Link - Simon Fraser University

Integrative and Comparative Biology, volume 50, number 4, pp. 643–661doi:10.1093/icb/icq068SYMPOSIUMIt’s About Time: Divergence, Demography, and the Evolutionof Developmental Modes in Marine InvertebratesMichael W. Hart 1,* and Peter B. Marko 2,†*Department of Biological Sciences, Simon Fraser University, Burnaby, BC V5A 1S6, Canada; † Department of BiologicalSciences, Clemson University, Clemson, SC 29634, USAFrom the symposium ‘‘Evolutionary Paths Among Developmental Possibilities: A Symposium Marking the Contributionsand Influence of Richard Strathmann’’ presented at the annual meeting of the Society for Integrative and ComparativeBiology, January 3–7, 2010, at Seattle, Washington.1 E-mail: mwhart@sfu.ca2 E-mail: pmarko@clemson.eduSynopsis Differences in larval developmental mode are predicted to affect ecological and evolutionary processes rangingfrom gene flow and population bottlenecks to rates of population recovery from anthropogenic disturbance and capacity forlocal adaptation. The most powerful tests of these predictions use comparisons among species to ask how phylogeographicpatterns are correlated with the evolution and loss of prolonged planktonic larval development. An important and largelyuntested assumption of these studies is that interspecific differences in population genetic structure are mainly caused bydifferences in dispersal and gene flow (rather than by differences in divergence times among populations or changes ineffective population sizes), and that species with similar patterns of spatial genetic variation have similar underlying temporaldemographic histories. Teasing apart these temporal and spatial patterns is important for understanding the causes andconsequences of evolutionary changes in larval developmental mode. New analytical methods that use the coalescent historyof allelic diversity can reveal these temporal patterns, test the strength of traditional population-genetic explanations forvariation in spatial structure based on differences in dispersal, and identify strongly supported alternative explanations forspatial structure based on demographic history rather than on gene flow alone. We briefly review some of these recentanalytical developments, and show their potential for refining ideas about the correspondence between the evolution oflarval developmental mode, population demographic history, and spatial genetic variation.IntroductionStrathmann (1985, 1990) argued that the physicaland biological chemistry of seawater has favoredunique adaptations in the life histories of benthicmarine animals, most notably the planktonic fertilizationof small, nutrient-poor eggs that subsequentlygrow (as well as develop) in the plankton as specializedfeeding larval forms before returning as largejuveniles to the benthic habitat used by adults.Such planktotrophic larvae occur in species of mostmajor marine animal lineages and in most marinecommunities, often living alongside closely relatedspecies that have evolved various combinationsof internal fertilization, lecithotrophic nutrition, simplifiedmorphogenesis, and benthic development.Subsequent efforts to understand the adaptivesignificance of such variation in developmentalmodes (Moran and Emlet 2001; Emlet and Sadro2006; Thiyagarajan et al. 2007; Byrne et al. 2008;Marshall and Keough 2009) have arisen largelyfrom Strathmann’s fundamental question of whetheror not the diversity of reproductive strategies andlarval modes of marine species primarily reflect theenergetic and evolutionary trade-offs between thesize and number of offspring. Understanding whetherthe benefits of producing many small planktotrophicoffspring are in fact balanced by losses tomortality during an extended larval period remainsa challenging issue for the study of larval ecology andevolution, and Richard Strathmann’s research, focusedin large part on characterizing the adaptationsof marine larvae to their planktonic environment,has been primarily responsible for creating aDownloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010Advanced Access publication June 16, 2010ß The Author 2010. Published by Oxford University Press on behalf of the Society for Integrative and Comparative Biology. All rights reserved.For permissions please email: journals.permissions@oxfordjournals.org.

644 M. W. Hart and P. B. Markocomprehensive framework for understanding thetrade-offs between larval transport and larval losses.A key feature of this framework is Strathmann’ssuccessful argument that planktonic larvae are generallyoverdispersed relative to the distribution ofsuitable benthic habitat, and that individual adultsgain few (if any) short-term benefits from the obligatelong-distance planktonic dispersal of their offspring(Strathmann 1985). However, most larvalbiologists would probably agree that variation inlarval developmental modes, and the expected variationin gene flow between populations that arisesfrom the dispersal potential of different developmentalmodes, has important consequences overlonger ecological and evolutionary timescales. Forthis reason, population geneticists have focused onmarine species as model systems for understandingthe impacts of life-history variation on realized dispersalor gene flow between populations. An importantreason for the focus on life-history correlates ofdispersal is the extraordinary expected variance inlarval dispersal potential among species with differentmodes of development and potentially among conspecificlarvae with the same mode of development.In species with slow-growing planktotrophic larvae,siblings are predicted to diffuse large distances fromeach other with a high variance in dispersion fromthe natal habitat, whereas offspring with benthic developmentare predicted to recruit to their natal habitatalong with many of their siblings. Other thingsbeing equal, such differences in mode of developmentand larval dispersal potential are expected tostrongly affect important demographic measures includingfunctional sex ratio, effective population size,gene flow, population persistence and colonization(i.e., metapopulation dynamics), probability of extinction,and rate of speciation. Most recently, conservationbiologists (Fortuna et al. 2009) havefocused on life-history differences as predictors ofpopulation connectivity, metapopulation dynamics,and capacity for recovery from, or local adaptationto, disturbance by humans.Is variation in larval developmental mode (and ingene flow by larval dispersal) the prime determinantof genetic drift and population differentiation?Marine phylogeographers have put extraordinaryeffort into answering this question through the analysisof the spatial distribution of genetic variationwithin and between populations. Because the effectsof developmental mode act in the context of theoverall phenotype of the organism and the historicalbiogeographical context of its environment, the mostpowerful tests of the relative effect of evolutionarychanges in mode of development on populationgenetic variation have used comparisons betweenclosely related species that share otherwise similarphenotypes, or members of the same community thatshare a similar biogeographical context (Bohonak1999).Starting with the earliest comparative studies ofmarine population genetic structure (Berger 1973),a steady stream of comparative analyses has yieldedresults consistent with the predicted effects of differentmodes of larval development (regardless of otherphenotypic or biogeographical effects), such as strongerpopulation differentiation, smaller effective populationsize, or striking phylogeographic breaks inspecies with lower larval dispersal potential (e.g.,Janson 1987; Waples 1987; McMillan et al. 1992;Duffy 1993; Hunt 1993; Shulman and Bermingham1995; Hellberg 1996; Arndt and Smith 1998; Toddet al. 1998; Kyle and Boulding 2000; Collin 2001;Watts and Thorpe 2006; Sherman et al. 2008; forreviews see Gooch 1975; Crisp 1978; Burton andFeldman 1981; Palumbi 1994; Bohonak 1999;Kinlan and Gaines 2003). At the same time, however,there has also been a slow accumulation of a numberof exceptions in which either unexpectedly stronggenetic differentiation has been found over relativelysmall geographic scales in species with planktoniclarvae (Koehn et al. 1980; Barber et al. 2000;Buonaccorsi et al. 2002; Taylor and Hellberg 2003a;Sotka et al. 2004; Marko and Barr 2007; Marko et al.2007) or an absence of genetic subdivision over largeareas has been found in species without a planktoniclife-history stage (Kyle and Boulding 2000; Marko2004; Ayre et al. 2009; for reviews see Burton andFeldman 1982; Burton 1983; Cunningham andCollins 1998). For the most part, strong differentiationin species with planktonic larvae has usuallybeen attributed to natural selection acting directlyon genetic markers (Koehn et al. 1980; Sotka et al.2004), nearshore oceanographic processes (Gilgand Hilbish 2003; Sotka et al. 2004; Marko andBarr 2007; Banks et al. 2007), habitat specificity(Ayre et al. 2009), or larval behavior (Warner andPalumbi 2003) whereas genetic homogeneity inspecies lacking broadly dispersing planktonic larvaecould reflect recent population expansions (Edmands2001; Marko 2004), stabilizing selection (Karl andAvise 1992), or unexpected dispersal capability(e.g., by rafting of adults, juveniles, or encapsulatedembryos attached to mobile substrates such as birdsor kelps) (Highsmith 1985; Helmuth et al. 1994).Although the degree of differentiation (or absenceof differentiation) in many of these cases is striking,it is important to note that many of these exceptionalstudies involve single species (a situationDownloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

Divergence genetics and modes of development 645analogous to an uncontrolled ecological experiment),and therefore lack the inferential power inherent incomparative studies. Not surprisingly, opinions differwith respect to what exactly the expected patternsof neutral genetic differentiation should be for anyparticular species (Collin 2003; Taylor and Hellberg2003b; Palumbi and Warner 2003; Warner andPalumbi 2003); follow-up comparative studies involvingsome of these exceptional cases have beenvery helpful in distinguishing between contemporaryand historical causes of phylogeographic patterns ofdifferentiation (Taylor and Hellberg 2006).One potential solution to the mixed evidence forcorrelated evolution of developmental mode and spatialpatterns of genetic variation (e.g., Bohonak 1999or Kinlan and Gaines 2003 compared to Ayre et al.2009) is to refocus marine phylogeographic analyseson identifying the biogeographic and demographicprocesses responsible for generating different phylogeographicalpatterns among species, particularlyamong those differing in mode of larval development.For the most part, large phylogeographicbreaks, or large pair-wise measures of populationdifferentiation, are typically interpreted in terms ofgeographically localized limits on gene flow (Cowenand Sponaugle 2009; Pelc et al. 2009). Although reducedgene flow in itself is a valid hypothesis thatcan explain patterns of population genetic subdivision,genetic variation within and between populationsat selectively neutral loci, of course, evolvesunder the combined effects of mutation, geneticdrift, and gene flow acting through time.Phylogeography has largely relied on problematicpost hoc interpretations (Carstens et al. 2009) toidentify spatial patterns that are consistent with hypothesesof geographical or taxonomic variation inone or more of these process-based parameters. Animportant recent source of insight into this problem,which promises to help overcome the limitations ofpost hoc interpretation, is the development of coalescentmethods for the joint characterization of demographicparameters and for testing hypotheses abouttheir magnitude and their significance for explainingpatterns of spatial genetic variation. These new methodsprovide a more general context for marine phylogeographyin which spatial patterns of geneticvariation arise through a birth–death process of mutation,vicariance, genetic drift, and dispersal, alljointly modeled using the coalescent. This frameworkcan be used to ask whether effective population sizes,rates of gene flow, and times of population divergenceare predicted by variation in mode of developmentand larval dispersal. In particular, specieswith similar dispersal potential and similarphylogeographical patterns are expected to have similarunderlying histories of gene flow, changes in effectivepopulation size, and population divergencetime. Conversely, species with different dispersal potentialand different phylogeographical patterns ofspatial variation observed in contemporary populationsare expected to differ mainly in the magnitudeof migration and gene flow between populations,rather than in their demographic histories of populationdivergence and their variation in effective populationsize. Testing these expectations requires thejoint estimation of gene flow (via high or low ratesof larval dispersal) along with other parameters(population divergence time, effective populationsize) that might contribute substantially to the spatialpattern of differentiation in comparison to interpretationsthat emphasize gene flow and larval-dispersalpotential. Characterizing these temporal patterns ofpopulation genetic variation, and confirming thatsuch temporal patterns covary in a predictable waywith mode of development, would greatly bolster theargument for the primacy of larval dispersal potentialas the main determinant of genetic variation amongpopulations in the oceans. Alternatively, if these predictionsare often rejected, then the coevolution ofpopulation genetic variation and larval dispersal maybe much weaker than has been suggested by comparativephylogeographical studies of spatial variationalone, and other phenotypic or historicaleffects may be much more important in shapingthe magnitude of neutral genetic differentiation andits evolutionary corollaries. Here, we briefly reviewthe development of these methods, give examples oftheir recent application in comparative studies ofmarine invertebrates, and look ahead to the prospectsfor fully using such methods to understandthe evolution of diverse larval forms and phylogeographicpatterns.Genetic differentiation in the sea: isit always all about gene flow?The large majority of comparative marine phylogeographicstudies have characterized spatial patternsof population structure using analogs of Wright’sF-statistics. Although gene flow or migration canbe estimated from F ST in several ways (Wright1951; Felsenstein 1976; Slatkin 1985), this inferencerequires several notorious simplifying assumptions,such as constant and equal population sizes, symmetricalrates of migration, and population allele frequenciesthat are in a dynamic equilibrium betweengene flow and genetic drift (i.e., that the F ST statisticitself has reached equilibrium). Each of theseDownloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

646 M. W. Hart and P. B. Markoassumptions has been widely criticized for lackingbiological realism in many natural situations, andviolation of these assumptions can potentially severelybias estimates of migration from F ST (Burtonand Feldman 1982; Bossart and Prowell 1998;Cunningham and Collins 1998; Waples 1998; Beerliand Felsenstein 1999, 2001; Whitlock and McCauley1999; Neigel 2002). The equilibrium assumption, forexample, is probably wrong for many species in temperateregions that have undergone large polewardrange extensions and massive demographic expansionssince the end of the last glacial maximum10,000 years ago. The time required (in generations)for F ST to reach equilibrium may be muchgreater than this, depending on migration rates andeffective population sizes (Crow and Aoki 1984;Whitlock 1992; Whitlock and McCauley 1999). Theimpact of violating this assumption depends on thehistory of isolation between populations; estimates ofmigration between populations will be biasedupward if populations have very recently separatedbut will be biased downwards if populations havenewly come into contact. Comparative studies thatattempt to use F ST -based methods such as analysis ofmolecular variance (AMOVA; Excoffier et al. 2005)to account for phylogeographic differences amongmarine species in terms of mode of larval developmentand rate of gene flow are thus potentially confoundedby violations of the equilibrium assumptiongiven that the approach to equilibrium depends primarilyon effective population size, rates of geneflow, and metapopulation dynamics (population divergencetimes), all factors expected to differ amongspecies with different modes of development.Although all of the variants of F ST remain usefulways to summarize patterns of population differentiation(Bohonak 1999; Whitlock and McCauley1999; Neigel 2002), estimates of migration fromnew analytical methods, largely those based uponneutral coalescent theory, have slowly been incorporatedinto marine phylogeographic studies becausethese newer methods have several desirable features(Kuhner 2009). Arguably the most important improvementin these methods is that they providejoint estimates of the population genetic parametersthat collectively produce spatial patterns of differentiation,especially effective population size, migration,population growth rates, and divergence time. Thisapproach is generally superior because it avoids posthoc interpretation of spatial patterns (i.e., F ST ) interms of unobserved demographic processes (i.e.,asymmetrical gene flow, ancient vicariance, recentrange extension, population expansion, or bottlenecks),and instead directly estimates parametersassociated with those demographic processes. LikeF ST -based approaches, both the precision and accuracyof coalescent estimates of these populationparameters are greatly improved by (or even criticallydependent on) combined analysis of multiple loci insingle datasets that use information from the variancein coalescent times across loci (Edwards andBeerli 2000).For datasets with multiple loci, this coalescent approachhas mainly been implemented using one ofthree methods implemented in user-friendly softwarepackages running on fast publicly-accessible computerclusters: MIGRATE-N, LAMARC, and IMA (andthe earlier related programs IM and MDIV). All ofthese methods provide maximum likelihood estimatesof population genetic parameters across agroup of likely gene trees simulated with MarkovChain Monte Carlo (MCMC) sampling of genealogies,and are thus dubbed ‘‘coalescent samplers’’(Kuhner 2009). Using a random gene-tree topologyas a starting point, MCMC methods repeatedly makesmall arbitrary topological changes, and assess thelikelihood of the resulting genealogy at each stepwithin a model of neutral coalescence backwards intime. Highly likely gene trees are retained and populationgenetic parameters are estimated across theentire sample of trees. Parameter estimation across alarge sampling of trees (rather than from a single‘‘best’’ tree) is important, given the high degreeof uncertainty in gene-tree topologies within singlespecies or between closely related species. MCMCsampling can be guided by either a likelihood orBayesian approach in MIGRATE-N and LAMARC, butonly a Bayesian framework is available with IMA.The main differences among these methods (andthat pertain most to our review) are the populationparameters that each method estimates, and thuswhat each method does and does not assumeabout the history of the population. All three methodsestimate the population genetic parameter andseparate migration rates (M ¼ N e m, where N e is theeffective population size and m is the probability ofmovement by an individual from one population tothe other per generation) between populations inboth directions (bidirectional migration). The parameter, equivalent to 4N e m (where m is the mutationrate), is essentially a proxy for relativepopulation size given that mutation rates can reasonablybe assumed to be equal among populationswithin a single species. Both MIGRATE-N andLAMARC make these estimates for a group of npopulations but IMA is limited to the analysis of asingle pair of populations. MIGRATE-N and LAMARCcan accommodate multiple populations becauseDownloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

Divergence genetics and modes of development 647Fig. 1 The isolation-with-migration model for populationhistories that differ in the magnitude of gene flow, divergencetime, or effective population size but have similar levelsof differentiation between the population samples.both methods assume that population dynamics havebeen stable for 4N e generations, meaning that botheffectively assume that shared polymorphisms betweenpopulations are only due to gene flow, andthat gene flow and genetic drift are in equilibrium.Therefore, MIGRATE-N and LAMARC are most appropriatefor cases in which populations are not expectedto share ancestral polymorphisms. In contrast, IMA isideally applied to pairs of population that have divergedrecently because, in addition to M and ,IMA also estimates for the ancestral populationplus the time of divergence (t) of that ancestral populationinto the two sampled descendant populations;this combination of parameters allows IMAto infer whether shared alleles between populationsrepresent either ancestral polymorphisms or recentgene flow (Hey and Nielsen 2004, 2007). In theory,IMA can therefore potentially discriminate betweensubstantially different population histories (Fig. 1)that may show similar amounts of genetic differentiation.Because the IMA method can be applied toonly two populations at a time (the isolationwith-migrationmodel was originally developed toanalyze recent speciation events), it assumes thatno other populations have influenced thedivergence of the sampled populations. The recentlyreleased IMA 2.0 (Hey 2010a) can accommodatemultiple ‘‘populations,’’ but because it requires anexplicitly defined phylogeny for those populations,its intended use is for groups of closely related taxawhose phylogenetic relationships are known, or canbe reliably inferred, from other methods (Liu andPearl 2007; Liu 2008; Hey 2010b).LAMARC has the option of adding a populationgrowthparameter (g) so that either positive or negativeexponential growth can be simulated (Kuhner2006); likewise, IM (the predecessor to IMA) hasa splitting parameter that allows for growth andunequal division of the ancestral population (thisparameter is not included in IMA) (Hey 2007).LAMARC can also estimate the per-site recombinationrate, which allows the user to account for intralocusrecombination between alleles. Alternatively (forother methods), contiguous sequence data from individualnuclear genes must be first tested for evidenceof intragenic recombination and trimmed(if necessary) to single blocks of nonrecombiningsequences for use. The more recently released programMIMAR (Becquet and Przeworski 2007), whichis very similar to the IMA family of programs, can accommodateintragenic recombination, which allowsthe use of longer nuclear sequences, and presumablyleads to less uncertainty in individual gene trees.To illustrate what we perceive to be some of theimportant differences among these methods, we havecompared the results from different analyses usingpreviously published data from the bay scallop,Argopecten irradians (Marko and Barr 2007). Bayscallops live along the southeast coast of the USAin seagrass beds found within semi-enclosed lagoonalbasins or sounds that are connected to shelf watersby gaps and inlets between adjacent barrier islands.We have re-analyzed mtDNA sequence data fromtwo adjacent basins (Bogue Sound and BackSound) that are connected by a deep-water channel,but are potentially isolated from each other by thepattern of tidal circulation through the inlet, whichcreates a characteristic hydrodynamic ‘‘wall’’ that isexpected to limit the exchange of water and veligerlarvae across the inlet (Stommel and Farmer 1952;Zimmerman 1981; Luettich et al. 1998, 1999; Brownet al. 2000; Hench et al. 2002). In the original treatmentof the data, migration rates were inferred fromF ST and with MIGRATE-N (using a maximum likelihoodsearch), but we have repeated the MIGRATE-Nanalysis using Bayesian inference and have also usedIMA. The estimate of gene flow (N e m) betweenBogue and Back Sounds from F ST was 33 migrantsper generation (Fig. 2), an inference that assumessymmetric or equal migration rates, equal populationsizes, plus a drift-migration equilibrium. OurMIGRATE-N analysis suggested that this assumptionwas probably not valid: both and N e m werehighly asymmetrical, with much higher immigrationinto the much larger Back Sound population (assumingan old population divergence followed by attainmentof an equilibrium between drift and gene flow).However, our IMA analysis suggested in turn thatthese latter assumptions are also probably notvalid, and gave qualitatively different inferencesabout the origin of the modest spatial differentiationbetween Bogue and Back Sounds; IMA estimatedgene flow to be zero in both directions and, mostDownloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

648 M. W. Hart and P. B. MarkoFig. 2 Different insights into demographic processes (especially gene flow, N e m)fromF ST (left), MIGRATE-N (center), or IMA (right)analyses of bay scallop mtDNA variation from the Pamlico Sound region of coastal North Carolina, USA. Data from Marko and Barr(2007). The MIGRATE analysis used search methods like those described by Marko and Barr (2007) but with the faster Bayesian estimator.The IMA analyses used methods like those in Marko et al. (2010), run on the BioHPC clusters at Cornell University. We carried out aseries of analyses in M mode to optimize mixing and to find a combination of search parameters and prior distributions that gavehigh-density coverage of the nonzero parts of the posterior distribution of demographic parameter values. The saved gene trees werethen used in L mode to test nested sets of demographic models less complex than the full model including three different values andtwo nonzero migration rates. We report parameter estimates in demographic units (population size as number of individuals, divergencetime as years) for the longest of multiple runs that used the best-fit model. We used several standard biogeographical estimatesof mtDNA mutation rates to convert the parameter estimates to demographic units. We used the 90% highest posterior density (HPD)to characterize the confidence interval around each parameter estimate.surprisingly, estimated relative population size to beseveral orders of magnitude larger than estimatesfrom MIGRATE-N. In short, MIGRATE-N inferred high(but asymmetrical) gene flow between small populationswhereas IMA inferred the opposite pattern of nomigration between very large populations that divergedfrom each other during the early Wisconsinglacial period (Fig. 2). Although we cannot be sure,the discrepancies between MIGRATE-N and IMA mayreflect a key difference between the two methods: ifthe separation of two populations was sufficientlyrecent (Fig. 2) that they still retain many sharedancestral polymorphisms, then estimates of geneflow and population size will be biased upwardsand downwards, respectively, by MIGRATE-N giventhat the method assumes that all shared haplotypesare due to gene flow.Examples like the bay scallop highlight both thegeneral advantages of the coalescent approach to understandingprocesses underlying spatial variationand the potential advantages of joint estimation ofall three types of historical parameters (, M, t).Currently the best approach to this type of problemfor recent population divergences appears to us to bethe IMA method because it includes all three criticalpopulation parameters. The main drawback to usingIMA is its limitation to single pairs of populations:gene flow from unsampled populations may bias estimatesof parameters for the sampled populations(Strasburg and Rieseberg 2010), as it will withmost other analytical approaches (Beerli 2004).Most researchers have employed a variety of strategiesto overcome this limitation (depending on thespatial distribution of populations), such as conductingseparate analyses for all adjacent samples or conductingall pair-wise analyses and then makingqualitative inferences about the overall patterns.Other approaches to this problem may soon deliveruser-friendly software that can rapidly and accuratelyestimate the same range of parameters for more thantwo populations or offer other conceptual advantages(MSBAYES: Hickerson et al. 2006; Hickerson andMeyer 2008; BEST: Liu 2008; POPABC: Lopes et al.2009; IMA 2.0: Hey 2010a, 2010b) and thus overcomethe most significant limitation of the IMAapproach. However, most of these newer programsare aimed either at phylogenetic problems withwell-defined populations or at species in which thegenealogies of the populations are known (i.e., IMA2.0) or migration can be assumed to be a negligiblefactor (i.e., MSBAYES, POPABC, BEST). Modeling populationvicariance with gene flow for multiple populationsis a complex problem that we can only hopeto be tackled in the future.Although the rate of application of IMA and othercoalescent samplers is rapidly increasing overall,many published studies focus on single-species datasetsof mtDNA sequences. So far, there are too fewapplications that analyze two or more sequenced lociin a comparative context, so it is not yet possibleDownloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

Divergence genetics and modes of development 649to ask how often, or under what circumstances, thedemographic history of marine invertebrate populationsis consistent with predictions about patterns ofspatial genetic differentiation based on similarities ordifferences in the mode of development and larvaldispersal. Instead, we review a few such recent studies(including some of our own) with the aim of (1)highlighting the unexpected and exciting insightsavailable from such analyses, and (2) encouragingmore larval ecologists to become marine phylogeographersand analyze genetic and developmental variationwithin the IMA framework.Congruent patterns of temporaldemography and mode of dispersalBecause coalescent analyses could reveal different demographichistories among species with similar levelsof spatial differentiation, or, alternatively, could accountfor differences among species in genetic diversityand differentiation, either through differences indivergence times, effective population sizes, or migrationrates, the insights from coalescent analysesseem most likely to erode the strong support forthe correlated evolution of larval forms and populationstructure evident in F ST -based and AMOVAbasedapproaches. However, some studies find verystrong similarities between patterns of spatial variationand underlying demographic histories. Manysuch results from comparative coalescent studieswould tend to reinforce rather than erode supportfor dispersal as the main determinant of spatial geneticstructure in marine communities. One creativeexample used mtDNA sequences from large samplesof a snail species (Littorina littorea) and its hostspecifictrematode parasite (Cryptocotyle lingua).These snail populations in the northwesternAtlantic are probably descended from northeasternAtlantic ancestral populations, but the mechanismand history of their range extension to the westernAtlantic have been controversial for more than 100years: by historical human-mediated introductionsince the European colonization of North America,or by an earlier nonanthropogenic mode of dispersal.Blakeslee et al. (2008) found similar spatial distributionsof mtDNA variation across the North Atlanticboth in host and parasite, and then used IM to testthe hypothesis of ancient versus historical divergencetime between North American and European populationsof the snail. They reported a single-locusestimate of t500 years, a result that is consistentwith a historical anthropogenic origin (and invasivestatus) for L. littorea in North America. Blakesleeet al. (2008) reported a similar divergence timebetween eastern and western trematode populationsfrom the same locations. The results confirm theexpected correspondence between similarities in dispersalpotential and spatial genetic differentiation betweentwo species that have a similar underlyingdemographic history. We reanalyzed the samemtDNA sequence data for both snails and trematodesin IMA: although we found population divergencetimes much older than any of the Holocenehuman colonizations of North America (Fig. 3), thetrans-Atlantic divergence times were similar forthe snail host and its trematode parasite. The olderdivergence times from IMA analyses agree with theconclusions of one of the earliest comparativepopulation genetic studies, in which Berger (1977)interpreted fixed allozyme differences at seven of12 loci between eastern and western Atlantic populationsas indications of relatively ancienttrans-Atlantic divergence in L. littorea (withFig. 3 Posterior probability distributions of population divergencetimes from IMA analysis of mtDNA of northwestern andnortheastern Atlantic populations of a marine gastropod (Littorinalittorea, black) and its obligate trematode parasite (Cryptocotylelingua, gray). Data from Blakeslee et al. (2008), methods asdescribed in Figure 2. The mode of each posterior distributionindicates the most probable population divergence time betweeneastern and western populations. Because we used uniform priordistributions, these most probable parameter estimates are alsomaximum likelihood estimates (MLE). The estimates for snailsand for trematodes are considered to be not significantlydifferent from each other because the MLE of divergence time forthe snail host falls well within the confidence interval around theMLE of divergence time for the trematode parasite.Downloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

650 M. W. Hart and P. B. Markoplanktonic larvae) compared to much more modestallozyme divergence across the North Atlantic in twoother Littorina species that lack planktonic larval dispersal.Later unpublished studies failed to replicateBerger’s result for L. littorea, and suggest thatBerger’s observations of trans-Atlantic allozyme differentiationin that species may have been flawed (seeCunningham 2008). However, none of these studiesstrongly support a geologically recent introduction ofNorth American populations of L. littorea associatedwith humans migrating from Europe. We are uncertainwhy the two methods (IM versus IMA) give suchdifferent absolute divergence times both for L. littoreaand C. lingua; data from nuclear loci for bothspecies might be useful in resolving the disagreementbetween the two results, but seem unlikely to changethe conclusion that the snail and trematode had similartrans-Atlantic population histories. Such hostparasiteor host-symbiont analyses seem to havegreat potential for testing the strength of the predictedrelationship between spatial genetic differentiationand underlying demographic history in whichdispersal potential is expected to be nearly identicalbetween some strongly interacting species due totheir close ecological associations.Incongruent patterns of temporaldemography and mode of dispersalTwo recent surveys of rocky-shore marine communitiesin southern Australia (Ayre et al. 2009)and the northeastern Pacific of North America(Marko et al. 2010) compared spatial patterns of differentiationwith underlying temporal historiesamong large numbers of species with differentlarval dispersal potential. Results from both studiestend to contradict a simple correlation betweenmode of development and rate of gene flow as theprimary determinant of spatial differentiation. Ayreet al. (2009) studied eight species using mtDNA sampledfrom populations within several hundred kilometersto the east and west of a significant regionalbiogeographic disjunction in New South Wales thatis associated with a historical geological dispersalbarrier (the Pleistocene land bridge betweenAustralia and Tasmania), unsuitable adult habitat(long sandy beaches near Wilson’s Promontory inVictoria), and divergent offshore currents thatshould tend to limit present-day gene flow acrossthe barrier via planktonic larval dispersal. In a barnacle(Catomerus polymerus), a limpet (Cellana tramoserica),a chiton (Plaxiphora albida), and a sea star(Meridiastra calcar) Ayre et al. (2009) found no differentationamong populations within each regionbut strong and statistically significant phylogeographicbreaks between eastern and western populationson either side of the biogeographic disjunction.Notably, all four species that share this phylogeographicbreak have planktonic larval development.In contrast, Ayre et al. (2009) found very strongpopulation differentiation, but no statistically significantadditional regional genetic differentiation,across the same biogeographic barrier in two specieswith benthic development: another gastropod(Haustrum vinosa) and another sea star (Parvulastraexigua). Could gene flow actually be greater acrossthis barrier among species with nonplanktoniclarvae? Possibly, but IMA estimates of gene flow havenot yet been carried out; Ayre et al. (2009) did notanalyze sequence data from these species under thefull isolation-with-migration model, and thus couldnot obtain reliable estimates of migration rates(C. Perrin, personal communication) for the fourspecies that showed regional differentiation. No IMAanalyses were carried out for the two species withbenthic development that showed no additional differentiationassociated with the biogeographic barrier, orfor the two other species they studied (another barnacle,Tetraclitella purpurascens, and another gastropod,Bembicium nanum) that have planktonic larval dispersaland geographic ranges that span the biogeographicbarrier but showed no mtDNA populationdifferentiation. However, IMA analyses of populationdivergence times by Ayre et al. (2009) showed that theages of this regional break in three of the four specieswere quite old (on the order of 100,000–1,000,000years), suggesting that in three of four species the presenceof the deep regional genetic break between easternand western populations could reflect relatively ancientseparation times rather than simply restricted geneflow. Whether this might also be true of the two specieswith benthic development (and no regional differentiationacross the biogeographic barrier) is not known.Ayre et al. (2009) concluded that the most likely causeof the regional phylogeographic break seemed to be thedistribution of suitable benthic habitat in the vicinityof the biogeographic barrier, and thus not simply thefacilitation or restriction of planktonic larval dispersaland gene flow. Although distinguishing between thesetwo alternative hypotheses (restricted gene flow versusvery old population divergence time) will requirebetter estimates of gene flow, presumably throughthe addition of more data (i.e., loci), the data and analysesfrom this suite of species shows no simple relationshipbetween mode of development, geneticdifferentiation, and gene flow.In the biogeographic region studied by Ayre et al.(2009), Puritz et al. (J. Puritz, C. Keever, J. Addison,Downloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

Divergence genetics and modes of development 651Fig. 4 Posterior probability distributions of population divergence time from IMA analysis of sea stars with (Meridiastra calcar) andwithout (Parvulastra exigua) planktonic larval development (J. Puritz, C. Keever, and J. Addison, unpublished data; methods as describedin Fig. 2). Analyses for M. calcar include mtDNA only; analyses for P. exigua include mtDNA data plus sequences for an intron froma nuclear locus encoding glucose phosphate isomerase (as in Keever et al. 2009). For each species, two divergence times areshown: between populations from New South Wales and Tasmania (separated by Bass Strait; black); and between the same New SouthWales populations and those from sites in South Australia to the west of a significant regional biogeographic barrier studied byAyre et al. (2009) (gray).M. Byrne, R. Toonen, R. Grosberg, M. Hart, unpublisheddata) have similarly tried to analyze regionalvariation in spatial structure and demographic historyfrom multiple loci in the same pair of broadlysympatric sea star species with different modes ofdevelopment (planktonic larvae in M. calcar; encapsulateddevelopment in P. exigua) using IMA analysesof mtDNA plus an intron from a nuclearprotein-coding locus. The results of this second analysisgenerally argue against simple predictions aboutthe evolution of spatial genetic structure basedmainly on the mode of development and gene flowvia larval dispersal. On one spatial scale, Puritz et al.estimated divergence times and other parametersfor population pairs of both species from the easternside of the biogeographic region (in New SouthWales and in Tasmania, which have been linkedduring Pleistocene low sea level stands by the continuouscoastline of the Tasmanian land bridge, andare now separated by Bass Strait). Both species showstrong and statistically significant spatial differentiationin AMOVA analyses (e.g., for mtDNA variation,M. calcar F CT ¼ 0.15; P. exigua F CT ¼ 0.73) on thisgeographic scale, with moderate mtDNA haplotypediversity in both species, but very high intron allelediversity in M. calcar and very low intron allele diversityin P. exigua (similar to or lower than mtDNAdiversity within populations of this species). Puritzet al. found several surprising features of the demographicprocesses underlying this qualitatively similarpattern of spatial differentiation. First, divergencetimes across Bass Strait (Fig. 4) were significantlydifferent between the two species: nearly an orderof magnitude younger for M. calcar (93,000 years)compared to P. exigua (673,000 years). More recentpopulation divergence might reflect the relative easeof colonization and population establishment ormore consistent genetic connectivity across this barrierin a species with planktonic larval dispersal(M. calcar), or the more persistent effects of ancientextirpation with poor colonization by species withoutplanktonic larvae (P. exigua). However, these twospecies do not differ in the rate of gene flow acrossBass Strait in spite of their obvious difference indispersal potential: in both cases, M ¼ 0 after eitherrecent (M. calcar) or more ancient (P. exigua) divergencebetween Australian and Tasmanian populations.This unexpected set of patterns suggests thatdifferences in mode of development might morestrongly affect metapopulation dynamics of extirpationand colonization (measured as t) than it wouldthe rates of gene flow between populations after divergence(measured as M). Such colonization differenceshave been long appreciated by larval ecologistsDownloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

652 M. W. Hart and P. B. Marko(Johannesson 1988) but the quantitative effects ofvariation in colonization rate (independent of differencesin gene flow after colonization) have beenlargely overlooked by marine phylogeographers, perhapsbecause the two effects could not previously bedifferentiated in analyses that used F-statistics interpretedmainly in terms of ongoing gene flow. Lastly,Puritz et al. found qualitatively different patterns ofN e variation between these two species: a two-fold tofive-fold increase in M. calcar N e (compared to ancestralN e ), and at least an order of magnitude decreasein P. exigua N e (Fig. 5). We are unsurewhether these two contrasting differences in t andN e are causally linked, or independently varying, inthese two species, but they strongly contradict asimple expectation of similar demographic processesunderlying similar spatial patterns of geneticdifferentiation.On a second geographic scale, Puritz et al. compareddemographic parameters for population pairsof both species from the far ends of the eastern andwestern biogeographic regions documented by Ayreet al. (2009). Because these eastern and westernsamples were from locations in South Australia andNew South Wales that are separated by an additionalcandidate biogeographic barrier (soft sedimentshores between Wilson’s Promontory and theregion around Adelaide to the west in SouthAustralia) outside of the immediate region of thephylogeographic break that was sampled on amuch finer scale by Ayre et al., the results ofPuritz et al. tend to complement, rather than test,the patterns reported previously. Both species showvery strong spatial differentiation on this geographicscale (for mtDNA variation, M. calcar F CT ¼ 0.58; P.exigua F CT ¼ 0.72), with reciprocally monophyleticallele or haplotype genealogies at one or both loci.In this comparison, east–west population divergencetimes were generally similar to those estimated byAyre et al. (2009) but significantly younger in M.calcar (223,000 years) than in P. exigua (4600,000years; Fig. 4), and older than the M. calcar divergencetimes across Bass Strait. Like the Bass Straitcomparison, migration rates were zero for both speciesand the two species differed in the pattern ofevolution of population size (expanding in M. calcar,Downloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010Fig. 5 Posterior probability distributions of effective population sizes (N e ) of sea stars as in Fig. 4. Each panel shows estimated ancestralN e (gray symbols and text), plus the estimate for extant populations from New South Wales (black, large symbols and text) andfrom either South Australia (above; small symbols and text) or from Tasmania (below). For P. exigua (right), N e estimates werelog-transformed in order to show both small (extant) and large (ancestral) N e values on the same scale. Note the seconddependent variable scale (gray) for lower probability values of ancestral N e estimates in P. exigua.

Divergence genetics and modes of development 653shrinking in P. exigua, relative to the ancestral population;Fig. 5). Because all estimates of migrationrates on these spatial scales were zero, the spatialdifferentiation and contrasting histories of populationgrowth are not likely to be explained by differencesin larval dispersal. One candidate explanationis based on differences in mating systems: P. exiguaindividuals are small sex-changing hermaphroditeswith fecundities on the order of 10 2 –10 3 per eggmass, whereas M. calcar individuals are large gonochoricbroadcast spawners with clutch sizes severalorders of magnitude larger than in P. exigua. SmallN e in P. exigua may reflect population bottlenecksexacerbated by the evolution of very small clutchsize. IMA analyses of variation in both species onsmaller spatial scales (within each of the three regionswe considered in this comparison) will beuseful in estimating the possible difference betweenspecies in local migration rates within biogeographicalregions (as suggested by Ayre et al. 2009).However, none of these comparisons so far suggesta primary role for gene flow via larval dispersal (ordifferences between species in mode of development)for generating large-scale phylogeographic structureor determining species differences in these spatialpatterns.In the northeastern Pacific, Marko et al. (2010)combined mtDNA with some nuclear intron orother noncoding nuclear DNA sequences for a similartaxonomically broad sample of co-occurring speciesfrom a temperate rocky-shore community. Thisstudy did not include a single putative regional biogeographicbarrier (although some strong phylogeographicbreaks occur among some of the speciessurveyed), but rather focused on inter-specific patternsof spatial genetic diversity caused by latePleistocene glaciations, especially the last glacial maximum(LGM) 20,000 years ago. Among the 14 speciesstudied, 12 showed statistically significantevidence of past population expansions as measuredby standard summary statistics that assume equilibriumrates of mutation, gene flow, and genetic drift(i.e., Tajima’s D, Fu’s F s , and Ramos-Osins andRozas’ R2; Tajima 1989; Ramos-Onzins and Rozas1992). Eleven species showed no strong populationdifferentiation, together suggesting a shared demographichistory of recent postglacial expansion ofthe population across the entire community.However, when analyzed with IMA, some speciesshowed consistently ancient inter-population divergencetimes that were much older than the LGM(Balanus glandula, Patiria miniata, and Xiphisteratropurpureus) whereas others displayed youngerpopulation divergence times (e.g., Nucella lamellosa,X. mucosus) consistent with post-LGM recolonization.Old population divergences were found in specieswith (B. glandula, P. miniata) and without(N. lamellosa, X. atropurpureus) long-lived planktoniclarvae, and IMA estimates of migration rates were notconsistently correlated with either the mode of reproductionor the age of the most significant populationdifferentiation. Notably, Marko et al. (2010) foundvery old (B. glandula) or very young (X. mucosus)multilocus population divergence times in specieswithout significant mtDNA population differentiation.Population separation times inferred with IMAwere consistent with demographic reconstructions ofchanges in past population size (from Bayesian skylineplots); many species exhibited evidence of largedemographic expansions in the past, but the timingof these expansions varied dramatically among speciesin an unpredictable manner with respect to developmentalmode. Marko et al. (2010) concludedthat there might be more evidence for long-termpopulation persistence (population-pairs with old divergencetimes) among species with planktonic larvaldispersal than among species with benthic embryosand larvae, but found no statistically significant associationbetween population history and mode ofdevelopment.In the same biogeographic region studied by Markoet al. (2010), we have tried to carefully compare regionalvariation in spatial structure and demographic historyin two broadly sympatric species with differentmodes of development (N. lamellosa, P. miniata)using IMA analyses of a large sample of mtDNA plussix anonymous noncoding nuclear DNA sequencemarkers for each species (T. McGovern, C. Keever, C.Saski, M. Hart, P. Marko, unpublished results). Thisdetailed comparison has revealed clear examples ofthree types of incongruence between spatial populationstructure and underlying demographic history thattend to reject the simple predicted effects of modeof development and larval dispersal on rates of geneflow and the evolution of population differentiation.The first example demonstrates how unexpectedpatterns of differentiation with respect to larval dispersalpotential can be more readily explained oncethe patterns are considered in the temporal contextprovided by IMA. Across Queen Charlotte Sound inBritish Columbia (between the northern end ofVancouver Island and the Haida Gwaii/Alexander archipelagosof northern BC and southeastern Alaska),P. minata shows a large and highly statistically significantmtDNA population genetic break (F CT 0.4) butN. lamellosa shows no significant genetic differentiationacross the same expanse. Given their differentmodes of development, this spatial pattern ofDownloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

654 M. W. Hart and P. B. Markodifferentiation appears to completely contradict theexpectation that the greater potential for gene flowby larval dispersal in P. miniata should erase the signatureof older vicariant events and that the geneticevidence of older vicariant events might be moreeasily preserved in N. lamellosa, which lacks planktoniclarvae (Pelc et al. 2009). However, IMA estimates of theage of this divergence between northern and southernpopulations on either side of Queen Charlotte Sound(Fig. 6) are about an order of magnitude older in thesea stars (280,000 years) than in the snails (15,000years, consistent with colonization and population divergenceat the end of the last glaciation), and implythat population structure on this geographic scale isa consequence of different responses to vicariantevents of very different ages (possibly including environmentaleffects of the same age as the most recentglaciation, plus one or more much older processes)rather than differences in dispersal ability. The multilocusestimate of gene flow across Queen CharlotteSound from IMA was significantly greater than zerofor P. miniata, but not substantially lower than comparisonsbetween populations in other parts of the speciesrange. This consistency across the species rangeindicates that the phylogeographic break reflects a relativelylong history of population separation, ratherthan a regional restriction on gene flow. For N. lamellosa,even though no significant mtDNA structure wasfound across Queen Charlotte Sound, IMA inferredzero gene flow from the multilocus data, similar toestimates of gene flow between most other pairs ofpopulation.The second example from this detailed comparisoninvolves similar demographic histories underlyinga different spatial pattern of differentiation.In N. lamellosa the most significant regionalmtDNA phylogeographic break (F CT 0.07) occursbetween British Columbian populations in northernand southern Vancouver Island (500 km to thesouth of the large break in P. miniata at QueenCharlotte Sound). On this spatial scale, populationsof P. miniata show no significant differentiation inDownloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010Fig. 6 Posterior probability distributions of population divergence times from IMA analysis of mtDNA and six anonymous nuclear lociof a northeast Pacific whelk (Nucella lamellosa) and a sympatric sea star (Patiria miniata). Data from Marko et al. (2010); methods as inFig. 2 and Marko et al. (2010). Population sampling for both species included locations north and south of a significant regionalphylogeographic break at Queen Charlotte Sound, British Columbia, Canada (in the sea stars), and north or south of a phylogeographicbreak on Vancouver Island (in snails). Note the slightly different time scales for the two comparisons. For snails (upper right), twoposterior distributions are shown for comparisons between a population south of Queen Charlotte Sound on Vancouver Island(San Josef Bay) and one of two populations from Ketchikan, Alaska (gray), or from Prince Rupert, British Columbia (black), to thenorth of Queen Charlotte Sound. Snail populations were sampled on both the eastern and western sides of Vancouver Island, and allfour posterior distributions are shown (lower right) for divergence times between eastern and western populations at the same latitude(gray), or between northern and southern populations along the wave-exposed western coast or along the wave-protected easterncoast (black) of Vancouver Island.

Divergence genetics and modes of development 655mtDNA. However, this qualitative difference in thespatial pattern of mtDNA differentiation hides a statisticallyindistinguishable history of population separationof 112,000 years in the sea stars and98,000–106,000 years in snails (for north–south divergencetimes between snail populations on the exposedwestern or protected eastern sides ofVancouver Island, respectively) (Fig. 6). Gene flowat this scale in both species appears to be low, butthe effective population sizes of the sea stars dwarfthose of the snails, and this difference may have beenimportant in slowing the evolution of spatial differentiationbetween sea star populations as measuredby AMOVA analyses. Given their different modes ofdevelopment, this temporally coincident vicariance isdifficult to explain in terms of limitation of dispersal(for the snails) and high rates of gene flow via planktoniclarvae (for the sea stars), and thus probablyrequires an explanation based on the environmentaleffects (such as Pleistocene ice) on adult populationpersistence in both species.In both of these examples, IMA estimated scaledmigration rates (N e m) that were similar for pairsof populations separated by a phylogeographicbreak and for pairs on the same side of such abreak. Overall, migration rates were lower (andmore often estimated to be zero) between snail populationsthan between sea star populations, but inboth species the migration rate across a significantphylogeographic break (Queen Charlotte Sound forsea stars, Vancouver Island for snails; Fig. 6) was notsignificantly lower than migration rates between populationson one or the other side of the break.In both cases, this result argues strongly against theprimary significance of gene flow via larval dispersal(or philopatric recruitment via encapsulated development)in the evolution of regional population geneticstructure.Obstacles to full exploitation ofnew analytical methodsHare (2001) reviewed the prospects for use of nucleargene trees in phylogeography at about the sametime as coalescent MCMC methods were first madewidely available. Hare noted the liabilities associatedwith reliance on a single gene tree (from mtDNA orother organellar genomes) for inferences about populationhistories, and the critical need for multipleloci in demographic inference. His review identifiedthree main obstacles at that time to the use of nuclearDNA sequences in phylogeography: lack ofstandardized methods for PCR amplification of nucleargenes from under-studied genomes; expectedhigh rates of recombination between alleles; and unknownbut possibly low rates of polymorphism.Progress in the short time since that review haslargely resolved these issues through the (1) developmentof relatively accessible methods for individualresearchers to clone and sequence anonymous nuclearloci, and through the development of partial orcomplete genome sequences from EST and genomiclibraries for hundreds of animal, plant, and microbialgenomes; (2) development of methods for quantifyingsite-specific recombination rates in sequencealignments, incorporating recombination into isolation-with-migrationmodels, and identifying blocksof nucleotide sites unaffected by real or apparentrecombination; and (3) demonstration of remarkablyhigh rates of polymorphism, with deep intraspecificcoalescent times, at many nuclear loci across a widevariety of organisms.Recent developments have also resolved a significantissue associated with coalescent populationmodels: the impact of demographic or spatial structurewithin the two population samples in IMAanalyses. A recent simulation study (Strasburg andRieseberg 2010) suggests that the IM model is notmuch affected by population structure withinsamples, and the development of IMA2 may furtherresolve such issues in favor of splitting some differentiatedsamples for analyses of more than two populations.The most significant remaining obstacle tosuccessful application of the IMA method across speciesand loci may be the fit of individual sequencealignments to the DNA sequence mutation modelimplemented in current versions of the software.Only two (out of many possible and more realistic)mutation models for DNA sequences are implementedin IMA and IMA2. Strasburg and Rieseberg(2010) showed that a poor fit between the mutationmodel and the real pattern of nucleotide substitutioncan be a significant source of error in IMA analyses.Microsatellites, which often contain many ‘‘imperfect’’repeats, may be particularly problematic inthis regard, given that IMA only allows the use ofthe stepwise mutational model (Kimura and Ohta1978).Since Hare’s (2001) review, the progressive developmentof more complex and realistic coalescentpopulation models has put increasing pressure onlarval ecologists (working as phylogeographers) tokeep pace with the data requirements of themodels by adding nucleotides (for greater resolutionof genealogies), loci (for capturing more of the populationhistory sampled among individuals), individuals(for capturing more of the coalescent historyof each locus), populations (for comprehensiveDownloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

656 M. W. Hart and P. B. Markounderstanding of population history across each species’range), and species (for comparative analyseswithin communities or within higher taxa). In spiteof a decade of advice from reviews, models, andsimulation studies to ‘‘add loci’’, a glance at therecent literature shows that many larval ecologistshave found it difficult to satisfy all five of these imperatives,and many have found it necessary insteadto trade-off some against others, most often byrelying on single-locus mtDNA analyses that usehighly reliable PCR methods for previously wellcharacterizedmitochondrial genes, and avoid thelabor, expense, and errors associated with cloningPCR products from heterozygotes at nuclear loci.It seems likely that many of these cases stem fromlack of knowledge of the nuclear genome of thestudy organism, or the perceived difficulty of developingPCR primers for polymorphic nuclear genesequences.The effects of dependence on mtDNA alone (andthe need for multiple loci in order to take full advantageof the insights from IMA and other coalescentMCMC methods) are apparent in the resultsfrom some recent comparative phylogeographicalstudies whose design and research questions weredirectly relevant to the goals of our review butwhich yielded inconclusive results from IM or IMAanalysis. Teske et al. (2007) reported variation inpatterns of mtDNA spatial differentiation (F ST ) insouthern Africa among five sympatric rocky-shorespecies that was correlated with the mode of larvaldevelopment in a predictable way (higher for specieswithout planktonic larvae). However, IM estimates ofpopulation parameters from those mtDNA sequencealignments were ambiguous in the sense that theposterior distributions of most estimated migrationrates and population divergence times had weaklydefined lower bounds (including zero) and upperbounds (the posterior probabilities for the highestparameter values failed to decline to zero withinthe range of the prior distribution). Similarly,Crandall et al. (2008) followed a heroic samplingdesign that included 1182 individuals and 133 populationsfrom 97 localities spanning the 7000 kmeast–west range of two closely related Nerita gastropods.Crandall et al. found striking differences betweenthe large-scale phylogeographic structures inmtDNA sequence alignments from these two species(which have similar high-dispersal modes of larvaldevelopment), but were unable to use IM to comparethe population demographic histories underlyingthose two different patterns of spatial variationbecause one of the sequence alignments stubbornlyresisted all efforts to converge on a reliable IM result.Instead, Crandall et al. used alternative methods(Bayesian skyline plots) to infer differences amongspecies and lineages in the history of variation ineffective population size but without jointly estimatingvariation in migration rates or population divergencetimes (as in IM). In an analogous study (ofecotypes within a snail species complex, rather thanamong species with different distributions or modesof development), Quesada et al. (2007) found thatmtDNA sequences alone could clearly identify fourgeographically distinct parallel divergences betweenthe two ecotypes, but three of those four populationdivergence times could not be estimated with anyprecision in IM (with poorly defined posterior distributionsas described above). In all three of thesecases, the inability to resolve the population demographichistories that underlie the observed patternsof mtDNA spatial differentiation does not invalidatethe characterization of those spatial patterns, but itdoes prevent us from discovering whether the spatialdifferentiation is a direct reflection of differences inrates of gene flow, or a more complex outcome ofdifferent population histories, divergence times, andpatterns of change in population size. It seems likelythat in all three of these cases (and others we havenot reviewed that did not focus on marine invertebratepopulations), mtDNA variation alone containedtoo little coalescent information for the jointestimation of many demographic parameters, andthat greater sampling of loci (with the same oreven less intensive sampling of nucleotides, individuals,and populations) could have sidestepped theselimitations on the coalescent information contentfrom mtDNA alone.An important but underappreciated considerationin the development of new nuclear sequence alignmentsfor coalescent population genetic analyses isthe potential bias in marker selection associatedwith a historic preference for highly variable markers.In the past, a standard procedural step in the processof development of genetic markers used by manypopulation geneticists was to ‘‘screen loci for polymorphism’’by focusing on loci with high levels ofallelic variation in test samples and discarding thosewith low levels of polymorphism. Although loci withlow levels of polymorphisms are not useful for studiesof individuality (e.g., kinship and parentage) orfor some methods that attempt to infer recent dispersalevents using multilocus linkage disequilibrium(e.g., STRUCTURE, BAYESASS), loci with low levels ofnucleotide and haplotype diversity can provide importantinformation about variance in the coalescentprocess. Because the relative mutation rates of lociare estimated separately in IMA (i.e., each locusDownloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

Divergence genetics and modes of development 657receives its own scalar of mutation rate), the effect ofdiscarding less variable loci depends on the priordistribution of mutation-rate scalars with and withoutsuch loci included. If the prior distributionswith, and without, less variable loci were substantiallydifferent, then their exclusion could bias the results(Hey 2010a) unless some other information(such as very low relative rates of nonsynonymoussubstitutions in coding sequences) can be used toargue for a nonneutral basis for the low polymorphism.The need for multiple loci in coalescent demographicanalyses, and the perils of reliance onmtDNA alone in animal phylogeography, is an increasinglydifficulty problem for larval ecologists toavoid. In the IMA2 documentation, Hey suggested asa rule of thumb that we should double the numberof loci for each additional population in orderto avoid the crippling limitations on the quality ofestimation of parameters as the number of parametersincreases approximately exponentially with thenumber of populations. Whether larval ecologistswill have to meet this or some less onerous standardin order to achieve reliable and precise coalescentMCMC results in future comparative analyses is anopen question, but there seems little doubt that thefuture of comparative phylogeography will be builtmainly on coalescent hypothesis-tests using multiplenuclear-sequence alignments (Hurt et al. 2009).The new marine phylogeographyAre marine invertebrate larvae entirely responsiblefor determining population genetic structure inmarine communities? Although relevant evidencehas been presented for 430 years, we suggest thatthe jury should continue deliberations. On the onehand, post hoc interpretations based on dispersalcapability for results from F ST and AMOVA analysesoften suggest a strong correlation. On the otherhand, coalescent methods will tend to give evidencethat often contradicts AMOVA-type results byaccounting for some large or small proportion ofpopulation differentiation in terms of variation indivergence times and changes in population sizesrather than strictly in terms of gene flow. Reachinga reliable verdict will depend on reanalysis of previouslypublished (mainly mtDNA) sequence data, andthe accumulation of new (multilocus) sequencealignments analyzed using improved coalescentsamplers.In this short review, we advocated comparativetests of the correlation between modes of development,spatial population genetic structure, and ratesof gene flow (jointly estimated with other populationdemographic parameters). However, additionalinsight could come from better within-species hypothesesof the spatial distribution of gene flowand population differentiation. One hypothesis thatmight bear repeated testing involves regional biogeographicpatterns like those studied by Ayre et al.(2009) and Marko et al. (2010): does IMA analysisof pairs of populations reveal higher rates of geneflow (N e m) between pairs on the same side of sucha phylogeographic break, and lower gene flow betweenpairs of populations separated from eachother by the biogeographic ‘‘barrier’’? The expectationis intuitively obvious, but our observations ofspatially uniform gene flow around a phylogeographicbreak in P. miniata suggests that vicarianceand not barriers to gene flow may account for thisphylogeographic structure. Many such analyses mightreveal whether this pattern is an anomaly or a typicalexplanation for unexpected population differentiationin those marine invertebrates with apparentlyhigh potential for larval dispersal. For example,Ayre et al. (2009) sampled on both sides of a distinctivebiogeographic barrier, and found patterns ofspatial variation on either side that were consistentwith high rates of gene flow within each region forspecies with planktonic larvae, but did not test thehypothesis of larval dispersal by estimating migrationrates and other demographic parameters for thosepairs of populations within each region (or comparesuch results between species with and without planktoniclarvae). We look forward to many such analysesas the product of increasingly powerful analyticalmethods and increasingly inexpensive DNA sequencingservices for compiling multilocus sequence alignmentsfor comparative phylogeography.We foresee a surprising but very happy convergenceof interests and research tools not previouslyshared in common by phylogeographers and comparativedevelopmental biologists (including larvalecologists), both intent on understanding the evolutionof marine animal development. Phylogeographershave a critical need for datasets composed ofsequence alignments from multiple nuclear loci forcoalescent MCMC demographic analyses, but oftenlack easy access to preliminary genomic data for thedesign of PCR and sequencing strategies other thanthose for mtDNA. In contrast, comparative developmentalbiologists require (among other things) genomicresources for understanding the evolution ofgene expression networks underlying differences indevelopmental programs that have led to the evolutionand loss of dispersing planktonic larval forms.An important driver of the explosive growth ofDownloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

658 M. W. Hart and P. B. Markocomparative genomics has been the development ofsecond-generation sequencing technology. In largepart at the urging of developmental geneticists, thishigh-throughput sequencing capacity has led to thesequencing of hundreds of complete eukaryotic genomes.In addition to providing information foranalysis of gene expression patterns (for comparativedevelopmental biology), these new complete genomesequences can provide initial insight into genomes ofnonmodel organisms that present interesting phylogeographicproblems, and can be used for the designof sampling strategies for phylogeographic studiesbased on PCR amplification of known target lociand high-throughput sequencing of PCR amplicons.In addition, these complete genome sequences can alsoserve as reference sequences for the computationalassembly of whole genome sequence data from individualorganisms in phylogeographic studies. Thesetwo complementary approaches to using secondgenerationsequencing in phylogeographic studieshave different design requirements and might havedifferent goals, but both approaches seem to holdconsiderable promise for fully exploiting the new analyticalmethods and for resolving some of the mostimportant problems in larval ecology: how far dolarvae go, how often, and for how long have theydone so?AcknowledgmentsThanks to Jason Addison, David Ayre, Maria Byrne,Rick Grosberg, Jody Hey, Carson Keever, TammyMcGovern, Cecile Perrin, Jon Puritz, Jenn Sunday,and (especially) April Blakeslee for sharing data,analyses, and ideas used in this article.FundingPart of this work was carried out by using the resourcesof the Computational Biology Service Unitfrom Cornell University, which is partially funded byMicrosoft Corporation. We were supported by SICB,the Divisions of Invertebrate Zoology and of Ecologyand Evolution, American Microscopical Society,US National Science Foundation (OCE-0550526),and Natural Sciences and Engineering ResearchCouncil of Canada. This is Technical ContributionNo. 5822 of the Clemson University ExperimentStation, supported by the CSREES/USDA.ReferencesArndt A, Smith MJ. 1998. Genetic diversity and populationstructure in two species of sea cucumber: differing patternsaccording to mode of development. Mol Ecol 7:1053–64.Ayre DJ, Minchinton TE, Perrin C. 2009. Does life historypredict past and current connectivity for rocky intertidalinvertebrates across a marine biogeographic barrier? MolEcol 18:1887–903.Banks SC, Piggott MP, Williamson JE, Bové U, Holbrook NJ,Beheregaray LB. 2007. Oceanic variability and coastal topographyshape genetic structure in a lopng-dispersing seaurchin. Ecology 88:3055–64.Barber PH, Palumbi SR, Erdmann MV, Moosa MK. 2000.Biogeography. A marine Wallace’s line? Nature 406:692–3.Becquet C, Przeworski M. 2007. A new approach to estimateparameters of speciation models with application to apes.Genome Res 17:1505–19.Beerli P. 2004. Effect of unsampled populations on theestimation of population sizes and migration rates betweensampled populations. Mol Ecol 13:827–36.Beerli P, Felsenstein J. 1999. Maximum-likelihood estimationof migration rates and effective population numbers in twopopulations using a coalescent approach. Genetics 152:763–73.Beerli P, Felsenstein J. 2001. Maximum likelihood estimationof a migration matrix and effective population sizes in nsubpopulations by using a coalescent approach. Proc NatlAcad Sci USA 98:4563–8.Berger E. 1973. Gene-enzyme variation in three sympatricspecies of Littorina. Biol Bull 145:83–90.Berger E. 1977. Gene-enzyme variation in three sympatricspeceis of Littorina. II. The Roscoff population, with anote on the origin of North American L. littorea. BiolBull 152:255–64.Blakeslee AM, Byers JE, Lesser MP. 2008. Solving cryptogenichistories using host and parasite molecular genetics: theresolution of Littorina littorea’s North American origin.Mol Ecol 17:3684–96.Bohonak AJ. 1999. Dispersal, gene flow, and population structure.Quart Rev Biol 74:21–45.Bossart JL, Prowell DP. 1998. Genetic estimates of populationstructure and gene flow: limitations, lessons and new directions.Trends Ecol Evol 13:202–6.Brown CA, Jackson GA, Brooks DA. 2000. Particle transportthrough a narrow tidal inlet due to tidal forcing and implicationsfor larval transport. J Geophys Res 105:24141–56.Buonaccorsi VP, Kimbrell CA, Lynn EA, Vetter RD. 2002.Population structure of copper rockfish (Sebastes caurinus)reflects postglacial colonization and contemporary patternsof larval dispersal. Can J Fish Aquat Sci 59:1374–84.Burton RS. 1983. Protein polymorphisms and genetic differentiationof marine invertebrate populations. Mar Biol Lett4:193–206.Burton RS, Feldman MS. 1981. Population genetics ofTigriopus californicus: II. Differentiation among neighboringpopulations. Evolution 35:1192–205.Burton RS, Feldman MS. 1982. Population genetics of coastaland estuarine invertebrates: does larval behavior influencepopulation structure? In: Kennedy VS, editor. Estuarinecomparisons. New York: Academic Press. p. 537–51.Downloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

Divergence genetics and modes of development 659Byrne M, Prowse TA, Sewell MA, Dworjanyn S,Williamson JE, Vaitilingon D. 2008. Maternal provisioningfor larvae and larval provisioning for juveniles in thetoxopneustid sea urchin Tripneustes gratilla. Mar Biol155:473–82.Carstens BC, Stoute HN, Reid NM. 2009. An informationtheoreticapproach to phylogeography. Mol Ecol 18:4270–82.Collin R. 2001. The effects of mode of development on phylogeographyand population structure of North AtlanticCrepidula (Gastropoda: Calyptraeidae). Mol Ecol 10:2249–62.Collin PL. 2003. Larvae retention: genes or oceanography?Science 300:1657.Cowen RK, Sponaugle S. 2009. Larval dispersal and marinepopulation connectivity. Annu Rev Mar Sci 1:443–66.Crandall ED, Frey MA, Grosberg RK, Barber PH. 2008.Contrasting demographic history and phylogeographical patternsin two Indo-Pacific gastropods. Mol Ecol 17:611–26.Crisp DJ. 1978. Genetic consequences of different reproductivestrategies in marine invertebrates. NATO Conf Ser IVMarine Sci 2:257–73.Crow JF, Aoki K. 1984. Group selection for a polygenic behavioraltrait: estimating the degree of population subdivision.Proc Natl Acad Sci USA 81:6073–7.Cunningham CW. 2008. How to use genetic data to distinguishbetween natural and human-mediated introductionof Littorina littorea to North America. Biol Invasions10:1–6.Cunningham CW, Collins TM. 1998. Beyond area relationships:extinction and recolonization in molecular marinebiogeography. In: DeSalle R, Schierwater B, editors.Molecular ecology and evolution: approaches and applications.Basel: Birkhauser Verlag. p. 297–321.Duffy JE. 1993. Genetic population structure in two tropicalsponge-dwelling shrimps that differ in dispersal potential.Mar Biol 116:459–70.Edmands S. 2001. Phylogeography of the marine copepodTigriopus californicus reveals substantially reduced interpopulationdivergence at northern latitudes. Mol Ecol10:1743–50.Edwards SV, Beerli P. 2000. Perspective: gene divergence,population divergence, and the variance in coalescencetime in phylogeographic studies. Evolution 54:1839–54.Emlet RB, Sadro S. 2006. Linking stages of life history:how larval quality translates into juvenile performance foran intertidal barnacle (Balanus glandula). Int Comp Biol46:334–46.Excoffier L, Laval G, Schneider S. 2005. Arlequin ver. 3.0: anintegrated software package for population genetics dataanalysis. Evol Bioinform Online 1:47–50.Felsenstein J. 1976. The theoretical population genetics of variableselection and migration. Annu Rev Genet 10:253–80.Fortuna MA, Albaladejo RG, Fernández L, Aparicio A,Bascompte J. 2009. Networks of spatial genetic variationacross species. Proc Natl Acad Sci USA 1006:19044–9.Gilg MR, Hilbish TJ. 2003. Geography of marine larvaldispersal: coupling genetics with fine-scale physical oceanography.Ecology 84:2989–98.Gooch JL. 1975. Mechanisms of evolution and populationgenetics. In: Kinne O, editor. Marine ecology, Vol. 2. PartI. London: Wiley. p. 349–409.Hare MP. 2001. Prospects for nuclear gene phylogeography.Trends Ecol Evol 16:700–6.Hellberg ME. 1996. Dependence of gene flow on geographicdistance in two solitary corals with different larval dispersalcapabilities. Evolution 50:1167–75.Helmuth B, Veit RR, Holberton R. 1994. Long-distancedispersal of a subantarctic brooding bivalve (Gaimardiatrapesina) by kelp-rafting. Mar Biol 120:421–6.Hench JL, Blanton BO, Luettich RA. 2002. Lateral dynamicanalysis and classification of barotropic tidal inlets. ContShelf Res 22:2615–31.Hey J. 2007. Introduction to the IM and IMA computer programs(http://genfaculty.rutgers.edu/hey/software).Hey J. 2010a. Isolation with migration models for more thantwo populations. Mol Biol Evol 27:905–20.Hey J. 2010b. The divergence of chimpanzee species andsubspecies as revealed in multi-population Isolationwith-Migrationanalyses. Mol Biol Evol 27:921–33.Hey J, Nielsen R. 2004. Multilocus methods for estimatingpopulation sizes, migration rates and divergence time,with applications to the divergence of Drosophila pseudoobscuraand D. persimilis. Genetics 167:747–60.Hey J, Nielsen R. 2007. Integration within the Felsensteinequation for improved Markov chain Monte Carlo methodsin population genetics. Proc Natl Acad Sci USA 104:2785–90.Hickerson MJ, Meyer CP. 2008. Testing comparative phylogeographicmodels of marine vicariance and dispersal usinga hierarchical Bayesian approach. BMC Evol Biol 8:322.Hickerson MJ, Stahl EA, Lessios HA. 2006. Test for simultaneousdivergence using approximate Bayesian computation.Evolution 60:2435–53.Highsmith RC. 1985. Floating algal rafting as potential dispersalmechanisms in brooding invertebrates. Mar EcolProg Ser 25:169–79.Hunt A. 1993. Effects of contrasting patterns of larval dispersalon the genetic connectedness of local populations of twointertidal starfish, Patiriella calcar and P. exigua. Mar EcolProg Ser 92:179–86.Hurt C, Anker A, Knowlton N. 2009. A multilocus test ofsimultaneous divergence across the Isthmus of Panamausing snapping shrimp in the genus Alpheus. Evolution63:514–30.Janson K. 1987. Allozyme and shell variation in two marinesnails (Littorina, Prosobranchia) with different dispersalabilities. Biol J Linn Soc 30:245–56.Johannesson K. 1988. The paradox of Rockall: why is abrooding gastropod (Littorina saxatilis) more widespreadthan one having a planktonic larval dispersal stage(L. littorea)? Mar Biol 99:507–13.Downloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

660 M. W. Hart and P. B. MarkoKarl SA, Avise JC. 1992. Balancing selection at allozymeloci in oysters: implications from nuclear RFLPs. Science256:100–2.Keever CC, Sunday J, Puritz JB, Addison JA, Toonen RJ,Grosberg RK, Hart MW. 2009. Discordant distribution ofpopulations and genetic variation in a sea star with highdispersal potential. Evolution 63:3214–27.Kimura M, Ohta T. 1978. Stepwise mutation model and distributionof allelic frequencies in a finite population.Proc Natl Acad Sci USA 75:2868–72.Kinlan BP, Gaines SD. 2003. Propagule dispersal in marineand terrestrial environments: a community perspective.Ecology 84:2007–20.Koehn RK, Newell RIE, Immerman F. 1980. Maintenance ofan aminopeptidase allele frequency cline by natural selection.Proc Natl Acad Sci USA 77:5385–9.Kuhner MK. 2006. LAMARC 2.0: maximum likelihood andbayesian estimation of population parameters. Bioinformatics22:768–70.Kuhner MK. 2009. Coalescent genealogy samplers: windowsinto population history. Trends Ecol Evol 24:86–93.Kyle CJ, Boulding EG. 2000. Comparative population geneticstructure of marine gastropods (Littorina spp.) with andwithout pelagic larval dispersal. Mar Biol 137:835–45.Liu L. 2008. BEST: Bayesian estimation of species trees underthe coalescent model. Bioinformatics 24:2542–3.Liu L, Pearl DK. 2007. Species trees from gene trees: reconstructingBayesian posterior Distributions of a species phylogenyusing estimated gene tree distributions. Syst Biol56:504–14.Lopes JS, Balding D, Beaumont MA. 2009. PopABC: a programto infer historical demographic parameters. Bioinformatics25:2747–9.Luettich RA Jr, Hench JL, Fulcher CW, Werner FE,Blanton BO, Churchill JH. 1999. Barotropic tidal andwind driven larval transport in the vicinity of a barrierisland inlet. Fish Oceanogr 8:190–209.Luettich RA Jr, Hench JL, Williams CD, Blanton BO,Werner FE. 1998. Tidal circulation and transport througha barrier island inlet. In: Spaulding M, Blumberg AF,editors. Estuarine and coastal modeling V. Reston, VA:ASCE. p. 849–63.Marko PB. 2004. ‘What’s larvae got to do with it?’ Disparatepatterns of post-glacial population structure in two benthicmarine gastropods with identical dispersal potential. MolEcol 13:597–611.Marko PB, Barr KB. 2007. Basin-scale patterns of mtDNAdifferentiation and gene flow in the Bay Scallop,Argopecten irradians concentricus Say. Mar Ecol Prog Ser349:139–50.Marko PB, Rogers-Bennett L, Dennis AB. 2007. MtDNApopulation structure and gene flow in lingcod (Ophiodonelongatus): limited connectivity despite long-lived pelagiclarvae. Mar Biol 150:1301–11.Marko PB, Hoffman JM, Emme SA, McGovern TM,Keever C, Cox LN. 2010. The expansion-contractionmodel of Pleistocene demography: rocky shores suffer asea change? Mol Ecol 19:146–69.Marshall DJ, Keough MJ. 2009. Does interspecific competitionaffect offspring provisioning? Ecology 90:487–95.McMillan WO, Raff RA, Palumbi SR. 1992. Populationgenetic consequences of developmental evolution in seaurchins (Genus Heliocidaris). Evolution 46:1299–312.Moran AL, Emlet RB. 2001. Offspring size and performancein variable environments: field studies on a marine snail.Ecology 82:1597–612.Neigel JE. 2002. Is F ST obsolete? Conser Genet 3:167–73.Palumbi SR. 1994. Genetic divergence, reproductive isolation,and marine speciation. Annu Rev Ecol Syst 25:547–72.Palumbi SR, Warner RR. 2003. Why gobies are like hobbits.Science 299:51–2.Pelc RA, Warner RR, Gaines SD. 2009. Geographical patternsof genetic structure in marine species with contrasting lifehistories. J Biogeogr 36:1881–90.Quesada H, Posada D, Caballero A, Moran P, Rolan-Alvarez E.2007. Phylogenetic evidence for multiple sympatric ecologicaldiversification in a marine snail. Evolution 61:1600–12.Ramos-Onzins SE, Rozas J. 1992. Statistical properties of newneutrality tests against population growth. Mol Biol Evol19:2092–100.Sherman CDH, Hunt A, Ayre DJ. 2008. Is life history abarrier to dispersal? Contrasting patterns of genetic differentiationalong an oceanographically complex coast. Biol JLinn Soc 95:106–11.Shulman MJ, Bermingham E. 1995. Early life histories, oceancurrents, and the population genetics of Caribbean reeffishes. Evolution 49:897–910.Slatkin M. 1985. Gene flow in natural populations. Annu RevEcol Syst 16:393–430.Sotka EE, Wares JP, Barth JA, Grosberg R, Palumbi SR. 2004.Strong genetic clines and geographical variation in geneflow in the rocky intertidal barnacle Balanus glandula.Mol Ecol 13:2143–56.Stommel H, Farmer HG. 1952. On the nature of estuarinecirculation, Part I. Woods Hole Oceanographic TechnicalReport 52–88. Woods Hole, Massachusetts.Strasburg JL, Rieseberg LH. 2010. How robust are ‘‘Isolationwith Migration’’ analyses to violations of the IM model?A simulation study. Mol Biol Evol 27:297–310.Strathmann RR. 1985. Feeding and nonfeeding larval developmentand life-history evolution in marine invertebrates.Ann Rev Ecol Syst 16:339–61.Strathmann RR. 1990. Why life histories evolve differently inthe sea. Amer Zool 30:197–207.Tajima F. 1989. Evolutionary relationship of DNA sequencesin finite populations. Genetics 105:437–60.Taylor MS, Hellberg ME. 2003a. Genetic evidence for localretention of pelagic larvae in a Caribbean reef fish. Science299:107–9.Taylor MS, Hellberg ME. 2003b. Larvae retention: genes oroceanography? Science 300:1657–8.Downloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010

Divergence genetics and modes of development 661Taylor MS, Hellberg ME. 2006. Comparative phylogeographyin a genus of coral reef fishes: biogeographic and geneticconcordance in the Caribbean. Mol Ecol 15:695–707.Teske PR, Papadopoulos I, Zardi GI, McQuaid CD,Edkins MT, Griffiths CL, Barker NP. 2007. Implicationsof life history for genetic structure and migrationrates of southern African coastal invertebrates: planktonic,abbreviated and direct development. Mar Biol 152:697–711.Thiyagarajan V, Pechenik JA, Gosselin LA, Qian PY. 2007.Juvenile growth in barnacles: combined effect of delayedmetamorphosis and sub-lethal exposure of cyprids tolow-salinity stress. Mar Ecol Prog Ser 344:173–84.Todd CD, Lambert WJ, Thorpe JP. 1998. The genetic structureof intertidal populations of two species of nudibranchmolluscs with planktotrophic and pelagic lecithotrophiclarval stages: are pelagic larvae ‘‘for’’ dispersal? J Exp MarBiol Ecol 228:1–28.Waples RS. 1987. A multispecies approach to the analysis ofgene flow in marine shore fishes. Evolution 41:385–400.Waples RS. 1998. Separating the wheat from the chaff: patternsof genetic differentiation in high gene flow species. J Hered89:438–50.Warner RR, Palumbi SR. 2003. Larvae retention: genes oroceanography? Science 300:1658.Watts PC, Thorpe JP. 2006. Influence of contrasting larvaldevelopmental types upon the population-genetic structureof cheilostome bryozoans. Mar Biol 149:1093–101.Whitlock MC. 1992. Non-equilibrium population structurein forked fungus beetles: extinction, colonization, and thegenetic variance among populations. Am Nat 139:952–70.Whitlock MC, McCauley DE. 1999. Short Review: indirectmeasures of gene flow - Fst does not equal 1/(4Nmþ1).Heredity 82:117–25.Wright S. 1951. The genetical structure of populations. AnnEugen 15:323–54.Zimmerman JTF. 1981. Dynamics, diffusion and geomorphologicalsignificance of tidal residual eddies. Nature290:549–55.Downloaded from icb.oxfordjournals.org at Simon Fraser University on November 2, 2010