Ecological characteristics explain neutral genetic variation of three coastal sparrow species

Eco‐phylogeographic approaches to comparative population genetic analyses allow for the inclusion of intrinsic influences as drivers of intraspecific genetic structure. This insight into microevolutionary processes, including changes within a species or lineage, provides better mechanistic understanding of species‐specific interactions and enables predictions of evolutionary responses to environmental change. In this study, we used single nucleotide polymorphisms (SNPs) identified from reduced representation sequencing to compare neutral population structure, isolation by distance (IBD), genetic diversity and effective population size (Ne) across three closely related and co‐distributed saltmarsh sparrow species differing along a specialization gradient—Nelson's (Ammospiza nelsoni subvirgata), saltmarsh (A. caudacuta) and seaside sparrows (A. maritima maritima). Using an eco‐phylogeographic lens within a conservation management context, we tested predictions about species' degree of evolutionary history and ecological specialization to tidal marshes, habitat, current distribution and population status on population genetic metrics. Population structure differed among the species consistent with their current distribution and habitat factors, rather than degree of ecological specialization: seaside sparrows were panmictic, saltmarsh sparrows showed hierarchical structure and Nelson's sparrows were differentiated into multiple, genetically distinct populations. Neutral population genetic theory and demographic/evolutionary history predicted patterns of genetic diversity and Ne rather than degree of ecological specialization. Patterns of population variation and evolutionary distinctiveness (Shapely metric) suggest different conservation measures for long‐term persistence and evolutionary potential in each species. Our findings contribute to a broader understanding of the complex factors influencing genetic variation, beyond specialist‐generalist status and support the role of an eco‐phylogeographic approach in population and conservation genetics.


| INTRODUC TI ON
Comparative population-level genomic analyses across species hold the power to expand our understanding of the effects of ecology, historical demography and contemporary connectivity on evolutionary patterns and processes.Comparative genomic studies often use a phylogeographic approach to describe spatial patterns of diversity and structure at a broad scale.The field of comparative phylogeography has historically focused on an adherence to concordance as a way of testing expectations of spatial or temporal patterns in genetic variation, such that the same extrinsic historical or biogeographic factors that shape species assemblages across a landscape also explain intraspecific genetic structure (Avise, 1992;Hewitt, 2004).Any observed discordance or species-specific differences in these patterns were treated as outliers requiring further investigation (Avise, 2009;Ornelas et al., 2013).
Recent arguments to move beyond strict expectations of concordance suggest examining intrinsic or species-specific differences in spatial genetic structure using an eco-phylogeographic approach.Eco-phylogeography is a concept integration which advocates that observed phylogeographic structure is a result of the interaction between organisms and their environment.Using an eco-phylogeographic approach requires that ecological traits be included in predictions of genetic variation in co-distributed species (Paz et al., 2015).This approach moves beyond abiotic or extrinsic factors to include ecological or life-history traits as predictors of genetic variation and divergence across taxa, which can provide insights about the role of these biotic traits at a microevolutionary level (Papadopoulou & Knowles, 2016;Paz et al., 2015).Indeed, studies have linked observed differences in spatial genetic structuring among species to several taxon-specific traits, for example, morphology, behaviour and life history (Papadopoulou et al., 2009;Paz et al., 2015;Qu et al., 2010;Thomaz & Knowles, 2020); habitat or microhabitat preferences (An et al., 2020;Beavis et al., 2011;Massatti & Knowles, 2014;Zhang et al., 2012); degree of ecological specialization (Dellicour et al., 2015;Rocha et al., 2002); and response to environmental (Moritz et al., 2012) and landscape (Dellicour et al., 2015) change.
An eco-phylogeographic approach to understanding species' patterns of spatial genetic structure may be especially relevant for those taxa highly influenced by climate change, as generalizations about the effects of climate-induced changes on entire ecological communities can be difficult to make (Massatti & Knowles, 2014).
Environmental changes do not impact all ecological niches equally, even within shared environments (Beavis et al., 2011;Langerhans & DeWitt, 2004).Thus, refined hypotheses that account for demographic history, current population conditions and speciesspecific habitat preferences or life-history traits will help elucidate the interactions these species have with their environment while also improving the predictive framework (Papadopoulou & Knowles, 2016).This improved understanding of species and environment interactions will aid conservation and management efforts by enabling better predictions of evolutionary responses to anthropogenic landscape alterations (Gagnaire, 2020) and will in turn inform how we approach and sustain effective management efforts.
Here we compare genetic diversity and population structure in three closely related, co-distributed and congeneric tidal marsh sparrow taxa in an eco-phylogeographic context, considering their evolutionary history, ecological specialization and contemporary distribution and population pressures.Our three focal taxa-the Acadian subspecies of the Nelson's sparrow (Ammospiza nelsoni subvirgata), the saltmarsh sparrow (A.caudacuta), and the nominate, northern subspecies of the seaside sparrow (A.maritima maritima) are tidal marsh endemics that breed along the Mid-to North Atlantic coast.These taxa differ with respect to their distribution and evolutionary history in tidal marsh ecosystems, which has led to a gradient of tidal marsh adaptation and ecological specialization.Each of these tidal marsh sparrow species is of conservation concern, due to their vulnerability to climate change-induced sealevel rise and habitat degradation/fragmentation, with two of them (saltmarsh and Nelson's sparrow) experiencing substantial population declines within the last two decades (Bayard & Elphick, 2011;Correll et al., 2017).We expect species-specific niches, marsh colonization history and resulting degree of ecological specialization, and contemporary demographic processes have resulted in differences in genetic variation and spatial genetic structuring at a microevolutionary level for each species.
The specialist-generalist variation hypothesis (SGVH) predicts that generalist species will harbour more genetic diversity and show less genetic differentiation across populations than those of specialist species (Li et al., 2014).The underpinning of this hypothesis suggests that since generalist species use a broader set of resources that may be more widely distributed across the landscape, they encounter fewer barriers to dispersal, resulting in higher gene flow among populations.Many recent studies present results inconsistent with these expectations; therefore, it is unclear how predictive this hypothesis is across wild populations (Griffith & Sultan, 2012;Martinossi-Allibert et al., 2017;Matthee et al., 2018;Titus & Daly, 2017).Following this framework, expectations of population structure and diversity across the three sparrow species may follow the species-specific level of ecological specialization consistent with evolutionary history with the marsh.Other factors, however, including contemporary demographic processes, the fragmented and patchy distribution of salt marsh, and fine-scale habitat features across the species ranges, may also influence genetic variation.Using known species-specific ecological and evolutionary differences in combination with information from prior studies, habitat data, and population genetic theory, we developed predictions for the following genetic metrics: neutral population structure, isolation by distance (IBD), genetic diversity and effective population size (N e ; summarized in Table 1 and described in methods) for each species.As these species are vulnerable to ongoing and future climate impacts, we take particular interest in the effect current population pressures TA B L E 1 Predictions of neutral population structure, isolation by distance (IBD), genetic diversity and effective population size for each of the three Ammospiza sparrows within an ecophylogeographic framework.North America and around Hudson and James Bays), although it may have occupied coastal habitats intermittently during interglaciation events spanning the last half million years (Walsh et al., 2021).
Compared to the saltmarsh sparrow and the seaside sparrow, the Acadian Nelson's sparrow has a broader ecological niche that includes brackish upriver marshes and hayfields in addition to salt marshes (Greenlaw, 1993;Nocera et al., 2007;Shriver et al., 2005) and is the least specialized with respect to tidal marsh adaptations (sensu Greenberg, 2006).The Acadian Nelson's sparrow is distributed in a patchy mosaic of available habitat within their range, existing at low densities in small, isolated marshes, along the mouth and tidally influenced portions of rivers (Walsh et al., 2015).Although the Acadian Nelson's sparrow is not as specialized as the other eastern Ammospiza sparrows, and generalist species are typically predicted to have higher genetic diversity than specialists as predicted by the SGVH (Li et al., 2014), its recent colonization history relative to the other two species may have resulted in stronger recent purifying selection that eliminated genetic diversity (Charlesworth, 1996;Charlesworth et al., 1993;Cvijovic et al., 2018).We also expected that, because its current distribution is limited to small disjunct habitat patches due to the fragmented nature of the habitat across its range, its smaller, more isolated populations relative to the other two species are likely subjected to greater effects of genetic drift, a source for both differentiation and genetic diversity loss.
Accordingly, we expected to find patterns of local population structure, a lack of IBD, relatively low genetic diversity, and a small N e .
The saltmarsh sparrow breeds along the Atlantic coast from southern Maine to Virginia (Greenlaw & Woolfenden, 2007;Nocera et al., 2007).Where its range overlaps with the Acadian Nelson's sparrow, the two taxa, which diverged approximately half a million years ago (Rising & Avise, 1993;Walsh et al., 2019), interbreed in a 200-km hybrid zone along the coast of New England (Hodgman et al., 2002;Walsh et al., 2016; Figure S1).The saltmarsh sparrow has a comparatively intermediate level of association and history with tidal marshes, which it is thought to have occupied obligately since its divergence from its closest ancestor, the seaside sparrow, 1.5-3 million years ago (Chan et al., 2006;Klicka et al., 2014;Walsh et al., 2021).Prior population genetic studies in portions of the species' range have identified some fine-scale population structure, in the face of broader gene flow (Walsh et al., 2012(Walsh et al., , 2023)).Despite being strictly obligate to tidal marsh ecosystems, habitat characteristics vary across the species' range.In the northern portion of its range, marshes tend to be smaller, more isolated and riverine, while south of the hybrid zone, marshes are generally larger, more continuous and coastal (Greenlaw, 1993;Walsh et al., 2015).Given these habitat differences and the effects of interspecific introgression in the northern populations (Walsh et al., 2016), we expected the potential for hierarchical population structure between the north and south and fine-scale differentiation among some local populations.
We also predicted greater gene flow among populations in the saltmarsh sparrow than the Nelson's sparrow overall, and the potential for IBD, due to a more continuous availability of tidal marshes within the species' range.Due to its wider distribution and longer time since colonization relative to the Nelson's sparrow, we predicted higher genetic diversity and a larger N e in the saltmarsh sparrow than the Nelson's sparrow, but lower and smaller than seaside sparrow.Alternatively, reduced genetic diversity and N e could also be expected, relative to both other species, due to the stronger recent population declines in the saltmarsh sparrow, which has seen the most dramatic population declines, estimated at 9% annually, with complete population collapse predicted within the next 50 years (Correll et al., 2017;Field et al., 2017).
The northern seaside sparrow, 1 of 7 recognized, extant seaside sparrow subspecies, breeds exclusively in tidal marshes from Virginia to Massachusetts, in sympatry with saltmarsh sparrow where their ranges overlap (Davis et al., 2021; Figure S1).The seaside sparrow has the longest association with tidal marshes (~4.5 million years; Walsh et al., 2021) and, accordingly, exhibits the greatest ecological specialization and adaptation to the ecosystem, as evidenced by only occurring on large, expansive coastal marsh complexes.Their niche position includes a narrower gradient of tidal marsh conditions and requirements for nesting than the other two species (low to high saltmarsh for seaside sparrow and high saltmarsh to brackish marsh for saltmarsh sparrow within our study area; Conway, 2019).
Additionally, seaside sparrows exhibit the strongest tidal marsh adapted phenotypes (Greenberg, 2006;Greenberg & Droege, 1990), including a large beak and dark, melanistic plumage, further evidence for their degree of ecological specialization.The northern seaside sparrow's population estimate in our sampling range is considerably higher than either of our other focal species, consistent with greater availability of tidal marsh habitat within that distribution (Wiest et al., 2016(Wiest et al., , 2019)), and prior genetic studies have found panmixia within this subspecies (Davis et al., 2021;Roeder et al., 2021), likely a result of large, stable and highly connected populations occurring in expansive, coastal marsh complexes.Therefore, we did not expect to find population structure in the seaside sparrow, nor IBD, and we predicted high genetic diversity and a large N e .The highest degree of salt marsh specialization in this species could also be accompanied by reduced relative genetic diversity; however, stable large population sizes and a long evolutionary time since colonization, during which mutations can accumulate, both point towards the greater likelihood of high genetic diversity.1; Data S1).Some sampling locations within species were grouped for analyses with the nearest sampling location due to small sample sizes (n < 3), resulting in 18 marsh-level populations for the Nelson's sparrow, 31 marsh-level populations for the saltmarsh sparrow and 16 marsh-level populations for the seaside sparrow (Data S1).We used these grouped marsh-level populations for all subsequent analyses except for Structure, in which we used all individual sampling locations (Data S1).

| Molecular methods
Genomic DNA was isolated using a Qiagen DNeasy Extraction Kit Indexing groups were combined and sequenced on three Illumina HiSeq 2500 lanes (read length 100 bp) at the Cornell University Institute for Biotechnology.

| Data processing, SNP discovery and filtering
All reads were trimmed to 97 bp and then filtered for quality using the FASTX-Toolkit (http:// hanno nlab.cshl.edu/ fastx_ toolkit).We removed any sequence containing a single base with a Phred quality score less than 10 (90% accuracy) and required 95% of bases within the resulting read to have a Phred quality score greater than 20 (99% accuracy).We used the process_radtags commands in StackS v 1.19 (Catchen et al., 2011) to demultiplex the remaining 97-bp sequences.
We applied additional filtering steps, retaining reads only if the following conditions were met: they passed the Illumina chastity filter (defined as the ratio of the brightest base intensity divided by the sum of the brightest and second brightest base intensities; requiring no more than 1 base in a cluster to have values lower than 0.6 in the first 25 cycles), they contained an intact SbfI RAD site, they contained one of the unique barcodes, and they did not contain Illumina indexing adapters.To accommodate differences in the length of inline barcodes, we trimmed all sequences at their 3′ end to the length of the shortest sequence (90 bp).
To maximize detection of within-species variation for populationlevel analyses, we called variants for the Nelson's sparrow, the saltmarsh sparrow, and the seaside sparrow separately.Sequences were assembled with StackS version 2.5 with the reference map pipeline and alignment to the saltmarsh sparrow reference genome (Ammospiza caudacuta; Genbank; GCA_027887145.1), which was generated with the Vertebrate Genomes Project using a combination of PacBio, Bionano and Hi-C sequencing technologies, following the protocols of Rhie et al. (2021).We used an increased stringency level to call genotypes (gt-alpha parameter set at 0.01 rather than default of 0.05).Sequences were aligned with bowtie2 version 2.2.2 (Langmead et al., 2009), using the end-to-end option to avoid reads discarded due to soft clipping.
Data for each species were exported as VCF files using the populations module of STACKS with minimum percentage individuals required to process a locus per species (-R) set to 70%, a single-SNP-per-locus (--write_single_snp) and minor allele count of at least the data set (--relatedness2 > 0.20; Table S1; Danecek et al., 2011;Manichaikul et al., 2010).
To identify F1 hybrids and misidentified individuals in the saltmarsh-Nelson's hybrid zone so we could exclude them from subsequent analyses of genetic differentiation and population structure, a VCF was created containing all Nelson's sparrows and saltmarsh sparrows with 977 SNPs common to both datasets.The program Structure version 2.3.4 was used to analyse this data set with 10 independent runs with the number of clusters (K) set to 2 for 100,000 MCMC iterations after 200,000 burn-in iterations (Pritchard et al., 2000).Lambda was defined by doing an initial run at K = 1 with a 10,000 generation burn-in followed by 10,000 iterations, allowing lambda to vary.The average lambda value across the run (0.84) was then used in the subsequent runs.The program CLUMPP version 1.1.2was used to compile independent runs into a consensus value for each individual (Jakobsson & Rosenberg, 2007).
After filtering individuals, VCFTools was used to remove any SNP that violated Hardy-Weinberg Equilibrium (--hwe) with an extremely low p-value (<.00001), which are likely artefactual (e.g.paralogs or null alleles), and to remove any SNPs that fell below the initial missingness and MAC filters (--mac 3 and --max-missing 0.7; see Table S1).For the saltmarsh sparrow data, we saw elevated SNP numbers at positions 27-29, which was driving patterns in principal components space for a single lane of sequencing (see Figure S3).A sequencing error on this lane causing artefactual SNPs across individuals at these positions was determined to be the likely cause of this pattern.Therefore, all SNPs at these positions (n = 631) were excluded from the final saltmarsh sparrow data set using VCFTools.
Both the Nelson's sparrow and the seaside sparrow data sets were assessed to confirm that this was not an issue.
Finally, outlier loci were identified for each species using pcadapt in R (Privé et al., 2020).We assessed the Scree plot to identify the appropriate number of principal components to retain-8 for Nelson's sparrow, 8 for saltmarsh sparrow and 9 for seaside sparrow.Any SNP identified as an outlier with the q-value method (α = .(Table S1 for a summary of SNP-retention by filtering step).

| Genetic differentiation and population structure
Within Nelson's, saltmarsh, and seaside sparrow, we calculated global fixation index (F ST ) estimates at the species and marsh-level and generated other basic descriptive statistics, including inbreeding coefficient (F IS ), observed heterozygosity (H O ), within population gene diversity (H S ) and overall gene diversity (H T ) using hierfstat version 0.5-7 in R (Goudet, 2005).F ST and F IS were calculated using Weir and Cockerham (1984) formula, with confidence intervals calculated in hierfstat using the boot.vcfunction with 100 bootstrap iterations.
Per-population global fixation estimates ( ̂ i ST ) were calculated in hierfstat as an average over all loci within the population, described in Weir and Goudet (2017).To visualize genetic clusters, we used Discriminant Analysis of Principle Components (DAPC) using the adegenet package (Jombart, 2008;Jombart et al., 2010).The optimal number of principal components to retain was determined per species and for groups within species (as identified by Structure, see below) using the cross-validation method (Jombart et al., 2010).The number of principal components retained for DAPC was 21 for the Nelson's sparrow, 56 for the saltmarsh sparrow and 50 for the seaside sparrow.
To identify patterns of genetic structure within each species, individuals were assigned to genetic clusters using the Bayesian clustering approach of Structure 2.3.4 (Pritchard et al., 2000).For each species, we first conducted 10 runs for each value of K = 1-10; each run consisted of a 200,000 generation burn-in followed by 100,000 iterations, using the LocPrior model with sampling location information (Hubisz et al., 2009).Based on observed hierarchical structure within saltmarsh sparrows, we subsequently performed the same Structure approach described above for 3 regional groups of populations within the saltmarsh sparrow: north, mid-range and south (Figure 2).For all Structure analyses, we used the admixture model and assumed correlated allele frequencies.Lambda was defined for each species by performing an initial run at K = 1 with a 10,000 generation burn-in followed by 10,000 iterations, allowing lambda to vary across the run.The average lambda value across the run was then used in subsequent analyses (Nelson's 0.42, saltmarsh 0.30 and seaside 0.43).
For each species, we determined the most likely value of K using a combination of the plateau in the estimated log probability of the data (Lnp(D)), the ΔK method of (Evanno et al., 2005) and examination of the bar plots; Structure output was visualized using the program Structure harveSter (Earl & vonHoldt, 2012).We averaged results across runs using the greedy algorithm implemented in the program clumpp (Jakobsson & Rosenberg, 2007).Admixture plots from Structure analyses for each sampling location were visualized on a map for each species using QGIS (QGIS Development Team, 2021).

| Testing for isolation by distance
We tested for patterns of isolation by distance within each species by performing a Mantel test (Mantel, 1967) between genetic distance and geographic distance (km) with 999 Monte Carlo simulations using the adegenet package (Jombart, 2008).We used marsh-level population groupings for this analysis rather than sampling location, as described above.For any species in which we found a significant isolation by distance pattern (p < .05),we further assessed spatial

| Effective population size
We calculated N e for each species, using NeEstimatorV2 (Do et al., 2014).NeEstimator is a tool for estimating contemporary effective population sizes using multi-locus genotypes from population samples.We calculated estimates with the single sample mode using bias corrected linkage disequilibrium method and PCrit value (lowest allele frequency to be included) of 0.05, including parametric 95% confidence intervals.We also calculated N e for population clusters as identified by Structure for Nelson's (5 population clusters) and saltmarsh sparrows (north, mid-range and south regional populations).Because we found no population structure in this subspecies of the seaside sparrow, we only calculated N e as a single group across all sampling locations.Input files were in a Genepop format for both species and population-level analyses.

| Conservation units
To further assess the population structure of the Nelson's sparrow and the saltmarsh sparrow in a conservation context, we tested populations for evolutionary significance using the pairing of a phylogenetic approach (NeighborNet networks) and a Shapley value (Fernandez-Fournier et al., 2021;Volkmann et al., 2014).This method uses a pairwise matrix of population genetic distances to construct a phylogenetic network (distance-based method with a neighbour-joining algorithm) from which a Shapley value is estimated.The Shapley value provides a ranking of evolutionary distinctiveness to each population, such that populations with higher ranks contribute more genetic diversity to the whole species/network (Fernandez-Fournier et al., 2021).For Nelson's and saltmarsh sparrows, populations were projected as a NeighborNet network (Bryant & Moulton, 2003) in R with the phanghorn package (Schliep, 2011) to visualize their relationships based on an Jost's D matrix generated using the mmod R package (Jost et al., 2018).Then, the Shapley value ranking was calculated for each sampling site in R using methods of -Fournier et al. (2021).This was not performed for seaside sparrow due to the lack of observed genetic structure.

| Habitat predictors of population structure via distance-based redundancy analysis (dbRDA)
To assess the influence of landscape and specific marsh habitat features on population structuring across the three species, we compiled 5 data points for each sampling location from a previously developed marsh patch-level GIS data layer (Wiest et al., 2019) to capture meaningful habitat differences (Data S2).Variables we predicted may be important in defining marsh habitat included marsh patch size (ha), percent of the marsh patch composed of high marsh vegetation (dominated by Spartina patens, Spartina alterniflora short form, Juncus gerardii, and/or Distichlis spicata; needed for Nelson's and saltmarsh sparrow nesting, while seaside relies more on low marsh habitat), the amount of surrounding marsh within a 150and 1000-m buffer of the patch as metrics of connectivity, and sea level rise trend (SLT; mm/year) as a way to capture tidal differences and climate change pressures suggesting habitat quality.We also calculated a measure of distance from each sampling location to the Atlantic coastline (km) using the measure tool in QGIS (QGIS Development Team, 2021) to capture habitat differences seen between coastal and inland marshes.Before analysis, we ensured sampling resolution matched between the marsh patch-level data and genetic sampling locations, grouping any unique genetic sampling locations which fell within a single marsh patch in the data layer.
We also excluded any sites that had <3 individuals from the analysis (removed 1 site from Nelson's, 2 sites from saltmarsh, and 6 sites from seaside).Since the patch-level data layer only included marshes within the United States, we excluded the two Canadian sampling locations from the Nelson's sparrow data set.
We used dbRDA to test the relationship between population genetic structure and marsh-specific habitat predictors for each of the three sparrow species.This method is a constrained ordination approach that allows for the use of non-Euclidian distance metrics in RDA (Legendre & Anderson, 1999).Here, we evaluated whether genetic differentiation among sites in the form of a pairwise F ST matrix, is driven by any of the 6 marsh-level habitat features described above.Multicollinearity was assessed before the analysis by testing for pairwise correlations between each of the habitat variables within each species separately.We calculated the variance inflation factor (VIF) for any Pearson correlation coefficient >0.7 between two variables and removed the one with the highest VIF.For Nelson's sparrow, we excluded proportion of high marsh and amount of marsh within the 150 m buffer which were correlated with SLT.We excluded marsh area (ha) for seaside sparrow and saltmarsh sparrow which were correlated with distance to coast in both species.
Each resulting variable was scaled by its range using decostand function and subsequent dbRDA performed using the capscale function in the R package vegan (Oksanen et al., 2022).Since we found significant IBD for saltmarsh sparrow, we performed a conditional dbRDA for saltmarsh sparrow to control for geographic distance.
Because Nelson's and seaside sparrow did not have significant IBD, conditional dbRDA was not run for these species.Finally, we used

| Genetic differentiation & population structure
Global F ST was highest among populations of Nelson's sparrow (0.020), followed by saltmarsh sparrow populations (0.016), with the lowest F ST found among seaside sparrows (0.003; Popham in midcoast Maine (−0.409;Table S3; Figure S5).F ST estimates were low across all the seaside sparrow sampling locations, ranging from lowest at Wallops Island, Virginia (−0.056) to highest on Long Island, New York (0.031; Table S4).
Results from the Structure analysis for Nelson's sparrow suggest K = 2 using the ∆K method; however, ∆K metric is not reliable on its own at K = 2, as it has been shown to identify K = 2 as the top level of hierarchical structure even when more subpopulations are present (Janes et al., 2017).The estimated log probability of We observed hierarchical population structure for saltmarsh sparrow (Figure 2).An initial Structure run indicated a K = 2 solution with a north-south split at Parker River, Massachusetts We found evidence for fine-scale population structure within each saltmarsh sparrow population region (Figure 2).Within the northern region, ∆K supported K = 3, but the plateau in the estimated Lnp(D) in combination with a secondary peak in ∆K at K = 6 and distinct groups within the bar plots suggest further hierarchical structure within this region (Figures 1 and 2; Figure S8).We Bay) near the upper range limit of the species (Figures 1 and 2).
Beyond K = 4, there were also subsequent peaks of ∆K and Lnp(D) at higher K values, suggesting the existence of even further hierarchical sub-structuring (Figure S13).Indeed, the bar plots continue to separate unique sites within the northern group at those higher K values, including in Great Bay in New Hampshire (Chapman's Landing and Lubberland Creek; Figure S13).
For the mid-range and southern regions, the evidence suggested K = 2 as the best model for each (Figures 1 and 2).The ∆K and Lnp(D) methods suggested K = 2 for the mid-range model; however, the plateau in Lnp(D) extended until K = 5 after a dip at K = 4 (there was also a secondary peak in ∆K at K = 5) suggesting that there could be further hierarchical structure within this region as well (Figure S10).Indeed, Structure is known to perform poorly in the presence of hierarchical structure and IBD, and the ∆K method will often point to a K = 2 solution even when substructure exists (Janes et al., 2017).Furthermore, the solution for K = 2 in this middle range pulled out the same individuals which were found distinct in the previous southern hierarchical run that had more of a "northern" signal (Figure S9; Figure 2).Bar plots of the higher K values in the middle range suggest that other sites within this region may be genetically distinct to some extent, including Monomoy, Nantucket and Great Marsh (Massachusetts), as well as Sachuset in Rhode Island (Figure S10).
For the southern region, Lnp(D) and ∆K suggested K = 2, with a single marsh near New York City (Four Sparrow Marsh) clustering out as distinct from the rest.Lnp(D) values remained stable until K = 5, and ∆K values also appeared to trend upwards at higher K values (including K = 5), suggesting that there may also be further substructure at this spatial scale (Figure S13).Marshes that cluster out at these higher levels of K include other New York City marshes (Sawmill, Idlewild) and a western Long Island marsh (Shirley) in the southern region (Figure S11).
Structure results for the northern subspecies of seaside sparrow indicated that populations are panmictic across the range sampled (Figure 1).Although there is a peak in ∆K at K = 2, this metric alone cannot correctly distinguish between K = 1 and K = 2.Both ∆K and the Lnp(D) values decreased from K = 2 with every additional K, suggesting K = 1 the best solution.The bar plots also suggested there is no population structure, supporting K = 1 as the appropriate model for these data (Figure S14; Figure 1).
Patterns in DAPC corresponded generally to the patterns observed from Structure.There was local population structure in the Nelson's sparrow, with specific marshes and regions splitting out as distinct.Regional structure was seen in the saltmarsh sparrow with three main groups corresponding to those found in Structure analysis (red, yellow, blue in Figure 1).DAPC within each hierarchical group for the saltmarsh sparrow show further substructure within each region, as seen in our STRUCTURE results (Figure S15).In the midrange, Monomoy and Nantucket cluster out between the rest of the Massachusetts sites and Connecticut/Rhode Island sites, and New York City Marshes cluster out in the southern region (Figure S15).
The seaside sparrow showed no clear structure from the DAPC, with one large cluster.Two small populations separated out slightly in the seaside sparrow DAPC (including 2 sites within the southern portion of the range, the Lower Chesapeake and Wallops Island); however, given lack of evidence for differentiation in our Structure run (∆K suggests K = 1) and relatively uniform Eigen values compared to the other two species, this is likely a sampling artefact or signature of DAPC overfitting, with the overall pattern consistent with panmixia (Figure 1).

| Genetic diversity
H O , H S and H T all showed a consistent pattern, with the highest values in the seaside sparrow (0.167, 0.162, 0.162 respectively), followed closely by the Nelson's sparrow (0.160, 0.156, 0.160 respectively), and much lower for the saltmarsh sparrow (0.086, 0.082, 0.083 respectively; Table 2).Global F IS was negative across all species (Table 2).
F IS was negative among most sampled Nelson's sparrow marshes; however, some smaller marshes had low but positive values (Table S2).F IS values were also negative across most saltmarsh sparrow sites.H O levels increased from north to south across the Nelson's sparrow sampling range, with higher levels of heterozygosity seen in the southern marshes within the hybrid zone than those in the north, which are outside the hybrid zone (Figure 3).Lubec, Maine, was an exception, with higher heterozygosity compared to surrounding marshes, more akin to those in the southern portion of the Nelson's sparrow range (Figure 3).
A pattern of increased H O in the hybrid zone was also present in the saltmarsh sparrow, with higher levels in the north of the range where there is interspecific gene flow and lower levels in the southern part of the range located outside of the hybrid zone (Figure 3).H O and F IS levels were similar among all sampled marshes for seaside sparrow, with no apparent geographical pattern across the range (Figure 3; Table S4).

| Isolation by distance
The Mantel test found support for IBD, with a statistically significant correlation between geographic and genetic distance in the saltmarsh sparrow (r = .198,p = .03;Figure 4), but we did not find support for IBD in the Nelson's (r = .145,p = .174)nor the seaside sparrow (r = −.197,p = .860).A correlogram of Mantel's r values across distance classes in the saltmarsh sparrow showed significant spatial autocorrelation at the 10, 20, 50 and 375 km distance classes; the X intercept was ~100 km, indicating that genetic similarity among adjacent marsh locations extended to this distance (Figure 4).

| Effective populations size
The seaside sparrow had the largest estimated N e (10,873), followed by the saltmarsh sparrow (N e = 973), then the Nelson's sparrow (N e = 638; Table 2).These species-level N e estimates, however, are likely biased for the latter two species due to population structure.
Within the saltmarsh sparrow, the mid-range population had the largest estimated N e (1388), followed by the southern population (752), and northern population (346; Table 3).Within the Nelson's sparrow population groups identified by Structure, we found that the Nova Scotia and Downeast Maine population was the largest (N e = 805), followed by Mid-Coast Maine (N e = 228) and the southern marshes from Spurwink, Maine to Parker River, Massachusetts (N e = 237), which had similar population sizes.The Penobscot Bay population in Maine (N e = 184) was the smallest estimated, besides the population in Lubec, Maine, which was too small to correctly estimate (Table 3).

| NeighborNet and Shapley conservation units
The NeighborNet tree for the Nelson's sparrow had many distinct branches, with some regional grouping among sites on the tree (Figure 5).The three sites with the highest Shapley values in the Nelson's sparrow were Parker River (Massachusetts), Lubec (Maine) and Weskeag (Maine; Figure S16).The tree for saltmarsh sparrow (Figure 5) exhibited some sharp branching in the northern end of the hybrid zone (sites 28-30 in Figure 5).Southern sites constituted the body of tree with some sites jutting out with longer branches, while the mid-range sites were between the north and south.The four sites with the highest Shapley values (Figure 5) in the saltmarsh sparrow were Weskeag (Maine), Maquoit, (Maine) and two New York City marshes-Idlewild and Four Sparrow (Figure S16).

| Habitat predictors of population structure (dbRDA)
Nelson's Sparrow dbRDA identified a best-fitting model with two habitat variables explaining genetic divergence among sites, including marsh area and sea-level rise trend.This model accounted for about 18% of genetic variation (constrained proportion = 0.185); however, within this model, only SLT was significant (F = 2.052; p = .004)while marsh area was not (F = 1.363; p = .127).For Nelson's Sparrow, SLT was significantly negatively correlated with proportion of high marsh vegetation (r = −.69) and marsh within a 150 m buffer (r = −.68), and therefore excluded from the analysis; however, this F I G U R E 3 Observed heterozygosity (H O ) values per marsh sampling location for each of the three Ammospiza sparrow species: Nelson's (top panel), saltmarsh (middle panel) and seaside (bottom panel).Sites are in geographic order south to north for each species, with the green bar representing sites that reside in the saltmarsh-Nelson's sparrow hybrid zone.
finding suggests the influence of these features on population structure cannot be excluded.
The conditional dbRDA controlling for geographic distance (due to identified IBD) within saltmarsh sparrow yielded a top model that included two habitat variables influencing population structure: distance to Atlantic coast and the amount of surrounding marsh within a 150 m buffer.This model explained 12% of genetic variation, with geographic distance further accounting for 14% (constrained proportion = 0.122; conditioned proportion = 0.135).Both distance to coast (F = 2.652, p = .016)and surrounding marsh within 150 m buffer (F = 1.798; p = .043)were significant predicters within the model.Distance to coast was also significantly correlated with marsh area (r = .73)and excluded from analysis; however, variation in marsh size may also be encapsulated within this variable.In the seaside sparrow dbRDA, the null model was the best model, suggesting none of the tested habitat features influence genetic differentiation in this species.

F I G U R E 4
Tests of isolation by distance (IBD) for each focal species: Nelson's sparrow (left), saltmarsh sparrow (centre) and seaside sparrow (right).The top panels show the relationship between geographic and genetic distance (km) for each species, respectively, with only saltmarsh sparrow resulting in a significant relationship (r = .198,p = .030).The middle panel illustrates the result of the mantel test, with the empirical value of the genetic: geographic distance correlation as the black diamond/line, plotted over a histogram of 999 Monte Carlo simulations run under the absence of spatial structure.The bottom panel (in saltmarsh sparrow only) depicts a mantel correlogram used to access spatial autocorrelation within the data, comparing mantel r across distance (km), with dark circles denoting significant distance classes.Species illustrations by Mackenzie Roeder.

Species
Population a Sample size too small to estimate.

TA B L E 3 Effective population size of
Structure groups for Nelson's sparrow (site-level) and saltmarsh sparrow (regionlevel).Effective population size was calculated using NeEstimator.
We found a lack of concordance between our results and predictions of the SGVH for population structure, genetic diversity and N e in relation to specialization status in Ammospiza sparrows.Our findings contribute to a growing body of literature reporting results inconsistent with this hypothesis (Griffith & Sultan, 2012;Matthee et al., 2018;Titus & Daly, 2017).For example, a literature review of wild bird populations found no significant effect of habitat specialization on any quantitative genetic estimators (Martinossi-Allibert et al., 2017).Instead of ecological specialization status, trends in our study and others highlight the role of habitat in structuring populations and support expectations derived from population genetic theory, including higher genetic diversity in larger populations and a role of demographic history and neutral genetic processes across contemporary species' distributions.

| Habitat and species distribution influence population structure
We found mash-specific habitat predictors and differences in regional landscape characteristics are important drivers of neutral genetic variation across the three investigated Ammospiza sparrow taxa.Additionally, these factors better predict patterns of population genetic structure across the three co-distributed species than degree of ecological specialization or categories assumed by the SGVH.The mechanism underlying the SGVH suggests that specialist species are likely adapted to a patchier landscape than generalists because they use a narrower set of resources that are more patchily distributed and less connected.Consequently, more specialized populations are subject to genetic drift as a result of limited gene flow among fragmented habitat patches (Li et al., 2014).The available tidal marsh habitat and the distribution of sparrows across the landscape do not match the habitat heterogeneity continuum assumed by the SGVH; therefore, we found our results contrary to its predictions.The geographic scale of our study, however, may not have been adequate for a thorough test of the SGVH, as the habitat and ecological differences across the tidal marsh gradient within the portion of the Nelson's and seaside sparrow distributions sampled may be too subtle for detecting a pattern.
Although the specialist seaside sparrow has the narrowest niche in tidal salt marsh, it also occupies the most continuous and homogenous habitat across its range.The saltmarsh and the more generalist Nelson's sparrow occupy increasingly patchier habitat across the landscape, and we found genetic differentiation and population structure concordant with those habitat expectations, rather than the specialist-generalist categories.Nelson's sparrow had the highest genetic differentiation and local population structuring, matching the patchy mosaic of habitat throughout its range.Conversely, the northern seaside sparrow was panmictic across its range with low F ST values, reflecting its relatively widespread distribution in expansive marsh complexes allowing for greater connectivity and gene flow across populations.The saltmarsh sparrow had hierarchical population structuring and IBD, consistent with regional differences in habitat distribution (Walsh et al., 2023).Moreover, the long, linear range of tidal marsh habitat is more continuous across the southern portion of the saltmarsh sparrow range than that of the Nelson's sparrow, allowing for greater gene flow among most neighbouring marshes.In addition, hybridization and widespread introgression with Acadian Nelson's sparrow in the North may also contribute to population structuring across the saltmarsh sparrow range.Indeed, both Nelson's and saltmarsh sparrow had higher marsh-level heterozygosity within the hybrid zone than in their allopatric ranges (Figure 5).Across species, therefore, our results point to the role of species and microhabitat distribution contributing to observed patterns of population structure.
Marsh-specific habitat predictors of genetic structure differed for each species, further exemplifying the importance of habitat distribution across the landscape in shaping genetic variation at a microevolutionary level.Meeting our expectations, for both Nelson's and saltmarsh sparrow, different habitat predictors were associated with genetic divergence, whereas no habitat predictors were associated with genetic divergence in the largely panmictic seaside sparrow, consistent with its occurrence on large, expansive and homogenous marsh complexes.Increased genetic differentiation of Nelson's sparrow populations across marshes with higher sea-level rise trends suggests that genetic variation may partition along a tidal gradient from coastal marshes to more sheltered, inland marshes with less tidal influx.Since sea-level rise trend (SLT) decreases with the proportion of high marsh vegetation within a marsh patch (r = −.69) and the amount of surrounding marsh within a small buffer (150 m; r = −.68); this inland-coastal gradient also encapsulates specific vegetative characteristics of the marsh (habitat quality) and the level of patch connectivity.This fits predictions that small, patchier and more isolated marshes differ in habitat within the Nelson's sparrow range and these characteristics are influential in population differentiation.For saltmarsh sparrow, distance of marsh to coastline and marsh connectivity (marsh buffer within 150 m) were important predictors of genetic structure, although these factors together explained less variation in the genetic data than did geographic distance (12% and 14% respectively).The distance to coastline measure may reflect a gradient of inland-coastal habitat differences related to population structuring; however, the F I G U R E 5 Map of sampling locations for each study species: Nelson's (top), saltmarsh (centre) and seaside (bottom) sparrows.Nelson's and saltmarsh sparrow maps include an inset of an unrooted neighbour-net tree derived from a Jost's D matrix between marshes with orange dots indicating sites with the highest Shapley values.Due to panmictic nature of seaside sparrows, no neighbour-net tree is displayed for this species.Numbers in the trees correspond to locations on the sampling map.Black hatching on maps indicates the saltmarsh-Nelson's sparrow hybrid zone, while grey shading illustrates the spatial extent of introgression.
connectivity measure and overall size of a patch (marsh area significantly was correlated with distance to coast; r = .76)may speak to the influence of genetic drift on patterns of genetic differentiation (i.e.drift greater in small marshes, leading to their divergence; Walsh et al., 2023).Local adaptation may influence observed genetic structure, especially in the presence of reduced geneflow; however, our data are limited to neutral markers and cannot speak directly to differential adaptation.Rather, our findings support a pattern of isolation by environment, whereby genetic differentiation is associated with environmental differences (Wang & Bradburd, 2014).
Habitat preferences have been linked to spatial genetic structuring in many systems, including funnel web spiders, montane sedges, two sister species of snowcock and two widespread magpies in China (An et al., 2020;Beavis et al., 2011;Massatti & Knowles, 2014;Zhang et al., 2012), suggesting that taxon-specific traits should not be overlooked as drivers of structure across the landscape.

| Degree of ecological specialization does not predict genetic diversity
Our findings were similar for genetic diversity as for population structure, with limited support for the influence of degree of ecological specialization.Environmental heterogeneity has been suggested as a mechanism for facilitating the quantity of genetic variation in a population, the niche width of individuals, and the maintenance of diversity at the landscape level, leading to differences in genetic diversity along a habitat specialization gradient (Kassen, 2002).Specialists are expected to evolve through stabilizing or directional selection, leading to depletion of overall genetic diversity, while generalists evolve under fluctuating selection to heterogenous environments, allowing for the maintenance of genetic variability (Kassen, 2002;Keightley & Hill, 1988).These predictions, however, were not supported by our findings, suggesting other factors, such as evolutionary or demographic history, rather than degree of ecological specialization are the drivers of genetic diversity patterns in Ammospiza sparrows.
We found that the northern seaside sparrow, considered the most adapted of the three taxa to the salt marsh environment or most specialized, had the highest heterozygosity and gene diversity of the three investigated species, contrary to the prediction of the SGVH that generalists harbour the most genetic diversity.The least specialized Nelson's sparrow had similar levels of genetic diversity to the northern seaside sparrow and higher genetic diversity levels than the comparatively more specialized saltmarsh sparrow; other factors, besides ecological specialization, however, could be driving this pattern.The northern seaside sparrow was the earliest of these species to colonize salt marsh (Walsh et al., 2021), providing time Nelson's sparrow relative to the saltmarsh sparrow was contrary to our predictions, given its shorter evolutionary history with the ecosystem, as well as its distribution in smaller, more isolated marshes.
One explanation for this incongruity could be historical population bottlenecks in the saltmarsh sparrow's recent past, suggested by previous demographic modelling (Walsh et al., 2019(Walsh et al., , 2021)), which could have reduced genetic diversity and resulted in a loss of rare alleles in the species.Additionally, the saltmarsh sparrow has seen the most dramatic population declines of the three species in recent decades (Correll et al., 2017); although, based on the recency of these declines, it is unlikely we would see direct effects to genetic diversity and heterozygosity this quickly (Anderson et al., 2010).
Contemporary distributions of the Ammospiza sparrows across the available saltmarsh habitat in conjunction with neutral population genetic theory help explain the observed patterns in genetic diversity and N e .The seaside sparrow had the largest N e of the three sparrows, followed by the saltmarsh sparrow, and the Nelson's sparrow.This pattern is consistent with expectations of population genetic theory based on how the bird species are spatially distributed across the landscape: small, isolated populations in the Nelson's sparrow range, somewhat larger and more connected populations across the variable habitat of the saltmarsh sparrow, and large populations across more continuous habitat of the seaside sparrow.
Neutral theory also suggests that larger populations have higher genetic diversity (Hague & Routman, 2016), a pattern we observed for seaside sparrow, but not for the Nelson's sparrow, which had the smallest N e (consistent with marsh isolation) despite higher levels of genetic diversity and heterozygosity than saltmarsh sparrows, likely due to additional factors described above.

| Conservation through an eco-phylogeographic lens
When a group of species face the same environmental gradient, they may show both shared and unique elements in their patterns of divergence (Langerhans & DeWitt, 2004).Comparing genetic structure and diversity of three Ammospiza sparrows differently adapted to a common tidal marsh gradient provides insight into shared elements, as well as species-specific patterns of divergence.
This is especially important in the context of conservation amidst climate-induced environmental changes, as these changes may impact ecological niches and species responses differently despite shared environmental history (Beavis et al., 2011).One important goal of species conservation is to preserve genetic variation, often through prioritization based on phylogenetic trees, with more isolated lineages ranking higher as a representation of raw material for future evolution (Bowen & Roman, 2005;Haake et al., 2008).
Setting conservation priorities at the population-level within species is also becoming increasingly important in the face of range fragmentation and decline in population connectivity, and these practices can follow the same principles of evolutionary distinctiveness at the species level (Fernandez-Fournier et al., 2021;Volkmann et al., 2014).To this end, the Shapely values obtained from our phylogenetic analyses identified marshes with relatively high evolutionary distinctiveness, suggesting they have conservation value due their elevated contributions to genetic diversity.
Conservation goals are often framed within the context of "the three Rs" (Shaffer & Stein, 2000)-representation, resiliency and redundancy-as essential population attributes to maintain viability and reduce extinction risk (Malcom & Carter, 2021;Smith et al., 2018).Effective conservation measures must maintain multiple populations across the species' range (redundancy) in ecologically representative settings, all of which are healthy, self-sustaining, genetically robust and resilient to environmental changes (Redford et al., 2011).Strategies to achieve conservation Conserve (New York) and some sites in Maryland within the southern part of the range.The distinctiveness of the New York marshes is likely due to the effects of genetic drift on these small and isolated sites (Walsh et al., 2023).Accordingly, these populations may be less resilient or prone to extirpation in the face of environmental change.
Other genetically distinct marshes, such as Weskeag, Maquoit and Spurwink (Maine), Nantucket (Massachusetts) marshes, or those in coastal Maryland may have unique habitat, perhaps supporting local adaptation.Protecting representative populations across the full ecological gradient of a species is imperative to conserve local adaptation and evolutionary potential (Crandall et al., 2000;Redford et al., 2011).The lack of genetically distinct sites in the panmictic seaside sparrow subspecies suggests that representation will be easier to achieve for this species and conservation should focus more on redundancy and resiliency.
Redundancy and resiliency can be achieved through protecting multiple, large marshes within each genetically distinct area and across species ranges to maintain populations that are robust and healthy across all three species.Managing for a mix of habitats (coastal and estuarine), especially for the saltmarsh and Nelson's sparrow, may be important to capture the heterogeneity in the landscape that influences population structure and diversity.
Connectivity is important for all three species, as it promotes genetically diverse and robust populations.As such, it is also important to conserve smaller marshes in locations within species ranges which may not have large marshes to aid in connectivity and capture environmental heterogeneity.It has been suggested that species that are ecologically intermediate between two extremes may be most seriously affected by environmental and land use change in the context of conservation genetics, as these species must occur in large population networks to sustain high levels of genetic diversity via geneflow and may be at high risk of inbreeding depression when this connectivity is suddenly lost (Habel & Schmitt, 2012).The saltmarsh sparrow, being intermediately adapted to tidal marshes compared to the seaside and Nelson's sparrows (although still a tidal marsh specialist) and occurring in a large population network where connectivity is integral to maintaining diversity, is the most seriously affected by rapid environmental changes, including habitat collapse and loss of connectivity (Correll et al., 2017).In the bigger picture, all three of these Ammospiza sparrows are considered specialists of marsh habitats (with seaside and saltmarsh more narrowly specialized on tidally influenced marsh systems; Correll et al., 2016); accordingly, they are all vulnerable to environmental degradation of marshes.
With a greater mechanistic understanding of spatial population structure and genetic diversity patterns, we can tailor conservation efforts to incorporate those drivers both among and within species.Generalist species are hypothesized to have an advantage over specialist species in the face of environmental change (Dennis et al., 2011), and specialists often receive more conservation focus given their unique natural history or strict association with rare environments which may be more easily protected (Dapporto & Dennis, 2013).Our results suggest species do not always fit predic-

Supported
FigureS1).Of our three focal taxa, the Acadian Nelson's sparrow has the shortest habitat association with tidal marsh, with roughly 5000 years since it split from its closest inland subspecies (which occur in freshwater marshes in the continental interior of northern

(
Qiagen, Valencia, California, USA) according to the manufacturer's protocol.Double-digested restriction-site associated DNA markers (ddRAD tags) were generated following the protocol ofPeterson et al., 2012, andas in Maxwell et al., 2021.We used 500 ng of DNA at a standardized concentration of 25 ng/μL for each individual.DNA concentrations were quantified using a Qubit fluorometer (Life Technologies, NY, USA).DNA was digested with SbfI and MspI (New England Biolabs, MA, USA) and the ends of the digested genomic DNA were ligated using T4 DNA ligase (New England BioLabs) to P1 and P2 adapters(Peterson et al., 2012).We pooled samples with unique P1 barcodes into different indexing groups post digestion/ ligation.Products were cleaned using 1.5× Agencourt AMPure XP beads (Beckman Coulter, CA, USA) and fragments between 400 and 700 bp were size-selected using Blue Pippin (Sage Science, MA, USA).We performed low cycle number PCR with Phusion High-Fidelity DNA Polymerase (New England Biolabs) to incorporate the full Illumina TruSeq primer sequences into the library.After a final AMPure clean-up to eliminate small fragments, libraries were visualized on a 1% agarose gel and on a fragment Bioanalyzer (Agilent Technologies, CA, USA) to determine fragment size distribution.
3 (--mac 3).At this stage,VCFTools (v.0.1.16)was used to remove individuals from the data set that were missing data for more than 25% of loci, had an average depth across all SNPs of <10× or had a high degree of relatedness with another individual retained in F I G U R E 1 Population structure for Nelson's sparrow (top), saltmarsh sparrow (centre) and seaside sparrow (bottom).The black hatching on maps indicates the saltmarsh-Nelson's sparrow hybrid zone, while grey shading illustrates the spatial extent of introgression.Species illustrations by Mackenzie Roeder.Left: Map of sampling sites for each species, showing site-level admixture pie charts from Structure analysis, for K = 5 for Nelson's sparrow, K = 1 for seaside sparrow, and independent hierarchical runs for northern (K = 4 and K = 1), mid-range (K = 2) and southern (K = 2) regions in saltmarsh sparrow (see Figure2).Size of admixture plots are proportional to the number of samples.Right: Discriminant analysis of principal components (DAPC) plots (LD1 and LD2) for each species, explaining 20.9% cumulative variance in Nelson's sparrow, 37.3% in saltmarsh sparrow and 48.9% in seaside sparrow.Inset shows Scree plot of eigenvalues for linear discriminant axes.Cluster colours correspond to the identified genetic groupings from Structure.
05) was then removed from each data set using VCFTools (Nelson's n = 229, saltmarsh n = 55, seaside n = 24), ensuring the data sets only contained neutral SNPs.Sizes of the final SNP data sets were 5852, 5006 and 2844 for Nelson's sparrow (171 individuals), saltmarsh sparrow (283 individuals) and the seaside sparrow (141 individuals) respectively autocorrelation using a multivariate autocorrelation function to create a correlogram of Mantel r values across distance classes (km) with the ecodist package version 2.0.7 (Goslee & Urban, 2007).
Results of Structure analysis for (a) Nelson's sparrow with K = 5, and (b) hierarchical Structure analysis for saltmarsh sparrow with K = 2 for range-wide (top) and with K = 3 for southern (bottom) sites.Sites were divided into three regions for subsequent Structure analysis (denoted by brackets, with K = 4 for northern with unique signal at Popham & Maquoit sites, and K = 2 for both mid-range and southern regions).
the function ordistep in the R program vegan with a backwards stepwise model selection to determine the best set of variables for each species (50 steps with 199 permutations) and assessed significance using multivariate permutation tests (F-statistic with 999 permutations; function anova.cca in vegan).
the data (Lnp(D)) plateaued at K = 5, and there was a secondary peak in ∆K at K = 5; these results in conjunction with examination of the bar plots suggest the best supported solution is K = 5 (FigureS6).Results show some areas of genetic connectivity, with patchy/localized population structure evidenced by differentiation of specific marshes such as Lubec (Maine) near the Canadian border, and other groups of marshes such as some Downeast Maine (Mendall/Penobscot/Weskeag), Midcoast Maine (Sheepscot/ Popham/Maquoit) and others in the southern portion of the range (Spurwink, Maine to Parker River, Massachusetts) distinct from the areas to the north (Figures1 and 2).

(Figure 1 ;
Figure2).In total, we were left with three hierarchical population regions within the saltmarsh sparrow (northern, mid-range, southern) for which we ran three independent Structure analyses (Figures S8, S10 and S11).
further subsetted this group into a Popham and Maquoit (Maine) hierarchical group (as these two sites were distinct at the K = 3 solution) and kept all other sites together as another hierarchal group.Subsequent hierarchical STRUCUTRE runs on these two groups showed that Popham/Maquoit (Maine) sites cluster as one population, with the bar plots showing relatively equal ancestry across both colours at both sites (Figures 1 and 2; Figure S12), while the rest of the northern region had a peak of ∆K and a plateau of Lnp(D) at K = 4 (Figures 1 and 2; Figure S13).These results show broad connectivity among northern sites, with a number of genetically distinct marshes, including Popham/Maquoit in mid-coast (Maine), the upriver marshes at Parker River (Massachusetts), and a few other mid-coast Maine marshes (Weskeag and Thomas for mutations to accumulate and possibly for the species to reach mutation-selection balance, which could partially explain the comparatively high levels of genetic diversity.The more recent colonization of the tidal marsh environment by the Acadian Nelson's sparrow(Walsh et al., 2021) would suggest recent or ongoing purifying selection, possibly explaining its lower observed genetic diversity than seaside sparrow.The greater genetic diversity of the Acadian success through representation, resiliency and redundancy looks different for each species of Ammospiza sparrow and is informed by their unique population genetics as viewed through an ecophylogeographic lens.Addressing the attribute of representation, Shapely values highlighted important marshes for conservation in Nelson's and saltmarsh sparrows, in a manner consistent with patterns of genetic structure and diversity.For Nelson's sparrow, Lubec, Weskeag (Maine) and Parker River (Massachusetts) populations exhibit notable evolutionary distinctiveness, genetic divergence and span the range of the species.In addition, Lubec (Maine) shows a unique genetic signal not seen anywhere else across the range and has high levels of heterozygosity.Sites with particularly high Shapely values for the saltmarsh sparrow occurred across the range, with Weskeag, Maquoit and Spurwink (Maine) in the north, Nantucket, Idlewild and Four Sparrow marsh (New York) in the mid-range, as well as Marine Nature tions based on broad generalization of specialist-generalist categories and that mechanistic understanding of species-specific patterns, based in the environmental context, are also critical for informing conservation strategies.As such, a move towards linking species assemblages in the context of divergence with species traits, occurrences', population trends and environment may aid in understanding species persistence amidst a backdrop of environmental change to develop directed conservation measures (WallisDeVries, 2014).AUTH O R CO NTR I B UTI O N SAIK conceived the project, with help from JW.Samples were collected by JW and MC.Sequence data were generated by JW and LM.Data analysis was performed by LM and JC, with guidance from JW.The manuscript ideas were developed by LM, with help from JW, JC, AIK and BJO; the manuscript was written by LM, with input and editing by all authors, primarily AIK.

Table S2 ;
Figure S4). Certain marshes showed elevated F ST estimates comparatively throughout the range of Nelson's sparrows including Parker River (Massachusetts), the Canadian populations (Wolfville and Yarmouth) and Narraguagus in Downeast Maine (TableS2; FigureS4).Saltmarsh sparrow F ST values ranged from highest in the Shinnecock Bay of Long Island, New York (0.229) to lowest at TA B L E 2 Species-level genetic metrics.Effective population size was calculated using NeEstimator.Other metrics were calculated using hierfstat.