwestern Atlantic. U.S. Fish and Wildlife Service Fishery Bulletin 60: 283–296. Dingerkus, G. and Uhler, L. 1977. Enzyme clearing of alcian blue stained small vertebrates for demonstration of cartilage. Stain Technology 52: 229–232. Fournier, D. A., Beacham, T. D., Riddell, B. E., and Busack, C. A. 1984. Estimating stock composition in mixed stock ﬁsheries using morphometric, meristic, and electrophoretic characteristics. Canadian Journal of Fisheries and Aquatic Sciences 41: 400–408.

Meristics

205

Grove, T. L., Berggren, T. J., and Powers, D. A. 1976. The use of innate tags to segregate spawning stocks of striped bass, Morone saxatilis. Estuarine Processes 1: 166–176. Haddon, M. and Willis, T. J. 1995. Morphometric and meristic comparison of orange roughy (Hoplostethus atlanticus: Trachichthyidae) from the Puysegur Bank and Lord Howe Rise, New Zealand, and its implications for stock structure. Marine Biology 123: 19–27. Henault, M. and Fortin, R. 1989. Comparison of morphometric and meristic characters among springand fall-spawning ecotypes of cisco (Coregonus artedii) in southern Quebec, Canada. Canadian Journal of Fisheries and Aquatic Sciences 46: 166–173. Hjort, R. C. and Schreck, C. B. 1982. Phenotypic differences among stocks of hatchery and wild coho salmon, Oncorhynchus kisutch, in Oregon, Washington, and California. U.S. National Marine Fisheries Service Fishery Bulletin 80: 105–119. Hubbs, C. L. and Lagler, K. L. 1958. Fishes of the Great Lakes Region, 2nd Ed. Cranbrook Institute of Science Bulletin 26: 1–213. James, F. C. and McCulloch, C. E. 1990. Multivariate analysis in ecology and systematics: panacea or Pandora’s box? Annual Review of Ecology and Systematics 21: 129–166. Lachenbruch, P. A. 1975. Discriminant Analysis. Hafner Press, New York. Lewis, R. M. 1957. Comparative studies of populations of the striped bass. U.S. Fish and Wildlife Service Special Scientiﬁc Report—Fisheries 204. Lindsey, C. C. 1981. Stocks are chameleons: plasticity in gill rakers of coregonid ﬁshes. Canadian Journal of Fisheries and Aquatic Sciences 38: 1497–1506. Lindsey, C. C. 1988. Factors controlling meristic variation. Fish Physiology 11B: 197–274. Lux, F. E. 1963. Identiﬁcation of New England yellowtail ﬂounder groups. U.S. Fish and Wildlife Service Fishery Bulletin 63: 1–10. MacCrimmon, H. R. and Claytor, R. R. 1985. Meristic and morphometric identity of Baltic stocks of Atlantic salmon. Canadian Journal of Zoology 63: 2032–2037. MacGregor, R. B. and MacCrimmon, H. R. 1977. Meristic variation among world hatchery stocks of rainbow trout, Salmo gairdneri Richardson. Environmental Biology of Fishes 1: 127–143. Marr, J. C. 1957. The problem of deﬁning and recognizing subpopulations of ﬁshes. U.S. Fish and Wildlife Service Special Scientiﬁc Report—Fisheries 208: 1–6. Martini, F. H., Lesser, M. P., and Heiser, J. B. 1998. A population proﬁle for hagﬁsh, Myxine glutinosa, in the Gulf of Maine. Part 2: Morphological variation in populations of Myxine in the North Atlantic Ocean. U.S. National Marine Fisheries Service Fishery Bulletin 96: 516–524. McElroy, D., Moran, P., Bermingham, E., and Kornﬁeld, I. 1992. REAP: an integrated environment for the manipulation and phylogenetic analysis of restriction data. The Journal of Heredity 83: 157–158. McHugh, J. L. 1951. Meristic variations and populations of northern anchovy (Engraulis mordax mordax). Bulletin of the Scripps Institute of Oceanography 6: 123–160. Murawski, W. S. 1958. Comparative study of populations of the striped bass, Roccus saxatilis (Walbaum), based on lateral-line scale counts. Master’s thesis. Cornell University, Ithaca, New York. Ni, I. H. 1982. Meristic variation in beaked redﬁshes, Sebastes mentella and S. faciatus, in the Northwest Atlantic. Canadian Journal of Fisheries and Aquatic Sciences 39: 1664–1685. Olney, J. E., Grant, G. C., Schultz, F. E., Cooper, C. L., and Hageman, J. 1983. Pterygiophoreinterdigitation patterns in larvae of four Morone species. Transactions of the American Fisheries Society 112: 525–531. Perlmutter, A., Miller, S. W., and Poole, J. C. 1956. The weakﬁsh (Cynoscion regalis) in New York waters. New York Fish and Game Journal 3: 1–43. Prager, M. H. and Fabrizio, M. C. 1990. Comparison of logistic regression and discriminant analyses for stock identiﬁcation of anadromous ﬁsh, with application to striped bass (Morone saxatilis) and American shad (Alosa sapidissima). Canadian Journal of Fisheries and Aquatic Sciences 47: 1570–1577.

206

John R. Waldman

Raney, E. C. 1957. Subpopulations of the striped bass, Roccus saxatilis (Walbaum) in tributaries of Chesapeake Bay. U.S. Fish and Wildlife Service Special Scientiﬁc Report—Fisheries 208: 85–107. Raney, E. C. and de Sylva, D. P. 1953. Racial investigations of the striped bass, Roccus saxatilis (Walbaum). Journal of Wildlife Management 17: 495–509. Raney, E. C. and Woolcott, W. S. 1955. Races of the striped bass, Roccus saxatilis (Walbaum) in the southeastern United States. Journal of Wildlife Management 19: 444–450. Roff, D. A. and Bentzen, P. 1989. The statistical analysis of mitochondrial DNA polymorphisms: c2 and the problem of small samples. Molecular Biology and Evolution 6: 539–545. Rounsefell, G. A. 1930. Contribution to the biology of the Paciﬁc herring, Clupea pallasii, and the condition of the ﬁshery in Alaska. Bulletin of the U.S. Bureau of Fisheries 45: 227–320. Royce, W. F. 1953. Preliminary report on a comparison of the stocks of yellowﬁn tuna. Proceedings of the Indo-Paciﬁc Fisheries Council 4, Section 2: 130–145. Royce, W. F. 1957. Statistical comparison of morphological data. U.S. Fish and Wildlife Service Special Scientiﬁc Report—Fisheries 208: 7–28. Schaefer, K. M. 1991. Geographic variation in morphometric characters and gill-raker counts of yellowﬁn tuna Thunnus albacares from the Paciﬁc Ocean. U.S. National Marine Fisheries Service Fishery Bulletin 89: 289–297. Schweigert, J. F. 1990. Comparison of morphometric and meristic data against truss networks for describing Paciﬁc herring stocks. American Fisheries Society Symposium 7: 47–62. Shepherd, G. 1991. Meristic and morphometric variation in black sea bass north of Cape Hatteras, North Carolina. North American Journal of Fisheries Management 11: 139–148. Smith, C. L. 1985. The inland ﬁshes of New York State. New York State Department of Environmental Conservation, Albany. Sokal, R. R. and Rohlf, F. J. 1981. Biometry. W. H. Freeman, San Francisco. Stephenson, R. L. 1999. Stock complexity in ﬁsheries management: a perspective of emerging issues related to population sub-units. Fisheries Research 43: 247–249. Strauss, R. E. and Bond, C. E. 1990. Taxonomic methods: morphology. In C. B. Schreck and P. B. Moyle (eds.), Methods for Fish Biology, American Fisheries Society, Bethesda, pp. 109–140. Swain, D. P. and Foote, C. J. 1999. Stocks and chameleons: the use of phenotypic variation in stock identiﬁcation. Fisheries Research 43: 113–128. Tåning, A. V. 1952. Experimental study of meristic characters in ﬁshes. Biological Reviews 27: 169–193. Templeman, W. T. 1981. Vertebral numbers in Atlantic cod, Gadus morhua, of the Newfoundland and adjacent areas, 1947–1971, and their use in delineating cod stocks. Journal of Northwest Atlantic Fishery Science 2: 21–45. Titus, K., Mosher, J. A., and Williams, B. K. 1984. Chance-corrected classiﬁcation for use in discriminant analysis: ecological applications. The American Midland Naturalist 111: 1–7. Tucker, J. W. and Larouche, J. L. 1984. Radiographic techniques in studies of young ﬁshes. American Society of Ichthyologists and Herpetologists Special Publication 1: 37–39. Van Winkle, W., Kumar, K. D., and Vaughan, D. S. 1988. Relative contributions of Hudson River and Chesapeake Bay striped bass stocks to the Atlantic coastal population. American Fisheries Society Monograph 4: 255–266. Vladykov, V. D. and Wallace, D. H. 1938. Is the striped bass (Roccus lineatus) of Chesapeake Bay a migratory ﬁsh? Transactions of the American Fisheries Society 67: 67–86. Waldman, J. R. and Andreyko, H. 1993. Variation in patterns of interdigitation among supraneurals, pterygiophores, and vertebral elements diagnostic for striped bass and white perch. Copeia 1993: 1097–1113. Waldman, J. R. and Fabrizio, M. C. 1994. Problems of stock deﬁnition in estimating relative contributions of Atlantic striped bass to the coastal ﬁshery. Transactions of the American Fisheries Society 123: 766–778.

Meristics

207

Waldman, J. R., Grossﬁeld, J., and Wirgin, I. 1988. Review of stock discrimination techniques for striped bass. North American Journal of Fisheries Management 8: 410–425. Winans, G. A. 1987. Using morphometric and meristic characters for identifying stocks of ﬁsh. NOAA (National Oceanic and Atmospheric Administration) Technical Memorandum NMFS (National Marine Fisheries Service) SEFC-199: 25–62. Winter, G. W., Schreck, C. B., and McIntyre, J. D. 1980. Meristic comparison of four stocks of steelhead trout (Salmo gairdneri). Copeia 1980: 160–162.

This Page Intentionally Left Blank

PART

Natural Marks— Environmental Signals

IV

This Page Intentionally Left Blank

CHAPTER

11

Parasites as Biological Tags K. MACKENZIE* AND P. ABAUNZA† *School of Biological Sciences, Department of Zoology, The University of Aberdeen, Aberdeen, Scotland, United Kingdom, †Instituto Español de Oceanografía, 39080 Santander, Spain

I. II. III. IV. V. VI. VII. VIII. IX. X.

Introduction General Principles Advantages and Limitations of Parasite Tagging Selection of Parasites for Use as Tags Methodology Collection of Hosts and Parasites Fixation and Preservation of Parasites Identiﬁcation of Parasites Parasite Genetics Interpretation of Results References

I. INTRODUCTION The ﬁrst publication describing the use of a naturally occurring parasite as a biological tag in a population study of marine ﬁsh dates back over 60 years (Herrington et al., 1939). Many original papers on the subject have been published since then, plus a number of reviews (Sindermann, 1961, 1983; Kabata, 1963; MacKenzie, 1983, 1987a, 2002; Lester, 1990; Moser, 1991; Williams et al., 1992; Arthur, 1997; MacKenzie and Abaunza, 1998). In this chapter, we describe procedures and methods for applying this method to stock identiﬁcation of marine ﬁsh.

II. GENERAL PRINCIPLES The basic principle underlying the use of parasites as tags in ﬁsh population studies is that ﬁsh can become infected with a parasite only when they come within the endemic area of that parasite. The endemic area is that geographic Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

211

212

K. MacKenzie and P. Abaunza

region in which conditions are suitable for the transmission of the parasite. If infected ﬁsh are found outside the endemic area, we can infer that these ﬁsh had been within that area at some time in their past history. Information on the maximum life span of the parasite in that particular host allows us to estimate the period of time since the ﬁsh left the parasite’s endemic area. The more parasites with different endemic areas that can be used, the more information that can be obtained about the past movements of ﬁsh populations, and hence stock structure.

III. ADVANTAGES AND LIMITATIONS OF PARASITE TAGGING The most efﬁcient approach to stock identiﬁcation is the multidisciplinary one in which the results of different tagging methods and techniques are compared and used to complement one another (e.g., see Campbell et al., 2002). However, each method has its own strengths and weaknesses, and the use of parasites as tags is recognized as having certain advantages over other methods such as artiﬁcial tagging (Williams et al., 1992) and genetic studies (MacKenzie, 2002). • Parasite tags are more appropriate for studies of small, delicate species of

•

• • •

ﬁsh, such as small clupeoids, deepwater species, and crustaceans, for which artiﬁcial tags can either be used with difﬁculty or not at all. Each specimen sampled represents a valid observation, whereas with artiﬁcial tags each individual must be sampled, tagged, and recaptured to obtain a valid observation. They are less expensive to use because samples can be obtained from routine sampling programs. The use of biological tags eliminates doubts concerning the possible abnormal behavior of artiﬁcially tagged hosts. Parasites can often be used to identify subpopulations of ﬁsh distinguished by behavioral differences, but between which there is still a considerable amount of gene ﬂow (“ecological stocks”; see MacKenzie, 2002).

We should also be aware of the following limitations of biological tagging, some of which were ﬁrst discussed by Sindermann (1983): • Lack of adequate information on the complex ecology and biology of

aquatic parasites can limit their efﬁcient use as tags. However, as research adds to our knowledge of the biology and ecology of marine parasites, their use as tags is becoming more efﬁcient. • The identiﬁcation of many parasite species is uncertain and subject to disagreement among taxonomists. The recent application of molecular

Parasites as Biological Tags

213

biology techniques to parasite taxonomy has resulted in the identiﬁcation of two or more “sibling species” in parasites which had previously been regarded as comprising a single species (Nascetti et al., 1986, 1993; Paggi et al., 1991; Renaud et al., 1983). Conversely, the same techniques have also shown that parasites which had previously been regarded as separate species may in fact be conspeciﬁc (Raibaut et al., 1986). • It is usually desirable to know the age of host individuals, but in some species of ﬁsh and invertebrates the techniques of age determination have not been validated.

IV. SELECTION OF PARASITES FOR USE AS TAGS The ideal tag parasite may be described as having the following features, according to the selection criteria suggested by Kabata (1963), Sindermann (1983), MacKenzie (1983, 1987a), and Williams et al. (1992). Parasites fulﬁlling all of these criteria are rarely encountered, so compromises usually have to be made. • It should have signiﬁcantly different levels of infection in the subject host

in different parts of the study area. Infection data can be analyzed according to prevalence, intensity and abundance of infection, as deﬁned by Bush et al. (1997). • It should persist in the subject host for a long period of time, the minimum time depending on the nature of the study. For stock identiﬁcation and recruitment studies, only parasites with life spans of more than one year should be used, whereas for studies of seasonal migrations, species with life spans of less than one year are acceptable. • Parasites with single-host life cycles, such as monogenetic trematodes and most parasitic crustaceans, are the simplest to use. Those with complex life cycles, such as digenetic trematodes, tapeworms, nematodes, and acanthocephalans, involving two or more stages in different hosts, are more difﬁcult to use because more information is required on the biotic and abiotic factors which inﬂuence the transmission of the parasite between hosts. Given this information they can, however, be used just as effectively. Køie (1983), in fact, suggested that digenetic trematodes have advantages as tags over other taxonomic groups of parasites because they tend to be highly speciﬁc to the primary host, which is usually a mollusc. The endemic area of a digenean is therefore largely determined by the geographic distribution of its mollusc host. • The level of infection should remain relatively constant from year to year. The effects of annual variations, however, can be nulliﬁed by following infection levels in single year-classes of the subject host over several years.

214

K. MacKenzie and P. Abaunza

• The parasite should be easily detected and identiﬁed. Examination of the

host should involve the minimum of dissection; otherwise time can become a limiting factor. • Parasites that are serious pathogens, particularly those that affect host behavior, should be avoided. Many different taxonomic groups of parasites have been used as tags for freshwater, anadromous, and marine ﬁsh and marine invertebrates, as described by Williams et al. (1992). Protozoans, myxosporeans, larval and adult helminthes, and parasitic crustaceans have all been used as tags for commercially important marine ﬁsh such as the Atlantic herring, Clupea harengus, and the Atlantic cod, Gadus morhua (see MacKenzie, 1987b; Hemmingsen and MacKenzie, 2001). The parasites most commonly used as tags are larval anisakid nematodes, probably because they are among the most common and widespread parasites of teleost ﬁsh (MacKenzie, 1987a; Sindermann, 1990; Quinteiro, 1990).

V. METHODOLOGY Two different approaches to the use of parasites as biological tags can be recognized. • A small number of parasite species are selected according to the criteria

previously outlined and a large number of host individuals are examined speciﬁcally for these species. Examples are the studies of Kabata (1963), Margolis (1963), Chenoweth et al. (1986) and MacKenzie (1990). The more information on the parasite fauna of the host that is available prior to the start of a biological tag study, the more efﬁcient the selection of parasites is likely to be. For this reason, it is useful to begin with a preliminary survey of the parasite fauna of the host in the proposed study area, as was done by Arthur and Arai (1980a,b), Gaevskaya and Kovaleva (1985), McGladdery and Burt (1985), Bower and Margolis (1991), and MacKenzie and Longshaw (1995). This type of approach is most applicable to studies of host species which are readily available for examination in large numbers. • Entire parasite assemblages can be analyzed using sophisticated statistical techniques. Examples are the studies of Lester et al. (1986, 1988, 2001), George-Nascimento and Arancibia (1992), Arthur and Albert (1993), Speare (1994, 1995), Sewell and Lester (1995), Boje et al. (1997), and George-Nascimento (2000). This type of approach can be applied to any host species but is particularly applicable to those which are large and valuable and are not readily available for examination in large numbers. For both approaches, summary statistics of levels of infection in terms of prevalence, mean intensity, and/or abundance should be expressed by the mean value

Parasites as Biological Tags

215

plus and minus the standard deviation, and the range. In some cases, the median intensity will also be an appropriate descriptor. The biological interpretations of mean, median, and other statistical parameters should be clearly described and understood to avoid misleading interpretations concerning the true nature of the infection (Rózsa et al., 2000). For example, in a highly overdispersed parasite distribution, the median intensity could provide more information about the level of infection than the mean. Various methods have been used for testing differences between samples for statistical signiﬁcance, with the Chi-squared test, Fisher’s exact test, and G-tests for differences between prevalence values being the most frequently used. When parasite distributions are aggregated, with highly skewed data and/or severe kurtosis, then nonparametric methods are more appropriate (Potvin and Roff, 1993; Stewart-Oaten, 1995). In such cases, account should be taken of what the test actually does (usually compares other characteristics of the distribution instead of means), and one should be conﬁdent about the assumption of the equality of variances in the original data set (Stewart-Oaten, 1995). The nonparametric Kruskal–Wallis and Mann–Whitney tests have been used frequently to test for signiﬁcance between prevalence and abundance values (see Brattey and Ni, 1992; Khan and Tuck, 1995). The parametric tests have broad validity for moderate nonnormal data (Stewart-Oaten, 1995). In this sense, t-tests and various ANOVA-type tests have been applied successfully to compare means of parasite infection levels (Bishop and Margolis, 1955; McClelland et al., 1990; Boily and Marcogliese, 1995). In all these cases, the data were logarithmically transformed to fulﬁl the assumptions of the parametric methods. Randomization methods (Manly, 1997) represent a solution to the problems of small sample size or unbalanced data (Potvin and Roff, 1993). Bush et al. (1997) strongly recommended that authors publish an appropriate statistical measure of how good the estimate is, such as conﬁdence limits or standard errors. For skewed distributions, and particularly if the sample is small, bootstrap conﬁdence intervals are a good solution. Rózsa et al. (2000) showed how to construct conﬁdence intervals for the mean and the median as well as a distribution-free comparison of mean intensities. Among the most frequently used methods applicable to the parasite assemblage approach is the use of a discriminant analysis. This operates on data sets for which prespeciﬁed well-deﬁned groups exist and describes the maximum differences among those groups (McGarigal et al., 2000). Examples of the application of this and related techniques can be found in Arthur and Albert (1993) and George-Nascimento (2000) (nonparametric discriminant analyses), and Lester et al. (1988) and Sewell and Lester (1995) (canonical multivariate analysis). Ordination and classiﬁcation techniques (i.e., detrended correspondence analysis, cluster analysis) were also used by George-Nascimento and Arancibia (1992), Speare (1995), and George-Nascimento (2000). A multivariate maximumlikelihood model was applied by Bailey et al. (1988) for stock composition

216

K. MacKenzie and P. Abaunza

analysis. These methods are described and summarized in Elliot (1979), Everitt and Dunn (1992), Johnson and Wichern (1992), Sokal and Rohlf (1995), Manly (1997), and McGarigal et al. (2000). Observed frequency distributions of parasites can provide information about parasite population dynamics and so help to explain different levels of infection in different host age groups (Anderson, 1978; Anderson and May, 1978; Pacala and Dobson, 1988; Brattey and Ni, 1992). Basically, the most informative way to quantify the occurrence of parasites is to describe the frequency distributions (Rózsa et al., 2000). To study the distributions or the dispersion patterns of parasites, GLMs (General Linear Models) should be used (Wilson et al., 1996). The best way to obtain a more precise adjustment of the mathematical model (with regard to parasite distributions, etc.) is by applying maximum-likelihood methods in the estimation of parameters (Brattey and Ni, 1992; Williams and Dye, 1994). Special care must be taken with analyses of data on long-lived parasites such as larval anisakid nematodes in ﬁsh. When intensity of infection is shown to be age-dependent, all analyses carried out to determine possible differences in the geographic distributions of different groups within a host population must be performed on similar host age groups (Abaunza and Villamor, 1994; Abaunza et al., 1995). Analyses of length rather than age are less accurate because of the effects of different environmental conditions on ﬁsh growth. In general, it is good practice, prior to any comparison of infection levels between areas, to analyze the effect of host biological factors such as sex, age, or length on the observed prevalences or intensities.

VI. COLLECTION OF HOSTS AND PARASITES Host samples should preferably be examined fresh. If this is not possible they should be deep-frozen or preserved in 10% buffered formal saline as soon as possible after capture. Frozen specimens should be packaged individually or in small groups so that large numbers do not have to be defrosted at the same time. In the laboratory each host specimen should ﬁrst be measured, weighed, sexed, and, where possible, aged. In the initial stages of a study, before selection of the most appropriate tag parasites, the entire protistan and metazoan parasite fauna of the host should be recorded, so a complete autopsy of each host is necessary. The autopsy requires examination of host tissues under a dissecting microscope at X10–25 followed by examination of smears at magniﬁcations up to X2000. The external surfaces should be examined ﬁrst; then the host should be dissected according to a clearly deﬁned procedure. The alimentary tract should be divided into sections, for example, stomach, pyloric caeca, anterior intestine, mid-intestine, and rectum, each of which should be opened longitudinally before examination. Numbers of metazoan parasites of each species should be noted.

Parasites as Biological Tags

217

Special precautions should be taken when collecting the following types of parasite. • Myxosporeans. These can make excellent tags, but care should be taken to

note the presence of not only spores, but also the less easily observed vegetative stages. • Adult cestodes and acanthocephalans. The scoleces of cestodes and the proboscides of acanthocephalans have important diagnostic features. When these parasites are found ﬁrmly attached to host tissues, they should be removed together with a piece of host tissue around the site of attachment to ensure that the entire worm remains intact. • Larval nematodes. Many of these occur scattered throughout the musculature and other soft tissues of ﬁsh. They can be detected by “candling” ﬁllets over a light box or by digesting them in a pepsin digest solution (Stern et al., 1958; Smith and Wootten, 1975). • Digenean metacercariae and cestode plerocercoids. These larval or juvenile stages are often encysted and must be removed from their cysts for identiﬁcation. This can be done either by dissection with needles or by digesting the cyst with a pepsin digest solution (Smith and Halton, 1983). The nature of the proboscid armament is an important diagnostic feature for trypanorhynch cestodes, but in plerocercoids the proboscides are usually inverted and must be everted by placing the plerocercoid in fresh water under coverslip pressure. This can only be done with fresh material; it is not usually possible to evert the proboscides in frozen specimens.

VII. FIXATION AND PRESERVATION OF PARASITES A reference collection of permanently mounted specimens of all parasite species recovered should be made. Good general guides to the methods involved are provided by Pritchard and Kruse (1982) and Ash and Orihel (1987), while the following are recommended for particular taxonomic groups. • Protozoa and Myxozoa: Canning and Lom (1986); Lom and Dykova (1992). • Helminths: Gibson (1984); Hendrix (1994). • Crustacea: Kabata (1970); McLaughlin et al. (1982).

VIII. IDENTIFICATION OF PARASITES Parasites are best examined when fresh and preferably live, but this is not always possible. Nomarski interference contrast microscopy is recommended for protozoans and small fresh helminths. Phase contrast is recommended for examining

218

K. MacKenzie and P. Abaunza

sporozoan protozoans and small crustaceans. Scanning electron microscopy is useful for conﬁrming the identity of protozoans and small helminths. Most of the published keys to identiﬁcation of parasite species are to be found in specialist papers scattered throughout the parasitological literature. The following taxonomic textbooks, apart from being useful in their own right, include references to most of these specialist publications. • Protozoa: Canning and Lom (1986); Lom and Dykova (1992). • Helminths: Golvan (1969), Schmidt (1986); Anderson (1992); Khalil

et al. (1994); Williams and Jones (1994); Gibson et al. (2002). • Crustacea: Kabata (1970, 1979).

IX. PARASITE GENETICS Genetic analyses of parasite populations may also provide a tool for host stock identiﬁcation. Beverley-Burton (1978) was the ﬁrst to use genetic techniques in this way, in the form of a study of the frequencies of different acid phosphatase alleles in larvae of the nematode Anisakis simplex. Subsequent studies have shown that several of the ascaridoid nematode genera infecting ﬁsh and marine mammals consist of two or more morphologically similar but genetically distinct sibling species (Nascetti et al., 1986, 1993; Paggi et al., 1991; Mattiucci et al., 1997). Some recent papers have incorporated this information in the analyses of levels of infection (Brattey and Bishop, 1992; Brattey and Ni, 1992; Boily and Marcogliese, 1995). Although no further genetic studies have been carried out on parasites for ﬁsh stock identiﬁcation, they have been used to conﬁrm the species composition of parasite faunas to ensure uniformity in data analyses.

X. INTERPRETATION OF RESULTS After selection of the tag parasites and collection and analysis of the data, the ﬁnal important stage in a biological tag study is the interpretation of the results in terms of host population biology. Initially the results of such a study may be open to more than one possible interpretation. For example, a decrease in the level of infection in a particular area could be explained by (1) loss of parasites from the same host population; (2) selective mortality of heavily infected hosts; (3) immigration of another host population with a lower level of infection; (4) emigration of the most heavily infected element of the original population; or (5) a combination of two or more of the above. All of these possibilities must be considered in the light of the available information on the biology of both host and parasite until all but one have been eliminated. The guidelines previously given are designed to assist with this process of elimination of the outset. As we acquire

219

Parasites as Biological Tags TABLE 11-1.

Life Cycles of Some Common Parasites Infecting Fish

Parasite species Derogenes varicus (Digenea) Cryptocotyle lingua (Digenea) Renicola sp. (Digenea) Prosorhynchoides gracilescens (Digenea) Lacistorhynchus tenuis (Cestoda) Diphyllobothrium spp. (Cestoda) Anisakis simplex (Nematoda) Hysterothylacium aduncum (Nematoda) Echinorhynchus gadi (Acanthocephala) Corynosoma spp. (Acanthocephala) Lernaeocera branchialis Parasitic Copepod)

First intermediate or primary host(s) Natica spp. (gastropod molluscs) Littorina spp. (gastropod molluscs) Turritella spp. (gastropod molluscs) Abra spp. (bivalve molluscs) Copepods

Second intermediate host(s)

Deﬁnitive or ﬁnal host(s)

Crustaceans and chaetognaths

Teleost ﬁsh: many spp.

Køie (1979)

Teleost ﬁsh

Piscivorous birds

Stunkard (1929)

Small teleost ﬁsh

Piscivorous birds

Gadoid ﬁsh

Wright (1956); MacKenzie (1985) Matthews (1974)

Key references

Copepods

Pelagic teleost ﬁsh Teleost ﬁsh

Euphausiids

Teleost ﬁsh

Crustaceans

Small teleost ﬁsh

Angler ﬁsh, Lophius piscatorius Elasmobranch ﬁsh Mudry and Dailey (1971) Piscivorous birds Vik (1974) and mammals Cetaceans Smith (1983); Køie et al. (1995) Large piscivorous, Køie (1993) teleost ﬁsh

Gammarid crustaceans Amphipods

None

Gadoid ﬁsh

Teleost ﬁsh

Seals

Valtonen et al. (1983) Valtonen (1983)

Flatﬁsh, lumpﬁsh

None

Gadid ﬁsh

Kabata (1960)

more information on the biology of marine parasites, the interpretation of infection data becomes more accurate. This is particularly true of parasite life cycles, where great advances in knowledge have been made in recent years. Even if the life cycle of a particular species of tag parasite is not fully known, it will usually be broadly similar to those of related species for which more information is available. Examples of parasite life cycles involving ﬁsh as hosts are shown in Table 11-1, and the entire procedure for using parasites as biological tags is illustrated in Figure 11-1.

220 FIGURE 11-1. Summary of procedures and methods for using parasites as biological tags for stock discrimination of marine ﬁsh.

Parasites as Biological Tags

221

ACKNOWLEDGMENT The contribution of one of us (P. Abaunza) was prepared under the projects IEO 105 and XUGA 30103A93.

REFERENCES Abaunza, P. and Villamor, B. 1994. Further considerations into the usefulness of parasites as biological tags for marine ﬁsh stock discrimination, with special reference on use of Anisakis simplex (L3). Working Document presented to the 1994 ICES Study Group on Stock Identiﬁcation Protocols for Finﬁsh and Shellﬁsh Stocks. Abaunza, P., Villamor, B., and Pérez, J. R. 1995. Infestation by larvae of Anisakis simplex (Nematoda: Ascaridata) in horse mackerel, Trachurus trachurus, and Atlantic mackerel, Scomber scombrus, in ICES Divisions VIIIb, VIIIc and IXa (N–NW of Spain). Scientia Marina 59: 223–233. Anderson, R. C. 1992. Nematode Parasites of Vertebrates. Their Development and Transmission. CAB International, Oxford, UK. 600 pp. Anderson, R. M. 1978. The regulation of host population growth by parasitic species. Parasitology 76: 119–158. Anderson, R. M. and May, R. M. 1978. Regulation and stability of host-parasite population interactions: I. Regulatory processes. Journal of Animal Ecology 47: 219–247. Arthur, J. R. 1997. Recent advances in the use of parasites as biological tags for marine ﬁsh. In T. W. Flegel and I. H. MacRae (eds.), Diseases in Asian Mariculture III. Fish Health Section, Asian Fisheries Society, Manila, pp. 141–154. Arthur, J. R. and Albert, E. 1993. Use of parasites for separating stocks of Greenland halibut (Reinhardtius hippoglossoides) in the Canadian northwest Atlantic. Canadian Journal of Fisheries and Aquatic Sciences 50: 2175–2181. Arthur, J. R. and Arai, H. P. 1980a. Studies on the parasites of Paciﬁc herring (Clupea harengus pallasi Valenciennes): survey results. Canadian Journal of Zoology 58: 64–70. Arthur, J. R. and Arai, H. P. 1980b. Studies on the parasites of Paciﬁc herring (Clupea harengus pallasi Valenciennes): a preliminary evaluation of parasites as indicators of geographical origin for spawning herring. Canadian Journal of Zoology 58: 521–527. Ash, L. A. and Orihel, T. C. 1987. Parasites: A Guide to Laboratory Procedures and Identiﬁcation. ASCP Press. American Society of Clinical Pathologists, Chicago. 328 pp. Bailey, R. E., Margolis, L., and Groot, C. 1988. Estimating stock composition of migrating juvenile Fraser River (British Columbia) sockeye salmon, Oncorhynchus nerka, using parasites as natural tags. Canadian Journal of Fisheries and Aquatic Sciences 45: 586–591. Beverley-Burton, M. 1978. Population genetics of Anisakis simplex (Nematoda: Ascaridoidea) in Atlantic salmon (Salmo salar) and their use as biological indicators of host stocks. Environmental Biology of Fishes 3: 369–378. Bishop, Y. and Margolis, L. 1955. A statistical examination of Anisakis larvae (Nematoda) in herring (Clupea harengus) of the British Columbia coast. Journal of the Fisheries Research Board of Canada 12: 571–592. Boily, F. and Marcogliese, D. J. 1995. Geographical variations in abundance of larval anisakine nematodes in Atlantic cod (Gadus morhua) and American plaice (Hipoglossoides platessoides) from the Gulf of St. Lawrence. Canadian Journal of Fisheries and Aquatic Sciences 52(suppl. 1): 105– 115.

222

K. MacKenzie and P. Abaunza

Boje, J., Riget, F., and Køie, M. 1997. Helminth parasites as biological tags in population studies of Greenland halibut [Reinhardtius hippoglossoides (Walbaum)], in the northwest Atlantic. ICES Journal of Marine Science 54: 886–895. Bower, S. M. and Margolis, L. 1991. Potential use of helminth parasites in stock identiﬁcation of ﬂying squid, Ommastrephes bartrami, in North Paciﬁc waters. Canadian Journal of Zoology 69: 1124–1126. Brattey, J. and Ni, I-H. 1992. Ascaridoid nematodes from the stomach of harp seals, Phoca groenlandica, from Newfoundland and Labrador. Canadian Journal of Fisheries and Aquatic Sciences 49: 956–966. Brattey, J. and Bishop, C. A. 1992. Larval Anisakis simplex (Nematoda: Ascaridoidea) infection in the musculature of Atlantic cod, Gadus morhua, from Newfoundland and Labrador. Canadian Journal of Fisheries and Aquatic Sciences 49: 2635–2647. Bush, A. O., Lafferty, K. D., Lotz, J. M., and Shostak, A. W. 1997. Parasitology meets ecology on its own terms: Margolis et al. revisited. Journal of Parasitology 83: 575–583. Campbell, N., MacKenzie, K., Mattiucci, S., Ramos, P., Pereira, A., and Abaunza, P. 2002. Parasites as biological tags in a population study of horse mackerel Trachurus trachurus. In Proceedings of the 10th International Congress of Parasitology—ICOPA X: Symposia, Workshops and Contributed Papers. Vancouver (Canada), August 4–9, 2002, pp. 217–222. Monduzzi Editore, International Proceedings Division, Bologna, Italy. Canning, E. U. and Lom, J. 1986. The Microsporidia of Vertebrates. Academic Press, London. 304 pp. Chenoweth, J. F., McGladdery, S. E., Sindermann, C. J., Sawyer, T. K., and Bierm, J. W. 1986. An investigation into the usefulness of parasites as tags for herring (Clupea harengus) stocks in the Western North Atlantic, with emphasis on use of the larval nematode Anisakis simplex. Journal of Northwest Atlantic Fisheries Science 7: 25–33. Elliot, J. M. 1979. Some Methods for the Analysis of Samples of Benthic Invertebrates, 2nd ed. Freshwater Biological Association Scientiﬁc Publication No. 25. 160 pp. Everitt, B. S. and Dunn, G. 1992. Applied Multivariate Data Analysis. Oxford University Press, New York. 304 pp. Gaevskaya, A. V. and Kovaleva, A. A. 1985. Parasitofauna of Trachurus picturatus and ecological and geographical features of its formation (in Russian). Ekologiya Morya 20: 80–84. George-Nascimento, M. 2000. Geographical variations in the jack mackerel Trachurus symmetricus murphyi populations in the southeastern Paciﬁc Ocean as evidenced from the associated parasite communities. Journal of Parasitology 86: 929–932. George-Nascimento, M. and Arancibia, H. 1992. Stocks ecológicos del jurel (Trachurus symmetricus murphyi Nichols) en tres zonas de pesca frente a Chile, detectados mediante comparación de su fauna parasitaria y morfometría. Revista Chilena de Historia Natural 65: 453–470. Gibson, D. I. 1984. Technology as applied to museums collections. Systematic Parasitology 6: 241–255. Gibson, D. I., Jones, A., and Bray, R. A. (eds.). 2002. Keys to the Trematoda. Vol. 1. CABI and The Natural History Museum, London. 521 pp. Golvan, Y. J. 1969. Systématique des Acanthocéphales (Acanthocephala Rudolphi 1801). Premiére partie. L’ordre des Palaeacantocephala Meyer 1931. Premier fascicule: la super-familie des Echinorhynchoidea (Cobbold 1876) Golvan et Houin 1963. Mémoires du Muséum national d’histoire naturelle, Série A, 57, fasc. Unique. 373 pp. Hemmingsen, W. and MacKenzie, K. 2001. The parasite fauna of the Atlantic cod, Gadus morhua L. Advances in Marine Biology 40: 1–80. Hendrix, S. S. 1994. Marine ﬂora and fauna of the eastern United States. Platyhelminthes: Monogenea. NOAA Technical Report NMFS 121. 106 pp.

Parasites as Biological Tags

223

Herrington, W. C., Bearse, H. M., and Firth, F. E. 1939. Observations on the life history, occurrence and distribution of the redﬁsh parasite Sphyrion lumpi. United States Bureau of Fisheries Special Report, No. 5: 1–18. Johnson, R. A. and Wichern, D. W. 1992. Applied Multivariate Statistical Analysis, 3rd ed. PrenticeHall International, New Jersey. 642 pp. Kabata, Z. 1960. On the speciﬁcity of Lernaeocera (Copepoda parasitica). Annals and Magazine of Natural History, Series 13(3): 133–139. Kabata, Z. 1963. Parasites as biological tags. ICNAF Special Publication No. 4: 31–37. Kabata, Z. 1970. Diseases of Fishes, Book 1: Crustacea as Enemies of Fishes. TFH Publications, Jersey City, NJ. Kabata, Z. 1979. Parasitic Copepoda of British Fishes. The Ray Society, London, UK. 468 pp. Khalil, L. F., Jones, A., and Bray, R. A. (eds.). 1994. Keys to the Cestode Parasites of Vertebrates. CAB International, Oxford, UK. 768 pp. Khan, R. A. and Tuck, C. 1995. Parasites as biological indicators of stocks of Atlantic cod (Gadus morhua) off Newfoundland, Canada. Canadian Journal of Fisheries and Aquatic Sciences 52(suppl. 1): 195–201. Køie, M. 1979. On the morphology and life-history of Derogenes varicus (Müller, 1784) Looss, 1901 (Trematoda, Hemiuridae). Zeitschrift für Parasitenkunde 59: 67–78. Køie, M. 1983. Digenetic trematodes from Limanda limanda (L.) (Osteichthyes, Pleuronectidae) from Danish and adjacent waters, with special reference to their life histories. Ophelia 22: 201– 228. Køie, M. 1993. Aspects of the life cycle and morphology of Hysterothylacium aduncum (Rudolphi, 1802) (Nematoda, Ascaridoidea, Anisakidae). Canadian Journal of Zoology 71: 1289–1296. Køie, M., Berland, B., and Burt, M. D. B. 1995. Development to third-stage larvae occurs in the eggs of Anisakis simplex and Pseudoterranova decipiens (Nematoda, Ascaridoidea, Anisakidae). Canadian Journal of Fisheries and Aquatic Sciences 52(suppl. 1): 134–139. Lester, R. J. G. 1990. Reappraisal of the use of parasites for ﬁsh stock identiﬁcation. Australian Journal of Marine and Freshwater Research 41: 855–864. Lester, R. J. G., Barnes, A., and Habib, G. 1986. Parasites of skipjack tuna, Katsuwonus pelamis: ﬁshery implications. Fishery Bulletin 83: 343–356. Lester, R. J. G., Sewell, K. B., Barnes, A., and Evans, K. 1988. Stock discrimination of orange roughy, Hoplostethus atlanticus, by parasite analysis. Marine Biology 99: 137–144. Lester, R. G. J., Thompson, C., Moss, H., and Barker, S. C. 2001. Movement and stock structure of narrow-banded Spanish mackerel as indicated by parasites. Journal of Fish Biology 59: 833–843. Lom, J. and Dykova, I. 1992. Protozoan Parasites of Fishes. Developments in Aquaculture and Fisheries Science (26). Elsevier, Amsterdam. 315 pp. MacKenzie, K. 1983. Parasites as biological tags in ﬁsh population studies. Advances in Applied Biology 7: 251–331. MacKenzie, K. 1985. The use of parasites as biological tags in population studies of herring (Clupea harengus L.) in the North Sea and to the north and west of Scotland. Journal du Conseil International pour l’exploration de la Mer 42: 33–64. MacKenzie, K. 1987a. Parasites as indicators of host populations. International Journal for Parasitology 17: 345–352. MacKenzie, K. 1987b. Relationships between the herring, Clupea harengus L., and its parasites. Advances in Marine Biology 24: 263–319. MacKenzie, K. 1990. Cestode parasites as biological tags for mackerel (Scomber scombrus L.) in the Northeast Atlantic. Journal du Conseil International pour l’Exploration de la Mer 46: 155–166. MacKenzie, K. 2002. Parasites as biological tags in population studies of marine organisms. Parasitology 124: S153–S163.

224

K. MacKenzie and P. Abaunza

MacKenzie, K. and Abaunza, P. 1998. Parasites as biological tags for stock discrimination of marine ﬁsh: a guide to procedures and methods. Fisheries Research 38: 45–56. MacKenzie, K. and Longshaw, M. 1995. Parasites of the hakes Merluccius australis and M. hubbsi in the waters around the Falkland Islands, southern Chile, and Argentina, with an assessment of their potential value as biological tags. Canadian Journal of Fisheries and Aquatic Sciences 52(suppl 1): 213–224. Manly, B. F. J. 1997. Randomization, Bootstrap and Monte Carlo Methods in Biology. 2nd ed. Chapman & Hall, London. 399 pp. Margolis, L. 1963. Parasites as indicators of the geographical origin of sockeye salmon, Oncorhynchus nerka (Walbaum), occurring in the North Paciﬁc Ocean and adjacent seas. Bulletin of the International North Paciﬁc Fisheries Commission 11: 101–156. Matthews, R. A. 1974. The life cycle of Bucephaloides gracilescens (Rudolphi, 1819) Hopkins, 1954 (Digenea: Gasterostomata). Parasitology 68: 1–12. Mattiucci, S., Nascetti, G., Cianchi, R., Paggi, L., Arduino, P., Margolis, L., Brattey, S., Webb, S., D’Amelio, S., Orecchia, P., and Bullini, L. 1997. Genetic and ecological data on the Anisakis simplex complex, with evidence for a new species (Nematoda, Ascaridoidea, Anisakidae). Journal of Parasitology 83: 401–416. McClelland, G., Misra, R. K., and Martell, D. J. 1990. Larval anisakine nematodes in various ﬁsh species from Sable Island Bank and its vicinity. In W. D. Bowen (ed.), Population Biology of Sealworm (Pseudoterranova decipiens) in Relation to Its Intermediate and Seal Hosts. Canadian Bulletin of Fisheries and Aquatic Sciences 222: 83–118. McGarigal, K., Cushman, S., and Stafford, S. 2000. Multivariate Statistics for Wildlife and Ecology Research. Springer, New York. 283 pp. McGladdery, S. E. and Burt, M. D. B. 1985. Potential of parasites for use as biological indicators of migration, feeding and spawning behaviour of northwestern Atlantic herring (Clupea harengus). Canadian Journal of Fisheries and Aquatic Sciences 42: 1957–1968. McLaughlin, P. A., Taylor, G. T., and Tracey, M. L. 1982. Systematic methods in research. In L. G. Abele (ed.), The Biology of Crustacea, 1. Systematics, the Fossil Record and Biogeography. Kluwer, Dordrecht, Belgium, pp. 29–63. Moser, M. 1991. Parasites as biological tags. Parasitology Today 7: 182–185. Mudry, D. R. and Dailey, M. D. 1971. Postembryonic development of certain tetraphyllidean and trypanorhynchan cestodes with a possible alternative life cycle for the order Trypanorhyncha. Canadian Journal of Zoology 49: 1249–1253. Nascetti, G., Paggi, L., Orecchia, P., Smith, J. W., Mattiucci, S., and Bullini, L. 1986. Electrophoretic studies on the Anisakis simplex complex (Ascaridida: Anisakidae) from the Mediterranean and North-East Atlantic. International Journal for Parasitology 16: 633–640. Nascetti, G., Cianchi, R., Mattiucci, S., D’Amelio, S., Orcchia, P., Paggi, L., Brattey, J., Berland, B., Smith, J. W., and Bullini, L. 1993. Three sibling species within Contracaecum osculatum (Nematoda, Ascaridida, Ascaridoidea) from the Atlantic arctic-boreal region: reproductive isolation and host preferences. International Journal for Parasitology 23: 105–120. Pacala, S. W. and Dobson, A. P. 1988. The relation between the number of parasites/host and host age: population dynamic causes and maximum likelihood estimation. Parasitology 96: 197–210. Paggi, L., Nascetti, G., Cianchi, R., Orecchia, P., Mattiucci, S., D’Amelio, S., Berland, B., Brattey, J., Smith, J. W., and Bullini, L. 1991. Genetic evidence for three sealworm species within Pseudoterranova decipiens (Nematoda: Ascaridida: Ascaridoidea) in the North Atlantic, and Norwegian and Barents Seas. International Journal for Parasitology 21: 195–212. Potvin, C. and Roff, D. A. 1993. Distribution-free and robust statistical methods: viable alternatives to parametric statistics? Ecology 74: 1617–1628. Pritchard, M. A. and Kruse, G. O. W. 1982. The Collection and Preservation of Animal Parasites. University of Nebraska Press, Lincoln. 141 pp.

Parasites as Biological Tags

225

Quinteiro, P. 1990. Helmintos parásitos de peces de las costas gallegas. Tesis Doctoral, Universidad de Santiago de Compostela. 243 pp. Raibaut, A., Berrebi, P., and Rousset, V. 1986. Utilisation de la génétique enzymatique pour comparer les copépodes parasites du genre Lernaeenicus (Pennellidae), selon la zone d’implantation dans l’hôte, Sardina pilchardus (Walbaum, 1792). Compte rendu hebdomadaire des séances de l’Académie des sciences, 303, Série III, No. 2: 127–129. Renaud, F., Gabrion, C., and Pasteur, N. 1983. Le complexe Bothriocephalus scorpii (Mueller, 1776): différenciation par électrophorèse enzymatique des espèces parasites du turbot (Psetta maxima) et de la barbue (Scophthalmus rhombus). Compte rendu hebdomadaire des séances de l’Académie des sciences, 296, Série III, No. 2: 127–129. Rózsa, L., Reiczigel, J., and Majoros, G. 2000. Quantifying parasites in samples of hosts. Journal of Parasitology 86: 228–232. Schmidt, G. D. 1986. Handbook of Tapeworm Identiﬁcation. CRC Press. Boca Raton, FL. 675 pp. Sewell, K. B. and Lester, R. J. G. 1995. Stock composition and movement of gemﬁsh, Rexea solandri, as indicated by parasites. Canadian Journal of Fisheries and Aquatic Sciences 52(suppl. 1): 225–232. Sindermann, C. J. 1961. Parasite tags for marine ﬁsh. Journal of Wildlife Management 25: 41–47. Sindermann, C. J. 1983. Parasites as natural tags for marine ﬁsh: a review. NAFO Scientiﬁc Council Studies 6: 63–71. Sindermann, C. J. 1990. Principal Diseases of Marine Fish and Shellﬁsh, 1, 2nd ed. Academic Press, San Diego. 521 pp. Smith, J. D. and Halton, D. W. 1983. The Physiology of Trematodes, 2nd ed. Cambridge University Press, Cambridge, UK. 446 pp. Smith, J. W. 1983. Anisakis simplex (Rudolphi, 1809, det. Krabbe, 1878) (Nematoda: Ascaroidea): morphology and morphometry of larvae from euphausiids and ﬁsh, and a review of the life-history and ecology. Journal of Helminthology 57: 205–224. Smith, J. W. and Wootten, R. 1975. Experimental studies on the migration of Anisakis sp. larvae (Nematoda: Ascaridida) into the ﬂesh of herring, Clupea harengus L. International Journal for Parasitology 5: 133–136. Sokal, R. R. and Rohlf, F. J. 1995. Biometry, 3rd ed. W.H. Freeman, New York. 887 pp. Speare, P. 1994. Relationships among black marlin, Makaira indica, in eastern Australian coastal waters, inferred from parasites. Australian Journal of Marine and Freshwater Research 45: 535–550. Speare, P. 1995. Parasites as biological tags for sailﬁsh Istiophorus platypterus from east coast Australian waters. Marine Ecology Progress Series 118: 43–50. Stern, J. A., Chakravarti, D., Uzmann, J. R., and Hesselholt, M. N. 1958. Rapid counting of Nematoda in salmon by peptic digestion. United States Fish and Wildlife Service, Special Scientiﬁc Report—Fisheries No. 255. 5 pp. Stewart-Oaten, A. 1995. Rules and judgements in statistics: three examples. Ecology 76: 2001–2009. Stunkard, H. W. 1929. The life cycle of Cryptocotyle lingua (Creplin), with notes on the physiology of the cercaria. Journal of Morphology 50: 143–183. Valtonen, E. 1983. Relationships between Corynosoma semerme and C. strumosum (Acanthocephala) and their paratenic ﬁsh hosts in the Bothnian Bay, Baltic Sea. Acta Universitatis Ouluensis, Series A, Scientiae Rerum Naturalium No. 155, Biologica No. 21: 1–32. Valtonen, E. T., van Maren, M. J., and Timola, O. 1983. A note on the intermediate hosts of Echinorhynchus gadi Zoega, in Müller (Acanthocephala) in the Baltic Sea. Aquila Series in Zoology 22: 93–97. Vik, R. 1964. The genus Diphyllobothrium. An example of the interdependence of systematics and experimental biology. Experimental Parasitology 15: 361–380.

226

K. MacKenzie and P. Abaunza

Williams, B. G. and Dye, C. 1994. Maximum likelihood for parasitologists. Parasitology Today 10: 489–493. Williams, H. H. and Jones, A. 1994. Parasitic Worms of Fish. Taylor & Francis, London, UK. 593 pp. Williams, H. H., MacKenzie, K., and McCarthy, A. M. 1992. Parasites as biological indicators of the population biology, migrations, diet and phylogenetics of ﬁsh. Reviews in Fish Biology and Fisheries 2: 144–176. Wilson, K., Grenfell, B. T., and Shaw, D. J. 1996. Analysis of aggregated parasite distributions: a comparison of methods. Functional Ecology 10: 592–601. Wright, C. A. 1956. Studies on the life history and ecology of the trematode genus Renicola Cohn, 1904. Proceedings of the Zoological Society of London 126: 1–49.

CHAPTER

12

Otolith Elemental Composition as a Natural Marker of Fish Stocks STEVEN E. CAMPANA Marine Fish Division, Bedford Institute of Oceanography, Dartmouth, Nova Scotia, Canada

I. Introduction II. Sampling and Assays A. Sample Preparation and Quality Control B. Statistical Analysis III. Case Studies A. Population Mixing of Atlantic Cod B. Natal Homing of Weakﬁsh C. Estuarine Contribution of Juvenile Snapper to the Adult Fishery D. Determination of River of Origin of Atlantic Salmon IV. Conclusion References

I. INTRODUCTION To the extent that groups of ﬁsh inhabit different environments, the otolith elemental composition often serves as a natural marker or tag of those groups. Two key properties of the otolith underlie the use of the otolith elemental composition as a natural marker: (1) unlike bone, the otolith is metabolically inert; therefore, newly deposited material is neither resorbed nor reworked after deposition (Campana and Neilson, 1985); and (2) trace element uptake onto the growing otolith reﬂects the physical and chemical environment (Fowler et al., 1995; Gallahar and Kingsford, 1996), albeit with signiﬁcant physiological regulation (Kalish, 1989; Farrell and Campana, 1996). Isotopic ratios of elements such as strontium (Kennedy et al., 1997) and oxygen (Thorrold et al., 1997a) are similarly inﬂuenced by environmental availability and temperature. Such Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

227

228

Steven E. Campana

environmental responses, recorded permanently in the otolith, imply that the otolith concentration of selected elements and isotopes (the “elemental ﬁngerprint”) can be used as a biological tag to discriminate among groups of ﬁsh which have spent at least part of their lives in different environments (Fig. 12-1). As a 40

BA

30

20

10

0 200

300

400

500

600

700

SR 3

Discriminant Function 2

40

BA

30

20

10

0 0.0

.5

1.0

1.5

2.0

2.5

3.0

MN

1

-1

-3 -6

-4

-2

0

2

4

6

8

Discriminant Function 1

700

600

SR

500

400

300 200 0.0

.5

1.0

1.5

2.0

2.5

3.0

MN

FIGURE 12-1. Example of the preparation of a multivariate otolith elemental ﬁngerprint for shad (Alosa sapidissima). The concentrations of three elements (Ba, Mn, and Sr) were measured in otoliths from about 60 shad collected in each of three river systems. When one element is plotted against another (left panels), there are varying degrees of differences among the three rivers (each river is represented by a different symbol). The differences in elemental composition among rivers become more evident when the individual elements are pooled into a multivariate ﬁngerprint (right panel). Full details of the shad study are available in Thorrold et al. (1998).

Otolith Elemental Composition as a Natural Marker of Fish Stocks

229

result, the elemental ﬁngerprint appears to be an excellent biological tracer of groups of ﬁsh, both in freshwater (Kalish, 1990; Northcote et al., 1992; Bronte et al., 1996; Kennedy et al., 1997, 2000, 2002; Limburg, 1998; Thorrold et al., 1998a) and saltwater environments (Edmonds et al., 1989, 1991, 1992, 1995; Gunn et al., 1992; Sie and Thresher, 1992; Campana et al., 1994, 1995, 1999; Campana and Gagné, 1995; Thresher et al., 1994; Proctor et al., 1995; Severin et al., 1995; Dove et al., 1996; Gillanders and Kingsford, 1996; Milton et al., 1997; Thorrold et al., 1997b, 1998b, 2001; Begg et al., 1998; Dufour et al., 1998; Newman et al., 2000; Volk et al., 2000; Gillanders, 2001, 2002; Secor et al., 2001). The presence of geographic variations in water temperature and chemistry, both of which can result in different otolith composition, suggests that otolith elemental ﬁngerprints should discriminate well among ﬁsh that have grown up in different environments. However, it is probably inappropriate to refer to the use of elemental ﬁngerprints as stock discriminators, since genetic differences are not implied and spatial heterogeneity in the stock environment can result in different ﬁngerprints for different stock components (Campana et al., 2000; Thorrold et al., 1998a). Perhaps more importantly, ontogenetic effects and agerelated differences in exposure history can result in very different ﬁngerprints for ﬁsh of different size classes from the same population (Edmonds et al., 1989; Hoff and Fuiman, 1993; Campana et al., 1995, 2000; Begg et al., 1998; Begg and Weidman, 2001). Since the elemental ﬁngerprint reﬂects the exposure of the individual ﬁsh to both the environment and its own physiology, it would be expected to differ among any groups of ﬁsh which have experienced different histories, whether or not the groups come from the same population. Logically, the presence of different ﬁngerprints could not be used to infer the length of time that the groups of ﬁsh remained separate, since even occasional residency in a different environment would have the potential to introduce a detectable difference in the elemental composition. By corollary, the absence of differences would not necessarily imply that the groups of ﬁsh are of common origin. As a result, it is fair to categorize otolith elemental ﬁngerprints as powerful discriminators of groups when differences exist, but of negligible value when differences cannot be detected. Where differences are detected, additional information would be required to determine if the groups actually corresponded to stocks or populations. Nevertheless, the presence of different ﬁngerprints among groups of ﬁsh of similar age necessarily implies different environmental histories. To the extent that populations or stocks of ﬁsh inhabit different environments, otolith elemental composition can then serve as an indicator of stock identity. Use of the ﬁngerprint as a long-term stock discriminator may be justiﬁed in instances where environmental differences among stock areas are larger than those within areas or across year-classes, and where the effect of size-related effects on the ﬁngerprint have been statistically removed. The assumption of long-term stability in the ﬁngerprint is probably met in some, but not all, stocks.

230

Steven E. Campana

In principle, the otolith elemental composition can be used to identify and track any groups of ﬁsh of different environmental history. In practice, there are three practical limitations to their use. The ﬁrst limitation is that many of the most easily measured elements are under strict physiological regulation, and thus unsuitable for use as environmental indicators. This list includes the major elements calcium, oxygen, and carbon (which make up the calcium carbonate matrix), as well as the minor (>100 ppm) elements Na, K, S, P, and Cl, although it excludes Sr (Thresher et al., 1994; Proctor et al., 1995; Schwarcz et al., 1998). Nevertheless, even physiologically regulated elements can prove useful as biological tracers of a group of ﬁsh, as long as the otolith concentrations of those elements vary signiﬁcantly among groups. The second limitation concerns the analysis of the less abundant trace (<100 ppm) elements, which appear to be more suitable as environmental indicators. While their lower concentrations makes them less likely to be osmoregulated by the ﬁsh, it also makes them more difﬁcult to assay with accuracy and without contamination during handling. Finally, few if any trace elements (even when normalized to Ca) are likely to be incorporated into the otolith in direct proportion to availability in the environment (Hanson and Zdanowicz, 1999). Both temperature and growth rate are known to be at least as inﬂuential as ambient concentration in modifying otolith elemental composition. With these caveats in mind, the elements most likely to serve as environmentally inﬂuenced stock markers include Sr, Ba, Mn, Fe, and Pb (and perhaps Li, Mg, Cu, and Ni), in which both ambient element:Ca concentrations and/or temperature produce signiﬁcant effects on otolith composition (Fowler et al., 1995; Farrell and Campana, 1996; Dove, 1997; Geffen et al., 1998; Campana, 1999; Bath et al., 2000; Milton and Chenery, 2001). The isotopes most useful as natural tags are Sr, S, Pb, and O (Kennedy et al., 1997, 2000; Spencer et al., 2000; Thorrold et al., 2001; Weber et al., 2002). Two types of elemental ﬁngerprinting are in general use: one based on whole dissolved otoliths, and the other based on analysis of the otolith core. Since the otolith grows continually throughout the life of the ﬁsh, the whole-otolith ﬁngerprint integrates across the entire lifetime and thus serves as a marker for groups of ﬁsh that have experienced different overall environmental exposures. The ﬁngerprint thus serves as a natural tag of these groups of ﬁsh if they were to mix with other groups shortly after characterization of their ﬁngerprints. In contrast, analysis of the otolith core is generally intended as a more direct measure of stock or nursery origin. Both of these approaches are discussed further below. The most robust application of whole-otolith ﬁngerprints is one which is targeted at questions of stock mixing or for tracking stock migrations, in which the ﬁngerprints are used as natural tags of predeﬁned groups of ﬁsh over short periods of time (Campana et al., 1995, 1999, 2000; Gillanders and Kingsford, 1996; Kennedy et al., 1997, 2000). Application of an elemental ﬁngerprint as a natural tag takes advantage of the fact that otolith size and composition cannot

231

Otolith Elemental Composition as a Natural Marker of Fish Stocks

change appreciably over a brief time period. Once the elemental ﬁngerprint of all potential source groups has been determined, ﬁsh should remain identiﬁable as to their source group, despite any mixing with other groups, until the elemental composition of later otolith growth has signiﬁcantly altered overall elemental composition. The ﬁngerprint would not be expected to remain stable over extended periods of time (e.g., years), since interannual variation in the habits and environment of the ﬁsh would eventually produce a detectable change in the overall elemental composition (Fig. 12-2). However, short-term stability is both expected and observed, particularly with respect to differences among groups (Fig. 12-3) (Campana et al., 1995; Kennedy et al., 1997; Thorrold et al., 1998b). An appealing feature of this application is that the elemental ﬁngerprint need not be linked to potential sources or locations in the environment.

4000

6

N Shelf

SW Gulf 5

Ba (μg/g)

Sr (μg/g)

3500

SE Gulf

4

SW Gulf

3000 3

N Shelf

E Shelf 2500

E Shelf

SE Gulf

2 1984

1988

1992

1996

1984

1988

1992

1996

3.3

1.8 1.6

1.2

LN Mg (μg/g)

Li (μg/g)

1.4

E Shelf SW Gulf

1.0

E Shelf 2.8

N Shelf SE Gulf

SE Gulf

.8 .6

N Shelf

SW Gulf 2.3

.4 1984

1988

1992

1996

1984

1988

1992

1996

FIGURE 12-2. Long-term variation in mean (±95% CI) elemental concentration in cod otoliths at four spawning sites off eastern Canada. Long-term stability was noted for some elements at some locations, but not others, nor did all elements at a given location necessarily change in tandem. E Shelf, eastern Scotian Shelf; N Shelf, northern Scotian Shelf; SE Gulf, southeast Gulf of St. Lawrence; SW Gulf, southwest Gulf of St. Lawrence. Full details of the cod study are available in Campana et al. (2000).

232

Steven E. Campana

1.0 E Shelf S Gulf

DF 2

.5

N Shelf

0.0 S NF

-.5

-1.0 -2.0

N Gulf

-1.0

0.0

1.0

DF 1 FIGURE 12-3. Short-term stability and speciﬁcity of the elemental ﬁngerprint as a marker of cod spawning aggregations off eastern Canada. Differences among groups were highly signiﬁcant, but remained relatively stable across adjacent years (1995: open symbol; 1996: ﬁlled symbol; 1997: hatched symbol). DF, discriminant function; E Shelf, eastern Scotian Shelf; N Shelf, northern Scotian Shelf; S Gulf, southern Gulf of St. Lawrence; N Gulf, northern Gulf of St. Lawrence; S NF, southern Newfoundland. Full details of the cod study are available in Campana et al. (2000).

Irrespective of the cause of the differences in elemental ﬁngerprints among the groups, the ﬁngerprints become the natural distinguishing feature of those groups at a given point in time. Accordingly, otolith elemental ﬁngerprints appear to be well suited as biological tracers of groups of ﬁsh, requiring relatively few assumptions for conﬁdent application to difﬁcult tracking or stock mixing situations. Use of otolith elemental ﬁngerprints as natural tags makes three central assumptions, all of which apply as much to genetic and morphometric stock mixture analyses as to otolith-based assays (Wood et al., 1989; Wirgin et al., 1997): Assumption 1: There are characteristic and reproducible markers for each group. If the elemental ﬁngerprints of the groups of interest do not differ signiﬁcantly, little more can be accomplished. However, group-speciﬁc variation in elemental composition appears to be common (Edmonds et al., 1989, 1991, 1992, 1995; Northcote et al., 1992; Campana and Gagné, 1995; Campana et al., 1995, 1999; Bronte et al., 1996; Thorrold et al., 1998a). Since smaller ﬁsh seem far more likely to contain elevated (or depressed) concentrations of any given element than larger ﬁsh (Edmonds et al., 1989; Hoff and Fuiman, 1993), it is important that differences in elemental concentration among groups not be confounded by size dif-

Otolith Elemental Composition as a Natural Marker of Fish Stocks

233

ferences among groups. Statistical removal of the effect of otolith weight on elemental concentration is a ready solution to this problem (Campana et al., 2000). Assumption 2: All possible groups contributing to the group mixture have been characterized. This assumption applies as much to genetic studies as it does to otolith elemental ﬁngerprints, with the implication being that uncharacterized groups of ﬁsh present in the mixture could be mistakenly interpreted as one or more of the reference groups (Wood et al., 1987, 1989; Wirgin et al., 1997). Careful selection of reference groups can help minimize this problem, particularly if they are sampled at a time when the groups are known to be discrete (e.g., on the spawning or nursery grounds). Assumption 3: The marker remains stable over the interval between characterization and mixing. Long-term stability of an environmentally induced marker would not be expected, nor has it been observed (Edmonds et al., 1995; Campana et al., 2000; Begg and Weidman, 2001). However, short-term stability over the interval between characterization (e.g., spawning group) and mixing is both expected and observed, particularly with respect to differences among groups (Campana et al., 1995, 2000; Kennedy et al., 1997, 2000; Thorrold et al., 1998a). In the case of Atlantic cod (Gadus morhua), the interval between characterization and mixing was less than 6 months, and thus much less than the 1- to 2-yr period required for a noticeable change in the elemental ﬁngerprint (Campana et al., 2000). Longer intervals may be possible in some instances, but the potential for drift in elemental composition becomes greater as the interval length is extended (Edmonds et al., 1995). By corollary, shorter interval lengths would presumably be required in analyses of young ﬁsh, in which the proportional annual change in otolith weight (and potentially, composition) would be more marked. In general, the period of relative stability can probably be approximated as the period during which the mean otolith weight increases by <5%. Analysis of the otolith core is generally intended as a more direct measure of stock origin than is the analysis of the whole otolith. As is the case with wholeotolith elemental ﬁngerprints, the presence of ﬁngerprint differences implies differences in the history of environmental exposure which may or may not correspond to genetic differences. In this application, however, the environmental exposure is limited to the period of growth represented by the otolith material that is assayed, whether that is the period around hatch, the ﬁrst few months of life, or some other period. The subsequent life history is not sampled and therefore is irrelevant. To the extent that spawning or nursery grounds are characterized by different temperature or chemical environments, this approach has proved effective in distinguishing among groups of ﬁsh with different origins (Kalish, 1990; Sie and Thresher, 1992; Campana et al., 1994; Thresher et al., 1994; Proctor et al., 1995; Severin et al., 1995; Dove et al., 1996; Gillanders and

234

Steven E. Campana

Kingsford, 1996; Milton et al., 1997; Thorrold et al., 1997b, 2001; Gillanders, 2002).

II. SAMPLING AND ASSAYS A distinctive and powerful feature of the ﬁeld of otolith chemistry is that the assays can either be restricted to some portion of the ﬁsh’s life history or integrated across the entire lifetime of the ﬁsh. In other words, the scale of sampling can be modiﬁed to address the hypothesis being tested, through analysis of either the entire otolith or through a targeted assay of a speciﬁc region. In general, analyses of whole otoliths are best suited for use as a natural tag, since the primary question is one of overall differences between groups of ﬁsh, irrespective of the portion of the lifetime which produced the difference. In contrast, microsampled or beam-based assays can target a particular range of ages or dates and thus take advantage of the chronological growth sequence recorded in the otolith to detect differences between groups at some earlier life stage. Currently, bulk and/or solution-based elemental assays are capable of better accuracy, precision, and/or sensitivity than are most beam-based assay techniques, a factor that must be considered given the exceedingly low concentrations of many otolith trace elements (Campana, 1999). Advantages of whole-otolith assays for use as a natural tag include ease of preparation, absence of error associated with sampling or identifying growth increments, and the availability of accurate and precise assay protocols. The major disadvantage is associated with the inability to take advantage of the chronological growth sequence recorded in the otolith. Atomic absorption spectrometry (AAS) (Hoff and Fuiman, 1993), inductively coupled plasma atomic emission spectroscopy (ICP-AES) (Edmonds et al., 1995), neutron activation analysis (Papadopoulou et al., 1980), and inductively coupled plasma mass spectrometry (ICPMS) (Edmonds et al., 1991; Dove et al., 1996) are among the techniques that have been used to analyze otoliths. However, it is ICPMS that has emerged as the instrument of choice for such assays due largely to its capability for rapid and accurate isotopic and elemental assays over a wide range of elements and concentrations. Isotope dilution ICPMS (ID-ICPMS), a variant of ICPMS often used to certify reference materials (Fassett and Paulsen, 1989), is the most accurate of the otolith analytical techniques currently available (Campana et al., 1999). Sample sizes required for most of the above assays are on the order of 5 to 10 mg of otolith material, although ICPMS units outﬁtted with high efﬁciency nebulizers are capable of handling otolith weights as low as 0.3 mg (Thorrold et al., 1998a). Beam-based assays target a particular age or date range in the sectioned otolith and thus can be used to identify nursery areas (Thresher et al., 1994; Milton

Otolith Elemental Composition as a Natural Marker of Fish Stocks

235

et al., 1997; Thorrold et al., 2001; Gillanders, 2002). The advantages of an agestructured approach are obvious, particularly since the beam sizes of the current generation of instruments approach the width of a typical daily increment. As a result, the assay can be limited to the time scale of interest, which for stock identiﬁcation purposes, would be the early life history. Disadvantages of the approach include the requirement for sectioning to expose the growth sequence, the potential for contamination from the sectioning and polishing procedure, and some degree of beam penetration into underlying growth layers. The assumption that elemental concentration is independent of the growth axis is discussed in Campana (1999). There are a wide variety of sophisticated instruments available for probed assays of the otolith, but the most frequently used include the energy-dispersive (ED-EM) and wavelength-dispersive (WD-EM) electron microprobes (Gunn et al., 1992), proton-induced X-ray emission (PIXE) (Sie and Thresher, 1992), and laser ablation ICPMS (LA-ICPMS) (Campana et al., 1994; Thorrold and Shuttleworth, 2000). In a detailed experimental comparison among the above instruments, Campana et al. (1997) noted that no one instrument type was sensitive to each element, nor was any one instrument preferred for use in all assays. In general, however, the minor elements such as Na and K could only be measured accurately with an electron microprobe, while the trace elements generally used in stock identiﬁcation studies required PIXE or LA-ICPMS. Sr was measured accurately and precisely with either WD-EM, PIXE, or LA-ICPMS. Assays for stable isotope ratios are examples of applications where a particular age or date range in the otolith is required, but beam-based assay techniques are either inappropriate or insufﬁciently sensitive. For these applications, the best alternative often involves microsampling or coring techniques that physically remove a portion of the otolith for subsequent analysis. Computerized micromilling machines have proved effective in some studies, whereby seasonal or annual growth zones visible in otolith cross sections are milled to a discrete depth and the powder collected for assay (Wurster et al., 1999). Controlled acid dissolution of overlying material has also been reported, although the acid apparently leached some material from the core (Dove et al., 1996). The advantages of microsampling include access to bulk analytical techniques of high sensitivity and accuracy: mass spectrometry for stable isotope ratios (Schwarcz et al., 1998), and ICPMS for trace element assays (Campana et al., 1995). The disadvantage is one of limited sampling resolution, since the temporal resolution of the extracted sample is seasonal at best. It appears unlikely that microsampling or coring would introduce contaminants that would confound stable isotope assays as long as the extracted samples were treated carefully. On the other hand, there is potential for contamination from the sampling process on trace element assays, despite the fact that Dove et al. (1996) reported no artifacts due to sectioning with an Isomet saw.

236 A. SAMPLE PREPARATION

Steven E. Campana

AND

QUALITY CONTROL

With analytical sensitivity comes the potential for contamination from unwanted sources. Factors such as the the mode of ﬁsh or otolith preservation, composition of the instruments used to remove the otolith from the ﬁsh, cleaning methods, handling, and even household dust are all potentially major modiﬁers of the perceived trace element composition (Milton and Chenery, 1998; Thresher, 1999). Preservation in ﬂuids such as ethanol and formalin appears to have the greatest potential for contamination, given the microchannel architecture of the otolith and the relative impurity of most preservatives. Therefore, trace element analysis of otoliths stored dry or frozen appears to be safest. Current protocols for handling and preparing otoliths are drawn from the water analysis literature and always involve isolation from skin, metallic instruments, and solutions that are of other than trace metal grade. In general, decontamination based on brushing and soniﬁcation in ultrapure water, followed by storage in acid-washed polyethylene vials, results in minimal contamination (Campana, 1999). Minor elements such as Na, K, Cl, and S appear to be affected by the water soniﬁcation stage (Proctor and Thresher, 1998), perhaps because these elements are incorporated by occlusion and are not lattice bound. However, it is equally probable that such poorly bound elements would be severely affected by exposure to any ﬂuid, including the endolymph if it shifts its composition during the death of the ﬁsh. As a result, such elements would probably not be well suited for use as stable biological tracers. Acid washing of otoliths does not appear to be necessary for elements such as Ba, Mg, Sr, and Li (Campana et al., 2000; Secor et al., 2001), despite the fact that it is an important step in the decontamination of sedimentladen forams. Complete protocols for handling and preparing otoliths for elemental assay are presented elsewhere (Campana et al., 2000). A simpliﬁed protocol is as follows: 1. Remove sagittal otolith pair from ﬁsh immediately after capture; alternatively, freeze ﬁsh, but do not store in liquid preservative. 2. Upon otolith removal, immediately remove all adhering tissue. Handling with metal forceps at this stage is acceptable. 3. Decontaminate otolith by sonifying in a series of distilled, deionized, reverse osmosis water baths (Super Q or Milli Q water) in acid-washed polyethylene vials. Brushing with an acid-washed nylon toothbrush under a ﬂow of Super Q water can be used to remove any adherent tissue before the ﬁrst soniﬁcation. All handling at this and subsequent stages must be with nonmetallic, acid-washed tools. 4. Dry decontaminated otoliths in a positive-ﬂow laminar-ﬂow fume hood (Class 100); store in dry, acid-washed polyethylene vials.

Otolith Elemental Composition as a Natural Marker of Fish Stocks

237

5. For assay by ICPMS, dissolve otolith in redistilled nitric acid to a concentration of no more than 0.1% w/v; isotope spikes should be added at this stage if analyzing with ID-ICPMS. 6. Randomize assay sequence. Of particular relevance to ICPMS, but applicable to all analytical techniques, is the likelihood of instrument drift (change in sensitivity) during the analysis of large numbers of samples or between instrument days. Since the estimated elemental concentration can be signiﬁcantly affected by this drift, despite the analysis of analytical standards, it is important that the analysis sequence be blocked and randomized so that the order of analysis for any one sample group is spread over the entire analysis sequence (Campana and Gagné, 1995). Use of ID-ICPMS minimizes (although it may not eliminate) instrumental drift. Differences in otolith elemental composition among groups of ﬁsh may be statistically signiﬁcant, but will not necessarily be large. Artifactual but signiﬁcant differences among groups of otolith elemental assays are not uncommon (Campana et al., 1997). Therefore, calibration of separate assay runs or laboratories against an otolith reference powder is highly recommended to ensure that any observed differences among runs are real rather than artifactual (Campana, 1999; Thresher, 1999; Yoshinaga et al., 2000).

B. STATISTICAL ANALYSIS While elemental concentrations are generally reported in terms of microgram per gram of otolith, many studies have noted size-speciﬁc concentrations for some elements which could otherwise be confused for stock-speciﬁc differences (Fig. 12-4). To insure that differences in ﬁsh length and/or otolith weight among samples do not confound any stock-speciﬁc differences in elemental composition, it is important to remove the effect of otolith weight from the statistical analysis. In steelhead trout (Oncorhynchus mykiss) for example, Mg varied signiﬁcantly with otolith weight, while Ba did not. Subtraction of the common within-group linear slope (derived from the ANCOVA) from the Mg data removed the trend from the element–otolith weight relationship. In instances where the element–otolith weight relationship is markedly nonlinear, alternative detrending procedures are possible (Campana et al., 2000). Detrended elemental concentrations should show no obvious residual relationship with otolith weight, but in any event, relative differences among detrended samples should be similar to those based on original data. Within-group distributions of elemental concentrations are sometimes skewed and thus must be transformed prior to statistical analysis. Each otolith is characterized by a suite of several elements; therefore, multivariate statistics are used

Steven E. Campana

12

40

9

30

Mg (μg/g)

Ba (μg/g)

238

6

20

10

3

0

0 1

2

3

4

Otolith weight (mg)

5

6

1

2

3

4

5

6

Otolith weight (mg)

FIGURE 12-4. Examples of the relationship between elemental concentration and otolith weight in steelhead trout. A signiﬁcant negative relationship was noted for Mg, but no relationship was evident for Ba. To insure that variations in otolith weight among samples do not confound the interpretation of differences among areas, the effect of otolith weight should be removed statistically from elements where the relationship exists.

to distinguish among samples. MANOVA is used to test for signiﬁcant differences among samples, while discriminant analysis can be used to prepare two-factor elemental ﬁngerprints for illustrative purposes. Discriminant analysis is not necessarily the best method to classify samples of unknown stock afﬁnity, since it performs poorly when the stock markers are similar. In contrast, stock composition analysis using a maximum likelihood-based method provides maximal discriminatory power in mixed stock situations (Wood et al., 1987; Campana et al., 1999; Gillanders, 2002). A simpliﬁed protocol for the statistical analysis of otolith elemental data is as follows: 1. Examine frequency histograms of the concentration of each element in each group. For elements where the distribution is nonnormal in most or all groups, transform that element appropriately (e.g., ln transform; transformation must be applied to that element in all groups). Remove clearly aberrant outliers (>5 SD away from the mean) from the transformed data if not associated with particularly small or large ﬁsh. 2. Visually and statistically assess each element within each group for a relationship with ﬁsh size or otolith weight. Where a relationship is evident in most or all groups, the effect of the relationship must be removed statistically by subtracting the common, within-group slope (obtained from the ANCOVA of the element with group as the factor and otolith weight as the covariate) from the observed value in each group. Nonlinear relationships must be removed differently, as in Campana et al. (2000).

Otolith Elemental Composition as a Natural Marker of Fish Stocks

239

3. Test for univariate differences in the concentration of each element across groups (e.g., through ANOVA). Error bar plots or box and whisker plots help visualize the intergroup differences. 4. Test for overall differences in the elemental ﬁngerprint among groups using MANOVA. 5. Use stepwise discriminant function analysis to identify the elements that contribute the most to ﬁngerprint differences among groups. Visually assess the differences among groups by plotting the ﬁrst two discriminant function axes against each other. Note that classiﬁcation of unknown ﬁsh using discriminant analysis can give highly inaccurate results and is not recommended. 6. Classify an unknown mixture using a maximum likelihood-based or Bayesian mixture analysis, using the known identity ﬁsh as the reference. Reference ﬁsh must be completely comparable to the unknown ﬁsh, as per the assumptions of the method discussed earlier.

III. CASE STUDIES

A. POPULATION MIXING

OF

ATLANTIC COD

The purpose of this study was to identify the Atlantic cod (Gadus morhua) populations and their proportions contributing to a densely populated overwintering ground off the coast of eastern Canada (Campana et al., 1999, 2000). The overwintering ground was adjacent to four large cod populations suspected of contributing to the overwintering grounds, so the otolith elemental ﬁngerprints of known-identity spawners were used as tracers to classify the unknown winter mixture. The initial phase of the study consisted of collecting otoliths from spawning individuals of all four populations on their respective spawning grounds for use as known-identity reference groups. Later in the same year, samples from throughout the overwintering ground were collected. All otoliths were dissolved and assayed for B, Li, Mg, Zn, Sr, Ba, and Pb using IDICPMS. Mn is monoisotopic; thus, it was assayed with conventional ICPMS using an internal standard. A cod otolith reference powder was used to insure analytical consistency. B, Zn, and Pb exhibited low reproducibility and/or concentrations near the limit of detection, and thus were not considered further. After statistical removal of size effects, signiﬁcant differences were noted in the elemental concentration of Li, Mg, Mn, Sr, and Ba among the spawning groups; therefore this suite of ﬁve elements was used in all multivariate analyses that followed. Using the spawning stock samples as the reference (known-identity) groups, a maximum likelihood-based stock mixture analysis was used to estimate the proportion of each spawning stock present in the overwintering mixture. The

240

Steven E. Campana

results were unequivocal and completely consistent with previous tagging studies.

B. NATAL HOMING

OF

WEAKFISH

The purpose of this study was to estimate the extent of natal homing in weakﬁsh (Cynoscion regalis), a species that spawns in estuaries along the eastern coast of North America (Thorrold et al., 1998b, 2001). By using the otolith elemental ﬁngerprints of one year-class of juvenile weakﬁsh (while on their nursery grounds) as reference marks, it was possible to identify and classify as to estuarine origin the same year-class of ﬁsh when they returned to the estuaries years later to spawn. Juvenile otoliths were collected from ﬁve major estuarine nursery grounds; these served as the known-identity reference groups. One sagittal otolith of each pair was assayed for the elements B, Mg, Zn, Sr, and Ba using ID-ICPMS, and Ca and Mn using ICPMS. The remaining otolith was assayed for the stable isotopes d13C and d18O using isotope ratio mass spectrometry. Due to low reproducibility and concentrations near the limit of detection, B and Zn were not considered further. Thus, the combination of the trace elements Mg, Mn, Sr, and Ba (each standardized to Ca concentration) and the stable isotopes d13C and d18O was collectively considered the elemental ﬁngerprint, and these differed signiﬁcantly among the estuaries. Signiﬁcant relationships between concentration and otolith weight were noted for some elements, but since they were not consistent across all estuaries, they were assumed to be unrelated to ﬁsh size. To compare the elemental ﬁngerprint of the returning adult to that of the juvenile, one of the adult otoliths from each pair was sectioned to expose the core corresponding to the juvenile region, which was then assayed with a UV laser ablation system coupled to an ICPMS (LA-ICPMS). The core of the remaining otolith from each pair was microsampled for stable isotope assay. Various univariate and multivariate analyses were carried out to conﬁrm the similarity of juvenile and adult ﬁngerprints. A maximum likelihood-based mixture analysis program was used to estimate the proportion of the returning adults at each estuary which originated as juveniles from that same estuary. The results indicated that the degree of natal homing was very high.

C. ESTUARINE CONTRIBUTION ADULT FISHERY

OF JUVENILE

SNAPPER

TO THE

The purpose of this study was to estimate the proportional contribution of several estuarine nursery grounds to the adult ﬁsh caught in the ﬁshery (Gillanders, 2002). Juvenile snapper (Pagrus auratus) were collected from 15 estuaries off the

Otolith Elemental Composition as a Natural Marker of Fish Stocks

241

southeastern coast of Australia for use as known-identity reference groups. Otoliths were sectioned through the core, then assayed for the elements Mg, Ca, Mn, Sr, and Ba using LA-ICPMS. Calcium was used as an internal standard to correct for variations in ablation yield, and a snapper otolith reference powder was used to insure analytical consistency. Otoliths from adult ﬁsh sampled from the commercial ﬁshery were similarly sectioned; ﬁsh aged as being of the same year-class as that of the juveniles were then assayed with LA-ICPMS. Both univariate and multivariate tests indicated that there were signiﬁcant ﬁngerprint differences among estuaries, based largely on Mn, Sr, and Ba. Maximum likelihood-based estimation indicated that most adult snapper recruited from nearby estuaries, and that juvenile exchange among estuaries was limited.

D. DETERMINATION ATLANTIC SALMON

OF

RIVER

OF

ORIGIN

OF

The purpose of this study was to reconstruct the movement of individual Atlantic salmon (Salmo salar) among and within streams based on the strontium isotopic composition of the otoliths (Kennedy et al., 1997, 2000). Juvenile salmon were collected from 29 sites in two river systems. Sagittal otoliths were dissolved for assay of 87Sr/86Sr using a thermal ionization mass spectrometer (TIMS). There were signiﬁcant differences in the isotopic composition of both vertebrae and otoliths among the various streams. Therefore, salmon from most of the streams could be clearly differentiated from one another based on Sr isotope composition. Deviations in the ﬁsh’s isotopic composition from that of the surrounding water indicated that some ﬁsh had migrated into a stream from another stream of different isotopic composition; in several instances, the source stream could be identiﬁed. This work set the stage for a subsequent study which reconstructed the juvenile life history based on micromilled otoliths of sea-run adults (Kennedy et al., 2002).

IV. CONCLUSION Speciﬁc elements and isotopes incorporated into the growing surface of the ﬁsh otolith reﬂect the physical and chemical characteristics of the ambient water, although not necessarily in a simplistic manner. Since ﬁsh that spend at least part of their lives in different water masses often produce otoliths of different elemental composition, the otolith elemental composition (“elemental ﬁngerprint”) can serve as an environmentally induced tag of groups of ﬁsh. These tags tend to be physically stable, reproducible, and different among stocks, but are not necessarily stable over long periods. Thus, they do not serve as a proxy for genetic

242

Steven E. Campana

identity. However, the ﬁngerprint is very stable over the short term, making it valuable as a seasonally stable biological tracer of predeﬁned groups of ﬁsh. Alternatively, the ﬁngerprint of the otolith core can be used as a marker for groups of ﬁsh hatched in different environments. Technological advancements in recent years have made otolith elemental ﬁngerprints a viable, and sometimes preferable, means for distinguishing among ﬁsh stocks.

REFERENCES Bath, G. E., Thorrold, S. R., Jones, C. M., Campana, S. E., McLaren, J. W., and Lam, J. W. H. 2000. Strontium and barium uptake in aragonitic otoliths of marine ﬁsh. Geochimica Cosmochimica Acta 64: 1705–1714. Begg, G. A., Cappo, M., Cameron, D. S., Boyle, S., and Sellin, M. J. 1998. Stock discrimination of school mackerel, Scomberomorus queenslandicus, and spotted mackerel, Scomberomorus munroi, in coastal waters of eastern Australia by analysis of minor and trace elements in whole otoliths. Fisheries Bulletin, U.S. 96: 653–666. Begg, G. A. and Weidman, C. R. 2001. Stable d13C and d18O isotopes in otoliths of haddock Melanogrammus aegleﬁnus from the northwest Atlantic Ocean. Marine Ecology Progress Series 216: 223–233. Bronte, C. R., Hesselberg, R. J., Shoesmith, J. A., and Hoff, M. H. 1996. Discrimination among spawning concentrations of Lake Superior lake herring based on trace element proﬁles in sagittae. Transactions of the American Fisheries Society 125: 852–859. Campana, S. E. 1999. Chemistry and composition of ﬁsh otoliths: pathways, mechanisms and applications. Marine Ecology Progress Series 188: 263–297. Campana, S. E., Chouinard, G. A., Hanson, J. M., and Fréchet, A. 1999. Mixing and migration of overwintering cod stocks near the mouth of the Gulf of St. Lawrence. Canadian Journal of Fisheries and Aquatic Sciences 56: 1873–1881. Campana, S. E., Chouinard, G. A., Hanson, M., Fréchet, A., and Brattey, J. 2000. Otolith elemental ﬁngerprints as biological tracers of ﬁsh stocks. Fisheries Research 46: 343–357. Campana, S. E., Fowler, A. J., and Jones, C. M. 1994. Otolith elemental ﬁngerprinting for stock identiﬁcation of Atlantic cod (Gadus morhua) using laser ablation ICPMS. Canadian Journal of Fisheries and Aquatic Sciences 51: 1942–1950. Campana, S. E. and Gagné, J. A. 1995. Cod stock discrimination using ICPMS elemental assays of otoliths. In D. H. Secor, J. M. Dean, and S. E. Campana (eds.), Recent Developments in Fish Otolith Research. University of South Carolina Press, Columbia, SC, pp. 671–691. Campana, S. E., Gagne, J. A., and McLaren, J. W. 1995. Elemental ﬁngerprinting of ﬁsh otoliths using ID-ICPMS. Marine Ecology Progress Series 122: 115–120. Campana, S. E. and Neilson, J. D. 1985. Microstructure of ﬁsh otoliths. Canadian Journal of Fisheries and Aquatic Sciences 42: 1014–1032. Campana, S. E., Thorrold, S. R., Jones, C. M., Günther, D., Tubrett, M., Longerich, H., Jackson, S., Halden, N. M., Kalish, J. M., Piccoli, P., de Pontual, H., Troadec, H., Panﬁli, J., Secor, D. H., Severin, K. P., Sie, S. H., Thresher, R., Teesdale, W. J., and Campbell, J. L. 1997. Comparison of accuracy, precision and sensitivity in elemental assays of ﬁsh otoliths using the electron microprobe, proton-induced X-ray emission, and laser ablation inductively coupled plasma mass spectrometry. Canadian Journal of Fisheries and Aquatic Sciences 54: 2068–2079. Dove, S. G. 1997. The incorporation of trace metals into the eye lenses and otoliths of ﬁsh. Ph.D. Thesis, University of Sydney, Sydney, Australia.

Otolith Elemental Composition as a Natural Marker of Fish Stocks

243

Dove, S. G., Gillanders, B. M., and Kingsford, M. J. 1996. An investigation of chronological differences in the deposition of trace metals in the otoliths of two temperate reef ﬁshes. Journal of Experimental Marine Biology and Ecology 205: 15–33. Dufour, V., Pierre, C., and Rancher, J. 1998. Stable isotopes in ﬁsh otoliths discriminate between lagoonal and oceanic residents of Taiaro Atoll (Tuaotu Archipelago, French Polynesia). Coral Reefs 17: 23–28. Edmonds, J. S., Caputi, N., Moran, M. J., Fletcher, W. J., and Morita, M. 1995. Population discrimination by variation in concentrations of minor and trace elements in sagittae of two Western Australian teleosts. In D. H. Secor, J. M. Dean, and S. E. Campana (eds.), Recent Developments in Fish Otolith Research. University of South Carolina Press, Columbia, SC, pp. 655–670. Edmonds, J. S., Caputi, N., and Morita, M. 1991. Stock discrimination by trace-element analysis of otoliths of orange roughy (Hoplostethus atlanticus), a deep-water marine teleost. Australian Journal of Marine and Freshwater Research 42: 383–389. Edmonds, J. S., Lenanton, R. C. J., Caputi, N., and Morita, M. 1992. Trace elements in the otoliths of yellow-eye mullet (Aldrichetta forsteri) as an aid to stock identiﬁcation. Fisheries Research 13: 39–51. Edmonds, J. S., Moran, M. J., Caputi, N., and Morita, M. 1989. Trace element analysis of ﬁsh sagittae as an aid to stock identiﬁcation: Pink snapper (Chrysophrys auratus) in Western Australia waters. Canadian Journal of Fisheries and Aquatic Sciences 46: 50–54. Farrell, J. and Campana, S. E. 1996. Regulation of calcium and strontium deposition on the otoliths of juvenile tilapia, Oreochromis niloticus. Comparative Biochemistry and Physiology 115A: 103–109. Fassett, J. D. and Paulsen, P. J. 1989. Isotope dilution mass spectrometry for accurate elemental analysis. Analytical Chemistry 61: 643A–649A. Fowler, A. J., Campana, S. E., Jones, C. M., and Thorrold, S. R. 1995. Experimental assessment of the effect of temperature and salinity on elemental composition of otoliths using solution-based ICPMS. Canadian Journal of Fisheries and Aquatic Sciences 52: 1421–1430. Gallahar, N. K. and Kingsford, M. J. 1996. Factors inﬂuencing Sr/Ca ratios in otoliths of Girella elevata: an experimental investigation. Journal of Fish Biology 48: 174–186. Geffen, A. J., Pearce, N. J. G., and Perkins, W. T. 1998. Metal concentrations in ﬁsh otoliths in relation to body composition after laboratory exposure to mercury and lead. Marine Ecology Progress Series 165: 235–245. Gillanders, B. 2002. Connectivity between juvenile and adult ﬁsh populations: do adults remain near their recruitment estuaries? Marine Ecology Progress Series 240: 215–223. Gillanders, B. M. 2001. Trace metals in four structures of ﬁsh and their use for estimates of stock structure. Fisheries Bulletin, U.S. 99: 410–419. Gillanders, B. M. and Kingsford, M. J. 1996. Elements in otoliths may elucidate the contribution of estuarine recruitment to sustaining coastal reef populations of a temperate reef ﬁsh. Marine Ecology Progress Series 141: 13–20. Gunn, J. S., Harrowﬁeld, I. R., Proctor, C. H., and Thresher, R. E. 1992. Electron probe microanalysis of ﬁsh otoliths—evaluation of techniques for studying age and stock discrimination. Journal of Experimental Marine Biology and Ecology 158: 1–36. Hanson, P. J. and Zdanowicz, V. S. 1999. Elemental composition of otoliths from Atlantic croaker along an estuarine pollution gradient. Journal of Fish Biology 54: 656–668. Hoff, G. R. and Fuiman, L. A. 1993. Morphometry and composition of red drum otoliths: changes associated with temperature, somatic growth rate and age. Comparative Biochemistry and Physiology 106A: 209–219. Kalish, J. M. 1989. Otolith microchemistry: validation of the effects of physiology, age and environment on otolith composition. Journal of Experimental Marine Biology and Ecology 132: 151–178.

244

Steven E. Campana

Kalish, J. M. 1990. Use of otolith microchemistry to distinguish the progeny of sympatric anadromous and non-anadromous salmonids. Fisheries Bulletin, U.S., U.S. 88: 657–666. Kennedy, B. D., Folt, C. L., Blum, J. D., and Chamberlain, C. P. 1997. Natural isotope markers in salmon. Nature 387: 766–767. Kennedy, B. P., Blum, J. D., Folt, C. L., and Nislow, K. H. 2000. Using natural strontium isotopic signatures as ﬁsh markers: methodology and application. Canadian Journal of Fisheries and Aquatic Sciences 57: 2280–2292. Kennedy, B. P., Klaue, A., Blum, J. D., Folt, C. L., and Nislow, K. H. 2002. Reconstructing the lives of ﬁsh using Sr isotopes in otoliths. Canadian Journal of Fisheries and Aquatic Sciences 59: 925–929. Limburg, K. E. 1998. Anomalous migrations of anadromous herrings revealed with natural chemical tracers. Canadian Journal of Fisheries and Aquatic Sciences 55: 431–437. Milton, D. A. and Chenery, S. R. 1998. The effect of otolith storage methods on the concentrations of elements detected by laser-ablation ICPMS. Journal of Fish Biology 53: 785–794. Milton, D. A. and Chenery, S. R. 2001. Sources and uptake of trace metals in otoliths of juvenile barramundi (Lates calcarifer). Journal of Experimental Marine Biology and Ecology 264: 47–65. Milton, D. A., Chenery, S. R., Farmer, M. J., and Blaber, S. J. M. 1997. Identifying the spawning estuaries of the tropical shad, terubok Tenualosa toli, using otolith microchemistry. Marine Ecology Progress Series 153: 283–291. Newman, S. J., Steckis, R. A., Edmonds, J. S., and Lloyd, J. 2000. Stock structure of the goldband snapper Pristipomoides multidens (Pisces: Lutjanidae) from the waters of northern and western Australia by stable isotope ratio analysis of sagittal otolith carbonate. Marine Ecology Progress Series 198: 239–247. Northcote, T. G., Hendy, C. H., Nelson, C. S., and Boubee, J. A. T. 1992. Tests for migratory history of the New Zealand common smelt (Retropinna retropinna) using otolith isotopic composition. Ecology of Freshwater Fish 1: 61–72. Papadopoulou, C., Kanias, G. D., and Kassimati, E. 1980. Trace element content in ﬁsh otoliths in relation to age and size. Marine Pollution Bulletin 11: 68–72. Proctor, C. H. and Thresher, R. E. 1998. Effects of specimen handling and otolith preparation on concentration of elements in ﬁsh otoliths. Marine Biology 131: 681–694. Proctor, C. H., Thresher, R. E., Gunn, J. S., Mills, D. J., Harrowﬁeld, I. R., and Sie, S. H. 1995. Stock structure of the southern blueﬁn tuna Thunnus maccoyi: an investigation based on probe microanalysis of otolith composition. Marine Biology 122: 511–526. Schwarcz, H. P., Gao, Y., Campana, S., Browne, D., Knyf, M., and Brand, U. 1998. Stable carbon isotope variations in otoliths of Atlantic cod (Gadus morhua). Canadian Journal of Fisheries and Aquatic Sciences 55: 1798–1806. Secor, D. H., Campana, S. E., Zdanowicz, V. S., Lam, J. W. H., Yang, L., and Rooker, J. R. 2002. Interlaboratory comparison of Atlantic and Mediterranean blueﬁn tuna otolith microconstituents. ICES Journal of Marine Science 59: 1294–1304. Secor, D. H., Rooker, J. R., Zlokovitz, E., and Zdanowicz, V. S. 2001. Identiﬁcation of riverine, estuarine, and coastal contingents of Hudson River striped bass based upon otolith elemental ﬁngerprints. Marine Ecology Progress Series 211: 245–253. Severin, K. P., Carroll, J., and Norcross, B. L. 1995. Electron microprobe analysis of juvenile walleye pollock, Theragra chalcogramma, otoliths from Alaska: a pilot stock separation study. Environmental Biology of Fishes 43: 269–283. Sie, S. H. and Thresher, R. E. 1992. Micro-PIXE analysis of ﬁsh otoliths: methodology and evaluation of ﬁrst results for stock discrimination. International Journal of PIXE 2: 357–379. Spencer, K., Shafer, D. J., Gauldie, R. W., and DeCarlo, E. H. 2000. Stable lead isotope ratios from distinct anthropogenic sources in ﬁsh otoliths: a potential nursery ground stock marker. Comparative Biochemistry and Physiology 127A: 273–284.

Otolith Elemental Composition as a Natural Marker of Fish Stocks

245

Thorrold, S. R., Campana, S. E., Jones, C. M., and Swart, P. K. 1997a. Factors determining d13C and d18O fractionation in aragonitic otoliths of marine ﬁsh. Geochimica Cosmochimica Acta 61: 2909–2919. Thorrold, S. R., Jones, C. M., and Campana, S. E. 1997b. Response of otolith microchemistry to environmental variations experienced by larval and juvenile Atlantic croaker (Micropogonias undulatus). Limnology and Oceanography 42: 102–111. Thorrold, S. R., Jones, C. M., Campana, S. E., McLaren, J. W., and Lam, J. W. H. 1998a. Trace element signatures in otoliths record natal river of juvenile American shad (Alosa sapidissima). Limnology and Oceanography 43: 1826–1835. Thorrold, S. R., Jones, C. M., Swart, P. K., and Targett, T. E. 1998b. Accurate classiﬁcation of juvenile weakﬁsh Cynoscion regali to estuarine nursery areas based on chemical signatures in otoliths. Marine Ecology Progress Series 173: 253–265. Thorrold, S. R., Latkoczy, C., Swart, P. K., and Jones, C. M. 2001. Natal homing in a marine ﬁsh metapopulation. Science 291: 297–299. Thorrold, S. R. and Shuttleworth, S. 2000. In situ analysis of trace elements and isotope ratios in ﬁsh otoliths using laser ablation sector ﬁeld inductively coupled plasma mass spectrometry. Canadian Journal of Fisheries and Aquatic Sciences 57: 1232–1242. Thresher, R. E. 1999. Elemental composition of otoliths as a stock delineator in ﬁshes. Fisheries Research 43: 165–204. Thresher, R. E., Proctor, C. H., Gunn, J. S., and Harrowﬁeld, I. R. 1994. An evaluation of electron probe microanalysis of otoliths for stock delineation and identiﬁcation of nursery areas in a southern temperate groundﬁsh, Nemadactylus macropterus (Cheilodactylidae). Fisheries Bulletin, U.S. 92: 817–840. Volk, E. C., Blakley, A., Schroder, S. L., and Kuehner, S. M. 2000. Otolith chemistry reﬂects migratory characteristics of Paciﬁc salmonids: using otolith core chemistry to distinguish maternal associations with sea and freshwaters. Fisheries Research 46: 251–266. Weber, P. K., Hutcheon, I. D., McKeegan, K. D., and Ingram, B. L. 2002. Otolith sulfur isotope method to reconstruct salmon (Oncorhynchus tshawytscha) life history. Canadian Journal of Fisheries and Aquatic Sciences 59: 587–591. Wirgin, I. I., Waldman, J. R., Maceda, L., Stabile, J., and Vecchio, V. J. 1997. Mixed-stock analysis of Atlantic coast striped bass (Morone saxatilis) using nuclear DNA and mitochondrial DNA markers. Canadian Journal of Fisheries and Aquatic Sciences 54: 2814–2826. Wood, C. C., McKinnell, S., Mulligan, T. J., and Fournier, D. A. 1987. Stock identiﬁcation with the maximum-likelihood mixture model: sensitivity analysis and application to complex problems. Canadian Journal of Fisheries and Aquatic Sciences 44: 866–881. Wood, C. C., Rutherford, D. T., and McKinnell, S. 1989. Identiﬁcation of sockeye salmon (Oncorhynchus nerka) stocks in mixed-stock ﬁsheries in British Columbia and southeast Alaska using biological markers. Canadian Journal of Fisheries and Aquatic Sciences 46: 2108–2120. Wurster, C. M., Patterson, W. P., and Cheatham, M. M. 1999. Advances in micromilling techniques: a new apparatus for acquiring high-resolution oxygen and carbon stable isotope values and major/minor elemental ratios from accretionary carbonate. Computers & Geosciences 25: 1159–1166. Yoshinaga, J., Nakama, A., Morita, M., and Edmonds, J. S. 2000. Fish otolith reference material for quality assurance of chemical analyses. Marine Chemistry 69: 91–97.

This Page Intentionally Left Blank

CHAPTER

13

Fatty Acid Proﬁles as Natural Marks for Stock Identiﬁcation O. GRAHL-NIELSEN University of Bergen, Bergen, Norway

I. Introduction II. Methodology III. Case Histories A. Three Stocks of Herring in Norwegian Waters B. North Sea Herring vs. Baltic Herring C. Investigation of Fraud in Fisheries D. Atlantic vs. North Sea Mackerel E. Striped Bass from East American Rivers F. Farmed Salmon Escapees G. Two Faroe Island Stocks of Cod H. Redﬁsh Stocks in the North Atlantic I. Identiﬁcation of Harp Seals on Foraging Migrations IV. Discussion References

I. INTRODUCTION In the lipids of marine animals about 20 fatty acids appear in relative amounts of more than 1%. A larger number are present in minor concentrations. Different species of ﬁsh have characteristic fatty acid proﬁles (Ackman, 1980), but the variability in the composition of the tissue fatty acids of ﬁsh is very large. Different tissues have different fatty acid proﬁles. Within each type of tissue the fatty acids are bound in a number of different lipid classes, phospholipids, triacylglycerides, wax esters, cholesterol esters, and so on, all with different proﬁles (Morris and Culkin, 1989). On top of these variations, the composition of tissue fatty acids may be inﬂuenced by factors such as age, maturity, condition, and reproductive cycle of the ﬁsh. External factors, such as water temperature and salinity, and probably also Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

247

248

O. Grahl-Nielsen

pressure, have been shown to have an effect on fatty acid composition (Greene, 1990). Numerous investigations have shown that the diet of ﬁsh has an impact on the fatty acid proﬁle in ﬁsh tissues (see Sargent et al., 1989). With so many causes for variation of the relative amounts of 20 plus variables, that is, fatty acids, systematic investigations in this ﬁeld were not possible until statistical, multivariate treatment of the analytical data (chemometry) was introduced by the author and researchers emerging from his group in the ﬁrst half of the 1980s (Grahl-Nielsen and Barnung, 1985; Vogt et al., 1986). With principal component analysis, one of the most used multivariate methods, it is possible to distinguish among the different internal and external causes of variation. The method also allows ranking of fatty acids according to their importance for the different variations. Differences in fatty acid proﬁle on the species level are obvious and can be detected without the use of multivariate methods (Ackman, 1980). However, ﬁsh and other marine animals, such as marine mammals are present in distinct populations, and it is a challenge for the chemometric approach to detect systematic differences in fatty acid proﬁles on this level. We have therefore carried out investigations on different populations of striped bass, herring, mackerel, cod, redﬁsh, salmon, and the pinniped, harp seal. Diet is considered to be the most important external factor inﬂuencing the composition of tissue fatty acids. This inﬂuence is mainly on neutral triacylglycerides, which dominate the lipids in muscle and many other tissues of ﬁsh. Polar, functional phospholipids are less prone to dietary inﬂuence. Since we wanted to keep the workup procedure as simple and quick as possible, we avoided extraction and fractionation of the lipids prior to analysis. Instead, we subjected the tissue sample to direct methanolysis, which extracts and converts all fatty acids in a sample to methyl esters, ready for gas chromatography, in one step. To keep the contribution of diet inﬂuenced triacylglycerides in the samples to a minimum, we used heart tissue in our method. In one investigation, lipids in heart tissue of herring were found to contain only 11% triacylglycerides (Grahl-Nielsen and Ulvund, 1990), and in another investigation between 10% and 17% (Meier, 1997). Another advantage with heart is that it is easily identiﬁed and retrieved from the ﬁsh without contamination from other tissues, except blood. However, blood is quickly washed away before subsamples of heart tissue are excised.

II. METHODOLOGY The heart is retrieved and blood is carefully removed by washing in seawater. The tissue is dried on ﬁlter paper, and subsamples of approximately 20 mg are cut from the tip of the heart and transferred to thick-walled glass tubes with teﬂon-

Fatty Acid Proﬁles as Natural Marks for Stock Identiﬁcation

249

lined screw caps. One-half milliliter of the methanolysis reagent, anhydrous methanol, containing 2 N HCl, is added, and the tubes are securely capped. The methanolysis is allowed to proceed for 2 hr at 90°C. All fatty acids in the tissue are then transformed to methyl esters dissolved in the methanol phase. After cooling, approximately half of the methanol is evaporated under a stream of nitrogen gas, and an equivalent amount of distilled water is added to lower the solubility of fatty acid methyl esters. The methanol–water phase is extracted twice with 1 ml hexane, each time mixing the phases in a whirl mixer (Sorvall) and separating them by centrifugation, if necessary. The hexane is withdrawn by a Pasteur pipette, carefully avoiding withdrawal of the bottom phase. One microliter of the combined extracts is gas chromatographed on a capillary column with a polar stationary phase under conditions allowing for proper separation of all fatty acid methyl esters between 14 : 0 and 24 : 1 within about 30 min. Details of chromatographic equipment and procedure are given in Table 13-1. To monitor the performance of the chromatographic system and to establish response factors for the various fatty acids, a mixture of known amounts of 20 fatty acid methyl esters is chromatographed for every ninth sample. The gas chromatographic output is retrieved by a lab data system, for example, Atlas, where the largest peaks, usually between 20 and 30, are selected and integrated. The smallest peaks, that is, those with areas of less than 0.1% of the total area of all peaks, are not included. It is our experience that the analytical uncertainty in the analysis of these small amounts contributes more noise than worthwhile information to the subsequent multivariate treatment. To correct for uneven losses in the chromatographic system, mainly the injector, the integrated areas of the peaks are divided by response factors established from chromatography of the mixture of standards. To obtain the combined information from all selected fatty acids simultaneously, they are subjected to TABLE 13-1. Chromatograph Autosampler Column

Carrier gas Injector Detector Temperature program Labdata system

Chromatographic Equipment and Settings Hewlett-Packard 5890A Hewlett-Packard 7673A 25 m ¥ 0.25 mm CP-WAX 52CB from Chrompack with 0.2 mm polyethylene glycol as stationary phase Helium at 20 psi Split/splitless in splitless mode at 60°C, split opened after 4 min Flame ionization at 330°C 90°C for 4 min, 30°C/min to 165°C, 3°C/min to 225°C, isothermal at 225°C for 10.5 min Atlas from Thermo LabSystems, UK

250

O. Grahl-Nielsen

TABLE 13-2. Fatty Acids in Muscle of Reference Herring from Atlantoscandic Stock and from Test Herring from Atlantoscandic and North Sea Stocks

Fatty acid 14.0 14:1n5 anteiso –15:0 15:0 16.0 16:1 n7 17:0 16:2 n6 16:3 n4 17:1 n9 16:4 n1 18:0 18:1 n9 18:1 n7 18:1 n5 18:2 n6 18:3 n3 18:4 n3 20:0 20:1 n11 20:1 n9 20:2 n6 20:4 n6 20:3 n3 20:4 n3 20:5 n3 22:0 22:1 n11 22:1 n9 22:5 n3 24:0 22:6 n3 24:1 n9

Reference herring (N = 10) 6.5 0.11 0.09 0.51 14 3.9 0.07 0.18 0.18 0.32 0.15 1.2 9 1.2 0.43 1.07 1.2 3.9 0.13 1.1 11 0.29 0.49 0.13 0.58 7.3 0.07 17 1.3 0.56 0.05 14 0.86

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.5 0.02 0.03 0.02 1 0.8 0.01 0.04 0.02 0.02 0.03 0.1 2 0.2 0.06 0.03 0.1 0.3 0.02 0.2 2 0.07 0.07 0.01 0.04 0.6 0.01 3 0.4 0.03 0.04 2 0.09

Atlantoscandic stock (N = 10) 7.4 0.12 0.09 0.51 14 4.4 0.07 0.22 0.17 0.33 0.23 1.1 10 1.2 0.44 1.2 1.3 3.9 0.15 1.11 11 0.29 0.37 0.14 0.58 6.9 0.07 18 1.1 0.55 0.03 11 0.9

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.6 0.01 0.03 0.05 1 0.7 0.01 0.05 0.02 0.02 0.07 0.2 3 0.1 0.07 0.1 0.2 0.6 0.01 0.11 2 0.04 0.05 0.03 0.07 0.8 0.01 3 0.3 0.05 0.01 1 0.1

North Sea stock (N = 10) 8 0.08 0.08 0.46 13 4.7 0.06 0.5 0.21 0.29 0.5 1.1 7.1 1.2 0.29 1.2 0.8 1.8 0.26 1.4 13 0.30 0.48 0.14 0.49 4.7 0.12 26 1.1 0.80 0.04 9 1.0

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

1 0.01 0.01 0.07 1 0.7 0.01 0.1 0.03 0.03 0.1 0.2 1.5 0.2 0.01 0.1 0.1 0.3 0.02 0.2 2 0.06 0.07 0.02 0.06 0.9 0.01 3 0.1 0.05 0.01 2 0.1

multivariate treatment based on principal component analysis (PCA). Their relative values, expressed as a percentage of their sum, are logarithmically transformed, thereby leveling out the large differences between the fatty acids. Other transformations of the data, such as division by the mean values, may also be employed. With each sample positioned in multidimensional space described by log-transformed variables, that is, fatty acids, the coordinates [principal compo-

Fatty Acid Proﬁles as Natural Marks for Stock Identiﬁcation

251

nents (PCs)] that describe the largest and second largest (and in some cases the third largest) variance among the samples are computed by way of a suitable program package such as SIRIUS (Kvalheim and Karstang, 1987). PCA is available in a large number of program packages for statistical analysis. In this manner, the relationship among the samples can be described in two or three dimensions instead of the original 20 plus, without considerable loss of the total original variance. For evaluation of the results, the samples are displayed in the coordinate system of PC1 vs. PC2, or in three-dimentional plots of PC1 vs. PC2 vs. PC3. To detect the fatty acids that are of importance to the identiﬁcation of the stocks, the fatty acids are displayed in the same coordinate system as the samples, resulting in a so-called biplot. When a fatty acid lies close to the samples from a particular stock, these samples contain relatively higher amounts of that fatty acid than samples from other stocks lying farther away in the plot. When the fatty acids that discriminate best among the stocks are detected in this manner, a recomputation of principal components based only on these fatty acids may be carried out, thereby enhancing the stock differences. The distinction between stocks may not necessarily be along one of the principal components that describe the largest, the second largest, or the third largest variation among the samples. In a set of reference samples, the direction in the multidimensional space that gives the best discrimination between the different stocks may be found by a method that is equivalent to discriminant analysis. New and dependent variables are then added to the data matrix. The ﬁrst of these is given the value +1 for all samples belonging to one stock and -1 for all samples belonging to the other stocks. The second dependent variable is given the value +1 for all samples belonging to the second stock and -1 for all samples belonging to the other stocks, and so on. New principal components, which describe the best covariance between the samples in the original matrix and the new variables can then be extracted. This method is a multivariate, supervised learning method, called partial least square (PLS) (Esbensen et al., 1994). The formation of PC plots, biplots, or reﬁned PLS plots is the ﬁrst step in the multivariate treatment of the analytical data. These plots are merely display methods, giving no quantitative information about differences or similarities. In many instances, this is sufﬁcient because clear-cut differences can be seen. Specimens of unknown stock identity may then be identiﬁed by using PC plots or biplots of the reference samples. A quantitative classiﬁcation of the samples may be carried out in a second multivariate step using SIMCA analysis (Soft Independent Modeling of Class Analogy) (Wold and Sjøstrøm, 1977), also available in the SIRIUS package. A principal component model is then created for the group of samples belonging to the same stock. The conﬁdence limit, based on a chosen signiﬁcance level, usually 95% or 99%, around the model is then determined. This is the maximum

252

O. Grahl-Nielsen

residual standard deviation, RSDmax, for the model. When a sample is subjected to identiﬁcation, its RSD with respect to the model is determined, and this will then tell if the sample belongs to the stock in question or, if not, how far removed it is from the stock.

III. CASE HISTORIES

A. THREE STOCKS

OF

HERRING

IN

NORWEGIAN WATERS

Herring from three stocks, a local stock from a Norwegian fjord, Trondheimsfjorden, Norwegian spring spawners, which are ﬁshed along the Norwegian coast betweeen 63° and 68°N, and autumn spawners from the North Sea, were sampled on the spawning grounds. Bone tissue from the jaw and heart tissue were subjected to chemometric analysis (Grahl-Nielsen and Ulvund, 1990). The fatty acid proﬁle of both tissues differed among the three stocks, but the heart tissue gave a somewhat better distinction (Fig. 13-1).

B. NORTH SEA HERRING

VS.

BALTIC HERRING

Herring from a North Sea stock and from a Baltic stock were sampled in July 1991 in the North Sea. The North Sea herring were approximately 50 nautical miles east of Shetland, and the Baltic herring, on foraging migration, were southwest of Southern Norway (Hornnes, 1993). The fatty acid proﬁle of heart tissue was different in herring from the two stocks, and the two stocks were clearly distinguished by PCA (Fig. 13-2).

C. INVESTIGATION

OF

FRAUD

IN

FISHERIES

Two different stocks of herring are targets of Norwegian ﬁsheries, Norwegian spring spawners, NVG, which are ﬁshed along the Norwegian coast betweeen 63° and 68°N, and the North Sea stock, NS, ﬁshed south of the 62nd parallel. The quota for 2000 for Norwegian purse seiners was 400,600 tonnes NVG herring and 72,500 tonnes NS herring. Since less effort is needed to ﬁll the vessels with NVG herring than with NS, it may be tempting for ﬁshermen to register catches of this stock as NS herring. One such case of fraud was recently detected, leading to a harsh punishment of an 8-week ban from ﬁshing. The most used methods for identiﬁcation of stocks of herring are morphological, the average NS herring being smaller and leaner than NVG herring. Mean

253

Fatty Acid Proﬁles as Natural Marks for Stock Identiﬁcation

PC 2 20%

PC 1 48% PC 2 21%

PC 1 50% PC 2 21%

PC 1 58% FIGURE 13-1. PC plots of herring from the North Sea stock (squares), from the Norwegian spring spawning stock (circles), and from a local stock in the Trondheimsfjord (diamonds), based on fatty acids in heart tissue. The percentage variance explained by the principal components is given in each case.

vertebrae number differs slightly between NVG and NS herring, but this method can be used only for identiﬁcation of batches of herring, and needs 100 individuals from each stock. There are also differences in the structure of scales and otoliths from the two stocks.

254

O. Grahl-Nielsen

PC 2 20%

14:0 18:4n3 16:1n7 18:3n3 20:4n6 16:0 22:6n3 18:1n7 18:0

18:1n11 20:1n9 22:1n9

22:1n11 20:1n11

PC 1 62% FIGURE 13-2. PC plot of samples of herring from the North Sea stock (squares) and of herring from the Baltic stock, sampled in the North Sea (circles), based on fatty acids in heart tissue. The fatty acids of largest importance for the discrimination between the stocks are displayed in the same coordinate system.

However, herring is processed on board vessels, and only ﬁllets, free of skin and bone, are usually landed and thus available for identiﬁcation. The challenge was to see if the fatty acid method could be applied to the muscle tissue of the ﬁllets, where triacylglycerols dominate the lipids. The possibility of testing the method occurred in a case where a Norwegian purse seiner landed both NVG herring and NS herring, but the authorities anticipated the possibility that the herring reported as being caught in the North Sea might actually be NVG herring. Ten ﬁllets from each batch were retrieved for chemometric analysis. As reference, 10 herring from the NVG stock were analyzed. Two samples, each weighing approximately 20 mg, were retrieved from white muscle in the front part above the side line, carefully avoiding remaining skin, bone, and brown tissue. The samples were methanolysed and the resulting fatty acid methyl esters were gas chromatographed. The relative amounts of 30 detected methyl esters (Table 13-1) were subjected to multivariate PCA. The resulting PC plot showed a distinct difference between the herring from the two stocks (Fig. 13-3). The herring reported as NVG were similar to reference NVG herring, while the herring reported as North Sea herring were clearly different. In this case, any suspicion about false reporting was unfounded.

255

Fatty Acid Proﬁles as Natural Marks for Stock Identiﬁcation

PC 2 8%

18:4n3 18:3n3

16:4n1 22:1n1120:1n9 14:0

16:2n6

18:1n5 14:1n5

22:5n3 20:0 22:0

20:1n11

20:5n3 22:6n3 18:1n9

PC 1 76% FIGURE 13-3. PC plot of herring from a catch in the North Sea (squares), a catch on the spawning ground of the Norwegian spring spawning stock (circles), and of reference herring from the Norwegian spring spawning stock (diamonds), based on fatty acids in the ﬁllet. Two parallel samples from each herring are shown. The most discriminating fatty acids are shown in the plot.

D. ATLANTIC

VS.

NORTH SEA MACKEREL

Specimens from the Atlantic stock, caught in the spawning areas off Ireland, and from the North Sea stock, caught on their spawning grounds, turned out to have different fatty acid proﬁles of heart tissue (unpublished, Institute of Marine Research, Bergen, Norway). In this case, space-ﬁlling PC models were formed for each of the two stocks. All mackerel from the Atlantic stock appeared, as expected, inside the model for this stock (Fig. 13-4). One of the mackerel caught in the North Sea was identiﬁed as belonging to the Atlantic stock. Another of the North Sea mackerel appeared outside both models, but the remaining mackerel caught in the North Sea fell inside the model for this stock.

E. STRIPED BASS

FROM

EAST AMERICAN RIVERS

Striped bass from spawning stocks in the Hudson River, the Roanoke River, and the Chesapeake Bay area had signiﬁcantly different fatty acid proﬁles of 13

256

Distance from North Sea model

O. Grahl-Nielsen

200

100

100

200

Distance from Atlantic model FIGURE 13-4. RSD plot of mackerel caught on spawning grounds off Ireland (squares) and from the North sea stock (circles), based on fatty acids in heart tissue. The outer limits of the Atlantic model (i.e., RSDmax = 116) and of the North Sea model (i.e., RSDmax = 108) are indicated in the plot.

selected fatty acids from heart tissue, although the spread among the individuals within stocks was considerable. In a training set of 49 striped bass, 45 initially were correctly classiﬁed (Fig. 13-5). Three striped bass fell outside their own models, but were closer to their own model than to the others. One striped bass from the Potomac River fell inside both its own model and the Roanoke model (Fig. 13-5) All striped bass from the Hudson River fell outside the Roanoke and Chesapeake models (Fig. 13-5), but inside their own model (not shown), and none of the Roanoke or Chesapeake striped bass fell inside the Hudson model (not shown). In a test set of 19 samples, 84% were assigned to the correct stock (Grahl-Nielsen and Mjaavatten, 1992).

F. FARMED SALMON ESCAPEES Wild salmon and farmed salmon that had escaped from net pens on the coast of western Norway were caught in the Os River. They were identiﬁed as wild and

257

Fatty Acid Proﬁles as Natural Marks for Stock Identiﬁcation

H

Distance from Chesapeake model

R R

200 H

R R

H

R

HH

H R

100

H

A

R H

R R

H

R

P

P P

AA

U

U

P

AU P A P

A A A

P

U

U

A U

P P U

U

U

U A

P

100

200

300

400

Distance from Roanoke model FIGURE 13-5. RSD plot of 49 striped bass from the Roanoke River, R, Rappahannock River, A, Potomac River, P, Upper Chesapeake Bay, U, and the Hudson River, H, based on fatty acids in heart tissue. The outer limits of the Roanoke model (i.e., RSDmax = 110) and of the Chesapeake model (including striped bass from the Rappahannock River, Potomac River, and Upper Chesapeake Bay) (i.e., RSDmax = 87) are indicated in the plot.

farmed, respectively, by morphology and growth patterns in shells and otoliths (Økland et al., 1991). Samples of tissue were removed from the same area of the brain on all salmon and subjected to methanolysis and gas chromatographic analysis, resulting in the relative amounts of fatty acids shown in Table 13-3 (Roseth, 1994). Multivariate analysis of the data indicated a clear-cut difference between the wild and farmed salmon (Fig. 13-6), even if there was a large spread among the individual salmon, particularly among the escapees. This variation was probably enhanced by relatively long storage in the freezer from the salmon that were caught until analysis. Another cause for variation of fatty acid composition between salmon within each group is that fatty acid composition differs in different parts of the brain, and it is difﬁcult to obtain well-deﬁned samples from the same lobe from all salmon.

G. TWO FAROE ISLAND STOCKS

OF

COD

Cod off the Faroe Islands appear in two different stocks, the Faroe Plateau stock and the Faroe Bank stock. The Faroe Bank cod spawn on an oceanic bank about

258

O. Grahl-Nielsen TABLE 13-3. Relative Amounts, as Percentage of Sum ± SD, of Fatty Acids in Brain Tissue of Wild Salmon and Farmed Salmon Escapees from Os River, Western Norwaya

Fatty acid 14.0 16:0 18.0 20:0 22:0 16:1n9 16:1 n7 18:1n11 18:1n9 18:1n7 20:1n9 22:1n11 22:1n9 24:1n9 18:2n6 18:3n3 18:4n3 20:2n6 20:4n6 20:5 n3 22:5n3 22:6n3

Escaped salmon (N = 16)

Wild salmon (N = 9) 0.4 18 7.3 0.1 0.3 1.8 2.0 0.9 22 3.3 1.9 0.3 0.5 9 0.11 0.1 0.11 0.01 1.3 5.5 2.7 22

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.1 2 0.7 0.01 0.1 0.1 0.2 0.2 2 0.2 0.5 0.2 0.1 1 0.02 0.1 0.03 0.01 0.1 0.5 0.3 3

<

<

< >

0.7 17 7 0.1 0.3 1.8 2.2 1.9 22 3.3 2.2 0.7 0.7 10 0.4 0.4 0.1 0.12 1.1 5.8 2.5 20

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.2 3 1 0.02 0.1 0.2 0.3 0.2 3 0.2 0.8 0.5 0.2 3 0.3 0.5 0.1 0.04 0.1 0.6 0.4 4

a

The cases in which the amounts of fatty acids in brain tissue were signiﬁcantly different (p < 0.01) between wild and escaped salmon are marked with inequality signs.

150 km southwest of Thorshavn, Faroe Islands. Within the 200 m contour, the bank is 40 ¥ 90 km, and the minimum depth is less than 100 m. The Bank is separated from the Faroe Plateau, the shelf around the islands, by the more than 800-m-deep Faroe Bank Channel. Because of anticyclonic currents around both the Bank and the Plateau, eggs and fry from the two stocks were kept apart. A portion of the progenies of the two stocks were reared under identical circumstances, with the same feeding regime, ever since hatching in 1994. Therefore, all the combined biotic inﬂuences and abiotic factors had been identical for the individuals in both stocks until 15 specimens from each stock were slaughtered 3 years and 8 months after hatching. The fatty acid proﬁle of their hearts was chemometrically determined. The proﬁle turned out to be rather similar in cod from the two stocks, but eight fatty acids and two components, related to cho-

259

Fatty Acid Proﬁles as Natural Marks for Stock Identiﬁcation

PC 2 20% 20:1n9 22:1n9 24:1n9

18:1n9 16:1n9 16:1n7 20:5n3 18:1n7 20:4n6 18:0 22:5n3 16:0

20:2n6 18:2n6

14:0

22:6n3

PC 1 67% FIGURE 13-6. PC plot of wild salmon (squares) and escaped salmon (circles), based on fatty acids in brain tissue.

lesterol, were signiﬁcantly (p < 0.01) different in the two stocks (Table 13-4). Principal component analysis showed complete distinction between the two stocks (Fig. 13-7) (Joensen et al., 2000). The results suggest that the fatty acid proﬁle of heart tissue is genetically controlled.

H. REDFISH STOCKS

IN THE

NORTH ATLANTIC

By way of fatty acid composition of heart tissue, redﬁsh of the species Sebastes mentella in the North Atlantic were found to belong to four different stocks (Joensen and Grahl-Nielsen, 2004). This ﬁnding was based on analysis of 220 redﬁsh caught during the summer and fall of 1999 in 11 different areas (Fig. 138). Altogether, 36 fatty acids, 2 cholesterol derivatives, the same derivatives that were important in distinguishing among two stocks of cod from Faroe Island waters, and 7 unidentiﬁed components were used in the multivariate statistics. Due to large individual variation in fatty acid composition, initial mapping of the redﬁsh was carried out on the average composition of ﬁsh from each area (Fig. 13-9). Redﬁsh from the waters around the Faroe Islands apparently belong to two different stocks. One was detected on the Faroe Plateau around the islands, that

260

O. Grahl-Nielsen TABLE 13-4. Relative Amounts, as Percentage of Sum ± SD, of Fatty Acids in Heart Tissue of Cod from the Faroe Bank and the Faroe Plateau in the North Atlantica

Fatty acid

Faroe Bank stock (N = 15)

14.0 15:0 16.0 16:1 n7 16:1n5 17:1 18:0 18:1 n9 18:1 n7 18:1 n5 18:2 n6 18:3 n3 18:4n3 20:1 n9 20:2 n6 20:4 n6 20:3n3 20:4n3 20:5 n3 22:1n11 22:1 n9 22:1n7 21:5n3 22:5n6 22:5n3 22:6 n3 24:1n9 C1 C2

0.9 0.19 14.5 0.9 0.6 0.35 6.0 12.7 3.6 0.29 2.3 0.41 0.23 2.0 0.25 2.9 0.09 0.37 9.7 0.9 0.21 0.16 0.11 0.38 1.01 31 2.4 1.17 3.5

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.1 0.02 0.7 0.1 0.2 0.05 0.3 0.4 0.1 0.02 0.2 0.04 0.02 0.2 0.02 0.2 0.01 0.02 0.8 0.2 0.02 0.03 0.01 0.03 0.04 2 0.1 0.06 0.2

Faroe Plateau stock (N = 15) > >

>

< <

>

<

> > >

0.81 0.17 13.9 0.87 0.55 0.32 6.0 12.2 3.6 0.30 2.56 0.48 0.24 1.8 0.25 3.0 0.08 0.39 9.9 0.8 0.18 0.14 0.12 0.42 0.99 32 2.2 1.08 3.3

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.06 0.01 0.4 0.06 0.08 0.03 0.3 0.4 0.2 0.02 0.09 0.02 0.03 0.1 0.02 0.2 0.01 0.02 0.7 0.1 0.01 0.01 0.01 0.03 0.04 1 0.2 0.05 0.2

a

The cases in which the amounts of fatty acids in brain tissue were signiﬁcantly different (p < 0.01) between the two stocks are marked with inequality signs.

is, areas F1, F4, and F5 in Figure 13-8, and one in the Bill Bailey Bank/Wyville Thomson Ridge area to the southwest of the islands, that is, areas F2 and F3 in Figure 13-8. These latter areas are separated from the Faroe Plateau by the more than 800 m deep Faroe Bank Channel. Redﬁsh from the Norwegian Sea belong to the same stock as the Faroe Plateau redﬁsh, while redﬁsh from the waters southeast and southwest of Iceland, IC1 and IR1, are related to the other Faroe Island stock (Fig. 13-9). Redﬁsh from the northwestern part of the Icelandic

15:0

PC 2 15%

14:0

18:2n6 18:3n3

c1 c2

22:5n6 24:1n9 22:1n9

18:1n9

PC 1 52% FIGURE 13-7. PC plot of cod from Faroe Plateau (squares) and from Faroe Bank (circles), based on the eight most discriminating fatty acids and two cholesterol derivatives, c1 and c2, in heart tissue. Two parallel samples of 15 cod from each stock are shown. The borderline between the two stocks is shown.

70

N1 IC2

IC1 IR1

F2 F3

60

N2

F5

F1 F4

IR2

50 -50

-40

-30

-20

-10

0

10

20

FIGURE 13-8. Eleven locations for catches of redﬁsh during the summer and fall of 1999.

261

262

O. Grahl-Nielsen

FIGURE 13-9. Plot of the average redﬁsh from 11 locations shown in Figure 13-7, based on fatty acids in heart tissue, in three dimensions, that is, PC1, accounting for 48% of the total variance, from left to right, PC2, accounting for 14% of the total variance, from front to back, and PC3, accounting for 10% of the total variance, vertical.

plateau, IC2, on the other hand, form a distinct stock. Sebastes mentella, from depths between 650 and 800 m in the Irminger Sea, IR2, also form a distinct stock. The same sample of redﬁsh was independently subjected to electrophoretic analysis of the enzyme MEP-1 in another laboratory (Institute of Marine and Fisheries Biology, University of Bergen). This analysis revealed the same stock structure of the redﬁsh from the waters around the Faroe Islands as was found by the fatty acid proﬁles (T. Johansen, personal communication). The electrophoretic method also conﬁrmed the ﬁnding by the fatty acid method that redﬁsh from the Norwegian coast belonged to the same stock as redﬁsh from the Faroe Island Plateau, and that redﬁsh from the areas to southeast and southwest of Iceland belonged to the same stock as redﬁsh from the Bill Bailey Bank/Wyville Thomson Ridge area to the southwest of the Faroe Islands. The electrophoretic method did not, however, distinguish redﬁsh from the northwestern part of the Icelandic plateau or redﬁsh from depths between 650 and 800 m in the Irminger Sea as separate stocks, as was done by the fatty acid method. Rather, they were found to belong to the stock extending from southwest of the Faroe Islands to areas to the south of Iceland. This suggests that

263

Fatty Acid Proﬁles as Natural Marks for Stock Identiﬁcation

PC 2 8%

E B G G FF F B FG EA B A E A AG C D D B D C C D C E

PC 1 88% FIGURE 13-10. PC plot of two parallel samples from the jawbone from each of six harp seals from the western population (squares) and from each of four harp seals from the eastern population (circles), based on four fatty acids: 14 : 0, 18 : 1n7, 22 : 5n3, and 22 : 6n3, with highest discrimination power between the two populations. Four parallel samples from each of seven test seals, A to G, are projected into the plot, but without any inﬂuence on PCs.

the fatty acid method has a higher resolving power than the electrophoretic method.

I. IDENTIFICATION

OF

HARP SEALS

ON

FORAGING MIGRATIONS

In the winter of 1986–1987, a large invasion of harp seals took place along the Norwegian coast down to the southernmost part, that is, the Skagerrak coast. Approximately 300,000 animals were estimated to have approached the coast, and 60,000 of these were trapped and drowned in gill nets. It was not known if the seals originated from the western population, with breeding areas in the Greenland Sea by Jan Mayen, or from the eastern population, breeding in the White Sea. Reference seals from the two populations were obtained from their respective breeding areas. To be identiﬁed were six seals which had drowned in gill nets on the Norwegian coast at Sunnmøre, approximately 63°30¢N. In addition, one seal from the Barents Sea was considered as a test specimen.

264

O. Grahl-Nielsen

TABLE 13-5. Relative Amounts, as Percentage of Sum ± SD, of Fatty Acids in Samples of Jawbone of Harp Sealsa Reference seals West (N = 6) 14:0 14:1n5 16.0 16:1n9 16:1n7 18:0 18:1n9 18:1n7 18:1n5 18:2n6 18:3n3 18:4n3 20:1n9 20:4n6 20:5n3 22:5n3 22:6n3

4.5 1.2 9.1 0.8 14.4 1.2 29 6.6 0.6 2.3 0.8 1.8 13 0.7 4.8 2.5 6.2

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.3 0.1 0.8 0.1 1.5 0.3 1 0.6 0.1 0.2 0.1 0.3 2 0.2 0.9 0.3 0.9

> > <

< <

> > < > >

East (N = 4)

A

B

C

D

E

F

G

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

4.4 0.6 14.3 0.9 11.1 3.1 30.4 8.2 0.7 2.2 0.6 0.9 14.6 1.0 2.6 1.4 3.0

4.5 0.5 18.2 1.1 13.6 3.6 29.8 9.9 0.7 1.8 0.5 0.7 9.5 1.0 1.9 0.9 1.8

5.7 0.5 20.5 1.0 6.6 6.3 28.5 6.9 0.6 1.9 0.4 0.4 16.1 1.3 1.3 0.6 1.7

4.9 1.6 12.4 1.0 15.4 4.1 32.5 7.8 0.8 1.7 0.6 0.7 12.7 1.0 1.2 0.6 1.0

5.8 0.9 14.6 1.0 12.2 2.8 28.3 9.1 0.7 1.9 0.8 1.3 11.4 0.9 2.8 1.8 3.9

7.1 0.7 18.6 1.2 11.4 3.2 28.6 9.8 0.9 1.8 0.6 0.9 10.0 0.8 1.7 0.8 1.9

4.9 1.3 13.1 1.0 16.1 2.5 28.8 9.3 0.9 1.7 0.5 1.2 11.8 0.8 2.7 1.1 2.3

3.6 0.9 13 0.7 15 5 29 8 0.6 2.1 0.5 1.2 9 3 5 1.4 2.3

0.8 0.3 3 0.2 3 3 1 1 0.1 0.3 0.2 0.4 4 2 1 0.2 0.6

Test seals

a Test seal A was from the Barents Sea and the other six were from the Norwegian coast, the average of four replicate analyses in each case. The cases in which the amounts of fatty acids in the jawbone were signiﬁcantly (p < 0.05) different between the western and eastern seals are marked with inequality signs.

A small piece of bone was removed from the rear part of the lower jaw. Flesh and blood were carefully removed. The bone sample was rinsed in water, dried, and crushed in a mortar. Parallel samples of 50 mg were subjected to chemometric analysis. The fatty acid proﬁles were different in seals from the two populations (Table 13-5). Principal component analysis of the reference seals from the two populations was carried out on the basis of the four most discriminating fatty acids. The seven test seals were matched against the reference seals, and all were found to belong to the eastern population (Fig. 13-9) (Grahl-Nielsen et al., 1993).

IV. DISCUSSION The cases presented are all from the author’s laboratory. Other investigations have demonstrated differences in tissue fatty acids on species and stock level. For example, Armstrong et al. (1994) distinguished ﬁve species of ﬁsh from temper-

Fatty Acid Proﬁles as Natural Marks for Stock Identiﬁcation

265

ate Australian waters by PCA of fatty acid composition in the ﬁllet. Within some of the species, seasonal and geographic differences in the fatty acid proﬁle were detected. Castell et al. (1995), using multivariate discriminant analysis, distinguished lobster (Homarus americanus) eggs from three Nova Scotia wild stocks. The fatty acid proﬁles of muscle lipids differed between cultured and wild sturgeon, Acipenser oxyrinchus desotoi (Chen et al., 1995). Silversand et al. (1996) used the fatty acids 18 : 2n6 and 20 : 1n9 in lipids from ovulated eggs to distinguish between wild and cultured turbot, Scophthalmus maximus. Pickova et al. (1997) distinguished Skagerrak and Baltic cod in terms of egg lipid fatty acid composition. Seaborn et al. (2000) used proﬁles of muscle fatty acids to differentiate wild striped bass and its hybrids from cultured hybrid striped bass. It turns out that the applications of fatty acid proﬁles for identiﬁcation purposes fall into two categories, based on different assumptions: One is that diet inﬂuences the composition of fatty acids in triacylglycerides in storage lipids in, for example, muscle tissue. In this case, differences in diet is the key identiﬁcation factor, which may be used on a short timescale. The other assumption is that the composition of fatty acids in membrane phospholipids is genetically controlled and stable over time. The phospholipid fatty acids may therefore be used as a natural mark over a longer timescale. A wide range of investigations have shown that the composition of fatty acids in ﬁsh tissue is inﬂuenced by the composition of fatty acids in the diet. Still, many of these investigations have been conducted with diets of anomalous compositions, often based on vegetable oils (Cowley et al., 1983; Bell et al., 1985; Leray and Pelletier, 1985; Lie et al., 1986; Anderson and Arthington, 1989). In addition, many of the investigations were carried out on juvenile stages of ﬁsh which are more easily affected than ﬁsh in their mature stages (Navarro et al., 1995; Muje et al., 1989). However, even in cases where natural diets have been used on mature ﬁsh, an inﬂuence on the fatty acid composition has been demonstrated. Kirsch et al. (1998) fed cod squid, Illex illecebrosus, for 6 weeks followed by mackerel for 8 weeks. The fatty acid composition of the total body lipids ﬁrst changed in the direction of the pattern found in squid, and thereafter in the direction of the pattern in mackerel. The change in pattern in the body lipids occurred during the ﬁrst three to ﬁve weeks after change in diet. However, despite changes in the fatty acid pattern, the pattern in the cod could be readily distinguished from that in the diets. The richer a tissue is in triacylglycerides, the closer is the resemblance of its fatty acid composition with that of the diet (Viga and Grahl-Nielsen, 1990). Diet-based, short-term identiﬁcation should therefore be based on intestinal fat or brown or white muscle tissue. Since these tissues make up the bulk of the bodies, whole, ground-up ﬁsh may also be suitable for this type of identiﬁcation.

266

O. Grahl-Nielsen

The majority of published investigations on identiﬁcations are based on dietinduced composition of fatty acids in tissues rich in triacylglycerides, that is, muscle tissue in cultured and wild sturgeon (Chen et al., 1995); total body lipids in Canadian ﬁsh and invertebrates (Budge et al., 2002); egg lipids in lobsters from Nova Scotia (Castell et al., 1995); muscle tissue in various Australian ﬁshes (Armstrong et al., 1994); and muscle lipids in cultured and wild striped bass (Seaborn et al., 2000). The drawback with diet-induced identiﬁcation is that observed differences in tissue fatty acid proﬁles from different specimens may not necessarily be due to differences in stock. This is because many ﬁsh species are opportunistic feeders. Selection of diet may therefore not be fenotypic. Still, the method may be advantageously used in cases where the purpose is to distinguish between groups of ﬁsh with clear-cut differences in diet, for example, in distinguishing between farmed and wild specimens. Contrary to the labile proﬁle of fatty acids in triacylglycerides, the proﬁle of fatty acids in phospholipids may serve as a natural mark for identiﬁcation of stocks. The condition is that the fatty acid proﬁle of these lipids is typical for a stock and stable over time. Our investigation of the Faroe stocks of cod (Joensen et al., 2000) showed that the between-stock difference in fatty acid composition in heart tissue, in which the lipids were made up of between 80% and 90% phospholipids, was caused by inborn factors. The stability of the fatty acid proﬁle then needs to be questioned. Compared with the composition of the fatty acids in triacylglycerides, the composition of fatty acids in phospholipids is generally expected to be less sensitive to diet. Pickova et al. (1997) demonstrated that fatty acid composition of total phospholipids in cod eggs was population-speciﬁc and diet-independent. They used egg phospholipid fatty acids to distinguish between Skagerrak and Baltic cod. Bandarra et al. (1997) have shown that the fatty acid composition of sardine phospholipids is hardly affected by seasonal change in dietary planktonic lipids. In rainbow trout, Oncorhynchus mykiss, dietary fatty acids are selectively incorporated into muscle phospholipids to obtain narrowly deﬁned physiological levels (Greene and Selivonchick, 1990). In fact, this was also the case for triacylglycerols. Owen et al. (1972) detected relatively little change in the fatty acids of phospholipids of liver and extrahepatic tissue in plaice, Pleuronectes platessa, subjected to different dietary regimes. However, even if dietary inﬂuence on fatty acid composition of phospholipids, and thus phospholipid-rich tissue, such as heart, might be negligible, other environmental factors may cause changes. We need to sample ﬁsh from different populations at intervals, preferably during an entire year, to see how stable the fatty acid proﬁle is under shifting dietary regimens and during changes in other environmental factors, such as temperature. Changing physiological conditions of the ﬁsh also need to be considered.

Fatty Acid Proﬁles as Natural Marks for Stock Identiﬁcation

267

Stock identiﬁcation methods based on natural marks, such as morphological and genetic analyses as presented in this volume, are carried out as frequency distributions, and a large number of specimens are needed. The examples discussed in the present chapter demonstrate the superiority of the fatty acid proﬁle method in that it is powerful enough for identiﬁcation of individual specimens. It may be applied in research and management of stocks of various ﬁsh species, in surveillance of commercial catches, and in detecting ﬁsh farm escapees.

REFERENCES Ackman, R. G. 1980. Fish lipids. In J. J. Connell (ed.), Advances in Fish Science and Technology. Fishing New Books, Farnham, UK, pp. 86–103. Anderson, A. J. and Arthington, A. H. 1989. Effect of dietary lipid on the fatty acid composition of silver perch (Leiopotherapon bidyanus) lipids. Comparative Biochemistry and Physiology 93B: 715–720. Armstrong, S. G., Wyllie, S. G., and Leach, D. N. 1994. Effects of season and location of catch on fatty acid composition of some australian ﬁsh species. Food Chemistry 51: 295–305. Bandarra, N. M., Batista, I., Nunes, M. L., Empis, J. M., and Christie, W. W. 1997. Seasonal changes in lipid composition of sardine (Sardina pilchardus). Journal of Food Science 62: 40–42. Bell, M. V., Henderson, R. J., and Sargent, J. R. 1985. Changes in the fatty acid composition of phospholipids from turbot (Scophthalmus maximus) in relation to dietary polyunsaturated fatty acid deﬁciencies. Comparative Biochemistry and Physiology 81B: 193–198. Budge, S. M., Iverson, S. J., Bowen, W. D., and Ackman, R. G. 2002. Among- and within-species variability in fatty acid signatures of marine ﬁsh and invertebrates on the Scotian Shelf, Georges Bank, and southern Gulf of St. Lawrence. Canadian Journal of Fisheries and Aquatic Sciences 59: 886–898. Castell, J. D., Boston, L. D., Miller, R. J., and Kenchington, T. 1995. The potential identiﬁcation of the geographic origin of lobster eggs from various wild stocks based on fatty acid composition. Canadian Journal of Fisheries and Aquatic Sciences 52: 1135–1140. Chen, I.-C., Chapman, F. A., Wei, C.-I., Portier, K. M., and O’Keefe, S. F. 1995. Differentiation of cultured and wild sturgeon (Acipenser oxyrinchus desotoi) based on fatty acid composition. Journal of Food Science 60: 631–635. Cowley, C. B., Wee, K. L., and Tacon, A. G. J., 1983. Effect of fatty acid intake on growth and fatty acid composition of liver and muscle of snakehead. Bulletin of the Japanese Society of Scientiﬁc Fisheries 49: 1573–1577. Esbensen, K., Schønkopf, S., and Midtgaard, T. 1994. Multivariate analysis in practice. In Camo. Computer-Aided Modelling AS. Trondheim, Norway, pp. 157–212. Grahl-Nielsen, O. and Barnung, T. N. 1985. Variations in the fatty acid proﬁle of marine animals caused by environmental and developmental changes. Mar. Environ. Res. 17: 218–221. Grahl-Nielsen, O., Mjaavatten, O., and Tvedt, E. 1993. Distinguishing between different populations of harp seal (Phoca groenlandica) by chemometry of the fatty acid proﬁles in jaw bone. Canadian Journal of Fisheries and Aquatic Sciences 50: 1400–1404. Grahl-Nielsen, O. and Mjaavatten, O. 1992. Discrimination of striped bass stocks: a new method based on chemometry of fatty acid proﬁle in heart tissue. Transactions of the American Fisheries Society 121: 307–314. Grahl-Nielsen, O. and Ulvund, K. A. 1990. Distinguishing populations of herring by chemometry of fatty acids. American Fisheries Society Symposium, 7: 566–571.

268

O. Grahl-Nielsen

Greene, D. H. S. 1990. Lipid metabolism in ﬁsh. In M. S. Stansby (ed.), Fish Oils in Nutrition. Van Nostrand Reinhold, New York, pp 226–246. Greene, D. H. S. and Selivonchick, D. P. 1990. Effects of dietary vegetable, animal and marine lipids on muscle lipid and hematology of rainbow trout (Oncorhynchus mykiss). Aquaculture 89: 165–182. Hornnes, H. K. 1993. Fettsyrer i sild. M.Sc. Thesis, Department of Chemistry, University of Bergen. Joensen, H., Steingrund, P., Fjallstein, I., and Grahl-Nielsen, O. 2000. Discrimination between two reared stocks of cod (Gadus morhua) from the Faroe Islands by chemometry of the fatty acid composition in the heart tissue. Marine Biology 136: 573–580. Joensen, H. and Grahl-Nielsen, O. 2004. Stock structure of Sebastes mentalla in the North Atlantic revealed by chemometry of the fatty acid proﬁle in hear tissue. ICES Journal of Marine Science 61: 113–126. Kirsch, P. E., Iverson, S. J., Bowen, W. D., Kerr, S. R., and Ackman, R. G. 1998. Dietary effects on the fatty acid signature of whole atlantic cod (Gadus morhua). Canadian Journal of Fisheries and Aquatic Sciences 55: 1378–1386. Kvalheim, O. M. and Kvarstang, T. V. 1987. A general-purpose program for multivariate data analysis. Chemometr. Intell. Lab. Syst. 2: 235–237. Leray, C. and Pelletier, X. 1985. Fatty acid composition of trout phospholipids: effect of (n-3) essential fatty acid deﬁciency. Aquaculture 50: 51–59. Lie, Ø., Lied, E., and Lambertsen, G. 1986. Liver retention of fat and of fatty acids in cod (Gadus morhua) fed different oils. Aquaculture 59: 187–196. Meier, S. 1997. Variasjon i fettsyresammensetningen i sild, Clupea harengus L. M.Sc. Thesis, Department of Chemistry, University of Bergen. Morris, R. J. and Culkin, F. 1989. Fish. In R. G. Ackman (ed.), Marine Biogenic Lipids, Fats and Oils, Vol. 2. CRC Press, Boca Raton, FL, pp. 145–178. Muje, P., Ågren, J. J., Lindqvist, O. V., and Hänninen, O., 1989. Fatty acid composition of vendace (Coregonus albula L.) muscle and its plankton feed. Comparative Biochemistry and Physiology 92B: 75–79. Navarrro, J. C., McEvoy, L. A., Amat, F., and Sargent, J. R. 1995. Effects of diet on fatty acid composition of body zones in larvae of the sea bass Dicentrarchus labrax: a chemometric study. Mar. Biol. 124: 177–183. Owen, J. M., Adron, J. W., Sargent, J. R., and Cowey, C. B. 1972. Studies on the nutrition of marine ﬂatﬁsh. The effect of dietary fatty acids on the tissue fatty acids of the plaice Pleuronectes platessa. Marine Biology 13: 160–166. Pickova, J., Dutta, P. C., Larsson, P.-O., and Kiessling, A. 1997. Early embryonic cleavage pattern, hatching success, and egg-lipid fatty acid composition: comparison between two cod (Gadus morhua) stocks. Canadian Journal of Fisheries and Aquatic Sciences 54: 2410– 2416. Roseth, I. E. 1994. Fettsyrer i laks. M.Sc. Thesis, Department of Chemistry, University of Bergen. Sargent, J., Henderson, R. J., and Tocher, D. R. 1989. The lipids. In J. E. Halver (ed.), Fish Nutrition. Academic Press, San Diego, pp. 153–218. Seaborn, G. T., Jahncke, M. L., and Smith, T. I. J. 2000. Differentiation between cultured hybrid striped bass and wild striped bass and hybrid bass using fatty acid proﬁles. North American Journal of Fisheries Management 20: 618–626. Silversand, C., Norberg, B., and Haux, C. 1996. Fatty acid composition of ovulated eggs from wild and cultured turbot (Scophthalmus maximus) in relation to yolk and oil globule lipids. Marine Biology 125: 269–278. Viga, A. and Grahl-Nielsen, O. 1990. Genotypic and phenotypic fatty acid composition in the tissues of salmon, Salmo salar. Comparative Biochemistry and Physiology 96B: 721–727.

Fatty Acid Proﬁles as Natural Marks for Stock Identiﬁcation

269

Vogt, N. B., Moksness, E., Sporstøl, S. P., Knutsen, H., Nordenson, S., and Kolset, K. 1986. SIMCA principal component analysis of fatty acid patterns in Day-1 and Day-8 cod (Gadus morhua) and haddock (Melanogrammus aegleﬁnus) eggs. Marine Biology 92: 173–182. Wold, S. and Sjøstrøm, M. 1977. SIMCA: a method for analyzing chemical data in terms of similarity and analogy. In B. R. Kowalski (ed.), Chemometrics: Theory and Applications. Symp. Ser. Am. Chem. Soc. 52: 243–282. Wold, S. 1978. Cross validatory estimation of the number of components in factor and principal models. Technometrics 20: 397–406. Økland, F., Lund, R. A., and Hansen, L. P. 1991. Rømt oppdrettslaks i vassdrag 1989 og 1990; tidspunkt for oppvandring i elver, og betydningen av oppdrettsnæringens omfang. NINA Oppdragsmelding 082: 1–16.

This Page Intentionally Left Blank

PART

Natural Marks— Genetic Analyses

V

This Page Intentionally Left Blank

CHAPTER

14

Chromosome Morphology RUTH B. PHILLIPS Washington State University, Vancouver, Washington, USA

I. Intraspeciﬁc Variation in Fish and Shellﬁsh Chromosome Number and Morphology A. Variation in Chromosome Number B. Variation in Chromosome Structure C. Variation in Chromosome Banding Patterns II. Standard Methods for Detecting Chromosome Variation A. Tissue Sampling and Treatment to Obtain Dividing Cells B. Colchicine Treatment, Hypotonic Treatment, and Fixation C. Slide Preparation D. Staining E. Chromosome Banding Techniques F. Photography and Analysis III. Molecular Cytogenetic Methods A. Fluorescence In Situ Hybridization (FISH) B. Application of FISH to Fish and Shellﬁsh Chromosomes IV. Discussion and Conclusions References

I. INTRASPECIFIC VARIATION IN FISH AND SHELLFISH CHROMOSOME NUMBER AND MORPHOLOGY Variation in chromosome number and morphology has the potential to be used to identify ﬁsh and shellﬁsh stocks, although chromosome markers have been applied to stock identiﬁcation in only a few cases. Their use has been limited because the karyotypes of relatively few species have been determined and the Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

273

274

Ruth B. Phillips

methods require examination of dividing cells. Compilations of chromosome data on ﬁsh species can be found in Gold et al. (1980), Ojima (1980), Sola et al. (1981), and Klinkhardt et al. (1995). Reviews of speciﬁc groups include Galetti et al. (2000) (marine ﬁshes), Rab and Collares-Pereira (1995) (European cyprinid ﬁshes), and Phillips and Rab (2001) (salmonid ﬁshes). References on shellﬁsh include Nakamura (1985) and Thiriot-Quievreux and Ayraud (1982).

A. VARIATION

IN

CHROMOSOME NUMBER

1. Polyploidy Chromosome number variation occurs in natural populations in the form of spontaneous polyploids, especially triploids. Polyploid species such as sturgeons with even numbers of chromosome sets (4n, 8n, 16n) (Birstein et al., 1993) are fertile, but odd-numbered ones such as triploids (3n) usually cannot form viable gametes. However, unisexual triploid ﬁshes that reproduce parthenogenetically occur in several groups of live-bearing ﬁsh of the family Poecillidae from the southern United States and Mexico. Among cyprinids, there is a diploid–triploid complex in the genus Rivulus in the southeastern United States and in the genus Rutilus on the Iberian peninsula (Collares-Pereira, 1985). The triploids usually have a higher temperature tolerance and survive better in harsh environments. Naturally occurring triploids are also found in shellﬁsh. For example, both triploids and diploids are found in different species of Korean Corbicula (Park et al., 2001). Triploidy has also been artiﬁcially induced in many ﬁsh and shellﬁsh species to obtain sterile animals, and some of these have been stocked. Triploid shellﬁsh are useful for aquaculture because of their sterility, superior growth, and improved meat quality (Guo and Standish, 1994; Yang et al., 2001). Tetraploids are also valuable for 100% producing triploids through mating with diploids (Guo et al., 1996; Guo and Allen, 1997). In ﬁshes, triploids usually are sterile, but effects on growth have been variable (reviewed in Tillmann, 2001; Felip et al., 2001; Cotter et al., 2002; Lilysestrom et al., 1999; and Sheenhan et al., 1999). Polyploids have extra haploid sets of chromosomes and can be detected by the increase in DNA content per cell with ﬂow cytometry as well as cytogenetically (reviewed in Thorgaard and Allen, 1987). If a species has only one chromosome pair with nucleolar organizer regions (NORs), then polyploid cells can be identiﬁed by the number of nucleolar organizers found per cell (Phillips et al., 1986) (Fig. 14-1). 2. Robertsonian Translocations The most common type of chromosome number variation that has been identiﬁed in ﬁshes results from Robertsonian translocations. In this type of chromo-

Chromosome Morphology

275

A FIGURE 14-1. Silver-stained cells from rainbow trout (Oncorhynchus mykiss) embryos of three different ploidy levels: (A) haploid cells, (B) diploid cells, and (C) triploid cells. From Phillips et al. (1986).

some rearrangement, two uniarmed chromosomes fuse to form a biarmed chromosome, or a biarmed chromosome undergoes ﬁssion to form two uniarmed chromosomes. There is no change in the amount of genetic material, but there is a change in chromosome number. Robertsonian translocations have been reported for many ﬁsh species (reviewed in Gold, 1979; Sola et al., 1981; Thorgaard and Allen, 1987; Hartley, 1987; Phillips and Ihssen, 1990; and Phillips and Rab, 2001), and intraspeciﬁc variation in chromosome number is especially common in salmonid ﬁshes. In contrast, there are no known examples of intraspeciﬁc variation in chromosome number in shellﬁsh (Guo, personal communication, Rutgers University).

B

C FIGURE 14-1. Continued

276

277

Chromosome Morphology

Different populations of rainbow trout vary in chromosome number between 2n = 58 and 2n = 64 (Thorgaard, 1976; 1983; Hartley and Horne, 1982) (Fig. 14-2). Hatchery stocks also differ in chromosome numbers, so in some cases stocked vs. wild populations can be distinguished on this basis. There are two evolutionarily distinct chromosome lineages in rainbow trout native to western Washington. Populations between and including the Elwha River have 2n = 60 chromosomes, while populations on the central Washington coast have 2n = 58 chromosomes (Ostberg and Thorgaard, 1999). The difference between the two groups is a Robertsonian translocation and a pericentric inversion. Another example of a Robertsonian polymorphism occurs in pink salmon on the northwest coast of North America and the eastern coast of Russia (Gorshkov and Gorshkova, 1981; Phillips and Kapuscinski, 1987). The even year-class has 2n = 52, but in many locations the most common chromosome number among odd-year ﬁsh is 2n = 53, with some 2n = 52 and 2n = 54 being found (Phillips and Kapuscinski, 1988). In southeastern Alaska and northern British Columbia, a second chromosome rearrangement (an inversion) has occurred, with ﬁsh heterozygous for this inversion (2n = 53) being the most numerous. 3. Tandem Translocations Tandem fusions can cause a change in chromosome number. In this type of translocation, there is a “head to tail” fusion of two uniarmed chromosomes and deactivation of one of the centromeres to produce a large uniarmed chromosome. Heterozygotes for such fusions usually produce inviable gametes. Tandem translocations have been important in the production of the highly derived karyotype found in Atlantic salmon (Hartley and Horne, 1984a,b; Ueda and Kobayashi, 1990). The North American (and European) Atlantic salmon populations have different chromosome numbers (2n = 54 and 2n = 58, respectively), and inspection of the karyotypes reveals that they differ by several independent tandem and fusion translocations (Roberts, 1970) (Fig. 14-3). European Atlantic salmon have been introduced into North America and some North American hatchery stocks are heterozygous for these rearrangements (Phillips, unpublished).

B. VARIATION

IN

CHROMOSOME STRUCTURE

1. Inversions Inversions have also been documented in a number of ﬁshes. including the goodeid Ilyodon fucidens (Turner et al., 1985), Atlantic salmon (Roberts, 1970; R. B. Phillips and S. E. Hartley, unpublished), and chum salmon, Oncorhynchus keta (Kulikova, 1971). In this type of rearrangement, two breaks occur in a

A

B FIGURE 14-2. Karyotypes from rainbow trout (Oncorhynchus mykiss) with different chromosome numbers. (A) Donaldson strain with 2n = 60 and (B) Goldendale strain with 2n = 61. The arrow shows the chromosome pair consisting of one metacentric and two acrocentrics produced by a chromosome ﬁssion. Chromosomes were stained with DAPI and the images reversed so that the DAPI bright bands at the centromeres appear dark. From Phillips (unpublished).

278

A

B FIGURE 14-3. Karyotypes from (A) European (2n = 58) and (B) North American (2n = 54) Atlantic salmon. Chromosomes were stained with DAPI and the images reversed so that the DAPI bright bands appear dark. An interstitial DAPI band can be seen near the end of the long arm of the largest metacentric pair in the European karyotype and two interstitial bands in the largest acrocentric chromosome pair in both karyotypes. These bands apparently mark the sites of tandem fusions. The karyotypes differ by several rearrangements including both tandem fusions and centric fusions.

279

280

Ruth B. Phillips

chromosome, and the piece is inverted before fusion. Pericentric inversions that involve the centromere result in a different chromosome arm ratio, so are easier to detect. As described in the previous section, Robertsonian translocations have been followed by inversions to produce derived karyotypes that are speciﬁc to certain populations of pink salmon. 2. Duplications and Deletions Small duplications or deletions can cause changes in the size of the short arms in subtelocentrics that are detectable without special staining methods. This type of variation appears to be very common in many ﬁsh species (reviewed by Sola et al., 1981). If the duplications involve repetitive DNA, they may be detected using special staining methods (see later). Differences in the amounts of ribosomal DNA (either 18S or 5S) may also cause a difference in the size of the short arm. These differences in copy number can be detected using various methods, as explained later. 3. Reciprocal Translocations Reciprocal translocations do not produce a change in chromosome number. In this type of translocation, portions of the chromosome are exchanged between two pairs. Robertsonian translocations and tandem fusions are easy to detect because they change the diploid chromosome number, but detection of reciprocal translocations requires markers for each chromosome arm or paint probes, neither of which is available for most ﬁshes at the present time.

C. VARIATION

IN

CHROMOSOME BANDING PATTERNS

Several banding techniques have been used to reveal intraspeciﬁc polymorphisms in animal chromosomes. Variation in the amount, location, and sequence of constitutive heterochromatin (the highly repetitive DNA found near centromeres and telomeres) can be detected by C banding and various ﬂuorescent banding techniques. Each species has several subclasses of these sequences that vary in the base sequence of the repeating unit (100–600 bp). C banding stains repetitive DNA, regardless of sequence; DAPI and Quinacrine stain primarily AT-rich repetitive DNAs; and CMA3 (chromomycin A3) stains GC-rich repetitive DNAs (reviewed in Comings, 1978; Sumner, 1982). In a number of species, these repetitive DNAs have been cloned and characterized (reviewed in Phillips and Reed, 1996; Phillips, 2001), so speciﬁc families of sequences can be detected using molecular cytogenetic methods. In addition to the highly repetitive DNAs, special staining methods can be used to detect the chromosomal location of the ribosomal DNA (rDNA) cistrons, which code for the 5.8S, 18S, and 28S ribosomal RNAs

Chromosome Morphology

281

and are found in multiple copies at the nucleolar organizer regions in eukaryotes, as described later. 1. Nucleolar Organizer Regions Variation in the number and size of the nucleolar organizer regions (NORs) can be detected staining with silver (Howell and Black, 1980) or CMA3, which stains NORs regardless of transcriptional activity in many ﬁsh species (Amemiya and Gold, 1986; Phillips and Ihssen, 1985b). There are a few notable exceptions including zebraﬁsh in which NORs are not stained by CMA3 (Gornung et al., 1997). In this species, CMA3 stains only a GC-rich repetitive DNA sequence that is found adjacent to the centromeres on 40% of the chromosomes (He et al., 1992; Phillips, unpublished). Currently, the most accurate technique for identifying the NORs is in situ hybridization with a ribosomal RNA probe. Stock variation in the number and size of NORs has been found in many ﬁsh species. Examples include lake trout (Phillips et al., 1989), Arctic char (Phillips et al., 1988), brown trout (Pendas et al., 1993; Castro et al., 1994), various cyprinids (reviewed in Gold, 1984 and Gold and Amemiya, 1986), and poecilids (Sola et al., 1990, 1992). Figure 14-4 shows variation in Arctic char. Variation in the number of NORs per cell can also be documented in interphase cells as shown in Figure 14-1 in which it was used as a method for identiﬁcation of triploidy. [Accurate assessment of polyploidy from NOR staining is only feasible in species that normally have only 1 pair of NORs per cell and must be done on cells from the same age ﬁsh, Phillips et al. (1986)]. Differences between hatchery and wild populations in Atlantic salmon have been detected using variation in NORs (Woznicki and Jankun, 1994a,b). Often closely related species with otherwise similar or identical karyotypes will have different chromosomal locations of NORs. For example, the NOR is located on the second largest chromosome pair in the eastern oyster, Crassostrea virginia, but on the smallest chromosome pair in C. giga, the Paciﬁc oyster (Xu et al., 2001). 2. Heterochromatin Additions and Deletions (C Bands) Variation in the total amount of repetitive DNA found near the centromeres and telomeres is very common in ﬁsh species and can be detected using C banding. Intraspeciﬁc variation in the size and location of C bands has been observed in natural populations of several ﬁshes including the iwana, Salvelinus leucomaenis (Ueda and Ojima, 1983a), Arctic char (Hartley, 1989), lake trout (Phillips and Ihssen, 1989), and various whiteﬁshes (Jankun et al., 1995) and can serve as population markers (Fig. 14-5). In a number of cases, variation in the amount of heterochromatin on certain chromosome pairs has been used to distinguish hatchery stocks from wild stocks [e.g., Atlantic salmon (Woznicki and Jankun, 1996)].

282

Ruth B. Phillips

FIGURE 14-4. Partial CMA3-stained karyotypes showing locations of NORs in individuals from three different stocks of Arctic char (Salvelinus alpinus). Top row: Northwest Territories; middle row: Labrador; bottom row: Scotland. From Phillips et al. (1988).

3. Variation in Sequence of Repetitive DNAs: Fluorochrome Bands In certain species, some of the heterochromatin blocks stain positively with various ﬂuorochromes, and polymorphisms in the number and location of these bands have been shown to be inherited (Phillips and Ihssen, 1986). Stockspeciﬁc variation in the number of these bands has been found for several ﬁsh species, including lake trout (Phillips and Ihssen, 1989), Arctic char (Pleyte et al., 1988), and iwana (Ueda and Ojima, 1983b). Minor changes in the amount and sequence of the AT-rich repetitive sequences in closely related ﬁsh species may result in bright staining of the heterochromatin

Chromosome Morphology

283

FIGURE 14-5. C-banded karyotype from a male lake trout (Salvelinus namaycush). Note the large C band on the short arm of the X which is missing on the Y chromosome (Phillips and Ihssen, 1985a; Reed and Phillips, 1995).

with quinacrine in one species, but not in another species. For example, most of the large blocks of telomeric heterochromatin on the metacentric chromosomes of lake trout stains with quinacrine, but similar sites in brook trout generally do not. Thus, the parental origin of most of the metacentric chromosomes in the hybrid splake can be identiﬁed with Q banding. Repetitive DNAs have been isolated and sequenced from a number of ﬁsh species (reviewed in Phillips and Reed, 1996). For example, the sequence of the AT-rich telomeric heterochromatin in lake trout has been isolated and can be detected using in situ hybridization (Reed and Phillips, 1995a). A very similar sequence is also found at the telomeres of some of the chromosomes in other salmonid species, including Arctic char (Hartley and Davidson, 1994; Reed and Phillips, 1997) and rainbow trout (Reed et al., 1997) (see section III, A). Recently, the vertebrate telomeric sequence (TTAGGG)n was localized to telomeres in four bivalve molluscs (Wang and Guo, 2001).

II. STANDARD METHODS FOR DETECTING CHROMOSOME VARIATION Chromosome methods for ﬁsh have been reviewed recently (Gold, et al., 1990; Thorgaard and Disney, 1990; Report of the First International Workshop on Fish Cytogenetic Techniques, 1992; Phillips and Reed, 2003). Methods for shellﬁsh are given in Thiriot-Quievreux and Ayraud (1982) and Leitao et al. (1999). In order to examine chromosomes, dividing tissue must be obtained. There are

284

Ruth B. Phillips

invasive methods that require sacriﬁce of the animal and noninvasive methods that do not. The most widely used methods in the ﬁrst category include preparations from anterior kidney of juvenile and adult ﬁsh, gills of shellﬁsh, and embryos of developing ﬁsh and shellﬁsh. Methods in the second category for ﬁnﬁsh include blood culture (Hartley and Horne, 1985) and regenerating tissue from ﬁnclips. For very small ﬁshes and shellﬁsh, cell lines can be started either from embryos, ﬁns, or gills. Kidney and embryo preparations are best when made after incubation of the cells or tissues in colchicine, a chemical that delays the formation of the spindle. For some species the number of dividing cells (mitotic index) is low and better preparations are obtained if ﬁsh are stimulated by injection of the mitogen phytohemaglutinin or cobalt chloride a few days prior to sacriﬁce. Increased rates of mitosis in embryos may be stimulated by raising the incubation temperature (heat shock).

A. TISSUE SAMPLING DIVIDING CELLS

AND

TREATMENT

TO

OBTAIN

Since it is important to obtain dividing cells in order to examine the chromosomes, cells with high mitotic activity must be examined. This means tissue sampling is limited to tissues with a high mitotic index, or cell division must be induced during in vitro culturing. Tissues with high mitotic index include embryos, anterior kidney (analogous to bone marrow in mammals), and regenerating tissues. In vitro stimulation is usually applied to lymphocyte cultures or tissue cultures using mitogens such as bacterial lipopolysaccharides or phytohemaglutinin.

B. COLCHICINE TREATMENT, HYPOTONIC TREATMENT, AND FIXATION Mitotically active cells must be blocked with colchicine in order to produce condensed chromosomes, exposed to a hypotonic solution to swell the cells to enhance spreading, and ﬁxed, usually in Carnoy’s ﬁxative (3:1, methanol:acetic acid). The suspension of ﬁxed cells is dropped onto slides and dried as described below.

C. SLIDE PREPARATION The method of slide preparation depends on whether solid tissue or cells have been ﬁxed. The method of Kligerman and Bloom (1977) is used for solid tissue, while the standard splash technique is used for ﬁxed cells. Slides made by the

Chromosome Morphology

285

splash technique are usually superior for banding purposes. To obtain good spreading of chromosomes it is important for the slides to dry slowly. We usually allow slides to dry on a slide warmer at 30°–40°C with adequate humidity. (To ensure adequate humidity, use the cover on the slide warmer and place a couple of wet paper towels under the cover with the slides.)

D. STAINING Although chromosome number can be determined without speciﬁc banding techniques, detection of many of the chromosome polymorphisms requires special staining methods (see section II, F). For normal staining, a solution of 10% Giemsa made up in a phosphate buffer at pH 6.8–7.0 is applied for 15 min.

E. CHROMOSOME BANDING TECHNIQUES 1. C Banding C bands represent constitutive heterochromatin, which is visualized by selective removal of the less compacted DNA with alkali reagents such as barium hydroxide and/or treatment with salt solutions followed by staining with Giemsa. A commonly used protocol is that of Sumner (1972). 2. Fluorochrome Banding: DAPI and Q Bands These bands can be revealed by staining of chromosome slides with a ﬂuorochrome solution and examination with the ﬂuorescence microscope (see Phillips and Hartley, 1988). Mounting in an antifade solution is important to avoid bleaching of slides. 3. NOR Banding Nucleolar organizer regions (NORs) are the sites of the ribosomal RNA genes (rDNA). Many of these sites are actively transcribed during the cell cycle so that they appear decondensed and may be visualized as secondary constrictions or gaps in the chromosome, if they are located at interstitial locations. The silver staining technique of Howell and Black (1980) detects active NORs because ribosomal proteins are associated with these regions and these are stained with silver. In most ﬁshes, the rDNA is GC-rich and NORs can be visualized regardless of activity in most species using CMA3 and mithromycin. An alternative (DPI, denaturation/propidium iodide staining) based on the thermal characteristics of rDNA has recently been described for the staining of NORs in ﬁshes (Rab et al., 1996). Bands produced with this technique directly correspond with those pro-

286

Ruth B. Phillips

duced by CMA3 staining. However, in some species, both of these techniques stain sites of GC-rich heterochromatin instead of NORs. Thus, for unambiguous identiﬁcation of NORs it is necessary to use in situ hybridization with probes speciﬁc to ribosomal DNA (rDNA). The rDNA probes will also identify small NORs which can be missed by the other methods (Reed and Phillips, 1995b). 4. Replication Banding Replication bands can be obtained in ﬁshes by adding BudR during the last 8 hr of cell culture (Fujiwara et al., 2001) or by injecting ﬁsh directly with it several hours before sacriﬁce (Report of the First International Workshop on Fish Cytogenetic Techniques, 1992). 5. Restriction Enzyme (RE) Banding Restriction enzymes have been used to reveal bands on ﬁsh chromosomes (Lloyd and Thorgaard, 1988). A solution containing the enzyme is applied to the slide for several hours and then washed off. Apparently these enzymes can digest tandemly repetitive DNA on chromosomes if the repeats contain a restriction site for the particular enzyme being used. After digestion, the C band which contains repetitive DNA cut by the enzyme will appear faintly stained as compared to C bands containing other repetitive DNAs that lack the proper sites. Thus, different restriction enzymes can be used to identify subclasses of heterochromatin (C bands). C bands with the same RE patterns may have completely different sequences, although they must have at least one restriction site in common. 6. Fluorescence In Situ Hybridization (FISH) A more direct approach to classifying and localizing repetitive DNAs is to isolate, clone, and sequence them. Individual clones can then be labeled as probes to identify the location of these repeats in the genome with FISH (Fig. 14-6) (see section III).

F. PHOTOGRAPHY

AND

ANALYSIS

The chromosomes are examined under the microscope and karyotypes are prepared. The conventional method of preparing karyotypes is to photograph the cells, make prints, and cut out and arrange the chromosomes. An alternative method is to capture digital images directly from the microscope or from negatives for computer-based image analysis and karyotyping. Chromosomes can be analyzed using software packages such as Adobe Photoshop, NIH image, or

Chromosome Morphology

287

FIGURE 14-6. Metaphase of zebraﬁsh chromosomes from the AB strain hybridized with a clone to a speciﬁc centromeric DNA (L22) labeled with Spectrum Orange (probe shown in white). This sequence is found at the centromeres of most, but not all chromosome pairs. The size of the bands is also variable.

software especially prepared for karyotyping, such as Cytovision (Applied Imaging, Inc.).

III. MOLECULAR CYTOGENETIC METHODS

A. FLUORESCENCE IN SITU HYBRIDIZATION (FISH) DNA sequences can be localized in metaphase and interphase cells using the technique of ﬂuorescence in situ hybridization (FISH). This method has revolutionized human cytogenetics in the past decade (reviewed in Lichter, 1997). Speciﬁc chromosome regions, entire chromosomes, or genomes of one species in interspeciﬁc hybrids can be “painted” using the appropriate probes. Chromosome paint probes are available for individual human chromosomes and

288

Ruth B. Phillips

chromosome arms, and probes are being produced for individual chromosome bands. Abnormalities in chromosome number and structure can be identiﬁed by examination of interphase cells with these probes. For example, if DNA from chromosome 21 is labeled with an orange ﬂuorochrome, individuals with trisomy 21 will have three orange spots in each interphase nucleus. Fetuses are routinely sexed using a red centromere probe for the X chromosome and a green probe to the repetitive DNA on the Y chromosome. Human paint probes have been used to identify homologous blocks of genes in many other mammals including carnivores, artiodactyls, and primates (reviewed in Chowdhary et al., 1998). The FISH technique involves several steps: probe preparation and denaturation, slide preparation and denaturation, hybridization of probes on the slide, posthybridization washes, detection, and visualization of label. Probes are prepared by labeling speciﬁc DNA sequences with biotin, dioxygenin, or, more recently, a nucleotide directly conjugated to a speciﬁc ﬂuorochrome. Until very recently, most FISH experiments used one or two probes at a time because direct labeling with different colored ﬂuorochromes was not sensitive enough for single copy genes. In order to localize two genes at a time, one would be labeled with biotin and the other with dioxygenin. The direct-labeled ﬂuorochromes were used only for localizing repetitive DNAs. However, it is now possible to localize single copy genes using direct labeling. This is the result of introduction of digital cameras which can detect lower light levels, marketing of a number of new ﬂuorochromes with restricted nonoverlapping emission spectra, and the availability of large insert clones from BAC and PAC libraries. The order of genes on a chromosome arm can be determined by in situ hybridization with probes to these BACs labeled with different colors (reviewed in Lichter, 1997). A new karyotyping method based on multicolor ﬂuorescence in situ hybridization (M-FISH) has been introduced for human chromosomes (Speicher et al., 1996). Chromosome-speciﬁc paint probes have been constructed using DNA from ﬂow-sorted chromosomes of each pair. These are labeled with a different combination of ﬂuorochromes so each chromosome pair can be visualized as a different color. The M-FISH protocol involves acquiring digital images separately with a CCD camera and combining them with software that generates a composite image in which each chromosome is pseudocolored based on its ﬂuorochrome composition. Large clones containing single genes obtained from libraries constructed with a variety of vectors, including yeast (YACs), bacterial artiﬁcial chromosomes (BACs), and P1 phage (PACs), are easily visualized on chromosomes with FISH. With M-FISH many probes can be localized in a single experiment. Thus, gene order on individual chromosome arms can be determined, or these can be combined to produce paint probes. The M-FISH technique could be useful for gene localization experiments utilizing in situ hybridization with ﬁsh chromosomes because probes to speciﬁc centromeres for chromosome identiﬁcation could be added to the ones for the genes being mapped.

289

Chromosome Morphology

B. APPLICATION OF FISH TO FISH SHELLFISH CHROMOSOMES

AND

The application of ﬂuorescence FISH to ﬁsh genetics has been reviewed recently (Phillips and Reed, 1996; Phillips, 2001). Different repetitive and moderately repetitive DNAs including ribosomal RNAs and histones (Pendes et al., 1994) have been localized to centromeres, telomeres, and sex chromosomes of several ﬁshes, and paint probes have been produced using PCR-based microdissection of chromosome regions. Centromeric sequences have been localized in the Paciﬁc oyster (Wang et al., 2001). Single copy genes have been localized in zebraﬁsh and rainbow trout using PAC or BAC clones containing speciﬁc genes as probes (Phillips, unpublished) (Fig. 14-7). In pufferﬁsh, the BAC clones are being used to help produce a framework genetic map (Catherine Ozouf-Costaz, personal communication). Rainbow trout probes have been shown to work on other salmonids including chinook salmon, lake trout, and Atlantic salmon. Such probes could be used to make a quick genetic map for other related species and to identify the chromosome arms involved in Robertsonian fusions. Currently, genome projects are under way for ﬁsh species, including zebraﬁsh, rainbow trout, Atlantic salmon, medaka, tilapia, catﬁsh, and pufferﬁsh. The large insert libraries prepared for these projects will produce reagents that could be used for paint probes for these species and other related ﬁsh species. With appropriate probes it should be possible to identify Robertsonian translocations in interphase cells if the two acrocentric chromosomes involved in the fusion are labeled with different colors. Intraspeciﬁc chromosome variation has been documented in rainbow trout, cutthroat trout, and Atlantic salmon, so in the future these probes might be used for stock identiﬁcation.

IV. DISCUSSION AND CONCLUSIONS Chromosome markers usually do not require as much development time as molecular markers, but they have the disadvantage of requiring living tissue. Once developed for a particular group, molecular methods are more efﬁcient because multiple loci can be scored on a single gel. In addition, although DNA variation of some type is present in virtually every species, intraspeciﬁc chromosome variation may not be present. In fact it appears that marine ﬁshes have much more stable karyotypes than freshwater ﬁshes (Singh et al., 1997), so chromosome number variation is rare in them. Although chromosome number variation has been observed in shellﬁsh, no intraspeciﬁc variation has been documented. Usually NOR staining will reveal intra-speciﬁc variation in almost any species, but it may not be stock-speciﬁc. Most species have a unique karyotype, so cytogenetic methods can be very valuable for analysis of hybrid zones between closely related species or

290

Ruth B. Phillips

FIGURE 14-7. Karyotype of rainbow trout chromosomes from the Donaldson strain hybridized with a BAC clone containing the Fgf6 gene labeled with Spectrum Orange (probe shown in white).

subspecies. For ﬁsh species with intraspeciﬁc chromosome number variation, hatchery ﬁsh may have a different chromosome number than wild ﬁsh, so in that case it would be possible to estimate the percentage of hatchery vs. wild ﬁsh and identify any hybrids between the two using cytogenetic methods. Such variation is especially common in salmonid ﬁshes, and often involves intraspeciﬁc translocation polymorphisms. In human genetics, molecular markers diagnostic for speciﬁc translocations involved in speciﬁc tumors and birth defects have been developed. These types of assays could make identiﬁcation of chromosome variation more feasible for population studies in the future.

REFERENCES Amemiya, C. T. and Gold, J. R. 1986. Chromomycin A3 stains nucleolar organizer regions of ﬁsh chromosomes. Copeia 1986(1): 226–231. Birnstein, V. J., Poletaev, A. I., and Goncharov, B. F. 1993. The DNA content in Eurasian sturgeon species determined by ﬂow cytometry. Cytometry 14: 377–383. Castro, J., Rodriguez, S., Arias, J., Sanchez, L., and Martinez, P. 1994. A population analysis of Robertsonian and AgNOR polymorphisms in brown trout (Salmo trutta). Theoretical Applied Genetics.

Chromosome Morphology

291

Chowdhary, B. P., Taudsepp, T., Fronicke, L., and Scherthan, H. 1998. Emerging patterns of comparative genome organization in some mammalian species as revealed by Zoo–FISH. Genome Research 8: 577–589. Collares-Pereira, M. J. 1985. The “Rutilus alburnoides (Steindachner, 1866) Complex” (Pisces, Cyprindae). II. First data on the karyology of a well-established diploid-triploid group. Arquivos do Museu Bocage. Series A III: 69–89. Comings, D. E. 1978. Mechanisms of chromosome banding and implications for chromosome structure. Ann. Rev. Tenet. 12: 25–46. Cotter, D., O’ Donovan, V., Drumm, A., Roche, N., Nigel, L. E., and Wilkins, N. P. 2002. Comparison of freshwater and marine performances of all female diploid and triploid Atlantic salmon (Salmo salar L.). Aquacul. Res. 33: 43–53. Felip, A., Piferrer, F., Znuy, S., and Carillo, M. 2001. Comparative growth performance of diploid and triploid European sea bass over the ﬁrst four spawning seasons. J. Fish Biol. 58: 76–88. Fujiwara, A., Nishida-Umehara, C., Sakamoto, T., Okamoto, N., Nakayma, I., and Abe, S. 2001. Improved ﬁsh lymphocyte culture for chromosome preparation. Genetica 111: 77–89. Galetti Jr., P. M., Aguilar, C. T., and Molina, W. F. 2000. An overview of marine ﬁsh cytogenetics. Hydrobiologia 420: 55–62. Gold, J. R. 1979. Fish cytogenetics. In Fish Physiology, Vol. 8, 1979, pp. 353–404. Gold, J. R. 1984. Silver staining and heteromorphism of chromosomal nucleolus organizer regions in north American cyprinid ﬁshes. Copeia 1984: 133–139. Gold, J. R. and Amemyia, C. T. 1986. Cytogenetic studies in North American minnows (Cyprinidae). XII. Patterns of chromosomal nucleolus organizer region variation among 14 species. Canadian Journal of Zoology 64: 1869–1877. Gold, J. R., Karel, W. J., and Strand, M. R. 1980. Chromosome formulae of North American ﬁshes. Progressive Fish Culturist 42: 10–23. Gold, J. R., Li, Y. C., Shipley, N. S., and Powers, P. K. 1990. Improved methods for working with ﬁsh chromosomes with a review of metaphase chromosome banding. Journal of Fish Biology 37: 563–575. Gornung, E., Gabrielli, I., Cataudella, S., and Sola, L. 1997. CMA3-banding pattern and ﬂuorescence in situ hybridization with 18S rRNA genes in zebraﬁsh chromosomes. Chromosome Research 5: 40–46. Gorshkov, S. A. and Gorshkova, G. V. 1981. Chromosome polymorphism of the pink salmon Onocrhynchus gorbuscha (Walb). Tsitologiya 23: 954–960. Guo, X. and Allen, S. K., Jr. 1994. Reproductive potential and genetics of the triploid Paciﬁc oyster, Crassostrea gigas (Thunberg). Biol. Bull. (Woods Hole). 187: 309–318. Guo, X., DeBrosse, G., and Allen, S. K., Jr. 1996. All-triploid Paciﬁc oysters (Crassostrea gigas Thunberg) produced by mating tetraploids and diploids. Aquaculture 142: 149–161. Guo, X. and Allen, S. K., Jr. 1997. Sex and meiosis in autotetraploid Paciﬁc oyster, Crassostrea gigas (Thunberg). Genome 40: 397–405. Hartley, S. E. 1987. The chromosomes of salmonid ﬁshes. Biological Reviews (Cambridge) 62: 197–214. Hartley, S. E. 1989. Chromosome and constitutive heterochromatin distribution in Arctic charr, Salvelinus alpinus (L.) (Pisces: Salmonidae). Genetica 79: 161–166. Hartley, S. E. and Davidson, W. S. 1994. Characterization and distribution of genomic repeat sequences from Arctic char (Salvelinus alpinus). In A. R. Beaumont (ed.), Genetics and Evolution of Aquatic Organisms. Chapman & Hall, London, UK, pp. 271–280. Hartley, S. E. and Horne, M. T. 1982. Chromosome polymorphism in the rainbow trout (Salmo gairdneri) Richardson. Chromosoma 87: 461–468. Hartley, S. E. and Horne, M. T. 1984a. Chromosome relationships in the genus Salmo. Chromosoma 90: 229–237.

292

Ruth B. Phillips

Hartley, S. E. and Horne, M. T. 1984b. Chromosome polymorphism and constitutive heterochromatin in Atlantic salmon, Salmo salar. Chromosoma 89: 377–380. Hartley, S. E. and Horne, M. T. 1985. Cytogenetic techniques in ﬁsh genetics. J. Fish Biol. 26: 575–582. He, L., Zhu, Z., Faras, A. J., Guise, K. S. Hackett, P. B., and Kapuscinski, A. R. 1992. Characterization of AluI repeats of zebraﬁsh (Brachydanio rerio). Molecular Marine Biology and Biotechnology 1: 125–135. Howell, W. M. and Black, D. A. 1980. Controlled silver-staining of nucleolus organizer regions with a protective colloidal developer: a 1-step method. Experientia 36: 1014–1015. Jankun, M., Rab, P., Vuorinen, J., and Luczynski, M. (1995). Chromosomal polymorphism in Coregonus lavaretus populations from two locations in Finland and Poland. Arch. Hydrobiol. Spec. Issues Adv. Limnol., 46: 1–11. Kligerman, A. D. and Bloom, S. E. 1977. Rapid chromosome preparations from solid tissues of ﬁshes. J. Fish. Res. Board Can. 34: 266–269. Klinkhardt, M., Tesche, M., and Greven, H. J. 1995. Database of Fish Chromosomes. Westarp Wissenschaften, Magdeburg. 237 pp. Kulikova, N. I. 1971. Intraspeciﬁc variability of karyotypes of the chum salmon (Oncorhynchus keta) (Walb.). Journal of Icthyology 11: 977–983. Lichter, P. 1997. Multicolor FISHing: what’s the catch? Trends in Genetics 13: 475–478. Lilysestrom, C. G., Wolters, W. R., Bury, D., Rezk, M., and Dunham, R. A. 1999. Growth, carcass traits, and oxygen tolerance of diploid and triploid catﬁsh hybrids. N. Am. J. Aqua. 61: 293– 303. Lloyd, M. A. and Thorgaard, G. H. 1988. Restriction endonuclease banding of rainbow trout chromosomes. Chromosoma 96: 171–177. Nakamura, H. 1985. A review of molluscan cytogenetic information based on CISMOCH—Computerized index system for molluscan chromosomes. Bivalvia, Polyplacophora and Cephalopoda. Venus Jpn. J. Malacol. 44: 193–225. Ojima, Y. 1980. Chromosomes in evolution of eukaryotic groups. Fish Cytogenetics, Vol. 1, Ch. 2. Ostberg, C. O. and Thorgaard, G. H. 1999. Geographic distribution of chromosome and microsatellite DNA polymorphisms in Oncorhynchus mykiss native to western Washington. Copeia, 1998). Park, G. M., Yong, T. S., Im-Kyung, I. L, and Chung, E. Y. Karyotypes of three species of Corbicula (Bivalvia: Veneroida) in Korea. J. Shellﬁsh Res. 19: 979–982. Pendas, A. M., Moran, P., and Garcia-Vasquez, E. 1993. Multichromosomal location of ribosomal genes and heterochromatin association in brown trout. Chromosome Res. 1: 63–67. Pendas, A. M., Moran, P., and Garcia-Vasquez, E. 1994. Organization and chromosomal localization of the major histone cluster in brown trout, Atlantic salmon and rainbow trout. Chromosoma 103: 147–152. Phillips, R. B. 2001. Application of ﬂuorescence in situ hybridization to ﬁsh genetics and genomics. Marine Biotechnology 3: S145–S152. Phillips, R. B. and Hartley, S. E. 1988. Fluorescent banding patterns of the chromosomes of the genus Salmo. Genome 30: 193–197. Phillips, R. B. and Ihssen, P. E. 1985a. Identiﬁcation of sex chromosomes in lake trout (Salvelinus namaycush). Cytogenetics and Cell Genetics 39(1): 14–18. Phillips, R. B. and Ihssen, P. E. 1985b. Chromosome banding in salmonid ﬁshes: nucleolar organizers in Salmo and Salvelinus. Can. J. Genet. Cytol. 27: 433–440. Phillips, R. B. and Ihssen, P. E. 1986. Inheritance of Q band chromosomal polymorphisms in lake trout (Salvelinus namaycush). J. Hered. 77: 93–97. Phillips, R. B. and Ihssen, P. E. (1989). Population differences in chromosome banding polymorphisms in lake trout (Salvelinus namaycush). Transactions of the American Fisheries Society 118: 64–73.

Chromosome Morphology

293

Phillips, R. B. and Ihssen, P. E. 1990. Genetic marking of ﬁsh using variation in chromosomes and nuclear DNA. In N. C. Parker, A. E. Giogri, R. C. Heidinger, D. B. Jerter Jr., E. D. Prince, and G. A. Winans (eds.), Fish Marking Techniques. American Fisheries Symposium 7: 499–513. Phillips, R. B. and Kapuscinski, A. R. 1987. A Robertsonian polymorphism in pink salmon (Oncorhynchus gorbuscha) involving the NOR region. Cytogenetics and Cell Genetics 44: 148–152. Phillips, R. B. and Kapuscinski, A. R. l988. High frequency of translocation heterozygotes in odd year populations of pink salmon (Oncorhynchus gorbuscha). Cytogenetics and Cell Genetics 48: 178–182. Phillips, R. B. and Rab, P. 2001. Chromosome evolution in the Salmonidae (Pisces): an update. Biological Reviews. 76: 1–25. Phillips, R. B. and Reed, K. M. 1996. Application of ﬂuorescence in situ hybridization (FISH) to ﬁsh genetics. Aquaculture 140: 197–216. Phillips, R. B. and Reed, K. M. 2003. Chromosome variation. In E. M. Hallerman (ed.), Genetic Principles and Practices for Fisheries Scientists. American Fisheries Society, Ch. 3. Bethesda, MD, pp. 37–58. Phillips, R. B., Pleyte, K. A., and Hartley, S. E. 1988. Stock-speciﬁc differences in the number and chromosome positions of the nucleolar organizer regions in Arctic char (Salvelinus alpinus). Cytogenet. Cell Genet. 48: 9–12. Phillips, R. B., Pleyte, K. A., and Ihssen, P. E. 1989. Patterns of chromosomal nucleolar variation in ﬁshes of the genus Salvelinus. Copeia 1980(1): 47–53. Phillips, R. B., Zajicek, K. D., Ihssen, P. E., and Johnson, O. 1986. Application of silver staining to the identiﬁcation of triploid ﬁsh cells. Aquaculture 54: 313–319. Pleyte, K. A., Phillips, R. B., and Hartley, S. E. l989. Q band chromosomal polymorphisms in Arctic char (Salvelinus alpinus). Genome 32: 129–133. Rab, P. and Collares-Pereira, M. J. 1995. Chromosomes of European cyprinid ﬁshes (Cyprinidae, Cypriniformes): a review. Folia-Zoologica 44: 193–214. Rab, P., Reed, K. M., Ponce De Leon, A., and Phillips, R. 1996. Denaturation/propodium iodide staining: a new method for detecting nucleolar organizer regions (NORs) in ﬁsh chromosomes. Biotechnic. Histochem. 71: 157–162. Reed, K. M. and Phillips, R. B. 1995a. Molecular characterization and cytogenetic analysis of highly repeated DNAs of lake trout, Salvelinus namaycush. Chromosoma 104: 242–251. Reed, K. M. and Phillips, R. B. 1995b. Molecular cytogenetic analysis of the double CMA3 chromosome in lake trout, Salvelinus namaycush. Cytogenetics and Cell Genetics 70: 104–107. Reed, K. M. and Phillips, R. B. 1997. Polymorphism of the nucleolus organizer region (NOR) on the putative sex chromosomes of Arctic char (Salvelinus alpinus) is not sex related. Chromosome Research 5: 221–227. Reed, K. M., Dorschner, M. O., and Phillips, R. B. 1997. Characteristics of two salmonid repetitive DNA families in rainbow trout (Oncorhynchus mykiss). Cytogenetics and Cell Genetics 79: 184–187. Report of the First International Workshop on Fish Cytogenetic Techniques, Concarneau, France, September, 1992. (Available from C. Ozouf-Costaz, Laboratorie d’Ichtyologie, MNHN 43 rue Cuvier, 75231, Paris Cedex 05, France. Roberts, F. L. 1970. Atlantic salmon (Salmo salar) chromosomes and speciation. Transactions of the American Fisheries Society 99: 105–111. Sheenhan, R. J., Shasteen, S. P., Suresh, A. V., Kapuscinski, A. R., and Seeb J. E. 1999. Better growth in all female diploid and triploid rainbow trout. Trans. Am. Fish. Soc. 128: 491–498. Singh, L. B., Nagpure, N. S., Singh, S. P., and Pandey, O. P. 1997. Hydrobiologia 420: 55–62. Sola, L., Cataudella, S., and Capanna, E. 1981. New developments in vertebrate cytotaxonomy. III. Karyology of bony ﬁshes: a review. Genetica 54: 285–328.

294

Ruth B. Phillips

Sola, L., Iaselli, V., Rossi, A. R., Rasch, E. M., and Monaco, P. J. 1990. Cytogenetics of bisexual/ unisexual species of Poecilia. I. C-bands, Ag-NOR polymorphisms and sex chromosomes in three populations of Poecilia latipinna. Cytogenet. Cell Genet. 53: 148–154. Sola, L., Rossi, A. R., Iaselli, V., Rasch, E. M., and Monaco P. J. 1992. Cytogenetics of bisexual/ unisexual species of Poecilia. II. Analysis of heterochromatin and nucleolar organizer regions in Poecilia mexicana mexicana by C-banding and DAPI, quinacrine, chromomycin A3 and silver staining. Cytogenet. Cell. Genet. 60: 229–235. Speicher, M. R., Ballard, G. S., and Ward, D. C. 1996. Karyotyping human chromosomes by combinatorial multi-ﬂuor FISH. Nat. Genet. 12: 368–375. Sumner, A. T. 1972. A simple technique for demonstrating centromeric heterochromatin. Exp. Cell Res. 75: 304–306. Sumner, A. T. 1982. The nature and mechanisms of chromosome banding. Cancer Genetics and Cytogenetics 6: 59–87. Thiriot-Quievreux, C. and Ayraud, N. 1982. Les caryotypes de quelques epseces de Bivalves et de Gasteropodes marins. Mar. Biol. 70: 165–172. Thorgaard, G. H. 1976. Robertsonian polymorphism and constitutive heterochromatin distribution in the chromosomes of the rainbow trout. Cytogenet. Cell Genet. 17: 174–1190. Thorgaard, G. H. 1983. Chromosomal differences among rainbow trout populations. Copeia 1983: 650–662. Thorgaard, G. H. and Allen, S. K. 1987. Chromosome manipulation and markers in ﬁshery management. In N. Ryman and F. Utter (eds.), Population Genetics and Fishery Management. Ch. 13: 319–331. University of Washington Press, Seattle, WA, and London. Thorgaard, G. H. and Disney, J. E. 1990. Chromosome preparation and analysis. Methods in Fish Biology, Ch. 6. American Fisheries Society, Bethesda, MD. Tillmann, B. 2001. Use of sterile triploid Atlantic salmon (Salmo salar L.) for aquaculture in New Brunswick, Canada. ICES-J. Mar. Sci. 58: 525–529. Turner, B. J., Grudzen, T. A., Adkinsson, K. P., and Worrell, R. A. 1985. Extensive chromosomal divergence within a single river basin in the goodeid ﬁsh, Ilyodon furcidens. Evolution. 39: 122–134. Ueda, T. and Kobayashi, J. 1990. Karyotype differentiation of Atlantic salmon, Salmo salar, especially the sequential karyotype change. La. Kromosomo II-58: 1967–1972. Ueda, T. and Ojima, Y. 1983a. Karyotypes with C banding patterns of two species in the genus Salvelinus of the family Salmonidae. Proc. Jap. Acad. 59: Ser B: 343–346. Ueda, T. and Ojima, Y. 1983b. Geographic and chromosomal polymorphisms in the iwana (Salvelinus leucomaenis). Proc. Jpn. Acad. 59(B)8: 259–262. Wang, Y. and Guo, X. 2001. Chromosomal mapping of the vertebrate telomeric sequence TTAGGG)n in four bivalve molluscs by ﬂuorescence in situ hybridization. J. Shellﬁsh Res. 20: 1187–1190. Wang, Y., Xu, Z., and Guo, X. 2001. A centromeric satellite sequence in the Paciﬁc oyster (Crassostrea gigas Thunberg) identiﬁed by ﬂuorescence in situ hybridization. Mar. Biotech. 3: 486–492. Woznicki, P. and Jankun, M. 1994a. Chromosome polymorphism of Atlantic salmon (Salmo salar) from the Dzwina River, Baltic Sea Basin: arm length and NOR location variation of the eighth chromosome. Can. J. Zool. 72: 364–367. Woznicki, P. and Jankun, M. 1994b. New cytotype of a highly polymorphic NOR bearing chromosome pair in Atlantic salmon Salmo salar (L.) Cytobios 79: 59–62. Xu, Z., Guo, X., Gaffney, P. M., and Pierce, J. 2001. Chromosomal location of the major ribosomal RNA genes in Crassostrea virginica and Crassostrea gigas. Veliger 44: 79–83. Yang, H. P., Li, L., and Guo, X. M. 2001. Preliminary study on inducing polyploidy in Japanese scallop (Patinopecten yessoensis) by cytochalasin B. Acta Zoologica Sinica 47: 459–464.

CHAPTER

15

Genetic Analysis: Allozymes M.-L. KOLJONEN* AND R. WILMOT† *Finnish Game and Fisheries Research Institute, Helsinki, Finland † United States Department of Commerce, National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Juneau, Alaska, USA

I. II. III. IV. V.

Introduction Electrophoresis Allele Frequencies Mixed-Stock Analysis Advantages and Limitations of the Allozyme Data in MSA VI. Factors Affecting the Reliability of Allozyme-Based MSA VII. Examples of Application of Allozymes to MixedStock Analysis References

I. INTRODUCTION Allozymes are the electrophoretic expression of alternative gene forms (alleles) of functionally similar enzymes produced by a gene or gene locus. The combination of two alleles at a particular gene locus determines the genotype of an individual. By determining the genotypic composition of an individual or a population over several loci, we can establish a multilocus genetic mark that may distinguish that individual or population from others (Pella and Milner, 1987). Population and stock are here deﬁned as a group of interbreeding individuals that are sufﬁciently isolated from other groups of individuals of the same species for some level of genetic differentiation to have occurred. Numerous inheritance studies on ﬁsh have established that the observed allozyme variation follows simple known rules of Mendelian inheritance according to which two alleles of each gene locus are inherited from both parents (May et al., 1980; Kornﬁeld et al., 1981). Because genes are inherited, the genetic structure of populations is relatively stable over generations and can Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

295

296

M.-L. Koljonen and R. Wilmot

be determined by standard population genetics methods. If the genetic differentiation is large enough between populations or stocks, analytical methodologies will permit determination of the stock proportions in mixed-stock ﬁsheries (Pella and Milner, 1987). Mixed-stock analysis (MSA) using allozyme variation was ﬁrst developed for Paciﬁc salmon (Oncorhynchus sp.) (Grant et al., 1980, Fournier et al., 1984, Milner et al., 1985). The greatest number of MSA studies have indeed been conducted on Paciﬁc salmon (see Shaklee and Phelps, 1990; Shaklee et al., 1990a; Utter, 1991; Begg et al., 1999), but it has also been used with lake trout (Salvelinus namaycush) (Perkins and Krueger, 1995, Marsden et al., 1989), brown trout (Salmo trutta) (Krueger and May, 1987), Atlantic salmon (Salmo salar) (Koljonen, 1995; Koljonen and Pella, 1997), and Dolly Varden (Salvelinus malma) (Krueger et al., 1999);

II. ELECTROPHORESIS Allozyme variation is studied by enzyme electrophoresis (Shaw and Prasad, 1970; Harris and Hopkinson, 1976; Siciliano and Shaw, 1976; Aebersold et al., 1987), a method based on the separation of electrically charged protein molecules, usually enzymes, in an electric ﬁeld. Mutation in a protein-coding gene locus causes a change in the DNA nucleotide order, which then changes the amino acid composition of the enzyme protein. Of the 20 potential amino acids that make up protein molecules, ﬁve carry an electrical charge—three a positive and two a negative one. Therefore, different protein molecules have different net electrical charges. In practice, about one-third of DNA nucleotide changes are noticeable as charge changes at the enzyme level. Each individual may have several gene loci producing slightly different forms (called isozymes) of the same functional enzyme. These different loci may be active in the same tissue or they may be expressed only in certain tissues. For example, the ﬁve loci of lactate dehydrogenase (LDH) enzyme are tissuespeciﬁc in salmonid ﬁshes, usually such that LDH-1* and LDH-2* are mainly active in muscle tissue, LDH-3* in the heart, LDH-4* in the liver, and LDH-5* in the eyes (Shaklee et al., 1973). In Paciﬁc salmon, samples of heart, liver, muscle, and eye are usually required for the acquisition of data on the full range of usable loci. As enzymes begin to degrade immediately after the death of organisms, fresh or frozen tissue is used for electrophoresis. The tissues are macerated in a buffer solution and introduced into a gel of starch, agarose, cellulose acetate, or other medium with absorbent wicks (Fig. 15-1). Each wick represents a single tissue of an individual ﬁsh, and usually 25 to 40 wicks are used per gel. An electric

297

Genetic Analysis: Allozymes 1 2

Frequency of allele A = 12/20 = 0.6 Frequency of allele B = 8/20 = 0.4

3 4

5 7

6 8

9

(+)

10

1 1

2

3

4

5

6

7

8

9 10

(-)

A

AA AB AA AA AB AB AB BB AB AB

D

Tissue extract

Specific staining procedures 1

B

(+) Power Supply 2

1 2 3 4 5 6 7 8 9 10 3 4 5 6 7 8 9 10

(-)

C FIGURE 15-1. Standard steps for obtaining allele frequency data from electrophoresis. (A) Crude protein is extracted from tissue. (B) Extract from each ﬁsh is introduced individually to gel by ﬁlter paper inserts. (C) Different allozymes move different distances in an electric ﬁeld. (D) Allozymes are made visible with speciﬁc stains, the genotypes (AA, AB, BB) are interpreted, and the allele frequencies are calculated. Modiﬁed from Utter et al. (1987). Reprinted by permission from Washington Sea Grant Program, University of Washington.

ﬁeld is applied over the gel, normally for 4 to 8 hr, and the enzyme molecules migrate through the gel at a rate related to their electrical charges. The shape and weight of the molecules may affect the migration rate to some extent. The gel is then sliced horizontally into thin slabs (usually 3–5). Each slab is individually stained for a different enzyme, and an enzyme-speciﬁc banding pattern appears that can be read and interpreted for each individual. For interpretation of the banding patterns, the molecular structure of the enzyme and the locus structure should be known (Utter et al., 1974, 1987). A standardized system for interpretation and locus nomenclature has been developed by Shaklee et al. (1990b). The end product of electrophoresis is the multilocus genotype of each individual organism. The simplest banding patterns occur for monomeric protein molecules, which are composed of single subunits. If a locus for monomeric proteins has two

298

M.-L. Koljonen and R. Wilmot

codominant alleles, they are usually designated by A and B. In diploid individuals, in which one allele is inherited from each parent, three potential genotypes are possible—AA, AB, and BB. The homozygote individuals with either AA or BB genotypes each produce a single band because of their different migration rates through the gel. The heterozygote individual with the AB genotype displays two bands since it produces both A and B proteins (Fig. 15-1). More than two alleles may occur at the same locus in the population, but only two of the alleles can exist at any locus in any diploid individual. If three alleles occur in the population, one individual could potentially have an AA, AB, BB, AC, BC, or CC genotype at one locus. The number of alleles at variable gene loci in ﬁsh populations is usually 2 to 3 in allozyme data, but it can be as high as 11 in pink (Oncorhynchus gorbuscha) and chum salmon (Oncorhynchus keta). In invertebrates, 6 to 8 alleles are more common, although there are often 8 to 15. In DNA microsatellite data on ﬁshes, the number of alleles easily exceeds 30. The interpretation of banding patterns becomes much more complex when we are dealing with a multimeric enzyme protein molecule composed of two or more protein subunits. An enzyme consisting of two subunits is called a dimer. If a heterozygote individual (AB) with two different alleles produces two types of protein subunit, three types of dimeric protein molecule (AA, BB, BB) could be formed, and thus a three-banded gel pattern will be displayed for all heterozygote individuals. The higher the number of loci, subunits, and alleles, the more complex and difﬁcult to interpret the banding patterns may become. A good overview of the interpretation of complex banding patterns has been given by Utter et al. (1987). Once several individual loci have been analyzed separately, the multilocus genotype for these loci can be formed for each individual. For example, a 7-locus genotype for one individual gets the form A1A2B1B1C2C2D1D3E1E2F2F3G1G1, where a letter designates the locus and the subindex the particular allele. Separate loci can usually be assumed to be independent variables, and allozymes are thought to be selectively relatively neutral. The use of multilocus genotypes causes the potential number of genotype combinations to increase very rapidly. The number of possible multilocus genotypes for loci with two alleles increases as 3L when several loci (L) are analyzed. For ﬁve loci with two alleles, the number of different potential genotypes is 243. When the number of alleles and loci is high, the number of possible genotypes will exceed the population size. The power of different identiﬁcation methods is indeed based on the differences in multilocus genotype distributions among populations.

Genetic Analysis: Allozymes

299

III. ALLELE FREQUENCIES The proportions of each allele at each locus—the allele frequencies—are used to describe and determine the population structure (Fig. 15-1). A population is assumed to be sufﬁciently reproductively isolated from other populations of the same species for some degree of genetic divergence to have taken place. It is extremely rare to ﬁnd ﬁxed or diagnostic allele differences between stocks of the same species, that is, only an A allele in one stock and only a B allele in another. Such differences usually denote separate species. Absolute classiﬁcation of individuals into different populations of the same species is therefore rarely possible with allozyme data. According to genetic theory, a known relationship called Hardy–Weinberg equilibrium should exist between allele frequencies and genotype frequencies in a diploid, random mating population where random union of gametes is assumed and the alleles are inherited according to Mendelian rules. In addition, migration, mutation, selection, or random changes due to very small population size (genetic drift) must not have a marked effect on the genotype frequencies of the population. The theory predicts that if a population is in Hardy–Weinberg equilibrium, the proportions of the genotypes can be determined from the allele frequencies. For a two-allelic locus where p is the frequency of allele A and q is the frequency of allele B (p + q = 1), the expected genotype frequencies can be computed with the formula (p + q)2. This gives the three terms: p2 for the proportion of genotype AA, 2pq for genotype AB, and q2 for genotype BB. For multiallelic loci where pi is the frequency of allele Ai the formula for expected genotype frequencies can be extended to the form (p1 + p2 + . . . + pm)2. This gives m(m + 1)/2 terms corresponding to the frequencies of both possible homozygotes, AiAi, and heterozygotes, AiAj. When independence of loci is assumed, the distributions for multilocus genotypes are products of the single-locus distributions. Signiﬁcant deviations from expected genotype distributions could mean that some of the assumptions are not true. The Hardy–Weinberg equilibrium test can then be run to determine whether marked mixing or substructure occurs in the populations studied. Mixing of genetically differentiated populations results in an excess of heterozygotes in the next generation. Subpopulation structure could cause a deﬁciency in the proportion of heterozygote individuals in relation to the expected numbers and is called the Wahlund effect. The most important feature of the Hardy–Weinberg theorem is that it enables the genotype distributions at a locus in a population to be expressed entirely in terms of allele frequencies. Assuming independence of loci and Hardy–Weinberg equilibrium, it is straightforward to compute the multilocus genotype distributions, including all the potential multilocus genotypes occurring in the populations from the allele frequencies at that locus. These multilocus genotype

300

M.-L. Koljonen and R. Wilmot

distributions are then used to evaluate probabilities of sources for mixture individuals in stock identiﬁcation methods. This advantage of using Hardy– Weinberg equilibrium can also be applied to other Mendelian inherited data such as DNA microsatellite data, but not to maternally inherited mitochondrial DNA data. Allele frequencies are also used to calculate numerous other statistics measuring the amount of genetic differentiation and diversity among populations, that is, genetic distances (Nei, 1987) and mean heterozygosities (Wright, 1969). Methods such as the gene diversity analysis of Chakraborty (1980) and the standardized allele frequency variance among populations, the FST analysis of Wright (1965) or Weir and Cockerham (1984), can be used to measure the signiﬁcance of population differentiation. FST can also be used to assess the probability of successful identiﬁcation of populations in mixed stock ﬁsheries. Simulations with the true baseline data and test analyses are recommended for estimating the potential resolution power of the data set before application to true situations.

IV. MIXED-STOCK ANALYSIS Successful MSA depends on the degree of genetic differentiation among stocks. Calculation of the above population genetic statistics will usually provide a good indicator of whether the degree of differentiation among stocks is sufﬁcient for MSA. Traditional methods of stock identiﬁcation based on maximum-likelihood statistics are not used to identify the stock origin of individual ﬁsh. They can, however, provide fairly reliable estimates of the proportions of stocks contributing to a mixture. The estimation methods include the EM algorithm (Millar, 1987), GIRLSEM (Masuda et al., 1991) and SPAM (Debevec et al., 2000). MSA assumes that allele frequency data are available for all possible major stocks contributing to the mixture (the baseline) and that the frequencies are in Hardy–Weinberg equilibrium. Maximum likelihood estimation (MLE) is based on the idea that, given the multilocus genotypic distributions in c baseline stocks (pi, i = 1, 2, . . . , c, c

Â pi = 1 ),

and the distribution observed in the catch mixture sample, the

i= 1

proportions of baseline stocks composing the stock mixture are best estimated as those for which the observed mixture genotype distribution is most probable. The probability of sampling a particular multilocus genotype h from any mixture c

is l h =

Â pi g hi, where g

hi

is the frequency of that genotype in the baseline stock

i= 1

i. The probability of the sample, or the likelihood function that is maximized with respect to the pi’s, is obtained as the product of the individual probabilities

301

Genetic Analysis: Allozymes

of the N observed genotypes: L = ’ lmh h , where mh is the observed number of h

ﬁsh of the multilocus genotype h in the catch sample. The baseline multilocus genotype frequencies used in the estimation process are calculated from the observed baseline allele frequencies by assuming Hardy–Weinberg genotypic distribution and independence among loci (Pella and Milner, 1987). A new method of MSA based on Bayesian statistics has been developed by Pella and Masuda (2001). According to the Bayesian method, an informative prior for genetic characters of the separate stocks in a mixture is derived from baseline samples. In addition, a neutral, low information prior is used for the stock proportions in the mixture. A Gibbs sampler—the data augmentation algorithm—is used to alternatively generate samples from the posterior distributions of genetic parameters of the baseline stocks and for the stock proportions in the mixture. The posterior distribution incorporates the information about genetic characters in the baseline stocks, including relatedness of stocks, with that in the stock mixture sample to better estimate genotypic composition of the separate stocks. The posterior distribution of stock proportions is then used as an estimator of stock composition. Given the inherent bias of MLE to underestimate predominant components, the Bayesian mode is a sensible estimator with no logical counterpart in the MLE method other than the MLE point estimate. The program also provides posterior probabilities of the source population for each individual in the mixture. These probabilities can be used to assign the individuals to a particular stock. Other programs currently available can also be used for plane individual assignment, for example, GENECLASS (Cornuet et al., 1999), WHICHRUN (Banks and Eichert, 2000) and STRUCTURE (Pritchard et al., 2000). GENECLASS and WHICHRUN assign individuals to a population without regard to the mixture composition other than assuming equal proportions of stocks in the mixture.

V. ADVANTAGES AND LIMITATIONS OF THE ALLOZYME DATA IN MSA Allozyme data have several advantages over other types of genetic data. The cost of analysis is low and a large number of samples can be analyzed in a relatively short time as compared with DNA methods. Laboratory analysis is simple and does not require complicated techniques. Genetic tags are inherited traits and thus the environment does not change their expression. Genotypes are discrete characteristics; interpretation is therefore relatively unambiguous and differences can be quantiﬁed. In most cases, loci can also be assumed to be independent (not correlated) variables, which simpliﬁes mathematical treatment. Genetic differences are relatively stable over time and from generation to generation (Waples, 1990).

302

M.-L. Koljonen and R. Wilmot

The advantages of genetic tags over external tags are the following: no costs associated with actual tagging, no loss of tags, and no bias due to possible effects of the external tags. Moreover, all ﬁsh are tagged for life, which enables studies to be conducted on ﬁshes that cannot be tagged by other methods, for example, wild ﬁsh in remote areas or newly hatched ﬁsh in releasing programs (Mathias et al., 1992). In genetic stock identiﬁcation, the time and place of sampling can be chosen more freely and precisely than with external tagging as they are not dependent on preceding tag and release programs. Further, genetic stock identiﬁcation does not depend on ﬁshermen for the return of external tags. The most obvious limitation to the wider use of allozyme analysis in stock identiﬁcation is that sufﬁcient reproductive isolation must occur among contributing stocks for temporally stable genetic differences in allele frequencies to arise. Statistically signiﬁcant differences in allele frequencies often occur, but quantitatively they may be too small for sufﬁcient stock separation. For species formed of several clearly different reproduction units, such as salmonids, allozyme analysis will usually provide a useful tool. Even a relatively small amount of gene ﬂow may dilute the differences, especially in the loci presumed to be selectively rather neutral. Where allozyme information alone is not sufﬁcient for accurate stock identiﬁcation, additional information may be necessary, for example, parasite infection rates and/or scale pattern characteristics (Rutherford et al., 1988; Wood et al., 1989; Wood et al., 1988; Wilmot et al., 1999; Pella et al., 1998) and smolt age distributions (Koljonen and Pella, 1997). For cases in which identiﬁcation of individual stocks has failed, groups of genetically similar stocks have been estimated accurately enough for management purposes (Seeb and Crane, 1999). Intentional changes in allele frequencies caused by controlled matings in hatchery production can create an identiﬁable population. These changes need not be dramatic, but a relatively small increase in the frequency of a rare allele may improve separation of the stock of interest (Seeb et al., 1986, 1990).

VI. FACTORS AFFECTING THE RELIABILITY OF ALLOZYME-BASED MSA Special interest should be focused on estimation reliability. The method based on maximum likelihood will always produce an estimate, but the usefulness of these estimates depends on the criteria set for their accuracy and precision. In MLE, errors in composition estimates may derive from the mixture sample or the baseline sample, or both. When the sample sizes are large, these two errors are additive in theory (Pella and Robertson 1979; Millar, 1987). The bias is greatest when genetically similar stocks differ greatly in abundance (Millar, 1987).

Genetic Analysis: Allozymes

303

The variation resulting from mixture sampling depends on the size of the sample and the stock composition of the sample. According to Wood et al. (1987), the critical sample size is about 40 ﬁsh per stock for ordinary allozyme data; below that, the reliability of the estimates is greatly reduced. In general, it is known that with small mixture samples the contributions of dominant stocks are underestimated and those of minor stocks overestimated. Baseline sampling tends to be more problematic than mixture sampling. The amount of variation and bias in the estimates of stock proportions may be affected by several factors: coverage of the baseline data, conformity to Hardy–Weinberg expectations, temporal variation in allele frequencies, sample sizes, number of loci, genetic differentiation among stocks, and strategies for pooling the individual stock proportion estimates. An important issue is whether all the contributing stocks are represented in the baseline data (Smouse et al., 1990). Errors in baseline sampling can also arise if individual stocks are composed of several breeding units, either natural or hatchery ones, or if variation in the allele frequencies of different year-classes is marked. This may cause deviations from the Hardy–Weinberg distributions in the stocks and may also mean that the allele frequencies observed are not representative of the stock as a whole. Such problems are most likely to occur in small natural stocks or in hatchery stocks with changing brood ﬁsh and breeding ranges. Waples (1990) analyzed the implications of temporal allele frequency changes for MSA and observed that, in general, temporal changes should be taken into account as potential sources of error. However, any effects of changes caused by genetic drift can be compensated for to a marked extent by collecting baseline data over several years. The importance of repeated sampling depends on the life history of the species concerned and on the degree of overlapping in the year-classes. The Bayes program captures information about allele frequencies from mixture samples. Therefore, if annual mixture sampling from ﬁsheries capturing the same source populations were analyzed sequentially by date of sampling, or even pooled for the season, the Bayesian method could be used to detect and estimate long-term changes in allele frequencies among contributing stocks. This would eliminate or reduce the baseline sampling needs. According to Wood et al. (1987), a critical value for the baseline sample size was about 40 ﬁsh per stock. The number of ﬁsh needed per stock, however, also depends on the level of genetic differentiation between the stocks. Both the number of loci and the level of differentiation between the stocks affect the discriminatory power of the baseline data. The number of loci that have been analyzed in studies of Paciﬁc salmonids has been relatively high (e.g. Shaklee et al., 1990a, 14–22 loci; Brodziak et al., 1992, 15 loci). In the event of small separation, performance of the analysis can be improved by increasing the number of

304

M.-L. Koljonen and R. Wilmot

loci or the size of the mixture sample. In some cases, however, only pooled estimates of stock proportions are obtained.

VII. EXAMPLES OF APPLICATION OF ALLOZYMES TO MSA Genetic results have been used for Paciﬁc salmonids both in season and postseason to determine ﬁshery openings and closures with a view to providing harvest beneﬁts or meeting conservation needs, to address catch allocation and equity issues among user groups and between countries, to provide data for in-season run-size updates, and to investigate migration patterns and timing (Shaklee et al., 1999). The Columbia River Chinook salmon (Oncorhynchus tshawytscha) gill-net ﬁshery has been regulated on the basis of allozyme information. Mixed-stock analysis is based on 27-stock, 22-loci allozyme baseline, and MLE (Shaklee 1991; Shaklee et al., 1999). The Chinook ﬁshery in the lower river was limited to ensure that adequate numbers of upper river stocks would return to their spawning grounds. The harvest rate in the lower river was set at 4.1% of the total upriver run. Since 1990, the results of in-season allozyme analyses together with daily counts of the total ﬁsh harvest have been the primary data used to manage the ﬁshery. Estimates of the cumulative impacts on upper river stocks are determined, and when the harvest rate approaches the maximum acceptable impact level (4.1%), based on preseason run-size predictions, the ﬁshery is closed. The pink salmon ﬁshery (Oncorhynchus gorbuscha) in the Fraser River provides another example of the systematic use of allozyme analysis in ﬁsheries management (Shaklee et al., 1999). In 1987, the Paciﬁc Salmon Commission began to use genetic stock mixture analysis to identify Fraser River pink salmon in catches from Alaska southward. In-season tissue samples were generally analyzed within 3 days, thus providing ﬁshery managers with current information on Fraser River pink salmon contributions to important ﬁsheries and also with a tool for the implementation of management measures to meet in-season catch and escapement requirements. Postseason estimates of actual Fraser River pink salmon catches by country and within user groups in the United States were compared with the allocated catches, and any shortfalls or overages in allocation by country or user group were to be remedied in future catch paybacks. More examples of the application of allozyme variation to MSA of Paciﬁc salmon are given in Table 15-1.

TABLE 15-1.

Examples of Mixed-Stock Analysis Studies on Paciﬁc Salmon

Species Chum salmon (O. keta)

Chinook salmon (O. tshawytscha)

Pink salmon (O. gorbuscha)

305

Sockeye salmon (O. nerka)

Region

No of loci/ populations

Eastern Bering Sea

20/77

Aleutian Islands

20/77

North Paciﬁc Ocean

20/77

Central Bering Sea

16/273

Yukon River (U.S./Canada) Fraser River (Canada)

19/31

Vancouver Island (Canada) Yukon River (U.S./Canada) Columbia River (U.S.) North Paciﬁc Ocean

22/21 10/14 20/77

Washington State Coast Fraser River (Canada)

14/23

Central Bering Sea

19/95

North Paciﬁc Ocean

14/165

Southcentral Alaska

6/13

Results Determined region of origin of chum salmon by catch in Pollock ﬁshery Determined region of origin of chum salmon by catch in commercial sockeye salmon ﬁshery Region of origin of chum salmon harvest by illegal high seas drift netters Region of origin of chum salmon seized from Russian trawler in U.S. waters Determined catch of Canadian origin chum salmon in U.S. ﬁshery at mouth of Yukon River Estimated abundance of Fraser River chum salmon in test ﬁshery in Johnstone Strait Estimated origin of chum salmon in mixed-stock ﬁshery near Vancouver Island Determined catch of Canadian origin Chinook salmon in U.S. ﬁshery at mouth of Yukon River Determined in-season origins of Chinook salmon in mixed-stock ﬁshery at mouth of Columbia River Region of origin of Chinook salmon harvest by illegal high seas drift netters Determined stock composition of Chinook salmon caught in coastal troll ﬁshery Determined in-season origins of pink salmon harvested in U.S. and Canadian mixed-stock ﬁsheries Region of origin of pink salmon seized from Russian trawler in U.S. waters Region of origin of sockeye salmon harvest by illegal high seas drift netters Determined origin of sockeye salmon in mixed-stock ﬁshery in Cook Inlet, Alaska

Reference Wilmot et al., 2000 Crane and Seeb, 2000 Wilmot et al., 1999 Wilmot et al., 2000 Kondzela et al., 2002 Wilmot et al., 1992 Beacham et al., 1985 Fournier et al., 1984 Wilmot et al., 1992 Shaklee et al., 1999 Wilmot et al., 1999 Wilmot et al., 2000 Milner et al., 1983 Shaklee et al., 1999 Kondzela et al., 2002 Wilmot et al., 2000 Grant et al., 1980

306

M.-L. Koljonen and R. Wilmot

REFERENCES Aebersold, P. B., Winans, G. A., Teel, D. J., Milner, G. B., and Utter, F. M. 1987. Manual for starch gel electrophoresis: A method for the detection of genetic variation. NOAA Technical Report NMFS 61, Department of Commerce, National Oceanic and Atmospheric Administration, National Marine Fisheries Service. 19 pp. Banks, M. A. and Eichert, W. 2000. WHICHRUN (Version 3.2): a computer program for population assignment of individuals based on multilocus genotype data. J. Hered. 91: 87–89. Beacham, T. D., Withler, R. E., and Gould, A. P. 1985. Biochemical genetic stock identiﬁcation of chum salmon (Oncorhynchus keta) in southern British Columbia. Can. J. Fish. Aquat. Sci. 42: 437–448. Begg, G. A., Friedland, K. D., and Pearce, J. B. 1999. Stock identiﬁcation—its role in stock assessment and ﬁsheries management: a selection of papers presented at a symposium of the 128th annual meeting of American Fisheries Society in Hartford, Connecticut, 23–27 August 1998. Fish. Res. 43 (1–3), 249 pp. Brodziak, J., Bentley, B., Bartley, D., Gall, G. A. E., Gomulkiewicz, R., and Mangel, M. 1992. Test of genetic stock identiﬁcation using coded wire tagged ﬁsh. Can. J. Fish. Aquat. Sci. 49: 1507–1517. Chakraborty, R. 1980. Gene-diversity analysis in nested subdivided populations. Genetics 96: 721–726. Cornuet, J.-M., Piry, S., Luikart, G., Estoup, A., and Solignac, M. 1999. New methods employing multilocus genotypes to select or exclude populations as origins of individuals. Genetics 153: 1989–2000. Crane, P. A. and Seeb, L. W. 2000. Genetic analysis of chum salmon harvested in the South Peninsula, post June ﬁshery, 1996–1997. Alaska Department of Fish and Game, Anchorage, Alaska. Regional Information Report No. 5J00–05. Debevec, E. M., Gates, R. B., Masuda, M., Pella, J., Reynolds, J., and Seeb, L. W. 2000. SPAM (Version 3.2): statistics program for analyzing mixtures. J. Hered. 91: 509–510. Fournier, D. A., Beacham, T. D., Riddell, B. E., and Busack, C. A. 1984. Estimating stock composition in mixed stock ﬁsheries using morphometric, meristic, and electrophoretic characteristics. Can. J. Fish. Aquat. Sci. 41: 400–408. Grant, W. S., Milner, G. B., Krasnowski, P., and Utter, F. M. 1980. Use of biochemical genetic variants for identiﬁcation of sockeye salmon (Oncorhynchus nerka) stocks in Cook Inlet, Alaska. Can. J. Fish. Aquat. Sci. 37: 1236–1247. Harris, H. and Hopkinson, D. A. 1976. Handbook of Enzyme Electrophoresis in Human Genetics. American Elsevier, New York. Koljonen, M.-L. 1995. Distinguishing between local and migrating Atlantic salmon (Salmo salar L.) stocks by genetic stock composition analysis. Can. J. Fish. Aquat. Sci. 52: 665–674. Koljonen, M.-L. and Pella, J. J. 1997. The advantage of using smolt age with allozymes for assessing wild stock contributions to Atlantic salmon catches in the Baltic Sea. ICES J. Mar. Sci. 54: 1015–1030. Kondzela, C. M., Hawkins, S. C., Guthrie III, M., and Wilmot, R. L. 2002. Origins of salmon seized from the F/V Petropavlovsk. (NPAFC Doc. 598) Auke Bay Fisheries Laboratory, Alaska Fisheries Science Center, NMFS, NOAA, 11305 Glacier Highway, Juneau, AK 99801–8626. 10 pp. Kornﬁeld, I., Beland, K. F., Moring, J. R., and Kircheis, F. W. 1981. Genetic similarity among endemic Arctic char (Salvelinus alpinus) and implications for their management. Can. J. Fish. Aquat. Sci. 38: 32–39. Krueger, C. C. and May, B. 1987. Stock identiﬁcation of naturalized brown trout in Lake Superior tributaries: differentiation based on allozyme data. Trans. Am. Fish. Soc. 116: 785–794. Krueger, C. C., Wilmot, R. L., and Everett, R. J. 1999. Stock origins of Dolly Varden collected from Beaufort Sea coastal sites of Arctic Alaska and Canada. Trans. Am. Fish. Soc. 128: 49–57.

Genetic Analysis: Allozymes

307

Marsden, J. E., Krueger, C. C., and May, B. 1989. Identiﬁcation of parental origins of naturally produced lake trout in Lake Ontario: application of mixed-stock analysis to a second generation. N. Am. J. Fish. Manag. 9: 257–268. Masuda, M., Nelson, S., and Pella, J. 1991. The computer programs for computing conditional maximum likelihood estimates of stock composition from discrete characters. USA-DOC-NOAANMFS, Auke Bay Laboratory, Alaska Fisheries Science Center, Juneau, AK. Mathias, J. A., Franzin, W. G., Craig, J. F., Babaluk, J. A., and Flannagan, J. F. 1992. Evaluation of stocking walleye fry to enhance a commercial ﬁshery in a large, Canadian prairie lake. N. Am. J. Fish. Manag. 12: 299–306. May, B., Stoneking, M., and Wright Jr., J. E. 1980. Joint segregation of biochemical loci in Salmonidae: II. Linkage associations from a hybridized salvelinus genome (S. namaycush x S. fontinalis). Genetics 95: 707–726. Millar, R. B. 1987. Maximum likelihood estimation of mixed stock ﬁshery composition. Can. J. Fish. Aquat. Sci. 44: 583–590. Milner, G. B., Teel, D. J., and Utter, F. M. 1983. Genetic stock identiﬁcation study. National Oceanic and Atmospheric Administration, Northwest and Alaska Fisheries Center, Seattle, WA. 33 pp. (and Appendix). Milner, G. B., Teel, D. J., Utter, F. M., and Winans, G. A. 1985. A genetic method of stock identiﬁcation in mixed populations of Paciﬁc salmon, Oncorhynchus spp. Mar. Fish. Rev. 47: 1–8. Nei, M. 1987. Molecular Evolutionary Genetics. Columbia University Press, New York. Pella, J. J. and Masuda, M. 2001. Bayesian methods for analysis of stock mixtures from genetic characters. Fish. Bull. 99: 151–167. Pella, J. J. and Milner, G. B. 1987. Use of genetic marks in stock composition analysis. In N. Ryman and F. Utter (eds.), Population Genetics and Fishery Management. Washington Sea Grant Program, University of Washington Press, Seattle, pp. 247–276. Pella, J. J. and Robertson, T. L. 1979. Assessment of composition of stock mixtures. Fish. Bull. 77: 387–398. Pella, J., Masuda, M., Guthrie, C., Kondzela, C., Gharrett, A., Moles, A., and Winans, G. 1998. Stock composition of some sockeye salmon, Oncorhynchus nerka, catches in southeast Alaska, based on incidence of allozyme variants, freshwater ages, and a brain-tissue parasite. NOAA Tech. Rpt NMFS 132. Auke Bay Laboratory, Alaska Fisheries Science Center, Juneau, AK. 23 pp. Perkins, D. L. and Krueger, C. C. 1995. Dynamics of reproduction by hatchery-origin lake trout (Salvelinus namaycush) at Stony Island Reef, Lake Ontario. J. Great Lakes Res. 21 (suppl. 1): 400–417. Pritchard J. K., Stephens, M., and Donnelly, P. 2000. Inference of population structure using multilocus genotype data. Genetics 155: 945–959. Rutherford, D. T., Wood, C. C., Jantz, A. L., and Southgate, D. R. 1994. Biological characteristics of Nass River sockeye salmon (Oncorhynchus nerka) and their utility for stock composition analysis of test ﬁshery samples. Can. Tech. Rep. Fish. Aquat. Sci. 1988. 72 pp. Seeb, J. E., Seeb, L. W., and Utter, F. M. 1986. Use of genetic marks to assess stock dynamics and management programs for chum salmon. Trans. Am. Fish. Soc. 115: 448–454. Seeb, L. W., Seeb, J. E., Allen, R. L., and Hershberger, W. K. 1990. Evaluation of adults returns of genetically marked chum salmon, with suggested future applications. Am. Fish. Soc. Symp. 7: 418–425. Seeb, L. W. and Crane, P. A. 1999. Allozymes and mitochondrial DNA discriminate Asian and North American populations of chum salmon in mixed-stock ﬁsheries along the south coast of the Alaska Peninsula. Trans. Am. Fish. Soc. 128: 88–103. Shaklee, J. B. 1991. Simulations and other analysis of the 1991 Columbia River spring chinook GSI baseline. WDF Tech. Rpt. 115. Washington Department of Fisheries, Olympia, WA. 40 pp.

308

M.-L. Koljonen and R. Wilmot

Shaklee, J. B. and Phelps, S. R. 1990. Operation of a large scale, multiagency genetic stock identiﬁcation program. Am. Fish. Soc. Symp. 7: 817–830. Shaklee, J. B., Kepes, K. L., and Whitt, G. S. 1973. Specialized lactate dehydrogenase isozymes: the molecular and genetic basis for the unique eye and liver LDHs of teleost ﬁshes. J. Exp. Zool. 185: 217–240. Shaklee, J. B., Busack, C., Marchall, A., Miller, M., and Phelps, S. R. 1990a. The electrophoretic analysis of mixed-stock ﬁsheries of Paciﬁc salmon. In Z. I. Ogita and C. L. Markert (eds.), Isozymes: Structure, Function, and Use in Biology and Medicine. Progress in Clinical and Biological Research, Vol. 344. Wiley-Liss, Inc. New York, pp. 235–265. Shaklee, J., Allendorf, F. W., Morizot, D. C., and Whitt, G. S. 1990b. Gene nomenclature for proteincoding loci in ﬁsh. Trans. Am. Fish. Soc. 119: 2–15. Shaklee, J. B., Beacham, T. D., Seeb, L., and White, B. A. 1999. Managing ﬁsheries using genetic data: case studies from four species of Paciﬁc salmon. Fish. Res. 43: 45–78. Shaw, C. R. and Prasad, R. 1970. Starch gel electrophoresis of enzymes—a compilation of recipes. Bioch. Gen. 4: 297–320. Siciliano, M. R. and Shaw, C. R. 1976. Separation and visualization of enzyme gels. In I. Smith (ed.), Cromatographic and Electrophoretic Techniques. Zone Electrophoresis. Interscience, Wiley, New York, Vol. 2, pp. 217–238. Smouse, P. E., Waples, R. S., and Tworek, J. A. 1990. A genetic mixture analysis for use with incomplete source population data. Can. J. Fish. Aquat. Sci. 47: 620–634. Utter, F. M. 1991. Biochemical genetics and ﬁshery management: an historical perspective. J. Fish Biol. 39: 1–20. Utter, F. M., Hodgins, H. O., and Allendorf, F. W. 1974. Biochemical genetic studies on ﬁshes: potentials and limitations. In D. Maklins (ed.), Biochemical and Biophysical Perspective in Marine Biology. Academic Press, San Francisco, Vol. 1, pp. 213–237. Utter, F., Aebersold, P., and Winans, G. 1987. Interpreting genetic variation detected by electrophoresis. In N. Ryman and F. Utter (eds,), Population Genetics and Fishery Management. Washington Sea Grant Program, University of Washington Press, Seattle, pp. 21–45. Waples, R. S. 1990. Temporal changes of allele frequency in Paciﬁc salmon: implications for mixedstock ﬁshery analysis. Can. J. Fish. Aquat. Sci. 47: 968–976. Weir, B. C. and Cockerham, C. C. 1984. Estimating F-statistics for the analysis of population structure. Evolution 38: 1350–1370. Wilmot, R. L., Everett, R. J., Spearman, W. J., and Baccus, R. 1992. Genetic stock identiﬁcation of Yukon River chum and chinook salmon—1987 to 1990. Prog Rep. U.S. Fish and Wildlife Service, Anchorage, AK. 132 pp. Wilmot, R. L., Kondzela, C. M., Guthrie, C. M., and Masuda, M. M. 1998. Genetic stock identiﬁcation of chum salmon harvested incidentally in the 1994 and 1995 Bering Sea trawl ﬁshery. N. Pac. Anadr. Fish Comm. Bull. 1: 285–299. Wilmot, R. L., Kondzela, C. M., Guthrie III, C. M., Moles, A., Martinson, E., and Helle, J. H. 1999. Origins of sockeye and chum salmon seized from the Chinese vessel Ying Fa. (NPAFC Doc.) Auke Bay Fisheries Laboratory, Alaska Fisheries Science Center, NMFS, NOAA, 11305 Glacier Highway, Juneau, AK 99801–8626. 20 pp. Wilmot, R. L., Kondzela, C. M., Guthrie III, C. M., Moles, A., Pella, J. J., and Masuda, M. 2000. Origins of salmon seized from the F/V Arctic Wind. (NPAFC Doc.) Auke Bay Fisheries Laboratory, Alaska Fisheries Science Center, NMFS, NOAA, 11305 Glacier Highway, Juneau, AK 99801–8626. 18 pp. Wood, C. C., McKinnel, S, Mulligan, T. J., and Fournier, D. A. 1987. Stock identiﬁcation with the maximum-likelihood mixture model: sensitivity analysis and application to complex problems. Can. J. Fish. Aquat. Sci. 44: 866–881.

Genetic Analysis: Allozymes

309

Wood, C. C., Oliver, G. T., and Rutherford, D. T. 1988. Comparison of several biological markers used in stock identiﬁcation of sockeye salmon (Oncorhynchus nerka) in northern British Columbia and southeast Alaska. Can. Tech. Rep. Fish. Aquat. Sci. 1624. 49 pp. Wood, C. C., Rutherford, D. T., and McKinnell, S. 1989. Identiﬁcation of sockeye salmon (Oncorhynchus nerka) stocks in mixed-stock ﬁsheries in British Columbia and southeast Alaska using biological markers. Can. J. Fish. Aquat. Sci. 46: 2108–2120. Wright, S. 1965. The interpretation of population structure by F-statistics with special regard to system of mating. Evolution 19: 395–420. Wright, S. 1969. Evolution and the Genetic of Populations, Vol. 2: The Theory of Gene Frequencies. University of Chicago Press, Chicago.

This Page Intentionally Left Blank

CHAPTER

16

Mitochondrial DNA ANTONIOS MAGOULAS Hellenic Centre for Marine Research, Institute of Marine Biology and Genetics, Heraklion, Crete, Greece

I. Introduction A. General Features of Mitochondrial DNA B. mtDNA as a Tool for Intraspeciﬁc Analyses II. Methods for the Analysis of mtDNA A. Restriction Analysis B. Polymerase Chain Reaction (PCR) Analysis III. Review of mtDNA Studies of Fish Populations IV. Critique References

I. INTRODUCTION

A. GENERAL FEATURES

OF

MITOCHONDRIAL DNA

1. Molecular Cåharacteristics Mitochondrial DNA (mtDNA) is a small, double-stranded circular DNA molecule contained in multiple copies in the mitochondria, which are cytoplasmic organelles found in all eukaryotic cells. There may be up to several thousand copies per cell, depending on the cell type. In higher animals, mtDNA typically is around 16,000 base pairs (bp) long, although there is some length variation. The largest mtDNA molecule found in higher animals is that of scallops (Placopecten magellanicus), which is more than 39,000 bp long (Snyder et al., 1987). Variability in the size can be found not only between species, but also intraspeciﬁcally (see heteroplasmy, section I, A, 3). The animal mitochondrial genome contains 13 genes coding for proteins, two genes coding for ribosomal RNAs (12S and 16S rRNA), 22 genes coding for transfer RNAs (tRNAs), and one noncoding control region (also called D-loop in vertebrates). The coding genes code for enzyme subunits involved in electron Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

311

312

Antonios Magoulas

transport and oxidative phosphorylation. The ribosomal RNAs and the transfer RNAs participate in protein translation on mitochondrial ribosomes. The control region is about 1,000 bp long and contains the origin of mtDNA replication. Mitochondrial gene order differs slightly among vertebrates, and the piscine gene order does not differ from the vertebrate consensus order (Meyer, 1993). Usually the differences found are due to tRNA gene translocations. Mitochondrial DNA has been characterized as an “extreme example of genetic economy” (Attardi, 1985) because there are no introns within the coding genes, no repetitive DNA, and essentially no spacer sequences between genes. 2. Evolution of mtDNA Despite the conservation of the mitochondrial gene content and order over long evolutionary time, the nucleotide sequence of mtDNA evolves rapidly. The divergence of mtDNAs of taxa that shared a common ancestor is believed to be 2% per million years and to remain linear for 8 to 10 million years (Moritz et al., 1987). Some portions of the control region evolve exceptionally rapidly and are very useful for high-resolution analysis of population structure. The slowly evolving protein coding genes like CO I, II, III, and cytochrome b are more suitable for comparisons at the interspeciﬁc or higher level. 3. Transmission Genetics Generally, somatic and germ cells of an individual animal contain a single type of mtDNA, a state known as homoplasmy. However, there is a number of reported cases of heteroplasmy, that is, of presence of more than one type of mtDNA in an individual. In most cases of heteroplasmy, the two mtDNA variants differ in their size (length), typically due to tandem duplications in some portion of the molecule, largely in the control region. But there also have been reports of siteheteroplasmy in which the two variants differ in their sequence due to point mutations. Generally, the existence of heteroplasmy does not constitute a serious complication in population genetic analysis. Animal mtDNA is usually considered to be nonrecombining, but even if recombination does occur as evidence has been reported recently (see Rokas et al., 2003 and references therein), usually no novel genotypes will be generated because the two recombining molecules will most probably be the same due to homoplasmy. Moreover, it is almost exclusively maternally inherited. Despite reports of several cases of paternal mtDNA leakage to the progeny (Rokas et al., 2003), it seems that the predominantly maternal inheritance of mtDNA, in combination with the fact that recombination usually does not produce new detectable variants of mtDNA, creates a linear evolutionary history of maternal transmission (matriarchal phylogeny) of mtDNA genotypes.

313

Mitochondrial DNA

B. mtDNA

AS A

TOOL

FOR INTRASPECIFIC

ANALYSES

Mitochondrial DNA has become a very effective molecule to use for the assessment of intraspeciﬁc genetic variation and genealogy. The basic idea behind using mtDNA for stock structure analysis is that if samples of individuals, taken from different areas or from the same area at different times, belong to the same stock, they will contain the same types of mtDNA molecules (mitotypes hereinafter) in similar frequencies. On the contrary, if substantial differences in the mitotype distribution between the samples are encountered, there is strong evidence that these samples belong to different stocks characterized by a certain degree of genetic isolation. Traditionally, the most commonly used method for population-level analysis and stock identiﬁcation was the restriction analysis of mtDNA, that is, analysis by means of the digestion enzymes called restriction endonucleases. This type of analysis assesses the mtDNA sequence variability by sampling small regions, randomly distributed over the molecule, which correspond to the recognition sites of the enzymes. The reliability and resolution of the method depends on the number of restriction enzymes used (the more the enzymes, the higher the reliability and resolution). The invention of polymerase chain reaction (PCR) in the late 1980s allowed for the in vitro ampliﬁcation of portions of the mtDNA molecule, even from minute amounts of total DNA, provided that two or more conserved regions had been identiﬁed and sequenced near the boundaries of the portion to be assayed (see Mullis et al., 1986; Saiki et al., 1988). The ampliﬁed region can subsequently be studied either by restriction analysis or, more importantly, by sequencing. The introduction of PCR ampliﬁcation revolutionized studies of intraspeciﬁc variation and ﬁnds an ever-increasing application. However, one should bear in mind that there is a trade-off between the detailed information that PCR provides for a portion of the mtDNA genome and the less detailed, but more inclusive information provided by restriction analysis across the entire mtDNA genome. Here both methods, restriction analysis and PCR-based analysis, will be presented brieﬂy.

II. METHODS FOR THE ANALYSIS OF mtDNA

A. RESTRICTION ANALYSIS 1. General Description of the Method Restriction endonucleases (RE) are enzymes that recognize speciﬁc short sequences, usually 4 to 6 bp long, on the DNA molecule and cleave the DNA in

314

Antonios Magoulas

a constant position within the recognition sequence. The process of cutting the DNA with the RE is called restriction or digestion. For a circular molecule, such as mtDNA, it is obvious that after digestion with a given RE, it will be cut in as many fragments as there are recognition sites for the enzyme. The sizes of the fragments are determined by the position of these sites on the molecule. The number and size of fragments produced after digestion can be detected after separation of the fragments by gel electrophoresis and appropriate visualization of the fragments. Thus, for each mtDNA molecule (or, equivalently, individual since, as mentioned, each individual usually bears one type of mtDNA) a restriction or digestion proﬁle is produced. The comparison of restrictionfragment proﬁles of several individuals is considered to be representative of the nucleotide differences of their whole mtDNA sequences. This is the main assumption behind restriction analysis, which is also known as RFLP (restriction fragment length polymorphism) analysis. To exemplify this, consider the case shown in Figure 16-1. Let x be an ancestral molecule that bears three recognition sequences of the EcoRI RE. Note that there is also another position, which is only one base removed from being an EcoRI recognition sequence. After digestion with this enzyme three linear fragments of different size will be produced. If these fragments are separated and visualized on a gel, the three fragments of proﬁle A will be produced. This proﬁle is considered as representing mitotype (or haplotype) A. But as mtDNA passes from generation to generation, it replicates repeatedly and each replication has a certain (small) probability of error, or mutation. So, it is possible that a molecule y, bearing a single base substitution in one of the recognition sequences, will arise and be ﬁxed in some individual in the future. This recognition site will not be recognized and cleaved by the enzyme (an event known as site loss), and proﬁle B of molecule y will contain only two bands. Of those two bands, the larger corresponds to the same fragment as the largest band of proﬁle A (we say that this is a shared band), and the smaller represents the fragment that resulted from the uniﬁcation of the two fragments of molecule x that were separated by the site that was lost. Thus the proﬁles/mitotypes A and B have one shared fragment (band) and three unshared fragments. On the other hand, it is possible that in another evolutionary lineage the site that was one base away from being a recognition sequence in molecule x may undergo the “right” mutation (C Æ G) to turn into an EcoRI recognition sequence in molecule z (a site gain). The resulting proﬁle/mitotype C will have two shared bands with A and no shared band with B. In this way, even with this kind of crude analysis, the comparison of the digestion proﬁles would provide evidence that mitotype A is more closely related to each of mitotypes B and C than those two mitotypes are related to each other. Thus, a ﬁrst phylogenetic network of the mitotypes can be constructed which gives to mitotype A an intermediate position between B and C.

GAA TT A

Site loss

y GAATTC TC T A A G

x

G A A T T C AC TA T C

z

Site gain

C TT AA

B

G

A

C Large size

C

A

B

Small size

FIGURE 16-1. Evolutionary changes of restriction sites and the resulting proﬁles after electrophoresis of the digests. 315

316

Antonios Magoulas

The accuracy and reliability of the method increases if several (usually from 5 to 20) REs are used. The mitotypes are then determined by taking into account the digestion proﬁles produced by all the REs used and are designated usually by a multiletter code in which each letter is the proﬁle of each separate RE used, with a speciﬁed order. With some extra effort (which is considerably time consuming) the position of the restriction sites on the molecule can be determined and thus a restriction map can be produced for each mitotype. The site changes interconnecting the mitotypes can then be determined by comparison of their restriction maps. Once the site changes have been determined, the nucleotide divergence between the mitotypes can be estimated on the basis of the number of their shared/not shared sites (Nei, 1987). If information on the site changes is available, a parsimonious phylogenetic network, that is, a network that minimizes the evolutionary changes (site gains or losses) through which the mitotypes are connected, can be constructed. 2. Raw Data Acquisition The procedures for the acquisition of raw data are schematically presented in Figure 16-2. The ﬁrst step is the extraction of DNA, usually from tissues such as ovary, liver, brain, and muscle. Depending on the kind of tissue available and on what kind of assays are to be conducted subsequently, mtDNA can be isolated either in highly puriﬁed condition (enriched mtDNA extraction) or in crude form (together with nuclear DNA, total DNA extraction). The extraction of enriched mtDNA is much more tedious and time consuming than total DNA extraction and requires fresh tissue. On the other hand, the subsequent analysis of enriched mtDNA is much easier than that of total DNA, the analysis of which involves the laborious method of Southern blot analysis and relies heavily on the availability of extra pure mtDNA for use as a probe for the hybridization step. It should be noted that the development of the PCR-based methods in combination with recent technological advances (e.g., automated high throughput genetic analyzers, which use ﬂuorescent dyes) tend to restrict the application of methods like those described in this section. However, they are brieﬂy explained here, also for historical reasons. a. Method Based on Enriched mtDNA Extraction and Direct Visualization of the Proﬁles mtDNA isolation. There are various protocols for the isolation of enriched mtDNA. One of the most widely applied is that of Lansman et al. (1981), which uses a CsCl density gradient centrifugation to separate the closed-circular mtDNA from the lineal nuclear DNA. The puriﬁed mtDNA, in addition to being used for

Mitochondrial DNA

317

FIGURE 16-2. Outline of restriction analysis of mtDNA. Methods for data collection.

individual scoring as described in the following paragraphs, can also be used as a probe in the Southern blot analysis of total DNA extracts (see later). A collection of several protocols for isolating enriched mtDNA can be found in White and Densmore (1992). Restriction digestion and electrophoresis. Several hundred REs are commercially available. The digestion conditions are usually speciﬁed by the manufacturer. After digestion, the produced fragments are separated by molecular weight by gel electrophoresis. At neutral pH, DNA is negatively charged and so, if placed in an electric ﬁeld, migrates toward the anode. The digests are loaded into media (agarose or polyacrylamide) that form a dense matrix, through the pores of which

318

Antonios Magoulas

smaller fragments migrate faster than larger fragments (for details on gel electrophoresis see Sambrook et al., 1989). The method of visualization of the digestion proﬁles on the gels is dictated by the purity and amount of mtDNA available. If there is enough DNA the fragments can be identiﬁed by direct staining with ethidium bromide. The minimum amount of DNA in a band detectable by this method is 2 ng (Dowling et al., 1996). If there is a limited amount of DNA, visualization is achieved by radioactively labeling the restriction fragments prior to electrophoresis and subsequent exposure of the gel to X-ray ﬁlm. Development of the ﬁlm (autoradiography) reveals the digestion proﬁle (see Sambrook et al., 1989). Usually for the determination of the size of the restriction fragments a molecular weight marker, that is, a pool of DNA fragments of known size, is run together with the digest under assay onto the gel. The sizes of the restriction fragments are determined by comparison of their migration on the gel with that of the marker fragments or “size standards.” b. Method Based on Total DNA Extraction and Southern Blotting Total DNA isolation. Fresh or frozen tissue is homogenized and all the cellular membranes lysed by means of a detergent, such as SDS (sodium dodecyle sulfate). Usually proteinase k is added concurrently to disintegrate the proteins. DNA is then separated from impurities by phenol/chloroform extractions and precipitated with ethanol. Several protocols for total DNA extraction are used nowadays (e.g., see Sambrook et al., 1989). Restriction digestion and electrophoretic separation. Total DNA is restricted and electrophoresed much the same way as in the case of enriched mtDNA. Again, double digestions should be performed if a site map is to be produced. Transfer and hybridization. After electrophoresis, the DNA on the gel is denatured by dipping the gel in a high pH solution, and then transferred (blotted) to nitrocellulose or nylon membranes. During the subsequent hybridization step, the membrane is incubated, under speciﬁed conditions, with a solution containing a radioactively or otherwise labeled probe. The probe is pure mtDNA from the same or a closely related species, which was obtained by cloning the molecule (or a portion of it), or by PCR ampliﬁcation, or by enriched mtDNA extraction. In the latter case, the mtDNA can be extracted from pooled tissues coming from several individuals if the animals under assay are small. The probe under the appropriate stringency will anneal only to its homologous strands on the membrane, that is, only to the mtDNA strands. If a radioactively labeled probe has been used, autoradiography will reveal the position of the DNA fragments on the original gels; if a nonradioactive tag (such as digoxigenin) has been used, the detection of the fragments is usually performed with an immune-enzymatic assay, according to the instructions of the manufacturers.

Mitochondrial DNA

319

3. Data Processing As shown in Figure 16-3, once the raw data have been collected, the ﬁrst thing to be done is to assign to each animal a composite pattern or mitotype based on the restriction pattern that it demonstrated for each RE used. There are two main general directions in which further analysis of data may proceed. The ﬁrst involves the geographic patterning of the mitotypes. The second takes also into account the evolutionary relationships between the mitotypes themselves (phylogeographic approach). These types of analysis can be combined to help obtain a better understanding of present population structure and of the past or present forces that may have shaped it.

FIGURE 16-3. Outline of restriction analysis of mtDNA. Processing of data.

320

Antonios Magoulas

a. Analysis of Geographic Partitioning of Mitotypes The ﬁrst type of analysis compares the mitotype frequencies in the different samples and estimates the diversity within and between the samples. mtDNA is thought to be one of the best genetic markers for the study of population structure, gene ﬂow, hybridization zones, and other questions at the population level. For stock assessment applications, the mitotype frequencies of two or more samples are compared. If there are statistically signiﬁcant differences between them, the samples are considered to belong to different stocks that exhibit a certain degree of genetic isolation. Statistical tests usually employed are the X2 test and the G-test, which has the advantage of hierarchical partitioning of heterogeneity within the samples and among samples or among groups of samples. Other conventional parameters of population genetics analysis can also be estimated. The index of mitotype diversity (h) corresponds to the gene diversity index of Nei (1987, eq. 8.5) and gives the probability that two different animals drawn at random from the population will have the same mitotype. Nei’s genetic distances (Nei, 1972) between pairs of populations can be estimated from their mitotype frequencies. The populations can then be clustered based on a distance matrix using any of the conventional clustering methods (UPGMA, Neighbour Joining, Fitch-Margoliash, etc.; see Swofford and Olsen, 1996, for details). There are several programs for population genetic analysis that can be used for assessing genetic variation and heterogeneity, such as GENEPOP (Raymond and Rousset, 1995) and Arlequin (Snyder et al., 2000). b. Phylogeographic Analysis In many cases, it is important to examine the evolutionary relationships and divergence among mitotypes because they may provide valuable information on population structure, demography, and history. This has led to the introduction of a new scientiﬁc discipline, phylogeography, which is the study of principles and processes governing the geographic distribution of genealogical lineages (Avise, 1994; Avise, 2000). To mention an example of the matters dealt with by the phylogeographic approach, the existence in a population of two groups of mitotypes that are separated from each other by high divergence (genetic gap), yet characterized by genetic similarity within each group, most likely indicates that the two groups evolved in populations that were isolated in the past. There are two main approaches for the assessment of phylogenetic relationships: parsimony and analysis based on distance matrices. For a detailed description of the methods used for the construction of phylogenetic trees, see Swofford and Olsen (1996).

Mitochondrial DNA

321

B. POLYMERASE CHAIN REACTION (PCR) ANALYSIS 1. General Description of the Method PCR uses the thermostable enzyme Taq polymerase to replicate a stretch of the DNA molecule, starting from minute amounts of DNA samples, even in crude form, provided that the appropriate primers are available. The primers are synthetic single-stranded oligonucleotides (around 20 bp in length) that are complementary to the regions ﬂanking the target segment. During each cycle of PCR the number of copies of the target segment is doubled. Advantages of PCR have rendered it the most popular tool for sequence analysis today: it is a very rapid method permitting the ampliﬁcation of DNAs from several hundreds of individuals within a day or so; it is much faster and cheaper than conventional cloning techniques; it is a robust method that can utilize tiny amounts of tissue; and, the tissue can be frozen, dried, or preserved in ethanol or formalin. The crucial step of the PCR method is the availability of the appropriate primers. The primers can be designed only if the sequences ﬂanking the target sequence in the species under consideration are known, so it would seem at ﬁrst glance that a sequencing step would be necessary before any PCR-based assay could be started. Fortunately, the conservatism of the mitochondrial gene order and of certain portions of the molecule provide the opportunity to circumvent this problem. All we need is to ﬁnd evolutionary conserved sequences bracketing evolutionary variable regions. For example, the tRNA and rRNA genes contain sequences that are sufﬁciently conserved to design primers, which can be used in a wide spectrum of animal species. There are already several such universal primers available, which were deﬁned by comparison of the mtDNA sequence of divergent species, such as human, frog, and ﬂy (Kocher et al., 1989). Another important step in PCR-based analysis is to choose the appropriate region of mtDNA to be ampliﬁed. For population structure analysis the control region is usually preferable because it is the most variable part of the mitochondrial genome and is expected to exhibit substantial polymorphism intraspeciﬁcally. However, it has been shown that even conservative protein-coding genes like those for cytochrome b tend to show intraspeciﬁc variation and can be used to identify ﬁsh stocks (Meyer, 1993, and references therein). 2. Raw Data Acquisition PCR ampliﬁcation is achieved usually by using crude preparations of total DNA as a source of template DNA. However, even simple homogenates of the tissues can sometimes be used efﬁciently. Once double-stranded PCR products have been produced from a number of specimens, there are two main alternatives for further analysis: Either the products will be analyzed using restriction analysis or

322

Antonios Magoulas

the products may be sequenced to provide the raw data for analysis. The ﬁrst analysis is much easier and cheaper, but the second provides the highest possible resolution. a. Restriction Analysis of PCR Products This analysis is performed much the same way described in the section on restriction analysis. Usually, in a preliminary stage several REs (preferably 4-cutters) are tested for the purpose of choosing the enzymes that can detect different mitotypes. This identiﬁcation is relatively difﬁcult because the ampliﬁed DNA will usually be less than 2,000 bp in length, around eight times smaller than the entire mtDNA, which is analyzed through traditional restriction analysis. Nevertheless, data obtained from PCR ampliﬁcation of different portions of the mtDNA molecule can be combined for restriction analysis. Once such enzymes have been identiﬁed, they can be used for the analysis of a large number of animals. After digestion and gel electrophoresis, the restriction proﬁles can be detected by direct ethidium bromide staining because the PCR product contains a large amount of mtDNA and minute amounts of impurities. b. Sequencing of the PCR Product The most commonly used method for DNA sequencing is that of the dideoxy chain-terminating technique, but there are several alternatives for sequencing a PCR product. Information about these techniques can be found in Sambrook et al. (1989). 3. Data Processing The data from the restriction analysis of the PCR products can be used to address population-level questions such as the existence of polymorphism and the geographic distribution of the mitotypes in the samples. Their usefulness for phylogenetic inferences depends on their size because it is possible that a very small (<500 bp) product will not provide a sufﬁcient number of informative restriction sites. Sequence data are extremely useful not only for the detection of differences in the geographic distribution of mitotypes and their genealogy, but also because they provide invaluable information on the forces that govern the evolution of mtDNA. As in the case of restriction analysis, both parsimony and distance-based phylogenetic analysis can be performed. For parsimony analysis, the nucleotide positions are usually the characters, which can have four different states (A, T, C, G). For distance-based analysis, the evolutionary divergence between the mitotypes can be estimated from the nucleotide differences in their sequences.

Mitochondrial DNA

323

III. REVIEW OF mtDNA STUDIES OF FISH POPULATIONS Very early RFLP analyses of mitochondrial DNA variation in terrestrial and freshwater species showed that mitotypes are usually strongly patterned geographically in conspeciﬁc populations. In addition, these studies found that restriction analysis of mtDNA was a more sensitive indicator of intraspeciﬁc differentiation than allozyme analysis (Avise, 1987, and references therein). The results of the ﬁrst analyses in marine species revealed a somewhat different situation. Whereas in some cases a continuously distributed marine species exhibited conspicuous geographic structuring [e.g., Limulus polyphemus (Saunders et al., 1986) and Opsanus tau (Avise et al., 1987)], in other cases there was an apparent lack of geographic structuring of mitotypes over large areas. For example, American eel (Anguilla rostrata) presented no mtDNA divergence in a 4,000-km stretch of shoreline (Avise et al., 1986). Also, Graves et al. (1984) found no signiﬁcant differences between Atlantic and Paciﬁc populations of the skipjack tuna (Katsuwonus pelamis) despite the fact that this is a highly vagile species. A lot of work has since been done on population structure of marine species based on restriction analysis. The application of PCR-based methods for the assessment of mtDNA variation has been introduced mainly during the 1990s in intraspeciﬁc population analysis. There are several reports that have demonstrated the utility and high resolution that these methods offer (see, e.g., references in Meyer, 1993; Ward, 2000; Feral, 2002). The following are some indicative examples of the numerous studies that have been conducted by using mtDNA as a tool to address questions related to population or stock structure in marine ﬁsh species. Smith et al. (1989) by RFLP analysis in Atlantic cod (Gadus morhua) argued for a lack of genetic divergence between eastern and western populations. On the contrary, Carr and Marshall (1991) ampliﬁed and sequenced a 298-bp region of the cytochrome b gene of this species and found differences in mitotypic frequencies between Newfoundland and Norwegian populations. In the same line, Dahle (1991) presented evidence for the existence of a discrete eastern Atlantic stock. It is worth noting that microsatellite DNA analysis revealed ﬁne-scale genetic differentiation off Nova Scotia, but not increased large-scale (east-west) differences (Bentzen et al., 1996; Ruzzante et al., 1997). This ﬁnding may be attributed to different evolutionary dynamics of microsatellite DNA (mainly higher mutation rate) compared to mitochondrial DNA. Martin et al. (1992) studied the population structure of the armorhead (Pseudopentaceros wheeleri) in the North Paciﬁc Ocean by PCR ampliﬁcation and sequencing the cytochrome oxidase I gene, and PCR ampliﬁcation and restriction analysis of the rRNA genes and the D-loop. They found no geographic partitioning of mitotypes, thus refuting the hypothesis of the existence of two distinct subspecies.

324

Antonios Magoulas

Working with PCR ampliﬁcation and sequencing a 612-bp fragment of cytochrome b gene, Finerty and Block (1992) found seven mitotypes in the blue marlin (Makaira nigricans) from the Atlantic and Paciﬁc. Those mitotypes are grouped into two phylogenetic assemblages, the frequencies of which differed greatly between the Atlantic and Paciﬁc Oceans. Several studies in yellowﬁn tuna (Thunnus albacares) have shown that there is more mtDNA differentiation within oceans than between oceans, with the between oceans differences being very small (Scoles and Graves, 1993; Ward et al., 1994). On the contrary, mtDNA variation in two other tuna species, albacore (Thunnus alalunga) and bigeye (Thunnus obesus), has shown strong differentiation between the Atlantic and Paciﬁc Oceans (Chow and Ushiama, 1995; Alvarado Bremer et al., 1998; Grewe and Hampton, 1998). Using RFLP analysis of mtDNA, Kotoulas et al. (1995) found evidence for discrete Mediterranean and southern-eastern (Gulf of Guinea) Atlantic stocks of swordﬁsh (Xiphias gladius). Later, Chow and Takeyama (2000) conﬁrmed the existence of these two stocks and added a third one in the Indo-Paciﬁc. Five different stocks in the Paciﬁc Ocean, the north Atlantic Ocean, the south Atlantic Ocean, and the Mediterranean Sea were detected by Alvarado Bremer et al. (1996) using a combined RFLP/sequencing technique. Finally, Reeb et al. (2000), by sequencing a portion of the control region, studied population structure of swordﬁsh in the Paciﬁc Ocean and found that northern and southern populations in the western Paciﬁc were signiﬁcantly divergent, while such differentiation was not observed in the east. Sedberry et al. (1996) analyzed by RFLP a portion of ND-1 gene in wreckﬁsh (Polyprion americanus) and found evidence for differentiation between the North and South Atlantic ﬁsh, but no stock separation was detected between eastern North Atlantic and western North Atlantic or between South Atlantic and South Paciﬁc Oceans. It should be noted that a study conducted a few years later by employing more variable nuclear genetic markers (microsatellite DNA) gave evidence for stock differentiation between the southern oceans (Ball et al., 2000). European eel (Anguilla anguilla) has long been considered a true panmictic species because of its extraordinary life history pattern, which involves migration to the Sargasso Sea for spawning. Indeed, the ﬁrst studies of mtDNA variation in this species did not detect any genetic structure (Avise et al., 1986; Lintas et al., 1998). However, a recent study using microsatellite DNA provided evidence for a weak but statistically signiﬁcant population structure, while the sequence analysis of cytochrome b locus, which was also employed, did not reveal signiﬁcant structure (Daemen et al., 2001). Independently, Wirth and Bernatchez (2001), by using analysis of microsatellite DNA, also concluded that panmixia might not be true in this species. In a recent study, Ball et al. (2003), by using mtDNA sequence and RFLP analysis, provided evidence for large-scale genetic differentiation of red porgy (Pagrus

Mitochondrial DNA

325

pagrus) in the Atlantic Ocean. Distinct stocks were suggested in the northeastern, northwestern, and southwestern Atlantic.

IV. CRITIQUE The marine realm in general does not present severe barriers to gene ﬂow such as those confronted by freshwater species. In addition, many marine species have extremely high dispersal capabilities, either as adults (migrations) or in pelagic premature stages (passive dispersal of eggs and larvae). Thus, freshwater species would be expected to exhibit greater overall population structure than marine species, and this seems to be the case in general. Nevertheless, speciﬁc life history characteristics, such as homing behavior to distinct spawning grounds, can lead to the evolution of different genetic stocks. Such life history characteristics may explain, for example, the marked mtDNA differences found between swordﬁsh (Xiphias gladius) samples from the Gulf of Guinea and the Mediterranean (Kotoulas et al., 1995). Despite the fact that this species is highly motile and there is free migration through the Gibraltar Straits, two divergent genetic stocks were established in the Mediterranean and the studied area of the Atlantic. It is known that there is one swordﬁsh spawning ground in the Mediterranean and another one in the Atlantic, and it is reasonable to assume that the two stocks owe their existence to the fact that the animals return to spawn on their natal spawning ground. If genetic differentiation is detected between two samples, then one could conclude that they belong to two different stocks, if it can be assumed that there is no selection acting on the genetic markers used for the analysis. On the contrary, if no genetic differentiation is found, then any model of population structure ranging from complete panmixia to the exchange of as little as 1% of individuals can be invoked (Ward, 2000). It is also possible that the sampling design, in terms of time and place, and sample sizes were not appropriate. The sample sizes are often too small to provide reliable conclusions about genetic differentiation. It has been proposed that a realistic target for sample size is 100 specimens per sample (Ward, 2000). A genetic discontinuity (genetic break) can arise in continuously distributed marine organisms because neighboring populations are prevented from exchanging genes by hydrographic or other environmental factors (salinity or temperature gradients, ocean, currents, etc.). For example, the genetic break found in Limulus polyphemus offshore of Cape Canaveral, Florida, coincides with the longrecognized transition area between warm-temperate and tropical faunas. This could mean that there are some ecological factors (e.g., salinity or temperature gradient) in this area that prevent the free genetic exchange between the populations from either size of the break (Saunders et al., 1986).

326

Antonios Magoulas

There are some potential difﬁculties when using mtDNA stock structure analysis, stemming mainly from its uniparental mode of inheritance. For example, if a major geographically localized mtDNA break is found in a continuously distributed species, it is commonly interpreted as delineating two different stocks in the sense that the corresponding populations have been genetically isolated for a considerable time span. However, as Neigel and Avise (1986) have shown, such breaks may also appear stochastically if the species is characterized by limited dispersal capability and gene ﬂow (such as demersal species with low mobility and no pelagic life stages). Thus, the existence of a geographic mtDNA break should be interpreted cautiously, as evidence for the existence of different stocks. The researcher should try to determine which factors (life history characteristics, environmental factors, etc.) could have acted as barriers to gene ﬂow, and thus given rise to different stocks, before deciding that these stocks are real. If two groups of highly divergent mitotype groups (phylads or phylogroups) are found coexisting in the same area, this could be interpreted as evidence that two different subspecies are present. But here also caution should be exercised because the same pattern could have arisen in a random-mating population by secondary mixing of formerly isolated subpopulations. Contrary to the case with nuclear DNA markers, because mtDNA is uniparentally inherited and nonrecombining, the signals of past separation events (i.e., two divergent assemblages of mitotypes) will coexist in the population for many generations after mixing. Thus, there is a danger to mistaking a randomly mating population as consisting of two different stocks or even of two different taxonomic units, although such stocks are nonexistent from either the management or the population genetics perspective. One way to decide if there are two distinct taxonomic entities coexisting, or there is only one panmictic assemblage which experienced a secondary contact in the past, is to examine whether there is random or nonrandom association between mitotypes and nuclear markers. A random association is evidence of a panmictic population, whereas nonrandom association may be indicating the existence of two subspecies. Another test for panmixia is to see whether the frequencies of nuclear markers are in agreement with Hardy–Weinberg expectations for a randomly mating population. The different subspecies hypothesis would be reinforced if an excess of homozygotes were found. But even if such information is not available, in some cases the knowledge of the hydrography and/or the geologic history of the region can help to extrapolate what is the most plausible explanation. In a study of mtDNA variation in European anchovy (Engraulis encrasicolus) populations, Magoulas et al. (1996) found that the whole Mediterranean Sea and Bay of Biscay are characterized by the coexistence (in varying proportions) of two highly divergent mtDNA phylads. One might be tempted to speculate that two different subspecies, or even sibling species, indistinguishable in terms of

Mitochondrial DNA

327

morphology, coexist in these areas. However, there are several lines of evidence that this is not the case. Geologic and hydrographic evidence suggest a one-way gene ﬂow from the Black Sea into the Mediterranean, through the straits of Bosporus and the Dardanelles. It seems, then, that the Black Sea population evolves independently because there is no introduction of genetic material from the Mediterranean, while at the same time there is an “enrichment” of Mediterranean populations with Black Sea genetic material. It appears reasonable to assume that the Black Sea population has not reached the stage of a distinct subspecies because in a previous study of allozyme variation in the Greek seas (Spanakis et al., 1989) no signiﬁcant deviations from Hardy–Weinberg equilibrium expectations were found. Moreover, the existence of several heteroplasmic individuals, all of which bore a phylad-A mitotype and a phylad-B mitotype (Magoulas and Zouros, 1993), indicates that mating between animals of phylad A and phylad B does take place. The examples of cod and eel that were exempliﬁed in the previous section demonstrate that identiﬁcation of stocks using genetics is not an easy task and different markers may provide contrasting evidence. For this reason, the selection of the marker that is appropriate for a given study is of crucial importance. For example, mtDNA is more appropriate for cases where differentiation in broad geographic scales is suspected, whereas microsatellite DNA is more appropriate in cases where ﬁne-scale differentiation is more likely. In conclusion, the usefulness of mtDNA in assessing intraspeciﬁc variation and stock structure is irrefutable, but as Avise (1987) has pointed out, it may be best to interpret mtDNA data on a case-by-case basis, taking into account the zoogeography, biology, and life history of the species. It is also important that information on the geologic history and the present environmental conditions of the area under consideration is taken into account.

REFERENCES Alvarado Bremer, J. R., Mejuto, J., Greig, T. W., and Ely, B. 1996. Global population structure of the swordﬁsh (Xiphias gladius L.) as revealed by analysis of the mitochondrial DNA control region. J. Exper. Mar. Biol. and Ecol. 197: 295–310. Alvarado Bremer, J. R., Stequert, B., Robertson, N. W., and Ely, B. 1998. Genetic evidence for interoceanic subvision of bigeye tuna (Thunnus obesus Lowe) populations. Mar. Biol. 132: 547–557. Attardi, G. 1985. Animal mitochondrial DNA: an extreme example of genetic economy. Intern. Rev. Cytol. 93: 93–145. Avise, J. C. 198). Identiﬁcation and interpretation of mitochondrial DNA stocks in marine species. Proceedings of the Stock Identiﬁcation Workshop, Nov. 1985, Panama City Beach, Florida, NOAA Technical Memorandum NMFS-SEFC-199. Avise, J. C. 1994. Molecular Markers, Natural History and Evolution. Chapman and Hall, New York. Avise, J. C. 2000. Phylogeography: The History and Formation of Species. Harvard University Press, Cambridge.

328

Antonios Magoulas

Avise, J. C., Helfman, G. S., Saunders, N. C., and Hales, L. S. 1986. Mitochondrial DNA differentiation and life history pattern in North Atlantic eels. Proc. Natl. Acad. Sci. USA 83: 4350–4354. Avise, J. C., Reeb, C., and Saunders, N. C. 1987. Geographic population structure and species differences in mitochondrial DNA of mouthbrooding marine catﬁshes (Ariidae) and demersal spawning toadﬁshes (Batrachoididae). Evolution 41(5): 991–1002. Ball, A. O., Sedberry, G. R., Zatcoff, M. S., Chapman, R. W., and Carlin, J. L. 2000. Population structure of the wreckﬁsh Polyprion americanus determined with microsatellite genetic markers. Mar. Biol. 137: 1077–1090. Ball, A. O., Sedberry, G. R., Wessel III, J. H., and Chapman, R. W. 2003. Large-scale genetic differentiation of Pargus pargus in the Atlantic. J. Fish. Biol. 62: 1232–1237. Bentzen, P., Taggart, C. T., Ruzzante, D. E., and Cook, D. 1996. Microsatellite polymorphism and the population structure of Atlantic cod (Gadus morhua) in the northwest Atlantic. Can. J. Fish. Aquat. Sci. 53: 2706–2721. Carr, S. M. and Marshall, H. D. 1991. Detection of intraspeciﬁc DNA sequence variation in the mitochondrial cytochrome b gene of Atlantic cod (Gadus morhua) by the polymerase chain reaction. Can. J. Fish. Aquat. Sci. 48: 48–52. Chow, S. and Ushiama, H. 1995. Global population structure of albacore (Thunnus alalunga) inferred by RFLP analysis of the mitochondrial ATPase gene. Mar. Biol. 123: 39–45. Chow, S. and Takeyama, H. 2000. Nuclear and mitochondrial DNA analyses reveal four genetically separated breeding units of the swordﬁsh. J. Fish Biol. 56: 1087–1098. Daemen, E., Cross, T., Ollevier, F., and Volckaert, F. A. M. 2001. Analysis of the genetic structure of European eel (Anguilla anguilla) using microsatellite DNA and mtDNA markers. Mar. Biol. 139: 755–764. Dahle, G. 1991. Cod, Gadus morhua L., populations identiﬁed by mitochodrial DNA. J. Fish. Biol. 38: 295–303. Dowling, T. E., Moritz, C., and Palmer, D. 1996. Nucleic acids III: analysis of fragments and restriction sites. In D. M. Hillis and C. Moritz (eds.), Molecular Systematics. Sinauer, Sunderland, MA, pp. 249–320. Feral, J. 2002. How useful are the genetic markers in attempts to understand and manage marine biodiversity? J. Exp. Mar. Biol. Ecol. 268: 121–145. Finerty, J. R. and Block, B. 1992. Direct sequencing of mitochondrial DNA detects highly divergent haplotypes in blue marlin (Makaira nigricans). Mol. Mar. Biol. Biotech. 1(3): 206–214. Graves, J. E., Ferris, S. D., and Dizon, A. E. 1984. Close genetic similarity of Atlantic and Paciﬁc skipjack tuna (Katsuonus pelamis) demonstrated with restriction endonucleases analysis of mtDNA. Mar. Biol. 79: 315–319. Grewe, P. M. and Hampton, J. 1998. An assessment of bigeye (Thunnus obesus) population structure in the Paciﬁc Ocean, based on mitochondrial DNA and DNA microsatellite analysis. CSIRO Marine Research, Hobart, Australia. Kocher, T. D., Thomas, K., Meyer, A., Edwards, S. V., Paabo, S., Villablanca, F. X., and Wilson, A. C. 1989. Dynamics of mitochondrial DNA evolution in animals: ampliﬁcation and sequencing with conserved primers. Proc. Natl. Acad. Sci. USA 86: 6196–6200. Kotoulas, G., Magoulas, A., Tsimenides, N., and Zouros, E. 1995. Marked mitochondrial DNA differences between Mediterranean and Atlantic populations of the swordﬁsh, Xiphias gladius. Mol. Ecol. 4: 473–481. Lansman, R. A., Shade, R. O., Shapira, J. F., and Avise, J. C. 1981. The use of restriction endonucleases to measure mitochondrial DNA sequence relatedness in natural populations. III Techniques and potential applications. J. Mol. Evol. 17: 214–226. Lintas, C., Hinaro, J., and Archer, H. 1998. Genetic variation of the European eel (Anguilla anguilla). Mol. Mar. Biol.Biotech 7: 263–269.

Mitochondrial DNA

329

Magoulas, A. and Zouros, E. 1993. Restriction-site heteroplasmy in anchovy (Engraulis encrasicolus) indicates incidental biparental inheritance of mitochondrial DNA. Mol. Biol. Evol. 10(2): 319–325. Magoulas, A., Tsimenides, N., and Zouros, E. 1996. Mitochondrial DNA phylogeny and the reconstruction of the population history of a species: the case of European anchovy (Engraulis encrasicolus). Mol. Biol. Evol. 13(1): 178–190. Martin, A. P., Humphreys, R., and Palumbi, S. R. 1992. Population genetic structure of the armorhead, Pseudopentaceros wheeleri, in the North Paciﬁc Ocean: application of the polymerase chain reaction to ﬁsheries problems. Can. J. Fish. Aquat. Sci. 49: 2386–2391. Meyer, A. 1993. Evolution of mitochondrial DNA in ﬁshes. In P. W. Hochachka and T. P. Mommsen (eds.), Biochemistry and Molecular Biology of Fishes, Vol. 2. Elsevier, Amsterdam, pp. 1–38. Moritz, C., Dowling, T. E., and Brown, W. M. 1987. Evolution of animal mitochondrial DNA: relevance for population biology and systematics. Ann. Rev. Ecol. Syst. 18: 269–292. Mullis, K. B., Faloona, F., Scharf, S. J., Saiki, R. K., Horn, G. T., and Erlich, H. A. 1986. Speciﬁc enzymatic ampliﬁcation of DNA in vitro: the polymerase chain reaction. Cold Spring Harbor Symposium on Quantitative Biology, 51: 263–273. Nei, M. 1972. Genetic distance between populations. Am. Natur. 106: 283–292. Nei, M. 1987. Molecular Evolutionary Genetics. Columbia University Press, New York. Neigel, J. E. and Avise, J. C. 1986. Phylogenetic relationships of mitochondrial DNA under various demographic models of speciation. In S. Karlin and E. Nevo (eds.), Evolutionary Processes and Theory. Academic Press, New York, pp. 515–534. Raymond, M. and Rousset, F. 1995. GENEPOP: population genetics software for exact test and ecumenicism. J. Heredity 83: 239. Reeb, C. A., Arcangelli, L., and Block, B. A. 2000. Structure and migration corridors in Paciﬁc populations of the Swordﬁsh Xiphius gladius, as inferred through analyses of mitochondrial DNA. Mar. Biol. 136: 1123–1131. Rokas, A., Ladoukakis, E., and Zouros, E. 2003. Animal mitochondrial DNA recombination revisited. TREE 18: 411–417. Ruzzante, D. E., Taggart, C. T., Cook, D., and Goddard, S. V. 1997. Genetic differentiation between inshore and offshore Atlantic cod (Gadus morhua L.) off Newfoundland: a test and evidence of temporal stability. Can. J. Fish. Aquat. Sci. 54: 2700–2708. Saiki, R. K., Gelfand, D. H., Stoffel, S., Scharf, S. J., Higuchi, R., Horn, G. T., Mullis, K. B., and Erlich, H. A. 1988. Primer-directed enzymatic ampliﬁcation of DNA with a thermostable DNA polymerase. Science 239: 487–491. Sambrook, J., Fritsch, E. F., and Maniatis. T. 1989. Molecular Cloning: A Laboratory Manual. Cold Spring Harbor Laboratory, Cold Spring Harbor, NY. Saunders, N. C., Kessler, L. G., and Avise, J. C. 1986. Genetic variation and geographic differentiation in mitochondrial DNA of the horseshoe crab, Limulus polephemus. Genetics 112: 613–627. Scoles, D. R. and Graves, J. E. 1993. Genetic analysis of the population structure of yellowﬁn tuna, Thunnus albacares, from the Paciﬁc Ocean. Fish. Bull. U.S.A. 91: 690–698. Sedberry, G. R., Carlin, J. L., Chapman, R. W., and Eleby, B. 1996. Population stucture in the pan-oceanic wreckﬁsh, Polyprion americanus (Teleostei: Polyprionidae), as indicated by mtDNA variation. J. Fish Biol. 49 (suppl A): 318–329. Smith, P. J., Jamieson, A., and Bishop, C. A. 1989. Mitochondrial DNA in the Atlantic cod, Gadus morhua: lack of genetic divergence between eastern and western populations. J. Fish Biol. 34: 369–373. Snyder, M., Fraser, A. R., Laroche, J., Gardner-Kepkay, K. F., and Zouros, E. 1987. Atypical mitochondrial DNA from the deep sea scallop Placopecten magellanicus. Proc. Nat. Acad. Sci. USA 84: 7595–7599.

330

Antonios Magoulas

Snyder, S., Roessli, D., and Excofﬁer, L. 2000. Arlequin: A Software for Population Genetics Data Analysis. Genetics and Biometry Laboratory, University of Geneva, Switzerland. Spanakis, E., Tsimenides, N., and Zouros, E. 1989. Genetic differences between populations of sardine, Sardina pilchardus, and anchovy, Engraulis encrasicolus, in the Aegean and Ionian Seas. J. Fish. Biol. 35: 417–437. Swofforf, D. L. and Olsen, G. J. 1996. Phylogeny reconstuction. In D. M. Hillis and C. Moritz (eds.), Molecular Systematics. Sinauer, Sunderland, MA, pp. 407–514. Ward, R. D. 2000. Genetics and ﬁsheries management. Hydrologia 420: 191–201. Ward, R. D., Elliot, N. G., Grewe, P. M., and Smolenski, A. J. 1994. Allozyme and mitochondrial DNA variation in yellowﬁn tuna (Thunnus albacares) from the Paciﬁc Ocean. Mar. Biol. 118: 531–539. White, P. S. and Densmore III, L. D. 1992. Mitochondrial DNA isolation. In A. R. Hoelzel (ed.), Molecular Genetic Analysis of Populations A Practical Approach. IRL Press. Oxford, pp. 29–58. Wirth, T. and Bernatchez, L. 2001. Genetic evidence against panmixia in the European eel. Nature 409: 1037–1040.

CHAPTER

17

Use of Nuclear DNA in Stock Identiﬁcation: Single-Copy and Repetitive Sequence Markers ISAAC WIRGIN* AND JOHN R. WALDMAN† *Department of Environmental Medicine, New York University School of Medicine, Tuxedo, New York, USA, †Hudson River Foundation for Science and Environmental Research, New York, New York, USA, Currently, Biology Department, Queens College, The City University of New York, New York, New York, USA

I. Introduction A. Genetic Stock Identiﬁcation B. Nuclear vs. Mitochondrial DNA C. Introduction to the Nuclear Genome II. Methodology Overview for Analysis of Nuclear DNA III. Single-Copy Coding Nuclear DNA IV. Single-Copy Noncoding Nuclear DNA V. Repetitive Nuclear DNA A. Minisatellites B. Microsatellites VI. Case History: Genetic Analyses of Atlantic Cod A. How Difﬁcult a Problem B. Allozyme and Blood Protein Studies C. RFLP and Sequence Analysis of mtDNA D. Single-Copy Coding nDNA Analysis E. Single-Copy Noncoding nDNA Analysis F. Minisatellite Analysis G. Microsatellite Analysis VII. Conclusions VIII. Summary Acknowledgments References

Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

331

332

Isaac Wirgin and John R. Waldman

I. INTRODUCTION

A. GENETIC STOCK IDENTIFICATION Marine ﬁshes often have high dispersive capabilities and exhibit far-ranging distributions. For many species, even those that as adults are benthic, early life stages are frequently planktonic and subject to passive transport by oceanic currents over long distances. These species often have high fecundity and large population sizes, characteristics that tend to reduce the rate of development of genetic partitioning among localized stocks. Genetic divergence of ﬁsh populations requires that gene ﬂow among units be minimal, and in freshwater and anadromous species is associated with physical barriers. But obvious mechanisms of reproductive isolation most often are absent in marine taxa. Alternatively, very strong divergent natural selection may be operative on a subset of genetic loci to cause signiﬁcant discontinuities in allelic or genotype frequencies. Yet, empirical evidence of genetic divergence of marine ﬁsh stocks is frequently observed, although usually weaker than among freshwater and anadromous taxa, indicating that more subtle mechanisms of reproductive isolation exist. These may entail biological mechanisms of postzygotic isolation or physical means of separation including differential transport by currents or major physical discontinuities in bottom topography. Genetic stock identiﬁcation may be considered a two-step process. First, diagnostic genetic markers must be identiﬁed, their use optimized, and then they must be applied in surveys to spatially or temporally distinct collections to determine the genetic similarity of potentially distinct stocks. These genetic markers must exhibit sufﬁcient levels of genetic variation to robustly evaluate the distinctiveness of potential stocks and to quantify the extent of their genetic differentiation. Population differentiation can be said to exist if the frequencies of alternative alleles or genotypes differ signiﬁcantly among these discrete collections. Markers with only very low levels of genetic variation usually do not provide sufﬁcient statistical power or conﬁdence in surveys plumbing for genetic stock structure. Usually differences in modal alleles are most effective in statistical tests of stock differences. Diagnostic polymorphic markers at moderate to high frequencies provide the most power in deﬁning population structure. Second, for estuarine or marine species, genetic stock identiﬁcation sometimes entails another component, mixed stock analysis. Successful application of genetic mixed stock analysis demands that individual stocks contributing to the mixed stock be discriminated with some degree of certainty. Inability to distinguish individual stocks contributing to the mixed stock limits the application of this approach. Statistical conﬁdence in mixed stock estimates is to a great degree dependent on the extent of resolution of contributing stocks, that sufﬁcient numbers of markers are used, and that sufﬁciently robust numbers of individu-

Use of Nuclear DNA in Stock Identiﬁcation

333

als from the spawning and mixed stocks are sampled. Thus. there are two major prerequisites to the use of genetic approaches in stock identiﬁcation: the identiﬁcation of multiple diagnostic markers and the logistical ability to apply these to the large number of individuals needed for statistical precision in real-world applications.

B. NUCLEAR

VS.

MITOCHONDRIAL DNA

Early DNA-based molecular stock identiﬁcation investigations on ﬁshes focused on mitochondrial DNA (mtDNA) rather than on nuclear DNA (nDNA), largely due to the technical simplicity and high sensitivity of mtDNA approaches compared to allozymes (reviewed in Waldman and Wirgin, 1994; O’Reilly and Wright, 1995). It was widely held that mtDNA approaches provided about an order of magnitude higher levels of intraspeciﬁc variation than exists in coding nDNA (Brown et al., 1979). However, this oversimplifcation did not consider the variation in DNA sequence evolution within each of these two genomes. Within both mtDNA and nDNA there are areas that evolve rapidly and others that evolve slowly. This variation in the rate of sequence change within a genome is particularly dramatic in nDNA. Analysis of mtDNA offers some advantages and disadvantages compared to nDNA approaches. Because mtDNA is usually maternally inherited and does not exhibit recombination, in the absence of additional mutations, all offspring will exhibit identical mtDNA haplotypes as their mothers. Additionally, because it is haploid, the effective population size of mtDNA is only one-fourth that of autosomal sequences in nDNA. Because of its smaller effective population size, the rate of lineage sorting should be much more rapid in mtDNA than nDNA. Thus, because of genetic drift, phylogeographic structure should occur four times faster and be more pronounced using mtDNA than nDNA polymorphisms (Hare, 2001). However, the picture of population structuring painted by mtDNA will only reﬂect the matrilineal history of that population, which could differ from that of the overall population if gender-speciﬁc dispersal behaviors exist. Importantly, no matter how intensively characterized, mtDNA still represents only a single genetic locus, a quality that limits the data analyses that can be applied and the usefulness of mtDNA data in stock identiﬁcation studies. Ideally, results from mtDNA and multilocus nDNA studies should be combined to gain a much more comprehensive picture of the structure and evolutionary history of individual populations or mixed stocks. Because nDNA is composed of more than 3 billion base pairs (bp) compared to 16,000 to 18,000 bp typically seen in mtDNA in ﬁshes, it provides far more potential diagnostic markers useful in discriminating stocks. Thus, nDNA offers an enormous quantity of bp that can be explored for variation. Improved genomic

334

Isaac Wirgin and John R. Waldman

DNA cloning strategies using artiﬁcial chromosomes from bacteria or yeast as vectors, with high-throughput automated DNA sequencing, now make it possible to compare long stretches of nDNA (millions of base pairs) from species of interest for informative single nucleotide polymorphisms (SNPs: base substitutions or additions/deletions). Such studies in humans and other taxa have revealed that the occurrence of SNPs is quite frequent, approximately 0.1% between any two randomly selected humans and even higher in some other taxa. The differences between taxa reﬂect the age of the species and, thus, comparisons in humans may present a conservative estimation of nucleotide diversity. To date, about 6 million SNPs have been revealed in the human genome, which means that SNPs occur approximately every 500 bp. This number of SNPs in humans will increase as more individuals from more semiisolated populations are analyzed. DNA sequence change typically is much more rapid in genomic areas with no known functional constraints. Whereas most of the mitochondrial genome is occupied by genes encoding functional protein products or tRNAs, with very little or an absence of intergenic sequence, only a small percentage (2–5%) of the nuclear genome is composed of encoding gene loci. Instead, most nDNA is noncoding, and these noncoding nDNA sequences can be divided into several different structural entities (single-copy or repetitive), each of which probably also differs in its rate of evolutionary change.

C. INTRODUCTION

TO THE

NUCLEAR GENOME

Nuclear DNA nucleotides occur across a genome containing regions that serve highly varied functions (coding for proteins, regulation of gene expression, and noncoding) and, therefore, which are under different evolutionary constraints. Because of this variation in function and constraint, some regions are highly conserved and others are much more labile. Nuclear DNA genes that encode for protein products in nonpolyploid genomes are usually single copy. There are two regions within individual genes termed exons and introns (Fig. 17-1). Exons are transcribed into mRNA and encode for proteins, tRNAs, and rRNAs. As a result, their nucleotide sequences are usually under tight selective restraints (with the exception of synonymous sites at the third nucleotide position within codons at which substitutions usually do not result in amino acid changes in the peptide). Within each gene there are several to many individual exons, each of which is separated by intervening intron sequences, which can be considerable in length. Thus, for each exon within a gene, there is an equal number minus one of introns. While the sequences within introns do play a functional role in the splicing of primary RNA transcripts and in the transcriptional regulation of expression of some genes, their functional roles in cellular homeostasis is probably much less

335

Use of Nuclear DNA in Stock Identiﬁcation chromosome of 1.5X108 nucleotide pairs, containing about 3000 genes

0.5% of chromosome, containing 15 genes Gene 2

Intergenic Region

Intergenic Region

Gene 13

one gene of 105 nucleotide pairs

5¢ regulatory region

Exon 1

Exon 11 3¢ untranslated region

Intron 7

DNA TRANSCRIPTION 5 5¢

3 3¢

primary RNA transcript

RNA SPLICING 5¢ 5

3 3¢ mRNA

FIGURE 17-1. A schematic of a eukaryotic chromosome illustrating the structure of an individual gene, including 12 exons, 11 introns, 5¢ regulatory region, and 3¢ untranslated region.

than that of exons. Additionally, the ﬁrst (5¢) and last (3¢) exons within a gene (although functional in transcriptional regulation and mRNA stability) are usually not completely translated into proteins and, thus, are probably under reduced selective constraint compared to other exons. Additionally, each gene has associated with it 5¢ and 3¢ sequences (outside of the exons), which play roles in the regulation of expression of that gene. Finally, there are very large intergenic sequences of unknown function between individual genes, which make up the vast bulk of the nuclear genome. In a Drosophila esterase gene, it was found that intron sequences were most variable, followed by the 5¢ distal untranscribed region, the 3¢ untranslated region, the exons, the 5¢ proximal untranscribed region, and, ﬁnally, the 5¢ untranslated noncoding regions were the least variable (reviewed in Zhang and Hewitt, 2003). Within the intergenic areas and, to a lesser degree, the noncoding areas associated with genes, are loci with highly repetitive DNA sequences termed microsatellites and minisatellites. Repetitive DNA sequences are sometimes seen in coding DNA and in regulatory regions of genes, but much less so than in noncoding areas. There are tens to hundreds of thousands of microsatellite loci interspersed throughout much of the nuclear genome and a lesser number of minisatellite loci whose distribution within the genome may be restricted to certain chromosomal regions. Microsatellite loci are comprised of very short (1–6 bp), tandem repeat sequences. The repeat sequences in minisatellites are longer than in microsatellites, typically 10 bp to 40 bp in length. A virtue of these

336

Isaac Wirgin and John R. Waldman

repetitive DNA loci are the high levels of allelic diversity and heterozygosity they exhibit. Today, microsatellite analysis is the favored genetic approach for elucidation of stock structure in ﬁshes because of the often high levels of polymorphisms revealed and its ease of laboratory and data analyses. Thus, one of the chief advantages of nDNA analysis for stock identiﬁcation purposes is that there is a spectrum of genetic divergence available from which a researcher can select the appropriate level of sensitivity. Moreover, results using microsatellites or single-copy nDNA may be analyzed within a Hardy–Weinberg framework, which may be useful in detecting mixtures of stocks through gametic disequilibrium analysis (Waples and Smouse, 1990) and which may provide additional resolution in stock composition analysis using mixture (Xu et al., 1994) or individual assignment models (Paetkau et al., 1995). We present an overview of the major classes of nDNA useful for stock identiﬁcation, proceeding from the most conserved (least sensitive) to the most rapidly evolving (most sensitive), and then review and compare the body of studies using these approaches on populations of Atlantic cod to demonstrate their strengths and weaknesses.

II. METHODOLOGY OVERVIEW FOR ANALYSIS OF NUCLEAR DNA In analyzing nDNA, total DNA must ﬁrst be isolated from individual specimens. Many tissues can serve as DNA sources, most often ﬁn clips, although in the past liver and blood were the primary sources of DNA. Tissues used may be fresh, previously frozen, preserved in a suitable buffer, or most frequently, an alcohol, usually ethanol. Also, retrospective studies sometimes are conducted to compare and contrast contemporary and historical population structure or overall levels of genetic diversity. In most cases, these investigations use archived scale samples (Miller and Kapuschinski, 1997), although hard body parts such as otoliths are also used (Ruzzante et al., 2001; Ludwig et al., 2002). Archived scales and otoliths offer the opportunity to evaluate the long-term stabilities of allelic frequencies over decadal increments in stocks or species that have experienced dramatic ﬂuctuations in abundance (Lage et al., 2001). There are many reliable techniques available to isolate suitable DNA for analysis, and the choice of technique depends on tissue sources, allowable hazard exposure to workers, cost, technique of DNA analysis, and anticipated storage time for isolated DNAs. A word of caution, however, is that the storage of tissues in formaldehyde usually presents nearly insurmountable obstacles to their analysis. Today, almost all DNA-based studies use the polymerase chain reaction (PCR) to generate sufﬁcient copies of the target DNA to analyze. The advent of PCR technology has greatly simpliﬁed the technical demands of visualizing targeted DNA sequences. PCR allows for the in vitro ampliﬁcation of 106 or more copies

Use of Nuclear DNA in Stock Identiﬁcation

337

of a target DNA fragment within several hours. PCR generates sufﬁcient copy numbers of target DNA sequences for their routine visualization through standard procedures such as staining with ethidium bromide. As a result, PCR permits routine analysis of DNA from single eggs and larvae, and from noninvasively secured tissues such as ﬁn clips and scales. Even partially degraded DNA from poorly preserved sources can be analyzed if sufﬁciently small PCR products are identiﬁed. Initially, a locus that harbors an informative DNA polymorphism is identiﬁed. Oligonucleotide primers (15–20 nucleotides in length), which ﬂank the intervening targeted DNA sequence, are designed by the investigator based on data in DNA sequence repositories, such as GenBank, or selected from the literature and commercially synthesized at a nominal cost. Design of these primers must be based on some prior knowledge of the DNA sequence of the targeted locus unless random ampliﬁed polymorphic DNA (RAPDs) or ampliﬁed fragment length polymorphism (AFLP) analyses are used. This knowledge can be obtained by cloning and sequencing of the targeted locus in the species of interest or by trial via transfer from a closely related taxon if the sequence has been sufﬁciently conserved in evolution. The size of the targeted DNA sequence should be somewhat limited in size (usually not exceeding 1,000 bp) to ensure reproducible ampliﬁcation in PCR. The smaller the size of the PCR product, the greater the ease of reproducible ampliﬁcation, particularly when DNA samples are of dubious quality. In PCR ampliﬁcation, the double-stranded total DNA is denatured by heat, an oligonucleotide primer pair anneals speciﬁcally with the now single-stranded targeted DNA sequence, and a heat-stable DNA polymerase is added which copies with great ﬁdelity the intervening DNA sequence. This process is repeated for 30 to 40 cycles, which allows for exponential ampliﬁcation of the intervening DNA sequence. As a result, sufﬁcient copies of targeted double-stranded DNA sequence are generated, permitting easy detection by ﬂuorimetric assays. Often, multiple PCR reactions, each amplifying different DNA targets, are conducted in a single reaction tube (multiplexing) in an approach that greatly conserves reagents and effort. After the targeted DNA sequence is ampliﬁed, several strategies are available for its analysis, depending on the molecular approach selected by the investigators. The PCR products may be sequenced manually or by use of an automated DNA sequencer to determine the exact sequence of a portion or most of the nucleotides within the PCR-ampliﬁed product. The use of an automated sequencer allows for routine DNA sequence analysis of 400 to 800 bp in a single run. A new technology called Pyrosequencing permits very rapid and accurate sequencing of short stretches of DNA (up to 40 bp) that contain previously deﬁned informative polymorphisms. Alternatively, the PCR product may be digested with restriction enzymes [restriction fragment length polymorphism

338

Isaac Wirgin and John R. Waldman

(RFLP) analysis], which allows for a comparison of the sequence of a small subset of the nucleotides (those recognized by the restriction enzyme) within the PCR product. Clearly, sequencing is a more efﬁcient and informative approach; however, in those instances in which one or a very limited number of diagnostic nucleotides have been previously identiﬁed, RFLP analysis can provide a costeffective alternative. In microsatellite or single-locus minisatellite analyses in which alleles differ in the number of tandem repeats at a locus, the PCR products are compared for differences in molecular size. At microsatellite loci, these differences in molecular size are very small, requiring the use of sensitive means of analysis. This can be done by electrophoresis (usually polyacrylamide gel), but most reliably and efﬁciently through use of an automated DNA sequencer. The use of automated sequencers allows for the easy discrimination of microsatellite alleles that differ in molecular size by as little as 1 to 2 bp.

III. SINGLE-COPY CODING NUCLEAR DNA In taxa with genomes that are nonpolyploid, genes that encode for protein products are usually, although not always, single-copy genes. Duplication of individual loci can and does occur, thus confounding results from analysis of single-copy genes. For example, pseudogenes frequently arise that have lost their coding function, but whose sequences are very similar to those of the actual gene from which they were derived. Similarly, polyploid genomes, such as those that occur in sturgeon species, can introduce problems in data analysis of single-copy and repetitive nDNA sequences. The exons, introns, and 5¢ and 3¢ regions immediately ﬂanking genes can serve as single-copy, coding nDNA targets (Fig. 17-1). In recent years, the use of single-copy nDNA has received little attention in population studies of ﬁshes or other organisms. However, there are advantages to their use not enjoyed by repetitive DNA markers, namely, that de novo generation of mutants, SNPs, is much easier to understand than that of repetitive DNA. This allows for more certainty in the models upon which much data analysis is based. Also, compared to microsatellites, the rate of mutational changes (substitutions per site) across the genome may be far more constant for single-copy sequences. Additionally, surveys of single-copy coding sequences combined with neutral repetitive (or noncoding single-copy sequence) markers allow for evaluations of the relative effects of selective and stochastic processes on genetic population structure. A primary concern with the use of single-copy coding nDNA sequences in population studies was that their levels of polymorphism were depauperate compared to repetitive or noncoding, single-copy DNA. Clearly, on an individual locus basis, they do not exhibit levels of allelic diversity comparable to repetitive

Use of Nuclear DNA in Stock Identiﬁcation

339

DNA loci such as microsatellites or minisatellites and, in fact, polymorphic singlecopy loci are often, although not always, biallelic. Yet, it has been estimated that the average gene contains approximately four SNPs with frequencies of at least a few percent in human populations, with about 240,000 to 400,000 common SNPs across the entire genome (Cargill et al., 1999). Also, it has been estimated that approximately three times more SNP loci than microsatellites are needed to estimate population genetic parameters with statistical conﬁdence (Brumﬁeld et al., 2003). However, data from human and other genome sequence projects along with increasingly automated and sophisticated means of their analyses suggest that these polymorphisms could yield a rich harvest of informative markers. For example, levels of nucleotide diversity in the human genome, based on comparisons among limited panels of individuals (usually 40), have been estimated at approximately 0.1% (International SNP MAP Working Group, 2001), with much higher levels seen in birds (0.23–0.25%) (Primmer et al., 2002) and in Drosophila (Moriyama and Powell, 1996). Thus, in humans, any two copies of the human genome will differ by about 3 million nucleotides or about one variant per 1,000 bases on average (International HapMap Consortium, 2003). It has been estimated that there are about 10 million SNP sites (with minor allele frequencies >1%) within the world’s human populations, or one variant site per every 300 bases. SNP sites constitute about 90% of DNA sequence variation in human populations (International HapMap Consortium, 2003). Comparative estimates of overall levels of SNPs in ﬁsh genomes will soon be available when the results of several genome-wide sequencing projects are made available. Thus, there are almost certainly sufﬁcient levels of diversity at single-copy loci to serve as sensitive targets for stock identiﬁcation studies. Comparative sequence data from these ﬁsh genome projects will allow for easier design of PCR primers at individual coding-gene loci. Multiplexing of PCRs from several single-copy loci and automated methods, such as Pyrosequencing, for screening multiple SNP loci simultaneously for allelic variants at previously deﬁned informative polymorphic loci will allow for rapid multiloci surveys of many individuals. The few studies in ﬁsh on single-copy coding nDNA loci used DNA hybridization techniques that by today’s standards are labor intensive and demanding in terms of DNA quality. Wirgin et al. (1992) used ﬁve single-copy proto-oncogene (genes that are highly conserved in animal taxa and are important in cancer development) probes such as K-ras and c-abl to distinguish among all four North American species of Morone (striped bass, white bass, white perch, and yellow bass). They found ﬁxed differences among the species at nine genetic loci, ensuring a high degree of certainty (97%) in distinguishing among parental species, F1 hybrids, and later generation hybrids. This type of nDNA also has been used to identify differences among salmonid stocks. Although single-copy coding genes reside at a single genetic locus, internal regions may be repeated within a locus, thus providing potentially informative

340

Isaac Wirgin and John R. Waldman

polymorphisms. Nuclear ribosomal RNA genes (rDNA) contain three coding regions that evolve slowly: 5.8S, 18S, and most of 28S rDNA. However, the internal and external transcribed spacer regions (ITS-1, ITS-2, 5¢ ETS) as well as portions of 28S rDNA evolve more rapidly, and the intergenic spacer region (IGS) evolves most rapidly (Phillips and Pleyte, 1991; Phillips et al., 1992). To date, ﬁxed differences in the 5¢ ETS were found among allopatric populations of Arctic charr, Salvelinus alpinus (Zhuo, 1991). Also, a polymorphism was found in the spacer region of rRNA genes that discriminates Atlantic salmon, Salmo salar, from North America and Europe (Cutler et al., 1991). Individualspeciﬁc variation in the IGS was found in ﬁve populations of lake trout, Salvelinus namaycush, with most of the variation occurring within, rather than between, populations (Zhuo et al., 1995). As will be described in detail later, populationdiagnostic polymorphisms at the pantophysin (PanI) locus have been described in Atlantic cod from the northeastern and northwestern Atlantic (Pogson and Fevoldsen, 2003) and have provided exciting results both in terms of stock identiﬁcation and in understanding those evolutionary processes that shape their population structure.

IV. SINGLE-COPY NONCODING NUCLEAR DNA The vast majority of the nuclear genome is noncoding and, thus, may evolve rapidly and provide a fertile substrate for stock identiﬁcation studies. For example, most of the nuclear genome is within potentially rapidly evolving intergenic regions. However, a major disadvantage to an approach based on singlecopy, noncoding nDNA is the need to identify polymorphic loci and develop and test species-speciﬁc probes. This labor-intensive process, as originally described for lesser snow goose, Anser caerulescens caerulescens by Quinn and White (1987), included development of a genomic DNA library from the species of interest in plasmid or viral vectors, selection of recombinant clones representative of nonrepetitive nDNA, and empirical testing of each cloned DNA in conjunction with various restriction enzymes to determine which combinations revealed informative polymorphisms. An almost inﬁnite number of clones can be generated, each most probably representative of different nDNA sequences. Unfortunately, there is no a priori way of determining which clone-restriction enzyme combinations will be informative, and so, each must be tested on individuals of the stocks being investigated. However, once informative clonerestriction enzyme combinations are identiﬁed, diagnostic clones can be sequenced, PCR primers designed, and the assay made PCR-based as described for striped bass, Morone saxatilis, in Wirgin et al. (1997). Alternatively, nucleotide sequence data from ﬁsh genome projects can be used to design PCR primers for noncoding nDNA regions. Also, most of the PCR products generated from RAPDs

Use of Nuclear DNA in Stock Identiﬁcation

341

or AFLPs analyses are from intergenic regions and can be sequenced to design PCR primers. Because of its technical complexity, this approach saw only limited use in stock identiﬁcation of ﬁshes. Wirgin and Maceda (1991) screened 20 clones developed from striped bass. Seven of these clones revealed polymorphisms across collections representative of most of the species’ distribution, a level of variation that was low but which is consistent with the limited genetic diversity found in striped bass through analyses of isozymes, mtDNA, and microsatellite DNA (Waldman et al., 1988; Wirgin et al., 1991). Among these samples, informative polymorphisms were found at a broad scale, for example, between ﬁsh from the Atlantic and Gulf of Mexico coasts, and at a ﬁner scale, for example, the Hudson River and Chesapeake Bay (Waldman et al., 1996). Karl and Avise (1992) used a modiﬁcation of the above approach to screen for restriction fragment length polymorphisms in single-copy nDNA sequences of American oysters, Crassotrea virginica. Pairs of 24 to 25 PCR primers were designed to amplify polymorphic intervening sequences, and the products were digested with restriction enzymes, separated electrophoretically, and visualized by ethidium bromide staining. This approach revealed a sharp genetic discontinuity between Atlantic and Gulf of Mexico oyster populations; this ﬁnding was consistent with results from an earlier mtDNA study on the same populations. Pogson and colleagues successfully used a slight modiﬁcation of this approach to explore for population structure in Atlantic cod, Gadus morhua using a panel of 17 single-copy probe-enzyme combinations (Pogson et al., 1995). Details of this study will be described in a later section.

V. REPETITIVE NUCLEAR DNA Analysis of repetitive nDNA sequences has become the major approach for ﬁsh stock identiﬁcation purposes and is especially useful in those instances where alternative approaches have revealed depauperate levels of genetic variation. Short, minisatellite (~10–40 bp) and very short, microsatellite (1–6 bp) DNA sequence motifs, sometimes referred to as VNTRs (variable number of tandem repeats), often are found as tandem repeats throughout the nuclear genome. Although functional or selective constraints on the evolution of repetitive nDNA are probably limited, a degree of concerted evolution of these repeats apparently exists among some taxa so that hierarchical groupings at the stock level based on the analysis of repetitive nDNA may emerge (Dowling et al., 1990). Additionally, the number of tandem repeats in the promoter, introns, or other ﬂanking regions of genes may be functionally important in regulating expression of coding genes.

342

Isaac Wirgin and John R. Waldman

A. MINISATELLITES Minisatellites were the original repetitive nDNA motifs analyzed in stock identiﬁcation studies, and their use revolutionized the ﬁeld of molecular evolution because of the high levels of genetic diversity they exhibited. In fact, rates of mutations at many minisatellite loci in humans exceed those at microsatellite loci (Armour et al., 2001). DNA ﬁngerprinting is a process in which the allelic products of many minisatellite loci are visualized simultaneously. The term DNA ﬁngerprinting was coined because proﬁles of variation at a panel of minisatellite loci could be used with high certainty to uniquely identify individuals within human populations. Minisatellites are less common than microsatellites in genomes, probably present at hundreds or thousands of different loci per genome (Armour et al., 2001). Unlike microsatellites, which are distributed throughout the entire genome at regular 10 kbp intervals, minisatellites are frequently clustered on chromosomes. DNA ﬁngerprinting examines allelic diversity at minisatellite loci which contain much longer repeat units (>10 bp) than those at microsatellite loci. Many hypervariable minisatellite loci show some variation in the precise sequence of repeat units. An allele at a typical hypervariable minisatellite locus is not perfectly repeated, but instead is a mixture of two or more repeat units (Armour et al., 2001). Minisatellite mutations probably result from a combination of processes including unequal recombination at meiosis, gene conversion, and slippage at replication forks (Yauk, 1998). Early studies of repetitive nDNA in ﬁshes were dependent on the efﬁcacy of heterologous DNA probes isolated and characterized in higher vertebrates to recognize repetitive sequences in ﬁsh DNA. Minisatellite probes such as Per, M13, and Jeffreys’ core probes 33.6 and 33.15 were evaluated and found to hybridize to most ﬁsh DNA and to reveal highly polymorphic patterns (Castelli et al., 1990). That is because the target minisatellite tandem repeat sequences are fairly long and conserved, and that universal DNA probes that were readily available were able to detect multilocus DNA ﬁngerprints in a variety of disparate taxa, including ﬁshes. Because of the large number of loci visualized on DNA ﬁngerprinting gels, there is uncertainty regarding assignment of minisatellite DNA bands to individual loci. Thus, it is inadvisable to use the Hardy–Weinberg model in analyzing allelic frequencies of these data. Similar to RAPDs and AFLPs, the extent of band sharing between individuals is calculated and then used as a measure of dissimilarity among individuals and between populations. Additionally, DNA ﬁngerprinting is not a PCR-based approach and so demands on tissue condition are much more stringent than for PCR-based approaches. However, if one family of minisatellite DNA fragments proves particularly informative in DNA ﬁngerprinting, the locus that encodes these bands can be cloned, ﬂanking sequences characterized, primers designed, and the assay for that one minisatellite locus can be

Use of Nuclear DNA in Stock Identiﬁcation

343

made PCR-based and even multiplexed with microsatellite loci or other minisatellite loci. In DNA ﬁngerprinting, relatively large amounts of highly intact DNAs are isolated and fragmented by digestion with one or more restriction enzymes. The digested DNA fragments are electrophoretically separated in agarose gels, transferred to hybridization membranes, and ﬁxed on the membranes. The DNA fragments on the membranes are visualized by hybridization with labeled DNA ﬁngerprinting probes and visualized by exposure to X-ray ﬁlm. The amount of DNA fragment sharing among individuals is then determined within a population, and these data are then used to calculate interpopulation genetic diversity. Heath et al. (1995) evaluated the ability of 12 different oligonucleotides containing minisatellite core sequences to hybridize to total DNA isolated from chinook salmon, Oncorhynchus tshawytscha. Of the 12 minisatellite probes, 7 produced highly variable ﬁngerprint-like banding patterns, but only 3 of these revealed clear, distinct DNA bands. Wirgin et al. (1991) analyzed minisatellite nDNA in striped bass with the heterologous M-13 and Per DNA ﬁngerprinting probes and found single DNA fragments that were diagnostic at the population level among selected Atlantic and Gulf of Mexico stocks. Prodohl et al. (1992) used multilocus DNA ﬁngerprint analysis to determine genetic variability within and among brown trout, Salmo trutta, populations in Ireland. Using Jeffreys’ et al. (1985) human minisatellite 33.6 probe, Prodohl et al. (1992) found signiﬁcant differences in overall levels of genetic heterozygosity within these populations, and band sharing among populations. Single-locus hypervariable minisatellite probes have been developed in several laboratories for use in salmonid species, including Atlantic salmon (Taggart and Ferguson, 1990; Bentzen et al., 1993). Applicability of one of the Atlantic salmon probes (3.15.34) to chinook salmon was demonstrated (Stevens et al., 1993) and polymorphisms were found among a limited number of hatchery families. Taylor et al. (1994) examined minisatellite variation among 42 populations of chum salmon, Oncorhynchus keta, over most of their range in the North Paciﬁc. Their single locus probe (Ssa I) hybridized to at least two linked minisatellite loci; evidence was found of three regional chum salmon groupings that were consistent with probable patterns of postglacial dispersal from three refugia. Three minisatellite probes, Ssa I (Bentzen et al., 1993) and 3.15.34 (Taggart and Ferguson, 1990) isolated from Atlantic salmon, and OtPBS1 isolated from chinook salmon, were used to discriminate among 10 sockeye salmon, Oncorhynchus nerka, populations in Russia, western Alaska, and British Columbia–Washington (Beacham et al., 1995). Signiﬁcant differences in allelic frequencies were observed with all three probes at regional, among-river, and within-river levels. Although it was not possible to classify individual ﬁsh to speciﬁc populations with a high degree of conﬁdence, accurate and precise estimates of the relative contributions of regional stocks to simulated mixed stocks were

344

Isaac Wirgin and John R. Waldman

obtained. Furthermore, contributions to a population of two subpopulations within a drainage system were estimated accurately with this approach. While analysis of minisatellite loci revealed high levels of variation in nDNA, the uncertainty in assignment of bands to individual loci, the inability to make the assay PCR-based, and the need for high-quality DNA have resulted in the apparent abandonment of this approach in stock identiﬁcation.

B. MICROSATELLITES 1. What Are Microsatellites? Microsatellite loci contain tandemly repeated motifs of 1 to 6 bases that are found in all prokaryote and eukaryote genomes investigated to date. That means that a simple DNA motif, for example, AC, is arranged head-to-tail and in perfect repeats usually without interruption by any other motif or base (Fig. 17-2). Microsatellites that are used in stock identiﬁcation studies typically contain di(AC)n, tri-(ACC)n, or tetranucleotide (GATA)n repeats. There may be from 5 to 100 tandem copies of a repeat motif at a single microsatellite locus. Microsatellite loci are abundant in all eukaryote genomes and it has been estimated that there are from 103 to 105 microsatellite loci dispersed at 7- to 10-100 kilobase pair (kb) intervals in the eukaryotic genome (Wright and Bentzen, 1994). Fish genomes may contain more microsatellite loci than most other invertebrate and vertebrate taxa (Zane et al., 2002). Mapping studies suggest more or less even distributions of microsatellites throughout genomes, although they are somewhat rarer within coding sequences. 2. Why Use Microsatellites? Microsatellite analysis has become the tool of choice in studies of molecular evolution, genetic stock identiﬁcation, and genetic mixed stock analysis. This is primarily due to the very high levels of genetic variation that are often detected at individual microsatellite loci, the large number of loci that can be screened, and their relative ease of analysis, particularly with automated DNA sequencers. Microsatellite loci present higher levels of genetic diversity than in most, if not all, other types of DNA typically analyzed, which may be particularly true for ﬁsh. For example, an average of 13.7 alleles at AC dinucleotide repeat loci was found across 27 ﬁsh species compared to 8.6 alleles in mammals and 7.7 alleles in reptiles (Neff and Gross, 2001). Estimates of microsatellite mutation rates vary considerably among taxa and varieties of microsatellites (Zhang and Hewitt, 2003). From pedigree analyses, microsatellite mutation rates were 10-3 to 10-4 in mice (Dallas, 1992), 1 ¥ 10-3 to 5.6 ¥ 10-4 in humans (Weber and Wang, 1993),

Use of Nuclear DNA in Stock Identiﬁcation

345

Panel A 1) CTAGCTACTGGAACACACACACACACACTGACTAGGATCGA 2) CTAGCTACTGGAACCACCACCACCACCTGACTAGGATCGA 3) CTAGCTACTGGAACCTACCTACCTACCTACCTACCTTGACTAGGATCGA Panel B 1) CTAGCTACTGGAACACACACACACACACTGACTAGGATCGA 2) CTAGCTACTGGAACACACACACACACACACTGACTAGGATCGA 3) CTAGCTACTGGAACACACACACACACACACACTGACTAGGATCGA Panel C

M 1 2 3 4

FIGURE 17-2. (A). Three different types of microsatellite repeats: (1) dinucleotide, (2) trinucleotide, and (3) tetranucleotide. Repeat units are underlined, and single-copy ﬂanking sequences from which PCR primers are designed are indicated in bold. (B). Variation among individual ﬁsh in number of dinucleotide AC repeats at a single microsatellite locus. Individual 1 has 8 AC repeats, individual 2 has 9 AC repeats, and individual 3 has 10 AC repeats. Single-copy ﬂanking sequences are indicated in bold. (C). Depiction of gel showing polymorphic microsatellite alleles in the three individuals in (B) and a fourth individual who is a heterozygote for alleles with 8 and 10 AC repeats. Lane M contains a molecular weight ladder; lane 1 contains allele from a homozygote for 8 AC repeats; lane 2, a homozygote for 9 AC repeats; lane 3, a homozygote for 10 AC repeats; and lane 4, a heterozygote for alleles with 8 and 10 AC repeats.

346

Isaac Wirgin and John R. Waldman

and 1.5 ¥ 10-4 in zebraﬁsh (Shimoda et al., 1999) events per locus per generation. This compares to rates of SNP mutations that in most taxa are on the order of 10-8 to 10-10 per locus per generation (Hancock, 2001). Also, because allele variants at microsatellite loci are codominant and are inherited in Mendelian fashion, traditional approaches, such as Hardy–Weinberg, can be used in initial data analysis. This is not the case with dominant/recessive RAPDs in which band sharing is typically used to quantify genetic relatedness of samples or for haploid mtDNA markers. In stock identiﬁcation studies, the extent to which genotypes conform to Hardy–Weinberg equilibrium confers information about population structure, presence of mixed stocks, and those processes (admixtures, migration, selective pressure) that may have contributed to deviations from expected Hardy–Weinberg genotype frequencies. Multilocus microsatellite surveys are powerful tools in determining population structure and evolutionary history of target species. 3. How and Why Do Microsatellites Vary Among Individuals? Panels of microsatellite loci have been variable, and sometime hypervariable, in all taxa investigated to date. The number of copies of repeats at a microsatellite locus often differs among individuals within a population and serves as the basis for microsatellite allelic variation. The origin of such high levels of variability is probably due to frequent slippage events during DNA replication (Schlotterer and Tauritz, 1992). The functional signiﬁcance of variation at microsatellite loci is largely unknown; consequently, microsatellites have usually been considered to be neutral genetic markers. However, evidence has accumulated that microsatellites serve functional roles as coding or regulatory elements (Kashi and Soller, 2001). The conserved presence among species of particular microsatellite motifs in the regulatory regions of speciﬁc genes, binding of proteins to these microsatellite sequences, and alterations in gene expression when the microsatellites are artiﬁcially removed provides strong support that they are sometimes important in regulation of gene transcription. Additionally, instability in the number of tandem repeats at individual microsatellite loci in humans, particularly trinucleotide repeats (Eisen, 2001), has been shown to be associated with progression of various types of human diseases, including some cancers (Rubinsztein, 2001). In some diseases in humans and possibly other animals, the number of repeats at a locus increases during disease progression. 4. Are Levels of Polymorphisms at Microsatellite Loci Predictable? Within a taxon, microsatellite loci differ greatly in their molecular characteristics, and it is not clear what effects these have on their levels or patterns of variation. Characteristics in which individual microsatellite loci vary include the following:

Use of Nuclear DNA in Stock Identiﬁcation

347

the number of bases in the core tandem repeat unit (2, 3, or 4), the number of tandem repeats at an individual locus (5 to 100), the actual DNA sequence of the core nucleotide repeat motif, and the DNA sequence in single-copy regions that ﬂank the microsatellite loci. Tandem repeats at a locus may be perfect (CTCTCT), imperfect (CTCTACT), or compound (CTCTGAGA), characteristics that also may affect their levels of variation. Recombination rate, transcription rate, age and sex, and efﬁciency of mismatch DNA repair probably all contribute to the differing rates and patterns of variation among individual microsatellite loci. Although it is impossible to predict which loci will exhibit high or low levels of allelic diversity a priori based on sequence, studies in yeast suggest that longer, perfect repeats provide higher levels of allelic diversity than loci with shorter, imperfect repeats (Wierdel et al., 1997; Petes et al., 1997), but that has not been empirically veriﬁed for ﬁshes. 5. How to Conduct Microsatellite Studies Di-, tri-, and tetranucleotide repeat loci are the microsatellites that are almost always used in population studies. The shorter dinucleotide repeats are much more numerous in vertebrate genomes than the longer tri- and tetranucleotide repeats and, thus, are somewhat easier to isolate (Tooth et al., 2000). However, allelic variants at dinucleotide repeat loci are more difﬁcult to resolve in gels because of frequent “stuttering” in PCR ampliﬁcation (O’Reilly and Wright, 1995) and because of the similar size of the variant alleles (minimally 2 bases). Thus, in the long run, the isolation of loci containing the longer 3 and 4-base repeat units probably warrants the extra effort. Typically, at least 4 to 10 loci microsatellite loci are screened in population studies. Because most loci are unlinked and inherited independently, the greater the number of loci screened, the greater the likelihood of selecting loci that reveal signiﬁcant allelic frequency differences among populations. Additionally, analysis of a larger number of loci may provide a more accurate picture of the evolutionary history of the populations. Also, when the extent of population differentiation using microsatellites is weak, the greater the number of loci analyzed, the more statistical power is gained in individual-based population assignment tests (Paetkau et al., 1995; Bernatchez and Duchesne, 2000) that are now often used to quantify the extent of genetic differentiation among populations. Similarly, the number of informative loci used is important in assignment power in mixed-stock analyses. Analysis of microsatellite polymorphisms is a PCR-based approach in which oligonucleotide primers are designed based on unique single-copy sequences ﬂanking the microsatellite repeats. PCR primer pairs are selected such that PCR products are of small molecular size (usually <300 bp), providing relative ease in ampliﬁcation from low-quality DNAs and also allowing for distinguishing

348

Isaac Wirgin and John R. Waldman

small differences in the molecular size of alleles among individuals by using polyacrylamide-gel electrophoresis or automated DNA sequencers. Ideally, each individual shows a single (homozygote) or two-band (heterozygote) DNA pattern, with one band inherited from each parent. Polymorphic alleles at a locus are usually characterized by their molecular sizes. For dinucleotide repeats, these will differ by two base units. The mutational process at microsatellite loci is largely unknown, thereby complicating the analysis of microsatellite data. It is not known if mutation always occurs in single steps to the next larger or smaller number of tandem repeat units at a locus (Stepwise Mutation Model—SMM) or if the generation of an inﬁnite number of alleles at a locus are possible (Inﬁnite Allele Model—IAM). If mutation occurs in single steps, is this process symmetrical so that gains or losses of single repeat units occur with equal probability? Empirical evidence indicates that mutations at microsatellite loci probably follow both scenarios and, therefore, applications of both models in data analyses may be most appropriate in assessing the distribution of genetic diversity within and between populations. If the SMM model applies, then alleles that are similar in size are of more recent common ancestry than if alleles are more dissimilar in size (i.e., number of tandem repeat units). If so, models upon which data analysis is founded can use this information in determining the relatedness of populations. It is fairly certain that there is a ﬁnite number of tandem repeats at an individual locus (although this number differs among loci); thus, homoplasy is almost certainly a common occurrence at individual loci (Estoup et al., 2002). Therefore, alleles that are identical in molecular size are not necessarily identical by descent due to convergent mutation(s). This information is also important in statistical analysis of microsatellite data. The relative sensitivity of different approaches in detecting genetic population structure is in part dependent on levels of genetic variation encountered. The higher the levels of variation, the greater the likelihood in detecting genetic dissimilarity among populations, although very high levels of polymorphism including many private alleles do not ensure the detection of signiﬁcant population differences. The popularity of microsatellite analysis stems in large part from the moderate to high levels of allelic diversity observed at individual microsatellite loci in almost all species investigated to date. For example, mean number of alleles at individual microsatellite loci in populations of freshwater (n = 13 species), anadromous (n = 7 species) and marine (n = 12 species) ﬁshes were 7.5, 11.3, and 20.6, respectively (DeWoody and Avise, 2000). For comparison, in striped bass, allozymes proved almost monomorphic in many Atlantic and Paciﬁc coast populations (Waldman et al., 1988), levels of base substitutions in mDNA were exceedingly low (Wirgin et al., 1989), single-copy, anonymous nDNA loci exhibited a maximum of two alleles/locus (Wirgin and Maceda, 1991; Wirgin et al., 1997) and band sharing in multilocus DNA ﬁngerprints was so high

Use of Nuclear DNA in Stock Identiﬁcation

349

that single DNA fragments were used in distinguishing some populations (Wirgin et al., 1991). In contrast, up to 33 alleles were observed at single microsatellite loci, and the mean number of alleles detected at four loci in striped bass from the Hudson River and four Canadian rivers was 19 (Robinson et al., 2004). 6. How to Isolate Microsatellites One of the major drawbacks to the use of microsatellites has been the need to develop a suite of PCR primers for each new taxonomic group investigated due to the fact that microsatellite loci are usually found in noncoding regions and, therefore, single-copy sequences ﬂanking the repeats used to design PCR primers evolve rapidly—often prohibiting the use of primers that for other markers may be conserved across taxa. This is unlike PCR-based analysis of mtDNA that often offers the luxury of the use of universal primers across diverse animal taxa (Kocher et al., 1989). For nDNA, RAPDs and AFLP analyses are two methods that also obviate the need to develop species-speciﬁc PCR primers by instead using very short, random primers which, because of their brevity, anneal many times to multiple sequences within any genome. In contrast, analysis of microsatellites requires the use of species-speciﬁc (or if lucky, genus-speciﬁc) PCR primers. In birds and cattle, a 50% success rate in cross-ampliﬁcation and polymorphism detection was reported in species that diverged 10 to 20 Ma (Primmer et al., 1996; Moore et al., 1991). In cross-ampliﬁcation tests among 10 species of sturgeon from 3 genera, 80% of PCR ampliﬁcations were successful at 6 loci isolated from Atlantic sturgeon, Acipenser oxyrinchus (King et al., 2001). Although microsatellite primers developed for one species may amplify diagnostic loci in closely related species, there is no guarantee that a locus that is highly polymorphic in one species will be likewise in a second species. For example, about 70% of the successful locus-species ampliﬁcations among the 10 sturgeon species were polymorphic (King et al., 2001). Therefore, microsatellite analysis often requires that microsatellite loci be identiﬁed, isolated, and characterized in the species of interest or a closely related taxon. This involves considerable time (3–6 months), effort, and expertise in molecular biology, although commercial vendors are available who will customize this service at a signiﬁcant cost for previously unexplored target species. Although PCR primers from closely related species often successfully amplify microsatellite loci, results usually are not as reliable and levels of genetic diversity not as high as when species-speciﬁc primers are used. However, because of the recent prevalency of microsatellite analysis, loci have been identiﬁed and PCR primers developed for a surprisingly large number and diversity of ﬁsh taxa. For example, microsatellite arrays have been isolated, characterized, and used in stock identiﬁcation studies in many families of anadromous and marine ﬁshes, including acipenseridae (Smith et al., 2002; King et al., 2001), clupeidae (O’Connell et al.,

350

Isaac Wirgin and John R. Waldman

1998), coregonidae (Patton et al., 1997), gadidae (Brooker et al., 1994; Lage and Kornﬁeld, 1999; Miller et al., 2000; O’Reilly et al., 2002), moronidae (Roy et al., 2000; Ely et al.), merlucidae (Moran et al. 1999), pleuronectidae (Bouza et al., 2002; Wirgin, 2003), salmonidae (Estoup et al., 1993; Slettan et al., 1993; Morris et al., 1996; McConnell et al., 1995), scombridae (Gold et al., 2002), and scorpaenidae (Buonaccorsi et al., 2002), among others. In short, this process involves creating a genomic DNA library with small (<1 kb) DNA fragments from the species of interest, isolating recombinant DNA clones (containing ﬁsh DNA fragments) from the library, and identifying those clones with microsatellite DNA sequences by hybridization with labeled DNA probes containing multiple microsatellite repeats. Usually, only 2% to 3% of clones from the library are found to harbor microsatellite loci (Zane et al., 2002). Single-copy DNA sequences immediately ﬂanking tandem repeats within the microsatellite loci are characterized by DNA sequencing, and these ﬂanking sequences are then used to design oligonucleotide primers that can reproducibly be used to PCR amplify the intervening microsatellite repeats. Primer speciﬁcity to complementary sequence is critical, such that in PCR only alleles from a single locus are ampliﬁed. Each set of PCR primers then must be empirically tested for its reproducibility and speciﬁcity in amplifying the microsatellite locus. Because levels of allelic diversity differ dramatically among individual microsatellite loci for unknown reasons, each locus is then screened by PCR with locus-speciﬁc primers in a small number of ﬁsh from the populations of interest to empirically determine if it reveals the requisite levels of genetic variation for the objective of the study. Once diagnostic loci have been identiﬁed, PCR conditions are empirically optimized for that locus by evaluating different experimental conditions in the ampliﬁcation process or perhaps through redesign of the primers. Today, PCRs of microsatellites are usually multiplexed such that two to ﬁve different microsatellite loci are ampliﬁed in a single PCR reaction and analyzed simultaneously. One primer from each locus pair is ﬂuorescently labeled, resulting in the product at each locus having a different color, or, alternatively, the molecular sizes of the products of some loci in the multiplex differ such that they do not overlap. The multilocus PCR reaction products are then loaded together in the gel or capillaries of the sequencer and its detection system allows discrimination of the allelic products of each locus by color or molecular size. Individual alleles at a locus differ in molecular size because of their varying numbers of tandem repeat units; gel or capillary-based sequencers provide the exact size of each allele at each locus. Sufﬁcient numbers of individuals from populations of interest are then screened at the 4 to 10 loci chosen for study, allelic composition at each locus is determined for each individual in the study, and genotype and allelic frequencies are compiled for each collection and statistically analyzed for heterogeneity. The more loci screened, the greater the conﬁdence in the population structure that is determined. In fact, the number of loci scored

Use of Nuclear DNA in Stock Identiﬁcation

351

may be more critical than increased allelic diversity per locus in studies of population assignment (Bernatchez and Duchesne, 2000). Because of the large effort needed in the isolation, optimization, and testing of microsatellites from new species under investigation, this may be a major drawback to their use in some population studies (Campbell et al., 2003).

VI. CASE HISTORY: GENETIC ANALYSES OF ATLANTIC COD

A. HOW DIFFICULT

A

PROBLEM?

Delineation of Atlantic cod stocks has been attempted for more than four decades with a variety of biochemical and molecular approaches. In fact, some of the earliest successful stock identiﬁcation studies using biochemical approaches were performed on cod populations in the Northeast Atlantic. The extent of genetic differentiation of ﬁsh stocks is highly dependent on the amount of gene ﬂow among stocks, their evolutionary age, and the size of populations. In this sense, cod present many of the same problems in delineating stock structure as seen with most marine species, that is, the potential for gene ﬂow is very high, some populations are probably relatively recently established because of postPleistocene colonization in portions of the species’ range, and cod are highly fecund and historically stock sizes were very large. Embryos and larval life stages are subject to passive transport by surface ocean currents and gyres to sites distant from where they were spawned. In cod, this problem may be particularly severe given their prolonged hatching (50–60 days at -1.5° to 0°C) and larval development times (Templeman, 1981). Additionally, mark-recapture studies of both European and North American populations have demonstrated that adult cod tend to exhibit extensive movements (Rasmussen, 1959; Templeman, 1974), including trans-Atlantic migrations (Gulland and Williamson, 1962). For example, in large tagging programs conducted over ﬁve decades in Newfoundland, a moderate percentage of returns were from the distant Scotian Shelf area (Taggart et al., 1995), thus offering the possibility of gene ﬂow among populations. A comparison of the results of genetic stock identiﬁcation studies in cod can be informative in identifying the advantages and limitations of the different molecular approaches in deﬁning stock structure.

B. ALLOZYME

AND

BLOOD PROTEIN STUDIES

In some instances, allozyme and blood protein analyses of a limited number of loci revealed signiﬁcant differences among cod populations (Cross and Payne,

352

Isaac Wirgin and John R. Waldman

1978), particularly among northeastern Atlantic stocks (Moller, 1968, 1969, Dahle and Jorstad, 1993). Baltic and North Sea populations and Norwegian coastal and Arctic stocks were distinguishable by allelic frequency differences at blood protein loci, such as hemoglobin and transferrin. However, when populations were sampled across the species’ range at larger numbers of allozyme loci, these differences were not seen (Mork et al., 1985). Similarly, Pogson et al. (1995) failed to ﬁnd signiﬁcant heterogeneity at 8 of 10 allozyme loci surveyed among 6 cod collections, of which 2 were from North America, 1 from Iceland, and 3 from northern Europe. Mean levels of gene ﬂow among these collections was high at Nm = 17.7 (a measure of gene ﬂow), a level sufﬁciently high to prevent the establishment of signiﬁcantly distinct genetic stocks. These results with allozymes suggested that gene ﬂow among most cod populations was extensive except for selected populations along the coast of northern Europe, or, that the effects of balancing selection had homogenized allelic frequencies at these functionally important protein loci. Also, these results illustrate the point that some genetic markers may be effective in distinguishing some populations, but not others.

C. RFLP

AND

SEQUENCE ANALYSIS

OF MTDNA

Using RFLP analysis of the entire mtDNA molecule, Smith et al. (1989) failed to ﬁnd signiﬁcant differences between even northeastern and northwestern Atlantic coast cod populations. Both RFLP analysis of mtDNA and direct sequence analysis of the mtDNA cytochrome b gene and mtDNA control region have been reported for cod from a variety of stocks in North America, northern Europe, and Iceland. Cytochrome b within mtDNA is moderately conserved and portions of the control region are the most rapidly evolving sequences within the mitochondrial genome. Using direct sequence analysis of cytochrome b, Carr and Marshall (1991) identiﬁed a relatively large number of mtDNA polymorphisms, but these variant haplotypes were in such low frequencies that they were of little utility in distinguishing Canadian stocks within the northern cod complex in Newfoundland (Pepin and Carr, 1993). Similarly, mtDNA cytochrome b haplotype frequencies were almost identical between bay and offshore collections on the Newfoundland continental shelf (Carr et al., 1995). Similarly, Arnason and Palsson (1996) conducted sequence analysis of mtDNA cytochrome b from a limited number of Norwegian cod from nine sampling localities representing three overall areas: Arctic, coastal, and middle. They found greater differences among individuals within populations than between populations, leading to the conclusion that net interlocality nucleotide divergence in mtDNA cytochrome b in these areas was absent. Also, Arnason et al. (2000) failed to detect signiﬁcant haplotype frequencies in mtDNA cytochrome b sequences among locales in Iceland or between stocks in Iceland and West Greenland despite the presence

Use of Nuclear DNA in Stock Identiﬁcation

353

of relatively high levels of mtDNA haplotype diversity. Even though high levels of haplotype and nucleotide diversity at synonymous sites were found, a different region of cytochrome b and an intergenic spacer in mtDNA also failed to reveal signiﬁcant spatial heterogeneity among cod collected from localities around the Faroe Islands (Sigurgislason and Arnason, 2003). Using direct sequence analysis of the more variable mtDNA control region, Arnason et al. (1992) found little evidence of population differentiation among cod from Iceland, Norway, and Newfoundland. In contrast, sequence analysis of mtDNA cytochrome b showed promise in distinguishing stocks in the northeastern Atlantic and Barents Sea (Carr and Crutcher, 1998). In summary, despite sufﬁcient levels of sequence variation and an effective molecular population size one-fourth that of allozymes, mtDNA was of little value in distinguishing most cod populations, except for some in the northeastern Atlantic, indicating that contemporary and/or historical gene ﬂow among populations must be great or that the portion of the mtDNA genome under study was under balancing selection. Thus, results with mtDNA and allozymes were congruent in that both successfully discriminated some stocks along the northern European coast, but failed to detect signiﬁcant population structure elsewhere.

D. SINGLE-COPY CODING NDNA ANALYSIS As mentioned previously, single-copy coding nDNA polymorphisms have also been used to distinguish cod stocks and to evaluate the importance of stochastic and selective processes in molding stock structure. One of the 11 cDNA clones originally used by Pogson et al. (1995) in Southern blot hybridization analysis of cod population structure turned out to code for a gene originally thought to be synaptophysin [but later determined to be the closely related pantophysin gene (PanI) (Pogson et al., 1995)]. PanI is involved in a variety of shuttling, secretory, and endocytotic recycling pathways. In further studies, Fevolden and Pogson (1997) reported strong differences among northeastern Atlantic cod stocks at the PanI locus, perhaps due to prolonged natural selection. Their results were consistent with those of Moller (1968, 1969), who found differences between Arctic and Norwegian coastal cod in hemoglobin and blood type E protein polymorphisms. In contrast, analysis of these stocks using allozyme (Mork et al., 1985; Mork and Giaever, 1999) and sequence analysis of mtDNA cytochrome b failed to detect signiﬁcant differences between Arctic and Norwegian coastal stocks or among individual Norwegian coastal stocks (Arnason and Palsson, 1996). In combination, these results suggested that the pronounced allelic differences observed at PanI may result from strong postsettlement selection pressure each generation, rather than historical processes that isolated populations. To distinguish between historical isolation vs. natural selection as the prime determinants

354

Isaac Wirgin and John R. Waldman

in sculpting structure between and within coastal and Arctic cod populations, Pogson and Fevolden (2003), using sequence analysis, compared geographic variation within PanI at nucleotides that are the direct targets of selection (nonsynonymous sites at which mutations lead to amino acid changes) and those that are not [synonymous and noncoding sites within or immediately ﬂanking PanI (silent sites)]. Surprisingly, they found signiﬁcant differences in the frequencies of silent substitutions at PanI among coastal sites. They concluded that recent diversifying selection had resulted in differentiation between Arctic and coastal populations, but historical isolation had also occurred among local stocks of cod along the Norwegian coast. Pogson (2001) also compared variation at PanI in cod from ﬁve geographically more distant locales: Nova Scotia, Newfoundland, Iceland, Balsfjord, and the Barents Sea. He found large differences in the frequencies of two PanI alleles among populations but could not attribute this to a simple model of spatially varying selection. Instead, some form of balancing selection for both PanI alleles was occurring within all populations. Jonsdottir et al. (2001) also observed signiﬁcant spatial differences at the PanI locus between cod collected from two spawning locales off southern Iceland and temporal stability between collections from two different years. In contrast, no variation was observed between the two locales using protein electrophoresis of hemoglobin.

E. SINGLE-COPY NONCODING NDNA ANALYSIS Using anonymous nDNA probes developed from an Atlantic cod coding DNA (cDNA) library and Southern blot hybridization analysis, Pogson et al. (1995) investigated the genetic structure among six cod populations previously studied by Mork et al. (1985) using allozyme analysis. The cDNA was made from messenger RNA and, therefore, it represented expressed (coding) DNA. However, the coding DNA (cDNA) probes hybridized to both expressed (coding) and unexpressed (noncoding) intron and ﬂanking nDNA sequences. Of the 17 cDNA probe/restriction enzyme combinations tested, 11 showed polymorphisms among the 6 Atlantic cod populations (2 North American, 1 Icelandic, 3 European). Of the 11 independent cDNA loci, 10 exhibited signiﬁcant heterogeneity among populations. Unlike allozyme studies, highly signiﬁcant differences were observed among all populations using these nDNA markers in a pattern consistent with an isolation by distance model. Estimates of gene ﬂow among these 6 populations using this nDNA approach were an order of magnitude lower than seen with allozyme analysis. Additionally, the magnitude of population differences revealed at these anonymous nDNA loci was signiﬁcantly higher than that observed at the allozyme loci. This study clearly revealed that levels of SNPs within the cod genome were substantial and proved promising in revealing signiﬁcant genetic population structure among North American sampling locales. However, its

Use of Nuclear DNA in Stock Identiﬁcation

355

reliance on Southern blot analysis required the isolation of relatively large amounts of highly intact DNA that would not be available from typical sample collections. Of course, it would be quite easy to make analyses of these nDNA loci PCR-based and thus relax the demands on DNA quantity and quality.

F. MINISATELLITE ANALYSIS The ﬁrst study of nDNA in cod used the DNA ﬁngerprinting approach. Dahle (1994) used the human-derived minisatellites probes, 33.6 and 33.15, to examine population structure among collections of Arctic cod and one sample of Norwegian coastal cod from a production pond. Both probes successfully hybridized to genomic DNA from cod and provided complex, individual-speciﬁc DNA banding patterns. The authors concluded “that because individual loci could be not be unambiguously distinguished by this method, multilocus DNA ﬁngerprinting has no advantage over allozyme analysis in population genetic studies.” Galvin et al. (1995) attempted to rectify this problem by developing a PCR-based assay to screen for allelic variation at a single minisatellite locus, Mmer-AMP2 that was previously isolated and characterized from a related gadoid, whiting, Merlangius merlangus. He applied this approach to collections of cod from two European (northern Norway and Irish Sea) and two Canadian locales (Scotian Shelf and Newfoundland.) Banding patterns were easy to interpret, unlike those observed with multilocus ﬁngerprinting, and a reasonable number of alleles (22) were seen among the 119 ﬁsh analyzed. Samples from Europe and North America showed highly signiﬁcant differences in allelic composition. The problem with the application of the single-locus minisatellite approach to other taxa is the difﬁculty, compared to microsatellites, in isolating the number of minisatellite loci needed for an intensive population study.

G. MICROSATELLITE ANALYSIS As mentioned previously, suites of microsatellite loci have been isolated from Atlantic cod and characterized by two groups of Canadian investigators. Initially, investigators at the Marine Gene Probe Lab at Dalhousie University isolated 64 microsatellite loci from an Atlantic cod genomic DNA library (Brooker et al., 1994) and characterized genetic diversity at a subset of these among Canadian cod stocks. Of these, ﬁve to six of the most hypervariable loci were selected for use in intensive screening of Canadian stocks. Highlights of these population studies have included the genetic discrimination of adult northern cod aggregations on offshore banks such as the Flemish Cap, southern reaches of the northern cod complex on Grand Banks, and northern reaches of the northern cod complex on Hamilton/Belle Isle and Funk Island Banks (Bentzen et al., 1996;

356

Isaac Wirgin and John R. Waldman

Ruzzante et al., 1998). Despite the catastrophic decline of cod stocks at spawning banks off Newfoundland and Labrador, and a two orders of magnitude variation in population size, these workers found long-term stability over three decades of genetic stock structure revealed by these microsatellite loci (Ruzzante et al., 2001). Analyses failed to reveal signiﬁcant allelic differences between contemporary samples collected in the 1990s compared to those in archived otoliths from cod sampled in the 1960s and 1970s. They hypothesized that if recovery does occur, it will be by population regrowth from historical stocks rather than migratory inﬂux from elsewhere. Studies by these same investigators also revealed signiﬁcant allelic differences between these adult northern cod populations (Newfoundland and Labrador) and those off the southern and eastern coasts of Nova Scotia (Ruzzante et al., 1998). More intensive analysis of this southern stock revealed signiﬁcant allelic differences among adult cod from Browns Bank, the Bay of Fundy, and the northeast peak of Georges Bank (Ruzzante et al., 1998, 1999). These studies deﬁnitively demonstrated that offshore banks in Canadian waters that are separated by submarine channels and trenches and at which spawning times for cod differ harbor genetically distinct populations of adult cod. Other studies investigated the extent of genetic differentiation between inshore and offshore wintertime aggregations of adult cod in Newfoundland. Inshore cod in Newfoundland exhibited signiﬁcantly higher levels of plasma antifreeze glycoprotein levels than cod from offshore banks (Goddard et al., 1994), thus providing a physiological mechanism potentially isolating these two aggregations. Microsatellite analysis provided evidence that cod overwintering in inshore Newfoundland bays with high blood antifreeze levels are genetically distinguishable from cod overwintering at the offshore northern Grand Banks, whereas inshore cod with low blood antifreeze levels are not (Ruzzante et al., 1996). Temporal stability of allelic frequencies at diagnostic loci over a 4-yr period in the inshore populations was also demonstrated (Taggart et al., 1998). Thus, intrinsic biological characteristics can serve as a mechanism to reproductively isolate geographically proximal or seemingly continuous cod populations. Studies also explored the genetic relationships among spatially or temporally separated larval cod aggregations at an offshore bank in Canadian waters. Intensive sampling of cod larvae over a 3-week period on the Western Bank of the Scotian Shelf provided evidence of several genetically distinct groups within the aggregation. Allelic differences within this larval aggregation segregated on the basis of age at length. These results suggest that the larval aggregation on the Western Bank originated from several distinct spawning events involving parents with heterogeneous allelic compositions (Ruzzante et al., 1996). Sibling relationships among these cod larvae were also investigated. Results demonstrated that these cohorts of cod larvae appeared to have come from large genetic pools and were not siblings (Herbinger et al., 1997).

Use of Nuclear DNA in Stock Identiﬁcation

357

Microsatellite analysis at these loci was also used to determine the origin of wintertime aggregations of cod in the Cabot Strait approaches to the Gulf of St. Lawrence. Potential source spawning populations included those in the southern Gulf of St. Lawrence, northern Gulf of St. Lawrence, Cape Breton Island, and other northern offshore banks described previously. Mixed-stock analysis revealed that cod spawned within the southern and northern Gulf of St. Lawrence contributed 46% to 71% of the ﬁsh sampled in the Cabot Strait wintertime collection and that contributions were temporally stable over a 2-yr period (Ruzzante et al., 1999, 2000). Contributions of other individual stocks did not exceed 13% to 14%. A second set of seven tri-and tetranucleotide repeat microsatellite loci was later developed by a second Canadian group (Miller et al., 2000). These loci, along with the PanI locus, were analyzed in 5,230 cod from 19 collections made in inshore bays and offshore grounds within the northern cod complex in Labrador and Newfoundland. Similar to earlier studies, they too found that inshore and offshore collections in the northern cod complex were distinct. However, with the exception of the collection from Gilbert Bay, Labrador, little difference was seen among collections from individual bays. Signiﬁcant structuring of stocks was observed among offshore samples, with at least three distinct offshore populations identiﬁed. These results agree with those of earlier studies (Bentzen et al., 1996; Ruzzante et al., 1998; Taggart et al., 1998) with the exception that no support for separate bay stocks was found except for that in Gilbert Bay, and that an isolation-by-distance population structure for cod in Newfoundland was invoked rather than a strict inshore–offshore division. Microsatelle analysis was also applied to cod from European waters usually using combinations of the microsatellite loci isolated by the two Canadian laboratories. Similar to some of the early blood protein studies, signiﬁcant genetic differentiation was found among many collections, sometimes even those from geographically proximate sites. For instance, Hutchinson et al. (2001) used 5 microsatellite loci (including 3 that were newly developed) on cod from 12 samples from around the United Kingdom to report the presence of 3 or 4 genetically distinct populations within the North Sea and a single stock off the southern Cornish coast. All of these European populations were signiﬁcantly different from those at the Canadian Scotian Shelf and Barents Sea. The stock in the Barents Sea was more divergent from the 12 European collections than was the Scotian Shelf stock despite its greater geographic proximity, and it exhibited less genetic diversity than other stocks. Knutsen et al. (2003) tested whether spawning adult cod collected from 6 sites along 300 km of the Norwegian Skagerrak coast exhibited signiﬁcant genetic diversity at 10 microsatellite loci. Collection sites were roughly equidistant, being about 60 km apart. Although ﬁsh were collected in or just off small islands and fjords, the predominant ocean current sweeps along this coast and could serve

358

Isaac Wirgin and John R. Waldman

as a possible vector for the passive drift of eggs and larvae. Despite this opportunity for admixture, highly signiﬁcant differentiation of all six collections was observed, with weak but signiﬁcant differentiation at all 10 loci. Using a combination of nine microsatellite loci, Nielsen et al. (2003) observed signiﬁcant population differentiation between cod from the North Sea and the Baltic Sea. Using an individual admixture model, they found intermediate genotypes in all samples from a transition zone between the two nonadmixed populations. Signiﬁcant gametic phase disequilibrium across loci suggested that the western Baltic Sea was the center of a hybrid, rather than a mechanical mixing zone, between cod from the high-saline North Sea and less-saline Baltic Sea populations. Very recently, microsatellite analysis at nine loci was used to distinguish cod from the three principal populations in the northeastern Atlantic Ocean, the North Sea, the Baltic Sea, and the northeastern Arctic Ocean—and to attempt to assign individual ﬁsh to their population of origin (Nielsen et al., 2001). Strong genetic differentiation among the three populations was observed and temporal stability of differentiation was conﬁrmed. The authors claimed that the probabilities of assigning individuals to the populations from which they originated and were sampled ranged from 97% to 100%. Analysis of historical patterns of abundance, behavior, and migration of cod from many locations in the inshore Gulf of Maine suggested the presence of at least four distinct subpopulations of cod, although contemporary population structure may be altered by extirpation of some spawning aggregations (Ames, 2004). However, Wirgin (2004), using seven microsatellite loci, observed only slight, although sometimes statistically signiﬁcant differences in allelic frequencies between cod from Georges Bank and sites in the inshore Gulf of Maine or among sites in Cape Cod Bay and Massachusetts Bay. Using a subet of these loci and Pan E’age et al. (2004) found signiﬁcant genetic differention between adult cod from Georges Bank and Nantucket Shoals, but not between Georges Bank and Browns Bank. Additionally, no signiﬁcant differences in microsatellite allelic frequencies were observed between winter and spring spawning aggregations of cod from an inshore site in the Gulf of Maine (Berlinsky and Kovach, personal communication). Microsatellite studies on Atlantic haddock Melanogrammus aegleﬁnus at offshore banks in the Northwest Atlantic off southern Canada (Browns Bank and Scotian Shelf) and the United States (northeastern Georges Bank and Nantucket Shoals), using three haddock-speciﬁc (Lage and Kornﬁeld, 1999) and one cod-speciﬁc locus, failed to ﬁnd evidence of discrete populations among the Browns Bank, Scotian Shelf, and Georges Banks sites, although the Nantucket Shoals samples were genetically discrete from all others (Lage et al., 2001). Thus, it is possible that the stock structure of groundﬁsh along the coast of southern Canada and the northeastern United States may not be as pronounced as among northern stocks off Newfoundland and Labrador due to bottom topography or prevalence of gyres, which serve to mix young life stages.

Use of Nuclear DNA in Stock Identiﬁcation

359

In summary, microsatellite analysis demonstrated that benthic topography, topographically induced eddies, and physiological adaptations may have induced signiﬁcant allelic frequency differentiation among cod in Canadian waters. Surprisingly, the same magnitude of genetic discontinuities was not observed among hypothesized cod stocks in U.S. waters despite their maintenance of allelic diversity/heterozygosity that was slightly greater than observed in cod from Canadian waters.

VII. CONCLUSIONS Our review of the history of biochemical and molecular stock identiﬁcation of Atlantic cod illustrates a number of points important to genetic stock identiﬁcation and to the use of nDNA in particular. The overall picture of stock structure of this species in the North Atlantic is that of regional sculpting to prevailing oceanographic conditions; that is, with the complicated bottom morphology of the northeastern Atlantic, cod show stronger partitioning into local stocks than do northwestern cod, which exist in perhaps an overall more homogeneous environment. Nonetheless, there are exceptions in the Northwest Atlantic where stock structure is indicated, including some that correlate with physiological differences that stem from different overwintering conditions. And even within an area there may be migratory stocks with large ranges and less migratory and, even highly localized stocks with small ranges. However, given this broad pattern, different approaches sometimes yielded different results when applied to the same putative stocks. It is important to keep in mind that some of these differences may stem simply from subtle differences among the specimens sampled despite attempts to represent the same putative stocks; that is, detectable differences in genetic characteristics may originate from variations among studies in year-classes, sizes, sex ratios, and dates and locations sampled. Also, larger sample sizes may, on their own, produce statistical signiﬁcance in borderline situations. But real differences do exist in the relative sensitivities of available approaches. Allozyme, blood protein, and mtDNA studies all failed to reveal the extent of stock structure in Atlantic cod as demonstrated with newer approaches, in particular microsatellite analysis. However, the spectrum of approaches applied to northwestern Atlantic cod appears to be greater than northeastern Atlantic cod because the older approaches yielded mostly satisfactory results in the northeast, but not in the northwest. This highlights the point made by Waldman (this volume) that negative results are not conclusive inasmuch as approaches that offer greater sensitivity may yield otherwise hidden stock structure. The issue in situations in which stock structure is suspected, but not found, is where to stop the search. For species of low importance to humans, economics often dictate a halt.

360

Isaac Wirgin and John R. Waldman

However, we believe that for species of such major value as cod, the most sensitive approaches will almost always be tried. Because of its wide spectrum of sensitivity and high number of potential markers available, analysis of nDNA now ranks among the most important ﬁsh stock identiﬁcation techniques. In order to gain a more precise understanding of the particular levels of sensitivities of the various nDNA approaches, there is a need for more comparative analyses among the nDNA approaches and among nDNA, mtDNA, and allozyme approaches. These comparative analyses would be most useful if they were conducted on the same specimens (Waldman et al., 1997; Waldman, 1999), which removes the effects of that potentially important confounding variable and, if not, then at minimum, with the same stocks. Fortunately, there is growing recognition of the value of redundancy of stock identiﬁcation approaches, and in particular, the complementarity of mtDNA and nDNA analyses (Begg and Waldman, 1999). Recent examples of the latter include Patton et al. (1997), Hansen et al. (1999), and Wirgin et al. (2002). Moreover, the analyses of multiple genetic loci, single-copy and repetitive nDNA and mtDNA in the same study provide a more comprehensive picture of population history, additional loci for individual-based population discrimination, and the ability to tease out the relative importance of selective and stochastic processes in molding population structure. The use of a suite of nonlinked nDNA markers in the elucidation of a genome-wide picture of population structure has been highly recommended for studies in human and other taxa (Brumﬁeld et al., 2003). In this regard, introns are nDNA sequences that are fertile for population studies, but very little attention has been paid to date on their use in genetic stock identiﬁcation. We feel that this should change in the future. It has already been demonstrated that introns offer relatively high levels of polymorphisms in ﬁsh (Quattro and Jones, 1999) and other aquatic taxa (Hare et al., 2002), although variant sites will usually be biallelic as opposed to multiallelic. In intraspeciﬁc comparisons, their mode of mutation will almost always be single nucleotide substitutions and, unlike microsatellites homoplasy, will almost never occur at these loci. In addition, PCR analysis of intronic polymorphisms can easily be developed at multiple introns within a single or a suite of different genes by using exon sequences conserved among species to design PCR primers whose ampliﬁcation products will be introns (Hassan et al., 2002). As analysis of several ﬁsh genomes are completed, conserved sequences at individual nDNA genes of known function can easily be identiﬁed and used to design PCR primers. Also, many toxicologically relevant genes have already been identiﬁed and characterized in multiple ﬁsh taxa, and these too can be used in the design of primers to amplify intron sequences. Intron sequences then can be analyzed by RFLP, or preferably DNA sequencing analysis, at least initially to identify informative SNP sites. For example, Fevolden and Pogson (1997) successfully used PCR and RFLP analysis

Use of Nuclear DNA in Stock Identiﬁcation

361

of an intron at the previously described PanI locus to distinguish between Northeast Arctic and Norwegian coastal cod and possibly among resident populations in different Norwegian fjords. Recent technical achievements have ushered in a time when nDNA and mtDNA approaches can routinely and accurately be applied to the large number of specimens needed in ﬁsheries management. Once diagnostic markers have been identiﬁed and their population applicability evaluated in a discovery mode, rapid screening in a production mode of diagnostic markers on large numbers of specimens can be highly automated, resulting in very short turnaround times (within one to several days). This potential rapid turnaround time for data makes these approaches useful for management schemes that are both dynamic and ﬂexible. Because all of these approaches are PCR-based, they allow for the analysis of very small tissue samples that are, even to a great extent, degraded. Thus, very young life stages (embryos and larvae) and air-dried, archived scales or other hard body parts can supply sufﬁcient amounts of suitable DNA for multilocus population studies. Noninvasively acquired ﬁn clips or any other tissue allows for investigations on threatened species.

VIII. SUMMARY An arsenal of molecular tools now exists for use in genetic stock identiﬁcation. Fish genomes are comprised primarily of nuclear DNA (nDNA) and, to a far less extent, mitochondrial DNA (mtDNA) (Table 17-1). Methods to intensively and reproducibly analyze large numbers of samples in both genomes are now available. Use of the polymerase chain reaction (PCR) permits the application of DNAbased techniques to the large number of specimens and early life stages that need to be routinely analyzed in ﬁsheries applications. Genetic approaches are used in stock identiﬁcation to discover DNA sequences that are polymorphic within a species, that exhibit signiﬁcant differences in the frequencies of their alleles or genotypes among populations, that can routinely and very accurately be screened to describe the architecture of population structure, and can be used to estimate the contributions of individual populations to ﬁsheries which harvest individuals from multiple populations. Within each genome there are areas that are highly functional (coding for proteins) while others are under no, or far less evolutionary constraint (noncoding), resulting in areas of DNA sequence that evolve rapidly and others that are much slower in their rates of change. Most nDNA is comprised of single-copy DNA sequences, but interspersed are many loci that contain repetitive DNA motifs that are very short (microsatellites) or somewhat longer (minisatellites) in length. Repetitive nDNA sequences have been frequently used successfully in genetic stock identiﬁcation studies, and single-copy nDNA sequences have attracted far less attention. Although minisatellites are effective

362

Table 17-1 A Comparison of Genetic Methods Used in Fish Stock Identiﬁcation

Technique

a

Function

Data interpretation

Tissue demands

Low

Known

Easy

Stringent

No

High

Low

Moderate Low–high

Unknown Known

Easy Easy

Stringent Relaxed

Not usually Yes

High High

Moderate Low or higha

High High Low Moderate Very high

Unknown Unknown Known Unknown Unknown

Difﬁcult Difﬁcult Easy Easy Difﬁcult

Relaxed Relaxed Relaxed Relaxed Stringent

Yes Yes Yes No No, but can bed

Low Moderate High High Moderate

Low Low Lowb High or Lowc High

Very high

Unknown

Easy

Relaxed

Yes

High

Low or highe

PCR-based

Reproducibility

Low for RFLP analysis; higher for sequence analysis. Low if taxon-speciﬁc primers can be developed from sequences conserved among other taxa. c High to discover informative polymorphic sites; low after assay has been made PCR-based. d No, but assay can be made PCR-based at individual minisatellite loci. e Low if taxon-speciﬁc PCR primers have already been developed; high if a battery of taxon-speciﬁc PCR primers need to be developed. b

Relative Cost

Isaac Wirgin and John R. Waldman

Allozymes and other proteins Mitochondrial DNA Whole genome RFLP Selected regions sequencing or RFLP Nuclear DNA RAPDs AFLPs Single-copy coding Singe-copy noncoding Minisatellites (DNA ﬁngerprinting) Microsatellites

Genetic variation

Use of Nuclear DNA in Stock Identiﬁcation

363

in elucidating population structure, because of technical limitations they are no longer favored in efforts to identify stocks. Microsatellite analysis is often the method of choice in the description of the structure of ﬁsh stocks because of the high levels of genetic variation that is revealed, data reliability, and the ﬂexibility that it sometimes offers in data analysis. However, there are instances when the complementary use of mtDNA and single-copy nDNA analyses offers additional power and certainty in the discrimination of stock structure or mixed-stock estimates beyond that provided by microsatellites alone. We present the more than 40-yr case history of stock identiﬁcation efforts of Atlantic cod, Gadus morhua, to illustrate the advantages and limitations of these approaches.

ACKNOWLEDGMENTS We acknowledge the support of NIEHS Center Grant ES00260, Marﬁn Project 99-MaR-011, and the Hudson River Foundation for Science and Environmental Research, Inc.

REFERENCES Ames, E. P. 2004. Atlantic cod structure in the Gulf of Maine. Fisheries 29: 10–28. Armour, J. A. L., Alegre, S. A., Miles, S., Williams, L. J., and Badge, R. M. 2001. Minisatellites and mutation processes in tandemly repetitive DNA. In D. B. Goldstein and C. Schlotterer (eds.), Microsatellites: Evolution and Applications. Oxford University Press, New York, pp. 24–33. Arnason, E. and Palsson, S. 1996. Mitochondrial cytochrome b DNA sequence variation of Atlantic cod, Gadus morhua, from Norway. Molecular Ecology 5: 715–724. Arnason, E., Palsson, S., and Arason, A. 1992. Gene ﬂow and lack of population differentiation in Atlantic cod, Gadus morhua L., from Iceland and comparison of cod from Norway and Newfoundland. Journal of Fish Biology 40: 751–770. Arnason, E., Petersen, P. H., Kristsinsson, K., Sigurgislason, H., and Palsson, S. 2000. Mitochondrial cytochrome b DNA sequence variation of Atlantic cod from Iceland and Greenland. Journal of Fish Biology 56: 409–430. Beacham, T. D., Brattey, J., Miller, K. M., Le, K. D., and Withler, R. E. 2002. Multiple stock structure of Atlantic cod (Gadus morhua) off Newfoundland and Labrador determined from genetic variation. ICES Journal of Marine Science 59: 650–665. Beacham, T. D., Withler, R. E., and Wood, C. C. 1995. Stock identiﬁcation of sockeye salmon by means of minisatellite DNA variation. North American Journal of Fisheries Management 15: 249–265. Begg, G. A. and Waldman, J. R. 1999. An holistic approach to ﬁsh stock identiﬁcation. Fisheries Research 43: 35–44. Bentzen, P., Taggart, C. T., Ruzzante, D. E., and Cook, D. 1996. Microsatellite polymorphism and the population structure of Atlantic cod (Gadus morhua) in the northwest Atlantic. Canadian Journal of Fisheries and Aquatic Sciences 53: 2706–2721. Bentzen, P., Taylor, E. B., and Wright, J. M. 1993. A novel synthetic probe for DNA ﬁngerprinting salmonid ﬁshes. Journal of Fish Biology 43: 313–316.

364

Isaac Wirgin and John R. Waldman

Bernatchez, L. and Duchesne, P. 2000. Individual-based genotype analysis in studies of parentage and population assignment: how many loci, how many alleles? Canadian Journal of Fisheries and Aquatic Sciences 57: 1–12. Bouza, C., Presa, P., Castro, J., Sanchez, L., and Martinez, P. 2002. Allozyme and microsatellite diversity in natural and domestic populations of turbot (Scopthalmus maximus) in comparison with other Pleuronectiformes. Canadian Journal of Fisheries and Aquatic Sciences 59: 1460– 1473. Brooker, A. L., Cook, D., Bentzen, P., Wright, J. M., and Doyle, R. W. 1994. Organization of microsatellites differs between mammals and cold-water teleost ﬁshes. Canadian Journal of Fisheries and Aquatic Sciences 51: 1959–1966. Brown, K. M., Baltazar, G. A., Weinstein, B. N., and Hamilton, M. B. 2003. Isolation and characterization of nuclear microsatellite loci in the anadromous marine ﬁsh Morone saxatilis. Molecular Ecology Notes 3: 414–416. Brown, W. M., George, Jr. M., and Wilson, A. C. 1979. Rapid evolution of animal mitochondrial DNA. Proceedings of the National Academy of Sciences USA. 76: 1967–1971. Brumﬁeld, R. T., Beerli, P., Nickerson, D. A., and Edwards, S. V. 2003. The utility of single nucleotide polymorphisms in inferences of population history. Trends in Ecology and Evolution 18: 249–256. Buonaccorsi, V. P., Kimbrell, C. A., Lynn, E. A., and Vetter, R. D. 2002. Population structure of copper rockﬁsh (Sebastes caurinus) reﬂects postglacial colonization and contemporary patterns of larval dispersal. Canadian Journal of Fisheries and Aquatic Sciences 59: 1374–1384. Campbell, D., Duchesne, P., and Bernatchez, L. 2003. AFLP utility for population assignment studies: analytical investigation and empirical comparison with microsatellites. Molecular Ecology 12: 1979–1991. Cargill, M. and 16 coauthors. 1999. Characterizaton of single-nucleotide polymorphisms in coding regions of human genes. Nature Genetics 22: 231–238. Carr, S. M. and Crutcher, D. C. 1998. Population genetic structure in Atlantic cod (Gadus morhua) from the North Atlantic and Barents Sea: contrasting or concordant pattern in mtDNA sequence and microsatelllite data? In I. Hunt von Herbling, I. Kornﬁeld, M. Tupper, and J. Wilson (eds.), The Implications of Localized Fishery Stocks. Northeast Regional Agricultural Engineering Service, New York, pp. 91–103. Carr, S. M. and Marshall, H. D. 1991. A direct approach to the measurement of genetic variation in ﬁsh populations: applications of the polymerase chain reaction to studies of Atlantic cod, Gadus morhua L. Journal of Fish Biology 39 (suppl. A): 101–107. Carr, S. M., Snellen, A. J., Howse, K. A., and Wroblewski, J. S. 1995. Mitochondrial DNA sequence variation and genetic stock structure of Atlantic cod (Gadus morhua) from bay and offshore locations on the Newfoundland continental shelf. Molecular Ecology 4: 79–88. Castelli, M., Philippart, J.-C., Vassart, G., and Georges, M. 1990. DNA ﬁngerprinting in ﬁsh: a new generation of genetic markers. American Fisheries Society Symposium 7: 514–520. Cross, T. F. and Payne, R. H. 1978. Geographic variation in Atlantic cod, Gadus morhua, off Eastern North America: a biochemical systematics approach. Journal of the Fisheries Research Board of Canada 35: 117–123. Cutler, M. G., Bartlett, S. E., Hartley, S. E., and Davidson, W. S. 1991. A polymorphism in the ribosomal RNA genes distinguishes Atlantic salmon (Salmo salar) from North America and Europe. Canadian Journal of Fisheries and Aquatic Sciences 48: 1655–1661. Dahle, G. 1994. Minisatellite DNA ﬁngerprints of Atlantic cod (Gadus morhua). Journal of Fish Biology 44: 1089–1092. Dahle, G. and Jorstad, K. E. 1993. Haemoglobin variation in cod: a reliable marker for Arctic cod (Gadus morhua L.). Fisheries Research 16: 301–311. Dallas, J. F. 1992. Estimation of microsatellite mutation rates in recombinant inbred strains of mouse. Mammalian Genome 3: 452–456.

Use of Nuclear DNA in Stock Identiﬁcation

365

DeWoody, J. A. and Avise, J. C. 2000. Microsatellite variation in marine, freshwater, and anadromous ﬁshes compared with other animals. Journal of Fish Biology 56: 461–473. Dowling, T. E., Moritz, C., and Palmer, J. D. 1990. Nucleic acids II: restriction site analysis. In D. M. Hillis and C. Moritz (eds.), Molecular Systematics. Sinauer Associates, Sunderland, MA, pp. 250–317. Eisen, J. A. 2001. Mechanistic basis for microsatellite instability. In D. B. Goldstein and C. Schlotterer (eds.), Microatellites: Evolution and Applications. Oxford University Press, New York, pp. 34–48. Estoup, A., Jame, P., and Cornuet, J.-M. 2002. Homoplasy and mutation model at microsatellite loci and their consequences for population genetics analysis. Molecular Ecology 11: 1591–1604. Estoup, A., Presa, P., Krieg, F., Vaiman, D., and Guyomard, R. 1993. (CT)n and (GT)n microsatellites: a new class of genetic markers for Salmo trutta L. (brown trout). Heredity 71: 488–496. Fevolden, S. E. and Pogson, G. H. 1997. Genetic divergence at the synaptophysin (SypI) locus among Norwegian coastal and north-east Arctic populations of Atlantic cod. Journal of Fish Biology 51: 895–908. Galvin, P., Sadusky, T., McGregor, D., and Cross, T. 1995. Population genetics of Atlantic cod using ampliﬁed single locus minisatellite VNTR analysis. Journal of Fish Biology 47 (suppl. A): 200–208. Goddard, S. V., Wroblewski, J. S., Taggart, C. T., Howse, K. A., Bailey, W. L., Kao, M. H., and Fletcher, G. L. 1994. Overwintering of adult Northern Atlantic cod (Gadus morhua) in cold inshore waters as evidenced by plasma antifreeze glycoprotein levels. Canadian Journal of Fisheries and Aquatic Sciences 51: 2834–2842. Gold, J. R., Pak, E., and DeVries, D. A. 2002. Population structure of king mackerel (Scomberomorus cavalla) around peninsular Florida, as revealed by microsatellite DNA. U.S. National Marine Fisheries Service Fishery Bulletin 100: 491–509. Gulland, J. A. and Williamson, G. R. 1962. Transatlantic journey of tagged cod. Nature 195: 921. Han, K., Li, L., Leclerc, G. M., Hays, A. M., and Ely, B. 2000. Isolation and characterization of microsatellite loci for striped bass (Morone saxatilis). Marine Biotechnology 2: 405–408. Hancock, J. M. 2001. Microsatellites and other simple sequences: genomic context and mutational mechanisms. In D. B. Goldstein and C. Schlotterer (eds.), Microsatellites: Evolution and Applications. Oxford University Press, New York, pp. 1–9. Hansen, M. M., Mensberg, K.-L. D., and Berg, S. 1999. Postglacial recolonization patterns and genetic relationships among whiteﬁsh (Coregonus sp.) populations in Denmark, inferred from mitochondrial DNA and microsatellite markers. Molecular Ecology 8: 239–252. Hare, M. P. 2001. Prospects for nuclear gene phylogeography. Trends in Ecology and Evolution. 16: 700–716. Hare, M. P., Cipriano, F., and Palumbi, S. R. 2002. Genetic variation on the deomography of speciation in allopatric dolphin species. Evolution 56: 804–816. Hassan, M., Lemaire, C., Fauvelot, C., and Bonhomme, F. 2002. Seventeen new exon-primed introncrossing polymerase chain reaction ampliﬁable introns in ﬁsh. Molecular Ecology Notes 2: 33–340. Herbinger, C. M., Doyle, R. W., Taggart, C. T., Lochmann, S. E., Brooker, A. L., Wright, J. M., and Cook, D. 1997. Family relationships and effective population size in a natural cohort of Atlantic cod (Gadus morhua) larvae. Canadian Journal of Fisheries and Aquatic Sciences 54 (suppl. 1): 11–18. Hutchinson, W. F., Carvalho, G. R., and Rogers, S. I. 2001. Marked genetic structuring in localised spawning populations of Gadus morhua in the North Sea and adjoining waters, as revealed by microsatellites. Marine Ecology-Progress Series 223: 251–260. The International HapMap Consortium. 2003. The international HapMap project. Nature 46: 789–796.

366

Isaac Wirgin and John R. Waldman

International SNP MAP Working Group. 2001. A map of human genome sequence variation containing 1.42 million single nucleotide polymorphisms. Nature 409: 928–933. Jeffreys, A. J., Wilson, V., and Thein, S. L. 1985. Hypervariable minisatellite regions in human DNA. Nature 314: 67–73. Jonsdottir, O. D. B., Danielsdottir, A. K., and Naevdal, G. 2001. ICES Journal of Marine Science 58: 114–122. Karl, S. A. and Avise, J. C. 1992. Balancing selection at allozyme loci in oysters: implications from nuclear RFLPs. Science 256: 100–102. Kashi, Y. and Soller, M. 2001. Functional roles of microsatellites and minisatellites. In D. B. Goldstein and C. Schotterer (eds.), Microsatellites: Evolution and Applications. Oxford University Press, New York, pp. 10–25. King, T. L., Lubinski, B. A., and Spidle, A. P. 2001. Microsatellite DNA variation in Atlantic sturgeon (Acipenser oxyrinchus oxyrinchus) and cross-species ampliﬁcation in the Acipenseridae. Conservation Genetics 2: 103–119. Knutsen, H., Jorde, P. E., Andre, C., and Stenseth, N. C. 2003. Fine-scaled geographical population structuring in a highly mobile marine species: the Atlantic cod. Molecular Ecology 12: 385– 394. Kocher, T. D., Thomas, W. K., Meyer, A., Edwards, S. V., Paabo, S., Villablanca, F. X., and Wilson, A. C. 1989. Dynamics of mitochondrial DNA evolution in animals: ampliﬁcation and sequencing with conserved primers. Proceedings of the National Academy of Sciences of the USA 86: 6196–6200. Lage, C. R. and Kornﬁeld, I. 1999. Isolation and characterization of microsatellite loci in Atlantic haddock (Melanogrammus aegleﬁnus). Molecular Ecology 8: 1351–1362. Lage, C., Kuhn, K., and Kornﬁeld, I. 2004. Genetic differentiation among Atlantic cod (Gadus morhua) from Browns Bank, Georges Bank, and Nantucket Shoals. Fishers Bulletin 102: 289–297. Lage, C., Purcell, M., Fogarty, M., and Kornﬁeld, I. 2001. Microsatellite evaluation of haddock (Melanogrammus aegleﬁnus) stocks in the northwest Atlantic Ocean. Canadian Journal of Fisheries and Aquatic Sciences 58: 982–990. Ludwig, A., Debus, L., Lieckfeldt, D., Wirgin, I., Benecke, N., Jenneckens, I., Williot, P., Waldman, J. R., and Pitra, C. 2002. When the American Atlantic sturgeon went east. Nature 419: 448– 449. McConnell, S. K., O’Reilly, P., Hamilton, L., Wright, J. M., and Bentzen, P. 1995. Polymorphic microsatellite loci from Atlantic salmon (Salmo salar): genetic differentiation of North American and European populations. Canadian Journal of Fisheries and Aquatic Sciences 52: 1863–1872. Miller, L. M. and Kapuscinski, A. R. 1997. Historical analysis of genetic variation reveals low effective population size in a northern pike (Esox lucius) population. Genetics 147: 1249–1258. Miller, K. M., Le, K. D., and Beacham, T. D. 2000. Development of tri-and tetranucleotide repeat microsatellite loci in Atlantic cod. Molecular Ecology 9: 238–239. Miller, K. M., Withler, R. E., and Beacham, T. D. 1996. Stock identiﬁcation of coho salmon (Oncorhynchus kisutch) using minisatellite DNA variation. Canadian Journal of Fisheries and Aquatic Sciences 53: 181–195. Moller, D. 1968. Genetic diversity in spawning cod along the Norwegian coast. Hereditas 60: 1–32. Moller, D. 1969. The relationship between Arctic and coastal cod in their immature stages illustrated by frequencies of genetic characters. Fiskeridirektoratets Skrifter, Serie Havundersokelser 15: 220–233. Moore, S. S., Sargent, L. L., King, T. J., Mattick, J. S., Georges, M., and Hetzel, J. S. 1991. The conservation of dinucleotide microsatellites among mammalian genomes allows the use of heterologous PCR primer pairs in closely related species. Genomics 10: 654–660. Moran, P., Lundy, C., Rico, C., and Hewitt, G. M. 1999. Isolation and characterization of microsatellite loci in European hake, Merlucius merlucius (Merlucidae, Teleostei). Molecular Ecology 8: 1357–1362.

Use of Nuclear DNA in Stock Identiﬁcation

367

Mork, J. and Giaever, M. 1999. Genetic structure of cod along the coast of Norway: results from isozyme studies. Sarsia 84: 157–168. Mork, J., Ryman, N., Stahl, G., Utter, F., and Sundness, G. 1985. Genetic variation in Atlantic cod (Gadus morhua) throughout its range. Canadian Journal of Fisheries and Aquatic Sciences 42: 1580–1587. Morris, D. B., Richard, K. R., and Wright, J. M. 1996. Microsatellites from rainbow trout (Oncorhynchus mykiss) and their use for genetic study of salmonids. Canadian Journal of Fisheries and Aquatic Sciences 53: 120–126. Moriyama E. N., and Powell, J. R. 1996. Intraspeciﬁc nuclear DNA variation in Drosophila. Molecular Biology and Evolution 13: 261–277. Neff, B. D. and Gross, M. R. 2001. Microsatellite evolution in vertebrates: inference from AC dinucleotide repeats. Evolution 55: 1717–1733. Nielsen, E. E., Hansen, M. M, Ruzzante, D. E., Meldrup, D., and Gronkjaer, P. 2003. Evidence of a hybrid-zone in Atlantic cod (Gadus morhua) in the Baltic and the Danish Belt Sea revealed by individual admixture analysis. Molecular Ecology 12: 1497–1508. Nielsen, E. E., Hansen, M. M., Schmidt, C., Meldrup, D., and Gronkjaer, P. 2001. Population of origin of Atlantic cod. Nature 413: 272. Nielsen, J. L., Gan, C., Wright, J. M., Thomas, W. K., and Morris, D. B. 1994. Biogeographic distributions of mitochondrial and nuclear markers for southern steelhead. Molecular Marine Biology and Biotechnology 3: 281–293. O’Connell, M., Dillon, M. C., Wright, J. M., Bentzen, P., Merkouris, S., and Seeb, J. 1998. Genetic structuring among Alaskan Paciﬁc herring populations using microsatellite variation. Journal of Fish Biology 53: 150–163. O’Reilly, P. T., McPherson, A. A., Kenchington, E., Taggart, C., Jones, M. W., and Bentzen, P. 2002. Isolation and characterization of tetranucleotide microsatellites from Atlantic haddock (Melanogrammus aegeleﬁnus). Marine Biotechnology 4: 418–422. O’Reilly, P. and Wright, J. M. 1995. The evolving technology of DNA ﬁngerprinting and its application to ﬁsheries and aquaculture. Journal of Fish Biology 47 (suppl. A): 29–55. Paetkau, D., Calvert, W., Stirling, I., and Strobeck, C. 1995. Microsatellite analysis of population structure in Canadian polar bears. Molecular Ecology 4: 347–354. Patton, J. C., Galloway, B. J, Fechhelm, R. G., and Cronin, M. A. 1997. Genetic variation of microsatellite and mitochondrial DNA markers in broad whiteﬁsh (Coregonus nasus) in the Colville and Sagavanirktok rivers in northern Alaska. Canadian Journal of Fisheries and Aquatic Sciences 54: 1548–1556. Pepin, P. and Carr, S. M. 1993. Morphological, meristic, and genetic analysis of stock structure in juvenile Atlantic cod (Gadus morhua) from the Newfoundland shelf. Canadian Journal of Fisheries and Aquatic Sciences 50: 1924–1933. Petes, T. D., Greenwell, P. W., and Dominska, M. 1997. Stabilization of microsatellite sequences by variant repeats in the yeast Saccharomyces cerevisiae. Genetics 146: 491–498. Phillips, R. B. and Pleyte, K. A. 1991. Nuclear DNA and salmonid phylogenetics. Journal of Fish Biology 39 (suppl. A): 259–275. Phillips, R. B., Pleyte, K. A., and Brown, M. R. 1992. Salmonid phylogeny inferred from ribosomal DNA restriction maps. Canadian Journal of Fisheries and Aquatic Sciences 49: 2345–2353. Pogson, G. H. 2001. Nucleotide polymorphism and natural selection at the pantophysin (PanI) locus in the Atlantic cod, Gadus morhua (L.). Genetics 157: 317:330. Pogson, G. H. and Fevolden, S.-E. 2003. Natural selection and the genetic differentiation of coastal and Arctic populations of the Atlantic cod in northern Norway: a test involving nucleotide sequence variation at the pantophysin (PanI) locus. Molecular Ecology 12: 63–74. Pogson, G. H., Mesa, K. A., and Boutilier, R. G. 1995. Genetic population structure and gene ﬂow in the Atlantic cod, Gadus morhua: a comparison of allozyme and nuclear RFLP loci. Genetics 139: 375–385.

368

Isaac Wirgin and John R. Waldman

Pogson, G. H., Taggart, C. T., Mesa, K. A., and Boutilier, R. G. 2001. Isolation by distance in the Atlantic cod, Gadus morhua, at large and small geographic scales. Evolution 55: 131–146. Primmer, C. R., Borge, T., Lindell, J., and Saetre, G. P. 2002. Single-nucleotide polymorphism characterization in species with limited available sequence information: high nucleotide diversity revealed in the avian genome. Molecular Ecology 11: 603–612. Primmer, C. R., Moller, A. P., and Ellegren, H. 1996. A wide-ranging survey of cross-species ampliﬁcation in birds. Molecular Ecology 5: 365–378. Prodohl, P. A., Taggart, J. B., and Ferguson, A. 1992. Genetic variability within and among sympatric brown trout (Salmo trutta) populations: multilocus DNA ﬁngerprint analysis. Hereditas 117: 45–50. Quattro, J. M. and Jones, W. J. 1999. Ampliﬁcation primers that target locus-speciﬁc introns in Actinopterygian ﬁshes. Copeia. 1999: 171–176. Quinn, T. W. and White, B. N. 1987. Identiﬁcation of restriction fragment-length polymorphisms in genomic DNA of the lesser snow goose (Anser caerulescens caerulescens). Molecular Biological Evolution 4: 126–143. Rasmussen, B. 1959. On the migration pattern of the West Greenland stock of cod. Annals of Biology Copenhagen 4: 123–124. Robinson, M., Courtenay, S., Benfey, T., Maceda, L., and Wirgin, I. 2004. Origin and movements of young-of-the-year striped bass in the southern Gulf of St. Lawrence, New Brunswick. Transactions of the American Fisheries Society 133: 411–425. Rubinsztein, D. C. 2001. Trinucleotide expansion mutations cause diseases which do not conform to classical Mendelian expectations. In D. B. Goldstein and C. Schlotterer (eds.), Microsatellites: Evolution and Applications. Oxford University Press, New York, pp. 80–97. Ruzzante, D. E. 1998. A comparison of several measures of genetic distance and population structure with microsatellite data: bias and sampling variance. Canadian Journal of Fisheries and Aquatic Sciences 55: 1–14. Ruzzante, D. E., Taggart, C. T., and Cook, D. 1996. Spatial and temporal variation in the genetic composition of a larval cod (Gadus morhua) aggregation: cohort contribution and genetic stability. Canadian Journal of Fisheries and Aquatic Sciences 53: 2695–2705. Ruzzante, D. E., Taggart, C. T., and Cook, D. 1998. A nuclear DNA basis for shelf-and bank-scale population structure in northwest Atlantic cod (Gadus morhua): Labrador to Georges Bank. Molecular Ecology 7: 1663–1680. Ruzzante, D. E., Taggart, C. T., and Cook, D. 1999. A review of the evidence for genetic structure of cod (Gadus morhua) populations in the NW Atlantic and population afﬁnities of larval cod off Newfoundland and the Gulf of St. Lawrence. Fisheries Research 43: 79–97. Ruzzante, D. E., Taggart, C. T., Cook, D., and Goddard, S. 1996. Genetic differentiation between inshore and offshore Atlantic cod (Gadus morhua) off Newfoundland: microsatellite DNA variation and antifreeze level. Canadian Journal of Fisheries and Aquatic Sciences 53: 634–645. Ruzzante, D. E., Taggart, C. T, Doyle, R. W., and Cook, D. 2001. Stability in the historical pattern of genetic structure of Newfoundland cod (Gadus morhua) despite the catastrophic decline in population size from 1964 to 1994. Conservation Genetics 2: 257–269. Ruzzante, D. E., Taggart, C. T., Lang, S., and Cook, D. 1999. Mixed-stock analysis of Atlantic cod (Gadus morhua) in the Cabot Strait approaches to the Gulf of St. Lawrence: a microsatellite DNA application. Canadian Stock Assessment Secretariat Research Document 99/55. Ruzzante, D. E., Taggart, C. T., Lang, S., and Cook, D. 2000. Mixed-stock analysis of Atlantic cod near the Gulf of St. Lawrence based on microsatellite DNA. Ecological Applications 10: 1090–1109. Schotterer, C. and Tauritz, D. 1992. Slippage synthesis of simple sequence DNA. Nucleic Acids Research 20: 211–215.

Use of Nuclear DNA in Stock Identiﬁcation

369

Shimoda, N., Knapik, E. W., and Ziniti, J. 1999. Zebraﬁsh genetic map with 2000 microsatellite markers. Genomics 58: 219–232. Sigurgislason, H. and Arnason, E. 2003. Extent of mitochondrial DNA sequence variation in Atlantic cod from the Faroe Islands: a resolution of gene genealogy. Heredity 91: 557–564. Slettan, A., Olsaker, I., and Lic, O. 1993. Isolation and characterization of (GT)n repetitive sequences from Atlantic salmon, Salmo salar L. Animal Genetics 24: 195–197. Smith, P. J., Birley, A. J., Jamieson, A., and Bishop, C. A. 1989. Mitochondrial DNA in the Atlantic cod, Gadus morhua: lack of genetic divergence between eastern and western populations. Journal of Fish Biology 34: 369–373. Smith, C. T., Nelson, R. J., Pollard, S., Rubidge, E., McKay, S. J., Rodzen, J., May, B., and Koop, B. 2002. Population genetic analysis of white sturgeon (Acipenser transmontanus) in the Fraser River. Journal of Applied Ichthyology 18: 307–312. Stevens, T. A., Withler, R. E., Goh, S. H., and Beacham, T. D. 1993. A new multilocus probe for DNA ﬁngerprinting in chinook salmon (Oncorhynchus tshawytscha), and comparisons with a single locus probe. Canadian Journal of Fisheries and Aquatic Sciences 50: 1559–1567. Taggart, J. B. and Ferguson, A. 1990. Hypervariable minisatellite DNA single locus probes for the Atlantic salmon, Salmo salar L. Journal of Fish Biology 37: 991–993. Taggart, C. T., Penney, P., Barrowman, N., and George, C. 1995. The 1954–1993 Newfoundland codtagging data base: statistical summaries and spatial-temporal distributions. Canadian Technical Report on Fisheries and Aquatic Sciences No. 2042. Taggart, C. T., Ruzzante, D. E., and Cook, D. 1998. Localized stocks of cod (Gadus morhua L.) in the northwest Atlantic: the genetic evidence and otherwise. In I. Hunt von Herbing, I. Kornﬁeld, M. Tupper, and J. Wilson (eds.), The Implication of Localized Fishery Stocks. Northeast Regional Agricultural Engineering Service, New York, pp. 65–90. Taggart, J. B., Verspoor, E., Galvin, P. T., Morán, P., and Ferguson, A. 1995. A minisatellite DNA marker for discriminating between European and North American Atlantic salmon (Salmo salar). Canadian Journal of Fisheries and Aquatic Sciences 52: 2305–2311. Taylor, E. B., Beacham, T. D, and Kaeriyama, M. 1994. Population structure and identiﬁcation of North Paciﬁc Ocean chum salmon (Oncorhynchus keta) revealed by an analysis of minisatellite DNA variation. Canadian Journal of Fisheries and Aquatic Sciences 51: 1430– 1442. Templeman, W. 1974. Migrations and intermingling of Atlantic cod (Gadus morhua) stocks of the Newfoundland area. Journal of the Fisheries Research Board of Canada. 31: 1073–1092. Templeman, W. 1981. Vertebral numbers in Atlantic cod, Gadus morhua, of the Newfoundland and adjacent areas, 1947–1971, and their use in delineating cod stocks. Journal of Northwest Atlantic Fisheries Sciences 2: 21–45. Tooth, G., Gaspari, Z., and Jurka, J. 2000. Microsatellites in different eukaryotic genomes: survey and analysis. Genome Research 10: 967–981. Waldman, J. R. 1999. The importance of comparative studies in stock analysis. Fisheries Research 43: 237–246. Waldman, J. R., Grossﬁeld, J., and Wirgin, I. 1988. Review of stock discrimination techniques for striped bass. North American Journal of Fisheries Management 8: 410–425. Waldman, J. R., Nolan, K., Hart, J., and Wirgin, I. I. 1996. Genetic differentiation of three key anadromous ﬁsh populations of the Hudson River. Estuaries 19: 759–768. Waldman, J. R., Richards, R. A., Schill, W. B., Wirgin, I. Wigin, and Fabrizio, M. C. 1997. An empirical comparison of stock identiﬁcation techniques applied to striped bass. Transactions of the American Fisheries Society 126: 369–385. Waldman, J. R. and Wirgin, I. 1994. Use of DNA analyses in the management of natural ﬁsh populations. In S. J. Garte (ed.), Molecular Environmental Biology, Lewis Publishers, Boca Raton, FL, pp. 29–64.

370

Isaac Wirgin and John R. Waldman

Waples, R. S. and Smouse, P. E. 1990. Gametic disequilibrium analysis as a means of identifying mixtures of salmon populations. American Fisheries Society Symposium 7: 439–458. Weber, J. L. and Wang, C. 1993. Mutation of human short tandem repeats. Human Molecular Genetics 2: 1123–1128. Wierdel, M., Dominska, M., and Petes, T. D. 1997. Microsatellite instability in yeast: dependence on the length of the microsatellite. Genetics 146: 769–779. Wirgin, I. 2003. Stock identiﬁcation and mixed-stock analyses of winter ﬂounder (Pleuronectes americanus). Final report to New York Sea Grant, Project Number R/XG-13. Stony Brook, New York. Wirgin, I. 2004. Stock structure of Atlantic cod in U.S. waters based on microsatellite analysis. Final report to the Marﬁn Program, Project Number 99-MaR-011. National Marine Fisheries Service, Gloucester, MA. Wirgin, I. I., Grunwald, C., Garte, S. J., and Mesing, C. 1991. Use of DNA ﬁngerprinting in the identiﬁcation and management of a striped bass population in the southeastern United States. Transactions of the American Fisheries Society 120: 273–282. Wirgin, I. I. and Maceda, L. 1991. Development and use of striped bass-speciﬁc RFLP probes. Journal of Fish Biology 39 (suppl. A): 159–167. Wirgin, I. I., Maceda, L., and Mesing, C. 1992. Use of cellular oncogene probes to identify Morone hybrids. Journal of Heredity 83: 375–382. Wirgin, I. I., Proenca, R., and Grossﬁeld, J. 1989. Mitochondrial DNA diversity among populations of striped bass in the southeastern United States. Canadian Journal of Zoology 67: 891–907. Wirgin, I. I., Waldman, J. R., Maceda, L., Stabile, J., and Vecchio, V. J. 1997. Mixed-stock analysis of Atlantic coast striped bass using nuclear DNA and mitochondrial DNA markers. Canadian Journal of Aquatic Sciences 54: 2814–2826. Wirgin, I. I., and Waldman, J. R. 1994. What DNA can do you for you. Fisheries 19(7): 16–27. Wirgin, I. I., Waldman, J. R., Maceda, L., Stabile, J., and Vecchio, V. J. 1997. Mixed-stock analysis of Atlantic coast striped bass using nuclear DNA and mitochondrial DNA markers. Canadian Journal of Aquatic Sciences 54: 2814–2826. Wirgin, I., Waldman, J., Stabile, J., Lubinski, B., and King, T. 2002. Comparison of mitochondrial DNA control region sequence and microsatellite DNA analyses in estimating population structure and gene ﬂow rates in Atlantic sturgeon Acipenser oxyrinchus. Journal of Applied Ichthyology 18: 313–319. Wright, J. M. and Bentzen, P. 1994. Microsatellites: genetic markers for the future. Reviews in Fish Biology and Fisheries 4: 384–388. Xu, S., Kobak, C. J., and Smouse, P. E. 1994. Constrained least squares estimation of mixed population stock composition from mtDNA haplotype frequency data. Canadian Journal of Fisheries and Aquatic Sciences 51: 417–425. Yauk, C. 1998. Monitoring for induced heritable mutations in natural populations: application of minisatellite DNA screening. Mutation Research-Review in Mutation Research. 411: 1–10. Zane, L., Bargelloni, L., and Patarnello, T. 2002. Strategies for microsatellite isolation: a review. Molecular Ecology 11: 1–16. Zhang, D.-X. and Hewitt, G. M. 2003. Nuclear DNA analyses in genetic studies of populations: practice, problems and prospects. Molecular Ecology 12: 563–584. Zhuo, L. 1991. Cytogenetic and molecular analysis of the ribosomal RNA genes in lake trout. Doctoral dissertation. University of Wisconsin, Milwaukee, Wisconsin. Zhuo, L., Reed, K. M., and Phillips, R. B. 1995. Hypervariability of ribosomal DNA at multiple chromosomal sites in lake trout (Salvelinus namaycush). Genome 38: 487–496.

CHAPTER

18

Random Ampliﬁed Polymorphic DNA (RAPD) P. J. SMITH National Institute of Water and Atmospheric Research Ltd., Wellington, New Zealand

I. II. III. IV. V. VI. VII.

Introduction DNA Extraction, Ampliﬁcation, and Separation DNA Fragment Scoring and Data Analyses Inheritance of Presence/Absence Fragments Species Identiﬁcation and Taxonomy Stock Discrimination Summary References

I. INTRODUCTION The development of the polymerase chain reaction (PCR), which ampliﬁes DNA, enables analyses to be carried out on very small quantities of tissue and offers a range of tools for the detection of genetic variation without the need for cloning and sequencing. Random ampliﬁed polymorphic DNA, or RAPD, is a method described independently by Williams et al. (1990) and Welsh and McClelland (1990), for the identiﬁcation of plant cultivars. Welsh and McClelland (1990) termed the technique AP-PCR, for arbitrarily chosen primers, but it has become known by the acronym RAPD. The RAPD technique allows the detection of DNA polymorphisms by amplifying randomly chosen regions of DNA by PCR with single arbitrary primers. Any section of DNA ﬂanked by a pair of primer sites, and less than ~5,000 base pairs apart, will be ampliﬁed by the RAPD technique. The ampliﬁed products are separated by gel electrophoresis and detected by direct staining with ethidium bromide or silver nitrate. The RAPD technique requires no speciﬁc probes and does not use radioisotopes. The comparative ease with which a large number of DNA primers can be screened for polymorphism has made the RAPD technique an attractive population tool. Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

371

372

P. J. Smith

RAPD primers detect polymorphisms caused by point mutations, which allow or disallow primer binding to the sample DNA, and by insertions or deletions between primer binding sites. Different primers bind to different DNA sites and thereby detect different RAPD polymorphisms. The primers do not discriminate between coding and noncoding regions and therefore sample the genome more randomly than tools such as allozymes, targeting proteins, or sequencing of small subregions of mitochondrial DNA. Initial applications of RAPDs focused on distinguishing plant cultivars (e.g., Hu and Quiros, 1991) and strains of laboratory mice (Welsh et al., 1991), but the potential of the technique was quickly recognized in invertebrate taxonomy (Black et al., 1992; Hadrys et al., 1992) and for the identiﬁcation of ﬁsh species (Dinesh et al., 1993). RAPDs have subsequently been applied to a range of marine and freshwater ﬁshes and invertebrates for both species and stock discrimination (see summary in Table 18-1).

II. DNA EXTRACTION, AMPLIFICATION, AND SEPARATION DNA can be extracted from fresh, frozen, and ethanol-preserved ﬁsh tissues and dried scales following standard DNA protocols (e.g., Kessing et al., 1990; Hillis and Moritz, 1990). DNA samples are ampliﬁed separately with oligonucleotide primers. Usually a random selection of 10-base primers, with a G + C content of 60% to 70%, is used for ampliﬁcation; use of A + T rich primers generally does not produce ampliﬁcation products (Nadeau et al., 1992; Patwary et al., 1993). More than 1,000 different 10-base primers are available ex stock from commercial manufacturers such as Operon Technologies in California. Use of oligonucleotide primers as short as ﬁve nucleotides in length has been shown to produce complex banding patterns, and the method has been termed DNA ampliﬁcation ﬁngerprinting, or DAF (Caetano-Anolles et al., 1991). Ampliﬁcation conditions and temperature proﬁles are chosen according to the type of thermocycler used (Devos et al., 1992; Penner et al., 1993). Serial dilutions of DNA samples are tested ﬁrst to determine optimum DNA concentration for ampliﬁcation. Ideally, DNA concentrations should be standardized with a spectrophotometer across all samples to ensure consistency among ampliﬁcations. In a typical ampliﬁcation, each reaction contains the template or sample DNA in 10 mM Tris HCl, pH 8.3, a single 10-base primer, KCl, MgCl2, equal volumes of deoxynucleotide triphosphates (dATP, dCTP, dGTP, and dTTP), and Taq DNA polymerase in PCR buffer. Control samples are ampliﬁed without DNA template. The thermocycler is programmed for 30 to 40 cycles of 1 min at 90–95°C to denature the DNA, 1 min at 35°–40°C to allow the primers to anneal to their complementary sequence, followed by a step of around 2 min at 70°–75°C

TABLE 18-1. Summary of Random Ampliﬁed Polymorphic DNA Population Studies on Commercially Important Fishes and Aquatic Invertebrates.a

Number of primers

Number of variable primers

Number of populations; specimens

Bangladesh

60

7

3; 87

Swedish lakes

20

3

3; nd

Spain

nd

7

6; 126

Brown trout, Salmo trutta

Spain: Atlantic and Mediterranean

20

9

4; 88

Bream, Abramis brama

Rivers Main and Danube, Europe

25

8

2; 133

Perch, Perca ﬂuviatilis

Baltic Sea: lake and brackish populations

20

3

4; 66

Northern Territory and Queensland, Australia

20

3

7; nd

Species Freshwater ﬁshes Hilsa shad, Tenualosa ilisha Arctic char, Salvelinus alpinus Brown trout, Salmo trutta

Freshwater invertebrates Redclaw, Cherax quadricarinatus

Region

Result

Reference

Three different populations; genetic distances 0.08–0.15 Strain speciﬁc fragments

Dahle et al., 1997 Nilsson and Schmitz, 1995 Royo et al., 1996

63.8% variance within populations, 5.6% populations within groups, 30.7% among groups Unique fragment in one Atlantic population; genetic distances 0.08–0.24; GST =0.237 Within and between river differentiation; genetic distances 0.008–0.063 GST = 0.147

373

Unique fragments observed in each population; genetic distances 0.14–0.34

Cagigas et al., 1999

Fuchs et al., 1998 Nesbo et al., 1998

Macaranas et al., 1995

(Continues)

374

TABLE 18-1. Continued

Species Prawn, Macrobrachium borelli Marine ﬁshes Indian mackerel, Rastrelliger kanagurata Fugu rubripes Red mullet, Mullus barbatus

Region Argentina

Number of primers 20

Number of variable primers 4

Number of populations; specimens 2; 35

Result Lower diversity in artiﬁcial pond than wild population

East coast India

35

1

3; 30

No population differentiation

China and Japan Mediterranean

20 29

nd 4

nd 8; 146

No population differentiation Differentiation between Ionian Sea/Aegean Sea and western Mediterranean; genetic distances 0.002–0.032 High diversity within populations; low diversity among populations; differentiation between marine and lagoon populations Changes in RAPD frequencies during acclimation from marine to freshwater Genetic sub division; GST =0.44

Sea bass, Dicentrarchus labrax

Mediterranean

8

8

9; 230

Sea bass, Dicentrarchus labrax

Mediterranean

40

15

4; 181

Striped bass, Morone saxatilis Paciﬁc cod, Gadus macrocephalus

Atlantic coast, United States Japanese coast and Bering Sea

40

8

5; 114

4

4

4; 108

Four-wing ﬂying ﬁsh. Hirundichthys afﬁnis

Central western Atlantic

17

3

6; 360

Differentiation between coastal and oceanic populations; genetic distances 0.012–0.109 Three stocks distinguished; 71% variance among regions,

Reference D’Amoto and Corach, 1996

Jayasankar and Dharmalingam, 1997 Lu et al., 1999 Mamuris et al., 1998a

Caccone et al., 1997

Allegrucci et al., 1995 Bielawski and Pumo, 1997 Saitoh, 1998

Gomes et al., 1998

Orange roughy, Hoplostethus atlanticus Antarctic toothﬁsh, Dissostichus mawsoni Marine invertebrates Scallop, Placopecten magellanicus Scallop, Pecten maximus

Southwest Paciﬁc New Zealand Antarctic Seas

24

7

4; 170

40

12

2; 42

High genetic similarity, but signiﬁcant population structure GST = 0.297

Parker et al., 2002

North Atlantic, Canada Irish Sea

40

15

5; 24

60

13

7: 126

Mendelian inheritance; no population differentiation 7% variance among regions, 93% variance within locations Signiﬁcant differentiation among populations separated by >55 km Positive correlation between population differentiation and geographic distance. No differentiation; genetic distances 0.002–0.006; GST = 0.05–0.15 Genetic differentiation between sea areas; genetic distances 0.032–0.070

Patwary et al., 1994 Heipel et al., 1998 Huang et al., 2000

Abalone, Haliotis rubra

Victoria Australia

50

14

9; 90

Lobster, Homarus gammarus

European Sea areas

36

9

7; 71

Lobster, Homarus americanus

East coast North America

42

8

3; 108

Prawn, Penaeus monodon

Andaman Sea and Gulf of Thailand

200

7

5; 100

Blue shrimp, Penaeus stylirostris

Gulf of California, Mexico

8

8

6; 78

375

a

nd = no data.

8% among populations, 21% within populations Three population groups distinguished; GST = 0.019

Localized populations; 15% variance among populations 85% within populations

Smith et al., 1997

Ulrich et al., 2001 Harding et al., 1997 Tassanakajon et al., 1997, 1998; Klinburga et al., 2001 Aubert and Lightner, 2000

376

P. J. Smith

to provide optimum polymerase activity to extend the annealed primers. After the last cycle, an additional step of 5–10 min at 70°–75°C can be used to allow complete extension of all the ampliﬁcation products. The length of the extension step is dependent on the length of target DNA that is being ampliﬁed. Ampliﬁcation products can be stored at 4°C for up to 1 month or at -20°C, or lower, for longer periods. The DNA fragments are either separated in agarose gels and detected with ethidium bromide under a UV light (312 nm) or in polyacrylamide gels and visualized with silver staining. Data are recorded with a Polaroid camera or captured with a digital system. Polyacrylamide gel electrophoresis and silver staining is the preferred method for visualizing small DNA products (<400 bp) produced from ampliﬁcations with short oligonucleotide primers (CaetanoAnolles et al., 1991; Dinesh et al., 1993). The PCR has been shown to be very sensitive to changes in concentration of primer, concentration of template, annealing temperature, and concentration of magnesium ions, all of which can affect the number and intensity of RAPD fragments (Devos and Gale, 1992; Ellsworth et al., 1993; Penner et al., 1993; Patwary et al., 1993). Furthermore, some RAPD primers have been shown to give more reproducible results than others (Penner et al., 1993). Most studies of RAPDs in aquatic species have noted that changes in PCR parameters, or quality of the DNA sample, dramatically altered RAPD patterns (Bardakci and Skibinski, 1994; and references in Table 18-1). Problems with reproducibility of RAPDs can be minimized by standardizing laboratory procedures such as using the same thermocycler and batch of reagents throughout the study, optimizing the quantity of DNA and other reagents used in PCR, and reamplifying with each primer to test for repeatability of DNA fragments. Despite these limitations, numerous studies have used short primers of arbitrary nucleotide sequence to amplify genomic DNA segments in commercially important ﬁshes and invertebrates (Table 18-1).

III. DNA FRAGMENT SCORING AND DATA ANALYSES The number of different ampliﬁcation products, or DNA fragments, for each primer depends on the primer sequence and the genome size of the test organism. Assuming that primer sites are randomly distributed throughout the genome for a typical vertebrate, 1 to 10 fragments are produced by each primer. The size of the fragments typically ranges from 250 to 3,000 bp, although smaller fragments may be detected with short primers when separated in polyacrylamide gels (Dinesh et al., 1993). Fragments are numbered with the primer code, and the fragment size (e.g., 15/650 or A1-1100) is determined from a size ladder in the gel. Only distinct, well-stained, and repeatable fragments are scored. Some primers produce weak fragments that are not repeatable in reampliﬁcations (e.g., Nilsson and Schmitz, 1995; Fuchs et al., 1998). These weak fragments (see Fig. 18-1) may

Random Ampliﬁed Polymorphic DNA (RAPD)

377

FIGURE 18-1. RAPD fragment proﬁles in three species of Nemadactylus (Teleostei: Cheilodactylidae). DNA had been ampliﬁed with the oligonucleotide primer E01, and the fragments separated in an agarose gel stained with ethidium bromide. Gel lanes 1–4 contain tarakihi Nemadactylus macropterus, lanes 5–8 king tarakihi Nemadactylus new sp., lane 9 porae Nemadactylus douglasii, and lane L contains a DNA size ladder (fragment sizes in descending order: 1000, 700, 500, 400, 300 bp). Note the very distinct proﬁle for porae in lane 9; several large weak fragments (>>1000 bp) in tarakihi and king tarakihi; and the presence/absence of small fragments (~500 bp) in lanes 1–4, and one small fragment in lanes 5–8. Data from Smith et al. (1996).

be produced by excessive PCR cycles; Bell and DeMarini (1991) have shown that increasing the number of PCR cycles above 30 can lead to nonspeciﬁc DNA products. However in preliminary ampliﬁcations with frozen ﬁsh tissue samples, I found that fewer than 40 cycles produced faint fragment patterns for many primers, necessitating 40 cycles as a standard technique. The addition of nonionic detergent to the PCR reaction can improve the yield of weak products (Patwary et al., 1994). Sometimes a smear of DNA stain is observed in lanes in the gel, which may indicate an excessive concentration of sample template DNA in the ampliﬁcation reaction (Bell and DeMarini, 1991; Devos and Gale, 1992). Most RAPD polymorphisms are scored by the presence or absence of an ampliﬁcation product (see Fig. 18-1). Standard genetic calculations are not applicable to RAPD data because most fragments are dominant. Heterozygosity cannot be calculated (other than by assuming samples to be in Hardy–Weinberg equilibrium) because homozygous individuals carrying two copies of the dominant allele cannot be distinguished from heterozygous individuals carrying one copy of the allele, and both genotypes appear as a single fragment. Occasionally, polymorphisms appear as the brightness of the fragment, but such polymorphisms are subjective to score in the absence of breeding studies (Hunt and Page, 1992; Rabouam et al., 1999). In the honey bee Apis mellifera codominant fragmentlength polymorphisms have been reported (Hunt and Page, 1992). An index of similarity (or differences) can be calculated as the number of shared (or unique)

378

P. J. Smith

fragments between pairs of individuals divided by the number of fragments scored for both individuals (e.g., Gilbert et al., 1990; Jeffreys and Morton, 1987; Lynch, 1990; Wetton et al., 1987). The similarity index (S) calculated after Wetton et al. (1987), is S = 2N ab / N a + N b where Nab is the number of fragments shared by both individuals a and b, and Na and Nb are the number of fragments in individuals a and b, respectively. The average percentage similarity is the average of all S values multiplied by 100 (Gilbert et al., 1990). Dendrograms of genetic similarity can be constructed by cluster analysis (e.g., Nei, 1987). Lynch and Milligan (1994) have developed a general theory for the analysis of population structure with dominant markers and provided formulae for the estimation of conventional population-genetic measures such as allele frequencies and gene diversity for RAPD data sets. The scored characters are the fraction of individuals with (1 - x) and without (x) the fragment. Assuming that there are two alleles per locus and that the population is in Hardy–Weinberg equilibrium, then x = q2 and x1/2 is the null allele frequency. However, there are some restrictions, and Lynch and Milligan (1994) suggest that analyses are limited to loci for which the frequency of the null phenotype is <3/N. This means that more individuals and more loci have to be surveyed than with conventional genetic markers and that there are biases in the data when comparing genetic parameters estimated with conventional methods. Nucleotide diversity can be estimated from RAPD data sets, but unless speciﬁc conditions are met then the estimates of phylogenetic relationships are error prone (Clark and Lanigan, 1993). Homology of fragments of the same size in different species should be conﬁrmed by techniques such as southern blots and hybridization (Smith et al., 1994) or sequencing.

IV. INHERITANCE OF PRESENCE/ABSENCE FRAGMENTS In the absence of breeding studies, the allelic nature of presence/absence RAPD fragments may be suspect. Occasionally nonparental bands have been reported in progenies from a wide range of species from humans (Riedy et al., 1992) to bees (Hunt and Page, 1992) to ﬂax rust (Ayliffe et al., 1994). Unlike the other genetic methods described in this volume, there are no simple internal checks that can be used to ﬁt RAPD gel phenotypes to a genetic model. For example, with allozymes each enzyme should conform to expected gel phenotypes in the homozygous and heterozygous state, with all alleles equally expressed; with mtDNA restriction fragment length polymorphisms (RFLPs), the size of the restricted fragments should add up to the size of the undigested fragment.

Random Ampliﬁed Polymorphic DNA (RAPD)

379

Rabouam et al. (1999) have shown that RAPD fragments not present in the source DNA can appear following ampliﬁcation because of interactions between primer and fragments during the denaturation and annealing steps of PCR ampliﬁcation. Such PCR artifacts could produce apparent fragment polymorphisms and indicate the need to undertake Southern blotting and hybridization to conﬁrm fragment homology (Rabouam et al., 1999). Ideally, the allelic nature of RAPD markers should be conﬁrmed through breeding studies, but these are impractical for many aquatic species. Nevertheless, where inheritance studies have been carried out with guppies (Foo et al., 1995), salmonids (Elo et al., 1997; Stott et al., 1997), catﬁsh (Liu et al., 1998), tilapia (Appleyard and Mather, 2000), and scallop (Patwary et al., 1994), results have demonstrated that RAPD polymorphisms conform with Mendelian expectations and are consistent with a dominance model. In the prawn Penaeus monodon, hatchery crosses have shown that most RAPD fragments were inherited as dominant Mendelian markers, but one primer produced a fragment in offspring that was not present in adults and that probably originated from algal or bacterial contaminants, indicating the need for careful preparation of DNA samples (Garcia and Benzie, 1995).

V. SPECIES IDENTIFICATION AND TAXONOMY Initial applications of RAPDs to ﬁsh species focused on species discrimination (Dinesh et al., 1993; Bardakci and Skibinski, 1994). Pooling of DNA samples from intraspeciﬁc individuals permits the rapid screening of a large number of primers for taxonomic studies and takes no more laboratory time than conventional allozyme screening. With this approach, the RAPD technique has been used to ﬁnd markers that distinguish closely related pairs of species of Cheilodactylidae (Smith et al., 1996; see Fig. 18-1) and blueﬁn tuna (P. J. Smith, unpublished) that could not be separated by conventional iso-electric focusing of muscle proteins. RAPD techniques have been applied to the phylogeny of freshwater ﬁshes [e.g., sturgeon (Comincini et al., 1998) and Anguillidae (Lehmann et al., 2000)], for distinguishing closely related species with similar morphologies [e.g., Barbus (Callejas and Ochando, 2001) and Cottus (Gasper et al., 2001)], and for identiﬁcation of strains and species used in aquaculture [e.g., carp (Zhang et al., 1998; Jackson et al., 2000; Sun et al., 2001) and tilapia (Naish et al., 1995; Xia et al., 1999; Degani et al., 2000)]. Parallel approaches have been applied to marine ﬁshes for the identiﬁcation and phylogeny of pufferﬁshes (Chen et al., 2001; Song et al., 2001a,b), identiﬁcation of closely related species of grouper Epinephelus (Bakar and Azizah, 2000), mullets Mullidae (Mamuris et al., 1999), and morwongs Cheilodactylidae (Smith et al., 1996). and identiﬁcation of ﬁllets (Schneider et al., 1997). Likewise, RAPDs have been applied as species markers for marine invertebrates: oysters (Klinburga et al., 2000a), octopii (Warnke et al.,

380

P. J. Smith

2000), penaeid prawns (Meruane et al., 1997, Song et al., 1999), Decapoda (Klinburga et al., 2000b; Lu et al., 2000), and bivalve and coral larvae (Coffroth and Milawka, 1995; Andre et al., 1999).

VI. STOCK DISCRIMINATION Population studies employing RAPDs have been undertaken on a range of marine and freshwater ﬁshes and invertebrates (Table 18-1). Several studies report the potential rather than the application of the technique, and sample sizes have been inadequate for rigorous population discrimination. Nevertheless, population differentiation has been reported in all freshwater species tested to date (Table 18-1), although this must be balanced against the tendency to report positive results. Population differentiation has been reported in most marine ﬁshes and invertebrates tested with RAPD markers, the few exceptions being based on small sample sizes and/or few loci (Table 18-1). An appreciation of the usefulness of RAPDs can be gained from parallel studies testing two or more classes of genetic markers on the same samples or samples from the same geographic range (Table 18-2). For several species, the stock structures revealed with RAPDs and either allozyme, or mtDNA, markers appear to be similar (Table 18-2). In some respects, this result is surprising as the methods measure different components of the genome, which may be subject to different evolutionary rates, and strengthens conclusions on genetic stock relationships. In contrast, greater population differentiation was found with allozymes and RAPDs than with mtDNA RFLPs in the orange roughy Hoplostethus atlanticus (Smith et al., 1997). A low level of polymorphism, and consequently no population differentiation, was found with RAPDs in the horseshoe crab, Limulus polyphemus, while signiﬁcant population differentiation was found with mtDNA COI sequences (Pierce et al., 2000). Occasionally, RAPDs have been used when low levels of polymorphism have been found with other methods, such as allozymes (Nilsson and Schmitz, 1995). Studies on red mullet and sea bass in the Mediterranean found a lack of concordance in genetic differentiation measured with allozymes and RAPDs (Mamuris et al., 1998a; Allegrucci et al., 1995). In the sea bass some allozyme markers reveal strong genetic differentiation between lagoon and coastal sites, suggesting that allozyme variation at some loci (which code for functional genes) is under the inﬂuence of selection, while RAPDs, which randomly sample the genome, might preferentially amplify repetitive noncoding regions (Allegrucci et al., 1995; 1997). Developments in molecular biology provide other tools for detecting genetic variation in noncoding regions of DNA, for example, microsatellite DNA (Bentzen et al., 1996) and introns (Chow and Takeyama, 1998), which, although

381

Random Ampliﬁed Polymorphic DNA (RAPD)

TABLE 18-2. Population Studies of Teleosts Employing Two or More Genetic Techniques, Including RAPDsa

Species and area

Genetic technique

Paciﬁc cod, Gadus macrocephalus, Japan

mtDNA (1000), RAPDs (24)

Red mullet, Mullus barbatus, Mediterranean Sea Sea bass, Dicentrarchus labrax, Mediterranean Sea

Allozymes (20), RAPDs (4)

Flying ﬁsh, Hirundichthys afﬁnis, Central West Atlantic Orange roughy Hoplostethus atlanticus, S.W. Paciﬁc

mtDNA RFLP (480), RAPDs (20)

Brown trout, Salmo trutta, Spain

Perch, Perca ﬂuviatilis, Baltic Sea a

Allozymes (28), RAPDs (126)

Results Both markers showed no genetic differentiation among coastal populations, but differentiation between coastal and Bering Sea population Correlation between geographic distance and genetic distance for RAPDs, but not allozymes East—west differentiation within the Mediterranean with both methods; greater differentiation observed with allozyme markers between lagoon and coastal populations Both markers in agreement, with 3 distinct populations

Reference Saitoh, 1998

Mamuris et al., 1998a,b

Allegrucci et al., 1994, 1995, 1997; Caccone et al., 1997

Gomes et al., 1998, 1999

mtDNA RFLP Similar population differentiation Smith et al., 1997 (2000), with allozymes and RAPDs (5/6 allozymes (11), pairwise comparisons signRAPDs (29) iﬁcant); less differentiation with mtDNA RFLPs (2/6 comparisons signiﬁcant) allozymes (25), Genetic relationships among 4 Cagigas et al., 1999 microsats. (3), populations in agreement for 3 RAPDs (9) markers; within-sample variation greater in microsatellites; expected heterozygosities allozymes (0.036); microsats (0.42); RAPDs (0.24) mtDNA (378), Both markers in agreement: Nesbo et al., 1998 RAPDs (12) distinct genetic populations in lake, sea, and anadromous perch

Numbers in parentheses refer to number of loci for allozymes, microsatellites, and RAPDs, and number of base pairs for mtDNA sequences or mtDNA RFLPs (restriction fragment length polymorphism.

382

P. J. Smith

technically more demanding and more expensive than RAPDs, are preferred for population studies because allelic variation is codominant. The development of species-speciﬁc microsatellite markers is laborious, and RAPDs can provide an alternative and faster approach to the isolation of microsatellites, using Southern blot procedures and hybridization with appropriate probes (Cifarelli et al., 1995; Garcia et al., 1996; Iyengar et al., 2000).

VII. SUMMARY RAPDs have been successfully applied in taxonomic studies, for identiﬁcation of ﬁsh ﬁllets, and for stock discrimination of marine and freshwater invertebrates and ﬁshes. The RAPD technique works with minute quantities of DNA and is applicable to identifying individual ﬁsh eggs and larvae. The RAPD technique is technically more demanding than allozyme electrophoresis, but is easily accommodated in a laboratory set up for simple DNA studies. The appeal of the RAPD method is its technical simplicity and speed compared with other DNA methods. However, the RAPD method has limitations. The fragment patterns are sensitive to PCR conditions and care is required to ensure that polymorphisms are repeatable. Ideally, the homology of fragments, particularly interspeciﬁc fragments, should be established. Most RAPD polymorphisms appear as presence or absence of dominant fragments, so that assumptions (biallelic systems in Hardy– Weinberg equilibrium) have to be made to estimate standard genetic parameters. Given these limitations and the more recent development of alternative molecular tools, future applications of RAPDs are likely to be for species discrimination and for population screening when microsatellites are not available and allozymes show low levels of polymorphism.

ACKNOWLEDGMENT The author was supported by the National Institute of Water and Atmospheric Research, Limited.

REFERENCES Allegrucci, G., Fortunato, C., Catudella, S., and Sbordoni, V. 1994. Acclimation to freshwater of the sea bass: evidence of selective mortality and allozyme genotypes. In A. Beaumont (ed.), Genetics and Evolution of Marine Organisms. Chapman & Hall, London, UK, pp. 486–502. Allegrucci, G., Caccone, A., Catudella, S., and Powell, J. R. 1995. Acclimation of the European sea bass to freshwater: monitoring genetic changes by RAPD polymerase chain reaction to detect DNA polymorphisms. Marine Biology 121: 591–599. Allegrucci, G., Fortunato, C., and Sbordoni, V. 1997. Genetic structure and allozyme variation in sea bass (Dicentrarchus labrax and D. punctatus) in the Mediterranean Sea. Marine Biology 128: 347–358.

Random Ampliﬁed Polymorphic DNA (RAPD)

383

Andre, C., Lindegarth, M., Jonsson, P. R., and Sundberg, P. 1999. Species identiﬁcation of bivalve larvae using random ampliﬁed polymorphic DNA (RAPD) differentiation between Cerastoderm edule and C. lamarcki. Journal of the Marine Biological Association of the UK 79: 563–565. Appleyard, A. S. and Mather, B. P. 2000. Investigation into the mode of inheritance of allozyme and random ampliﬁed polymorphic DNA markers in tilapia Oreochromis mossambicus (Peters). Aquaculture Research 31: 435–445. Aubert, H. and Lightner, D. V. 2000. Identiﬁcation of genetic populations of the paciﬁc blue shrimp Penaeus stylirostris of the Gulf of California, Mexico. Marine Biology 137: 875–885. Ayliffe, M. A., Lawrence, G. J., Ellis, J. G., and Pryor, A. J. 1994. Heteroduplex molecules formed between allelic sequences cause nonparental RAPD bands. Nucleic Acids Research 2: 1632–1636. Bakar, T. A. and Azizah, M. N. S. 2000. Fingerprinting of two species of the grouper, Epinephelus off the coast of Pulau Pinang. Asia Paciﬁc Journal of Molecular Biology and Biotechnology 8: 177–179. Bardakci, F. and Skibinski, D. O. F. 1994. Application of the RAPD technique in tilapia ﬁsh: species and subspecies identiﬁcation. Heredity 73: 117–123. Bell, D. A. and DeMarini, D. M. 1991. Excessive cycling converts PCR products to random length higher molecular weight fragments. Nucleic Acids Research 19: 5079. Bentzen, P., Taggart, C. T., Ruzzante, D. E., and Cook, D. 1996. Microsatellite polymorphism and the population structure of Atlantic cod (Gadus morhua) in the northwest Atlantic Canadian Journal of Fisheries and Aquatic Science 53: 2706–2721. Bielawski, J. P. and Pumo, D. E. 1997. Randomly ampliﬁed polymorphic DNA (RAPD) analysis of Atlantic coast striped bass. Heredity 78: 32–40. Black, W. C., DuTeau, N. M., Puterka, G. J., Nechols, J. R., and Pettorini, J. M. 1992. Use of the random ampliﬁed DNA polymerase chain reaction (RAPD-PCR) to detect DNA polymorphisms in aphids (Homoptera: Aphididae). Bulletin Entomological Research 82: 151–159. Caccone, A., Allegrucci, G., Fortunato, C., and Sbordoni, V. 1997. Genetic differentiation within the European sea bass (D. labrax) as revealed by RAPD-PCR assays. Journal of Heredity 88: 316–324. Caetano-Anolles, G., Bassam, G. J., and Gresshof, P. M. 1991. High resolution DNA ampliﬁcation ﬁngerprinting using very short arbitrary oligonucleotide primers. Biotechnology 9: 553–556. Cagigas, M. E., Vazquez, E., Blanco, G., and Sanchez, J. A. 1999. Combined assessment of genetic variability in populations of brown trout (Salmo trutta L.) based on allozymes, microsatellites and RAPD markers. Marine Biotechnology 1: 286–296. Callejas, C. and Ochando, M. D. 2001. Molecular identiﬁcation (RAPD) of the eight species of the genus Barbus (Cyprinidae) in the Iberian Peninsula. Journal of Fish Biology 59: 1589–1599. Chen, C., Shi, T., Sun, S., Yu, H., Sun, Z., and Kong, J. 2001. Identiﬁcation and phylogenetic relationships among four species of pufferﬁsh in Fugu as determined by RAPD markers. Marine Fisheries Research 22: 32–36. Chow, S. and Takeyama, H. 1998: Intron length variation observed in the creatine kinase and ribosomal protein genes of the swordﬁsh Xiphias gladius. Fisheries Science 64: 397–402. Cifarelli, R. A., Gallitelli, M., and Cellini, F. 1995. Random ampliﬁed hybridisation microsatellites (RAHM): isolation of a new class of microsatellite-containing DNA clones. Nucleic Acids research 23: 3802–3803. Clark, A. G. and Lanigan, C. M. S. 1993. Prospects for estimating nucleotide divergence with RAPDs. Molecular Biology and Evolution 10: 1096–1111. Coffroth, M. A. and Milawka, J. M. 1995. Identiﬁcation of marine invertebrate larvae by means of PCR-RAPD species-speciﬁc markers. Limnology and Oceanography 40: 181–189. Comincini, S., Lanfredi, M., Rossi, R., and Fontana, F. 1998. Use of RAPD markers to determine the genetic relationships among sturgeons (Acipenseridae, Pisces). Fisheries Science 64: 35–38. Dahle, G., Rahman, M., and Eriksen, A. G. 1997. RAPD ﬁngerprinting used for discriminating among three populations of Hilsa shad (Tenualosa ilisha). Fisheries Research 32: 263–269.

384

P. J. Smith

D’Amoto, E. and Corach, D. 1996. Genetic diversity of populations of the freshwater shrimp Macrobrachium borelli (Caridea: palaemonidae) evaluated by RAPD analysis. Journal of Crustacean Biology 16: 650–655. Degani, G., Jackson, K., Goldberg, D., and Yehuda, Y. 2000. Application of RAPD in study of genetic variations in Cichlidae in Israel. Journal of Aquaculture in the Tropics 15: 219–227. Devos, K .M. and Gale, M. D. 1992. The use of random ampliﬁed DNA markers in wheat. Theoretical Applied Genetics 84: 567–572. Dinesh, K. R., Lim, T. M., Chua, K. L., Chan, W. K., and Phang, V. P. E. 1993. RAPD analysis; an efﬁcient method of DNA ﬁngerprinting in ﬁshes. Zoological Science 10: 849–854. Ellsworth, D. L., Rittenhouse, K. D., and Honeycutt, R. L. 1993. Artifactual variation in randomly ampliﬁed polymorphic DNA (RAPD) banding patterns. BioTechniques 14: 214–217. Elo, K., Ivanoff, S., Vuorinen, J.A., and Piironen, J. 1997. Inheritance of RAPD markers and detection of interspeciﬁc hybridization with brown trout and Atlantic salmon. Aquaculture 152: 55–65. Foo, C. L., Dinesh, K. R., Lim, T. M., Cahn, W. K., and Phang, V. P. E. 1995. Inheritance of RAPD markers in the guppy ﬁsh Poecilia reticulata. Zoological Science 12: 535–541. Fuchs, H., Gross, R., Stein, H., and Rottmann, O. 1998. Application of molecular genetic markers for the differentiation of bream (Abramis brama L.) populations from the rivers Main and Danube. Journal of Applied Ichthyology 14: 49–55. Garcia, D. K. and Benzie, J. A. H. 1995. RAPD markers of potential use in penaeid prawn (Penaeus monodon) breeding programs. Aquaculture 130: 137–144. Garcia, D. K., Dhar, A. K., and Aicivar-Warren, A. 1996. Molecular analysis of a RAPD marker (B20) reveals two microsatellites and differential mRNA expression in Penaeus vannamei. Molecular Marine Biology and Biotechnology 5: 71–83. Gasper, B. R., Koppelman, J. B., Schwery, S. P., and Eckdahl, T. T. 2001. Genetic variation among Cottus species in Missouri revealed by RAPD analysis. American Midland Naturalist 146: 264–270. Gilbert, D. A., Lehman, N., O’Brien, S. J., and Wayne, R. K. 1990. Genetic ﬁngerprinting reﬂects population differentiation in the California Channel Island fox. Nature 344: 764–766. Gomes, C., Dales, R. B. G., and Oxenford, H. A. 1998. The application of RAPD markers in stock discrimination of the four-wing ﬂying ﬁsh, Hirundichthys afﬁnis, in the central western Atlantic. Molecular Ecology 7: 1029–1039 Gomes, C., Oxenford, H. A., and Dales, R. B. G. 1999. Mitochondrial DNA d-loop variation and implications for stock structure of the four-wing ﬂying ﬁsh, Hirundichthys afﬁnis, in the central western Atlantic. Bulletin of Marine Science 64: 485–500. Hadrys, H., Balick, M., and Schierwater, B. 1992. Applications of random ampliﬁed polymorphic DNA (RAPD) in molecular ecology. Molecular Ecology 1: 55–63. Harding, G. C., Kenchington, E. L., Bird, C. J., Pezzack, D. S., and Landry, D. C. 1997. Genetic relationships among subpopulations of the American lobster (Homarus americanus) as revealed by random ampliﬁed polymorphic DNA. Canadian Journal of Fisheries and Aquatic Sciences 54: 1762–1771. Heipel, D. A., Bishop, J. D., Brand, A. R., and Thorpe, J. P. 1998. Population genetic differentiation of the great scallop Pecten maximus in western Britain investigated by random ampliﬁed polymorphic DNA. Marine Ecology Progress Series 162: 163–171. Hillis, D. M. and Moritz, C. 1990. Molecular Systematics. Sinauer Associates, Sunderland, MA. Hu, J. and Quiros, C. F. 1991. Identiﬁcation of broccoli and cauliﬂower cultivars with RAPD markers. Plant Cell Reports 10: 505–511. Huang, B. X., Peakall, R., and Hanna, P. J. 2000. Analysis of genetic structure of blacklip abalone (Haliotis rubra) populations using RAPD, minisatellite and microsatellite markers. Marine Biology 136: 207–216. Hunt, G. J. and Page, R. E. 1992. Patterns of inheritance with RAPD molecular markers reveal novel types of polymorphism in the honey bee. Theoretical and Applied Genetics 85: 15–20.

Random Ampliﬁed Polymorphic DNA (RAPD)

385

Iyengar, A., Piyapattanakorn, S., Stone, D. M., Howell, D. A., Baynes, S. M., and Maclean, N. 2000. Identiﬁcation of microsatellite repeats in turbot (Scophthalmus maximus) and Dover sole (Solea solea) using RAPD based technique: characterization of microsatellite markers in Dover sole. Marine Biotechnology 2: 49–56. Jackson, K., Goldberg, D., Yehuda, Y., and Degani, G. 2000. Molecular variation in koi (Cyprinus carpio) of various color patterns. Israeli Journal of Aquaculture 52: 151–158. Jayasankar, P. and Dharmalingam, K. 1997. Analysis of RAPD polymorphisms in Rastrelliger kanagurta off India. Naga, the ICLARM Quarterly: July–Dec. 1997: 52–56. Jeffreys, A. J. and Morton, D. B. 1987. DNA ﬁngerprinting of dogs and cats. Animal Genetics 18:1–15. Kessing, B., Croom, H., Martin, A., McIntosh, C., McMillan, W. O., and Palumbi, S. 1990. The Simple Fool’s Guide to PCR. University of Hawaii, Honolulu. Klinburga, S., Boonyapakdee, A., and Pratoomchat, B. 2000a. Development of speciesspeciﬁc markers of the tropical oyster Crassostrea belcheri in Thailand. Marine Biotechnology 2: 476–484. Klinburga, S., Boonyapakdee, A., and Pratoomchat, B. 2000b. Genetic diversity and species diagnostic markers of mud crabs (genus Scylla) in Eastern Thailand determined by RAPD analysis. Marine Biotechnology 2: 180–187. Klinburga, S., Siludjai, D., Wudthijinda, W., Tassanakajon, A., Jaraybhand, P., and Menasveta, P. 2001. Genetic heterogeneity of the giant tiger shrimp (Penaeus monodon) in Thailand revealed by RAPD and mitochondrial DNA RFLP analyses. Marine Biotechnology 3: 428–438. Lehmann, D., Hettwer, H., and Taraschewski, H. 2000. RAPD-PCR investigations of systematic relationships among four species of eels (Teleostei: Anguillidae), particularly Anguilla anguilla and A. rostrata. Marine Biology 137: 195–204. Liu, Z., Li, P., Argue, B. J., and Dunham, R. A. 1998. Inheritance of RAPD markers in channel catﬁsh (Ictalurus punctatus), blue catﬁsh (I. furcatus), and their F1, F2 and backcross hybrids. Animal Genetics 29: 58–62. Lu, R., Shi, T., Liu, X., Chen, C., and Kong, J. 1999. Identiﬁcation of three populations (Chinese Fugu rubripes, Japanese Fugu rubripes and Fugu pseudommus) using RAPD markers. Transactions of Oceanology and Limnology 4: 38–42 (in Chinese with English abstract). Lu, R., Qiu, T., Xiang, C., Xie, H., and Zhang, J. 2000. RAPD and AFLP techniques for the analysis of genetic relationships in two genera of Decapoda. Crustaceana 73: 1027–1036. Lynch, M. 1990. The similarity index and DNA ﬁngerprinting. Molecular Biology and Evolution 7: 478–484. Lynch, M. and Milligan, B. G. 1994. Analysis of population genetic structure with RAPD markers. Molecular Ecology 3: 91–99. Macaranas, J. M., Mather, P. B., Hoeben, P., and Capra, M. F. 1995. Assessment of genetic variation in wild populations of the redclaw crayﬁsh (Cherax quadricarinatus, von Martens 1868) by means of allozyme and RAPD-PCR markers. Marine and Freshwater Research 46: 1217–1218. Mamuris, Z., Apostolidis, A. P., Theodororou, A. J., and Triantaphyllidis, C. 1998a. Application of random ampliﬁed polymorphic DNA (RAPD) markers to evaluate intraspecifc genetic variation in red mullet (Mullus barbatus). Marine Biology 132: 171–178. Mamuris, Z., Apostolidis, A. P., and Triantaphyllidis, C. 1998b. Genetic protein variation in red mullet (Mullus barbatus) and striped red mullet (M. surmultus) populations from the Mediterranean Sea. Marine Biology 130: 353–360. Mamuris, Z., Stamatis, C., Bani, M., and Triantaphyllidis, C. 1999. Taxonomic relationships between four species of the Mullidae family revealed by three genetic methods: allozymes, random ampliﬁed polymorphic DNA and mitochondrial DNA. Journal of Fish Biology 55: 572–587. Meruane, J., Takagi, M., and Taniguchi, N. 1997. Species identiﬁcation and polymorphisms using RAPD-PCR in penaeid prawns Penaeus japonicus and Metanephrops ensis. Fisheries Science 63: 149–151.

386

P. J. Smith

Nadeau, J. H., Bedigian, H. G., Bouchard, G., Denial, T., Kosowsky, M., Norberg, R., Pugh, S., Sargent, E., Turner, R., and Paigen, B. 1992. Multilocus markers for mouse genome analysis: PCR ampliﬁcation based on single primers of arbitrary sequence. Mammalian Genome 3: 55–64. Naish, K. A., Warren, M., Bardakci, F., Skibinski, D. O. F., Carvalho, G. R., and Mair, G. C. 1995. Multilocus DNA ﬁngerprinting and RAPD reveal similar genetic relationships between strains of Oreochromis niloticus (Pisces: Cichlidae). Molecular Ecology 4: 271–274. Nei, M. 1987. Molecular Evolutionary Genetics. Columbia University Press, New York. 512 pp. Nesbo, C. L., Magnhagen, C., and Jakobsen, K. S. 1998. Genetic differentiation among stationary and anadromous perch (Perca ﬂuviatilis) in the Baltic Sea. Hereditas 129: 241–249. Nilsson, J. and Schmitz, M. 1995. Random ampliﬁed polymorphic DNA (RAPD) in Arctic char. Nordic Journal of Freshwater Research 71: 372–377. Parker, R. W., Paige, K. N., and DeVries, A. L. 2002. Genetic variation among populations of the Antarctic toothﬁsh: evolutionary insights and implications for conservation. Polar Biology 25: 256–261. Patwary, M. U., Mackay, R. M., and van der Meer, J. P. 1993. Revealing genetic markers in Gelidium vagum (Rhodophyta) through the random ampliﬁed polymorphic DNA (RAPD) technique. Journal Phycology 29: 216–222. Patwary, M. U., Kenchington, E. L., Bird, C. J., and Zouros, E. 1994. The use of random ampliﬁed polymorphic DNA markers in genetic studies of the sea scallop Placopecten magellanicus (Gmelin, 1791). Journal of Shellﬁsh Research 13: 547–553. Penner, G. A., Bush, A., Wise, R., Kim, W., Domier, L., Kasha, K., Laroche, A., Scoles, G., Molnar, S. J., and Fedak, G. 1993. Reproducibility of random ampliﬁed polymorphic DNA (RAPD) analysis among laboratories. PCR Methods and Applications 2: 341–345. Pierce, J. C., Tan, G., and Gaffney, P. M. 2000. Delaware Bay ands Chesapeake Bay populations of the horseshoe crab Limulus polyphemus are genetically distinct. Estuaries 23: 690–698. Rabouam, C., Comes, A. M., Bretagnolle, V., Humbert, J.-F., Periquets, G., and Bigots, Y. 1999. Features of DNA fragments obtained by random ampliﬁed polymorphic DNA (RAPD) analysis. Molecular Ecology 8: 493–503. Riedy, M. F., Hamilton, W. J., and Aquadro, C. F. 1992. Excess of non-parental bands in offspring from known primate pedigrees assayed using RAPD PCR. Nucleic Acids Research 20: 918. Royo, L. J., Cristobal, C., Canon, J., and Dunner, S. 1996. Analysis of genetic variation of brown trout (Salmo trutta fario) populations using RAPD markers in pools of DNA. Animal Genetics 27: 17–42. Saitoh, K. 1998. Genetic variation and local differentiation in the Paciﬁc cod Gadus macrocephalus around Japan revealed by mtDNA and RAPD markers. Fisheries Science 64: 673–679. Schneider, M., Mandorf, T. H., and Rubach, K. 1997. Species identiﬁcation of ﬁshes with the DNA analysis and RAPD technique. Deutsche Lebensmittel-Rundscahu 93: 137–140. Smith, J. J., Scott-Craig, J. S., Leadbetter, J. R., Bush, G. L., Roberts, D. L., and Fulbright, D. W. 1994. Characterization of random ampliﬁed polymorphic DNA (RAPD) products from Xanthomas campestris and some comments on the use of RAPD products in phylogenetic analysis. Molecular Phylogenetics and Evolution 3: 135–145. Smith, P. J., Roberts, C. D., McVeagh, S. M., and Benson, P. G. 1996. Genetic evidence for two species of tarakihi (Teleostei: Cheilodactylidae Nemadactylus) in New Zealand waters. New Zealand Journal of Marine and Freshwater Research 30: 209–220. Smith, P. J., Benson, P. G., and McVeagh, S. M. 1997. A comparison of three genetic methods for stock discrimination of orange roughy, Hoplostethus atlanticus: allozymes, PCR ampliﬁed mitochondrial DNA and random ampliﬁed polymorphic DNA. Fishery Bulletin 94: 800–811. Song, L., Xiang, J., Zhou, L., Zhang, S., and Liu, R. 1999. Studies of random ampliﬁed polymorphic DNA (RAPD) markers on genomic DNA polymorphism in six species of marine shrimp. Oceanologia et Limnologia Snicia 30: 62–67.

Random Ampliﬁed Polymorphic DNA (RAPD)

387

Song, L., Liu, B., Xiang, J., and Qian, P.-Y. 2001a. Molecular phylogeny and species identiﬁcation of pufferﬁsh of the genus Takifugu (Tetraodontiformes, Tetraodontiade). Marine Biotechnology 3: 398–406. Song, L., Liu, B., Wang, Z., Li, H., Xiang, J., and Qian, P.-Y. 2001b. Phylogenetic relationships among pufferﬁsh of genus Takifugu by RAPD analysis. Chinese Journal of Oceanology and Limnology 19: 128–134. Stott, W., Ihssen, P. E., and White, B. N. 1997. Inheritance of RAPD molecular markers in lake trout Salvelinus namaycush. Molecular Ecology 6: 609–613. Sun, J., Lou, Y., and Yao, J. 2001. Application of RAPD technology to analyze the genetic diversity of three breeds of red carp. Shangai Shuichan Daxue Xuebao 10: 207–212. Tassanakajon, A., Pongsomboon, S., Rimphanitchayakit, V., Jarayabhand, P., and Boonsaeng, V. 1997. Random ampliﬁed polymorphic DNA (RAPD) markers for determination of genetic variation in wild populations of the black tiger prawn (Penaeus monodon) in Thailand. Molecular Marine Biology and Biotechnology 6: 110–115. Tassanakajon, A., Pongsomboon, S., Jarayabhand, P., Klinbunga, S., and Boonsaeng, V. 1998. Genetic structure in wild populations of the black tiger shrimp (Penaeus monodon) using randomly ampliﬁed polymorphic DNA analysis. Journal of Marine Biotechnology 6: 249–254. Ulrich, I., Muller, J., Schutt, C., and Buchholz, F. 2001. A study of population genetics in the European lobster Homarus gammarus (Decapoda, Nephropidae). Crustaceana 74: 825–837. Warnke, K., Soeller, R., Blohm, D., and Saint-Paul, U. 2000. Rapid differentiation between Octopus vulgaris Cuvier (1797) and Octopus mimus Gould (1852) using randomly ampliﬁed polymorphic DNA. Journal of Zoological Systematics and Evolutionary Research 38: 119–122. Welsh, J. and McClelland, M. 1990. Fingerprinting genomes using PCR with arbitrary primers. Nucleic Acids Research 18: 7213–7218. Welsh, J., Peterson, C., and McClelland, M. 1991. Polymorphisms generated by arbitrarily primed PCR in the mouse: application to strain identiﬁcation and genetic mapping. Nucleic Acids Research 19: 303–306. Wetton, J. H., Carter, R. E., Parkin, D. T., and Walters, D. 1987. Demographic study of a wild house sparrow population by DNA ﬁngerprinting. Nature 327: 147–149. Williams, J. G. K., Kubelik, A. R., Livak, K. J., Rafalski, J. A., and Tingey, S. V. 1990. DNA polymorphisms ampliﬁed by arbitrary primers are useful genetic markers. Nucleic Acids Research 18: 6531–6535. Xia, D., Cao, Y., Wu, T., and Wang, T. 1999. A study on genetic variation of tilapia ﬁsh with RAPD analysis and its application to heterosis. Journal of Fisheries of China 23: 27–32. Zhang, H., Rongzong, L., Zhang, X., Chen, T., Xiao, T., and Jinheng, L. 1998. Assessment of population genetic variation of grass carp and common carp using RAPD ﬁngerprints. Acta Hydrobiologica Sinica 22: 168–173.

This Page Intentionally Left Blank

CHAPTER

19

Ampliﬁed Fragment Length Polymorphism (AFLP) ZHANJIANG ( JOHN) LIU The Fish Molecular Genetics and Biotechnology Laboratory, Department of Fisheries and Allied Aquacultures and Program, of Cell and Molecular Biosciences, Aquatic Genomics Unit, Auburn University, Auburn, Alabama, USA

I. Introduction II. Background of AFLP Technology A. Nucleus, Chromosomes, Genomes, and Genomic DNA B. Molecular Basis of Genetic Variation C. Other Genetic Variations at the Molecular Level D. Molecular Analysis Related to Development of AFLP Technology E. The Procedures and Principles of AFLP Analysis F. The Power of AFLP Analysis G. Molecular Basis of AFLP Polymorphism H. Inheritance of AFLP Markers III. Methods for AFLP Analysis A. Selection of Proper Tissue for Genomic DNA Isolation B. Isolation of Genomic DNA C. AFLP Analysis D. Genotyping AFLP Gels E. Data Analysis F. Applications of AFLP in Fish Studies IV. Conclusions References

I. INTRODUCTION Multilocus DNA ﬁngerprinting technologies based on polymerase chain reactions (PCR) are of enormous value for the study of genetic variation in natural Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

389

390

Zhanjiang ( John) Liu

populations. These ﬁngerprinting technologies, such as random ampliﬁed polymorphic DNA (RAPD; Williams et al., 1990; Welsh and McClelland, 1990; Smith, this volume, Chapter 18) and ampliﬁed fragment length polymorphism (AFLP; Vos et al., 1995), allow rapid generation of large amounts of genetic data. Genetic ﬁngerprinting using these technologies does not require prior knowledge, making them “ready to be used” technologies for any species, without previous genetic information. The ﬁngerprints may be used as a tool to identify a speciﬁc DNA sample or to assess the relatedness between samples. Conserved common bands deﬁne relatedness, while polymorphic bands deﬁne differentiation in phylogenetic and population genetic analyses. AFLP technology combines the advantages of restriction enzyme ﬁngerprinting using restriction fragment length polymorphism (RFLP) and those of PCRbased ﬁngerprinting using RAPD. AFLP is based on the selective ampliﬁcation of a subset of genomic restriction fragments using PCR. DNA is digested with restriction enzymes, and double-stranded DNA adaptors are ligated to the ends of the DNA fragments to generate primer-binding sites for ampliﬁcation. The sequence of the adaptors and the adjacent restriction site serve as primer binding sites for subsequent ampliﬁcation of the restriction fragments by PCR. Selective nucleotides extending into the restriction sites are added to the 3¢ ends of the PCR primers such that only a subset of the restriction fragments is recognized. Only restriction fragments in which the nucleotides ﬂanking the restriction site match the selective nucleotides will be ampliﬁed. The subset of ampliﬁed fragments is then analyzed by denaturing polyacrylamide-gel electrophoresis to generate the ﬁngerprints. To fully appreciate the advantages and applications of AFLP, this chapter is written to describe the course of the technology development in relation to several other existing technologies, the procedures and principles, the molecular basis of polymorphism, and the potential power for genetic analysis using AFLP. As detailed below, AFLP is a technology that provides robustness, reliability, and efﬁciency. Its simultaneous analysis of hundreds of loci using only a single primer combination offers robust power of differentiation. AFLP is also advantageous because markers are inherited in Mendelian fashion; it does not require prior genetic information and is therefore adaptable to genetic analysis of any species. AFLP truly provides the multilocus coverage and potential for genome-wide coverage for analysis of genetic variations. For comparisons of AFLP with other marker systems, readers are referred to other chapters of this volume and a review on applications of DNA markers in ﬁsheries and aquaculture (Liu and Cordes, 2004).

391

Ampliﬁed Fragment Length Polymorphism (AFLP)

II. BACKGROUND OF AFLP TECHNOLOGY

A. NUCLEUS, CHROMOSOMES, GENOMES,

AND

GENOMIC DNA

Within each living cell, there is a nucleus in which chromosomes are located. Individual species harbor a ﬁxed number of chromosome pairs (2n) with ﬁxed shapes, sizes, and centromere location. These chromosome morphologies are commonly known as the karyotypes. All somatic cells in a diploid organism harbor identical chromosome pairs that are randomly shared into a single chromosome set during meiosis to produce eggs and sperm. Upon fertilization of an egg (n) by a sperm (n), the embryo recovers the diploid state with two sets of chromosomes. The numbers of chromosome pairs vary greatly among different species of vertebrate animals. The entire genetic material of an organism is known as its genome. Strictly speaking, animal genomes are composed of their nuclear and mitochondrial genomes although the term genome is often used to indicate only the nuclear genome (herein genome and nuclear genome are used interchangeably unless when speciﬁed for comparative discussions). Each chromosome is a portion of the genome and all the chromosomes compose the entire genome. Although all chromosomes maintain their own integrity, they each can be viewed as a segment of the genome. The total length of genomic DNA thus is equal to the sum of all chromosomal DNA. In their natural existence, the physical pieces of DNA in each cell are equal to the number of chromosomes. It must be emphasized that such entire chromosomal DNA is essentially impossible to be obtained for routine molecular analysis. Chromosomal DNA is randomly broken during genomic DNA extraction even under the most sophisticated preparation by the most skilled researchers. Most often, millions of cells are used in a single DNA extraction. Therefore, genomic DNA used in molecular analysis represents multiple copies of the genome with multiple overlapping segments, simply because the breakage points are random and different in each cell genome. Genomic sizes are measured in terms of base pairs (bp). Genomic sizes of aquatic and amphibian animals exhibit the greatest variation, ranging from 100 million bp to almost 100 billion bp. Genomic DNA must ﬁrst be cut into small pieces for molecular analysis. Geneticists have limited capacity of making direct analysis of large segments of DNA. Although chromosomes or chromosome segments can be directly analyzed through a special gel electrophoresis known as pulse ﬁeld electrophoresis, little genetic information can be obtained from such analysis concerning genetic variation. In contrast, resolution of differentiation can be drastically increased when DNA is cut into small segments. Restriction enzymes are site-speciﬁc “molecular scissors” for DNA. They recognize speciﬁc sequences 4 to 8 bp long. In a restriction digest reaction,

392

Zhanjiang ( John) Liu

restriction enzyme is mixed with genomic DNA and incubated under speciﬁc buffer and temperature conditions as required by the restriction enzyme. Within usually 1 hr of incubation, a restriction enzyme “searches” through the entire DNA lengths for its speciﬁc 4 to 8 bp recognition sequences. The genomic DNA is then “cut” by the enzyme whenever the proper recognition sequences are found. The cutting frequency of any restriction enzyme is directly related to the length of its recognition sequences. On average, a restriction enzyme with 4-bp recognition sequences should cut DNA once every 256 bp (1/44); a restriction enzyme with 6-bp recognition sequences should cut DNA once every 4096 bp (1/46); and a restriction enzyme with 8-bp recognition sequences should cut DNA once approximately every 64,000 bp (1/48). These cutting frequencies should be considered when choosing restriction enzymes. This is particularly true for AFLP, which will be discussed later in the chapter. Assuming a ﬁsh genome of 1 billion base pairs (109 bp), a 4-bp cutter will digest the genome into approximately 4 million segments; a 6-bp cutter will digest the genome into about one quartermillion segments; while an 8-bp cutter will digest the genome into just 15,000 segments. In addition to the length of recognition sequences, the genomic content also affects the cutting frequency of a restriction enzyme. For instance, AT-rich genomes are in favor of restriction enzyme with AT-rich recognition sequences, while GC-rich genomes are in favor of restriction enzyme with GC-rich recognition sequences.

B. MOLECULAR BASIS

OF

GENETIC VARIATION

In the long history of evolution, genomes have evolved in each species to have a ﬁxed number of chromosomes whose shape and sizes are constant. The number of genes and gene locations on each chromosome are also constant such that genetic linkage maps can be constructed. Such structural and organizational order is maintained by accurate inheritance of genes from generation to generation. However, just as constant as the inheritance of genes and traits from parents to progenies, mutations are also constant events. It was because of mutations that analysis of traits and genes was possible. Mutations can happen spontaneously or under induction of adverse environmental cues such as radiation, UV light, or chemical mutagens. Spontaneous mutations occur at a very low rate of 1 ¥ 10-5 - 2 ¥ 10-6 per gene per generation. Assuming the average gene size of 2,000 bp, this low spontaneous mutation rate translates into only one to ﬁve base mutations throughout the entire genome of 1 billion base pairs per generation. However, through the long process of evolution, many mutations have accumulated. The basic idea behind genetic analysis and stock identiﬁcation lies in accumulation of different mutations in reproductive isolated populations.

Ampliﬁed Fragment Length Polymorphism (AFLP)

393

Mutations are random events and can happen to any part of the genome, although mutation hot spots are often reported. As a result, mutations are accumulated in evolution more often in noncoding regions. First, because the nonprotein coding regions of the genome accounts for the vast majority of the entire genome, most mutations occur by chance in these regions; second, nature has placed great selection pressure for advantageous mutations and neutral mutations, but against deteriorating mutations inside the protein coding sequences. In the coding regions, silent mutations are the most predominant with only base substitutions without causing changes of the amino acid encoded by the gene sequences. No matter what the driving forces are, mutations can be categorized at the molecular level as caused by deletions, insertions, inversions, base substitutions, and rearrangements. Deletions are losses of, while insertions are additions of, DNA bases of variable sizes ranging from a single base to long stretches of DNA. Base substitutions are changes of a speciﬁc base to any other of the three bases. For instance, base A can be mutated to any of C, G, or T. Mutations from purines (A and G) to purines or from pyrimidines (C and T) to pyrimidines are called transitions; mutations from purines to pyrimidines, or vice versa, are called transversions. Transitions are the most frequent mutations because the chemical reactions involved in such mutations are more likely to occur. In relation to molecular analysis, deletions and insertions are expected to cause changes of fragment lengths of at least one base pair (Fig. 19-1a and b), while base substitutions generally do not affect fragment sizes unless the base substitutions cause gaining or losing of restriction sites (Fig. 19-1c). A base substitution within the restriction enzyme recognition site causes loss of the restriction site, and therefore, leads to loss of the restriction fragment (Fig. 19-1d). In contrast, a single base change may lead to gaining of a new site for the restriction enzyme. For instance, the recognition sequences for restriction enzyme Eco RI are GAATTC. If the original sequence was GgATTC, a single base change of the second G into A would generate a new restriction site for Eco RI. This would lead to gaining of a restriction site and production of an additional fragment (Fig. 19-1e). Rearrangements do not affect fragment lengths (Fig. 19-1f) unless the rearranged fragments contain restriction enzyme sites (Fig. 19-1g). Inversion can be viewed as a special form of rearrangement.

C. OTHER GENETIC VARIATIONS

AT THE

MOLECULAR LEVEL

In addition to the mechanisms of mutations previously mentioned, several other highly mutable sequences should also be noted because they may account for a signiﬁcant portion of polymorphism as revealed by AFLP analysis. The ﬁrst is the microsatellite sequences. As detailed in previous chapters, microsatellites are simple sequence repeats of 1 to 6 bp. High levels of mutation rates can happen

394

Zhanjiang ( John) Liu

(a) Insertion causes increased fragment size

(c) Base changes in nonrestriction sites that do not affect fragment sizes

(b) Deletion causes decreased fragment size

(d) Base changes lead to elimination of sites

G

G

A

A

(e) Base changes lead to gaining of new sites

(f) Rearrangements do not affect fragment size if restriction sites are not involved

G A

(g) Rearrangements affect fragment size if restriction sites are involved

FIGURE 19-1. The molecular basis of polymorphism. Insertion of a piece of DNA between two restriction sites (arrow) leads to increased fragment size (a). Deletion of a piece of DNA leads to decreased fragment size (b). Base substitution generally does not lead to fragment size changes (c) unless the substitution is within the restriction site that destroys the restriction site (d) or a new restriction site is gained due to the base mutation (e). Rearrangements of DNA segments do not affect restriction fragment sizes (f ) if the restriction sites are not involved in the rearranged fragments, but do affect fragment sizes if the rearranged fragments contain the restriction site (g).

at the microsatellite loci. In some cases, mutation rates can be as high as 0.2% per locus per generation (Crawford and Cuthbertson, 1996; Levinson and Gutman, 1987). Such a high mutation rate is believed to be caused by slippage of DNA polymerase with the repeated microsatellite sequences. As a result, microsatellite repeats can either expand or shrink. The differences in repeat numbers of microsatellite sequences cause changes of fragment lengths. In a sense, this type of mutation is a special form of insertion or deletion. Due to large numbers of microsatellite loci existing in ﬁsh and their high mutation rates, their

395

Ampliﬁed Fragment Length Polymorphism (AFLP)

contribution to the overall polymorphism of genomes should not be neglected no matter what approaches are used for genetic variation analysis. Similarly, unequal crossing over of minisatellite and satellite sequences may also contribute to a signiﬁcant level of genetic variations among genomes.

D. MOLECULAR ANALYSIS RELATED AFLP TECHNOLOGY

TO

DEVELOPMENT

OF

Three major technological advances made it possible to come into the genomics era. The ﬁrst was the discovery of restriction enzymes in 1973, which offered the ability to scientists to cut enormously large pieces of DNA into small segments for handling and analysis. The second was the development of sequencing technologies in 1977 that provided the high resolution of DNA at a single base level. The third was the invention of polymerase chain reaction (PCR) in 1986, which offered the ability to enhance sensitivity dramatically. The discovery of restriction enzymes led to a new era of molecular biology. Applications of restriction enzymes resulted in the entire ﬁeld of recombinant DNA; coupled with hybridizations, restriction enzymes allow procedures like Southern blot analysis possible, which was the basis for the emergence of DNA ﬁngerprinting techniques such as RFLP. Nucleotide sequencing capabilities provided not only the reality of revealing genetic differences at a single base level, but also conditions for technology advances involving PCR. As a result of these major milestones, analysis of genetic variation has experienced a technological revolution in recent years. As a matter of fact, the theoretical understanding of genomic variations today is not too much different from that of 20 years ago. The difference is the technological advances that revolutionized the ability to detect genetic variations. AFLP methodology was developed by combining several of these technological advances; it was based on RFLP and PCR while resolved by sequencing gel electrophoresis. RFLP was the most popular approach for analysis of genetic variation in the entire 1980s. As indicated by its name, RFLP was based on DNA fragment length difference after digesting genomic DNA with one or more restriction enzymes. Most typically, genomic DNA is digested by one or more restriction enzymes and separated on an agarose gel. To adapt to further handling, the DNA in the gel must be transferred to a solid support such as nitrocellulose or nylon membranes. The speciﬁc DNA locus with potential fragment length difference is characterized by hybridization to a probe, a radioactively labeled DNA, or RNA molecules with sequence similarities to the locus of interest. After hybridization, the nonspeciﬁc probes must be washed away, leaving only hybridized probes to the speciﬁc locus. The membrane is then exposed to a piece of X-ray ﬁlm for autoradiography to visualize the DNA bands. Despite its popularity, RFLP is able to detect only large

396

Zhanjiang ( John) Liu

shifts of DNA fragment sizes. Therefore, it can detect only insertions and deletions of large sizes and gaining or loss of restriction sites. It is unable to detect the vast majority of point mutations, nor deletions or insertions involving small sizes of segments because of its low resolution using agarose gel electrophoresis. As a result, polymorphic rates were low at most loci. The efforts involved in RFLP marker development have been enormous. RFLP attempts to detect genetic variation one at a time in each locus. The low polymorphic rates, when coupled to expensive and laborious processes, made application of RFLP limited. It should be particularly noted that one could not use RFLP without previously known genetic information such as availability of probes or sequence information, which are often not available for most ﬁsh or other wildlife species. The desire to have an analysis to cover a large proportion of genomes rather than one locus at a time was growing enormously as the genomics era emerged. Coming into the 1990s, scientists were desperate to develop approaches that offer both sensitivity and efﬁciency to analysis of genetic variation. Particularly in the genomics era, understanding genomic variation on a genome-wide scale was greatly demanded. The invention of PCR in the mid-1980s prompted many to create ways of genetic analysis using this powerful tool. This led to the development of RAPD in 1990 (Williams et al., 1990; Welsh and McClelland, 1990). RAPD approach was based on the fact that short oligonucleotide primers can bind to DNA at the predicted odds. For instance, every 1 million (410) bp should contain one sequence that matches with a primer of 10 nucleotide long. Therefore, a genome of 1 billion bp should contain 1,000 perfect binding sites for the 10-bp primer on each of its two strands of DNA. The 2,000 perfect binding sites plus many more subperfect binding sites (with 9 of 10 matched, or 8 of 10 matched, and so on) would make it possible to amplify DNA using a single arbitrary short primer. The conditions for this special PCR include the following: The annealing temperature must be low because of the short primer; the short primer must bind to the opposite strands of DNA with its 3¢ end facing each other; and the two binding sites must be close enough to allow a successful PCR using Taq DNA polymerase, which often travels only several kilobases. It turns out that all these conditions can be met and often multiple bands can be ampliﬁed. Any deletion/insertion existing between the two successful primers would produce a polymorphic band. Additionally, base substitutions at primer binding sites can also cause gain or loss of ampliﬁed bands. Because about a dozen bands can be analyzed simultaneously which had been ampliﬁed from random sites in the genome, RAPD rapidly gained popularity for analysis of genetic variation. What makes it more powerful is the fact that it does not require any previously known genetic information. Additionally, there are numerous possibilities of a 10 nucleotide combination; thus, RAPD primers can never be exhausted. RAPD markers are particularly adapted for efﬁcient DNA ﬁngerprinting of genotypes for genetic variation. However, its usefulness is limited by low reproducibility because of low

397

Ampliﬁed Fragment Length Polymorphism (AFLP)

annealing temperatures during PCR. This dilemma cannot be resolved because RAPD relies on short arbitrary primers, while short primers require low stringent PCR. AFLP combines the strengths of RFLP and RAPD. It is a PCR-based approach and thus requires a small amount of DNA; it does not require any prior genetic information or probes; it overcomes the problem of low reproducibility of RAPD. AFLP is capable of producing much greater numbers of polymorphic bands than RAPD in a single analysis, signiﬁcantly reducing the cost and making genetic analysis of closely related populations possible (Qi and Lindhout, 1997). The utilization of AFLP markers in genetic linkage mapping (Meksem et al., 1995; Cho et al., 1996; Mackill et al., 1996) and analyses of genetic resource pools (Folkertsma et al., 1996; Travis et al., 1996; Keim et al., 1997) has facilitated progress that would otherwise take a much longer time using other technologies. It is particularly well adapted for stock identiﬁcation because of its robust nature of analysis. Simultaneous analysis of hundreds of loci not only provides reliability for phylogenetic analysis, but also provides conﬁdence and certainty to the researcher before experimentation for choice of this technology. The other advantage of AFLP is the balance of its ability to reveal genetic variation as well as genetic conservation. In this regard, it is superior to microsatellites for applications in stock identiﬁcation. Microsatellites often possess large numbers of alleles, too many to obtain a clear picture with small samples. Identiﬁcation of stocks using microsatellites, therefore, would require very large sample sizes. For instance, if 30 ﬁsh are analyzed, each of the 30 ﬁsh may exhibit distinct genotypes at a few microsatellite loci, making it difﬁcult to link the relatedness without any commonly conserved genotypes. In closely related populations, AFLP can readily reveal commonly shared bands, which deﬁne the common roots in a phylogenetic tree, and polymorphic bands that deﬁne branches in the phylogenetic tree.

E. THE PROCEDURES

AND

PRINCIPLES

OF

AFLP ANALYSIS

Genetic variations are widely spread among genomes of even very closely related individuals. The matter is how to reveal the very minor differences among genomes. In principle, AFLP can be viewed as a multilocus or genome-wide RFLP analysis (Fig. 19-2). The technique starts with restriction digestion of genomic DNA using two restriction enzymes, most often, Eco RI and Mse I. Eco RI recognizes a 6-bp sequence of GAATTC, while Mse I recognizes a 4-bp sequence of AATT. For a genome of 1 billion bp, Eco RI digestion should produce about 250,000 fragments, while Mse I digestion should produce 4 million fragments. Because the 4-bp cutter Mse I cut DNA at a frequency 16 times greater than Eco RI, essentially all Eco RI fragments should be further digested by Mse I. The

398

Zhanjiang ( John) Liu

Step 1

Digest genomic DNA with Eco RI and Mse I

For a genome of 109 bp, you expect 250,000 Eco RI fragments (109/4,000) and 4 million Mse I fragment (109/256) Add Eco RI (

Step 2

) and Mse I (

) adaptors

The majority of fragments should be Mse I-Mse I fragments, some Eco R I-Mse I fragments, few, if any, Eco R I-Eco R I fragments. This step is to add adaptors with known sequences to create PCR primer binding sites. (For a genome of 109 bp, you expect 2x 250,000 Eco RI-Mse I fragments).

Step 3

preselection PCR amplification

A C

A C

A C

A subset of 1/16 (1/4 x 1/4) of all Eco RI-Mse I fragments are amplified due to the additional arbitrary base on each primer.

Step 4

selective PCR amplification

A subset of 1/4,096 (1/64X1/64) of all Eco RI-Mse I fragments are amplified due to the additional three arbitrary bases ( Eco RI side, Mse I side) on the PCR primers.

Step 5

Analyze PCR products on denaturing gel electrophoresis

Data collection and computational analysis

FIGURE 19-2. Schematic presentation of AFLP analysis. Step 1: genomic DNA is digested by Eco RI and Mse I into many segments of various sizes; Step 2: adaptors are ligated to the ends of the DNA fragments; Step 3: selection ampliﬁcation of a subset of the restriction fragments by adding an extra arbitrary base at the 3¢ end of the PCR primers, which leads to 1/16 fragments to be ampliﬁed; Step 4: selective ampliﬁcation of a subset of the restriction fragments by adding three extra arbitrary base at the 3¢ end of the PCR primers, which leads to 1/4,096 fragments to be ampliﬁed; Step 5: PCR products are resolved on a sequencing gel.

double enzyme digest would produce approximately 500,000 Eco RI-Mse I fragments (one original Eco RI fragment now is cleaved by many Mse I sites, leaving both ends as Eco RI-Mse I fragments), and about 4 million Mse I-Mse I fragments.

Ampliﬁed Fragment Length Polymorphism (AFLP)

399

The second step of AFLP analysis is to add adaptors on both ends of digested DNA fragments. The Eco RI-Mse I fragments must be ampliﬁed by PCR to be detected because they represent a small amount of DNA. However, there is no sequence information about these fragments. The ﬁrst challenge is to “create” two stretches of known sequences to each of these fragments for PCR. This can be achieved simply by connecting a short segment of DNA with known sequences on the Eco RI end and a short segment of DNA with different known sequences on the Mse I end. These short segments of DNA with known sequences are called Eco RI adaptors and Mse I adaptors. They are called adaptors because they harbor speciﬁc end sequences allowing them each to be perfectly paired and ligated to the double-digested Eco RI-Mse I fragments. After ligation, each Eco RI-Mse I fragment now harbors known sequences on both ends, allowing PCR ampliﬁcation of these segments to be possible by using primers with the same sequences as the adaptors. The third step of AFLP is the preselection PCR ampliﬁcation. In the 500,000 Eco RI-Mse I fragment pool, one can imagine that many of these Eco RI-Mse I fragments must exhibit size difference or length polymorphism even between two highly related individuals. However, 500,000 fragments are too many to be resolved in any kind of gel electrophoresis, simply because too many bands have to be placed per unit length of gel and the incremental size difference between any two adjacent bands is so small that a smear would result after gel electrophoresis. A well-resolved sequencing gel can display several hundred bands. This indicates that somehow the 500,000 bands must be reduced approximately 2,000 times to reach the resolvable goal of a couple hundred bands. Vos et al. (1995) intelligently fulﬁlled this challenge by adding additional arbitrary bases at the 3¢ end of the PCR primers. As each extra arbitrary base is added, the PCR primer can match to only 1/4 subset of the fragments because at each base of DNA, there are four possibilities: the base can be A, C, G, or T. When a given base is added to the 3¢ end of the PCR primer, only 1/4 of the total fragments are ampliﬁable. When a single base is added to the 3¢ end of both PCR primers, only a subset of 1/16 of the total fragments will be ampliﬁable. When two additional bases are added to each PCR primer, the reduction power is now 256 (16 ¥ 16). When three additional bases are added to each 3¢ end of the two PCR primers, the reduction power now is 4,096 (64 ¥ 64). Now with a reduction power of 4,096, the original 500,000 fragments should become about 100 bands. These bands can then be displayed on sequencing gels. The preselective PCR ﬁrst reduce the Eco RI-Mse I fragments to a subset containing 1/16 of the original fragments. The selective PCR further reduce the number of bands by amplifying only a subset of the preselective PCR products. AFLP chooses to analyze only the Eco RI-Mse I fragments. This is achieved by labeling only Eco RI primers. Since the Mse I primer is not labeled, all the ampliﬁed Mse I-Mse I fragments are not visible on the sequencing gel.

400 F. THE POWER

Zhanjiang ( John) Liu

OF

AFLP ANALYSIS

It is possible to scan the entire genome for examination of all 500,000 Eco RI-Mse I fragments by use of all possible combinations of the selective bases. That would take 64 Eco RI primers and 64 Mse I primers or 4,096 primer combinations. However, it is probably never necessary to perform such exhausted analysis. Since over 100 loci can be analyzed by a single primer combination, a few primer combinations should display thousands of ﬁngerprints. For genetic resource analysis, the number of primer combinations required for construction of phylogenetic trees/dendrograms depends on the level of polymorphism in the populations, but probably takes no more than 10 primer combinations. The potential power of AFLP in the study of genetic variation is enormous. In principle, any combination of 6-bp cutter with a 4-bp cutter in the ﬁrst step can be used to determine potential fragment length polymorphism. In the above tour through the procedures, Eco RI and Mse I were used as restriction enzymes to examine the 500,000 Eco RI-Mse I fragments. Theoretically, 4,096 primer combinations compose a complete genome-wide scan of the fragment length polymorphism using the two restriction enzymes. As hundreds of restriction endonucleases are commercially available, the total power of AFLP for analysis of genetic variation can never be exhausted.

G. MOLECULAR BASIS

OF

AFLP POLYMORPHISM

AFLP analysis is an advanced form of restriction fragment length analysis. Therefore, the molecular basis causing RFLP would also cause AFLP. First, any deletions and/or insertions between the two restriction enzymes, for example, between Eco RI and Mse I in the preceding example, will cause shifts of fragment sizes. Second, base substitution at the restriction sites will lead to loss of restriction sites and thus a size change. However, only base substitutions in all Eco RI sites and 1/8 of Mse I sites are detected by AFLP, since only Eco RI primer is labeled and AFLP is designed to analyze only the Eco RI-Mse I fragments. Third, base substitutions leading to gaining of new restriction sites may also produce AFLP. Once again, gaining Eco RI sites always leads to production of AFLP, and gaining Mse I sites must be within the Eco RI-Mse I fragments to produce new AFLP. In addition to the common mechanisms involved in polymorphism of RFLP and AFLP, AFLP also scans for any base substitutions at the ﬁrst three bases immediately after the two restriction sites. Considering large numbers of restriction sites for the two enzymes (250,000 Eco RI sites and 500,000 Mse I sites immediately next to Eco RI sites), a complete AFLP scan would also examine over 2 million bases immediately adjacent to the restriction sites.

401

F1-4

F1-3

P2

F1-2

P1

F1-1

Ampliﬁed Fragment Length Polymorphism (AFLP)

Backcross progeny (F1-1xP2)

1

2

3

4

5

6

A B C D

E

FIGURE 19-3. Inheritance of AFLP markers as dominant markers. Two parents each has three bands: Bands A, C, and D for P1, and bands B, D, and E for P2. All the bands show up in F1 except band B, which is heterozygous in P2, and therefore, are subject to segregation. When the ﬁrst F1 individual (F1-1) is backcrossed with P2, P1-speciﬁc bands are segregating; all the bands from P2 show up in backcross progenies except the heterozygous band B, which is segregating in a 1 : 1 ratio.

H. INHERITANCE

OF

AFLP MARKERS

AFLP markers are inherited in a Mendelian fashion as dominant markers (Fig. 19-3). Similar to the traditional meaning of dominance in genetics, one dose is enough to determine phenotype. All parental homozygous bands will be shown in F1, while heterozygous parental bands will segregate in F1. In Figure 19-3, each parent has three bands. Band D was not polymorphic, leaving four polymorphic bands. Band A was inherited from the ﬁrst parent into F1. When F1 was backcrossed with the second parent (P2), this band segregated in the backcross progeny in a 1 : 1 ratio. Band B was in the second parent, but absent in the F1, indicating that it was heterozygous at the locus in the parent and was not segregated into this particular F1 individual. This notion is supported by the fact that this band segregated in a 1 : 1 ratio in the backcross progeny when the F1 was backcrossed with the P2 parent. The behavior of band C is very similar to that of band A. Band E demonstrates the typical behavior of a dominant marker. Although this band was heterozygous in F1 (inherited from P2) and should segregate in backcross progenies, all six backcross progenies possess it. This is because all six individuals should receive one allele from the backcross parent P2, while some of the six individuals also receive the allele from F1. Therefore, despite the same phenotype (band), their genotypes differ; some bands were ampliﬁcation products from one allele (heterozygous), while the others were ampliﬁcation products from both alleles (homozygous). Dominant markers provide relatively less genetic information since homozygous and heterozygous individuals cannot be differentiated; they each produce a band at the locus though

402

Zhanjiang ( John) Liu

band intensities may vary, depending on allele numbers. Although double alleles often produce double amounts of PCR products, homozygous alleles and heterozygous alleles cannot be distinguished with certainty. As detailed later, caution must be exercised when scoring AFLP as codominant markers.

III. METHODS FOR AFLP ANALYSIS

A. SELECTION OF PROPER TISSUE DNA ISOLATION

FOR

GENOMIC

Various tissues are suitable for extraction of genomic DNA. Most often, highquality genomic DNA can be isolated from ﬁsh blood, liver, head kidney, trunk kidney, and brain. Muscles are ﬁne, but yield is generally low. Sampling of many of these tissues is lethal to the ﬁsh and thus may not be desirable. Collection of blood is particularly useful not only because the animals can be saved, but also because blood can provide extremely high-quality DNA. Fin clips can also be used as a nonlethal source of tissues for DNA isolation, but the quality of DNA is not as good as that of blood. Blood cell concentration may vary greatly among different species of ﬁsh. Most often, 0.1 to 0.5 ml blood is ample to provide enough DNA for thousands of AFLP reactions. DNA isolated from blood samples has always generated extremely clean AFLP proﬁles. Therefore, DNA isolation using blood will be described below. If other tissues are used, the only difference is that the tissues are chopped to small pieces with a razor blade. The pieces of tissues are then quickly frozen in liquid nitrogen and ground with a mortar and pestle. The tissue powder is then lysed in the DNA digestion buffer with proteinase K, as detailed below.

B. ISOLATION

OF

GENOMIC DNA

We have used two methods for DNA isolation: the traditional proteinase K-phenol chloroform extraction method (Strauss, 1990) and use of the genomic DNA isolation kit. For the conventional methods, DNA extraction buffer containing salt and detergent (100 mM NaCl, 10 mM Tris, pH 8.0, 25 mM EDTA, 0.5% SDS) is supplemented with proteinase K at 100 mg/ml immediately before sample collection. Because proteinase K degrades itself in the buffer, early addition of the enzyme into buffer should be avoided. Collected blood should be sharply expelled into the buffer to disperse it as much as possible. This is particularly important when large volumes of blood samples are used. Hydrophobicity immediately after lysis otherwise would slow down enzymatic digestions of cellular proteins. Samples in the lysis buffer are stable at room temperature for at least

Ampliﬁed Fragment Length Polymorphism (AFLP)

403

2 yr. The proteinase K digestion is incubated at 55°C overnight. DNA is extracted by phenol and chloroform (1 : 1) and precipitated with 2 vol ethanol. Multiple phenol chloroform extractions are not essential, but cleaner DNA can be obtained by an additional extraction. We have used the genomic DNA isolation kit PUREGENE, manufactured by Gentra Systems (Minneapolis, MN), and obtained satisfactory results. Many other DNA isolation kits are also available commercially.

C. AFLP ANALYSIS AFLP analysis kits are commercially available. The following kit, manufactured by GIBCOBRL (Life Technologies, MD, now a part of Invitrogen), is highly recommended: The AFLP Analysis System I AFLP Starter Primer Kit (catalog numbers 10544-013, 10483-014). We have used numerous kits from GIBCOBRL and are extremely happy with their user-friendliness. Kits from other sources should work as well. Kits are recommended because AFLP requires large numbers of reagents that are organized all together in a sequential order in a kit. Use of kits not only avoids a lot of trouble, but also perhaps reduces costs. The only missing items from the kits are labeled primers suitable to the user’s detection system. If manual sequencing systems are still used, Eco RI PCR primer needs to be labeled using g-P33 ATP. If automated sequencers are used, proper labels should be selected according to the ﬂuorescence used by the laser detector in the sequencer. Although any automated sequencers are ﬁne for fragment analysis, the sequencers manufactured by LI-COR (Lincoln, NE) are extremely efﬁcient and reliable for fragment analysis. Custom-labeled primers can be ordered from manufacturers of respective automated sequencers. In addition to researchers using labeled primers, an adopted direct labeling of PCR products by the incorporation of alpha-labeled nucleotides during PCR was used in a recent stufy (Reineke and Karlovsky, 2000). The reliability of this modiﬁed protocol is not known at present. The detailed protocols of AFLP reactions are well described in the kit from manufacturers. No modiﬁcations are necessary. In a typical reaction, 250 ng DNA is digested by 10 units of Eco RI and Mse I, respectively, in the presence of proper buffer. The digestion reaction is incubated at 37°C for 2 hr and the restriction enzymes are inactivated by heating at 70°C for 15 min. Eco RI and Mse I adaptors are ligated to the restriction fragments by using T4 DNA ligase. The ligation reaction is incubated at 20°C for 2 hr. After ligation, the samples are diluted 1 : 10 to increase the size of the restriction fragment library for further analysis. This sample is enough for analysis of at least 100 primer combinations. For most genetic resource or population studies, one probably does not need to go back to this step anymore.

404

Zhanjiang ( John) Liu

Preselection and selection ampliﬁcation using PCR follow strictly the protocols of the kits (e.g., GIBCOBRL catalog number 10544-013, 10483-014). The choice of primer combinations should be made by the researcher. Our experience indicates that T- rich selection bases should be avoided because they do not produce clean AFLP proﬁles (Liu et al., 1998). Whenever possible, starting with AT/GC, balanced selection bases are advised. Primers with more AT-rich selective nucleotides should be used if too few bands are produced. In contrast, if too many bands are produced, use of primers with more GC-rich selective nucleotides may improve the results. Genomic size should also be considered when choosing the number of selection bases to be added to the PCR primer. With small genomes, perhaps two selection bases are enough. For genomes greater than 1 billion bp, three selection bases should be used. It is also possible to use three selection bases in one PCR primer, but two selection bases in the other, as used by Seki et al. (1999). For every base added to either PCR primer, the subset of the fragments to be ampliﬁed is reduced fourfold. Gel electrophoresis on a denaturing sequencing gel is conducted to resolve the ampliﬁed AFLP fragments. The reliable region of fragment length tends to be within 60 to 600 bp. Although longer AFLP fragments can be produced, most fragments are between 60 and 350 bp. This is expected because Mse I is a 4-bp cutter which produces small fragments (theoretically 256 bp on average). Exclusion of the largest and smallest 10% of AFLP bands was found to signiﬁcantly enhance reproducibility (Bagley et al., 2001).

D. GENOTYPING AFLP GELS AFLP markers are inherited as dominant markers. Because of the dominance nature of AFLP, they are scored as presence/absence type of markers in genotyping. Each band is treated as a locus (not an allele). Although the true alternative allele must be somewhere in the gel with a different fragment size, practically one has no way of knowing where the alternative allele of the band is. In some cases, complementary phases of bands are observed, indicating they may be the alternative alleles of the locus, but in the absence of molecular evidence, each band is still scored as a separate locus. Therefore, the total number of AFLP loci under analysis is inﬂated about twofold because all the alleles are treated as loci. Under this treatment, the presence of one band is treated as one allele at the locus, and the absence of the band is treated as the alternative allele. In strictly controlled mating systems, it is possible to score AFLP markers as codominant markers. In such cases, the scoring is based not only on length polymorphism, but also on intensity polymorphism. The rationale is that two alleles in homozygotes should produce twice the amount of PCR products as that produced from a single allele in heterozygotes. As a matter of fact, computer

Ampliﬁed Fragment Length Polymorphism (AFLP)

405

software is available for quantitative scoring of intensity polymorphism. AFLP-Quantar Pro, marketed by Keygene products B. V. in the Netherlands, is an example. Despite its success, we like to recommend caution in intensity polymorphism, simply because of the nonlinear nature of PCR at high rounds of cycles. For identiﬁcation of stocks and population analysis, intensity polymorphism should be discouraged because scoring may be extremely difﬁcult with samples from random mating populations. The term informative AFLP is used to indicate only polymorphic AFLP bands in genetic linkage mapping analysis. In the case of linkage mapping, only polymorphic bands are expected to segregate and thus provide genetic linkage information. Therefore, commonly shared nonpolymorphic bands are not scored. For population studies, all the bands are actually informative. In fact, the commonly shared bands are extremely important since they deﬁne the common ancestor or roots for dendrogram grouping. In fact, the shared bands are used to calculate the Nei¢s similarity F values (Nei and Li, 1979) as detailed later. Of course, the polymorphic bands provide information about differentiation or branches for dendrogram grouping. Therefore, all AFLP bands need to be scored for population genetic analysis.

E. DATA ANALYSIS AFLP ﬁngerprints are scored as binary data. The presence of a band is scored as “1” and absence of the band as “0”. The binary data can be converted to similarity values using the formula Sxy = 2Nxy/(Nx + Ny). Nx and Ny refer to the number of DNA fragments generated by the AFLP analysis in individual x and y, respectively, whereas Nxy is the number of DNA fragments shared by the two individuals (Nei and Li, 1979). For intrapopulation analysis, similarity is calculated as the average of Sxy across all possible pairwise comparisons of individuals within a population. For interpopulation analysis, similarity between populations is calculated according to the formula: Sij = 1 + S¢ij - [(Si + Sj)/2] (Lynch, 1990), where S¢ij is the average similarity between randomly paired individuals from populations i and j; Si and Sj are the coefﬁcients calculated from the within-population comparisons for population i and j, respectively. Genetic distance can be calculated using S¢ij with the equation D¢ij = -ln[S¢ij(SiSj)1/2] (Lynch, 1991). Initially, similarity values of each individual within and among populations can be used to construct dendrograms to visualize population structure to see if most of the individuals of a population are clustered together. Genetic distance can be used to construct phenetic trees among populations concerning relatedness of the populations. The conﬁdence of branch support can then be evaluated by way of bootstrap analysis, with at least 1,000 replications using the PAUP software package (version 4; Swofford, 2001).

406

Zhanjiang ( John) Liu TABLE 19-1. Statistic, Software, and Selected References Using AFLP for Population Studies Statistic Software References Similarity values Sxy = 2Nxy/[Nx + Ny] Clustan32 VAX-VMS (Sneath and Sokal, 1973) NTSYS-PC (Rohlf, 1997) PHYLIP 3.5c (Felsenstein, 1993) POPGENE Version 1.31 (Yeh et al., 1997) NTSYS-PC Treecon (Van de Peer and De Wachter, 1994) Folkertsma et al., 1996 Han et al., 2000 Pejic et al., 1998 Wong et al., 1999 Keiper and McConchie, 2000 Semblat et al., 1998 Muluvi et al., 1999 Cardoso et al., 2000 Breyne et al., 1999 Genetic distance D = -ln[ Jxy/(JxJy)1/2] Genetic distance Gdxy = 1 - [2Nxy/Nx + Ny]

As shown in Table 19-1, various computer software packages are available for dendrogram construction such as POPGENE, the Clustan32 VAX-VMS program, NTSYS-PC, PHYLIP, and TREECON. NTSYS-PC is quite popular for population genetic analysis using AFLP. To date, however, most of the population genetic studies have been conducted in plants and microorganisms, but the same principles should be applicable for population genetic studies in ﬁsh. Statistical difﬁculties do exist for estimation of allele frequencies with dominant markers such as RAPD and AFLP markers. Unlike codominant markers such as microsatellite and RFLP, heterozygotes cannot be directly distinguished from the dominant homozygote phenotype (band) at individual locus with AFLP analysis. Readers who are interested in advanced statistical complications of dominant markers and potential alternative statistical treatments are referred to the cited references (Lynch, 1990; 1991; Lynch and Milligan, 1994; Isabel et al., 1995; 1999; Szmidt et al., 1996; Zhivotovsky, 1999; Krauss, 2000).

F. APPLICATIONS

OF

AFLP

IN

FISH STUDIES

AFLP is well adapted for many types of genetic analysis such as molecular systematics, analysis of population structures, migration, hybrid identiﬁcation,

Ampliﬁed Fragment Length Polymorphism (AFLP)

407

strain identiﬁcation, parentage identiﬁcation, genetic resource analysis, genetic diversity, reproduction contribution, endangered species protection, molecular ecology, marker-assisted selection, and genome mapping. Despite the advantages of AFLP, published literature on its application for the analysis of genetic variation of ﬁsh populations is still limited (Seki et al., 1999; Jorde et al., 1999; Sun et al., 1999; Chong et al., 2000; Kai et al., 2002; Mickett et al., 2003). Many AFLP analyses in ﬁsh so far are limited to genetic linkage analysis (Liu et al., 1998; Kocher et al., 1998; 1999; Grifﬁths and Orr, 1999; Agresti et al., 2000; Robison et al., 2001; Rogers et al., 2001; Liu et al., 2003), and analysis of parental genetic contributions involving interspeciﬁc hybridization (Young et al., 2001) and meiogynogenesis (Felip et al. 2000). In a recent study of the black rockﬁsh (Sebastes inermis), Kai et al. (2002) used AFLP to distinguish three color morphotypes in which diagnostic AFLP loci were identiﬁed as well as loci with signiﬁcant frequency differences. In such reproductive isolated populations, it is likely that “ﬁxed markers” of AFLP can be identiﬁed to serve as diagnostic markers. Fixed markers are associated most often with relatively less migratory, reproductive isolated populations (Kucuktas et al., 2002). With highly migratory ﬁsh species, ﬁxed markers may not be available. However, distinct populations are readily differentiated by differences in allele frequencies. For instance, Chong et al. (2000) used AFLP for the analysis of ﬁve geographic populations of Malaysian river catﬁsh (Mystus nemurus) and found that AFLP was more efﬁcient for the differentiation of subpopulations and for the identiﬁcation of genotypes within the populations than RAPD, although similar clusters of the populations were concluded with either analysis.

IV. CONCLUSIONS AFLP analysis is a robust, multilocus PCR-based DNA ﬁngerprinting technique that can provide the most efﬁcient, reliable, and economical analysis of population genetics. AFLPs are nuclear DNA markers inherited in Mendelian fashion, in contrast to environmental markers and mitochondrial markers. As compared to other nuclear markers such as RFLP and RAPD, AFLPs provide a much greater level of polymorphism and a much wider genomic coverage. AFLP probably is also superior to microsatellites for population genetic studies because of its ability to display hundreds of loci simultaneously. However, AFLP markers are inherited as dominant markers. Caution should be exercised for transfer of information across laboratories. The needs for special equipment such as sequencers may limit its wide application. These disadvantages can be compensated for by the robustness of the multilocus AFLP analysis, which not only provides high levels of polymorphism, but also provides a great level of band sharing required to establish relatedness among populations. Most importantly, AFLP (and also RAPD)

408

Zhanjiang ( John) Liu

analysis does not require any previous knowledge and thus is suitable to population genetic analysis of any species. Because of these advantages, the application of AFLP in ﬁsh population genetic studies is increasing. In time, its wide application in the studies of ﬁsh population genetics is inevitable.

ACKNOWLEDGMENTS Research in my laboratory is supported by grants from USDA NRI Animal Genome and Genetic Mechanisms Program, USDA NRI Basic Genome Reagents and Tools Program (USDA/NRICGP 200335205-12827), Mississippi-Alabama Sea Grant Consortium (Project NA86RG0039-4), Alabama Department of Conservation, USAID, and BARD (US-2954-97).

REFERENCES Agresti, J. J., Seki, S., Cnaani, A., Poompuang, S., Hallerman, E .M., Umiel, N., Hulata, G., Gall, G. A. E., and May, B. 2000. Breeding new strains of tilapia: development of an artiﬁcial center of origin and linkage map based on AFLP and microsatellite loci. Aquaculture 185: 43–56. Bagley, M. J., Anderson, S. L., and May, B. 2001. Choice of methodology for assessing genetic impacts of environmental stressors: polymorphism and reproducibility of RAPD and AFLP ﬁngerprints. Ecotoxicology 10: 239–244. Breyne, P., Rombaut, D., Van Gysel, A., Van Montagu, M., and Gerats, T. 1999. AFLP analysis of genetic diversity within and between Arabidopsis thaliana ecotypes. Mol. Gen. Genet. 261: 627–634. Cardoso, S. R. S., Eloy, N. B., Provan, J., Cardoso, M. A., and Ferreira, P. C. G. 2000. Genetic differentiation of Euterpe edulis Mart. Populations estimated by AFLP analysis. Mol. Ecol. 9: 1753–1760. Cho, Y. G., Blair, M., Panaud, W., and McCouch, S. R. 1996. Cloning and mapping of varietyspeciﬁc rice genomic DNA sequences: ampliﬁed fragment length polymorphisms (AFLP) from silver-stained polyacrylamide gels. Genome 39: 373–378. Chong, L. K., Tan, S. G., Yusoff, K., and Siraj, S. S. 2000. Identiﬁcation and characterization of Malaysian river catﬁsh, Mystus nemurus (C&V): RAPD and AFLP analysis. Biochem. Genet. 38: 63–76. Crawford, A. M. and Cuthbertson, R. P. 1996. Mutations in sheep microsatellites. Genome Res. 6: 876–879. Felip, A., Martinez-Rodriguez, G., Piferrer, F., Carrillo, M., and Zanuy, S. 2000. AFLP analysis conﬁrms exclusive maternal genomic contribution of meiogynogenetic sea nass (Dicentrarchus labrax L.). Mar. Biotechnol. 2: 301–306. Felsenstein, J. 1993. PHYLIP (Phylogeny Inference Package) version 3.5c. Department of Genetics, University of Washington, Seattle, WA. Folkertsma, R. T., van der Voort, J. N. R., de Groot, K. E., van Zandvoort, P. M., Schots, A., Gommers, F. Helder, J., J., and Backer, J. 1996. Gene pool similarities of potato cyst nematode populations assessed by AFLP analysis. Mol. Plant-Microbe Interactions 9: 47–54. Grifﬁths, R. and Orr, K. 1999. The use of ampliﬁed fragment length polymorphism (AFLP) in the isolation of sex-speciﬁc markers. Mol. Ecol. 8: 671–674. Han T. H., de Jeu, M., van Eck, H., and Jacobsen, E. 2000. Genetic diversity of Chilean and Brazilian Alstroemeria species assessed by AFLP analysis. Heredity 84: 564–569.

Ampliﬁed Fragment Length Polymorphism (AFLP)

409

Isabel, N., Beaulieu, J., and Bousquet, J. 1995. Complete congruence between gene diversity estimates derived from genotypic data at enzyme and random ampliﬁed polymorphic DNA loci in black spruce. Proc. Natl. Acad. Sci. USA 92: 6369–6373. Isabel, N., Beaulieu, J., Theriault, P., and Bousquet, J. 1999. Direct evidence for biased gene diversity estimate from dominant random ampliﬁed polymorphic DNA (RAPD) ﬁngerprints. Mol. Ecol. 8: 477–483. Jorde, P. E., Palm, S., and Ryman, N. 1999. Estimating genetic drift and effective population size from temporal shifts in dominant gene marker frequencies. Mol. Ecol. 8: 1171–1178. Kai, Y., Nakayama, K., and Nakabo, T. 2002. Genetic differences among three colour morphotypes of the black rockﬁsh, Sebastes inermis, inferred from mtDNA and AFLP analyses. Mol. Ecol. 11: 2591–2598. Keim, P., Kalif, A., Schupp, J., Hill, K., Travis, S. E., Richmond, K., Adair, D. M., Hugh-Jones, M., Kuske, C. R., and Jakcson, P. 1997. Molecular evolution and diversity in Bacillus anthracis as detected by ampliﬁed fragment length polymorphism markers. J. Bacteriol. 179: 818–824. Keiper, F. J. and McConchie, R. 2000. An analysis of genetic variation in natural populations of Sticherus ﬂabellatus [R. Br. (St. John)] using ampliﬁed fragment length polymorphism (AFLP) markers. Mol. Ecol. 9: 571–581. Kocher, T., Lee, W., Sobolewska, H., Penman, D., and Andrew, B. 1998. A genetic linkage map of a cichlid ﬁsh, the tilapia (Oreochromis niloticus). Genetics 148: 1225–1232. Krauss, S. L. 2000. Patterns of mating in Persoonia mollis (Proteaceae) revealed by an analysis of paternity using AFLP: implications for conservation. Australian J. Botany 48: 349–356. Kucuktas, H., Wagner, B. K., Shopen, R., Gibson, M., Dunham, R. A., and Liu, Z. J. 2002. Genetic analysis of Ozark Hellbenders (Cryptobranchus alleganiensis bishopi) utilizing RAPD markers. Proc. Ann. Conf. SEAFWA 55: 126–137. Levinson, G. and Gutman, G. A. 1987. High frequency of short frameshifts in poly-CA/GT tandem repeats borne by bacteriophage M13 in Escherichia coli K–12. Nucleic Acid Res. 15: 5323–5338. Liu, Z., and J. Cordes 2004. DNA marker technologies and their applications in aquaculture genetics. Aquaculture, in review. Liu, Z, Nichols, A., Li, P., and Dunham, R. A. 1998. Inheritance and usefulness of AFLP markers in channel catﬁsh (Ictalurus punctatus), blue catﬁsh (I. furcatus), and their F1, F2, and backcross hybrids. Mol. Gen. Genet. 258: 260–268. Liu, Z., Li, P., Kucuktas, H., Nichols, A., Tan, G., Zheng, X., Argue, B. J., and Dunham, R. A. 1999. Development of ampliﬁed fragment length polymorphism (AFLP) markers suitable for genetic linkage mapping of catﬁsh. Trans. Am. Fish. Soc. 128: 317–327. Liu, Z. J., Karsi, A., Li, P., Cao, D., Ju, Z., and Dunham, R. 2003. A genetic linkage map of channel catﬁsh (Ictalurus punctatus) constructed by using AFLP markers and an interspeciﬁc hybrid resource family. Genetics, in press. Lynch, M. 1990. The similarity index and DNA ﬁngerprinting. Mol. Biol. Evol. 7: 478. Lynch, M. 1991. Analysis of population genetic structure by DNA ﬁngerprinting. In T. Burke, G. Dorf, A. J. Jeffreys, and R. Wolff (eds.), DNA Fingerprinting: Approaches and Applications. Birkhauser Verlag, Basel. Lynch, M. and Milligan, B. G. 1994. Analysis of population genetic structure with RAPD markers. Mol. Ecol. 3: 91–99. Mackill, D. J., Zhang, Z., Redona, E. D., and Colowit, P. M. 1996. Level of polymorphism and genetic mapping of AFLP markers in rice. Genome 39: 969–977. Meksem, K., Leiser, D., Peleman, J., Zabeau, M., Salamini, F., and Gebhardt, C. 1995. A highresolution map of the vicinity of the R1 locus on chromosome V of potato based on RFLP and AFLP markers. Mol. Gen. Genet. 249: 74–81. Mickett, K., Morton, C., Feng, J., Li, P., Simmons, M. Cao, D., Dunham, R., and Liu, Z. J. 2003. Assessing genetic diversity of domestic populations of channel catﬁsh (Ictalurus punctatus) in Alabama using AFLP markers. Aquaculture, in press.

410

Zhanjiang ( John) Liu

Muluvi, G. M., Sprent, J. I., Soranzo, N., Provan, J., Odee, D., Folkard, G., McNicol, J. W., and Powell, W. 1999. Ampliﬁed fragment length polymorphism (AFLP) analysis of genetic variation in Moringa oleifera Lam. Mol. Ecol. 8: 463–470. Nei, M. 1978. Estimation of average heterozygosity and genetic distance from a small number of individuals. Genetics 89: 583–590. Nei, M. and Li, W. H. 1979. Mathematical model for studying genetic variation in terms of restriction endonucleases. Proc. Natl. Acad. Sci. USA 76: 5269–5273. Pejic, I., Ajmone-Marsan, P., Morgante, M., Kozumplick, V., Castiglioni, P., Taramino, G., and Motto, M. 1998. Comparative analysis of genetic similarity among maize inbred lines detected by RFLPs, RAPDs, SSRs, and AFLPs. Theor. Appl. Genet. 97: 1248–1255. Qi, X. and Lindhout, P. 1997. Development of AFLP markers in barley. Mol. Gen. Genet. 254: 330–336. Reineke, A. and Karlovsky, P. 2000. Simpliﬁed AFLP protocol: replacement of primer labeling by the incorporation of alpha-labeled nucleotides during PCR. Biotechniques 28: 622–623. Robison, B. D., Wheeler, P. A., Sundin, K., Sikka, P., and Thorgaard, G. H. 2001. Composite interval mapping reveals a major locus inﬂuencing embryonic development rate in rainbow trout (Oncorhynchus mykiss). J. Hered. 92: 16–22. Rohlf, F. J. 1997. NTSYS-PC. Numerical Taxonomy and Multivariate Analysis System, Version 2.00. Exeter Software, Setauket, New York. Rogers, S. M., Campbell, D., Baird, S. J., Danzmann, R. G., and Bernatchez, L. 2001. Combining the analyses of introgressive hybridisation and linkage mapping to investigate the genetic architecture of population divergence in the lake whiteﬁsh (Coregonus clupeaformis, Mitchill). Genetica 111: 25–41. Seki, S., Agresti, J. J., Gall, G. A. E., Taniguchi, N., and May, B. 1999. AFLP analysis of genetic diversity in three populations of ayu Plecoglossus altivelis. Fish. Sci. 65: 888–892. Semblat, J. P., Wajnberg, E., Dalmasso, A., Abad, P., and Castagnone-Sereno, P. 1998. Highresolution DNA ﬁngerprinting of parthenogenetic root-knot nematodes using AFLP analysis. Mol. Ecol. 7: 119–125. Sneath, P. H. A. and Sokal, R. R. 1973. Numerical Taxonomy: The Principles and Practice of Numerical Classiﬁcation. W. H. Freeman, San Francisco, CA. Strauss, W. M. 1990. Preparation of genomic DNA from mammalian tissue. In F. M. Ausubel et al. (eds.), Current Protocols in Molecular Biology. Wiley , New York, pp. 2.2.1–2.2.2. Sun, Y., Song, W., Zhong, Y., Zhang, R., Abatzopoulos, T. J., and Chen, R. 1999. Diversity and genetic differentiation in Artemia species and populations detected by AFLP markers. Int. J. Salt Lake Res. 8: 341–350. Swofford, D. L. 2001. PAUP*. Phylogenetic Analysis Using Parsimony (*and Other Methods). Version 4. Sinauer, Sunderland, MA. Szmidt, A. E., Wang, X. R., and Lu, M. Z. 1996. Empirical assessment of allozyme and RAPD variation in Pinus sylvestris (L.) using haploid tissue analysis. Heredity 76: 412–420. Travis, S. E., Maschinski, J., and Keim, P. 1996. An analysis of genetic variation in Astragalus cremnophylax var. cremnophylax, a critically endangered plant, using AFLP markers. Mol. Ecol. 5: 735–745. Van de Peer, Y. and Wachter, De R. 1994. TREECON for Windows: a software package for the construction and drawing of evolutionary trees for the Microsoft Windows environment. Comput. Appl. Biosci. 10: 569–570. Vos, P., Hogers, R., Bleeker, M., Reijans, M., van de Lee, T., Hornes, M., Frijters, A., Pot, J., Peleman, J., Kuiper, M., and Zabeay, M. 1995. AFLP: a new technique for DNA ﬁngerprinting. Nucl. Acids Res. 23: 4407–4414. Welsh, J. and McClelland, M. 1990. Fingerprinting genomes using PCR with arbitrary primers. Nucl. Acids Res. 18: 7213–7218.

Ampliﬁed Fragment Length Polymorphism (AFLP)

411

Williams, J. G. K., Kubelik, A. R., Livak, K. J., Rafalski, J. A., and Tingey, S. V. 1990. DNA polymorphisms ampliﬁed by arbitrary primers are useful as genetic markers. Nucl. Acids Res. 18: 6531–6535. Wong, H. L., Yeoh, H. H., and Lim, S. H. 1999. Customisation of AFLP analysis for cassava varietal identiﬁcation. Phytochem. 50: 919–924. Yeh, F. C., Yang, R. C., and Boyle, T. 1997. POPGENE, version 1.21: software microsoft Windowbased freeware for population genetic analysis. University of Alberta, Canada. Young, W. P., Ostberg, C. O., Keim, P., and Thorgaard, G. H. 2001. Genetic characterization of hybridization and introgression between anadromous rainbow trout (Oncorhynchus mykiss irideus) and coastal cutthroat trout (O. clarki clarki). Mol. Ecol. 10: 921–930. Zhivotovsky, L. A. 1999. Estimating population structure in diploids with multilocus dominant DNA markers. Mol. Ecol. 8: 907–913.

This Page Intentionally Left Blank

PART

Applied Marks

VI

This Page Intentionally Left Blank

CHAPTER

20

Internal and External Tags J. A. JACOBSEN* AND L. P. HANSEN† *Faroese Fisheries Laboratory, Tórshavn, Faroe Islands, †Norwegian Institute for Nature Research, Oslo, Norway

I. II. III. IV.

Introduction Overview of Tagging History The Concept of Stock Identiﬁcation The Practical Aspects of Tagging A. General Concerns When Planning a Tagging Study B. Capture C. Handling and Holding D. Tag Types and Tagging Methods E. Effects of Handling and Tagging V. Applications of Mark-Recapture Studies for Stock Identiﬁcation A. Stock Identiﬁcation of Atlantic Salmon B. Stock Structure of Northeast Atlantic Mackerel C. Stock Structure of Herring D. Assessment of Atlantic Salmon E. In Situ Tagging of Deep-Sea Redﬁsh (Sebastes Spp.) VI. Conclusion References

I. INTRODUCTION External and internal tags have been used for centuries as markers on marine and freshwater ﬁshes for the purpose of identiﬁcation and information retrieval. We focus on the use of conventional tagging methods (i.e., conventional tags and deliberately applied marks) and their applications for identifying stocks, excluding electronic tags and natural marks. When referring to a mark-recapture study or marking/tagging in general, we assume that either a mark, a tag, or both have been applied to the ﬁsh. Further, the review is limited to ﬁnﬁsh. Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

415

416

J. A. Jacobsen and L. P. Hansen

The aim of this chapter is to provide a summary and critique of conventional tagging methods and applications in stock identiﬁcation. We do not attempt a full review of all aspects of tagging; rather we provide a description of a few case studies where mark-recapture studies have been used for stock identiﬁcation purposes, concluding with a summary discussion.

II. OVERVIEW OF TAGGING HISTORY External and internal tags have been used for centuries as markers on marine and freshwater ﬁshes for the purpose of identiﬁcation and information retrieval. From the earliest application on salmonids and subsequently to most ﬁsh species, these markers have been divided into various categories, depending on the authors’ choices, for example by methods of attachment (external and internal) and design (buttons and disks, dangling, and others) (Jakobsson, 1970; Nielsen, 1992), whether they are natural or artiﬁcially made tags (Jones, 1976; McFarlane et al., 1990), their style of identiﬁcation (style of marking) (Nielsen, 1992), and whether they are applied in mass marking or as individual identiﬁable tags (Laird and Stott, 1978). The ﬁrst successful large-scale marking of Atlantic salmon was done in the 1800s with the Atkins tag, but it was not until the 1890s that Petersen conducted the ﬁrst successful tagging of marine ﬁsh, and shortly after, the well-known Petersen disc was applied with great success on plaice in Danish waters, mainly through longer retention time of the tag (Jakobsson, 1970). Tagging of other demersal species such as cod was initiated in the beginning of the twentieth century, and the number of ﬁsh and species tagged accelerated after the 1920s (Jakobsson, 1970). Tagging of marine pelagic ﬁsh presented many difﬁculties, however. The ﬁrst tagging experiments of herring with barbed hooks in the late 1890s were unsuccessful. The advances in handling and tagging techniques as well as the development of the magnetic body cavity tag during these experiments resulted in the ﬁrst successful ﬁeld tagging of herring in Alaska in the early 1930s (Rounsefell and Dahlgren, 1933). The method was further enhanced by the development of an electronic tag detector installed at ﬁsh plants and through the invention of a tagging gun to speed up the tagging process. From the 1950s onward several pelagic ﬁsh species have been successfully tagged with both external tags and internal magnetic steel tags (body cavity tags and hypodermically implanted tags) (Jakobsson, 1970). Larger marine pelagic species such as tuna were successfully tagged after the introduction of the loop spaghetti tag and the dart streamer tag in the late 1950s (Mather, 1963; Miyake, 1990). The latter developed into the well-known T-bar anchor tag used in many tagging programs today.

Internal and External Tags

417

Since the early 1960s, a variety of subcutaneous tags has been developed and used initially on salmonids. Many of these involve the use of magnetic and electronic detection devices. The small magnetized stainless steel coded wire tag (CWT) is hypodermically implanted into suitable tissue such as the snout or cheek cartilage (Johnson, 1990) or the nape (Buckmeier, 2001). Its small size and ease of use have facilitated many large-scale tagging programs (Schweigert et al., 2001; Weitkamp and Neely, 2002). A more recent subcutaneous tag that is individually discernible and externally visible without the use of an external device is the Visible Implant (VI) tag (Bergman et al., 1992; Haw et al., 1990). Artiﬁcially applied marks, either externally or internally (including subcutaneous attachment), have been extensively used in ﬁsh marking for over 100 years (McFarlane et al., 1990). Mutilation is generally achieved by clipping or holepunching ﬁns or other body parts of ﬁsh; however, identiﬁcation of individual ﬁsh is limited. Since the 1930s, mutilation, and especially ﬁn clipping, has been used as a secondary mark to identify ﬁsh with internal or subcutaneous tags. Other marks, such as branding, dyes, and paints (tattoos), have been used as secondary marks or primary marks to batch-mark ﬁsh in aquaculture and husbandry (McFarlane et al., 1990).

III. THE CONCEPT OF STOCK IDENTIFICATION Numerous deﬁnitions of the stock concept have been proposed throughout the ﬁsheries literature and were recently reviewed by Begg and Waldman (1999). In its present form, the stock concept essentially describes characteristics of a population unit assumed homogeneous for a particular management purpose. Consequently, the working deﬁnition of the stock concept consists of semidiscrete groups of ﬁsh with some deﬁnable attributes of interest to managers (Begg et al., 1999). Begg and Waldman (1999) advocated a holistic approach to ﬁsh stock identiﬁcation, that is, one involving a broad spectrum of complementary techniques, because this would maximize the likelihood of correctly deﬁning ﬁsh stocks. In the management of stocks, it is essential to have some knowledge of the stock structure, including the total extent of the stock (distribution), migration, spawning areas, and temporal and spatial degree of overlap with other stocks. In anadromous ﬁsh, stock identiﬁcation can be used to gather information on, for example spawning stocks, separation of stock components or stock structure (local vs. shared or hatchery vs. wild), determination of stock components in mixed stock ﬁsheries, the extent of escape of farmed ﬁsh in areas with ﬁsh farms, and management of ranching operations. Marking of ﬁsh can be used to infer stock structure. For stock identiﬁcation purposes, best results occur when putative stocks are marked when they are

418

J. A. Jacobsen and L. P. Hansen

geographically discrete in order to determine whether they subsequently intermingle (Cushing, 1981). Alternatively, stock mixtures may be marked to ﬁnd out if ﬁsh later separate geographically. The success of mark-recapture for stock identiﬁcation purposes is dependent on representative tagging and recapture efforts (Begg and Waldman, 1999; Cushing, 1981). Mark-recapture studies have been used in stock identiﬁcation applications since the beginning of the twentieth century. In many cases, stock identiﬁcation has been a part of tagging studies, either as the primary goal or auxiliary to the objectives. Hilborn et al. (1990) and Hammer and Blankenship (2001) suggested that ﬁsh marking provides useful data for critical management decisions designed to manage weak or depleted stocks, strengthen the resolve of governments in dealing with reduced ﬁshing opportunities, assist with decisions to eliminate inefﬁcient programs, provide estimation of ﬁsh migration in jurisdictional disputes, and determine the success of artiﬁcial enhancement efforts.

IV. THE PRACTICAL ASPECTS OF TAGGING This section deals with the practical aspects of tagging (i.e., planning, sampling design, capture, types of tags, ﬁeld techniques, holding, handling, and anesthetics). An extensive literature on conventional tags and marks as well as tagging methods is available (e.g., Jakobsson, 1970; Jones, 1979; Kohler and Turner, 2001; Laird and Stott, 1978; McFarlane et al., 1990; Nielsen, 1992; Rounsefell and Everhart, 1953; Thorsteinsson, 2002). This section provides a summary of methodologies in their application for identifying stocks.

A. GENERAL CONCERNS WHEN PLANNING

A

TAGGING STUDY

The success of tagging experiments depends on many factors, the most important being handling and tagging mortality of the marked individuals, loss of tags and marks, and failure to detect or report marked ﬁsh during the recovery process. There have been many attempts to deﬁne the perfect mark or tag [see, e.g., Ricker (1956), Laird and Stott (1978), and Nielsen (1992)]. In summary, the ideal mark and marking method would make any ﬁsh identiﬁable to at least a group (preferably individually) to anyone examining it, be permanent on any ﬁsh, inexpensive, easy and fast to apply, preferably without the use of anesthetics, and have no effect on growth, mortality, behavior, vulnerability to ﬁshing gear, or the commercial value of the ﬁsh. Obviously, it will rarely be possible to obtain a perfect tag or mark meeting all these requirements. For many purposes, one or more of the above requirements are unnecessary, or can be relaxed with little loss of information, depending on the objectives of the study. If the duration of a study

419

Internal and External Tags

is short, it is obvious that a (cheap) tag or mark with lower retention rate that is easy and fast to apply could be used, and in a closed scientiﬁc study, the ease of recognition for untrained people might not be applicable.

B. CAPTURE An important part of any marking experiment is the way in which ﬁsh are caught and handled. Some ﬁsh species are relatively tough and can withstand rough treatment, such as cod. Other species, especially many small and medium-sized pelagic ﬁsh, are quite delicate, and unless precautions are taken, may be severely damaged or killed. Generally, ﬁsh with loose scales suffer most from capture and handling during tagging; indeed, descaling is considered most severe compared to other types of damage such as scars (Hartt, 1963; Mattson et al., 1990). When tagging salmon in the Paciﬁc, Hartt (1963) observed less damage on the ﬁsh when using purse seines compared to ﬁsh caught with gill nets and long-line. Trawls usually cause more damage to the ﬁsh than most other types of gear (Jones, 1979). A newly developed live ﬁsh cage (Fish-Lift) attached to the end of the trawl, instead of a cod end, and capable of bringing live ﬁsh on deck might improve survival of tagged ﬁsh (Holst and McDonald, 2000).

C. HANDLING

AND

HOLDING

In planning tagging procedures, handling time for the ﬁsh should be short. This will enhance survival and thus optimize return rate as well as the economics of the project. For inexperienced people it is recommended to practice tagging methods on dead ﬁsh, and if surgical implants are planned, to keep tagged ﬁsh under observation for some time. Many tagging methods have been tested by controlled survival and tag retention experiments (e.g., Buckmeier and Irwin, 2000; McAllister et al., 1992; Nakashima and Winters, 1984; Strand et al., 2002). In rivers with ﬁsh traps, it is possible to estimate the survival of tagged ﬁsh. The lower return rates of tagged ﬁsh compared to untagged ﬁsh might be due to mortality caused by handling and tagging or to lower growth rate of tagged individuals (Hansen, 1988; Hansen and Jonsson, 1988; Wertheimer et al., 2002).

D. TAG TYPES

AND

TAGGING METHODS

Tags can either be externally or internally attached. The decision on which tag type to use depends on the objectives of the study.

420

J. A. Jacobsen and L. P. Hansen

External tags usually have shorter life expectancy than internal tags (McFarlane et al., 1990). Atlantic herring have been recaptured after 15 years carrying an internal CWT (Jakobsson, 1970). For recovery purposes, however, an internal tag is usually not detectable unless a secondary mark/tag is employed, which is often not noted by the ﬁshermen. Therefore, some loss must be expected due to the failure to detect the presence of CWTs (Mattson et al., 1990). Fin clipping of salmon as a secondary tag is not always enough to alert commercial ﬁshermen, although the interested angler might suspect a tagged ﬁsh. If the objective is to tag ﬁsh over many years, with expected recapture after one or more years, tags with high retention rates are required, while in a study of short duration (days or weeks) higher tag loss rate may be tolerated without compromising the objectives. Often, large-scale studies require relatively cheap tags with high retention rates such as internal CWTs. However, these have to be recovered in dedicated recovery programs or during manual or automatic screening of catches. It is questionable how long tagged ﬁsh should be held prior to release. Some authors recommend releasing the ﬁsh once they have been tagged, while others argue for longer periods, from a few hours to several days (Beaumont et al., 2002; Martinelli et al., 1998). This is also dependent on the tagging site, whether on board a vessel or ashore. It might be considered more stressful to keep ﬁsh in onboard tanks on vessels in rough weather (Clay, 1990). Simonsen and Treble (2001) found that holding tagged Greenland halibut for 5 hr or more in observation tanks ashore reduced mortality of the released ﬁsh signiﬁcantly. A detailed list of commonly used anesthetics in ﬁsh marking and recommendations for use was given by Thorsteinsson (2002) and should be consulted as part of any tagging project when surgical methods are used.

E. EFFECTS

OF

HANDLING

AND

TAGGING

Many authors have investigated the effects of various tags and tagging procedures on ﬁsh (Parker et al., 1990). Mortality due to the tagging process is one of the critical parameters. Mortality due to increased predation of externally tagged ﬁsh should also be considered (Parker et al., 1963; Strand et al., 2002), although Maynard et al. (1996) reported that the main cause of higher predation of marked ﬁsh was trauma associated with tagging itself and not the tag type or attachment. Otterå et al. (1998), however, did not ﬁnd any difference in predation rate of externally tagged small cod and controls. It has often been noted that the small individuals suffer more from handling during tagging (mainly scale loss) as a result of their vigor compared to larger individuals (Hansen and Jacobsen, 2003; Hartt, 1963). Increased recovery rates associated with ﬁsh size at tagging were observed for sableﬁsh, Anoplopoma

421

Internal and External Tags

ﬁmbria, tagged off the west coast of Canada (Saunders et al., 1990), for groundﬁsh species (Atlantic cod, haddock, pollock and American plaice) in the Northwest Atlantic (Fowler and Stobo, 1999), and for cod in the Northeast Atlantic (Beverton and Bedford, 1963). Other effects include decreased readability due to the possibility of fouling of external parts in long-time studies (Henderson-Arzapalo et al., 1999; Jones, 1979; Thorstad et al., 2001), and deterioration of external parts of some plastic dart or anchor tags when the ﬁsh were at liberty (Henderson-Arzapalo et al., 1999; Mattson et al., 1990; Pepperell, 1990). Marks, such as the visible implant (VI) and various types of paints and dyes, usually have very little effect on the ﬁsh and can be read without use of intrusive methods. They can also be applied on very small ﬁsh. Disadvantages include their low visibility to ﬁshermen and their reported shorter retention rate, although recent enhancements of VI marks seem to indicate higher retention rates (Olsen and Voellestad, 2001; Rikardsen et al., 2002).

V. APPLICATIONS OF MARK-RECAPTURE STUDIES FOR STOCK IDENTIFICATION A. STOCK IDENTIFICATION

OF

ATLANTIC SALMON

Mark-recapture was used in stock identiﬁcation of Atlantic salmon (Salmo salar) at West Greenland in the late 1960s and early 1970s (Horsted, 1988; Parrish and Horsted, 1980). The commercial exploitation of salmon at west Greenland began in the mid-1960s after local ﬁshermen noticed salmon in abundance. Catches rose quickly with increasing ﬁshing effort as vessels from Norway, Denmark, Sweden, Faroes, as well as Greenland became involved. The general pattern is that salmon from the Northwest Atlantic is mainly conﬁned to the western areas while salmon from the Northeast Atlantic is found in both the eastern and the western part of the Atlantic. Tags applied to salmon smolts leaving home rivers also indicated that the salmon being caught at Greenland originated from Europe and North America (Parrish and Horsted, 1980). Since then, due to the involvement of both European and North American salmon stocks, attention has been focused on estimating the proportions of North American and European Atlantic salmon in commercial catches. These investigations were based on discriminant analyses of scale characters whereby Atlantic salmon could be classiﬁed into a western and eastern group with apparently high accuracy (Reddin and Friedland, 1999). To determine the origin of the Atlantic salmon caught on ﬂoating long-lines in the Faroese ﬁshery in the southern Norwegian Sea, a tagging program using external Carlin tags was carried out during the winter season (November–March)

422

J. A. Jacobsen and L. P. Hansen

TABLE 20-1. Estimated Proportions (%) of Atlantic Salmon Returning to Different Homewater Countries Tagged North of the Faroes during Three Fishing Seasons (Winter 1992/1993–1994/1995)a Tag reporting rate

Exploitation rate

Country of origin

Number recaptured

Min

Max

Min

Max

Norway Scotland Russia Canada Ireland Denmark England Sweden Spain Iceland

47 12 6 4 9 2 1 4 1 1

0.40 0.80 0.60 0.65 0.60 0.40 0.40 0.55 0.60 0.80

0.60 1.00 0.80 0.85 0.80 0.60 0.60 0.75 0.80 1.00

0.50 0.10 0.10 0.15 0.50 0.14 0.15 0.55 0.55 0.40

0.80 0.30 0.15 0.28 0.75 0.34 0.35 0.90 0.85 0.60

Total

87

Estimated number recaptured 145 67 69 25 21 17 8 8 2 2 364

Simulation ¢-5% Mean(%) ¢+95% 27.2 8.8 7.6 1.6 2.5 0 0.6 0 0 0

39.6 19.2 18.3 6.9 5.7 4.7 2.3 2.3 0.6 0.6

51.7 32.5 30.5 13.6 9.4 11.8 4.7 7.1 1.8 1.7

100.2

a About 5,500 ﬁsh were tagged with Carlin tags. Conﬁdence limits (95%) were based on 1,000 simulations using Monte Carlo simulation (“@Risk”). Recoveries were adjusted for homewater exploitation rates and tag reporting rates. From Hansen and Jacobsen (2003).

from 1992 to 1995 (Hansen and Jacobsen, 2003). The areas north of the Faroes seem to be one of the main feeding areas for Atlantic salmon in the sea (Jacobsen and Hansen, 2000; Jacobsen and Hansen, 2001), and recaptures from this area were returned from most of the regions bordering the North Atlantic. Estimates of the proportions of salmon from their countries of origin are given in Table 20-1. The recoveries were adjusted for homewater exploitation rates and tag reporting rates. About half of the tags were recovered from Norway and signiﬁcant numbers were recovered from Scotland and Russia, with less from Ireland, Canada, Sweden, Denmark, England, Spain, and Iceland (Hansen and Jacobsen, 2003). The recapture rates were generally low, with an increasing trend correlated to size of ﬁsh, that is 1SW, 2SW, and 3+SW ﬁsh (Fig. 20-1), caused mainly by higher tagging mortality of smaller individuals suffering high scale loss due to increased vigor during the tagging process. Further, the recapture rates of ﬁsh with the hook left in when released were signiﬁcantly higher than of ﬁsh where the hook was removed prior to release, which might indicate that active removal of the hook was lethal for some ﬁsh. Fish were tagged without use of anesthetics and kept in observation tanks up to 2hr before release (Hansen and Jacobsen, 2003). Separation of stock components in this area has been used in the assessment of Atlantic salmon by the International Council for the Exploration of the Sea (ICES) in recent years (ICES, 2002).

423

Internal and External Tags

FIGURE 20-1. Recapture rates by sea age (size groups) of Atlantic salmon tagged and released north of the Faroes during 1992/1993–1994/1995. The ﬁgures show the number of observations. From Hansen and Jacobsen (2003).

The main problems with mark-recapture studies of Atlantic salmon have been the heterogeneity of recaptures and the lack of simultaneous large-scale marking in the whole distributional area, which might bias the information gained. Factors like nonreporting of tags, tagging mortality, and low exploitation resulting in low numbers of returns may reduce the power in the derivation of total catches of single stocks using tag recaptures. Further, many countries use CWT while others use external Carlin- or Leatype tags with noncomparable shedding and detection rates, further complicating a quantitative analyses of the data.

B. STOCK STRUCTURE

OF

NORTHEAST ATLANTIC MACKEREL

A dispute on stock components of Atlantic mackerel (Scomber scombrus) in the Northeast Atlantic has been ongoing for decades. The perception of mackerel migration has changed over time, from the old hypothesis that mackerel undertook relatively short migrations from their spawning grounds off the coast to deeper waters nearby, where they hibernated during winter, to the present recognition of mackerel as a highly migratory species (Iversen, 2002). The current ﬁshery ranges from Gibraltar, along the western continental slope, north into the Norwegian Sea and eastward into the North Sea (ICES, 2003b). It has been debated whether the Northeast Atlantic mackerel consists of two stocks or spawning components, one western and one North Sea component, or three spawning components, named after their main spawning areas: southern, western, and North Sea (Iversen, 2002). The three components are treated as one unit in management, named the Northeast Atlantic mackerel. This is because it is impossible to allocate catches in the Norwegian Sea and North Sea to different components during the second half of the year (ICES, 1995).

424

J. A. Jacobsen and L. P. Hansen

Several mark-recapture studies have been applied to answer the question of stock identity, allocation of catches, and other biological issues that have been unresolved since the mid-1950s (Hamre, 1970; Iversen, 2002; Iversen and Skagen, 1989; Uriarte and Lucio, 2001). The most recent studies, in 1997 and 1998, had the objectives of clarifying the migration pattern of adult mackerel from the southern and western areas and determining the spatial recruitment pattern of juveniles from the nursery areas (Uriarte et al., 2001). Adult recoveries show that almost all adult mackerel, regardless of whether tagged in the southern or western areas, follow the same northward migration from the spawning grounds in late spring and summer along the west of the British Isles to the feeding grounds north of Faroes, the Norwegian Sea, and the northern part of the North Sea (Fig. 20-2). Recaptures of juveniles suggested, that in general, they remain close to the areas where they are tagged (Uriarte et al., 2001). Further, the presence of adult mackerel of the southern component on the western spawning grounds during spring questioned the present assumption of separate spawning components in these areas (Uriarte et al., 2001). Unfortunately, no tagging was done at the summer feeding grounds in the oceanic part of the distribution areas further north. Also, the use of two different tag types, one internal and one external, complicated the subsequent analysis. Further, the heterogeneity of the recaptures, with no or little ﬁshery and, hence, recaptures in some of the nursery areas, excluded an unbiased analysis of the origin and migration in these areas. However, it is clear from recent studies, that the main issue in management of mackerel today, that is, whether the southern and western components are separate stock units, still remains to be solved. Genetic analysis seems to indicate that there is a separate southern component (Nesbø et al., 2000).

C. STOCK STRUCTURE

OF

HERRING

Results from the tagging of Atlantic herring (Clupea harengus) in the Northeast Atlantic (Dragesund et al., 1980; Hamre, 1989; Jakobsson, 1970) provide an example of long-distance ﬁsh migrations and stock identiﬁcation. The term Atlanto-Scandian herring was used as a common name for three stocks: Norwegian spring spawners, Icelandic spring spawners, and Icelandic summer spawners, with a small component of spring spawners on the eastern banks of the Faroes (Fridriksson and Aasen, 1950; Jakobsson, 1970). The objective was to test if at least part of the stock of herring caught during summer off the north coast of Iceland was identical to the Norwegian winter herring. Jakobsson (1970) reviewed these tagging experiments and stated that the main ﬁndings supported the original view that they originated from the same stock. Schweigert et al. (2001) recently applied coded wire tagging technology in Paciﬁc herring (Clupea pallasi) to investigate stock structure and migration (Fig.

Internal and External Tags

425

FIGURE 20-2. Recaptures of adult Atlantic mackerel in the Northeast Atlantic from tagging in 1994 (white arrow) in the Bay of Biscay (redrawn from Uriarte and Lucio, 2001). Recaptures grouped into seasons, winter (12/21–3/31), spring (3/21–6/21), summer (6/21–9/21), and autumn (9/21–12/21).

20-3). Since the 1930s tagging of Paciﬁc herring using internal metal tags and external Floy tags has been carried out in British Columbia to understand stock structure and mixing rates of populations. Unfortunately, uncertainty in some tag recovery locations and low rates of tag return limited the utility of these studies. In 1999, a new tagging program was initiated employing coded wire microtags to mark Paciﬁc herring on the spawning grounds in order to monitor the

426

J. A. Jacobsen and L. P. Hansen

FIGURE 20-3. The central nape (neck) tag site and brass needle support are shown as a herring is about to have the needle inserted prior to tag insertion. From Schweigert et al. (2001).

movement and mixture of ﬁsh interannually (Schweigert et al., 2001). Tank experiments indicated high rates of survival and low tag-shedding rates, and ﬁeld trials indicated the feasibility of cost-effective application of large numbers of tags during the short spawning season (250,000 tags applied over 28 days). A detailed description of methodologies for capturing, holding, tagging, and releasing tagged herring was developed (Schweigert et al., 2001). Tag recovery rates of 1% to 2% in 2000 from the 1999 releases greatly exceeded the returns from previous tagging programs and indicated a high degree of homing or ﬁdelity to the area of release, but also produced a number of remarkable strays (Schweigert et al., 2001). The automated coded wire tagging technology appears to provide a useful tool for large-scale marking experiments on smaller pelagic species for stock identiﬁcation purposes.

427

Internal and External Tags

Moores and Winters (1984) reported on the migration patterns of Atlantic herring in western Newfoundland waters elucidated from eight tagging experiments from 1975 to 1980 involving both spring and autumn spawners. A total of 43,700 external tags were applied from which 1,062 recaptures were reported to the end of 1981. The spatial and temporal distribution of the returns indicated that the western Newfoundland herring populations are discrete from stocks in adjacent areas, although extensive mixing occurs during part of the year, particularly outside the spawning season. The Strait of Belle Isle appeared to be an important summer feeding area for the western Newfoundland herring populations and, to a lesser degree, for herring from northeastern Newfoundland and the southern Gulf of St. Lawrence. These results were utilized in establishing the boundaries for management of the western Newfoundland stock complex. These boundaries were considered appropriate for the current herring ﬁsheries which exploit both spring and autumn spawners (Moores and Winters, 1984).

D. ASSESSMENT

OF

ATLANTIC SALMON

In the management of Atlantic salmon (Salmo salar) in the North Atlantic, the stock composition must be known to estimate the contribution to the ﬁshery from various regions. The present assessment model provides catch advice to ﬁsheries managers based on a forecast of pre-ﬁshery abundance (ICES, 2002). CWT and Carlin tag studies have been extensively used and form the basis for estimating the exploitation rate of various European salmon stocks using the runreconstruction model (Lassen et al., 1988; Potter and Dunkley, 1993; Rago et al., 1993). This approach has been developed for estimating levels of exploitation of Atlantic salmon stocks in sequential ﬁsheries. The run-reconstruction model backcalculates the stock size at the beginning of a ﬁshery, taking the number of returning spawners as the starting point. Mainly CWT and Carlin tag studies are used to estimate the number of ﬁsh removed by ﬁsheries and hence the levels of exploitation of the extant stock. However, the method is limited to stocks from closely monitored rivers because of the need to know the total numbers of tagged ﬁsh returning to home rivers.

E. IN SITU TAGGING

OF

DEEP-SEA REDFISH (SEBASTES SPP.)

Stock identiﬁcation of redﬁsh stock complex in the Irminger Sea and its adjacent waters is one important unresolved issue within the NEAFC (North-East Atlantic Fisheries Commission). Many nations ﬁsh on various stocks or stock components in the area without knowing the stock boundaries of the different components (ICES, 2003a). For tagging of ﬁsh in situ, a newly developed

428

J. A. Jacobsen and L. P. Hansen

FIGURE 20-4. The newly developed Underwater Tagging Equipment (UTE) for tagging of ﬁsh in situ, by Star-Oddi (http://www.star-oddi.com, Iceland).

Underwater Tagging Equipment (UTE, Fig. 20-4) by Star-Oddi in collaboration with the Marine Research Institute in Reykjavik (http://www.star-oddi.com, Iceland) is expected to bring a new dimension to the research of deepwater species in general, and results from tagging of deep-sea redﬁsh with the new system may help resolve the stock identiﬁcation issue in the near future. The UTE is located at the end of the belly section of a trawl net (both bottom and pelagic) and performs the whole tagging operation underwater, from the surface down to more than 1,000 m. Since UTE prevents the need for hauling the ﬁsh to the surface for tagging and release (most deepwater species do not survive being brought to the surface), the handling time for each ﬁsh is short. By tagging the ﬁsh in its natural environment, stress factors such as changes in pressure and temperature can be avoided, reducing tagging mortality. The ﬁsh to be tagged is enclosed by a grid that diverts the ﬁsh into the tagging place. The ﬁsh is viewed through an onboard video camera, and the tagging gun is moved into position. A knife makes a small incision into the skin of the ﬁsh and the tag is pressed into its body cavity. Both data storage tags and anchor tags can be used. A small plastic tube hangs outside to allow for identiﬁcation (Fig. 20-5). After tagging, the ﬁsh is released into open water. The whole process takes less than 30 sec after the ﬁsh enters the device. The device has been tested in tank experiments on cod without damaging their health, and successful tagging has been made on deep-sea redﬁsh in situ (StarOddi), but survival of the tagged ﬁsh was not examined. Long-term studies are needed to verify the success of the method, especially regarding long-term tagging mortality and shedding of tags, since the incision wound is not closed after tagging. Wagner et al. (2000) found that insertion of dummy radio transmitters in rainbow trout, when compared to sham-operated controls, had negative effects

Internal and External Tags

429

FIGURE 20-5. Deep-sea redﬁsh (Sebastes sp.) tagged in situ. A small plastic tube hangs outside the ﬁsh to allow its identiﬁcation (Star-Oddi, http://www.star-oddi.com, Iceland).

on healing incisions, probably because of increased pressure on the wound from the tag, thus, increasing the chance that the tag eventually migrates out through the wound since the wound is not closed as part of the tagging process.

VI. CONCLUSION Identiﬁcation of ﬁsh stocks is necessary for a number of reasons including allocation of catch among competing ﬁsheries (nationally or internationally), management of highly migratory stocks, recognition and protection of nursery and spawning areas, and development of optimal harvest and monitoring strategies. Tagging is one of several techniques that have been used to analyze stock composition, separation, and identity in time and space, providing the basis for subsequent regulation and management, and the examples given here show that conventional tagging has been and can still be used for stock identiﬁcation of ﬁsh. However, there are some limitations. The main problems with the various mark-recapture studies have been the heterogeneity of tagging and recaptures and the lack of simultaneous large-scale marking in the whole distributional area.

430

J. A. Jacobsen and L. P. Hansen

Such large-scale marking is often difﬁcult due to the nature of the ﬁshing ﬂeet in which the ﬁshery might be limited by national boundaries such that sampling merely reﬂects the ﬁshing effort in space and time and not biological distribution of the ﬁsh. For stock identiﬁcation the use of internal tags requires the setup of a screening program, whereas external tags can be recovered in dedicated sampling programs and by commercial and recreational ﬁshermen. The choice of tag and tagging method should be decided based on the purpose of the investigation, the species to be tagged, size, and costs. Hilborn et al. (1990) argue that marking will play an important role in meeting the current challenges of ﬁsheries management. Stock identiﬁcation is a multidisciplinary ﬁeld, as shown in this volume, and encompasses many techniques (Begg and Waldman, 1999; Cadrin and Friedland, 1999; Waldman et al., 1997). It continues to evolve along with ﬁsheries management and conservation requirements (Begg et al., 1999). Because a large proportion of world ﬁsheries occur on mixed stocks, it is essential to continuously develop new tagging technologies to quantify the various stock components that comprise these ﬁsheries. Fish marking, in combination with other state-of-the-art techniques, continues to be one means of obtaining such information, if well designed.

REFERENCES Beaumont, W. R., Cresswell, B., Hodder, K. H., Masters, J. E., and Welton, J. S. 2002. A simple activity monitoring radio tag for ﬁsh. Hydrobiologia 483: 219–224. Begg, G. A., Friedland, K. D., and Pearce, J. B. 1999. Stock identiﬁcation and its role in stock assessment and ﬁsheries management: an overview. Fish. Res. 43: 1–8. Begg, G. A. and Waldman, J. R. 1999. An holistic approach to ﬁsh stock identiﬁcation. Fish. Res. 43: 35–44. Bergman, P. K., Haw, F., Blankenship, H. L., and Buckley, R. M. 1992. Perspectives on design, use, and misuse of ﬁsh tags. Fisheries 17: 20–25. Beverton, R. J. H. and Bedford, B. C. 1963. The effects of the return rate of condition of ﬁsh when tagged. ICNAF Spec. Publ. 4: 106–116. Buckmeier, D. L. 2001. Coded wire tag insertion sites for small ﬁngerling black bass. N. Am. J. Fish. Manage. 21: 696–698. Buckmeier, D. L. and Irwin, E. R. 2000. An evaluation of soft visual implant tag retention compared with anchor tag retention in channel catﬁsh. N. Am. J. Fish. Manage. 20: 296–298. Cadrin, S. X. and Friedland, K. D. 1999. The utility of image processing techniques for morphometric analysis and stock identiﬁcation. Fish. Res. 43: 129–139. Clay, D. 1990. Tagging demersal marine ﬁsh in subzero temperatures along the Canadian Atlantic coast. Am. Fish. Soc. Symp. 7: 147–151. Cushing, D. H. 1981. Fisheries Biology. A Study in Population Dynamics. University of Wisconsin Press, Madison, WI. Dragesund, O., Hamre, J., and Ulltang, Ø. 1980. Biology and population dynamics of the Norwegian spring spawning herring. J. Cons. Int. Explor. Mer. 177: 43–71.

Internal and External Tags

431

Fowler, G. M. and Stobo, W. T. 1999. Effects of release parameters on recovery rates of tagged groundﬁsh species. Can. J. Fish. Aquat. Sci. 56: 1732–1751. Fridriksson, Á. and Aasen, O. 1950. The Norwegi–Icelandic herring tagging experiment. FiskDir. Skr. Ser. HavUnders. 9: 1–34. Hammer, S. A. and Blankenship, H. L. 2001. Cost comparison of marks, tags, and mark-with-tag combinations used in salmonid research. N. Am. J. Aquacult. 63: 171–178. Hamre, J. 1970. Internal Tagging Experiments of Mackerel in the Skagerak and the North-eastern North Sea. ICES CM 1970/H:25, 11 pp. Hamre, J. 1989. Life history and exploitation of the Norwegian spring spawning herring. Proceedings 4th Soviet-Norwegian Symposium, Bergen, 12–16 June 1989, 5–39. Hansen, L. P. 1988. Effects of carlin tagging and ﬁn clipping on survival of Atlantic salmon (Salmo salar L.) released as smolts. Aquaculture 70: 391–394. Hansen, L. P. and Jacobsen, J. A. 2003. Origin, migration and growth of wild and escaped farmed Atlantic salmon, Salmo salar L., in oceanic areas north of the Faroe Islands. ICES J. Mar. Sci. 60: 110–119. Hansen, L. P. and Jonsson, B. 1988. Salmon ranching experiments in the River Imsa: effects of dipnetting, transport and chlorbutanol anaesthesia on survival. Aquaculture 74: 301–305. Hartt, A. C. 1963. Problems in tagging salmon at sea. ICNAF Spec. Publ. 4: 144–155. Haw, F., Bergman, P. K., and Fralick, R. D. 1990. Visible implanted ﬁsh tag. Am. Fish. Soc. Symp. 7: 311–315. Henderson-Arzapalo, A., et al. 1999. An evaluation of six internal anchor tags for tagging juvenile striped bass. N. Am. J. Fish. Manage. 19: 482–493. Hilborn, R., Walters, C. J., and Jester, D. B., Jr. 1990. Value of ﬁsh marking in ﬁsheries management. Am. Fish. Soc. Symp. 7: 5–7. Holst, J. C. and McDonald, A. 2000. FISH-LIFT: a device for sampling live ﬁsh with trawls. Fish. Res. 48: 87–91. Horsted, S. A. 1988. Future investigations on the ocean life of salmon. In D. Mills and D. Piggins (eds.), Atlantic Salmon: Planning for the Future. Croom Helm, London, pp. 512–523. ICES 1995. Report of the Working Group on the Assessment of Mackerel, Horse Mackerel, Sardine and Anchovy. ICES CM 1995/Assess:2. ICES 2002. Report of the Working Group on North Atlantic Salmon. ICES CM 2002/ACFM:14, 299 pp. ICES 2003a. Report of the North Western Working Group. ICES CM 2003a/ACFM:24, 393 pp. ICES 2003b. Report of the Working Group on the Assessment of Mackerel, Horse Mackerel, Sardine and Anchovy. ICES CM 2003b/ACFM:07, 572 pp. Iversen, S. A. 2002. Changes in the perception of the migration pattern of Northeast Atlantic mackerel during the last 100 years. ICES Marine Science Symposia 215: 382–390. Iversen, S. A. and Skagen, D. W. 1989. Migration of western mackerel to the North Sea 1973–1988. ICES CM 1989/H:20, 9 pp. Jacobsen, J. A. and Hansen, L. P. 2000. Feeding habits of Atlantic salmon at different life stages at sea. In D. Mills (eds.), The Ocean Life of Atlantic Salmon: Environmental and Biological Factors Inﬂuencing Survival. Fishing News Books, Blackwell Science, Oxford, UK, pp. 170–192. Jacobsen, J. A. and Hansen, L. P. 2001. Feeding habits of wild and escaped farmed Atlantic salmon, Salmo salar L., in the Northeast Atlantic. ICES J. Mar. Sci. 58: 916–933. Jakobsson, J. 1970. On ﬁsh tags and tagging. Oceanography and Marine Biology 8: 457–499. Johnson, J. K. 1990. Regional overview of coded wire tagging of anadromous salmon and steelhead in Northwest America. Am. Fish. Soc. Symp. 7: 782–816. Jones, R. 1976. The use of marking data in ﬁsh population analysis. FAO Fisheries Technical Paper, 1–142.

432

J. A. Jacobsen and L. P. Hansen

Jones, R. 1979. Material and methods used in marking experiments in ﬁsheries research. FAO Fisheries Technical Paper, 1–134. Kohler, N. E. and Turner, P. A. 2001. Shark tagging: a review of conventional methods and studies. Environ. Biol. Fish. 60: 191–223. Laird, L. M. and Stott, B. 1978. Marking and tagging. In T. Bagenal (eds.), Methods for Assessment of Fish Production in Fresh Waters. Blackwell Science, Oxford, UK, pp. 84–100. Lassen, H., Jensen, J. M., and Hansen, L. P. 1988. Simulating North Atlantic salmon marine life history. ICES CM 1988/M:18, 12 pp. Martinelli, T. L., Hansel, H. C., and Shively, R. S. 1998. Growth and physiological responses to surgical and gastric radio transmitter implantation techniques in subyearling chinook salmon (Oncorhynchus tshawytscha). Hydrobiologia 371/372: 79–87. Mather, F. J. 1963. Tag and tagging techniques for large pelagic ﬁshes. ICNAF Spec. Publ. 4: 288–293. Mattson, M. T., Waldman, J. R., Dunning, D. J., and Ross, Q. E. 1990. Abrasion and protrusion of internal anchor tags in Hudson River striped bass. Am. Fish. Soc. Symp. 7: 121–126. Maynard, D. J., Frost, D. A., Waknitz, F. W., and Prentice, E. F. 1996. Vulnerability of marked age-0 steelhead to a visual predator. Trans. Am. Fish. Soc. 125: 330–333. McAllister, K. W., McAllister, P. E., Simon, R. C., and Werner, J. K. 1992. Performance of 9 external tags on hatchery-reared rainbow trout. Trans. Am. Fish. Soc. 121: 192–198. McFarlane, G. A., Wydoski, R. S., and Prince, E. D. 1990. Historical review of the development of external tags and marks. Am. Fish. Soc. Symp. 7: 9–29. Miyake, P. M. 1990. History of the ICCAT tagging program, 1971–1986. Am. Fish. Soc. Symp. 7: 746–764. Moores, J. A. and Winters, G. H. 1984. Migration patterns of Newfoundland west coast herring, Clupea harengus, as shown by tagging studies. J. Northw. Atl. Fish. Sci. 5: 17–22. Nakashima, B. S. and Winters, G. H. 1984. Selection of external tags for marking Atlantic herring (Clupea harengus harengus). Can. J. Fish. Aquat. Sci. 41: 1341–1348. Nesbø, C. L., Rueness, E. K., Iversen, S. A., Skagen, D. W., and Jakobsen, K. S. 2000. Phylogeography and population history of Atlantic mackerel (Scomber scombrus L.): a genealogical approach reveals genetic structuring among the eastern Atlantic stocks. Proc. R. Soc. Lond. B 267: 281–292. Nielsen, L. A. 1992. Methods of Marking Fish and Shellﬁsh. American Fisheries Society Special Publication 23, Bethesda, MD. Olsen, E. M. and Voellestad, L. A. 2001. An evaluation of visible implant elastomer for marking agebrown trout. N. Am. J. Fish. Manage. 21: 967–970. Otterå, H., Kristiansen, T. S., and Svåsand, T. 1998. Evaluation of anchor tags used in sea-ranching experiments with Atlantic cod (Gadus morhua L.). Fish. Res. 35: 237–246. Parker, N. C., Giorgi, A. E., Heidinger, R. C., Jester, D. B., Jr., Prince, E. D., and Winans, G. A. 1990. Fish-Marking Techniques. American Fisheries Society Symposium 7, Bethesda, Maryland. Parker, R. R., Black, E. C., and Larkin, P. A. 1963. Some aspects of ﬁsh-marking mortality. ICNAF Special Publication 4: 117–122. Parrish, B. B. and Horsted, S. A. (eds.) 1980. ICES/ICNAF joint investigations on North Atlantic salmon. Rapp. P.-v. Réun. Cons. int. Explor. Mer 176: 1–146. Pepperell, J. G. 1990. Australian cooperative game-ﬁsh tagging program, 1973–1987: Status and evaluation of tags. Am. Fish. Soc. Symp. 7: 765–774. Potter, E. C. E. and Dunkley, D. A. 1993. Evaluation of marine exploitation of salmon in Europe. In D. Mills (eds.), Salmon in the sea and new enhancement strategies. Fishing News Books, Blackwell Science, Oxford, pp. 203–219. Rago, P. J., Reddin, D. G., Porter, T. R., Meerburg, D. J., Friedland, K. D., and Potter, E. C. E. 1993. A continental run reconstruction model for the non-maturing component of the North American Atlantic salmon: analysis of ﬁsheries in Greenland and New Foundland-Labrador, 1974–1991. ICES CM 1993/M:24.

Internal and External Tags

433

Reddin, D. G. and Friedland, K. D. 1999. A history of identiﬁcation to continent of origin of Atlantic salmon (Salmo salar L.) at west Greenland, 1969–1997. Fish. Res. 43: 221–235. Ricker, W. E. 1956. Uses of marking animals in ecological-studies: the marking of ﬁsh. Ecology 37: 665–670. Rikardsen, A. H., Woodgate, M., and Thompson, D. A. 2002. A comparison of Floy and soft VIalpha tags on hatchery Arctic charr, with emphasis on tag retention, growth and survival. Environ. Biol. Fish. 64: 269–273. Rounsefell, G. A. and Dahlgren, E. H. 1933. Tagging experiments on the Paciﬁc herring, Clupea pallasii. J. Cons. Int. Explor. Mer. 8: 371–384. Rounsefell, G. A. and Everhart, W. H. 1953. Fishery Science—Its Methods and Applications. Wiley, New York. Saunders, M. W., McFarlane, G. A., and Beamish, R. J. 1990. Factors that affect the recapture of tagged sableﬁsh off the west coast of Canada. Am. Fish. Soc. Symp. 7: 708–713. Schweigert, J., Flostrand, L., Slotte, A., and Tallman, D. 2001. Application of coded wire tagging technology in Paciﬁc herring to investigate stock structure and migration. ICES CM 2001/O:12, 4 pp. Simonsen, C. S. and Treble, M. A. 2001. Tagging mortality of Greenland halibut, Reinhardtius hippoglossoides (Walbaum). Sci. Counc. Res. Doc. NAFO, 1–15. Strand, R., Finstad, B., Lamberg, A., and Heggberget, T. G. 2002. The effect of Carlin tags on survival and growth of anadromous Arctic charr, Salvelinus alpinus. Environ. Biol. Fish. 64: 275–280. Thorstad, E. B., Økland, F., and Heggberget, T. G. 2001. Are long term negative effects from external tags underestimated? Fouling of an externally attached telemetry transmitter. J. Fish Biol. 59: 1092–1094. Thorsteinsson, V. 2002. Tagging Methods for Stock Assessment and Research in Fisheries. Report of Concerted Action FAIR CT.96.1394 (CATAG). Marine Research Institute Technical Report, 1–179. Uriarte, A. et al. 2001. Spatial pattern of migration and recruitment of Northeast Atlantic mackerel. ICES CM 2001/O:17, 40 pp. Uriarte, A. and Lucio, P. 2001. Migration of adult mackerel along the Atlantic European shelf edge from a tagging experiment in the south of the Bay of Biscay in 1994. Fish. Res. 50: 129–139. Wagner, G. N., Stevens, E. D., and Byrne, P. 2000. Effects of suture type and patterns on surgical wound healing in rainbow trout. Trans. Am. Fish. Soc. 129: 1196–1205. Waldman, J. R., Richards, R. A., Schill, W. B., Wirgin, I., and Fabrizio, M. C. 1997. An empirical comparison of stock identiﬁcation techniques applied to striped bass. Trans. Am. Fish. Soc. 126: 369–385. Weitkamp, L. and Neely, K. 2002. Coho salmon (Oncorhynchus kisutch) ocean migration patterns: insight from marine coded-wire tag recoveries. Can. J. Fish. Aquat. Sci. 59: 1100–1115. Wertheimer, A. C., Thedinga, J. F., Heintz, R. A., Bradshaw, R. F., and Celewycz, A. G. 2002. Comparative effects of half-length coded wire tagging and ventral ﬁn removal on survival and size of pink salmon fry. N. Am. J. Aquacult. 64: 150–157.

This Page Intentionally Left Blank

CHAPTER

21

Electronic Tags MARK B. BAIN Center for the Environment, Cornell University, Ithaca, New York, USA

I. II. III. IV.

Introduction The Technology Telemetry Methods Conclusions References

I. INTRODUCTION A ﬁsh stock has been traditionally deﬁned as a group of ﬁsh of a given species occupying a deﬁned geographic area at a particular time. From a practical management perspective, a ﬁsh stock is a collection of ﬁsh of a species that use speciﬁc localities and habitats. Such a collection of ﬁsh may or may not have genetic integrity, but distinctive genomes correlate with distinct behaviors and population properties that guide management. As a management unit, some consistency in population, habitat, and behavioral properties is important because these factors relate to the yield of the ﬁshery on a stock. Typically, ﬁshery managers distinguish stocks on the basis of genetic, meristic, and morphometric characteristics. When these data suggest multiple stocks, information on the geographic variation in population properties and behaviors will usually resolve the need for stock-speciﬁc management. Information on movement patterns can also identify stock boundaries, indicate times of stock mixing that confuse genetic results, and conﬁrm that population differences among distinct habitats are due to local environmental causes rather than genetic causes. The use of electronic tags for understanding and deﬁning behaviors that determine stock distribution, movements, and habitat use provides a potentially valuable tool for ﬁsheries management. Good information on ﬁsh behavior complements the genetic and population information needed for ﬁsh stock identiﬁcation and management, and behavioral knowledge helps design effective management measures. Some examples of stock deﬁnition investigations demonstrate the value of assembling genetic, population, and behavior information on a species dispersed Stock Identiﬁcation Methods Copyright © 2005 by Academic Press. All rights of reproduction in any form reserved.

435

436

Mark B. Bain

across a region or series of habitats. Walleye pollock (Therogra chalcogramma) of the north Paciﬁc Ocean supports one of the largest single-species ﬁsheries in the world. Recognized differences in growth and size among ocean regions (e.g., Aleutian basin, southeast portion of the Bering Sea) could be due to a stock structure in this species or regional environmental differences. Hinckley (1987) investigated the spatial and temporal distribution of walleye pollock in combination with population properties such as length at age, fecundity, egg-stage periodicity, and length–weight relationships. Clear differences in spatial distribution combined with distinct population attributes suggested a multiple stock structure. Movement information from standard marking studies indicated no mixing among regions and the existence of distinct spawning areas, thereby making a complete case for multiple stock management. In the case of New Zealand barracouta (Thrysites atun), genetic analyses indicated regional variations that could be the product of gradual variation suggestive of weak stock differentiation (Gauldie and Johnston, 1980). Multiple stock structure for this species was supported, however, with movement data which showed that migrations associated with spawning interacted with geographically targeted ﬁshing to produce a mixed stock catch that was the source of genetic samples. Otherwise, individuals from distinct stocks of ﬁsh are separated in distribution across the shallow waters surrounding New Zealand (Hurst and Bagley, 1989). Many mackerel species are highly mobile ﬁsh with large seasonal migrations. King mackerel (Scomberomorous cavalla) have been managed as two stocks (Gulf of Mexico, Atlantic) in southeastern U.S. coastal waters with a seasonally shifting stock mixing zone along the east and west coasts of southern Florida (Sutter et al., 1990). This species lacks genetically distinct stock differences, but the two stocks have widely separated centers of distribution, regionally separated ﬁsheries and management, and limited mixed distributions on the margin of each stock’s range. In coastal waters of Australia, mackerel movement ﬁndings show sharp species differences (Begg et al., 1997). The ﬁshery for spotted mackerel (Scomberomorous munroi) is supported by a single, well-dispersed stock while the school mackerel (Scomberomorous queenslandicus) is comprised of several, locally restricted stocks that display little movement. The highly migratory Atlantic blueﬁn tuna (Thunnus thynnus) has been observed to travel between the Gulf of Mexico and the Mediterranean Sea in conventional ﬁsh-marking studies. However, the tuna has been regarded as being distributed in east and west Atlantic stocks. The International Commission for the Conservation of Atlantic Tunas designated separate east and west management stocks. Recent analyses of trans-Atlantic movements (Block et al., 1998, 2001) show that Atlantic blueﬁn tuna move among the management stocks in signiﬁcant numbers, and that tuna from western Atlantic waters are being harvested under the relatively higher catch quotas in the eastern Atlantic Ocean. Overall, behaviors that shape distribution, movements, and habitat use

Electronic Tags

437

can often confound interpretations of genetic and population data. Therefore, a good understanding of behavior completes the context for stock management. The use of electronic tags that transmit signals or information on an animal is termed telemetry. Telemetry is an especially useful technique for revealing complex behaviors because knowing the movement history of individual ﬁsh can explain much more than simple marking and recapture results. Using ultrasonic telemetry, Zeller (1998) was able to identify four coral trout (Plectropomus leopardus) spawning sites on a portion of the Great Barrier Reef (Australia) that would have produced very confusing mark and recapture data. Coral trout maintain strict ﬁdelity to their spawning sites but are otherwise widely distributed in many sparsely inhabited areas. Although coral trout spawning aggregations seemed to provide an opportunity for stock assessment, the telemetry also revealed that less than one-third of the adult coral trout participate in annual spawnings, so censusing the aggregations would be misleading. A contrasting set of ﬁndings were obtained using telemetry on razorback suckers (Xyrauchen texanus) in the Green River (Utah; Modde and Irving, 1998). Few individuals of this endangered species are available to study so the efﬁciency of learning from individual ﬁsh movements is important. The razorback suckers move considerably within the Green River system, occupying multiple spawning sites and foraging areas, indicating the river supports one small population. The occurrence of the ﬁsh in widely separated habitats suggested greater abundance and potentially mixed stocks until the telemetry ﬁndings revealed that single ﬁsh could account for multiple, widely distributed occurrences. White sharks (Carcharodon carcharias) have been studied almost entirely in coastal waters, and the species has been regarded as primarily a continental shelf inhabitant. Once telemetry methods were used (Boustany et al., 2002), it became clear their life history includes a long and extensive oceanic migration period with regular use of deepwater habitats. These cases show how important telemetry studies can be in understanding complex life histories and behaviors, and how telemetry ﬁndings can change what seems to be established understanding of a species.

II. THE TECHNOLOGY Telemetry technology has advanced greatly in the past 5 years with the integration of multiple types of transmitters and sensors, but all systems rely on communicating information from ﬁsh to biologist by way of energy transmissions in sound [ultrasonic telemetry, 20–300 kilohertz (kHz)] or radio [radio telemetry, ca. 20–300 megahertz (MHz); satellite UHF radio signal, 401.650 MHz] frequencies. The basic ultrasonic and radio telemetry methods are described in Winter (1983), and there are numerous studies cited in Winter’s review

438

Mark B. Bain

employing these techniques for ﬁsheries investigations. Recent advancements in aquatic telemetry have yielded transmitters that combine ultrasonic and radio transmissions, incorporate sensed data (e.g., water pressure and temperature), allow within-tag data storage, and use satellites for data transmission. These new and integrated capabilities are reviewed by Lucas and Baras (2000) and CATAG (2003). In addition, manufacturers of telemetry equipment have been making regular improvements in features and capability, and their company web sites provide details on current technologies [see CATAG (2003) and Lucas and Baras (2000) for web addresses]. Manufacturers will provide detailed advice and instructions for study and equipment needs: optimized sets of transmitter tags, signal detectors (antennas, hydrophones), receivers, and data recorders. Both radio and ultrasonic transmitter tags are sized to the ﬁsh so that tag weight is no greater than 2% of the typical study ﬁsh weight in water. The maximum tag weight comes from the capability of freshwater ﬁsh to adjust buoyancy with the swim bladder (7% of a ﬁsh volume, 25% adjustment range, 1.75% body weight; Alexander, 1966), and observations of ﬁsh performance (Gallepp and Magnuson, 1972; Stasko and Pincock, 1977; Winter, 1996). The largest determinant of tag weight is battery size, so larger ﬁsh can be tagged with larger tags that transmit with stronger signals and last longer in ﬁeld use. In practice, most investigators use the largest tag acceptable for their study species and from this constraint, they determine the tag detection range and expected transmission duration. Often, this information describes the key factors that limit study design or the suitability of telemetry for a speciﬁc study need. Transmitter tags are typically coated in biologically and signal inert material for either internal or external attachment to ﬁsh. Ultrasonic transmitters emit pulsed (beeping) acoustic signals in water, and these signals are detected with a hydrophone (listening device immersed in the study waters) and a variable frequency receiver. Listening for ultrasonic tag signals is done manually with a hydrophone and boat or by a ﬁxed hydrophone with an automated signal recording device linked to a computer for data storage. Ultrasonic transmitters are the preferred telemetry gear for brackish or marine waters and deep (>5m) habitats where radio telemetry cannot be used (discussed later). Tagged ﬁsh can often be detected at distances up to 3km, but the signal can only be detected in water (i.e., no aircraft or land-based reception). Vegetation, sharp water density changes (e.g., thermoclines), entrained air (turbulence), boat engines, and high suspended material concentrations can degrade signal transmission. Acoustic signals can also be reﬂected off rock, water density gradients, and underwater structures, although experienced ﬁeld investigators can often learn to identify and circumvent these complications. Individual ﬁsh are generally identiﬁed by the transmission frequency of the tag they carry. However, when many tagged ﬁsh are aggregated, it is difﬁcult or impossible to isolate each signal. In these situations, ultrasonic transmitters can be constructed to send signals that

Electronic Tags

439

vary in pulse rate (pulses per second), pulse interval (varied times between pulses), pulse duration (length of pulse time, can be combinations of variable duration), and combinations of these signal attributes. Pulse attributes can also be used to transmit sensor information on water depth (pressure sensor) and temperature being experienced by the tagged ﬁsh. Ultrasonic tags with detection ranges of from 1 to 3km typically have durations of 10 to 50 months, and these tags can be used with ﬁsh that are at least 0.4 to 1.2 kg (Fig. 21-1) in weight. Overall, ultrasonic telemetry works well in many conﬁned marine environments because the tag transmits signals well in saltwater but the detection range is limited. Radio transmitters emit radio frequency signals from wire antennas through water and into the atmosphere. The signals are received with a tuned antenna of varied sizes and designs, including aircraft-mounted antenna. A radio signal receiver is used to detect and amplify the signals for listening with headphones or recording by a computer. Radio telemetry is preferred in low-conductivity freshwaters, shallow (<5m) habitats, and turbulent waters. The detection range of radio-tagged ﬁsh varies by ﬁsh depth as radio signals are rapidly diminished by water, especially if dissolved solid concentrations are high (high conductivity). Consequently, ﬁsh tagged with radio transmitters occupying habitats at water depths greater than about 5m will be difﬁcult or impossible to detect from boats or aircraft. When tagged ﬁsh are in surface waters, boat-mounted antennae can generally detect ﬁsh from 1 to 3km away but aircraft can often detect radio signals from over many kilometers. Power and radio transmissions, land vegetation and buildings, and industrial facilities can degrade signal detection by signal interference. Individual ﬁsh are identiﬁed by the radio signal frequency, which is determined by tuning the receiver frequency settings. As with ultrasonic tags, radio transmitters can be built to pulse the signal transmission to communicate information on water depth and temperature. For ﬁsh from 1 to 2kg in weight, radio tags would be used that have an expected duration of about 12 to 18 months (Fig. 21-2). Overall, radio telemetry works well with ﬁsh that reside in surface waters in freshwater environments. Radio telemetry is especially effective for highly mobile ﬁsh in rivers with little deepwater (migrating anadromous species) because aircraft surveys will often locate ﬁsh that have dispersed outside the survey range of boat-mounted ﬁeld crews. Telemetry technology advancements in recent years have been built around the basic ultrasonic and radio transmitters. Tags are now constructed with both radio and ultrasonic transmitters for studies of ﬁsh in estuarine waters and diadromous species. Sensors linked to variable signal patterns for remote data detection have been expanding beyond water depth and temperature to include activity and physiological data such as heart rate, swimming activity, and tail-beat frequency. Light sensors have been added to telemetry tags to detect sunrise and sunset so that latitude and longitude can be estimated using day length and time

FIGURE 21-1. General relations between ultrasonic telemetry tag weight (g), transmission duration (months, m), and minimum acceptable ﬁsh size (kg). Three sets of relations are shown for tags differing in signal detection range: 0.5, 1, and 3km. [Developed using ultrasonic tag speciﬁcations from Sonotronics, Tucson, Arizona (http://www.sonotronics.com) and tag weight limit discussed in the text.]

440

FIGURE 21-2. General relations between radio telemetry tag weight (g), transmission duration (m), and minimum acceptable ﬁsh size (kg). [Developed using radio tag speciﬁcations from Advanced Telemetry Systems, Inc., Isanti, Minnesota (http://www.atstrack.com) and tag weight limit discussed in the text.]

441

442

Mark B. Bain

of relative midday. However, the accuracy of geolocation using this technique is about ±1° (Welch and Eveson, 1999). On-tag data storage has been added (1–2 Mbytes) to record frequent readings on ﬁsh activity, physiology, and the local environment for later recovery. Now far beyond simple signal issuing tags, these digital devices are being called archival tags for their data recording function or platform terminal transmitters for their variety of data detection, storage, and transmission components. With the development of archival tags, some means of retrieving the recorded data was needed. Tags have been developed that transmit data periodically to a satellite receiver that communicates with a ground-based computer monitor. The satellite relay option for telemetry emerged with the deployment of the ARGOS satellite by the French Centre National d’Etudes Spatiales and the U.S. National Aeronautics and Space Administration with the National Oceanic and Atmospheric Administration. While these telemetry systems have been used mainly in studies of large surfacing animals, they allow remote and continuous monitoring of movement, behaviors, and local environments. Most studies employing this technique have involved terrestrial animals, but several studies illustrate the methods and capabilities for walruses (Born and Knutsen, 1992), sea turtles (Renaud and Carpenter, 1994; Gitschlag, 1996; Hughes et al., 1998; Luschi et al., 1998), whales (Mate et al., 1997, 1998), porpoises (Read and Westgate, 1997), and ﬁsh in shallow freshwaters (Eiler, 1995). Recapturing tagged ﬁsh is sometimes possible, and even if recaptures are few, the amount of data retrieved from one tag can be highly valuable. Very recently, though, an alternative for marine ﬁsh has been developed: the pop-up, satellite transmitting, archival tag. The tags are tethered to the ﬁsh externally with a short wire segment that is easily corroded with low power input at a scheduled time (Block et al., 1998). The tags are buoyant, surface with an antenna above the waterline, and begin transmitting all data to a satellite via radio signals. Currently, pop-up archival tags are large and expensive, making them suitable for study of only very large marine ﬁsh. In time, however, the direction of technology development will likely bring this level of telemetry capacity to many ﬁsh biology problems.

III. TELEMETRY METHODS Once decisions are made about the proper telemetry system for a species and habitat, the method of attaching transmitters to ﬁsh needs attention. The best attachment method will depend on characteristics of the species under study, size of the tag, the type of transmitter, and the duration of the monitoring period. Discussions with experienced investigators will often provide the best basis for deciding about attachment method. Winter (1983) provides an extensive list of published investigations using a wide array of attachment methods, and recent

Electronic Tags

443

information can be found in CATAG (2003). External transmitter attachments are commonly made by wiring the tag through the dorsal surface of the ﬁsh at the base of the dorsal ﬁn. This attachment technique works best with ﬁsh having heavy dorsal musculature, large scales or bony plates, and when the tag weight is small relative to the ﬁsh weight (much less than 2%). External attachment avoids surgical procedures used for internal implants, is quick to accomplish in the ﬁeld, and involves relatively little handling stress to the ﬁsh. The disadvantages are mostly related to loss of tags over time such as when ﬁsh rub off the tags tearing tissue and scales, species occupy vegetated and high cover habitats, and when tags interfere with swimming. Internal tag attachments require surgery because the tags are implanted in the body cavity. With practice, surgical implants can be done quickly but this attachment method requires suturing the implant incision and often sedation of the ﬁsh. The body wall of ﬁsh does not interfere with ultrasonic signal transmission, but optimum radio transmission requires the transmission antennae be left on the outside of the ﬁsh. Thus the wire antenna will protrude from the suture wound and could permit internal infection through the persistent opening. Despite the appearance of signiﬁcant tissue damage and infection risk, internal tag implants have resulted in almost no ﬁsh mortality in my experience with the technique in single year studies (Bain et al., 1990; Bain and Boltz, 1992). Regardless of the attachment technique selected, the procedure should be practiced on preserved or fresh ﬁsh. Once the tag is attached to a ﬁsh, most investigators now recommend releasing the ﬁsh immediately on recovery of swimming ability rather than holding newly tagged ﬁsh in tanks for hours or days to observe their behavior. Locating tagged ﬁsh or tracking is rarely difﬁcult if the approximate location of tagged ﬁsh is known. Both ultrasonic and radio transmissions can be identiﬁed from a distance of 1km or more using boats. When two or more signal directions are established, triangulation with maps, compasses, and global positioning system devices will allow the ﬁsh location to be estimated. Often, a ﬁeld investigator will then approach the tagged ﬁsh location in a boat until the signal properties (high signal strength and lack of directional variation) indicate the investigators are close or above a tagged ﬁsh. Field experience readily allows investigators to conﬁdently identify tagged ﬁsh locations. This skill can be mastered with tags placed in likely ﬁsh locations on lines attached to marked buoys. The practice tags can be retrieved and used later for ﬁsh tagging since all tags have magnetic on and off switches. Field practice also indicates the accuracy of locating tagged ﬁsh because some variation in location designation occurs due to signal interference, reﬂected signals, and lack of directional transmission cues at close distances. The frequency of tracking will generally be determined by study objectives and the ability to estimate ﬁsh locations within detection range (typically a few kilometers). Highly mobile species in large waters will require frequent

444

Mark B. Bain

monitoring to avoid losing track of the study ﬁsh. Brill et al. (2002) were able to monitor the movements of Atlantic blueﬁn tuna in the North Atlantic using ultrasonic tags, but they had to follow one ﬁsh at a time continuously for a maximum of 48hr with a tracking boat. Remote, computer signal monitors (hydrophone or radio antennas, receiver, and computer) can be used to monitor the movement of tagged ﬁsh through narrow (<2km) passes and river locations, but with ﬁeld experience most investigators ﬁnd the effort to maintain monitoring stations counterproductive. Wide-ranging ﬁsh in large waters can also be located with aircraft surveys if radio transmitters are used, and the ﬁsh remain at shallow depths of freshwaters. Overall, most investigators seem to quickly master locating tagged ﬁsh if proper transmitters are used for the habitats under study.

IV. CONCLUSIONS The acquisition of thorough information on ﬁsh behaviors that shape distributions, movements, and habitat use substantially improves the value and interpretation of genetic and population data for stock management. Telemetry is an excellent, high-resolution technique to obtain behavioral information in many situations. For marine ﬁshery species, two key aspects of telemetry systems have traditionally limited the utility of this research technique: very poor radio signal transmission in salt water, and the limited detection range (1–3km in water) of ultrasonic telemetry for open-water studies. However, the recent technological advances reported here have circumvented these impediments for large ﬁsh, and new ﬁshery-relevant information on well-known species is now being reported. Traditional radio and ultrasonic telemetry has had broad utility for species migrating into freshwaters or making regular movements within enclosed coastal habitats. Performance capabilities and limitations are well known, selecting equipment and procedures are typically straightforward, and proper study designs no longer need to be characterized as trial and error. Potential investigators and ﬁshery agencies can proceed conﬁdently with telemetry if reasonable efforts are made to review the basic information presented here, communicate with investigators experienced with similar species and habitats, and explore equipment design options with well-established suppliers. Overall, marine aquatic telemetry should be viewed as a potentially useful research technique with rapidly growing capability to understand distributions and migrations, and to reveal complexities of very speciﬁc behaviors.

ACKNOWLEDGMENTS The author’s telemetry studies have been supported by grants from the Tennessee Valley Authority, The Hudson River Foundation, New York Sea Grant Institute, and the U.S. Army Corps

Electronic Tags

445

of Engineers. Christopher Solomon assisted with assembling literature on marine applications of telemetry, and Jordan Gass edited the manuscript. Sonotronics and Advanced Telemetry Systems, Inc. (Christopher Kochanny) reviewed the use of product data presented in the ﬁgures.

REFERENCES Alexander, R. M. 1966. Physical aspects of swim bladder function. Biological Reviews 41: 141–176. Bain, M. B., Webb, D. H., Tangedal, M. D., and Mangum, L. N. 1990. Movements and habitat use by grass carp in a large mainstream reservoir. Transactions of the American Fisheries Society 119: 553–561. Bain, M. B. and Boltz, S. E. 1992. Effect of aquatic plant control on the microdistribution and population characteristics of largemouth bass. Transactions of the American Fisheries Society 121: 94–103. Begg, G. A., Cameron, D. S., and Sawynok, W. 1997. Movements and stock structure of school mackerel (Scomberomorous queenslandicus) and spotted mackerel (S. munroi) in Australian eastcoast waters. Australian Journal of Marine and Freshwater Research 48: 295–301. Block, B. A., Dewar, H., Farwell, C., and Prince, E. D. 1998. A new satellite technology for tracking the movements of Atlantic blueﬁn tuna. Proceedings of the National Academy of Sciences USA 95: 9384–9389. Block, B. A., Dewar, H., Blackwell, S. B., Williams, T. D., Prince, E. D., Farwell, C. J., Boustany, A., Teo, S. L. H., Seitz, A., Walli, A., and Fudge, D. 2001. Migratory movements, depth preferences, and thermal biology of Atlantic blueﬁn tuna. Science 293: 1310–1314. Born, E. W. and Knutsen, L. O. 1992. Satellite-linked radio tracking of Atlantic walruses (Odobenus rosmarus rosmarus) in northeastern Greenland, 1989–1991. Z. S¸getierkunde 57: 275–287. Boustany, A. M., Davis, S. F., Pyle, P., Anderson, S. D., Le Boeuf, B. J., and Block, B. A. 2002. Expanded niche for white sharks. Nature 415: 35–36. Brill, R., Lutcavage, M., Metzger, G., Bushnell, P., Arendt, M., Lucy, J., Watson, C., and Foley, D. 2002. Horizontal and vertical movements of juvenile blueﬁn tuna (Thunnus thynnus) in relation to oceanographic conditions of the western North Atlantic, determined with ultrasonic telemetry. Fishery Bulletin 100: 155–167. CATAG. 2003. Improvements in tagging methods for stock assessment and research in ﬁsheries. Concerted Action for Tagging of Fishes Project, Marine Research Institute, Directorate of Fisheries, Reykjavik, Iceland. http://www.hafro.is/catag/ Eiler, J. H. 1995. A remote satellite-linked tracking system for studying Paciﬁc salmon with radio telemetry. Transactions of the American Fisheries Society 124: 184–193. Gitschlag, G. R. 1996. Migration and diving behavior of Kemp’s ridley (Garman) sea turtles along the U.S. southeastern Atlantic coast. Journal of Experimental Marine Biology and Ecology 205: 115–135. Gallepp, G. W. and Magnuson, J. J. 1972. Effects of negative buoyancy on the behavior of the bluegill, Lepmois macrochirus Reﬁnesque. Transactions of the American Fisheries Society 101: 507–512. Gauldie, R. W. and Johnston, A. J. 1980. The geographical distribution of phosphoglucomutase and glucose phosphate isomerase alleles of some New Zealand ﬁshes. Comparative Biochemistry and Physiology 66B: 171–183. Hinckley, S. 1987. The reproductive biology of walleye pollock, Theragra chalcogramma, in the Bering Sea, with reference to spawning stock structure. Fishery Bulletin 85: 481–498. Hughes, G. R., Luschi, P., Mencacci, R., and Papi, F. 1998. The 7000-km oceanic journey of a leatherback turtle tracked by satellite. Journal of Experimental Marine Biology and Ecology 22: 209–217.

446

Mark B. Bain

Hurst, R. J. and Bagley, N. W. 1989. Movements and possible stock relationships of the new Zealand barracouta, Thrysites atun, from tag returns. New Zealand Journal of Marine and Freshwater Research 23: 105–111. Lucas, M. C. and Baras, E. Baras. 2000. Methods for studying spatial behaviour of freshwater ﬁshes in the natural environment. Fish and Fisheries 1: 283–316. Luschi, P., Hays, G. C., Del, S. C., Marsh, R., and Papi, F. 1998. The navigational feats of green sea turtles migrating from Ascension Island investigated by satellite telemetry. Proceedings of the Royal Society of London Series B Biological Sciences 265: 2279–2284. Mate, B. R., Nieukirk, S. L., and Kraus, S. D. 1997. Satellite-monitored movements of the northern right whale. Journal of Wildlife Management 61: 1393–1405. Mate, B. R., Gisiner, R., and Mobley, J. 1998. Local and migratory movements of Hawaiian humpback whales tracked by satellite telemetry. Canadian Journal of Zoology 76: 863–868. Modde, T. and Irving, D. B. 1998. Use of multiple spawning sites and seasonal movements by razorback suckers in the middle Green River, Utah. North American Journal of Fisheries Management 18: 318–326. Read, A. J. and Westgate, A. J. 1997. Monitoring the movements of harbour porpoises (Phocoena phocoena) with satellite telemetry. Marine Biology 130: 315–322. Renaud, M. L. and Carpenter, J. A. 1994. Movements and submergence patterns of loggerhead turtles (Caretta caretta) in the Gulf of Mexico determined through satellite telemetry. Bulletin of Marine Science 55: 1–15. Stasko, A. B. Pincock, and D. G. 1977. Review of underwater biotelemetry with emphasis on ultrasonic techniques. Journal of the Fisheries Research Board of Canada 34: 1261–1285. Sutter, F. C., III., Williams, R. O., and Godcharles, M. F. 1990. Movement patterns and stock afﬁnities of king mackerel in the southeastern United States. Fishery Bulletin 89: 315–324. Welch, D. W. and Eveson, J. P. 1999. An assessment of light-based geoposition estimates from archival tags. Canadian Journal of Fisheries and Aquatic Sciences 56: 1217–1327. Winter, J. D. 1983. Underwater biotelemetry. In L. A. Nielsen and D. L. Johnson (eds.), Fisheries Techniques. American Fisheries Society, Bethesda, MD, pp. 371–395. Winter, J. D. 1996. Advances in underwater biotelemetry. In B. R. Murphy and D. W. Willis (eds.), Fisheries Techniques. American Fisheries Society, Bethesda, MD, pp. 555–590. Zeller, D. C. 1998. Spawning aggregations: patterns of movement of the coral trout Plectropomus leopardus (Serranidae) as determined by ultrasonic telemetry. Marine Ecology Progress Series 162: 253–263.

CHAPTER

22

Otolith Thermal Marking ERIC C. VOLK, STEVEN L. SCHRODER, AND JEFFREY J. GRIMM Washington Department of Fish and Wildlife, Olympia, Washington, USA

I. Introduction II. Inducing Otolith Marks A. Temperature Effects on Otoliths B. Organizing Pattern Information III. Errors in Mark Recovery IV. Applications V. Conclusion References

I. INTRODUCTION A basic challenge for ﬁsh biologists is to benignly mark or tag juvenile ﬁsh so their origins can be identiﬁed upon recapture. Because juvenile ﬁshes often experience high natural mortality, large numbers must be marked to obtain meaningful data on growth, migration, and survival. The use of individual tags on small ﬁshes, like newly emerged salmonids, is often not practical because of expense and physiological impacts. Even the most successful method for individual extrinsic tagging, the coded-wire tag ( Jefferts et al., 1963), cannot meet current demands to identify vast numbers of supplemented salmon in the North Paciﬁc, and questions remain about the biological impacts of tags on juvenile ﬁsh. Thus, developing techniques to mark salmonids en masse is essential to keep pace with ﬁsh-marking needs currently facing ﬁsheries managers and researchers, especially in view of the growing number of salmon released from hatcheries. Over the past two decades, a novel approach to mass-marking juvenile salmon has evolved among Paciﬁc Rim nations. It relies on short-duration temperature ﬂuctuations to alter otolith structural features as a means to induce speciﬁc codes within developing otoliths. The use of temperature changes to induce otolith marks is attractive since they seem benign; they can be delivered to large numbers of incubating embryos and larvae simultaneously; they are permanent; and the Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

447

448

Eric C. Volk et al.

technology for administering and recovering the induced mark is simple, inexpensive, and commonly available. Several projects converged on the notion of using altered water temperatures to induce unique increment patterns onto ﬁsh otoliths. Brothers (1985) published the ﬁrst results showing that the sensitivity of otolith increment characteristics to temperature changes could be used to induce speciﬁc patterns in otoliths of lake trout (Salvelinus namaycush), functioning as a permanent identiﬁer for that ﬁsh. The practical application of this mark was described in Bergstedt et al. (1990). At the same time, Mosegaard et al. (1987) showed the potential of using thermally induced otolith patterns for marking Atlantic salmon (Salmo salar). Volk et al. (1990) demonstrated its feasibility for Paciﬁc salmon, which represented the logical arena for extensive application of the technique in view of the large hatchery salmon program on the Paciﬁc Rim. In 1987, the Washington Department of Fish and Wildlife began marking large numbers of hatchery salmon on a regular basis and created a laboratory where thermally marked otoliths could be processed. Fisheries agencies in Alaska, Canada, Japan, and Russia (Akinicheva and Rogatnykh, 1996) soon followed with similar programs, and scientists from several other U.S. states and European countries have experimented with marking juvenile salmonids via temperature ﬂuctuations. To date, otolith thermal marking has been applied to six species of Paciﬁc salmon (Oncorhynchus spp.), rainbow trout/steelhead (O. mykiss) (Van der Walt and Faragher, 2002), Atlantic salmon (S. salar), brown trout (S. trutta), cutthroat trout (S. clarki), and lake trout (Salvelinus namaycush). In this chapter, we review the underlying basis for thermal marking, discuss how thermal marks are applied, describe methods for creating mark codes to organize information on otoliths, and note some of the challenges and drawbacks of the technique. We also discuss current applications of the technique, focused on salmonid ﬁshes on the Paciﬁc Rim.

II. INDUCING OTOLITH MARKS

A. TEMPERATURE EFFECTS

ON

OTOLITHS

Otoliths are primarily (~95%) composed of calcium carbonate as the aragonite mineral morph (more rarely vaterite or calcite) (Carlstrom, 1963; Degens et al., 1969), an array of trace elements (Calaprice et al., 1975; Edmonds et al., 1995; Campana and Gagné,1995), and an organic protein that functions as a template for the deposition of calcium carbonate (Degens et al., 1969; Dunkelberger et al., 1980). Although organic matrix is distributed throughout the growing otolith, the characteristic dark and light bands observed in sectioned otoliths reﬂect the bipartite nature of otolith increments, each consisting of a calcium-rich compo-

Otolith Thermal Marking

449

nent, translucent when viewed with transmitted light, and an organically rich component, optically dense under transmitted light. Nomenclature suggested in Secor et al. (1995) refers to the organically rich portion of otolith increments as D-zones, or discontinuous zones, and the translucent portion, dominated by crystalline calcium carbonate, as L-zones, or incremental zones. We adopt this convention here; however, we also refer to any otolith increment with an enhanced D-zone as a dark, or optically dense increment, and regions consisting of several otolith increments as dark or light zones. These terms are used for simplicity and reﬂect the general appearance of otolith marks when viewed with transmitted light. The basis for otolith thermal marking rests in the fundamental relationship between environmental temperature ﬂuctuations and the appearance of regularly deposited otolith increments (Brothers, 1981; Campana and Neilson, 1982, 1985; Neilson and Geen, 1985). Some impetus for otolith marking arose from observations in nature (Brothers, 1990). The idea behind using short-term temperature manipulations to mark juvenile ﬁsh otoliths is to alter the appearance of D- and L-zones in one or more increments to produce an obvious pattern of events. A dramatic illustration at the heart of thermal marking is the comparison of otolith increment characteristics resulting from diurnally variable temperature regimes against those produced by relatively constant water temperature. Figure 22-1 shows an otolith from a juvenile coho salmon (O. kisutch) that was incubated shortly after hatching in water intentionally depressed 4°C for 8hr daily, followed by constant temperature rearing. Under the diurnally ﬂuctuating thermal regime, otolith D-zones were dramatically contrasted, producing an obvious series of dark increments. The otolith region corresponding to nonﬂuctuating water temperatures showed poorly contrasted increments that were far less obvious than those produced by the diurnally changing temperature regime. Early experiments with Chinook salmon (O. tshawytscha) showed that exposures to rapid temperature declines of 3.5°C for as little as 30 min produced obvious effects on otoliths and that these became more pronounced as exposure times increased (Volk et al., 1999). A simple otolith mark pattern using 2-hr duration chilled water events is shown in Figure 22-2. Longer exposure periods on three salmonid species showed that sudden temperature declines, lasting 4– 24 hr, produced distinctively dark increments on the day that the temperature decline occurred (Mosegaard et al., 1987; Brothers, 1990; Volk et al., 1994). Figure 22-3 shows the dramatic impacts from a series of 8-hr thermal depressions recorded in a juvenile coho salmon otolith (lapillus). Thermal declines lasting 48hr to 15 days on juvenile chum salmon (O. keta) also produced a very dark otolith increment on the day of the temperature decline, followed by optically dense otolith zones whose width was proportional to the duration of the exposure to cool temperatures (Volk et al., 1990).

FIGURE 22-1. Sectioned otolith from a juvenile coho salmon exposed to two different water temperature regimes. The region marked (A) represents a ﬂuctuating regime where water temperature was lowered 4°C daily for 8hr over a 21-day period. The region marked (B) corresponds to a thermal regime where diurnal ﬂutuations were <0.5°C. Scale bar = 50mm.

FIGURE 22-2. Otolith mark on the sagittal otolith of a juvenile Chinook salmon. The mark pattern, indicated by dots, is represented by two series of three events near the center of the photograph. Each event was created with 2-hr exposures to water chilled 4°C below ambient. Photograph was taken with transmitted light. Scale bar = 50mm. 450

Otolith Thermal Marking

451

FIGURE 22-3. Thermal marks created in a juvenile coho salmon lapillus otolith with a series of 8hr thermal depressions 4°C below ambient. Photograph was taken with transmitted light. Scale bar = 50mm.

Elevating water temperature above ambient also produces predictable impacts on otoliths. Brothers (1990) showed that short immersions in warm water result in wider L-zones in the otoliths of lake trout, presumably due to growth rate effects of warmer water, and Bergstedt et al. (1990) showed impacts of 14-hr heating cycles on lake trout otoliths. This effect is demonstrated with a coho salmon otolith in Figure 22-4, where two lighter zones created by temperatures elevated 2.5°C over ambient for 4 days surround a dark zone created on return to cooler, ambient temperature water. A regular series of extended warm water events alternating with cooler, ambient water can create very clear patterns on otoliths (Fig. 22-5). Effects of the magnitude of temperature changes on otolith increments are less well studied, and observations typically are qualitative and subjective. Nonetheless, it does appear that at least to a point, impacts of sudden temperature changes on otoliths increase with their magnitude (Volk et al., 1990; Negus, 1999).

FIGURE 22-4. Thermal mark on a juvenile chum salmon sagitta induced with two warm water events, approximately 2.5°C above ambient. In (A), the arrow indicates the location of the mark. Light zones correspond to warm water periods and the intervening dark zone was created on return to cooler, ambient water. Scale bar = 50mm. (B) High-magniﬁcation view of the same mark. Scale bar = 10mm. Both photographs were taken with transmitted light. 452

FIGURE 22-5. The effects of alternating exposure to warm and cold incubation temperatures on the sagittal otolith of a juvenile coho salmon. Exposure times were 2 days each for elevated warm or ambient cold cycles, with temperatures elevated about 2.5°C above ambient. Photograph (A) shows the entire mark pattern and photograph (B) shows the alternating density differences for exposures to warm water (light zone) and colder ambient water (dark zone). Both photographs were taken with transmitted light. Scale bars = 50mm (A) and 10mm (B). 453

454

Eric C. Volk et al.

These examples demonstrate that even relatively minor temperature changes may have obvious impacts on otoliths, and producing patterns with periodic events should be quite simple. However, an important aspect of inducing patterns on otoliths with periodic water temperature manipulations is an appreciation of the ambient thermal regime’s impact on increment characteristics, against which induced patterns must be recognized. Otolith mark recognition can be thought of as a signal-to-noise ratio problem where the induced mark signal must be distinguished against the background incremental “noise” created by the ambient thermal regime. Where ambient temperatures are relatively uniform with little diurnal or short-term ﬂuctuation, otolith increments typically will be poorly contrasted, and even small temperature change effects on increments will be obvious against this background. Conversely, where ambient thermal regimes ﬂuctuate, producing highly contrasted otolith increments with dark D-zones, much more dramatic temperature changes will be necessary to make induced increment effects stand out against this background. This problem is easy to visualize if one imagines inducing marks to the two regions depicted in Figure 22-1. Clearly, a subtle mark will be easier to identify in the outer region, amidst a constant temperature background pattern, than against the highly contrasted increments produced by ﬂuctuating temperatures. Several early thermal marking efforts failed to induce recognizable patterns due to insufﬁcient temperature change duration or magnitude in the face of ﬂuctuating ambient regimes (Volk et al., 1999). For marking with chilled water, our experience shows that exposure times of less than 12 hr should be used only where ambient temperature regimes ﬂuctuate little and that 24-hr or greater exposures are preferable for creating distinctive marks. Ultimately, if one uses a sufﬁciently large temperature decline or increase lasting at least 24hr, most background structural noise can be overshadowed by the mark event.

B. ORGANIZING PATTERN INFORMATION The fact that relatively subtle environmental events are recorded in otolith increment patterns is both the basis of otolith thermal marking and one of the major challenges for readers recovering the marks. Any strategy that helps readers decipher induced patterns amidst ambient increment characteristics increases success and efﬁciency. Once one has devised a system where groups of ﬁsh can receive temperature changes on a predetermined schedule, there are clearly a great variety of options that can be pursued to create different marks. These range from a simple set of regularly repeated thermal events, to more complex patterns and combinations of patterns. The selection of a scheme depends on the goals of the study or application. Generally speaking, the use of a pattern rather than one or two discrete events not only makes a mark easier to recognize, but also helps

Otolith Thermal Marking

455

avoid the problem of spurious events confusing the distinction between marked and unmarked ﬁsh. Even when thermal regimes are fairly constant, unaccounted for increments reminiscent of intended thermal events appear in otoliths, perhaps because of handling or the artiﬁcial and irregular light cycles found in hatcheries. Both Volk et al. (1990) and Bergstedt et al. (1990) have noted slight discrepancies between the number of thermal events ﬁsh were exposed to and the number of thermal marks observed in otoliths. Hatchery incubated salmonids are particularly well suited for thermal marking because incubation and yolk absorption stages are protracted, large numbers of ﬁsh are concentrated in these facilities, and otoliths begin growing in embryos. As a result, there is a lengthy period during which multiple marks may be administered, including the pre- and posthatch otolith zones. The frequent occurrence of an otolith “check” mark associated with hatching in salmonids conveniently separates these two regions and provides an opportunity to encode different information in each region (Volk et al., 1990; Hagen et al., 1995). Species whose otolith does not form until near hatching offer less ﬂexibility for pattern induction, since time to perform appropriate water temperature manipulations is more limited. Otolith thermal marking has been applied almost exclusively to salmonid ﬁshes. Over the past 10 years, a variety of strategies have been employed to place recognizable patterns on otoliths using periodic thermal events. Projects with limited scope and few marks to identify may employ a simple series of evenly spaced water temperature changes of 24 to 48-hr duration (Bergstedt et al., 1990; Volk et al., 1990; Hagen et al., 1995). An example of such a mark, employing several series of daily temperature depressions in both the pre- and posthatch regions of a juvenile Chinook salmon otolith is shown in Figure 22-6. However, applications of otolith thermal marking for identifying salmon stocks in the North Paciﬁc have grown enormously in the past decade, and it has become clear that systems for organizing information on the otolith are necessary to avoid mark duplication. The most widely used system for inducing and describing thermal mark patterns is the RBr notation (Munk and Geiger, 1998; Hagen, 1999). The other coding system in use is a variation on a common bar code symbology (Volk et al., 1994). Both of these systems use induced band number, relative spacing between bands, and the position of groups of induced bands relative to one another in speciﬁc otolith regions to encode information. Naturally, one can imagine any number of ways to create patterns on otoliths, and these two systems were developed to create a standardized approach. In theory, there are truly an enormous number of patterns available, given the time needed to induce marks and the space available on otoliths to do so. Certainly, small-scale studies conducted in limited geographic regions would have no problem with pattern duplication. However, practical limitations associated with hatchery operations, ﬁsh development, and visual recognition may place important limits on the actual

456

Eric C. Volk et al.

FIGURE 22-6. Thermal marks in the sagittal otolith of a juvenile Chinook salmon. Three sets of ﬁve events were induced into the pre- and posthatch regions by daily exposures to cooler water for 8hr followed by a return to constant temperature, ambient water. Scale bar = 50mm.

number of available patterns (Hagen, 1999). The rapidly growing number of stocks being marked on an annual basis in the North Paciﬁc has created some potential mark duplication between countries and ﬁsh stocks (Urawa et al., 2001). Apparent conﬂicts often can be resolved through inspection and measurement of the otolith mark image.

III. ERRORS IN MARK RECOVERY Along with the rapidly expanding application of otolith thermal marking has come an increasing awareness that otolith mark recovery is not error-free. Decoding errors may arise from poor otolith marks, natural mimics of induced patterns, poor preparations, and simple recognition or clerical mistakes associated with data entry. In general, evaluations of error rates in mark recovery do not accompany studies that employ the marking technique, and most investigators use

Otolith Thermal Marking

457

agreement between multiple readers to discover obvious errors in recognition or data entry and to uncover those specimens that have truly confusing patterns. The mark code for confusing specimens may either be resolved by majority or simply identiﬁed with a mark status as undetermined. A few studies have sought to evaluate error rates in mark recovery. Bergstedt et al. (1990) reported that among juvenile lake trout held 6 months after marking, technicians accurately recognized marked and unmarked controls in 85% to 98% of the samples. Most errors occurred when unmarked ﬁsh were assigned a mark code. Hagen et al. (1995) accurately identiﬁed 64% to 100% of known marked and unmarked adult pink salmon (O. gorbuscha) otoliths. In another study, samples of 50 marked and unmarked juvenile Chinook salmon otoliths were scored three times by four readers (Volk et al., 1994). In this case, only one error was noted in 600 observations. This included not only the recognition of marked or unmarked status, but also distinguishing between any of 10 possible mark codes in the sample. Double-blind tests with emergent pink salmon fry showed greater than 99% success rates distinguishing marked hatchery from wild fry (Joyce and Evans, 2001). In an extensive study using 1,852 known mark status adult Chinook and coho salmon otoliths, Volk et al. (1999) reported error rates of 2% and 6% for marked and unmarked controls, with much higher error rates from a group where mark quality was known to be poor. Their results emphasized the inﬂuence of mark quality on error rates and suggested that the presence of a pattern was more easily discerned than its absence, highlighting the importance of the background structure in pattern recognition. In general, it appears that error rates have diminished as investigators have learned how to more effectively induce clear otolith marks. Unfortunately, in practical applications, it is difﬁcult to evaluate the magnitude of errors because known mark status specimens generally are not available and conclusions are usually reached by consensus among multiple readers. Blick and Hagen (2002) have discussed the use of agreement measures and latent class models for evaluating the precision of otolith mark determinations where agreement indexes such as kappa may be useful as a relative measure of the reliability of determinations with two independent readings. With a third reader, latent class models can provide estimates of the error rates for each reader. Where key assumptions can be met, these methods provide a means to evaluate the precision of mark group composition where no error-free standards exist.

IV. APPLICATIONS During nearly two decades of use, applications of otolith thermal marking in studies of salmonid ﬁshes have been very diverse. Initially, otolith thermal

458

Eric C. Volk et al.

marking found most of its utility in the research arena, where relatively small groups of juvenile ﬁsh were uniquely marked as a means to study their distribution or behavior (Quinn et al., 1994) or to distinguish hatchery origin juveniles from their naturally spawned counterparts (Hargreaves et al., 2001). Thermal marking is used extensively to evaluate the success of salmon recovery programs in the Paciﬁc Northwest, where various supplementation efforts are employed to augment critically depleted stocks. In view of domestication concerns and interactions between hatchery and naturally spawned salmon, many of these programs place embryos in streamside incubation vessels to allow natural “emergence” and migration. Without the ability to thermally mark otoliths in embryonic salmonids, it would be impossible to evaluate the success of these endeavors. In the State of Washington, for example, 11 chum salmon recovery programs use thermal marks to evaluate fry-to-adult survival, straying, and whether inadvertent domestication has resulted because of the cultural strategies being employed. While the use of thermal marking for research has been exceedingly important on a local scale, there can be little question that the most extensive and economically important application of the method has been in the stock assessment and ﬁsheries management arena. Clearly, the reason for this rests in the simplicity of inducing marks to all individuals in hatchery populations and the huge advantages conferred by such a high mark rate. Simply put, when mark rates are near 100% for a hatchery stock, their contribution can be estimated with great precision from a fairly small sample. Sampling design remains a critical issue; however, the need for very large sample numbers resulting from low mark rates is eliminated. Hagen et al. (1995) demonstrated how otolith marking could be used as an in-season estimator of stock composition for ﬁsheries management purposes in Alaska, since relatively small sample sizes could be processed within 24 hr to provide critical management information through the ﬁshing season. Continued use of this technique to manage the Prince William Sound pink salmon ﬁshery has demonstrated greater precision of hatchery contribution estimates with far smaller sample sizes and much faster results than traditional coded-wire tagging programs could provide ( Joyce and Evans, 2001). Similar success was demonstrated for in-season management of Canadian and U.S. sockeye (O. nerka) stocks in southeastern Alaska ( Jensen and Milligan, 2001). Otolith thermal marking is providing unprecedented data on high seas distribution and migratory characteristics of Paciﬁc salmon, much of which may have important implications for predicting speciﬁc run sizes (Kawana et al., 2001). Obviously, there are speciﬁc goals associated with each thermal marking effort, but important unforeseen beneﬁts commonly arise when hatchery salmon populations have been marked. For instance, many ﬁsheries in the Paciﬁc Northwest are managed to protect weak or underperforming wild stocks as they com-

Otolith Thermal Marking

459

mingle with hatchery ﬁsh that are intended for harvest. Where insufﬁcient data exist on contribution rates of hatchery stocks, ﬁsheries cannot occur, with unnecessary and adverse consequences for harvesters. Hargreaves et al. (2001) provided an excellent example of how a recreational Chinook salmon ﬁshery was maintained without undue harm to wild stocks because of abundance and distributional data provided by an existing thermal marking program. Marking efforts already in place also provide an opportunity to estimate the abundance of hatchery adults on spawning grounds where their interactions with wild or naturally spawning ﬁsh are of increasing concern to many (Rawson et al., 2001; Joyce and Evans, 2001). Since stream surveys for a variety of biological data are already occurring, if hatchery ﬁsh were marked, extremely important data would be value added to these efforts. There is little doubt that in the Paciﬁc Northwest, an organized drive to otolith mark all hatchery salmon production would provide huge beneﬁts for science and management. We currently have the ability to do so at a reasonable cost.

V. CONCLUSION It is remarkable that periodic changes in water temperature have profound enough effects on growing otoliths that regulated changes in water temperature can be used to mark ﬁsh on such a grand scale. Thermal marking has been successful with a variety of salmonid species, and at this writing, large-scale programs in Canada, Japan, Russia, and the United States are thermally marking some 20% of hatchery-produced salmon in the North Paciﬁc Ocean (Urawa et al., 2001). An estimated 1.1 billion thermally marked juveniles were released from Paciﬁc Rim hatcheries in 2003, more than 90% of which were pink and chum salmon. The impact of this marking technique on stock assessment for speciﬁc ﬁsheries management issues has been signiﬁcant, and because many hatchery stocks carry otolith marks, collected samples from a variety of settings are more routinely checked for marks. This will continue to produce important information on the distribution, abundance, and migration of Paciﬁc salmon. Thermal marking has drawn considerable attention from ﬁsheries researchers who typically work with smaller groups of ﬁsh, but need a number of distinctive marks. Some studies would not be possible without thermal marking, such as the evaluation of eyed salmon egg plants in remote incubators, where embryos must be marked in a hatchery. Aside from the one-time capital equipment costs for distributing and altering the temperature of incubation water, marking ﬁsh costs very little, and recovery of marks requires no specialized equipment. Generally, analysis costs are comparable to other techniques where tags must be dissected and read. It can be employed on a small scale by anyone willing to learn the basics of otolith preparation.

460

Eric C. Volk et al.

Like any marking method, there also are drawbacks associated with otolith thermal marking. Primary among these is the need to modify existing hatchery infrastructure in order to deliver thermal events to incubating ﬁsh. This involves considerations of power supply, water availability, and space—important and potentially limiting resources in a ﬁsh culture facility. Otolith marking may also require some level of detailed scheduling of thermal events where a number of different codes must be induced among ﬁsh fertilized on different dates. Also, we do not have complete control over the recording of information within the otolith, and environmental perturbations occurring around the marking schedule may be recorded, producing confusion and possible errors on recovery. As a result, it is important to use postmarking “voucher” otoliths to document the actual pattern induced. One of the interesting aspects of this marking approach is that much of the physiology and chemistry behind otolith increment formation and the impacts that temperature and other variables have on the structure and composition of otoliths is not completely understood. This notwithstanding, cause-andeffect observations at the heart of otolith thermal marking are clear. Finally, otolith thermal marking requires lethal sampling, which precludes its use in some circumstances. The application of otolith thermal marks to salmon will continue to grow throughout the Paciﬁc Rim. A better understanding of stock-speciﬁc abundance, distribution, and migratory characteristics of Paciﬁc salmon carries signiﬁcant economic and scientiﬁc importance. While increasing efforts to identify hatchery-produced salmonids using otolith thermal marking is a positive development, it also is clear that with a ﬁnite number of mark patterns available, some objectives may be compromised if duplicate or indistinguishable mark patterns exist. One of the challenging impediments to avoiding mark duplication has been the conﬂicting goals and commitments to marking hatchery production in different regions. For instance, in Alaska and Japan, ﬁsheries agencies are ﬁrmly committed to the application of otolith thermal marks as a normal operating principle, and the primary goal of those efforts surrounds management of economically important ﬁsh stocks. As a result, application of marks is well funded, broadly implemented, and well coordinated. On the other hand, most thermal marking applications in Washington State, Oregon, and California are principally associated with efforts to evaluate local stock rebuilding efforts, where sampling typically occurs near or in spawning rivers. These efforts usually involve the use of many different codes to identify different treatment or release groups, with little regard to pattern duplication except within the localized study arena. The problem of multiple marks is often exacerbated by limited facilities available with which to perform the marking. Moreover, planning often occurs in-season as run size information becomes available. In these scenarios, the ratio of induced patterns to numbers of ﬁsh released is high. While the total number of marked ﬁsh released by these states is very small compared to other countries and Alaska

Otolith Thermal Marking

461

(about 2.5% in 2002), they may contribute inordinately to the problem of mark duplication and confusion among high seas recoveries. The high value of otolith thermal marking to stream basin scale research suggests that this dilemma will continue. In an attempt to document and coordinate the widespread application of otolith thermal marks, a salmon marking working group was established under the Committee for Scientiﬁc Research and Statistics of the North Paciﬁc Anadromous Fish Commission (Urawa et al., 2001). This group maintains a database of existing thermal marks and attempts to coordinate mark induction to avoid duplication. In principle, countries annually submit speciﬁc mark plans for induction to the upcoming brood year so that obvious conﬂicts might be resolved prior to the commencement of marking. Following the marking season, summaries of actual marks induced are submitted and entered into the database. Users can query the database through the Internet, with a link from the NPAFC web site. Speciﬁc information on each nation’s mark groups and induced patterns is summarized, including a digital image of most mark patterns. This provides a ready source of information for determining the origins of an unknown pattern. We encourage anyone embarking on thermal marking programs to report his or her efforts to this body. The long-term utility of this marking technique depends on participatory oversight among its users.

REFERENCES Akinicheva, E. G. and Rogatnykh, A. Yu. 1996. The salmon marking experiment at ﬁsh hatcheries: Thermal marking. Journal of Ichthyology 36: 659–664. Bergstedt, R. A., Eshenroder, R. L., Bowen, C., Seelye, J. G., and Locke, J. C. 1990. Mass-marking of otoliths of lake trout sac fry by temperature manipulation. Am. Fish. Soc. Symp. 7: 216–223. Blick, D. J. and Hagen, P. T. 2002. The use of agreement measures and latent class models to assess the reliability of classifying thermally marked otoliths. Fishery Bulletin 100(1): 1–10. Brothers, E. B. 1981. What can otolith microstructure tell us about daily and subdaily events in the early life history of ﬁsh? Rapports et Procés-Verbaux des Réunions, Conseil International pour l’Exploration de la Mer 178: 393–394. Brothers, E. B. 1985. Otolith Marking Techniques for the Early Life History Stages of Lake Trout. Final report to Great Lakes Fishery Commission. October, 1985. 34 pp. Brothers, E. B., 1990. Otolith marking. Am. Fish. Soc. Symp. 7: 183–202. Calaprice, J. R., Lapi, L. A., and Carlson, L. J. 1975. Stock identiﬁcation using X-ray spectrometry and multivariate techniques. Int. North Pac. Fish. Comm. Bull. 32: 81–101. Campana, S. E. and Neilson, J. D. 1982. Daily growth increments in otoliths of starry ﬂounder (Platichthys stellatus) and the inﬂuence of some environmental variables in their production. Can. J. Fish. Aquat. Sci. 39: 937–942. Campana, S. E. and Neilson, J. D. 1985. Microstructure of ﬁsh otoliths. Can. J. Fish. Aquat. Sci. 42: 1014–1032.

462

Eric C. Volk et al.

Campana, S. E. and Gagné, J. A. 1995. Cod stock discrimination using ICPMS elemental assays of otoliths. In D. H. Secor, J. M. Dean, and S. E. Campana (eds.), Recent Developments in Fish Otolith Research. University of South Carolina Press, Columbia, SC, pp. 671–691. Carlstrom, D. 1963. A crystallographic study of vertebrate otoliths. Biol. Bull. 124: 441–463. Degens, E. T., Deuser, W. G., and Haedrich, R. L. 1969. Molecular structure and composition of ﬁsh otoliths. Mar. Biol. 2: 105–113. Dunkelberger, D. G., Dean, J. M., and Watabe, N. 1980. The ultrastructure of the otolithic membrane and otolith in the juvenile mummichog, Fundulus heteroclitus. J. Morph. 163: 367–377. Edmonds, J. S., Caputi, N, Moran, M. J., Fletcher, W. J., and Morita, M. 1995. Population discrimination by variation in concentrations of minor and trace elements in sagittae of two western Australian teleosts. In D. H. Secor, J. M. Dean, and S. E. Campana (eds.), Recent Developments in Fish Otolith Research. University of South Carolina Press, Columbia, SC, pp. 665–669. Hagen, P. 1999. A modeling approach to address the underlying structure and constraints of thermal mark codes and code notations. NPAFC Doc. 395. 12 pp. Alaska Department of Fish and Game. Juneau, Alaska, 99801–5526. Hagen, P., Munk, K., Van Alen, B., and White, B. 1995. Thermal mark technology for in-season ﬁsheries management: A case study. Alaska Fishery Research Bulletin 2: 143–155. Hargreaves, B., Luedke, W., and Till, J. 2001. Application of thermal mass marking in British Columbia, Canada. NPAFC Technical Report 3: 19–22. Jefferts, K. B., Bergman, P. K. and Fiscus, H. F. 1963. A coded wire identiﬁcation system for macroorganisms. Nature (London) 198: 460–462. Jensen, K. A. and Milligan, P. A. 2001. Use of thermal mark technology for the in-season management of transboundary river sockeye ﬁsheries. NPAFC Technical Report 3: 37–38. Joyce, T. L. and Evans, D. G. 2001. Using thermally marked otoliths to aid the management of Prince William Sound pink salmon. NPAFC Technical Report 3: 35–36. Kawana, M., Urawa, S., Hagen, P. T., and Munk, K. 2001. High seas ocean distribution of Alaskan hatchery pink salmon estimated by otolith marks. NPAFC Technical Report 3: 27–30. Mosegaard, H., Steffner, N. G., and Ragnarsson, B. 1987. Manipulation of otolith microstructure as a means of mass-marking salmonid yolk sac fry. Proceedings of the Fifth Congress of European Ichthyologists 1985: 213–220. Munk, K. M. and Geiger, H. J. 1998. Thermal marking of otoliths: the RBr coding structure of thermal marks. NPAFC Doc. 367. 19 pp. Alaska Department of Fish and Game, Juneau, Alaska. Negus, M. T. 1999. Thermal marking of otoliths in lake trout sac fry. North American Journal of Fisheries Management. 19(1): 127–140. Neilson, J. D. and Geen, G. H. 1985. Effects of feeding regimes and diel temperature cycles on otolith increment formation in juvenile chinook salmon Oncorhynchus tshawytscha. Fishery Bulletin 82: 91–101. Quinn, T. P., Dittman, A. H., Peterson, N. P., and Volk, E. C. 1994. Spatial distribution, survival and growth of sibling groups of juvenile coho salmon (Oncorhynchus kisutch) in an experimental stream channel. Can. J. Zool. 72: 2119–2123. Rawson, K., Kraemer, C., and Volk, E. C. 2001. Estimating the abundance and distribution of locally hatchery-produced Chinook salmon throughout a large river system using thermal mass marking of otoliths. NPAFC Technical Report 3: 31–34. Secor, D. H., Dean, J. M., and Campana, S. E. (eds.). 1995. Recent Developments in Fish Otolith Research. University of South Carolina Press, Columbia, SC. 735 pp. Stevenson, D. K. and Campana, S. E. (eds.). 1992. Otolith microstructure examination and analysis. Can. Spec. Publ. Fish. Aquat. Sci. 117: 126. Urawa, S., Hagen, P. T., Meerburg, D., Rogatnykh, A., and Volk, E. 2001. NPAFC Technical Report 3: 13–15.

Otolith Thermal Marking

463

Van Der Walt, B. and Faragher, R. A. 2002. Thermal marking of rainbow trout (Oncorhynchus mykiss) otoliths. New Zealand Journal of Marine and Freshwater Research 36: 8883–8888. Volk, E. C., Schroder, S. L., and Fresh, K. L. 1990. Inducement of unique otolith banding patterns as a practical means to mass-mark juvenile Paciﬁc salmon. American Fisheries Society Symposium 7: 203–215. Volk, E. C., Schroder, S. L., Grimm, J. J., and Ackley, H. S. 1994. Use of a bar code symbology to produce multiple thermally induced marks. Trans. Am. Fish. Soc. 123: 811–816. Volk, E. C., Schroder, S. L., and Grimm, J. J. 1999. Otolith thermal marking. Fisheries Research 43: 205–219.

This Page Intentionally Left Blank

PART

Stock Identiﬁcation Data Analysis

VII

This Page Intentionally Left Blank

CHAPTER

23

Experimental Design and Sampling Strategies for Mixed-Stock Analysis MARY C. FABRIZIO NOAA-Fisheries, Northeast Fisheries Science Center, James J. Howard Marine Sciences Laboratory, Highlands, New Jersey, USA

I. Introduction II. Sampling the Source Stocks A. Source Stock Sampling Strategy B. Source Stock Sample Size III. Sampling the Mixture A. Mixture Sampling Strategy B. Mixture Sample Size IV. Number of Features Necessary for Reliable Classiﬁcation V. Reliability of Features for Stock Delineation A. Allometric and Age Relations B. Sex Effects C. Temporal Stability of Features D. Allelic Designations and Scoring Ambiguities VI. Power to Detect Stock Differences VII. Conclusions References

I. INTRODUCTION Commercial ﬁsheries management relies on formal quantitative assessments that assume ﬁshery statistics derived from catches or landings are representative of a unit stock. If the harvest is obtained from a population comprising a mixture of Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

467

468

Mary C. Fabrizio

stocks, the relative contribution of each stock must be determined so that estimates of exploitation, mortality, and harvest rates can be apportioned correctly in a stock assessment. Additionally, population models used to estimate recruitment, production, and growth are predicated on the treatment of a single stock (Ricker, 1975). Stock recruitment analyses based on observations from overexploited, mixed-stock populations result in erroneous conclusions about “stock” size and productivity (Ricker, 1973; Hilborn, 1985). The purpose of this chapter is to explore sampling issues affecting the precision and accuracy of composition estimates for stock mixtures. In this chapter, a stock refers to a group of conspeciﬁcs that are reproductively isolated in space or time from other conspeciﬁcs; a population consists of individuals from two or more stocks that intermingle during a particular time. For example, during the summer, the striped bass Morone saxatilis population along the New England coast comprises individuals from the Hudson River and Chesapeake Bay stocks. The term group refers to an aggregation of conspeciﬁcs of indeterminate stock structure. Although most of the issues discussed in this chapter are derived from stock studies of ﬁsh, the sampling considerations are germane to studies of any harvested aquatic resource that exhibits stock structure, including marine mammals (Akin, 1988), mollusks (Dillon and Manzi, 1992; King et al., 1994), and crustaceans (Seeb et al., 1990). Resource management questions that focus on the determination of stock composition must meet several requirements before an analysis of stock mixtures can yield useful results. Here, I assume that the mixture comprises two or more putative stocks that have been identiﬁed previously using descriptive approaches (such as tagging) or other exploratory analyses. The ﬁrst step is to characterize stock differences by obtaining baseline information from source stocks. Baseline information includes characters or features that are genotypic, phenotypic, or a combination of both. Typically, ﬁsh are collected from source stocks during the spawning period in spawning areas to ensure that sampled ﬁsh are of known stock origin. An important consideration at this point is the existence of sufﬁcient (detectable) differences in the measured characters among stocks. Once these differences are established, the same set of characters is measured from ﬁsh sampled from the mixed-stock population. The measurement of features and the statistical techniques selected for analysis depend on the species under study and the statistical properties of the features. Guidelines for these decisions are discussed elsewhere in this book. In this chapter, I discuss sampling issues that inﬂuence the performance of classiﬁcation functions, design issues concerning the number and reliability of features selected for stock delineation, and the importance of determining statistical power to detect stock differences. In the material that follows, I refer to two distinct approaches to stock composition analysis—one using classiﬁcation models (such as linear or quadratic discriminant functions, and logistic regres-

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

469

sions), and the other using mixture models. These approaches differ in the manner in which stock compositions are estimated. With classiﬁcation models, individual ﬁsh are assigned to a source stock based on the estimated probability of inclusion, whereas mixture models provide the most likely combination of source stocks that would produce the observed mixture (individuals are not classiﬁed). Classiﬁcation models are well studied statistically and have been useful in mixed-stock analyses relying on phenotypic characters. Mixture models are typically applied to genotypic data, but have been extended (using conditional maximum-likelihood methods) to apply to continuous data, discrete data, or a combination of both (Fournier et al., 1984). The ability to consider multiple data types is a notable advantage. For example, Wood et al. (1989) found that along with genetic markers, age composition and parasite prevalence were necessary to delineate stocks of sockeye salmon (Oncorhynchus nerka). Unlike classiﬁcation-based models, the mixture model approach can be applied reliably even if some of the data are missing from the mixed stock sample (Millar, 1987). Classiﬁcation models and mixture models, along with their necessary assumptions, are thoroughly discussed elsewhere in this volume (see Chapter 24, this volume).

II. SAMPLING THE SOURCE STOCKS

A. SOURCE STOCK SAMPLING STRATEGY The sampling design selected by investigators has tremendous inﬂuence on the ability to discriminate stocks, the reliability of the classiﬁcation algorithm, and the accuracy of estimates resulting from a mixed-stock analysis. Questions such as “Are all the substocks or minor stocks represented in the baseline collection?” “Was the within-stock variation adequately represented by the baseline samples?” and “Which stocks or regional stocks should be considered?” seem elementary, yet few investigations have adequately addressed these sampling issues. Potential biases in mixed-stock analyses have been reported when sampling issues have been examined, either through simulation or directly (Wood et al., 1987; Mulligan et al., 1988; Brodziak et al., 1992; Waldman and Fabrizio, 1994). In this section, I describe sampling and biological issues to consider when designing a sampling plan to characterize source stocks (i.e., the baseline). 1. All Stocks in the Mixture Are Included in the Baseline The typical mixed-stock analysis of genetic variation requires the assumption that all stocks present in the mixture were sampled as part of the baseline data (Pella and Milner, 1987). If this is not the case, biases in allocation of the mixed-stock

470

Mary C. Fabrizio

sample will arise (Smouse et al., 1990; Brodziak et al., 1992). This type of bias increases as the contribution of the unsampled source stock increases (Smouse et al., 1990). If the contribution of an unsampled stock is substantial, then the unconditional likelihood approach developed by Smouse et al. (1990) can be used to detect ﬁsh from an unsampled source stock. Validation studies using the mixture model for genetic stock analysis indicated that inclusion of too many source stocks in the baseline sample produces biases as well (Brodziak et al., 1992). Exclusion of source stocks that constitute only a minor component (<3%) of the mixture may improve the accuracy of stock contribution estimates (Brodziak et al., 1992). Similarly, omission of stocks is warranted when the spawning stock is small and is expected to contribute only negligibly to a mixed stock ﬁshery, as is the case with some pink salmon (Oncorhynchus gorbuscha) stocks in British Columbia (Beacham et al., 1985). Similar biases associated with inclusion or exclusion of stocks in the baseline were found for stock composition estimates based on analysis of phenotypic characters (Waldman and Fabrizio, 1994). The question of which stocks should be included in the baseline is best determined with independent data, such as tag returns (Waldman and Fabrizio, 1994). How a stock is included in the baseline data also affects classiﬁcation outcome. For example, Waldman et al. (1997) found that stock composition estimates were most affected when ﬁsh from the Roanoke River were considered a discrete stock instead of being grouped with geographically proximate stocks. Proper sampling for stock delineation studies of ﬁsh found over large oceanic areas includes sampling ﬁsh from locations throughout the range of a stock. For example, Bermingham et al. (1991) sampled Atlantic salmon (Salmo salar) from only a few contiguous rivers in Maine and southern Canada, yet they used these samples to characterize Atlantic salmon in North America. With this geographically limited sampling, the diversity of mtDNA haplotypes for North American stocks was probably underestimated. 2. Source Stocks Comprising Few Breeders For semelparous ﬁshes, another aspect of the sampling strategy that should be considered is the need to sample ﬁsh in two or more years to adequately estimate allele frequencies for source stocks (Waples, 1990). This is particularly important because only a few year-classes may contribute to spawning (e.g., semelparous Paciﬁc salmon in the genus Oncorhynchus), and allele frequencies may vary among year-classes. Such variation in allele frequencies, which may arise from genetic drift, may be nontrivial, especially when the number of spawners is low (Parkinson, 1984; Jacobson et al., 1986; Waples and Teel, 1990). In these situations, collections in a single year will not adequately represent the extent of allelic variation in the stock, and development of baseline data will

471

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

require multiyear collections to ensure all year-classes contributing to the mixture are represented (Waples, 1990; Grewe et al., 1994). 3. Number of Source Stocks When stocks contribute equally to the mixture and the genetic distance between stocks is equal, the performance of the mixture model is not signiﬁcantly affected by the number of source stocks considered, as long as the number of stocks ranges between three and nine (Wood et al., 1987). However, this special case (equal contribution, equal genetic distances) is unlikely to be encountered in ﬁsheries applications (Wood, 1989). In general, as the differences among stocks increases, a greater number of source stocks can be considered in a mixed-stock analysis (Xu et al., 1994). However, empirical studies show that the performance of classiﬁcation or mixture models deteriorates as the number of source stocks considered increases. Wood (1989) postulated that the decreased performance is due to an increased likelihood of observing (genetically) similar stocks as additional source stocks are considered. In a genetic mixed-stock analysis with mixture samples of known origin, Brodziak et al. (1992) reported greater accuracy of stock contribution estimates with fewer source groups (Fig. 23-1). This general ﬁnding

0.05 Brodziak et al. 1992 Waldman and Fabrizio 1994

Average bias

0.04

0.03

0.02

0.01

0 1

2

3

4

5

6

7

8

Number of stock groups Figure 23-1. Effect of the number of stock groups on the average bias of stock composition estimates from a mixed-stock analysis (Brodziak et al., 1992) and a linear discriminant function analysis (Waldman and Fabrizio, 1994). Average bias was calculated as the mean of the absolute values of the bias (data from Table 6, Brodziak et al., 1992; and Table 4, Waldman and Fabrizio, 1994). In all cases, the total number of samples classiﬁed remained constant (N = 2,550, Brodziak et al., 1992; N = 348, Waldman and Fabrizio 1994), but the number of stock groups varied.

472

Mary C. Fabrizio

of decreasing classiﬁcation accuracy with increasing number of source stocks was also reported for linear discriminant models using morphometric characters (Fig. 23-1; Waldman and Fabrizio, 1994) and scale pattern descriptors (Myers et al., 1987). 4. Identiﬁcation of Composite Stocks For some stocks, appropriate characterization of the baseline data may involve grouping of samples from various spawning streams into “regional” or composite stocks (Pella and Milner, 1987; Wood, 1989; Utter and Ryman, 1993; Waldman and Fabrizio, 1994). This practice should not be considered merely a statistical convenience, because hierarchical groupings are suggested by studies of genetic variation at the species level (e.g., Parkinson, 1984; Beacham et al., 1995). In other cases, ﬁsh from two or more stocks may share identical haplotypes or may not exhibit signiﬁcant differences in haplotype frequencies, suggesting pooling of these groups (Waldman et al., 1996). When the number of putative stocks is large (more than about 20), stock contribution estimates are less reliable, and this approach becomes necessary (Wood et al., 1989). Using similarities in life history characteristics of Altantic salmon from Quebec, Labrador, and Newfoundland, Power (1981) postulated the existence of ﬁve major stock groups, although he estimated the existence of more than 500 stocks in this area. In this case, use of composite stocks appears to be reasonable because sufﬁcient numbers of ﬁsh from all spawning sites may not be obtainable. This is because spawning sites in rivers with reduced visibility would be difﬁcult to locate or because remote sites may be inaccessible (Grant et al., 1980). Similarly, a study of the mixed-stock ﬁshery in British Columbia for chum salmon (Oncorhynchus keta) could entail sampling 880 rivers—clearly, not a feasible enterprise (Beacham et al., 1987). A genetic survey of chum salmon from 83 of those rivers revealed ﬁve regional groups comprising stocks that were more similar within a region than between regions (Beacham et al., 1987). Campana and Casselman (1993) found markedly better discrimination of cod (Gadus morhua) stocks when the number of groups to be resolved by linear discriminant function analysis was reduced from 19 distinct spawning stock groups to three large, but geographically distant, composite groups. However, pooling of substocks into composite stocks should be restricted to substocks that exhibit a high degree of similarity (Wood, 1989; Smouse et al., 1990) because unreliable stock composition estimates result when pooling of substocks is based on geographic or political boundaries (Mulligan et al., 1988). Advantages of pooling minor stocks are larger sample sizes, reduced dimensionality of the classiﬁcation problem, and better resolution of stock contribution estimates to the mixed-stock sample (Smouse et al., 1990). In addition, Millar (1987) suggested that pooling of similar stocks may reduce the variance and bias of stock composition estimates from mixture models.

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

473

Composite stocks may be identiﬁed using similarity dendrograms which may also be used as a preliminary indication of the reliability of mixture estimates for stocks forming the composite (Wood, 1989). Based on stock groupings suggested by similarity dendrograms, Wood et al. (1989) found that both accuracy and precision of stock composition estimates improved when such composite stocks were considered. They postulated that the collective contribution of the composite stock was estimated more accurately because most of the bias in estimating stock composition of a mixture occurs among groups of similar stocks (Wood et al., 1989). The decision to include (or exclude) substocks in a composite stock may be problematic when phenotypic characters are used to discriminate stocks unless such decisions are corroborated by evidence of similarity (or dissimilarity) among substocks (e.g., from tagging studies). For example, classiﬁcation accuracy decreased when Roanoke River striped bass were considered as a distinct source stock for the Atlantic coast mixed-stock population (Fabrizio, 1987; Margraf and Riley, 1993; Waldman and Fabrizio, 1994). To improve classiﬁcation accuracy, Roanoke River ﬁsh were grouped with ﬁsh from the Chesapeake Bay to form a composite stock that was well distinguished from the Hudson River stock (Fabrizio, 1987; Waldman and Fabrizio, 1994). In other studies, the Roanoke River stock was disregarded as a potential source stock because evidence from previous tagging studies suggested that its contribution to the mixed-stock ﬁshery was insigniﬁcant (VanWinkle et al., 1988; Margraf and Riley, 1993). The formation of a composite stock from substocks that individually contribute little to a mixed population may be useful (Xu et al., 1994), but care should be exercised to ensure that the contribution to the mixture remains insigniﬁcant (see section III). When selecting sampling sites for characterizing composite or regional stocks, it is important to include sites to represent each of the components or substocks in the baseline samples (Waldman and Fabrizio, 1994). Although exclusion of some substocks from the composite (baseline) stock may only minimally affect correct classiﬁcation rates for baseline samples, large variations in stock composition estimates (e.g., 29% vs. 45%) may be obtained when classiﬁcation rules are applied to a mixture sample (Waldman and Fabrizio, 1994). As more substocks are added to the composite stock, the contribution of the composite stock to the mixed-stock sample may change.

B. SOURCE STOCK SAMPLE SIZE Although it is not possible to suggest a guideline for necessary sample size, several factors affecting this determination can be explored. To a large extent, more ﬁsh from each of the source stocks should be sampled when differences among stocks are low. Some guidelines for genetic stock delineations have been published, but

474

Mary C. Fabrizio

may be inadequate if applied to different species or different suites of characters. Sample size guidelines for linear discriminant analyses are available (Fig. 23-2; Lachenbruch, 1968), and it has been generally observed that a greater number of samples from source stocks decreases the bias of the correct classiﬁcation rate (Waldman and Fabrizio, 1994). Adequate source stock sample size to permit testing of model assumptions (such as conformance to Hardy–Weinberg expectations for studies based on genotypes) should be ensured. Genetic data often are used to estimate frequencies of particular alleles in the population, and the observed genotypic frequencies are tested against the expected frequency of occurrence based on assumptions of random mating and random assortment of alleles. Typically, conformance to Hardy–Weinberg expectations is tested with goodness-of-ﬁt statistics (either Chi-square or likelihood ratio statistics). Fairbairn and Roff (1980) studied the power of the Chi-square test to detect deviations from Hardy–Weinberg equilibrium and found that these tests have low power with typical sample sizes (<200 ﬁsh) used for stock delineations. Another problem with Chi-square and similar goodness-of-ﬁt tests occurs when the test statistic is estimated using large-sample approximations but applied to small samples. When the number of alleles is large, many of the possible genotypes are not observed in samples or the observed frequencies are small (Chakraborty, 1993). In these situations, investigators often group the data using the rule of thumb that no more than 20% of the expected cell frequencies are less than 5, and no expected cell frequencies are less than 1 (Cochran, 1954). Pooling of frequency data is not advised unless there is little loss of information or natural groupings can be readily discerned (Agresti, 1990). Alternatively, a Monte Carlo approach may be used to provide an approximation of the probability of the estimated Chi-square statistic when sample sizes are small (e.g., Roff and Bentzen, 1989). However, neither approach corrects the problem of inadequate sampling. Large-sample approximations of goodness-of-ﬁt statistics are invalid and instead, conditional tests should be used to estimate exact sampling distributions (Chakraborty, 1993). Specialized software that provides goodnessof-ﬁt testing should be evaluated for appropriateness of the test statistic prior to application. Failing appropriate goodness-of-ﬁt tests, investigators should seek to increase sample size to ensure more accurate estimation of genotypic frequencies for less common genotypes. In the past, sample sizes of 50 to 100 ﬁsh from each source stock were used for mixed-stock analyses (e.g., Milner et al., 1985), but this sample size may be too small to provide an adequate representation of genetic variability within stocks (Waples and Aebersold, 1990). When fewer than 50 ﬁsh are sampled from each source stock, stock composition estimates from a mixed-stock analysis have increased bias and decreased precision (Wood et al., 1987). Even when source stock samples exceed 50 ﬁsh, the mixture model performs poorly if stock

Gamma = 0.05

80 D2=1

70

P =0.358

Number of samples

D2=2

60

D2=3

P =0.207

50 40 P =0.115

30 20 10 0 0

5

10

15

20

25

Number of features Gamma = 0.01

450

P =0.319

400 Number of samples

350 300 P =0.169

250 200 150

P =0.077

100 50 0 0

5

10

15

20

25

Number of features Figure 23-2. Necessary sample size as a function of the number of features considered in a linear discriminant function analysis applied to the two-group problem (data from Table 2, Lachenbruch, 1968). Sample size is the number of samples required from each group (N1 = N2). The top panel depicts three functions corresponding to an average expected error rate (P bar) of 0.358, 0.207, and 0.115, when the Mahalanobis distance (D2) is 1, 2, or 3. These expected error rates will be within 0.05 of the optimum rate (gamma = 0.05). The bottom panel represents outcomes with a smaller tolerance (i.e., gamma = 0.01) and depicts the sample size necessary to obtain an average expected error rate of 0.319, 0.169, and 0.077. 475

476

Mary C. Fabrizio

separation is low and few loci are considered (Wood et al., 1987). Recognizing this, contemporary genetic stock identiﬁcation programs may include multiyear sampling and have extremely large sample sizes in the baseline (e.g., 21,000 coho salmon, Oncorhynchus kisutch, from source stocks in British Columbia; Beacham et al., 2001). Failure to detect an allele in a sample from the baseline stocks is a common problem in mixed-stock analyses (Waples and Aebersold, 1990). Consider the case of a single locus with two alleles, A and a; three genotypes are possible: AA, Aa, and aa. In a study of stock structure, many such loci may be considered, and the number of possible genotypes is 3L, where L is the number of loci. Screening for 5 loci could yield a possibility of 243 different multilocus genotypes, and 10 loci could produce more than 59,000 possibilities (Waples and Aebersold, 1990). The question of adequate representation of the source stock samples is serious. In addition, Wood et al. (1987) noted that sampling variability is an important consideration when estimating genotypic frequencies for loci with three or more alleles. For mixed-stock analyses, large sample sizes from source stocks increase the likelihood of detecting rare haplotypes that may be stock speciﬁc (Epifanio et al., 1995). In situations where genetic divergence between stocks is high, fewer ﬁsh can be sampled as the frequency of common haplotypes stabilizes with sample sizes of 50 to 70 ﬁsh (Epifanio et al., 1995). Insufﬁcient sampling of source stocks may lead to inaccurate characterization of the stocks such that it may be impossible to allocate a portion of the ﬁsh from the mixed-stock sample to the source stocks under consideration. Theoretically, ﬁsh from the stock mixture that carry an allele not detected in the source stocks have a zero probability of originating from the sampled source stocks (Waples and Aebersold, 1990). In some cases, these ﬁsh may compose a sizable portion of the mixture. For example, a pilot study of mtDNA haplotypes revealed that 19% of the mixed-stock sample of American shad (Alosa sapidissima) could not be attributed to putative stocks from any of the 10 rivers sampled in the midAtlantic region (Brown and Chapman, 1991 in Epifanio et al., 1995). Either the source stocks were inadequately sampled, or the unallocated ﬁsh may have originated in rivers not sampled by the investigators (see section II, A). In either case, adequate sampling of source stocks should not be overlooked.

III. SAMPLING THE MIXTURE

A. MIXTURE SAMPLING STRATEGY Mixtures of stocks are typically obtained from a ﬁshery that intercepts or harvests individuals originating from any number of source stocks. Although much atten-

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

477

tion has been focused on appropriate sampling designs for characterizing source stocks, sampling the mixed-stock harvest requires understanding the potential geographic and temporal variability of the mixture. Highly migratory ﬁshes such as striped bass, American shad, and Paciﬁc salmon species may exhibit stockspeciﬁc migration patterns and ﬂuctuations in year-class strength; these and possibly other factors virtually ensure that stock mixtures for these species are temporally dynamic, even within a season. For example, the contribution of source stocks to the mixed-stock ﬁshery for Atlantic coast striped bass varies along the coast and by season (Fabrizio, 1987). Similar geographic and seasonal variability is thought to characterize the Atlantic salmon harvest off west Greenland (Bermingham et al., 1991). Because annual variation in stock composition is likely (e.g., Fabrizio, 1987; Myers et al., 1987; Brown et al., 1999), mixture samples collected over several seasons or years should not be pooled (Mulligan et al., 1988). Likewise, samples from mixtures occurring in a variety of locations should be treated separately. Recently, gear selectivity was hypothesized to affect stock composition estimates from ﬁsheries using several gear types (Beacham et al., 2000), indicating that mixture samples from dissimilar ﬁsheries should not be pooled. Investigators wishing to make assumptions about homogeneity of stock composition estimates among mixture samples should provide evidence for the temporal stability of these estimates.

B. MIXTURE SAMPLE SIZE Simulation studies using mixed-stock analysis have advanced our understanding of the effect of mixture sample size on the performance of the mixture model. As sample size of the mixture decreases, stock composition estimates have greater bias and lower precision (Fig. 23-3; Wood et al., 1987; Mulligan et al., 1988). Beacham et al. (1995) reported a similar loss in precision, with only a minor effect on accuracy (Fig. 23-4), but this was likely due to the restricted sample sizes considered (200 vs. 50 ﬁsh sampled from the mixture). For a given mixture sample size, higher accuracy is obtained when each stock contributes equally to the mixture; bias increases as proportions deviate from equality (Fig. 23-4; Beacham et al., 1995). It has long been known that mixture models perform best when each source stock contributes equally to the mixture (Mulligan et al., 1988). With the mixture model, contributions of dominant stocks are underestimated and those from minor stocks are overestimated as mixture sample size decreases (Wood et al., 1987). Because the number of ﬁsh from any one stock in the mixture may be small, sampling error should be considered more troublesome for the mixture sample than for the source stocks (Wood et al., 1989).

0.035

0.16

Average bias

0.14 0.12

0.025

0.1 0.02 0.08 0.015 0.06 0.01

0.04

0.005

Standard deviation

Standard deviation Average bias

0.03

0.02 0

0 30

90

180

300

Number of fish in mixture Figure 23-3. Effect of mixture sample size on average bias and precision of stock composition estimates for a three-stock problem (data from Table 4, Mulligan et al., 1988). In this example, all three stocks contribute equally to the mixture. The standard deviation of the distribution of stock composition estimates is an indicator of the precision of the estimates. As the number of ﬁsh in the mixture increases, stock composition estimates have lower bias and are less variable (more precise).

0.16 50 fish in mixture 0.14

200 fish in mixture

Average bias

0.12 0.1 0.08 0.06 0.04 0.02 0 100% A

50% A - 50% B

100% B

Mixture composition

Figure 23-4. Effect of mixture sample size (N = 50 and N = 200) on accuracy of mixed stock composition analysis for a two-stock problem (data from Table 8, Beacham et al., 1995). Five mixtures of known composition were analyzed ranging from 100% stock A and 0% stock B to 0% stock A and 100% stock B. In this application, the average bias of the stock composition estimates did not change appreciably as mixture sample size decreased from 200 to 50 ﬁsh. Highest accuracy was obtained when stocks A and B contributed equally to the mixture.

478

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

479

As with most applications of asymptotic models to ﬁnite samples, greater precision is achieved with larger sample sizes (Beacham et al., 1985; Beacham et al., 1987; Beacham et al., 1995). But how large is large enough? Brodziak et al. (1992) achieved allocation errors below 3% with a sample of 220 Chinook salmon (Oncorhynchus tshawytscha) from a mixed-stock ﬁshery. On the other hand, a sample of 328 Atlantic salmon was considered too small to characterize the mixed-stock ﬁshery off West Greenland (Bermingham et al., 1991). For some stocks of chum salmon, Beacham et al. (1987) found that precise stock composition estimates were realized only when mixture sample sizes exceeded 600 ﬁsh. A speciﬁc rule regarding sufﬁciency of mixture sample size is not possible. Instead, sample sizes for mixed-stock analysis should be determined for each species studied using a simulation approach that permits understanding of the effect of stock contribution on the accuracy of allocation and precision of the stock composition estimates (e.g., Winans et al., 2001). Precision of stock composition estimates is also affected by the magnitude of stock differentiation. In general, the reliability of stock composition estimates will be determined in large part by the actual stocks comprising the mixture and the degree of similarity among stocks (Wood et al., 1989). In an extreme situation— a mixture comprising ﬁsh from a single source stock which is genetically distinct—Wood et al. (1989) showed that mixture samples consisting of as few as 11 ﬁsh may be sufﬁcient to estimate stock composition. Conversely, for a mixture comprising one stock that shares characteristics with many other stocks, unreliable stock composition estimates were obtained for sample sizes up to 300 ﬁsh (Wood et al., 1989). Wood et al. (1989) speculated that even if 1,000 ﬁsh were sampled from the mixture, stock contribution estimates would remain unreliable. Other simulations of mixtures have shown that a sizable portion of the mixture may be allocated to stocks that did not contribute to the mixture (e.g., Pella and Milner, 1987), indicating that similarities among stocks lead to misclassiﬁcation. The performance of the mixture model was found to be limited by the degree of stock separation, as the accuracy and precision of stock composition estimates reﬂected the allocation uncertainty associated with the least identiﬁable stock in the mixture (Wood et al., 1989).

IV. NUMBER OF FEATURES NECESSARY FOR RELIABLE CLASSIFICATION In most cases, the problem concerning the number of features necessary for a mixed-stock analysis is a statistical one. In general, as additional features are considered in mixed-stock analysis, the accuracy and precision of the stock composition estimates increase (e.g., Wood et al., 1989). This is also true when discriminant functions are used to identify stock of origin. However, Hand (1981)

480

Mary C. Fabrizio

cautioned against the “curse of dimensionality”: as more features are considered, the performance of the discriminant rule deteriorates and the misclassiﬁcation rate of unknowns increases. A good example of this was documented by Campana and Casselman (1993) with Fourier shape descriptors of cod otoliths. The correct classiﬁcation rate of the discriminant rule increased as more variables were included. But when test samples of known origin were classiﬁed, classiﬁcation accuracy was found to be unacceptably low. Although the classiﬁcation algorithm appeared well “tuned” to the training samples, test samples exhibited unacceptably high misclassiﬁcation rates. Another pitfall to avoid when applying discriminant function analysis is the increased likelihood of obtaining spurious discrimination of stocks as the number of features increases (Misra and Easton, 1999). Most investigations are characterized by a large number of potential features for analysis, and the utility of any one of the features for stock identiﬁcation may be unknown or untested. Thus, in a typical application, the number of features needs to be reduced to a subset that is internally consistent and uncorrelated. There is, however, a minimum number of features necessary when using the maximum-likelihood mixture model to examine stock differences. The number of observable states of the markers (e.g., genotypes) is the maximum number of stocks that can be resolved by the mixture model (Pella and Milner, 1987; Mulligan et al., 1988). For example, with two loci having two alleles each, there are nine possible genotypes and therefore the mixture problem cannot be parameterized for more than nine stocks. With genetic characters, classiﬁcation accuracy should (Smouse et al., 1982; Bernatchez and Duchesne, 2000) but may not always (Brodziak et al., 1992) increase with the number of loci examined. Classiﬁcation accuracy is likely to improve if the additional loci are highly variable for one or a few stocks and monomorphic for the remainder of stocks (see Brodziak et al., 1992). Care must be taken to ensure that oversampling of loci does not occur. In this situation, rare genotypes may be revealed in ﬁsh sampled from the mixture, and these ﬁsh will not be accurately allocated to the source stocks. The likelihood of observing rare genotypes increases with the number of alleles examined (Brodziak et al., 1992). In the two-stock problem, simulation studies show that for a given number of loci, improvements to the correct classiﬁcation of unknowns are insigniﬁcant when the number of alleles per locus exceeds six (Bernatchez and Duchesne, 2000). It has been argued that mtDNA investigations are potentially more useful for stock identiﬁcation studies than allozyme analyses because there are more mtDNA haplotypes than there are alleles from allozyme-encoding loci (Epifanio et al., 1995). However, mtDNA haplotype distributions do not follow Hardy–Weinberg (H–W) expectations because H–W expectations are applicable to diploid characters only (Epifanio et al., 1995). Thus, expected frequencies of

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

481

haplotypes cannot be calculated. In addition, the physical linkage of sites on the mtDNA molecule prevents independent segregation of sites, resulting in a lack of independence among polymorphic sites (Xu et al., 1994; Epifanio et al., 1995). Haplotype information derived from some restriction enzymes may be redundant (due to lack of independence) and therefore less useful in resolving stock differences (Epifanio et al., 1995). Only recently has this characteristic been identiﬁed as a possible hindrance to resolution of stocks (Ward et al., 1997). Moreover, mtDNA haplotype frequencies may be similar among groups within a species. For example, yellowﬁn tuna (Thunnus albacares) in the Atlantic, Paciﬁc, and Indian oceans exhibited genotypic variation at a single allozyme locus but only limited variation among mtDNA haplotypes (Ward et al., 1997). Stock identiﬁcation studies based on mtDNA variation often report a large number of mtDNA haplotypes. If the genetic divergence between source stocks is low, Xu et al. (1994) recommended that the number of mtDNA haplotypes exceed the number of source stocks by 1.5 to 2.5 times. Estimation of a large number of haplotypes is considered desirable because theoretically, as the number of characters is increased, putative stocks may be identiﬁed as collections of unique haplotypes (Epifanio et al., 1995). For example, Danzmann et al. (1991) detected a high degree of differentiation of mtDNA haplotypes between hatchery and wild brook trout (Salvelinus fontinalis) when a large number of restriction enzymes (11) were used. However, it is also recognized that sampling variation becomes a serious problem as the number of characters exceeds the number of individuals sampled (Epifanio et al., 1995). In many instances, as more haplotypes are identiﬁed, the ability to resolve stock differences decreases because too many rare haplotypes are revealed (Xu et al., 1994; Epifanio et al., 1995). For American shad on the east coast of the United States, 116 mtDNA haplotypes were found using 14 restriction enzymes, but only 8 haplotypes occurred in frequencies exceeding 1%, and the remaining 108 haplotypes were considered “rare” (Epifanio et al., 1995). It is generally believed that rare haplotypes are of limited use in stock delineation studies because they usually do not enhance stock resolution (Xu et al., 1994; Ward et al., 1997). The same problem of reduced stock resolution may be encountered in studies that detect too many rare genotypes.

V. RELIABILITY OF FEATURES FOR STOCK DELINEATION The selection of features for a mixed-stock analysis entails more than the determination of “what to measure” (e.g., otolith shape), and the selected methodology affects reliability. For example, studies relying on scale characteristics should ensure that scale sampling techniques are consistent across ﬁsh, particularly with respect to area of the body sampled (Myers et al., 1987). In an analysis of cod

482

Mary C. Fabrizio

otolith shape, Campana and Casselman (1993) concluded that either the otolith nucleus or centroid could be used as the reference point for Fourier analysis because comparable results were obtained. But Castonguay et al. (1991) noted that if the otolith centroid is selected as the reference, phase angles are unreliable and should be ignored. Although both studies used otolith shape descriptors, methodologies differed and indicate the need to fully investigate the effect of protocols on the reliability of features. The reliability of features can also be affected by the instruments selected to measure the features if measurement accuracy varies among instruments or among laboratories using the same instrument. Campana et al. (1997) noted that the accuracy and sensitivity of various instruments to the elemental composition of otoliths differed signiﬁcantly and could explain some of the variation in measures of elemental composition collected at different laboratories. In a recent review of the use of otolith elemental composition for stock identiﬁcation, Thresher (1999) called for studies to elucidate handling effects, protocols to ensure data quality (e.g., to eliminate or minimize contamination of samples), and validation studies to provide independent conﬁrmation of stock identity. Features used to delineate stocks should be (statistically) independent. Signiﬁcant correlations between features indicate possible dependence. When features are correlated, either some features are omitted from consideration or a transformation is pursued to remove or at least minimize correlations among features. Principal components (PCs) provide appropriate transformations by removing correlations among features and retaining useful variation. This approach was used by Waldman et al. (1997) to obtain uncorrelated morphometric measures for striped bass. An added advantage of PC analysis is that it can also reduce the dimensionality of the problem because in many cases, fewer PCs are needed to describe most of the variation in the original features. Principal component factor scores, which are by deﬁnition uncorrelated, can be used in maximum-likelihood mixed-stock analyses. Wood et al. (1989) applied this transformation to correlated scale pattern data from sockeye salmon to resolve stocks using the mixture model. Ideally, selected features will exhibit sufﬁcient differences among stocks to permit classiﬁcation and allocation. Some features may be more reliable for one set of stocks, while other features have greater resolving power when additional, but different stocks are considered (Wood et al., 1989). If a stock is difﬁcult to distinguish from others and the “problem” stock comprises more than 80% or less than 20% of the mixture, then stock composition estimates for the mixture are unreliable (Beacham et al., 1985). One approach to handling “problem” stocks is to tag individuals from one of the problem stocks and apply the extended mixture model developed by Brodziak (1993). Simulation studies indicated that this strategy improves the accuracy and precision of stock composition estimates (Brodziak, 1993). Resolution of stocks may be enhanced by continued improve-

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

483

ments in molecular techniques, namely, the development of a suite of markers using minisatellite or microsatellite probes of nuclear DNA. Minisatellite regions of nuclear DNA are hypervariable, noncoding segments of DNA consisting of 15 to 60 base pairs (Bruford et al., 1992); these segments are also known as VNTRs or variable number tandem repeats (Bentzen et al., 1993). Microsatellites are similar except that these repeat sequences typically consist of 2 to 4 base pairs. Loci detected by microsatellite probes may be more variable than mtDNA loci and prove useful in differentiating “problem” stocks (e.g., Ferguson and Danzmann, 1998; O’Connell and Wright, 1997). In the 1990s, ﬁshery scientists began using minisatellite and microsatellite DNA probes to study intraspeciﬁc variation in ﬁshes (e.g., Taggart and Ferguson, 1990; Turner et al., 1991; Bentzen and Wright, 1993; Bentzen et al., 1993; Dahle, 1994; Leung et al., 1994; Fontaine et al., 1997). Some minisatellite probes can be used to determine DNA ﬁngerprints that are speciﬁc to individual ﬁsh, but this level of resolution is not useful for delineation of stocks (Dahle, 1994). Minisatellite loci have provided enhanced stock delineation in a few cases (e.g., Beacham et al., 1995); however, technical complexities and high costs of the technique appear to reduce the utility of minisatellite probes for stock identiﬁcation problems (O’Connell and Wright, 1997; Beacham et al., 2001). Most contemporary studies rely on analysis of microsatellite alleles to describe stock differentiation, or more recently, to resolve stock mixtures (e.g., Beacham et al., 2001). Use of microsatellite DNA probes has become more common, particularly for stock identiﬁcation problems that have been previously intractable. However, for closely related stocks, simulations and empirical studies indicate that even the new markers will not reliably discriminate among stocks (Ferguson and Danzmann, 1998; Scribner et al., 1998).

A. ALLOMETRIC

AND

AGE RELATIONS

Some features, such as morphometric measures, meristic counts, otolith shape descriptors, and otolith microchemistry, are inﬂuenced by ﬁsh size and age. Stock delineation algorithms based on size-related features will perform poorly when the effect of size is large and ﬁsh of unknown origin are different in size from those in the training samples. Several approaches have been used to remove the effects of size, including expression of characters as a ratio of ﬁsh length (Casselman et al., 1981; King, 1985); adjustments by geometric mean regressions (Meng and Stocker, 1984); covariate adjustments based on ﬁsh length (Waldman and Fabrizio, 1994), otolith length (Campana and Casselman, 1993), or otolith weight (Campana and Gagné, 1995); and use of PC scores (Prager and Fabrizio, 1990). Reist (1986) developed speciﬁc guidelines for removing size effects using regression residuals and allometric adjustments, but the success of these methods

484

Mary C. Fabrizio

in completely removing size effects is poor (Cadrin, 2000). For example, in one study, the efﬁcacy of using covariate adjustments to remove size effects was tested and found to be low: despite size adjustments of otolith shape variables for cod, persistent size effects remained (Campana and Casselman, 1993). This was evidenced by the fact that misclassiﬁed cod were assigned to groups with similar growth rates (Campana and Casselman, 1993). In a recent review, Cadrin (2000) suggested two methods for effectively removing size effects from phenotypic characters: Burnaby’s (1966) growth-invariant discriminant analysis (and modiﬁcations described by Klingenberg, 1996), and a technique known as multiple-group PC analysis, or MGPCA. Unlike PC analysis, which can be applied only to a single group, MGPCA allows the user to extract within-group size effects when two or more groups are considered (Thorpe, 1988). MGPCA followed by discriminant analysis has not yet been applied to the mixed-stock problem. Burnaby’s technique has long been recognized as an effective method for removing size effects from morphometric data (Reyment et al., 1984; Rohlf and Bookstein, 1987; Klingenberg, 1996) and was applied to a mixed-stock analysis by Fabrizio (1987). Another approach involves transformation of the allocation rule based on a link function between individuals from known stocks and individuals whose stock of origin is unknown (Biernacki et al., 2002). This technique appears promising, but has not yet been applied to ﬁsh stocks. Whereas size effects may be removed from the features used to delineate stocks or from the allocation rule, age effects remain largely unexplored. Observed age effects are sometimes considered spurious (Campana and Gagné, 1995). Misra and Carscadden (1984) noted differences among age classes for several meristic counts in capelin (Mallotus villosus) from the northwest Atlantic Ocean. Interestingly, when capelin age classes were evaluated individually, counts that varied between ages were some of the best features for discriminating stocks within an age class (Misra and Carscadden, 1984). This observation supports the premise that these meristic features are environmentally inﬂuenced, as ﬁsh from putative stocks spawn and develop in areas differing in depth and water temperature. Annual ﬂuctuations in temperature and perhaps other environmental factors contribute to the observed differences in traits among age classes within a stock (Misra and Carscadden, 1984). Although herring (Clupea harengus) otolith shape did not vary among age classes of adult ﬁsh, adult otolith shapes differed from those of juvenile herring (Bird et al., 1986). Both age and year-class effects have been noted in otolith shape descriptors of Atlantic mackerel (Scomber scombrus) (Castonguay et al., 1991) and haddock (Melanogrammus aegleﬁnus) (Begg and Brown, 2000), making these features ineffectual for mixed-stock analyses of these species unless shape characteristics and allocation rules are recalculated each year for each age group present in the mixed-stock sample. One way to explore the effects of age is to construct a classiﬁcation rule using age classes instead of stocks. When features are related to age, age-class assign-

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

485

ments can be made with fairly high correct classiﬁcation rates. For example, using otolith shape variates, over 80% of Atlantic cod were classiﬁed to their correct age class (Campana and Casselman, 1993).

B. SEX EFFECTS Few reports exist on the heterogeneity of features among sexes, most likely because of the lack of a relation between expression of most characters and sex. For example, male and female herring (Bird et al., 1986) did not exhibit differences in otolith shape, nor did Atlantic mackerel (Castonguay et al., 1991). The elemental composition of cod otoliths adjusted for otolith weight differences was not affected by sex (Campana and Gagné, 1995). Sex effects were, however, reported for a few otolith shape variables for Atlantic cod (Campana and Casselman, 1993) and for morphometric characters in chum salmon (Fournier et al., 1984). When the sex ratio of the mixture sample is close to unity, sex effects may be inconsequential. Alternatively, data from males and females may be treated and analyzed separately (Fournier et al., 1984), or a single sex may be examined (e.g., female king mackerel, Scomberomorus cavalla, DeVries et al., 2002). Separate treatment of males and females is also warranted for morphometric features obtained from sexually dimorphic species such as American lobster, Homarus americanus, pink salmon, and chum salmon (Cadrin, 2000).

C. TEMPORAL STABILITY

OF

FEATURES

The expression of phenotypic and genetic features may be inﬂuenced by environmental factors, genetic drift, or selection and may therefore be subject to large temporal variations. In some cases, physiology may mediate environmental effects on the expression of a feature; for example, hormones mediate calcium and trace element accretion rates in otoliths (Edmonds et al., 1995). A trait is less useful for stock delineation when temporal variation within the stock exceeds variation among stocks. For example, temporal variation in ﬁve meristic characters in Atlantic salmon was signiﬁcant for all traits examined over a 10-yr period, and was large relative to the known geographic variation among populations (Blouw et al., 1988). Use of temporally varying traits in mixed-stock analyses introduces an error that is most severe when the variation through time equals or exceeds the variation among stocks. In another example, trace element composition of otoliths varied among stocks within a given year, but also varied between years (Edmonds et al., 1995), making stock composition analysis from multiyear samples unreliable. Some scale attributes exhibit signiﬁcant temporal variation that is thought to be associated with conditions affecting growth. For example,

486

Mary C. Fabrizio

circuli counts and increment widths of sockeye salmon scales reﬂect growth patterns due to changes in climate or population density (Wood et al., 1989), and fewer scale circuli in European stocks of Atlantic salmon are associated with decreasing growth through time (Reddin, 1986). In the presence of such variation, stock composition estimates can be severely biased because scale characters from ﬁsh in the reference stocks will not be representative of those from ﬁsh in the mixture collected in different years (Wood et al., 1989). Examination of temporal variability of features is now recognized as a necessary component of stock identiﬁcation studies that rely on baseline samples collected in two or more years (e.g., Rooker et al., 2001; Swain et al., 2001; Gillanders, 2002). Temporal stability is also critical to establish for features derived from historical samples such as microsatellite allele frequencies from DNA recovered from archived scales (Nielsen et al., 2001) or otoliths (Ruzzante et al., 2001). Use of temporally variable features is not recommended for long-term monitoring of stocks because use of these features requires (potentially costly) renewal of baseline information each year. In such a case, the characterization of source stocks is a “moving target.” Most genetic markers, such as allele frequencies, are temporally stable (Utter et al., 1980), but some markers may exhibit temporal instabilities (Waples, 1990; Waples and Teel, 1990; Grewe et al., 1994). When changes through time are reported for allele frequencies, they have been ascribed to genetic drift resulting from small population sizes, possibly unpure samples (Phelps et al., 1994), purposeful or inadvertent selection of hatchery ﬁsh (Grewe et al., 1994), or existence of substocks within a putative stock (Kondzela et al., 1994). Based on a simulation study of chinook salmon stocks, short-term (1 to 5yr) changes in allele frequencies did not differ in magnitude from those expected over longer time periods (10–25 yr) (Waples and Teel, 1990). Furthermore, signiﬁcant temporal changes in allele frequencies were not due to small sample sizes; instead, these changes were more pronounced with larger samples (Waples and Teel, 1990). This simulation was based on observed temporal changes in allelic frequencies for hatchery stocks of Chinook salmon, and extensions of these results are limited to species with similar life history patterns (namely, overlapping generations with complete turnover of the spawning population each year). Nevertheless, this study emphasizes the importance of testing for temporal stability in features considered for a mixed-stock analysis. Like allele frequencies, mtDNA haplotypes also appear to exhibit temporal stability. For example, no differences in the frequencies of mtDNA length genotypes were found among striped bass year-classes or between samples from different years, implying that these genotypes were stable over several generations and useful for mixed-stock analysis (Wirgin et al., 1993). Haplotype frequencies of mtDNA length fragments were stable for Atlantic sturgeon (Acipenser oxyrinchus) during a 5-yr period of study (Waldman et al., 1996). Temporal variation in allele

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

487

frequencies and mtDNA haplotype frequencies for red drum (Sciaenops ocellatus) in the Gulf of Mexico also were not signiﬁcant (Gold et al., 1993). However, Grewe et al. (1994) argued that for a given population, mtDNA haplotype frequencies are more likely to exhibit genetic drift than allelic frequencies, thus extending the need to examine temporal variation in mtDNA characters, particularly when the effective population size is small.

D. ALLELIC DESIGNATIONS

AND

SCORING AMBIGUITIES

Isoloci, which are pairs of duplicated loci, are common in salmonid ﬁshes, but the electrophoretic patterns of isoloci may be ambiguous (Waples and Aebersold, 1990). Variations in the manner that isoloci are scored have been shown to have a large effect on the estimation of stock composition from a mixed-stock sample (Smouse et al., 1990; Waples and Aebersold, 1990). If such a pair consists of two alleles each, then nine two-locus genotypes are possible, but only ﬁve phenotypes can be discerned electrophoretically (Waples and Aebersold, 1990). Usually, allele frequencies for isoloci are estimated as a single average frequency by assuming that the frequencies of each allele arise equally from each loci in the pair. However, this introduces bias in phenotype frequency estimates for the source stock when true frequencies among loci are not equal. Based on a simulation study, Waples and Aebersold (1990) demonstrated that the precision of stock composition estimates improved when a maximum-likelihood estimate of the allele frequencies for each loci was calculated prior to mixed-stock analysis. Another technique for treating isoloci is to attribute the variation to a single locus, as Smouse et al. (1990) did for Chinook salmon; however, this approach may not be suitable for other isoloci and may result in loss of information for stock separation. When studied in detail, isolocus systems can display complex, tissuespeciﬁc patterns of variation that may be useful in stock delineation (Shaklee and Phelps, 1992). Another problem with allelic designations arises from improper handling of tissue samples for allozyme analyses. In a mixed-stock analysis of Chinook salmon stocks, a rare allele that was present in only 2 of 111 source stocks was found in 49% of ﬁsh in the mixture sample (Brodziak et al., 1992). The rare allele detected in the mixture sample was actually an artifact associated with poor tissue quality (Brodziak et al., 1992). The discrepancy was detected because ﬁsh from the mixture sample carried coded wire tags and their stock of origin was known. An extremely large number of alleles may be resolved at minisatellite and microsatellite loci (e.g., Scribner et al., 1994, Beacham et al., 2001), and as a result, the estimation of allele frequencies for source stocks may be unreliable for all but the more common alleles (Pella and Masuda, 2001). In this situation, the sampling error of the baseline samples cannot be ignored. (A fundamental

488

Mary C. Fabrizio

assumption of the conditional mixture model is that allele frequencies from source stocks are known without error.) With so many alleles to consider, the relative frequency of some alleles will be low (near zero) and difﬁcult to estimate precisely when source stock samples are limited (Pella and Masuda, 2001). Two solutions have been proposed. The ﬁrst uses an unconditional likelihood method to estimate stock compositions from a mixture. This method eliminates the need to assume error-free allele frequencies of source stocks because allele frequencies are correctly treated as sample estimates (Smouse et al., 1990). The second uses a Bayesian method to delineate stocks in a mixture (Pella and Masuda, 2001). The Bayesian approach uses information on genetic similarities among stocks and exploits the hierarchical grouping suggested by allelic similarities—more common alleles are expected to be present in all or most stocks from the region, less common alleles are expected to be present in stocks from a subregion, and rare alleles are expected to be associated with individual stocks. Accurate resolution and interpretation of allelic fragment size from studies using microsatellite and minisatellite DNA probes are difﬁcult, and allelic designation was recognized as a potential problem in early studies. Minisatellite and microsatellite alleles that vary only slightly in size may escape detection, or two identically sized minisatellite alleles may have different sequences (Bentzen and Wright, 1993; O’Connell and Wright, 1997). In some studies, measurement errors exceeded the length of the repeated unit (see Scribner et al., 1994). Fragments that are close in size may be difﬁcult to distinguish due to crowding on the gel, and the number of alleles (fragments) may be ambiguous (Taggart and Ferguson, 1990; Bentzen et al., 1993; Dahle, 1994). Fragments (bands) are scored according to their molecular weight using a “bin” or size-class procedure, which is a technique to transform continuous information to discrete data (Scribner et al., 1994; Beacham et al., 1995). In this approach, bins are considered to represent “alleles,” but occasionally, more than one fragment (band) may be detected within a given bin (Beacham et al., 1995). An additional problem arises in multilocus minisatellite analyses in which some bands cannot be associated with speciﬁc loci (Scribner et al., 1994). Microsatellite alleles may be so numerous that researchers may opt to combine the low-frequency alleles with adjacent alleles, resulting in fewer genotypes but little loss in the ability to discriminate stocks (e.g., Small et al., 1998; Beacham et al., 2001). Allelic scoring errors decrease discriminatory power in general, but the effect of scoring errors on classiﬁcation accuracy has not been fully studied (Bernatchez and Duchesne, 2000).

VI. POWER TO DETECT STOCK DIFFERENCES Peterman (1990) ﬁrst called attention to the question of the power of statistical procedures used to delineate stocks and challenged practitioners to calculate sta-

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

489

tistical power when the null hypothesis could not be rejected. Since then, several studies have included a power analysis when stock differences could not be supported by the data (e.g., Buonaccorsi et al., 2001). In the classical (experimental) approach to statistical testing, investigators select a small alpha value (e.g., 0.05) to ensure that the null hypothesis is rejected only when there is strong evidence against it. This may not be a prudent approach in a risk-averse situation where the desired outcome is the conservation of stocks. High costs are incurred if the null hypothesis is not rejected when it is false because true stock differences will be undetected (Dizon et al., 1995). With respect to stock identiﬁcation, failure to reject the null hypothesis of no difference among stocks results in management of groups or populations as if they constituted a unit stock (Dizon et al., 1995). If stock differences are real, but the power to detect differences is low, management of multiple stocks as a single unit is likely to lead to overﬁshing of less productive stocks and possible underutilization of productive stocks. Other outcomes include the sequential “ﬁshing down” and loss of less abundant stocks. The rate of mixing and the proportion of the ﬁshing effort attributable to each stock are important factors in determining the magnitude of the loss in yield or overharvesting under equilibrium conditions (Fox, 1977). In risk-averse situations, it would be prudent to maintain high power to detect differences among stocks (Dizon et al., 1995). Although procedures for estimating statistical power of some linear models have been explored (Self and Mauritsen, 1988), most work to date has focused on specialized models with speciﬁc applications beyond the problem of classiﬁcation (Self et al., 1992). Similarly, power calculations for maximum-likelihood mixture models typically used to discern genetic differences among stocks have not been described. It has been suggested that Monte Carlo simulation can be used to investigate power when formulae do not exist (Peterman, 1990). A power analysis requires stipulation of the magnitude of the difference considered meaningful by the investigator (Toft and Shea, 1983). In stock identiﬁcation studies this can be difﬁcult and is often subjective (Dizon et al., 1995). A percentage of the total variation within a species may be associated with the individual stocks, but the biological meaning and management signiﬁcance of the partitioning of the intraspeciﬁc variation is unclear. For example, genetic studies can often distinguish individual ﬁsh (based on a genetic “ﬁngerprint”) (Dizon et al., 1995) or families of ﬁsh (parents and offspring) (Letcher and King, 1999), but this level of resolution is usually not relevant to ﬁshery management objectives. The challenge is to identify the level of resolution that is signiﬁcant for management purposes. This will become more pertinent as the newer DNA probing techniques are used to study stock structure in ﬁshes.

490

Mary C. Fabrizio

VII. CONCLUSIONS Stock structure, when it exists, can be detected with a wide variety of techniques, but techniques should be selected carefully and interpretations made with caution. Not all studies of stock structure in ﬁshes uncover useful variation. For marine ﬁshes, most of the genetic variability occurs within and not between putative stocks (Smith et al., 1990). This is particularly true for large, migratory species such as tunas and mackerels which exhibit high levels of genetic similarity attributable to gene ﬂow between groups (Smith et al., 1990). In the absence of a power analysis, it may be difﬁcult to judge the adequacy of selected methods that reveal a lack of stock structure. Another aspect to consider when genetic studies reveal little variation is the importance or role of observed phenotypic variability. Is stock structure “real” when genotypic, but not phenotypic, differences are detected? Indeed, phenotypic characters are likely to reﬂect environmental ﬂuctuations and not biological adaptation. Paciﬁc ocean perch (Sebastes alutus) provide a good example of such a case. This species appears to be genetically similar throughout its range, exhibiting only gradual differences in allele frequencies from north to south (Seeb and Gunderson, 1988). Although lacking genetic differences, local groups of Paciﬁc ocean perch exhibit differences in growth and maturation rates (Seeb and Gunderson, 1988). The response of local groups to ﬁshing pressure may therefore vary and resource managers may wish to treat these groups as separate stocks for management purposes. On the other hand, some multivariate techniques used to differentiate stocks are quite sensitive to small differences among groups and may erroneously indicate the existence of stocks when discrete samples are collected along a cline (Bowering, 1988). Thus, the validity of using a certain set of features does not depend on the underlying mechanism of expression, whether it be genotypic or phenotypic, but rather, the approach and cautions exercised by the investigator play a large role in determining the reliability of the study and its implications. In this chapter, I argued that the experimental design and sampling strategy critically affect the reliability of a mixed-stock analysis. Because most of the observed variation in genotypic (e.g., 96% for pink salmon; Beacham et al., 1985) or phenotypic characters is common to all stocks, resolution is based on a small amount of detectable variation. Sampling adequacy is acutely important. Factors such as representativeness of the samples, sample size, and underlying biological and statistical assumptions must be thoroughly investigated before reliable stock analyses can be obtained. With respect to how well samples represent the source stock, care must be exercised to obtain ﬁsh of known stock origin. Stray ﬁsh should be avoided, and collection of samples during the appropriate time and in the appropriate place is important to assure sampling discrete stocks. In any case, sampling issues that are under the control of the investigator should not be overlooked as possible ways to increase the reliability of stock delineations.

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

491

A number of factors that are not under the investigator’s direct control— “problem stocks,” minor stocks, and unknown stocks—can affect the bias and precision of stock composition estimates. “Problem stocks,” that is, stocks with high similarities, are difﬁcult to delineate reliably, and when present in the baseline (or mixture), lead to poor performance of the mixture model (Wood et al., 1987; Wood, 1989). The reliability of stock composition estimates is affected by the magnitude of the difference among stocks, more so than the number of source stocks considered or the number of ﬁsh in the mixture sample (Wood, 1989). The investigator cannot alter the degree of stock separation, but can control the number of features considered, how the mixture is sampled, and the number of ﬁsh in the mixture sample. These attributes can be adjusted to maximize accuracy of stock composition estimates. However, even large mixture samples may not resolve accurately with the mixture model when similarity is high (Wood et al., 1987). The second factor, minor stocks, concerns stocks that contribute little or nothing to the mixture. It has been shown through simulation studies that the contribution of “missing” stocks will be overestimated by the mixture model (Mulligan et al., 1988). These types of errors can lead to overestimation of harvest rates for stocks with small spawning runs (e.g., sockeye salmon; Cass and Wood, 1994). Unless the contribution of minor stocks is the primary concern, it may be more appropriate to form composite or regional stocks. Unknown stocks, the third issue, refers to stocks present in the mixture but not in the baseline sample. Fish from unsampled stocks will affect the accuracy of stock composition estimates because the proportion will be attributed to similar stocks that were sampled in the baseline (Brodziak et al., 1992). One version of the mixture model (Smouse et al., 1990) allows detection of unsampled stocks. This situation should be suspected in cases where samples from limited geographic areas are used to describe large regional (or composite) stocks. In general, source stock samples may be more reliable for classiﬁcation when population characteristics of the mixture ﬁsh are closely reﬂected by the source stocks. For example, matching length frequencies, age classes (Waples, 1990), or sex ratios will eliminate additional variability due to these factors. Although this would appear to apply to phenotypic features only, it is true for genotypic features in the presence of genetic drift or when populations are small and allele frequencies vary through time. Estimates of stock composition are most useful if conﬁdence intervals are provided (Mulligan et al., 1988), but estimation of variance for the mixture model is not straightforward. In addition, variance can be underestimated with the classiﬁcation model. A jackknife approach or bootstrap resampling (Efron, 1982) can be used effectively to estimate conﬁdence intervals for stock composition estimates.

492

Mary C. Fabrizio

Simulation studies can provide insights on the sensitivity of the mixture model to sampling decisions and characteristics of the stocks. These types of studies allow investigators to assess the reliability of the model used to classify ﬁsh and to determine if simpler sampling strategies or fewer features can yield equally reliable results. Millar (1990) published a computer program that performs simulation studies to determine the effects of sample size, number of stocks, and number of features, allowing investigators to examine the reliability of the model. Such resampling approaches are recommended to understand the effect of randomness in sampling ﬁsh stocks.

ACKNOWLEDGMENTS Discussions with Mary Burnham Curtis (U.S. Fish and Wildlife Service) and Tim King (USGS) greatly improved my understanding of genetic markers for mixed-stock analyses. Steve Cadrin (NOAA-Fisheries) provided helpful comments on an earlier version of this manuscript; Jeff Pessutti (NOAA-Fisheries) drafted the ﬁgures; and Ann Zimmerman (USGS) and Judy Berrien (NOAAFisheries) assisted in library research for this chapter. Much of this material was prepared while I was a research biologist with the USGS Great Lakes Science Center, Ann Arbor, Michigan.

REFERENCES Agresti, A. 1990. Categorical Data Analysis. John Wiley, New York. Akin, P. A. 1988. Geographic variation in tooth morphology and dentinal patterns in the spinner dolphin, Stenella longirostris. Marine Mammal Science 4: 132–140. Beacham, T. D., Gould, A. P., Withler, R. E., Murray, C. B., and Barner, L. W. 1987. Biochemical genetic survey and stock identiﬁcation of chum salmon (Oncorhynchus keta) in British Columbia. Canadian Journal of Fisheries and Aquatic Sciences 44: 1702–1713. Beacham, T. D., Withler, R. E., and Gould, A. P. 1985. Biochemical genetic stock identiﬁcation of pink salmon (Oncorhynchus gorbuscha) in southern British Columbia and Puget Sound. Canadian Journal of Fisheries and Aquatic Sciences 42: 1474–1483. Beacham, T. D., Withler, R. E., and Wood, C. C. 1995. Stock identiﬁcation of sockeye salmon by means of minisatellite DNA variation. North American Journal of Fisheries Management 15: 249–265. Beacham, T. D., Le, K. D., Raap, M. R., Hyatt, K., Luedke, W., and Withler, R. E. 2000. Microsatellite DNA variation and estimation of stock composition of sockeye salmon, Oncorhynchus nerka, in Barley Sound, British Columbia. Fishery Bulletin 98: 14–24. Beacham, T. D., Candy, J. R., Supernault, K. J., Ming, T., B. Deagle, Schulze, A., Tuck, D., Kaukinen, K. H., Irvine, J. R., Miller, K. M., and Withler, R. E. 2001. Evaluation and application of microsatellite and major histocompatibility complex variation for stock identiﬁcation of coho salmon in British Columbia. Transactions of the American Fisheries Society 130: 1116–1149. Begg, G. A. and Brown, R. W. 2000. Stock identiﬁcation of haddock Melanogrammus aegleﬁnus on Georges Bank based on otolith shape analysis. Transactions of the American Fisheries Society 129: 935–945.

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

493

Bentzen, P., Taylor, E. B., and Wright, J. M. 1993. A novel synthetic probe for DNA ﬁngerprinting salmonid ﬁshes. Journal of Fish Biology 43: 313–316. Bentzen, P. and Wright, J. M. 1993. Nucleotide sequence and evolutionary conservation of a minisatellite variable number tandem repeat cloned from Atlantic salmon, Salmo salar. Genome 36: 271–277. Bermingham, E., Forbes, S. H., Friedland, K., and Pla, C. 1991. Discrimination between Atlantic salmon (Salmo salar) of North American and European origin using restriction analyses of mitochondrial DNA. Canadian Journal of Fisheries and Aquatic Sciences 48: 884–893. Bernatchez, L. and Duchesne, P. 2000. Individual-based genotype analysis in studies of parentage and population assignment: how many loci, how many alleles? Canadian Journal of Fisheries and Aquatic Sciences 57: 1–12. Biernacki, C., Beninel, F., and Bretagnolle, V. 2002. A generalized discriminant rule when training populations and test populations differ on their descriptive parameters. Biometrics 58: 387–397. Bird, J. L., Eppler, D. T., and Checkley, Jr., D. M. 1986. Comparisons of herring otoliths using Fourier series shape analysis. Canadian Journal of Fisheries and Aquatic Sciences 43: 1228–34. Blouw, D. M., Saxon, S. D., and Chadwick, E. M. P. 1988. Temporal variation of meristic traits within an Atlantic salmon (Salmo salar) stock, and implications for stock identiﬁcation. Canadian Journal of Fisheries and Aquatic Sciences 45: 1330–1339. Bowering, W. R. 1988. An analysis of morphometric characters of Greenland halibut (Reinhardtius hippoglossoides) in the northwest Atlantic using a multivariate analysis of covariance. Canadian Journal of Fisheries and Aquatic Sciences 45: 580–585. Brodziak, J. 1993. An extension of stock composition analysis to include marking data. Canadian Journal of Fisheries and Aquatic Sciences 50: 251–257. Brodziak, J., Bentley, B., Bartley, D., Gall, G. A. E., Gomulkiewicz, R., and Mangel, M. 1992. Tests of genetic stock identiﬁcation using coded wire tagged ﬁsh. Canadian Journal of Fisheries and Aquatic Sciences 49: 1507–1517. Brown, B. L., P. E. Smouse, J. M. Epifanio, and C. J. Kobak. 1999. Mitochondrial DNA mixed-stock analysis of American shad: coastal harvests are dynamic and variable. Transactions of the American Fisheries Society 128: 977–994. Bruford, M. W., Hanotte, O., Brookﬁeld, J. F. Y., and Burke, T. 1992. Single-locus and multilocus DNA ﬁngerprinting. In A. R. Hoelzel (ed.), Molecular Genetic Analysis of Populations: A Practical Approach. Oxford University Press, New York, pp. 225–269. Buonaccorsi, V. P., Starkey, E., and Graves, J. E. 2001. Mitochondrial and nuclear DNA analysis of population subdivision among young-of-the-year Spanish mackerel (Scomberomorus maculatus) from the western Atlantic and Gulf of Mexico. Marine Biology 138: 37–45. Burnaby, T. P. 1966. Growth-invariant discriminant functions and generalized distances. Biometrics 22: 96–110. Cadrin, S. X. 2000. Advances in morphometric identiﬁcation of ﬁshery stocks. Reviews in Fish Biology and Fisheries 10: 91–112. Campana, S. E. and Casselman, J. M. 1993. Stock discrimination using otolith shape analysis. Canadian Journal of Fisheries and Aquatic Sciences 50: 1062–1083. Campana, S. E. and Gagné, J. A. 1995. Cod stock discrimination using ICPMS elemental assays of otoliths. In D. H. Secor, J. M. Dean, and S. E. Campana (eds.), Recent Developments in Fish Otolith Research. University of South Carolina Press, Columbia, SC, pp. 671–691. Campana, S. E. and 18 coauthors. 1997. Comparison of accuracy, precision, and sensitivity in elemental assays of ﬁsh otoliths using the electron microprobe, proton-induced X-ray emission, and laser ablation inductively coupled plasma mass spectrometry. Canadian Journal of Fisheries and Aquatic Sciences 54: 2068–2079.

494

Mary C. Fabrizio

Cass, A. J. and Wood, C. C. 1994. Evaluation of the depensatory ﬁshing hypothesis as an explanation for population cycles in Fraser River sockeye salmon (Oncorhynchus nerka). Canadian Journal of Fisheries and Aquatic Sciences 51: 1839–1854. Casselman, J. M., Collins, J. J., Crossman, E. J., Ihssen, P. E., and Spangler, G. R. 1981. Lake whiteﬁsh (Coregonus clupeaformis) stocks of the Ontario waters of Lake Huron. Canadian Journal of Fisheries and Aquatic Sciences 38: 1772–1789. Castonguay, M., Simard, P., and Gagnon, P. 1991. Usefulness of Fourier analysis of otolith shape for Atlantic mackerel (Scomber scombus) stock discrimination. Canadian Journal of Fisheries and Aquatic Sciences 48: 296–302. Chakraborty, R. 1993. A class of population genetic questions formulated as the generalized occupancy problem. Genetics 134: 953–958. Cochran, W. G. 1954. Some methods of strengthening the common c2 tests. Biometrics 10: 417–451. Dahle, G. 1994. Minisatellite DNA ﬁngerprints of Atlantic cod (Gadus morhua). Journal of Fish Biology 44: 1089–1092. Danzmann, R. G., Ihssen, P. E., and Hebert, P. D. N. 1991. Genetic discrimination of wild and hatchery populations of brook charr, Salvelinus fontinalis (Mitchell), in Ontario using mitochondrial DNA analysis. Journal of Fish Biology 39(suppl. A): 69–77. DeVries, D. A., Grimes, C. B., and Prager, M. H. 2002. Using otolith shape analysis to distinguish eastern Gulf of Mexico and Atlantic Ocean stocks of king mackerel. Fisheries Research 57: 51–62. Dillon, R. T., Jr., and Manzi, J. J. 1992. Population genetics of the hard clam, Mercenaria mercenaria, at the northern limit of its range. Canadian Journal of Fisheries and Aquatic Sciences 49: 2574–2578. Dizon, A. E., Taylor, B. L., and O’Corry-Crowe, G. M. 1995. Why statistical power is necessary to link analyses of molecular variation to decisions about population structure. American Fisheries Society Symposium 17 (Evolution and the Aquatic Ecosystem: Deﬁning Unique Units in Population Genetics): 288–294. Edmonds, J. S., Caputi, N., Moran, M. J., Fletcher, W. J., and Morita, M. 1995. Population discrimination by variation in concentrations of minor and trace elements in sagittae of two Western Australian teleosts. In D. H. Secor, J. M. Dean, and S. E. Campana (eds.), Recent Developments in Fish Otolith Research. University of South Carolina Press, Columbia, SC, pp. 655–670. Efron, B. 1982. The jackknife, the bootstrap and other resampling plans. Regional Conference Series in Applied Mathematics 38. SIAM, Philadelphia, PA. Epifanio, J. M., Smouse, P. E., Kobak, C. J., and Brown, B. L. 1995. Mitochondrial DNA divergence among populations of American shad (Alosa sapidissima): how much variation is enough for mixed-stock analysis? Canadian Journal of Fisheries and Aquatic Sciences 52: 1688–1702. Fabrizio, M. C. 1987. Growth-invariant discrimination and classiﬁcation of striped bass stocks by morphometric and electrophoretic methods. Transactions of the American Fisheries Society 116: 728–736. Fairbairn, D. J. and Roff, D. A. 1980. Testing genetic models of isozyme variability without breeding data: can we depend on the c2? Canadian Journal of Fisheries and Aquatic Sciences 37: 1149–1159. Ferguson, M. M. and Danzmann, R. G. 1998. Role of genetic markers in ﬁsheries and aquaculture: useful tools or stamp collecting? Canadian Journal of Fisheries and Aquatic Sciences 55: 1553–1563. Fontaine, P.-M., Dodson, J. J., Bernatchez, L., and Slettan, A. 1997. A genetic test of metapopulation structure in Atlantic salmon (Salmo salar) using microsatellites. Canadian Journal of Fisheries and Aquatic Sciences 54: 2434–2442. Fournier, D. A., Beacham, T. D., Riddell, B. E., and Busack, C. A. 1984. Estimating stock composition in mixed stock ﬁsheries using morphometric, meristic, and electrophoretic characteristics. Canadian Journal of Fisheries and Aquatic Sciences 41: 400–408.

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

495

Fox, W. W., Jr. 1977. Some effects of stock mixing on management decisions. Report of the International Whaling Commission 27. Cambridge, England. Gillanders, B. M. 2002. Temporal and spatial variability in elemental composition of otoliths: implications for determining stock identity and connectivity of populations. Canadian Journal of Fisheries and Aquatic Sciences 59: 669–679. Gold, J. R., Richardson, L. R., King, T. L., and Matlock, G. C. 1993. Temporal stability of nuclear gene (allozyme) and mitochondrial DNA genotypes among red drums from the Gulf of Mexico. Transactions of the American Fisheries Society 122: 659–668. Grant, W. S., Milner, G. B., Krasnowski, P., and Utter, F. M. 1980. Use of biochemical genetic variants for identiﬁcation of sockeye salmon (Oncorhynchus nerka) stocks in Cook Inlet, Alaska. Canadian Journal of Fisheries and Aquatic Sciences 37: 1236–1247. Grewe, P. M., Krueger, C. C., Aquadro, C. F., and May, B. 1994. Stability of allozyme and mitochondrial DNA markers among three year-classes of lake trout propagated from Seneca Lake, New York. North American Journal of Fisheries Management 14: 467–474. Hand, D. J. 1981. Discrimination and Classiﬁcation. Wiley, New York. Hilborn, R. 1985. Apparent stock recruitment relationships in mixed stock ﬁsheries. Canadian Journal of Fisheries and Aquatic Sciences 42: 718–723. Jacobson, L. D., Torblaa, R. L., and Morman, R. H. 1986. Temporal variation in allelic frequencies of sea lamprey ammocoetes. Copeia 1986: 199–202. King, D. P. F. 1985. Morphological and meristic differences among spawning aggregations of northeast Atlantic herring, Clupea harengus L. Journal of Fish Biology 26: 591–607. King, T. L., Ward, R., and Zimmerman, E. G. 1994. Population structure of eastern oysters (Crassostrea virginica) inhabiting the Laguna Madre, Texas, and adjacent bay systems. Canadian Journal of Fisheries and Aquatic Sciences 51(suppl. 1): 215–222. Klingenberg, C. P. 1996. Multivariate allometry. In L. F. Marcus, M. Corti, A. Loy, G. J. P. Naylor, and D. E. Slice (eds.), Advances in Morphometrics. NATO ASI Series A: Life Sciences, Vol. 284. Plenum Press, New York, pp. 23–49. Kondzela, C. M., Guthrie, C. M., Hawkins, S. L., Russell, C. D., Helle, J. H., and Gharrett, A. J. 1994. Genetic relationships among chum salmon populations in southeast Alaska and northern British Columbia. Canadian Journal of Fisheries and Aquatic Sciences 51(suppl. 1): 50–64. Lachenbruch, P. A. 1968. On expected probabilities of misclassiﬁcation in discriminant analysis, necessary sample size, and a relation with the multiple correlation coefﬁcient. Biometrics 24: 823–834. Letcher, B. H. and King, T. L. 1999. Targeted stock identiﬁcation using multilocus genotype “familyprinting.” Fisheries Research 43: 99–111. Leung, F. C., Welt, M., Quesenberry, R. D., and Shen, X-Z. 1994. DNA ﬁngerprinting and cloning of hypervariable minisatellite repeats in salmonids. Canadian Journal of Fisheries and Aquatic Sciences 51(suppl. 1): 258–266. Margraf, F. J. and Riley, L. M. 1993. Evaluation of scale shape for identifying spawning stocks of coastal Atlantic striped bass (Morone saxatilis). Fisheries Research 18: 163–172. Meng, H. J. Stocker, and M. 1984. An evaluation of morphometrics and meristics for stock separation of Paciﬁc herring (Clupea harengus pallasi). Canadian Journal of Fisheries and Aquatic Sciences 41: 414–422. Millar, R. B. 1987. Maximum likelihood estimation of mixed stock ﬁshery composition. Canadian Journal of Fisheries and Aquatic Sciences 44: 583–590. Millar, R. B. 1990. A versatile computer program for mixed stock ﬁshery composition estimation. Canadian Technical Report of Fisheries and Aquatic Sciences 1753. 32 pp. Milner, G. B., Teel, D. J., Utter, F. M., and Winans, G. A. 1985. A genetic method of stock identiﬁcation in mixed populations of Paciﬁc salmon, Oncorhynchus spp. Marine Fisheries Review 47(1): 1–8.

496

Mary C. Fabrizio

Misra, R. K. and Carscadden, J. E. 1984. Stock discrimination of capelin (Mallotus villosus) in the northwest Atlantic. Journal of Northwest Atlantic Fisheries Science 5: 199–205. Misra, R. K. and Easton, M. D. L. 1999. A note on the number of morphometric characters used in ﬁsh stock delineation studies employing a MANOVA. Fisheries Research 42: 191–194. Mulligan, T. J., McKinnell, S., and Wood, C. C. 1988. Uncertainty in stock composition estimates of oceanic steelhead trout using electrophoretic characteristics: comments on a recent study. Canadian Journal of Fisheries and Aquatic Sciences 45: 432–442. Myers, K. W., Harris, C. K., Knidsen, C. M., Walker, R. V., Davis, N. D., and Rogers, D. E. 1987. Stock origins of chinook salmon in the area of the Japanese mothership salmon ﬁshery. North American Journal of Fisheries Management 7: 459–474. Nielsen, E. E., Hansen, M. M., and Bach, L. A. 2001. Looking for a needle in a haystack: discovery of indigenous Atlantic salmon (Salmo salar L.) in stocked populations. Conservation Genetics 2: 219–232. O’Connell, M. and Wright, J. M. 1997. Microsatellite DNA in ﬁshes. Reviews in Fish Biology and Fisheries 7: 331–363. Parkinson, E. A. 1984. Genetic variation in populations of steelhead trout (Salmo gairdneri) in British Columbia. Canadian Journal of Fisheries and Aquatic Sciences 41: 1412–1420. Pella, J. J. and Milner, G. B. 1987. Use of genetic marks in stock composition analysis. In N. Ryman and F. Utter (eds.), Population Genetics and Fishery Management. University of Washington Press, Seattle, WA, pp. 247–276. Pella, J. and Masuda, M. 2001. Bayesian methods for analysis of stock mixtures from genetic characters. Fishery Bulletin 99: 151–167. Peterman, R. M. 1990. Statistical power analysis can improve ﬁsheries research and management. Canadian Journal of Fisheries and Aquatic Sciences 47: 2–15. Phelps, S. R., LeClair, L. L., Young, S., and Blankenship, H. L. 1994. Genetic diversity patterns of chum salmon in the Paciﬁc northwest. Canadian Journal of Fisheries and Aquatic Sciences 51(suppl. 1): 65–83. Power, G. 1981. Stock characteristics and catches of Atlantic salmon (Salmo salar) in Quebec, and Newfoundland and Labrador in relation to environmental variables. Canadian Journal of Fisheries and Aquatic Sciences 38: 1601–1611. Prager, M. H. and Fabrizio, M. C. 1990. Comparison of logistic regression and discriminant analyses for stock identiﬁcation of anadromous ﬁsh, with application to striped bass (Morone saxatilis) and American shad (Alosa sapidissima). Canadian Journal of Fisheries and Aquatic Sciences 47: 1570–1577. Reddin, D. G. 1986. Discrimination between Atlantic salmon (Salmo salar L.) of North American and European origin. Journal du Conseil International pour l’Exploration de la Mer 43: 50–58. Reist, J. D. 1986. An empirical evaluation of coefﬁcients used in residual and allometric adjustment of size covariation. Canadian Journal of Zoology 64: 1363–1368. Reyment, R. A., Blackwith, R. E., and Campbell, N. A. 1984. Multivariate Morphometrics, 2nd ed. Academic Press, New York. Ricker, W. E. 1973. Two mechanisms that make it impossible to maintain peak period yields from stocks of Paciﬁc salmon and other ﬁshes. Journal of the Fisheries Research Board of Canada 30: 1275–1286. Ricker, W. E. 1975. Computation and interpretation of biological statistics of ﬁsh populations. Bulletin of the Fisheries Research Board of Canada 191. 382 pp. Roff, D. A. and Bentzen, P. 1989. The statistical analysis of mitochondrial DNA polymorphisms: c2 and the problem of small samples. Molecular Biology and Evolution 6: 539–545. Rohlf, F. J. and Bookstein, F. L. 1987. A comment on shearing as a method for “size correction.” Systematic Zoology 36: 356–367.

Experimental Design and Sampling Strategies for Mixed-Stock Analysis

497

Rooker, J. R., Secor, D. H., Zdanowicz, V. S., and Itoh, T. 2001. Discrimination of northern blueﬁn tuna from nursery areas in the Paciﬁc Ocean using otolith chemistry. Marine Ecology Progress Series 218: 275–282. Ruzzante, D. E., Taggart, C. T., Doyle, R. W., and Cook, D. 2001. Stability in the historical pattern of genetic structure of Newfoundland cod (Gadus morhua) despite the catastrophic decline in population size from 1964 to 1994. Conservation Genetics 2: 257–269. Scribner, K. T., Arntzen, J. W., and Burke, T. 1994. Comparative analysis of intra- and interpopulation genetic diversity in Bufo bufo, using allozyme, single-locus microsatellite, minisatellite, and multilocus minisatellite data. Molecular Biology and Evolution 11: 737–748. Scribner, K. T., Crane, P. A., Spearman, W. J., and Seeb, L. W. 1998. DNA and allozyme markers provide concordant estimates of population differentiation: analyses of U.S. and Canadian populations of Yukon river fall-run chum salmon (Oncorhynchus keta). Canadian Journal of Fisheries and Aquatic Sciences 55: 1748–1758. Seeb, L. W. and Gunderson, D. R. 1988. Genetic variation and population structure of Paciﬁc ocean perch (Sebastes alutus). Canadian Journal of Fisheries and Aquatic Sciences 45: 78–88. Seeb, L. W., Seeb, J. E., and Polovina, J. J. 1990. Genetic variation in highly exploited spiny lobster Panulirus marginatus populations from the Hawaiian archipelago. Fishery Bulletin 88: 713–718. Self, S. G. and Mauritsen, R. H. 1988. Power/sample size calculations for generalized linear models. Biometrics 44: 79–86. Self, S. G., Mauritsen, R. H., and Ohara, J. 1992. Power calculations for likelihood ratio tests in generalized linear models. Biometrics 48: 31–39. Shaklee, J. B. and Phelps, S. R. 1992. Chinook salmon NADP+-dependent cytosolic isocitrate dehydrogenase: electrophoretic and genetic dissection of a complex isozyme system and geographic patterns of variation. Biochemical Genetics 30: 455–489. Small, M. P., Withler, R. E., and Beacham, T. D. 1998. Population structure and stock identiﬁcation of British Columbia coho salmon, Oncorhynchus kisutch, based on microsatellite DNA variation. Fishery Bulletin 96: 843–858. Smith, P. J., Jamieson, A., and Birley, A. J. 1990. Electrophoretic studies and the stock concept in marine teleosts. Journal du Conseil International pour l’Exploration de la Mer 47: 231–245. Smouse, P. E., Spielman, R. S., and Park, M. H. 1982. Multiple-locus allocation of individuals to groups as a function of the genetic variation within and differences among human populations. American Naturalist 111: 445–463. Smouse, P. E., Waples, R. S., and Tworek, J. A. 1990. A genetic mixture analysis for use with incomplete source population data. Canadian Journal of Fisheries and Aquatic Sciences 47: 620–634. Swain, D. P., Frank, K. T., and Maillet, G. 2001. Delineating stocks of Atlantic cod (Gadus morhua) in the Gulf of St. Lawrence and Cabot Strait areas using vertebral number. ICES Journal of Marine Science 58: 253–269. Taggart, J. B. and Ferguson, A. 1990. Hypervariable minisatellite DNA single locus probes for the Atlantic salmon, Salmo salar L. Journal of Fish Biology 37: 991–993. Thorpe, R. S. 1988. Multiple group principal component analysis and population differentiation. Journal of Zoology 216: 37–40. Thresher, R. E. 1999. Elemental composition of otoliths as a stock delineator in ﬁshes. Fisheries Research 43: 165–204. Toft, C. A. and Shea, P. J. 1983. Detecting community-wide patterns: estimating power strengthens statistical inference. American Naturalist 122: 618–625. Turner, B. J., Elder, Jr., J. F., and Laughlin, T. F. 1991. Repetitive DNA sequences and the divergence of ﬁsh populations: some hopeful beginnings. Journal of Fish Biology 39(suppl. A): 131–142. Utter, F. M., Campton, D., Grant, S., Milner, G., Seeb, J., and Wishard, L. 1980. Population structures of indigenous salmonid species of the Paciﬁc northwest. In W. J. McNeil and D. C.

498

Mary C. Fabrizio

Himsworth (eds.), Salmonid Ecosystems of the North Paciﬁc. Oregon State University Press, Corvallis, OR, pp. 285–304. Utter, F. and Ryman, N. 1993. Genetic markers and mixed stock ﬁsheries. Fisheries 18(8): 11–21. VanWinkle, W., Kumar, K. D., and Vaughan, D. S. 1988. Relative contributions of Hudson River and Chesapeake Bay striped bass stocks to the Atlantic coastal population. In L. W. Barnthouse, R. J. Klauda, D. S. Vaughan, and R. L. Kendall (eds.), Science, Law, and Hudson River Power Plants: A Case Study of Environmental Impact Assessments. American Fisheries Society, Monograph 4, Bethesda, MD, pp. 255–266. Waldman, J. R. and Fabrizio, M. C. 1994. Problems of stock deﬁnition in estimating relative contributions of Atlantic striped bass to the coastal ﬁshery. Transactions of the American Fisheries Society 123: 766–778. Waldman, J. R., Hart, J. T., and Wirgin, I. I. 1996. Stock composition of the New York Bight Atlantic sturgeon ﬁshery based on analysis of mitochondrial DNA. Transactions of the American Fisheries Society 125: 364–371. Waldman, J. R., Richards, R. A., Schill, W. B., Wirgin, I., and Fabrizio, M. C. 1997. An empirical comparison of stock identiﬁcation techniques applied to striped bass. Transactions of the American Fisheries Society 126: 369–385. Waples, R. S. 1990. Temporal changes of allele frequency in Paciﬁc salmon: implications for mixedstock ﬁshery analysis. Canadian Journal of Fisheries and Aquatic Sciences 47: 968–976. Waples, R. S. and Aebersold, P. B. 1990. Treatment of data for duplicated gene loci in mixed-stock ﬁshery analysis. Canadian Journal of Fisheries and Aquatic Sciences 47: 2092–2098. Waples, R. S. and Teel, D. J. 1990. Conservation genetics of Paciﬁc salmon. I. Temporal changes in allele frequency. Conservation Biology 4: 144–156. Ward, R. D., Elliott, N. G., Innes, B. H., Smolenski, A. J., and Grewe, P. M. 1997. Global population structure of yellowﬁn tuna, Thunnus albacares, inferred from allozyme and mitochondrial DNA variation. Fishery Bulletin 95: 566–575. Winans, G. A., Viele, D., Grover, A., Palmer-Zwahlen, M., Teel, D., and VanDoornik, D. 2001. An update of genetic stock identiﬁcation of chinook salmon in the Paciﬁc northwest: test ﬁsheries in California. Reviews in Fisheries Science 9: 213–237. Wirgin, I., Maceda, L., Waldman, J. R., and Crittenden, R. N. 1993. Use of mitochondrial DNA polymorphisms to estimate the relative contribution of the Hudson River and Chesapeake Bay striped bass stocks to the mixed ﬁshery on the Atlantic coast. Transactions of the American Fisheries Society 122: 669–684. Wood, C. C. 1989. Utility of similarity dendrograms in stock composition analysis. Canadian Journal of Fisheries and Aquatic Sciences 46: 2121–2128. Wood, C. C., McKinnell, S., Mulligan, T. J., and Fournier, D. A. 1987. Stock identiﬁcation with the maximum-likelihood mixture model: sensitivity analysis and application to complex problems. Canadian Journal of Fisheries and Aquatic Sciences 44: 866–881. Wood, C. C., Rutherford, D. T., and McKinnell, S. 1989. Identiﬁcation of sockeye salmon (Oncorhynchus nerka) stocks in mixed-stock ﬁsheries in British Columbia and southeast Alaska using biological markers. Canadian Journal of Fisheries and Aquatic Sciences 46: 2108–2120. Xu, S., Kobak, C. J., and Smouse, P. E. 1994. Constrained least squares estimation of mixed population stock composition from mtDNA haplotype frequency data. Canadian Journal of Fisheries and Aquatic Sciences 51: 417–425.

CHAPTER

24

An Introduction to Statistical Algorithms Useful in Stock Composition Analysis MICHAEL H. PRAGER AND KYLE W. SHERTZER Population Dynamics Team, Center for Coastal Fisheries and Habitat Research, National Oceanic and Atmospheric Administration, Beaufort, North Carolina, USA

I. The Problem and Its Terminology II. Algorithms A. Discriminant Analysis B. Logistic Regression C. Artiﬁcial Neural Networks D. Finite Mixture Distribution (FMD) Methods III. The Importance of Prior Knowledge A. Priors and Discriminant Analysis B. Priors and Logistic Regression C. Priors and Neural Networks D. Priors and FMD Methods IV. Discussion References

I. THE PROBLEM AND ITS TERMINOLOGY In many ﬁsheries, catches include ﬁsh that are conspeciﬁc but that originate in several spawning stocks. Because the population effects of ﬁshing—and thus the choice of suitable management approaches—depend on which stock or stocks are harvested, estimates of stock composition of catches are needed. This need has given rise to the set of techniques often labeled stock identiﬁcation. The focus of applications is usually on proportions in the harvest rather than on origin of individual ﬁsh; consequently, a more precise description of this work is stock composition analysis. Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

499

500

Michael H. Prager and Kyle W. Shertzer

We deﬁne stock composition analysis as estimation of the stock composition of a mixed-stock sample (usually, some part of the harvest) taken from a known number J of component stocks. The proportion that originates in any given stock j is represented Pj. Thus SJj=1Pj = 1; within this constraint, any particular Pj may equal zero. The process by which the Pj are estimated constitutes the stock composition analysis. Data used for such analyses are observations on characteristics of individual specimens; typical characteristics may include morphometrics, meristics, genetic characters, or chemical signatures (reviewed in Begg and Waldman, 1999). When we refer to characteristics here, we assume that they have been quantiﬁed in some reasonable way so that statistics (such as mean and variance) can be computed for the entire set (matrix) of observed characteristics. The probabilistic nature of the methods considered here is needed to account for overlap in the distribution of characteristics from different populations. When there is no overlap (as when using tags), the origin of each ﬁsh can be established with certainty, and much simpler methods can be used to deﬁne the stock composition of the catch. This discussion also assumes the availability of a training sample of individuals whose stock membership is known. The training sample is used to ﬁt a model by means of the investigator’s choice of algorithm, a word used here to denote a statistical method or group of related methods. The ﬁtted model is then used to estimate the stock composition of a mixed-stock sample (or samples) of the catch. The estimation can, but need not always, involve estimating the probability of stock membership of each individual in the mixed-stock sample. If in the course of estimation each individual is assigned membership in a particular stock, the method can be considered a classiﬁcation method. Classiﬁcation methods are a subset of all methods useful for stock composition analysis, because stock composition can be estimated without performing a classiﬁcation. The terminology of stock composition analysis is speciﬁc to ﬁshery science, but the general problem is not. Analyzing characteristics of individual objects in a mixture to estimate the mixture’s proportions is a general statistical problem known as ﬁnite mixture analysis. Constituent ﬁsh stocks in a mixed harvest are just one example of constituent classes or statistical populations of mixed objects; here, we tend to use the term class when describing algorithms generally, and stock when describing ﬁsheries applications. Estimating stock composition is therefore a special case of the general problem of estimating mixing proportions. We continue this chapter by introducing some algorithms useful for stock composition analysis. We then discuss issues involved in estimating performance of various algorithms, either in an absolute sense or relative to one another on a particular data set. We close with a few general comparative remarks.

An Introduction to Statistical Algorithms Useful in Stock Composition Analysis

501

II. ALGORITHMS Many statistical algorithms have been proposed or used for stock composition analysis. We consider four—discriminant analysis, logistic regression, artiﬁcial neural networks, and ﬁnite mixture distributions—each of which may incorporate more than one observed characteristic. Variants of each algorithm exist, but since we focus on the conceptual basis of each algorithm, our treatment of such variants is generally brief.

A. DISCRIMINANT ANALYSIS (DA) Among classiﬁcation schemes, discriminant analysis (McLachlan, 1992; Johnson and Wichern, 1998; Hastie et al., 2001) boasts the longest history. Its two most common forms are linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA). Linear discriminant analysis was the ﬁrst formal statistical method used for stock composition analysis (Hill, 1959), and the method has been used many times and in many variations. One of its advantages is the wide availability of well-tested and ﬂexible software, as discriminant analysis forms an important component of most major statistics packages. The fundamental assumption of LDA is that observed characteristics follow a multivariate normal distribution with common variance–covariance structure among stocks. If the characteristics vector is represented X, we can write this distribution for stock j as fj(X) ~ MVN(mj,S) where mj is the mean characteristics vector for stock j and S is the common variance–covariance matrix. Typically, S and mj, j = {1, . . . , J} are estimated from the training set, and those estimates are used in forming discriminant functions. Classiﬁcation in LDA is determined from stock-speciﬁc linear discriminant functions, computed from three types of information: an individual’s characteristics vector x; estimates of mj and S; and a set of prior probabilities, or priors, pj, j = {1, . . . , J}. The priors are the analyst’s estimates (which may be subjective) of the probabilities that a randomly chosen ﬁsh originates in each of the j stocks. The discriminant function for stock j, evaluated for individual x, is d j (x) = x T Â -1 m j -

1 T -1 m j Â m j + log (p j ) 2

(1)

where the notation MT indicates the transpose of vector or matrix M. The quantity -dj(x) measures the distance from individual x to the center of stock j (the function is conventionally in negative distance for computational reasons). In classiﬁcation, each individual is assigned membership to the stock that maximizes dj(x), which is the stock with the closest center. This is also the stock with the largest posterior probability (eq. 3 of Pella and Masuda, this volume, Chapter 25).

502

Michael H. Prager and Kyle W. Shertzer

The J discriminant functions thus deﬁne decision boundaries that classify an individual, from its observed characteristics, to the most likely stock of origin. The decision boundary between stocks j and k occurs where dj(x) = dk(x), and it is this decision boundary that, when solved for x, is linear in the observed characteristics. If there are two measured characteristics, the decision boundary is a line; if three, a plane, and so on. Quadratic discriminant analysis (QDA) is often considered preferable for problems in which the variance–covariance structure differs by class (Misra, 1985), as QDA does not assume equality of variance among classes (Kendall et al., 1983). However, estimates from QDA generally are of higher variance than those from LDA because of the additional parameters that must be estimated. The quadratic discriminant function for stock j, evaluated for individual x, is 1 1 T log Â j - (x - m j ) Â -j 1 (x - m j ) + log (p j ) (2) 2 2 where Sj is the stock-speciﬁc variance–covariance matrix and |Sj| is its determinant. Classiﬁcations and posterior probabilities of class membership are computed as in LDA. The decision boundary in QDA is deﬁned as in LDA, but the resulting boundary is quadratic (curved) in the observed characteristics. For that reason, decision boundaries for LDA and QDA generally differ (Fig. 24-1). In using either linear or quadratic discriminant analysis to estimate stock composition, one can proceed in two slightly different ways. The usual procedure, which we term discrete classiﬁcation, is to classify each individual in the (mixed-stock) sample into the class for which its estimated membership probability is highest. The estimate of stock composition is then formed from the relative numbers of individuals classiﬁed into each class. In the second procedure, which we term nondiscrete classiﬁcation, the probability of membership in a particular class is summed across all individuals. The estimate of stock composition is then obtained from the sums for each class divided by the total sample size. As a simple example, consider a two-stock problem in which three ﬁsh are in the mixed-stock sample. Let the estimated probabilities of membership in class I for the three ﬁsh be {0.55, 0.45, 0.8}. The discrete estimate of mixing proportions would be 2/3 from class I and 1/3 from class II. The nondiscrete estimate would be 0.6 from class I and 0.4 from class II. The discrete estimate is obtained because two of the three ﬁsh are thought more likely to belong to class I; the nondiscrete is the mean of the three probabilities given. Although discrete and nondiscrete classiﬁcation usually produce similar estimates, it seems logical to prefer nondiscrete classiﬁcation. There is no necessity to round estimated membership probabilities to whole numbers, as done in discrete classiﬁcation, when the objective is to estimate mixing proportions. d j (x) = -

503

An Introduction to Statistical Algorithms Useful in Stock Composition Analysis

4

X2

2

0

–2

–4 –4

–2

0

2

4

X1 FIGURE 24-1. Example of decision boundaries between stock j (ﬁlled circles) and stock i (open circles) from linear (dotted line) and quadratic (solid line) discriminant analysis based on two 0.6 ˆ observed characteristics (X1, X2). The prior probabilities are pj = pi = 0.5; the means are m j = Ê Ë 0.75¯ and

-0.8 ˆ ; mi = Ê Ë -0.5 ¯

the

variance–covariance

matrices

are

1.5 Sj =Ê Ë -0.25

-0.25ˆ 1.25 ¯

and

1.8 - 0.2 ˆ ; and, with equal sample sizes, the pooled matrix S for the linear boundary is Si = Ê Ë -0.2 2.0 ¯ the mean of Sj and Si.

Several other variants of discriminant analysis have been applied to stock identiﬁcation; these include polynomial discriminant analysis (Cook and Lord, 1978), age-invariant discriminant analysis (Fabrizio, 1987), jackknife discriminant analysis (Small et al., 1998), and stepwise discriminant analysis (Palma and Andrade, 2002). Correction matrices, which can be computed from classiﬁcation results on a test data set, are frequently used to improve mixture estimates from discriminant analyses (Cook and Lord, 1978; Pella and Robertson, 1978) and

504

Michael H. Prager and Kyle W. Shertzer

might be used to correct estimates from other classiﬁcation-based methods as well. Millar (1987) demonstrated that the use of classiﬁcation with correction is a special case of maximum-likelihood ﬁnite mixture distribution methods (described below).

B. LOGISTIC REGRESSION Logistic regression (Aldrich and Nelson, 1984; Hosmer and Lemeshow, 1989; Agresti, 2002) is a type of generalized linear model (McCullagh and Nelder, 1989). It was suggested for stock identiﬁcation by Prager and Fabrizio (1990), who found the method promising. Its chief theoretical advantage is that it assumes neither multivariate normality of input data nor equality of variances and is appropriate for a wide variety of distributions (Kendall et al., 1983). It can also handle input data that are continuous, categorical, or a mix rather than continuous only, as in discriminant analysis. Logistic regression is applied most often to problems with a binary response, as when analyzing mixtures of two source stocks. But its use is not limited to binary problems, and indeed logistic regression has been applied to stock identiﬁcation problems with more than two stocks (Prager and Fabrizio, 1990; Waldman et al., 1997). In binary logistic regression, the response (Yi) for ﬁsh i takes one of two values: Yi = 0 implies membership in the ﬁrst stock, and Yi = 1 implies membership in the second. The probability that ﬁsh i with N measured characteristics xi = (xi1, xi2, . . . , xiN) belongs to the second stock is estimated by the continuous logistic function p, P(Y i = 1 | x i ) = p(x i ) =

exp(z i ) 1 + exp(z i )

(3)

where z i = b 0 + b1 x i1 + b 2 x i2 + . . . + b N x iN

(4)

and the b’s are parameters to be estimated (Fig. 24-2). The analysis deﬁned by eq. 4 is called multiple logistic regression, which refers not to the number of stocks, but to analyzing more than one characteristic per ﬁsh (i.e., N > 1). A suitable transformation, accomplished by use of a link function, converts the model of eq. 3 into one that is linear. Several standard links exist for binary data, such as logit, probit, or log-log (Agresti, 2002); we present here the logit link because it is most often applied in polytomous logistic regression (more than two classes). The logit link is log ÊË

p(x i ) ˆ = b 0 + b1 x i1 + b 2 x i2 + . . . + b N x iN 1 - p(x i ) ¯

(5)

505

An Introduction to Statistical Algorithms Useful in Stock Composition Analysis

1.0

0.8

p

0.6

0.4

0.2

0.0 –6

–4

–2

0

2

4

6

z FIGURE 24-2. Logistic transformation shown in eq. 3. In logistic regression for stock identiﬁcation, z is an unbounded linear combination of the measured characteristics. The value of the transformation p is the estimated probability of membership in the second of two stocks.

After transformation, errors are assumed to be distributed binomially, and parameters can readily be estimated by maximum likelihood using standard statistical software. If there are more than two classes, binary logistic regression can be extended to polytomous logistic regression (Hosmer and Lemeshow, 1989; Agresti, 2002). The major difference is that now errors are assumed to be distributed multinomially rather than binomially. Classiﬁcation among J stocks requires J - 1 link functions, which is no different from the binary case. For stock membership j = 1, . . . , J, P(Y i = j | x i ) = p j (x i ) =

exp(z ij ) Â hJ = 1 exp(z ih )

(6)

where zij are analogous to eq. 4 but with parameters b0j, b1j, . . . , bNj. The problem is constrained by the requirement that the probabilities of stock membership sum

506

Michael H. Prager and Kyle W. Shertzer

to one. Given J - 1 estimates of pj, the Jth estimate must be pJ(xi) = 1 - SJ-1 j=1pj(xi). That constraint is implemented by deﬁning parameters for the Jth stock to be zero (i.e., b0J, b1J, . . . , bNJ = 0). As a result, eq. 6 simpliﬁes because exp(ziJ) = 1, and the pj’s can be estimated under transformation by a link function. As in the binary case, there are J - 1 unique logit equations, p j (x i ) ˆ = b 0 j + b1 j x i1 + b 2 j x i2 + . . . + b Njx iN log Ê Ë p J (x i ) ¯

(7)

Computer programs for logistic regression are readily available in the major statistics software packages. In using such software, the analyst usually wishes to specify that the stock designations are nominal rather than ordinal values. Careful reading of the software’s documentation may be needed to effect that choice, which is not always the default.

C. ARTIFICIAL NEURAL NETWORKS The term artiﬁcial neural network (ANN) is not a precise one, but refers to a group of computational algorithms that sift through and combine many models to arrive at a model of optimum (in some sense) complexity (Ripley, 1996; Hastie et al., 2001). Unlike the other classiﬁcation schemes presented here, ANNs are nonparametric. They require no assumptions about the distribution of data nor the particular functional relationship between model input and output. This can be a major advantage when such assumptions would be violated. Nonetheless, the success of such methods still depends on similarity of the data in the mixed-stock sample to the data in the training set. Artiﬁcial neural networks have been developed in analogy to the structure of the human brain. Like the brain, an ANN consists of interconnected layers that process information provided by neurons. The input layer performs computations on the input data, and the results, along with a constant (bias), are then passed to a hidden layer. That procedure is iterated among a series of hidden layers until ﬁnally results are passed to the output layer. The number of hidden layers and the number of neurons in each can be adjusted to reﬂect the complexity of the problem. Once the architecture is established, values of network parameters are chosen as those that minimize some ﬁtting criterion, a task usually accomplished with a learning algorithm. Neural networks have proved useful in numerous ﬁelds, such as artiﬁcial intelligence, image compression, medical diagnosis, nondestructive testing, signal processing, and terrain classiﬁcation, to name only a few. To our knowledge, the ﬁrst published application in stock composition analysis was by Prager (1984, 1988) to estimate stock composition of striped bass and American shad. That study used

An Introduction to Statistical Algorithms Useful in Stock Composition Analysis

Input layer

Hidden layer

507

Output layer

wab

wbk

X1

X2

X3

Bias bb

Bias bk

FIGURE 24-3. Example of an artiﬁcial neural network with a single hidden layer. The input variables (Xa) are the measure characteristics. Their weighted (wab) values plus a bias term (bb) are processed by the hidden layer, and in turn, those results are processed similarly by the output layer. The ﬁnal result is probabilities of stock membership, among two stocks in this example.

the group method of data handling (Ivakhnenko and Ivakhnenko, 1974), a type of neural network based on polynomials. More recent studies have compared the performances of ANNs and LDA and have found ANNs to be slightly more successful, at least on the particular data sets analyzed (Taylor and Beacham, 1994; Thorrold et al., 1998; Wells et al., 2000). In the context of stock composition analysis, the input variables for an ANN are the measured characteristics; the output is the stock classiﬁcation. Figure 24-3 illustrates a neural network with a single hidden layer that classiﬁes a sample into two component stocks based on three characteristics. Each neuron in the hidden layer sums the weighted (wab) input signals (X1, X2, X3) and adds a bias term (bb). The result is then processed by a hidden-layer function (fH) to produce an input signal for the output layer. The procedure is the same at the output layer, but with an output-layer function (fO). Figure 24-3 can be expressed mathematically as Ê Ê ˆˆ Y k = fO b k + Â w bk f H bb + Â wab X a Ë Ë ¯¯ b

a

(8)

508

Michael H. Prager and Kyle W. Shertzer

where Yk is the probability of membership in stock k. Any monotonic smooth function can be used for fH or fO, and they need not be the same, but the most popular by far is the logistic function of eq. 3. Because of the complexity and specialized nature of the calculations, application of neural networks is generally made with specialized software. Because the methods are not standardized, different programs may offer different procedures and different results. For that reason, it may not be possible to duplicate existing results unless the same software is used, a situation that differs markedly from the other methods described here.

D. FINITE MIXTURE DISTRIBUTION METHODS Discriminant analysis, logistic regression, and neural networks can be considered classiﬁcation-based algorithms because their focus is on an estimated classiﬁcation of each individual (to stock), at least probabilistically. Those methods estimate, for each individual, the probability of membership in each class, and the estimates of composition—the desired results of ﬁsh stock composition analysis—are derived from the estimated membership probabilities of the individuals. The ﬁnal set of algorithms discussed here, methods based on ﬁnite mixture distributions (FMD), does not share that focus on individuals. (Here ﬁnite refers to the number of classes in the mixture.) Although the probability of an individual’s group membership can be estimated from ﬁnite mixture methods, the primary focus is on estimating the composition of a mixed sample (mixture distribution). This difference in focus is important. Maximum-likelihood estimation of ﬁnite mixture distributions has been the subject of several books in the statistical literature (Wolfe, 1970; Everitt and Hand, 1981). The methods were introduced to the ﬁsheries literature to separate size compositions into age classes (Cassie, 1954; Bhattacharya, 1967). Application to stock composition analysis came later (Fournier et al., 1984; Pella and Milner, 1986; Millar, 1987; Wood et al., 1987). The methods are simplest to apply if one assumes that characteristics follow a multivariate normal distribution with a common covariance matrix among classes, the same assumption used in linear discriminant analysis. However, FMD methods can be adapted to a wide variety of distributions and can accommodate unequal covariance matrices (e.g., DeVries et al., 2002). A ﬁnite mixture distribution (f) describes the distribution of a vector (x) of N observed characteristics. The mixture distribution f can be expressed as a weighted sum of its J component probability distributions gj, j = 1, . . . , J (where as before J is the number of stocks). For example, a mixture of multivariate normal distributions with unequal variances can be written,

509

An Introduction to Statistical Algorithms Useful in Stock Composition Analysis J

f (x | p, m, Â) =

Â p j g j (x | m j , Â j )

(9)

j= 1

Here p = (p1, p2, . . . , pJ-1) are the J - 1 independent mixing proportions, or weights, of the component distributions such that 0 < pj < 1 and pJ = 1 - SJ-1 j=1pj. The mj and Sj are the mean vector and variance–covariance matrix of the jth component distribution. In a stock composition analysis, the mixed-stock sample is from the mixture distribution (f) of characteristics. The goal is to estimate mixing proportions (pj); to do so, one must ﬁrst estimate (from the training set) the parameters of the component distribution ( gj) of each constituent stock. The mixture distribution observed in the mixed-stock sample depends both on the parameters (shapes) of the stock-speciﬁc component distributions and on the mixing proportions in the mixed-stock sample (Fig. 24-4).

0.6

f(x)

0.4 0.2 0.0 –4

–2

0

2

4

2

4

x 0.6

f(x)

0.4 0.2 0.0 –4

–2

0 x

FIGURE 24-4. Mixture distributions with the same component distributions in different proportions. Each mixture is of two univariate normal distributions: stock one with m1 = -1.0, s1 = 1.0, and stock two with m2 = 1.5, s2 = 0.5. The mixing proportions are p1 = 0.25, p2 = 0.75 in the top panel, and p1 = 0.75, p2 = 0.25 in the bottom panel.

510

Michael H. Prager and Kyle W. Shertzer

Practical Application of FMD Methods Software for mixture problems seems less readily available than software for ﬁtting discriminant functions, logistic regressions, and even neural networks, a situation that may have slowed the wide adoption of FMD methods in ﬁsheries. However, R. Millar has developed Fortran software, HISEA, that, as of the date of writing, was available on the World Wide Web. (http://www.stat.auckland.ac.nz/~millar/mixedstock/code.html). Millar’s software implements the FMD method under the assumption of multivariate normality with constant variance. Other suitable software (e.g., EMMIX by G. McLachlan) may also be freely available. It may not be obvious that commercial software for discriminant analysis can also be used to obtain maximum-likelihood FMD estimates of stock composition. Thus, any analyst with access to standard statistical packages can explore the properties of FMD estimates. To proceed in this way, it is simplest to assume that the measured characteristics are multivariate normal with a common covariance matrix across stocks. Under that assumption, the equations for LDA and FMD are identical, with the mixing proportions to be estimated by FMD corresponding to the the correct (but unknown) priors in LDA. By using an EM algorithm (Dempster et al., 1977), maximum-likelihood estimates of the mixing proportions can be obtained under the FMD model. As noted by Millar (1987), “. . . in constructing a classiﬁcation rule, one is actually doing all of the work required to construct the likelihood function, so from there it would be a matter of simply running a maximization program to obtain the maximum likelihood estimates.” The procedure is as follows: 1. Fit a linear discriminant function to the training sample. 2. Obtain a starting guess for the priors. In the absence of other information, one can use equal priors, that is, set pj = 1/J for all j. 3. Using the current priors and the discriminant function ﬁt in step 1, make a nondiscrete estimate of the mixture proportions of the mixed-stock sample. 4. Revise the priors to equal the current estimated mixture proportions. 5. Repeat steps 3 and 4 until the composition estimates converge. Without doubt, this procedure is more tedious than using software written speciﬁcally for the estimation of mixing proportions. However, if such software is not readily available, or if an investigator wishes to take the ﬁrst steps into using FMD methods, this iterative procedure may prove useful.

An Introduction to Statistical Algorithms Useful in Stock Composition Analysis

511

III. THE IMPORTANCE OF PRIOR KNOWLEDGE Each classiﬁcation-based method described above (discriminant analysis, logistic regression, neural networks) either makes implicit use of properties of the training sample or requires additional assumptions to estimate stock composition. In the case of discriminant analysis, the composition estimates are conditional on the prior probabilities of stock membership (the pj in eq. 1) speciﬁed by the analyst. Perhaps because standard software offers defaults for these priors, their importance is often overlooked. Nonetheless, the reliance on priors—or equivalent information from the training sample—is a major limitation of the classiﬁcation-based methods. For that reason, we emphasize here the role of priors in discriminant analysis and the role of analogous information in other classiﬁcation algorithms.

A. PRIORS

AND

DISCRIMINANT ANALYSIS

Priors, as used by discriminant analysis, are a priori estimates of the probability that an individual in the mixed-stock sample is a member of each component stock. By a priori, we mean that the individual is chosen at random from the mixed sample and that nothing further is known about it: its characteristics have not yet been observed. If we assume that the stocks are present in the mixedstock sample in proportion to their presence in the mixture under consideration (except for sampling error), the paradox involved in using discriminant analysis for this problem becomes clear. The priors, which are required to make an estimate, are precisely what we are trying to estimate: the relative contribution of stocks in the mixture. This paradox does not occur in some other ﬁelds that use classiﬁcation methods because the structure of their questions is fundamentally different. For example, a typical medical application might be to estimate the probability of a patient’s contracting some disease, given observations about general health and family history. In that example, the proportion of the general population that will contract the disease is well known and can be used as the prior. The focus of such a study is estimation about the individual. In such applications, reliable priors are readily available, and classiﬁcation methods are appropriate. Notably, their use does not entail the same paradox as in stock composition studies. In discriminant analyses, priors are speciﬁed explicitly, and composition estimates cannot be made without them. Statistics packages commonly offer two simple tactics for generating priors. The ﬁrst tactic is to base priors on the proportions observed in the training set. Under that approach, the prior for a given stock is set equal to its relative predominance in the training sample. The unstated

512

Michael H. Prager and Kyle W. Shertzer

assumption is that the composition of the training sample is a good estimate of the composition of the mixed population. The second tactic is to set the priors equal: with three stocks, the prior for each would be set to 1/3. Neither tactic has a valid theoretical basis, and neither is particularly likely to result in accurate priors. When priors are inaccurate, resulting estimates of mixture proportions are biased toward the priors, although correction matrices may reduce such biases. Frequently, studies that use discriminant analysis provide estimates of classiﬁcation error rates. In interpreting such estimates, it is important to understand that the error rates of an uncorrected discriminant estimator depend on the actual (and unknown) mixed-stock composition. Such error rates are generally smallest when the priors are accurate (i.e., when they correspond to the actual stock composition) and can be much higher when the priors are inaccurate. Reported error rates can also be biased low if the training-set data have been used to estimate the error rates.

B. PRIORS

AND

LOGISTIC REGRESSION

In logistic regression, explicit priors are not speciﬁed by the analyst. However, a logistic regression estimator is derived from the distributions observed in the training sample. It therefore provides the least biased estimates when the mixedclass sample is of the same composition as the training sample. To translate the preceding into ﬁsheries language, a logistic-regression estimator will be biased toward the stock composition of its training sample and will be increasingly inaccurate as the composition of the mixed-stock sample differs from that of the training sample. The use of correction matrices to reduce bias in mixing estimates from logistic regression seems feasible. However, we do not know that the speciﬁc subject has been studied. Unfortunately, the bias just described was overlooked by Prager and Fabrizio (1990). When an adjustment was made to their experimental procedure to compensate, the same authors found the performance of logistic regression to be similar to that of linear discriminant analysis, even on nonnormal data (M. H. Prager, C. D. Jones, and M. C. Fabrizio, unpublished study).

C. PRIORS

AND

NEURAL NETWORKS

Estimation of stock composition by neural networks is also conditional on the composition of the training sample. Therefore, bias of these estimators should be similar to that from logistic regression. The ultimate performance of such a method also depends on the particular data and speciﬁc method involved. As

An Introduction to Statistical Algorithms Useful in Stock Composition Analysis

513

with logistic regression, it seems that the use of correction matrices with neural networks has not yet been studied, at least in ﬁshery science.

D. PRIORS

AND

FMD METHODS

Finite mixture distribution methods do not require priors, implicit or explicit. It stands to reason that, in most cases, they should produce estimates with lower bias than uncorrected classiﬁcation-based algorithms. This may be accompanied by increased variance.

IV. DISCUSSION The four methods presented here are familiar techniques in stock composition analysis. However, other classiﬁcation methods exist and may hold added promise for ﬁsheries. One nonparametric technique is tree-based regression (often called CART), applied to stock classiﬁcation by Weigel et al. (2002). A quite different approach is analysis of tagging data for stock composition (Schwartz, this volume, Chapter 28). Evaluation of statistical algorithms for stock composition analysis is not a simple task, whether undertaken on simulated or real data. What has been underemphasized in some such evaluations is how an estimator performs as the composition of the mixed-class sample varies from that of the training sample. One would expect the error of uncorrected discriminant analyses to become considerably worse as the true composition varies from the priors. A similar deterioration in performance of uncorrected logistic regression or neural network estimates will occur as the true composition varies from that of the training sample. The error rate of FMD methods may deteriorate as the composition of the mixed-stock sample becomes quite different from that of the training sample, but variation in the composition of the mixed-stock sample does not constitute violation of a major assumption, as it does with the classiﬁcation-based methods. The analyst should always be aware of the chosen algorithm’s assumptions and the degree to which they are met. Gray (1994) showed that in FMD applications, model misspeciﬁcation can lead to poor estimation. For example, if unequal variances are assumed equal, or if skewed distributions are assumed normal, FMD results may be strongly biased (Gray, 1994). Similar problems can arise with any algorithm if assumptions are violated. Moreover, bias and variance are statistical properties, and there is no assurance that a speciﬁc algorithm will be more accurate or precise than another in a particular application. To some degree, properties of algorithms in speciﬁc applications can be examined through Monte Carlo simulation studies.

514

Michael H. Prager and Kyle W. Shertzer

From the preceding, it can be seen that for the stock composition problem in ﬁsheries, uncorrected methods appear least desirable, and FMD methods appear more appropriate than methods based on classiﬁcation, provided assumptions are met. Millar (1990) concluded that corrected classiﬁcation estimators are as useful as FMD methods when the number of stocks is small (two or three), but recommended use of direct maximum-likelihood estimation (FMD methods) for more complex problems. We concur with that recommendation, and we further recommend that when error rates are estimated, they should be reported based on a range of priors (or their equivalent), if used, and over a range of (simulated) stock compositions.

ACKNOWLEDGMENTS Correspondence and conversations with J. Pella helped to clarify the relationship between discriminant analysis and the ﬁnite mixture problem. Insight was also gained during an unpublished simulation study conducted with M. Fabrizio and C. D. Jones. We thank R. Millar for sharing his HISEA software, and we thank D. Ahrenholz, P. Hanson, and J. Waters for reviewing the manuscript. This work was supported by the Virginia Sea Grant Program and the Southeast Fisheries Science Center of the U.S. National Marine Fisheries Service through the NOAA Center for Coastal Habitat and Fisheries Research.

REFERENCES Agresti, A. 2002. Categorical Data Analysis, 2nd Ed. Wiley, New York. 734 pp. Aldrich, J. H. and Nelson, F. D. 1984. Linear probability, logit, and probit models. Sage University Paper series on Quantitative Applications in the Social Sciences, 07–045. Sage Publications, Beverly Hills, CA. Begg, G. A. and Waldman, J. R. 1999. An holistic approach to ﬁsh stock identiﬁcation. Fish. Res. 43: 35–44. Bhattacharya, C. G. 1967. A simple method of resolution of a distribution into Gaussian components. Biometrics 23: 115–135. Cassie, R. M. 1954. Some uses of probability paper in the analysis of size frequency distributions. Austral. J. Mar. Freshw. Res. 5: 513–522. Cook, R. C. and Lord, G. E. 1978. Identiﬁcation of stocks of Bristol Bay sockeye salmon, Oncorynchus nerka, by evaluating scale patterns with a polynomial discriminant method. Fish. Bull. 76: 415–423. Dempster, A. P., Laird, N. M., and Rubin, D. B. 1977. Maximum likelihood from incomplete data via the EM algorithm. J. Royal Statist. Soc., Series B 39: 1–38. DeVries, D. A., Grimes, C. B., and Prager, M. H. 2002. Using otolith shape analysis to distinguish eastern Gulf of Mexico and Atlantic Ocean stocks of king mackeral. Fish. Res. 57: 51–62. Everitt, B. S. and Hand, D. J. 1981. Finite Mixture Distributions. Chapman & Hall, London. 143 pp. Fabrizio, M. C. 1987. Growth-invariant discrimination and classiﬁcation of striped bass stocks by morphometric and electrophoretic methods. Trans. Am. Fish. Soc. 116: 728–736.

An Introduction to Statistical Algorithms Useful in Stock Composition Analysis

515

Fournier, D. A., Beacham, T. D., Riddell, B. E., and Busack, C. A. 1984. Estimating stock composition in mixed stock ﬁsheries using morphometric, meristic, and electrophoretic characteristics. Can. J. Fish. Aquat. Sci. 41: 400–408. Gray, G. 1994. Bias in misspeciﬁed mixtures. Biometrics 50: 457–470. Hastie, T., Tibshirani, R., and Friedman, J. 2001. The Elements of Statistical Learning. Springer, New York. 552 pp. Hill, D. R. 1959. Some uses of statistical analysis in classifying races of American shad (Alosa sapidissima). Fish. Bull. 59: 269–286. Hosmer, D. H. and Lemeshow, S. 1989. Applied Logistic Regression. Wiley, New York. 307 pp. Ivakhnenko, A. G. and Ivakhnenko, N. A. 1974. Long-term prediction of random processes by GMDH algorithms using the unbiasedness criterion and balance-of-variables criterion. Sov. Autom. Control 7(4): 40–45. Johnson, R. A. and Wichern, D. W. 1998. Applied Multivariate Statistical Analysis, 4th Ed. Prentice Hall, Upper Saddle River. 816 pp. Kendall, M., Stuart, A., and Ord, J. K. 1983. The Advanced Theory of Statistics, Vol. 3, 4th Ed. Charles Grifﬁn, London. 780 pp. McCullagh, P. and Nelder, J. A. 1989. Generalized Linear Models, 2nd Ed. Chapman & Hall, London. 532 pp. McLachlan, G. J. 1992. Discriminant Analysis and Statistical Pattern Recognition. Wiley, New York. 544 pp. Millar, R. B. 1987. Maximum likelihood estimation of mixed stock ﬁshery composition. Can. J. Fish. Aquat. Sci. 44: 583–590. Millar, R. B. 1990. Comparison of methods for estimating mixed stock ﬁshery composition. Can. J. Fish. Aquat. Sci. 47: 2235–2241. Misra, R. K. 1985. Quadratic discriminant analysis with covariance for stock delineation and population differentiation: a study of beaked redﬁshes (Sebastes mentella and S. fasciatus). Can. J. Fish. Aquat. Sci. 42: 1672–1676. Palma, J. and Andrade, J. P. 2002. Morphological study of Diplodus sargus, Diplodus puntazzo, and Lithognathus mormyrus (Sparidae) in the Eastern Atlantic and Mediterranean Sea. Fish. Res. 57: 1–8. Pella, J. J. and Milner, G. B. 1986. Use of genetic marks in stock composition analysis. In N. Ryman and F. Utter (eds.), Population Genetics and Fishery Management. University of Washington Press, Seattle, pp. 247–276. Pella, J. J. and Robertson, T. L. 1978. Assessment of composition of stock mixtures. Fish. Bull. 77: 387–398. Prager, M. H. 1984. The Group Method of Data Handling: Applications in Oceanography and Fishery Science. Ph.D. dissertation, University of Rhode Island, Kingston. Prager, M. H. 1988. Group method of data handling: a new method for stock identiﬁcation. Trans. Am. Fish. Soc. 117: 290–296. Prager, M. H. and Fabrizio, M. C. 1990. Comparison of logistic regression and discriminant analyses for stock identiﬁcation of anadromous ﬁsh, with application to striped bass (Morone saxatilis) and American shad (Alosa sapidissima). Can. J. Fish. Aquat. Sci. 47: 1570–1577. Ripley, B. D. 1996. Pattern Recognition and Neural Networks. Cambridge University Press, Cambridge. 403 pp. Small, M. P., Withler, R. E., and Beacham, T. D. 1998. Population structure and stock identiﬁcation of British Columbia coho salmon, Oncorhynchus kisutch, based on microsatellite DNA variation. Fish. Bull. 96: 843–858. Taylor, E. B. and Beacham, T. D. 1994. Population structure and identiﬁcation of North Paciﬁc Ocean chum salmon (Oncorhynchus keta) revealed by an analysis of minisatellite DNA variation. Can. J. Fish. Aquat. Sci. 51: 1430–1442.

516

Michael H. Prager and Kyle W. Shertzer

Thorrold, S. R., Jones, C. M., Swart, P. K., and Targett, T. E. 1998. Accurate classiﬁcation of juvenile weakﬁsh Cynoscion regalis to estuarine nursery areas based on chemical signatures in otoliths. Mar. Ecol. Prog. Ser. 173: 253–265. Waldman, J. R., Richards, R. A., Schill, W. B., Wirgin, I., and Fabrizio, M. C. 1997. An empirical comparison of stock identiﬁcation techniques applied to striped bass. Trans. Am. Fish. Soc. 126: 369–385. Weigel, D. E., Peterson, J. T., and Spruell, P. 2002. A model using phenotypic characteristics to detect introgressive hybridization in wild westslope cutthroat trout and rainbow trout. Trans. Am. Fish. Soc. 131: 389–403. Wells, B. K., Thorrold, S. R., and Jones, C. M. 2000. Geographic variation in trace element composition of juvenile weakﬁsh scales. Trans. Am. Fish. Soc. 129: 889–900. Wolfe, J. H. 1970. Pattern clustering by multivariate mixture analysis. Multivar. Behav. Res. 5: 389–398. Wood, C. C., McKinnell, S., Mulligan, T. J., and Fournier, D. A. 1987. Stock identiﬁcation with the maximum-likelihood mixture model: sensitivity analysis and application to complex problems. Can. J. Fish. Aquat. Sci. 44: 866–881.

CHAPTER

25

Classical Discriminant Analysis, Classiﬁcation of Individuals, and Source Population Composition of Mixtures JEROME PELLA AND MICHELE MASUDA United States Department of Commerce, National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Auke Bay, Alaska, USA

I. Introduction II. Statistical Theory III. Classiﬁcation of Mixture Individuals by Their Posterior Probabilities IV. Estimation of Posterior Probabilities A. Classiﬁcation-Based vs. Direct Estimation for Source Prior B. Conditional vs. Unconditional Estimation of Source Measurement Distributions C. Plug-in vs. Predictive Classiﬁcation V. Classiﬁcation-Based Conditional Estimation A. Models for Source Measurements, Discriminant Functions, and Their Conditional Estimation B. Maximum-Likelihood Estimation of Source Prior VI. Direct Estimation A. Conditional Estimation of Source Measurement Distributions B. Unconditional Estimation of Source Measurement Distributions VII. Applications to the Tree Point Sockeye Salmon Blind Mixture Samples Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

517

518

Jerome Pella and Michele Masuda

VIII. Summary Appendix References

I. INTRODUCTION The idea of stocks, or populations, is central to ﬁsheries management. Whenever possible, separate populations should be accommodated in the management plans of ﬁsheries that harvest them individually or in mixtures. If their individuals occur in mixtures, estimating the source population proportions composing the mixtures is invaluable for assessing the separate population sizes and yields. Additionally, identifying, as best possible, the source populations of sampled mixture individuals can be useful to research and forensic issues. Multivariate measurements from individuals, having possibly categorical, discrete, and continuous components, are to be the basis of estimation or identiﬁcation. Although the special case of unordered categorical measurements from genotypic assays is excluded as it is considered in other chapters, the estimation methods are easily adapted to include it. If the measurement distributions were distinctive of the populations, source identiﬁcation for individuals would be certain, and the sampling theory for multinomial proportions and percentages (Cochran, 1963) would sufﬁce for composition estimation of population mixtures. Typically, the measurement distributions of the source populations overlap, and so source identiﬁcation of mixture individuals includes this uncertainty in addition to the multinomial sampling variation. When measurements of individuals from the separate populations and the mixture are available, discriminant and classiﬁcation analysis is a methodology often used for assigning the population sources of mixture individuals. Discriminant analysis comprises the deriving of classiﬁcation rules, collectively called the classiﬁer, from the measurements of individuals in so-called learning, training, or baseline samples that are obtained from each of the possible separate source populations. Based on the measurements of a mixture individual whose source population is unknown, the classiﬁer assigns it to one population of those considered as possible candidates. Classiﬁcation analysis will refer to applying the classiﬁer to the measurements on each individual from the mixture sample. Traditional statistical methods for developing classiﬁcation rules commonly deﬁne a discriminant function of an individual’s measurements for each source population, and the classiﬁer identiﬁes the source population as that for which the discriminant function value is greatest. The discriminant function that is most relevant for population mixture analysis is the posterior probability of an individual’s source population, given its measurements, or for brevity, its posterior source probability. As will be seen, monotone, order-preserving transformations

Classical Discriminant Analysis

519

of these posterior source probabilities also produce useful forms for a discriminant function. Simplest among mathematical models for the posterior source probabilities, or their monotone transforms, are linear functions of the individual’s measurements. Among classiﬁcation methods using linear discriminant functions, two are of note: (1) linear discriminant analysis (LDA) that assumes multivariate normal distributions for measurements with means varying among the source populations, but with common covariance matrix, and (2) logistic discriminant analysis (LGA) that makes no distributional assumptions, but models the logarithms for posterior source odds, a monotone transform of the odds, as a linear function of the measurements. Quadratic discriminant analysis (QDA) is a generalization of LDA which allows for the source population covariance matrices, as well as their means, to differ. In QDA, the discriminant functions are quadratic functions of the individual’s measurements. With the recent availability of ample computing power, many new and general approaches to discriminant and classiﬁcation analyses have been developed, and these offer promise for improvements to population source identiﬁcation by the classical methods—LDA, QDA, and LGA—to be described. For example, mixture discriminant analysis (MDA) (Hastie and Tibshirani, 1996) is an extension of LDA in which the measurement distributions for each source population are assumed to be a mixture of multivariate normal components with common covariance matrix and different means. Therefore, MDA can be applied in situations where the measurements within populations have complex, even multimodal distributions. Several of the newer techniques, for example, neural networks and nonparametric methods, are described in other chapters. Earliest ﬁsheries applications used LDA with morphological characters of ﬁsh and growth patterns of their scales. Hill (1959) separated American shad (Alosa sapidissima) populations, and Fukuhara et al. (1962), Amos et al. (1963), Dark and Landrum (1964), Anas (1964), and Mason (1966) identiﬁed continent of origin of Paciﬁc salmon from the high seas. Later, Anas and Murai (1969) compared LDA and QDA with scale features of Paciﬁc salmon. Cook and Lord (1978) appear to have been ﬁrst in ﬁsheries to explore alternatives to LDA and QDA when they applied what seems to have been a kernal-based discriminant analysis to sockeye salmon (Oncorhynchus nerka) using a polynomial function of scale measurements. Prager and Fabrizio (1990) appear ﬁrst to have promoted LGA for ﬁsheries applications, nearly two decades after an earlier proponent (Anderson, 1972, 1974, 1979, 1982) described it. Even today, adoption of LGA has been slow to develop despite its advantages. LDA predominates in current ﬁsheries applications because its software is widely available and generally performs well, and so it remains a standard by which other methods are compared. The long-standing application of LDA by the Paciﬁc Salmon Commission in its regular scale pattern analyses of annual Fraser River sockeye salmon returns attests to its trustworthiness (Cook and Guthrie, 1987; Gable and Cox-Rogers,

520

Jerome Pella and Michele Masuda

1993). QDA is less used than LDA, probably because gains in performance are often marginal against LDA, and LGA is used even less, probably because it is less familiar and its software is less developed. Most, but not all, ﬁsheries applications of discriminant and classiﬁcation analysis are concerned with estimation of source population proportions composing a mixture, rather than the identiﬁcation of the source population of particular individuals. Mistakes during classiﬁcation, or misclassiﬁcations, are well known to lead to biased apparent mixture composition estimates (Worlund and Fredin, 1962; Cook and Lord, 1978, and others), and so the numbers of mixture individuals classiﬁed to each of the possible source populations usually require further analysis to remove such bias. The bias is largely caused by incorrect values for the so-called prior probabilities, hereafter termed the prior source probabilities, that are needed for classiﬁcation analysis using the discriminant functions. The correct prior source probabilities would be the source population proportions composing the mixture. Unfortunately, these source proportions are unknown. In this regard, ﬁsheries applications differ importantly from many others, where primary interest is in classiﬁcation, and knowledge of the prior probabilities is already available. Medical diagnoses provide examples in which reasonably well-known diseased proportions of the population constitute the essential prior probabilities by which to assess the probability of diseases in individual patients. The difference in goals and assumed knowledge must not be overlooked when utilizing methods and software developed outside ﬁsheries, whether the purpose is to estimate the source population composition of a mixture or to classify the individuals to their sources. In particular, all software for discriminant analysis with classiﬁcation of individuals requires values for the prior source probabilities, and these values may be directly requested of the user, indirectly (and usually inappropriately) computed from other available information such as the learning sample sizes, or implicitly (and usually inappropriately) assumed to be equal. Unless the source population measurement distributions are distinct, classiﬁcation rules with fewest mistakes generally depend critically on knowing something about the prior source population probabilities, that is, the unknown source population proportions composing the mixture. For example, mixture individuals may be assigned to a population despite its unexpected absence from the mixture unless the prior reﬂects it is missing. Obviously, the classiﬁcation of the mixture sample individuals provides information about the source population proportions. Regrettably, a circularity is evident: The rules ought to depend on the outcome of the classiﬁcation, which depends on the rules. Initially, a simple adjustment or correction for the bias in apparent source composition from misclassiﬁcations was derived (Worlund and Fredin, 1962; Cook and Lord, 1978). This correction depended on performance of the classiﬁer when applied to known individuals of each of the sources. Although precision of these corrected source

Classical Discriminant Analysis

521

composition estimates (Pella and Robertson, 1979) could be assessed, the method could result in infeasible, negative estimates for source proportions. Later, Millar (1987) and Wood et al. (1987) recognized that feasible corrected source composition estimates were the maximizing solution of a likelihood function involving the counts assigned to the sources from the mixture and learning samples, and that a technically sound solution to estimation of feasible source proportions would always be obtained by constrained maximization of this likelihood function. The conditional solution implemented by Fournier et al. (1984), Millar (1987), and Wood et al. (1987) used only the learning samples for estimation of the source measurement distributions, and did not make use of additional information they recognized present in the mixture sample. Because this omission can be important when learning sample sizes are limiting, we describe algorithms by which this information can be extracted simply. This introductory sketch of discriminant and classiﬁcation analyses and statistical methods for integrating mixture information about the missing source prior is expanded next. In particular, classical methods using discriminant functions, including LDA, LGA, and QDA are detailed. Much attention is devoted to estimation of the source proportions in the mixture, which are essential for use of the discriminant functions. The methods will be illustrated with subsets of sockeye salmon scale pattern data collected in 2002 for a Paciﬁc Salmon Treaty experiment regarding the Tree Point ﬁshery in Southeast Alaska. The Tree Point ﬁshery, near the international border, catches sockeye salmon bound for spawning grounds of three population groups: Canadian (1) Nass River and (2) Skeena River, and (3) southern Southeast Alaska. Under the most recent treaty annex, the Tree Point ﬁshery is constrained in its allowable annual catch of Nass River sockeye salmon. Although the total sockeye salmon catch is known, that from the Nass River can only be estimated from measurements of sampled individuals. Geographic differences in spawning grounds make for variation in scale patterns among populations that have been used since the 1980s for identiﬁcation of their members in mixed catches (Fig. 25-1). However, a switch to newer genetic technology, which can resolve greater population detail than scale patterns, is contemplated if independent assessments of the group contributions with the two sets of characters and analytical methods are either consistent or the disagreements are resolved in favor of the proposed genetic microsatellites. This initial experiment included matched scale and tissue sampling from individuals of the three stock groups and the creation of their blind mixtures: Research interest included both the estimation of source population composition of the blind mixtures as well as the source identiﬁcation of the comprising individuals. Individual identities were revealed after the determinations were submitted by the independent scale and genetics laboratories. The blind to the source populations of the mixture individuals was removed before performance of the scale pattern analyses reported in this chapter.

522

Jerome Pella and Michele Masuda

(a) FIGURE 25-1. Histograms of four scale measurements from age 1.2 sockeye salmon selected by stepwise discriminant analysis for classiﬁcation of individuals from the Nass River and Skeena River in Canada, and from Southeast Alaska. Vertical lines represent mixture individuals that were incorrectly assigned despite posterior source probabilities in excess of 0.8.

II. STATISTICAL THEORY A mixture of ﬁsh is composed of several populations, say c, and a researcher wishes to determine either the population composition of the mixture, the source populations of individual ﬁsh from the mixture, or both. A sample of ﬁsh is taken from the mixture, and a number, say d, of measurements, denoted by the vector X, are made for each individual. Unless the populations are unique in their measurements, the identiﬁcation of sources for individuals will be uncertain. Usually, the measurements are quite variable within populations and the source populations’ measurement distributions overlap. Even if the measurements were unique among populations, the population composition would be uncertain because of multinomial sampling variation in the source population proportions. Therefore, the researcher will need to use probabilistic methods in order to effectively

Classical Discriminant Analysis

523

FIGURE 25-1. Continued

achieve the goals and describe the uncertainty in assignments and source composition. If a complete list of the possible populations composing the mixture is available and separate samples of ﬁsh and their measurements have been obtained from each population, the methodology is called supervised classiﬁcation with learning samples. If only a sample from the mixture is available, the methodology is called unsupervised classiﬁcation or cluster analysis. We consider only supervised classiﬁcation for the purpose of estimating population proportions and assigning individuals to their sources, and direct the reader concerned with cluster analysis to general coverage in Seber (1984) and the recent review of model-based methods by Fraley and Raftery (2002).

III. CLASSIFICATION OF MIXTURE INDIVIDUALS BY THEIR POSTERIOR PROBABILITIES The individuals from the c source populations form a mixture composed of proportions p = (p1, p2, . . . , pc)¢. These proportions represent the source

524

Jerome Pella and Michele Masuda

probabilities of a random individual sampled from the mixture before observing its measurements, and collectively are called the source prior. The assumption is implicit that the individual’s measurements do not inﬂuence the probability that it is drawn. The population-speciﬁc probability distributions of measurements from the separate populations, or source measurement distributions, are denoted as fi(X), i = 1, 2, . . . , c. These functions effectively describe the relative frequencies of individuals with measurement X in the separate populations. If the source measurement distributions were known, for example, if very large learning samples were available for the sources, Bayes’ theorem of mathematical statistics provides the posterior source probabilities of an individual with measurements X: P(i | X ) =

p i f i (X ) c

i = 1, 2, . . . , c

(1)

Â p k f k (X ) k =1

Essentially, the posterior source probabilities are the proportions of the mixture individuals with measurement X that come from each source population, and these proportions clearly deﬁne and quantify our uncertainty regarding an individual’s source based on its measurements. Notice that the posterior source probability of eq. 1 is unchanged by dividing the numerator and denominator by fr(X), the measurement distribution value of the rth source chosen as a reference. Bayes’ theorem requires both the source prior and either the source measurement distributions, fi(X), i = 1, 2, . . . , c, or else the ratios, fi(X)/fr(X), i = 1, 2, . . . , c. If the costs of misclassiﬁcations are equal for the sources, as seems reasonable for ﬁshery applications, the expected misclassiﬁcation rate is minimized by assigning the individual to the population i* for which the posterior probability is greatest: P(i*| X ) = max i{P(i | X ), i = 1, 2, . . . , c} This rule, which assigns the individual to the maximum a posteriori (MAP) source, is called the Bayes classiﬁer. The Bayes classiﬁer is biased in that individuals with measurement X are never classiﬁed to their other possible source populations. An alternative unbiased classiﬁer with greater expected misclassiﬁcation rate is stochastic, and assigns the source of each mixture individual by a random draw with probabilities equal to the posterior source probabilities of eq. 1. The stochastic classiﬁer is used to advantage in the stochastic expectationmaximization (SEM) algorithm (Diebolt and Ip, 1996) by iteratively generating the sources of the mixture individuals, which can be treated as known within a repetition, and used in very simple estimation of the source proportions and source measurement distributions.

525

Classical Discriminant Analysis

If no information were available by which to specify the source prior, the subjective, equiprobable prior, p = (1/c, . . . , 1/c), is inevitable. With the equiprobable prior, the prior probabilities cancel from eq. 1, and the mixture individuals are assigned to the source i* for which the measurement is most frequent: f i* ◊(X ) = max i { f i (X ), i = 1, 2, . . . , c} Wood et al. (1987) termed this classiﬁer as the dominant likelihood rule, and its common use reﬂects more on its convenience than its performance. The equiprobable prior will often be seen to have been doubtful after the mixture individuals have been assigned to their sources by the dominant likelihood rule: The apparent source composition from the assignments will deviate from the approximately equal proportions expected.

IV. ESTIMATION OF POSTERIOR PROBABILITIES

A. CLASSIFICATION-BASED SOURCE PRIOR

VS.

DIRECT ESTIMATION

FOR

Usually both the source prior and the source measurement distributions are unknown and have to be estimated in order to use the Bayes classiﬁer. Two general approaches to estimation of the mixture proportions have been developed: (1) earlier, classiﬁcation-based methods, and (2) more recent, direct methods. In classiﬁcation-based estimation, the mixture individuals are initially classiﬁed by their estimated posterior probabilities, commonly assuming the prior source probabilities are equal. Then, the apparent source proportions of the mixture are effectively corrected for misclassiﬁcation via maximum-likelihood estimation applied to the counts of mixture individuals classiﬁed to the separate sources. The likelihood function includes the confusion matrix—a probability matrix that quantiﬁes overlap among the source measurement distributions—which is treated as known even though it is estimated. In direct estimation, the probability distribution of the mixture individuals’ measurements is modeled explicitly, and its parameters, which include the source proportions, can be estimated by maximum-likelihood or Bayesian methods.

B. CONDITIONAL VS. UNCONDITIONAL ESTIMATION SOURCE MEASUREMENT DISTRIBUTIONS

OF

The source measurement distributions, or their ratios, are modeled to facilitate their estimation. These models include parameters whose values are unknown.

526

Jerome Pella and Michele Masuda

The learning samples and the mixture sample are used to estimate the source prior and the source measurement parameters. Two implementations are possible by which to accomplish the estimation: (1) conditional estimation, which uses only the learning samples to estimate the source measurement distributions or their confusion matrix; or (2) unconditional estimation, which uses the information in both the learning and mixture samples for this purpose. Unconditional estimation makes more efﬁcient use of the information in the mixture sample and provides an advantage mainly when some of the learning samples are small.

C. PLUG-IN

VS.

PREDICTIVE CLASSIFICATION

The classiﬁcation of individuals by their posterior source probabilities, computed from estimates of the source prior and source measurement parameters, can be accomplished by two methods, either plug-in or predictive classiﬁcation. If the point estimates for the unknowns are substituted into the posterior source probabilities at eq. 1 and the mixture individuals are classiﬁed to the MAP source, the Bayes classiﬁer is termed a plug-in classiﬁer. A plug-in classiﬁer uses the estimated values for the unknowns as if they equaled the true values and does not acknowledge their uncertainty. The original proposal for including uncertainty about unknowns—source prior and source measurement distributions—in the assignments is called predictive classiﬁcation and originated from Bayesian statisticians (Aitchison and Dunsmore, 1975; Geisser, 1993). In implementations of predictive classiﬁcation, described later, Bayesian methods are used to draw large samples of the unknown parameters that reﬂect their uncertainty. Given the parameter values from each Bayesian draw, the posterior source probabilities are computed for each of the mixture individuals. Then, each mixture individual is classiﬁed stochastically to their source, with probabilities equal to the posterior source probabilities. The long run relative frequencies among draws with which each individual is assigned to the possible sources are its posterior source probabilities computed to include the uncertainty in source proportions and source measurement parameters. The Bayes classiﬁer with MAP principle can be applied to these predictive relative frequencies to classify the individuals with ﬁnality if that is a goal. The predictive approach is implemented here by the SEM algorithm (Diebolt and Ip, 1996), which opens the opportunity to easily update estimates or guesses of unknown parameters with mixture sample information. For frequentists, bootstrap resampling can be viewed as an approximation to Bayesian estimation (Sec. 8.4, Hastie et al., 2001). Bootstrap resampling of the learning and mixture samples and calculation of parameter estimates can be substituted for Bayesian draws. In bootstrap resampling, the goal—either estimation of source proportions or classiﬁcation of individuals to their source—affects how samples are drawn. If estimation of source proportions is the goal, both the learn-

Classical Discriminant Analysis

527

ing and mixture samples are resampled. If classiﬁcation of individuals is the goal, only the learning samples are resampled. Next, methods for estimating the source measurement distributions, or their ratios, that are needed in the Bayes classiﬁer are described.

V. CLASSIFICATION-BASED CONDITIONAL ESTIMATION Information about the source measurement distributions will usually come mainly from learning samples obtained from the separate populations. The distribution functions for the measurements can be approximated by models of varying complexity or number of effective parameters. As complexity increases, apparent misclassiﬁcation rates from the Bayes classiﬁer applied to the learning samples decrease. However, actual misclassiﬁcation rates from independent test samples initially decrease, but then increase with greater model complexity. The misclassiﬁcation rates from the learning samples also are commonly optimistic throughout, increasingly so as complexity increases. Underlying this behavior with complexity is a trade-off between bias and variance (Hastie et al., 2001). At too low complexity, the approximating model disagrees with, or is biased for, the actual measurement distributions; and at too high complexity, it is ﬁt too closely to the learning samples and generalizes poorly to the test samples. At low complexity, the ﬁt of the approximating model is stable among possible samples from the source populations, but at high complexity, the ﬁt becomes unstable. The analyst’s choice should be an intermediate complexity which minimizes the misclassiﬁcation rates in the test samples if these are available. Two extremes in complexity are nearest neighbor and linear models (Hastie et al., 2001). Nearest neighbor models have large numbers of effective parameters and are complex from this viewpoint. Classical discriminant functions and the statistical models from which they are derived have relatively few parameters and so are comparatively simple. These simple functions of the measurements and the associated models are described next.

A. MODELS FOR SOURCE MEASUREMENTS, DISCRIMINANT FUNCTIONS, AND THEIR CONDITIONAL ESTIMATION 1. Normal Theory Discrimination Most ﬁsheries applications have employed linear methods for discrimination and classiﬁcation analysis based on the multivariate normal distribution. The most popular method is linear discriminant analysis (LDA). Underlying LDA is the assumption that the measurements of individuals, X, are distributed as the multivariate normal in the source populations with distinct means, mi, i = 1, 2, . . . , c,

528

Jerome Pella and Michele Masuda

distinguishing them, but with a shared covariance matrix, S. Speciﬁcally, the probability density functions for the d measurements of X in the separate populations are as follows: fi (X; m i , S ) =

1 (2 p)

d 2

S

1 exp ÈÍ- (X - m i )¢ S -1 (X - m i )˘˙ Î 2 ˚

12

(2)

and so the posterior source probability for the ith source is

P(i | X ) =

p i f i (X; m i , m) c

=

Â p k f k (X; m k , m) k =1

{ [ Â p {exp[ -

]}

p i exp - 12 (X - m i )¢ S -1 (X - m i )

c

k

k =1

1( X 2

]}

,

- m k )¢ S -1 (X - m k )

(3)

i = 1, 2, . . . , c To compare two populations, say the ith and jth, as possible sources for an individual with measurements X, the posterior odds, or equivalently, the logarithm of this ratio: P(i | X ) ˘ È pi ˘ 1 = log Í ˙ - (m i + m j )¢ S -1 (m i - m j ) + X ¢S -1 (m i - m j ) log ÈÍ ˙ Î P( j | X ) ˚ Îpj ˚ 2

(4)

is sufﬁcient. This log-odds transformation is easily seen to be a linear function of the components of X. When the log-odds equals zero, the posterior source probabilities are equal, and X lies on a decision boundary between the two source populations. The locus of all possible X for which the log-odds equals zero constitutes the decision boundary, which can be a point (d = 1), a straight line (d = 2), a plane (d = 3), or a hyperplane (d > 3). Either the numerator or denominator of the log-odds is a convenient linear discriminant function: d i (X ) = log (p i ) -

1 ¢ -1 m i S m i + m i ¢ S -1 X 2

i = 1, 2, . . . , c,

(5)

and classifying the individual to the source population with the largest linear discriminant function value is equivalent to classifying it to the source population with the highest posterior probability, as the Bayes classiﬁer does. Also, at the decision boundary for two populations, i and j, their discriminant functions are equal, di(X) = dj(X). The discriminant functions at eq. 5 include the parameters, mi and S, that are estimated from their learning sample counterparts: 1 mˆ i = X i = Ni

Ni

Â X ij j= 1

i = 1, 2, . . . , c

(6)

529

Classical Discriminant Analysis

Sˆ =

c

Ni

ÂÂ

(X ij - X i )(X ij - X i )¢ (N ∑ - c)

i= 1 j= 1

c

where N ∑ =

ÂNi

(7)

i= 1

Equation 5 also includes the prior probabilities, or mixture proportions. Typical software for LDA will set these prior probabilities to equal values, p = (c-1, c-1, . . . , c-1)¢, or proportional to learning sample sizes, p = (N1/N•, N2/N•, . . . , Nc/N•)¢, but neither is appropriate for the ﬁshery mixture under study. However, the counts of mixture individuals classiﬁed by the discriminant functions with incorrect p will be used later to estimate the unknown mixture proportions. A nonlinear method closely related to LDA is quadratic discriminant analysis (QDA) in which the measurements are also assumed to be distributed as the multivariate normal, and the source populations have not only distinct means, mi, i = 1, 2, . . . , c, but also covariance matrices, Si, i = 1, 2, . . . , c. The discriminant functions corresponding to eq. 5 are quadratic in the measurements: d i (X ) = log (p i ) -

1 1 log S i - (X - m i )¢ S -i 1 (X - m i ) i = 1, 2, . . . , c 2 2

(8)

The means are estimated from eq. 6 and the source covariance matrices by the following: Sˆ i =

Ni

Â (X ij - X i )(X ij - X i )¢

(N i - 1)

i = 1, 2, . . . , c

j= 1

The nonlinear decision boundaries between populations i and j require solving the discriminant functions for X such that di(X) = dj(X). The complexity or number of parameters of QDA is considerably greater than LDA, especially as the dimension of X increases, because of the added parameters of the covariance matrices. Experience with a wide range of classiﬁcation problems has shown both LDA and QDA perform remarkably well (Hastie et al., 2001), and that only when the differences among source population covariance matrices is very large would QDA be expected to perform materially better than LDA (e.g., Anas and Murai, 1969; Michie et al., 1994). Software for performing QDA is commonly available. A third classical method, logistic discrimination, also known as logistic regression, makes few distributional assumptions and also results in a linear discriminant function similar to eq. 5, but with far fewer parameters. 2. Logistic Theory Discrimination The multivariate normal assumption is commonly implausible for some measurements. Counts of meristic characters, such as gill rakers, ﬁn rays, and scale circuli, and categorical variables, such as age and presence of parasites, are clearly

530

Jerome Pella and Michele Masuda

nonnormal variables, but even continuous variables may have skew or heavytailed distributions. Often, the measurements comprise a list of continuous and discrete variables. To better model such data, logistic discrimination can be recommended. Advantages of the logistic method are that few distributional assumptions are needed, it is applicable to either discrete variables, continuous variables, or their mixture, and it is relatively robust to outliers as compared to normal theory discrimination. Disadvantages include a modest loss in efﬁciency of, at worst, about 30% in asymptotic error rate if the multivariate normal assumption were valid (Efron, 1975; Hastie et al., 2001), and some increase in computational difﬁculty for estimation of parameters. The 30% efﬁciency loss means the logistic method would do as well as LDA with 30% more data. The logistic model leaves the measurement distributions in the source populations unspeciﬁed, but makes the assumption that the logarithms of ratios of these distributions, or logodds transformations, are linear functions of the measurements, namely, d f1 (X ) ˘ log ÈÍ = a + b1 j X j 1 Â Î f c (X ) ˚˙ j= 1

M

(9) d

f c -1 (X ) ˘ = a c -1 + Â b c -1, j X j log ÈÍ Î f c (X ) ˚˙ j= 1 This assumption turns out to be exactly true for several important cases and probably a good approximation in many others. In particular, these equations are true for the multivariate normal distributions with common covariance matrix, multivariate independent dichotomous variables, or mixtures of both that are not necessarily independent. Also, certain censored or truncated forms of these distributions are covered (Seber, 1984). In the case of multivariate normal distributions, the parameters of eq. 9 are deﬁned as 1 (m i + m c )¢ S -1 (m i - m c ) i = 1, 2, . . . , c - 1 2 b i = (b i1 , b i2 , . . . , b i,d )¢ = S -1 (m i - m c ) i = 1, 2, . . . , c - 1

ai = -

(10)

The c - 1 equation pairs of eq. 10 contain only (c - 1)(d + 1) logistic parameters related to the measurement distributions, as compared to the cd + d(d + 1)/2 normal parameters. If the numerators and denominators of the measurement distribution ratios, fi(X)/fc(X), i = 1, 2, . . . , c - 1, are multiplied by the respective mixture proportions, pi and pc, and divided by the sum comprising the c

denominator of eq. 1, that is,

Â p k f k (X ) , the logarithm of the resulting odds k =1

ratio is linear in the components of X, just as eq. 5 was, but with

531

Classical Discriminant Analysis d P(i | X ) ˘ È pi ˘ + a + b X d i (X ) = log ÈÍ = log i Â i, j j ÍÎ p c ˙˚ Î P(c | X ) ˙˚ j= 1

i = 1, 2, . . . , c - 1 (11)

Notice that in the case of the multivariate normal distributions, eq. 11, with the parameters deﬁned by eq. 10, agrees with eq. 4. The decision surfaces are linear as described for LDA, and the right-hand side of eq. 11 can be considered to be a discriminant function. The Bayes classiﬁer assigns the individual with measurement X to the population for which di(X), i = 1, 2, . . . , c - 1, is positive and largest, or to i = c, if none are positive. The posterior source probabilities for measurement X and mixture proportions p can be shown from eq. 1 and eq. 9 to be: d Ê ˆ P(i | X ) = p i expÁ a i + Â b i, j X j ˜ Ë ¯ j= 1

È Ípc + Î

c -1

d Ê ˆ˘ p exp a b k , jX j ˜ ˙ + Á k k Â Â Ë ¯˚ k =1 j= 1

c -1 d È Ê ˆ˘ P(c | X ) = p c ◊ 1 Í p c + Â p i expÁ a i + Â b i, j X j ˜ ˙ Ë ¯˚ Î i= 1 j= 1

(12)

i = 1, 2, . . . , c - 1

Anderson (1982, 1984) describes maximum-likelihood estimation of the logistic parameters for several sampling designs, including the usual situation here in which separate source population learning samples are drawn. In separate source population sampling, technical issues allow only a quasi-likelihood function to be derived for which the estimators have asymptotic properties similar to those for maximum likelihood. The maximizing estimates of the quasi-likelihood function are usually computed by polytomous or multivariate logistic regression (Hosmer and Lemeshow, 1989; Anderson, 1982), hence the method is sometimes called logistic regression. This regression approach is a ﬁnesse of the actual estimation problem if X is continuous rather than discrete, in which case continuous variables are viewed as partitioned into intervals. In the regression analysis, X is regarded as ﬁxed, and the counts of individuals with this measurement, or occurring in the measurement interval, from the c source populations are assumed to have a multinomial distribution with probabilities given by eq. 12. The marginal distribution of X is unspeciﬁed and ignored, and can be thought of as estimated non-parametrically by its empirical sample distribution function. Nonlinear optimization methods are required to compute the maximumlikelihood estimates for the regression model. The Newton–Raphson algorithm was originally recommended for maximum-likelihood computations from a set of learning samples (Day and Kerridge, 1967; Anderson, 1972), and its application for the logistic regression is described in detail for the case of two source populations (Sec. 4.4.1, Hastie et al., 2001). Anderson (1982) shows that these logistic regression estimates for the intercepts must be corrected for sample sizes.

532

Jerome Pella and Michele Masuda

The intercept estimated by the logistic regression is a+i = ai + ln(Ni/Nc), that is, ln(Ni/Nc) must be subtracted from the estimated intercept. Estimates for the slope coefﬁcients, bi,j, are not affected. Evaluation of sampling variation in estimates of the logistic parameters can be accomplished by computation of the maximum-likelihood estimates from large numbers of resampled learning samples. Because nonlinear optimization methods are needed to compute the maximum-likelihood estimates, fast algorithms are a premium. Anderson (1982) changed his recommendation to quasi-Newton methods for their speed, robustness to initial starting values of the parameters, and circumvention of the Hessian inversion required by the Newton–Raphson algorithm. With large numbers of measurements per individual (d), or less commonly, large numbers of populations in the attempted mixture analysis (c), rapid inversion of the Hessian matrix becomes problematic because of its size. This limitation may not be severe in many ﬁsheries applications, and in any case, can be avoided by judicious selection of variables. However, quasi-Newton methods, which avoid the inversion, could be useful for repeated ﬁtting of the logistic regression from resamplings. Apparently, the Fortran and Algol software developed by Anderson, now deceased, for computing the estimates by either algorithm is lost. The Newton–Raphson algorithm, programmed in Fortran by Dr. Joseph Schafer, University of Pennsylvania, is used as a subroutine later.

B. MAXIMUM-LIKELIHOOD ESTIMATION

OF

SOURCE PRIOR

In order to compute the posterior source probabilities of individuals in the mixture sample, their prior source probabilities—the source population proportions of the mixture—are the ﬁnal critical parameters for which estimates are needed. A naîve estimate of these source proportions would be the apparent composition from classiﬁcation of mixture individuals to the several possible sources by the estimated discriminant functions of the previous section. Early analysts recognized that the apparent source proportions could be materially biased by misclassiﬁcations. Worlund and Fredin (1962) and Cook and Lord (1978) described the original correction for this bias, which involved the classiﬁcation, or confusion, matrix: F = (f ij )

i, j = 1, 2, . . . , c

where fij denotes the probability that the classiﬁer assigns an individual of source population i to source population j. The diagonal elements of F represent probabilities of correct assignment, and the off-diagonal elements, of misclassiﬁcation. The expected fraction of individuals assigned to source population j when the source composition of the mixture equals p is

533

Classical Discriminant Analysis c

lj =

Â pif ij

j = 1, 2, . . . , c

(13)

i= 1

With L = (l1, l2, . . . , lc)¢, the bias of the apparent source proportions from the assignments equals L - p. The linear (in p) system at eq. 13 can be written as L = F¢p, and if F is invertible, p = (F¢)-1L. Estimates of L and F can be substituted for the unknowns in order to estimate p. An unbiased estimate of L is provided by the observed proportions of the mixture individuals classiﬁed to the several source populations. Methods by which to estimate F are various and include plug-in, apparent, leaving-one-out, K-fold cross-validation, bootstrapping, and test samples (Hand, 1981; Efron, 1983; Seber, 1984; Hastie et al., 2001). All the methods estimate the proportions of individuals from the known sources that are classiﬁed, correctly or incorrectly, to each of the possible sources. Test samples provide unbiased estimates, but require the test sample individuals be withheld during calculation of the discriminant functions. Likely, better rules could be developed by including the test samples with the learning samples. Unless learning samples are very large and inexpensive, portions set aside as test samples are not practical. In K-fold crossvalidation, the learning samples are divided randomly into K equal-sized subsamples, and each in turn serves as a test sample for discriminant functions derived from the remaining combined K-1 subsamples. Larger values for K result in more variable estimates of classiﬁcation rates, and smaller values, more biased estimates. Reasonable compromise values for K are 5 or 10 (Hastie et al., 2001). Leave-one-out cross-validation (Lachenbruch and Mickey, 1968) is probably the most commonly used method of estimating classiﬁcation rates and can be viewed as K-fold cross-validation, with K equal to the total learning sample size and the separate learning samples considered to be a single sample. The leaving-one-out method provides nearly unbiased, if variable, estimates. The apparent estimate is obtained by simply classifying the individuals of the learning samples with the estimated discriminant functions, and it is optimistic in understating misclassiﬁcation rates, especially so for small learning samples. Bootstrapping can be used to estimate the optimism in the apparent estimate (Efron, 1983). The plug-in estimate of classiﬁcation rates is obtained by assuming the source measurement distributions equal their estimated counterparts from substitution of parameters by estimates in models and integrating over regions in the measurement space enclosed by the decision boundaries. The plug-in estimate is known to be too optimistic if learning samples are not large. The bias is reduced and ignorable at large sample sizes, and any of these methods can be used then. If the estimated confusion matrix ˆ = (fˆ ) F ij

i, j = 1, 2, . . . , c

534

Jerome Pella and Michele Masuda

is invertible, an estimate for p is computed as ˆ ¢)-1 Lˆ pˆ = (F

(14)

This estimate for p is that named the classiﬁcation matrix correction procedure of Cook and Lord (1979). Underlying sampling models for the source populations and the mixture were described by Pella and Robertson (1979), who then developed the covariance matrix for the estimator. However, the solution, pˆ, can be infeasible with negative values for some source proportions and so must be constrained to feasible values in order to be useful. The covariance matrix from Pella and Robertson (1979) will overstate imprecision of a constrained estimator, that is, variances too large if, as is common, any estimated source population proportions are near zero. Millar (1987) and Wood et al. (1987) showed that a likelihood function for p from the mixture classiﬁcation could easily be maximized to satisfy the necessary constraints, that is, nonnegative source proportions summing to 1. The original measurements of the mth mixture individual, Xm, m = 1, 2, . . . , M, can be viewed as transformed by classiﬁcation to a source indicator array, Zm = (zm1, zm2, . . . , zmc), which has a “1” at the jm position corresponding to the assigned population for individual m, and “0”s elsewhere. The probability that the mth ﬁsh is assigned to the jth population, that is, that zmj = 1, equals l j from eq. 13. Then the likelihood function for p from Z1, Z2, . . . , ZM is M

L(p) =

c

’ ’l j

z mj

M

=

m = 1 j= 1

’ lj m =1

M

m

c

c

c

Ê ˆ Ê ˆ = ’ Á Â p i f ijm ˜ = ’ Á Â p i f ij ˜ Ë ¯ Ë ¯ m = 1 i= 1 j= 1 i= 1

mj

(15)

M

where m j =

Â z mj

is the count of individuals assigned to the jth source

m =1

population. The likelihood function at eq. 15 can be most simply maximized with the iterative expectation-maximization (EM) algorithm (Millar, 1987), which is reliable and easy to code, even if somewhat slow. The EM algorithm is guaranteed to converge to the maximizing p provided that all the elements of the initial guess, p(0), are positive and sum to 1, and that the likelihood function is concave in p. The following equations are iterated (h = 0, 1, . . .) to convergence: p(i h +1) =

1 c Âmj M j= 1

p(i h) f ij c

Â pk

( h)

f kj

=

1 c Â m j ◊ P( h) (i | j, F) M j= 1

i = 1, 2, . . . , c

k =1

where convergence is usually assumed to occur if changes in all components of p(h) and p(h+1) become small, for example, <10-6. Millar (1987) and Wood et al.

Classical Discriminant Analysis

535

(1987) recognized that this maximum-likelihood estimate, which is always feasible, agrees with that from the classiﬁcation matrix correction procedure at eq. 14 when the latter is feasible, and so established that the maximum-likelihood method was a sound and improved replacement. To be useful, the uncertainty in the estimate of p must be quantiﬁed. Even if the uncertainty in F were negligible, the asymptotic covariance matrix for the conditional maximum-likelihood estimate of p, which equals the inverse of the negative Hessian of the logarithm of the likelihood function, is nearly always inaccurate, tending to overstate variation in pˆ because the underlying theory is valid only away from boundaries of its feasible values (Fournier et al., 1984, Millar, 1987). The inﬁnitesimal jackknife estimator of variance (Millar, 1987) can be used to determine uncertainty in the estimation of p caused by multinomial sampling variation for source populations in the mixture. However, this variance estimator does not account for added uncertainty if F is uncertain and must be estimated. In that event, bootstrap resampling of learning and mixture samples can best be used to evaluate the precision of the classiﬁcation-based conditional maximum-likelihood estimate. If the LDA assumptions are made, this bootstrapping can be accomplished with the Fortran program HISEA (Millar, 1990). In HISEA, samples are drawn with replacement from and size equal to the available samples, the discriminant functions are estimated from the resampled source samples, the resampled mixture and source samples are classiﬁed, and the apparent estimate of F is used to compute the conditional maximum-likelihood estimate of p, say pˆ*. The resampling and estimation are repeated B times to provide a bootstrap sample of estimates for p, pˆ*1, pˆ*2, . . . , pˆ*B. The sample average and standard deviation of the elements of these bootstrap maximum-likelihood estimates provide Monte Carlo estimates for the mean and standard error of the maximum-likelihood estimate, pˆ, from the original source and mixture samples. In general, a minimum of 25 repetitions of such bootstrap sampling are necessary for reliable standard errors for the maximum-likelihood estimate of p, and 1,000 repetitions are needed for conﬁdence intervals (Efron and Tibshirani, 1986). The source code for HISEA has been modiﬁed by us to output the individual bootstrap estimates and the corresponding posterior source probabilities for individuals. Both Millar (1987) and Wood et al. (1987) observe that the classiﬁcation-based method discards information. However, Millar (1987) notes an important beneﬁt from doing so: The estimation of the classiﬁcation matrix, F, and so of p, is free of assumptions about the source populations’ measurement distributions. Several considerations affect the worth of this beneﬁt. First, if classiﬁcation of individuals to their source populations is a goal, the probabilities of correct assignments must vary among individuals, depending on X, even those assigned to the same source. The classiﬁcation-based maximum-likelihood method discards information by which to evaluate the uncertainty in the individual assignments. For

536

Jerome Pella and Michele Masuda

determining probabilities of correct assignments, the analyst would be well advised to estimate the source populations’ measurement distributions, or their ratios, directly maximize the likelihood function for the mixture individuals’ measurements, as shown later, and use the source composition estimate for p in computing the posterior source probabilities at eq. 1. Second, as is evident, better rules of assignment, that is, F nearer the identity matrix, result in greater precision in estimation of p (Pella and Robertson, 1979). The best rules for classiﬁcation will usually depend on accurate estimates of the source populations’ measurement distributions, at least near the decision boundaries in the measurement space. Therefore, the distribution-free beneﬁt is best enjoyed by a careful effort to estimate as accurately as possible the measurement distributions for development of classiﬁcation rules. Third, and last, measurement distributions for many variables are reasonably approximated by the normal distribution, and if not, the multivariate logistic model may well do better. Therefore, the distribution assumptions will not usually deter direct estimation with its more efﬁcient use of information.

VI. DIRECT ESTIMATION

A. CONDITIONAL ESTIMATION OF SOURCE MEASUREMENT DISTRIBUTIONS If a mixture sample of M individuals with measurements X1, X2, . . . , XM is available and the source populations’ measurement distributions, fi(X), i = 1, 2, . . . , c, are known, the mixture likelihood function for p from the measurements is M

L(p) =

È

c

˘

’ ÍÎÂ pi f i (X m )˙˚

(16)

m = 1 i= 1

c

where p must satisfy the constraints that 0 £ pi £ 1, i = 1, 2, . . . , c, and

Â pi = 1. i= 1

This formulation was ﬁrst used in mixed population analysis by Grant et al. (1980) and Milner et al. (1981) for genotype measurements and Fournier et al. (1984) and Millar (1987) for combinations of discrete and continuous measurements, including genotypes. If the LDA assumptions hold, fi(X) = f(X; mi, S), i = 1, 2, . . . , c, are given by eq. 2 with the mi and S known. If the logistic assumptions hold, eq. 16 can be rewritten as c f i (X m ) ˘ f i (X m ) ˘ È M ÈM ˘ M È c L(p) = Í ’ f r (X m )˙ ’ Í Â p i μ ’ ÍÂ pi ˙ ˙ ( ) f X f Îm = 1 ˚ m = 1 Î i= 1 r m ˚ r (X m ) ˚ m = 1 Î i= 1

(17)

537

Classical Discriminant Analysis

and the logarithms of the odds ratios are given by eq. 9 with the ai and bi,j known. Anderson (1979) refers to this model as a multivariate logistic compound, another term for a mixture of the odds ratios. The conditional maximumlikelihood estimate of p, say pˆ is obtained by directly maximizing eq. 16 or 17 with the unknown source measurement parameters replaced by their estimates from the learning samples. The EM algorithm can be used to ﬁnd the value of p that maximizes the likelihood functions. Equation 16 is maximized by iterating (h = 0, 1, . . .) the equation system: p(i h +1) =

1 M Â M m =1

p(i h) f i (X m ) c

Â pk

( h)

=

f k (X m )

1 M ( h) Â P (i | X m ) M m =1

i = 1, 2, . . . , c

k =1

until convergence. Equation 17 is maximized by iterating the equation system: p(i h +1) =

1 M Â M m =1

p(i h) [ f i (X m ) f r (X m )] c

Âpj

( h)

[ f j (X m ) f r (X m )]

=

1 M ( h) Â P (i | X m ) M m =1

i = 1, 2, . . . , c

j= 1

Equation 12 can be used for computing the P(h)(i| Xm). At each iteration h, the EM algorithm can be said to allocate a fraction of each mixture individual to each source. The fractions are equal to the posterior source probabilities, P(h)(i| Xm), computed at each iteration. The subsequent value of p(h+1) is adjusted for the allocations of the preceding iteration and equals the arithmetic average of the allocated fractions for mixture individuals. Later, algorithms are described in which each individual is allocated as an entity to one of its possible sources. As before, the EM algorithm is guaranteed to converge to the maximizing p provided that all the elements of the initial guess, p(0), are positive and sum to 1, and that the likelihood function is concave in p. Generally, a single maximum will exist when the source populations’ measurement distributions all differ. These population differences in measurements are always sought by researchers before analysis, so a single maximum is nearly certain. Substitution for p with its maximum-likelihood estimate in eq. 1 provides the maximum-likelihood estimates of the posterior source probabilities of the mixture individuals. Bootstrap resampling of the learning and mixture samples can be used in the evaluation of precision of pˆ. The resampling and computation of the maximumlikelihood estimates of p can be performed by the program HISEA for the LDA assumptions and by our program CONDLGA for the LGA assumptions. The source code for HISEA has been modiﬁed by us to output individual bootstrap estimates for source proportions and the corresponding posterior source probabilities for individuals. The recipe for CONDLGA is provided in the appendix.

538

Jerome Pella and Michele Masuda

Skewed bootstrap distributions are typical of estimated source proportions, with higher concentrations near zero and declining relative frequencies at larger values. Such distributions can cause difﬁculty in determining frequentist conﬁdence intervals with accurate coverage (Reynolds, 2001).

B. UNCONDITIONAL ESTIMATION MEASUREMENT DISTRIBUTIONS

OF

SOURCE

Information on the source measurement distributions in the mixture sample is not used in conditional estimation. If some of the learning samples are small, use of this information can be important to improved estimation. For example, if the sources of some mixture individuals are nearly certain, as would be indicated by posterior source probabilities near 1, their measurements could as well be combined with those of the learning samples for estimation of the source measurement distributions. The extraction of this mixture sample information for mixture individuals, particularly those with less certain sources, can be accomplished by unconditional maximum-likelihood or Bayesian estimation methods. 1. Maximum Likelihood for LDA Assumptions When only mixture sample information is available, the maximum-likelihood estimates of the source proportions and measurement parameters were derived by Day (1969) and are the same as those obtained by application of the EM algorithm (McLachlan, 1982). The estimates, modiﬁed to include the source population samples, are computed iteratively from the following system: pˆ i = mˆ i =

1 M ˆ Â P(i | X m ) M m =1 M + N∑

Â

Pˆ (i | X m ) ◊ X m

m =1

Sˆ =

i = 1, 2, . . . , c

1 M + N∑

M + N∑

Â

Pˆ (i | X m )

m =1 c M + N∑

Â Â

i= 1 m = 1

Pˆ (i | X m ) ◊ (X m - mˆ i ) (X m - mˆ i )¢ (18)

The mixture and learning samples are combined in the estimation formulas for the source measurement parameters, mˆ i and Sˆ , by setting, for m > M, Pˆ(i|Xm) equal to 1 if the measurement, Xm, is from the ith population learning sample and equal to 0 otherwise. The posterior source probabilities, Pˆ(i|Xm), for m £ M can be computed by eq. 3. Initial guesses for the means, mi, i = 1, 2, . . . , c, and covariance matrix, S, can be obtained from the learning samples, and an initial guess of

Classical Discriminant Analysis

539

p = (c-1, c-1, . . . , c-1)¢ can be used to begin the computations. Revised estimates of all unknowns are obtained from the left side of eq. 18, and these are substituted into the right side. The process is repeated to convergence. Convergence to a local maximum is guaranteed, and in most applications with substantive learning samples, the maximum is probably unique. Bootstrapping can be used to determine the precision of the estimates. The learning and mixture samples are resampled and the maximum-likelihood estimates are computed iteratively by eq. 18 for each resampling. 2. Bayesian Approach for LDA Assumptions Bayesian estimation presents an alternative to bootstrapping that provides a probability distribution—the Bayesian posterior distribution—for the unknown source proportions and measurement parameters as well as posterior source probabilities for mixture individuals. The reader is alerted that the term Bayesian posterior distribution is a central concept in Bayesian statistics and refers to a probability distribution for arbitrary unknowns, given the data. The common term of this chapter, posterior source probability, also derives from Bayes’ theorem, but it represents a simple probability value given the data, X, and presumed knowledge of the values for p and the fi(X). A Bayesian posterior distribution is possible for the posterior source probability of an individual when the latter depends on unknowns, such as p or the fi(X), that are estimated from wider information than just the individual’s measurements. The Bayesian credible interval, also called the posterior interval, is obtained from quantiles of the Bayesian posterior distribution, and is the counterpart to the frequentist conﬁdence interval. The credible intervals are direct probability statements about the unknowns’ values rather than the often misconstrued indirect probability statements of frequentist conﬁdence intervals regarding coverage of parameter values under imaginary sampling repetitions. Bayesian methods for estimation of source population proportions and individual assignments are a practical alternative to a direct unconditional maximumlikelihood method. The direct unconditional maximum-likelihood method has been described for genotypic measurements (Pella and Milner, 1987), where it is known to have the technical problem that the likelihood function may well have multiple maxima (Smouse et al., 1990). In Bayesian estimation, the collection of unknowns—the source population proportions, p, and parameters of the source probability densities of measurements, q—are viewed as random variables, and the measurements, X, as ﬁxed. The Bayes posterior distribution summarizes the information for p and q in the learning and mixture samples. The posterior distribution is obtained by integrating the product of a prior probability density, p(p,q), for the unknowns and the likelihood of the measurements given the unknowns, namely,

540

Jerome Pella and Michele Masuda

M È c ˘ p(p, q | X 1 , . . . , X m ) = a ◊ p(p, q) ◊ ’ Í Â p i f i (X m | q)˙ ˚ m = 1 Î i= 1

where the constant of proportionality, a, has a value such that the posterior density integrates to 1, and the notation fi(Xm|q) makes explicit the dependence of the source measurement distributions on the unknown parameters, q. The joint prior density for p and q is assumed here to be a product of individual priors, p(p,q) = p(p)p(q|Y), where Y represents the learning sample measurements available from the separate source populations preceding the mixture sample. Analytical derivation of the posterior distribution is impractical, but a large number of samples drawn from it can be used to describe its characteristics essentially without error. Pella and Masuda (2001) developed a Markov chain Monte Carlo (MCMC) algorithm for genetic measurements, which is easily modiﬁed to accommodate other kinds of data once the form of the joint probability density for the data is speciﬁed. A unit Dirichlet prior distribution (equal parameters summing to 1) for the source proportions is used because of its interpretation as the information provided by just a single additional mixture individual that is neutral with regard to its source, the ease with which MCMC samples of the posterior can be obtained with its use, and the close agreement of the mean and covariance matrix for p to those of frequentist methods if the population sources of the mixture individuals were known. The mixture prior of q|Y depends on the probability densities for the measurements and requires a Bayesian analysis of the learning samples. The necessary tools are well described in many texts on Bayesian statistics, for example, Gelman et al. (1995) and Tanner (1996). Using LDA assumptions and the noninformative Jeffrey’s prior for the multivariate normal parameters (see p. 81, Gelman et al., 1995), the posterior distributions from the learning samples are the normal-inverse-Wishart distribution. The MCMC sampling can be begun at ﬁxed or random choices for p and q = (m1, m2, . . . , mc, S)¢. In program UCONLDA developed for this sampling, random starting values are drawn from distributions for the unknown parameters. At any cycle of the MCMC sampling, each mixture individual is randomly assigned to one of the source populations with probabilities equal to the posterior source probabilities at eq. 1, the latter computed from the most recent draws of p and q. Given the counts of individuals assigned to the source populations, say, m1, m2, . . . , mc, the posterior distribution of p is the Dirichlet with parameters equal to the augmented counts, m1 + c-1, m2 + c-1, . . . , mc + c-1. The mixture individuals assigned to each source population can be combined with those in the learning sample to provide an updated posterior distribution for q. A draw of the next p and q from their updated posterior distributions brings the process back to random assignment of the mixture individuals to source populations. A burn-in sample needs to be discarded. Recommendations for burn-in and running mul-

Classical Discriminant Analysis

541

tiple chains to assure convergence to the posterior distribution for the unknowns can be found in Pella and Masuda (2001). The recipe for this algorithm can be found in the appendix, and it is implemented by our program UCONLDA. 3. Maximum Likelihood for LGA Assumptions Anderson (1979) derived the likelihood function for unconditional estimation of source proportions and the logistic parameters of the source populations. The likelihood function is nonlinear in the parameters, and Anderson (1982) indicates a Fortran program for its maximization was available, probably for the case of two-population mixtures. The program was not found by our limited search of the Internet.

VII. APPLICATIONS TO THE TREE POINT SOCKEYE SALMON BLIND MIXTURE SAMPLES The learning sample sizes for Nass River, Skeena River, and Southeast Alaska for age 1.2 were 199, 199, and 200 individuals, respectively; and for age 2.2, 200, 32, and 148 individuals. The age 2.2 sample for the Skeena River populations is relatively small, and unconditional estimation should perform better for this age group. Four scale pattern variables, two discrete and two continuous, of the many (>50) available were selected by stepwise discriminant analysis of the learning samples for age 1.2 as being useful. Histograms of their distributions (Fig. 25-1) show considerable overlap, but obvious differences in the locations occur. To emphasize differences among populations, the learning sample measurements are plotted on the ﬁrst two linear discriminant variables, which are axes oriented in the original space of measurements to maximize separation of the populations (Hastie et al., 2001) (Fig. 25-2). The decision planes corresponding to equiprobable sources are projected as lines onto this transformed space (Fig. 25-2). Details for age 2.2 learning samples will not be described here. The blind mixture sample of 195 age 1.2 sockeye salmon was comprised of 69 (35.4%) Nass River, 76 (39.0%) Skeena River, and 50 (25.6%) Southeast Alaska individuals; and that of the 128 age 2.2 sockeye salmon was composed of 77 (60.2%), 20 (15.6%), and 31 (24.2%) individuals of those respective populations. The age 1.2 sockeye salmon population composition does not differ from equal proportions as much as does that of age 2.2, and so use of a discriminant function with equiprobable prior would perform better for the former age group than the latter. If the individuals had been perfectly identiﬁable to their sources, sampling theory (Cochran, 1963) tells us that the standard errors of estimated source composition for age 1.2 would be 3.4%, 3.5%, and 3.1%, for Nass River, Skeena River, and Southeast Alaska, and for age 2.2, 4.3%, 3.2%, and 3.8%,

542

Jerome Pella and Michele Masuda

4

second linear discriminant

2

0

–2

–4 –4

–2

0

2

first linear discriminant FIGURE 25-2. Age 1.2 learning samples plotted on the ﬁrst and second discriminant or canonical variable, Nass River (+), Skeena River (䉭), Southeast Alaska ( ). The linear decision surfaces are indicated by the lines (- - -).

•

respectively. The standard errors for the methods applied next are considerably larger, thereby reﬂecting the added variation caused by overlap of measurement distributions and uncertainty in their parameters. Four of the methods for assessing source composition have been applied to the Tree Point mixed population samples of age 1.2 and 2.2 sockeye salmon (Table 25-1). The four methods are labeled as follows: Method 1: LDA-classiﬁcation of mixture individuals, bootstrap resampling for the conditional maximum-likelihood estimate of p, assuming a multivariate normal mixture (MVN), computations by the program HISEA (the 4th estimator of Millar, 1990) Method 2: Direct estimation, bootstrap resampling for the conditional maximum-likelihood estimate of p, assuming a multivariate normal mixture, computations by the program HISEA (the 5th estimator of Millar, 1990)

543

Classical Discriminant Analysis TABLE 25-1. Actual Source Population Composition (%) of Blind Mixtures of Sockeye Salmon of Age Groups 1.2 (Mixture Sample Size M = 195) and 2.2 (M = 128), and Distributional Statistics for Estimates Derived by Four Methodsa Age 1.2 Sockeye Salmon Estimation method

Nass River 35.4

Means (left) and standard deviations (right) Method 1: LDA classiﬁcation, 34.6 bootstrap, conditional, MVN, HISEA Method 2: Direct, bootstrap, 37.6 conditional, MVN, HISEA Method 3: Direct, bootstrap, 34.6 conditional, MLG, CONDLGA Method 4: Direct, MCMC, Bayes 37.7 unconditional, MVN, UCONLDA 2.5 (left) and 97.5 (right) percentiles Method 1: LDA classiﬁcation, bootstrap, conditional, MVN, HISEA Method 2: Direct, bootstrap, conditional, MVN, HISEA Method 3: Direct, bootstrap, conditional, MLG, CONDLGA Method 4: Direct, MCMC, Bayes, unconditional, MVN, UCONLDA

Skeena River 39.0

Alaska 25.6

6.0

39.3

5.3

26.1

5.0

5.8

38.5

4.9

23.9

4.7

5.7

39.1

4.5

26.4

5.2

5.6

39.7

4.7

22.7

4.6

23.1

46.3

29.2

49.4

17.0

36.3

27.0

49.2

29.1

48.2

15.3

33.9

23.3

46.3

30.4

48.3

16.8

36.8

26.9

49.0

30.4

49.1

14.3

32.0

Age 2.2 Sockeye Salmon Estimation method

Nass River 60.2

Means (left) and standard deviations (right) Method 1: LDA classiﬁcation, 52.7 bootstrap, conditional, MVN, HISEA Method 2: Direct, bootstrap, 58.3 conditional, MVN, HISEA Method 3: Direct, bootstrap, 54.9 conditional, MLG, CONDLGA Method 4: Direct, MCMC, Bayes, 59.5 unconditional, MVN, UCONLDA 2.5 (left) and 97.5 (right) percentiles Method 1: LDA classiﬁcation, bootstrap, conditional, MVN, HISEA Method 2: Direct, bootstrap, conditional, MVN, HISEA

Skeena River 15.6

Alaska 24.2

10.1

21.7

11.3

25.6

5.7

6.6

15.2

6.0

26.5

4.6

7.1

16.6

7.1

28.5

5.0

5.9

14.0

5.0

26.6

4.4

29.8

72.0

2.1

48.3

14.9

36.9

44.2

70.1

5.4

29.0

17.6

36.1

544

Jerome Pella and Michele Masuda

TABLE 25-1. Continued Age 2.2 Sockeye Salmon Estimation method Method 3: Direct, bootstrap, conditional, MLG, CONDLGA Method 4: Direct, MCMC, Bayes, unconditional, MVN, UCONLDA

Nass River 60.2

Skeena River 15.6

Alaska 24.2

40.5

68.6

4.9

32.4

18.9

38.6

47.5

70.6

5.3

24.9

18.3

35.6

a Method 1: LDA-classiﬁcation of mixture individuals, bootstrap resampling for the conditional maximum-likelihood estimate of p, assuming a multivariate normal mixture (MVN), computations by the program HISEA; Method 2: direct estimation, bootstrap resampling for the conditional maximum-likelihood estimate of p, assuming a multivariate normal mixture, computations by the program HISEA; Method 3: direct estimation, bootstrap resampling for the conditional maximumlikelihood estimate of p, assuming a multivariate logistic compound (MLG), computations by the program CONDLGA; and Method 4: direct estimation, MCMC sampling from Bayesian unconditional posterior distribution of unknowns including p, assuming a multivariate normal mixture (MVN), computations by the program UCONLDA. An initial sample of 30,000 draws of the unknowns was made from the Bayes posterior distribution. Trace plots and the Raftery and Lewis (1996) diagnostic, computed by the Fortran program GIBBSIT, for these samples indicated convergence. Their second halves, 15,000 samples, are used to compute the descriptive statistics.

Method 3: Direct estimation, bootstrap resampling for the conditional maximum-likelihood estimate of p, assuming a multivariate logistic compound (MLG), computations by the program CONDLGA Method 4: Direct estimation, MCMC sampling from Bayesian unconditional posterior distribution of unknowns including p, assuming a multivariate normal mixture (MVN), computations by the program UCONLDA. The programs can be obtained by either contacting the authors or via the Internet (click on the link under Stock Identiﬁcation of the Auke Bay Laboratory homepage). In general, all the methods provide averages that approximate reasonably well the actual source compositions, and the 95% conﬁdence intervals for three maximum-likelihood methods or 95% posterior intervals for the Bayesian method, cover the actual source compositions. Generally speaking, Method 1 produced composition estimates with less precision than the others, more noticeably for age 2.2 in which the mixture sample and learning samples were smaller. The discard of information by classiﬁcation-based estimation noted by Wood et al. (1987) and Millar (1990) explains its poorer performance. The beneﬁt in precision of unconditional estimation by Method 4 compared to the other direct esti-

545

Classical Discriminant Analysis

mation methods, Methods 2 and 3, was evident but not impressive. Evidently, the learning samples would need to be smaller in order for unconditional estimation to excel conditional estimation. The individuals of age 1.2 were classiﬁed by the maximum of their average posterior source probabilities, that is, the MAP principle applied to the averages. Because both microsatellites and scale pattern data were obtained for the individuals, the assignments can be compared. For scale patterns, the average posterior probabilities were computed by Method 3 using program CONDLGA, and for microsatellites, by the Bayesian method and the implementing program BAYES described by Pella and Masuda (2001). The comparison is limited to 157 of the 195 individuals because laboratory results are incomplete at this time (Fig. 253). The scale patterns and the genetic data appear largely independent as indi-

1.0

scale patterns

0.8

0.6

0.4

0.4

0.6

0.8

1.0

microsatellites FIGURE 25-3. Comparison of maximum average posterior source probabilities from scale patterns (program CONDLGA) and genetic microsatellites (program BAYES) for age 1.2 mixture individuals. The individuals, whose sources are now known, were assigned by the maximum of the average posterior probability, and these assignments by the two kinds of information were either both correct ( ), both incorrect (䊊), only microsatellites were correct (䉭), or only scale patterns were correct (+).

•

546

Jerome Pella and Michele Masuda

cated by the posterior probabilities, that is, any correspondence between posterior probabilities is obviously weak. The assignments from the two kinds of data were both correct for 113 (72.0%) individuals, both incorrect for only 2 (1.3%) individuals; only microsatellites were correct for 37 (23.6%) individuals, and only scale patterns were correct for 5 (3.2%) individuals. As the maxima for individuals rise, relatively more of them are expected to be correctly assigned. For example, for average posterior probabilities from scale patterns between 0.4 to 0.6, 12 of 23 (52.2%) assignments were correct, between 0.6 to 0.8, 37 of 55 (67.3%) assignments were correct, and between 0.8 to 1, 70 of 79 (88.6%) assignments were correct. That 9 individuals were incorrectly assigned from scale patterns despite an average posterior probability in excess of 0.8 is slightly suspicious until the measurements of these individuals are compared to those of the learning samples (Fig. 25-1). Clearly, the misclassiﬁed individuals appear from their scale patterns to be more like the learning samples of the incorrect sources than actual sources. Small populations within the three groups—Nass River, Skeena River, and Southeast Alaska—could easily have measurements more similar to the other groups than their actual group. Given the complexity of stock structure hidden by grouping, the good performance of the methodology, even to the detail of identifying individuals’ sources, is encouraging.

VIII. SUMMARY Classical discriminant and classiﬁcation analysis has been used since the late 1950s to assess the unknown source population proportions composing ﬁsh mixtures. Although the unknown source proportions are the usual focus of study, the source labels of the mixture individuals are sometimes of interest as well. Measurements of individuals in learning samples from the separate sources are used to determine rules, or discriminant functions, by which to classify measured individuals of unknown sources. Linear discriminant functions covered in this chapter are based on the assumption that the measurement distributions in the source populations are multivariate normal with common covariance matrix, or else their distributions are unspeciﬁed, but the log odds between populations are assumed to be linear. The source labels of the mixture individuals are usually assigned with uncertainty because the measurement distributions of the sources overlap. Among mixture individuals with matching measurements, the proportions composing the mixture from the separate sources limit the certainty with which they can be classiﬁed. These proportions, which vary with the measurements, are called the posterior source probabilities of the mixture individuals. The Bayes classiﬁer, which has the lowest misclassiﬁcation rate, assigns any unlabeled mixture individual to the source for which the posterior source probability is greatest. However, the posterior source probabilities are functions of

Classical Discriminant Analysis

547

unknowns that include the source proportions, also known as prior source probabilities, as well as parameters of the separate source measurement distributions. A circularity is evident in that the intent of classifying the mixture individuals is to determine the source proportions and source labels, but the optimal classiﬁer for the mixture individuals depends on the unknown source proportions. Solutions to this conundrum are of two kinds: classiﬁcation-based and direct estimation. In classiﬁcation-based estimation, the mixture individuals and the learning or other test individuals are initially classiﬁed to the candidate sources using the Bayes classiﬁer with an arbitrary proposal for the source prior. Because the apparent source composition from this mixture classiﬁcation is biased, the source proportions are best estimated by maximizing a likelihood function from the counts assigned to the various sources. In direct estimation, two forms are described: conditional and unconditional estimation. In both forms, iterative algorithms alternately (1) compute the posterior source probabilities of mixture individuals from current estimates of source proportions and the source measurement parameters, and (2) revise the parameter estimates from the current posterior source probabilities. Both the source proportions and the source measurement parameters are revised during unconditional direct estimation, but only the source proportions are revised during conditional direct estimation. Unconditional estimation extracts mixture sample information about source measurement parameters, which is ignored by conditional estimation. The enhanced precision from unconditional estimation will mainly occur if some of the learning samples are small. The uncertainty in the unknown source proportions, source measurement parameters, and posterior source probabilities can be described by bootstrap resampling or Bayesian methods. Either a plug-in or predictive Bayes classiﬁer can be applied to the mixture individuals, if their source labels are needed. The plug-in version replaces unknown parameters by their point estimates, and the predictive version uses averages of distributions computed for the posterior source probabilities.

ACKNOWLEDGMENTS Joseph L. Schafer, Department of Statistics, The Pennsylvania State University, provided us his Fortran code of the iterative Newton–Raphson algorithm to estimate the logistic regression parameters. Russell Millar made the source code for his HISEA program available to Michael Prager, who passed it to us. The scale pattern data for sockeye salmon related to the Tree Point ﬁshery were made available to us by Richard Bloomquist, Glen Oliver, and Tim Zadina of the Alaska Department of Fish and Game. The microsatellites were assayed and the data processed under the direction of Terry Beacham of the Canadian Department of Fisheries and Oceans Paciﬁc Biological Station, and John Candy of that laboratory provided the average posterior probabilities computed with program BAYES that are used in Figure 25-3. Our sincere thanks are extended to all for their favors.

548

Jerome Pella and Michele Masuda

APPENDIX Section 1. Bootstrap resampling and conditional maximum-likelihood estimation of source population proportions from logistic discriminant analysis. 1. Resample the mixture and source population samples. 2. Compute the maximum-likelihood estimates of (c - 1)(d + 1) logistic regression parameters of the logarithms of ratios of measurement distributions from the resampled source population samples. Correct the intercept coefﬁcient estimates, denoted by aˆ +i, i = 1, 2, . . . , c - 1, computed by the maximum-likelihood program as aˆ i = aˆ +i - ln(Ni/Nc), i = 1, 2, . . . , c - 1. Denote these resample corrected estimates of intercept coefﬁcients and estimates of slope coefﬁcients by aˆ i. and bˆ i,j, i = 1, . . . , c - 1, j = 1, . . . , d. 3. Set p(0) = (c-1, c-1, . . . , c-1), the equiprobable prior, and set the iteration t = 0. 4. Compute the posterior source population probabilities of the resampled mixture individuals with measurements X1, X2, . . . , XM as d ˆ Ê pˆ (i t ) expÁ aˆ i + Â bˆ i, j X m , j ˜ ¯ Ë j= 1

Pm( t ) (i | X m ) =

c -1

( t)

pˆ c

d Ê ˆ + Â p k expÁ aˆ k + Â bˆ k , j X m , j ˜ Ë ¯ k =1 j= 1

p(ct )

Pm( t ) (c | X m ) =

c -1

( t)

pc

i = 1, 2, . . ., c - 1

( t)

d Ê ˆ + Â p k expÁ aˆ k + Â bˆ k , j X m , j ˜ Ë ¯ k =1 j= 1

m = 1, 2, . . ., M

( t)

5. Compute p(t+1) as p(i t +1) =

1 M ( t) Â P (i | X m ) M m =1

i = 1, 2, . . . c

6. Set t = t + 1 and return to step 4. Cycle the calculation of p(t), t = 1, 2, . . . , T, to convergence. 7. Return to step 1. Section 2. MCMC sampler for direct Bayesian estimation for normal measurement distributions with common covariance matrix. Assume that the vector of d measurements for a ﬁsh from the ith source population, X, is distributed as the multivariate normal density with mean mi, and covariance matrix, S, that is,

549

Classical Discriminant Analysis

fi ( X | m i , S ) =

1 (2 p)

d 2

S

12

1 exp ÈÍ- ( X - m i )¢ S -1 ( X - m i )˘˙ Î 2 ˚

Use the noninformative Jeffrey’s prior for the normal model parameters (see p. 81, Gelman et al., 1995) and the low-information Dirichlet prior of Pella and Masuda (2001) for the source population proportions. 1. Draw a Dirichlet random vector p ~ D(1/c, . . . ,1/c). 2. Draw S|X ~ Inv-WishartN-c(S-1), where S =

c

Ni

Â Â (X ij - X i )(X ij - X i )

¢

is

i= 1 j= 1

the pooled sum of cross products matrix. ¯ i, (1/Ni)·S], i = 1, 2, . . . , c. 3. Draw normal random vectors mi|S, X ~ N[X 4. Compute the posterior source probabilities for each of the M mixture i = 1, 2, . . . , c; p i f i (X m | m i , S ) individuals, P(i | X m ) = c m = 1, 2, . . . , M. Â p k f k (X m | m k , S) k =1

5. Draw the source of each mixture individual with a single draw from the multinomial distribution with probabilities P(i|Xm), i = 1, 2, . . . , c. If the assigned source population of individual m is i, set the ith element, zmi, of the vector zm to “1” and the other (c - 1) elements, zmj, j π i, to “0”. Denote the number of mixture individuals assigned to the ith population by M

mi =

Â z mi ,

i = 1, 2, . . . , c.

m =1

6. Combine the mixture individuals assigned to the ith source population with those of the learning sample to form an augmented learning sample of size N*i= Ni + mi, denote the collection of learning and assigned measurements by X*i, i = 1, 2, . . . , c, and denote the collection of c such augmented learning samples by X*. 7. Draw S|X* ~ Inv-WishartN+M-c(S*-1), where S* =

c N* i

Â Â (X *ij - X *i ) i= 1 j= 1

¯ *i )¢ is the pooled sum of cross products matrix. (X*ij - X 8. Draw normal random vectors mi|S, X* ~ N[X¯ *i, (1/N*i)·S], i = 1, 2, . . . , c. 9. Draw a Dirichlet random vector p ~ D(m1 + 1/c, . . . , mc + 1/c). 10. Go to step 4 and repeat until speciﬁed sample size is achieved. If the learning samples are large and parameter uncertainty in the source measurement probability densities is negligible, steps 2, 3, 6, 7, and 8 can be omitted and the maximum-likelihood estimates can be used throughout the sampling.

550

Jerome Pella and Michele Masuda

REFERENCES Aitchison, J. and Dunsmore, I. R. 1975. Statistical Prediction Analysis. Cambridge University Press, London. 273 pp. Amos, M. H., Anas, R. E., and Pearson, R. E. 1963. Use of discriminant function in the morphological separation of Asian and North American races of pink salmon, Oncorhynchus gorbuscha (Walbaum). Int. North Pac. Fish. Comm., Bull 11: 73–100. Anas, R. E. 1964. Sockeye salmon scale studies. Int. North Pac. Fish. Comm., Annual Rep. 1963: 158–162. Anas, R. E. and Murai, S. 1969. Use of scale characters and a discriminant function for classifying sockeye salmon (Oncorhynchus nerka) by continent of origin. Int. North Pac. Fish. Comm., Bull 26: 157–192. Anderson, J. A. 1972. Separate sample logistic discrimination. Biometrika 59: 19–35. Anderson, J. A. 1974. Diagnosis by logistic discriminant function: further practical problems and results. Appl. Statist. 23: 397–404. Anderson, J. A. 1979. Multivariate logistic compounds. Biometrika 66: 17–26. Anderson, J. A. 1982. Logistic discrimination. pp. 169–191, In P. R. Krishnaiah and L. N. Kanal (eds.), Handbook of Statistics, Vol. 2, Classiﬁcation, Pattern Recognition, and Reduction of Dimensionality. North-Holland, Amsterdam, pp. 169–191. Anderson, J. A. 1984. On the existence of maximum likelihood estimates in logistic regression models. Biometrika 71: 1–10. Cochran, W. G. 1963. Sampling Techniques, 2nd Ed. Wiley, New York. 413 pp. Cook, R. C. and Lord, G. E. 1978. Identiﬁcation of stocks of Bristol Bay sockeye salmon, Oncorhynchus nerka, by evaluating scale patterns with a polynomial discriminant method. Fishery Bulletin 76: 415–423. Cook, R. C. and Guthrie, I. 1987. In-season stock identiﬁcation of sockeye salmon (Oncorhynchus nerka) using scale pattern recognition. pp. 327–334, In H. D. Smith, L. Margolis, and C. C. Wood (eds.), Sockeye Salmon (Oncorhynchus nerka) Population Biology and Future Management. Can. Spec. Publ. Fish. Aquat. Sci. 96: 327–334. Dark, T. A. and Landrum, B. J. 1964. Analysis of 1961 red salmon morphological data. Int. North Pac. Fish. Comm., Annual Rep. 1962: 110–115. Day, N. E. and Kerridge, D. F. 1967. A general maximum likelihood discriminant. Biometrics 23: 313–323. Day, N. E. 1969. Estimating the components of a mixture of normal distributions. Biometrika 56: 463–474. Diebolt, J. and Ip, E. H. S. 1996. Stochastic EM: method and application. pp. 259–273, In W. R. Gilks, S. Richardson, and D. J. Spiegelhalter (eds.), Markov Chain Monte Carlo in Practice. Chapman & Hall, London, pp. 259–273. Efron, B. 1975. The efﬁciency of logistic regression compared to normal discriminant analysis. J. Amer. Statist. Assoc. 70: 892–898. Efron, B. 1982. The jackknife, the bootstrap, and other resampling plans. Soc. Ind. Appl. Math. CBMS-Natl. Sci. Found. Monogr. 38. Efron, B. 1983. Estimating the error rate of a prediction rule: improvement on cross-validation. J. Amer. Statistical Association 78: 316–331. Efron, B. and Tibshirani, R. 1986. Bootstrap methods for standard errors, conﬁdence intervals, and other measures of statistical accuracy. Statistical Science 1: 54–77. Fournier, D. A., Beacham, T. D., Riddell, B. E., and Busack, C. A. 1984. Estimating stock composition in mixed stock ﬁsheries using morphometric, meristic, and electrophoretic characteristics. Can. J. Fish. Aquat. Sci. 41: 400–408.

Classical Discriminant Analysis

551

Fraley, C. and Raftery, A. E. 2002. Model-based clustering, discriminant analysis, and density estimation. J. Amer. Stat. Soc. 97: 611–631. Fukuhara, F. M., Murai, S., LaLanne, J. J., and Sribhibhadh, A. 1962. Continental origin of red salmon as determined from morphological characters. Int. North Pac. Fish. Comm., Bull 8: 15–109. Gable, J. and Cox-Rogers, S. 1993. Stock identiﬁcation of Fraser River sockeye salmon: methodology and management application. Paciﬁc Salmon Commission Tech. Rep. No. 5. 36 pp. Geisser, S. 1993. Predictive Inference: An Introduction. Monographs on Statistics and Applied Probability 55. Chapman and Hall, New York. 264 pp. Gelman, A., Carlin, J., Stern, H. S., and Rubin, D. B. 1995. Bayesian Data Analysis. Chapman and Hall, New York. 526 pp. Grant, W. S., Milner, G. B., Krasnowski, P., and Utter, F. M. 1980. Use of biochemical genetic variants for identiﬁcation of sockeye salmon (Oncorhynchus nerka) stocks in Cook Inlet, Alaska. Can. J. Fish. Aquat. Sci. 37: 1236–1247. Hand, D. J. 1981. Discrimination and Classiﬁcation. Wiley, New York. 218 pp. Hastie, T. and Tibshirani. 1996. Discriminant analysis by Gaussian mixtures. J. Royal Statist. Soc. B. 58: 155–176. Hastie, T., Tibshirani, R., and Friedman, J. 2001. The Elements of Statistical Learning. Data Mining, Inference, and Prediction. Springer, New York. 533 pp. Hill, D. R. 1959. Some uses of statistical analysis in classifying races of American shad (Alosa sapidissima). Fish. Bull., U.S. 59: 269–286. Hosmer, D. W. Jr. and Lemeshow, S. 1989. Applied Logistic Regression. Wiley New York. 307 pp. Lachenbruch, P. A. and Mickey, M. R. 1968. Estimation of error rates in discriminant analysis. Technometrics 10: 1–11. Mason, J. E. 1966. Sockeye salmon scale studies. Int. North Pac. Fish. Comm., Annual Rep. 1964: 117–118. McLachlan, G. J. 1982. The classiﬁcation and mixture maximum likelihood approaches to cluster analysis. In P. R. Krishnaiah and L. N. Kanal (eds.)., Handbook of Statistics, Vol. 2, Classiﬁcation, Pattern Recognition, and Reduction of Dimensionality. North-Holland, Amsterdam. pp. 199–208. Michie, D., Spiegelhalter, D. J., and Taylor, C. C. (eds.) 1994. Machine Learning, Neural and Statistical Classiﬁcation. Ellis Horwood Series in Artiﬁcial Intelligence, Ellis Horwood Hertfordshire, England. 290 pp. Also available free at www.amsta.leeds.ac.uk/~charles/statlog/. Millar, R. B. 1987. Maximum likelihood estimation of mixed stock ﬁshery composition. J. Fish. Aquat. Sci. 44: 583–590. Millar, R. B. 1990. A versatile computer program for mixed stock ﬁshery composition estimation. Can. Tech. Rep. Fish. Aquat. Sci. 1753: iii + 29p. Milner, G. B., Teel, D. J., Utter, F. M., and Burley, C. L. 1981. Columbia River stock identiﬁcation study: validation of genetic method. Annual report of research (FY80). NWAFC, NOAA, Seattle, Washington. Pella, J. J. and Robertson, T. L. 1979. Assessment of composition of stock mixtures. Fishery Bull 77: 387–398. Pella, J. J. and Milner, G. B. 1987. Use of genetic marks in stock composition analysis, In N. Ryman and F. Utter (eds.), Population Genetics and Fishery Management. University of Washington Press, Seattle, pp. 247–276. Pella, J. and Masuda, M. 2001. Bayesian methods for analysis of stock mixtures from genetic characters. Fish. Bull. 99: 151–167. Prager, M. H. and Fabrizio, M. C. 1990. Comparison of logistic regression and discriminant analyses for stock identiﬁcation of anadromous ﬁsh, with application to striped bass (Morone saxatilis) and American shad (Alosa sapidissima). Can. J. Fish. Aquat. Sci. 47: 1570–1577.

552

Jerome Pella and Michele Masuda

Prentice, R. L. and Pyke, R. 1979. Logistic disease incidence models and case-control studies. Biometrika 66: 403–411. Raftery, A. E. and Lewis, S. M. 1996. Implementing MCMC. In W. R. Gilks, S. Richardson, and D. J. Spiegelhalter, (eds.), Markov Chain Monte Carlo in Practice. Chapman & Hall, London, pp. 115–130. Reynolds, J. H. 2001. SPAM Version 3.5: User’s Guide Addendum. Addendum to Special Publication 15, Alaska Department of Fish and Game, Commercial Fisheries Division, Gene Conservation Lab, Anchorage, Alaska. 63 pages. Available from http://www.cf.adfg.state.ak.us/geninfo/research/ genetics/Software/SpamPage.htm. Ripley, B. D. 1996. Pattern Recognition and Neural Networks. Cambridge University Press, Cambridge, UK. 403 pp. Seber, G. A. F. 1984. Multivariate Observations. Wiley, New York. 686 pp. Smouse, P. E., Waples, R. S., and Tworek, J. A. 1990. A genetic stock-mixture analysis for use with incomplete source population data. Can. J. Fish. Aquat. Sci. 47: 620–634. Tanner, M. A. 1996. Tools for Statistical Inference. Methods for the Exploration of Posterior Distributions and Likelihood Functions, 3rd Ed. Springer, New York. 207 pp. Wood, C. C., McKinnell, S., Mulligan, T. J., and Fournier, D. A. 1987. Stock identiﬁcation with the maximum-likelihood mixture model: sensitivity analysis and application to complex problems. Can. J. Fish. Aquat. Sci. 44: 866–881. Worlund, D. D. and Fredin, R. A. 1962. Differentiation of stocks. In N. J. Wilimovsky (ed.), Symposium on Pink Salmon. H. R. MacMillan Lectures in Fish, University of British Columbia, Vancouver, pp. 143–153.

CHAPTER

26

Neural Networks Used in Classiﬁcation with Emphasis on Biological Populations SAUL B. SAILA University of Rhode Island, Graduate School of Oceanography, Narragansett, Rhode Island, USA

I. II. III. IV.

Introduction Neural Tasks Neural Network Classiﬁcation Applications Some Comparisons of Neural Networks with Statistical Methods in Classiﬁcation V. Neural Network Types for Classiﬁcation A. Supervised Learning: MLFN Example B. Unsupervised Learning: SOM Example VI. Discussion and Suggestions References

I. INTRODUCTION Neural networks are one of several types of artiﬁcial intelligence tools for which Saila (1996) has provided a brief introduction from a ﬁshery science perspective. A more speciﬁc introduction to neural network architectures can be found in a book by Dayhoff (1990). Zurada (1990) provides a comprehensive and uniﬁed text for anyone with a good background in mathematics and statistics, and an intuitive approach to neural networks is given by Smith (1993). Neural network development and applications have increased in several areas of science to the extent that neural networks are the most actively used of the available artiﬁcial intelligence tools. This chapter is conﬁned primarily to introducing ﬁshery scientists to neural network methodology and applications in biological classiﬁcation. However, this is preceded by a very short introduction to neural networks and some of its terminology. Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

553

554

Saul B. Saila

For the purposes at hand, neural networks are deﬁned as models of real-world systems which are built by training a set of parameters. These parameters, which are called weights, describe a model that forms a mapping from a set of given values known as inputs to an associated set of values known as outputs. The process of ﬁnding the weights for the correct values is called training, and it is usually carried out by passing sets of input–output pairs through the model and adjusting the weights in order to minimize the error between the answer the network provides and the desired output. Once the weights have been set by suitable training, the model is able to produce answers for input values which are not included in the training data. The models do not refer to the training data after they have been trained, and in some sense they are a functional summary of the training data. The number of inputs to the network need not be equal to the number of outputs. Thus, the neural network can be described as performing a mapping from a set of variables onto another set of a different size. The total set of combinations of possible values a set of variables can take is referred to as its space. Therefore, the input variables can take any set of values in input space. Each set of values can be conceived as a point in this space, and the neural network can be thought of as a machine which learns the route from each point in the input space to the correct point in the output space.

II. NEURAL TASKS Neural networks are used primarily to learn two types of tasks: 1. Classiﬁcation. Classiﬁcation involves tasks where the input is a description of an object to be recognized. An example might be the microelemental composition of the ﬁrst-year growth zone of an otolith of a juvenile Atlantic blueﬁn tuna from a given geographic area. The output might be a classiﬁcation of the source (Gulf of Mexico or Mediterranean Sea) to which the object (ﬁsh) belongs. As reported by Swingler (1996), a classiﬁcation task is one for which the outputs cannot be arrayed along a meaningful continuum, such as the Atlantic blueﬁn tuna sources just indicated; that is, each possible output of the classiﬁcation network is a separate entity, discrete from all others. 2. Continuous numeric functions. These functions describe the relationship between different sets of variables from a real physical or biological system. A few examples might include a recruitment index, a survey index of catch-per-unit effort, and a series of sea surface temperature anomalies. The following deﬁnition was provided by Swingler (1996): A continuous function is one for which the target outputs fall along a

Neural Networks Used in Classiﬁcation with Emphasis on Biological Populations

555

meaningful continuum where each possible output of the network has its place along the continuum. There may be many such continua for a multidimensional output space. A relatively large number of neural network application areas have been concerned with function mapping in contrast to classiﬁcation. Although a distinction is made between these two areas, it should be recognized that the two tasks can be implemented into a single neural framework, such as the multilayer feedforward network (MLFN), which will be described in further detail in a later section. Neural networks may also be dichotomized based on learning as follows: 1. Supervised networks. These networks classify patterns and make predictions and decisions according to other patterns of inputs and outputs which have been “learned” through training. They output the most reasonable answer based on the variety of learned patterns. In a supervised network, one must show the network how to make predictions, classiﬁcations, or decisions by providing it with a large number of correct inputs and outputs it can learn. Learning as used herein is deﬁned as the process of adapting or modifying connection weights in response to stimuli being provided at the input and optionally at the output buffer. The network gradually conﬁgures itself to achieve the desired input/output mapping. Such learning is usually some variation of three basic types: a. Hebbian learning, where a connection weight on an input path to a processing element (neuron) is incremented if both the input is high and the desired output is high, b. Delta rule learning, based on reducing the error between the actual output of a processing element and the desired output by modifying incoming connection weights, and c. Competitive learning, in which processing elements compete among themselves, and the one which yields the strongest response to a given input modiﬁes itself to become more like that input. 2. Unsupervised networks. Unsupervised networks can classify a set of training patterns without showing it correct outputs in the sample patterns. The Kohonen Self-Organizing Map (SOM) network (Kohonen, 1987) is the most popular type of unsupervised network. This network learns to make classiﬁcations by making its own clustering scheme for patterns. The patterns are clustered into categories based on proximity to each other. This network type is suitable for applications in which the outputs are not known. Although unsupervised networks do not classify as accurately as supervised networks, they are believed to be very useful

556

Saul B. Saila

in exploratory classiﬁcation and in reducing large and heterogeneous data sets into more parsimonious groups. The objectives of this report are to brieﬂy introduce neural networks (especially those with applications to biological classiﬁcation problems), to provide some examples of neural network classiﬁcation with emphasis on the biological sciences, to make some comparisons of neural networks with statistical classiﬁcations, to indicate some advantages and disadvantages of the neural network approach to biological classiﬁcation, and, ﬁnally, to suggest a role for neural network classiﬁcation in stock identiﬁcation work.

III. NEURAL NETWORK CLASSIFICATION APPLICATIONS Neural networks have been utilized in a wide variety of disciplines and applications. Some typical applications include prediction, process control, optimization, medical diagnostics, signal processing, robotics, pattern recognition, and classiﬁcation. A summary table from Maren (1992) includes applications of several network types. However, there are many more presently available. In this study, only neural network classiﬁcation with a few network types is considered. A search of aquatic sciences and ﬁsheries abstracts from 1992 to date (December, 2002) revealed a total of 415 references under the subject heading of “neural networks.” A search of Water Resources abstracts revealed considerably more references under the same topic heading. A selection from these and a few other sources (Index Medicus and some older references) was made, and those that contained an additional term “classiﬁcation” in the title or abstract are listed in Table 26-1. Only a limited number of references involving the physical sciences or other disciplines outside of biology are included in this table. However, the list of references to neural net classiﬁcation in the biological sciences is believed to be reasonably complete. It is clear that ﬁsheries and other aquatic sciences applications are a small proportion of the total number of references found in the search mentioned above. Eleven citations in Table 26-1 address classiﬁcation as related to some ﬁsh species. However, it is evident that neural network classiﬁcation has been applied to diverse objects and to many disciplines. This suggests that their potential for ﬁsh stock identiﬁcation may be considerably higher than indicated in the citations.

IV. SOME COMPARISONS OF NEURAL NETWORKS WITH STATISTICAL METHODS IN CLASSIFICATION The premise of White’s (1989) comprehensive review article is that learning procedures used to train neural networks are inherently statistical techniques. From

Neural Networks Used in Classiﬁcation with Emphasis on Biological Populations

557

TABLE 26-1. Some Applications of Neural Networks Classiﬁcation Methods, with Emphasis on the Biological Sciences

Discipline Geological Biological Biological Physical Biological Biological Physical Biological Biological community Biological Biological Biological Physical Biological Biological Physical Biological Capacity Biological Geological Biological Biological Biological Biological biological Biological Biological Biological Biological Biololgical Biological Physical Physical

Reference

Classiﬁed organism or object

Alexandrou and Pantzartzis (1993) Aurelle et al. (1999) Balfoort et al. (1992) Bankert (1994) Bankman et al. (1991) Bartone et al. (1996) Bischof et al. (1992) Brosse et al. (2001) Chon et al. (1996)

Seaﬂoor classiﬁcation Brown trout Algae Clouds EEG wave form Bacteria Images Fish assemblage classiﬁcation Benthic invertebrate

Eberhart et al. (1989) Grus and Zimmerman (1997) Hansen et al. (2001) Hara et al. (1995) Haralabous and Georgakarakos (1993) Haralabous and Georgakarakos (1996) Key et al. (1989) Lein (1995)

EEG classiﬁcation Antibody repertoire Stock identiﬁcation Sea ice Species identiﬁcation Species identiﬁcation Clouds Environmental carrying

Mitra et al. (1994) Ping and Chung (1995) Potter et al. (1991) Potter et al. (1993) Ramani and Patrick (1992) Rzempoluck (1997) Saila, S.B. (1998) Simpson et al. (1993) Smits et al. (1992) Stewart et al. (1994) Storbeck and Daan (2001) Wilson et al. (1991) Withler et al. (1994) Wu et al. (1994) Yhann and Simpson (1995)

Fingerprints Geological mapping Atlantic salmon scales Atlantic salmon scales Fish identiﬁcation EEG Blueﬁn tuna otoliths Dinophyceae Algae Sonar images Fish species EEG Chinook salmon Submerged objects Cloud segmentation

this, it follows that statistical theory can provide useful insights into the properties, advantages, and disadvantages of different neural network learning methods. White’s concluding remarks state that applications of neural network models to new and existing data sets hold the potential for fundamental advances in empirical understanding across a broad spectrum of sciences. To realize these advances,

558

Saul B. Saila

White believes that statistics and neural network modeling must work hand-inhand. It is with this important goal in mind that the following brief comparisons are made. There are several reported applications of and empirical comparisons between multilayer neural network classiﬁers and statistical classiﬁcation techniques. Some of these are brieﬂy indicated herein. Reibnegger et al. (1991) applied a linear discriminant function and a neural network technique to a medical classiﬁcation problem. Their comparison of methods was based on information theoretic concepts, and their conclusion was that the neural network approach (which employed a backpropagation algorithm) classiﬁed at least as well as the statistical techniques used for comparison, namely, linear discriminant analysis, statistical classiﬁcation (clustering), and regression trees. It was further stated that neural networks showed a unique ability to detect features held in the input data which were not explicitly formulated in the input. Another comparison between neural network and statistical methods is provided by Benediktsson et al. (1993), which involved a classiﬁcation of multisource remote sensing data. Their conclusions, also based on information theoretic reliability measures, reveal differences. However, they conclude that if the neural network could be trained by representative training samples, the three-layer network which was used would outperform the statistical methods. Ruck et al. (1990) proved that every fully connected feedforward network, when backpropagation trained as a classiﬁer, approximates the Bayes optimal discriminant function. The linear discriminant function is a form of Bayes rule that applies when the measurement vector from a multivariate normal distribution and all the groups are assumed to have identical covariance matrices. This suggests that the classiﬁcation problem can be completely solved by applying Bayes rule to obtain the minimum classiﬁcation error. A practical form of Bayes rule for optimal classiﬁcation is to assign to group i if P (X|Gi) P (Gi) > P (X|Gj) P(Gj) for all j π i, where P (X|Gi) is the conditional probability, the probability of getting a particular set of measurements X given that the object comes from group i. P (Gi) is the probability that the object comes from group i in the absence of any other information, and the same deﬁnitions apply to the quantities subscripted by j. The problem with Bayes rule that makes it virtually useless is the volume of data required to estimate P (X|Gi). For example, suppose one has measurements on 5 variables, each consisting of 10 categories. Then for even one group, it is required to estimate 50 relative frequencies, each requiring sample sizes of the order of 200. This level of sampling would have to occur for each group. It seems evident that using Bayes rule is not practical. Saila (1998) provides further information related to Bayes rule and the classiﬁcation problem. Walsh et al. (1996) found that the stepwise linear discriminant function did not outperform neural network classiﬁcation using the feedforward backpropagation network and a probabilistic neural network for polar cloud and

Neural Networks Used in Classiﬁcation with Emphasis on Biological Populations

559

surface classiﬁcation. Theoretical accuracy of the classiﬁers was determined using the bootstrap method. However, the overall accuracy differences among the methods were small. For data and comparisons more relevant to the subject matter of this chapter, Potter et al. (1991) concluded that their neural network application gave more accurate results in separating North American and European origin Atlantic salmon groups than discriminant analysis (85.8% correct vs. 80.7% correct). In their later study (Potter et al., 1993), it was demonstrated that a genetic algorithm was less successful than a neural network for the same classiﬁcation problem. In a study of farm-raised vs. native chinook salmon, Withler et al. (1994) achieved essentially equal classiﬁcation accuracies (99%) with discriminant analysis and a neural network. In addition to speciﬁc comparisons mentioned above, there are some general attributes of neural networks worth mentioning. They include the following. 1. Multisource Data Within the past decade, advances in computer, space, and marine technologies have made it possible to acquire large amounts of data about the marine environment, as well as data regarding individuals and populations. Therefore, there may be many kinds of data available from different sources which could contribute to some types of classiﬁcation problems. These kinds of data are termed multisource data. Our interest is in using these data to get higher accuracy in classiﬁcation. Conventional multivariate statistical classiﬁcation methods cannot be used satisfactorily in processing multisource data for several reasons. One is that multisource data may include nonnumerical data, since the data are multitype. These data also may not be in common units, and therefore, scaling problems may arise. Another problem with statistical classiﬁcation is that data sources may not be equally reliable. This means weighting data according to reliability, but most statistical classiﬁcation methods do not provide for this. The above suggests that alternatives to conventional multivariate classiﬁcation are needed to classify multisource data. Neural networks offer such an alternative. 2. Distributional Assumptions Neural networks are distribution free, and no prior knowledge is needed about the statistical distribution of the classes in the data source(s) in order to apply the method. Neural networks have an advantage over statistical classiﬁcation when no knowledge of the distribution functions exists or when the data are non-Gaussian. Indeed, neural networks have an inherent ability to model nonlinear problems without forcing modelers to make restrictive assumptions.

560

Saul B. Saila

3. Training Use of neural networks generally requires computationally intensive training. Most statistical classiﬁcation algorithms do not require such large amounts of time. However, this time requirement does not seriously affect classiﬁcation in biology because real time recognition of classes is not usually required. To perform well, neural network models must be trained by representative training samples. Although statistical classiﬁcation methods also require representative training samples, the latter tend not to be as sensitive to representative data as neural network models. This is a possible advantage of statistical models over neural networks. 4. Fault Tolerance Partial destruction of a network leads to a degradation of performance. However, some networks will tolerate considerable damage, in contrast to statistical classiﬁcation which fails completely when any program or input damage occurs. In summary, statistical classiﬁcation methods utilize a programming approach which follows rules, provides a formally speciﬁable solution, cannot generalize, and are not error tolerant. In contrast, the neural network classiﬁcation approach learns from data, follows rules that are not visible, is able to generalize, and tolerates some noise.

V. NEURAL NETWORK TYPES FOR CLASSIFICATION Most of the studies in Table 26-1 focused on classiﬁcation using supervised learning types of networks. Although there is a fairly large number of such learning types, the MLFN, using the backpropagation algorithm, is widely used. Backpropagation was ﬁrst derived by Rumelhart et al. (1986), and it is perhaps the most widely used option among the several types of network systems currently available. Strictly speaking, backpropagation is a learning algorithm and not a network type. However, it is frequently described as such. It is regrettable but common that the terminology of a rapidly evolving subject area such as neural networks has not yet stabilized.

A. SUPERVISED LEARNING: MLFN EXAMPLE As indicated earlier, a neural network consists of objects called neurons (nodes, processing elements) and weighted paths connecting these neurons. Each neuron has an activity represented by a real number. This activity value is computed as

Neural Networks Used in Classiﬁcation with Emphasis on Biological Populations

561

a nonlinear-bounded monotone-increasing function of a sum of the activities of the other nodes that are directly connected to it. The typical neural network described in Table 26-1 has three (rarely four or more) processing layers (sometimes called slabs): an input layer, one (sometimes more) hidden layer, and an output layer. A simple diagram of a three-layer network is shown in Figure 26-1. Using the notation and logic of Walsh et al. (1996), a brief description of a MLFN with backpropagation follows: Let the activity of node K be denoted by VK and a weight on a path from node L to node K be designated by WKL. Then Vk = f (SWKL V L )

(1)

where f is a nonlinear function. The establishment of the appropriate weights is termed learning. The neural network may be conceived as a nonlinear vector-valued function: O = f (I)

(2)

where O is a vector with one component for each activity of an output node, and I is a vector with one component for the activity of each input node. In supervised learning mode (such as the illustration), for each possible input vector I, an associated output vector O is speciﬁed. The function of the learning algorithm is to change the value of the weights so that f(I) is a good approximation of O.

Output buffer

............. .............

Hidden layer

Input buffer

.............

FIGURE 26-1. An example of a three-layer neural network.

562

Saul B. Saila

Backpropagation (Rumelhart et al., 1986) refers to the process of iteratively determining the weights WKL that locally minimize the global error E. N

E=

Â

O L - f (I)I

2

(3)

L =1

The algorithm is a special case of gradient search in which the weights are initialized as small random numbers and repeatedly updated at the ith iteration according to the rule: DW = - hE, WW +1 = Wn + DW

(4)

where W is the vector composed of the weights, E is the gradient of the global error, and h is the learning rate. Further details concerning this network type are provided by Hecht-Neilsen (1990) and by Gallant (1994).

B. UNSUPERVISED LEARNING: SOM EXAMPLE In supervised learning techniques, a training data set is presented to a network together with a desired network output. The network then learns to produce the correct output by virtue of the learning rules applied. In a sense, the network is trained to perform a task resembling discriminant analysis; that is, a given input data vector has to be transformed into an output vector representing the class membership of the input data vector. In unsupervised learning, a series of input vectors are presented to the network. However, no output vectors are provided. Instead, in a self-organizing process driven only by the input data and the learning scheme applied, the network adjusts its internal structure without reference to any external “teacher.” It can be shown that appropriately designed self-organizing networks are capable of performing tasks similar to cluster analysis or principal components analysis; that is, structures hidden in a complex array of input data can be visualized and extracted. The goal of this section is to brieﬂy introduce ﬁshery scientists without much prior knowledge of unsupervised learning to the Kohonen network, which is the most famous neural network model geared toward unsupervised learning. Selforganizing networks are believed to be an effective way for reducing the dimensionality of complex multivariate data sets by producing easily comprehensible “maps” of essential features. An example of results from such an approach is provided by Chon et al. (1996) in a study of benthic macroinvertebrate communities, which include spatiotemporal changes in community patterns. It is also believed that unsupervised classiﬁcation may be useful in other applications, such as exploratory analysis of stock mixtures. For example, if a sample of otoliths of

Neural Networks Used in Classiﬁcation with Emphasis on Biological Populations

563

adult Atlantic blueﬁn tuna from international waters in the middle Atlantic were examined for the trace elemental composition of the early growth stages of the otoliths, this approach could contribute to a better understanding of the Atlantic blueﬁn spawning ﬁdelity issue, which is unresolved at present. The Kohonen SOM exploit the idea of competitive learning. In this type of learning, the neurons compete for the privilege of learning, and only the one or at most a few neurons are allowed to adjust their weights in response to a presentation of input data. SOMs are designed to map or adaptively project input signals of arbitrary dimensionality onto a structured set of processing units in such a way that topological (i.e., neighborhood) relations of the input patterns and of the representation patterns are kept similar. The concept of the Kohonen SOM is illustrated in Figure 26-2. Note that this is a two-layer network. Assume that there is a linear array of k neurons. This number is chosen according to the problem under consideration. For example, if one were to explore the neighborhood relations in a data set consisting of four trace elements per case (individual otolith), one would use four input neurons. Next, assume that there is a second m ¥ n two-dimensional array of additional neurons. Each of the m ¥ n layer of neurons is connected with each of the input neurons, and the connection strength of each second-layer neuron j to each input neuron i is represented by a scalar weight factor w ( j,i). It is important to recognize that the neurons in the second layer are characterized by their physical position in the layer, and thereby, neighborhood relations are deﬁned for each of these neurons.

Input neurons

Information flow

Kohonen layer

FIGURE 26-2. Schematic of the basic structure of a Kohonen-type neural network (SOM). Note that, for simplicity, only a few of the connections between input neurons and second-layer neurons are shown.

564

Saul B. Saila

The weights (w[ j,i]) are ﬁrst assigned randomly. The input data must be preprocessed in a manner that assures that input values with greatly different numerical coefﬁcients are avoided. This can be achieved by computing the mean values and standard deviation for each variable and using these to calculate z scores. For example, a linear projection of the values onto the range 0.1–0.9 can be accomplished as follows: First, the minimum (xmin) and the maximum (xmax) are determined. Each value xold is then transformed by x new = ( x old - x min ) / ( x max - x min )

(5)

This linearly projects the values onto the interval [0, 1]. The ﬁnal step consists of multiplying these transformal values by 0.8 and adding 0.1. If a preprocessed input vector is sent through the network, corresponding activity occurs in the network. The activity ai of each second-layer neuron is given by the scalar product between the current vector w( j,i) and the input vector x(i). Each second-layer neuron computes a weighted sum: K

aj =

Â [ w( j, i) ¥ x i ]

(6)

i= 1

In this technique, the neuron responding maximally to a given input vector is chosen to be the neuron, the weight vector of which has the smallest Euclidean distance to the input vector. Thus, this neuron is allowed to “learn” after presentation of a given input vector. Learning is achieved by changing the weights of this maximally activated neuron in a way that the Euclidean distance between its weight vector and the input vector is further decremented. In other words, the new weight factor, W( j,i) of the neuron is given as W ( j, i)( t +1) = W ( j, i) + g(t) [ xi - w( j, i)t ]

(7)

where g(t) is the fractional increment of the correction. Typically, g(t) is large (near 1.0) for early steps of the learning process and then decreases continuously to approach 0.0. The ﬁnal result of this interactive process is that neurons in the second layer, which are in mutual physical proximity, will collectively respond only and maximally to input vectors which are “neighbors” in the multidimensional input data space. Thus, a mapping of a multidimensional data space onto a two-dimensional grid of neurons is achieved in a way that neighborhood relations remain preserved to a great extent. Although only a few references to unsupervised classiﬁcation were found, they indicated considerable promise for this approach in the context of biological classiﬁcation. Lebninger and Stancl (1992) compared the performance of the Kohonen network with k-nearest neighbor clustering and reported superior performance with the network model. The Kohonen network also equaled

Neural Networks Used in Classiﬁcation with Emphasis on Biological Populations

565

multilayer feedforward networks and discriminant analysis in their comparison. Reibnegger et al. (1993) also evaluated the SOM vs. some statistical clustering techniques. They found that the grouping of data by the Kohonen network was comparable to statistical clustering and principal component analysis. A recent reference (Brosse et al., 2001) describes the use of a nonsupervised network to analyze the spatial occupancy of several ﬁsh species in the lateral zone of a lake.

VI. DISCUSSION AND SUGGESTIONS Because classiﬁcation in a ﬁshery science context has frequently been associated with classical statistics, and especially discriminant analysis, it is signiﬁcant that each of the studies cited in Table 26-1 used neural network technologies, either instead of or along with classical statistical approaches, to develop classiﬁer systems. It is also signiﬁcant that this chapter on neural networks for biological classiﬁcation was solicited for inclusion for the ICES Stock Identiﬁcation Methods Working Group Report. This discussion and the suggestions herein are based on limited experience, and they reﬂect the author’s best judgment and opinions. These items are not listed in order of perceived importance. 1. Both neural networks and classical statistical approaches can be used in the development of classiﬁers, and both approaches are valid in the appropriate context. Just as some statistical approaches are better than others, some neural network approaches may occasionally be superior to a more classical approach to classiﬁcation. However, it is believed the relative merits of both should be critically evaluated on a case-by-case basis. 2. A primary advantage that classical statistical classiﬁcation methods have over neural network models is, in general, if the distribution functions of the classes are known, these methods perform very well. However, in some cases, for example, in multisource classiﬁcation, the distribution functions are not always known. Therefore, neural network models may be more appropriate. Neural networks also tend to have an advantage over classical statistical methods by their ability to model nonlinear problems better without forcing modelers to make restrictive assumptions. 3. The advice of competent statisticians should be sought when critically evaluating the performance of neural networks. Assessment of performance during learning (training) is particularly important for neural networks because decisions must be made as to when to stop training. The ultimate performance comparison of both statistical and neural network modes is equally important. Neither problem has

566

Saul B. Saila

TABLE 26-2. Some Names and Vendors of Commercially Available Software Suitable for Neural Network Classiﬁcation Product BrainMaker and BrainMake Pro/MMX Neuroshell2 Professional

Neural Works Professional

Neuro Solutions, V 3.02

MATLAB Neural Network Toolbox

Vendor California Scientiﬁc Software 10024 Newtown Road Nevada City, CA 95959 Ward Systems Group, Inc. Reception Park West 5 Hillcrest Drive Frederick, MD 21703 Neuralware, a subsidiary of Aspen Technology 202 Park West Drive Pittsburgh, PA 15075 Neuro Solutions 1800 North Main Street Suite D4 Gainesville, FL 32609 The Mathworks, Inc. 24 Prime Park Way Natick, MA 01960

Comments User friendly, contains complete documentation User friendly, contains 16 architectures and source code generator Professional development system, supports many paradigms Object-oriented simulation environment for neural network design and applications Powerful functions for network design and application

received adequate consideration to date, and progress will require cooperation between biologists and statisticians. 4. Each input and output of a neural network will vary along a continuous scale or be made up of a set of discrete counts. Each variable type requires a different coding method. There are several steps required to convert a set of raw data into a neural network training set. Among these are data scaling, dimensionality reduction, and outlier removal. 5. Some software and their vendors are listed in Table 26-2. This list is not complete but includes a sample giving some indication of the range of available systems and some indication of their capabilities. In summary, it has been demonstrated by diverse examples that neural networks are an efﬁcient method for discriminating groups and classifying individuals within those groups. They seem to be limited primarily by the quality of the data used to perform the analyses. Although the applications of neural networks in the published ﬁshery literature are still limited, it should be recognized that several interesting stock identiﬁcation studies using neural networks are in progress. These include work at the University of Connecticut by C. Crivello using microsatellite data and neural networks to classify larval and juvenile winter ﬂounder (Pseudopleuronectes americanus) populations. Somewhat similar work is also being done at New York University School of Medicine by I. Wirgin and colleagues. Another study using

Neural Networks Used in Classiﬁcation with Emphasis on Biological Populations

567

microelemental separation of larval winter ﬂounder stocks by inductively coupled plasma mass-spectrometry and neural networks is in progress by S. B. Moran and colleagues at the University of Rhode Island’s Graduate School of Oceanography. It seems clear that neural networks are here to stay as an important tool for stock identiﬁcation, and their increasing utilization by ﬁshery scientists provides evidence for this statement.

REFERENCES Alexandrou, D. and Pantzartzis, D. 1993. A methodology for acoustic sea ﬂoor classiﬁcation. IEEE J. Ocean Eng. 18(2): 81–86. Aurelle, D., Lek, S., Giraudel, J.-L., and Berrebi, P. 1999. Microsatellites and artiﬁcial neural network tools for the discrimination between natural and hatchery brown trout (Salmo trutte, L.) in Atlantic populations. Ecological Modelling 120: 313–324. Balfoort, H. W., Snock, T., Smits, J. K., Breedveld, L. W., Hofstreet, J. W., and Ruigelberg, J. 1992. Journal of Plankton Research 14(4): 575–589. Bankert, R. L. 1994. Cloud classiﬁcation of AVHRR imagery in maritime regions using a probabilistic neural network. J. of Applied Meteorology 33: 909–918. Bankman, I. N., Sigillito, V. G., Wise, R. A., and Smith, P. C. 1991. Detection of the EEG K-complex with neural networks. In Proc. IEEE Symposium on Computer-Based Model Systems. IEEE Computer Society Press, Baltimore, Maryland, pp. 280–287. Bartone, S., Gincomine, M., Ruggerio, M., Piccarito, C., and Calagore, L. 1996. Automated systems for identiﬁcation of hetrotrophic marine bacteria on the basis of their fatty acid composition. Appl. Environ. Microbiology 62(6): 2,122–22,1322 Benediktsson, J. A., Swain, P. H., and Ersoy, O. K. 1993. Conjugate-gradient neural networks in classiﬁcation of multisource and very-high-dimensional remote sensing data. International Journal of Remote Sensing 14: 2883–2903. Bischof, M., Schneider, M., and Ping, A. 1992. Multi-spectral classiﬁcation of Landsat images using neural networks. IEEE Trans. Geosci. and Remote Sensing 39(3): 482–490. Brosse, S., Giraudel, J.-L., and Lek, S. 2001. Utilization of non-supervised neural networks and principal component analysis to study ﬁsh assemblages. Ecological Modelling 141(1–3): 159–166. Chon, T.-S., Park, Y. S., Moon, K. H., and Cha, E. Y. 1996. Patternizing communities by using an artiﬁcial neural network. Ecological Modelling 90: 69–78. Dayhoff, J. E. 1990. Neural Network Architectures: An Introduction. Van Nostrand Reinhold, New York. Eberhart, R. C., Dobbins, R. W., and Webber, W. R. S. 1989. Case Net: A neural network tool for EEG waveform classiﬁcation. In Proc. IEEE Symposium on Computer-Based Medical Systems. IEEE Computer Society Press, Minneapolis, Minnesota, pp. 60–68. Gallant, S. I. 1994. Neural Network Learning and Expert Systems. The MIT Press, Cambridge, Massachusetts. Grus, F.-H. and Zimmerman, C. W. 1997. Identiﬁcation and classiﬁcation of autoantibody repertoires (Eastern blots) with a pattern recognition algorithm by an artiﬁcial neural network. Electrophoresis 18(7): 1,120–1,125. Hansen, M. M., Kenchington, E., and Nielsen, E. E. 2001. Assigning individual ﬁsh to populations using microsatellite DNA markers. Fish and Fisheries 2(2): 93–112. Hara, Y., Atkins, R. C., Shin, R. J., Kong, J. A., Yueh, S. H., and Wok, R. K. 1995. Application of neural networks for sea ice classiﬁcation in polarmetric SHR images. IEEE Trans. Geoscience Remote Sens. 33(3): 740–748.

568

Saul B. Saila

Haralabous, J. and Georgakarakos, S. 1993. Fish-school species identiﬁcation using a neural network. ICES Council Meeting Papers, Copenhagen, Denmark. ICES-CM-1993/B:8. Haralabous, J. and Georgakarakos, S. 1996. Artiﬁcial neural networks as a tool for species identiﬁcation of ﬁsh schools. Fisheries and Plankton Acoustics. ICES Journal of Marine Science 53(2): 173–180. Hecht-Neilsen, R. 1990. Neuro Computing. Addison Wesley, Reading, Massachusetts. Key, J., Maslanik, A., and Schweiger, A. J. 1989. Classiﬁcation of image AVHRR and SNMR Arctic data with neural networks. Photogramm. Eng. Remote Sens. 55: 1,331–1,338. Kohonen, T. 1987. Self-organization and Associative Memory. Springer-Verlag, Berlin. Lebninger, H. and Stancl, F. 1992. Comparing the performance of neural networks to well established methods of multivariate data analysis: The classiﬁcation of mass spectral data. Fresenius J. Anal. Chem. 344: 186–189. Lein, J. K. 1995. Mapping environmental carrying capacity using an artiﬁcial neural network: A ﬁrst experiment. Land. Degrad. Rehab. 6(1): 17–28. Maren, A. J. 1992. Introduction to neural networks. In A. J. Maren, C. T. Harston, and R. M. Pap (eds.), Handbook of Neural Computing Applications. Academic Press, San Diego, pp. 1–12. Mitra, S., Pal, S. K., and Kundu, M. K. 1994. Fingerprint classiﬁcation using a fuzzy multi-layer perceptron. Neural Computing and Applications 2(4): 227–233. Ping, A. and Chung, C.-J. F. 1995. Neural network approach for geological mapping: Technical background and case study. Can. J. Remote Sens/Can. Teledect. 20(2): 293–301. Potter, E. C. E., Kell, L., and Reddin, D. C. 1991. The use of a neural network to distinguish North American and European salmon (Salmo salar) using scale characteristics. International Council for the Exploration of the Sea, Anadromous and Catadromous Fish Committee C/M 1991/M:10. Potter, E. C. E., Kell, L., and Reddin, D. G. 1993. The discrimination of North American and European salmon using a genetic algorithm and by neural network. International Council for the Exploration of the Sea. Anadromous and Catadromous Fish Committee CM 1993/M: 13. Ramani, N. and Patrick, P. H. 1992. Fish detection and identiﬁcation using neural networks. Some laboratory results. IEEE Journal of Oceanic Engineering 17(4): 364–368. Reibnegger, G., Weiss, G., and Wachter, H. 1993. Self-organizing neural networks as a means of cluster analysis in clinical chemistry. Eur. J. of Clin. Biochem. 31: 311–316. Reibnegger, G., Weiss, L., Werner-Felmayer, G., Judmaier, G., and Wachter, H. 1991. Neural networks as a tool for utilizing laboratory information: comparison with linear discriminant analysis and with classiﬁcation and regression trees. Proc. Natl. Acad. Sci. USA 88: 11,426–11,430. Ruck, D. W., Rogers, S. K., Kabrisky, M., Oxley, M., and Suter, B. W. 1990. The multilayer perception as an approximation to a Bayes optimal discriminant function. IEEE Trans. Neural Networks 1(4): 296–298. Rumelhart, D. E., Hinton, G. E., and Williams, R. J. 1986. Learning representations by backpropagating errors. Nature 363: 533–536. Rzempoluck, E. J. 1997. Neural network classiﬁcation of EEG during camouﬂaged object identiﬁcation. Int. J. Med. Inf. 44(3): 69–75. Saila, S. B. 1998. Some methodologies related to stock identiﬁcation with special reference to Atlantic blueﬁn tuna. In J. Beckett (ed.), Proceedings of the ICCAT Symposium, San Miguel, The Azores, Portugal. Saila, S. B. 1996. Guide to some computerized artiﬁcial intelligence tools. In B. A. Megrey and E. Moksness (eds.), Computers in Fisheries Research. Chapman & Hall, London, pp. 8–40. Simpson, R., Culverhouse, P. F., Williams, R., and Ellis, R. 1993. Classiﬁcation of Dinophyceae by artiﬁcial neural networks. pp. 183–190. In J. J. Smayda and Y. Shimuzo (eds.), Toxic Phytoplankton Blooms in the Sea. Elsevier, Amsterdam, 183–190. Smith, M. 1993. Neural Networks for Statistical Modeling. Van Nostrand Reinhold, New York.

Neural Networks Used in Classiﬁcation with Emphasis on Biological Populations

569

Smits, J. R. M., Breedveld, L. W., Berkson, M. W. J., Rateman, G., Balfoort, H. W., Snock, J., and Hofstreet, J. W. 1992. Pattern classiﬁcation with artiﬁcial neural networks: Classiﬁcation of algae, based on ﬂow cytometer data. Anal. Chem. Acta 251(1): 11–25. Stewart, W. K., Jiang, M., and Marre, M. 1994. A neural network approach to classiﬁcation of sidescan sonar imagery from a mid-ocean ridge area. IEEE J. Ocean Eng. 19(2): 214–224. Storbeck, F. and Daan, B. 2001. Fish species recognition using computer vision and a neural network. Fisheries Research 51(2–3): 11–15. Svaerdstroem, A. 2003. Neural network feature vectors for sonar-type classiﬁcation. J. Acoust. Soc. Amer. 81(5): 2,656–2,665. Swingler, K. 1996. Applying Neural Networks: A Practical Guide. Academic Press, London. Walsh, R. M., Sangupta, S. K., Goroch, A. K., Rabindra, P., Rangaraj, N., and Navur, M. S. 1996. Polar cloud and surface classiﬁcation using AVHRR imagery: An intercomparison of methods. J. of Appl. Meteorology 31: 409–420. White, H. 1989. Learning in artiﬁcial neural networks: A statistical perspective. Neural Computing 1: 435–454. Wilson, K., Webber, W. R. S., Lesser, R. P., Fisher, R. S., Eberhart, R. C., and Dobbins, R. W. 1991. Detection of epileptiform spikes in the EEG using a pattern-independent neural network. In Proc. IEEE Symposium on Computer-Based Medical Systems. IEEE Computer Society Press, Baltimore, Maryland, pp. 264–271. Withler, R. E., Beacham, T. D., Watkins, R. F., and Stevens, T. A. 1994. Identiﬁcation of farm-reared and native chinook salmon (Oncorhynchus tshawytscha) on the east coast of Vancouver Island, British Columbia, using the nuclear DNA probe B2-2. Can. J. Fish. Aquatic Sci. S1(suppl. 1): 267–272. Wu, J., Smith, J. S., and Lucas, J. 1994. An underwater object classiﬁcation system using Fournier descriptors and neural networks. Underwater Technology 20(1): 25–31. Yhann, S. R. and Simpson, J. J. 1995. Application of neural networks to AVHRR cloud segmentation. IEEE Transactions in Geoscience and Remote Sensing 33(3): 590–604. Zurada, J. M. 1990. Introduction to Artiﬁcial Neural Systems. West Publishing Co., St. Paul, Minnesota.

This Page Intentionally Left Blank

CHAPTER

27

Maximum-Likelihood Estimation of Stock Composition JON BRODZIAK National Marine Fisheries Service, Northeast Fisheries Science Center, Woods Hole, Massachusetts, USA

I. II. III. IV. V.

Introduction Likelihood Formulation Extended-Likelihood Model Numerical Computation of MLEs Utility of the Likelihood Models A. A Simulation Study to Compare Two Approaches for Summarizing Stock Group Contributions B. Proportion Simplex Sampling Algorithm C. A Blind Test of the Standard Model D. Utility of the Extended Model VI. Discussion References

I. INTRODUCTION Stock composition can be a signiﬁcant uncertainty in the management of mixedstock ﬁsheries of highly migratory or transboundary resources. When a mixture of stocks with different productivities is harvested together, stocks with low productivity may be overharvested. Conversely, forgone yield may be substantial for high-productivity stocks if mixed-stock ﬁshing effort is set too low. In this context, mixed-stock ﬁsheries are similar to multispecies ﬁsheries where different life histories imply different target harvest rates across species. Last, allocation of harvest among user groups is often predicated upon knowing the stock composition of the catch. Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

571

572

Jon Brodziak

Measuring stock composition in mixed-stock ﬁsheries is often more challenging than determining species composition in multispecies ﬁsheries, since identifying species is generally much simpler than determining where ﬁsh come from. One of the key challenges to effective management of mixed-stock ﬁsheries is estimation of the relative contribution of each stock to the harvest. This is the focus of stock composition analysis. Stock composition analysis uses either markrecapture techniques or differences in the frequency distributions of population characteristics among stocks to estimate the composition of a mixture of stocks (Fournier et al., 1984; Millar, 1987; Pella and Milner, 1987). For example, statistical analyses of allozyme frequency distributions have routinely been used to estimate the contribution of Paciﬁc salmon stocks in mixed-stock ocean ﬁsheries (Carvalho and Hauser, 1994). Seasonal ﬂuctuations in the spatial distribution of stocks along with changes in the magnitude and distribution of ﬁshing effort can further complicate stock composition analysis.

II. LIKELIHOOD FORMULATION The structure of likelihood-based estimation depends on the likelihood principle. This principle states that all of the information obtainable from a ﬁxed set of observations about parameters of interest is contained in the likelihood function. Or alternatively, two or more likelihood functions contain the same information about the parameters of interest if they are proportional to each other given the data. A maximum-likelihood estimator (MLE) produces a parameter estimate that makes the observed data most probable with respect to a given likelihood function. Such estimators are founded on the principle that statistical inference should be based only upon observed data that can be used in the likelihood function. Given observed data (D), one can estimate a set of parameters (P) using the likelihood (L) of the observed data. The probability of observing the data conditioned on the parameters is the sampling density f(D|P). The likelihood of the parameters conditioned on the observed data is any function of P that is proportional to f(D|P): L(P D) μ f (D P)

(1)

While an MLE has the desirable theoretical properties of consistency, asymptotic efﬁciency, and asymptotic normality, it is important to recognize that these large sample properties do not necessarily hold for small sample sizes. Maximum-likelihood estimation of stock composition has been considered by several authors (Grant et al., 1980; Milner et al., 1981, 1985; Fournier et al., 1984; Millar, 1987; Pella and Milner, 1987; Brodziak, 1993). These references provide a more thorough exposition of the assumptions and historical develop-

573

Maximum-Likelihood Estimation of Stock Composition

ment of the statistical model than will be presented here. In this section, I formulate the likelihood function for a random sample of ﬁsh taken from a mixedstock ﬁshery for the purpose of stock composition analysis. The population characteristics used to estimate stock composition are a primary component of the likelihood framework. The chosen characteristics can be discrete or continuous random variables but must provide some differentiation among constituent stocks. It is important that the variables are measured in a consistent manner according to a standard protocol, especially if different facilities process samples (White and Shaklee, 1991). Inconsistency in the measurement of characteristics can lead to inaccurate estimates of stock composition (Brodziak et al., 1992), and quality assurance techniques may be needed to ensure consistency. Formal likelihood-based curvature methods can also be applied to evaluate the utility of alternative sets of population characteristics selected for stock composition analysis (Gomulkiewicz et al., 1990, Millar, 1991). In this case, the goal is to select a set of population characteristics that will produce a peaked likelihood surface for maximal discrimination among constituent stocks (Fig. 27-1). It is generally assumed that population characteristics used for maximum-likelihood estimation of stock composition are statistically independent (see, e.g., Fournier et al., 1984). Independence provides computational convenience because the joint probability density function (pdf) of the characteristics is then the product of the probability density functions of the individual characteristics. Of course, independence may not be a reasonable assumption when, for example, a particular set of characteristics are highly correlated within all constituent stocks. In this case, more intensive sampling would likely be needed to characterize the covariance structure within the constituent stocks. Alternatively, A

B

1

1 MLE

1 1

P2

P1

MLE P2

P1

0

0 FIGURE 27-1. Use of the curvature of the likelihood function to select sets of population characteristics for stock composition analysis. Well-deﬁned maximum-likelihood estimates (MLE) have a sharply peaked likelihood surface and high curvature (A). Poorly deﬁned MLEs have a ﬂat peak and low curvature (B).

574

Jon Brodziak

principal components analysis can be applied to a set of highly correlated characteristics to determine a reduced set of linearly independent combinations of these characteristics. If the characteristics have different measurement scales, they should be standardized prior to the principal components analysis. Two simplifying assumptions will be made without too great a loss in generality. First, it will be assumed that all population characteristics are independent multinomial random variables. Note that the distribution of any continuous population characteristic can be discretized to be a multinomial through the integration of its pdf over a suitable partition of its range so this assumption is not restrictive. Given n data points, a reasonable choice for the number of equiprobable multinomial cells (m) for discretization is roughly m = 2n0.4. This choice is based on Mann and Wald’s (1942) recommendation with an a = 0.05 signiﬁcance level (Moore, 1986). Second, it will be assumed that the distribution of each characteristic within each stock (the baseline data) has been determined a priori. This is, in effect, the best-case scenario where adequate data are available for all potentially contributing stocks. Given these assumptions, consider an independent and identically distributed random sample of N ﬁsh taken from the mixed-stock ﬁshery. Let Ci denote the set of measured characteristics of the ith ﬁsh. Further, let Pj be the unknown proportion of ﬁsh in the sample from the jth stock and let Xij be the conditional probability of observing the characteristics Ci if the ﬁsh had originated in the jth stock. The probability that the ith ﬁsh came from the jth stock is then PjXij, and the likelihood of observing the set of characteristics of the ith ﬁsh is S

LC i =

Â X ijPj

(2)

j= 1

Ignoring constants, the joint likelihood of observing the entire mixed-stock ﬁshery sample is N Ê S ˆ LC = ’ Á Â X ijP j ˜ Ë ¯ i= 1 j= 1

(3)

This will be called the standard-likelihood model.

III. EXTENDED-LIKELIHOOD MODEL Data used for stock composition analysis are either population characteristics that vary among stocks or artiﬁcial marks that identify stock of origin (e.g., Parker et al., 1990). Population characteristics used for stock identiﬁcation may reﬂect underlying genetic differences among stocks. Some characteristics, however, may

Maximum-Likelihood Estimation of Stock Composition

575

be inappropriate because they are not stable through time. The use of artiﬁcial marks may be useful for estimating contributions of selected stocks. However, these unique identiﬁers may be of limited utility for estimating the contributions of all stocks to a mixed-stock ﬁshery unless extensive (and expensive) marking programs are conducted. In this section, an estimation approach that combines population characteristics and unique identiﬁers for stock composition analysis is described (Brodziak, 1993). This is a general approach to address the uncertainty caused by a lack of detectable genetic and/or phenotypic separation among constituent stocks. In particular, this approach can be used to help discriminate stocks that have similar population frequency characteristics. In this case, artiﬁcial marks are applied to improve estimation accuracy for problem clusters consisting of stocks with similar characteristics. In brief, I describe the structure of the maximum-likelihood estimation model and show how it can be used to resolve problem clusters. Application of the extended-likelihood model requires marking data that uniquely identiﬁes the stock of origin of a marked ﬁsh. For the extended model, it is assumed that the process of ﬁsh marking is independent of the population characteristics. It is also assumed that a ﬁxed proportion of ﬁsh are marked. Let Fj be the frequency of observing a mark within the jth stock and assume that, for all stocks, Fj is known and Fj < 1. The use of a marking frequency formally requires that the stock be sampled for marks with replacement, and contrasts the use of a hypergeometric model for sampling marks from a ﬁnite population without replacement. However, it is assumed that the total number of ﬁsh captured in the mixed-stock ﬁshery is far greater than the number of ﬁsh sampled from the mixture so that the distinction between sampling with and without replacement becomes unimportant (cf. Geiger, 1990). To obtain the extended-likelihood model, consider an independent and identically distributed random sample of N ﬁsh taken from the mixed-stock ﬁshery. Assume that each ﬁsh has been examined for a mark in addition to having its population characteristics measured. Given the marking frequencies, the probability of observing a mark from stock j is FjPj and the probability of observing a mark on a given ﬁsh, denoted by F, is S

F=

Â F jP j

(4)

j= 1

Whenever a mark is detected, it is assumed to be interpreted without error. Suppose that there are a total of Mj marked ﬁsh from the jth stock in the mixture. Let M and U be the number of marked and unmarked ﬁsh in the mixture sample, respectively. The likelihood of observing M marked ﬁsh and U unmarked ﬁsh, LM, is a product of N multinomial trials where either a mark from one of the S stocks is observed or no mark is observed. This gives

576

Jon Brodziak

U

S

L M = (1 - F) ◊ ’ (F jP j )

Mj

(5)

j= 1

Because marking is independent of the measurement of the population characteristics, the joint likelihood of the extended model, L, is simply the product of LC and LM, so that L = LC · LM.

IV. NUMERICAL COMPUTATION OF MLEs To ﬁnd the vector of stock contributions that maximizes the standard or extended mixture likelihood, PMAX, two iterative schemes are presented. It should be noted, however, that other methods of numerical solution can be applied (cf. Fournier et al., 1984; Pella and Milner, 1987). The method for numerical solution of the standard model is based on the expectation maximization (EM) algorithm (Dempster et al., 1977). This method produces a sequence of estimates, for example, P(1), P(2), P(3), . . . , that eventually converge to PMAX due to the likelihood increasing property of the EM algorithm. An initial condition P(0) for the sequence of estimates is needed and a natural choice is Pj(0) = 1/S for all j. The iterative equations to compute P(n+1) given P(n) are, for j = 1 to S,

P (j n +1)

È ˘ Í N X P ( n) ˙ 1 ij j ˙ = ÍÂ S ˙ N Í i= 1 ( ) Í Â X isPs n ˙ Î s=1 ˚

(6)

Each iteration of eq. 6 sets the next estimate of the jth stock contribution to be the mixture sample mean of the conditional probability that each ﬁsh originated in the jth stock based on the current contribution estimates. To calculate PMAX for the extended model, I use an iterative scheme which is analogous to the method of “successive substitutions” (Hasseblad, 1969). I derive this scheme by calculating the Lagrange multiplier associated with the constraint that stock contributions sum to 1 and applying the result to the necessary conditions for a critical point of the extended-likelihood model (cf. Everitt and Hand, 1981). This leads to iterative equations to compute P(n+1) as

P (j n +1) =

where j = 1, . . . , S.

1 - F ( n) N + M - 2N ◊ F ( n)

È ˘ Í N F jP (j n) X ijP (j n) ˙ ÍM j ˙ +Â Í ˙ 1 - F ( n) i = 1 S X isPs( n) ˙ Í Â Î ˚ s=1

(7)

Maximum-Likelihood Estimation of Stock Composition

577

A natural choice for an initial condition P(0) is P (j 0) =

Mj U + N S ◊N

(8)

The two iterative schemes are related in that, if no marked ﬁsh are found in the mixture sample and the probability of observing a marked ﬁsh is very small (F(n) ª 0), then eq. 7 is equivalent to eq. 6. The iterative formulae in eq. 7 are derived from ﬁrst-order conditions to maximize the extended likelihood and are equivalent to those obtained if the EM algorithm was applied to this case.

V. UTILITY OF THE LIKELIHOOD MODELS The standard-likelihood model has been reported to perform better than alternatives based on linear regression (Mulligan et al., 1988) and classiﬁcation methods (Millar, 1987, 1990; Wood et al., 1987, appendix). These studies used Monte Carlo simulation to evaluate the accuracy and precision of alternative estimators for stock composition analysis. Overall, the conclusion was that the maximum-likelihood estimator performed better than the alternatives.

A. A SIMULATION STUDY TO COMPARE TWO APPROACHES SUMMARIZING STOCK GROUP CONTRIBUTIONS

FOR

One important use of stock composition analysis is to calculate catch statistics for a mixed-stock ﬁshery. Often this requires summarizing stock contributions by one or more groups of stocks. For example, it may be necessary to estimate the total catch of Atlantic salmon originating in Canadian rivers in the mixed-stock subsistence ﬁshery off of Greenland. In this case, estimates of the contribution from the group of Canadian stocks might be constructed either from individual estimates for each stock or from an aggregate estimate for the entire group. Two general methods of computing stock group contribution estimates have been proposed for genetic stock identiﬁcation: the pool and allocate procedure and the allocate and sum procedure (Wood et al., 1987). In the context of genetic stock identiﬁcation, the pool and allocate procedure averages the allele frequencies within a group to produce a composite stock. The composite stock’s allele frequencies can be pooled using either weighted or unweighted average, where the weighting factor is proportional to the baseline sample size from each stock. In this case, the composite stock is assumed to adequately represent the genetic proﬁle of the entire stock group. A stock contribution is then estimated for each composite stock to complete the pool and allocate procedure. In contrast,

578

Jon Brodziak

the allocate and sum procedure estimates stock contributions for each individual stock. These estimates are then summed to estimate stock group contributions. The question of whether to use the pool and allocate or the allocate and sum procedure arises in the application of stock composition analysis to mixed-stock ﬁsheries. Wood et al. (1987) addressed this question in a general context with simulation studies. They recommended the allocate and sum procedure for two reasons. First, it was easier to implement since no composite stock group needed to be formed before the mixture analysis. Second, it was more ﬂexible since contribution estimates can be summed at various grouping levels. Gall et al. (1989) also addressed the question of whether to use the pool and allocate or the allocate and sum procedure. They concluded that the answer depended on the particular mixed-stock ﬁshery application. In practice, the answer depends on which procedure performs best for summarizing stock groups for ﬁsheries management. To address this question for stock composition analyses of chinook salmon ﬁsheries south of Cape Falcon, Oregon, I conducted a series of simulation experiments (Brodziak, 1990) using genetic data reported in Gall et al. (1989) and Bartley et al. (1992). For the purpose of describing how the allocate and sum and the pool and allocate procedures were applied, consider a particular stock group consisting of C stocks. Let K be the indicator set for this group; that is, if the integer j is in K, then the jth stock is in the stock group. Let qAS denote the contribution estimate of the stock group using the allocate and sum procedure, and let qSA denote the contribution estimate of the stock group using the pool and allocate procedure. Suppose that a total of L genetic loci are measured and let fjk be the vector of allele frequencies for the jth stock at the kth locus. The baseline data for the jth stock is its set of L allele frequency vectors Fj = {fj1, fj2, . . . , fjL}. The allocate and sum procedure applies the baseline data for the individual stocks in the stock group to compute contribution estimates for each stock; that is, for each j in K, a stock contribution estimate qj is computed based on the individual stock’s baseline data Fj. These estimates are then summed to compute qAS: qAS =

Â qj

(9)

jŒ K

In contrast, the pool and allocate procedure computes the baseline data for a composite stock F = {f1, f2, . . . , fL}, where fk denotes the allele frequency vector for the composite stock at the kth locus. Unweighted averaging was used to compute each vector in F where fk =

1 C

Â f jk

(10)

j ŒK

The stock group contribution estimate qSA is then computed using the composite stock baseline data F.

579

Maximum-Likelihood Estimation of Stock Composition

TABLE 27-1. Loci Used in Simulation Study to Evaluate Relative Performance of Stock Group Summarizing Proceduresa Locus acronym AAT1, AAT2, AAT3, mAAT1, mAAT2 ADA1 AH1, mAH1, mAH2, mAH3, mAH4 CK4 DPEP1 GPI2, GPIH GR HAGH IDH2, IDH3, IDH4 LDH4, LDH5 mDH1-2, mDH3-4, mDHP1, mDHP2, mMDH1, mMDH2 MPI PDPEP2 PEPLT PGDH PGM1, PGM2 SOD1 TAPEP1 TPI3, TPI4

Enzyme system Aspartate aminotransferase Adenosine deaminase Aconitate hydratase Creatine kinase Glycyl leucine peptidase Glucose phospate isomerase Glutathione reductase Hydroacylglutathione hydrolase Isocitrate dehydrogenase Lactate dehydrogenase Malate dehydrogenase Mannose phosphate isomerase Phenylalanyl proline peptidase Prolyl leucine peptidase Phosphoglucokinase Phosphoglucomutase Superoxide dismutase Leucylglycylglycine peptidase Triosephosphate isomerase

a

Locus acronyms follow those used by Gall et al. (1989) and Bartley et al. (1992).

Both summarizing procedures were applied to simulated mixed-stock ocean chinook salmon ﬁshery samples to compare their performance. The baseline data consisted of 38 polymorphic loci (Table 27-1) measured for each of 45 stocks (Table 27-2) reported in Gall et al. (1989). The 45 stocks were grouped into 10 stock groups based on genetic identity analyses (Table 27-2). Stock composition estimates were computed using the expectation-maximization algorithm. Average maximum absolute estimation errors for stock group contributions were used to compare the relative performance of the allocate and sum and the pool and allocate procedures for estimating the contributions of the 10 stock groups. Average estimation errors for the three California chinook stock groups were also evaluated to compare the performance of the two procedures for three important ﬁshery management units south of Cape Falcon, Oregon. There were no a priori assumptions made about the possible mixture composition of the chinook ﬁshery. Instead, proportion simplex sampling was used to evaluate the performance of the two summarizing procedures over all possible ﬁshery compositions. Proportion simplex sampling provides a method to compute expectations over the set of possible mixed-stock ﬁshery compositions. This approach can be used

580

Jon Brodziak

TABLE 27-2. Stock Groups Used in Simulation Study to Evaluate Relative Performance of Stock Group Summarizing Proceduresa Stock group Sacramento/San Joaquin Rivers Eel/California Coastal Rivers

Klamath/Trinity Rivers

Smith/Southern Oregon Rivers

Mid-Oregon Rivers Northern Oregon Rivers Upper Columbia River Spring Upper Columbia River Fall Lower Columbia River Spring Lower Columbia River Fall

Source stocks within group Coleman hatchery, Feather River hatchery, Upper Sacramento River, Merced River hatchery, Nimbus hatchery Middle Fork Eel River, South Fork Eel River, Salmon Creek, Mattole River, Redwood Creek (South Fork Eel River), Hollow Tree Creek, Van Duzen River, Mad River hatchery, North Fork Mad River, Redwood Creek, Redwood Creek L agoon Salmon River, Irongate hatchery, Bogus Creek, Camp Creek, Horse Linto Creek, Shasta River, Trinity River hatchery, South Fork Trinity River Blue Creek, Omagar Creek, Rowdy Creek hatchery, Middle Fork Smith River, Chetco River hatchery, Rogue River Applegate River, Rock Creek hatchery Elk River hatchery, South Fork Coquille River, Millacoma River, Morgan Creek hatchery Fall Creek hatchery, Trask River Tucannon hatchery, Leavenworth hatchery Priest Rapids, Lyons Ferry Cowlitz River, Lewis River Spring Creek

a

Genetic identity analyses to form stock groups are reported in Gall et al. (1989) and Bartley et al. (1992).

to address uncertainty in stock composition by enabling the analyst to consider the set of all possible mixed-stock ﬁshery compositions or a subset. The constraints on the mixing proportions are S

Â Pj = 1

and P j ≥ 0

(11)

j= 1

These constraints deﬁne the proportion simplex D, which is an (S-1) dimensional linear subspace of RS. The proportion simplex contains all possible combinations of mixed-stock ﬁshery composition. In general, any mixed stock ﬁshery composition can be represented as an S-dimensional vector P = (P1, P2, . . . , PS) whose elements satisfy eq. 1. Given a uniform sampling grid D* of the simplex D consisting of N(D*) vectors, the (unconditional) expected value of any quantity D that depends on P can be approximated by a sum over the sampling grid

581

Maximum-Likelihood Estimation of Stock Composition

E [D] = Ú D(P)dP ª D

1

Â

N (D* ) P ŒD *

D(P)

(12)

Here the quantity D could be a measure of estimation accuracy across stocks. Alternatively it might represent a ﬁshery performance measure, such as landings, or a risk measure such as the bycatch of a rare or endangered ﬁsh stock. Regardless of the application, the construction of a uniform sampling grid D* is the key to numerically calculating the expectation in eq. 12. The algorithm for constructing a uniform sampling grid D* is related to a problem in combinatorial theory (Brodziak, 1990). Without loss of generality, assume that the sampling grid consists of discrete points (vectors in RS) and that neighboring points are equally spaced. The fundamental insight is that the sampling grid can be constructed from an enumeration of all possible compositions of a positive integer I into S parts. Here the composition of a positive integer I into S parts is an S-dimensional vector composed of nonnegative integers that sum to I. For example, consider the ﬁve compositions of the integer 4 into 2 parts. These are: 4 = 4 + 0; 4 = 0 + 4; 4 = 3 + 1; 4 = 1 + 3; 4 = 2 + 2. These ﬁve compositions correspond to ﬁve 2-dimensional vectors: (4,0), (0,4), (3,1), (1,3), and (2,2). Upon division by I = 4, these vectors produce ﬁve points: (1,0), (0,1), (0.75,0.25), (0.25,0.75), and (0.5,0.5). These points are the sampling grid D* for two stocks with a mesh of 0.25 (Fig. 27-2). To specify the algorithm, let R denote an S part composition of I, where the elements of vector R = (Rs, R2, . . . , RI) satisfy I = Rs + R2 + . . . + RI. The algorithm to construct the sampling grid D* consists of eight steps (with comments in parentheses).

B. PROPORTION SIMPLEX SAMPLING ALGORITHM 1. 2. 3. 4. 5. 6. 7. 8.

Set R1 = I and Rk = 0 for k = 2, 3, . . . , S (Initialize the composition) Set h = min{k | k in {1, 2, . . . , S} and Rk π 0} (Compute the index) Set keep = Rh (Store indexed element) Set Rh = 0 (Zero indexed element) Set R1 = keep - 1 (Use Rh to compute new R1) Set Rh+1 = Rh+1 + 1 (Increase next element) Set Pk = Rk/I for k = 1, 2, . . . , S (Compute new grid point P) If RS π I, then return to (step 2); else stop. (Stop when RS = I)

Each iteration produces one new point P in the sampling grid D*. Computations with this point P (as in eq. 12) can be executed after step 7 prior to the end of the loop at step 8. The total number of points generated is J(I,S), the number of S compositions of I, where

582

Jon Brodziak

Stock contribution P2

1.00

D 0.75

0.50

0.25

0.00 0.00

D

*

0.25

0.50

0.75

1.00

Stock contribution P1 FIGURE 27-2. Proportion simplex D for 2 stocks and discrete sampling grid D* with a mesh of 0.25.

J (I, S) =

(I + S - 1)!

I! (S - 1)!

(13)

This gives an exact count of the number of compositions in the grid which can help determine the computational cost of a given simulation. The difference between elements of neighboring sampling points is 1/I; this is the mesh of the sampling grid. To construct a sampling grid with a particular mesh size, say 1/K for ﬁxed S, one can set I = K - S. For the simulation to compare stock group summarizing procedures, proportion simplex sampling was conducted with I = 2 and S = 10. For each point, a total of 10 simulated mixed-stock ﬁsheries were generated by randomly assigning stock group and individual stock of origin based on the stock composition at that point. Allelic frequencies were then assigned to simulated ﬁsh based on their stock of origin for the allocate and sum procedure and based on their stock group of origin for the pool and allocate procedure. To assess the effect of mixture sample size on both procedures, mixture samples of 50, 100, 250, and 500 ﬁsh were generated. Thus, for each mixture sample size, a total of 550 separate mixtures were simulated for stock composition analysis. Simulation results show that both summarizing procedures produce accurate estimates of chinook salmon stock group contributions (Fig. 27-3). Although the

Maximum absolute error

0.08

0.06

Pool and allocate Allocate and sum

0.04

0.02

0.00 0

100

200

300

400

500

600

500

600

Mixed-stock fishery sample size Standard error of maximum absolute error

0.035

0.030 Pool and allocate Allocate and sum

0.025

0.020

0.015

0.010

0.005

0.000 0

100

200

300

400

Mixed-stock fishery sample size FIGURE 27-3. Comparison of accuracy and precision of stock group summarizing procedures for mixed-stock chinook salmon ﬁsheries.

583

584

Jon Brodziak

pool and allocate procedure is slightly more accurate than the allocate and sum procedure at low to moderate mixed-stock ﬁshery sample sizes, the two procedures have comparable precision over the range of simulated data (Fig. 27-3). In contrast, the pool and allocate procedure provides more accurate estimates of the contributions of California-origin stock groups (Fig. 27-4). While both procedures produce accurate estimates of the important Sacramento/San Joaquin stock group, the allocate and sum procedure overestimates the Klamath/Trinity and California coastal stock group contributions. This suggests that the pool and allocate procedure would be preferable to the allocate and sum procedure for estimating stock group contributions of major California chinook salmon stock groups for ﬁsheries management.

C. A BLIND TEST

OF THE

STANDARD MODEL

In a cross-validation study based on coded-wire tagged chinook salmon with a known stock of origin, Brodziak et al. (1992) found that stock group contribution estimates were very accurate using the standard model. This blind test was conducted on a mixed-stock ﬁshery sample of 220 ﬁsh using 35 polymorphic allozyme loci, 37 baseline samples from source stocks, and ﬁve groups of source stocks (Sacramento River, Coastal California rivers, Klamath River, Southern Oregon rivers, and Mid-Oregon rivers). To ensure the integrity of the test, stock composition estimates were calculated without any prior knowledge of the mixed-stock ﬁshery sample. This test demonstrated that the standard model could produce accurate stock contribution estimates for chinook salmon using actual ﬁsheries data. In particular, maximum stock group estimation errors were very small for the allocate and sum (3.8%) and pool and allocate (2.3%) procedures (Table 27-3). In similar analyses using coded-wire tagged ﬁsh, Brodziak et al. (1992) found the estimation accuracy depended on the concordance between the stocks present in the baseline data and the mixture sample and on the number of population characteristics used in the analysis. Overall, the utility of the standard model has been well established.

D. UTILITY

OF THE

EXTENDED MODEL

Brodziak (1993) examined the potential utility of the extended model and found that it substantially reduced estimation error in simulation studies based on genetic data reported in Wood (1989). Although the extended model can be shown to produce asymptotic variance estimates of stock contributions that are no larger than those for the standard model, further work to evaluate the utility of this method would be desirable. An example of applying this estimation

Average stock group contribution error

0.015

(A) Pool and allocate 0.010

0.005

0.000

-0.005 Sacramento Klamath Coastal

-0.010

-0.015 0

100

200

300

400

500

600

Mixed-stock fishery sample size 0.015

Average stock group contribution error

(B) Allocate and sum 0.010

0.005

0.000

-0.005 Sacramento Klamath Coastal

-0.010

-0.015 0

100

200

300

400

500

600

Mixed-stock fishery sample size FIGURE 27-4. Comparison of accuracy of (A) pool and allocate and (B) allocate and sum procedures for major California chinook salmon stock groups.

585

586

Jon Brodziak TABLE 27-3. Results of a Blind Cross-Validation Test of the Accuracy of Maximum Likelihood Estimates of Stock Composition Using 35 Polymorphic Loci to Compute the Contributions of 5 Stock Groups to a Mixed-Stock Fishery Sample Comprised of 220 Chinook Salmon Marked with Coded-Wire Tagsa (Brodziak et al. 1992) Stock group

Actual contribution

Estimate (± SD)

Error

Using the allocate and sum procedure Sacramento 0.223 0.246 Coastal 0.036 0.048 Klamath 0.486 0.498 S. Oregon 0.246 0.208 Mid-Oregon 0.009 0.000

(±0.162) (±0.079) (±0.050) (±0.124) (±0.131)

0.024 0.012 0.012 -0.038 -0.009

Using the pool and allocate procedure Sacramentob 0.219 0.242 Coastal 0.037 0.038 Klamath 0.489 0.485 S. Oregon 0.247 0.236 Mid-Oregon 0.009 0.000

(±0.028) (±0.021) (±0.051) (±0.037) (±0.031)

0.023 0.001 -0.004 -0.011 -0.009

a

From Brodziak et al., 1992. One sample from the Sacramento stock group had an improbable genotype relative to the pool and allocate baseline data. This ﬁsh was excluded from the stock composition analysis.

b

approach to a hypothetical mixed-stock salmon ﬁshery is reported in Brodziak (1993). In that example, simulated allelic data were taken from Wood (1989) who deﬁned a set of hypothetical mixtures of ﬁve stocks (A, B, C, D, and E) where stocks B, C, and D form a problem cluster with similar population characteristics (allele frequencies). Wood’s mixtures included only stocks A, B, and C; thus, D and E were present in the baseline data but not in the mixture. To resolve the problem cluster, stock C was marked with marking fractions of 0%, 10%, 20%, . . . , 100%. A total of 100 simulations were conducted for each combination of marking fraction and mixture composition. For each simulation, baseline data sets of 100 ﬁsh were simulated for each of the ﬁve stocks using allele frequency data in Wood (1989, Table 27-1). Similarly, a mixture sample of 300 ﬁsh was simulated using the mixture compositions in Wood (1989, Table 27-2). The results of applying the estimation approach were reported as mean angular deviation of the estimates from the true stock composition (Wood et al., 1987). In this case, a larger angular deviation corresponds to a larger stock composition estimation error.

Maximum-Likelihood Estimation of Stock Composition

587

The simulation results showed that marking stock C can appreciably improve the accuracy and precision of stock composition estimates (see Fig. 1, Brodziak, 1993). This was true even where stock C was not present in the mixture because in this case, the lack of recovery of marks from stock C reduces the likelihood that the stock was actually present in the mixture. Diminishing returns in the reduction of estimation error as marking frequency increased also suggested the potential for optimizing the marking frequency to achieve an acceptable estimation error.

VI. DISCUSSION Maximum-likelihood estimators have become a standard tool for stock composition analysis. Evaluating the quality of estimators and data is an important ongoing process for stock assessment. In particular, demonstrating the accuracy of stock composition analysis is vital for providing conﬁdence in the results for ﬁsheries management (Brodziak et al., 1992). Objective identiﬁcation of potentially contributing stocks to a mixed-stock ﬁshery through tagging or more formal stock composition analysis is another important step. In some cases, it may be possible to reduce the number of stocks used for stock composition analysis based on auxiliary information that certain stocks do not contribute to a particular mixed-stock ﬁshery. In general, it is important to be able to assess the consequences of management alternatives under stock composition uncertainty. This is especially true when the composition of the mixed-stock ﬁshery varies substantially. In such cases, proportion simplex sampling can be a useful tool for calculating the expected accuracy of an estimator or a ﬁshery performance measure. In my example, the expected performance of the pool and allocate procedure for estimating stock group contributions was moderately better than the allocate and sum procedure based on all stock groups and averaged over the entire set of possible mixed-stock chinook salmon ﬁsheries. However, the expected performance for estimating contributions of key California stock groups, such as the Klamath River stocks (see Fraidenburg and Lincoln, 1985), was more accurate using the pool and allocate procedure. This example underscores the need to critically examine the choice of summarizing procedure for management applications of stock composition analysis. Besides estimation methodology, it is also important to evaluate the quality of basic data used for stock composition analysis of mixed-stock ﬁsheries. The selection of population characteristics for stock composition analysis can be performed in an objective manner (Gomulkiewicz et al., 1990; Millar, 1991). The expected estimation performance of alternative sets of population characteristics can be rigorously evaluated using proportion simplex sampling and Monte Carlo

588

Jon Brodziak

simulation. In my example, a large set of genetic loci were used (38) for simulated stock composition analyses. In practice, it is important to choose the smallest set of loci that can reliably estimate key stock group contributions to minimize costs. I advocate using simulation and maximum likelihood-based measures of curvature to identify the best set of characteristics for stock composition analysis. In some cases, increasing the information base for stock composition analysis may be necessary to achieve management goals due to a lack of detectable genetic and/or phenotypic stock separation. Although further investigation of population genetic structure is recommended in such cases, the existence of high levels of gene ﬂow among constituent stocks may preclude genetic separation (see Waples, 1998). In the extended model example, using artiﬁcial marks resolved a problem cluster of stocks with similar population characteristics. In general, this approach can be used to improve the resolution of stock composition analysis. In particular, the extended model could be applied to estimate the relative contributions of wild vs. hatchery-origin salmon in mixed-stock Paciﬁc salmon ﬁsheries where ﬁsh marking programs and genetic stock identiﬁcation have become an essential component of ﬁsheries management (Hilborn et al., 1990; Carvalho and Hauser, 1994).

REFERENCES Bartley, D., Bentley, B., Brodziak, J., Gomulkiewicz, R., Mangel, M., and Gall, G. A. E. 1992. Geographic variation in population genetic structure of chinook salmon from California and Oregon. Fish. Bull., U.S. 90: 77–100. Brodziak, J. 1990. Theoretical aspects of genetic stock identiﬁcation. Ph.D. dissertation, University of California. Davis, California. Brodziak, J., Bentley, B., Bartley, D., Gall, G. A. E., Gomulkiewicz, R., and Mangel, M. 1992. Tests of genetic stock identiﬁcation using coded wire tagged ﬁsh. Can. J. Fish. Aquat. Sci. 49: 1507–1517. Brodziak, J. 1993. An extension of stock composition analysis to include marking data. Can. J. Fish. Aquat. Sci. 50: 251–257. Carvalho, G. and Hauser, L. 1994. Molecular genetics and the stock concept in ﬁsheries. Rev. Fish. Biol. Fisheries. 4: 326–350. Dempster, A. P., Laird, N. M., and Rubin, D. B. 1977. Maximum likelihood from incomplete data via the EM algorithm. J. R. Stat. Soc. Ser. B 39: 1–38. Everitt, B. S. and Hand, D. J. 1981. Finite Mixture Distributions. Chapman & Hall, London. Fournier, D., Beacham, T. D., Riddell, B. E., and Busack, C. A. 1984. Estimating stock composition in mixed-stock ﬁsheries using morphometric, meristic, and electrophoretic characteristics. Can. J. Fish. Aquat. Sci. 41: 400–408. Fraidenburg, M. and Lincoln, R. 1985. Wild chinook salmon management: an international conservation challenge. N. Amer. J. Fish. Man. 5: 311–329. Gall, G. A. E., Bentley, B., Panattoni, C., Childs, E., Qi, C., Fox, S., Mangel, M., Brodziak, J., and Gomulkiewicz, R. 1989. Chinook mixed ﬁshery project 1986–1989. California Department of Fish and Game, Rancho Cordova, California. 192 pp. Geiger, H. J. 1990. Parametric bootstrap conﬁdence intervals for estimating contributions to ﬁsheries from marked salmon populations. Am. Fish. Soc. Symp. 7: 667–676.

Maximum-Likelihood Estimation of Stock Composition

589

Gomulkiewicz, R. S., Brodziak, J. K. T., and Mangel, M. 1990. Ranking loci for genetic stock identiﬁcation by curvature methods. Can. J. Fish. Aquat. Sci. 47: 611–619. Grant, W. S., Milner, G. B., Krasnowski, P., and Utter, F. M. 1980. Use of biochemical genetic variants for identiﬁcation of sockeye salmon (Oncorhynchus nerka) in Cook Inlet, Alaska. Can. J. Fish. Aquat. Sci. 37: 1236–1247. Hasseblad, V. 1969. Estimation of ﬁnite mixtures of distributions from the exponential family. J. Am. Stat. Assoc. 64: 1459–1471. Hilborn, R., Walters, C., and Jester, Jr., D. 1990. Value of ﬁsh marking in ﬁsheries management. Amer. Fish. Soc. Symp. 7: 5–7. Mann, H. B. and Wald, A. 1942. On the choice of the number of class intervals in the application of the chi-squared test. Ann. Math. Statist. 13: 306–317. Millar, R. 1987. Maximum likelihood estimation of mixed stock ﬁshery composition. Can. J. Fish. Aquat. Sci. 44: 583–590. Millar, R. 1990. Comparison of methods for estimating mixed stock ﬁshery composition. Can. J. Fish. Aquat. Sci. 47: 2235–2241. Millar, R. B. 1991. Selecting loci for genetic stock identiﬁcation using maximum likelihood, and the connection with curvature methods. Can. J. Fish. Aquat. Sci. 48: 2173–2179. Milner, G. B., Teel, D. J., Utter, F. M., and Burley, C. L. 1981. Columbia River stock identiﬁcation study: validation of genetic method. Annual Report of Research (FY80). NWAFC, NOAA, Seattle, Washington. Milner, G. B., Teel, D. J., Utter, F. M., and Winans, G. A. 1985. A genetic method of stock identiﬁcation in mixed populations of Paciﬁc salmon, Oncorhynchus spp. Mar. Fish. Rev. 47: 1–8. Moore, D. S. 1986. Tests of chi-squared type. In R. D’Agostino and M. Stephens (eds.), Goodnessof-Fit Techniques. Marcel Dekker, New York, pp. 63–95. Mulligan, T. J., McKinnell, S., and Wood, C. C. 1988. Uncertainty in stock composition estimates of oceanic steelhead trout using electrophoretic characteristics: comments on a recent study. Can. J. Fish. Aquat. Sci. 45: 432–442. Parker, N., Giorgi, A., Heidinger, R., Jester Jr., D., Prince, E., and Winans, G. (eds.). 1990. Fish marking techniques. Amer. Fish. Soc. Symp. 7, Bethesda, Maryland. 879 pp. Pella, J. J. and Milner, G. B. 1987. Use of genetic marks in stock composition analysis. In N. Ryman and F. M. Utter (eds.), Population Genetics and Fisheries Management. University of Washington Press, Seattle, pp. 247–276. Waples, R. 1998. Separating the wheat from the chaff: patterns of genetic differentiation in high gene ﬂow species. J. Hered. 89: 438–450. White, B. A. and Shaklee, J. B. 1991. Need for replicated electrophoretic analyses in multiagency genetic stock identiﬁcation (GSI) programs: Examples from a pink salmon (Oncorhynchus gorbuscha) GSI ﬁsheries study. Can. J. Fish. Aquat. Sci. 48: 1396–1407. Wood, C. C., McKinnell, S., Mulligan, T. J., and Fournier, D. A. 1987. Stock identiﬁcation with the maximum likelihood method: sensitivity analysis and application to complex problems. Can. J. Fish. Aquat. Sci. 44: 866–881. Wood, C. C. 1989. Utility of similarity dendrograms in stock composition analysis. Can. J. Fish. Aquat. Sci. 46: 2121–2128.

This Page Intentionally Left Blank

CHAPTER

28

Estimation of Movement from Tagging Data CARL JAMES SCHWARZ Statistics and Actuarial Science, Simon Fraser University, Burnaby, British Columbia, Canada

I. Introduction II. Discrete Time/Discrete Stock Models A. Arnason–Schwarz Model B. Harvest Models C. Model Fitting D. Example III. Continuous Time/Space Models A. Theory B. Example IV. Summary and the Future V. References

I. INTRODUCTION Capture-recapture or tag-recovery studies are the primary method for estimating migration rates. In their simplest usage, ﬁsh are tagged at various locations and simple tabulations of subsequent recoveries show where the tagged ﬁsh have moved. However, unless recovery rates are equal in all reporting locations, the relative number of recoveries are uninformative about the real underlying migration rates. More sophisticated methods have been developed. These can be subdivided into two broad (somewhat overlapping) categories. First are methods for estimating movement among discrete stocks measured in discrete time. Second are methods for estimating movement within a single stock in continuous space and time. The discrete stock/discrete time models are based on stratiﬁed capturerecapture methods. As to be expected, the study of movement rates has not been Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

591

592

Carl James Schwarz

restricted to movements of ﬁsh. In particular, much development has occurred in applications to bird migration, driven in large part by the triannual EURING conferences. These methods extend the capture-recapture methods developed for estimating survival and abundance (Cormack, 1964; Jolly, 1965; Seber, 1965) to populations that are spatially stratiﬁed. Schaefer (1951), Chapman and Junge (1956), Darroch (1961), Plante et al. (1998), and Schwarz and Taylor (1998) dealt extensively with the two sample case. Arnason (1972, 1973) extended these methods to three sample times while Brownie et al. (1993) and Schwarz et al. (1993) completely generalized the problem. In the past 10 years, there has been an explosion of effort in applying the Arnason–Schwarz multistate models not only to geographic movement, but also to any movement among discrete states (Nichols and Kaiser, 1999; Lebreton and Pradel, 2002). While not used extensively in ﬁsheries, Schwarz (2003) uses this model to estimate movement rates of walleye within a lake. A related model is used for tag returns typically from exploited ﬁsh stocks. The key differences between the Arnason–Schwarz model and this submodel based on harvesting ﬁsh is that now recoveries take place over an extended period of time between release occasions. Many of the tags that are captured are not reported, and the tag reporting rate is unknown. Hilborn (1990) and Schwarz et al. (1993) used this type of data to estimate migration rates in a ﬁshery context. In the second category are models for continuous time and space. Here movement takes place within a single stock over time and space. While this can be recast in terms of the Arnason–Schwarz model or the related tag-recovery model (e.g., Hilborn, 1990), an alternate approach is to use diffusion models. This approach has a long history in ecology (Skellam, 1951), with Beverton and Holt (1957) among the ﬁrst to apply it in a ﬁsheries context. In these models, tagging and recovery typically take place irregularly over large spatial and temporal scales, and the stock is not discretized into independent, nonoverlapping parts. Sibert et al. (1999) applied diffusion models to estimating tuna movement.

II. DISCRETE TIME/DISCRETE STOCK MODELS The stock is ﬁrst stratiﬁed into discrete, nonoverlapping components. This partitioning can be done in many ways. For example, the components could correspond to geographic location, stock units, or any changeable attribute such as weight class or condition factor. (Stratiﬁcation by age is not usually considered in the same context as movement models because of the ﬁxed, nonprobabilistic movement between age classes.) Each and every ﬁsh that is captured should be readily classiﬁed into one and only one stratum. Without loss of generality, suppose there are K strata corresponding to geographical locations numbered 1, . . . , K.

Estimation of Movement from Tagging Data

593

Releases occur at regularly spaced intervals (say, yearly) for a total of T years in the study. At each release time point, a total of Rsi ﬁsh are tagged and released in stratum s in year i. It is assumed that releases are instantaneous to ensure that ﬁsh are alive and present in area s at the time of release. Usually releases take place in all strata. Each ﬁsh released requires an individually labeled tag so that the location of release and recovery can be determined.

A. ARNASON–SCHWARZ MODEL While the study protocol and data requirements for this model are more rigorous than for tag-recovery models (next section), advances in tagging methodology, particularly radio tags, self-marking tags (e.g., whale sighting), and genetic tags will create new opportunities to collect data for this rich set of models. Immediately before the release of tags in year i, recaptures take place in every stratum. Because population estimation is not a goal of the study, only tagged ﬁsh need be counted and examined. Recoveries are assumed to take place before releases so that released ﬁsh are not immediately removed from the study. Tagged ﬁsh have their tag numbers recorded, after which they can be removed from the study (e.g. a harvest) or rereleased. Studies with only removals (i.e., no ﬁsh returned to the population) can be analyzed using the Arnason–Schwarz model, but multiple captures allow more complex models to be examined. For example, the assumption that movement in year i only depends on current location and not upon past movement patterns can be assessed (Brownie et al., 1993). Usually, recapture effort occurs in all strata at all time points. It is possible to have recaptures extending for several years after releases have ended, which can provide some information about migration rates early in the study. However, unless the lifetime of the ﬁsh is long and recapture rates are low, there is little to be gained from such an extended effort. All tags from ﬁsh that are recaptured are assumed to be read. Modern practice is to construct a capture history vector for each released ﬁsh. The capture history (Lebreton et al., 1992) is a vector, w, of length T. Notation is not yet completely standardized, but one common notation has components wi: Ï0 ﬁsh not seen at time i Ô w i = Ìs ﬁsh seen in stratum s at time i and released ÔÓ - s ﬁsh seen in stratum s at time i and removed The positive/negative entries distinguish between ﬁsh released back to the population and ﬁsh removed through, for example, instantaneous exploitation. For

594

Carl James Schwarz

example, the history vector (3, 0, 2, 0, -1, 0) represents a ﬁsh released in stratum 3 in year 1, not seen in year 2, seen in stratum 2 in year 3, not seen in year 4, and seen (and removed) in year 5 from stratum 1. There are four sets of parameters needed to model these data: fsi

ysti

Apparent survival rate, which is the probability that a ﬁsh alive in year i in stratum s will survive and remain in the study until year i + 1. Permanent migration out of the study area is indistinguishable from mortality. Movement rates conditional upon survival from year i to year i + 1. This is the probability that a ﬁsh will survive from year i to year i + 1, and K

move from stratum s to stratum

Â y stt = 1 . t =1

psi lsi

Recapture rate. This is the probability that a ﬁsh which is alive in year i in stratum s will be captured. All captured ﬁsh have their tag read. Loss on capture rate. This is the probability that a ﬁsh captured in year i in stratum s will be removed from the population and not released.

The Arnason–Schwarz model does not specify the timing of the migration, which can take place any time between year i and year i + 1. The movement parameters are also net movement rates that are conditional upon the ﬁsh surviving for the year. Lastly, the parameters say nothing about possible movement patterns between the years, that is, a ﬁsh may have moved from stratum 1 to stratum 4 via stratum 2 or stratum 3 between the 2 years. Similarly, the survival terms measure net survival “averaged” over all destinations. Because the migration rates sum to 1 over the destination, it is possible to combine the migration and survival parameters into one parameter measuring the product of survival and movement, but there are some advantages to the current parameterization. These parameters are used to construct a probability expression for each history conditional upon the initial release. These expressions are complex and must account for all the (hidden) paths when a ﬁsh is not seen. For example, the history vector (0, 2, 0, 2, -1) has probability (assuming only two strata): 1 1 12 2 22 2 2 22 2 2 2 21 1 1 [ f 22y 21 2 (1 - p 3 ) f 3y 3 + f 2y 2 (1 - p 3 ) f 3y 3 ] ¥ p 4 (1 - l 4 ) f 4 y 4 p 5l 5

where the term in square brackets represents the two possible paths between being seen in stratum 2 at time 2 and stratum 4 at time 4 without being seen at time 3. Construction of these expressions is complex (Brownie et al., 1993), but, as shown by Brownie et al. (1993) and Schwarz et al. (1993) results can more easily be found using matrix notation.

595

Estimation of Movement from Tagging Data

B. HARVEST MODELS In these studies, recoveries of tags occur between release periods over an extended period of time. Typically, these are from commercial or recreational harvests and, in most cases, the tag reporting rate is less than 100%. Because the timing of the migration event is unknown, the recovery of a tagged ﬁsh in stratum s between year i and year i + 1 may not provide information on the ﬁnal destination of the ﬁsh in year i + 1. Unlike with the Arnason–Schwarz model, tag return rates are typically less than 100% and unknown. A capture-history vector can also be constructed for each released ﬁsh, except that now there will be at most two nonzero entries, the last entry is always a losson-capture, and it must be remembered that the terminal recovery takes place between 2 years. For example, the history vector (0, 2, 0, -1, 0) represents a ﬁsh that was tagged and released in stratum 2 in year 2 and not seen until harvested between years 4 and 5 in stratum 1. Because each ﬁsh can be seen at most twice, an alternate display of matrices of release and recovery can also be constructed, as shown in Schwarz et al. (1993). Schwarz et al. (1993) developed the basic methodology that extended the Brownie et al. (1985) band-recovery models for birds. The notation below differs from that originally used by Schwarz et al. (1993) but is now consistent with the standardized notation in the Arnason–Schwarz model. The parameters for this model are as follows: fsi

ysti

The apparent survival rate which is the probability that a ﬁsh alive in year i in stratum s will survive and remain in the study until year i + 1. Fish that permanently migratel out of the study area are indistinguishable from mortality. Movement rates conditional upon survival from year i to year i + 1. This is the probability that a ﬁsh will survive from year i to year i + 1, and K

move from stratum s to stratum t.

Â y sti = 1 . t =1

fist

Recovery rate. This is the probability that a ﬁsh alive in stratum s in year i will be harvested in stratum t between year i and year i + 1, and that its tag will be seen, read, and reported.

The key difference in the parameterizations engendered by recoveries taking place during an extended period between release points and nonreporting of tags is that now the f parameter includes a mortality component and a tag-reporting component. As before, the probability of a capture history can be developed. It is similar, but not exactly the same as in the Arnason–Schwarz model. For example, the history (0, 2, 0, -1, 0) would have a probability of

596

Carl James Schwarz

1 11 11 2 21 1 12 21 2 22 2 21 11 2 22 2 22 21 [ f 22y 21 2 f 3y 3 f 4 + f 2y 2 f 3y 3 f 4 + f 2y 2 f 3y 3 f 4 + f 2y 2 f 3y 3 f 4 ]

Schwarz et al. (1993) show how these expressions can be formed using matrices.

C. MODEL FITTING While the expressions for the capture histories are different for the two protocols, the subsequent model ﬁtting follows a common methodology. The likelihood is constructed from the expressions for each history using either a product of multinomial distributions (Brownie et al., 1993) with use of the expressions shown in this chapter, or a product of Poisson distributions based on the number of observed recaptures (Hilborn, 1990). In the latter case, model development typically expresses the number of tags expected to be alive in the next time period given the distribution in the current time period using matrix expressions for movement and survival which form the mean of the Poisson distribution. Both are asymptotically equivalent. Numerical methods are used to ﬁnd the value of the parameters that maximize the likelihood function for a given set of data, that is, the maximum-likelihood estimates. Modern practice is to start with a global model (e.g., allowing parameters to vary over time and space) and use this model to assess goodness-of-ﬁt (Pradel, et al., 2003). This goodness-of-ﬁt statistic also provides a measure of overdispersion that may arise from nonindependent movement of ﬁsh (e.g., tagged ﬁsh move in schools). This overdispersion factor can be used to adjust standard errors for model failure. A series of models (e.g., assuming equal movement rates across time) are ﬁt. Traditionally, likelihood ratio tests are used to “select” the most parsimonious model, and its estimates and conditional standard errors are reported. However, this ignores the uncertainty in the estimates attributable to model uncertainty. Modern practice is to use the Akaike Information Criteria (Akaike, 1973) to determine the relative weight to be given to each of the ﬁtted models (Burnham and Anderson, 2002). Rather than trying to select the best model, the AIC weights are used to combine the estimates from the suite of models. A standard error that includes model uncertainty can also be computed. Easy-to-use software is available for the Arnason–Schwarz models: MARK (White and Burnham, 1999) and M-SURGE (Choquet et al., 2003). I am unaware of any similar generalized software for the harvest models, but Schwarz et al. (1993) show how to ﬁt these using the Arnason–Schwarz model. Bayesian methods have been extensively developed for the Arnason–Schwarz model (Dupuis, 1995; Dupuis et al., 2002). Unfortunately, easy-to-use software is not yet generally available for ﬁtting under this paradigm. The major problem of using these models with ﬁshery data are tag loss, nonreporting of tags, and noncomprehensive release and recovery efforts.

Estimation of Movement from Tagging Data

597

Tag loss is typically assumed to be a known function of time or is assessed by double tagging a subset of released ﬁsh using a second tag or a batch mark (e.g., a ﬁn clip). The former is preferred because batch marks provide no information on when the ﬁsh was released. The returns from these doubly tagged ﬁsh can be used to estimate the tag loss rate which is then incorporated into the likelihood function (e.g., Barrowman and Myers, 1996) Reporting rates of tags is typically assumed known or is assessed using reward tags (which are assumed to be returned at a 100% rate). Again, adjustments to the likelihood function are straightforward. Fishery data often have holes, that is, releases and/or recoveries do not take place in all strata in all years. These data are still useful, but restrict the complexity of models that can be ﬁt. For example, it may be necessary to assume equal movement rates over time to account for missing data (e.g., Hilborn, 1990; Schwarz and Ganter, 1995). Other restrictions can also be applied to deal with sparse data. For example, under the general formulation, movement can occur between any pair of strata between time points. Hilborn (1990) assumed that movement could only occur between geographically adjacent strata, or, the probability of movement could be related to the distance between the strata. The models as presented are self-contained and use only the tagging data to estimate parameters. However, ﬁshery data typically also have much additional data that are not available for other species. For example, catch, effort, and mortality rates are often estimated from other data. This additional information can be used: for example Hilborn (1990) assumed that catchability (recovery rate) was proportional to effort. Before one embarks on an extensive tagging program, it should be evaluated to ensure that estimates have adequate precision and low bias. This is most easily done using simulation capabilities built into the major software packages, or as outlined by Xiao (1996). Finally, these models are in a state of active research particularly for bird species. Williams, et al. (2002, Chapter 17.3) discuss in detail the latest research, and new work regularly appears in the EURING conference proceedings. Fishery researchers need to be aware of this other body of work. For example, McGarvey and Feenstra (2002) estimated movement working backwards from recaptures, but seem to be unaware of similar work by Pradel (1996) in reading capture-recapture history vectors backward.

D. EXAMPLE Anganuzzi et al. (1994) estimated movement rates of Paciﬁc halibut (Hippoglossus stenolepis) from mark-recapture data and illustrate many of the concepts explained above.

598

Carl James Schwarz

FIGURE 28-1. Approximate regulatory areas for halibut management set by the International Paciﬁc Halibut Commission off the coast of British Columbia, Canada, and Alaska, Washington, and Oregon, United States. Releases of tagged ﬁsh only occurred in Areas 3a and 3b. Recoveries of tagged ﬁsh occurred in all areas. To reduce the number of movement parameters to be estimated, it was assumed that no northward movement occurred and that movement to a new area more than two areas away was not possible (e.g. yearly movement rate from 3b to 2b was set to zero).

Brieﬂy, the coast of Washington State, British Columbia, and Alaska has been divided into ﬁve contiguous regulatory areas (see Anganuzzi et al., 1994, their Fig. 28-1). Approximately 30,000 halibut were tagged and released in each of 1980 and 1981 in only 2 areas as indicated. Fish were captured from short hauls, measured, marked with spaghetti tags, and released. Ten years of recoveries came from the long-line ﬁshery. The data consist of history vectors of length 10, with at most two nonzero entries representing releases in either year 1 or year 2 (limited to two areas) and potential harvest in years 2 to 10. The length at time of tagging was also measured. Anganuzzi et al. (1994) built up the probability of each history in stages. First, immediately after tagging, they allowed for a nonzero tag loss or tagging acute mortality, which was assumed constant over time and space. Next, they allowed for movement (and noncapture) between the release-year stratum and the recov-

599

Estimation of Movement from Tagging Data

ery year stratum. With only 2 years of releases that did not occur in all areas and 10 years of recoveries, it is not possible to ﬁt a general model that allows movement rates to vary by year. Instead, the authors assumed that either movement occurred only once (in the year immediately after release), or that movement rates were constant over time. Additionally, to reduce the number of parameters, only southward movements were allowed and movements more than two areas away were also not allowed—corresponding parameters were set to zero. Between release and recovery, ﬁsh must also survive. There were several forces of mortality modeled. Long-line ﬁshing mortality was modeled as a function of predicted length (from a deterministic Bertalanffy growth model with known parameter values). Sport ﬁshing was modeled as a function of the known sport ﬁshing effort. Out sources of mortality were modeled as a known function of predicted length. Finally, in the year of harvest, the ﬁsh were captured in the long-line ﬁshery (using the mortality function above) using known effort data and potentially area-length speciﬁc length vulnerabilities as a function of predicted length. The reporting rate was assumed to be known and equal for all ﬁsh in the same recovery area. Model selection used AIC to distinguish among models, but model averaging was not used. Estimates of yearly movement rates from the model with the smallest AIC value are shown in Table 28-1. Over 80% of ﬁsh were estimated to remain in the stratum in any year, with little out-migration more than 1 stratum away As noted by the authors, they assumed values for growth, ﬁshing mortality, and tag-return rates, and made strong assumptions about homogeneity of movement over space and time with additional restrictions on the direction and maximum movements allowed. They recommended that future studies have releases and recoveries occurring in all year–strata combinations to reduce the TABLE 28-1. Estimated Yearly Movement Rates between Halibut Management Areas Shown in Figure 28-1 from the Anganuzzi et al. (1994) Model 4 Receiving management area

Donating management area

2a 2b 2c 3a 3b

2a

2b

2c

3a

3b

1.00 0.04 ** ** **

* 0.96 0.16 0.06 **

* * 0.84 0.04 0.04

* * * 0.90 0.14

* * * * 0.82

*Only southward movement was allowed so these parameters were set to zero. **Movement only allowed to a maximum of two areas away from donating area so these parameters were set to zero.

600

Carl James Schwarz

number of assumptions that must be made. As lengths were also measured for recaptured ﬁsh, it would be possible to use these values to estimate the parameters of the growth model rather than assuming they are known from other studies.

III. CONTINUOUS TIME/SPACE MODELS

A. THEORY In these models, movement over a large spatial area is of interest rather than interchanges between discrete stock units. Sibert et al. (1999) is a typical example where the movement of tuna through the southwest Paciﬁc Ocean is studied. The basic idea can be summarized by a differential equation (Sibert et al., 1999): ∂N ∂ Ê ∂N ˆ ∂ Ê ∂N ˆ ∂ ∂ (uN ) (vN ) - ZN = D + D ∂t ∂ x Ë ∂ x ¯ ∂y Ë ∂y ¯ ∂ x ∂y where N is the number of tagged ﬁsh alive at a point (x, y) at time t in the study area. The ﬁrst two terms represent movement of ﬁsh through diffusion in the x and y directions. The next two terms represent directed movement in the x and y directions in response to some gradient. The last term represents mortality from all causes. In general, the diffusion parameter, D, the directed movement parameters, u and v, and the mortality parameter, Z, vary in space and time. In order to apply the model to data, it must be ﬁrst discretized to a regular spatial and temporal grid. (It now enters the realm of the previous models except that there is a very regular spatial structure to the strata.) The most general model has four parameters at each grid point in space and time, which is clearly too many to ﬁt. At this point, simpler models are assumed. For example, movement parameters are equal over regions or seasons, and ﬁshing mortality is a known function of ﬁshing effort. The observed number of tag returns are related to the predicted number of tag returns using (typically) a Poisson distribution: t - Z t Fij C tij ~ Poisson(C˜ tij ) where C˜ tij = N tij 1 - e ij t b tij Z ij

(

)

and Ctij, C˜ tij Ntij Ztij

are the observed and expected catch at grid (i, j) at time t is the number of tags present at grid (i, j) at time t according to the discretized solution to the differential equation; is the total mortality at grid (i, j) at time t;

Estimation of Movement from Tagging Data

Ftij btij

601

is the ﬁshing mortality at grid point (i, j) at time t; and is the reporting rate at grid point (i, j) at time t.

The latter two parameters are often assumed to be known (e.g., reporting rates of 100%) or a function of known covariates (e.g., ﬁshing mortality proportional to effort). The likelihood function is constructed as the product of the above over all grid points in space and time. Standard numerical methods are used to maximize the likelihood function. As pointed out in Sibert et al. (1999), the discretization process needs to be carefully implemented to obtain meaningful results. Model selection is typically done using likelihood ratio tests, but AIC methods, as discussed in the previous section, should be used along with model averaging over competing models that have a similar ﬁt. Tag loss and nonreporting rates are handled in the same way as previous sections. I am unaware of any general software packages that implement the above models, and these must be constructed for each problem. It would be possible, through careful coding and assuming the data are not too sparse, to ﬁt the Arnason–Schwarz model to such data.

B. EXAMPLE Sibert et al. (1999) has a very clear exposition on such models to estimate movement rates of skipjack tuna (Katsuwonus palamis) in the southwest Paciﬁc Ocean (Fig. 28-2). Between 1977 and 1980 over 94,000 tuna were tagged and released, and about 5,000 tags were returned. The entire region was subdivided into one-degree square blocks and time was discretized into monthly intervals. The one-degree squares were further grouped into 10 larger regions and time was further grouped into single-, two-, or four-season blocks. Within region–season combinations, movement parameters (over the one-degree grid) were set equal. These parameters were also held constant over years. They further assumed that natural mortality was assumed constant over time and space, that ﬁshing mortality was a simple function of known ﬁshing effort, and that reporting rates were 100%. Sibert et al. (1999) considered single, two, or four season models for the movement parameters with varying start points for the seasons. The best ﬁtting model selected was that with two seasons (starting in March and September). It has 66 parameters—three movement parameters (diffusion D and north-south and east-west rates (u and v) for each of the 10 regions for each of two seasons plus one parameter for natural mortality plus ﬁve parameters for the catchability coefﬁcients for the ﬁve ﬂeets operating. Results from this model are shown in Figure 28-3. There is considerable spatial variability in both the directed and random components of movement. Figure 28-3 also shows a

602

Carl James Schwarz

FIGURE 28-2. Map of the southwest Paciﬁc Ocean showing the location of the 10-region model area used for the skipjack tuna analysis. The regions are subdivided into one-degree blocks. Reprinted from Siebert et al. (1999, Fig. 2, p 929)

consistent eastward and southward directed movement in the SeptemberFebruary season for ﬁve of the regions. The diffusive movements are larger for most regions in this season compared to the diffusive movements in the MarchAugust season. Numerical estimates can be found in Siebert et al. (1999). Sibert et al. (1999) found good agreement between the observed and expected recoveries. It would be interesting to compare their results with the discrete time/discrete space formulation.

IV. SUMMARY AND THE FUTURE The discrete time/discrete space and the continuous time/continuous space approaches are two poles in how the underlying process of movement is viewed. The ﬁrst approach is likely preferable when data are collected on a gross scale with larger, clear separations among stocks. For example, Schwarz et al. (1993) and Brownie et al. (1993) had data that could only be collected on large geographic scales and on a yearly basis. The second approach is more appropriate for very ﬁnely collected data, for example, exact locations and times of release and recovery which, with GPS, are now feasible to collect. Of course, this type

FIGURE 28-3. Estimated movement pattern for model 2 of Sibert et al. (1999). In this model, movement differs between the two seasons. The length of the arrows is proportional to the resultant directed movement component (u, v), and the areas of the circles are proportional to the random movement component (D). The arrow in the legend inset represents directed movement of 100 Nmi·month-1; the circle represents random movement of 3,000Nmi2·month-1. Geographic coordinates are the same as in Figure 28-2. Reprinted from Siebert et al. (1999, Fig. 8, p. 933).

603

604

Carl James Schwarz

of data can always be discretized later. Consequently, it is recommended for the planning of movement studies to try to collect data at as ﬁne a scale as possible. Another key point is that adequate spatial and temporal coverage is essential to allow models to be ﬁt that are not so restrictive (e.g., constant migration rates over time). Fishery data are somewhat unique in that much auxiliary information is available (such as ﬁshing effort) that is just not possible to collect for other species. The usual formulations of the discrete time/space models are “self-contained” in that they rely only on the observed recovery data, but whenever possible, this new information should be incorporated to improve precision. For some species, multiple sources of information can be found, for example, resighting records and recovery records. The integration of multiple sources of information has been well studied in the bird context (e.g., Lindberg et al., 2001) in relation to estimation of survival rates, but has not been used for estimation of movement rates. Technology may provide new, exciting sources of data on movement. For example, archival tags can record, store, and report on the location of ﬁsh on a daily basis. In the past, ﬁsh had to physically be handled to read tags, but newer tags can be interrogated at a distance. These new data streams can be used to construct the entire trajectory of a ﬁsh—the problem of missing data between release and recovery no longer exists. New analysis methods are starting to come on stream to deal with this type of data. For example, Manly and Chatterjee (1993), Sibert and Fournier (2001), and Sibert et al. (2003) use a random walk as the basis for modeling the trajectory. In the former case, the underlying space was discretized into zones to cast it into a more familiar mark-recapture context, while in the latter cases, a Kalman ﬁlter is used to relate the observed data to predicted positions from the movement model. More effort will be needed to fully exploit this rich source of data.

V. REFERENCES Akaike, H. 1973. Information theory as an extension of the maximum likelihood principle. In “Second International Symposium on Information Theory” (B.N. Petrov, and F. Csaki, Eds),. Akademiai Kiado, Budapest, pp. 267–281. Anganuzzi, A., Hilborn, R., and Skalski, J. R. 1994. Estimation of size selectivity and movement rates from mark-recovery data. Canadian Journal of Fisheries and Aquatic Sciences 51: 734–742. Arnason, A. N. 1972. Parameter estimation from mark-recapture experiments on two populations subject to migration and death. Researches in Population Ecology 13: 97–113. Arnason, A. N. 1973. The estimation of population size, migration rates, and survival in a stratiﬁed population. Researches in Population Ecology 15: 1–8. Barrowman, N. J. and Myers, R. A. 1996. Estimating tag-shedding rates for experiments with multiple-tag types. Biometrics 52: 1410–1416.

Estimation of Movement from Tagging Data

605

Beverton, R. J. H. and Holt, S. J. 1957. On the dynamics of exploited ﬁsh populations. U.K. Ministry of Agriculture and Fisheries, Fish. Investigations (ser. 2), 19. Brownie, C., Hines, J. E., Nichols, J. D., Pollock, K. H., and Hestbeck, J. B. 1993. Capture-recapture studies for multiple strata including non-Markovian transition probabilities. Biometrics 49: 1173–1187. Burnham, K. P. and Anderson, D. R. 2002. Model selection and inference: a practical informationtheoretic approach. Springer Verlag, New York. Chapman, D. G. and Junge, C. O. 1956. The estimation of the size of a stratiﬁed animal population. Annals of Mathematical Statistics 27: 375–389. Choquet, R., Reboulet, A. M., Pradel, R., Gimenez, O., and Lebreton, J.-D. 2003. User’s manual for M-SURGE 1.0. Mimeographed document, CEFE/CNRS, Montpellier (ftp://ftp.cefe.cnrsmop.fr/biom/Soft-CR) . Cormack, R. M. 1964. Estimates of survival from the sighting of marked animals. Biometrics 51: 429–438. Darroch, J. N. 1961. The two-sample capture-recapture census when tagging and sampling are stratiﬁed. Biometrika 48: 241–260. Dupuis, J. A. 1995. Bayesian estimation of movement and survival probabilities from capturerecapture data. Biometrika 82: 761–772. Dupuis, J. A., Badia, J., Maublanc, J.-L., and Bon, R. 2002. Survival and spatial ﬁdelity of mouﬂon (Ovis gmelini): A Bayesian analysis of an age-dependent capture-recapture model. Journal of Agricultural Biological and Environmental Statistics 7: 277–298. Hilborn, R. 1990. Determination of ﬁsh movement patterns from tag-recoveries using maximum likelihood estimators. Canadian Journal of Fisheries and Aquatic Sciences 47: 635–643. Jolly, G. M. 1965. Explicit estimates from capture-recapture data with both death and immigration— stochastic model. Biometrika 52: 225–247. Lebreton, J.-D. and Pradel, R. 2002. Multi-stratum recapture models: modeling incomplete individual histories. Journal of Applied Statistics 29: 353–369. Lebreton, J.-D., Burnham, K. P., Clobert, J., and Anderson, D. R. 1992. Modeling survival and testing biological hypotheses using marked animals. A uniﬁed approach with case studies. Ecological Monographs 62: 67–118. Lindberg, M. S., Kendall, W. L., Hines, J. E., and Anderson, M. G. 2001. Combining band recovery data and Pollock’s robust design to model temporary and permanent emigration. Biometrics 57: 273–281. Manly, B. F. J. and Chatterjee, C. 1993. A model for mark-recapture data allowing for animal movement. In J.-D. Lebreton and P. M. North, (eds.), Marked Individuals in the Study of Bird Populations., Birkhäuser Verlag, Basel, pp. 309–322. McGarvey, R. and Feenstra, J. E. 2002. Estimating rates of ﬁsh movement from tag recoveries: conditioning by recapture. Canadian Journal of Fisheries and Aquatic Sciences 59: 1054–1064. Nichols, J. D. and Kaiser, A. 1999. Quantitative studies of bird movement: a methodological review. Bird Study 46 (suppl.): s289–s298. Plante, N., Rivest, L.-P., and Tremblay, G. 1998. Stratiﬁed capture-recapture estimation of the size of a closed population. Biometrics 54: 47–60. Pradel, R. 1996. Utilization of capture-mark-recapture for the study of recruitment and population growth rates. Biometrics 52: 371–377. Pradel, R., Wintrebert, C. M. A., and Gimenez, O. 2003. A proposal for a goodness-of-ﬁt test to the Arnason–Schwarz multisite capture-recapture model. Biometrics 59: 43–53. Schaefer, M. B. 1951. Estimation of the size of animal populations by marking experiments. U.S. Fish and Wildlife Service Fisheries Bulletin 69: 191–203. Schwarz, C. J. 2003. Estimating the Number of Walleye in Mille Lacs Lake, Minnesota. Prepared for the Minnesota Department of Natural Resources. 88 pp.

606

Carl James Schwarz

Schwarz, C. J. and Arnason, A. N. 1990. Use of tag-recovery information in migration and movement studies. American Fisheries Society Symposium 7: 588–603. Schwarz, C. J. and Ganter, B. 1995. Estimating the movement among staging areas of the barnacle goods (Branta leucopsis). Journal of Applied Statistics 22: 711–725. Schwarz, C. J. and Taylor, C. G. 1998. The use of the stratiﬁed Petersen estimator in ﬁsheries management with an illustration of estimating the number of pink salmon (Oncorhynchus gorbuscha) that return to spawn in the Fraser River. Canadian Journal of Fisheries and Aquatic Sciences 55: 281–296. Schwarz, C. J., Schweigert, J., and Arnason, A. N. 1993. Estimating migration rates using tag recovery data. Biometrics 49: 177–194. Seber, G. A. F. 1965. A note on the multiple recapture census. Biometrika 52: 249–259. Sibert, J. R., Hampton, J., Fournier, D. A., and Bills, P. J. 1999. An advection–diffusion-reaction model for the estimation of ﬁsh movement parameters from tagging data, with application to skipjack tuna (Katsuwonus pelamis). Canadian Journal of Fisheries and Aquatic Science 56: 925–938. Sibert, J. and Fournier, D. A. 2001. Possible methods for combining tracking data with conventional tagging data. In J. R. Sibert and J. Nielsen, (eds.), Electronic Tagging and Tracking in Marine Fisheries, Kluwer Academic Publishers, The Netherlands, pp. 443–456. Sibert, J., Musyl, M. K., and Brill, R. W. 2003. Horizontal movements of bigeye tuna (Thunnus obesus) near Hawaii determined by Kalman ﬁlter analysis from archival tagging data. Fisheries Oceanography 12: 141–151. Skellam, J. G. 1951. Random dispersal in theoretical populations. Biometrika 38: 196–218. White, G. C. and Burnham, K. P. 1999. Program MARK: Survival estimation from populations of marked animals. Bird Study 46 (suppl.): s120–s139. Williams, B. K., Nichols, J. D., and Conroy, M. J. 2002. Analysis and Management of Animal Populations. Academic Press, London. Xiao, Y. 1996. A framework for evaluating experimental designs for estimating rates of ﬁsh movement from tag recoveries. Canadian Journal of Fisheries and Aquatic Sciences 53: 1272–1280.

PART

VIII

Application of Stock Identiﬁcation Data in Resource Management

This Page Intentionally Left Blank

CHAPTER

29

Stock Identiﬁcation for Conservation of Threatened or Endangered Species MICHAEL A. BANKS Coastal Oregon Marine Experiment Station, Hatﬁeld Marine Science Center, Department of Fisheries and Wildlife, Oregon State University, Newport, Oregon, USA

I. Introduction II. Winter-Run Chinook of the California Central Valley: Identiﬁcation, Admixture, and Bias III. Spring-Run Chinook of California Central Valley: Increased Identiﬁcation Difﬁculties IV. Snake River Sockeye and Chinook, Oregon–Washington and Idaho: Broodstock Shortfalls, Disease and Fishery Impact Assessment V. Stock Identiﬁcation and U.S.–Canada Salmon Wars: Equalizing Interception Rates; Politics and Resolution for GSI VI. Supportive Breeding of Pink Salmon at Risk of Extinction: Genetic Marker and Statistical Measure Comparison VII. Applications in Alaska (AK): GSI Foundation, Polymorphic Microsatellites and Zero Frequencies, Endangered Species Interception, and Perspective VIII. Application in the High Seas: Problems Caused by Missing Stocks IX. Conclusions and Prospects for the Future References

I. INTRODUCTION Fisheries, water diversions, and other environmental shifts have the potential to result in signiﬁcant loss of speciﬁc ﬁsh stocks, necessitating careful management Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

609

610

Michael A. Banks

wherever possible. Impacts often affect complex mixtures of different stocks, however, making effective management critically dependent on precise methods for stock identiﬁcation. This is particularly important in contexts involving threatened and endangered species because conservation measures require focused targeting in order to achieve and monitor rebuilding goals. As a result, the application of genetic and statistical methods toward stock identiﬁcation has continued to develop and improve in the past few decades, largely in response to a number of new challenges. These fall into three primary areas: 1. Continued interest in sustaining harvest and economic yield despite extreme variability and changes in habitat availability, ocean productivity, stock strengths, and effectiveness of ﬁshing technologies. 2. Increasing interaction between farmed and hatchery-reared ﬁsh with their counterparts from wild populations. 3. The increasing number of endangered or threatened species cases needing special attention for protection and rebuilding. This chapter focuses on various cases of the last mentioned area, as the former two are already covered in this book. Cases are selected to span major regions around the Northern Paciﬁc Rim (Fig. 29-1) but also for their unique contributions to the ﬁeld of genetic stock identiﬁcation (GSI).

II. WINTER-RUN CHINOOK OF THE CALIFORNIA CENTRAL VALLEY: IDENTIFICATION, ADMIXTURE, AND BIAS Winter-run Chinook (Oncorynchus tschawytscha) is one among four distinct runs that return to spawn in the Central Valley of California, each with their own unique combination of life history traits. In recent times, this region has also hosted the greatest number of people and agricultural water uses found in the United States. High demand for water by people and increased drought conditions have signiﬁcantly challenged freshwater needs for salmon. This is particularly true for winter run because their juvenile ocean-bound migration peaks in spring. Winter-run returns from the ocean once numbered several hundred thousand but their ﬁrst major setback was caused by the construction of Shasta Dam in the early 1950s, because access to their spawning beds of the upper McCloud and Pitt rivers was completely ﬂooded by this dam (Yoshiyama, 1999). Lower numbers relocated to spawn in the upper main stem of the Sacramento River and perhaps upper tributaries. However, poor ocean productivity, sustained harvest, and drought conditions of the 1980s reduced the number of returning spawners to seriously low numbers of less than 200 by the early 1990s. Concern for the projected loss of this unique life history type proceeded toward Endangered

Stock Identiﬁcation for Conservation of Threatened or Endangered Species

611

Figure 29-1. Map of the North Paciﬁc indicating location of various cases discussed in this chapter.

Species Act (ESA) listing (NOAA Fisheries, 1994), and various activities to lessen the chances of extinction were initiated. Protection and rebuilding plans raised a number of challenges for stock identiﬁcation methods. First, protein electrophoresis data available at the time (Bartley et al., 1992) did not provide any means for discrimination of winter run from the other three life history types of the Sacramento River. Given signiﬁcant risk

612

Michael A. Banks

of complete loss of winter run from the natural system, it was decided to develop a captive broodstock that could serve as a gene bank for this stock. But how could one be sure that Chinook sampled for this endeavor were indeed winters? Indistinguishable phenotypic traits among two or three other life history types that occur in the upper Sacramento River at the time of winter-run maturation presented a need for rigorous individual-based genetic identiﬁcation. Second, rearing conditions for the few number of families that could be formed in the early days of this captive broodstock project necessitated combining of families together within the same tank so as to sustain sufﬁcient densities for normal feeding behavior. Once these juveniles reached maturity for spawning, how could they be mated without risking brother–sister mating or even mating between alternate life history types that may have been inadvertently selected for the winter-run captive broodstock project? Fortunately, molecular genetic and statistical tools were developed to meet these challenges. Banks et al. (1999) developed the ﬁrst Chinook salmon microsatellites and demonstrated that the distribution of their large number of characters (alleles) among subpopulations from the Central Valley drainage allowed reliable discrimination among alternate life history types (Banks et al., 2000). Likewise, individual-based methods for population assignment were developed (Paetkau et al., 1995; Cornuet et al., 1999; Banks and Eichert, 2000) and ensured that individuals selected for broodstock rearing were indeed winter run [100% accuracy can be achieved for winter-run identiﬁcation with as few as ﬁve carefully chosen microsatellites (Banks et al., 2003)]. Further, these identiﬁcation tools were applied to protection measures on occasion where winter run might be at risk for loss, for example, at water diversions (Banks et al., 1996) or in ocean ﬁsheries (Hedgecock et al., 2001). Sampling problems associated with relative proportions of numbers of winter Chinook in contrast to the more common fall and late-fall run ﬁsh presented another challenge to the statistical aspects of the individual-based methods described above. Population assignment is achieved through compounded frequency differences among genotypes assessed across runs. None of the genotypes observed or expected, however, was unique to any speciﬁc run. Let us consider an example at a single locus to demonstrate the relative proportion problem. The frequencies of alleles observed among runs might be such that genotype LL is expected at a frequency of 0.64 in winter run and less than 0.0025 in either spring, fall, or late-fall runs. Thus, assigning all individuals with genotype LL to winter run appears rigorous with chances of misidentiﬁcation of nonwinters to winter likely at less than 1%. However, one needs to acknowledge that the endangered winter run has strikingly low numbers in comparison to the other runs, so any ﬁsh randomly drawn from the ocean or at a water diversion site is about 500 or more times likely fall or late-fall than winter run. Thus, realistically, a ﬁshery harvest of 10,000 is likely to have only 20 winters, 12 or 13 of which would be

Stock Identiﬁcation for Conservation of Threatened or Endangered Species

613

expected to have genotype LL. A total of 227 (fraction 0.0025 of 90,980) LL genotypes are expected among the remaining fall or late-fall, however, providing a misidentiﬁcation problem. These “rare” fall or late-fall LL Chinook outnumber bona ﬁde LL winters just because there are so many more fall or late-fall ﬁsh. This occurs because individual-based methods described above assume equal proportions among potential source populations. Scenarios that combine rare endangered populations along with thriving stocks clearly refuted this assumption. Fortunately, invoking multiple loci and Bayes’ rule provided a solution (Shoemaker et al., 1999). Population-based mixed-stock analysis techniques (discussed below) provide Bayesian priors that can then be used to modify relative proportions (frequencies) among baseline populations to overcome bias caused by strong differences in contributing stock ratios. Development continues in these aspects for improving genetic stock identiﬁcation methods (Pella and Masuda, 2001; Blanchong et al., 2002; Brun et al., 2003; Sahu and Cheng, 2003). The overall status for winter-run recovery is good. Improved ocean productivity, ﬁshery closures, altered water diversion strategies, habitat restoration, supplementation from the captive broodstock program, and a number of other measures have resulted in increasing population estimates [Peterson model escapement estimates rose from 820 in 1996 to 12,797 for 2001 (California Department of Fish and Game (2002)].

III. SPRING-RUN CHINOOK OF CALIFORNIA CENTRAL VALLEY: INCREASED IDENTIFICATION DIFFICULTIES Like winter Chinook, spring-run Chinook from the California Central Valley were once very robust and comprised the major component of a 12 million-pound inriver ﬁshery between 1870 and 1920 (Clark 1940). Spring run was indeed the preferred stock owing to their greater abundance, higher fat content, and bright ocean-fresh condition. Although returning to fresh water in March, spring Chinook ascend higher elevation in tributary streams to ﬁnd deep cool pools for holding through the summer (without feeding) before spawning in the fall. They too have sustained substantial setback owing to human development and interests in their freshwater habitat. Initially, hydraulic mining and dam construction associated with the gold rush of the 1850s and then extensive damming and diversion of most of the Central Valley’s tributaries have reduced the availability of higher elevation habitat to just a few tributaries. Proposals for endangered species listing of spring followed shortly after winter but were put on hold because of strong collaboration between ﬁshing, farming, agency, environmental, and other enterprises. It was reasoned that listing of spring run as an endangered species would result in complete closure of the ocean ﬁshery, causing substantial economic and political backlash. Signiﬁcant outreach in key watersheds was

614

Michael A. Banks

proposed to achieve the same ends as would be mandated by the federal ESA process. This scenario is well described in Bingham et al. (2000) and indeed delayed the formal process, but spring-run Chinook were eventually added to the federal list of threatened species (NOAA Fisheries, 1998). Although the microsatellites described above provided sufﬁcient statistical power for reliable winter-run identiﬁcation, the closer genetic relationship between spring and the more common fall and late-fall life histories made unambiguous identiﬁcation of spring run more challenging. While just seven of the microsatellites described in Banks et al. (2000) provided unambiguous identiﬁcation for all of the winter-run genetic variation, these same markers could identify only about 70% of the spring life history type with rigor (Fig. 29-2). Thus, a second round of microsatellite isolation was initiated, focusing speciﬁcally on candidates that had unique genotype distribution among spring run (Greig et al., 2003). Development of new statistical methods (Banks et al., 2003; Rosenberg et al., 2003) also assisted this quest for increased stock identiﬁcation power. Although research continues, recent data raise the unambiguous component for spring-run identiﬁcation to greater than 99% (Banks and Jacobson, 2004).

Figure 29-2. Factorial correspondence analysis using data from seven microsatellites determined to have greater information content for winter run discrimination (Banks et al., 2000. Data for 500 samples per Chinook life history type were simulated using WHICHLOCI (Banks et al., 2003) and analyzed using GENETIX (Belkhir et al., 2002). Square plots represent individual salmon samples colored in accord with their life history type and positioned on the graph in relation to their genetic relationship. Discrete and independent clustering of winter (light gray) demonstrates that these loci do indeed allow clear discrimination of winter. Substantial overlap between spring (dary gray), fall (white), and late fall (black), however, demonstrates limited power for discrimination among these other three life history types using this particular set of microsatellites.

Stock Identiﬁcation for Conservation of Threatened or Endangered Species

615

IV. SNAKE RIVER SOCKEYE AND CHINOOK, OREGON–WASHINGTON AND IDAHO: BROODSTOCK SHORTFALLS, DISEASE AND FISHERY IMPACT ASSESSMENT Endangered species listings for salmon of the Snake River, one of the major tributaries of the Columbia River system, include sockeye, spring/summer Chinook, and fall Chinook. The endangered sockeye (Oncorhynchus nerka) migrate an impressive 1,450 km from the ocean through Oregon–Washington and pass a total of eight dams before reaching their namesake, Redﬁsh Lake, Idaho, to spawn. These were the ﬁrst salmonids to receive ESA listing (Waples et al., 1991a), and since that time, only nine adults have returned to the lake. The extreme nature of genetic sampling and mating choices in this scenario likely presents the salmonid equivalent of challenges faced by the California condor restoration program (Snyder and Snyder, 2000), but the two or three sockeye that returned in 2003 joined a well-stocked captive broodstock. How so? Anadromous sockeye of Redﬁsh Lake commingle with landlocked kokanee and a third O. nerka population known as “resident beach spawners.” Allozyme data did not provide for discrimination among these three populations. Thus, how could one choose mating counterparts to supplement the few returning sockeye in the captive broodstock project? Cummings et al. (1997) were able to use a mixed DNA ﬁngerprinting technique followed by development of a single locus probe to resolve differences among these three stocks. Fortunately, distinction was sufﬁcient to determine that juveniles captured on their outmigration from the lake were indeed most likely from the anadromous sockeye stock and thus appropriate to provide sufﬁcient number for the captive broodstock program. Disease problems associated with rearing of eggs and juveniles in the captive broodstock program also required careful stock-speciﬁc consideration before arriving at a difﬁcult decision. The infectious hematopoietic necrosis virus (IHNV), for which there is no known effective treatment, was identiﬁed in approximately 80,000 eggs in rearing for a planned release into Redﬁsh Lake in spring 2002. After extensive review and consultation among the broodstock oversight committee, university, state, and federal agency expertise, these juveniles were destroyed because there was no evidence that Redﬁsh sockeye stock had ever been exposed to IHNV. Chinook that spawn in the Snake River, while not as rare as Redﬁsh Lake sockeye, face similar dam and reservoir obstruction in their navigation toward upper elevation spawning grounds. Likewise, they have sustained precipitous declines in the past few decades. These stocks have traditionally been thought to be comprised of three different spawning populations delineated after the primary time of their freshwater spawn return: spring, summer, and fall. However, genetic

616

Michael A. Banks

analysis pending ESA listing petitions (NOAA Fisheries, 1991) revealed that spring and summer shared very similar life history and allozyme allele frequency traits (Mathews and Waples, 1991), while fall returnees were strikingly different (Waples et al., 1991b). Thus, only two evolutionary signiﬁcant units (ESUs) were designated (spring–summer and fall). An interesting note here is that unlike the early allozyme ﬁndings in California’s Central Valley that could not resolve differences among the four runs there, genetic relationships among Snake River Chinook were resolved using protein electrophoresis. In fact, recent allozyme data now resolve differences among the California runs adequately (Winans et al., 2001; Waples et al., 2004). In the Snake River, Marshall et al. (2000) took matters a step further by using allozyme electrophoresis to resolve candidate non–Snake River falls from the original samples used in the status review characterization. Toward similar ends as the individual-based methods discussed above, but through alternate means, Marshall et al. (2000) iterated through mixed-stock analysis results (using methods discussed below) followed by individual-based removal of contaminant samples to result in 100% singlelineage samples (based on pairwise genotype matching for more diagnostic markers). The Columbia River system supports a great many different salmon stocks. This and a number of other issues confound assessment of the impact of ﬁsheries on the above three endangered and other threatened stocks from this region. Yet, precise estimates of endangered species loss in this harvest are critical to the effectiveness of federal biological opinion documents and stock rebuilding plans. Primary difﬁculties all result from the fact that endangered stocks, by deﬁnition, have very low numbers in comparison to nonthreatened, often hatcherysupplemented stocks. As a result, recoveries from physical tagging methods for endangered stocks are very minimal, and signiﬁcant bias problems have been identiﬁed when using traditional genetic data and mixed stock analysis techniques for identifying rare stock components (Pella and Masuda, 2001; Scribner et al., 1998). These all limit conﬁdence in how to assess and test the relative merits of alternate ﬁshery management strategies. In summary, Snake River sockeye are thought to be least impacted by ﬁsheries in the Paciﬁc Fishery Management Council (PFMC) area because ﬁshing strategy there primarily uses hook-and-line gear targeted toward Chinook and coho, and troll catches have not exceeded 100 sockeye in any year since 1985. The relative likelihood that these 100 may comprise signiﬁcant loss of Snake River sockeye is minimal given the much greater abundances of sockeye runs originating in the Fraser River, Puget Sound, and other stocks of the Columbia River basin. Three lines of evidence indicate minimal ﬁshery impact on spring/summer Chinook of the Snake River: spawning return for spring run to freshwater peaks well before initiation of the ﬁshery, and both coded wire tag and genetic stock identiﬁcation methods afﬁrm minimal impact (Waples et al., 2004). Allozyme-based GSI

Stock Identiﬁcation for Conservation of Threatened or Endangered Species

617

methods applied to the spring-run harvest in winter gill-net ﬁsheries in the lower Columbia River between 1987 and 1992, however, demonstrated Snake River harvest of up to 20% (Shaklee et al., 1999). Snake River fall Chinook have a broad marine distribution, with evidence for harvest ranging from southern California through to Alaska (PFMC, 1995). As a result, there is substantial uncertainty around loss estimates in ocean ﬁsheries, but most estimates are around 1% or less (Berkson, 1991). Concerns for Chinook along with depressed coho stocks have constrained ﬁsheries substantially in recent years. Evidently, Snake River endangered salmon still pose a number of unresolved issues. Resource interests ranging from ocean ﬁsheries, hydroelectric power, agriculture, and conservation concerns have pushed for improved understanding and resolution for better management. Recent focus has turned to DNA-based markers with projects headed by scientists at the NOAA Fisheries, Northwest Fisheries Science Center, University of Idaho, the Columbia River Inter-Tribal Fish Commission, Eagle Hatchery on the Snake River (Idaho Department of Fish and Game), and the Hatﬁeld Marine Science Center (Oregon State University). It will be worth noting further developments in the application of GSI and the challenges that result from the spiking of this large watershed with turbines and other human-derived complications.

V. STOCK IDENTIFICATION AND U.S.–CANADA SALMON WARS: EQUALIZING INTERCEPTION RATES; POLITICS AND RESOLUTION FOR GSI Extensive migration of salmon during their ocean life stage, often through several state and country borders, poses signiﬁcant challenges to management of ﬁshery allocation and often mounts signiﬁcant political opposition. Annual harvests of close to 500 million pounds of salmon in Oregon, Washington, British Columbia, and southeast Alaska have triggered either cooperation or discord between U.S. and Canadian ﬁshery management parties. Each of these ﬁsheries harvest some component of salmon that originated in rivers outside their own territories, which are called “interceptions.” Factors contributing to disagreement are many but center largely on disproportional interception components, ﬂuctuation of various stock abundances, alternate value and migratory patterns among the ﬁve salmonid species, and competing status among the various ﬂeets that operate in these ﬁsheries. This brief summary of factors contributing to disagreement in equitable allocation covers many complications. For example, Chinook, coho, and sockeye have higher value ($1 to $2.25 per pound) and are therefore preferred to chum and pink salmon, which have lower value ($0.15 to $0.60 per pound). Yet these species are not evenly distributed among the three ﬁshing territories. Chum and

618

Michael A. Banks

pinks comprise the major component of southeast Alaska ﬁsheries; chum, pinks, and sockeye predominate in British Columbia; and Washington and Oregon harvest all ﬁve species in approximately equal proportions. Further, their abundance and survival rates do not vary in concert among the territories. Depressed Chinook and coho stocks in Washington/Oregon have raised major concern for interceptions in British Columbia and Alaska, yet relatively high abundance of salmon in Alaskan waters has reduced willingness to curtail ﬁsheries in more northern waters. The Paciﬁc Salmon Treaty of 1985, created by the Paciﬁc Salmon Commission, headquartered in Vancouver, B.C., had two primary motives: conservation (to provide for optimum production along with prevention of overﬁshing) and equity (to ensure that each party receives beneﬁts equivalent to production of salmon originating in their waters). Indeed, achievements in the ﬁrst 10 years were many, including the following: changes in Canadian and Alaskan ﬁshing seasons that reduced interceptions, reallocation of Fraser River pink and sockeye salmon that resulted in increased Canadian harvests but also preserved the U.S. ﬁshery (owing to stock enhancement), coho conservation plans, and the formation of numerous technical committees that have resulted in effective management targeted toward speciﬁc species. However, failure of the Chinook rebuilding program has forced matters to a crisis in the past decade. An impasse in the mid1990s revolved around Canadian ﬁsheries being hit on both southern and northern borders by U.S.-based interceptions. The emphasis by the United States on conservation, given depressed stocks in their southern regions, was viewed by Canada as a diversionary tactic to avoid resolving the equity issue. Canada, insisting on reduced interceptions from both sides, demonstrated willingness to overﬁsh their own stocks to press for U.S. concessions. Public outcry along with the notorious collapse of the Newfoundland cod ﬁshery resulted in a marked aboutface in August 1995 with agreement between the Canadian ﬁshery minister and Washington State governor to reduce salmon ﬁsheries in both waters (Huppert, 1995). Annexes and related agreements ﬁnalized in the Paciﬁc Salmon Treaty of June 30, 1999, substantially changed objectives and structure for Chinook assessment and management. Previous quota or ceiling ﬁsheries were replaced by the following: 1. Aggregate abundance-based management (AABM) for stocks that occur in large areas and affect complex aggregates of many stocks. These are managed to achieve speciﬁc and varied harvest rates in accord with ﬂuctuating abundance estimates; larger catches are allowed during more abundant years and are constrained when abundance is down. AABM ﬁsheries include Southeast Alaska troll, net, and sport ﬁsheries; Northern British Columbia troll and Queen Charlotte Islands sport ﬁsheries; and West Coast Vancouver Island troll and outside sport ﬁshery.

Stock Identiﬁcation for Conservation of Threatened or Endangered Species

619

2. Individual stock-based management (ISBM) for stocks that occur in marine waters close to their river of origin or directly in rivers. ISBM is aimed at harvesting hatchery-produced salmon or species other than Chinook. While these ﬁsheries may involve some Chinook harvest for depressed stocks, they will fall under the “general obligation” that losses will be at reduced exploitation rates relative to the “base period” of 1979–1982. ISBM ﬁsheries include central British Columbia troll, net, and sport ﬁsheries; southern British Columbia marine net, troll, and sport ﬁsheries (other than West Coast Vancouver Island troll and outside sport ﬁshery); and all net, sport, and troll marine and freshwater ﬁsheries in Washington, Oregon, and the Snake River basin in Idaho (Paciﬁc Salmon Commission, 1999/2000). Details given in this section emphasize the continued need for better determination of the origin of stocks. The Chinook Technical Committee, one of many under the Paciﬁc Salmon Commission, is initiating innovations toward these ends. They have contracted with seven ﬁsheries genetics research laboratories distributed from California to Alaska, including Fisheries and Oceans in Vancouver, Canada, in a collaborative project to characterize genetic variation among major Chinook stocks that contribute to these ﬁsheries. The major goal of this collaboration is to develop and standardize state-of-the-art molecular and statistical techniques to maximize the resolution of stock identiﬁcation among Chinook across the Paciﬁc Northwest. Overall, this work will provide tools to better understand variation in ocean migration, dispersal, and residence of stocks, and resolve more equitable allocations among ﬁshery parties.

VI. SUPPORTIVE BREEDING OF PINK SALMON AT RISK OF EXTINCTION: GENETIC MARKER AND STATISTICAL MEASURE COMPARISON Fall-run pink salmon of the Dungeness River in Puget Sound, Washington State, is an example of a depressed stock at risk of further loss through ﬁshery harvest that is under the administration of ISBM. Pink salmon are unique in their ﬁxed return after a 2-yr ocean residence and maturation, a strategy that is so strict that odd- and even-year spawning populations are reproductively isolated. Runs that occur within the same year but differ temporarily in terms of spawning time, however, have much lower genetic distinction. Summer and fall runs of the Dungeness River presented an interesting problem for stock identiﬁcation because of two primary issues. First, genetic distinction between summer and fall run is low [on the order of 2% (Olsen et al., 2000)]. Second, summer run outnumbered fall run at the ratio of 15 : 1; summer run maintained roughly 6,000 potential

620

Michael A. Banks

breeders per year, yet fall-run numbers had reduced to lower than 400 returns per year in the late 1980s. Thus, a supportive breeding project was initiated to restore fall numbers. Although spawning times and locations for these two runs are largely isolated, an overlap in August combined with overwhelming greater numbers of summer run meant that using only timing and location for sampling candidate fall-run broodstock would involve a likelihood of sampling summer run that was too large to ignore. Olsen et al. (2000) compared the utility of 25 allozymes, 7 microsatellites, and a number of statistical measures to maximize conﬁdence that candidates taken for broodstock were indeed fall run. Shriver et al. (1997) had established that markers with large allele frequency differences among populations, which he calls “population speciﬁc alleles,” were most useful for estimating population identity. Olsen et al. (2000), however, argues that in the absence of selection, allele frequencies differ among populations randomly, driven largely by forces of genetic drift. Thus, marker types with fewer characters (alleles) and lower heterozygosity, such as allozymes, are not necessarily less informative for estimating population afﬁliation than more polymorphic markers such as microsatellites. Findings made in this study were interesting. While estimators of mean allele frequency such as the commonly applied genetic distance measure, Fst or q (Weir and Cockerham, 1984), were approximately the same for allozymes and microsatellites, the latter were consistently more useful for population assignment, indicating that q was not a good predictor of marker utility. However, the allele frequency differential (d) of Shriver et al. (1997), which measures cumulative rather than mean variance in allele frequencies, was larger for microsatellites and thus a better predictor. The relationship between number of loci used and number of baseline samples correctly assigned was also revealing. Using relatively few of the more diagnostic microsatellites was sufﬁcient to assign most baseline samples, and this proportion of correctly assigned samples did not improve much when using more loci; but the likelihood of assignment of individuals (i.e., the conﬁdence in the assignment call) always improved when adding more loci, including a noticeable increase with the addition of allozyme information.

VII. APPLICATIONS IN ALASKA (AK): GSI FOUNDATION, POLYMORPHIC MICROSATELLITES AND ZERO FREQUENCIES, ENDANGERED SPECIES INTERCEPTION, AND PERSPECTIVE Alaska has no threatened or endangered salmonids originating in its waters and therefore should not qualify for discussion under the general topic of this chapter. However, exclusion of Alaska’s ﬁsheries, its management structure, and its expertise would be a signiﬁcant oversight for a number of reasons. First, about 80%

Stock Identiﬁcation for Conservation of Threatened or Endangered Species

621

of the total Northwest salmon ﬁshery harvest is taken in Alaskan waters (Meacham and Clark, 1994), often in very complex mixed ﬁsheries. It thus is no surprise that the primary population-based mixed-stock analysis method applied across the board for genetic stock identiﬁcation was developed by Alaska’s ﬁshery research expertise. Earlier versions were developed for application using allozyme data at the Northwest and Alaska Fisheries Center, NOAA Fisheries Auke Bay Lab (Fournier et al., 1984; Millar, 1987; Pella and Milner, 1987). More recent versions were further developed, with several rounds of modiﬁcations to accommodate microsatellite data by the Genetics Laboratory, Alaska Department of Fish and Game, under the name Statistical Package for Analysis of Mixtures (SPAM; Debevec et al., 2000). Recent issues of concern in application have occurred because of the great number of characters (alleles) that typify microsatellites identiﬁed to be more powerful in individual-based assignment methods. In many mixed ﬁshery contexts, there are often many more of these alleles per locus than number of contributing stocks—described as an “embarrassment of riches” by Pella and Masuda (2001). This results in a common occurrence of zero frequencies for certain alleles not observed among some of the contributing stocks (often simply as a result of sampling constraints). One approach to be considered is to assign rare alleles into a few, more common “bin-alleles” that would be less likely to have zero frequency among contributing populations. This, however, is difﬁcult to achieve without loss of information that may enhance stock discrimination (Pella and Masuda, 2001). Earlier mixed-stock analysis versions could not handle zero frequencies, forcing constraint on study design of mixed-stock ﬁshery analysis. A newly written application, called Genetic Mixture Analysis (GMA; Kalinowski, 2003), was speciﬁcally written to overcome this difﬁculty, and SPAM (Version 3.7) also incorporates changes that resolve this issue. A second major reason why applications in Alaskan ﬁsheries should feature in a chapter focusing on GSI challenges from ESA-listed species is because of the large migration distances traveled by salmon during the ocean life stage. While Alaska might not have any of its own ESA-listed stocks, its harvest of 80% of all Paciﬁc Northwest salmonid catch inevitably must intercept some ESA-listed stocks originating from more southern states. Some issues relating to this were covered in section V, but further issues from Alaska’s perspective are given here, per Meacham and Clark (1994): Salmon stocks in Alaska depreciated substantially during the ﬁrst half of the twentieth century, with President Dwight Eisenhower declaring Alaska a federal disaster area in 1953. At statehood in 1959, total salmon harvest was about 25 million salmon, 20% of today’s current sustainable production. Only through focused, local, and consistent restoration and management have numbers returned to those championed today. Despite a number of proposals, Alaska has chosen not to develop any hydroelectric schemes, in contrast to a number of more southerly located states. Consequently, the relative

622

Michael A. Banks

numbers of reasonable harvest vs. endangered species loss is argued from an Alaskan viewpoint. For example, it is estimated that Snake River fall Chinook salmon are taken in the southeastern Alaskan ﬁshery in a ratio of about 1 in 5,000 ﬁsh. Thus, any reduction to prevent loss of Snake River Chinook will cost Alaskan ﬁsheries on the order of 5,000 ﬁsh per every ESA-listed Chinook saved. Alaska is not slow to point out that the relative number proportions for Snake River fall Chinook salmon at risk for loss associated with hydropower systems on the Columbia River are much more biased toward ESA-listed species loss, yet power turbine operations continue unabated.

VIII. APPLICATION IN THE HIGH SEAS: PROBLEMS CAUSED BY MISSING STOCKS The High Seas Salmon Program at the University of Washington, Seattle, has been a stronghold of research directed toward ﬁshery management since its founding by William F. Thompson (ﬁrst director of the UW Fisheries Research Institute) in the early 1950s. Understanding migration patterns of various salmon stocks throughout the Paciﬁc Rim and how these patterns affect their risk for overexploitation by ﬁsheries in the high seas has remained a primary research objective. Scientiﬁc and political achievements through collaboration mediated by the International North Paciﬁc Fisheries Commission [INPC, now the North Paciﬁc Anadromous Fish Commission (NPAFC)] have been substantial. First, large tagging programs from 1955 through 1978 demonstrated extensive interception of both Asian and North American salmon stocks in the high seas ﬁshery, providing data for restrictions on high seas drift-net ﬁsheries and their eventual banning by international law in 1993. Primary federal programs responsible for research related to international legislation of Paciﬁc Salmon management include The Ocean Carrying Capacity and Stock Identiﬁcation programs stationed at the Auke Bay NOAA Fisheries Laboratory. This research is based on genetic characterization of ﬁshery samples and comparison with extensive genetic baselines (primarily allozyme based) as well as a suite of mixed-stock analysis techniques. For example, Wilmot et al. (2000) describe an assessment of the origins of salmon seized from the F/V Arctic Wind. This vessel was sighted with approximately 4 miles of drift nets in waters at 45°24¢ N, 171°58¢E, central Paciﬁc Ocean, despite the global moratorium on large-scale drift nets [Section 307(a)(M) of the Magnuson–Stevens Fishery Conservation and Management Act 16 U.S.C. 1857(a)(M), United Nations General Assembly Resolution 44-225, Large Scale Pelagic Driftnet Fishing and Its Impact on Living Marine Resources of the World’s Oceans and Seas]. While most of the chum salmon analyzed from the F/V Arctic Wind were likely of Asian origin, 70% of the sockeye and about half of the Chinook were most likely of North American origin.

Stock Identiﬁcation for Conservation of Threatened or Endangered Species

623

Many of the problems and solutions posed by endangered species considerations discussed here are relevant in the high seas, but the likelihood of ﬁshery sample origin of stocks other than those characterized in baseline data is particularly important in this context. Early mixed-stock analysis techniques such as conditional maximum likelihood (Fournier et al., 1984; Millar, 1987; Pella and Milner, 1987) had no explicit way of dealing with this problem. They assumed that relative frequencies of markers among baseline stocks were known without error, and variance in composition estimates were assessed through simulation and via bootstrapping. In addition, no algorithms existed to make use of ﬁshery sample data to upgrade or inform baseline relative frequency data. The unconditional maximum-likelihood method (Pella and Milner, 1987) and unconditional least squares method (Xu et al., 1994) presented a ﬁrst solution by providing estimates of both stock proportions and relative genotype frequencies among baseline source population data through optimizing a criterion of ﬁt to counts. No method to resolve global optima in these approaches exists, however (Smouse et al., 1990), and none of the above three methods makes use of relative frequencies among stocks to improve estimates of relative frequencies in separate stocks. Further, the accuracy of these methods fails increasingly when true stock contributions are uneven; as some stocks become more abundant, their contributions are underestimated, while rare or even absent stocks are overestimated. Pella and Masuda (2001), in the application of Bayesian methods, present an appealing new innovation to deal with these issues. Their approach uses Markov Chain Monte Carlo estimation to make use of information gained through assessing relatedness among contributing stocks, along with using information from the mixed ﬁshery sample to improve baseline data for contributing stocks (including cases where certain stocks harvested in a ﬁshery may be absent in starting baselines). Pella and Masuda (2001) illustrate their methods with two applications. The ﬁrst example considers harbor porpoise mitochondrial DNA (mtDNA) data (Rosel et al., 1999) from four different summer breeding populations and the question of relative presence of each of these breeding populations in a wintering group. As is typical for much mtDNA data, frequency distribution among harbor porpoise haplotypes had a number of common haplotypes shared among populations and many more rare haplotypes sporadically distributed among breeding populations. Such a distribution presents a challenge for mixed-stock estimation because rare haplotypes observed in the mixed samples have alternate most likely sources when they are absent from baseline samples. Indeed, ﬁndings in comparing results attained from different methods afﬁrmed biased estimation using earlier methods and demonstrated more realistic estimations attained from the Bayes method of Pella and Masuda (2001), although improvement came at the expense of wider conﬁdence limits. The second example considers steelhead data from allozymes, microsatellites, and mtDNA from two populations. These data

624

Michael A. Banks

are complicated because only one of these populations (above a barrier) could be sampled separately. Apparently, some individuals survive the falls separating these populations to recruit into the second population located beneath the barrier. Thus, this second population was admixed. In resolving alternate origins of components in the lower (admixed) population, allozyme and microsatellite data both provided similar estimates of the above barrier component that agreed with an earlier assessment of 0.25 (based on one allozyme marker that was ﬁxed for a unique allele above the barrier), but mtDNA estimates were higher, with substantially larger conﬁdence limits. Other examples using this method include assessing the origin of pink and chum salmon seized from the F/V Petropavlocsk in the Bering Sea 7 miles into the U.S. Exclusive Economic Zone (Kondzela et al., 2002). The vessel captain claimed that these salmon were purchased from Russian ﬁshermen off the east coast of Kamchatka; however, estimates using conditional maximum likelihood from earlier versions of SPAM (Debevec et al., 2000) included about a 10% component, most likely of North American origin. Estimates of 100% Asian stock origin attained using the same data, but using methods of Pella and Masuda (2001), again exposed the bias problem of earlier methods. The honest F/V Petropavlocsk captain was thus off the hook! A second example using mtDNA data to assess the rigor of component estimates concerns rookery representation in a mixedstock aggregation of sea turtles on foraging grounds (Bolker et al., 2003). Here, accuracy of assessment is important to conservation applications for endangered or threatened stocks. Are substantial numbers or the endangered breeding populations really there or not? As for porpoises, sea turtle mtDNA data have a few common haplotypes and a large number of rare haplotypes, posing a challenge to traditional maximum-likelihood methods. True to earlier form, the method of Pella and Masuda (2001) resulted in assessment of considerably wider conﬁdence intervals, precluding management conclusions that endangered rookeries presence was either not there at all or in substantially larger numbers than earlier estimates. Bolker et al. (2003) state explicitly that because this hinges on sampling uncertainty for rare haplotypes, conﬁdence intervals yielded by Bayesian methods are appropriate, and conﬁdence intervals from maximum-likelihood methods can be badly overoptimistic (from the perspective of rare ESA-listed species).

IX. CONCLUSIONS AND PROSPECTS FOR THE FUTURE In summary, challenges posed by endangered and threatened species to stock identiﬁcation methods have resulted in substantial molecular, genetic, and statistical advances. Microsatellites, with their inherent large numbers of characters, have largely provided for information shortcomings but also raised further statistical and sampling error complications. A number of new approaches and

Stock Identiﬁcation for Conservation of Threatened or Endangered Species

625

statistical innovations have met these challenges. It is important to note, however, that details covered above are a reﬂection of the current state of the art. New challenges continue to emerge demanding continued technological and statistical improvements. Thus, the ﬁeld of GSI is changing rapidly, making it well worth noting its emerging developments. For example, a number of ﬁsheries genetics laboratories are currently tooling up with equipment that allows high-throughput characterization of a marker type named single nucleotide polymorphisms (SNPs). These are distributed throughout all genomes and can be located in either gene coding or noncoding regions. Their maximum number of characters (alleles) is four because there are only four different nucleotides in DNA, but SNPs most often occur in only two different forms (biallelic). So they would likely have the same sort of information content as allozymes and intuitively seem less useful than microsatellites (Luikart, 2003). Equipment advances (at a cost), however, champion data gathering at more than a million SNP genotypes per day. Thus, one could then harness the same amount of information from a much more extensive region of the genome through characterizing half as many SNPs as the total number of microsatellite alleles, although linkage associated with the recently discovered haplotype blocks presents an important consideration (Wall and Pritchard, 2003). Further, population sample size requirements for biallelic loci are very much more forgiving, and the number of populations one might choose to represent in baselines are likely much more expansive given resources to characterize a million SNPs a day! One high-throughput SNP characterization technology also boasts similarly high rates for microsatellite characterization (Van Den Boom et al., 2001). Time will show which of these developments are indeed realized. Population genomics is another new ﬁeld likely to bear fruit useful for GSI. Broadly, this is the simultaneous study of multiple genetic loci or genome regions to enable a better understanding of the forces that affect evolutionary processes such as mutation, genetic drift, migration, and natural selection. Black et al. (2001) narrowed this concept further in describing the importance of using population genomics to separate out locus-speciﬁc effects such as selection, mutation, assortive mating, and recombination from genome-wide effects such as drift through reduced effective population size (bottlenecks), gene ﬂow through migration, and inbreeding. This distinction is important from an evolutionary perspective because neutral loci will be similarly affected by demography and evolutionary history, but loci under selection will reﬂect different forces and reveal different patterns of variation. Luikart et al. (2003) designate these loci with alternate patterns of variation as “outliers” and draw attention to much recent interest in the literature in using the population genomics approach to identify “outliers” and thus adaptive molecular evolution. The authors’ paper provides useful exposition of potential bias caused if one inadvertently uses outlier loci to draw evolutionary inferences and describe a number of new statistical and

626

Michael A. Banks

software developments for identifying locus effects. On the other hand, these same outliers are likely choice candidate loci for discrimination among populations for ﬁshery management. Thus, population genomics may provide boons in ﬁelds other than those currently considered. In short, molecular genetics and statistical innovations are hot on the chase to overcome new challenges raised in their application toward stock discrimination for ﬁshery management. The ﬁeld is a fascinating exposition of the relative strengths and weaknesses posed by new methods combined with unique sensitivities that promote complementary application of both new and old.

REFERENCES Banks, M., Eichert, W., and Olsen, J. 2003. Which genetic loci have greater population assignment power? Bioinformatics 19: 1436–1438. Banks, M. A., Baldwin, B. A., and Hedgecock, D. 1996. Research on chinook salmon (Oncorhynchus tshawytscha) stock structure using microsatellite DNA. Bulletin of National Research Institute of Aquaculture: Suppl. 2: 5–9. Banks, M. A., Blouin, M. S., Baldwin, B. A., Rashbrook, V. K., and Fitzgerald, H. A. et al., 1999. Isolation and inheritance of novel microsatellites in chinook salmon (Onchorhynchus tschawytscha). Journal of Heredity 90: 281–288. Banks, M. A. and Eichert, W. 2000. WHICHRUN (Version 3.2): a computer program for population assignment of individuals based on multilocus genotype data. Journal of Heredity 91: 87–89. Banks, M. A. and Jacobson, D. 2004. Which genetic markers and GSI methods are more appropriate for deﬁning marine distribution and migration of salmon? North Paciﬁc Anadromous Fish Commission Technical Note 5: in press. Banks, M. A., Rashbrook, V. K., Calavetta, M. J., Dean, C. A., and Hedgecock, D. 2000. Analysis of microsatellite DNA resolves genetic structure and diversity of chinook salmon (Oncorhychus tshawytscha) in California’s Central Valley. Canadian Journal of Fisheries and Aquatic Sciences 57: 915–927. Bartley, D., Gall, G. A. E., Bentley, B., Brodziak, Gomulkiewicz, J. R. et al. 1992. Geographic variation in population genetic structure of chinook salmon from California and Oregon. Fishery Bulletin, U.S. 90: 77–100. Belkhir, K., Borsa, P., Chikhi, L., Raufaste, N., and Bonhomme, F. 2002. GENETIX 4.04, logiciel sous Windows TM pour la génétique des populations. Laboratoire Génome, Populations, Interactions: CNRS UMR 5000, Université de Montpellier II, Montpellier, France. (http://www.univmontp2.fr/%7Egenetix/genetix/genetix.htm Berkson, J. 1991. Coded wire tag analysis on Snake River spring, summer, and fall chinook. Letter dated February 19, 1992, to NMFS, ETSD from Portland, Oregon. Columbia River Inter-Tribal Fish Commission, Portland. Bingham, N. and Harthorne, A. 2000. Spring-run Chinook Salmon Work Group: A cooperative approach to watershed management in California. In E. E. Knudsen, C. R. Steward, D. D. MacDonald, J. E. Williams, and D. W. Reiser (eds.), Sustainable Fisheries Management. Lewis, New York. Black, W., Baer, C., Antolin, M., and DuTeau, N. 2001. Population genomics: genome-wide sampling of insect populations. Annual Review of Entomology 46: 441–469. Blanchong, J. A., Scribner, K. T., and Winterstein, S. R. 2002. Assignment of individuals to populations: Bayesian methods and multi-locus genotypes. Journal of Wildlife Management 66: 321–329.

Stock Identiﬁcation for Conservation of Threatened or Endangered Species

627

Bolker, B., Okuyama, T., Bjorndal, K., and Bolten, A. 2003. Sea Turtle stock estimation using genetic markers: accounting for sampling error of rare genotypes. Ecological Applications 13: 763–775. Brun, M., Sabbagh, D., Kim, S., and Dougherty, E. R. 2003. Corrected small-sample estimation of the Bayes error. Bioinformatics 19: 944–951. California Department of Fish and Game. 2002. Sacramento Winter-Run Chinook Salmon. 2000–2001 Biennial Report prepared for the Fish and Game Commission by the Habitat Conservation Division. Clark, F. 1940. California salmon catch records. California Fish and Game 26: 49–66. Cornuet, J.-M., Piry, S., Luikart, G., Estoup, A., and Solignan, M. 1999. New methods employing multilocus genotypes to select or exclude populations as origins of individuals. Genetics 153: 1989–2000. Cummings, S., Brannon, E., Adams, K., and Thorgard, G. 1997. Genetic analysis to establish captive breeding priorities for endangered Snake River sockeye salmon. Conservation Biology 11: 662–669. Debevec, E. M., Gates, R. B., Masuda, M., Pella, J., and Reynolds, J. et al. 2000. SPAM (Version 3.2): statistics program for analyzing mixtures. Journal of Heredity 91: 509–510. Fournier, D., Beacham, T., Riddell, B., and Busack, C. 1984. Estimating stock composition in mixed ﬁsheries using morphometric, meristic, and electrophoretic characteristics. Canadian Journal of Fisheries and Aquatic Sciences 41: 400–408. Greig, C., Jaconson, D., and Banks, M. A. 2003. New tetranucleotide microsatellites for ﬁne-scale discrimination among endangered Chinook salmon (Oncorhynchus tshawytscha). Molecular Ecology Notes 3: 376–379. Hedgecock, D., Banks, M., Rashbrook, V., Dean, C., and Blankenship, S. 2001. Applications of population genetics to conservation of chinook salmon diversity in the Central Valley. Fish Bulletin 179: 45–70. Huppert, D. 1995. Understanding the U.S.–Canada Salmon Wars: Why the Paciﬁc Salmon Treaty Has Not Brought Peace? http://www.wsg.washington.edu/salmon/huppertreport.html Kalinowski, S. 2003. Genetic Mixture Analysis 1.0., Department of Ecology, Montana State University, Bozeman MT. Kondzela, C., Hawkins, S., Guthrie III, C., and Wilmot, R. 2002. Origins of sockeye and chum salmon seized from the the F/V Petropavlovsk (NPAFC doc. 598). Auke Bay Fisheries Laboratory, Alaska Fisheries Science Center, NOAA Fisheries, Juneau, AK. Luikart, G., England, P., Tallmon, D., Jordan, S., and Taberlet, P. 2003. The power and promise of population genomics: from genotyping to genome typing. Nature Reviews Genetics 4: 981–994. Marshall, A. R., Blankenship, H. L., and Commor, W. P. 2000. Genetic characterization of naturally spawned Snake River fall-run chinook salmon. Transactions of the American Fisheries Society 129: 680–698. Matthews, G. and Waokes, R. 1991. Status review for Snake River spring and summer chinook salmon. NOAA Technical Memorandum NMFS-F/NWC-200. Meacham, C. and Clark, J. 1994. Paciﬁc salmon management—the view from Alaska. Alaska Fishery Research Bulletin 1: 76–80. Millar, R. B. 1987. Maximum likelihood estimation of mixed stock ﬁshery composition. Canadian Journal of Fisheries and Aquatic Science 44: 583–590. NOAA Fisheries, 1991. Status review for Snake River spring and summer Chinook salmon. NOAA Technical Memorandum NMFS-F/NWC-200, June 1991. NOAA FISHERIES. 1994. Endangered and threatened species; status of Sacramento River winter-run chinook salmon. Federal Register vol. 59 no. 2. pp. 440–450, January 4, 1994. NOAA FISHERIES. 1998. Endangered and threatened species: West Coast chinook salmon; listing status change; proposed rule. Federal Register 63 # 45. p. 11482.

628

Michael A. Banks

Olsen, J. B., Bentzen, P., Banks, M. A., Shaklee, J. B., and Young, S. 2000. Microsatellites reveal population identity of individual pink salmon to allow supportive breeding of a population at risk of extinction. Transactions of the American Fisheries Society 129: 232–242. Paciﬁc Fishery Management Council. 1995. Preseason report III: Analysis of Council-adopted management measures for 995 ocean salmon ﬁsheries (with appendices). Paciﬁc Salmon Commission, 1999/2000. Annual Report for 1999/2000. Appendix D, Annex IV, Chapter 3. Paetkau, D., Calvert, W., Stirling, I., and Strobeck, C. 1995. Microsatellite analysis of population structure in Canadian polar bears. Molecular Ecology 4: 347–354. Pella, J. and Masuda, M. 2001. Bayesian methods for analysis of stock mixtures from genetic characters. Fisheries Bulletin 99: 151–167. Pella, J. and Milner, G. B. 1987. Use of genetic marks in stock composition analysis, In N. Rayman and F. M. Utter (eds.), Population Genetics and Fisheries Management. University of Washington Press, Seattle, WA. Rosel, P., France, S., Wang, J., and Kocher, T. 1999. Genetic structure of harbor porpoise (Phocoena phocoena) populations in the Northwestern Atlantic based on mitochondrial and nucelar markers. Molecular Ecology 8: S41–S54. Rosenberg, N., Li, L., Ward, R., and Pritchard, J. 2003. Informativeness of genetic markers for inference of ancestry. American Journal of Human Genetics 73: 402–1422. Sahu, S. K. and Cheng, R. 2003. A fast distance-based approach for determining the number of components in mixtures. Canadian Journal of Statistics 31: 3–22. Scribner, K. T., Crane, P. A., Spearman, W. J., and Seeb, L. W. 1998. DNA and allozyme markers provide concordant estimates of population differentiation: analyses of US and Canadian populations of Yukon River fall-run chum salmon (Oncorhynchus keta). Canadian Journal of Fisheries and Aquatic Sciences 55: 1748–1758. Shaklee, J. B., Beacham, T., Seeb, L. W., and White, B. 1999. Managing ﬁsheries using genetic data: case studies from four species of Paciﬁc salmon. Fisheries Research 43: 45–78. Shoemaker, J. S., Painter, I. S., and Weir, B. S. 1999. Bayesian statistics in genetics: a guide for the unitiated. Trends in Genetics 15: 354–358. Shriver, M. D., Smith, M. W., Jin, L., Marcini, A., and Akey, J. M. et al. 1997. Ethnic-afﬁliation estimation by use of population-speciﬁc DNA markers. American Journal of Human Genetics 60: 957–964. Smouse, P., Waples, R., and Tworek, J. 1990. A genetic mixture analysis for use with incomplete source population data. Canadian Journal of Fisheries and Aquatic Sciences 47: 620–634. Snyder, N. and Snyder, H. 2000. The California Condor: A Saga of Natural History and Conservation. Academic Press, San Diego, CA. Van den Boom, D., Jurinke, C., McGinniss, M., and Berkenkamp, S. 2001. Microsatellites: perspectives and potentials of mass spectrometric analysis. Expert Reviews of Molecular Diagnostics. 1: 383–393. Wall, J. and Pritchard, J. 2003. Haplotype blocks and linkage disequilibrium in the human genome. Nature Reviews Genetics 4: 587–597. Waples, R. S., Johnson, O., and Jones, R. 1991. Status review for Snake River sockeye salmon. NOAA Technical Memorandum, Northwest Center, Seattle. Waples, R. S., Jones, R., Beckham, B., and Swan, G. 1991. Status review for Snake River fall chinook salmon: NOAA Technical Memorandum, NMFS-F/NWC-201. Waples, R. S., Teel, D. J., Myers, J. M., and Marshall, A. R. 2004. Life history divergence in Chinook salmon: historic contingency and parallel evolution. Evolution, in press. Weir, B. S. and Cockerham, C. C. 1984. Estimating F-statistics for the analysis of population structure. Evolution 38: 1358–1370.

Stock Identiﬁcation for Conservation of Threatened or Endangered Species

629

Wilmot, R., Kindzela, C., Guthrie III, C., Moles, A., and Pella, J. et al. 2000. Origins of sockeye and chum salmon seized from the F/V Arctic Wind (NPAFC doc). Auke Bay Fisheries Laboratory, Alaska Fisheries Science Center, NOAA Fisheries, 11305 Glacier Highway, Juneau, Alaska. Winans, G. A., Viele, D., Grover, A., Palmer-Zwahlen, M., and Teel, D. et al. 2001. An update of genetic stock identiﬁcation of chinook salmon in the Paciﬁc Northwest: test ﬁsheries in California. Reviews in Fisheries Science 9: 213–237. Xu, S., Kobak, C., and Smouse, P. 1994. Constrained least squares estimation of mixed population stock composition from mtDNA haplotype frequency data. Canadian Journal of Fisheries and Aquatic Sciences 51: 417–425. Yoshimaya, R. M. 1999. A history of salmon and people in the Central Valley region of California. Reviews in Fisheries Science 7: 197–239.

This Page Intentionally Left Blank

CHAPTER

30

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice CORNELIUS HAMMER* AND CHRISTOPHER ZIMMERMANN† *Federal Research Centre for Fisheries, Institute for Baltic Sea Fisheries, Rostock, Germany, † Institute for Sea Fisheries, Hamburg, Germany

I. Introduction II. Where Information of Any Kind Would Help: Deepwater Stocks in the North Atlantic III. Stocks May Be Merged When Data Sampling Is Reduced—Example: Herring in the North Sea IV. When Biological Properties Change—Example: Arctic Cod V. When Information on Stocks is Underutilized— Example: Cod in the Baltic Sea VI. Stock Structure and Genetics—Example: Deepwater Redﬁsh in the Northwest Atlantic VII. Example for a More Holistic Approach: Horse Mackerel in European Waters VIII. New Approaches in Science: Redﬁsh Again IX. Summary References

I. INTRODUCTION There has been ample description in this book of how stocks are deﬁned, and whoever has made his way bravely to this chapter from the start should be sufﬁciently puzzled. It could at least be expected that, on their well-intended way to better understanding of what stocks are, ﬁsheries managers have accumulated doubt. It is understandable and acceptable if ﬁsheries managers do not consider Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

631

632

Cornelius Hammer and Christopher Zimmermann

this satisfactory and demand clearer deﬁnitions from science. It also has become clear from browsing through this volume’s Table of Contents that there are a number of valid approaches to characterize or to deﬁne a stock. The question remains: Is this of any help for ﬁsheries managers? The answer is typically scientiﬁc: yes and no. The heterogeneous approaches to identify stocks reﬂect the complexity of the nature of biological units. However, a second issue is with the management itself and lies in the speciﬁc deﬁnitions of stock in managerial concepts. While in an introductory chapter of this book (Waldman, Chapter 2) deﬁnitions of management and stock units have been given, it must be stressed that ﬁsheries managers and scientists often mean different things when talking about stocks. There has been ample discussion and a number of deﬁnitions of what a stock is from a scientiﬁc and managerial point of view. No matter what these deﬁnitions are speciﬁcally based on and even though it is difﬁcult to agree on what constitutes a stock (Gauldie, 1991), the common basis is that stocks are relatively homogenous and self-contained populations whose losses by emigration and accessions by immigration, if any, are negligible in relation to the rate of growth and mortality (Anon. 1960, cited in Jamieson, 1974). According to Ihssen et al. (1981), a stock is an intraspeciﬁc group of randomly mating individuals with temporal and spatial integrity, or (Carvalho and Hauser, 1994) stocks are discrete biological entities with varying degrees of temporal and spatial integrity. Thus, a stock is considered a group of individuals with self-sufﬁcient recruitment, even though it may gain from recruits of other stocks due to speciﬁc hydrographic conditions and the local drift patterns. Waples (1998) deﬁned a stock as a group of organisms whose demographic/genetic trajectory is largely independent from other such groups. The latter deﬁnition is, however, of little help unless it is clearly deﬁned what demographic/genetic trajectories are. This is complex enough but the stock concept is further complicated because the boundaries are by nature diffuse and stocks and species have the natural tendency to expand their distribution area. This tendency is a prerequisite for the radiation of species and the occupation of new habitats, for instance after a major ecological catastrophe or in periods of climatic change. It is essential for the ecology and evolution that species tend to expand their distribution area constantly. Naturally this expansion leads to some degree of mixing of stocks and, in marine species, there is usually an exchange of individuals associated with some degree of gene ﬂow. The key question is therefore, as Waples (1998) put it, how much gene ﬂow occurs among stocks? In general, as Roques et al. (2002) phrased it, marine species have been typically characterized by very weak or no obvious genetic population structure over large geographic areas, a feature potentially attributed to large effective population sizes, high potential for dispersal, and weak physical barriers to gene ﬂow (Avise et al., 1987; Gold et al., 1994).

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice

633

In other words, stocks are biological units with soft boundaries in terms of space and time. The best part of this book has dealt with the question of how to determine these ﬂexible boundaries and shall neither be discussed nor summarized here. However, apart from the deﬁnition of the singularity and distinction of stocks, an important feature of a stock (and the most important feature for the management) is its speciﬁc reaction to exploitation. It is an integral assumption that a stock responds independently to harvesting, no matter how narrow or wide the boundaries around a stock are set. For instance, the stock development of saithe (Pollachius virens) in the North Sea and west of Scotland and Rockall area was very similar throughout the 1990s, even though the degrees of ﬁshing pressure were different in both areas. As a result, the concept of two independent stocks was abandoned, both were united for assessment purposes and considered one stock (ICES, 2001a). This seems to make sense biologically since saithe is a semidemersal or semipelagic ﬁsh and is known to undertake fast and relatively large migrations in the northern North Sea area and adjacent waters. For scientists, the core of the stock concept is biological. If stocks are to a large degree independent biological entities, management can assign exploitation rates and patterns to them, with the underlying assumption that there is a stockspeciﬁc sustainable yield (e.g. Gulland, 1983, Hilborn and Walters, 1992). While this is compelling for biologists, it is less so for managers. Often stocks are deﬁned by managers as groups of ﬁsh exploited or harvested in a speciﬁc area or by a speciﬁc method (Carvalho and Hauser, 1994). A number of different stock deﬁnitions have been proposed for practical use, as follows. Harvest stocks are, according to Gauldie (1988), locally accessible ﬁsh resources in which ﬁshing pressure on one resource has no effect on the abundance of ﬁsh in another contiguous resource. This deﬁnition could be identical with the biological stock deﬁnition, and, in contrast to most stock deﬁnitions, this concept does not imply any genetic or phenotypic difference (Carvalho and Hauser, 1994). It only describes a group of individuals whose abundance depends to a very much larger degree on recruitment and mortality, especially that caused by ﬁshing, than on immigration and emigration (Carvalho and Hauser, 1994). Such a stock deﬁnition may become problematic when ﬁsh as a part of a larger stock are locally exploited. The effect on the population in the contiguous areas may not be signiﬁcant, not be noticed, or disappear in the statistical noise due to compensatory immigration. However, especially in highly structured stock complexes, this may well have severe inﬂuence on the entire stock, for example, on the adaptability to environmental changes. An example is the Downs herring in the North Sea and will be discussed later. More contradiction could be in using the terms ﬁshery stocks (Smith et al., 1990) or targeted stock, which by managers are often deﬁned as a group of ﬁsh exploited in a speciﬁc area or by a certain method, that is, the ﬁsh caught in a certain management unit. However, mainly in the older ﬁshery biological

634

Cornelius Hammer and Christopher Zimmermann

publications, the terms ﬁshery stock or ﬁsheries unit stock are also used by biologists, without being clear whether they are supposed to be different from a biological stock (e.g., Anon., 1960, cited in Jamieson, 1974). One of the main problems at present is that the management units have traditionally grown and are not adjusted to either the changes in distribution of stocks or to the change of scientiﬁc perception of the particular stock boundaries. Once stocks are acknowledged by statistical records and catch records per area and species have been established, it is often difﬁcult to abandon the stock deﬁnition and to redeﬁne it according to the real biological properties. One other reason for the inﬂexibility of the management units is the allocation of Total Allowable Catches (TACs). These are set in most cases annually and refer scientiﬁcally to a biological stock unit, but managerially to an area. A suggested TAC corresponds to a particular ﬁshing mortality (which is a scientiﬁc measure for the ﬁshing pressure exerted on a stock) and applies to the accessible stock (Holt, 1959), which may or may not be conﬁned to a distinct area in the sea and to a biological entity. In many cases, both coincide. However, in many other cases they do not. And, as Jamieson (1974) put it, the ﬁshery biologists hopefully regard the unit stocks as units of management. There are numerous examples for the mismatch of biological stock deﬁnition and management units to which a TAC applies. In this context it should be kept in mind that the discrepancy in understanding what a stock may be is nothing but new and exists since the stock concept, TACs and management units were introduced. In summary, the core of the confusion is that different stock concepts came into use and were used simultaneously without causing all too many problems or confusion for managers, politicians, and scientists. This occurred simply because the stocks were relatively large at the beginning and the ﬁshing pressure relatively low. As a result, semantic differences of what each party meant when talking about a stock were not as important as they are nowadays, where in all too many cases the stocks are low and the ﬁshing pressure far too high. However, the problems faced nowadays by managers, politicians, and scientists are manifold and are not simply solved by introducing clear and binding deﬁnitions.

II. WHERE INFORMATION OF ANY KIND WOULD HELP: DEEPWATER STOCKS IN THE NORTH ATLANTIC There are a number stocks in the North Atlantic with a mismatch of the level of knowledge about the stocks and their exploitation. For most deepwater species, this mismatch is outstanding. It is widely recognized that deepwater ecosystems, including deepwater ﬁsh stocks, are highly vulnerable to exploitation (e.g., Merrett and Haedrich, 1997), can be depleted very quickly (Koslow et al., 2000),

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice

635

and will recover slowly, or even extremely slow (ICES, 2001a). As a result, ﬁshing such stocks resembles to some degree the nature of mining rather than exploitation, and by no means resembles the nature of sustainable harvesting. Unfortunately, stocks are often exploited before information is available on the biology, productivity, and distribution of the species and before time-series of ﬁsheries data are available for stock assessments. Consequently, reliable information on the state of stocks and the potential for ﬁsheries frequently lags behind exploitation. It is now generally agreed in the scientiﬁc community that because of the general sensitivity of these stocks and the inadequate knowledge of the biology of deepwater species, initial exploitation rates should be very low (ICES, 2001b). The general approach should be not to signiﬁcantly exploit these resources until good progress has been made in understanding their biology and stock dynamics (Large et al., 2001). Since 1994 the International Council for the Exploration of the Sea (ICES) concentrated on collating background information on what was known about deepwater species and ﬁsheries in the ICES area (ICES, 1994, 1996). This work continues and the data collated have been sufﬁcient to attempt preliminary assessments of some species. Descriptions of deepwater ﬁsheries were prepared, available ﬁsheries-independent survey data were worked up and archived, steps were taken to ensure that deepwater species which were landed or discarded were recorded, and the biology of both targeted and by-catch species was investigated. However, although the quality and quantity of ﬁsheries and biological data available for assessments have improved considerably in recent years (Gordon, 1999, Magnússon et al., 1997; Menez et al., 2001), there are still a number of important areas where further work is required (Large et al., 2001). These are primarily basic biological and stock-speciﬁc data of the fundamental biological processes such as growth, feeding, maturation, and fecundity. Knowledge on the deepwater species still lags considerably behind that of the commercially exploited shelf-based species. Areas where our current knowledge is particularly poor are recruitment processes and their variation, stock identity, ﬁsh migration, and ﬁsh behavior. There have been only few studies on the stock structure of deepwater ﬁsh species in the ICES area (ICES, 2000a; Menezes et al., 2001). For assessment purposes, stock units have been deﬁned on the basis of current knowledge of species distribution and similarity of observed catch-rate trends between ICES areas (ICES, 1998). Thus, stock units are currently individual or groups of ICES subareas or occasionally ICES divisions. This is not ideal because the ICES statistical areas are devised for the ﬁshery on the continental shelf and are, in many instances, inappropriate for deepwater ﬁsheries (Large et al., 2001). For example, ICES subarea VI (Fig. 30-1) is divided into two divisions. Division VIa covers the shelf along the continental margin and VIb the Rockall Plateau. Division VIa, however, includes both the Rockall Trough and a part of the Faroe-Shetland

90

60

NAFO

Svalbard

IBSFC

NEAFC

I

IIb

45

XIVa

Latitude N

75

Barents Sea

regulatory area

70

IIa

Bear Isl.

Greenland Va

Norwegian Sea

65

Iceland

XIVb

Faroes

Vb

Reykjanes Ridge

North Sea

VI

Rockall Bank

60

III

IV

Hatton Bank

Baltic Sea

55

Irminger Sea

XII

North America Newfoundland

VII

St. Lawrence

50

Europe

Bay of Biscay Grand Banks

50

40

30

20

40

Mediterranean Sea

10

0

35

31 10

20

Longitude E Lat N

Longitude W 60

IX

Azores

NAFO

45

VIII

X

IIa

62

30

Vb1

Gulf of Bothnia

Vb2 29

West of Scotland

IVa(W) IVa(E) IIIa

VIa(N)

VIb

VIIc

VIIk

28

60 58

Gulf of Riga

Central Baltic

25

56

26

24 VIIa(N)

VIIa(S) VIIg VIIf Celtic Sea

VIIe

IVc VIId

54

IIIc (22)

Irish Sea

VIIb

VIIj

27

IIIb (23)

North Sea IVb

VIa(S) West of Ireland & Porcupine Bank

Western Baltic

32

52

Southern North Sea & Channel

50 Zi03

VIIh

Lon W

12

8

4

0

4

8

12

16

20

48

Longitude E

FIGURE 30-1. Fisheries convention areas (upper panel: NAFO, Northwest Atlantic Fisheries Organisation; NEAFC: Northeast Atlantic Fisheries Commission; IBSFC, International Baltic Sea Fisheries Commission); ICES areas and selected ﬁshing grounds in the North Atlantic (middle panel, 200 and 1000m depth contour drawn); and ICES Divisions in the Western Area, North Sea, and Baltic Sea as used for herring stock assessments (lower panel, 200m depth contour drawn). Map source: Gebco. 636

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice

637

Channel. The deepwater ﬁsh faunas of these two areas have little in common (Gordon, 2001). Division VIb extends westward from the Rockall Plateau and is contiguous with subarea XII at longitude 18°W and, in doing so, bisects the Hatton Bank, with a rapidly developing deepwater ﬁshery that takes place in international waters. Subarea XII covers a vast area of the northeastern Atlantic that includes large parts of the Mid-Atlantic and Reykjanes Ridges. While it may be reasonable to assume a stock separation between the slopes of the Rockall Trough and Mid-Atlantic Ridges, the Hatton Bank probably has more afﬁnity with the Rockall area. However, a proportion of the landings from subarea XII cannot be readily attributed to the Hatton Bank and are therefore excluded by the ICES Study Group from the assessments of the Rockall area. Thus, there is an urgent need to reconﬁgure some existing ICES areas to become biologically meaningful in terms of the distribution of deepwater species. Methods such as otolith microchemistry may be a useful tool for stock discrimination. The European Commission funded such a study on black scabbard as a contribution to the BASBLACK project (EC DGXIV 97/84) and is currently funding a FAIR project entitled “Otolith microchemistry as a means of identifying stocks of deepwater demersal ﬁsh” (EC FAIR 98/4365). The objective is to use the chemical signal embedded in otoliths to discriminate between stocks of deepwater species. DNA studies may also be useful in this context (ICES, 2000b). Studies of the population genetics of the red (blackspot) seabream and alfonsinos using DNA-based analyses have been carried out in the Azores (EC DGXIV 97/081).

III. STOCKS MAY BE MERGED WHEN DATA SAMPLING IS REDUCED—EXAMPLE: HERRING IN THE NORTH SEA Stock identity of North Sea herring has been a great issue since the mid-1960s. Indeed, the interest in this subject is even greater today. The results of the ICES working group dealing with such issues (ICES, 1965) acknowledged the existence of three major spawning components in the North Sea, that is, the Buchan herring, spawning close to the east Scottish Coast in late summer or early autumn (August, September), the Bank herring, spawning on or at the slopes of the Dogger Bank during early autumn (September, October), and the Downs herring with its spawning grounds in the central and Eastern Channel and the Southern Bight, which are frequented for spawning in winter (December, January) (Fig. 30-2). Differences in spawning season are the most prominent distinctive features of the three stocks. At a closer look, however, there were also consistent differences in meristic and morphometric characteristics (Zijlstra, 1969) which indicate that the majority of the herring return to spawn with the stock to which they recruited,

638 Lat N

Cornelius Hammer and Christopher Zimmermann

61

60

59

58

57

56

55

54

53

52

Total North Sea Autumn Spawner Catch (IIIa, IV & VIId)

700 51

Lon W 2

1

0

1

2

3

catch and quotas ('000)

600 500

4 4005

North Sea Catch (IV & VIId) 6

7 North 8 Sea 9 TAC 10(IV 11 & VIId)Lon E

300 200 100

"Downs" Catch (IVc & VIId) "Downs" TAC (IVc & VIId)

0 1990

1992

1994

1996 year

1998

2000

2002

FIGURE 30-2. Herring catches in the North Sea, 2001, by ICES statistical rectangle (divisions IVc and VIId are highlighted); comparison of total North Sea autumn spawner catches, herring catches in the North Sea, catches in Divisions IVc and VIId, and TACs for the North Sea and IVc and VIId 1989–2002 (inset). From ICES, 2002, redrawn.

suggesting that the majority of the recruits spawn on their parental ground, that is, they behave exactly as the deﬁnitions for stocks require. Based on a number of biological parameters, such as number of gill rakers, mean egg size (Burd, 1974), migration pattern, and spawning time (HardenJones, 1968), it was concluded in the 1970s that after full recruitment to the

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice

639

spawning group, a year-class of North Sea herring appears to maintain its integrity and association to different stock units until its extinction. Although Buchan, Bank, and Downs herring mix during the feeding season, there is no evidence of any signiﬁcant interchange between the fully recruited spawning groups. According to the stock concept as described above, this should be sufﬁcient evidence to treat the spawning entities as truly independent stocks from an assessment and managerial point of view. However, with the depletion of the stocks in the late 1970s, the collection of biological information also subsided, and at present it is not entirely clear if the recruits of the combined North Sea autumn spawning stock contribute to the Downs Herring component or if, vice versa, this component contributes to the recruitment of the combined North Sea stock. If this is an uncertainty today, it will have probably also been one in former times when the Downs herring was considered a separate stock, but apparently it did not matter as long as sufﬁcient biological information was collected from the Downs herring ﬁshery (age structure, maturity at age, weight at age, etc.) to maintain an independent assessment. With the decline of the herring ﬁshery in the Channel area, the collection of biological data was considerably reduced with a little time lag. Having too little information to substantiate an independent assessment, the Downs herring stock assessment was integrated in the major North Sea stock assessment in the mid1990s (ICES, 1995), with a separate sub-TAC for this stock component. During the fourth quarter, the spawning aggregations of the Downs herring are extraordinarily dense and, for this reason, easily accessible for the ﬁshery. These ﬁsh aggregations encourage the ﬁshery to overexploit the spawning concentrations and to exceed the separate sub-TAC. There is evidence that exceeding catches are concealed by misreporting in the North Sea. Under these circumstances a depletion of the Downs herring is easily possible, which could lead to a loss of genetic diversity. The Downs herring is one of the southernmost herring stocks in the Northeast Atlantic, and a loss of this stock could have severe implications for the adaptability toward environmental changes (e.g., increase of temperature) of the whole North Sea herring complex. The North Sea herring management is presently regarded as an example for sensible ﬁsheries management. However, apparently even a successful management regime based on TACs and technical measures still bears the potential of a depletion of substocks or stock components. The discrepancy in stock deﬁnition and perception from the scientiﬁc and the managerial side has another irritating side effect, as the difference between both concepts is used as an argument not to adhere to scientiﬁcally proposed TACs. As a result, a tendency prevails to allow higher TACs in view of the larger area for which the TAC is now valid. For most exploited stocks in the world, regular or irregular assessments provide a picture on the stock size and the recruitment into the spawning stock biomass. Both are the cornerstones for the scientiﬁc

640

Cornelius Hammer and Christopher Zimmermann

proposals for a TAC for this particular stock. Therefore, it should be expected that internationally agreed, ratiﬁed, and ofﬁcially enforced TACs are set in accordance with those scientiﬁcally proposed. However, it is well known that this has not always been the case. For instance, for herring in the central Baltic Sea, a TAC of 117,000 tonnes was proposed by the respective Advisory Committee for Fisheries Management (ACFM) of ICES (ICES, 2000c, 2001b). A total of 476,000 tonnes were eventually internationally agreed. As a consequence of the overexploitation of the stock and the subsequent further decline, only 95,000 tonnes were proposed in the following year (2000) and still 405,000 tonnes were agreed, decreasing to 60,000 tonnes proposed for the following year (2001) and 300,000 tonnes agreed (ICES, 2001b). (For details on TACs and assessments of herring stocks, see http://www.clupea.net/stocks). Similar examples could be given for many other stocks.

IV. WHEN BIOLOGICAL PROPERTIES CHANGE— EXAMPLE: ARCTIC COD In a perfect world, science would have simple and unequivocal means to separate stocks. Biological parameters relevant for the assessment of ﬁsh stocks could then be attributed to speciﬁc stocks, which would be the prerequisite to explore changes in biological properties. However, as these means for separation on a single stock level are still not available, science can only estimate to which degree changes in biological parameters are caused (or masked!) by stock mixing (or stock movements) or by “real” factors effecting a deﬁned stock. Note, however, that scientists, using biological parameters for the separation of stocks, must be aware that the stock deﬁnition might not be valid anymore in due course as these parameters can drastically change within a few years. The following section intends to illustrate these problems, which clearly lie beyond stock separation, but have a similar important inﬂuence on assessment and management. Managers come together to decide on the fate of ﬁsh stocks. Even though it is willingly accepted by them that ﬁsh stocks are biological entities, they perceive a stock as a given asset and assessments, models, and parameters are reluctantly tolerated as means of description. The outcome is usually crudely condensed to very few numbers, such as tonnage spawning stock biomass, landings, and number of recruits. The fundamental perception of managers of ﬁsh stocks is not very much different from that of cattle on the meadows or trees in the forest, thus being a fairly static asset which exists to be pruned. However, there are considerable differences. The core of the dilemma for managers is that what is true about a stock and understood today may not be true any longer next year. Biological properties of stocks often change, which is sometimes caused by climatic effects, changes in the predator–prey relationships,

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice

641

or is due to impacts of the ﬁshery in the ecosystem or on the stock itself. For instance, the age at which ﬁsh become mature shifts, often signiﬁcantly, toward maturing at younger ages with increasing ﬁshing pressure on the stock. Once this happens, the spawning stock comprises more age groups than before, increasing numerically the spawning stock and reducing numerically the stock of juveniles. Such shifts depend not only on the exerted ﬁshing pressure but also on nutritional factors, thus often being dependent on the ﬁshing of other prey stocks. In negotiations for TACs, a stock is qualiﬁed only by its size, that is, the tonnage of the spawning stock biomass. This parameter is not enough. In analogy with forests being affected by acid precipitation and weakened in their production or cattle being affected by infections or chronic accumulations of pesticides, ﬁsh stocks need an additional qualiﬁer with regard to stock composition and production. An example for such a stock is the Northeast Arctic cod. The Northeast Arctic cod stock is one of the most important cod stocks in the world and has supported a large Norwegian and Russian ﬁshery for many years. Landings from the 1940s to the mid-1970s have been on average about 800,000 tonnes annually (approx. 440,000–1.5 million tonnes) (ICES, 2002b). However, such an exploitation rate was unsustainable and the spawning stock biomass declined in the late 1940s and throughout the 1950s from more than 1 million tonnes to an average of approximately 200,000 tonnes between the mid-1950s and the late 1980s. Some larger year-classes brought the stock temporarily up in the 1990s. However, ﬁshing mortality in 1997–2001 was among the highest observed and well above the precautionary and even the limit reference points. The excessive catches again drove the stock rapidly below the safe biological limit of 500,000 tonnes, and the stock was outside safe biological limits in 2001 and 2002. In 2000 the spawning stock biomass (SSB) had declined to about 22,000 tonnes only. At the same time, the TACs were set considerably higher than recommended by ICES and approximately 415,000 tonnes were caught. For the following year the catches increased to 426,000 tonnes, and the SSB even increased slightly to 298,000 tonnes (ICES, 2001b) (Fig. 30-3). Such exploitation is hardly sustainable for a cold-water adapted cod stock if recruitment and growth is not exceptionally good. However, in Arctic cod this was not the case. The recruitment throughout the 1990s was only a little above average. In such a situation of apparent mismatch of SSB size, growth, recruitment, and catches, a decrease of the SSB would not be surprising, even if the models predict some likelihood for a moderate increase of the SSB in the long term. However, in 2001–2002 the SSB markedly increased again, as is obvious in Figure 30-3. The unforeseen reason for this was that the majority of the spawning stock comprises ﬁrst-time spawners and that the maturity at age had shifted toward younger individuals. For this reason, the SSB grew by inclusion of age groups that hitherto only had a much smaller percentage of maturing individuals. Such biological shift of the maturity also occurs with other cod stocks and a shift toward

642

Cornelius Hammer and Christopher Zimmermann

Spawning Stock Biomass and Landings (Mill. t)

1.4 Landings

1.2

Spawning Stock Biomass

1.0 0.8 0.6 0.4 0.2 0 1952

1960

1968

1976 Year

1984

1992

2000

FIGURE 30-3. Spawning stock biomass and landings of Northeast Arctic cod until 2002.

younger age groups enlarges, of course, the SSB and leads to an increase of the SSB. Naturally, the same may occur in the opposite direction, leading to a sudden reduction of the SSB. The sudden increase in the SSB is, of course, welcomed by all parties interested in the ﬁsh. However, the scientiﬁc surveys in the North Arctic cod (and many other stocks and species!) showed that the eggs and larvae of ﬁrst-time spawners are less viable than those of other mature ﬁsh and that the overall spawning period is reduced when the spawning stock consists of fewer age groups. Both these factors can reduce the reproductive potential of the stock for the same biomass. For these reasons, not only the absolute size of the SSB is of importance but also the age structure of the parental animals. A spawning stock that consists predominantly of young ﬁsh that are only at the beginning of the biologically reproductive period is less productive and produces eggs with a lower survival rate. As a result, the stock becomes more vulnerable if the age structure is reduced to a predominance of relatively young parental ﬁsh. In the case of the Arctic cod, a cannibalism problem also occurred in the mid1990s. Due to the absence of the preferred prey, which is the capelin, the adult cod foraged increasingly on young individuals of the same species. If such an effect is not noticed during extensive surveys (which requires regular stomach

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice

643

analyses) the year-class predicting models will develop a bias and thus will predict inaccurate developments.

V. WHEN INFORMATION ON STOCKS IS UNDERUTILIZED—EXAMPLE: COD IN THE BALTIC SEA There are a number of rather clear stock deﬁnitions available (see above), and if there are morphometric, meristic, physiological, temporal, and/or spatial differences between two groups of ﬁsh of the same species, there should be enough evidence to deﬁne stocks, separate them, and manage them accordingly. This is at least what could be expected as evidenced by this volume, dealing primarily with stock identiﬁcation. So much for the theory. Reality is often notably different. In the Baltic Sea, for example, there are two main cod stocks, the western and eastern stock, geographically separated by Bornholm Island, but there still seems to be a considerable exchange of individuals between the two areas due to passive or active immigration of 0-group ﬁsh from the western to the eastern area (Otterlind, 1966; Hinrichsen et al. 1999; Oeberst, 2001) and a spawning migration into the adjacent areas, also eastward to the main spawning grounds of the cod stock of the central Baltic Sea (Berner, 1967, 1981; Bagge et al., 1974; Bagge and Thurow, 1993) (Fig. 30-4). Berner (1980a,b) summarizes that the cod stock in the southeastern part of the spawning ground of the western stock is always a mixture of the two main cod stocks and thus, the spawning migration occurs in two directions. Once ﬁsh reach the area of the contiguous stock, there apparently is no backward migration, since the homing behavior of cod is not well developed (Otterlind, 1965, 1985; Bagge, 1983). Even though this exchange seems to be considerable, the two stocks differ from each other by clear morphometric and meristic characteristics, by hemoglobin types, otolith structure, allele frequencies of some enzyme coding loci (e.g., Aro, 2000), and by spawning area and time (e.g., Aro, 1989). The differences are so signiﬁcant that both are described as the subspecies Gadus morhua morhua in the western Baltic Sea, and Gadus morhua callarias in the eastern (Kändler, 1944; Sick, 1961; Otterlind, 1962; Jamieson and Otterlind, 1969). For this reason, it must be assumed that, despite the migratory exchange, these well-distinguished entities are clearly separated in their biology and form two different stocks, which, according to all deﬁnitions, should be managed separately. This assumption, however, is not the case despite the biological knowledge available and even though the ﬁshery in the Baltic Sea was the ﬁrst in the Atlantic region that was regulated by an international convention (1931) (Thurow, 1974a,b). The regulation is remarkable since during those days, the catches generally were low and in the range of 4,000 to 6,000 tonnes only (Thurow, 1974a,b).

644

Cornelius Hammer and Christopher Zimmermann 12

14

16

18

20

22

24

26 Lat N

10

Gulf of Bothnia 60

60 Gulf of Finland

Skagerrak

58 Go tla nd De ep

58

Kattegat 56

56 Bornholm Deep

Gdansk Deep

Spawning area Nursery area Spawning migration Drift of young fish

54 Baltic cod

Transition area cod 10

12

14

16

18

54

100 m depth contour

20

22

24 Lon E 26

FIGURE 30-4. Spawning grounds and migration of western and central Baltic cod. After Bagge and Thurow (1993), modiﬁed by Aro (2000).

Since then, the catches have increased, of course. The average landings have reached almost 40,000 tonnes since the early 1970s. Moreover, while the average age of the cod in the catch for both regions was 5 to 6 years in the 1930s (Kändler, 1944), it is now on average 3 years ICES (2001b), and large quantities of even younger individuals are discarded. Approximately 80% of the catches in the western Baltic are cod at the age 2 to 3. At the end of the year, even age group 1 grows into the catches. However, only 10% of the 2-year-old ﬁsh and only 65% of the 3-year-old ﬁsh are mature. For this reason, between 35% and 90% of the cod catches legally and ofﬁcially registered are juveniles, which certainly cannot be considered sensible management. So far, the western and central Baltic stock have been treated as one when it comes to the allocation of the TAC, set by the International Baltic Sea Fisheries Commission (IBSFC), and with regard to speciﬁc ﬁshery regulations. Even though there is a separate TAC calculated for each stock, they have been united. Only

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice

645

very recently (2002), international negotiations have started aiming at separating the TACs and implementing a differentiated management, to no avail. The reason for not changing the management units and introducing a more effective management is simply that if this were the case and a new management system was implemented, some of the many member countries negotiating the issue would have to cope with negative side effects and probably decreased national quota or restrictions to ﬁshing grounds. For this reason, an obviously insensible and counterproductive management system remains in practice even though better knowledge is available. Nevertheless, a change of the management units in the Baltic Sea is deemed very urgent since ﬁshing pressure on the declining central Baltic stock is now diverted in an unsustainable manner to the western stock to compensate for lost ﬁshing opportunities. In addition, ﬁshing pressure is very likely to increase for the western and central Baltic Sea cod stocks with the dramatic decrease of the North Sea cod stock and the introduction of stricter ﬁshing regulations there in 2003. It is hoped that with the inclusion of Poland, Latvia, Lithuania, and Estonia in the European Community, new and better management practices will become possible. Fishery negotiations from 2005 on will take place between the Community and Russia, and a great part of the managerial problems will not be discussed in the IBSFC any longer but bilaterally between the Community and Russia. The example of the cod stocks in the Baltic shows that for stock management, it is not always enough to have a clear stock deﬁnition and scientiﬁc separation with all additional biological information on stock identiﬁcation, boundaries, migration, exchange, and so on. The political decision to manage the entire resource sensibly and according to the precautionary principles is a prerequisite. However, this example also shows that in some cases all necessary biological information is readily at hand for the managers and only needs to be utilized.

VI. STOCK STRUCTURE AND GENETICS—EXAMPLE: DEEPWATER REDFISH IN THE NORTHWEST ATLANTIC There are two key questions in managing ﬁsh stocks that frequently occur: First, is this a stock? And second, if this is a stock and it is depleted, how quickly will it rebuild and/or be replenished by immigration from outside? Usually biological data and assessment records will help. However, in widespread or highly migratory species, biological and assessment data are often not available or are not conclusive. In such cases a remedy has been sought in biochemical and genetic studies. The basic underlying assumption is that if there are biochemical or genetic differences between species, there should also be some difference between the stocks. Genetic studies are therefore often regarded as a remedy to the rather

646

Cornelius Hammer and Christopher Zimmermann

vague and incomprehensive biological stock deﬁnitions. Genetic studies seem a priori to be convincing since they are expected to produce a measured result. This result is expected to have the quality of a deﬁned quantity which in the test is either measured or it is not. In other words, either a threshold is passed, then there are two stocks, or this is not the case. Such an approach is appealing because the impression is given that eventually there is a method that provides an answer to an either-or question. In both instances, the answers are clear, and from a managerial point of view such a concept is supposedly an appropriate help in decision making, since managers have (understandably) other things to do than to contemplate answers such as “yes, but only to some degree, and to what degree depends on the circumstances.” As a result, much effort has been invested in biochemical and genetic studies on very different analytical levels. And indeed, this approach has produced usable results as, for instance, for some freshwater and anadromous salmonids (e.g. Ward et al., 1994) and Newfoundland cod (Ruzzante et al., 1996, 1998). However, in many other cases, conclusions for managers and nongeneticists are harder to draw from the studies than initially expected. Even though the above questions are usually the motive for ﬁnancing particular analyses, the answers are often soft, or at least difﬁcult to interpret. In an attempt to narrow the results down to answer the question whether the sampled entities are different stocks, the typical answer is again, and correctly, “to some degree,” since this is, unfortunately, how nature is. However, from a managerial point of view, this answer is either extremely unsatisfactory and difﬁcult to handle, or by contrast, it is used to support many different managerial options, driven by particular political interests. An example of such a dilemma is the allocation of the deepwater redﬁsh (Sebastes mentella) quota in the Northwest Atlantic. This species is widely distributed from west of Greenland to the Faroe Islands and further up to the Norwegian coast at depths of 300 to 750 m. Also, the juveniles are widely dispersed, and extensive migration between feeding, copulating, and larval release areas is undertaken (Reinert and Lastein, 1992). To manage the stocks and to allocate TACs, two stocks are assumed to exist, mainly based on morphometric analyses and the meristic traits between the ﬁsh from the two areas (Reinert and Lanstein, 1992). Along this line, some genetic evidence points in the direction of separate redﬁsh “types,” even between ﬁsh in the upper and deeper pelagic regions of the Irminger Sea (Johansen et al., 2000). In this context, the authors do not deﬁne what to their understanding is a “type” in relation to a “stock.” The investigation aimed toward the question of whether the phenotypical appearance of different redﬁsh types is genetically induced, and the authors came to the conclusion that there is some evidence for it. But it remains as unclear as it was before whether these types are members of different stock. Biochemical and genetic studies of these ﬁsh by Nedreaas and Nævdal (1989, 1991), Nedreaas et al. (1994), and Duschenko (1997) have shown that there is

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice

647

little difference at the level of allozymes and hemoglobin. And genetic investigations on a wider geographic range give evidence for distinctly different “population units,” as Roques et al. (2002) phrased it, with eastern (Norway and Barents Sea), pan-oceanic, and western (Gulf of St. Lawrence and offshore Newfoundland) units. The authors comment that their most important observation is the lack of genetic difference among samples within the pan-oceanic zone, from the Faroe Islands to the Grand Banks. As a result, the managers of the redﬁsh stock are still left in the dark, and the answer to their question of whether there are distinct redﬁsh stocks in the North Atlantic is again “to some degree.” Even attempts to discriminate S. mentella from other sympatric congeners (S. fasciatus and S. marinus) have, according to Roques et al. (2002), met with varying success, hampered our understanding of these species’ ecology, life history, and population structure, and led to the combined management of all taxa as a single “species.” Even though Roques et al. (2002) remain somewhat unspeciﬁc about what their understanding is of a “population unit,” it may well be assumed that a pan-oceanic population unit is what ﬁsheries biologists and managers refer to as stock. Taking the different approaches into account, it may be summarized that on a genetic level there are no differences within the pan-oceanic stock (e.g., Greenland–Irminger Sea–Iceland–Faroe Islands), even though there are clear morphometric and weak genetic differences between ﬁsh sampled at different locations within this area. All this does not sound promising. The core of the dilemma is that the genetic differences between species are generally small, and between stocks even very small. However, very small differences can be detected with modern analytical techniques, and the problem would not exist if there would be less genetic exchange, that is, less gene ﬂow between the groups. If the level of genetic differentiation is per se low and genetic drift is only moderately high, it may be expected that it is extremely difﬁcult to ﬁnd statistically signiﬁcant differences. Ward et al. (1994) found that 60% of marine stocks they investigated showed low levels of genetic differentiation, whereas gene ﬂow between the stocks was relatively high. High gene ﬂow is chieﬂy a result of the unstructured environment of the sea, enabling large active migrations and passive larval drift over a wide area. As such, the marine environment is generally a less structured environment than the terrestric environment. Moreover, there are some intrinsic problems involved, which have been described and summarized well by Waples (1998): Difﬁculties arise because the genetic “signal” indicating stock structure is relatively weak for high gene ﬂow species, and consequently various sources of noise in the analysis assume relatively greater importance than would normally be the case. It is often unclear whether apparently weak genetic divergence truly reﬂects the population structure in marine organisms or is, instead, an artifact of low analytical resolution (Roques et al., 2002).

648

Cornelius Hammer and Christopher Zimmermann

If the signal is weak, the usual statistical approach is to increase the sample size. In marine biology this is not easy to achieve, but is still easier than to acquire more ship time for a research vessel and to increase the numbers of stations where samples are taken. Moreover, increasing station numbers does not necessarily help. To statistically prove differences, a number of sampling requirements must be met. Without going into depth here, basically sampling requirements are very restrictive (and are frequently violated). Samples must be multinominal, that is, the samples must have been drawn randomly from a population of inﬁnite size. In a statistical sense, this means that at one randomly chosen station one ﬁsh is taken at one randomly chosen depth. In reality, hauls are made in a statistical rectangle where ﬁshing is possible or when vessel restrictions allow for it (weather, bottom structure, working time on board and so on). Great numbers of ﬁsh are taken from each haul and are believed to be representative not only of the area swept by the net but also of a given area, for example, the statistical rectangle, that is, the samples are representative by deﬁnition. The sample size from each station depends on the size of the catch and usually the processing restrictions on board. The catches may be large or very small, but both station location and analyzed ﬁsh from the haul are far from being randomly acquired. Such violation of this prerequisite can lead to signiﬁcant test results even if there are no differences between the populations. This type of error is important to consider because it is almost always the case that sampling protocols violate the assumptions (Waples, 1998). Sampling from biological populations is typically constrained in time and space, and often there are individuals in the population that have no chance of appearing in the sample. In most cases, therefore, the question is not whether the assumption about random sampling is violated at all, but only how badly, and what the consequences are (Waples, 1998). It is obvious that large areas at sea can only be sampled with expensive research vessels in a cost-effective way, usually with stratiﬁed surveys, and that it is in most cases impossible to meet all statistical requirements. In evaluating the results as well as in formulating research proposals and allocating costly ship time, these restrictions must be taken into account. In addition, there is a problem related to the sampling and the sample size. It is commonplace in statistics that the likelihood of ﬁnding statistically signiﬁcant differences between two samples increases with the number of samples taken. Unfortunately, the sample size needs to be large in population genetics because the differences that are sought are small from the outset (low thresholds) and, in addition, are overlaid by individual variation (noise). Statistical power is determined not only by the magnitude of the differences between populations, but also by data richness (sample size, number of samples, and number of independent characteristics measured). However, only the former is biologically meaningful for stock identiﬁcation, but the latter can have profound inﬂuence on the power of the statistical test (Waples, 1998). In other words, not all statistically

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice

649

signiﬁcant test results indicate biologically important differences (Waples, 1998). Statistical signiﬁcance should therefore not be the only basis in decision making for management. Conceptually, stocks are managed separately if the test proves there are signiﬁcant differences; if not, they are managed jointly. This is realized in conservation biology rather than in ﬁshery management, an approach that is, as Waples (1991) put it, appealing in its simplicity, but not without problems. The basic difﬁculty is that there is little reason to expect a direct relationship between statistical signiﬁcance and biological signiﬁcance. Each statistical result should therefore be accompanied by a biological plausibility check. To complicate matters, biochemical and genetic studies have undergone development as well. The use of hypervariable genetic markers, such as microsatellite loci, along with the development of new statistical methods, have signiﬁcantly improved the understanding of genetic population structure in marine organisms (Roques et al., 2002). Such studies reveal signiﬁcant genetic heterogeneity where a lack of structure was previously reported using other markers, such as allozymes or mitochondrial DNA (Roques et al., 2002). It has become evident that genetic studies may have great potential for population studies and are capable of demonstrating how closely species are related (e.g., Johansen et al., 2000, 2002) but greatly depend on the method employed and on sampling. It is decisive that thresholds and levels of resolution are deﬁned. In many cases, population structure analyzed on the microsatellite level coincides well with the speciﬁc ecology, for instance, for Newfoundland cod (Ruzzante et al., 1996, 1998). In other cases, results are less evident, and stock identiﬁcation requires additional meristic or morphological evidence. For stock identiﬁcation it remains important that these additional identiﬁers are congruent with genetic results. In regard to the redﬁsh in the North Atlantic it is hoped that based on recent genetic investigations, clearer guidance is given for management. Regarding stock level and TAC management, the long uncertainty about the stock structure of redﬁsh in the Northwest Atlantic was the reason for the somewhat helpless statement that “research continues to clarify the genetic relationships amongst the various forms, but regardless of future advances in that area, the morphological similarities among species and forms, and the overlapping distributions among them will continue to present difﬁculties for assessment and management of these resources” (ICES, 2001a). In summary, genetic studies are in many cases useful, but should be treated with great care. The expectations projected for this approach should be adjusted to the nature of the method, that is, analyzing differences on a very low threshold level. The lower this level is, the greater the inﬂuence of noise and biasing factors becomes. For this reason, the results of genetic studies should always be weighted against or with other biological parameters. For this

650

Cornelius Hammer and Christopher Zimmermann

reason, management decisions generally should, for this reason, not be based solely on the results of genetic studies but should be treated as one indicator among many.

VII. EXAMPLE FOR A MORE HOLISTIC APPROACH: HORSE MACKEREL IN EUROPEAN WATERS It is evident from the chapter on the use of genetic markers in stock identiﬁcation that relying on it as a single source of information can be helpful, misleading, or effectively wrong. The recommendation to couple the genetic information with other biological information is the modern approach. According to Carvalho and Hauser (1994), many diverse characteristics and methods have been used to analyze stock structure in exploited species with regard to ecology, tagging, parasite distribution, physiological and behavioral traits, morphometrics and meristics, calciﬁed structures, cytogenetics, immunogenetics, blood pigments, and molecular genetic tools (Ihssen et al., 1981; Kumpf et al., 1987), providing in theory a basis for tackling the stock deﬁnition problem from a wider angle. An example of such a more holistic approach in stock identiﬁcation and separation is currently being undertaken in an international project on stock separation of horse mackerel (Trachurus trachurus) in European waters, ﬁnanced by the European Commission. Stock deﬁnition for horse mackerel is still doubtful and largely based on managerial criteria rather than on scientiﬁc knowledge. Management areas for which TACs are applicable are based mainly on traditional ﬁshing practices. Horse mackerel, however, are known to migrate over great distances, and for management of the stocks, it is not clear how well management areas and TACs ﬁt the biology of the ﬁsh and its natural stock structure (Fig. 305). It may well be that, in the absence of easy-to-apply discrimination methods between stocks, horse mackerel in the Northeast Atlantic will be managed as one stock in the future, which would have signiﬁcant implications, for example, for the access of different ﬂeets harvesting this stock in areas that have been inaccessible to these ﬂeets up to now. The overall objective of this project (www.homsir.com) is the biological stock identiﬁcation of horse mackerel throughout its whole range, from the Northeast Atlantic to the Mediterranean Sea. This goal is being approached by integrating both established and innovative procedures such as genetic markers, other biological tags (morphometry, parasites), tagging experiments, and life history traits (growth, reproduction, and distribution). The research is therefore a model for use of improved multidisciplinary tools for ﬁsh stock identiﬁcation, ultimately to allow for enhanced management of horse mackerel in EU waters in short, medium and long term. The basic principles of this approach are as follows:

651

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice W 15

10

5

0

5

Spawning area Feeding areas: North Sea component

Lon E

10

Jul

Western component

Aug

Southern component Sep

Lat N

60

Oct-Nov ? Aug 55 Oct-Dec

? Oct

May-Jul, peak: Jun Dec-Apr 50 Apr-Jul, peak:Jun

45

Oct-Dec seasonal movements unclear ? peak: Apr

peak: May-Jun

Bergen

01

NO

Aberdeen

40

02*

Jan-Aug

IR

05

DE

Dublin

Hamburg

XX Nation 00 Sampling Site * additional sampling for life history traits

03

IR

peak: Feb-Mar

06

ES

07*

35

ES

21

ES

Vigo

08

IT

ES

Rome Lisbon

09* PT10 ES17

PT

14

20

Santander

PT

12* IT18

ES

13*

IT

15

GR

19

16*

GR

Crete

IT

11

ES

FIGURE 30-5. Horse mackerel: Schematic outline of assumed migration routes, spawning, and feeding areas for the three stocks; sampling locations of the EU project HOMSIR 2000 and 2001 (inset). A 200m depth contour is drawn. Map source: Gebco. From ICES (1992), redrawn.

• Identiﬁcation of horse mackerel populations and stocks from the Mediter-

ranean Sea and Northeast Atlantic using different molecular genetic markers (allozyme electrophoresis, mitochondrial DNA sequencing, DNA microsatellite, SSPC)

652

Cornelius Hammer and Christopher Zimmermann

• Evaluation of parasitic infection levels by means of quantifying

•

• •

•

epizootiological parameters (prevalence, mean intensity, and/or abundance of metazoan infection) and interpretation of the parasitology results in terms of host population biology (e.g., migration routes, feeding and spawning areas) Application of geometric morphometrics on the horse mackerel body and otoliths, as a tool for identiﬁcation of intraspeciﬁc variation, that is, testing of body and otolith shape differences among areas, by applying multivariate statistical techniques Use of physical tagging in selected areas to infer patterns of movement and degree of mixing among horse mackerel stocks or populations Examination of the different population parameters: growth rate, maturity ogive, fecundity, age structure, distribution, and abundance, to identify horse mackerel management units Assessment of the effects of extrinsic factors (ﬁshing pressure), space and time, in the population parameters observed

VIII. NEW APPROACHES IN SCIENCE: REDFISH AGAIN Apart from the promising results of the use of parasites as biological tags (see MacKenzie and Abaunza, this volume, Chapter 11), otolith shape analysis is a particularly innovative approach. This approach generally falls into the category of morphometric measurements that are commonly used to investigate phenotypic differences between species and stocks, for example, redﬁsh (Power and Ni, 1985), Atheriniids (Creech, 1992), and horse mackerel (Murta, 2000). In addition to body morphometrics and meristic features, otolith shape analysis has become a popular tool for purposes of species and stock identiﬁcation. In numerous studies, otolith shapes were shown to be species-speciﬁc as well as population- or stockspeciﬁc (Messieh, 1972; McKern et al., 1974; Neilson et al., 1985). In many cases, geographic variations in otolith shapes could be related to stock differences (Bird et al., 1986; Castonguay et al., 1991; Campana and Casselmann, 1983; Friedland and Redding, 1994; Begg and Brown, 2000; Turan, 2000, Stransky, 2002). The outcome of a sophisticated analysis can look like the scatter in Figures 30-6 and 30-7. These ﬁgures represent data of North Atlantic redﬁsh species, and it is obvious that there are differences on the species level. By contrast, it is also obvious that stock separation for managerial purposes can not be expected from this method, which will be disappointing. However, in contrast to the redﬁsh species, the analyses for horse mackerel are on a stock level and are not yet ﬁnalized (Stransky, 2004). It can be expected that differentiation on a stock level demands more data, although each species may bear its own surprises. No matter what the outcome, it must be kept in mind that analyses by means of

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice

Otolith breadth (mm)

8

653

a

7

6

East Greenland Irminger Sea Barents Sea

5

Otolith weight (mg)

9

10

220 200 180 160 140 120 100 80 60 40 20 0

11 Otolith length (mm)

12

13

b

East Greenland Irminger Sea Barents Sea 9

10

11

12

13

Otolith length (mm)

FIGURE 30-6. Otolith length against otolith breadth (a) and otolith length against otolith weight (b) for S. mentella otoliths, collected in three areas in the North Atlantic (Stranky, 2001).

shape, even though sophisticated, are relatively inexpensive once the otoliths and basic biological parameters have been collected for assessment purposes. Once the measuring routine is established, shape analysis may one day become a kind of by-product of routine data collection.

IX. SUMMARY It has been shown that there is a marked difference in the perception and application of the concept of “stock” between ﬁshery biologists and ﬁsheries managers. In other words, both do not necessarily speak the same language. For a particular stock, the TAC advice given by ﬁshery biologists refers to biological entities,

654

Cornelius Hammer and Christopher Zimmermann 3

S. mentella S. viviparus

2

Dimension 2

S. marinus 1 0 -1 -2 -3

-3

-2

-1

0 1 Dimension 1

2

3

FIGURE 30-7. Multidimensional scaling (MDS) plot of the Euclidean distances between otolith shape (Fourier) descriptors for S. mentella, S. viviparus and S. marinus otoliths, collected in the Barents Sea (Stransky, 2001).

while the ofﬁcial TAC in most cases refers to a management unit, and both do not necessarily match. This occasional mismatch has existed ever since the stock and TAC concepts have been implemented, and it is clear that, in order to avoid more misunderstanding, the managerial units need to be adapted as far as possible to biological realities. However, even if this were the case, such a management regimen is not able to cope with the fact that in a number of cases stocks migrate and mix with stocks from other management areas and are caught together. One has to acknowledge that the whole concept of stock separation and the idea of science supporting sensible management by deﬁning biological entities has basic limitations. Such a concept could work only if methods exist to identify members of different stocks in the catches of mixed aggregations. Science is in need not only of better deﬁning stock characteristics and boundaries but also of developing tools for stock separation from the catches. These developments will not be possible without far more basic research and data collection on a routine basis, that is, much greater ﬁnancing, which is where politicians join the game. What is needed? 1. An agreement between ﬁshery management and ﬁshery science as to what exactly both are perceiving as stock. 2. Management areas that match natural stock boundaries. 3. More biological information about stocks in question, which can be broken down into the following major requirements:

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice

655

a. Clear deﬁnitions of thresholds for genetic differences between stocks and a clear deﬁnition of what a genetic difference between stocks is. b. Regular biological sampling for stock-speciﬁc parameters: age, length, weight, maturity, and meristics. c. Research and development of new stock separators: for example, discrimination by shape analysis of otoliths, radiochemistry, and trace elements. d. Tagging experiments for migration analyses. e. In some cases, better utilization of existing information

REFERENCES Aro, E. 1989. A review of ﬁsh migration patterns in the Baltic. Rapp. P.-v. Réun. Cons. int. Explor. Mer 190: 72–96. Aro, E. 2000. The Spatial and Temporal Distribution Patterns of Cod (Gadus morhua callarias L.) in the Baltic Sea and Their Dependence on Environmental Variability—Implications for Fishery Management. Ph.D. Thesis, Finnish Game and Fisheries Research Institute, Helsinki, 2000. Avise, J. C., Reeb, C. A., and Saunders, N. C. 1987. Geographic population structure and species differences in mitochondrial DNA of mouthbrooding marine catﬁshes (Ariidae) and demersal spawning toadﬁshes (Batrachoidae). Evolution 41: 991–1002. Bagge, O. 1983. Migration of transplanted cod. ICES CM 1983/J:16. 12 pp. Bagge, O., Tiews, K., Lamp, F., and Otterlind, G. 1974. German, Swedish and Danish cod tagging experiments in the Baltic 1968–1969. Rapp. P.-v. Réun. Coms. Int. Explor. Mer 166: 22–39. Bagge,O. and Thurow, F., 1993. The Baltic cod stocks, ﬂuctuations and possible causes. ICES Cod and Climate Symposium 14. 20 pp. Begg, G. A. and Brown, R. W. 2000. Stock identiﬁcation of haddock Melanogrammus aegleﬁnus on Georges Bank, based on otolith shape analysis. Trans. Am. Fish. Soc. 129: 935–945. Berner, M. 1967. Results of cod tagging in the western and central Baltic in the period 1962–1965. ICES CM 1967/F:5. 10 pp. Berner, M. 1980a. Results of the cod tagging experiment in the Arkona Sea during 1972 and 1973. Fischerei-Forschung 11: 19–25 Berner, M. 1980b. Calculation and analysis of migration parameters for Baltic cod on the basis of tagging experiments from 1959 to 1975 in the area from Mecklenburg Bay to the Bornholm Sea (ICES SD 22-25). Fischerei-Forschung 18: 31–49. Berner, M. 1981. Dislocation parameters of tagging experiments on cod in the Baltic (subdivision 22-25) from 1959–1975. ICES CM 1981/ J:15. 26 pp. Bird, J. L., Eppler, D. T., and Checkleay, D. M. 1986. Comparisons of herring otoliths using Fourierseries shape analysis. Trans. Am. Fish. Soc. 129: 935–945. Burd, A. C. 1974. The North-East Atlantic herring and the failure of an industry. In F. R. HardenJones (ed.), Sea Fisheries Research. Elek Science, London, UK. 510 pp. Campana, S. E. and Casselmann, J. M. 1993. Stock discrimination using otolith shape analysis. Can. J. Fish. Aquat. Sci. 50: 1062–1083. Carvalho, G. R. and Hauser, L. 1994. Molecular genetics and the stock concept in ﬁsheries. Rev. Fish Biol. Fisheries 4: 326–350. Castonguay, M. P., Simrad, P., and Gagnon, P. 1991. Usefulness of Fourier analysis of otolith shape for Atlantic mackerel (Scomber scombrus) stock discrimination. Can. J. Fish. Aquat. Sci. 48: 296–302.

656

Cornelius Hammer and Christopher Zimmermann

Creech, S. 1992. A multivariate morphometric investigation of Atherina boyeri Risso, 1810, and A. presbyter Cuvier, 1829 (Teleostei, Atherinidae): morphometric evidence in support of the two species. J. Fish Biol. 41: 341–353. Duschenko, V. 1997. Polymorphism of NADP-dependent malate-dehydrogenase in Sebastes mentella (Scorpaenidae) from the Irminger Sea. J. Ichthyol. 27: 121–131. Friedland, K. D. and Reddin, D. G. 1994. Use of otolith morphology in stock discrimination of Atlantic salmon (Salmo salar). Can. J. Fish. Aquat. Sci, 51: 91–98. Gauldie, W. S. 1988. Tagging and genetically isolated stocks of ﬁsh: a test of one stock hypothesis and the development of another. J. Appl. Ichthyol. 4: 168–173. Gauldie, W. S. 1991. Taking stock of genetic concepts in ﬁsheries management. Can. J. Fish. Aquat. Sci. 48: 722–731. Gold, J. R., Richardson, L. R., Furman, C., and Sun, F. 1994. Mitochondrial DNA diversity and population structure in marine ﬁsh species from the Gulf of Mexico. Can. J. Fish. Aquat. Sci. 51: 205–214. Gordon, J. D. M. (ed.). 1999. Developing Deep-Water Fisheries: Data for the Assessment of Their Interaction with and Impact on a Fragile Environment. Final Consolidated Report of European Commission FAIR Contract 95 0655. 1090 pp. (available on www.sams.ac.uk) Gordon, J. D. M. 2001. Deep-water ﬁsheries at the Atlantic Frontier. Continental Shelf Research 21: 987–1003. Gulland, J. A. 1983. Fish Stock Assessment: A Manual of Basic Methods. Wiley, Chichester, UK. 223 pp. Harden-Jones, F. R. 1968. Fish Migration. Edward Arnold, London, UK. 325 pp. Hilborn, R. and Walters, C. J. 1992. Quantitative Fisheries Stock Assessment. Kluwer, Boston. 592 pp. Hinrichsen, H.-H., Böttcher, U., Oeberst, R., Voss, R., and Lehmann, A. 1999. Drift patterns of cod early life stages in the Baltic: exchange between the western and eastern stock, a physical modelling approach. ICES Counc. Meet. Pap./Y04. 23 pp. Holt, S. J. 1959. Working Group Report II. Population Terminology. Proceedings of the World Scientiﬁc Meeting on the Biology of Sardines and Related Species. FAO, 1: 30. ICES. 1965. The North Sea Herring Report of the North Sea Working Group to the Herring Committee of ICES. Coop. Res. Rep. Int. Coun. Explor. Sea, No. 37. 57 pp. ICES. 1992. Report of the study group on stock identity of mackerel and horse mackerel. ICES C.M. 1992/H:4. ICES. 1994. Report of the Study Group on the Biology and Assessment of Deep-Sea Fisheries Resources. ICES C.M 1995/Assess: 4. 91 pp. ICES. 1995. Report of the Herring Assessment Working Group South of 62°N. ICES CM/Assess: 13. 303 pp. ICES. 1996. Report of the Study Group on the Biology and Assessment of Deep-Sea Fisheries Resources. ICES CM 1996/Assess: 8. 145 pp. ICES. 1998. Report of the Study Group on the Biology and Assessment of Deep-Sea Fisheries Resources. ICES CM 1998/ACFM: 12.172 pp. ICES. 2000a. Report of the Study Group on the Biology and Assessment of Deep-Sea Fisheries Resources. ICES CM 2000/ACFM: 8. 206 pp. ICES. 2000b. Report of the Working Group on the Application of Genetics in Fisheries and Mariculture. ICES CM 2000/F: 03. 53 pp. ICES. 2000c. Report of the ICES Advisory Committee on Fishery Management, 1999. ICES Coop. Res. Rep. 236. 405 pp. ICES. 2001a. Report of the ICES Advisory Committee on Fisheries Management 2000, ICES Cooperative Research Report, No. 242. 911 pp. ICES. 2001b. Report of the ICES Advisory Committee on Fishery Management, 2001. ICES Coop. Res. Rep. 246. 895 pp.

The Role of Stock Identiﬁcation in Formulating Fishery Management Advice

657

ICES. 2002a. Report of the Baltic Fisheries Assessment Working Group. ICES CM 2002/ACFM: 17. 543 pp. ICES. 2002b. Report of the Arctic Fisheries Working Group, 16–25 April 2002, ICES CM 2002/ACFM: 18. 317 pp. Ihssen, P. E., Booke, H. E., Casselman, J. M., McGlade, J. M., Payne, N. R., and Utter, F. M. 1981. Stock identiﬁcation: materials and methods. Can. J. Fish. Aquat. Sci. 38: 1838–1855. Jamieson, A. 1974. Genetic “tags” for marine ﬁsh stocks. In F. R. Harden-Jones (eds.), Sea Fisheries Research. pp. 91–99, Elek Science, London, UK, pp. 91–99. Jamieson, A. and Otterlind, G. 1969. The use of cod blood protein polymorphisms in the Belt Sea, the Sound and the Baltic Sea. ICES CM 1969, special meeting on the biological and serological identiﬁcation of ﬁsh stocks, no. 40. Johansen, T., Danìelsdòttir, A.K., Meland, K., and Nævdal, G. 2000. Studies of the genetic relationship between deep-sea and oceanic Sebastes mentella in the Irminger Sea. Fish. Res. 49: 179–192. Johansen, T., Daníelsdóttir, A. K., and Nævdal, G. 2002. Genetic variation of Sebastes viviparus Krøyer in the North Atlantic. J. Appl. Ichthyol. 18: 177–180. Kändler, R. 1944. Untersuchungen über den Ostseedorsch während der Forschungsfahrten mit dem R.F.D. “Poseidon” in den Jahren 1925–1938. Ber. Dt. Wiss. Komm. 11: 137–245. Koslow, J. A., Boehlert, G., Gordon, J. D. M., Haedrich, R. L., Lorance, P., and Parin, N. 2000. Continental slope and deep-sea ﬁsheries: implications for a fragile ecosystem. ICES J. Mar. Sci. 57: 548–557. Kumpf, H. E., Vaught, R. N., Grimes, C. B., Johnson, A. G. and Nakamura, E. L. 1987. Proceedings of the Stock Identiﬁcation Workshop. NOAA Tech. Memo. NMFS-SEFC, 199. Seattle: US Department of Commerce. 228 pp. Large, P. A., Hammer, C., Bergstad, O. A., Gordon, J. D. M., and Lorance, P. 2001. Options for the Assessment and Management of Deep-Water Species in the ICES Area. NAFO SCR Doc., 01/93. 16 pp. Magnússon, J. V., Berstad, O. A., Hareide N. R., Magnússon, J., and Reinert, J. 1997. Ling, blue ling and tusk of the northeast Atlantic. Tema Nord 535. 61 pp. McKern, J. L., Horton, H. F., and Koski, K. V. 1974. Development of steelhead trout (Salmo gairdneri) otoliths and their use for age analysis and for separating summer from winter races and wild from hatchery stocks. J. Fish. Res. Board. Can. 31: 1420–1426. Menez, G. A., Rogers, A., Krug, H., Mendonca, A., Stockley, B., Isidro, E., Pinho M. R., and Fernandes, A. 2001. Seasonal changes in biological and ecological traits of demersal and deep-water ﬁsh species in the Azores. Final report, European Commission DGXIV/C/1 Study Contract 97/081. 162 pp. and Appendices. Merrett, N. R. and Haedrich, R. L. 1997. Deep-Sea Demersal Fish and Fisheries. Chapman & Hall, London, UK. 282 pp. Messieh, S. N. 1972. Use of otoliths in identifying herring stock in the Southern Gulf of St. Lawrence and adjacent waters. J. Fish. Res. Board Can. 29: 1113–1118. Murta, A. G. 2000. Morphological variation of horse mackerel (Trachurus trachurus) in the Iberian and North African Atlantic: implications for stock identiﬁcation. ICES J. Mar. Sci. 57: 1240–1248. Neilson, J. D., Geen, G. H., and Chan, B. 1985. Variability in dimension of salmonids otolith nuclei: implications for stock identiﬁcation and microstructure interpretation. Fish. Bull. 83: 81–89. Nedreaas, K. and Nævdal, G. 1989. Studies of North-East Atlantic species of redﬁsh (genus Sebastes) by protein polymorphism. J. Cons. Int. Explor. Mer 46: 76–93. Nedreaas, K. and Nævdal, G. 1991. Identiﬁcation of 0-group and I-group redﬁsh (genus Sebastes) using electrophoresis. ICES J. Mar. Sci. 48: 91–99. Nedreaas, K., Johansen, T., and Nævdal, G. 1994. Genetic studies of redﬁsh (Sebastes sp.) from Iceland and Greenland waters. ICES J. Mar. Sci. 51: 461–467. Oeberst, R. 2001. The importance of the Belt Sea cod for the eastern Baltic cod stock. Arch. Fish. Mar. Res. 49: 83–102.

658

Cornelius Hammer and Christopher Zimmermann

Otterlind, G. 1962. Zoogeographical aspects of the southern Baltic. ICES CM 1962, 103. Otterlind, G. 1965. Transportation experiments with cod in the Baltic Belt Seas, a preliminary report. ICES CM/ Gadoid Fish Committee, 163. 6 pp. Otterlind, G. 1966. Problems concerning cod in the Baltic. ICES CM 1966/D:18. 15 pp. Otterlind, G. 1985. Cod migration and transplantation experiments in the Baltic. Z. Angew. Ichthyol. 1: 3–16. Power, D. J. and Ni, I.-H. 1985. Morphometric differences between golden redﬁsh (Sebastes marinus) and beaked redﬁshes (S. mentella and S. fasciatus). J. Northw. Atl. Fish. Sci. 6: 1–7. Reinert, J. and Lastein, L. 1992. Stock identiﬁcation of S. marinus L. and S. mentella, Travin in the Northeast Atlantic, based on meristic counts and morphometric measurements. ICES CM 1992/G: 29. Roques, S., Sévigny, J.-M., and Bernatchez, L., 2002. Genetic structure of deep-water redﬁsh, Sebastes mentella, populations across the North Atlantic. Mar. Biol. 140: 297–307. Ruzzante, D. E., Taggart, C. T., and Cook, D. 1996. Spatial and temporal variation in the genetic composition of the larval cod (Gadus morhua) aggregation: cohort contribution and genetic stability. Can. J. Fish. Aquat. Sci. 53: 2695–2705. Ruzzante, D. E., Taggart. C. T., Cook, D., and Goddard, S. V. 1998. Genetic differentiation between inshore and offshore Atlantic cod (Gadus morhua) off Newfoundland: a test and evidence of temporal stability. Can. J. Fish. Aquat. Sci. 54: 2700–2708. Sick, K. 1961. Haemoglobin polymorphism in whiting and cod. ICES CM 1961, 128. Smith, P. J., Jamieson, A., and Birley, A. J. 1990. Electrophoretic studies and stock concepts in marine teleosts. J. Cons. Int. Explor. Mer 47: 231–245. Stransky, C. 2001. Preliminary results of a shape analysis of redﬁsh otoliths: comparison of areas and species. NAFO SCR Doc. 01/14, Ser. No. 4382. 10 pp. Stransky, C. 2002. Otolith shape analysis of Irminger Sea redﬁsh (Sebastes mentella): preliminary results. NAFO SCR Doc. 02/35, Ser. No. 4646. 9 pp. Stransky, C. 2004. Stock separation and growth of redﬁsh (genus Sebastes) in the North Atlantic by means of shape and elemental analysis of otoliths. Ph.D. dissertation, University of Hamburg 2004, 132 pp. (available under http://www.sub.uni-hamburg.de/opus/volltexte/2004/2173/ Thurow, F. 1974a. Changes in population parameters of cod in the Baltic. Rapp. P.-v. Réun. Cons. Int. Explor. Mer 166: 85–93. Thurow, F. 1974b. Fischerei. In L. Magaard and G. Reinheimer (eds.), Meereskunde der Ostsee. Springer-Verlag, Berlin, Heidelberg, New York. 269 pp. Turan, C. 2000. Otolith shape and meristic analysis of herring (Clupea harengus) in the North East Atlantic. Arch. Fish. Mar. Res. 48: 213–225. Waples, R. S., 1991. Paciﬁc salmon, Oncorhynchus spp., and a deﬁnition of “species” under the Endangered Species Act. Mar. Fish. Rev. 53: 11–22. Waples, R. S. 1998. Separating the wheat from the chaff: patterns of genetic differentiation in high gene ﬂow species. J. Heredity 89: 438–450. Ward, R. D., Woodward, M., and Skibinski, D. O. F. 1994. A comparison of genetic diversity levels in marine, freshwater, and anadromous ﬁshes, J. Fish Biol. 44: 213–232. Zijlstra, J. J. 1969. On the “racial” structure of the North Sea autumn-spawning herring. J. Cons. Int. Explor. Mer 33: 67–80.

CHAPTER

31

Identifying Fish Farm Escapees PEDER FISKE,* ROAR A. LUND,* AND LARS P. HANSEN† *Norwegian Institute for Nature Research, Trondheim, Norway, †Norwegian Institute for Nature Research, Oslo, Norway

I. Introduction II. Protocol for Monitoring Methods A. Morphology and Morphometry B. Scale and Otolith Pattern Recognition C. Biochemical and Physiological Markers D. Genetic Markers E. Large-Scale Group Marking in Farms III. Perspectives in Aquatic Practices References

I. INTRODUCTION The production of farmed Atlantic salmon (Salmo salar) and rainbow trout (Oncorhynchus mykiss) has increased considerably over the past 20 years, and Atlantic salmon in net pens in the sea now by far outnumber their wild conspeciﬁcs (Gross, 1998). The salmonid species are produced in captivity both within and outside their natural distributions. Some of the farmed salmonids escape into the wild, and this may lead to both mixing with local stocks of the same species and to the spread of salmonid species into new areas, with potentially harmful effects on other species. To assess the problem with potential mixing with local stocks, it is important to be able to identify the extent of farmed escapees in the stocks. Furthermore, if the number of escapees is high, they may bias estimates of wild populations if not accounted for. Today, other species are domesticated for use in aquaculture industry (halibut, cod, turbot, etc.). If net pens stationed in the sea become the dominant method for production of these new species, we expect that escapees of those species also may interact with wild populations of the same species and potentially with other species. There is Stock Identiﬁcation Methods Copyright © 2005 by Elsevier. All rights of reproduction in any form reserved.

659

660

Peder Fiske et al.

therefore a need to develop methods for the identiﬁcation of escapees in these new aquaculture species as well. With the acceptance of the stock concept in ﬁsh, there was a need to enhance and further develop stock identiﬁcation methods. At present, a number of those are in current use and have proved useful tools in stock assessment and ﬁsheries management (e.g., Begg et al., 1999). Similarly, with increasing numbers of escaped farmed Atlantic salmon in ﬁsheries and spawning populations in some areas, it soon became desirable to distingush between wild and escaped farmed salmon. This process started in the mid-1980s, and for a simple and quick identiﬁcation a combination of external morphology and scale pattern is used routinely to survey the amount of farmed escapees in catches of salmon (Fiske et al., 2001; Hansen et al., 1993; Hansen et al., 1999; Lund et al., 1991). There has also been developed other techniques that can be used, but most of them are very labor-consuming or expensive. In this chapter, we describe the methods that are available for use, with main focus on those in current use. We also discuss the potential for progress. Even though we focus on salmonids (especially Atlantic salmon), most of the methods can probably be applied for other species after they have been “calibrated” to the target species.

II. PROTOCOL FOR MONITORING METHODS

A. MORPHOLOGY

AND

MORPHOMETRY

To identify farmed Atlantic salmon from wild salmon on a general basis, Lund et al. (1989) examined morphology of adult salmon from six ﬁsh farms and wild salmon caught in mixed-stock sea ﬁsheries and in riverine angling ﬁsheries. The wild salmon was identiﬁed from scale characteristics (see later). The frequent defects seen on the ﬁns of farmed salmonids are supposed to be caused primarily by aggressive behavior and biting during periods of insufﬁcient food rations. Such defects primarily appear at the pectoral, dorsal, and tail ﬁns and readily make the ﬁns shortened. 1. Defects of Fin Tissue The epidermal tissue on undamaged ﬁns runs to the tip of the ﬁn rays on all ﬁns of the salmon. The ﬁn tissue of the reared salmon is frequently fringed and appears as thickened after healing of wounds. This appearance is rare on ﬁns of wild salmon (Fig. 31-1, Fig. 31-3), although ﬁn damage frequently may be seen on wild salmon after being in contact with net gears or after handling during or after catching. Accordingly, ﬁn damage on wild ﬁsh usually appears as fresh wounds (Fig. 31-1).

Identifying Fish Farm Escapees

661

FIGURE 31-1. Dorsal ﬁn of a wild salmon with normal and evenly arched or straight ﬁn rays. The ﬁn has a tear in the epidermal tissue after handling.

2. Fin Ray Defects In reared ﬁsh, damage on ﬁn tissue frequently causes the ﬁn rays to grow together and lose evenly arched or straight rays as seen on wild ﬁsh. Thus, healed ﬁns often appear with ﬁn rays fused at the outer parts of the ﬁns (Fig. 31-2). They lose their normal shape and frequently appear wavy (Fig. 31-4). In worst cases, ﬁn rays may be torn to a humplike cartilage where the rays are completely eviscerated (Fig. 31-7). This state appears primarily at the dorsal and pectoral ﬁns. Minor defects on ﬁns of farmed ﬁsh can easily be missed without careful inspection. In particular, minor defects on pectoral ﬁns are easily missed unless the ﬁns are inspected by touch (thumb and foreﬁnger). Thus, ﬁn ray defects may be recognized as a breakage or a wavy shape on the two outermost ﬁn rays. The tail ﬁn of wild salmon (Fig. 31-5) normally forms marked tips where the two outermost ﬁn rays are the longest. On farmed ﬁsh, these tips are often strikingly worn and rounded (Fig. 31-6). However, in reared ﬁsh escaped at an early sea stage, tail ﬁn erosion may be regenerated and may look like the tail of wild salmon. Such appearance may also occur at the pectoral ﬁns of reared ﬁsh.

FIGURE 31-2. Damaged dorsal ﬁn of a reared salmon with wavy ﬁn rays and lack of epidermal tissue.

FIGURE 31-3. Pectoral ﬁn of a wild salmon with normal, evenly arched ﬁn rays.

662

FIGURE 31-4. A defective pectoral ﬁn ray of a reared salmon with wavy rays.

FIGURE 31-5. A normal tail ﬁn of a wild salmon. The ﬁn has pronounced pointed tips.

663

FIGURE 31-6. A damaged tail ﬁn of a reared salmon where the tail tips are worn and rounded.

FIGURE 31-7. A damaged dorsal ﬁn of a reared salmon where the ﬁn rays and the epidermal tissue are eroded and have fused into a raised lump (bud ﬁn).

664

Identifying Fish Farm Escapees

665

FIGURE 31-8. A shortened gill cover where the gills of the ﬁsh are visible.

3. Gill Cover Shortening Gill cover shortening (Fig. 31-8) may on occasion be comprehensive in farmed populations but is rarely seen on wild salmon. This defect is assumed to be a result of the environment, that is, high temperatures during hatching or the fry stage. Gill cover shortening is assumed to be nonregenerative. To be accepted as an identiﬁcation criterion of farmed ﬁsh, the gill cover should be eroded to the extent of visibility of the gills when the gill cover is naturally closed. 4. Undershot Jaw Breeding defects may occasionally be expressed as snout and jaw deformations, mostly seen as an undershot jaw. Salmon normally have an overshot jaw when the mouth is closed. This defect is very rare in wild salmon. 5. Heavy Pigmentation The pattern of external pigmentation of the salmon is assumed to be heritable. However, environmental conditions in rearing and food quality probably affect external pigmentation considerably. In different Norwegian wild salmon populations, counts of pigments in the area below the lateral line from the front edge

666

TABLE 31-1.

Occurence (%) of defects on the ﬁns of different groups of almon Dorsal ﬁn

Salmion type

N

Epidermal tissue damage

Adult salmon from farms Recaptures of hatchery smolts Wild salmon from the river lmsa Wild salmon from drift net ﬁsheries Wild salmon from bagnets and rod ﬁsheries Escaped farmd salmon from drift net ﬁsheries Presmolt from the lms hatchery

172 521 40 85 1927 32 50

100 – – 79 – 100 –

Pectoral ﬁn

Fused ﬁn rays

Wavy ﬁn rays

Bud ﬁn

Wavy ﬁn rays or bud ﬁn

Epidemal tissue damage

Fused ﬁn rays

Wavy ﬁn rays

Bud ﬁn

Wavy ﬁn rays or bud ﬁn

54 61 23 4 – 56 70

63 68 17 2 0.7 75 92

28 5 0 0 0.2 9 2

91 73 17 2 0.9 84 94

100 – – 82 – 88 –

19 4 0 2 – 6 20

90 15 8 2 0.2 31 90

4 0 0 0 0 6 0

94 15 8 2 0.2 37 90

-: not investigated. The table is similar to table 1 in Lund et al. (1989).

Peder Fiske et al.

Identifying Fish Farm Escapees

667

of the dorsal ﬁn and forward to the back edge of the gill cover showed an average variation on fresh-run wild ﬁsh between three and six spots. The numbers counted at the same area on farmed salmon (nonripened) were signiﬁcantly higher (average 17 spots). Deformed ﬁns were much more frequent among farmed salmon than among wild salmon (Table 31-1). Furthermore, shortened gill covers and snout/jaw defects also occurred in higher frequencies among salmon from ﬁsh farms than among wild salmon, and farmed salmon had a higher number of pigment spots in the foremost part of the body than wild salmon. As a general rule, Lund et al. (1989) recommended that a ﬁsh showing two or more of the characteristics discussed (moderate ﬁn deformities, each ﬁn counted as one character, more than 15 pigment spots below the sideline in the foremost part of the body) could be identiﬁed as a farmed ﬁsh. The occurrence of one single ﬁn showing severe deformity in the form of total erosion, or a humplike eroded ﬁn (bud ﬁn), or the occurrence of a shortened gill cover was enough to classify the ﬁsh as a farmed salmon. This study was performed with farmed salmon sampled in the 1980s. Farming practice has changed markedly since then, and the occurrence of ﬁn deformities in farmed salmon may have decreased as a consequence of more reﬁned rearing techniques. Thus, morphological characteristics may be less suitable for identifying farmed salmon at present. 6. Fin Measurements Fins of farmed salmon are often eroded. To distinguish between wild and farmed salmon, Lund et al. (1989) developed several discrimination models based on measurements of ﬁns (the dorsal ﬁn, both pectoral ﬁns, and the tail ﬁn lobes) relative to ﬁsh length. Measurements were carried out on three groups of adult salmon of known origin: wild, ranched, and farmed. The results were variable and adult returns of ranched ﬁsh caused the main problems, perhaps because these ﬁsh tended to regenerate ﬁns during their marine journey. This was not so for farmed ﬁsh that had recently escaped from farms. When morphological characters such as distorted ﬁn rays (especially the dorsal ﬁn) and shortened gill covers were added to the model, the discriminant power increased. Different dicriminant models, that is, combinations of types of ﬁns and groups of ﬁsh, tested on independent groups of salmon of known origin classiﬁed farmed salmon with high precision (usually up to 100%). The best models for classifying ranched salmon reached scores ranging from 72% to 83%. However, there was a tendency between models for improved classiﬁcation success on one group of salmon reducing classiﬁcation success on other groups. Available material did not permit the models to be tested on independent samples of wild salmon. However, withinmodel classiﬁcation of wild salmon was moderately high (80%—83%). In terms of ﬁn types, the dorsal ﬁn had the most discriminative power between the groups

668

Peder Fiske et al.

FIGURE 31-9. Location of 23 morphometric landmarks (points): 21 trusses (dashed lines) and 2 ﬁn areas (dotted lines). Landmarks are described by Winans (1984), except dorsal point at least depth of caudal peduncle (12), origin of pectoral ﬁn (13), distal tips of pectoral (14), dorsal (16), pelvic (17), and anal ﬁns (18), point on ventral surface of body directly below anterior dorsal ﬁn origin (15), and ventral point on lower jaw directly below kype tit (23), both based on a line drawn perpendicular to the horizontal axis of the ﬁsh, anteriormost edge of orbit (19), posterior tip of maxallary (20), and dorsal-most (21) and anterior-most points of kype (22). Landmarks 20–23 were not measured on parr. The body was divided into four regions (head, anterior trunk, posterior trunk, and tail) to localize morphological features. The ﬁgure is redrawn from Fleming et al. (1994).

of salmon, because this ﬁn often is the most eroded on both farmed and ranched salmon. Both the proportion of a ﬁsh’s life history and number of generations spent in culture may inﬂuence a farmed ﬁsh’s morphological divergence from wild ﬁsh (Fleming et al., 1994). Cultured salmon generally have smaller heads and ﬁns and different body shapes than wild salmon (Fleming et al., 1994). A thorough morphological study on ripened ﬁsh detected differences among Atlantic salmon with wild, sea-ranched, and farmed backgrounds (Fleming et al., 1994). These groups all had wild parents, and morphological differences could thus be related to the captive environment. Fleming and co-workers used a suite of morphological measurements (Fig. 31-9) based on distances among “landmarks” described by Winans (1984). Both principal component analysis and discriminant analysis separated the groups well. Furthermore, salmon from a ﬁfth-generation cultured population were different from all other groups in the study, but most similar to those experiencing captive conditions their whole life. Although this study demonstrated clear morphological effects of domestication, it only compared the farmed ﬁsh with one wild population. Thus, to develop a more general method one has to take variation among wild populations into account, and study ﬁsh from several ﬁsh farms. Image-processing techniques have a potential for stock identiﬁcation based on morphological, scale, and otolith characteristics, although the methods have not been commonly applied in ﬁshery science (Cadrin and Friedland, 1999).

Identifying Fish Farm Escapees

B. SCALE

AND

669

OTOLITH PATTERN RECOGNITION

Scales and otoliths are often used in age and growth studies of salmonids because they reﬂect growth at different life stages (Dahl, 1910; Jonsson, 1976; Jonsson and Stenseth, 1977; Tesch, 1968). Growth patterns differ among salmon stocks, and detailed scale patterns have been used to distinguish between different groups (different geographic areas or cultivated vs. wild origin) of salmon (Hiilivirta et al., 1998; Reddin and Friedland, 1999). In hatcheries, ﬁsh grow faster and frequently with different patterns than in the wild. This phenomenon has been utilized to distinguish between wild and reared individuals. Several studies have been carried out separating hatcheryreared ﬁsh from wild ﬁsh. Using scale circuli spacing and scale shape, Ross and Pickard (1990) discriminated wild and hatchery-origin striped bass (Morone saxatilis) with 87% to 91% accuracy. Furthermore, using circuli spacing in scales of wild barramundi, Lates calcarifer (Bloch), Barlow and Gregg (1991) were able to discriminate between hatchery and wild ﬁsh with up to 83% accuracy. Whereas hatchery-reared ﬁsh are often released early in their life cycle and ranched salmon are released at the smolt stage, farmed salmon may escape at all life stages. Friedland et al. (1994) proposed a method to distingish between farmed, ranched, and wild-origin Atlantic salmon using circuli spacing and scale texture. When using three-group quadratic discriminant function models, they were able to classify the three groups with an efﬁciency of 74%. When the models were simpliﬁed to two groups, farmed and wild, the efﬁciency increased to 90%. Lund and Hansen (1991) compared scales of adult farmed salmon from six ﬁsh farms with salmon caught in ﬁsheries before the onset of ﬁsh farming, and with sea-ranched ﬁsh. They arrived at six criteria that identiﬁed farmed ﬁsh from wild ﬁsh. A total of 97% of farmed adults in the study showed at least one of these criteria, while the same ﬁgure was only 1.6% for scales from wild salmon: 1. Smolt Size Farmed ﬁsh were larger at smolting than wild ﬁsh. When the backcalculated length at the smolt stage was larger than 95% of the observations of wild smolts from the same area, the ﬁsh were characterized as farmed. 2. Smolt Age Smolt age was difﬁcult to determine in ﬁsh from ﬁsh farms and in sea-ranched ﬁsh. Smolt age was generally overestimated, and this was especially pronounced for ﬁsh with known smolt age of 1. Overall, only 27% of farmed ﬁsh and 23% of sea-ranged ﬁsh were aged correctly. Similar ﬁgures for wild ﬁsh were unavailable because their real age was not known. Smolt age outside the range

670

Peder Fiske et al.

of 95% of wild ﬁsh in the area was characteristic of farmed ﬁsh. Given that smolts today are produced in 0.5 years, this criterion probably is even more useful than when this study was performed. Naturally produced Atlantic salmon from northern areas usually have smolt ages higher than 2 years, and ﬁsh of farmed origin could thus be distinguished relatively easily. Smolts from more southern regions, however, often have smolt ages of 1 or 2 years, and farm escapees could thus be more difﬁcult to distinguish from wild ﬁsh using smolt age as a criterion. 3. Transition from Freshwater to Saltwater The transition zone from freshwater to saltwater was sharp and easily deﬁned on the scales of wild ﬁsh (Fig. 31-10), while this zone was often large and very diffuse on the scales of farmed and ranched salmon (Fig. 31-11). Diffuse transition between freshwater and saltwater was characteristic of farmed ﬁsh. 4. Sea Winter Band Sea age was correctly classiﬁed in only 53% of farmed ﬁsh, while 99% of wild ﬁsh and 92% of sea-ranched ﬁsh of known age were correctly classiﬁed. Most of the incorrectly aged ﬁsh were overaged by 1 year because summer growth checks were incorrectly interpreted as winter bands. Furthermore, farmed ﬁsh showed major deviation from the normal growth pattern. Irregular locations of sea winter bands, for example, backcalculated lengths at ﬁrst sea annulus <35cm or backcalculated lengths at second sea annulus <1.55 times the backcalculated length at the ﬁrst sea annulus, were characteristic of farmed ﬁsh. 5. Summer Checks The occurrence of summer growth checks in the ﬁrst and second sea annuli was higher in the scales of farmed (67%) and sea-ranched (40%) salmon compared to wild salmon (18%). In all groups, one check was most common among ﬁsh with checks, but 26% of the scale samples from farmed salmon had more than one check. More than one summer growth check was thus characteristic of farmed ﬁsh. 6. Replacement Scales Handling of ﬁsh leads to loss of scales. Frequency of scale loss during the freshwater stage did not differ among groups of ﬁsh. At the sea stage, farmed and sea-ranched ﬁsh showed higher average levels of replacement scales than wild salmon. Changes in aquaculture practices in recent years have led to less han-

Identifying Fish Farm Escapees

671

FIGURE 31-10. A scale from a wild salmon caught in a marine ﬁshery in mid-Norway in July, 1970. The scale is aged 3.1+ and has a clear transition between the freshwater and saltwater zones and distinct annuli in both zones. F, scale focus, R, freshwater zones, S, sea zone. Figure from Lund and Hansen (1991).

dling of the ﬁsh, and probably to less loss of scales. In present screenings of scales from salmon ﬁsheries, the frequency of replacement scales is less used than previously. In the original study, more than 15% replacement scales at the sea stage in a sample was characteristic of farmed ﬁsh.

672

Peder Fiske et al.

FIGURE 31-11. A scale from a farmed salmon with known age 1.1+. The scale has a diffuse transition between the freshwater and seawater zones and the age of the ﬁsh is read to 3.1+. F, scale focus, R, freshwater zones, S, sea zones, RO, runout. Figure from Lund and Hansen (1991).

At present, we assign scale samples to either wild, uncertain origin, or farmed based on the above criteria. In the Norwegian monitoring program for farmed salmon in ﬁsheries and stocks, ﬁshermen are asked to examine the ﬁsh for morphological abnormalities, and these observations are used as additional information when scales are classiﬁed to come from ﬁsh with either a farmed or a wild

Identifying Fish Farm Escapees

673

FIGURE 31-12. Patterns of mean circuli spacing versus circuli pair for farmed, ranched and wild-origin Atlantic salmon. Smooth line is simple adjacent averaging. Figure from Friedland et al. (1994).

background. This method is conservative for identifying the number of farmed escapees in samples from ﬁsheries because specimens that are classiﬁed as of uncertain origin are not counted. However, the method is subjective because classiﬁcation is based on the judgment of personnel reading the scales. 7. Image Processing In the search for more objective methods, Friedland et al. (1994) used image processing to group scales from wild, farmed, and sea-ranched Atlantic salmon. They used many of the same scale samples as were used in the original study (Lund and Hansen, 1991). Scales of wild salmon had wider-spaced circuli than scales from farmed and ranched salmon (Fig. 31-12), and the texture of the scales from farmed and ranched salmon also differed from wild salmon scales. Discriminant analysis classiﬁed close to 90% of the scales correctly when scales from farmed and wild salmon were compared. Image processing thus provides an objective method of classifying scales to farmed or wild origin. However, the method is more resource demanding than manual scale reading and thus is not presently used in the surveillance of scales from salmon catches.

674

Peder Fiske et al.

8. Otoliths Otoliths are used interchangeably with scales for age determination and backcalculation of growth patterns. Otoliths can provide more accurate age determination than scales ( Jonsson, 1976), although they are more difﬁcult to obtain from routine surveys because they are more laborious to collect. A comparison of otoliths from wild and farmed salmon parr detected that naturally produced ﬁsh showed a seasonal pattern in the otoliths, with alternating opaque (summer) zones and hyaline (winter) zones. Hatchery-reared parr, on the other hand, had otoliths that were completely opaque, having indistinct checks only, or having one or more hyaline zones close to the edge, but no zone where the ﬁrst winter zone would be expected (Hindar and L’abée-Lund, 1992). Only 2.8% of the otoliths were erroneously classiﬁed in Hindar and L’abée-Lund’s (1992) study, with mostly hatchery ﬁsh being classiﬁed as wild. Otoliths can thus be used much in the same way as scales to identify farm escapees. Because of the more laborious collection process for otoliths than for scales, we suggest that this method be used for salmon parr caught in rivers and not for screening large samples from ﬁsheries.

C. BIOCHEMICAL

AND

PHYSIOLOGICAL MARKERS

Practically all farmed Atlantic salmon smolts are vaccinated against furunculosis and other diseases with oil-adjuvanted bacterins. Today this vaccination is performed manually by injection into the abdominal cavity. This injection leads to intraabdomial adhesions (Fig. 31-13) that can be detected by inspection of the opened abdominal cavity (Lund et al., 1997). Among adult salmon, adhesions were present in 94% to 100% of the ﬁsh from vaccinated groups and were completely absent in unvaccinated controls. This marker thus provides a useful method for identifying farmed salmon, provided that the vaccination practices are continued in the future. However, new vaccines are continuously being developed, and one of the main goals of new vaccines is to reduce the amount of adhesions. According to the vaccine producers, the vaccines that are used today lead to less severe adhesions; more than one-third of the ﬁsh today have adhesions that are difﬁcult to detect for a layperson. Furthermore, detecting adhesions requires experience, and even though the method provides a fast and easy way of detecting farmed salmon in mixed samples, it also requires that trained personnel are present when the ﬁsh are processed and that all ﬁsh are opened carefully. To obtain the pink color of the ﬂesh, farmed salmon are fed with carotenoids. The detection of synthetic carotenoids was suggested as a method of separating wild and farmed individuals (Craik and Harvey, 1987) and has been applied in

Identifying Fish Farm Escapees

675

FIGURE 13-13. Fibrous adhesions between the internal organs and the abdomial wall postvaccination. Figure from Lund et al. (1997).

Scotland (Webb and Youngson, 1992). Canthaxanthin in salmon-eyed ova or newly hatched fry may be used as an indicator of the prevalence of escaped farm progeny (Poole et al., 2000). However, the occurrence in older fry should be interpreted with care, since canthaxanthin occurs naturally in freshwater diets and was found at high levels in adult brown trout from Irish rivers (Poole et al., 2000). Earlier, synthetic canthaxanthin was used as the main source of pigmentation in the farming industry, but later synthetic astaxanthin has become the most commonly used source, although both carotenoids are used in commercial diets today (Buttle et al., 2001). Astaxanthin also is the main pigment of wild salmon. This carotenoid exists in three optical isomers, and the ratio of the isomers differs between naturally produced and synthetic astaxanthin. The distribution of the different isomers has been used to distinguish farm escapees from wild ﬁsh (Lura and Økland, 1994). Astaxanthin also is transferred to the eggs and can be detected in offspring; the method has thus been used to assess spawning success of escaped farmed salmon (Lura and Sægrov, 1991a,b; Sægrov et al., 1997). However, the distribution of different isomers is dependent on the time of escape, and salmon that escape early in the sea stage can be difﬁcult to separate from wild ﬁsh with the use of this method (Lura and Økland, 1994).

676

Peder Fiske et al.

Identiﬁcation of site of origin based on composition of trace elements in otoliths or other bone structures has been successful in a few striking studies (Kennedy et al., 1997, 2000; Thorrold et al., 2001). Waters in different streams and rivers often have distinct trace elements. Some of these elements deposit in bone structures of growing ﬁsh and thus represent a chemical signature of the environment at the time of structure formation. Atlantic salmon (Salmo salar L.) fry from 8 of 10 tributary streams in the Connecticut River could be identiﬁed based on differences in stable isotopes of strontium (Kennedy et al., 1997, 2000). The same studies also showed that the differences in stable isotopes of strontium in water samples were very similar to samples taken from the otoliths of ﬁsh from the same locations. Geochemical methods have also been used to assess origin and movement in several other organisms (Bearhop et al., 1999; Campana, 1999; Campana et al., 2000; Campana and Thorrold, 2001; Chamberlain et al., 2000; DiBacco and Levin, 2000; Hobson, 1999; Volk et al., 2000). Although this method has not yet been tested on salmon from ﬁsh farms compared to ﬁsh of wild origin, it has the potential of being useful in the future. Because the method is relatively expensive, it is unlikely to replace traditional methods for surveys of large numbers of samples. In stock identiﬁcation, fatty acid composition in heart tissue has proved to be a useful tool in some marine ﬁsh species such as herring (Clupea harengus) (GrahlNielsen and Ulvund, 1990) and striped bass (Grahl-Nielsen and Mjaavatten 1992). Furthermore, Joensen et al. (2000) were able to distinguish between two reared stocks of Atlantic cod (Gadus morhua). This difference was suggested to be of a purely genetic nature, and the method may therefore have potential to discriminate between wild stocks. The potential for this method to discriminate between wild and farmed salmon has not yet been assessed.

D. GENETIC MARKERS The recent development of genetic markers has provided useful methods for stock identiﬁcation. By combining information from highly polymorphic DNA markers and new statistical methods (“assignment test”), one can assign probabilities that different samples originate from different populations (Manel et al., 2002; Nielsen et al., 1997; Paetkau et al., 1995). Such methods require good samples from the possible candidate populations. Since farmed Atlantic salmon consists of relatively few strains (Gross, 1998), we see the potential of using this method for samples taken in rivers that are genetically divergent from farmed strains. In samples from mixed-stock ﬁsheries, however, detailed knowledge of many potential stocks will be necessary for successful identiﬁcation, and the costs of obtaining such knowledge may reduce this method’s use.

677

Identifying Fish Farm Escapees

E. LARGE-SCALE GROUP MARKING

IN

FARMS

The methods mentioned above, with the possible exception of microchemistry of otoliths, can class ﬁsh as either farmed or wild, but they cannot identify the origin of farmed ﬁsh. To do this, one has to apply marks to the ﬁsh. To address this issue, Heggberget et al. (2000) assessed different potential tagging methods for group marking of farmed salmon. They evaluated coded wire tags as the most promising alternative. Coded wire tags have been used for mass marking of Paciﬁc salmon (Oncorhynchus spp.) since the early 1970s (Weitkamp and Neely, 2002), for marking of both hatchery-reared and wild ﬁsh. The method thus has great potential for group marking of large numbers of ﬁsh, but no marking method has yet been implemented routinely for marking farmed salmon. Until the industry decides on a method for marking of smolts put in sea pens, we have limited ability to trace escaped ﬁsh to their site of origin.

III. PERSPECTIVES IN AQUATIC PRACTICES Salmon farming apparently will continue to increase, and even if production is not increased from levels we see today, the potential for escaped farmed salmon interacting with wild conspeciﬁcs will still be upheld. Furthermore, domestication of other marine species most likely will increase. Thus, continuous monitoring of ﬁsheries and stocks for the occurrence of farmed escapees will be needed, and methods for identifying the farmed origin of ﬁsh also will need to be used and developed. Different methods for identiﬁcation of farmed escapees cover different needs, and choice of method should be evaluated against the purpose of the study. If one needs a certain veriﬁcation of the origin of a sample, a combination of methods should be applied. However, for monitoring purposes, easy ﬁeld and laboratory methods are needed, preferably without needing to kill the ﬁsh. At present, we judge the combination of morphological examination and manual scale pattern evaluation to be the most cost-effective method for monitoring escaped farmed salmon.

REFERENCES Barlow, C. G. and Gregg, B. A. 1991. Use of circuli spacing on scales to discriminate hatchery and wild barramundi, Lates calcarifer (Bloch). Aquaculture and Fisheries Management 22: 491–498. Bearhop, S., Thompson, D. R., Waldron, S. et al. 1999. Stable isotopes indicate the extent of freshwater feeding by cormorants Phalacrocorax carbo shot at inland ﬁsheries in England. Journal of Applied Ecology 36: 75–84.

678

Peder Fiske et al.

Begg, G. A., Friedland, K. D., and Pearce, J. B. 1999. Stock identiﬁcation and its role in stock assessment and ﬁsheries management: an overview Fisheries Research 43: 1–8. Buttle, L. G., Crampton, V. O., and Williams, P. D. 2001. The effect of feed pigment type on ﬂesh pigment deposition and colour in farmed Atlantic salmon, Salmo salar L. Aquaculture Research 32: 103–111. Cadrin, S. X. and Friedland, K. D. 1999. The utility of image processing techniques for morphometric analysis and stock identiﬁcation. Fisheries Research 43: 129–139. Campana, S. E. 1999. Chemistry and composition of ﬁsh otoliths: pathways, mechanisms and applications. Marine Ecology Progress Series 188: 263–297. Campana, S. E., Chouinard, G. A., Hanson, J. M., Fréchet, A., and Brattey, J. 2000. Otolith elemental ﬁngerprints as biological tracers of ﬁsh stocks. Fisheries Research 46: 343–357. Campana, S. E. and Thorrold S. R. 2001. Otoliths, increments, and elements: keys to a comprehensive understanding of ﬁsh populations. Canadian Journal of Fisheries and Aquatic Sciences 58: 30–38. Chamberlain, C. P., Bensch, S., Feng, X., Åkesson, S., and Andersson, T. 2000. Stable isotopes examined across a migratory divide in Scandinavian willow warblers (Phylloscopus trochilus trochilus and Phylloscopus trochilus acredula) reﬂect their African winter quarters. Proceedings of the Royal Society of London B 267: 43–48. Craik, J. C. A. and Harvey, S. M. 1987. A biochemical method for distinguishing between wild and farmed salmonid ﬁshes by their carotenoid pigmentation. Journal of Forensic Science Society 27: 47–55. Dahl, K. 1910. Alder og vekst hos laks og ørret belyst ved studiet av ders skjæl. Centraltrykkeriet, Kristiania. DiBacco, C. and Levin, L. A. 2000. Development and application of elemental ﬁngerprinting to track the dispersal of marine invertebrate larvae. Limnology and Oceanography 45: 871–880. Fiske, P., Lund, R. A., Østborg, G. M., and Fløystad, L. 2001. Rømt oppdrettslaks i sjø- og elveﬁsket i årene 1989–2000. NINA Oppdragsmelding 704: 1–26. Fleming, I. A., Jonsson, B., and Gross, M. R. 1994. Phenotypic divergence of sea-ranched, farmed, and wild salmon. Canadian Journal of Fisheries and Aquatic Sciences 51: 2808–2824. Friedland, K. D., Esteves, C., Hansen, L. P., and Lund, R. A. 1994. Discrimination of Norwegian farmed, ranched and wild-origin salmon, Salmo salar L., by image processing. Fisheries Management and Ecology 1: 117–128. Grahl-Nielsen, O. and Mjaavatten, O. 1992. Discrimination of striped bass stocks: a new method based on chemometry of fatty acid proﬁle in the heart tissue. Transactions of the American Fisheries Society 121: 307–314. Grahl-Nielsen, O. and Ulvund, K. A. 1990. Distinguishing populations of herring by chemometry of fatty acids. American Fisheries Society Symposium 7: 566–571. Gross, M. R. 1998. One species with two biologies: Atlantic salmon (Salmo salar) in the wild and in aquaculture. Canadian Journal of Fisheries and Aquatic Sciences 55(suppl. 1): 131–144. Hansen, L. P., Jacobsen, J. A., and Lund, R. A. 1993. High numbers of farmed Atlantic salmon, Salmo salar L., observed in oceanic waters north of the Faroe Islands. Aquaculture and Fisheries Management 24: 777–781. Hansen, L. P., Jacobsen, J. A., and Lund, R. A. 1999. The incidence of escaped farmed Atlantic salmon, Salmo salar L., in the Faroese ﬁshery and estimates of catches of wild salmon. ICES Journal of Marine Science 56: 200–206. Heggberget, T. G., Skilbrei, O., Thorstad, E. B., Moen, V., and Johnsen, B. O. 2000. Tagging of hatchery produced salmon in Norway–-potential methods, costs and effects. NINA Oppdragsmelding 640: 1–25 (in Norwegian with English summary). Hiilivirta, P., Ilkonen, E., and Lappalainen, J. 1998. Comparison of two methods for distinguishing wild from hatchery-reared salmon (Salmo salar Linnaeus, 1758) in the Baltic Sea. ICES Journal of Marine Science 55: 981–986.

Identifying Fish Farm Escapees

679

Hindar, K. and L’abée-Lund, J. H. 1992. Indentiﬁcation of hatchery-reared and naturally produced Atlantic salmon, Salmo salar L., juveniles based on examination of otoliths. Aquaculture and Fisheries Management 23: 235–241. Hobson, K. A. 1999. Tracing origins and migration of wildlife using stable isotopes: a review. Oecologia 120: 314–326. Joensen, H., Steingrund, P., Fjallstein, I., and Grahl-Nielsen, O. 2000. Discrimination between two reared stocks of con (Gadus mohua) from the Faroe Islands by chemometry of the fatty acid composition in the heart tissue. Marine Biology 136: 573–580. Jonsson, B. 1976. Comparison of scales and otoliths for age determination in brown trout, Salmo trutta L. Norwegian Journal of Zoology 24. Jonsson, B. and Stenseth, N. C. 1977. A method for estimating ﬁsh length from otolith size. Reports of the Institute of Freshwater Research, Drottningholm 56: 81–86. Kennedy, B. P., Blum, J. D., Folt, C. L., and Nislow, K. H. 2000. Using natural strontium isotopic signatures as ﬁsh markers: methodology and application. Canadian Journal of Fisheries and Aquatic Sciences 57: 2280–2292. Kennedy, B. P., Folt, C. L., Blum, J. D., and Chamberlain, C. P. (1997). Natural isotope markers in salmon. Nature 387: 766–767. Lund, R. A. and Hansen, L. P. 1991. Identiﬁcation of wild and reared Atlantic salmon, Salmo salar L., using scale characters. Aquaculture and Fisheries Management 22: 499–508. Lund, R. A., Hansen, L. P., and Järvi, T. 1989. Identiﬁsering av oppdrettslaks og villlaks ved ytre morfologi, ﬁnnestørrelse og skjellkarakterer. NINA forskningsrapport 001: 1–54. Lund, R. A., Midtlyng, P. J., and Hansen, L. P. 1997. Post-vaccination intra-abdominal adhesions as a marker to identify Atlantic salmon, Salmo salar L., escaped from commercial ﬁsh farms. Aquaculture 154: 27–37. Lund, R. A., Økland, F., and Hansen, L. P. 1991. Farmed Atlantic salmon (Salmo salar) in ﬁsheries and rivers in Norway. Aquaculture 98: 143–150. Lura, H. and Sægrov, H. 1991a. Documentation of successful spawning of escaped female Atlantic salmon, Salmo salar, in Norwegian rivers. Aquaculture 98: 151–159. Lura, H. and Sægrov, H. 1991b. A method of separating offspring from farmed and wild Atlantic salmon (Salmo salar) based on different ratios of optical isomers of astaxanthin. Canadian Journal of Fisheries and Aquatic Sciences 48: 429–433. Lura, H. and Økland, F. 1994. Content of synthetic astaxanthin in escaped farmed Atlantic salmon, Salmo salar, ascending Norwegian rivers. Fisheries Management and Ecology 1: 205–216. Manel, S., Berthier, P., and Luikart, G. 2002. Detecting wildlife poaching: identifying the origin of individuals with Bayesian assingment tests and multilocos genotypes. Conservation Biology 16: 650–659. Nielsen, E. E., Hansen, M. M., and Loeschke, V. 1997. Analysis of microsatellite DNA from old scale samples of Atlantic salmon Salmo salar: a comparison of genetic composition over 60 years. Molecular Ecology 6: 487–492. Paetkau, D., Calvert, W., Stirling, I., and Strobeck, C. 1995. Microsatellite analysis of population structure in Canadian polar bears. Molecular Ecology 4: 347–354. Poole, W. R., Webb, J. H., Matthews, M. A., and Youngson, A. F. 2000. Occurrence of canthaxanthin in Atlantic salmon, Salmo salar L., fry in Irish rivers as an indicator of escaped farmed salmon. Fisheries Management and Ecology 7: 377–385. Reddin, D. G. and Friedland, K. D. (1999). A history of identiﬁcation to continent of origin of Atlantic salmon (Salmo salar L.) at west Greenland, 1969–1997. Fisheries Research 43: 221–235. Ross, W. R. and Pickard, A. 1990. Use of scale patterns and shape as discriminators between wild and hatchery striped bass stocks in California. American Fisheries Society Symposium 7: 71–77. Sægrov, H., Hindar, K., Kålås, S., and Lura, H. 1997. Escaped farmed Atlantic salmon replace the original salmon stock in the River Vosso, western Norway. ICES Journal of Marine Science 54: 1166–1172.

680

Peder Fiske et al.

Tesch, F. W. 1968. Age and growth. In W. E. Ricker (ed.), Methods of Assessments of Fish Production in Freshwater. Blackwell Scientiﬁc, Oxford, UK. Thorrold, S. R., Lakoczy, C., Swart, P. K., and Jones, C. M. 2001. Natal homing in a marine ﬁsh metapopulation. Science 291: 297–299. Volk, E. C., Blakley, A., Schroder, S. L., and Kuehner, S. M. 2000. Otolith chemistry reﬂects migratory characteristics of Paciﬁc salmonids: using otolith core chemistry to distinguish maternal associations with sea and freshwaters. Fisheries Research 46: 251–266. Webb, J. H. and Youngson, A. F. 1992. Reared Atlantic salmon, Salmo salar L., in the catches of a salmon ﬁshery on the western coast of Scotland. Aquaculture and Fisheries Management 23: 393–397. Weitkamp, L. and Neely, K. 2002. Coho salmon (Oncorhynchus kisutch) ocean migration patterns: insight from marine coded-wire tag recoveries. Canadian Journal of Fisheries and Aquatic Sciences 59: 1100–1115. Winans, G. A. 1984. Multivariate morphometric variability in Paciﬁc salmon: technical demonstration. Canadian Journal of Fisheries and Aquatic Sciences 41: 1150–1159.

INDEX

Note: The letter f designates illustrations; t designates tables.

A Abalone. See Haliotus rubra Abaunza, P., 211 Abramis brama (bream) random ampliﬁed polymorphic DNA (RAPD) and, 373 Absence fragments, random ampliﬁed polymorphic DNA (RAPD) and, 378–379 Abundance. See Distribution and abundance data Acanthocephalans parasites as biological markers. See also Parasites as biological tags, 217 Acanthurus triostegus (convict surgeonﬁsh) ELS and genetic population structure of, 96 Accessible stock deﬁned, 634 Acipenser oxyrinchus (Atlantic sturgeon) chromosome morphology and, 274 fatty acid proﬁle and, 265 microsatellites and, 349 mitochondrial DNA and, 486–487 nuclear DNA and, 349 sampling and mixed stock analysis (MSA) in, 486–487 Acipenseridae, microsatellites and, 349 Adaptive traits, 47–51 landmark-based morphometric identiﬁcation and, 167–168 life history traits and, 63 norms of reaction and, 68–70

Adhesions of abdominal cavity in, post vaccination, 673–674, 674f Advisory Committee for Fisheries Management (ACFM), 640 Age classes vs. stocks, mixed stock analysis (MSA) and, 484–485 Age to maturity, 126–135 Gadus morhua (Atlantic cod), 134f norms of reaction and, 69–70 recruitment and, 135–137, 136f Age-invariant discriminant analysis, 503 Aggregate abundance-based management (AABM), 618–619 Alaska, threatened and endangered species conservation and, 620–622 Albacore tuna. See Thunnus alalunga Allele frequencies, 299–300 sampling and mixed stock analysis (MSA) in, 474, 486 temporal stability of, 486 threatened and endangered species conservation and, 620–622 Alleles, 295. See also Allozymes designation of, and scoring ambiguities in, 487–488 frequencies of, 299–300, 474, 486 Hardy–Weinberg equilibrium and, 299–300, 303, 327 inﬁnite allele model (IAM) in, 348 isoloci in, 487–488 sampling and mixed stock analysis (MSA) in, detection/identiﬁation in, 476

681

682 Allometric and age relations, sampling and mixed stock analysis (MSA) in, 483–485 Allozyme analysis, 295–309 advantages and limitations of, 301–302 allele frequencies and, 299–300 alleles and, 295 applications for, and examples of, 304, 305t banding patterns in, 297–298 Bayesian statistics used with, 301, 303 Clupea harengus (Atlantic herring), 99 deﬁning and describing, 295–296 early life stages (ELSs) and, 94–95 electrophoresis studies for, 296–298, 297f expectation maximization (EM) algorithm in, 300 Gadus morhua (Atlantic cod) and, 351–352 gene loci, isozymes and, 296 genotype identiﬁcation using (AA, AB, BB), 298 Gibbs sampler and, 301 GIRLSEM estimation in, 300 Hardy–Weinberg equilibrium and, 299–300, 303, 327 lactate dehydrogenase (LDH) and, 296 maximum likelihood estimation (MLE) in, 300–301, 302 mixed stock analysis (MSA) using, 296, 300–301 multimeric enzymes, multilocus genotypes and, 298 Oncorhynchus gorbuscha (pink salmon), 298, 304, 305t Oncorhynchus nerka (sockeye salmon), 305t Oncorhynchus tshawytscha (chinook salmon), 304, 305t Oncorhynchus keta (chum salmon), 298, 305t Oncorhynchus sp. and, 296 reliability of, factors affecting, 302–304 Salmo salar (Atlantic salmon), 296 Salmo trutta (brown trout), 296 Salvelinus malma (Dolly Varden), 296 Salvelinus namaycush (lake trout) and, 296 sampling and mixed stock analysis (MSA) in, 487–488 sampling error and, 302–303 software analysis programs for, 301 SPAM estimation in, 300 statistical analysis in, 300–301

Index threatened and endangered species conservation and, 621 Trachurus trachurus (horse mackerel), 651 variations in ﬁsh populations of, 295–296 Alosa sapdissima (American shad), 64 artiﬁcial neural networks (ANNs) and, 519 mitochondrial DNA and, 481 otolith, 228f sampling and mixed stock analysis (MSA) in, 481 American lobster. See Homerus americanus American oyster. See Crassotrea virginica American plaice. See Hippoglossoides platessoides American shad. See Alosa sapdisima Ampliﬁcation of DNA, 391 ampliﬁed fragment length polymorphism (AFLP) and, 399, 404 random ampliﬁed polymorphic DNA (RAPD) and, 372–376 Ampliﬁed fragment length polymorphism (AFLP), 389–411 advantages and applications for, 406–408 advantages and applications of, 390 ampliﬁcation of DNA for, 399, 404 analysis methods for, 402–407 banding in, 401–402, 401f base pairs (bp) and, 391 data analysis of, 405–406 digestion of DNA for, 397–399 DNA ﬁngerprinting and, 389–390, 395–397 Eco R1 restriction enzyme in, 393, 398–399, 400, 403 genotyping gels in, 404–405 inheritance of markers identiﬁed by, 401–402 isolation of genomic DNA for, 402–403 kits for analysis of, GIBCOBRI, LI-COR, etc., 403 ligation reaction in, 403 limitations to, 407–408 microsatellite DNA and, 393–395, 397 molecular analysis related to development of, 395–397 molecular basis of genetic variation and, 392–393 Mystus nemurus (river catﬁsh), 407 nuclear DNA and, 337, 341 nucleus, chromosomes, genomes and genomic DNA in, 391–392

Index phenetic tree construction using data from, 405 polymerase chain reaction (PCR) and, 389–390, 395–397, 399, 403–404 polymorphism and, 393–395, 394f, 400, 401f power of, potential for uses of, 400 procedures and principles of, 397–399, 398f random ampliﬁed polymorphic DNA (RAPD) and, 390, 396–397, 406 restriction enzymes and, 390–392, 393, 398–399, 400, 403 restriction fragment length polymorphism (RFLP) and, 390, 395–397, 406 Sebastes inermis (black rockﬁsh), 407 sequencing in, 399 software programs for analysis of data from, 406, 406t stock identiﬁcation applications for, 406–407 technology of, 390 term informative, 405 tissue selection for, 402 Analytical design, 1, 2–3, 8 Anchor tags, 416 Anchovy. See Engraulis anchoita, Engraulis encrasicolus Anguilla sp., Leptocephalus sp. (eels), 31, 66 A. anguilla (European eel), 33, 324 A. japonica (Japanese eel), 99–101, 100f A. rostrata (American eel), 32, 33, 323 failed adults and differentiation in, 33–35 “freshwater” and, contingent thinking, 35–36 genetic studies of, 36 life cycle of, 35–36 Mediterranean vs. North Atlantic origin of, 32 migration and population structure in, 31–37 mitochondrial DNA studies in, 36 speciation or population structure division in, debate over, 32–33 Anisakis simplex (nematode), 218 Anoplopoma ﬁmbria (sableﬁsh), tagging and, 420–421 Anser caerulescens caerulescens (snow goose), nuclear DNA and, 340 Antartic toothﬁsh. See Dissostichus mawsoni Antennas used with electronic tracking tags, 439

683 Apis mellifera (honeybee), random ampliﬁed polymorphic DNA (RAPD) and, 377–378 AP-PCR. See Random ampliﬁed polymorphic DNA (RAPD) Archival (retrieved) tags, 442 Arctic char. See Salvelinus alpinus Argyrosomus inodorus (silver kob), commercial catch per unit effort (CPUE) data in, 125f Armorhead. See Pseudopentaceros wheeleri Arnason–Schwarz multistate models, movement estimation from tagging data and, 592, 593–594 Artiﬁcial neural networks (ANN), 506–508, 507f, 512–513, 553–569 advantages of, 565–567 Alosa sapidissma (American shad), 519 applications for, 553–554, 555–556, 566–567 Bayesian theorem and, 524, 558–559 classiﬁcation of stocks using, 518–519, 520, 554, 556–560, 557t competitive learning in, 555 CONDLGA program in, 537–538, 545, 545f confusion matrices in, 533–534 delta rule learning in, 555 development of, 553–554 Dirichlet prior in, 549 distributional assumptions in, 559 estimation, 525–527 Bayesian approach to LDA assumptions in, 539–541 classiﬁcation-based conditional, 527 classiﬁcation-based vs. direct, 525 conditional vs. unconditional, 525–526, 525 conditional, 536–538 direct methods for, 536–541 Markov chain Monte Carlo (MCMC) algorithm in, 540, 544, 548–549 maximum likelihood for LDA assumptions in, 538–539, 541 plug in vs. predictive classiﬁcation in, 526–527 unconditional, 538–541 expectation maximization (EM) algorithm in, 300, 510, 534–535, 537 fault tolerance of, 560 Hebbian learning in, 555

684 Artiﬁcial neural networks (ANN) (continued) HISEA program (Fortran) and, 535, 537 inputs to, 554, 566 K-fold cross validation in, 533 Kohonen network. See also Self-organizing map (SOM) networks, 555, 562–565, 563f learning algorithms for, 555 linear discriminant analysis (LDA) in, 527–529, 530, 531, 535, 536 bootstrap sampling and, 548 logistic discriminant analysis (LGA) in, 519, 521, 537–538 logistic regression (LG) in, 531–532 logistic theory discrimination in, 529–532 MAP principle in, 545 Markov chain Monte Carlo (MCMC) algorithm in, 540, 544, 548–549 maximum likelihood estimation (MLE) in, 531, 532–536 microsatellite DNA analysis in, 566–567 mixture discriminant analysis (MDA) in, 519 modeling using, 554 multilayer feedforward network (MLFN) in, 555, 560–562, 561f multisource data in, 559 multivariate logistic compound (MLG) in, 544 multivariate normal mixture (MVN) in, 544 normal theory discrimination in, 527–529 numeric functions (continuous) in, 554–555 Oncorhynchus nerka (sockeye salmon), 519, 521, 522–523f, 541–546, 542f, 543–544t, 545f Oncorhynchus spp. (Paciﬁc salmon), 519 posterior probabilities in, classiﬁcation of mixtures, 523–525 posterior probabilities in, estimation of, 525–527 probability in, 522–523 Pseudopleuronectes americanus (winter ﬂounder), 566–567 quadratic discriminant analysis (QDA) in, 529 self-organizing map (SOM) networks and, 555–556, 562–565, 563f software for, commercially available, 566, 566t statistical analysis vs., 522–523, 565–567

Index statistical methods of classiﬁcation vs., 556–560 stochastic expectation maximization (SEM) in, 524 supervised vs. unsupervised, 555–556, 560–562 tasks for, 554–556 training in, 554, 560 UCONDLA program for, 540–541, 544 uncertainty in, 522–523, 535 weights and weighting in, 554 Astaxanthin and pink ﬂesh pigmentation, 675 Atlantic blueﬁn tuna. See Thunnus thynnus Atlantic cod. See Gadus morhua Atlantic herring. See Clupea harengus Atlantic mackerel. See Scomber scombrus Atlantic salmon. See Salmo salar Atlantic silverside. See Mendidia menidia, 48 Atlantic sturgeon. See Acipenser oxyrinchus, 349 Atomic absorption spectrometry (AAS), otolith analysis, 234 Audioradiography development of ﬁlms, 318

B Bacterial artiﬁcial chromosomes (BACs), 288, 290f Bain, Mark B., 435 Baltic Sea, cod management in, 643–645, 644f Banding, ampliﬁed fragment length polymorphism (AFLP) and, 401–402, 401f Banding pattern variations, chromosomal, 280–283 Banks, Michael A., 609 Barbus sp., random ampliﬁed polymorphic DNA (RAPD) and, 379 Barracuda. See Thrysites atun Barramundi. See Lates calcarifer BASBLACK project, 637 Base pairs (bp), 391 Bayes theorem, 301, 303, 524, 558–559 LDA and, 539–541 movement estimation from tagging data and, 596 Beam-based assays, otolith, 234–235 Begg, Gavin A., 119 Bigeye tuna. See Thunnus obesus Biochemical markers, 30

685

Index Biodiversity adaptive traits and, 47–51 cogradient variation and, genetic vs. environmental variation in, 51 “common garden” experiments, genetic vs. environmental characters, 50, 69 countergradient variation and, genetic vs. environmental variation in, 51 ﬁtness and, 53–54 gene ﬂow between groups and, 47–51 genetic drift and, 47–51 intraspeciﬁc patterns of, factors underlying, 47–51 microsatellite DNA and, 47–51 migration and, 37 natural selection in, 47–51 phenotype plasticity, 48–51, 55–58, 67–70 phenotypic modulation and, 49–51, 49f reciprocal transplant experiments and, genetic vs. environmental variation, 50 Biological stock deﬁned, 634 Biological variation and management difﬁculties, 640–643, 642f Biomass, spawning stock (SSB), 641–642 Black rockﬁsh. See Sebastes inermis Blood protein studies, Gadus morhua (Atlantic cod) and, 351–352 Blue marlin. See Makaira nigricans Blue shrimp. See Penaeus stylirostris Blue whiting. See Micromesistius poutassou Blueﬁn tuna. See Thunnus thynnus Blueﬁsh. See Pomatomus saltatrix Bluegill sunﬁsh. See Lepomis macrochirus Body shape. See Morphometric traits Body size. See also Morphometric traits sampling and mixed stock analysis (MSA) in, 484 spawning vs. egg amount/viability and, 59–60 Booke, H.E., 37 Bootstrap sampling and logistic discriminant analysis (LDA), 548 Brachydanio rerio (zebraﬁsh), 52, 287f, 289 Bream. See Abramis brama Breed/brood stock, 470–471 sampling and mixed stock analysis (MSA) in, 470–471 spawning stock biomass (SSB) in, 641–642 supportive breeding programs in, 619–620

threatened and endangered species conservation and, 615–617 Brodziak, Jon, 571 Brown trout. See Salmo trutta

C C banding, 281, 283f, 285 Cadrin, Steven X., 1, 2, 153, 173, 185 Calciﬁed structure texture and spacing patterns, 185–195 aquaculture releases, escapees and, 191–192 bias in, 187–189 body parts collected for, 193 case studies in, 189–192 character selection in, 187–189 circuli spacing data in, 187–189, 190 development of techniques for, 185–186 environmental inﬂuences on, 193 Fourier series analysis in, 187 growth rates and, 186 image processing tools in, 192–193 interpretation of, 186, 192–193 Lates calcarifer (barramundi), 191, 668 measurement techniques used for, 186 methodology for, 187–189 Morone saxatilis (striped bass), 190–191, 191f Oncorhynchus kisutch (coho salmon), 190 Oncorhynchus mykiss (rainbow trout), 190, 191 Oncorhynchus nerka (sockeye salmon), 188f, 189 optical density proﬁles in, 187, 188f Salmo salar (Atlantic salmon), 190, 191–192, 192f silver perch, 191 Campana, Steven E., 227 Cancer magister (Dungeness crab), ELS and genetic population structure of, 96 Cannibalism in Arctic cod, 642–643 Canthaxanthin and pink ﬂesh pigmentation, 675 Capelin. See Mallotus villosus Capture and handling of ﬁsh for tagging, 419 Carassius carassius (crucian carp), 55 Carcharodon carcharias (white shark), electronic tracking tags and, 437 Carlin tags, 427

686 Carotenoids and pink ﬂesh color, escapee (aquaculture) identiﬁcation and, 674–675 CART (tree-based regression), 513 Castle–Hardy–Weinberg equilibrium, 10. See also Hardy–Weinberg equilibrium Catﬁsh. See Mystus nemurus Certiﬁcates of origin, 21 Cestodal parasites as biological markers. See also Parasites as biological tags, 217 Cestode pleorocercoid parasites as biological markers. See also Parasites as biological tags, 217 Cherax quadricarinatus (redclaw) random ampliﬁed polymorphic DNA (RAPD) and, 373 Chi-square tests, sampling and mixed stock analysis (MSA) in, 474 Chromosome morphology, 273–294 Acipenser oxyrincus desotoi (sturgeon), 274 applications for, 289–290 Atlantic salmon (Salmo salar), 275, 277, 279f, 281, 289 bacterial artiﬁcial chromosomes (BACs) and, 288, 290f banding pattern variations in, 280–283 blood and tissue sampling for, 284 Brachydanio rerio (zebraﬁsh), 287f, 289 C banding in, 285 catﬁsh, 289 colchicine treatment, hypotonic treatment and ﬁxation of cells for, 284 DAPI banding in, 285 dividing cells for, methods for obtaining, 284 duplications and deletions in, 280 ﬂuorescence in situ hybridization (FISH) in, 286, 287f, 287–289, 290f ﬂuorochrome banding in, 282–283, 285 heterochromatin additions and deletions (C bands) in, 281, 283f hybrids and, 290 Ilyodon fucidens (goodeid), 277 intraspecies variation in, 273–283 inversions in, 277–280 lake trout, 282, 283 medaka, 289 methods for detecting, 283–287 molecular cytogenic methods in, 287–289 multicolor FISH (M-FISH) in, 288–289

Index nucleolar organizer regions (NORs) and, 274, 281, 282f, 285–286, 290 number of, intraspecies variation in, 273–277 Oncorhynchus gorbushcha (pink salmon), 277 Oncorhynchus keta (chum salmon), 277 Oncorhynchus mykiss (rainbow trout), 275f, 276f, 277, 278f, 283, 289, 290f, 289 Oncorhynchus sp. (salmon), 275, 279f phage artiﬁcial chromosomes (PACs) and, 288 photography and analysis in, 286–287 polyploidy in, 274, 275f pufferﬁsh, 289 Q banding in, 283, 285 reciprocal translocations in, 280 replication banding in, 286 restriction enzyme (RE) banding in, 286 ribosomal DNA (rDNA) in, 280–281, 285–286 Rivulus genus (cyprinids), 274 Robertsonian translocations and, 274–277, 278f Rutilus genus (cyprinids), 274 Salvelinus alpinus (arctic char), 281, 282f, 282, 283 Salvelinus leucomaenis (iwana), 281, 282 Salvelinus namaycush (lake trout), 281 sequencing variations in repetitive DNA and, 282–283 shellﬁsh, 275, 289, 290 slide preparation for, 284–285 staining for, 285 structural differences in, 277–280 tandem translocations in, 277, 279f tilapia, 289 triploidy in, 274 whiteﬁsh, 281 yeast artiﬁcial chromosomes (YACs) and, 288 Chromosomes, 391–392 Cichlasoma managuense, 56 Cichlid. See Geophagus sp. Circuli. 187–189, 190. See also Calciﬁed structure texture, spacing patterns and Classiﬁcation algorithms and sampling, number of features necessary for, 479–481 Classiﬁcation models, mixed stock analysis (MSA) and, 468

687

Index Classiﬁcation of stocks, 518–519, 520 artiﬁcial neural networks (ANNs) and, 554, 556–560, 557t discrete vs. nondiscrete, 502–504 plug in vs. predictive classiﬁcation in, 526–527 statistical analysis in, 518–519, 520 Clupea harengus (Atlantic herring), 59 allozyme analysis and, 99 ELS stock deﬁnition in, 97–99, 98f fatty acid proﬁle and, 251t, 252–255, 254f, 675 genetically based identiﬁcation of, 99 life history traits and, changes in, 138 management issues and, 637–640, 638f metapopulation concept and, 108–109 sampling and mixed stock analysis (MSA) in, 484 sampling of, 637–640, 638f spatial segregation of spawning stock and, 142 spawning discretenes of various stocks and, 131–132 tagging, mark-release studies on, 416, 424–427, 426f year-class phenomenon in, 26–27 Clupea pallasi (Paciﬁc herring) tagging, mark-release studies on, 424–427 Clupeidae, microsatellites and, 349 Clustan32 AFLP data analysis software, 406 Cod hatchery enhancement, migration and, 19–21, 20t Cod. See Gadus macrocephalus; G. morhua Coded wire tag (CWT), 417, 420, 427 Cogradient variation and, genetic vs. environmental variation, 51 Colchicine treatment, chromosome morphology and, 284 Commercial catch per unit effort (CPUE) data, 125f, 125 “Common garden” experiments, genetic vs. environmental characters, 50, 69 norms of reaction and, 69–70 Competitive learning, artiﬁcial neural networks (ANNs) and, 555 Composite stock identiﬁcation, sampling and mixed stock analysis (MSA) in, 472–473 Conditional estimation, 536–538 Conditional vs. unconditional estimation, 525–526

CONDLGA program, 537–538, 545, 545f Conferences on stock identiﬁcation, 1–2 Confusion matrices, 533–534 Congruence of results, 13–14 Conservation biology, 11 Conservation. See Threatened or endangered species conservation Contingent concept in migration, 35–36 Continuous time/space models, movement estimation from tagging data and, 592, 600–602 Convict surgeonﬁsh. See Acanthurus triostegus Coregonidae, microsatellites and, 350 Cottus sp., random ampliﬁed polymorphic DNA (RAPD) and, 379 Countergradient variation and, genetic vs. environmental variation, 51 Crassotrea virginica (American oyster), nuclear DNA and, 341 Crucian carp. See Carassius carassius Crustacea parasites as biological markers. See also Parasites as biological tags, 214, 217, 218 Cultus Lake, homing mechanism experiments in, 25 Curvature methods, maximum likelihood estimation (MLE) and, 573, 573f Cushing, D.H., 29 Cutthroat trout. See Salmo clarki Cymatogaster aggregata (surfperch), 62 Cynoscion regalis (weakﬁsh) metapopulation concept and, 108–110, 109f otolith analysis, 240 Cyprinodont ﬁsh. See Rivulus marmoratus Cytogenic methods, chromosome morphology and, 287–289

D Dahl, K., 20, 21 Dannevig, G.M., cod hatchery enhancement program, 20–21, 20t DAPI banding, chromosome morphology and, 285 Dart streamer tags, 416 Deepwater stocks of North Atlantic, management issues and, 634–637 Deﬁnition of stock, 7–16. See also Stock identiﬁcation

688 Deﬁnition of stock (continued) Anguilla japonica (Japanese eel), using ELS, 99–101, 100f Clupea harengus (Atlantic herring), using ELS, 97–99, 98f contingent concept in, 90, 108 early life stages (ELSs) and, 89–91 genotypic, 90, 91–97, 92f Homerus americanus (American lobster), using ELS, 101–102, 103f management issues and, 632, 633 member-vagrant hypothesis in, 90–91, 94, 110 phenotypic, 90, 91–97, 92f spatial population models and, 108–110 stronger vs. weaker, 90 congruence of results in, 13–14 conservation biology and, 11 depth of stock division in, 13–14 early life stages (ELSs) and, 91–97 environmental vs. genetic markers and, 46 evolutionarily signiﬁcant unit (ESU) concept in, 11 evolving concepts of, 9–11 fundamental issues of, 11–14 genetically based identiﬁcation in, 9–15 harvest stock in, 9, 15 management of stocks and, 7 negative results (Type 1 errors) and null hypotheses in, 12–13 pattern or structure of resources in, 8 population dynamics and, 7 precision and, degree of, 8 single-approach studies in, 14 stock composition analysis in, 8 stock discrimination vs. identiﬁcation in, 8 unit stock concept in, 7–8, 14, 17–44 Deletions of chromosomes/DNA, 280, 393 Delta rule learning, artiﬁcial neural networks (ANNs) and, 555 Depth of stock division, 13–14 Dicentrarchus labrax (sea bass) random ampliﬁed polymorphic DNA (RAPD) and, 374, 380, 381t Diet fatty acid proﬁle and, 248, 265–267 morphometric traits and, 55–56 Differences among stocks, statistical signiﬁcance of, 482–483

Index Digenean metacercariae parasites as biological markers. See also Parasites as biological tags, 217 Digestion, digestion proﬁle in ampliﬁed fragment length polymorphism (AFLP), 397–399 in restriction analysis of mtDNA, 314 Dinucleotide (AC) repeats, microsatellites and, 344, 345f Direct estimation, 536–541 Directed migration. See Parent stream migration theory Dirichlet prior, 549 Discrete stock/discrete time models Arnason–Schwarz multistate models in, 592, 593–594 Bayesian theorem and, 596 example of, Hippoglossus stenolepis (Paciﬁc halibut), 597–600, 598f, 599t harvest models in, 595–596 model ﬁtting in, 596–597 movement estimation from tagging data and, 591–600 software for (MARK, M-SURGE), 596 stratiﬁcation of stock in, 592–593 Discrete vs. nondiscrete classiﬁcation in, 502–504 Discriminant analysis (DA), 501–504, 503f, 511–512, 517–552 parasites as biological tags and, 215–216 Disease impact, threatened and endangered species conservation and, 615–617 Dissostichus mawsoni (Antarctic toothﬁsh) random ampliﬁed polymorphic DNA (RAPD) and, 375 Distribution and abundance data, 121–126, 122f, 121 aggregate abundance-based management (AABM) in, 618–619 commercial catch per unit effort (CPUE) data in, 125f ﬁsheries-dependent sources of data for, 121, 125–126, 125f Gadus morhua (Atlantic cod), 123f Generalized Additive Models (GAMs) in, 121, 123f Geographic Information Systems (GIS) in, 121 individual stock-based manaegment (ISBM) in

Index Melanogrammus aegleﬁnus (haddock), 122f Scomberomorus munroi (spotted mackerel), 124f sources of information on, 121–122 spatial population models of, 121 Distributional assumptions, artiﬁcial neural networks (ANNs) and, 559 Diversity. See Biodiversity DNA, 13. See also Genetically based identiﬁcation mitochondrial. See Mitochondrial DNA DNA ampliﬁcation ﬁngerprinting (DAF), 372 DNA ﬁngerprinting, 342–344 ampliﬁed fragment length polymorphism (AFLP) and, 389–390, 395–397 Dolly Varden. See Salvelinus malma Dover sole. See Microstomus paciﬁcus Driftnet ﬁshing, United Nations resolution 44–225 (driftnet ﬁshing) and, 622–623 Drosophila sp. studies in nuclear DNA, 335, 339 Dungeness crab. See Cancer magister Duplications of chromosomes/DNA, 280, 393

E Early life stages (ELSs) in identiﬁcation, 89–117 allozyme analysis and, 94–95 Anguilla japonica (Japanese eel), stock deﬁnition using, 99–101, 100f Clupea harengus (Atlantic herring), stock deﬁnition using, 97–99, 98f contingent and vagrant stock in, 90–91, 108 ecosystem framework for, in stock identiﬁcation, 105–106 future directions for use of, 104–110 genotypic vs. phenotypic stock in, 90, 91–97, 92f, 108 geostatistics and, 104–105 Homerus americanus (American lobster), stock deﬁnition using, 101–102, 103f larval morphometrics and, 91, 92, 96–97 life history traits and, 96–97, 104–105 management and, 89–90 otoliths and, 94–95 phenotypic traits and, 96–97 planktonic stage in, 91, 92

689 planktonic transport stage in, 93–94, 106–108 population structures and, 91–93, 93f, 95–96 space and time separation in, 95 spatial population models and stock concept using, 108–110 spatial segregation of stocks and, 142 spawning discreteness and, 94–95 stock concept and, 91–97 “stock” deﬁned for, 89–91 vagrants and member-vagrant hypothesis in, 90–91, 94, 110 Eco-R1 restriction enzyme, ampliﬁed fragment length polymorphism (AFLP) and, 393, 398–399, 400, 403 Ecosystems, 105–106 early life stages (ELSs) and, in stock identiﬁcation, 105–106 larval stage in, 105–106 plankton in, 105–106 Eel. See Anguilla anguilla; Leptocephalus brevirostrum Effects of handling and tagging on ﬁsh, 420–421 Eggs. See Spawning Eisenhower, Dwight D., 621 Electronic tag readers, 416, 417 Electronic tags, 435–446. See also Tagging Electrophoresis allozyme analysis and, 296–298, 297f fatty acid proﬁle and, 262–263 mitochondrial DNA and, 317–318 Elemental ﬁngerprint. See Otoliths, 241 Elliptical Fourier analysis, outline-based morphometric identiﬁcation and, 177 Endangered Species Act (ESA) and, 610–611 Endangered species. See Threatened or endangered species conservation Energy-dispersive electron microprobes (ED-EM), otolith analysis, 235 Engraulis anchoita (anchovy) ELS and phenotypic traits of, 97 Engraulis encrasicolus (European anchovy) mitochondrial DNA and, 326–327 Enriched mtDNA extratraction, 316–318, 317f Environmental inﬂuences on identiﬁcation, 45–86, 632

690 Environmental inﬂuences on identiﬁcation (continued) calciﬁed structure texture, spacing patterns and, 193 cogradient variation and, genetic variation vs., 51 “common garden” experiments and, genetic variation vs., 50, 69 countergradient variation and, genetic variation vs., 51 deﬁning “stock” and, 46 life history traits in, 59–70 management issues and, 632 meristic traits and, 51–54, 197 morphometric traits and, 54–58 norms of reaction and, 67–70 otolith composition and, 46, 229, 233–234, 241–242, 449 phenotypic modulation and, 49–51, 49f plasticity of phenotypes and, 48–51, 55–58, 67–70 reciprocal transplant experiments and, genetic variation vs., 50 strengths and weakness of, 46 Enzymes. See also Allozymes restriction. See Restriction enzymes Epinephelus (grouper) random ampliﬁed polymorphic DNA (RAPD) and, 379 Escapees from aquaculture farms, 659–679 adhesions of abdominal cavity in, post vaccination, 673–674, 674f biochemical and physiological markers of, 673–675 calciﬁed structure texture, spacing patterns and, 191–192 carotenoids and pink ﬂesh color in, 674–675 fatty acid proﬁles as natural markers, 675 ﬁn measurement in, 666–667 ﬁn ray defects in, 661–665, 662f, 663f, 664f, 666t ﬁn tissue defects in, 660–661, 661f genetic markers in, 676 gill cover shortening in, 665, 665f identiﬁcation of, importance of, 659–660 modeling in, 668–669 monitoring methods for, 660–676 morphology and morphometriy in, 660–667 morphometric landmarks in, 667, 668f

Index Oncorhynchus mykiss (rainbow trout), 659–660 otolith analysis and, 675 perspectives in aquatic practices and, 676–677 pigmentation variations in, 665–666 replacement scales in, 672–673, 673f Salmo salar (Atlantic salmon), 659–660 scale and otolith pattern recognition in, 667–673 sea winter band and, 669–671 smolt size and age in, 669 summer checks in, 671–672 tagging, 676 trace elements in, 675 transition zone, salt-to-freshwater, 669, 670f, 671f undershot jaw in, 665 Esox lucius (Northern pike), 55 Estimation, 526–549 artiﬁcial neural networks (ANNs) and, 525–527 classiﬁcation-based conditional, 527 classiﬁcation-based vs. direct, 525 conditional vs. unconditional, 525–526 conditional, 536–538 direct methods for, 536–541 Markov chain Monte Carlo (MCMC) algorithm in, 540, 548–549, 544 maximum likelihood for LDA assumptions in, 538–539 maximum likelihood for LGA assumptions in, 541 movement estimation from tagging data, 591–606 plug in vs. predictive classiﬁcation in, 526–527 unconditional, 538–541 EURING conferences, movement estimation from tagging data and, 592, 597 European anchovy. See Engraulis encrasicolus Evolution and population development, 18 adaptive traits and, 47–51 ﬁtness in, 53–54 gene ﬂow between groups and, 47–51 genetic drift and, 47–51 microsatellite DNA and, 47–51 mitochondrial DNA and, 312, 315f natural selection and, 31, 46, 47–51, 59 norms of reaction and, 67–70

Index phenotype plasticity, 48–51, 55–58, 67–70 phenotypic modulation and, 49–51, 49f sexual selection and, 46 Evolutionarily signiﬁcant unit (ESU), 11 Exons, 334–336, 335f Expectation maximization (EM) algorithm, 300, 534–535, 537 allozyme analysis and, 300, 510 maximum likelihood estimation (MLE) and, 576 Experimental design. See Mixed stock analysis, experimental design and sampling strategies for Extended likelihood model, 574–576 Extinction, morphometric traits and, 57 Extraction of DNA, for random ampliﬁed polymorphic DNA (RAPD), 372–376

F Fabrizio, Mary C., 467 Failed adults eel theory and differentiation, 33–35 FAIR project, 637 Fatty acid proﬁles as natural markers, 247–269, 675 Acipenser oxyrincus desotoi (sturgeon), 265 applications for, 265 case histories in, 252–264, 252 Clupea harengus (Atlantic herring), 251t, 252–255, 253f, 254f, 675 composition of, 247–248, 265 diet and composition of, 248, 265–267 electrophoretic analysis of, 262–263 factors inﬂuencing, 264–267 fraud investigation using, 252–255, 255f Gadus morhua (Atlantic cod), 257–259, 260t, 261f, 675 gas chromatography used in, 249–250, 250t harp seal migration and identiﬁcation using, 263–264, 263t Homerus americanus (American lobster), 265 Illex illecebrosus (squid), 265 MEP-1 enzyme analysis and, 262 methodology for, 248–252 Morone saxatilis (striped bass), 255–256, 257f Oncorhynchus mykiss (rainbow trout), 266 Oncorhynchus sp., 256–257, 258t, 259f partial least square (PLS) analysis in, 251

691 phospholipids and, 266 Pleuronectes platessa (plaice), 266 principal component analysis (PCA) for, 250–251 sample preparation for, 248–249 Scomberomorus sp. (mackerel), 255, 256f Scopthalmus maximus (turbot), 265 Sebastes mentalla (redﬁsh), 259–263, 261f, 262f soft independent modeling of class analogy (SIMCA) analysis in, 251–252 triglycerides and, 265–266 variations in, 247–248, 251, 264–267 Fault tolerance, artiﬁcial neural networks (ANNs) and, 560 Fecundity/fertility, life history traits and, 65–67 Fin measurement, escapee (aquaculture) identiﬁcation and, 666–667 Fin ray defects, escapee (aquaculture) identiﬁcation and, 661–665, 662f, 663f, 664f, 666t Fin tissue defects, escapee (aquaculture) identiﬁcation and, 660–661, 661f Fingerprinting, DNA. See DNA ﬁngerprinting Fingerprinting, DNA ampliﬁcation (DAF), 372 Fingerprinting, nuclear, 342–344 Finite distribution methods, 508–510 Finite mixture distributions (FMD), 508–510, 509f, 513, 514 Fisheries convention areas, North Atlantic, ICES, 635–637, 636f Fishery stock deﬁned, 633–634 Fishing and harvesting, 633 aggregate abundance-based management (AABM) in, 618–619 ﬁshery stock deﬁned for, 633–634 harvest stock deﬁned for, 633 ICES ﬁsheries convention areas, 635–637, 636f individual stock-based manaegment (ISBM) in, 619 norms of reaction and, 70 Paciﬁc Salmon Treaties, 618–619 recruitment data for, 135–137, 136f targeted stock deﬁned for, 633–634 threatened and endangered species conservation and, 615–617 threatened and endangered species conservation and, politics and, U.S./Canada salmon wars, 617–619

692 Fishing and harvesting (continued) Total Allowable Catches (TAC) allocation in, 634, 639–640, 641, 653 Trachurus trachurus (horse mackerel), 652 U.S. Exclusive Economic zone and, 624 United Nations resolution 44–225 (driftnet ﬁshing) and, 622–623 Fishing mortality rates, 7 Fiske, Peder, 659 Fitness, in evolution, 53–54 Fixation and preservation of parasites, 217 Flodevigen cod hatchery, 19–21 Flounder. See Limanda ferruginea; Paralichthys dentatus; Pseudopleuronectes americanus Fluorescence in situ hybridization (FISH), chromosome morphology and, 286, 287f, 287–289, 290f Fluorochrome banding, 282–283, 285 Flying ﬁsh. See Hirundichthys afﬁnis Foote, Chris J., 45 Fourier series analysis calciﬁed structure texture, spacing patterns and, 187 outline-based morphometric identiﬁcation and, 173, 175, 176–178, 177f, 178f Fraser River, salmon migration and natural tag identiﬁcation in, 22–25, 23f Fraud investigation using fatty acid proﬁle, 252–255, 255f Friedland, Kevin D., 1, 2, 173, 185

G Gadidae, microsatellites and, 350 Gadus sp. Arctic cod, management issues and, 640–643, 642f Baltic cod, management issues and, 643–645, 644f biological variations and, 640–643, 642f Gadus macrocephalus (Paciﬁc cod) random ampliﬁed polymorphic DNA (RAPD) and, 374, 381t Gadus morhua (Atlantic cod) age to maturity data, 134f allozymes in, 351–352 blood protein studies in, 351–352 distribution and abundance data, 123f ELS and genetic population structure of, 95–96

Index fatty acid proﬁle and, 257–259, 260t, 261f genetic analyses performed on, 351–359 life history traits and, changes in, 137–138, 140f microsatellites in, 355–359 minisatellites in, 355 mitochondrial DNA and, 323, 352–353 nuclear DNA and, 341, 351–359, 362 otolith analysis in, 231f, 232f, 233, 239–240 Pan1 alleles in, 354, 357, 361 recruitment data on, 136f sampling and mixed stock analysis (MSA) in, 472 single nucleotide polymorphisms (SNPs) and, 354–355 single-copy coding nuclear DNA in, 353–354 single-copy noncoding nuclear DNA in, 354–355 spawning data on, 133f stock identiﬁcation in, problems with, 351 Galaxias platei, 57 Gas chromatography, fatty acid proﬁle and, 249–250, 250t Gasterosteus sp. (stickleback), 55, 66 G. aculeatus (threespine stickleback), 56 Gene diversity indexes, mitochondrial DNA and, 320 Gene ﬂow between groups and biodiversity, 47–51 GENECLASS allozyme analysis software, 301 Generalized Additive Model (GAM), 121 Gadus morhua (Atlantic cod), distribution data on, 123f life history traits and, 121, 123f Generalized Linear Model (GLM), 130 Genetic drift, 47–51 Genetic ﬁngerprinting. See DNA ﬁngerprinting Genetic gaps identiﬁed by mtDNA, 320 Genetic mixture analysis (GMA), 621 Genetically based identiﬁcation, 9–15, 45–86, 90 adaptive traits and, 47–51 allozymes. See Allozymes alternatives to, 11–12 ampliﬁed fragment length polymorphism (AFLP), 389–411 chromosome morphology in, 273–294 Clupea harengus (Atlantic herring), 99 cogradient variation and, environmental variation vs., 51

Index “common garden” experiments and, environmental variation vs., 50, 69 countergradient variation and, environmental variation vs., 51 deﬁning “stock” and, 46 DNA in, 13 eels, 36 escapee (aquaculture) identiﬁcation and, 676 gene ﬂow between groups and, 47–51 genetic drift and, 47–51 genotypic vs. alternative approaches to, 11–12 Hardy–Weinberg equilibrium and, 299–300, 303, 327 intraspeciﬁc patterns of diversity in, factors underlying, 47–51 life history traits in, 46, 59–71, 120 meristic traits and, 46, 51–54, 197–207, 198f microsatellite DNA, 13, 46, 47–51 mitochondrial DNA in, 46, 311–330 molecular basis of genetic variation and, 392–393 molecular markers and, 47–51 morphometric traits in, 46, 54–58, 71, 153–172 morphotypes and, 49 natural selection and, 46, 47–51, 71 natural tags in, 23–24, 23f, 38 norms of reaction and, 67–70 nuclear DNA, single-copy and repetitive sequence markers, 331–369 otoliths and, 229 parasites as biological tags and, 218 phenetic tree construction using data from, 405 phenotypic modulation and, 49–51, 49f philopatry in, 23–24, 23f plasticity of phenotypes and, 48–51, 55–58, 67–70 protein allozymes in, 46, 48–51 quantitative traits in, 46 random ampliﬁed polymorphic DNA (RAPD), 371–387 reciprocal transplant experiments and, environmental variation vs., 50 Sebastes sp. (redﬁsh), 645–650 sexual selection and, 46 strengths and weakness of, 46 Trachurus trachurus (horse mackerel), 651

693 Genomes, 391–392 molecular basis of genetic variation and, 392–393 nuclear, 334–336 population genomics and, 625–626 Genomic DNA, 391–392 Genotypes, 90, 91–97, 92f allozyme analysis and, identiﬁcation using (AA, AB, BB), 298 mitochondrial DNA and, 312 Genotypic vs. alternative approaches to identiﬁcation, 11–12 Genotyping gels, in ampliﬁed fragment length polymorphism (AFLP), 404–405 Geographic Information Systems (GIS), life history traits and, 121 Geolocation homing mechanisms, 25 Geophagus brasiliensis (cichlid), 55 Geophagus steindachneri (cichlid), 55 Geostatistics, 104–105 early life stages (ELSs) and, 104–105 Gibbs sampler, 301 Gilbert, C.H., 22–23, 26, 37 Gill cover shortening, escapee (aquaculture) identiﬁcation and, 665, 665f GIRLSEM estimation, 300 Global impacts, cod hatchery enhancement and, 19–21, 20t Global positioning satellite (GPS), movement estimation from tagging data and, 602, 604 Gobies, 62 Gonadosomatic index (GSI), 66 Goodness-of-ﬁt testing, statistical analysis, 474 Grahl-Nielsen, O., 247 Grassi, G.B., 32, 36 Great Barrier Reef Marine Park, metapopulations and, 109 Grimm, Jeffrey J., 447 Grouper. See Epinephelus, 379 Growth rate, 7, 126–131 age at maturity as, 64–65 calciﬁed structure texture, spacing patterns and, 186 curves of, comparisons of, 128, 129f environmental inﬂuences on, 60 Generalized Linear Model (GLM) in, 130 genetic inﬂuences on, 61 Lethrinus miniatus, 129f Melanogrammus aegleﬁnus (haddock), 129f

694 Growth rate (continued) norms of reaction and, 67–70 Scomberomorus queenslandicus (school mackerel), 129f sex speciﬁc, 127–128 variation among populations, 62–63 von Bertalanffy equation for, 60, 128–130, 129f GSI Foundation, threatened and endangered species conservation and, 620–622 Guppy. See Poecilia reticulata

H Haddock. See Melanogrammus aegleﬁnus Hake. See Merluccius bilinearis Halibut. See Hippoglossus stenolepis Haliotus rubra (abalone), random ampliﬁed polymorphic DNA (RAPD) and, 375 Hammer, Cornelius, 631 Handling of ﬁsh for tagging, 419 Hansen, Lars P., 415, 659 Haplotypes, mitochondrial DNA and, 314 Harbor porpoise, mitochondrial DNA analysis in, 623 Hardy–Weinberg equilibrium, 299–300, 303, 327 microsatellites and, 346 mitochondrial DNA and, 480–481 random ampliﬁed polymorphic DNA (RAPD) and, 377, 382 sampling and mixed stock analysis (MSA) in, 474, 480–481 Hare, Jonathan A., 89 Harp seal migration and identiﬁcation, fatty acid proﬁle and, 263–264, 263t Harvest models, movement estimation from tagging data and, 595–596 Harvest stock, 9, 15, 633 Harvesting. See Fishing and harvesting Hatchery releases, 19–21, 20t. See also Escapees from aquaculture farms calciﬁed structure texture, spacing patterns and, 191–192 cod, in Norwegian waters, 19–21, 20t effectiveness of, 21 escapees from aquaculture farms and, 659–679 mixed stock analysis (MSA) and, 467–469

Index otolith thermal marking and, 458–459, 460–461 salmon and, 24 transplant experiments and, 26 Hebbian learning, artiﬁcial neural networks (ANNs) and, 555 Helminthes parasites as biological markers. See also Parasites as biological tags, 214, 217, 218 Hematopoietic necrosis virus (IHNV), 615–617 Herring. See Clupea harengus; C. pallasi Heterochromatin additions and deletions (C bands), 281, 283f Hippoglossoides platessoides (American plaice) morphometric variation in, 58 Hippoglossus stenolepis (Paciﬁc halibut) movement estimation from tagging data and, 597–600, 598f, 599t Hirundichthys afﬁnis (ﬂying ﬁsh) random ampliﬁed polymorphic DNA (RAPD) and, 374, 381t HISEA program (Fortran), 535, 537 Hisla shad. See Tenualosa ilisha Hjort, Johan, 19, 21, 37 Holding of ﬁsh for tagging, 419, 420 Homerus americanus (American lobster) ELS stock deﬁnition using, 101–102, 103f fatty acid proﬁle and, 265 morphometric traits in, 96, 102 random ampliﬁed polymorphic DNA (RAPD) and, 375 sampling and mixed stock analysis (MSA) in, 485 Homing mechanisms, 25, 38 member-vagrant hypothesis in, 30 otolith analysis, 240 parent stream theory in, directed migration vs. local wandering, 27, 28–30 salmon, 38 year-class phenomenon in, 26–27 Honeybee. See Apis mellifera Hoplostethus atlanticus (orange roughy), 67 random ampliﬁed polymorphic DNA (RAPD) and, 375, 380, 381t Horse mackerel. See Trachurus trachurus Horseshoe crab. See Limulus polyphemus Huntsman, A.G., 27–28 Hutchings, Jeffrey A., 45

695

Index Hybridization ampliﬁed fragment length polymorphism (AFLP) and, 406–407 chromosome morphology and, 290 mitochondrial DNA and, 318 Hydrographic containment, 29

I Illex illecebrosus (squid), fatty acid proﬁle and, 265 Ilyodon fucidens (goodeid), 56 chromosome morphology and, 277 Image processing tools calciﬁed structure texture, spacing patterns and, 192–193 outline-based morphometric identiﬁcation and, 173, 174 In situ tagging methods, 427–429 Indian mackerel. See Rastrelliger kanguarat Individual stock-based management (ISBM) in, 619 Inductively couple plasma mass spectrometry (ICPMS), 234, 237, 239, 240 Inductively coupled plasma atomic emission spectroscopy (ICP-AES), 234 Infectious hematopoietic necrosis virus (IHNV), 615–617 Inheritance and AFLP markers, 401–402 Interdisciplinary analysis in stock identiﬁcation, 3 Internal and external tags. See Tagging International Baltic Sea Fisheries Commission (IBSFC), 644–645 International Council for the Exploration of the Sea (ICES), 2, 17 deepwater stock identifcation/management, 635–637 ﬁsheries convention areas, 635–637, 636f International North Paciﬁc Fisheries Commision (INPC), 622 salmon migration and spawning, 22 Introns, 334–336, 335f Inversions, chromosome morphology and, 277–280 Isoloci, 487–488 Isotope dilution ICPMS (ID-ICPMS), otolith analysis, 234, 237 Isozymes. See also Allozymes

J Jackknife discriminant analysis, 503 Jacobsen, J.A., 415 Jones, Harden, 25, 27, 29, 30, 34 Jordan’s rule phenomenon, 197

K K/M ratio, 60 Kappa index, otolith thermal marking and, 457 Karl Avise, 48 Katsuwonus palamis (skipjack tuna) mitochondrial DNA and, 323 movement estimation from tagging data and, 601–602, 602f, 603f K-fold cross validation, 533 King mackerel. See Scomberomorus cavalla Klamath River, transplant experiments of salmon into, 26 Kohonen network. See also Self-organizing map (SOM) networks, 555, 562–565, 563f Koljonen, M.L., 295

L Lactate dehydrogenase (LDH), allozyme analysis and, 296 Lake trout. See Salvelinus namaycush Landmark-based morphometric identiﬁcation, 153–172, 667, 668f adaptive traits and, 167–168 applications for, 169 case studies in, 153–154, 154f choice of characters to note in, 156–160, 157f criteria for, 157–158 deformations of, 158–159 development of, 153–154 distances between, measurement of, 158–159, 159f, 159f functional morphology and hypotheses regarding, 168–169 geometric analysis in, 158–159, 163f interpretation of differences in, 164–169 Limanda ferrunginia (yellowtail ﬂounder), 157–158, 157f linear distance analysis in, 158–159, 160f measuring devices for, 156 methodological protocols in, 155–163

696 Landmark-based morphometric identiﬁcation (continued) Oncorhynchus gorbushcha (pink salmon), 167 Oncorhynchus keta (chum salmon), 167 Oncorhynchus kisutch (coho salmon), case study in, 165–166 Oncorhynchus sp. (salmon) case studies in, 164–169 Oncorhynchus tshawytscha (chinook salmon), 167 phenotypic identiﬁcation using, 154–155, 169 pooled group PCA in, 161–163 population structures and, 169 principal components analysis (PCA) in, 160–163, 162f Salmo salar (Atlantic salmon), case study in, 164–165 Salmo trutta (brown trout), 167 Salvelinus alpinus (arctic char), 168 sampling in, 155–156 signiﬁcance (statistical) differences in, 162–163, 163f statistical analysis in, 160–163 Larval stage, 91, 92, 105–106. See also Planktonic transport Anguilla japonica (Japanese eel), distribution of, 100–101, 100f Clupea harengus (Atlantic herring), distribution of, 97–99, 98f ecosystem contribution of, 105–106, 106t Homerus americanus (American lobster), 102 meristic traits and, 199 morphometric traits of, 96–97 Laser ablation ICPMS (LA-ICPMS), otolith analysis, 235, 240, 241 Lates calcarifer (barramundi) calciﬁed structure texture, spacing patterns and, 191, 668 otolith and scale analysis in, 668 Learning algorithms for artiﬁcial neural networks (ANNs), 555 Lepomis gibbosus (pumpkinseed sunﬁsh), 58, 64 Lepomis macrochirus (bluegill sunﬁsh), 60, 65 Leptocephalus brevirostrum (eels), 32 Lethrinus miniatus, growth rate curves for, 129f Life cycle and parent stream theory, 29

Index Life cycles of common ﬁsh parasites, 219t Life history traits, 59–70, 71, 119–150. See also Early life stages (ELSs) in identiﬁcation age at maturity as, 64–65 age, growth, mortality data in, 126–131 applications for, 142–144 deﬁnition of, 119–120 differences in, 120 distributioin and abundance data in, 121–126, 122f early life stages (ELSs) and, 96–97, 104–105 environmental inﬂuences on, 59–60 factors affecting, 143–144 genetic inﬂuences on, 61–62, 120 geostatistics and, 104–105 gonadosomatic index (GSI) as, 66 growth rate vs. mortality ratio (K/M ratio), 60 life history traits and, 46 management using, 143–144 mixing of stocks and, 140–142 natural selection and, 59, 63 norms of reaction and, 67–70 parameters of, 120–137 phenotyping using, 143 plasticity of phenotypes and, 67–70 recruitment, 135–137, 136f reproduction, spawning, maturity as, 59–70, 131–135 reproductive effort, fecundity, egg size as, 65–67 spatial population models and, 121 spatial stability/variation in, 140–142, 141f temporal stability/variation in, 137–140, 139f–140f variation among populations and, 62–70 Likelihood formulation, maximum likelihood estimation (MLE) and, 572–574 Limanda ferruginea (yellowtail ﬂounder) landmark morphometry for, 157–158, 157f, 159f life history traits and, changes in, 137–138, 139f stock identiﬁcation of, 104–105 Limulus polyphemus (horseshoe crab) mitochondrial DNA and, 323, 325 random ampliﬁed polymorphic DNA (RAPD) and, 380

Index Linear discriminant analysis (LDA), 501–504, 503f, 510, 519, 520, 521, 527–529, 530, 531, 535, 536 Bayesian approach to LDA assumptions in, 539–541 maximum likelihood for LDA assumptions in, 538–539 Liu, Zhanjiang (John), 389 Lobster. See Homerus americanus Logistic discriminant analysis (LGA), 519, 521, 537–538 bootstrap sampling and, 548 maximum likelihood for LGA assumptions in, 541 Logistic regression (LG), 504–506, 505f, 512, 531–532 Logistic theory discrimination, 529–532 Loop spaghetti tags, 416 Lund, Roar A., 659

M MacKenzie, K., 211 Mackerel. See Rastrelliger kanguarat; Scomber scombrus; Scomberomorus sp.; Trachurus trachurus Macrobrachium borelli (prawn) random ampliﬁed polymorphic DNA (RAPD) and, 374 Magnetic tags, 416, 417 Magnuson Fishery Conservation and Management Act of 1976, 9 Magoulas, Antonios, 311 Mahalanobis distance, in sampling size, 475 Makaira nigricans (blue marlin) mitochondrial DNA and, 324 Mallotus villosus (capelin), 59 sampling and mixed stock analysis (MSA) in, 484 Management of stock, 7, 417–418, 447–448, 518, 631–658 accessible stock deﬁned for, 634 Advisory Committee for Fisheries Management (ACFM) in, 640 aggregate abundance-based management (AABM) in, 618–619 BASBLACK project and, 637 biological stock deﬁned for, 634 biological variations and, 640–643, 642f Clupea harengus (Atlantic herring), 637–640, 638f

697 deepwater stocks of North Atlantic and, 634–637 difﬁculties in, 653–654 DNA studies in, 637 early life stages (ELSs) and, 89–90 environmental inﬂuences on, 632 FAIR project and, 637 ﬁshery stock deﬁned for, 633–634 Gadus sp. (Artic cod), 640–643, 642f Gadus sp. (Baltic cod), 643–645, 644f genetic variability and, Sebastes sp. (redﬁsh), 632, 633, 645–650 harvest stock deﬁned for, 633 harvesting and, 633 holistic approach to, Trachurus trachurus (horse mackerel), 650–652, 651f individual stock-based manaegment (ISBM) in, 619 International Baltic Sea Fisheries Commission (IBSFC) in, 644–645 International Council for the Exploration of the Sea (ICES) and, 635–637 maximum likelihood estimation (MLE) and, 571–572 merged stock and sampling in, 637–640, 638f methods for, 632 mixed stock analysis (MSA) and, 468, 490–492 otolith analysis in, 652–653, 653f, 654f otolith thermal marking and, 458–461 Pollachius virens (saithe), 633 principles of, 654–655 Sebastes sp. (redﬁsh), 645–650, 652–653, 653f, 654f spatial and temporal instability of stock and, 632, 633 spawning stock biomass (SSB) in, 641– 642 stock identiﬁcation in, 631–658 tagging and, 417–418, 435–437, 447–448 targeted stock deﬁned for, 633–634 threatened and endangered species conservation and, 609–629 Total Allowable Catches (TAC) allocation in, 634, 639–640, 641, 653 underutilized information and, Baltic cod example, 643–645, 644f MAP principle, artiﬁcial neural networks (ANNs) and, 545

698 MARK software, movement estimation from tagging data and, 596 Markers, 417 Marking experiments, 24–25, 417 Markov chain Monte Carlo (MCMC) algorithm, 540, 544, 548–549 Mark-recapture programs, 12, 28, 418, 421–429. See also Tagging electronic tracking tags and, 442 mass marking. See Otolith thermal marking movement estimation from tag data in, 591–606 Marlin. See Makaira nigricans Mass marking. See Otolith thermal marking Masuda, Michele, 517 Maternal inheritance (matriarchal phylogeny), mitochondrial DNA and, 312 Maturity, age at, 64–65 Maximum likelihood estimation (MLE), 300–301, 302, 531, 532–536, 571–589 applications for, 577–587, 579t, 580t, 587–588 blind test of standard model in, 584, 586t curvature methods for, 573, 573f expectation maximization (EM) algorithm, 576 extended likelihood model for, 574–576 extended model utility in, 584–587 likelihood formulation for, 572–574 Monte Carlo simulation and, 577 numerical computation of, 576–577 Oncorhynchus tshawytscha (chinook salmon), 578–579, 584 population characteristics and, 573–574 proportion simplex sampling algorithm for, 581–584, 582f, 583f, 585f Salmo salar (Atlantic salmon), 577–578 sampling for, 579–584 utility of models using, 577–587 Measurement tools calciﬁed structure texture, spacing patterns and, 186 landmark-based morphometric identiﬁcation and, 156 meristic traits and, 202 otolith, 235 sampling and mixed stock analysis (MSA) in, 482 Medaka, chromosome morphology and, 289

Index Melanogrammus aegleﬁnus (haddock) distribution and abundance data, 122f growth rate and, 129f life history traits and, changes in, 137–138 microsatellites and, 358 recruitment data on, 136f sampling and mixed stock analysis (MSA) in, 484 stock identiﬁcation of, 104–105 Member-vagrant hypothesis, 30, 90–91 early life stages (ELSs) and, 94, 110 Mendidia menidia (Atlantic silverside), 48, 61 MEP-1 enzyme analysis, fatty acid proﬁle and, 262 Meristic traits, 51–54, 197–207, 198f analysis and interpretation of, 197 applications for, 204 counts used in, accuracy of, 199 deﬁnition of, 197 environmental inﬂuences on, 51–52, 197 ﬁshes studied using, 198–199 ﬁtness and, 53–54 genetic inﬂuences on, 52–53, 197 history and development of, 198–199 Jordan’s rule phenomenon and, 197 larval stage and, 199 life history traits and, 46 measurement tools for, 202 Morone saxatilis (striped bass), 199, 203–204 multivariate analysis and, 201–203 Oncorhynchus gairdneri (rainbow trout), 200 Oncorhynchus kisutch (coho salmon), 199 Scomberomorus (Spanish mackerel), 200f serial incrementation of most body parts used in, 199 signiﬁcant (statisical) difference in, 199–203, 200f, 201t standardization of technique in, 199 stock identiﬁcation through, 199–203 substock presence and, complication of, 203 variation among populations and, 53–54 Merlangius merlangus (whiting), nuclear DNA and, 355 Merluccius bilinearis (silver hake), life history traits and, changes in, 137–138 Merlucidae, microsatellites and, 350 Metapopulations, 99, 108–110 spatial models of, 108–110 Methodology for stock identiﬁcation, 2

Index Micromesistius poutassou (blue whiting) planktonic transport in, modeling and measurement of, 107 Microsatellite DNA, 13, 46, 47–51, 335–336, 338, 344–351, 345f, 362, 393–395 acipenseridae, 349 advantages to use of, 344–346 ampliﬁed fragment length polymorphism (AFLP) and, 393–395, 397 artiﬁcial neural networks (ANNs) and analysis, 566–567 clupeidae, 349 deﬁnition of, 344 dinucleotide (AC) repeats in, 344, 345f Gadus morhua (Atlantic cod) and, 355–359 Hardy–Weinberg equilibrium in, 346 inﬁnite allele model (IAM) in, 348 isolation of, 349–351, 349 Melanogrammus aegleﬁnus, 358 polymerase chain reaction (PCR) in, 347–351 polymorphisms in, predictability of, 346–347, 348–349 Pseudopleuronectes americanus (winter ﬂounder), 566–567 random ampliﬁed polymorphic DNA (RAPD) and, 380, 382 recombination and transcription in, 347 repeat motifs in, 347 sampling and mixed stock analysis (MSA) in, 483, 487–488 stepwise mutation model (SMM) in, 348 tetranucleotide (GATA) repeats in, 344, 345f threatened and endangered species conservation and, 620–622, 624–625 Trachurus trachurus (horse mackerel), 651 trinucleotide (ACC) repeats in, 344, 345f variations in, 346 Microstomus paciﬁcus (Dover sole) ELS and genetic population structure of, 96 Migration and unit stock concept, 17–44 alternative pathways in, 39 biochemical markers in, 30 biodiversity and, 37 circuits of, for salmon, 28 cod hatchery enhancement and, 19–21, 20t contingent concept in, 35–36 eel problem in, 31–37 evolutionary concepts of, 18 homing mechanisms in (geolocation), 25

699 hydrographic containment and, 29 ICES protocols and, 17–18 member-vagrant hypothesis in, 30, 90–91 modern migration theory and, 18, 21 natural selection and, 31 natural tags in identiﬁcation of, 22–24, 23f, 38 ocean studies and, 27–28 otolith thermal marking and, 458–459 panmixia theory in, 18 parent stream theory in, directed migration vs. local wandering, 21–24, 27, 28–30 population structures and, 18–19, 19f, 30 reproduction and, 31 sampling and mixed stock analysis (MSA) in, migratory ﬁshes and, 477 tagging and marking experiments and, 24–25 transplant experiments and, 26 triangle concept of, 29–30, 30f year-class phenomenon in, 26–27 Minisatellite DNA. See also Microsatellite DNA, 335–336 Gadus morhua (Atlantic cod) and, 355 nuclear DNA and, 342–344 sampling and mixed stock analysis (MSA) in, 483, 487–488 Mitochondrial DNA, 46, 311–330, 649 Acipenser oxyrinchus (Atlantic sturgeon), 486–487 Alosa sapdiissima (American shad), 481 Anguilla anguilla (European eel), 324 Anguilla rostrata (American eel), 323 audioradiography development of ﬁlms for, 318 case studies in, review of, 323–325 cautions in use of, 327 coding genes and, 311–312 data processing for, 319–320, 319f, 322 eel studies, 36 Engraulis encrasicolus (European anchovy), 326–327 enriched extraction and, 316–318, 317f evolution of, 312, 315f Gadus morhua (Atlantic cod) and, 323, 352–353 gene diversity indexes and, 320 genetic gaps identiﬁed by, 320 haplotype identiﬁcation through, 314 harbor porpoise, 623

700 Mitochondrial DNA (continued) Hardy–Weinberg equilibrium, 327, 480–481 hybridization and, 318 intraspecies analysis using, 313 Katsuwonus pelamis (skipjack), 323 Limulus polyphemus, 323, 325 Makaira nigricans (blue marlin), 324 maternal inheritance (matriarchal phylogeny) of, 312 mitotypes and, 313, 320, 326 molecular characteristics of, 311–312 mutations identiﬁed through, 314 nuclear DNA and vs., 333–334 Oncorhynchus gorbuscha (pink salmon), 624 Oncorhynchus keta (chum salmon), 624 Oncorhynchus mykiss (rainbow/steelhead trout), 623–624 Opsanus tau, 323 Pagrus pagrus (red porgy), 324–325 phylads or phylogroup identiﬁcation using, 326 phylogenetic network identiﬁcation using, 314 phylogeographic relationships formed from, 319, 320 polymerase chain reaction (PCR) and, 313, 321–322 Polyprion americanus (wreckﬁsh), 324 probes used in, 316 Pseudopentaceros wheeleri (armorhead), 323 restriction analysis of, restriction endonucleases (REs), 313–320, 315f, 322 restriction diagnosis and electrophoresis in, 317–318 restriction fragment length polymorphism (RFLP) analysis in, 314, 324, 352 ribosomal RNA (rRNA) and, 311, 321 Salvelinus fontinalis (brook trout), 481 sampling and mixed stock analysis (MSA) in, 480–481, 486–487 Sciaenops ocellatus (red drum), 487 shared bands identiﬁed through, 314 site gains and, 314 size of, maximum in Placopecten magellanicus, 311 Southern blot analysis and, 316, 318 temporal stability of, 486–487 threatened and endangered species conservation and, 623–624

Index Thunnus alalunga (albacore tuna), 324 Thunnus albacares (yellowﬁn tuna), 324, 481 Thunnus obesus (bigeye tuna), 324 total DNA extraction and, 316–318, 317f Trachurus trachurus (horse mackerel), 651 transfer of, 318 transfer RNA (tRNA) and, 311, 321 transmission of, 312 variation, genetic discontinuity and, 325–326 Xiphias gladius (swordﬁsh), 324, 325 Mitotypes, mitochondrial DNA and, 313, 320, 326 Mixed stock analysis (MSA), 467–498 Acipenser oxyrinchus (Atlantic sturgeon), 486–487 age classes vs. stocks in, 484–485 allele frequency and, 474, 486 allometric and age relations within, 483–485 allozyme analysis and, 296, 300–301, 487–488 applications for, 467–469 body size of individuals and, 484 characterizing differences in, 468 classiﬁcation algorithms, number of features necessary for, 479–481 classiﬁcation models in, 468 classiﬁcation of stocks in, 518–519, 520 Clupea harengus (Atlantic herring), 484 differences among stocks in, statistical signiﬁcance of, 482–483 discriminant analysis (DA) in, 517–552 experimental design and sampling strategies for, 467–498 factors affecting, 490–492 ﬁnite mixture distributions (FMD) in, 509–510, 509f Gadus morhua (Atlantic cod), 472 genetic mixture analysis (GMA) and, 621 hatchery releases and, 467–469 Homerus americanus (American lobster), 485 isoloci in, 487–488 life history traits and, 140–142 Mallotus villosus (capelin), 484 management and, 468, 490–492 maximum likelihood estimation (MLE) and, 571–589 measurement tools used in, 482 Melanogrammus aegleﬁnus (haddock), 484 microsatellite DNA and, 483, 487–488 minisatellite DNA and, 483, 487–488

Index mixture models in, 469 Oncorhynchus gorbuscha (pink salmon), 470 Oncorhynchus nerka (sockeye salmon), 541–546, 542f, 543–544t, 545f Oncorhynchus tshawytscha (chinook salmon), 486, 487 otolith analysis in, 239–240 posterior probabilities in, classiﬁcation of mixtures, 523–525 power to detect stock differences and, 488–489 prior knowledge required for proper application of statistical analysis to, 511–513, 520–521 reliability of features for stock delineation in, 481–488 sampling of, 468 Acipenser oxyrinchus (Atlantic sturgeon), 486–487 age classes vs. stocks in, 484–485 all stocks in mixture included in baseline, 469–470 allele detection/identiﬁation in, 476 allele frequency and, 474, 486 allelic designations and scoring ambiguities in, 487–488 allometric and age relations within, 483–485 allozyme analysis and, 487–488 Alosa sapdiissima (American shad), 481 annual variations in stocks and, 477 body size of individuals and, 484 Chi-square tests in, 474 classiﬁcation algorithms, number of features necessary for, 479–481 Clupea harengus (Atlantic herring), 484 composite stock identiﬁcation in, 472–473 differences among stocks in, statistical signiﬁcance of, 482–483 factors affecting, 490–492 goodness-of-ﬁt testing in, 474 Hardy–Weinberg equilibrium, 474, 480–481 Homerus americanus (American lobster), 485 Mahalanobis distance, in sampling size, 475 Mallotus villosus (capelin), 484 management and, 490–492 measurement tools used in, 482

701 Melanogrammus aegleﬁnus (haddock), 484 microsatellite DNA and, 483, 487–488 migratory ﬁshes and, 477 minisatellite DNA and, 483, 487–488 mitochondrial DNA and, 480–481, 486–487 number of source stock included in, 471–472, 471f Oncorhychuns kitsutch (coho salmon), 476 Oncorhynchus tshawytscha (chinook salmon), 479, 486, 487 power to detect stock differences and, 488–489 reliability of features for stock delineation in, 481–488 Salvelinus fontinalis (brook trout), 481 Sciaenops ocellatus (red drum), 487 Scomber scombrus (Atlantic mackerel), 484 Scomberomorus cavalla (king mackerel), 485 sex effects and, 485 size of sample in, 473–476, 475f, 477–479, 478f source stocks comprising few breeders in, 470–471 statistical analysis in, 474, 482–483 strategy for, 476–477 substock inclusion/exclusion in, 472–473 temporal stability of features and, 485–487 Thunnus albacares (yellowﬁn tuna), 481 Sciaenops ocellatus (red drum), 487 Scomber scombrus (Atlantic mackerel), 484 Scomberomorus cavalla (king mackerel), 485 sex effects and, 485 source stock sampling strategy for, 469–473, 469 statistical analysis in, 500, 517–552 statistical package for analysis of mixtures (SPAM) in, 621, 624 temporal stability of features and, 485–487 Mixture discriminant analysis (MDA), 519 Mixture models, mixed stock analysis (MSA) and, 469 Modeling Arnason–Schwarz multistate models in, 592, 593–594 artiﬁcial neural networks (ANNs) and, 554 classiﬁcation models in, 468

702 Modeling (continued) continuous time/space models in, 592, 600–602 discrete stock/discrete time models for, 591–600 elliptical Fourier analysis in, 177 escapee (aquaculture) identiﬁcation and, 668–669 extended likelihood model for, 574–576 Fourier series analysis in, 173, 175, 176–178, 177f, 178f, 187 Generalized Additive Models (GAMs) in, 121, 123f, 123 Generalized Linear Model (GLM) in, 130 Geographic Information Systems (GIS) in, 121 harvest models in, 595–596 inﬁnite allele model (IAM) in, 348 MARK and M-SURGE software for, 596 mixed stock analysis (MSA) and, 468–469 mixture models in, 469 movement estimation from tagging data and, 591–600 multistate models for, 592, 593–594 multivariate analysis in, 178–179 outline-based morphometric identiﬁcation and, 174–178, 175f planktonic transport, 106–108 polynomial functions in, 176 soft independent modeling of class analogy (SIMCA) analysis in, 251–252 spatial population models in, 108–110, 121 stepwise mutation model (SMM) in, 348 Modern migration theory, 18, 21 Molecular basis of genetic variation, 392–393 Molecular cytogenic methods, chromosome morphology and, 287–289 Molecular markers, 47–51 Monitoring methods, 660–676 Monitoring stations used with electronic tracking tags, 444 Monte Carlo simulation, maximum likelihood estimation (MLE) and, 577 Morone saxatilis (striped bass), 59 calciﬁed structure texture, spacing patterns and, 190–191, 191f fatty acid proﬁle and, 255–256, 257f meristic traits and, 199, 203–204 nuclear DNA and, 339 otolith analysis in, 668

Index outline-based morphometric identiﬁcation and, 180 random ampliﬁed polymorphic DNA (RAPD) and, 374 Moronidae, microsatellites and, 350 Morphometric traits, 54–58, 71 adhesions of abdominal cavity in, post vaccination, 673–674, 674f carotenoids and pink ﬂesh color in, 674–675 choice of characters to note in, 156–160, 157f early life stages (ELSs) and, 96–97 environmental inﬂuences on, 54–56 escapee (aquaculture) identiﬁcation and, 660–667 extinction vs., 57 ﬁn measurement in, 666–667 ﬁn ray defects in, 661–665, 662f, 663f, 664f, 666t ﬁn tissue defects in, 660–661, 661f food sources and, 55–56 genetic inﬂuences on, 56 gill cover shortening in, 665, 665f gonadosomatic index (GSI) in, 66 landmark method in. See Landmark morphometric identiﬁcation landmarks in, 153, 667, 668f larval stage, 96–97 life history traits and, 46 measuring devices for, 156 multivariate analyses and, 58 outline method in. See Outline-based morphometric identiﬁcation pigmentation variations in, 665–666 plasticity of phenotypes and, 55–58 replacement scales in, 672–673, 673f sampling in, 155–156 sea winter band and, 669–671 smolt size and age, 669 summer checks in, 671–672 Trachurus trachurus (horse mackerel), 652 transition zone, salt-to-freshwater, 669, 670f, 671f undershot jaw in, 665 variation among populations and, 56–58 Morphotypes, 49 Mortality rates, 7, 126–131 Movement estimation from tagging data, 591–606 applications for, 602–604 Arnason–Schwarz multistate models in, 592, 593–594

703

Index Bayesian theorem and, 596 continuous time/space models in, 592, 600–602 example of, Katsuwonus palamis (skipjack tuna), 601–602, discrete stock/discrete time models for, 591–600 Arnason–Schwarz multistate models in, 592, 593–594 Bayesian theorem and, 596 example of, Hippoglossus stenolepis (Paciﬁc halibut), 597 harvest models in, 595–596 model ﬁtting in, 596–597 software for (MARK, M-SURGE), 596 stratiﬁcation of stock in, 592–593 future uses of, 602–604 global positioning satellite (GPS) and, 602, 604 harvest models in, 595–596 Hippoglossus stenolepis (Paciﬁc halibut), 597–600, 598f, 599t model ﬁtting in, 596–597 multistate models for, 592 software for (MARK, M-SURGE), 596 tagging program selection and, 597 M-SURGE software, movement estimation from tagging data and, 596 Mullet. See Mullus barbarus Mullus barbarus (red mullet), random ampliﬁed polymorphic DNA (RAPD) and, 374, 380, 381t Multicolor FISH (M-FISH), chromosome morphology and, 288–289 Multilayer feedforward network (MLFN), 555, 560–562, 561f Multimeric enzymes, allozyme analysis and, 298 Multisource data, artiﬁcial neural networks (ANNs) and, 559 Multistate models, movement estimation from tagging data and, 592, 593–594 Multivariate analysis meristic traits and, 201–203 morphometric traits and, 58 outline-based morphometric identiﬁcation and, 178–179 parasites as biological tags and, 215–216 Multivariate logistic compound (MLG), 544 Multivariate normal mixture (MVN), 544

Mussels, outline-based morphometric identiﬁcation and, 180 Mutations, 314 inﬁnite allele model (IAM) in, 348 mitochondrial DNA and, 314 molecular basis of genetic variation and, 392–393 stepwise mutation model (SMM) in, 348 Mutilation or markers, tagging and, 417 Mystus nemurus (river catﬁsh) ampliﬁed fragment length polymorphism (AFLP) and, 407 chromosome morphology and, 289 Myxosporeal parasites as biological markers. See also Parasites as biological tags, 214, 217

N Natural markers, 22–24, 23f, 38 fatty acid proﬁles as, 247–269 otoliths as, 227–245. See also Otoliths parasites as, 211–226. See also Parasites as biological tags planktonic transport and, 107–108 salmon, 22–24, 23f Natural selection, 31, 46, 47–51, 70, 71 age at maturity as, 64–65 life history traits and, 59, 63 norms of reaction and, 67–70 Negative results (Type 1 errors) in stock identiﬁcation methods, 12–13 Nematodal parasites as biological markers. See also Parasites as biological tags, 217 Neural networks. See Artiﬁcial neural networks New Zealand barracuda. See Thrysites atun Normal theory discrimination, 527–529 Norms of reaction, life history traits and, 67–70 North Atlantic BASBLACK project and, 637 deepwater stocks of, and management of, 634–637 FAIR project and, 637 ICES ﬁsheries convention areas, 635–637, 636f North East Atlantic Fisheries Commision (NEAFC), 427 North Paciﬁc Anadromous Fish Commission (NPAFC), 622 otolith thermal marking and, 461

704 North Paciﬁc Fisheries Commission, salmon migration study, 28 Northern pike. See Esox lucius Norway, cod hatchery enhancement program, 20–21, 20t NTSYS-PC AFLP data analysis software, 406 Nuclear DNA advantages of, and applications for, 333–334 ampliﬁed fragment length polymorphism (AFLP), 337, 341, 342 Anser caerulescens caerulescens (snow goose), 340 applications for, 359–361 cloning strategies for, 333–334 Crassotrea virginica (American oyster), 341 DNA ﬁngerprinting and, 342–344 Drosophila sp. studies in, 335 exons and introns in, 334–336, 335f Gadus morhua (Atlantic cod), 341, 351–359, 362 Merlangius merlangus (whiting), 355 methodology of analysis of, 336–338 microsatellite (minisatellite) DNA and, 335–336, 338, 344–351, 345f, 355–359, 362 Acipenser oxyrinchus (Atlantic sturgeon), 349 acipenseridae, 349 advantages to use of, 344–346 clupeidae, 349 coregonidae, 350 deﬁnition of, 344 dinucleotide (AC) repeats in, 344, 345f DNA ﬁngerprinting and, 342–344 gadidae, 350 Hardy–Weinberg equilibrium in, 346 inﬁnite allele model (IAM) in, 348 isolation of, 349–351 Melanogrammus aegleﬁnus, 358 merlucidae, 350 methodology for analysis of, 347–351 minisatellites and, 355, 342–344 moronidae, 350 pleuronectidae, 350 polymerase chain reaction (PCR) in, 347–351 polymorphisms in, predictability of, 346–347, 348–349 recombination and transcription in, 347 repeat motifs in, 347 salmondidae, 350

Index scombridae, 350 scorpaenidae, 350 stepwise mutation model (SMM) in, 348 studied species, 343 tetranucleotide (GATA) repeats in, 344, 345f trinucleotide (ACC) repeats in, 344, 345f variations in, 346 mitochondrial DNA vs. 333–334 Morone saxatilis (striped bass), 339 noncoding, 340–341, 354–355 nuclear genome in, 334–336 Oncorhynchus nerka (sockeye salmon), 343–344 Oncorhynchus tshawytscha (chinook salmon), 343 Oncorhynchus keta (chum salmon), 343 polymerase chain reaction (PCR) in, 336–337, 339, 340, 342–343, 347–351, 360, 361 primers in, 337 pyrosequencing in, 337–338 random ampliﬁed polymorphic DNA (RAPDs) in, 337, 341, 342 repetitive, 341–351 restriction fragment length polymorphism (RFLP) and, 337–338, 341, 360 sampling for, 336–337 single nucleotide polymorphisms (SNPs) and, 334, 338–340, 354–355, 360, 625 single-copy coding, 338–340, 353–354 single-copy noncoding, 340–341 tissues used for, 336 variable number of tandem repeats (VNTRs) in, 341 variations in, 336 Nuclear DNA, single-copy and repetitive sequence markers, 331–369, 331 Nucleolar organizer regions (NORs), chromosome morphology and, 274, 281, 282f, 285–286, 290 Nucleus, in genetics, 391–392 Null hypotheses, 12–13 congruence of results in, 13–14 Numeric functions (continuous), artiﬁcial neural networks (ANNs) and, 554–555

O Ocean studies of migration, 27–28 Oncorhynchus sp., 56 allozyme analysis and, 296 artiﬁcial neural networks (ANNs) and, 519

Index chromosome morphology and, 275 circuits of oceanic migration in, 28 fatty acid proﬁle and, 256–257, 258t, 259f ﬁshing politics and, U.S./Canada salmon wars, 617–619 hatchery release programs for, 24 High Seas Salmon Program, 622–624 homing mechanisms in, 38 hydrographic containment and, 29 International North Paciﬁc Fisheries Commision (INPC), 622 landmark-based morphometric identiﬁcation and, case studies in, 164–169 mark-recapture programs for, 28 natural tags in identiﬁcation of, 22–24, 23f North Paciﬁc Anadromous Fish Commission (NPAFC), 622 North Paciﬁc Fisheries Commission migration study of, 28 O. gairdneri, 200 O. keta (chum salmon), 57, 58 allozyme analysis and, 298, 305t chromosome morphology and, 277 landmark-based morphometric identiﬁcation and, 167 mitochondrial DNA and, 624 nuclear DNA and, 343 otolith thermal marking and, 449, 452f threatened and endangered species conservation and, 617–619, 624 O. kitsutch (coho salmon), 56, 61, 68 calciﬁed structure texture, spacing patterns and, 190 landmark-based morphometric identiﬁcation and, case study in, 165–166 meristic traits and, 199 sampling and mixed stock analysis (MSA) in, 476 threatened and endangered species conservation and, 617–619 O. gorbuscha (pink salmon), 57 allozyme analysis and, 298, 304, 305t chromosome morphology and, 277 landmark-based morphometric identiﬁcation and, 167 mitochondrial DNA and, 624 otolith thermal marking and, 457, 458 sampling and mixed stock analysis (MSA) in, 470 supportive breeding programs for, 619–620

705 threatened and endangered species conservation and, 617–620, 624 O. lapillus (coho salmon) otolith thermal marking and, 449, 450f, 451f, 453f, 457 O. mykiss (rainbow trout), 60 calciﬁed structure texture, spacing patterns and, 190, 191 chromosome morphology and, 283, 289, 290f chromosome morphology and, polyploidy, 275f, 276f, 277, 278f farm raised, 659–660 fatty acid proﬁle and, 266 tagging and, 428–429 microsatellite DNA and, 623–624 otolith thermal marking and, 448 otolith analysis in, 237 O. nerka (sockeye salmon), 57, 69 allozymes of, 305t artiﬁcial neural networks (ANNs) and, 519, 521, 522f–523f, 541–546, 542f, 543–544t, 545f calciﬁed structure texture, spacing patterns and, 188f, 189 nuclear DNA and, 343–344 otolith thermal marking and, 458 threatened and endangered species conservation and, 615–619 O. tshawytscha (chinook salmon), 56, 61 allozyme analysis and, 304, 305t landmark-based morphometric identiﬁcation and, case study in, 167 maximum likelihood estimation (MLE) and, 578–579, 584 nuclear DNA and, 343 otolith thermal marking and, 449, 450f, 455, 456f, 457, 459 sampling and mixed stock analysis (MSA) in, 479, 486, 487 threatened and endangered species conservation and, 610–614, 611f, 614f, 615–619 otolith thermal marking and, 448 Paciﬁc Salmon Treaties, 618–619 parent stream theory in, directed migration vs. local wandering, 21–24, 27, 28–30 tagging and marking experiments on, 24–25 transplant experiments and, 26 year-class phenomenon in, 26–27

706 Opsanus tau, mitochondrial DNA and, 323 Optical density proﬁles, calciﬁed structure texture, spacing patterns and, 187, 188f Orange roughy. See Hoplostethus atlanticus Otolith analysis, 227–245 Alosa sapidissima (shad), 228f analysis of, 233–234 Atlantic salmon (Salmo salar), 241 atomic absorption spectrometry (AAS) in, 234 beam-based assays for, 234–235 case studies using, 239–241 characteristic and reproducible markers for each group assessed by, 232–233 characterization of all group mixtures for, 233 composition of, as natural marker, 227, 230–232, 231f Cynoscion regalis (weakﬁsh), 240 daily incrementation in, 36 early life stages (ELSs) and, 94–95 elemental ﬁngerprint of, 228–229, 228f, 241–242 energy-dispersive electron microprobes (ED-EM) for, 235 environmental inﬂuences and geographic variations on, 46, 229, 233–234, 241–242 escapee (aquaculture) identiﬁcation and, 667–673, 675 FAIR project and, 637 Gadus morhua (Atlantic cod), 231f, 232f, 233, 239–240 genetic inﬂuences on, 229 homing mechanisms and, 240 incrementation rate of, 36 inducement of marks in, methods for, 448–456, 448 inductively couple plasma mass spectrometry (ICPMS) in, 234, 237, 239, 240 inductively coupled plasma atomic emission spectroscopy (ICP-AES) in, 234 isotope dilution ICPMS (ID-ICPMS) in, 234, 237 isotopic analysis of, 227–228 laser ablation ICPMS (LA-ICPMS) in, 235, 240, 241 Lates calcarifer (barramundi), 668 measurement tools for, 235 microsampling techniques for, 235

Index mixing of stocks and, 239–240 Morone saxatilis (striped bass), 668 Oncorhynchus mykiss (steelhead trout), 237 outline-based morphometric identiﬁcation and, 173, 175–183 Pagrus auratus (snapper), 240–241 patterns in, organizing information from, 454–456 population composition analysis using, 240–241 protocol for, 236–237 proton induced X ray emission (PIXE) in, 235 river of origin determination from, 241 sample preparation and quality control in, 236–237 sampling and assays for, 234–239 sampling and mixed stock analysis (MSA) in, 486 Sebastes sp. (redﬁsh), 652–653, 653f, 654f short-term stability in, 231, 232f stability of, 233 statistical analysis in, 237–239, 238f temperature effects on, in thermal marking, 448–454 thermal ionization mass spectrometer (TIMS) in, 241 thermal marking of, 447–463, 450f, 451f, 452f, 453f, 456f applications for, 455, 457–461 drawbacks and limitations of, 460 early experiments in, 449 errors in mark recovery and, 456–457 hatchery use of, 455, 458–459, 460–461 inducement of marks in, methods for, 448–456 kappa index in, 457 management and, 458–461 method for, 447–448 migration tracking and, 458–459 Oncorhynchus gorbushcha (pink salmon), 457, 458 Oncorhynchus keta (chum salmon), 449, 452f Oncorhynchus lapillus (coho salmon), 449, 450f, 451f, 453f, 457 Oncorhynchus mykiss (rainbow/steelhead trout), 448 Oncorhynchus nerka (sockeye salmon), 458 Oncorhynchus spp. (Paciﬁc salmon), 448

707

Index Oncorhynchus tshawytscha (chinook salmon), 449, 450f, 455, 456f, 457, 459 organizing pattern information from, 454–456 RBr notation in, 455 Salmo clarki (cutthroat trout), 448 Salmo salar (Atlantic salmon), 448 Salmo trutta (brown trout), 448 Salvelinus namaycush (lake trout), 448 systems for, 455–456 temperature effects on otoliths and, 448–454 trace elements in, 227, 675 wavelength dispersive electron microprobes (WD-EM) for, 235 whole-otolith ﬁngerprinting in, 230–232, 231f, 234 Outline-based morphometric identiﬁcation, 173–183 alternative methods vs., 181 applications for, 181 biological signiﬁcance of, 179 case studies in, 180–181 elliptical Fourier analysis in, 177 Fourier series analysis in, 173, 175, 176–178, 177f, 178f image processing tools for, 173, 174 interpretation of data from, 179 methods for, 174–179 Morone saxatilis (striped bass), 180 multivariate analysis in, 178–179 mussels, 180 otolith shape and, 173, 175–183 polynomial functions in, 176 scale shape and, 173 Scomberomorus cavalla (king mackerel), 181 statistical model ﬁtting in, 174–178, 175f Theragra chalcogramma (walleye pollock), 180 Oyster. See Crassotrea virginica

P Paciﬁc Paciﬁc Paciﬁc Paciﬁc Paciﬁc Paciﬁc

cod. See Gadus macrocephalus halibut. See Hippoglossus stenolepis herring. See Clupea pallasi Salmon Commission, 519 Salmon Treaties, 618–619 salmon. See Oncorhynchus spp

Pagrus auratus (snapper) otolith analysis, 240–241 Pagrus pagrus (red porgy) mitochondrial DNA and, 324–325 Pan1 alleles, Gadus morhua (Atlantic cod) and, 354, 357, 361 Panmixia theory, 18 Paralichthys dentatus (summer ﬂounder) ELS and phenotypic traits of, 96–97 Parasites as biological tags, 226 acanthocephalans as, 217 advantages and limitations to, 212–213 Cestode pleorocercoid as, 217 cestodes as, 217 collection of hosts and parasites for, 216–217 criteria for use of, 212, 213–214 crustacea as, 214, 217, 218 development and history of, 211 Digenean metacercariae as, 217 discriminant analysis (DA) in, 215–216 ﬁxation and preservation of parasites for, 217 general principles of, 211–212 genetics of parasites and, 218 helminthes as, 214, 217, 218 identiﬁcation of parasites in, 217–218 interpretation of results from, 218–221 life cycles of common ﬁsh parasites in, 219t methodolgy for, 214–216 multivariate analysis in, 215–216 myxosporeans as, 214, 217 nematodes as, 217 procedures and methods ﬂowchart for, 220f protozoans as, 214, 217, 218 selection of parasites to use as, 213–214 statistical analysis in, 214–216 study of parasite characteristics for, 214 Trachurus trachurus (horse mackerel), 652 uses for data from, 216 Parent stream migration theory, 21–24, 27, 28–30 hydrographic containment and, 29 life cycle and, 29 population thinking and, 28–30 triangle concept of, 29–30, 30f Partial least square (PLS) analysis, fatty acid proﬁle and, 251 Pattern or structure of resources, 8 Pecten maximus (scallop), random ampliﬁed polymorphic DNA (RAPD) and, 375

708 Pella, Jerome, 517 Penaeus monodon (prawn) random ampliﬁed polymorphic DNA (RAPD) and, 375, 379 Penaeus stylirostris (blue shrimp) random ampliﬁed polymorphic DNA (RAPD) and, 375 Perca ﬂavescens (yellow perch), 67 Perca ﬂuviatilis (perch) random ampliﬁed polymorphic DNA (RAPD) and, 373, 381t Perch. See Perca ﬂavescens; P. ﬂuviatilis Petropavlocks ship, salmon ﬁshing, 624 Phage artiﬁcial chromosomes (PACs), 288 Phenetic trees, 405 Phenotypes, 90, 91–97, 92f diversity of, intraspeciﬁc patterns of, factors underlying, 47–51 early life stages (ELSs) and, 96–97 landmark-based morphometric identiﬁcation in, 154–155, 169 life history traits and, 143 modulation in, 49–51, 49f phenetic tree construction using data from, 405 plasticity of, 48–51, 55–58, 67–70 Phillips, Ruth B., 273 Philopatry. See Parent stream migration theory Phospholipids, fatty acid proﬁle and, 266 Photography, chromosome morphology and, 286–287 Phylads or phylogroup identiﬁcation using mtDNA, 326 PHYLIP AFLP data analysis software, 406 Phylogenetic networks, mitochondrial DNA and, 314 Phylogeographic relationships formed from mtDNA, 319, 320 Phylogroup identiﬁcation using mtDNA, 326 Pigmentation variations carotenoids and pink ﬂesh color in, 674–675 escapee (aquaculture) identiﬁcation and, 665–666 Placopecten magellanicus (scallop) mitochondrial DNA and, 311 random ampliﬁed polymorphic DNA (RAPD) and, 375 Plaice. See Hippoglossoides platessoides Plankton stage, 91, 92, 105–106 ecosystem contribution of, 105–106, 106t

Index Planktonic transport, 93–94, 106–108 early life stages (ELSs) and, 106–108 markers used in, 107–108 measurement of, 106–108 modeling of, 106–108 Plastic tube marker, 429f Plasticity of phenotypes, 48–51, 55–58, 67–70 Plectropomus leopardus (coral trout) electronic tracking tags and, 437 Pleuronectes platessa (plaice) fatty acid proﬁle and, 266 Pleuronectidae, microsatellites and, 350 Plug in vs. predictive classiﬁcation in, 526–527 Poecilia reticulata (guppy), 55, 64 Pogson, 48 Politics of ﬁshing ICES ﬁsheries convention areas, 635–637, 636f threatened and endangered species conservation and, U.S./Canada salmon wars, 617–619 Pollachius virens (saithe) management issues and, 633 Pollock. See Therogra chalcogramma Polymerase chain reaction (PCR) ampliﬁed fragment length polymorphism (AFLP) and, 389–390, 395–397, 399, 403–404 mitochondrial DNA and, 313, 321–322 nuclear DNA and, 336–337, 339, 340, 342–343, 347–351, 360, 361 random ampliﬁed polymorphic DNA (RAPD) and, 371, 376 Polymorphism, 393–395, 394f ampliﬁed fragment length polymorphism (AFLP) and, 393–395, 394f, 400, 401f microsatellites and, predictability of, 346–347, 348–349 random ampliﬁed polymorphic DNA (RAPD) and, 377–378 Polynomial discriminant analysis, 503 Polynomial functions, outline-based morphometric identiﬁcation and, 176 Polyploidy, chromosome morphology and, 274, 275f Polyprion americanus (wreckﬁsh) mitochondrial DNA and, 324 Pomatomus saltatrix (blueﬁsh) ELS and spawning discreteness in, 95 Pooled group PCA, 161–163

709

Index POPGENE AFLP data analysis software, 406 Population dynamics, 7 Population genomics, threatened and endangered species conservation and, 625–626 Population structures, 1 allele frequencies in, 299–300, 474, 486 allozyme analysis and, 295–296 ampliﬁed fragment length polymorphism (AFLP) and, 406–407 deepwater stock of North Atlantic, 635–637 early life stages (ELSs) and, 91–93, 93f, 95–96 eel problem in, 31–37 landmark-based morphometric identiﬁcation and, 169 life history traits and, 140–142, 141f maximum likelihood estimation (MLE) and, 573–574 migration and, 18–19, 19f, 30 molecular basis of genetic variation and, 392–393 nuclear DNA and, 333 random ampliﬁed polymorphic DNA (RAPD) and, 378 Trachurus trachurus (horse mackerel), 652 Populations, 518 biodiversity and, 37 cod hatchery enhancement and, 19–21, 20t contingent concept in, 35–36, 90, 108 deﬁned, 468 early life stages (ELSs) and, 91–93, 93f evolution and development of, 18 life history traits of, variations among, 62–70 member-vagrant hypothesis in, 30, 90–91, 94, 110 meristic traits of, variations among, 53–54 meta-, 99, 108–110 mixed stock analysis (MSA) and, 467–469 morphometric traits of, variations among, 56–58 natural selection and, 31 newly emerging, 12 norms of reaction and, 69–70 renewal efforts in, 37–38 sampling of, 468 spatial models of, 108–110 sub-, 12

unit stock concept in (see Unit stock concept), 17 year-class phenomenon in, 26–27 Porgy. See Pagrus pagrus Posterior probabilities, classiﬁcation of mixed stocks, 523–525 Prager, Michael H., 499 Prawn. See Macrobrachium borelli; Penaeus monodon Precision of measurement, 8 Presence fragments, random ampliﬁed polymorphic DNA (RAPD) and, 378–379 Primers, nuclear DNA and, 337 Principal component analysis (PCA), 160–163, 162f fatty acid proﬁle and, 250–251 Prior knowledge required for proper application of statistical analysis, 511–513, 520–521 Pritchard, A.L., 25 Probability, artiﬁcial neural networks (ANNs) and, 522–523 Probes used in mtDNA analysis, 316 Proportion simplex sampling algorithm, 581–584, 582f, 583f, 585f Protein allozymes, 46, 48–51 Protocols for stock identiﬁcation (ICES), 2, 17 Proton-induced X ray emission (PIXE), otolith, 235 Protozoal parasites as biological markers. See also Parasites as biological tags, 214, 217, 218 Pseudopentaceros wheeleri (armorhead) mitochondrial DNA and, 323 Pseudopleuronectes americanus (winter ﬂounder) artiﬁcial neural networks (ANNs) and, 566–567 microsatellite DNA and, 566–567 Pufferﬁsh chromosome morphology and, 289 random ampliﬁed polymorphic DNA (RAPD) and, 379 Pumpkinseed sunﬁsh. See Lepomis gibbosus Pyrosequencing, nuclear DNA and, 337–338

Q Q banding, chromosome morphology and, 283 Quadratic discriminant analysis (QDA), 501–504, 503f, 519, 520, 521, 529

710 R Radio transmitter tags, 428–429, 437–442 Rainbow trout. See Oncorhynchus mykiss Random ampliﬁed polymorphic DNA (RAPD), 337, 341, 342, 371–387 Abramis brama (bream), 373 advantages of, 382 ampliﬁed fragment length polymorphism (AFLP) and, 390, 396–397, 406 Apis mellifera (honeybee), 377–378 Barbus sp., 379 Cherax quadricarinatus (redclaw), 373 Cottus sp., 379 development of, 371–372 Dicentrarchus labrax (sea bass), 374, 380, 381t Dissostichus mawsoni (Antarctic toothﬁsh), 375 DNA ampliﬁcation ﬁngerprinting (DAF) and, 372 DNA extraction, ampliﬁcation and separation for, 372–376 Epinephelus (grouper), 379 fragment scoring and data analyses for, 376–378, 377f Gadus macrocephalus (Paciﬁc cod), 374, 381t Haliotus rubra (abalone), 375 Hardy–Weinberg equilibrium, 377, 382 Hirundichthys afﬁnis (ﬂying ﬁsh), 374, 381t Homerus americanus (American lobster), 375 Hoplostethus atlanticus (orange roughy), 375, 380, 381t inheritance of presence/absence fragments in, 378–379 Limulus polyphemus (horseshoe crab), 380 Macrobrachium borelli (prawn), 374 microsatellite DNA and, 380, 382 Morone saxatilis (striped bass), 374 Mullus barbarus (red mullet), 374, 380, 381t nuclear DNA and, 337, 341, 342 Pecten maximus (scallop), 375 Penaeus monodon (prawn), 375, 379 Penaeus stylirostris (blue shrimp), 375 Perca ﬂuviatilis (perch), 373, 381t Placopecten magellanicus (scallop), 375 polymerase chain reaction (PCR) and, 376 polymorphisms in, 377–378 population structures and, 378 population studies of teleosts using, 381t pufferﬁsh, 379

Index Rastrelliger kanguarat (Indian mackerel), 374 restriction fragment length polymorphism (RFLP), 378, 380 Salmo trutta (brown trout), 373, 381t Salvelinus alpinus (Arctic char), 373 similiarity index for, 378 species identiﬁcation and taxonomy using, 379–380 stock discrimination using, 380–382 studies using, 373–375t Tenualosa ilisha (Hilsa shad), 373 thermocyclers used for, 372 tilapia, 379 Rastrelliger kanguarat (Indian mackerel) random ampliﬁed polymorphic DNA (RAPD) and, 374 RBr notation, otolith thermal marking and, 455 Reaction norms. See Norms of reaction Reciprocal translocations, chromosome morphology and, 280 Reciprocal transplant experiments and, genetic vs. environmental variation, 50 Recombination, microsatellites and, 347 Recruitment, 135–137, 136f Gadus morhua (Atlantic cod), 136f Melanogrammus aegleﬁnus (haddock), 136f Theragra chalcogramma (walleye pollock), 136–137 Red drum. See Sciaenops ocellatus Red mullet. See Mullus barbarus Red porgy. See Pagrus pagrus Redclaw. See Cherax quadricarinatus Redﬁsh. See Sebastes sp. Renewal efforts, 37–38 Repetitive sequence markers. See Nuclear DNA, single-copy and repetitive sequence markers, 331 Replacement scales, escapee (aquaculture) identiﬁcation and, 672–673, 673f Replication banding, chromosome morphology and, 286 Reproduction, 7, 131–135 age at maturity and, 64–65 body size vs. egg amount/viability and, 59–60 effort of, 65 gonadosomatic index (GSI) in, 66 isolation of stocks and, 131–132 life history traits and, 59–70

Index migration patterns and, 31 spatial segregation of stocks and, 141–142 Restriction analysis, mitochondrial DNA and, 313–320, 315f, 322 Restriction endonucleases (REs), mitochondrial DNA and, 313 Restriction enzyme (RE) banding, chromosome morphology and, 286 Restriction enzymes, 390–391, 393, 398–399, 400, 403 Restriction fragment length polymorphism (RFLP), 314, 324 ampliﬁed fragment length polymorphism (AFLP) and, 390, 395–397, 406 mitochondrial DNA and, 352 nuclear DNA and, 337–338, 341, 360 random ampliﬁed polymorphic DNA (RAPD) and, 378, 380 Reviews of stock identiﬁcation, 1–2 Ribosomal DNA (rDNA), 280–281, 285–286, 311, 321 River catﬁsh. See Mystus nemurus Rivulus genus (cyprinids) chromosome morphology and, 274 Rivulus marmoratus (cyprinodont ﬁsh), 52 Robertsonian translocations, chromosome morphology and, 274–277, 278f Rockﬁsh. See Sebastes inermis Roughy. See Hoplostethus atlanticus Rutilus genus (cyprinids), chromosome morphology and, 274

S Sableﬁsh. See Anoplopoma ﬁmbria Sacramento River, transplant experiments of salmon into, 26 Saila, Saul B., 553 Saithe. See Pollachius virens Salmo clarki (cutthroat trout), otolith thermal marking and, 448 Salmo salar (Atlantic salmon), 57, 61, 62, 68 allozyme analysis and, 296 calciﬁed structure texture, spacing patterns and, 190, 191–192, 192f chromosome morphology and, 275, 277, 281, 289 farm raised, 659–660 landmark-based morphometric identiﬁcation and, case study in, 164–165

711 maximum likelihood estimation (MLE) and, 577–578 otolith analysis, 241 otolith thermal marking and, 448 tagging, mark-release studies on, 416, 421–423, 422t, 423f, 427 Salmo trutta (brown trout), 67 allozyme analysis and, 296 landmark-based morphometric identiﬁcation and, 167 otolith thermal marking and, 448 random ampliﬁed polymorphic DNA (RAPD) and, 373, 381t Salmon Treaties, 618–619 Salmon. See Oncorhynchus sp., Salmo sp. Salmondidae, microsatellites and, 350 Salvelinus alpinus (arctic char), 47–48, 56 chromosome morphology and, 281, 282f, 282, 283 landmark-based morphometric identiﬁcation and, 168 random ampliﬁed polymorphic DNA (RAPD) and, 373 Salvelinus fontinalis (brook trout) mitochondrial DNA and, 481 sampling and mixed stock analysis (MSA) in, 481 Salvelinus leucomaenis (iwana; white spotted char), 60, 68 chromosome morphology and, 281, 282 Salvelinus malma (Dolly Varden) allozyme analysis and, 296 Salvelinus namaycush (lake trout) allozyme analysis and, 296 chromosome morphology and, 281, 282, 283 otolith thermal marking and, 448 Sampling, 1, 2–3, 155–156, 467–498, 648 Acipenser oxyrinchus (Atlantic sturgeon), 486–487 all stocks in mixture included in baseline, 469–470 allele detection/identiﬁation in, 476 allele frequency and, 474, 486 allelic designations and scoring ambiguities in, 487–488 allometric and age relations within, 483–485 allozyme analysis and, 302–303, 487–488 Alosa sapdiissima (American shad), 481 annual variations in stocks and, 477

712 Sampling (continued) body size of individuals and, 484 bootstrap, and LDA, 548 Chi-square tests in, 474 chromosome morphology and, 284 classiﬁcation algorithms, number of features necessary for, 479–481 Clupea harengus (Atlantic herring), 484, 637–640, 638f composite stock identiﬁcation in, 472–473 differences among stocks in, statistical signiﬁcance of, 482–483 errors in, 637–640, 638f factors affecting, 490–492 fatty acid proﬁle and, 248–249 Gadus morhua (Atlantic cod), 472 Gibbs sampler and, 301 goodness-of-ﬁt testing in, 474 Hardy–Weinberg equilibrium, 474, 480–481 Homerus americanus (American lobster), 485 landmark-based morphometric identiﬁcation and, 155–156 Mahalanobis distance, in sampling size, 475 Mallotus villosus (capelin), 484 maximum likelihood estimation (MLE) and, 579–584, 579 Melanogrammus aegleﬁnus (haddock), 484 merged stock and size of sample in, 637–640, 638f microsatellite DNA and, 483, 487–488 migratory ﬁshes and, 477 minisatellite DNA and, 483, 487–488 mitochondrial DNA and, 480–481, 486–487 nuclear DNA and, 336 number of source stock included in, 471–472, 471f Onchorhychus kitsutch (coho salmon), 476 Oncorhynchus gorbuscha (pink salmon), 470 Oncorhynchus tshawytscha (chinook salmon), 479, 486, 487 otolith analysis in, 234–239 parasites as biological tags and, 216–217 populations and, 468 power to detect stock differences and, 488–489 proportion simplex sampling algorithm for, 581–584, 582f, 583f, 585f reliability of features for stock delineation in, 481–488 Salvelinus fontinalis (brook trout), 481

Index Sciaenops ocellatus (red drum), 487 Scomber scombrus (Atlantic mackerel), 484 Scomberomorus cavalla (king mackerel), 485 sex effects and, 485 size of sample in, 473–476, 475f, 477–479, 478f, 637–640, 638f source stock strategy for, 469–473 source stocks comprising few breeders in, 470–471 statistical analysis in, 474, 482–483 strategy for, 476–477 substock inclusion/exclusion in, 472–473 temporal stability of features and, 485–487 Thunnus albacares (yellowﬁn tuna), 481 Sardine. See Sardinops sagax Sardinops sagax (sardine) ELS and spawning discreteness in, 94 stock identiﬁcation of, 104–105 Sargasso Sea and eels, 32–35 Satellite relay for data gathering from electronic tracking tags, 442 Scales, 173. See also Otoliths outline-based morphometric identiﬁcation and, 173 replacement scales in, 672–673, 673f summer checks in, 671–672 transition zone, salt-to-freshwater, 669, 670f, 671f Scallop. See Pecten maximus; Placopecten magellanicus Schmidt, J., 32–33 School mackerel. See Scomberomorus queenslandicus Schroder, Steven L., 447 Schwarz, Carl James, 591 Sciaenops ocellatus (red drum) mitochondrial DNA and, 487 sampling and mixed stock analysis (MSA) in, 487 Scomber scombrus (Atlantic mackerel) sampling and mixed stock analysis (MSA) in, 484 tagging, mark-release studies on, 423–424, 425f, 423 Scomberomorus sp. (mackerel) fatty acid proﬁle and, 255, 256f S. cavalla (king mackerel) genetic analysis of stocks of, 436 outline-based morphometric identiﬁcation and, 181

Index sampling and mixed stock analysis (MSA) in, 485 spawning discreteness of various stocks, 132 S. munroi (spotted mackerel) distribution and abundance data, 124f, 124 genetic analysis of stocks of, 436 S. queenslandicus (school mackerel) genetic analysis of stocks of, 436 growth rate curves for, 129f, 129 S. (Spanish mackerel), 200f Scombridae, microsatellites and, 350 Scopthalmus maximus (turbot), fatty acid proﬁle and, 265 Scorpaenidae, microsatellites and, 350 Sea bass. See Dicentrarchus labrax Sea winter band, escapee (aquaculture) identiﬁcation and, 669–671 Sebastes inermis (black rockﬁsh) ampliﬁed fragment length polymorphism (AFLP) and, 407 Sebastes mentalla (redﬁsh) fatty acid proﬁle and, 259–263, 261f, 262f, 259 Sebastes sp. (redﬁsh) genetic variability in, 645–650, 645 management issues and, 645–650, 645 management issues and, 652–653, 653f, 654f, 652 otolith analysis in, 652–653, 653f, 654f, 652 tagging, mark-release studies on, 427–429, 429f, 427 Secor, D.H., 17 Self-organizing map (SOM) networks, 555–556, 562–565, 563f Sensors used with electronic tracking tags, 439, 442 Sequencing variations in repetitive DNA, 282–283 Sex effects, sampling and mixed stock analysis (MSA) in, 485 Sexual selection, 46 Shad. See Alosa sapdisima; Tenualosa ilisha Shared bands, mitochondrial DNA and, 314 Shellﬁsh, chromosome morphology and, 275, 289, 290 Shertzer, Kyle W., 499 Shrimp. See Penaeus stylirostris Silver hake. See Merluccius bilinearis

713 Silver kob. See Argyrosomus inodorus Silver perch, calciﬁed structure texture, spacing patterns and, 191 Silverside. See Mendidia menidia Similiarity index, random ampliﬁed polymorphic DNA (RAPD) and, 378 Simplex sampling algorithm, 581–584, 582f, 583f, 585f Single nucleotide polymorphisms (SNPs), 334, 338–340, 354–355, 625 Gadus morhua (Atlantic cod) and, 354–355 threatened and endangered species conservation and, 625 Single-approach studies, 14 Single-copy and repetitive sequence markers. See Nuclear DNA, single-copy and repetitive sequence markers, 331 Site gains, mitochondrial DNA and, 314 Size of sample, 473–476, 475f, 477–479, 478f Skipjack tuna. See Katsuwonus palamis Smith, P.J., 371 Smolt size and age, escapee (aquaculture) identiﬁcation and, 669 Snow goose. See Anser caerulescens caerulescens Soft independent modeling of class analogy (SIMCA) analysis, fatty acid proﬁle and, 251–252 Software analysis programs allozyme analysis and, 301 artiﬁcial neural networks (ANNs) and, 566, 566t movement estimation from tagging data and (MARK, M-SURGE), 596 Sole. See Microstomus paciﬁcus Source stocks. See Mixed stock analysis; sampling Southern blot analysis, mitochondrial DNA and, 316, 318 Spacing patterns. See Calciﬁed structure texture and spacing patterns Spaghetti tags, 416 SPAM estimation, 300 Spatial and temporal instability of stock management issues and, 632, 633 Spatial population models, 108–110 distribution and abundance data, 121 Generalized Additive Models (GAMs) in, 121, 123f Geographic Information Systems (GIS) in, 121 life history traits and, 121

714 Spatial variation in life history parameters, 140–142, 141f Spawning, 131–135 Anguilla japonica (Japanese eel), occurence and distribution of, 100–101, 100f biological variations and, 640–643, 642f body size vs. 59–60 Clupea harengus (Atlantic herring), occurence and distribution of, 97–99, 98f Clupea harengus (herring), discreteness of various stocks, 131–132 discreteness of, 94–95, 131–132 early life stages (ELSs) and, 94–95 Gadus morhua (Atlantic cod), 133f Homerus americanus (American lobster), 102 life history traits and, 59, 65–67 parent stream theory in, directed migration vs. local wandering, 21–24, 27, 28–30 planktonic transport stage in, 93–94 Scomberomorus cavalla (king mackerel), discreteness of various stocks, 132 spatial segregation of stocks and, 141–142 spawning stock biomass (SSB) in, 641–642 supportive breeding programs in, 619–620 threatened and endangered species conservation and, 615–617 Spawning stock biomass (SSB), 641–642 Spectroscopy and otolith analysis, 234, 241 Spotted mackerel. See Scomberomorus munroi Staining, chromosome morphology and, 285 Statistical analysis, 2–3, 474, 499–516, 517–552 age-invariant discriminant analysis in, 503 algorithms used in, 501–510 allozyme analysis and, 300–301 ampliﬁed fragment length polymorphism (AFLP) and, 405–406 artiﬁcial neural networks (ANN) in, 506–508, 507f, 512–513, 553–569 Bayesian statistics and, 301, 303, 558–559 Chi-square tests in, 474 discrete vs. nondiscrete classiﬁcation in, 502–504 discriminant analysis (DA) in, 215–216, 501–504, 503f, 511–512, 517–552 elliptical Fourier analysis in, 177 expectation maximization (EM) algorithm, 576 extended likelihood model for, 574–576 fatty acid proﬁle and, 251 ﬁnite distribution methods in, 508–510

Index ﬁnite mixture distributions (FMD) in, 508–510, 509f, 513, 514 Fourier series analysis in, 173, 175, 176–178, 177f, 178f, 187 geostatistics in, 104–105 Gibbs sampler and, 301 GIRLSEM estimation in, 300 goodness-of-ﬁt testing in, 474 jackknife discriminant analysis in, 503 landmark-based morphometric identiﬁcation and, 160–163 linear discriminant analysis (LDA) in, 501–504, 503f, 510, 519, 520, 521 logistic regression in, 504–506, 505f, 512 maximum likelihood estimation (MLE) in, 300–301, 302, 571–589 meristic traits and, 199–203, 200f, 201t Monte Carlo simulation and, 577 multivariate analysis in, 178–179, 201–203, 215–216 otolith, 237–239, 238f outline-based morphometric identiﬁcation and, 174–178, 175f parasites as biological tags and, 214–216 partial least square (PLS) analysis in, 251 polynomial discriminant analysis in, 503 polynomial functions in, 176 pooled group PCA in, 161–163 power to detect stock differences and, 488–489 principal components analysis (PCA) in, 160–163, 162f prior knowledge required for proper application of, 511–513, 520–521 quadratic discriminant analysis (QDA) in, 501–504, 503f, 519, 520, 521 sampling and mixed stock analysis (MSA) in, 474, 482–483 selection of appropriate algorithms in, 513 soft independent modeling of class analogy (SIMCA) analysis in, 251–252 software for, 301, 406, 406t SPAM estimation in, 300 statistical package for analysis of mixtures (SPAM) in, 621, 624 stepwise discriminant analysis in, 503 stock composition analysis and, 499–500 tree-based regression or CART in, 513 Statistical package for analysis of mixtures (SPAM), 621, 624

715

Index Steelhead trout. See Oncorhynchus mykiss Stepwise discriminant analysis, 503 Stepwise mutation model (SMM), microsatellites and, 348 Stickleback. See Gasterosteus sp. Stochastic expectation maximization (SEM), 524 Stock composition analysis, 8, 499–500 Stock Concept International Symposium (1980), deﬁning stocks, 9–10 Stock discrimination vs. identiﬁcation, 8 Stock identiﬁcation, 1–6, 417–418, 499–500. See also Deﬁnition of stock accessible stock deﬁned for, 634 ampliﬁed fragment length polymorphism (AFLP) and, 406–407 applications for, 3 biological stock deﬁned for, 634 biological variations and, 640–643, 642f conservation biology and, 11 deepwater stocks of North Atlantic and, 634–637 deﬁnition of stocks in, 7–16, 499–500 early life stages (ELSs) in, 89–117, 89 environmental inﬂuences on, 45–86 escapee (aquaculture) identiﬁcation and, 659–679 ﬁshery stock deﬁned for, 633–634 Gadus morhua (Atlantic cod) and, 351 genetic inﬂuences on, 9–15, 45–86 harvest stock deﬁned for, 633 interdisciplinary analysis in, 3 life history traits in, 119–150 management and, importance of, 631–658 management of stocks and, 7 methodology for, 2 protocols for (ICES), 2, 17 random ampliﬁed polymorphic DNA (RAPD) and, 380–382 reviews of, 1–2 statistical analysis in, 499–516 stock composition analysis and, 8, 499–500 stock discrimination vs., 8 tagging and, 417–418 targeted stock deﬁned for, 633–634 unit stock concept in, 7–8, 14, 17–44 Streamer tags, 416 Striped bass. See Morone saxatilis Stronger vs. weaker stock, 90 STRUCTURE allozyme analysis software, 301

Sturgeon. See Acipenser oxyrinchus Subcutaneous tags, 417 Subpopulations, 12 meristic traits and, 203 sampling and mixed stock analysis (MSA) in, inclusion/exclusion in, 472–473 Summer checks, escapee (aquaculture) identiﬁcation and, 671–672 Summer ﬂounder. See Paralichthys dentatus Supervised vs. unsupervised artiﬁcial neural networks (ANNs), 555–556, 560–562 Supportive breeding programs, 619–620 Surfperch. See Cymatogaster aggregata Swain, Douglas P., 45 Swordﬁsh. See Xiphias gladius

T T bar anchor tags, 416 Tagging, 24–25. 415–433, 426f, 447–448, See also Marking experiments Anoplopoma ﬁmbria (sableﬁsh), 420–421 applications for, 429–430, 447–448, 457–459 archival (retrieved) tags in, 442 capture and handling of ﬁsh for, 419 Carcharodon carcharias (white shark), 437 Carlin tags in, 427 Clupea harengus (Atlantic herring), 416, 424–427, 426f Clupea pallasi (Paciﬁc herring), 424–427 coded wire tag (CWT) in, 417, 420, 427 dart streamer tags in, 416 effects of handling and, on ﬁsh, 420–421 electronic tag readers and, 416, 417 electronic tracking tags in, 435–446 advantages of, 444 antennas used in, 439 archival (retrieved) tags in, 442 Carcharodon carcharias (white shark), 437 distance of telemetry signal in, 438–439, 443–445 limitations of, 444 mark-recapture programs and, 442 monitoring stations used with, 444 Plectropomus leopardus (coral trout), 437 satellite relay for data gathering from, 442 sensors used in, 439, 442 size of tags used in, determining parameters for, 438, 440f, 441f

716 Tagging (continued) technology used in, 437–442 telemetry in, 437, 442–444 transmitters for, 437–442, 443 ultrasonic vs. radio transmitters for, 437–442 Xyrauchen texanus (razorback suckers), 437 errors in recovery and, 456–457 escapee (aquaculture) identiﬁcation and, 676 history and development of, 415–417 holding of ﬁsh for, 419, 420 in situ methods for, 427–429 internal and external tags for, 415–433 life span of tags in, external vs. internal, 420 loop spaghetti tags in, 416 magnetic tags in, 416, 417 management and, 435–437, 447–448, 458–461 mark-recapture methods and, 418, 421–429 mass marking. See Otolith thermal marking, below movement estimation from tagging data and, 591–606 mutilation or markers for, 417 Oncorhynchus mykiss (rainbow trout), 428–429 otolith thermal marking in, 447–463, 450f, 451f, 452f, 453f, 456f applications for, 455, 457–459, 459–461 drawbacks and limitations of, 460 early experiments in, 449 errors in mark recovery and, 456–457 hatchery use of, 455, 458–459, 460–461 inducement of marks in, methods for, 448–456 kappa index in, 457 management and, 458–461 method for, 447–448 migration tracking and, 458–459 Oncorhynchus nerka (sockeye salmon), 458 Oncorhynchus gorbushcha (pink salmon), 457, 458 Oncorhynchus keta (chum salmon), 449, 452f Oncorhynchus lapillus (coho salmon), 449, 450f, 451f, 453f, 457 Oncorhynchus mykiss (rainbow/steelhead trout), 448 Oncorhynchus spp. (Paciﬁc salmon), 448

Index Oncorhynchus tshawytscha (chinook salmon), 449, 450f, 455, 456f, 457, 459 organizing pattern information from, 454–456 RBr notation in, 455 Salmo clarki (cutthroat trout), 448 Salmo salar (Atlantic salmon), 448 Salmo trutta (brown trout), 448 Salvelinus namaycush (lake trout), 448 systems for, 455–456 temperature effects on otoliths and, 448–454 planning a tagging study and, general concerns in, 418–419 plastic tube marker in, 429f Plectropomus leopardus (coral trout), 437 practical aspects of, 418–421 radio transmitter type, 428–429 RBr notation in, 455 Salmo salar (Atlantic salmon), 416, 421–423, 422t, 423f, 427 Scomber scombrus (Atlantic mackerel), 423–424, 425f Sebastes sp. (redﬁsh), 427–429, 429f stock identiﬁcation concept and, 417–418 subcutaneous tags for, 417 T bar anchor tags in, 416 tag selection for, 418–419 techniques for, 416 telemetry in, 437, 442–444 Thunnus sp., 416 Trachurus trachurus (horse mackerel), 652 types of tags used in, 416, 419–420 Underwater Tagging Equipment (UTE) for, 428–429, 428f Visible Implant (VI) tag in, 417, 421 Xyrauchen texanus (razorback suckers), 437 Tandem translocations, chromosome morphology and, 277, 279f Targeted stock deﬁned, 633–634 Telemetry, electronic tracking tags and, 437, 442–444 Temperature effects on otoliths, in thermal marking, 448–454 Temporal stability of features, 137–149, 139f–140f management issues and, 632, 633 sampling and mixed stock analysis (MSA) in, 485–487

Index Tenualosa ilisha (Hilsa shad), random ampliﬁed polymorphic DNA (RAPD) and, 373 Term informative AFLP, 405 Tetranucleotide (GATA) repeats, microsatellites and, 344, 345f Texture. See Calciﬁed structure texture and spacing patterns Theragra chalcogramma (walleye pollock) genetic analysis of stocks of, 436 outline-based morphometric identiﬁcation and, 180 recruitment data on, 136–137 Thermal ionization mass spectrometer (TIMS), 241 Thermal otolith marking. See Otoliths, thermal marking of Thermocyclers, random ampliﬁed polymorphic DNA (RAPD) and, 372 Thompson, William F., 622 Threatened or endangered species conservation, 609–629 aggregate abundance-based management (AABM) in, 618–619 Alaska and, 620–622 allozyme analysis and, 621 applications of, 610 broodstock shortfalls, disease, ﬁshery impact assessment in, 615–617 Endangered Species Act (ESA) and, 610–611 ﬁshing politics and, U.S./Canada salmon wars, 617–619 future of, 624–626 genetic mixture analysis (GMA) and, 621 GSI Foundation and, 620–622 hematopoietic necrosis virus (IHNV) tracking, 615–617 High Seas Salmon Program, 622–624 individual stock-based manaegment (ISBM) in, 619 interception of endangered species and, 620–622 International North Paciﬁc Fisheries Commision (INPC) in, 622 management of, 609–610 microsatellite DNA and, 620–622, 624–625 missing stocks and, 622–624 mitochondrial DNA and, 623–624 North Paciﬁc Anadromous Fish Commission (NPAFC) in, 622

717 Onchorhychuns kitsutch (coho salmon), 617–619 Oncorhynchus tshawytscha (chinook salmon), 610–614, 611f, 614f, 615–619 Oncorhynchus gorbuscha (pink salmon), 617–620, 624 Oncorhynchus keta (chum salmon), 617–619, 624 Oncorhynchus nerka (sockeye salmon), 615–619 Paciﬁc Salmon Treaties, 618–619 population genomics and, 625–626 single nucleotide polymorphisms (SNPs), 625 statistical package for analysis of mixtures (SPAM) in, 621, 624 supportive breeding programs in, 619–620 U.S. Exclusive Economic zone and, 624 United Nations resolution 44–225 (driftnet ﬁshing) and, 622–623 zero allele frequencies and, 620–622 Thrysites atun (New Zealand barracuda), genetic analysis of stocks of, 436 Thunnus sp. tagging of, 416 T. alalunga (albacore tuna), mitochondrial DNA and, 324 T. albacares (yellowﬁn tuna) mitochondrial DNA and, 324, 481 sampling and mixed stock analysis (MSA) in, 481 T. obesus (bigeye tuna), mitochondrial DNA and, 324 T. thynnus (Atlantic blueﬁn tuna), genetic analysis of stocks of, 436–437, 436 Tilapia sp., 66 chromosome morphology and, 289 random ampliﬁed polymorphic DNA (RAPD) and, 379 Toothﬁsh. See Dissostichus mawsoni Total DNA extraction, 316–318, 317f Total Allowable Catches (TAC) allocation, 634, 639–640, 641, 653 Trace elements, 227, 675 escapee (aquaculture) identiﬁcation and, 675 Trachurus trachurus (horse mackerel) management issues and, 650–652, 651f Training, artiﬁcial neural networks (ANNs) and, 554, 560 Transcription, microsatellites and, 347

718 Transfer RNA (tRNA), 311, 321 Transition zone, salt-to-freshwater, escapee (aquaculture) identiﬁcation and, 669, 670f, 671f Translocations, DNA, 274–277 reciprocal, 280 Robertsonian, 274–277, 278f tandem, 277, 279f Transmitters, electronic tracking tags and, 437–442, 443 Transplant experiments and migration, 26 Tree-based regression or CART, 513 TREECON AFLP data analysis software, 406 Triangle migration/life cycle concept, 29–30, 30f Triglycerides, fatty acid proﬁle and, 265–266 Trinucleotide (ACC) repeats, microsatellites and, 344, 345f Triploidy, chromosome morphology and, 274 Trout. See Salmo trutta Tsukamoto, K., 37 Tucker, D.W., 33–34 Tuna. See Katsuwonus palamis; T. alalunga, T. albacares, T. obesus, T. thynnus

Index natural tags in identiﬁcation of, 22–24, 23f, 38 ocean studies and, 27–28 panmixia theory in, 18 parent stream theory in, directed migration vs. local wandering, 21–24, 27, 28–30 population structures and, 18–19, 19f, 30 tagging and marking experiments and, 24–25 transplant experiments and, 26 year-class phenomenon in, 26–27 United Nations resolution 44–225 (driftnet ﬁshing) and, 622–623

V Vaccination, adhesions of abdominal cavity in, post vaccination, 673–674, 674f Vagrant. See Member-vagrant hypothesis Variable number of tandem repeats (VNTRs), nuclear DNA and, 341 Visible Implant (VI) tag, 417, 421 Volk, Eric C., 447 von Bertalanffy growth equation, 60, 128–130, 129f

U U.S. Endangered Species Act, 11 U.S. Exclusive Economic zone, 624 UCONDLA program, 540–541, 544 Ultrasonic transmitters, electronic tracking tags and, 437–442 Uncertainty, artiﬁcial neural networks (ANNs) and, 522–523, 535 Unconditional estimation, 538–541 Undershot jaw, escapee (aquaculture) identiﬁcation and, 665 Underwater Tagging Equipment (UTE), 428–429, 428f Unit stock concept, 7–8, 14, 17–44 biochemical markers in, 30 biodiversity and, 37 cod hatchery enhancement and, 19–21, 20t development of, 18 eel problem in, 31–37 evolutionary concepts of, 18 ICES protocols and, 17–18 modern migration theory and, 18, 21

W Waldman, John R., 1, 2, 7, 197, 331 Walleye pollock. See Theragra chalcogramma Wavelength dispersive electron microprobes (WD-EM), otolith analysis, 235 Weaker vs. stronger stock, 90 Weakﬁsh. See Cynoscion regalis Weights and weighting, artiﬁcial neural networks (ANNs) and, 554 Whale sharks, 62 WHICHRUN allozyme analysis software, 301 White spotted char. See Salvelinus leucomaenis Whiteﬁsh, chromosome morphology and, 281 Whiting. See Micromesistius poutassou Wilmot, R., 295 Winter band, escapee (aquaculture) identiﬁcation and, 669–671 Winter ﬂounder. See Pseudopleuronectes americanus Wirgin, Isaac, 331 Wreckﬁsh. See Polyprion americanus

719

Index

X Xiphias gladius (swordﬁsh) mitochondrial DNA and, 324, 325 Xyrauchen texanus (razorback suckers) electronic tracking tags and, 437

Yellow perch. See Perca ﬂavescens Yellowﬁn tuna. See Thunnus albacares Yellowtail ﬂounder. See Limanda ferruginea

Y

Z

Year-class phenomenon, migration, 26–27 Yeast artiﬁcial chromosomes (YACs), 288

Zebraﬁsh. See Brachydanio rerio Zimmerman, Christopher, 631