Changes in genetic diversity and differentiation in Red‐cockaded woodpeckers (Dryobates borealis) over the past century

Abstract Red‐cockaded woodpeckers (RCW; Dryobates borealis) declined after human activities reduced their fire‐maintained pine ecosystem to <3% of its historical range in the southeastern United States and degraded remaining habitat. An estimated 1.6 million RCW cooperative breeding groups declined to about 3,500 groups with no more than 10,000 birds by 1978. Management has increased RCW population abundances since they were at their lowest in the 1990s. However, no range‐wide study has been undertaken since then to investigate the impacts of this massive bottleneck or infer the effects of conservation management and recent demographic recoveries. We used mitochondrial DNA sequences (mtDNA) and nine nuclear microsatellite loci to determine if range‐wide demographic declines resulted in changes to genetic structure and diversity in RCW by comparing samples collected before 1970 (mtDNA data only), between 1992 and 1995 (mtDNA and microsatellites), and between 2010 and 2014 (mtDNA and microsatellites). We show that genetic diversity has been lost as detected by a reduction in the number of mitochondrial haplotypes. This reduction was apparent in comparisons of pre‐1970 mtDNA data with data from the 1992–1995 and 2010–2014 time points, with no change between the latter two time points in mtDNA and microsatellite analyses. The mtDNA data also revealed increases in range‐wide genetic differentiation, with a genetically panmictic population present throughout the southeastern United States in the pre‐1970s data and subsequent development of genetic structure that has remained unchanged since the 1990s. Genetic structure was also uncovered with the microsatellite data, which like the mtDNA data showed little change between the 1992–1995 and 2010–2014 data sets. Temporal haplotype networks revealed a consistent, star‐like phylogeny, suggesting that despite the overall loss of haplotypes, no phylogenetically distinct mtDNA lineages were lost when the population declined. Our results may suggest that management during the last two decades has prevented additional losses of genetic diversity.


| INTRODUC TI ON
Most population genetic studies provide a snapshot of the current status of the species under investigation. Studies that adopt a historical perspective mainly do so using phylogenetic or phylogeographic concepts that provide insights about long-term historical changes that may have occurred over evolutionary time scales (Avise, 2000).
Advancements in molecular techniques have increased the feasibility of extraction and amplification of DNA from low yield and low-quality sources such as historical museum specimens (Leonard, 2008b). These techniques have important practical applications within conservation genetics as they permit direct assessments of changes that have occurred over time through comparisons between historical and contemporary samples (D'Elia, Haig, Mullins, & Miller, 2016;Draheim, Baird, & Haig, 2012). Temporal changes in genetic diversity are of interest as the genetic history of a species can assist managers in predicting responses to stochastic and anthropogenic demographic fluctuations. Likewise, monitoring contemporary genetic diversity is important for evaluating the effects of current management and determining where to focus future efforts (Schwartz, Luikart, & Waples, 2007).  (Allen, Krieger, Walters, & Collazo, 2006;Peet & Allard, 1993). The LLP ecosystem harbors some of the most species-rich communities in temperate North America (Mitchell, Hiers, O'Brien, Jack, & Engstrom, 2006;Peet & Allard, 1993) and is dependent on frequent natural fires every 1-10 years (Drew, Kirkman, & Gholson, 1998). Less than 3% of the estimated 24 million hectares of historical LLP ecosystem now remain as it went from a nearly continuous distribution across the southeastern coastal plains and adjacent areas to a highly fragmented condition due to habitat loss from timber cutting, other land use changes, and degradation of remaining habitat resulting from fire suppression (Allen et al., 2006;Kirkman & Jack, 2018;Mitchell et al., 2006). Based on historical accounts and the extent of habitat loss, it is clear that RCW underwent a massive population bottleneck between 1870 and 1930, which coupled with further declines driven by fire suppression from 1960 to 1980 resulted in extirpation of the species in the most northern regions of the species' range in Missouri, Maryland, Tennessee, and Kentucky (Conner, Rudolph, & Walters, 2001).
Red-cockaded woodpeckers are monogamous, territorial cooperative breeders where male (and less often female) offspring frequently stay and assist with incubation and feeding of nestlings and fledglings (Haig, Walters, & Plissner, 1994), and therefore also delay their own dispersal and breeding (Walters, 1991). Most female and some male juveniles disperse during their first year, and though RCW do not disperse very far, it is generally females that disperse farther (Daniels & Walters, 2000;Kesler & Walters, 2012;Walters, 1991). RCWs are unusual in that they excavate roost and nest cavities in old (i.e., >80-120 years old), living pine trees. The excavation process is complex and apparently involves fungi introduced to the excavation by the birds (Jusino, Lindner, Banik, Rose, & Walters, 2016;Jusino, Lindner, Banik, & Walters, 2015), and typically takes many years to complete (Harding & Walters, 2004). As a result, cavities and the availability of old pines suitable for cavities limit current populations. Also, RCW rarely form new breeding groups by excavating cavities for new territories, but rather most nonbreeding adult helper individuals wait to fill breeding vacancies on their already established territories that contain a set of completed cavities (termed cavity tree clusters; Walters, Copeyon, & Carter, 1992). This promotes population stability, but also constrains population size and rates of population growth (Walters, 1991). In response, techniques for constructing artificial cavities in living pine trees were developed (Allen, 1991;Copeyon, 1990) as a means to sustain existing territories with natural cavity limitations and to create new viable territories at recruitment clusters, a management technique that has been highly successful in increasing the numbers of breeding groups in a population (Conner et al., 2001;Walters, Robinson, Starnes, & Goodson, 1995). Additional management techniques include habitat restoration through prescribed burning and translocation of birds among populations. The first successful translocations of individual birds were conducted in 1986 at Savannah River Site (Franzreb, 1999;Haig, Belthoff, & Allen, 1993a). Since the mid-1990s, birds have been translocated annually from select larger donor populations to augment size and growth of smaller, isolated recipient populations (Costa & DeLotelle, 2006).
Historical RCW population numbers have been estimated at more than 1.6 million cooperative breeding groups (Conner et al., 2001) decreasing to 3,500 groups and <10,000 total birds by 1978 (Jackson, 1978) shortly after they became one of the first species protected under the U.S. Endangered Species Act. RCW continued to decline during the 1970s and 1980s (James, 1995;USFWS, 2003). By the early 2000s, RCW populations had increased to an estimated 5,627 active territories and approximately 14,000 birds (USFWS, 2003), indicating population stabilizations and increases in some populations due to new management programs (Costa, 2004;Rudolph, Conner, & Walters, 2004). Today, there are at least 7,800 active territories (W. McDearman, personal communication) in response to successful recovery management.

K E Y W O R D S
Dryobates borealis, endangered species, microsatellite, mitochondrial DNA sequences, Redcockaded woodpecker, temporal change We sampled range-wide from available RCW museum specimens, previously collected blood samples, and contemporary RCW populations to determine the extent that genetic diversity and structure have changed in RCW populations over the past century. The first and most recent range-wide population genetic studies of RCWs were conducted in the early 1990s when population abundances were at their lowest (Haig, Belthoff, & Allen, 1993b;Haig, Bowman, & Mullins, 1996;Haig, Rhymer, & Heckel, 1994;Stangel, Lennartz, & Smith, 1992). These studies applied allozyme, DNA fingerprints, and randomly amplified polymorphic DNA (RAPD) markers and found low genetic diversity, especially in smaller populations. However, intensive management efforts have been applied to this species for over two decades since then (Baker, 1999;Ferraro, McIntosh, & Ospina, 2007;Leonard 2008a), and mitochondrial and microsatellite markers developed for RCWs are now available (Alstad 2010;Fike, Athrey, Bowman, Leberg, & Rhodes, 2009). Thus, an updated assessment of genetic structure and diversity in RCWs may illustrate if genetic patterns have changed over time and provide insight into F I G U R E 1 Maps highlighting the spatial distribution of Red-cockaded Woodpecker sampling locations throughout the southeastern United States. Individual maps illustrate differences in sample sets for mtDNA versus microsatellite data sets as well as differences among pre-1970s data, data from 1992-1995, and data from 2010-2014. Open circles highlight areas where fewer than five individual samples were available, whereas filled circles show locations where data for five or more samples existed. Dotted outlines show the boundaries for three regional sampling groupings defined by locations of samples from the pre-1970s data or for comparable groupings of available sample locations from later years (see Supporting Information Appendix S1 for more information). Region 1: Western; Region 2: Eastern; Region 3: Florida.

| Sample collections
We assembled a collection of RCW blood and tissue samples to use as a source of DNA for this study (Figure 1). The samples encompassed three distinct time periods (Supporting Information Appendix S1), thereby permitting us to take a temporal perspective and identify changes in genetic structure and diversity over the past century. First, we acquired toepad tissue samples from 50 historical museum specimens located in 13 university and natural history

| DNA extraction
DNA extraction protocols varied depending on sample type (museum vs. blood vs. buccal swab samples). DNA was obtained from buccal swabs using the protocol outlined in Vilstrup et al. (2018).
Blood samples were extracted using the DNeasy Blood and Tissue kit (Qiagen, Inc.) with 10-20 µl blood as input and elution in 100 µl AE buffer preheated to 37°C prior to the elution spin. Museum samples were initially soaked in ddH 2 0 for 24 hr to remove potential inhibitors and then extracted using the same general protocol used for the buccal swabs, with the exception that tissue digestion occurred in a three-step process that included: (a) a preliminary overnight incubation with proteinase K as per the buccal swab protocol, (b) a manual grinding step using plastic mortars, and (c) an additional overnight incubation after addition of 10 µl Proteinase K. DNA extractions for the museum samples were performed in a separate laboratory from the blood and buccal swab samples.

| Mitochondrial DNA and microsatellite data
Three mitochondrial markers (Cytochrome b, Cytochrome oxidase subunit I, and the control region) were initially screened for a subset of samples spanning the species' range. We found very low genetic diversity at the Cytochrome b and COX1 marker, and therefore, chose to only sequence part of the hypervariable region of the con-  Table S1).

| Changes in genetic diversity over time
Data sets included in this study contained varying sample sizes and differences in the locations from which samples originated ( Figure 1 and Supporting Information Appendix S1). Thus, given our goal of identifying changes in genetic diversity over time, we performed a number of analyses using various partitions, groupings, and subsets of the data to help equalize data sets and make them as comparable as possible among different time points (see group definitions below). For the mtDNA data, we used Arlequin version 3.5 (Excoffier & Lischer, 2010) to quantify the observed number of haplotypes in each group (A mtDNA ), haplotype diversity (H), and nucleotide diversity (π). Although haplotype diversity and nucleotide diversity are unbiased estimators whose values are not affected by sample size, the observed number of haplotypes is potentially correlated with sample size (Kalinowski, 2004). Given that the pre-1970 and 1992-1995 data were derived from smaller numbers of samples relative to the 2010-2014 data, we used the program HP-Rare (Kalinowski, 2005) to obtain rarefied estimates of the number of haplotypes for each group (Ar mtDNA ) by assuming sample sizes associated with the smallest group of interest in a particular comparison (Tables 1-4).
Genetic diversity was quantified using three approaches for aggregating samples into groups to promote the ability to make comparisons among time points. First, the genetic diversity parameters described above from each time period were analyzed en mass to provide an overall sense for whether genetic diversity has changed among the pre-1970 data, the 1992-1994 data, and  (Table 1). Finally, the range of RCW spans 10 distinct level III ecoregions that define a variety of biomes across the southeastern United States (Omernick, 1995  pre-1970 mtDNA data (Figure 1), and for each of eight ecoregions (Supporting Information Appendix S5) where sample sizes were >5. In all microsatellite analyses, individuals were included only if genotypes (nonmissing data) were available at 5 or more out of the 9 loci.

| Changes in genetic structure over time
We compared the magnitude of genetic structure (F ST ) at differ- TA B L E 3 Microsatellite genetic diversity in Red-cockaded Woodpeckers across data sets and within three regions ( Figure 1; 1: Western, 2: Eastern, 3: Florida) at three different time points Arlequin 3.5 (Excoffier & Lischer, 2010). Analyses of mtDNA included information on molecular differences between individual haplotypes as quantified by the proportion of nucleotides that differ between DNA sequences. Because of differences in sampling locations and sample sizes associated with the three mtDNA data sets (pre-1970, 1992-1994, and 2010-2014) described above, differentiation was calculated for various comparable partitions of the data similar to our analyses of genetic diversity. For the mtDNA data, we quantified genetic differentiation for each data set among the three ad hoc regions (Figure 1), among collection areas where five or more samples were available (Figure 1), and among ecoregions where aggregated sample sets had sample sizes of 5 or more.
Microsatellite data sets (1992-1994 and 2010-2014) were analyzed using the locus-by-locus analysis option in Arlequin, resulting in a measure of differentiation comparable to Weir and Cockerham's (1984) unbiased estimator of F ST (Weir and Cockerham's θ). As with analyses of the mtDNA data, separate estimates of differentiation were quantified for the 3 ad hoc regions, ecoregions where aggregated samples had sample sizes of 5 or more, and sample locations where data from 5 or more individuals were available.
We used two additional approaches to test for changes in dif- as recommended by Falush, Stephens, and Pritchard (2003), using values of K (the assumed number of clusters) ranging from 1 to 5.
Five replicate analyses were performed for each value of K using an initial burn-in of 10 5 Markov-Chain Monte Carlo steps followed by recording for 10 6 steps. We calculated the average likelihood associated with runs from each value of K and assumed that values of K with the highest average likelihood score reflected the true number of genetic clusters.

| RE SULTS
We detected 66 unique mtDNA haplotypes across all individuals included in our data sets (Genbank Accession Numbers MK253579-MK253645). Our analyses suggested that mitochondrial haplotypes have been lost in RCWs over time. Across the complete data sets, no consistent changes in π or H were apparent, however, rarefied estimates of haplotype richness (Ar mtDNA ) that accounted for different sample sizes from each time period were lower in the 1992-1995 and 2010-2014 data sets relative to the pre-1970 data ( Table 1).
Analyses of the Western, Eastern, and Florida regions (Figure 1)   Table 2). The microsatellite data, which were restricted to information from 1992-1995 and 2010-2014, revealed no overt changes in genetic diversity patterns at any spatial scale during these time periods (Tables 3 and 4).
Based on the mtDNA data, quantitative measures of genetic differentiation were lower at all spatial scales for the pre-1970s data relative to data sets representing 1992-1995 and 2010-2014 (Table 1). Nonsignificant F ST values were calculated for the pre-1970 data, and point estimates were slightly negative reflecting the absence of genetic structure. By contrast, and regardless of the sample groupings used for analysis, estimates of F ST for the 1992-1995 and 2010-2014 mtDNA data were significantly larger than zero and relatively comparable between the two time periods (Table 5). The microsatellite data likewise illustrated the existence of genetic structure for the 1992-1995 and 2010-2014 data, with the magnitude of differentiation also relatively comparable between those two time periods (Table 5).
Temporal haplotype networks generated for the mtDNA data

| D ISCUSS I ON
The availability of historical museum samples for use in retrospective comparisons with contemporary samples allows researchers to determine the extent of genetic changes that occur within a species over time. However, the inability to plan the specific sampling locations of historical samples poses its own set of challenges.
Completely balanced data sets are unlikely, and it can be difficult to obtain historical data from all regions of interest (Draheim et al., 2012). However, in our study, we were able to use a number of different hierarchical spatial groupings of the samples that were collected at different time points (Figure 1 and Supporting Information Appendix S1) to help illustrate the degree that RCW genetic structure and diversity have changed over the last century.
Overall, our mtDNA data point to changes in genetic differentiation patterns and the existence of greater genetic diversity in RCW prior to the population declines that it experienced in the early to mid-20th century (Table 1). This loss of diversity is consistent with many other investigations that have documented a loss of alleles in declining species of management interest (Wandeler, Hoeck, & Keller, 2007)  The loss of diversity was only observed in our mtDNA data set because the microsatellite loci did not reliably amplify from the museum specimens that we examined (Tables 3 and 4 (Tympanuchus cupido) that were known to have been reduced to a population of ~50 individuals in 1993 (Bateson et al., 2014).
Likewise, comparisons of historical and contemporary specimens of the Mauritius Kestrel (Falco punctatus) identified a similar loss of microsatellite alleles (Groombridge, Jones, Bruford, & Nichols, 2000 provisioned with artificial cavities, sustain and increase habitat by compatible forest and prescribed fire management programs, and translocate individuals to bolster population sizes or re-establish extirpated populations throughout its range (Conner et al., 2001;Costa & DeLotelle, 2006;Kulhavy et al., 1995). Given that we have observed minimal changes in diversity parameters between the 1992-1995 and 2010-2014 time periods, it remains feasible that these management actions helped halt the loss of diversity detected in comparisons with the earlier time period (Tables 1,2). Indeed, in our analyses, the Western Region was characterized by a decline in the 1992-1995 mtDNA data followed by an apparent recovery to historical haplotype abundance levels in 2010-2014. We note that Region 1 includes the West Gulf Coastal Plain (WGCP) ecoregion (Supporting Information Figure S1), which also demonstrated an increase in mtDNA haplotypes between 1992-1995 and 2010-2014 when examined separately (Table 2) have been artificially moved throughout their range (Allen, Franzreb, & Escano, 1993;Carrie, Conner, Rudolph, & Carrie, 1999;Costa & DeLotelle, 2006;Cox & McCormick, 2016;Connor, Rodolph, & Bonner, 1995;Franzreb, 1999;Haig et al., 1993a;Rudolph, Conner, Carrie, & Schaefer, 1992). While more detailed analyses are still required, it remains feasible that the cumulative effects of translocations over the decades following their initiation may have reversed the loss of genetic diversity that was detected in this subset of the species' range.
Based on the mtDNA analysis, RCW have lost ~25%-30% of their haplotypes over the past century (Ar mtDNA from "All Data" row in Table 1). This loss did not appear to include any phylogenetically distinct lineages ( Figure 2) that might correspond to separate Evolutionarily Significant Units (Crandall, Bininda-Emonds, Mace, & Wayne, 2000;Moritz, 1994). Instead, the temporal haplotype networks created for each time period had relatively similar star-like topologies, and primary differences among the three networks could be attributed to the differences in sample sizes among time points (i.e., more haplotypes revealed in the 2010-2014 data due to a ~fivefold larger number of analyzed individuals).
Prior analyses from the 1990s highlighted significant genetic structure among RCW populations. Stangel et al. (1992) estimated an F st of 0.14 in allozyme analyses of 26 populations while Haig, Rhymer, et al. (1994) estimated an F st of 0.19 using RAPD markers from 14 populations. Haig et al. (1996) later revisited the previous RAPD analyses and included additional populations from Florida, resulting in an F st of 0.21 that was comparable to that detected in the earlier investigation (Haig, Rhymer, et al., 1994). Our analyses of new microsatellite and mtDNA data based on samples from a similar time frame (samples from 1992-1995) produced similar results and illustrated significant genetic structure at multiple different spatial scales (Table 5). However, our use of data from more contemporary (2010-2014) and historical (pre-1970) samples provided new insights that were not apparent from the single time point snapshots achieved in prior studies. Specifically, the mtDNA data show that changes in genetic structure have occurred in concert with the loss of genetic diversity that occurred prior to the 1990s (Table 5).
Historically, RCW appears to have had continuous populations and panmictic genetic structure based on the negative point estimates of F ST that were identified in the pre-1970s mtDNA data. This widespread panmictic population has transitioned more recently into a discontinuously distributed species with isolated populations and reduced gene flow based on the low, but significant, population dif- Our study is among a growing list of investigations that use historical DNA samples to identify genetic changes that have occurred in natural populations. These approaches are powerful, as they may provide insights not only about trends over time, but can also help determine if management actions are having the desired effects at the genetic level (Frankham et al., 2017;Haig et al., 2011;Schwartz et al., 2007). At this time, there is a need to better determine the effects of RCW management actions beyond the demographic impacts that have been previously documented. In future investigations, closer examination of over 20 years of translocation data, habitat information, and other demographic factors may provide a better understanding of recent genetic changes that have occurred in RCW. Such analyses may help identify secondary repercussions of management actions beyond the beneficial demographic impact that has been previously documented.

ACK N OWLED G M ENTS
We are grateful to the museums that provided toe pad clippings: The History (Tamaki Yuri).. We also appreciate the many biologists who swabbed nestlings for us throughout the range of the Red-cockaded