Hierarchical analysis of genetic structure in the habitat-specialist Eastern Sand Darter (Ammocrypta pellucida)

Quantifying spatial genetic structure can reveal the relative influences of contemporary and historic factors underlying localized and regional patterns of genetic diversity and gene flow – important considerations for the development of effective conservation efforts. Using 10 polymorphic microsatellite loci, we characterize genetic variation among populations across the range of the Eastern Sand Darter (Ammocrypta pellucida), a small riverine percid that is highly dependent on sandy substrate microhabitats. We tested for fine scale, regional, and historic patterns of genetic structure. As expected, significant differentiation was detected among rivers within drainages and among drainages. At finer scales, an unexpected lack of within-river genetic structure among fragmented sandy microhabitats suggests that stratified dispersal resulting from unstable sand bar habitat degradation (natural and anthropogenic) may preclude substantial genetic differentiation within rivers. Among-drainage genetic structure indicates that postglacial (14 kya) drainage connectivity continues to influence contemporary genetic structure among Eastern Sand Darter populations in southern Ontario. These results provide an unexpected contrast to other benthic riverine fish in the Great Lakes drainage and suggest that habitat-specific fishes, such as the Eastern Sand Darter, can evolve dispersal strategies that overcome fragmented and temporally unstable habitats.


Introduction
Specialized microhabitat dependence presents a formidable challenge to species conservation in changing environments. For some species, the coupling of microhabitat specialization with increased habitat degradation and fragmentation can initiate or accelerate declines in population size and, ultimately, local extirpation. Microhabitat specialization provides an extreme example of local adaptation and raises questions about mechanisms that allow the persistence of such specialized life histories in variable environments. Characterization of genetic structure and gene flow among fragmented habitats can yield important information for the conservation of such microhabitatdependent species. Specifically, a hierarchical analysis can reveal the relative importance of large-scale historical processes (e.g., climatic, hydrological, geographic) and more contemporary fine-scale processes (e.g., in-stream barriers) in shaping overall patterns of genetic variation (Wiens 1997;Monaghan et al. 2002).
Molecular genetic methods can provide nonlethal means to successfully characterize many aspects of ecosystem processes and population connectivity for species at risk, including landscape effects on genetic substructure This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. (Cook et al. 2007;Caldera and Bolnick 2008), historical influences on contemporary population structure (Poissant et al. 2005;Stepien et al. 2007;Boizard et al. 2009), colonization patterns and alternative dispersal pathways (M€ akinen et al. 2006), and species introductions (Dlugosch and Parker 2008;Beneteau et al. 2012). Quantifying range-wide population connectivity provides valuable information on species dynamics and aids in the identification of isolated populations requiring special conservation attention (Manel et al. 2003;Cook et al. 2007;Storfer et al. 2007). Most importantly, genetic identification of fine-scale dispersal and gene flow patterns among fragmented populations may indicate natural or assisted recolonization potential for extirpated habitat patches (Bohonak 1999;Palsbøll et al. 2007).
Connectivity among populations depends on speciesspecific dispersal capabilities (Watanabe et al. 2010) and barriers to dispersal, which may disrupt gene flow by limiting among-population movements (McGlashan and Hughes 2001;Poissant et al. 2005;Johansson et al. 2008). Populations in freshwater ecosystems often show low levels of connectivity and high levels of genetic divergence as these ecosystems commonly rely on linear corridors of stream connectivity (Ward et al. 1994). The array of connectivity pathways among freshwater habitats ranging from small streams to lakes provides a variety of potential dispersal barriers for aquatic organisms (Caldera and Bolnick 2008). For habitat-specific fishes, such as darters, the loss or degradation of specialized habitats may disrupt not only within-river genetic connectivity but also natural metapopulation dynamics (Turner and Trexler 1998).
The Eastern Sand Darter (Ammocrypta pellucida) is a small benthic riverine fish federally listed as threatened in Canada and listed as threatened in many states in its American distribution (Grandmaison et al. 2004; Committee on the Status of Endangered Wildlife in Canada (COSEWIC) 2011). A. pellucida exhibits a unique burying behavior associated with sandy substrates, which may limit its potential for passive drift dispersal (Daniels 1989), but does exhibit a nonbenthic larval period (Simon and Wallus 2006). Tagging studies on A. pellucida showed no evidence for adult movement among sand bars during the summer months (Finch 2009). Those findings in conjunction with the patchy distribution of riverine sand bar habitat are expected to promote genetic divergence among adult assemblages. However, early life-stage dispersal and/or mixing of separate sand bar populations during the winter months has been suggested, but not tested, and both possibilities could facilitate mixing among sand bar populations (Simon and Wallus 2006). At a larger scale, the species range encompasses a patchy network of inhabited and uninhabited rivers, with the loss of suitable habitat largely attributed to anthropogenic pressures (Grandmaison et al. 2004;COSEWIC 2009). In the last century, A. pellucida has experienced a nearly 40% reduction in distribution, including extirpation from three Canadian river systems: Catfish Creek, Big Otter Creek (Lake Erie drainage), and the Ausable River (Lake St. Clair drainage).
Here, we assess the degree of population divergence for A. pellucida across its natural range. Using data from 10 microsatellite loci from individuals sampled from 39 sites, we aim to (1) characterize contemporary population connectivity through analyses of genetic structure and dispersal and (2) determine the relative influence of historic (postglaciation) colonization patterns versus current connectivity processes on drainage-level genetic structure. In general, we expect high genetic structure among sand bar sites for A. pellucida, even at small spatial scales, due to the species' restricted distribution to sandy substrate habitats. Additionally, we expect high levels of genetic divergence among regions as a result of population isolation and decline (Grandmaison et al. 2004; Committee on the Status of Endangered Wildlife in Canada (COSEWIC) 2011), although postglacial recolonization from different refugia and from different patterns of historic connectivity can also affect present-day genetic structure in A. pellucida. Consequently, the combination of habitat specialization and fragmentation within rivers, historic genetic patterns of connectivity and declining population sizes in most rivers reinforces the conservation and evolutionary importance of characterizing genetic structure among these populations.

Sampling
Sampling efforts focused on rivers recently reported to harbor A. pellucida populations (Grandmaison et al. 2004;Fisheries and Oceans Canada 2012), with targeted sampling directed to sand bars at depositional river bends. Hierarchical sampling definitions used in this study include sample sites (e.g., HR1), within rivers (e.g., Hocking River), within drainages (e.g., Ohio drainage). Sampling occurred in four drainages across the species range ( Fig. 1): (1) Ohio drainage (Little Muskingum River, Hocking River, Salt Creek, Red River, Licking River); (2) Wabash drainage (Eel River, East Fork White River, Deer Creek, Big Creek); (3) Great Lakes drainage (Maumee River, Grand River, Thames River, Sydenham River); and (4) St. Lawrence drainage (Richelieu River, Riv ere au Saumon, Champlain Canal). Ohio and Wabash drainages were categorized as separate drainages because the sampled rivers within those drainages are separated by over 1000 km. Fish were caught with a bag seine net (dimensions: wings 15 9 3 m with 0.64 cm mesh and 1.5 9 1.5 9 1.5 m bag with 0.32 cm mesh) or using a Missouri trawl specialized for benthic fish collection. Upon collection, a small pelvic fin clip was taken from each fish and preserved in 95% ethanol for subsequent DNA analysis. After a short recovery period in freshwater tanks, fish were then returned to their original habitats.

DNA extraction and genotyping
Fish were genotyped at 10 microsatellite loci, five of which were developed specifically for A. pellucida (Esd3, Esd13, Esd17, Esd18, Esd25) and an additional five loci from other darter species (Esc132b, EosC6, EosC112, EosD107, EosD11; see Table S1). Microsatellite loci discovery and primer development included the extraction of DNA followed by enrichment for repeat sequences using a protocol adapted from Fischer and Bachman (1998). Briefly, genomic DNA was digested with RsaI and ligated to MluI adapter-primer complexes (5 0 -CTCTTG CTTACGCGTGGACTA-3, 5 0 -pTAGTCCACGCGTAAGCA AGAGCACA-3 0 ). DNA fragments were hybridized with 5 0 -biotinylated oligo (GACA 4 ) probes, captured with streptavidin-coated beads (Roche, Indianapolis, USA), and enriched using polymerase chain reactions (PCR). The resulting enriched DNA library was inserted into TOPO vectors and transformed into One Shot competent Escherichia coli cells (Invitrogen, Burlington, ON, Can-ada). Clone inserts were amplified using M13 universal forward and reverse primers and sequenced at Genome Quebec Innovation Centre (McGill University, Montreal, Canada). Microsatellite primer pairs were designed and optimized for polymorphism and ease of PCR amplification. PCR amplification of all ten microsatellite loci used in this study was performed in 12.75 lL reactions containing approximately 50-100 ng template DNA, 0.25 lL of 0.5 lmol/L dye-labeled forward primer, 0.25 lL of 0.5 lmol/L reverse primer, 200 lmol/L of each dNTP, various concentrations of MgCl 2 (Table S1), and 0.25U Taq DNA polymerase (Applied Biosystems, Foster City, CA) in a 19 PCR buffer. PCR thermal cycler profiles consisted of an initial denaturing period at 94°C for 120 sec followed by 35 cycles of 94°C for 30 sec, various annealing temperatures for each primer (Table S1) for 45 sec, 30 sec at 72°C, and 90 sec at 72°C at the final extension period. Dyelabeled PCR products were visualized on a LiCor 4300 DNA analyzer (LiCor Biosciences, Inc. Lincoln, Nebraska, USA). Individual genotypes were determined by scoring allele sizes using GENE IMAGIR 4.05 software (Scanalytics Inc. Fairfax, VA, USA).

Marker validation
Genotype data for each site were tested for the presence of null alleles, allele scoring error, and large allele dropout using MICROCHECKER 2.2.3 (Van Oosterhout et al. 2004). All pairs of microsatellite loci were analyzed for linkage disequilibrium in ARLEQUIN 3.01 (Excoffier et al. 2005). Departures from Hardy-Weinberg equilibrium (HWE) were assessed for all possible locus-by-site combinations using the Markov chain Monte Carlo (MCMC) method (100,000 dememorization steps; 1,000,000 Markov chain steps) in ARLEQUIN. HWE departure significance and all subsequent pairwise comparisons were adjusted for multiple simultaneous tests using sequential Bonferroni corrections (Rice 1989).

Genetic differentiation
Genetic differentiation was estimated by calculating pairwise F ST values (Weir and Cockerham 1984) among all sites in ARLEQUIN. To quantify genetic differentiation among rivers, sites within each river were combined and mean pairwise F ST estimates were calculated among rivers using ARLEQUIN. Global F ST values were calculated among all rivers within each of the four drainages (to compare levels of divergence among rivers within drainages), with significance determined by jackknifing across all loci at the 95% confidence interval in FSTAT (Goudet 2001). Allelic richness (A R ), number of alleles (A), observed heterozygosity (H O ), expected heterozygosity (H E ), inbreeding coefficient (F IS ) were also calculated in FSTAT.

Isolation by distance
Rivers with more than three sampling sites were tested for adherence to an isolation by distance (IBD) model of migration-drift equilibrium. IBD was determined using the association between linearized genetic differentiation (F ST /(1 À F ST ); Slatkin 1995) and hydrological distances (km) among sites and the shortest hydrological distances between sites, with a Mantel test for significance (9999 permutations) in GENALEX 6.0 (Peakall and Smouse 2006). However, A. pellucida prefers shallow, sandy habitats so hydrological distances were determined at the drainage scale using two methods: shallow water restriction (assumes individuals avoid open water and calculates shoreline distances through lakes) and open-water dispersal (uses the shortest water distances among rivers including dispersal through open water).

Hierarchical genetic analysis
Analysis of molecular variance (AMOVA) was used to hierarchically partition genetic variation within each drainage into three levels: among rivers, among sites within rivers, and within sites using ARLEQUIN. We also identified the number of population genetic clusters using the Bayesian-based clustering program STRUCTURE 2.3.4 (Pritchard et al. 2000). When the model of K = 1 could be rejected, we used the DK method to select K (Evanno et al. 2005) as implemented in STRUCTURE HAR-VESTER (Earl and von Holdt 2012), and the process was repeated on all recovered Ks in a hierarchical approach as described in Roy et al. (2012) see Fig. S1. STRUCTURE runs were performed in five iterations for each K, each with a 100,000 burn-in, 1,000,000 Markov chain Monte Carlo (MCMC) generations, allele frequencies correlated, and admixture allowed. The number of genetic clusters was allowed to range from K = 1 (range-wide panmixia) to the total number of rivers plus one (K = 17). Runs were compiled using full searches in CLUMPP1.1.2 (Jakobsson and Rosenberg 2007) and plotted with DI-STRUCT1.1 (Rosenberg 2004). To explore within-river structure, we performed additional full dataset STRUC-TURE runs using the site of capture as a location prior. We also performed STRUCTURE runs on smaller riverspecific datasets to further resolve within-river structure. To visualize the relative divergence of the sites and rivers, we performed a principal coordinate analysis (PCoA) using a pairwise matrix of F ST values in GENALEX. We used BARRIER 2.2 (Manni et al. 2004) and the landscape genetic approach of Monmonier's maximum difference algorithm across the range to identify breaks in gene flow patterns among geographically close sites. In BARRIER, pairwise estimates of F ST were mapped onto a matrix of the population geographic coordinates (latitude and longitude), and the Monmonier's maximum difference algorithm identified which of the borders between neighboring populations exhibited anomalous genetic divergence relative to spatial separation.

Contemporary versus historic influences
As the genetic signature from historic colonization patterns may persist and affect estimation of contemporary connectivity patterns, population genetic structure should be analyzed to test for possible large-scale patterns consistent with historic gene flow patterns (Duvernell et al. 2008). To determine the potential influence of historic drainage connectivity on contemporary genetic structure, we tested the relative partitioning of genetic variance identified by historic versus contemporary groups of sites using AMOVA. The contemporary site grouping (based on present-day drainage connectivity) has three groups: (1) sites in the Great Lakes drainage; (2) sites in the Ohio and Wabash drainages; and (3) sites in the St. Lawrence drainage. Under the historic connectivity hypothesis, drainages were grouped based on preglaciated patterns of water drainage (Underhill 1986;Mandrak and Crossman 1992): (1) Great Lakes and Wabash drainage sites; (2) Ohio drainage sites; and (3) St. Lawrence drainage sites. The proportion and significance of genetic variance partitioned within and among groups for each hypothesized grouping pattern was assessed hierarchically using AM-OVA in ARLEQUIN. If the historic group model explains more variance than the contemporary river connectivity model, then historic effects still influence the structure of genetic connectivity across the range. We used the corrected Akaike information criterion (AIC c ; Burnham and Anderson 1998) to identify the best-fit model based on variance explained (Halverson et al. 2007).

Dispersal
To quantify patterns of dispersal among the sites, rivers, and drainages, we performed a self-exclusion analysis in GENECLASS. Individual fish were excluded/assigned to sites, rivers, and drainages using the Bayesian method of Rannala and Mountain (1997) and the Paetkau et al. (2004) Monte Carlo simulation as implemented in GENE-CLASS 2 (Piry et al. 2004) with a = 0.05, using 100,000 simulated individuals. A fish was considered excluded from a site, river, or drainage of capture if the Bayesian probability was less than 0.05, and assigned if the Bayesian probability was equal to or greater than 0.05: This results in conservative exclusion outcomes.

Sampling and marker assessment
A total of 1051 individuals were collected from 39 sites in 16 rivers across the entire species range from June 2010 to November 2011 (Fig. 1). Across sites, microsatellite allelic richness ranged from 2.64 to 5.87 (Table 1). Significant departures from HWE were found in eight of 390 possible locus-by-site combinations following Bonferroni correction (P < 0.001; Table S1). Five sites were monomorphic at locus Esd3 (HRc1, HRc2, HRm3, HRm1, LK), while site CC was monomorphic at locus EosC6. Seven of the locus-by-site deviations from HWE were attributed to null alleles by MICROCHECKER; however, no single locus had more than two sites deviating from HWE, and we conclude that null alleles are not substantially influencing our results. Significant (P < 0.001) linkage disequilibrium was determined for five of 390 possible locus-by-locus combinations over all the sites, with no two loci identified in linkage disequilibrium for more than one site indicating that our marker loci likely are not linked.

Genetic differentiation
Within-river pairwise F ST values among sites were generally low and nonsignificant (after Bonferroni correction), although some (<10%) between-site F ST values were substantial and significant ( Table 2). The pairwise exact tests of allele frequency distribution differences resulted in a higher proportion of significant between-site differences (51% significant; Table 2); this is likely due to the much higher sensitivity of the exact test. Pairwise F ST values among rivers within each drainage were substantially higher (0.021-0.18; Table 3) and all but three pairwise F ST values were highly statistically significant after Bonferroni correction (88%; Table 3). F ST values were even higher when rivers were compared among drainages (Table 3)

Isolation by distance
Due to limited numbers of within-river sample sites, IBD was only assessed in three rivers (Maumee, Grand and Thames rivers) in the Great Lakes drainage and one river (Hocking River) in the Ohio drainage. Significant withinriver IBD was found for the Maumee River (R 2 = 0.61, P = 0.039); however, no significant IBD was found in the Hocking River. Low F ST values among all sites in the Thames and Grand rivers resulted in a lack of IBD correlation for both rivers (R 2 = 0.035, P = 0.21 and R 2 = 0.021, P = 0.21, respectively). Mantel tests of IBD among rivers within each drainage showed that both the Ohio drainage (R 2 = 0.18, P = 0.004) and Great Lakes drainage (R 2 = 0.80, P = 0.0001, straight line and R 2 = 0.79, P = 0.0001, shallow water distances) had significant IBD (Fig. 2). Neither the Wabash drainage (R 2 = 0.79, P = 0.125) nor St. Lawrence drainage (R 2 = 0.52, P = 0.084) adhered to an IBD pattern of divergence.

Range-wide genetic structure
Individual AMOVAs for each drainage revealed low among-site (within river) genetic variance: Ohio drainage  However, the among-river genetic variance component was 15-20 times higher in the three drainages: Ohio drainage (6.50%, P < 0.0001), Great Lakes drainage (6.29%, P < 0.0001), St. Lawrence drainage (10.52%, P < 0.0001). The highest proportion of genetic variance in all analyses was attributed to the within-sites component: Ohio drainage (93.09, P < 0.0001), Great Lakes drainage (93.39%, P < 0.0001), St. Lawrence drainage (89.02%, P = 0.116). The Wabash drainage was excluded from the AMOVA analysis because the within-river sample sites were not replicated (Table 1). STRUCTURE revealed two cluster patterns with approximately equal likelihood (based on DK; Fig. S1). At K = 2, STRUC-TURE grouped sites from the Wabash drainage with the Great Lakes and St. Lawrence drainages, while the Ohio drainage sites grouped separately (Fig. 3A). Our hierarchical analysis also recovered seven clusters (K = 7) with population delineation at chiefly the river level (Fig. 3A). STRUCTURE runs using location priors did not produce differing results. STRUCTURE runs on reduced datasets from individual systems also supports genetic structure at the river level. Principal coordinate analysis (PCoA) revealed a similar delineation among sites in the Ohio drainage versus the remaining range-wide sites along the first axis (Fig. 3B). The PCoA also showed a clear division of the St. Lawrence drainage from other sites (Fig. 3B). PCoA further supported the STRUCTURE results, as the Wabash sites clustered with the Great Lakes sites (Fig. 3B). BARRIER identified three major genetic breaks: The first separated the Ohio drainage from the rest of the range, and the second genetic barrier isolated the Champ-lain Canal site from all other sites (Fig. 1). The third genetic barrier isolated the St. Lawrence drainage from the Great Lakes drainage (Fig. 1).

Contemporary versus historic influences
AMOVA results for both historic and contemporary connectivity models yielded highly significant models; however, a greater proportion of the among-group genetic variance was explained when the groups reflected the historic connection between the Wabash River and Great   Lakes drainages (8.15%, P < 0.0001), as opposed to contemporary connectivity alone (5.09%, P < 0.0001). The DAIC c between the two AMOVAs was 12.6, highly supportive for the historic model (AIC c = 2763) versus the contemporary model (AIC c = 2686) of genetic variation. Both AMOVA analyses revealed substantial and very similar components of the genetic variance attributed to within-river variation (historic = 87%, P < 0.0001 and contemporary = 88%, P < 0.001).

Dispersal
GENECLASS assignment resulted in a total of 120 individuals conservatively excluded from their "site of capture", ranging from 3.0% to 33% of the individuals caught at a given site (Fig 4). Of the fish excluded from their site of capture, most were assigned to another site within the same river they were captured from, or to an adjacent river (Table 2). A total of 20 fish failed to assign to any site sampled within the study, of those most were captured in the Grand River (Tables 2 and 4).

Discussion
Freshwater fish species inhabiting formerly glaciated regions commonly exhibit genetic signatures that reflect the influence of historical glacial refugia and recolonization patterns (Costello et al. 2003;Poissant et al. 2005;Stepien et al. 2007;Boizard et al. 2009;Shikano et al. 2010;Walter et al. 2012). Our data reveal the persisting influence of historic, postglacial drainage patterns on large-scale (range wide) patterns of genetic divergence. On the other hand, our analyses show little or no evidence for contemporary connectivity (i.e., gene flow) among drainages perhaps, not surprisingly, given the large hydrological distances between most rivers in the study, the limited dispersal capabilities of this small benthic fish, and unsuitable habitat separating some of the drainages. However, our data show extensive genetic connectivity among habitat patches within all sampled rivers, regardless of anthropogenic barriers (e.g., low Grand River genetic differentiation despite separation of sites by a dam). The high dispersal among sand bars identified in our analysis challenges previous conclusions regarding the sedentary nature of A. pellucida. In general, the nature of the freshwater "landscapes" promotes low genetic structure within rivers and higher genetic structure among rivers in freshwater fishes (M€ akinen et al. 2006;Cook et al. 2007;Shikano et al. 2010). Published exceptions to this pattern, where sculpins and darters exhibit high withinriver genetic structure, are attributed to anthropogenic barriers to dispersal (H€ anfling and Weetman 2006;Beneteau et al. 2009). For A. pellucida, we did not expect low levels of within-river genetic structure as their suitable habitat is fragmented within rivers (both naturally and anthropogenically). However, our analyses clearly indicate substantial movement of individuals among the sampled habitats. The lack of genetic structure within the sampled rivers likely reflects species-specific dispersal that counteracts patchy habitat distribution.

S C 2 S C 3 H R m 1 H R m 2 H R m 3 H R c 1 H R c 2 S M S J S y d T H u 1 T H u 2 T H u 3 T H d 1 T H d 2 T H d 3 G R u 1 G R u 2 G R u 3 G R d 1 G R d 2 G R d 3 R
For A. pellucida, a combination of long and short within-river dispersal (or "stratified dispersal") may contribute to the lack of within-river genetic structure and this, in turn, would act to buffer individual sand bar populations from genetic drift effects and loss of genetic diversity (Bronnenhuber et al. 2011). Hydrological distances among sites within rivers were generally not positively correlated with genetic differentiation. This lack of IBD is consistent with stratified within-river dispersal, restricting genetic differentiation among sample sites, a pattern that is apparent in the Thames and Grand rivers. Generally speaking, within-river IBD is expected unless dispersal distances are larger than the spatial extent of the study area or if sufficient long-distance dispersal events occur to swamp genetic drift effects (McGlashan and Hughes 2001). Disruptions to within-river IBD could also result from recurring population bottlenecks, preventing migration-drift equilibrium, as suggested for other darter species (Turner and Trexler 1998;Johnson et al. 2006). However, we observed no evidence for low genetic diversity or elevated F IS values (see Table 1). Thus, we conclude that the lack of within-river genetic structure in populations likely reflects primarily stratified dispersal. Within-river movements of adults may be directly influenced by their dependence on a temporally unstable habitat (depositional sand bars). That is, fish are forced to Table 4. Summary of GENECLASS exclusion/assignment results for all hierarchically sampled Eastern Sand Darters. Individuals were considered excluded from "site of capture" if their Bayesian probability was less than 0.05, those excluded individuals were then assigned to another site(s) if P > 0.05 for a given site. A total of 20 individuals could not be assigned to any site; therefore, their origin is unknown.  disperse when their preferred habitat is locally lost or degraded, a plausible scenario for sand deposition-based habitat. Within-river movement would be further promoted by a largely nonbenthic larval stage where downstream larval drift could facilitate gene flow within rivers (Simon and Wallus 2006). Independent of the mechanism behind within-river dispersal, the high genetic connectivity demonstrated here indicates that reintroduction efforts using fish taken from the same river would hold little genetic risk as the fish are already well mixed and might be better characterized as assisted dispersal rather than reintroduction.

River
At the drainage scale, among-river IBD patterns in the Ohio and Great Lakes drainages suggest that hydrological distances restrict genetic connectivity among rivers and that the river populations appear to be at, or near, dispersal-drift equilibrium. Very few among-river migrants were identified range wide, indicating that dispersal among rivers is infrequent and results in little amongriver gene flow. No difference in the IBD relationship was observed for shallow water versus straight-line dispersal pathways in the Great Lakes drainage (the only drainage with a large lake in this study), suggesting that open-lake habitat does not represent a major barrier to the dispersal-drift equilibrium for A. pellucida. Based on the pattern of among-river genetic divergence, reintroduction plans should give preference for populations as geographically close to the reintroduction site as possible, assuming fish from the same river cannot be used.
As expected, among-river genetic structure was substantial and significant in all sampled drainages, similar to other darter species within the Great Lakes drainage (Greenside Darter, Beneteau et al. 2009;Rainbow Darter, Etheostoma caeruleum, Haponski et al. 2009). A variety of biotic and abiotic habitat characteristics likely restrict the ability of darters to disperse through freshwater drainages (Jackson et al. 2001), including fast river flow and unsuitable habitat (Cook et al. 2007;Zamudio et al. 2009). We found only three exceptions to substantial among-river genetic divergence in A. pellucida: The first occurred between the Thames and Sydenham rivers, and the other two were among rivers in the Wabash drainage. Low genetic differentiation between the Thames and Sydenham river populations can be explained by either dispersal between the spatially close river mouths in the shallow Lake St. Clair or headwater connections from natural floods or anthropogenic fish movement. The genetic similarity between these two rivers most likely reflects a headwater connection or human-mediated transfer, as suggested for Greenside Darter (Beneteau et al. 2009). The relatively high genetic connectivity among rivers in the Wabash drainage may result from few anthropogenic barriers (e.g., dams, weirs), lower stream flow rates and shorter hydrological distances separating rivers. Unfortunately, our data do not allow us to conclusively identify the source of the anomalous genetic connectivity among rivers.
The genetic divergence of A. pellucida in the Ohio River drainage from the remainder of the species range is likely a result of long-term isolation. Much of the Ohio River drainage, including the sites in our study, were never glaciated, whereas the remaining sites in our study (i.e., Wabash, Great Lakes, and St. Lawrence drainages) were covered most recently by the Wisconsinan continental ice sheet (Trautman 1981, Burr andPage 1986). Following the Wisconsinan glacial retreat (approximately 14,000 years ago), A. pellucida would have colonized the Wabash and Great Lakes/St. Lawrence drainages from the Mississippian refugium (Underhill 1986;Mandrak and Crossman 1992) The genetic similarity between sites in the Wabash and Great Lakes drainages likely reflects the historical connection of the Great Lakes and Wabash drainages following the end of the Wisconsinan glacial period, when excess water from the glacial Lake Maumee (ancestor of presentday Lake Erie) drained into what is now the Wabash River (Underhill 1986). This historic connection between the Wabash and Great Lakes drainages has been previously hypothesized to be a major connection for aquatic organisms recolonizing the Great Lakes from the Mississippian refugium (Underhill 1986;Mandrak and Crossman 1992) and to have driven genetic similarities between mussel populations in the Wabash and Great Lakes drainages (Graf 2002;Elderkin et al. 2007). Another important genetic influence of glacial colonization pathways on populations involves isolated, or "disjunct", species range patterns (Witt et al. 2011). A major genetic break identified in this study occurred between the St. Lawrence drainage and the remainder of the species range. A. pellucida are thought to have colonized Lake Champlain and the St. Lawrence River from the Mississippian glacial refugium through either the Mohawk River of the glacial Lake Iroquois (present-day Lake Ontario), 12,000-13,500 years ago, and subsequently through Lampsilis Lake (presentday St. Lawrence River), 8500-10,000 years ago (Underhill 1986). Alternatively, A. pellucida in the St. Lawrence drainage may have derived from an Atlantic Coastal refugium (Committee on the Status of Endangered Wildlife in Canada (COSEWIC) 2011). The reduced genetic diversity exhibited by the St. Lawrence drainage populations, coupled with their genetic divergence and low connectivity, indicate that these populations merit increased conservation attention. The genetic divergence of the Quebec ESD also supports the recent identification of two conservation units in Canada (termed "designatable units"), the Quebec and Ontario population have separate status and recovery plans (COSEWIC 2009 Our study emphasizes the blending of contemporary and historic influences on the genetic structure of A. pellucida populations throughout the species range. Based on the pattern of among-river genetic divergence, supplementation and reintroduction plans for extirpated systems with currently suitable habitat (Dextrase et al. 2014) should give preference for not only geographically proximal populations, but also those with contemporary and historical genetic connections. This study highlights the influence of historic drainage connectivity and not only reveals genetic cohesiveness between previously connected drainages (e.g., the Wabash-Maumee historical connection) but also provides insight into the negative genetic effects of range isolation in disjunct drainages (e.g., St. Lawrence drainage). Small-scale analyses showed an unexpected lack of genetic structure at the within-river level, consistent with substantial and ongoing dispersal and hence connectivity. The within-river dispersal likely results from the temporal instability of specialized habitat (sand bars) possibly combined with larval drift. Our hierarchical range-wide analysis of the genetic structure in a habitat-specific species clearly demonstrates that speciesspecific life-history traits, such as dependence on specific habitats, can strongly affect genetic diversity patterns, particularly when the preferred habitat is fragmented and temporally unstable.

Supporting Information
Additional Supporting Information may be found in the online version of this article: Figure S1. Hierarchical analysis of population genetic structure using STRUCTURE and STRUCTURE Harvester. Recovered K's were separately analyzed until K = 1 could not be rejected. Table S1. Characterization of ten microsatellite markers used for genetic analysis of Ammocrypta pellucida.