Population Genetics of Seaside Sparrow (Ammodramus maritimus) Subspecies along the Gulf of Mexico

Seaside Sparrows (Ammodramus maritimus) along the Gulf of Mexico are currently recognized as four subspecies, including taxa in Florida (A. m. juncicola and A. m. peninsulae) and southern Texas (Ammodramus m. sennetti), plus a widespread taxon between them (A. m. fisheri). We examined population genetic structure of this “Gulf Coast” clade using microsatellite and mtDNA data. Results of Bayesian analyses (Structure, GeneLand) of microsatellite data from nine locations do not entirely align with current subspecific taxonomy. Ammodramus m. sennetti from southern Texas is significantly differentiated from all other populations, but we found evidence of an admixture zone with A. m. fisheri near Corpus Christi. The two subspecies along the northern Gulf Coast of Florida are significantly differentiated from both A. m. sennetti and A. m. fisheri, but are not distinct from each other. We found a weak signal of isolation by distance within A. m. fisheri, indicating this population is not entirely panmictic throughout its range. Although continued conservation concern is warranted for all populations along the Gulf Coast, A. m. fisheri appears to be more secure than the far smaller populations in south Texas and the northern Florida Gulf Coast. In particular, the most genetically distinct populations, those in Texas south of Corpus Christi, occupy unique habitats within a very small geographic range.


Introduction
Isolated populations of organisms inhabiting patchily distributed habitat often form genetically or morphologically distinct populations that may be recognized as species or subspecies [1][2][3]. From a conservation standpoint, protecting multiple small populations is important to preserving genetic diversity, but can be difficult in practice [4]. The use of subspecies to identify ecologically or morphologically distinct populations has a long history in ornithology [3,5], but issues regarding diagnostic criteria and the role of subspecies in conservation are continually debated and refined (e.g., [6,7]). Furthermore, many subspecific designations predate the modern statistical analyses used to evaluate them [8]. In the United States, the Endangered Species Act (ESA) considers ''distinct population segments'' as potentially eligible for special protections. Thus, it is important to understand whether taxonomically recognized subspecies can reasonably be considered distinct population segments with unique genetic or ecological characteristics [3].
Although the ESA does not specifically define diagnostic criteria [6], population genetic techniques allow a more refined way to identify unique populations and to infer patterns of gene flow among them [3,5,6,9]. Several recent studies have found at least some degree of concordance with population genetic data and subspecific taxonomy (e.g., Melospiza melodia [10], Buteo lineatus [11], Myioborus miniatus [12], Chondestes grammacus [13]), but the vast majority of named avian subspecies have not been evaluated using modern techniques and remain in place largely due to ''historical inertia'' [8].
The Seaside Sparrow (Ammodramus maritimus) is largely endemic to -and often abundant in -salt marshes along the Atlantic and Gulf coasts of the United States; seven extant subspecies are currently recognized [14] based on plumage, song, or for some Atlantic birds, mtDNA (reviewed in [15]). The presence of geographic variation in plumage [15] and song [16] provides a basis for inferring limited gene flow among populations, and some subspecies (e.g., A. m. nigrescens, A. m. mirabilis) have been considered full species in the past [17]. The taxonomic history of the Seaside Sparrow is complex, however, and has been complicated by the difficulty of interpreting subtle plumage characters, within-population plumage variation, and by limited numbers of specimens in fresh basic plumage in museum collections. Furthermore, even obvious differences in plumages among bird populations do not always correlate with genetic differentiation (see e.g., [18]). Based on studies of mtDNA, Ammodramus m. maritimus, A. m. macgillivraii, A. m. mirabilis, A. m, nigrescens, and presumably A. m. pelonota form an ''Atlantic'' clade (A. m. pelonota has not been analyzed due to lack of specimens; [19,20]). Ammodramus m. nigrescens and A. m. pelonota have become extinct since the 1980s, although the distinctness of the latter has been questioned [21]. The ''Gulf Coast'' clade has a more convoluted taxonomic history [15,[22][23][24], but most recently consists of four subspecies (based on plumage characteristics), from west to east along the Gulf Coast: A. m. sennetti (not sampled by Avise and Nelson [20]), A. m. fisheri, A. m. juncicola, and A. m. peninsulae [14]. The latter two populations are currently listed as Threatened in Florida [25,26].
Given that at least one subspecies has gone extinct and other named subspecies have quite narrow distributions, a modern genetic analysis is needed to enable an assessment of conservation risk for Gulf Coast Seaside Sparrows. In this paper we: (1) describe population genetic structure of Seaside Sparrows along the Gulf Coast; (2) ask whether genetically distinct populations align with current subspecific taxonomy, and (3) discuss potential conservation risks based on a more refined understanding of the distribution and genetic variation among these populations.

Study area
We captured 374 Seaside Sparrows with mist nets at nine locations across the northern coast of the Gulf of Mexico during 2012-2013 (Table 1, Fig. 1). Locations were chosen to include all four currently recognized subspecies, and to sample throughout the range of the subspecies with the broadest geographical distribution (A. m. fisheri). Captured birds were banded with numbered aluminum bands; a blood sample was taken from the brachial vein and stored in Queen's lysis buffer [27].

Analytical methods
Microsatellite data were checked for evidence of null alleles and potential scoring problems with MICROCHECKER [38], and withinpopulation HWE and LD with GENEPOP [39]. Population sample characteristics (allele richness, observed vs expected heterozygosity) were summarized with GENALEX ( [40]; Table 3). Pairwise population differentiation analysis (F ST ; 5,000 permutations to test for significance) and Factorial Correspondence Analysis (FCA; used to visualize degree of difference among population samples) were performed in GENETIX [41]. Jost's D ST ( [42]; as an alternative to F ST ) was estimated in SMOGD [43]. We tested three subsets of the data for isolation by distance (IBD) in IBDWS [44]  . Effective genetic population size (N e ) was estimated for all nine populations using the LD method with a random-mating model, 0.05 as the lowest allele frequency used, and a jackknife approach to estimating 95% CIs in the program NEESTIMATOR [45]. We chose this LD method because it is widely-used, incorporates a bias-correction for calculating confidence intervals, and requires relatively few assumptions regarding the samples [46]. An AMOVA of ND2 sequences was performed in ARLEQUIN (v. 3.5.1.2; [47]), using genetic distances (F ST ) and pairwise differences (100 permutations) for comparisons. A 95% confidence interval haplotype network was constructed with TCS (v 1.21; [48]).
We used two Bayesian clustering programs to evaluate the distribution of genetic structure within our study area. For Gulfwide analyses we used a reduced number of individuals from locations 4 and 5 in order to maintain a more balanced number of samples from each of the nine locations. In the program STRUCTURE [v2.3.4; [49,50] we ran an admixture model (50,000 burn-in, 1,000,000 iterations) with correlated allele frequencies, using sampling locations (locprior), and otherwise default settings We performed 20 runs for each hypothesis of K = 1-11. We processed STRUCTURE output with STRUCTURE HARVESTER [51], used CLUMPP [52] to condense data from multiple runs, and created a graphic of population assignments with DISTRUCT [53].
We used GENELAND [54] to explore spatial models for both the entire Gulf-wide dataset and two further subsets of the data. For the Gulf-wide analyses we used a no-admixture and uncorrelated allele frequencies model because IBD can cause problems for some of these algorithms [55]. We performed 10 runs for each hypothesis of K = 1-11 (1,000,000 iterations; thinning = 1,000). Two additional GENELAND analyses were run to further explore data from locations in southern Texas and Florida. To examine an apparent hybrid zone near Corpus Christi, Texas we used a correlated allele model and the admixture model for hybrid zones (populations 1-3; 5 runs for K = 1-3; 1,000,000 iterations, thinning = 1,000). To examine weak differentiation (based on F ST ) of the two named subspecies sampled in Florida (locations 8 and 9) we used a correlated allele model (5 runs for K = 1-2; 1,000,000 iterations, thinning = 1,000). For all GENELAND analyses we set coordinate uncertainty to 0.3 to allow for the possibility of different individuals captured at the same point to be assigned to different populations. Because GENELAND algorithms may produce ''ghost'' populations (i.e., inferred populations containing no sampled individuals), especially at large spatial scales, we limited our acceptance of GENELAND results to the most parsimonious estimation of K that did not include ''ghost'' populations [54].
To further explore the apparent admixture zone near Corpus Christi, Texas, we used two Bayesian approaches to estimate historic (MIGRATE-N; [56,57]) and more recent (BAYESASS; [58]) geneflow among the three Texas sampling areas. MIGRATE-N (v. 3.6.4; [56,57]) estimates scaled population sizes (h = 4N e m, where m is the mutation rate) and migration rates (M = m/m, where m = migration rate). We used a Bayesian framework and a Brownian motion microsatellite model, with a proposed connection matrix that included estimation of all h and M parameters with the exception of migration between locations 1 and 3 (i.e., a 'stepping stone' model), as direct migration between these two locations is biologically unreasonable considering the sedentary nature of Seaside Sparrows along the Gulf Coast [14]. Following a series of preliminary runs to explore the possible ranges of priors (uniform distribution), we used the following start and search parameters: both h and M were estimated using F ST , MCMC runs used one  Table 1  Griscom [62] noted that sennetti and fisheri from this area may intergrade, even though it is the type locality for sennetti. Subspecific identities based on summary by Robbins [22]. ''Site #'' refers to localities shown in Fig. 1 long chain, recording every 100 steps, and visiting 4610 6 geneologies. The program BAYESASS (v3.0) estimates recent immigration rates by estimating the proportion of migrant ancestry within population samples and individuals [58]. We followed the recommendations of Wilson and Rannala [58] and the program user's manual by running several trial runs to find search parameters that produced acceptable acceptance rates (0.2-0.6) for estimates of migration rate (M), individual migrant ancestry (A), and inbreeding coefficients (F). We subsequently performed multiple runs with different seeds to insure comparable results were obtained between runs. We ran 10 7 iterations, with 10 6 as burn-in, with a sampling interval of 10 3 . Mixing parameters were dM = 0.1, dA = 0.45, and dF = 0.45. We examined the trace outputs of the log-probabilities to insure the MCMC had converged using the program TRACER (v.1.6.0 [59]). Estimates of ancestry for each individual were summarized by adding the estimated ancestry coefficients (source, 1 st generation or 2 nd generation migrant from each population) to characterize the likelihood of migrant ancestry.

Results
We found no evidence of technical problems relating to microsatellite genotyping, and no evidence of within-location LD or deviation from HWE. The average number of alleles per locus was lowest (6.1) in A. m. sennetti from location 1, but similar (range = 9.0-11.4) among all other locations ( Table 3). Estimates of N e for the nine population samples were generally high; most confidence intervals included infinity, which Do et al. [45] interpret as a lack of LD signal to use for estimation of N e (Table 3). Although this may also be interpreted to mean the population sizes are indeed quite large, high values could also be due to other factors, such as migration and selection [60]. In our study, few populations are likely to be completely isolated, and immigration from adjacent areas seems likely. A majority of pairwise F ST values were statistically significant (P,0.001); both F ST and D est were consistently higher in all comparisons involving location 1 (southernmost A. m. sennetti), and lowest within samples encompassing A. m. fisheri (Table 4). The FCA shows that populations 3-7 are most similar to each other, but that populations 1, and [8+9] are distinct (Fig. 2).
All six locations sampled for ND2 contained at least one unique haplotype ( Table 2, Fig. S2 in File S1). Location 1 was unique in missing the most frequent haplotype found in the other five locations, and also in having only two haplotypes in similar proportions. Overall population structure was significant for the six locations considered (w ST = 0.1833, P,0.0001), and was also evident when excluding location 1 as an outlier with a very different distribution of haplotype frequencies (w ST = 0.0219, P = 0.0585). All pairwise comparisons of w ST involving location 1 were significant (P = 0.0180 for Pop 1 vs Pop 2; all other P, Table 2. ND2 haplotype frequencies for selected population samples of Seaside Sparrow along the coast of the Gulf of Mexico (see Fig. 1 for sampling localities).

Bayesian inference of K
Inspection of plots of summary statistics provided by STRUCTURE indicated a sufficient burn-in length and number of post-burn-in iterations. For the Gulf-wide analyses, STRUCTURE indicated the best support for K = 3, based both on evaluation of mean estimated ln probabilities for each hypothesis (Figs. 3 and 4), and also using the method of Evanno et al. [61] (not shown). The hypothesis of K = 4 had the next best support, and differed only in treating location 2 (Aransas Co., TX) as a separate population. GENELAND found a best estimate of K = 3 that corresponded well with STRUCTURE output (Fig. 4), and the next best GENELAND model also suggested K = 4, also treating location 2 separately (not shown).
The admixture analysis in GENELAND of locations 1-3 suggests that birds sampled in Aransas Co. have mixed ancestry of A. m. sennetti from further south and A. m. fisheri from further north (not shown), and this is congruent with the broader-scale output of the STRUCTURE analysis (see Fig. 4, location 2). A correlated allele model in GENELAND consistently recovered a single population among the two samples from Florida (populations 8 and 9), consistent with the low levels of differentiation observed via F ST and D est .
The MIGRATE-N analysis of the admixture zone between A. m. sennetti and A. m. fisheri suggests that our samples from Aransas Co., Texas have historically experienced unequal gene flow from populations to the north and south. Migration rates (M) between populations 1 and 2 are roughly similar in each direction. Estimated migration rates between populations and 2 and 3 are asymmetric, with population 3 contributing a greater number of migrants into population 2 (Fig. 5). Recent gene flow among the three Texas sampling locations (as estimated in the program BAYESASS) also appears asymmetric, in that, in contrast to populations 2 and 3, population 1 has not received many recent migrants (Table 5), and a large proportion (ca. 0.8) of individuals in populations 2 and 3 have estimated ancestry coefficients consistent with 2 nd generation (or greater) migrant (Fig. 6), as would be expected when gene flow is high [58].

Discussion
At the broadest geographic scale, our microsatellite and mtDNA analyses show that genetically differentiated Seaside Sparrow populations along the coast of the Gulf of Mexico are not entirely consistent with current subspecific taxonomy. Our microsatellite data are consistent with the recognition of three populations: A. m. cf. sennetti (south of Corpus Christi, TX -see below), A. m. fisheri, and a combined group including A. m. juncicola and A. m. peninsuale. Ammodramus m. cf. sennetti from our southernmost site in Texas (location 1) are significantly diverged from all other population samples, including the apparent hybrid A. m. sennetti 6 fisheri population in Aransas Co., Texas (location 2). The two Florida subspecies we sampled (A. m. juncicola and peninsulae) are not significantly differentiated from each other at the 14 loci we sampled, and form a single cluster diverged from the other subspecies in Gulf-wide Bayesian analyses. The subspecies with the largest geographic range (fisheri) exhibits relatively weak geographic structuring and IBD throughout its range from the mid-Texas coast to eastern Mississippi.
That Seaside Sparrows sampled near Corpus Christi, Texas (location 2) show genetic evidence of admixture is concordant with Griscom's [62] discussion regarding lighter and darker plumages of Seaside Sparrows between Nueces Bay and Matagorda Bay (a distance of ca. 130 km; see Fig. 5): ''the series is sufficiently large, so that one extreme passes into the other extreme by a perfect gradation'' (p. 106). Saltmarsh habitat dominated by Spartina and Juncus (i.e., the usual habitat of Seaside Sparrows along the Gulf Coast) reaches its southern limit in Texas [63], and thus from an ecological perspective, the admixture zone of A. m. sennetti and fisheri aligns with a shift to different coastal plant communities  from north to south. The birds we sampled in Cameron Co. (location 1) were found in habitat dominated by Borrichia and Batis -a plant community quite unlike any other Seaside Sparrow habitat. This transition in plant communities in the Corpus Christi area is also concordant with Webb's [64] delineation of a ''Rio Grande'' biogeographic zone for mammals. From a conservation standpoint, the existence of a morphologically, ecologically, and genetically distinct population restricted to a coastal area from Cameron Co. north to somewhere south of Corpus Christi should be of concern. The coastline distance between locations 1 and 2 is ca 200 km, and we do not know (1) where within that range the population genetic characteristics shift to those of A. m. cf. sennetti from location 1.; and (2) how much of this part of the Texas coast is occupied by Seaside Sparrows. Our data suggest that the geographic range of this taxon is smaller than that most recently described by Oberholser [65] as between Refugio and Cameron counties. Additional data are needed to better characterize and delineate the apparent admixture zone between Cameron Co. and Brazoria Co, and to explore whether the Aransas population (population 2) might be better considered an independent taxon. Based on the foregoing, we suggest that a taxonomic reassessment of ''A. m. sennetti'' is needed. The type locality of sennetti is Corpus Christi [59], ca 40 km southwest of our sampling location 2 (Aransas Bay). Griscom [62] was either unaware of, or did not accept populations south of Corpus Christi, and restricted the range of A. m. sennetti to Nueces and Copano Bays (see Fig. 5). Microsatellite allele and ND2 haplotype frequencies of the birds we sampled in this area (location 2; , 20 km east of Copano Bay) are more similar to allele frequencies found throughout the range of A. m. fisheri than to frequencies found in A. m. cf. sennetti from location 1 to the south. This prompts the question of whether the type specimen of A. m. sennetti is in fact of mixed ancestry, or whether the range of A. m. fisheri has expanded southward since the late 1800s, when A. m. sennetti was described. In light of this, until the type specimen of A. m. sennetti is analyzed from a genetic standpoint, we refer to the birds from Cameron Co., Texas (location 1) as A. m. cf. sennetti.
The distinction between A. m. juncicola and A. m. peninsulae has long been debated, generally because plumage variation between the two is slight relative to plumage variation within each area [15]. Pairwise estimates of F ST and D ST for these two populations (8 and 9; ca 170 km apart) are marginally greater than several estimates for populations of A. m. fisheri separated by greater geographic distances. GENELAND analyses -which are explicitly designed to assess fine-scale genetic structure -failed to distinguish between the two Florida populations we sampled. Based on our relatively limited genetic sampling, however, we are hesitant to make any taxonomic suggestions regarding these taxa, and genome-wide sampling, for example, could discover important differences between the populations (see [66]). Nonetheless, even as a combined group, they are genetically (this study) and ecologically [16,24,67] distinct from other populations of Seaside Sparrow. Moreover, marsh habitat is restricted in the region to an estimated maximum of 275 and 376 km 2 for A. m. juncicola and peninsulae, respectively [25,26]. Within this range, occurrence of both taxa is often spotty, even in seemingly suitable habitat ( [24], SW pers. obs.), and additional data are needed to better understand the size, distribution, and demography of these populations. The limited extent of salt marsh within the geographic range of these taxa suggests that marsh loss in this region could decimate these populations [25,26].
We note that the western panhandle of Florida (Escambia Co. to approximately Bay Co.) remains a gap in our knowledge, at least partially because the limited amount of saltmarsh habitat means that there are very few Seaside Sparrows in the area [23,24]. Although some published range maps depict this area as within the range of A. m. juncicola (e.g., [14,20,68]), this view may have begun as a misinterpretation of the range described (albeit somewhat ambiguously) by Griscom and Nichols [69]. This interpretation was abandoned by Robbins [22], and we consider  Numbers below STRUCTURE output correspond to the nine sampled populations of Seaside Sparrow (see Fig. 1). GENELAND output has been cropped to remove much of the area (Gulf of Mexico) not occupied by Seaside Sparrows. Note that no birds were sampled from the western Florida panhandle, and no population assignment for birds in this area is possible based on our data. Axes (latitude and longitude) are only relevant to GENELAND output.  [25,26] as the most reasonable interpretation of the original descriptions for those taxa [23,69]. We caution against interpreting our Fig. 4 as supporting the existence of A. m. juncicola and/or A. m. peninsulae in the western Florida panhandle because in this case it is a matter of the software (GENELAND) ''filling in the gaps.'' Analyses of samples from the western Florida panhandle are needed to understand the population genetic affinities of these birds.
In contrast to the case of A. m. sennetti and the situation in western Florida, the case of A. m. fisheri seems relatively straightforward from Brazoria Co., Texas eastward. Population genetic characteristics of Seaside Sparrows sampled from Brazoria Co. to Jackson Co., Mississippi (populations 3-7) were found to be relatively homogenous and essentially panmictic. This broad distribution and large (combined) population size suggests this taxon is of relatively low conservation concern at this time.

Summary
In light of recent extinctions (A. m. nigrescens and A. m. pelonota) and current Federal Endangered status (A. m. mirabilis) of taxa within the Seaside Sparrow, our results support the consideration of Seaside Sparrows along the northern coast of the Gulf of Mexico as comprising three ''distinct population segments:'' (1) Ammodramus m. cf. sennetti, (2) A. m. fisheri, which ranges from at least Brazoria Co., TX, east to Jackson Co., MS, and (3) a combined A. m. juncicola + A. m. peninsulae group which ranges from Bay Co. south to Pasco Co., FL. Additional data are needed to better understand the situation in Texas south of Brazoria Co., but at the very least our data indicate that a highly distinct taxon (A. m. cf. sennetti) occurs south of about Corpus Christi. This distinct and isolated taxon may be in need of additional protection and management.

Supporting Information
File S1 Figures S1 (showing plots of isolation by distance for microsatellite and mitochondria genotype data) and S2 (showing and ND2 haplotype network). (DOCX) Table 5. Estimated proportion of self-recruits (m; i.e., the proportion of individuals estimated to have originated within the population) in each of the three Texas population samples, as estimated by the program BAYESASS. The 95% credible set is calculated as the mean 6 SD*1.96. Location numbers are as in Fig. 1. doi:10.1371/journal.pone.0112739.t005