genetic stock identiﬁcation of steelhead in the columbia ......674 winans et al. figure 1.—map...

672

North American Journal of Fisheries Management 24:672–685, 2004q Copyright by the American Fisheries Society 2004

Genetic Stock Identification of Steelhead in the Columbia RiverBasin: An Evaluation of Different Molecular Markers

GARY A. WINANS,* MELANIE M. PAQUIN, AND DONALD M. VAN DOORNIK

National Marine Fisheries Service, Northwest Fisheries Science Center,2725 Montlake Boulevard East, Seattle, Washington 98112, USA

BRUCE M. BAKER

Washington Department of Fish and Wildlife,600 Capital Way North, Olympia, Washington 98501, USA

PERRY THORNTON


DAN RAWDING AND ANNE MARSHALL

Washington Department of Fish and Wildlife,600 Capital Way North, Olympia, Washington 98501, USA

PAUL MORAN


STEVEN KALINOWSKI

Montana State University, Department of Ecology, Bozeman, Montana 59717, USA

Abstract.—Protein genetic markers (allozymes) have been used during the last decade in a geneticstock identification (GSI) program by state and federal management agencies to monitor stocksof steelhead Oncorhynchus mykiss in the Columbia River basin. In this paper we report new datafor five microsatellite and three intron loci from 32 steelhead populations in the three upriverevolutionarily significant units (ESUs) and compare the performance of allozyme, microsatellite,and intron markers for use in GSI mixture analyses. As expected, microsatellites and introns hadhigh total heterozygosity (HT) values; but there was little difference among marker classes in themagnitude of population differentiation as estimated by Wright’s fixation index (FST), which rangedfrom 0.041 (microsatellite loci) to 0.047 (allozyme loci) and 0.050 (intron loci). For allozyme andmicrosatellite loci, the relationships among populations followed the patterns of geographic prox-imity. In computer-simulated mixture analyses, GSI estimates were more than 85% correct to thereporting group, the exact percentage depending on the marker data set and target group. Micro-satellite loci provided the most accurate estimate (83%) in the 100% upper Columbia River ESUsimulation, whereas simulation estimates for the 32-locus allozyme baseline were 93–94% for the100% middle Columbia River ESU and two Snake River management groups. The simulationsalso showed that the estimates improved substantially up to a sample size of 50 fish per population.Technical advances will concomitantly increase the number of useful microsatellite loci and therate of laboratory throughput, making this class of molecular marker more valuable for GSI mixtureanalyses in the near future. In the meantime, we recommend that steelhead management in theColumbia River rely on both allozyme and microsatellite data for GSI procedures.

Steelhead Oncorhynchus mykiss were once veryabundant in the Columbia River, peak run sizesbeing put as high as 500,000 adults (TAC 1997).As with other species of Pacific salmon, steelhead

* Corresponding author: [email protected]

Received March 13, 2003; accepted September 23, 2003

population sizes were reduced to the tens of thou-sands in the last century by hydroelectric devel-opment, in-river and ocean fisheries, and streamhabitat degradation. In the 1990s, concern aboutthe reduction in the number and abundance ofsteelhead populations prompted an evaluation ofpopulation viability and abundance under the pur-view of the U.S. Endangered Species Act. Based

673MOLECULAR MARKERS FOR STEELHEAD

on available life history and genetic data, the Na-tional Marine Fisheries Service identified fourevolutionarily significant units (ESUs) of steel-head in the basin (Busby et al. 1996). Three ESUsare found entirely above Bonneville Dam (riverkm 235 [measuring from the river’s mouth]): themiddle Columbia River ESU (roughly from Bon-neville Dam to and including the Yakima Riverbut excluding the Snake River), the upper Colum-bia River ESU (upstream of the Yakima River),and the Snake River ESU. The steelhead in theseESUs are considered inland-type steelhead O. m.gairdneri (as distinct from the coastal form O. m.irideus; Busby et al. 1996) and are almost exclu-sively stream-maturing populations.

Further biological diversity is recognized withinthese upriver populations of the Columbia Riverbasin. Fishery managers have designated two ‘‘runtypes,’’ A and B, based on bimodal peaks in therun counts of adults at Bonneville Dam. By defi-nition, adult A-run steelhead enter freshwater firstand pass Bonneville Dam prior to August 25,whereas B-run steelhead pass after that date. A-run steelhead are predominantly age-1 ocean fishwhile B-run steelhead adults are age-2 ocean fish(Busby et al. 1996). The B-run stocks are thoughtto have higher growth rates and to be larger thanA-run fish at a given age. B-run steelhead arethought to exist only in the Clearwater, MiddleFork Snake, and South Fork Snake rivers. Fisheryagencies manage populations by ESU and, in somecases, by run group (TAC 1997).

Since 1997, protein genetic markers (allozymes)have been used in a genetic stock identification(GSI) program by state and federal managementagencies to monitor steelhead stocks in the basin(Rawding et al. 1997). From April to September,fin clips and opercular punches have been takennonlethally from returning adults at BonnevilleDam for GSI analysis to monitor passage dates bystock. Upriver, steelhead are incidentally caughtin an in-river fishery (zone 6; see Figure 1) forChinook salmon O. tshawytscha in August andSeptember. To estimate and monitor stock com-position by ESU, tissues (muscle, heart, liver, andretinal) are taken from incidentally killed steelheadfor genetic analysis. For both GSI applications,allozyme data from mixture specimens are as-sessed against a baseline of 32 or more loci fromeach of 39 populations (Rawding et al. 1997).

While allozyme baselines are still being used inGSI applications (Seeb and Crane 1999; Wilmotet al. 2000; Winans et al. 2001), incipient baselinesbased on DNA markers are growing in number and

geographic complexity (e.g., Beacham et al. 2001).The potential advantages of these new molecularmarkers (e.g., single-nucleotide polymorphisms ofmitochondrial DNA and nuclear genes as well asnuclear microsatellites) are enhanced populationdiscrimination and technical suitability for highlaboratory output (e.g., number of loci per fish perworking day; Park and Moran 1994). Small tissuerequirements per specimen reduce fish-handlingtime and stress. Although heralded as the nextboon for population genetic research (Avise 1994;Lewontin 1991), the accumulation of geographi-cally complete baselines of DNA-based loci thatmatch the preexisting allozyme baselines for Pa-cific salmon (Seeb and Crane 1999; Teel et al.1999) has been slow (but see Beacham et al. 2001).There are also indications that some of the earlypredictions that DNA markers would provide moreaccurate population discrimination may not be ful-filled. For example, Allendorf and Seeb (2000)found concordance in the amount of genetic var-iation within and between populations of sockeyesalmon O. nerka for allozyme, nuclear DNA, andmitochondrial DNA markers and Scribner et al.(1998) found no difference in the ability of allo-zymes, mitochondrial DNA, and microsatelliteloci to discriminate between two groups comprisedof eight populations of chum salmon O. keta.

In this paper, we report new data for five mi-crosatellite and three intron loci from 32 steelheadpopulations in the three upriver ESUs (Figure 1).Our purposes are to (1) describe and compare thepatterns and levels of variability among allozyme,microsatellite, and intron markers; (2) demonstratethe accuracy and precision of ESU estimates insimulated mixtures using various marker data sets;(3) identify subsets of ‘‘important’’ allozyme locifor use in GSI applications; and (4) recommend aset of loci for future GSI applications in steelheadmanagement, monitoring, and conservation in theColumbia River basin.

Methods

Allozymes.—The allozyme baseline data set iscomposed of information collected over the last10–15 years, primarily by the Washington De-partment of Fish and Wildlife. When the NationalMarine Fisheries Service evaluated the coastwidestatus of steelhead under the purview of the En-dangered Species Act (Busby et al. 1996), fisheryagencies constructed a standardized allozymebaseline with data from several sources (Wapleset al. 1993; Phelps et al. 1994). These data haveserved as the in-river genetic baseline for GSI pro-

674 WINANS ET AL.

FIGURE 1.—Map showing the locations of 32 collections of steelhead in the Columbia River basin (delimitedby the white line). Numbers correspond to the populations listed in Table 1. Main-stem dams are indicated by bars.The Zone 6 fishery is located upstream of Bonneville Dam (river kilometer 235, measuring from the river’s mouth)and below McNary Dam (river kilometer 470).

cedures. Included in this data set are eight newpopulations, as indicated in Table 1.

All allozyme data were obtained by standardstarch gel electrophoresis (Aebersold et al. 1987)under the laboratory conditions described in Wa-ples et al. (1993). Population statistics have beendescribed previously (Waples et al. 1993; Phelpset al. 1994). Distance metrics and genetic diversity(i.e., Wright’s fixation index, FST) calculationswere obtained with BIOSYS (Swofford and Se-lander 1981). Population differences were depictedby means of multidimensional scaling (MDS) ofgenetic distance metrics in NTSYS-pc (Rohlf1994). A principal components analysis (PCA)was conducted on the correlation matrix amongthe frequencies of the most common allozyme al-leles and, in some cases, those of secondary ortertiary alleles present at a frequency of at least5% (e.g., P0.95 alleles).

DNA loci.—DNA was extracted from tissues ar-chived at 2808C using a Qiagen DNeasy DNAisolation kit. A 1-mm 3 5-mm piece of fin clip ormuscle tissue was added to lysis buffer and in-

cubated at 558C for 5.5–8 h (see Qiagen rodenttail and animal tissue protocol). We amplified fivemicrosatellite loci by means of polymerase chainreaction (PCR): Onem8 using primer sets 59-AA-CATTCTGGGATGACAGGGTA-39 and 59-CTGTTCTGCTCC AGTGAAGTGGA-39 (Scribner etal. 1996), Onem14 using primer sets 59-AGAAACATGAGAACAGTCTAGGT-39 and 59- CCTTATGAGTTTGGTCTCCATGT-39 (Scribner et al.1996).

Ots4 using primer sets 59-GACCCAGAGCACAGCACAA-39 and 59-GGAGGACACATTTCAGCAG-39 (Banks et al. 1999), Ssa289 using primersets 59-CTTTACAAATAGACAGACT-39 and 59-TCATACAGTCACTATCATC-39 (McConnell etal. 1995), and a new microsatellite locus (P53ms)using primer sets (59-TGACACATATCCTCGCTTTCTCC-39 and 59-CAACTCTCTTGGTGAGGC-39. Polymerase chain reactions were performedin 10-mL reactions with 5–25 ng/mL of DNA, 0.2–0.4 mM of each primer, 1.75–2.5 mM MgCl2, 0.2–0.4 mM deoxynucleotide triphosphates, and 0.5–2.5 units Promega Taq DNA polymerase. Initial


TABLE 1.—Sampling information for 32 populations of steelhead in the Columbia River basin. Population locationsare shown in Figure 1. Data were previously described by the Washington Department of Fish and Wildlife (WDFW;Phelps et al. 1994, 1996, 2000) and the National Marine Fisheries Service, (NMFS; Waples et al. 1993) except for the8 populations designated as ‘‘new.’’

Populationcode Population name Locationa

Number of fish

Allozymedata

DNAdata

Year(s)collected Agency

Upper Columbia River

12

Wells HatcheryWells ‘‘wild’’

Upper ColumbiaUpper Columbia

9044

4843

19911995, 1998

WDFWWDFW (new)

Middle Columbia River

345678

Upper Klickitat RiverBowman CreekLittle Klickitat RiverLower Klickitat RiverUmatilla HatcheryBeech Creek

KlickitatKlickitatKlickitatKlickitatUmatillaJohn Day Dam

3091041107586

105

155542215647

1991, 199419911991199419961996

WDFWWDFWWDFWWDFWNMFS (new)NMFS (new)

91011121314

Satus RiverToppenish CreekWapatox TrapRoza TrapChandler TrapTouchet River

YakimaYakimaYakimaYakimaYakimaWalla Walla

33317537017537399

585644555953

1989–199019901987198919871995

WDFWWDFWWDFWWDFWWDFWWDFW (new)

A run, Snake River

151617181920212223

Lower Tucannon RiverUpper Tucannon RiverLyons Ferry HatcheryAsotin CreekChesnimnus CreekDeer CreekWallowa HatcheryLick CreekCamp Creek

TucannonTucannonTucannonSnakeGrande RondeGrande RondeGrande RondeImnahaImnaha

14318410010020020020019299

483815524360465047

1989–19901989–1990199119951989, 19901989, 19901990–19911989–19901990

NMFSNMFSNMFSWDFW (new)NMFSNMFSNMFSNMFSNMFS

24252627

Little Sheep CreekLittle Sheep HatcheryBargamin CreekPahsimeroi Hatchery

ImnahaImnahaSnakeSnake

20020046

100

44474548

1989–19901990–199119991990

NMFSNMFSWDFW (new)NMFS

B run, Snake River

2829303132

Loon CreekSecesh RiverDworshak HatcherySelway Riverb

Lochsa Riverc

Mid Fork SalmonSouth Fork SalmonClearwaterClearwaterClearwater

6530

20083

176

4830404748

199919991989, 199119901989, 1990

NMFS (new)WDFW (new)NMFSNMFSNMFS

a River unless indicated otherwise.b At Gedney Creek.c At Fish Creek.

denaturation of 2–5 min at 958C was followed byone of two thermalcycler profiles, depending onthe locus. For Onem8 (608C annealing tempera-ture), P53ms (618C annealing temperature), andOts4 (548C annealing temperature), initial dena-turation was followed by 32 cycles of 40 s at 948C,40 s at the specified annealing temperature, and40 s at 728C. For Ssa289 (548C annealing tem-perature) and Onem14 (588C annealing tempera-ture), initial denaturation was followed by 32 cy-cles of 70 s at 948C, 70 s at the specified annealingtemperature, and 70 s at 728C. A final extensionof 5 min at 728C was added to the end of each

thermalcycler profile. The forward primer fromeach pair was fluorescently labeled with one offive dyes (FAM, NED, HEX, VIC, or PET; AppliedBiosystems), and electrophoresis was conductedon an Applied Biosystems Model 3100 automatedsequencer. Fragment size was determined and ge-notyping analysis performed with Genescan 3.6and Genotyper 3.6, respectively (Applied Biosys-tems). Restriction fragment length polymorphismswere analyzed to detect variation in the intron re-gions of three loci: GnRH, IK, and RAG39 (seeBaker et al. 2002 for primer sequences, PCR con-ditions, and electrophoretic conditions). The PCR

676 WINANS ET AL.

products for each locus were digested with a spe-cific restriction enzyme: MseI for GnRH, DraI forIK, and FokI for RAG39 (see Baker et al. 2002).Fragment lengths were determined by visuallycomparing the electrophoresed fragments with a100-base-pair DNA ladder from New England BioLaboratories.

Computer simulations.—We analyzed computer-generated (i.e., simulated) population mixtures toevaluate the ability of different sets of loci to es-timate the proportions of each ESU in a mixture.In each simulation, a mixture of 400 fish (i.e., 400multilocus genotypes) was drawn from the base-line data and the stock composition of this mixturewas estimated using a bootstrapped baseline dataset. All sampling was done with replacement andassumed random mating and the independent as-sortment of loci. This sampling was repeated 1,000times. Averages and standard deviations were re-corded for each stock proportion estimate. In thefirst set of simulations, 100% of the fish in themixture were from one ESU or management group.In the second set of simulations, we used threemixture scenarios that mimic the mixture com-positions seen in past GSI work in the ColumbiaRiver. In the third set of simulations, we exploredthe effect of baseline sample size on the accuracyof the estimates of mixture proportions. The al-lozyme baseline data set is currently much largerthan the microsatellite baseline data set (an aver-age of 100 individuals per population for the al-lozyme data versus 48 individuals per populationfor the microsatellite data), and we wanted to es-timate how much stock composition estimatesmight improve if the microsatellite data set wereincreased. In these simulations, we varied the base-line sample size from 10 to 300 individuals byrandomly drawing alleles from the Rannala andMountain probability distribution (which is verysimilar to the allele frequencies in the actual base-line data) for mixtures composed solely of indi-vidual ESUs.

The mixture analyses were conducted with Ge-netic Mixture Analysis (GMA; available at http://www.montana.edu/kalinowski/GMA/kalinowskipGMA.htm). In several instances, simulations werealso conducted with the SPAM (Debevec et al.2000) computer package (results not shown here).Both computer programs produce conditionalmaximum likelihood (Millar 1987) estimates ofmixture proportions, but they use different meth-ods for estimating the probability of sampling anallele from a baseline population. The importantdifference between these programs is that SPAM

uses the maximum likelihood estimate of the prob-ability of sampling the ith allele from a baselinepopulation (ni/n, where ni is the number of copiesof the ith allele that have been observed in a base-line sample of n genes) and GMA uses a Bayesianestimate of this probability ({ni 1 [1/k]}/[n 1 1],where k is the number of alleles at the locus; Ran-nala and Mountain 1997). The distinction betweenthese two estimates is that the maximum likelihoodprobability will equal zero for an allele not foundin a baseline sample while the Bayesian proba-bility will equal a small nonzero number. In ourexperience, the Bayesian approach completelyeliminates the ‘‘unclassified’’ multilocus geno-types that are identified in mixture results basedon SPAM.

Locus evaluation and selection criteria.—Toevaluate whether a subset of allozyme loci wouldprovide adequate population differentiation in amixture analysis, we constructed three differentsubsets of 10 loci each. For each subset, we chose10 loci as a baseline for simulation analyses basedon the largest FST or total heterozygosity (HT ; i.e.,the mean expected heterozygosity over popula-tions) values or on the weighting coefficients alongthe first three components of a PCA. We then com-pared these results with the full data sets for eachmarker group.

Results

We collected data for all marker sets for 32 pop-ulations (Table 1). The average number of fish an-alyzed for allozymes was 158 per population. Theaverage number of fish analyzed for DNA vari-ability was 44 per population. Three populationsfrom the lower Columbia River ESU (Wind, Pan-ther, and Skamania rivers) used in the WashingtonDepartment of Fish and Wildlife allozyme baselinefor GSI analyses (Kassler et al. 2002) were notincluded because of their absence or low occur-rence (,1%) in mixture analyses to date (Winans,unpublished data). Allele frequency data are avail-able for both allozyme (Washington Departmentof Fish and Wildlife, Genetics Section, 600 CapitalWay North, Olympia, Washington 98501–1091)and DNA loci (http://www.nwfsc.noaa.gov/data-sets/steelheadpallelepdata.xls).

Allozymes

The allelic variability at 32 loci is summarizedin Table 2. The allozyme loci were characterizedby generally low levels of variation. Only 8 alleles(from 7 loci) were variable at the P0.95 level in50% or more populations (Table 2). Of the 32 al-


TABLE 2.—Summary of genetic variation at 32 loci in populations of steelhead. In the column headings, A is thenumber of alleles; FST is Wright’s fixation index; PC loci are loci or alleles with large weights on principal componentaxes 1–3 (PC1–3) of 17 common alleles and 9 secondary or tertiary alleles; HT is total heterozygosity; and a P0.95population is one with a secondary or tertiary (parentheses) allele with a frequency $ 0.05 (for microsatellite loci, thenumber of alleles at a locus with a frequency $ 0.05 in at least one population is given in brackets).

Locus A

Meanallele

frequency FST

PCloci HT

No. ofP0.95

populations

Allozymes

mAAT-1*sAAT-1,2*sAAT-3*ADA-1*ADA-2*ADH*

233354

0.9830.9700.9940.9830.9700.988

0.0260.0100.0460.0370.0320.020

0.0320.0580.0130.0340.0580.024

471342

sAH*ALAT*FH*GAPDH*G3PDH*GPI-A*

433343

0.7600.9480.9860.9440.9870.993

0.0400.0440.0260.0660.0590.012

PC2

PC3PC1a

0.3750.0980.0280.1050.0260.026

32 (5)13 (1)2

14 (1)40

GPI-B1*GPI-B2*mIDHP*sIDHP-1*sIDHP-2*LDH-B1*

424452

0.9930.9990.9850.9950.2090.999

0.0430.0000.0290.0040.0270.015

a

a

a

0.0140.0010.0300.0100.6480.002

1020

32 (32)0

LDH-B2*LDH-C*sMDH-A1,2*sMDH-B1,2*sMEP-1*MPI*NTP*PEPA*

32462344

0.4080.9980.9950.9750.9880.9430.7210.879

0.0770.0110.0020.0200.1070.0650.0540.064

PC1a

a

PC1PC1PC2PC1

0.4870.0050.0100.0500.0230.1090.4100.215

32004 (1)3

1332 (1)28 (3)

PEPD-1*PEPLT*PGK-2*PGM-2*sSOD-1*TPI-3*Average

424334

0.9800.9910.5830.9900.9040.979

0.0420.0260.0360.0220.0640.0760.047

PC1

PC3PC1PC1

0.0390.0170.5110.0190.1780.0410.116

42

32 (6)27 (16)3

Microsatellites

Onem8*Onem14*

2110

0.190b

0.551c0.0440.027

0.8420.593

32 [15]32 [7]

Ots4*P53ms*Ssa289*Average

10198

0.436d

0.583e

0.648f

0.0360.0540.0420.041

0.7010.5870.5060.646

32 [8]32 [8]32 [5]

Introns

IK*GnRH*RAG3*Average

222

0.5680.9010.896

0.0720.0590.0360.050

0.4550.1670.1790.267

322229

a Dropped due to low variability.b Allele *149.c Allele *154.d Allele *123.e Allele *160.f Allele *106.

678 WINANS ET AL.

lozyme loci, 23 (72%) had 3–4 alleles per locus.The maximum number of alleles per locus was 6(sMDH-B1,2*). A total of 109 allozyme alleleswere scored. Of the 172 tests for conformity toHardy2Weinberg expectations in the 8 new pop-ulations (Table 1), only 2.5% were statistically sig-nificant. For the remaining 24 samples, previoustest statistics were all approximately within thelimits of what would be expected by chance alone(,5%; see Phelps et al. 1994, 1996, 2000; Wapleset al. 1993), and the incidence of departure fromHardy2Weinberg expectations appeared to be ran-domly distributed among populations and geneloci. All chi-square comparisons of allele fre-quencies locus by locus (across populations) werestatistically significant (P , 0.001), with the ex-ceptions of sMDH-A1,2* (P 5 0.017) and GPI-B2*(P 5 0.59). The average FST for allozymes was0.047, 10 loci having an FST greater than 0.05,indicating moderate levels of geographic differ-entiation (Hartl 1981). The HT values ranged from0.014 (GPI-B1*) to 0.648 (sIDH-P2*), averaging0.116.

Microsatellites

Allelic variability at five microsatellite loci issummarized in Table 2. Sixty-eight alleles werescored over the five loci, ranging from 8 alleles atSsa289 to 21 alleles at Onem8 (Table 2). From 5 to15 alleles per locus had a frequency of at least 0.05(i.e., P0.95 level; Table 2). There were 21 (13%)statistically significant departures from Hardy–Weinberg proportions. Almost half (9) of these weredue to heterozygote deficiencies in three YakimaRiver populations (Chandler, Rosa, and Wapatoxtraps) and may be due to downstream admixture ofjuveniles from semi-isolated populations. Allelefrequency data from these collections will be usedtentatively until further sampling is conducted. Theremainder of the Hardy–Weinberg departures ap-peared to be randomly distributed across popula-tions and loci. All chi-square contingency tests ofallele frequencies over all populations were statis-tically significant (P , 0.001). The mean FST formicrosatellite loci was 0.041; HT ranged from 0.506(Ssa289) to 0.842 (Onem8), averaging 0.646.

Introns

Two alleles were scored at each intron locus, fora total of 6 alleles (Table 2). Each locus was var-iable at the P0.95 level in essentially all populations,and the allele frequencies were statistically sig-nificant among all populations (P , 0.001). Thelevel of genotypic departure from Hardy–Weinberg

expectations (6%) was close to that expected bychance alone (i.e., 5%). The mean FST for intronloci was 0.05; HT ranged from 0.167 (GnRH) to0.455 (IK), averaging 0.267.

Variation among Populations

Patterns of differentiation were generally con-gruent among marker classes (Figures 2–4). Forexample, MDS plots of allozyme and microsatel-lite data showed distinct clusterings of the middleColumbia River ESU and Snake River B-run pop-ulations (Figures 2, 3). Although the Snake RiverA-run populations clustered together, both datapoints for the upper Columbia River ESU fell with-in this group for the allozyme and microsatellitedata sets on the first two axes. Only along the thirdMDS axis (which was based on microsatellite loci)was the upper Columbia River ESU distinct (Fig-ure 3). A scatter plot of PCA scores based on al-lozyme frequencies and an MDS plot based onNei’s genetic distance (Nei 1978) (neither graphshown) were highly congruent with the ESU re-lationships seen in the allozyme plot. The distri-bution of populations in the MDS plot based onintron variation was similar to the patterns fromthe other markers but showed less of a geographicpattern with respect to population differentiation(Figure 4).

Simulations

100% ESU analyses.—The Snake River B-rungroup was the most readily identified over all sim-ulations, and the upper Columbia River ESU hadthe lowest estimates (Table 3). Allozymes provid-ed the most accurate estimates for the middle Co-lumbia River ESU simulation (94%) and bothSnake River reporting groups (93%), whereas mi-crosatellite data had the highest estimate for theupper Columbia River ESU (83%). Over all sce-narios, estimates based on intron data ranged from47% to 77%. For each scenario, the all-loci dataset provided the most accurate estimates (Table 3).It is worth noting that the estimates made by meansof the standard algorithm procedure (i.e., SPAM,which does not accommodate rare alleles) weresubstantially lower (averaging only 72% for theall-loci analyses) and that 13–28% of the mixturegenotypes were identified as unknowns.

Realistic-mixture analyses.—Past GSI work inthe Bonneville Dam monitoring program and inthe Zone 6 bycatch surveys has shown that theSnake River ESU (runs A and B together) is thenumerically dominant ESU (Kassler et al. 2002).With these results in mind, we used values of 70,


FIGURE 2.—Multidimensional scaling plot of steelhead populations based on chord distances (Cavalli-Sforza andEdwards 1967) derived from 32 allozyme markers. Triangles represent evolutionarily significant units (ESUs) fromthe upper Columbia River, circles ESUs from the middle Columbia River, and squares ESUs from the Snake River(open squares refer to run A and closed squares to run B).

FIGURE 3.—Multidimensional scaling plot of steelhead populations based on chord distances (Cavalli-Sforza andEdwards 1967) derived from five microsatellite markers. See the caption to Figure 2 for an explanation of thesymbols used.

680 WINANS ET AL.

FIGURE 4.—Multidimensional scaling plot of steelhead populations based on chord distances (Cavalli-Sforza andEdwards 1967) derived from three intron markers. See the caption to Figure 2 for an explanation of the symbolsused.

TABLE 3.—Results of four simulated mixture scenarios consisting entirely of one evolutionarily significant unit (ESU)or other management group. Analyses were based on four sets of loci (introns, microsatellites, allozymes, and all data).Mean (SDs in parentheses) estimates of percent compositions were calculated from 1,000 bootstrapped mixtures with400 fish per mixture. Estimates were made for individual populations (n 5 32) and pooled by ESU or other managementgroup.

Genetic material Population

Management group

UpperColumbia

River

MiddleColumbia

River

A run,SnakeRiver

B run,SnakeRiver

Introns (3 loci, 6 alleles) Upper Columbia RiverMiddle Columbia RiverA run, Snake RiverB run, Snake River

47 (21)4 (4)3 (6)1 (2)

25 (19)73 (17)23 (16)5 (9)

23 (18)18 (18)63 (20)17 (14)

5 (6)5 (5)

11 (12)77 (15)

Microsatellites (5 loci, 68 alleles)

Allozymes (32 loci, 109 alleles)

Upper Columbia RiverMiddle Columbia RiverA run, Snake RiverB run, Snake RiverUpper Columbia RiverMiddle Columbia River

83 (5)2 (2)4 (3)0 (0)

61 (7)0 (0)

5 (3)90 (4)9 (4)3 (2)8 (4)

94 (3)

11 (4)8 (3)

85 (5)7 (3)

25 (7)6 (3)

1 (1)0 (1)2 (1)

90 (3)6 (3)0 (1)

All data (40 loci, 183 alleles)

A run, Snake RiverB run, Snake RiverUpper Columbia RiverMiddle Columbia RiverA run, Snake RiverB run, Snake River

1 (1)1 (1)

84 (4)0 (1)1 (1)1 (0)

4 (3)0 (1)4 (2)

95 (2)4 (2)0 (1)

93 (4)6 (3)

11 (4)5 (2)

94 (2)3 (2)

2 (2)93 (3)1 (1)0 (0)1 (1)

96 (2)


FIGURE 5.—Histograms of percent of mixture by steel-head population management group for three simula-tions (A–C) using four different marker sets (3 introns,5 microsatellites, 32 allozymes, and all loci).

80, and 90% for this ESU to determine how thethree marker classes would perform under morerealistic scenarios (Figure 5). Based on the pre-viously discussed simulations, it is not surprisingthat the microsatellite markers provided the bestestimates for the upper Columbia River ESU whenthe true values were 5% or 12% (note that theintron markers also did well; Figure 5a). Allozymedata provided the most accurate estimates for themiddle Columbia River ESU when it was presentat proportions of 5, 8, or 25%. In general, theintron and microsatellite markers underestimatedthe Snake River groups, whereas the allozyme datawere accurate or perhaps slightly positively biasedfor these two groups (Figure 5). Analyses usingall-loci yielded the best results for the middle Co-lumbia and Snake River groups.

Effect of Baseline Sample Size

Larger baseline samples produced better esti-mates of mixture proportions (Figure 6). However,

increasing sample sizes produced diminishing re-turns, and in most cases sampling more than 50individuals produced little improvement. Althoughthe five microsatellite loci used in this analysishave many more alleles per locus than the allo-zyme loci, the microsatellite loci did not requireor benefit from having larger sample sizes than theallozyme loci.

The mixtures that were composed entirely of fishfrom the upper Columbia River populations andthat were analyzed with allozyme loci showed thegreatest benefit from larger sample sizes. In thiscase, stock proportion estimates improved as base-line sample sizes increased beyond 200 individ-uals. It is not clear what aspects of the data effectdifferent relationships between discriminatorypower and sample size, but the relative indistinc-tiveness of the upper Columbia River collectionsfrom the Snake River collections with regard toallozyme variation is probably a factor.

Criteria for Selecting Subsets of Allozyme Loci

The three subsets of allozyme loci (plus the fivemicrosatellite loci) identified by FST, PCA, and HT

yielded similar results in the 100% middle Colum-bia River ESU and 100% Snake River A-run sce-narios; the estimates were 2–4% lower than theall-loci results (Table 4). However, the HT loci re-sults equaled the all-loci results for the 100% up-per Columbia River ESU simulation and were only1 percentage point less in the Snake River B-runsimulation (Table 4). Previous simulations showedthat a sample size of 50–100 individuals per pop-ulation was sufficient for baseline data (Figure 6).We reran the simulations using the HT allozymesplus the five microsatellite loci but assumed theallele frequencies were based on 100 fish per pop-ulation for the DNA markers. The estimate for asimulated mixture consisting entirely of upper Co-lumbia River ESU fish was 90%, 6 percentagepoints greater than the all-loci estimate. A simu-lation of the 100% Snake River B run led to a 98%estimate, 2 percentage points greater than the all-loci estimate.

Discussion

Relative to other species of Pacific salmon,steelhead have been shown to have high levels ofheterozygosity and moderate levels of genetic dif-ferentiation, with a strong geographic component(Busby et al. 1996; Scribner et al. 1998; Beachamet al. 1999). Here, new microsatellite data fromthe Columbia River basin corroborate earlier find-ings.

682 WINANS ET AL.

FIGURE 6.—Effects of eight sample sizes on estimates of mixture proportions in four simulated mixture scenarios.Mean estimates are based on 1,000 mixtures, each containing 400 fish.

As expected, microsatellite loci had a greaternumber of alleles per locus and larger HT valuesthan are typically observed at allozyme loci. In-trons had intermediate HT values and low allelicdiversity. Despite these differences in total vari-ation, there was little difference among markerclasses in the magnitude of population differen-tiation as estimated by FST, which ranged from0.041 (microsatellite loci) to 0.050 (intron loci).This result confirms findings of an earlier studyreporting that different classes of markers havesimilar FST values (Allendorf and Seeb 2000). Re-gardless of marker type, statistically significant

differences among populations were found for allloci. And with the exception of intron variability,genetic variability followed geographic patterns.The latter result has important practical conse-quences. It is unrealistic to expect to sample allcontributing source populations in a mixture forGSI analysis. It is also statistically unwieldy touse baselines that contain multilocus data for alarge number of populations (e.g., N 5 100–200).But when populations are genetically related bygeographic proximity, not all populations in anarea need to be included in the genetic baselineand ‘‘representative’’ populations that include ma-


TABLE 4.—Estimates comparable to those reported inTable 3 but based on allozyme loci (identified in Table 2)together with five microsatellite loci. Allozyme loci werechosen based on the values of FST and HT or because theyhad alleles with large weights on axes 1–3 in a principalcomponents analysis (PCA; Table 2). See Table 3 for ad-ditional details.

Criterion

UpperColumbia

River

MiddleColumbia

River

A run,SnakeRiver

B run,SnakeRiver

All dataFST

a

PCAa

HTa

HTb

84 (4)83 (5)83 (5)84 (5)90 (5)

95 (1)93 (2)93 (3)93 (3)

94 (1)90 (4)90 (4)90 (3)

96 (3)93 (3)94 (2)95 (2)98 (2)

a N ø 48 fish/locus/population.b N 5 100 fish/locus/population.

jor regional productivity will produce accurate re-gional GSI estimates (Beacham et al. 2001).

Genetic differences among steelhead popula-tions at allozyme and microsatellite loci are suf-ficient for accurate GSI estimation. In general, al-location to ESU is more than 85% correct, al-though it can be more or less accurate than thatdepending on the particular data set and target ESU(Table 3). Empirically determined biases (e.g., es-timates for the upper Columbia River ESU are like-ly to include individuals from the Snake RiverESU; Table 3) can be partially explained by theoverlapping genetic similarities seen in the MDSplots (Figures 2–4). It should be noted that we didnot evaluate the effect of year-to-year variation inmarker frequencies, which could be important inboth GSI discrimination and locus selection. Inthis regard, (Waples et al. 1993) reported manycases in which between-year differences withinpopulations were small compared with the differ-ences between populations (although exceptionswere noted).

Our initial results indicate that the greater thenumber of alleles used in a GSI analysis the betterthe resolution among ESUs. For example, the all-loci analyses yielded the most accurate estimatesof the single-composition simulations (Table 3).However, the gain was minimal in most cases com-pared with estimates using locus subsets. For ex-ample, the estimate for the 100% upper ColumbiaRiver ESU was 83% with microsatellite loci (63independent alleles) and 84% using all data (143independent alleles). For the 100% Snake River Brun, the results based on all data were only 3 per-centage points greater than the allozyme estimate(67 independent alleles; 93%). The simulationsbased on subsets of allozyme loci provide an in-

teresting perspective. We found that the results ob-tained with 10 ‘‘HT’’ allozyme loci plus the 5 mi-crosatellite loci (89 independent alleles) are com-parable to those obtained with all genetic data (Ta-ble 4). We also found that better results wereobtained using these same loci having assumedthat the microsatellite sample size was 100 fish(Table 4). Recent theoretical work has shown thatthe greater the number of alleles used in simulatedGSI analyses the better the estimates of stock pro-portions (Kalinowski, in press). This work is basedon allele distributions with similar distribution pat-terns. Our results suggest that the empirical pat-terns and levels of allelic variation may stronglyinfluence the resolution obtained for a particularallele set in a GSI application.

We evaluated whether the allozyme data setcould be simplified by identifying ‘‘important’’subsets of allozyme loci to use in mixture analyses.Of the three criteria, the loci selected by total het-erozygosity (HT) performed better in two of fourscenarios. This finding translates into dropping 22loci from the electrophoretic screening protocolwith a loss of only 2–4% accuracy for the abovescenarios. In contrast, Scribner et al. (1998) re-ported that the values of the likelihood ratio (G),FST, and HT failed to isolate critical loci for GSIanalyses of two management groups of chum salm-on in the Yukon River. Evaluations of variousmethodologies and statistics for identifying sub-sets of discriminatory loci will continue as GSIapplications increasingly use DNA-based loci withcomplex allele distributions and technical require-ments.

Are DNA-based loci better than protein-geneticmarkers for GSI? Perhaps. Here we show that thevariability in allozymes and microsatellites bearsa similar genetic signature with respect to the mi-croevolutionary events that produce spatial geneticrelationships among populations (Scribner et al.1998; Allendorf and Seeb 2000). From a practicalperspective, either or both data sets can be usedreliably for similar and accurate GSI estimations.From a technical perspective, DNA markers arebecoming more cost effective, diverse, and auto-mated. Importantly, extremely small tissue sam-ples can be taken from juveniles as well as adultfish nonlethally—a vast advance over protein tis-sue requirements. Still, there are attendant prob-lems with the technology. For example, standard-ization of allele scoring within and between lab-oratories may be more difficult for microsatellitemarkers than for allozymes due to the larger num-ber of alleles per locus. The large number of alleles

684 WINANS ET AL.

at microsatellite loci may also lead to other prob-lems. For example, it has been shown that lociwith more than 30 alleles can lead to serious ge-notyping errors and inconsistencies (O’Reilly etal. 1998). We also note that the future of steelheadGSI may not lie totally with microsatellites: lessdeveloped molecular technologies may providefurther easily automated procedures. For example,techniques to measure single nucleotide insertpolymorphisms can provide data for biallelicloci—and these techniques are amenable to ex-tremely high throughput for a large number of loci(Kalinowski, unpublished).

Which loci do we recommend? We suggest thatmixture analysis of steelhead in the Columbia Riv-er include both allozyme and microsatellite data.The detection of fish from the upper ColumbiaRiver ESU is best accomplished with microsatel-lite loci; the two other upriver ESUs (three man-agement units) are best resolved with a combi-nation of allozyme and microsatellite data. Non-lethally sampling adult fish for allozymes in thesetwo applications is not an issue, as an opercularpunch or fin clip is sufficient (Van Doornik et al.1999). The inconvenience of maintaining two fieldand laboratory protocols, however, will predict-ably end when a refined and expanded data set ofDNA-based loci is completed.

Acknowledgments

Stevan Phelps (WDFW) helped initiate the orig-inal research project and manage earlier portionsof the allozyme baseline. David Kuligowski(NMFS) performed a portion of the protein elec-trophoresis. Norm Switzler, Janine Rhone, Bill In-gram, Cherril Bowman, and Keith Sweeney par-ticipated in the data collection at WDFW. Bill Ho-pley (WDFW) and Paul Aebersold (NMFS) pro-vided administrative and technical support; andKathleen Neeley and Jennifer Burke (NMFS)helped with the illustrations. Reference to tradenames does not imply endorsement by NMFS.

References

Aebersold, P. B., G. A. Winans, D. J. Teel, G. B. Milner,and F. M. Utter. 1987. Manual for starch gel elec-trophoresis: a method for the detection of geneticvariation. NOAA Technical Report NMFS 61.

Allendorf, F. W., and L. W. Seeb. 2000. Concordanceof genetic divergence among sockeye salmon pop-ulations at allozyme, nuclear DNA, and mitochon-drial DNA markers. Evolution 54:640–651.

Avise, J. C. 1994. Molecular markers, natural history,and evolution. Chapman and Hall, New York.

Baker, J., P. Bentzen, and P. Moran. 2002. Molecular

markers distinguish coastal cutthroat trout fromcoastal rainbow trout/steelhead and their hybrids.Transactions of the American Fisheries Society 131:404–417.

Banks, M. A., M. S. Blouin, B. A. Baldwin, V. K. Rash-brook, H. A. Fitzgerald, S. M. Blankenship, and D.Hedgecock. 1999. Isolation and inheritance of nov-el microsatellites in Chinook salmon (Oncorhynchustschawytscha) (correction). Journal of Heredity90(3):U1–U1.

Beacham, T. D., J. R. Candy, K. J. Supernault, T. Ming,B. Deagle, A. Schulze, D. Tuck, K. H. Kaukinen,J. R. Irvine, K. M. Miller, and R. E. Withler. 2001.Evaluation and application of microsatellite and ma-jor histocompatibility complex variation for stockidentification of coho salmon in British Columbia.Transactions of the American Fisheries Society 130:1116–1149.

Beacham, T. D., S. Pollard, and K. D. Le. 1999. Pop-ulation structure and stock identification of steel-head in southern British Columbia, Washington, andthe Columbia River based on microsatellite DNAvariation. Transactions of the American FisheriesSociety 128:1068–1084.

Busby, P. J., T. C. Wainwright, G. J. Bryant, L. J. Lier-heimer, R. S. Waples, F. W. Waknitz, and I. V. La-gomarsino. 1996. Status review of West Coast steel-head from Washington, Idaho, Oregon, and Cali-fornia. NOAA Technical Memorandum NMFS-NWFSC-27.

Cavalli-Sforza, L. L., and A. W. F. Edwards. 1967. Phy-logenetic analysis: models and estimation proce-dures. Evolution 21:550–570.

Debevec, E. M., R. B. Gates, M. Masuda, J. Pella, J. J.Reynolds, and L. W. Seeb. 2000. SPAM (version3.2): Statistics program for analyzing mixtures.Journal of Heredity 91:509–511.

Hartl, D. L. 1981. A primer of population genetics. Sin-auer Associates, Sunderland, Massachusetts.

Kalinowski, S. T. In press. Genetic polymorphism andmixed stock fisheries analysis. Canadian Journal ofFisheries and Aquatic Sciences.

Kassler, T. W., D. Rawding, A. Marshall, B. Baker, andJ. B. Shaklee. 2002. Genetic mixed-stock analysisof Columbia River steelhead at Bonneville Dam andin the Zone 6 fishery, 1997–2001. Washington De-partment of Fish and Wildlife, Report to NorthwestFisheries Science Center, Seattle.

Lewontin, R. C. 1991. Twenty-five years ago in genet-ics: electrophoresis in the development of evolu-tionary genetics—milestone or millstone? Genetics128:657–662.

McConnell, S. K., P. O’Reilly, L. Hamilton, J. M.Wright, and P. Bentzen. 1995. Polymorphic micro-satellite loci from Atlantic salmon (Salmo salar):genetic differentiation of North American and Eu-ropean populations. Canadian Journal of Fisheriesand Aquatic Sciences 52:1863–1872.

Millar, R. B. 1987. Maximum likelihood estimation ofmixed-stock fishery composition. Canadian Journalof Fisheries and Aquatic Sciences 44:583–590.

Nei, M. 1978. Estimation of average heterozygosity and


genetic distance from a small number of individuals.Genetics 89:583–590.

O’Reilly, P. T., C. Herbinger, and J. M. Wright. 1998.Analysis of parentage determination in Atlanticsalmon (Salmo salar) using microsatellites. AnimalGenetics 29:363–370.

Park, L. K., and P. Moran. 1994. Development in mo-lecular genetic techniques in fisheries. Reviews inFish Biology and Fisheries 4:272–299.

Phelps, S. R., B. M. Baker, and C. A. Busack. 2000.Genetic relationships and stock structure of YakimaRiver basin and Klickitat River basin steelhead pop-ulations. Washington Department of Fish and Wild-life, Fisheries Management Program, Progress Re-port, Olympia.

Phelps, S. R., B. M. Baker, P. L. Hulett, and S. A. Leider.1994. Genetic analysis of Washington steelhead:initial electrophoretic analysis of wild and hatcherysteelhead and rainbow trout. Washington Depart-ment of Fish and Wildlife, Fisheries ManagementProgram Report 94-9, Olympia.

Phelps, S. R., S. A. Leider, P. L. Hulett, B. Baker, andT. Johnson. 1996. Genetic analyses of Washingtonsteelhead: preliminary results incorporating 36 newcollections from 1995 and 1996. Washington De-partment of Fish and Wildlife, Fisheries Manage-ment Program, Progress Report, Olympia.

Rannala, B., and J. L. Mountain. 1997. Detecting im-migration by using multilocus genotypes. Proceed-ings of the National Academy of Sciences of theUnited States of America 94:9197–9201.

Rawding, D., S. Phelps, A. Marshall, and C. W. Hopley.1997. Genetic stock identification of steelhead inColumbia River Zone 6 fishery and at BonnevilleDam. Washington Department of Fish and Wildlife,Annual Report of Research to the Northwest Fish-eries Science Center, Seattle.

Rohlf, F. J. 1994. NTSYS-pc: numerical taxonomy andmultivariate analysis system, version 1.80. ExeterSoftware, Setauket, New York.

Scribner, K. T., P. A. Crane, W. J. Spearman, and L. W.Seeb. 1998. DNA and allozyme markers provideconcordant estimates of population differentiation:analyses of U.S. and Canadian populations of Yu-kon River fall-run chum salmon (Oncorhynchusketa). Canadian Journal of Fisheries and AquaticSciences 55:1748–1758.

Scribner, K. T., J. R. Gust, and R. L. Fields. 1996. Iso-lation and characterization of novel salmon micro-satellite loci: cross-species amplification and pop-ulation genetic applications. Canadian Journal ofFisheries and Aquatic Sciences 53:833–841.

Seeb, L. W., and P. A. Crane. 1999. Allozymes andmitochondrial DNA discriminate Asian and NorthAmerican populations of chum salmon in mixed-stock fisheries along the south coast of the AlaskaPeninsula. Transactions of American Fisheries So-ciety 128:88–103.

Swofford, D. L., and R. B. Selander. 1981. BIOSYS-1:a FORTRAN program for the comprehensive anal-ysis of electrophoretic data in population geneticsand systematics. Journal of Heredity 72:281–283.

TAC (Technical Advisory Committee). 1997. ColumbiaRiver fish management plan: 1996 all-species re-view. Technical Advisory Committee, U.S. v.Oregon, Report 222, Portland, Oregon.

Teel, D. J., P. A. Crane, C. M. Guthrie III, A. R. Marshall,D. M. Van Doornik, W. D. Templin, N. V. Varnav-skya, and L. W. Seeb. 1999. Comprehensive allo-zyme database discriminates Chinook salmonaround the Pacific Rim. North Pacific AnadromousFish Commission, Document 440, Anchorage, Alas-ka.

Van Doornik, D. M., G. A. Winans, and D. J. Teel. 1999.Allozyme studies of Pacific salmonids with nonle-thal sampling of fin tissues. North American Journalof Fisheries Management 19:678–686.

Waples, R. S., O. W. Johnson, P. A. Aebersold, C. K.Shiflett, D. M. Van Doornik, D. J. Teel, and A. E.Cook. 1993. A genetic monitoring and evaluationprogram for supplemented populations of salmonand steelhead in the Snake River basin. Annual Re-por of Research to the Bonneville Power Admin-istration, Portland, Oregon.

Wilmot, R. L., C. M. Kondzela, C. M. Guthrie, III, A.Moles, J. J. Pella, and M. Masuda. 2000. Originsof salmon seized from the FV Arctic Wind. AukeBay Laboratory, Alaska Fisheries Science Center,Juneau.

Winans, G. A., D. Viele, A. Grover, M. Palmer-Zwahlen,D. Teel, and D. M. Van Doornik. 2001. An updateof genetic stock identification of Chinook salmonin the Pacific Northwest: test fisheries in California.Reviews in Fisheries Science 9:213–237.

genetic stock identiﬁcation of steelhead in the columbia ......674 winans et al. figure 1.—map...

Documents