Surrounding landscape, habitat and hybridization dynamics drive population structure and genetic diversity in the Saltmarsh Sparrow

Determining factors that shape a species’ population genetic structure is beneficial for identifying effective conservation practices. We assessed population structure and genetic diversity for Saltmarsh Sparrow ( Ammospiza caudacuta ), an imperiled tidal marsh specialist, using 13 microsatellite markers and 964 individuals sampled from 24 marshes across the breeding range. We show that Saltmarsh Sparrow populations are structured regionally by isolation-by-distance, with gene flow occurring among marshes within ~110 to 135 km of one another. Isolation-by-resistance and isolation-by-environment also shape genetic variation; several habitat and landscape features are associated with genetic diversity and gen-etic divergence among populations. Human development in the surrounding landscape isolates breeding marshes, reducing genetic diversity, and increasing population genetic divergence, while surrounding marshland and patch habitat quality (proportion high marsh and sea-level-rise trend) have the opposite effect. The distance of the breeding marsh to the Atlantic Ocean also influences genetic variation, with marshes farther inland being more divergent than coastal marshes. In northern marshes, hybridization with Nelson’s Sparrow ( A. nelsoni ) strongly influences Saltmarsh Sparrow genetic variation, by increasing genetic diversity in the population; this has a concomitant effect of increasing genetic differentiation of marshes with high levels of introgression. From a conservation perspective, we found that the majority of population clusters have low effective population sizes, suggesting a lack of resiliency. To conserve the representative breadth of genetic and ecological diversity and to ensure redundancy of populations, it will be important to protect a diversity of marsh types across the latitudinal gradient of the species range, including multiple inland, coastal, and urban populations, which we have shown to exhibit signals of genetic differentiation. It will also require maintaining connectivity at a regional level, by promoting high marsh habitat at the scale of gene flow (~130 km), while also ensuring “stepping stone” populations across the range.


INTRODUCTION
Globally, 14% of avian species are currently threatened or endangered (IUCN, 2021) and many more are in decline (BirdLife International 2018, Rosenberg et al. 2019).Understanding the population genetic structure of these species is of critical importance for effective conservation (Haig et al., 2011).Armed with this knowledge, managers can gain an understanding of connectivity patterns among populations and identify isolated populations that may warrant protection (Kleinhans and Willows-Munro 2019).Moreover, knowledge of population structure can help managers make informed conservation decisions regarding translocations for genetic rescue (Grosser et al. 2017), facilitate the identification of appropriate source individuals for captive breeding and recolonization in extirpated parts of the range (Pellegrino et al. 2015, Monti et al. 2018), or enable the accurate release of seized, illegally captured individuals back to their population of origin (Domínguez et al. 2017).Population genetic structure can also inform the designation of evolutionary significant units or revision of taxonomic designations for improved management (Andersen et al. 2017, Davis et al. 2021).Understanding population genetic structure and genetic diversity can further aid in prioritizing limited conservation resources toward the most essential or most threatened populations (Botero-Delgadillo et al. 2020, Mendelsohn et al. 2020, Ruegg et al. 2020) and inform species status assessments and recovery plans (Smith et al. 2018).Determining species population genetic structure, therefore, is a critical foundation for effective conservation management programs.
Species become structured into genetically (and often demographically) distinct populations typically due to one of, or a combination of, 3 mechanisms.The simplest and most common driver of population structure is isolation-bydistance (IBD: Wright 1943), whereby the species' range extends beyond its single-generation dispersal capabilities.This results in geographically proximate individuals being more genetically similar, whereas geographically distant individuals are more genetically divergent, due to genetic drift and/or local adaptation.Even geographically proximate individuals, however, may become isolated due to isolation-by-resistance (IBR: McRae 2006), by which inhospitable landcover or habitat features restrict the dispersal of individuals among areas separated by these features.As these barriers reduce or eliminate dispersal altogether, subsequent gene flow is similarly affected.Consequently, genetic drift exerts a stronger force within the isolated areas, increasing population divergence through time.Alternatively, selection may play a role in structuring species into local populations, by a process known as isolation-by-environment (IBE; Wang and Bradburd 2014).
Here, natural and/or sexual selection acts on individuals to prevent dispersers or their offspring from successfully mating in a new environment.These drivers of population structure are not mutually exclusive, and the relative influence of each mechanism may differ across the species range.
Disentangling which of these mechanisms drives current population genetic structure is crucial for species of conservation concern, as it helps determine the most effective management actions for improving connectivity, gene flow, and genetic diversity (Lombal et al. 2020, Gousy-Leblanc et al. 2021).For populations structured by IBD, the genetic uniqueness of populations is correlated with their geographic isolation and, therefore, it is important to ensure the persistence of populations throughout their geographic range (Taylor et al. 2021).Additionally, if reintroductions to historical range areas are warranted, gaps should be filled with source individuals or offspring from adjacent populations within the genetic neighborhood (Wright 1946) or species dispersal distance.If IBR is found to be driving population structure, the creation of additional patches of suitable habitat or corridors may be important to maintain for continued connectivity among populations (van Rees et al. 2018, García et al. 2021).In species with population structure driven by IBE, results may be site specific and indicate environmental factors are shaping genetic fitness.In these instances, genetic rescue actions should usually avoid translocations among locally-adapted populations (Zimmerman et al. 2019, Botero-Delgadillo et al. 2020); however, certain populations may harbor important reservoirs of genetic adaptation that could be useful for future natural or assisted gene flow under changing environmental conditions (Oh et al. 2019).When planning management actions that consider environmental change, it may be necessary to prioritize conservation actions toward populations experiencing faster rates of change, as selection may make some populations more vulnerable than others (Bay et al. 2018).Hence, conservation management decisions grounded in the patterns and processes of a species' population genetic structure should improve outcomes for species of conservation concern.
Here, we sought to determine the mechanisms driving population structure in an imperiled passerine, the Saltmarsh Sparrow (Ammospiza caudacuta), a tidal marsh obligate that breeds along the northeastern Atlantic coast of the United States, from Maine to Virginia (Greenlaw et al. 2020).The Saltmarsh Sparrow is adapted to living in this harsh tidal salt marsh environment (Greenberg 2006).Female Saltmarsh Sparrows require high marsh for nesting and their reproduction is tied to the lunar tidal cycle, as their ground nests are highly vulnerable to flooding events during spring tides (Gjerdrum et al. 2005, Greenberg et al. 2006, Shriver et al. 2007, Bayard and Elphick 2011).Along the New England coast, the Saltmarsh Sparrow's breeding range overlaps with that of the more generalist Acadian subspecies of the congeneric Nelson's Sparrow (Ammospiza nelsoni subvirgatus).There the two species interbreed in a narrow hybrid zone, characterized by a mosaic of marsh microhabitats and concomitant patchy distribution of pure and admixed individuals (Hodgman et al. 2002, Walsh et al. 2015, 2016a, Maxwell et al. 2021).Despite considerable introgression within and surrounding this hybrid zone, the two species remain distinct and first generation (F1) hybrids are rare (Walsh et al. 2016a,b, Maxwell et al. 2021).As a result of marsh loss and degradation due to human modification, development, and sea-level rise, the Saltmarsh Sparrow has experienced 9% annual population declines and is at risk of extinction by 2060 (Correll et al. 2017, Field et al. 2017).Consequently, the species is considered globally endangered under the IUCN Red List (Bird Life International 2020) and is currently the subject of a Species Status Assessment by the United States Fish and Wildlife Service (USFWS) as part of the process to determine if listing under the Endangered Species Act (ESA) is warranted.
With the aim of informing conservation practices for this species, our research objectives were to (1) characterize the population genetic structure of the Saltmarsh Sparrow across the majority of its breeding range; (2) estimate genetic diversity and effective population sizes; and (3) determine the relative influence of geography, habitat and land cover, and introgression on Saltmarsh Sparrow population genetic struc-ture and diversity, to evaluate the relatives roles of IBD, IBR and IBE on genetic variation.

Study System and Sample Collection
Saltmarsh Sparrows (n = 964) were sampled from 24 marshes along the northeastern coastline of the United States over a 9-year period, between 2007 and 2015 (Figure 1; Supplementary Material Table 1).Sampling covered ~60% of the global breeding range of Saltmarsh Sparrows, with sites in Maine (n = 6), New Hampshire (n = 4), Massachusetts (n = 2), Rhode Island (n = 2), Connecticut (n = 3), New York (n = 5), and New Jersey (n = 2).A portion of the samples analyzed in this study were collected in 2007-2008 for a study evaluating fine-scale genetic structure in Saltmarsh Sparrows in the northern portion of their range (Walsh et al. 2012) and in 2012-2013 while studying patterns of introgression between Saltmarsh Sparrows and their sister species, the Nelson's Sparrow (Walsh et al. 2015).The majority of the remaining samples were collected during a 3-yr (2011-2013) study investigating survival and fecundity of Saltmarsh Sparrows (Ruskin et al. 2017a, 2017b, Field et al. 2018).We captured adult individuals using mist nets and banded each individual with uniquely numbered aluminum USGS bands.Whenever possible, blood samples (10 µL) were drawn from the cutaneous ulnar vein using a non-heparinized capillary tube and stored at room temperature on filter cards (Whatman or Nobuto) for later genetic analysis.When blood sampling was not possible, we pulled the 2 outer tail feathers (R6) and stored feathers in a -20ºC freezer.

Microsatellite Genotyping
DNA was extracted from blood samples using a DNeasy Blood Kit (Qiagen, Valencia, CA) by following manufacturer's instructions.For tail feathers, we isolated the calamus and followed the same protocol as with a standard tissue extraction, except with the addition of 10 μL of dithiothreitol to the lysis buffer and a 48-hr incubation.We amplified DNA using fluorescent dye-labeled primers for 16 microsatellite loci in 3 multiplexes: Aca01, Aca04, Aca05, Aca08, Aca11, Aca12, Aca17 (Hill et al. 2008), Escμ1 (Hanotte et al. 1994), Asμ15 (Bulgin et al. 2003), Ammo006, Ammo011, Ammo015, Ammo020, Ammo027, Ammo034, and Ammo037 (Kovach et al. 2015).Amplification of the Hill et al. (2008), Hanotte et al. (1994), andBulgin et al. (2003) primers followed conditions of Walsh et al. (2012).Amplification of the Kovach et al. (2015) primers followed conditions by Walsh et al. (2015) and Kovach et al. (2015).We conducted 2 replicates for feather samples to reduce genotyping error associated with lower quality samples.Amplified products were electrophoresed on an automated DNA sequencer (ABI 3130 genetic analyzer, Applied Biosystems, Foster City, CA) and individual genotypes were scored manually using PEAKSCANNER software (Applied Biosystems).We manually entered genotypes into CONVERT format (Glaubitz 2004) and used PGDSpider (Lischer and Excoffier 2012) to format the data for input into the programs described in the following section.We assessed potential genotyping errors with MICRO-CHECKER (Van Oosterhout et al. 2004).To identify loci potentially under selection, we performed F ST outlier tests in LOSITAN (Antao et al. 2008).We excluded 2 loci from our analyses (Ammo015, Ammo034) as they were potentially under selection (Supplementary Material Figure 1).We further excluded locus Ammo020 as it had missing genotypes for 15% of the individuals and was not in Hardy-Weinberg equilibrium.None of the samples used in this study had more than 2 missing genotypes at the retained loci.

Breeding Marsh Level Genetic Diversity
To assess genetic diversity within our 24 sampled breeding marshes, we calculated estimates of expected and observed heterozygosities in GENEPOP v. 3.4 (Raymond and Rousset 1995) and F IS estimates in FSTAT, version 2.9.3 (Goudet 1995).We estimated allelic and private allelic richness using the rarefaction method, which corrects for sample size differences among populations, using the program HP-RARE (Kalinowski 2005).We estimated effective population size (N e ) for each breeding marsh in NeEstimator, version 2.01 (Do et al. 2014), using the linkage disequilibrium method and random mating model (Supplementary Material Table 1; Hill 1981, Waples 1989, Waples and Do 2008).Rare alleles with frequencies <0.05 were excluded.We estimated confidence intervals via the parametric method.We also estimated N e for groups of breeding marshes based on the results of our population genetic structure analyses described below.We further assessed whether the below identified groups were in Hardy-Weinberg equilibrium using the unbiased Hardy-Weinberg exact test (Guo and Thompson 1992) as implemented in GENEPOP, using default MCMC parameters.Significance was assessed at the 0.05-level following false-discovery rate correction (Benjamini and Hochberg 1995) in R (R Core Team 2020).

Population Structure
We assessed the population genetic structure across our study area using the Bayesian clustering approach of STRUCTURE, version 2.3.4 (Pritchard et al. 2000).We first conducted 10 runs for each value of K = 1-24 for all marshes; each run consisted of a 50,000 generation burn-in followed by 100,000 iterations.In most cases where hierarchical genetic structuring exists, STRUCTURE only captures the major structure in the data (Evanno et al. 2005, Vähä et al. 2007).Thus, we used a hierarchical approach to explore patterns of genetic structuring within the Saltmarsh Sparrow.Individuals assigned to each cluster (K) identified in our initial STRUCTURE run were analyzed separately in a second structure analysis, again with a 50,000 burn-in period followed by 100,000 iterations, and so forth for a total of three hierarchical tiers.In each tier of analysis, we performed ten replicate runs for each possible K.For all runs, we used the admixture model, assumed correlated allele frequencies (Falush et al. 2003), and employed the LOCPRIOR model (Hubisz et al. 2009).We determined the most likely number of population clusters (K) based on cumulative evidence from the peak (or where values began to plateau) STRUCTURE mean probability estimates (ln Pr(X|K) and the peak second-order rate of change of the probability estimates (ΔK) method (Evanno et al. 2005), as well as examination of the assignment bar plots.STRUCTURE output was summarized using STRUCTURE HARVESTER (Earl and vonHoldt 2012) and CLUMPAK (Kopelman et al. 2015).
To evaluate genetic differentiation among breeding marshes as well as population genetic clusters identified from STRUCTURE analyses, we calculated pairwise F ST values and performed significance testing using 1,000 permutations in FSTAT, version 2.9.3 (Goudet 1995).As an alternate metric to the fixation index, we calculated allelic differentiation using pairwise Jost's D (Jost 2008), which is a relevant metric for conservation considerations (Jost et al. 2018).We generated heat maps for D in the base package in R to visualize the patterns among breeding marshes and STRUCTURE clusters (Figure 2).We also conducted a principle coordinates analysis (PCoA) in GenAlEx 6.5 (Smouse and Peakall 2012) using both the F ST and D matrices, to visualize the clustering patterns among populations, for comparison with STRUCTURE results.To evaluate gene flow, we also estimated recent (past few generations) migration rates among the population clusters (as identified in STRUCTURE) with BAYEASS 1.3 (Wilson and Rannala 2003), using default settings (100,000 MCMC steps with a 1,000 step burn-in).
To evaluate whether genetic variation correlated with geography, we tested for isolation-by-distance effects in R, version 4.0.3(R Core Team, 2020) by comparing matrices of great-circle geographic distance and genetic distance (expressed as linearized F ST ; Rousset 1997; Slatkin 1995) using a Mantel test (Mantel 1967) with 10,000 permutations, implemented in the R package VEGAN, version 2.5-7 (Oksanen et al. 2020).We also identified the extent of spatial genetic structure, using global spatial autocorrelation in the program GenAlEx 6.5.This approach uses pairwise geographic and pairwise genetic distance matrices to calculate an autocorrelation coefficient (r) for each of a series of predetermined distance classes (Smouse and Peakall 1999).Significant spatial genetic structure is inferred when r deviates from a random distribution of genotypes.Because selection of distance classes can influence the intercept values, we performed this test 2 ways: (1) with 11 distance classes, with narrower bins at nearby distances and wider bins at farther ones (25, 50, 75, 100, 150, 200, 250, 300, 350, 400, 450, and 503 km) and (2) 9 distance classes with even sample sizes.We determined significance with 9,999 random permutations and 10,000 bootstrap estimates of r.

Landscape Data
We compiled data for each breeding marsh for a suite of landscape variables that were expected to represent important patch-level (within a marsh) and between-marsh connectivity variables.Patch delineation was based on an existing marsh patch layer (Wiest et al. 2019) created in ArcGIS v 9.3 (ESRI 2009).Using this GIS layer, we obtained landscape variables for each sampling location.These included 4 patch-level variables (patch size [ha], patch perimeter [m], annual sea-level trend [changes in mean sea level over 30+ consecutive years, in mm yr -1 ], and proportion of high marsh vegetation) and 6 connectivity variables measured within a 1,000-m buffer surrounding the marsh patch (proportion of surrounding natural lands, of developed lands, of agricultural lands, of surrounding roads, of open water, and of neighboring marsh).We calculated 3 additional patch-level characteristics, which we hypothesized to be important in classifying marsh habitat.These included distance of the marsh to the Atlantic coastline (m), average mean high water (average of high water heights over the National Tidal Datum Epoch, from NOAA), and a proximity index to quantify the connectivity of a patch to neighboring marshes (essentially a function of the size of the marsh and the distance of the marsh to the next nearest marshland; see Gustafson and Parker 1994).We also included a binary dummy categorical variable in our analyses to code for whether a site fell within or outside of the Saltmarsh-Nelson's sparrow hybrid zone, between South Thomaston, ME, and Newburyport, MA, as defined by Hodgman et al. (2002) (within hybrid zone = 1; outside of hybrid zone = 0).

Landscape Predictors of Population Structure
We evaluated the relationship between genetic differentiation, landscape features, and geographic distance using distance-based redundancy analysis (dbRDA; Legendre and Anderson 1999).This approach performs multivariate multiple regression on non-Euclidean distance matrices; in this case, to estimate the percentage of genetic distance explained by a chosen landscape predictor.To evaluate whether genetic differentiation among Saltmarsh Sparrow breeding sites (F ST ) is driven by landscape features, we calculated pairwise ecological distances among each breeding marsh for each of the environmental predictors described above.For this analysis, we tested the pairwise correlation among all variables and excluded one of each pair with a Pearson correlation coefficient >0.7 (patch size, patch perimeter, proportion of surrounding roads, and average mean high water were removed).Because our data are of mixed types, we used the Gower method (Gower 1971), which is suitable for both quantitative data and dichotomous variables and has also been extended for fuzzy-type data such as proportions (Pavoine et al. 2009).We calculated pairwise distances for each variable independently and scaled them by range using the dist.ktabfunction in the R package ADE4 (Dray and Dufour 2007).We performed dbRDA using the capscale function in the R package VEGAN.This approach uses a backward stepwise model selection with ordistep function to identify the variables that best explain genetic differentiation.We assessed the significance of the best set of predictors using multivariate F-statistics with 999 permutations.To account for geographic variation among locations, we also performed a conditional dbRDA that further assessed the influence of the predictor variables on genetic distance while controlling for the pairwise geographical distances among the marshes.We additionally visualized the environmental space and differentiation of the breeding marshes in a principal components analysis (PCA) in R, using the GGBIPLOT package (Vu 2011).

Landscape Predictors of Genetic Diversity
To understand how landscape factors may be influencing genetic diversity in Saltmarsh Sparrows, we used linear modelling with an information theoretic approach.First, we tested whether the residual error distributions of each of our landscape variables met normality assumptions for linear modeling when regressed on allelic richness using the R package OLSRR (Hebbali 2020).As our residual errors approximated normal expectations, we then performed linear regression using sets of our environmental predictors as independent variables and our estimates of allelic richness as the dependent variable.We eliminated correlated variables if they were within the same model.We fit linear models based on 4 hypotheses predicting drivers of local genetic diversity: (1) bigger areas can support more individuals, with patch size (model 1a) and proportion of high marsh (model 1b) on a patch as covariates; (2) sealevel rise is reducing marsh area on a patch and, as a result, genetic diversity, with sea-level trend (2a) and proportion of high marsh (2b) as covariates; (3) the surrounding landscape influences gene flow and thereby genetic diversity, with proportion of surrounding marsh (3a) positively correlated with diversity, and development + agriculture (3b), proportion of surrounding open water (3c), proximity index (3d), and distance to coastline (3e) as covariates negatively influencing diversity; and (4) introgression affects gene flow, increasing diversity; with the dummy variable for in versus outside of hybrid zone (4a) as well as distance from hybrid zone (4b), as covariates.We also included a model with the top variables from each of the above hypotheses (5), as well as a null model (intercept only).Models were run using the MASS package in R (Venables and Ripley 2002) and model fit was assessed by AIC c and model weight (w i ) (Burnam and Anderson 2002).AIC c was calculated using the R package AICmodavg (Mazerolle 2020) and models within 2 AIC c were considered competitive.

Breeding Marsh Level Genetic Diversity
Observed heterozygosity was relatively high across sparrow breeding marshes (range: 0.754-0.869;Supplementary Material Table 1).Observed heterozygosity was greater than expected in two marshes, Jones Creek and Four Sparrow, consistent with negative F IS values for these sites.Inbreeding coefficient values were positive, but low, for all remaining sites, ranging from 0.001 to 0.091, with the highest value at the Little River marsh (Supplementary Material Table 1).Allelic richness differed among breeding marshes (2-way ANOVA F-statistic: 2.631 on 23 DF, P-value: 0.0001; Supplementary Material Figure 2); 2 New York City marshes had the lowest allelic richness relative to the other breeding marshes (Idlewild and Four Sparrow), while 2 Maine marshes within the hybrid zone (Little River and Spurwink) had the highest.N e estimates varied widely among marshes, from 11 for Four Sparrow (New York City) to 1,633 at ATT/Forsythe NWR, New Jersey (Supplementary Material Table 2).In general, the smaller N e estimates (<100) were for marshes that were small, isolated, or part of smaller marsh complexes, while the larger estimates were for marshes that showed higher connectivity (based on F ST ) and were part of larger marsh complexes (see population structure results below).N e was inestimatable (infinity) for a few marshes, likely due to high connectivity with other marshes, as part of a larger metapopulation.
Microchecker found one potential genotyping error within a single Idlewild individual at locus Ammo037, which had considerably smaller allele sizes than typically found in saltmarsh sparrow at this locus.We retained the genotype as it had little impact on our population-level allele frequencybased analyses; however, if the genotype was indeed an error, then the private allelic richness in the Idlewild marsh may be slightly inflated.

Population Genetic Structure
For the structure run across all 24 marshes, the mean ln Pr(X/K) plateaued at K = 5-7, with another peak at K = 11 (Supplementary Material Figure 3A); the strongest genetic signal in ΔK was found for 2 population groupings, with smaller peaks at 5, 7, and 11 (Supplementary Material Figure 3B).Examination of the bar plots for K =2 revealed 2 populations generally corresponding to breeding marshes occurring within, versus outside of, the hybrid zone (Figure 3A).At higher K, the bar plots indicated support for further structure, with marshes and marsh groupings clustering out fairly clearly up to K = 7, with less clear clustering revealed at higher K's (Supplementary Material Figure 4).We interpreted these findings to indicate a hierarchical structure, with the highest level of structure between the northern and southern groups.
Because of the north-south signal and the evidence for hierarchical structuring, further analyses were needed to tease apart the genetic structure on a finer scale.Analyses conducted on just the marshes south of the hybrid zone (New Jersey to Monomoy) identified 4 distinct genetic clusters (Figure 3B), with both ln Pr(X/K) and ΔK plateauing and peaking, respectively, at K = 4 (Supplementary Material Figure 5).The bar plots indicated some similarity across most populations as well as some distinct clustering among neighboring marshes, for example, Sawmill and Four Sparrow, Idlewild and the Long Island marshes (Marine Nature Center and Wertheim), and the Connecticut marshes (East River, Hammonasset, and Barn Island).Examination of the bar plots at K = 5 and 6 suggested further structuring, with separation of each New York City marsh and the 2 Rhode Island marshes (Sachuest and Chafee; Supplementary Material Figure 6).A third tier of analysis identified a total of 6 distinct population clusters within this region, as follows.Within the northern mid-Atlantic (New Jersey to Long Island), ln Pr(X/K) platueaed at K = 4, while ΔK peaked at 3 (Supplementary Material Figure 7).Examination of bar plots supported 4 clusters, separating each of the 3 New York City marshes (Sawmill Creek, Four Sparrow, and Idlewild) from the New Jersey marshes (OC-Mullica and ATT), which shared genetic similarity with those east of New York City on Long Island, New York (Marine Nature Center and Wertheim; Figure 3C; Supplementary Material Figure 8).In southern New England (Connecticut to Monomoy), ln Pr(X/K) and ΔK both supported 2 clusters, comprised of the Connecticut marshes (Hammonasset, East River and Barn Island), which were similar to Monomoy, Massachusetts, and distinct from the Rhode Island (Chafee and Sachuest) marshes (Figure 3D; Supplementary Material Figures 9 and 10).
Analyses within the marshes in the northern hybrid zone indicated 2 or 3 genetically distinct clusters, with a plateau in ln Pr(X/K) at K =3 and the ΔK suggesting K= 2 (Supplementary Material Figure 11).At K = 2, the strongest signal distinguished several individual sparrows, particularly in 3 southern Maine marshes (Little River, Jones Creek and Spurwink), from the sparrows in the rest of the breeding marshes, possibly representing individuals with introgressed Nelson's Sparrow alleles (Supplementary Material Figure 12).When 3 putative clusters were considered (Figure 3E), the Great Marsh on the Massachusetts/New Hampshire border (Parker River, Hampton and Fairhill) displayed a distinct signal shared with the coastal Maine marshes (Eldridge, Furbish and Scarborough).Similarly, the Great Bay marshes in New Hampshire (Chapman's Landing and Lubberland Creek) exhibited a distinct signal shared in part with one of the Maine marshes (Eldridge), which was fairly equally admixed between the 2 clusters.The likely introgressed individuals in Little River, Jones Creek and Spurwink make up the third distinct grouping.This pattern shown by the bar plots at K = 3 provided the most biological relevance, and we considered this the highest level of structure within this region (Figure 3E).Thus, at the finest level of structure, evidence suggests 9 genetic clusters across the study area (Table 1), with some local connectivity among them.Hardy-Weinberg exact tests on those 9 clusters determined that the Connecticut/Monomoy population was out of Hardy-Weinberg equilibrium at 2 loci (Aca08 and Ammo006).Two additional populations were out of equilibrium for locus Aca08 (New Jersey/Long Island, New York and the southern Maine cluster), as was the Sawmill Creek population for locus Aca04.
Pairwise F ST and Jost's D values showed similar patterns of population differentiation among the 24 breeding marshes, although D tended to be an order of magnitude larger than F ST .The 3 NYC marshes were significantly differentiated from all other marshes, including each other; the 3 upriver marshes in southern Maine (Little River, Jones Creek, and Spurwink) were significantly differentiated from all other marshes, except each other and nearby Scarborough (in the case of Jones Creek and Spurwink).Sachuest, RI and the two Great Bay marshes (Chapmans and Lubberland Creek) were also significantly differentiated from many, but not all, marshes (Figure 2; Supplementary Material Tables 2  and 3).In general, both metrics tended to show larger dif-ferences for distant marshes, although there were some low, nonsignificant differentiation values between some of the larger marshes in the Great Marsh complex (Parker River, Hampton and Scarborough) and those in Connecticut, New Jersey and Long Island.At the level of the 9 population clusters, the New York City and southern Maine marshes were the most differentiated (Supplementary Material Table 4).The most genetically similar marshes with the lowest pairwise F ST values were those within either the Connecticut/ Monomoy or New Jersey/Long Island, New York clusters.Those, along with the marshes in the Great Marsh cluster, were also the ones with the lowest Jost's D (Figure 2; Supplementary Material Table 5).Gene flow analysis with  PCoA showed generally similar clustering patterns for F ST and Jost's D, with populations differentiated roughly by latitude along axis 1, such that New York City, New Jersey and Long Island marshes were on one end of the axis, Great Marsh in the middle, and southern Maine at the other end (Supplementary Material Figures 13 and 14).Four Sparrow and Idlewild also showed separation along axis 2, as did Lubberland Creek.Marshes within STRUCTURE clusters tended to cluster together in the PCoA, with some overlap between New Jersey/Long Island and Connecticut and Rhode Island Marshes.Separation among marshes from different clusters was a little greater in the PCoA based on Jost's D, with greater spread along axis 2 as well.
The genetic clusters with the smallest estimated N e included the 3 distinct New York City marshes (Idlewild, Four Sparrow, and Sawmill Creek), as well as the smaller Rhode Island marshes (Chafee and Sachuest) and southern Maine marshes (Little River, Jones Creek, and Spurwink), all of which were estimated to contain <100 individuals (Table 1).The genetic clusters encompassing the largest marshes had the highest estimated N e (Mullica-OC in New Jersey and the Long Island, New York marshes, and the Great Marsh complex combined with the Furbish and Scarborough, Maine marshes), both with over 2,000 individuals.
Genetic distance was significantly correlated with geographic distance, indicating a signal of IBD (Figure 4A).Analysis of spatial autocorrelation among sparrow breeding marshes indicated genetic similarity among neighboring marshes, as well as some intermediately distanced marshes, with significant positive spatial autocorrelation occurring at the 25, 50, and 200 km distance classes for the variable distance scenario (Figure 4B; Supplementary Material Table 6A) and at 27, 48, 82 (marginally), and 229 km distance classes for the even sample size scenario (Figure 4C; Supplementary Material Table 6B).The extent of positive spatial genetic structure, as indicated by the x-intercepts ranged from 113 to 132 km.The significant genetic correlations in the 200 and 229 km distance classes suggest genetic similarity among some marshes at these greater distances, corresponding to shared signals in the structure plots between nonadjacent sites, such as Long Island and Connecticut/Rhode Island and Connecticut and Monomoy.

Landscape Predictors of Population Structure
Tests of the effects of landscape variables on genetic divergence among breeding marshes resulted in a best-fitting model that retained four significant landscape variables and explained approximately 15% of the overall variation in genetic differentiation among Saltmarsh Sparrow breeding marshes (marginal test constrained proportion: 0.1536; conditioned proportion: 0.0469; conditional test constrained proportion: 0.1405; Table 3).Marginal tests revealed a highly significant effect of surrounding development increasing population divergence (biplot score = 0.8062) and this variable remained highly significant after controlling for the effect of distance among the marshes (biplot score = 0.6946).Conversely, the higher the proportion of marshland surrounding a focal marsh the lower the genetic divergence (biplot scores = −0.2444and −0.4305 for the marginal and conditional tests, respectively).The distance to the Atlantic coast also influenced genetic differentiation, with more inland marshes more differentiated (marginal and conditional biplot scores = 0.2969 and 0.2285, respectively).For the marginal tests, the proportion of high marsh was also a significant predictor of genetic divergence (biplot score = 0.4127); however, it was not significant after controlling for geographic distance.Instead, after controlling for IBD, sea-level trend was a significant predictor of genetic divergence in the best supported model (biplot score = −0.3640).
When we visualized the landscape characteristics of each marsh in multidimensional space and grouped by population genetic cluster, there was a clear distinction in the habitat of most of the nine genetic clusters (Supplementary Material Figure 15).The majority (72.6%) of the variance in the habitat characteristics among the marshes was explained by the first 3 principal components.The first principal component explained 30.2% of the variation and primarily depicts the strong latitudinal gradients of sea-level trend (PC1 eigenvalue = −0.4787)and proportion of high marsh (PC1 eigenvalue = 0.4320).PC1 was additionally associated with the proportion of surrounding natural (PC1 eigenvalue = 0.4360) and agricultural lands (PC1 eigenvalue = 0.4172).This axis roughly differentiates the population clusters from south to north.PC2, which explained 23.5% of the variance in the data, was most strongly associated with how isolated the marshes are, as the proportion of surrounding open water (PC2 eigenvalue = −0.5891)and development (PC2 eigenvalue = 0.4281) were most closely associated with this axis, as well as the distance to the Atlantic coast (PC2 eigenvalue = 0.4575).Individual marshes, rather than population clusters per se, tended to be distinguished on this axis.PC3 (18.9% variance explained) is more of a connectivity axis, as the most important variables along that axis were the proximity index (PC3 eigenvalue = 0.4819) and the proportion of marshland surrounding the target marsh (PC3 eigenvalue = 0.4894), a pattern that was driven largely by OC-Mullica, in New Jersey.

Landscape Predictors of Genetic Diversity.
The best supported models (ΔAICc < 2) for predicting landscape drivers of genetic diversity were a model that included  only the distance from the hybrid zone and an additive model that included proportion of agriculture, development, and open water surrounding the marsh as well as distance of the marsh to the coastline (Table 4).Distance from the hybrid zone proved a better predictor than simply whether a site was in the hybrid zone.Models including the proportion of high marsh on a patch and sea-level trend also had greater support than the null model.Parameter estimates indicated that distance from hybrid zone, proportion of development surrounding the marsh, and sea-level trend were negatively related to allelic richness, while proportion of agriculture surrounding the marsh and proportion of high marsh on a patch had positive effects on allelic richness.Proportion of water surrounding the marsh and distance of marsh to the coastline, although in one of the well-supported models, were not informative variables, as the confidence intervals for the beta coefficients overlapped zero (Supplementary Material Table 7).

DISCUSSION
We sought to understand the factors shaping population genetic structure and diversity in the imperiled Saltmarsh Sparrow, given the importance of this knowledge for effective conservation of the species.We found that genetic variation is driven in part by IBD and also by several habitat and landscape factors, suggesting that IBR and IBE may also play a role.We discuss the influence of geography and habitat on Saltmarsh Sparrow genetic structure and its conservation implications below.

Breeding Marsh Level Genetic Diversity
Genetic diversity metrics estimated from microsatellite markers may not accurately reflect genome-wide diversity for many reasons, including ascertainment bias associated with selecting highly polymorphic loci.Consequently, microsatellite-based heterozygosity estimates may fail to capture true differences among populations (Väli et al. 2008).
Allelic richness of microsatellite loci; however, correlate well with genome-wide single-nucleotide polymorphism (SNP) diversity and therefore provide a useful proxy for genomewide diversity and a reflection of a population's demographic history (Fischer et al. 2017).While genetic diversity was not substantially reduced in any marsh, it did vary significantly across marshes, as measured by allelic richness.Diversity was lowest in 2 New York City marshes (Four Sparrow and Idlewild), which are surrounded by development and therefore likely experiencing elevated effects of genetic drift from isolation, and highest in two southern Maine marshes (Little River and Spurwink), in the hybrid zone.Notably, all but one population (Wertheim) contained some private alleles, consistent with signals of local population variation.The 2 New York City marshes with the lowest genetic diversity also had among the highest private allelic richness of the southern marshes, further consistent with the influence of isolation and drift.Whereas in the northern hybrid zone populations, Lubberland Creek, Little River, and Jones Creek marshes had the greatest relative amount of private allelic richness, likely due to Nelson's sparrow introgression introducing unique alleles into these breeding populations.Accordingly, these marshes also had among the highest allelic differentiation, measured by Jost's D, from other marshes, highlighting their conservation value.Interestingly, elevated private allelic richness in Barn Island may be indicative of introgression in a few of the sampled individuals, as further evidenced by signals in the STRUCTURE bar plots (Fig. 3A) that are similar with the patterns of introgression evident in Spurwink, Little River, and Jones Creek.While this is south of the limit of Nelson's Sparrow distribution, introgression can occur well beyond the hybrid zone (Walsh et al. 2016b).
Estimates of N e were <100 for 5 of the nine population genetic clusters, suggesting that they are below the recommended threshold for short-term persistence, and potentially at future risk of inbreeding depression (Frankham et al. 2014).Only three genetic clusters (New Jersey/Long Island, New York, Connecticut/Monomoy, and the Great Marsh complex of Massachuestts/New Hamphsire/Maine) have breeding populations large enough for likely long-term persistence (N e > 1,000), according to recommendations of Frankham et al. (2014).The low precision in the estimates for larger populations, often with confidence intervals extending to infinity, is typical and may suggest that additional samples and/or loci are needed to better estimate N e in the larger populations (Neel et al. 2013) For comparison with our genetic data, we also calculated N e based on prior estimates of patch-level abundance (Wiest et al., 2019) and a conservative ratio of N e to census size of 0.2 (Supplementary Material Table 8; Frankham 1995, Frankham et al. 2014).The estimates of N e from abundance tended to be lower than our genetic estimates of N e .Both types of estimates have separate sources of uncertainty, making it difficult to evaluate this comparison.Nevertheless, while the actual values estimated for N e may not be precise, the overall patterns among both estimation methods are consistent-the largest breeding populations in our study area encompass the New Jersey/Long Island, Connecticut/Monomoy, and Great Marsh/Furbish/Scarborough clusters, while the most imperiled populations, with the smallest N e are the three isolated New York City marshes.The Rhode Island, southern Maine, and Great Bay/Eldridge populations have slightly larger N e than those in New York City, though they are below the levels recommended for long-term persistence.These findings of low N e corroborate known population declines of the species (Correll et al. 2017).Summing together the genetic N e estimates across the nine clusters gives a total of 6729 for the region, which represents ~60% of the species breeding range.Doubling this estimate to account (generously) for the remainder of the range yields a total range-wide N e of 13,458.Applying the 0.2 ratio, this would correspond to a census population size of 67,290 for the species-consistent with the abundance-based estimates of 53,000 (37,000-69,000) and 60,000 (40,000-80,000) by Wiest et al. (2016Wiest et al. ( , 2019) ) for the same period.

Population Genetic Structure
Prior work has also examined Saltmarsh Sparrow population structure based on microsatellite loci (Walsh et al. 2012).Here, we have expanded the sampling of that paper, adding 560 new samples and increasing the number of sampled marshes from 9 to 24 to fill geographical sampling gaps and encompass the full breeding range of the caudacuta subspecies (Greenlaw et al. 2020; but see Walsh et al., 2017a regarding the validity of that subspecies).These additional data refine our understanding of Saltmarsh Sparrow population genetic structure.Accordingly, we identified 5 additional subpopulations-three genetically distinct breeding marshes in New York City, marshes along the Connecticut coast, and a group of marshes in southern Maine that may have extensive Nelson's sparrow introgression-for a total of 9 population clusters.We further identified additional breeding marshes that clustered with the 5 groups defined previously by Walsh et al. (2012)-Long Island (Wertheim), Rhode Island, Great Bay (Chapman's Landing), Great Marsh/Furbish, and Scarborough.Otherwise, we found identical population clusters where our sites overlapped among the 2 studies, except we suggest Scarborough groups more closely with the Great Marsh and Furbish cluster rather than on its own, consistent with a high level of admixture between these areas.
Our results also lend support to the hypothesis of Walsh et al. (2012) that habitat and position along the inland/coastal gradient influence population structure in Saltmarsh Sparrows.
We found population structure in the Saltmarsh Sparrow to be driven in part by IBD.Gene flow is fairly localized, with connectivity primarily among neighboring marshes and extending to a scale of 110-135 km.This pattern is consistent with known breeding site fidelity in this species: females return to nest on the same marshes, often within meters of their prior nests, in subsequent years (Benvenuti et al. 2018).Genetic structure, however, is not influenced solely by geography, as marshes did not cluster strictly with neighboring marshes.Some nearby marshes were genetically differentiated, and some marshes were similar across distances larger than 135 km.These patterns are due to 1) isolation of some small, fragmented marshes; and 2) the influence of habitat characteristics and an IBR or IBE effect discussed in the following secion.At the broadest scale, genetic variation partitions into 2 groups, north and south of the Cape Cod peninsula.These geographical areas differ in 2 primary factors that influence genetic differentiation: habitat and presence of Nelson's sparrow.First, the proportion of high marsh is strongly correlated with latitude, with more high marsh available in the northern marshes when compared with the southern marshes.With the exception of Monomoy Island, the southern marshes are composed of <50% high marsh (mean = 22%), while northern marshes are over 50% high marsh (mean = 75%).Second, the breeding range of Nelson's Sparrow is not known to extend south of Parker River (the southernmost breeding marsh in the northern population; Hodgman et al. 2002;Walsh et al. 2016b).Thus, ecological divergence among Saltmarsh Sparrows between northern and southern populations may be driving genetic structure due to habitat differences, along with introgression of Nelson's sparrow alleles in the northern populations.At the level of the 9 population clusters, landscape (including fragmentation), habitat, and other patchlevel influences further drive the patterns of differentiation observed.

Landscape Predictors of Genetic Diversity and Population Structure
Several habitat factors were found to influence both genetic diversity (as measured by allelic richness) within a breeding marsh as well as genetic divergence among marshes, lending support to IBR and IBE as mechanisms influencing genetic variation in Saltmarsh Sparrows.Firstly, the surrounding landscape had a strong influence on both genetic diversity and divergence, with those landscape cover types most dissimilar to sparrow habitat positively associated with signs of genetic isolation, suggesting a role for IBR.Specifically, the proportion of developed lands surrounding the marsh was an influential variable in our models of both genetic diversity and genetic divergence.Genetic diversity declined and genetic divergence increased significantly with increasing marsh isolation due to surrounding human development.The small, urban New York City marshes (especially Four Sparrow and Idlewild) are the most isolated in this regard, being surrounded by dense residential and commercial buildings, major highways, and John F. Kennedy International Airport.Accordingly, immigration into these marshes was negligible.Thus, we hypothesize that developed lands in the landscape matrix serve to isolate populations from gene flow, even from nearby marshes, resulting in increased genetic divergence and loss of genetic diversity due to drift.Urban development contributes to marsh habitat loss, is known to be detrimental to salt marsh integrity, and has been shown to genetically isolate populations of other tidal marsh bird species (Bertness et al. 2002, Bromberg and Bertness 2005, Major et al. 2014).Additionally, urban marshes likely have different selection pressures than rural ones and sparrows may be locally adapted to those metropolitan breeding sites (Teyssier et al. 2020, Caizergues et al. 2021, Salmón et al. 2021), which may exacerbate the genetic divergence of these marshes from the rest of the population, consistent with a role for IBE.Supporting this notion, Kocek et al. (2022) found that sparrows in New York city marshes exhibited differential nesting strategies relative to sparrows nesting in nonurban marshes.In this way, both IBR and IBE could be operating together in influencing genetic variation.
In contrast to the negative influence of development, we found that the proportion of marsh in the landscape matrix surrounding a breeding marsh reduces genetic differentiation among marshes.Thus, there is more likely to be genetic exchange among proximate marshes, as compared to marshes that are isolated by development or open water.The marsh clusters we found to have the lowest among-patch genetic divergence-Connecticut/Monomoy and New Jersey/Long Island, New York-include breeding sites with some of the highest ranked (of those studied) proportion of surrounding marshland and proximity index.Given these findings about the influence of surrounding landscape, it is likely that connectivity among Saltmarsh Sparrow populations was greater in the past, prior to large-scale fragmentation from urban development.Yet, the population structure we observed can only partly be attributed to genetic drift acting in isolation in a few small marshes fragmented from a larger population, as other local features were found to be associated with genetic similarity.These features (discussed below), in combination with high breeding site fidelity (Benvenuti et al. 2018), have led to variable patterns of gene flow and differentiation among marshes.
The position of the marsh along the inland/coastal gradient influenced genetic variation.Genetic divergence among marshes increased significantly with increasing distance from the coast.As for urban marshes, this divergence may be driven by isolation and drift, and/or by local adaptation.Inland or upriver marshes are less saline and typically have a dampened tidal regime compared to coastal marshes, thereby differing in ecology and selection pressures (Clark et al. 2022, Walsh et al. 2019a, Walsh et al. 2019b, Benham and Cheviron 2020, Mikles et al. 2020).The most inland marshes studied here were Sawmill Creek and the Great Bay marshes of Chapman's Landing and Lubberland Creek, although Sawmill does not follow the pattern of reduced salinity.Consistent with earlier work of Walsh et al. (2012), we found that the Great Bay marshes do not receive migrants from nearby coastal marshes, suggesting that inland marshes are isolated from gene flow with adjacent breeding sites, and may function as source populations for downriver marshes.
Within a marsh, the amount of high marsh was found to have a positive influence on both genetic diversity and divergence.Thus, as habitat differences between the northern and southern marshes may account for their broadscale differentiation, similarly, finer scale differences in amount of high marsh habitat drive patterns of differentiation across the 9 population clusters.While not a variable in the top models, high marsh habitat was among the genetic diversity models receiving moderate support and with a non-zero beta coefficient, suggesting a role for this habitat in supporting robust populations.Although Saltmarsh Sparrows may be present and feed in areas of low marsh, high marsh that is not inundated from daily tides is required for successful nesting and is the best proxy for the vegetation structure features important for high-quality nesting habitat (Gjerdrum et al. 2008, Meiman et al. 2012, Meiman and Elphick 2012).High marsh is indicative of high-quality nesting sites and female sparrows may seek out better quality nesting habitat following failed nesting attempts (DiQuinzio et al. 2001, Benvenuti et al. 2018, Kocek 2022).Therefore, sites with more high marsh better support local breeding productivity and may also attract dispersers from nearby marshes, thereby minimizing effects of genetic drift and increasing genetic diversity within these marshes.
The local relative rate of sea-level rise was found to negatively influence genetic divergence, after controlling for geographic distance, and it was also an influential covariate in our genetic diversity models.Similar to high marsh, the trend in sea-level rise reflects an aspect of patch quality.Sea-level trend affects Saltmarsh Sparrow productivity and demographic rates (Gjerdrum et al., 2008, Shriver et al. 2016) and may also influence population structure, if it in turn influences individual sparrows' dispersal patterns or site-specific fitness.This pattern would suggest another possible mechanism for IBE.Alternatively, there could be a confounding influence of latitude, as sea-level trend and proportion of high marsh are strongly correlated with latitude (r = 0.72 and -0.91, respectively), such that marshes at lower latitudes have less relative amounts of high marsh and are experiencing greater increases in mean sea level each year.The latitudinal gradients of sealevel trend and proportion of high marsh, along with the relative amount of natural and agricultural lands in the nearby landscape matrix, explained a large proportion of the habitat differences among our population clusters.
Lastly, interspecific hybridization with Nelson's Sparrow has a strong influence on the patterns of genetic variation we observed.Introgression with Nelson's Sparrow increases both genetic diversity and divergence, with distance from the hybrid zone as the sole variable in the top model explaining genetic diversity.Genetic diversity was highest in southern Maine marshes of Little River and Spurwink, which tend to have relatively high rates of introgression (Walsh et al. 2016b).These two marshes have large areas of high marsh habitat and are surrounded primarily by natural and agricultural lands, possibly making them more attractive to Nelson's Sparrows, which have a broader ecological niche in brackish and inland marshes and are less adapted to the tidal, coastal marshes (Nocera et al. 2007, Shriver et al. 2020).These sites within the southern Maine marsh cluster also exhibited the greatest genetic divergence, after Four Sparrow and Idlewild.From a conservation standpoint, hybridization is not posing a threat to the persistence of Saltmarsh Sparrows and may even be beneficial to the adaptive evolution of both species (Walsh et al. 2017b).

Conservation Management Implications
Saltmarsh Sparrows are declining rapidly (Shriver et al. 2016, Correll et al. 2017) and under continued threat due to sealevel rise and increased storm effects (Bayard andElphick 2011, Shriver et al. 2016).Given the pending listing decision for this species under the ESA, our study revealed important information for the evaluation of the species regarding long-term persistence.As part of the review of the species for consideration for listing under the federal Endangered Species Act, the USFWS uses the "3Rs" framework (Shaffer and Stein 2000).This framework examines resiliency, redundancy, and representativeness with an eye toward evaluating likelihood of species persistence.In these terms, resiliency and redundancy together represent factors that determine a species' long-term persistence, such that resiliency is achieved by populations that are large enough to survive stochastic events and redundancy implies there are enough populations that at least some will persist in the event of a catastrophic event (ensuring you don't have "all your eggs in one basket").Representation refers to conserving the breadth of populations of a species in the array of environmental conditions in which they occur, including the genetic variation among them, thereby conserving the capacity for adaptation to changing environmental conditions.
For the Saltmarsh Sparrow, most populations exhibit low effective population sizes indicating a lack of resiliency and accordingly, a potentially limited capacity to recover from stochastic events, at least for some local populations.The 3 New York City marshes-Sawmill, Idlewild and Four Sparrow-are the most at risk, with some smaller marshes (e.g., Sauchest, RI), potentially following suit.While there is some level of redundancy on the current landscape, rapid population declines suggest concern about redundancy is warranted, and some potentially distinct genetic variation (e.g., associated with inland/upriver marshes) may occur in a limited number of small marshes.To help ensure long-term persistence in the face of these threats, it will be imperative to monitor and pursue conservation actions on at-risk marshes, with the aim of improving conditions to bolster population sizes.Additionally, such restoration would have the added benefit of maintaining gene flow among populations and mitigating local loss of genetic diversity.
From a standpoint of representation, across their breeding range, Saltmarsh Sparrows occupy a broad latitudinal gradient, which is correlated with high marsh habitat availability and sea level trend, as well as other abiotic variables such as seasonality, temperature and precipitation.Because the breeding population is not panmictic across that latitudinal gradient, it will be important to ensure the persistence of breeding populations from throughout the latitudinal range to conserve the representative breadth of genetic and ecological diversity (Malcom and Carter 2021).Accordingly, we suggest that ensuring the persistence of both urban and inland breeding populations, along with coastal ones, is essential to retain the full suite of Saltmarsh Sparrow genetic diversity, as we found important signals of genetic divergence associated with these differing marsh types.While it is likely that those genetic patterns are due at least in part to isolation and genetic drift, especially on urban marshes, we cannot rule out the possibility of local adaptation, further emphasizing the importance of their conservation.Moreover, protecting multiple urban and inland breeding sites will support both the conservation aspects of representativeness and redundancy for this species (Wolf et al. 2015).
We also recommend considering hybridization as another important facet of representation for this species, which may also serve to increase the resiliency of Saltmarsh Sparrow populations, because hybridization increases genetic diversity and may provide adaptive capacity in events of rapid environmental change.The lack of a framework for inclusion of hybrid species protection in the ESA is outdated in the genomics era (Wayne and Shaffer, 2016, vonHoldt et al. 2018, Draper et al. 2021); however, prioritizing conservation of genetically "pure" Saltmarsh Sparrows is not warranted (Walsh et al. 2017b).
To support connectivity of Saltmarsh Sparrow populations at a regional level, it will be important to promote suitable breeding marsh habitat at the scale of gene flow, which we found to be approximately 110-135 km, and to ensure sufficient "stepping-stones" to maintain gene flow across the region.Our finding that higher proportions of high marsh and greater amounts of marshland in the surrounding landscape reduces isolation and increases genetic diversity further highlight the importance of minimizing the loss of this globally rare and important ecosystem, especially in areas with low proximity to other nearby marshland.Restricting additional conversion of marsh land, maintaining natural or agricultural land buffers around breeding marshes, and promoting marsh restoration are critically important for the conservation of this species.

FIGURE 1 .
FIGURE 1. Saltmarsh Sparrow sampling locations.Point colors indicate whether the sampled marsh is within the Saltmarsh x Nelson's hybrid zone (red) or not (blue).

FIGURE 2 .
FIGURE 2. Heat maps showing pairwise differentiation with Jost's D among 24 breeding marshes (top panel) and nine population clusters (bottom panel) of Saltmarsh Sparrow.Lighter yellow colors indicate greater connectivity, whereas darker orange and red indicate greater differentiation.

FIGURE 3 .
FIGURE 3. Composite ancestry proportion results of hierarchical Bayesian clustering analysis with program STRUCTURE, showing genetic structure of Saltmarsh Sparrows across 24 marshes.Marshes are ordered south to north, from left to right on the bar graphs, and the location of the southern boundary of the hybrid zone is indicated by the dashed vertical line in (A).Each vertical bar represents a single individual, and the different colors indicate putative proportional genetic ancestry.(A) Results across the entire study area, given two assumed genetically distinct groupings.(B) Results for the southern marshes outside of the hybrid zone, for 4 assumed genetically distinct groupings.(C) The northern Mid-Atlantic (New Jersey/New York) marshes, with 4 assumed groups.(D) The southern New England (Connecticut/Rhode Island/Monomoy, MA) marshes, with 2 assumed groups.(E) Results for the northern marshes within the hybrid zone, for 3 genetically distinct groupings.

FIGURE 4 .
FIGURE 4. Spatial genetic structure of Saltmarsh Sparrows.(A) Mantel test of isolation-by-distance.Mantel correlation statistics are annotated in the chart inset.Plot shows the simple linear regression of the pairwise geographic distances among individual Saltmarsh Sparrow breeding marshes (given as great-circle distances) and their pairwise population genetic distances (calculated as linearized Fst (Fst/(1-Fst)), with the best-fitting regression line to illustrate their relationship.Correlograms showing spatial correlation of genetic distance (Fst) among Saltmarsh Sparrows sampled from 24 marshes.The genetic correlation coefficient (r) was evaluated at 11 pre-defined distance classes (x-axis) in (B) and 9 distance classes with even sample sizes in (C).Mean correlation among pairs of populations is indicated for each distance class (solid line) and the bootstrap confidence intervals are indicated by the error bars.Dashed lines denote the 95% confidence intervals of r values within which the null hypothesis of no spatial autocorrelation is supported.Correlations are significant for the 25, 50, and 200 km distance bins in (B) and the 27, 48, 82, and 229 distance bins in (C).The x intercepts indicate that spatial structuring of Saltmarsh Sparrows extends to 113-130 km.

TABLE 1 .
Effective population size estimates, rounded to the nearest whole number, for the 9 hierarchical breeding marsh clusters of Saltmarsh Sparrow identified in STRUCTURE.Where relevant, cluster names also include in parentheses abbreviations used in subsequent figures and tables.BAYEASS across the 9 clusters revealed negligible (<1%) immigration rates into each of New York City marshes, as well as the Rhode Island, Great Bay and southern Maine marsh complexes.The larger marsh complexes received immigrants primarily from neighboring marshes: with migration rates of 15% and 17% from Sawmill and Four Sparrow into New Jersey/Long Island, 14% from Idlewild to Connecticut/ Monomoy, and 12% and 16% from Great Bay and southern Maine into Great Marsh (Table2).

TABLE 2 .
Recent generation migration rates estimated with BAYESASS among 9 population clusters of Saltmarsh Sparrow.Standard errors in parentheses Italic font for values on the diagonal indicate the percent residency to each marsh.

TABLE 3 .
Distance-based redundancy analyses of landscape drivers of population genetic differentiation as measured by F ST among 24 breeding marshes of Saltmarsh Sparrows.Marginal tests assess the influence of the individual landscape variables, whereas the conditional tests assess the influence of the landscape variables after controlling for the effect of geographic distance between marshes.Dashed lines indicate that the variable was not influential in that test (P > 0.05).

TABLE 4 .
Multiple linear regression results of best-fitting models describing landscape drivers of population genetic diversity in Saltmarsh Sparrows as measured by allelic richness,