Genetic sampling

We collected 16,420 genetic samples across the range of sage-grouse in western North America spanning 11 U.S. states and two Canadian provinces. Feathers that had been dropped by sage-grouse during breeding activity were collected from 2005 to 2015. In addition, a small number of blood samples collected by other researchers as part of separate radio telemetry and GPS tracking field research were made available to us for use in this study. The spatial distribution of our sampling was optimized as described in Hanks et al. [25]. Briefly, following a pilot sample, we used a latent spatial model to identify spatial trends in genetic variation. We then identified geographic areas exhibiting the greatest transition in allele frequencies but where we lacked samples and increased our sample resolution in these areas in subsequent years.

Microsatellite genotyping

The laboratory methods used to generate the genetic data for this study have been described previously [26, 27] and detailed methods are in S1 Appendix. Briefly, DNA was extracted and used in the amplification of 15 microsatellite loci, 12 of which were developed for sage-grouse [28–30]. We conducted extensive quality control to ensure the reliability of all genotypes and identified and removed multiple recaptures of the same individual. The data used in this study are available as a USGS data release [31].

Sample optimization

We sampled some regions of the range with greater intensity than others. As heterogeneity in sampling intensity has been shown to affect the analysis of spatial genetic structure [18, 32, 33], we thinned the full dataset to achieve a more homogenous spatial representation. To thin the dataset, we followed the approach of Row et al. [27]. We used a hierarchical clustering approach to first define clusters, by grouping spatially proximate sampling locations using a hierarchical clustering analysis (hclust function with complete method) in R [34]. Each location was assigned to its own cluster, with the algorithm combining the two most proximal clusters iteratively until a single cluster remained. Distances between clusters were recomputed after each join by the Lance-Williams dissimilarity formula [35]. Finally, we used the cutree function in R to define clusters separated by a threshold distance of >50 km, representing two times the maximum average summer to winter movement distance for any population in Wyoming [36]. Once clusters were defined, we randomly chose ten individuals per cluster for those clusters that had >10 individuals.

We conducted a group-based principal component analysis (PCA) by computing the mean component scores for all individuals located within each state/province/regional group using the R packages stats and gstudio [37] and plotted the results using ggplot2 [38]. Previous genetic and genomic research has shown that sage-grouse from the Bi-State region (along the border between California and Nevada in the southern part of the range in both states, Fig 1) and from Washington are substantially differentiated from other parts of the sage-grouse range and were likely isolated for long periods of time [39–42]. Therefore, we grouped all individuals from the Bi-State region together to separate them from other individuals from Nevada and California. Our group-based PCA revealed (see Results) that, as has been found in previous analyses, the Bi-State and the Washington sage-grouse, are significantly differentiated from all other sage-grouse (S1 Fig). Since we were interested in large-scale patterns of population structure and gene flow across the core of the range (i.e., not influenced by isolated peripheral populations that are known to be genetically distinct), we removed the Washington and Bi-State populations from subsequent analyses with one exception—the analysis of estimated effective migration rates (i.e., gene flow) across the entire range of sage-grouse.

Examining population genetic structure

To investigate large-scale trends in genetic structure, we conducted an individual-based spatial principal component analysis (sPCA) calculated in the R package adegenet [43]. Additionally, we used the Bayesian clustering program STRUCTURE v2.3.4 [44] using the admixture model, correlated allele frequencies, and allele frequency distribution parameter (λ) set to 1. Analyses were run with non-informative priors and burn-in of 1,000,000 iterations followed by 1,000,000 Markov Chain Monte-Carlo (MCMC) iterations. For initial analysis, we completed ten replicates for each value of K from 1–30. The best-supported value of K was determined using two methods as implemented in STRUCTURE HARVESTER [45]. First, we plotted the mean and standard deviation of ln Pr(X|K) at each value of K and evaluated the point at which those values plateau. Second, we calculated and plotted ΔK following Evanno et al. [46]. Finally, we used CLUMPP v 1.1.2 [47] to average posterior probabilities of individual assignment to each of the K clusters across all ten STRUCTURE replicates. In CLUMPP we used 10,000 repeats of the greedy method with greedy option two and the pairwise matrix similarity statistic, G.

We examined spatial autocorrelation in our samples using the mantel.correlog function in the vegan package [48] inputting AMOVA genetic distance calculated using the genetic_distance function in R package gstudio [37]. We computed a multivariate Mantel correlogram (as described in [35]) with 80 distance classes, limiting the correlogram to the distance classes that included all points. A Pearson correlation was used in calculation of the Mantel statistic, with 999 permutations to test for significance, and a Bonferroni correction of P-values for multiple testing including progressive correction of multiple-testing (as described in [35]; briefly, the test for the first distance class: no correction, the test for the second distance class: correct for 2 simultaneous tests, the test for the k-th distance class: correct for k simultaneous tests).

Identification of subpopulation centers

Sage-grouse are continually distributed over large portions of their range and the genetic structure of the species follows a pattern of IBD [49] across continuous habitats [42]. Consequently, in many parts of the range, discrete genetic boundaries likely do not exist and defining hard boundaries for genetic populations is problematic [18]. Therefore, our objective was to identify genetic subpopulation centers, which would theoretically represent the core of each distinct genetic group. To do this, we used two independent methods. Our first approach involved assigning individuals to a genetic group based on their maximum posterior probability score from the standard STRUCTURE analysis above, then using adehabitatHR functions kernelUD and getverticeshr in R to calculate the 25%, 50%, and 75% kernel density estimate (KDE) of these genetic groups based on the maximum population assignment. Our second approach used a spatial iterative bifurcation process (SIBP) composed of three steps for each iteration: (1) division of the individuals into two clusters, (2) spatial interpolation of individual posterior probabilities of membership for each of the two clusters, and (3) selection of individuals constituting the ’subpopulation center’ for each of the two clusters. In step one, we analyzed the entire sample at K = 2 using STRUCTURE and CLUMPP with all parameter settings as described above. In step two, we spatially interpolated the individual posterior probabilities (where values range from 0–1, and total 1) of individual subpopulation membership to each of the K = 2 clusters using the default settings of the krig function in the fields package [50] in R, following Fedy et al. [6]. The result of the fitted model is an interpolative surface derived from irregularly spaced data. From the kriged posterior probabilities of cluster membership we generated spatial isoclines of probability of membership to each of two subpopulations based on the spatial arrangement of samples and their respective strength of membership. In step three, we divided individuals into two groups based on the kriged isoclines such that all individuals within each of the two ≥ 0.70 isoclines were subset into their respective groups and retained. Samples located within the < 0.70 isoclines for either of the two clusters (i.e., not members of the two subpopulations or groups) were dropped from further iterations. We repeated this process of analysis, krig, subset for retained groups until no individuals remained within the ≥ 0.70 isoclines. In naming subpopulation centers, we switched between the binary nomenclature of 1|2 and a|b to indicate each subpopulation center’s lineage. For example, the primary subdivision of all samples was 1 and 2. The secondary subdivision was 1a|1b, and 2a|2b. We also assigned names to subpopulation centers based on geographic regions.

Quantification of genetic diversity and divergence

We used gstudio [37] in R to characterize genetic diversity within each subpopulation center using the following metrics: observed (H_o) and expected (H_e) heterozygosity, average number of alleles (A), effective number of alleles (Ae), number of alleles with at least 5% frequency (A₉₅), and the inbreeding statistic F_IS. In GENEPOP v4.5.1 [51], we tested for deviations from Hardy-Weinberg proportions (HWP) and for gametic disequilibrium among loci correcting for multiple tests for significance using Bonferroni corrected P-values. We used GENODIVE [52] to quantify divergence among subpopulation centers using both Ɵ_ST [53] and Jost’s D [54].

Estimated effective migration surfaces

We used the Estimating Effective Migration Surfaces program (EEMS; [55]) to quantify genetic differentiation resulting from differences in effective migration. To do this, we used EEMS to model rates of migration among demes using a stepping-stone approach [56]. The program overlays the sample range with a tight systematic grid and approximates the expected genetic differentiation among individuals under an IBD model. Then, comparing actual genetic differentiation among individuals to the IBD model, the program algorithms identify areas of divergence between actual and modeled differentiation. Those areas where actual differentiation is greater than predicted by the IBD model are areas of lower effective migration, whereas areas where actual differentiation is less than predicted by the IBD model are areas of greater effective migration. Using EEMS, we identified geographic areas across the species’ range with lower effective migration than expected under IBD (resistance to gene flow), with approximately equivalent effective migration (neutral effect on gene flow), or with greater effective migration (facilitation of gene flow). We plotted the EEMS-output spatial data layer using the rEEMSplots package [55] and the ggplot2 package [38] in R. In EEMS, we fit models to, and averaged the results for, deme counts of 500 and 1,000, repeating this process three times at each deme count to ensure model convergence across random starting points. We used 1,000,000 MCMC iterations burn-in, 2,000,000 subsequent iterations, and thinned by 9,999 iterations. We repeated this analysis for the thinned dataset including Washington and the Bi-State populations.

Microsatellite genotyping and sample thinning

Genotyping and quality control yielded 6,725 individuals from 1,364 breeding areas or leks (median = 3, IQR = 4, range = 1–62 individuals per lek) collected across the entire range from 2005 to 2015. The sample genotypes we used were the same as those refined for use in Cross et al. [26] and Row et al. [27], now including samples from Washington. After thinning of the data to avoid heterogeneous sample distribution, the number of individuals was reduced to 2,134 from 927 leks (median = 2, IQR = 2, range = 1–13 individuals per lek). Further removal of the individuals from the Bi-State population and Washington state resulted in 2,091 individuals from 915 leks (median = 2, IQR = 2, range = 1–13 individuals per lek).

Determination of population genetic structure

Our PCA grouped by state/province/region (S1 Fig) showed the populations in Washington and in the Bi-State were more divergent than average from other groups. Therefore, we withheld these individuals from all other analyses (except the second analysis of effective migration). In the sPCA, we retained eigenvalues 1–3, as they were the most divergent from the other axes and characterized the global trends in variance and spatial autocorrelation. The first eigenvalue split the northeast (Montana, North and South Dakota and northeastern Wyoming) and the remainder of the range (S2A Fig). The second eigenvalue highlighted a difference between southwestern Wyoming, Colorado, and Utah and the rest of the range (S2B Fig). The third eigenvalue separated Utah from much of the rest of the range, yet also showed regional differentiation in the Dakotas and southeastern Montana and the westernmost portion of the range (S2C Fig). When combined, these first three eigenvalues (Fig 2) did not reveal distinct population boundaries, but rather a pattern of population cores and transition zones between those cores. However, at least four groups are apparent: one group included most of Montana, North Dakota, South Dakota, and northeastern Wyoming (Fig 2; pink), which transitions to a differentiated group including southwestern Wyoming, northeastern Utah, and Colorado (Fig 2; purple). Further south, a shorter transition to the green and brown in central and southern Utah was detected (Fig 2). The remainder of the range—the greater Great Basin into Idaho and southwestern Montana—grouped together (Fig 2; turquoise).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Spatial principal components analysis color plot of the first three eigenvectors for 15-locus microsatellite data from Greater Sage-grouse samples collected across the species’ range.

The colors summarize the first three components by translating each score into a channel of color (red, green, and blue). The dotted red line represents the Continental Divide. Tan polygons represent Priority Areas for Conservation and dotted black lines represent the seven sage-grouse Management Zones (MZ). State names are represented by the following abbreviations California (CA), Colorado (CO), Idaho (ID), Montana (MT), Nevada (NV), North Dakota (ND), Oregon (OR), South Dakota (SD), Utah (UT), Washington (WA), and Wyoming (WY).

https://doi.org/10.1371/journal.pone.0274189.g002

In our standard STRUCTURE analysis, the most likely value of K given our data was equivocal. The several common approaches to selecting the most appropriate K each supported a different value. The point at which the mean ln Pr(X|K) began to plateau was approximately at K = 6 (S3A and S3C Fig). This was also the point at which the rate of change in ln Pr(X|K) began to diminish, and so it was the value of K for which we observed the second greatest ΔK statistic value (S3B Fig). However, it could also be argued that the values of ln Pr(X|K) plateaued at K = 12, and the maximum value occurred at K = 21 (S3A and S3D Fig). The ΔK indicated K = 2 had the greatest support as the uppermost hierarchical level of structure (S3B Fig). At K = 6, individuals were divided into groups that were mostly contiguous, such that most geographically proximal individuals were assigned to the same genetic cluster (S4A Fig), with the six groups representing (1) the greater Great Basin (Oregon, California, Nevada), (2) Idaho and southwestern Montana, (3) central and southern Utah, (4) southwestern/central Wyoming, Colorado, and northeastern Utah (5) northeastern Wyoming with the Dakotas and southeastern Montana, and (6) central/northern Montana. As in the sPCA, contact zones between groups showed increased admixture within individuals compared to those closer to group centroids. Individual assignment was particularly ambiguous for individuals throughout much of Nevada and the Great Basin. At K = 12, geographic discreteness was reduced with multiple groups co-occurring (S4B Fig).

We observed statistically significant positive autocorrelation among pairs of sage-grouse samples separated by < 50 km. Some, but not all, distance classes above 450 km showed significant negative autocorrelation. Pairwise samples between 50 and 450 km were not significantly autocorrelated.

Identification of subpopulation centers

We identified 25%, 50%, and 75% KDE for individuals assigned to each of the K = 6 genetic clusters (Fig 3). The KDE revealed non-overlapping but non-contiguous (25% KDE) or largely non-overlapping and contiguous (50% KDE) areas that included one group in northeastern Montana (dark blue; Fig 3), a second including southeastern Montana, northeastern Wyoming, North and South Dakota (light blue; Fig 3), a third including southwestern Wyoming, Colorado, and a small portion of Utah (orange; Fig 3), a fourth including most of Utah and a sliver of southwestern Wyoming (yellow; Fig 3), a fifth covering much of Nevada, Oregon, southwestern Idaho, and extreme northeastern California (salmon; Fig 3), and a sixth that included central Idaho and southwestern Montana, dipping slightly into northeastern Nevada and northwestern Utah (brown; Fig 3). The 75% KDE revealed contact zones between groups. Three groups converged in southwestern Wyoming along the border where Wyoming, Utah and Idaho converge, (brown, orange, yellow) and south-central Idaho (salmon, brown, and yellow).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Boundaries of Greater Sage-grouse subpopulation centers at K = 6 from a STRUCTURE analysis using a 25% (darkest polygon), 50% (medium polygon), and 75% (lightest polygon) kernel density estimate to determine genetically distinct groups.

The six different colors represent the six clusters identified in the STRUCTURE analysis. The dotted red line represents the Continental Divide. Tan polygons represent Priority Areas for Conservation and dotted black lines represent the seven sage-grouse management zones (MZ). State names are represented by the following abbreviations California (CA), Colorado (CO), Idaho (ID), Montana (MT), Nevada (NV), North Dakota (ND), Oregon (OR), South Dakota (SD), Utah (UT), Washington (WA), and Wyoming (WY).

https://doi.org/10.1371/journal.pone.0274189.g003

We discovered twelve nested subpopulation centers across the species’ contiguous range after six successive rounds of analysis (Fig 4) using our SIBP method. The primary division in genetic structure loosely aligned near the Continental Divide (S5A and S6A Figs) bifurcating the range into eastern and western groups (Pop 1 and Pop 2, Fig 4A). Secondary subdivisions (S5B and S6B Figs) divided the western group (Pop1, Fig 4A) into a Great Basin group (Great Basin Pop1a, Fig 4A), and a group east of the Great Basin (Pop 1b, Fig 4A), and separated the eastern group (Pop 2, Fig 4A) into a north central Montana group (Great Plains Pop 2b, Fig 4A) and a group covering the Dakotas, southeastern Montana, and eastern Wyoming (Powder River Pop 2a, Fig 4A). Tertiary subdivision (S5C and S6C Figs) bifurcated a group in Oregon (Central OR Pop1a2, Fig 4A) from the rest of western Great Basin (from which no kriged posterior probabilities were ≥0.70) and divided the group east of the Great Basin into two groups (Pop 1b1 and Pop 1b2, Fig 4A). Quaternary subdivision (S5D and S6D Figs) separated a group in southwestern Montana, eastern Idaho, western Wyoming, northern Utah (Central Rockies Pop1b1a, Fig 4A), from one in northwestern Colorado and southern Wyoming (Wyoming Basin Pop1b1b, Fig 4A) as well as two groups in central Utah (Pop 1b2a and 1b2b, Fig 4A). Quintenary (S5E and S6E Figs) and senary (S5F and S6F Figs) subdivisions further bifurcated the central Utah group into six subpopulation centers (Fig 4A). Each successive round of substructure analysis resulted in a cut of 3–52% of the total starting samples with a mean cut of 19% (those individuals not within the ≥0.70 isocline S1 Table). After seven rounds of analysis (the senary round), no individuals remained within the 0.70 isocline (S1 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4.

(A) Diagram showing how Greater Sage-grouse subpopulation centers were identified using 6 rounds of spatial iterative bifurcation process (for detailed interpolated maps and a summary of individuals included in each bifurcation, see S5 and S6 Figs) and (B) the distribution of individuals in each subpopulation center with color and symbol representing the 12 subpopulation centers. Open grey symbols represent individuals that occur outside of the subpopulation centers. The dotted red line represents the Continental Divide. Tan polygons represent Priority Areas for Conservation and dotted black lines represent the seven sage-grouse management zones (MZ). State names are represented by the following abbreviations California (CA), Colorado (CO), Idaho (ID), Montana (MT), Nevada (NV), North Dakota (ND), Oregon (OR), South Dakota (SD), Utah (UT), Washington (WA), and Wyoming (WY).

https://doi.org/10.1371/journal.pone.0274189.g004

Quantification of genetic diversity

Among all samples, there was an average of 16.53 alleles per locus ranging from eight at TUT3 to 33 at MSP11, with an expected heterozygosity of 0.845, and an F_IS of 0.075 (Table 1). After grouping samples by SIBP, no single locus was out of HWP in more than one subpopulation center; however, there were four loci in the Great Basin group, one in the Central Rockies group, two in the Powder River group, and one in the Great Plains group that were out of HWP. No locus pairs were in significant gametic disequilibrium in more than one subpopulation center. Within subpopulation centers the average number of alleles per locus ranged from 4.87 in Wildcat-Horn to 13.60 in the Great Basin subpopulation center (mean = 8.77 ± 3.44 [SD]). Across all subpopulation centers, H_e averaged 0.747 ± 0.16 [SD] (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Measures of genetic diversity across 15 microsatellite loci within each of the spatial iterative bifurcation process (SIBP) subpopulation centers detected for Greater sage-grouse.

https://doi.org/10.1371/journal.pone.0274189.t001

Divergence among the 12 subpopulation centers ranged widely (Ɵ_ST: 0.021–0.156, Jost’s D: 0.084–0.444; S2 Table). Divergence (S2 Table) measured using Ɵ_ST was greatest among subpopulation centers in Utah (Uintah, Wildcat-Horn, Sheeprock-Tintic, and Southwest Desert). Yet when using Jost’s D, divergence was greatest between the Great Plains subpopulation center and centers in Utah (Sheeprock-Tintic, Uintah, Wildcat-Horn), between the Central Rockies and Wildcat-Horn and between centers in Utah (Sheeprock-Tintic and Uintah). Divergence was least when measured using Ɵ_ST between the subpopulation centers in Oregon (Central OR and Great Basin) and when measured using Jost’s D between the centers in Montana (Great Plains and Powder River).

Estimated effective migration surfaces

We estimated effective migration rates for the entire range of sage-grouse including Washington and the Bi-State, and then separately excluding samples from those two areas. As expected, analysis of the entire range revealed areas of constrained migration between Washington and the rest of the range and to a lesser extent between the Bi-State population and the rest of the range (S7 Fig). When excluding Washington and the Bi-State, effective migration was greatest within the Powder River (southeastern Montana and the Dakotas), diminishing slightly moving northward into the Great Plains (Fig 5, positive log migration, orange). Other areas of greater migration include within the northern part of the Wyoming Basin subpopulation center, the Central OR subpopulation center, and stretching across the Great Basin connecting the Great Basin subpopulation center to the Central Rockies subpopulation center. Low migration rates (Fig 5, negative log migration, shaded blue) were found in northern Idaho, between eastern Montana (Great Plains and Powder River subpopulation centers) and southwestern Montana (Central Rockies subpopulation center), through western Wyoming, extending south through far eastern Idaho, into all but the northwesternmost corner of Utah, and all of Colorado. Additional areas of low migration included central Wyoming between the northern part of the Powder River subpopulation center and the southern portion of the Wyoming Basin center stretching south into Colorado.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. The 12 Greater Sage-grouse subpopulation centers identified using the spatial iterative bifurcation process overlaid on the effective migration surface.

The color and symbol represent each of the 12 subpopulation centers. Open grey symbols represent individuals outside of subpopulation centers. Effective migration rates are shown on a log scale where the zero value (white) indicates the mean effective migration rate, positive values (orange) indicate greater than average effective migration and negative values (blue) indicate those less than average. The dotted red line represents the Continental Divide. Tan polygons represent Priority Areas for Conservation and dotted black lines represent the seven sage-grouse management zones (MZ). State names are represented by the following abbreviations California (CA), Colorado (CO), Idaho (ID), Montana (MT), Nevada (NV), North Dakota (ND), Oregon (OR), South Dakota (SD), Utah (UT), Washington (WA), and Wyoming (WY).

https://doi.org/10.1371/journal.pone.0274189.g005

Comparing patterns among analyses

Sage-grouse habitat is contiguous in some areas of the range and fragmented and isolated in others (Fig 1) and the extent to which IBD might impact our analysis was unknown, although a previous genetic study suggested that IBD is a prominent pattern across the range [42]. Although we used STRUCTURE and sPCA to define major patterns of genetic difference and to identify centers of genetic differentiation through SIBP (all of which are influenced by IBD), we also used EEMS that was designed specifically for instances in which broad-scale IBD exists and compared all results.

All analyses supported differentiation and low effective migration between an east and west group that split loosely along the Continental Divide (Figs 2 and 5), except where the topographical abruptness of the Divide is less severe (e.g., between southwestern Montana and eastern Idaho, and within WY). Divisions below the initial east/west split varied by analysis with sPCA resolving the fewest groups followed by STRUCTURE and SIBP. Within the eastern group, northern Montana (Great Plains) and southeastern Montana/Dakotas/northeastern Wyoming (Powder River) were further separated by STRUCTURE (Fig 3, S4A Fig) and the SIBP (Fig 4), yet only minimally with sPCA (shown as differentiated with the 3^rd eigenvector). Our EEMS analysis indicated slightly elevated effective migration between these two groups (Fig 5), which is consistent with the spatially predicted connectivity in that area described in Row et al. [27]. Within the western group, three or four additional groups were defined using sPCA and STRUCTURE with additional divisions detected using SIBP. Idaho and a Nevada/Oregon group comprise two (STRUCTURE) or three genetic (SIBP) groups (Figs 3 and 4), yet sPCA revealed only subtle differentiation and EEMS showed elevated effective migration among these groups (Figs 2 and 5). The Wyoming Basin group (southwestern Wyoming and Colorado) was consistently recognized as one group throughout all analyses. Effective migration within this group varied substantially, however, with high migration in southwestern Wyoming and lower migration in Colorado (Fig 5). In contrast, there was no consistency in the classification of samples in Jackson Hole, Wyoming (isolated population in the northwestern part of the state). This group clusters with samples from Idaho and Montana (Central Rockies subpopulation center) in the SIBP (Fig 4), it aligns with southwestern Wyoming in the sPCA (Fig 2), and it lies outside all polygons in the KDE (Fig 3). This population occurs in an area of local low effective migration in the EEMS analysis (Fig 5) and has been found to be isolated previously [57] yet not as genetically distinct as Washington or the Bi-State [39]. The degree of differentiation and isolation in central and southern Utah was also unclear but was nested in classification (i.e., the group was either considered one large group or six small groups within it) unlike Jackson Hole (which was assigned to completely different groups depending on the analysis). The SIBP analysis (Figs 4 and 5) revealed 6 subpopulation centers in Utah with increased differentiation (S2 Table), whereas the standard STRUCTURE analysis (Fig 3, S4A Fig) and to a lesser extent spatial PCA (Fig 2) showed one group differentiated from the rest of the range. Sage-grouse in central and southern Utah occur in small, isolated patches of sagebrush likely highly affected by genetic drift which has led to increased differentiation. Translocations from multiple areas in Utah (including the northeast and northwest) into central Utah (specifically into Strawberry-Anthro) complicate the interpretation of the structure found here [58–60] and may explain why the differentiation in the Strawberry-Anthro subpopulation center was reduced and its genetic diversity greater than others in that area.

Improving resolution and extent

Our analysis provides a comprehensive view of population genetic structure and gene flow across the entire range of sage-grouse. Past evaluations of connectivity for sage-grouse have examined population structure and gene flow both range-wide [42] and at smaller scales [5, 6, 40, 57, 61–64]. The only other range-wide assessment of population structure included over 1,000 samples largely from hunter-harvested birds, yet the spatial resolution was limited as samples were consolidated by hunt unit or county [42]. We found general agreement with that study yet found greater differentiation in central Utah (6 subpopulation centers here vs. 3 clusters prior), and greater connectivity in Colorado (1 subpopulation center here vs. 2 clusters prior). These differences are most likely due to increased resolution and spatial continuity of sampling in this study but could be due to shifts in genetic structure over time, as ~20 years (6–10 generations) have passed since the original study.

While our results are similar to those of others conducted at the state level or larger scales, there are subtle differences which are likely due to differing study extents. This study incorporated the genotypes used in Fedy et al. [6] who examined population structure in Wyoming and Cross et al. [63] who did similar research in Montana and the Dakotas. In most of our analyses we documented two or three groups as opposed to the four found previously in Wyoming [6]. Importantly, this study documents a prominent division between northeastern and southwestern Wyoming representing a previously unrecognized east-west split separating all sage-grouse into two main groups (Fig 5, S2A, S4A and S7 Figs). This division was largely along the boundary between management zone (MZ)1 and MZ2 (Figs 1 and 3). In Montana and the Dakotas, we found the same three groups as Cross et al. [63], yet they found additional substructure in their hierarchical analysis some of which can be discerned in our STRUCTURE analysis at K = 12 (S4B Fig). This analysis of sage-grouse genetic variation range-wide, at high resolution, and with sampling homogeneity removes artifacts that may be introduced by irregular sampling or truncation of sampling extent at state and provincial boundaries rather than biological boundaries such as a species’ distribution.

Incorporation with other range-wide connectivity products

This research builds upon two recent studies that used the same data set to examine different aspects of connectivity. The first developed a genetic network for sage-grouse, characterized the network connectivity structure, and identified nodes serving as hubs of genetic connectivity such that these nodes could be prioritized based on their importance for maintaining network structure and connectivity [26]. The second evaluated functional connectivity by determining the specific factors driving genetic divergence within sage-grouse MZs, determining the landscape and habitat features that most impact gene flow within individual zones [27]. When individual resistance models were stitched together across MZs, a wide-ranging functional connectivity surface was produced [27]. This study takes the next step in evaluating connectivity by describing and quantifying large-scale patterns of population genetic structure and gene flow.

In the network study Cross et al. [26] found that breeding areas in the C. J. Strike Reservoir Watershed in southwestern Idaho, and in the Bighorn Lake and Upper Green-Slate watersheds in southwestern and north central Wyoming respectively, were important for maintaining genetic connectivity range-wide. These regions exhibited elevated effective migration rates (Fig 5), yet only one (C. J. Strike Reservoir) was within a center of subpopulation differentiation, indicating that areas outside of subpopulation centers are likely important for maintaining overall connectivity, while the subpopulation centers serve as generators of and reservoirs for genetic variation. Our map of effective migration was in general agreement with the functional connectivity map from Row et al. [27], yet the spatial patterns inferable from our map are much coarser as our effective migration map is based solely on effective migration rates (the pattern) and not based on the landscape variables (the process) driving that pattern. Nonetheless, our areas of greatest effective migration concur with those of greatest functional connectivity described in Row et al. [27]. This is of particular note in central Wyoming east of the Continental Divide where we found lower than expected migration rates and Row et al. [27] found constrained functional connectivity between MZs 1 and 2 that defined the east-west division of sage-grouse. In this same region, Cross et al. [26] also identified lek clusters essential to maintaining range-wide connectivity (hubs of connectivity identified in the top 1% of betweenness centrality). Our findings of lowered effective migration in Utah and Colorado also mirrored the work of Row et al. [27] who detected diminished functional connectivity as evidenced by fewer and/or more diffuse modeled gene flow.

Relating genetic structure to current management boundaries

The distribution and connectivity of sage-grouse have been subdivided previously based on biologically relevant criteria. For example, floristic zones were used to define the seven MZs [65] and the populations and subpopulations delineated by Connelly et al. [66] used a combination of habitat, lek locations, and the degree of presumed fragmentation. More recent work has defined hierarchical clusters based on seasonal habitat and sage-grouse movements [22–24]. While our genetic groups are consistent with some of these boundaries (e.g., MZ1 and MZ2) our findings suggest less population subdivision than suggested by the boundaries for MZs 3, 4, or 5 (our Great Basin genetic subpopulation center [Fig 5] spans all three of these MZs). Differences between the subdivisions described here and in other efforts may lie in the fact that genetic data measures functional movement: dispersal followed by reproduction, rather than structural connectivity based on habitat distribution and Euclidean distance. Several studies have examined whether sage-grouse males or females disperse farther with conflicting results, possibly due to study extent and the number of individuals marked [62, 67–69]. If males disperse farther than females and fail to reproduce, their movements are irrelevant to gene flow. As such, our connectivity assessment and characterization of subpopulation centers may differ from those based purely on habitat characteristics and movement data yet are important for management. As genetic variation and connectivity are vital for the ability to cope with current and future environmental stochasticity and anthropogenic alterations, understanding how genes flow across the landscape is a vital component of sage-grouse conservation.

Implications for sage-grouse management

Previous studies have highlighted the genetic distinctiveness in the Bi-State and the Washington populations suggesting that these groups may warrant being managed separately [39, 42], and we reaffirmed that here. We identified more subtle subdivision across the rest of the range of sage-grouse that was formerly unknown including a distinct east-west split generally along the Continental Divide. To a large extent most of our analyses recognized two groups in the east (Powder River and Great Plains) and at least four in the west (Wyoming Basin, greater Great Basin, Central Rockies, and Utah), which could be used to help define large scale management units. The 6 higher-order subpopulation centers of differentiation that resulted from the first three rounds of SIBP are core areas of differentiation that may be important to protect to maintain maximum genetic diversity across the range. The 6 finest scale subpopulation centers, occurring in the fragmented habitat in Utah, are areas with smaller numbers of isolated sage-grouse that could be monitored to ensure that genetic diversity in those areas is maintained or increased and to ensure that connectivity with other subpopulation centers is not lost. Additionally, the areas outside of subpopulation centers where different genetic groups converge (e.g., south-central Idaho and southwestern Wyoming) could be priorities for maintaining overall connectivity, especially where they overlap areas previously identified as important to sage-grouse connectivity [26, 27]. The data and analyses presented provide a baseline for monitoring future changes in sage-grouse connectivity and genetic diversity resulting from landscape changes.