Evaluating wildlife translocations using genomics: A bighorn sheep case study

Abstract Wildlife restoration often involves translocation efforts to reintroduce species and supplement small, fragmented populations. We examined the genomic consequences of bighorn sheep (Ovis canadensis) translocations and population isolation to enhance understanding of evolutionary processes that affect population genetics and inform future restoration strategies. We conducted a population genomic analysis of 511 bighorn sheep from 17 areas, including native and reintroduced populations that received 0–10 translocations. Using the Illumina High Density Ovine array, we generated datasets of 6,155 to 33,289 single nucleotide polymorphisms and completed clustering, population tree, and kinship analyses. Our analyses determined that natural gene flow did not occur between most populations, including two pairs of native herds that had past connectivity. We synthesized genomic evidence across analyses to evaluate 24 different translocation events and detected eight successful reintroductions (i.e., lack of signal for recolonization from nearby populations) and five successful augmentations (i.e., reproductive success of translocated individuals) based on genetic similarity with the source populations. A single native population founded six of the reintroduced herds, suggesting that environmental conditions did not need to match for populations to persist following reintroduction. Augmentations consisting of 18–57 animals including males and females succeeded, whereas augmentations of two males did not result in a detectable genetic signature. Our results provide insight on genomic distinctiveness of native and reintroduced herds, information on the relative success of reintroduction and augmentation efforts and their associated attributes, and guidance to enhance genetic contribution of augmentations and reintroductions to aid in bighorn sheep restoration.


| INTRODUC TI ON
Population restoration using translocation, which involves moving live individuals from one area to another, is an important tool for restoring biodiversity (IUCN/SSC, 2013;Weeks et al., 2011).
Reintroduction starts a new population within formerly occupied landscapes, whereas augmentation adds individuals to an indigenous or reintroduced population (IUCN/SSC, 2013). The goal of augmenting a small, genetically isolated population is frequently to enhance viability by increasing the number of individuals in certain demographic groups or increasing genetic diversity to improve reproductive fitness, termed genetic rescue (IUCN/SSC, 2013;Tallmon et al., 2004).
Successful survival and breeding of translocated animals can depend on many factors, including environmental similarity between source and release areas in case of local adaptation, reproductive attributes of the species, habitat quality of the target area, distance from other conspecific populations, number of individuals moved, and management program duration (Bleich et al., 2018;Griffith et al., 1989;Groombridge et al., 2012;Magdalena Wolf et al., 1998).
Managers can control some of the factors that influence survival and breeding of translocated animals to enhance the probability of translocation success. However, the relative survival and reproduction of translocated individuals in augmentations and reintroductions and the long-term genetic effects of translocation efforts may vary widely by species and population of interest. Thus, it is beneficial to enhance understanding of the multi-generational genetic effects of reintroduction and augmentation efforts in wild populations (Moraes et al., 2017;White et al., 2018).
While reintroductions can successfully establish a new wildlife population, previous studies have found that some populations thought to be the result of reintroduction efforts were in fact the result of recolonization (Kruckenhauser & Pinsker, 2004;Statham et al., 2012;Stewart et al., 2017). In this case, animals from nearby areas naturally dispersed and established a population in previously occupied terrain, meaning that costly reintroduction efforts may not have been necessary. Further, some studies have suggested that matching environmental attributes between the source population and the reintroduction area is important to establishment of translocated animals, whereas others have found that sourcing from a native, rather than reintroduced, population is more important (Malaney et al., 2015;Olson et al., 2013). Thus, evaluating the genetic source of populations thought to be reintroduced can provide insight into whether similar translocation efforts would be productive in other areas or whether enhancement of habitat connectivity would be more important to encourage recolonization (Stewart et al., 2017). Further, evaluating the genetic differences between populations that were the result of reintroduction events and their founding source can teach us about the many influences on the reintroduced population's evolution (Jamieson, 2011;White et al., 2018). For example, the literature frequently recommends large founder populations for reintroductions to represent the genetic profile of the source and minimize inbreeding (Jamieson & Lacy, 2012). However, the necessary founder size for reintroductions may vary widely by species, population, and reintroduction area of interest. Genetic evaluation of reintroduced populations can help address these uncertainties in reintroduction planning.
In addition, augmentation of existing populations may not result in genetic contribution to the target recipient population. Previous studies have found that the sex of animals moved and species mating strategy can influence whether translocated animals breed with the resident population (Miller et al., 2009;Mulder et al., 2017;Sigg et al., 2005). For example, no translocated males contributed to a resident desert tortoise (Gopherus agassizii) population; resident and translocated females only produced progeny with resident males (Mulder et al., 2017). Thus, determining what factors influence animal breeding following augmentation can help biologists avoid implementation of costly augmentation efforts that fail to result in genetic contribution (Armstrong & Seddon, 2008;Fischer & Lindenmayer, 2000). In addition, evaluating unassisted genetic connectivity among populations can be used to assess whether a population needs a translocation. For example, dispersal was greater than expected among reintroduced elk (Cervus elaphus) populations originally thought to be genetically isolated (Hicks et al., 2007). If it is determined that dispersal provides sufficient genetic connectivity to certain populations, wildlife managers could devote resources to augmentation efforts for other populations that are genetically isolated.
After a translocation event, selection, genetic drift, unassisted gene flow, and mutation can influence population evolution on different timescales, making it difficult to identify relative influences on population viability . Thus, assessing how translocations affected multiple isolated populations with contrasting translocation histories and situations could help us understand the variation in population evolution (Moraes et al., 2017;White et al., 2018). Genomic techniques can be used to evaluate the success of past translocations and plan future efforts (Bell et al., 2019).
Specifically, genomic analyses can help evaluate the genetic effects of previous translocations and assess mean kinship to inform new translocation efforts Jahner et al., 2018).
Additionally, comparing long-term genomic effects of translocations in multiple wild populations would further our understanding of this conservation tool and inform future strategies for genetic management of species with fragmented distributions and genetically isolated populations.
Bighorn sheep (Ovis canadensis) in North America continue to face challenges found in many other fragmented wildlife populations, such as low recruitment, poor population growth rates, and widespread disease issues resulting in periodic die-offs that reduce populations an average of 48% with subsequent prolonged periods of poor lamb survival (Cassirer et al., 2013(Cassirer et al., , 2018Manlove et al., 2016;. Prior to European expansion across the American west in the late 19th and early 20th centuries, there were an estimated 1.5 to two million bighorn sheep (Seton, 1929).
However, market hunting, competition with domestic sheep, and new respiratory diseases introduced by comingling domestic sheep and goats resulted in a decline by 1960 to fewer than 20,000 bighorn sheep in scattered patches across North America (Buechner, 1960;Cassirer et al., 2018;Norris, 1877;Valdez & Krausman, 1999;Whittlesey et al., 2018aWhittlesey et al., , 2018b. Continued concerns regarding resident pathogens in bighorn sheep populations and disease spillover from domestic sheep that may cause epizootic events have resulted in management to keep populations small and isolated (Butler et al., 2018;Cassirer et al., 2016).
Translocation has been the primary management tool used to reintroduce bighorn sheep to previously occupied areas and augment the size and genetic diversity of existing populations. Over 1,460 translocations of 21,500 bighorn sheep have taken place across the indigenous range of this species, and some populations have been studied as examples for genetic rescue (Hogg et al., 2006;Olson et al., 2012;Poirier et al., 2019;Wild Sheep Working Group, 2015).
Despite these efforts, restoration of the species continues to be a challenge, as there are still large areas of unoccupied historic range and many small populations with poor demographic performance populations managed and expected to be isolated have genetic contributions due to natural dispersal. Secondly, we evaluated genetic differences between reintroduced herds and their documented founding source to determine (a) whether the population was established via translocation or recolonization and (b) what factors influenced reintroduction success (i.e., lack of signal for recolonization), such as environmental attributes or origin of the source herd. Thirdly, we evaluated genetic differences between reintroduced herds and their founding source to compare alternative hypotheses about which factors influenced the evolution of herds after reintroduction, including founder population size and unassisted/assisted gene flow between herds. Finally, we evaluated relative contributions of past augmentations (i.e., reproductive success of translocated individuals) to determine what factors influenced augmentation success, including the number and sex of translocated animals. We synthesized this information to inform risk and benefit assessments of future augmentations and reintroductions. The results of this study provide insights into long-term genomic consequences of different translocation approaches used for species restoration and may help inform future genetic management of bighorn sheep. We sought to sample at least 20-25 individuals per population, based on sample size simulations that determined sampling less than this number would introduce an unacceptable level of uncertainty to estimates of genomic kinship between populations (Flesch et al., 2018). We evaluated seven native herds, including one population of 80-90 animals with no augmentations (Galton), three small to moderately sized populations (80-200 animals) with augmentation attempts (Spanish Peaks, Taylor Hilgard, and Stillwater), and three large, continuous populations (380-3,800 animals), including Beartooth-Absaroka, Castle Reef, and Glacier (Table S1). Castle Reef is a geographic portion of a large, spatially structured population (a collection of subpopulations that occupy distinct geographic areas but are linked by animal movement) and is expected to have connectivity across four administrative units We also evaluated nine reintroduced herds. For eight of those herds, one or all of the initial founding sources were included in this study ( Figure 2; Table S2). We considered translocations within 3 years of the first reintroduction event to an unoccupied area to be part of the potential founding source for all populations except for Wild Horse Island. For this population, we considered two translocations that were 8 years apart as founding events, as the first reintroduction event consisted of only two animals. Founder size in reintroduced populations ranged from eight to 53 bighorn sheep.

| Study populations
Based on a generation length of 6 years, estimated using the mean age of reproductively active females, there were five to 11 generations since establishment of the reintroduced study populations (Hogg et al., 2006;Johnson et al., 2011)  serve as a same-species outgroup, but later learned that some translocations from the previously mentioned Whiskey Mountain herd occurred in this region.
As a true outgroup, we evaluated a different subspecies, Sierra Nevada bighorn sheep (Ovis canadensis sierrae; Buchalski et al., 2016;Wehausen et al., 2005). We included five outgroup samples following a similar outgroup approach by Sim et al. (2016). Three Sierra Nevada samples originated from the native Sawmill population, and two samples came from the reintroduced Wheeler population, which was founded by Sawmill individuals (Table S2). We gathered records for all known translocations received by study populations, including translocations that originated from areas not included in the study (Table S2). Where both the source and recipient populations were in this study, approximately zero to eight generations occurred between augmentation and genetic sampling.

| Genomic dataset and quality control
We genotyped bighorn sheep samples using the Illumina High Density (HD) Ovine array, also referred to as a SNP (single nucleotide polymorphism) chip. For genotyping, we used samples with a minimum of 300 ng of DNA, a minimum DNA concentration of 20 ng/µl, and a 260 nm/280 nm ratio of 1.0 to 1.5. The Ovine array is composed of 606,006 SNPs and was originally developed for domestic sheep with a density of one SNP per 4.279 kb, but its development included five bighorn sheep and four Dall's sheep (Ovis dalli; Kijas et al., 2009Kijas et al., , 2014. Speciation between domestic and bighorn sheep occurred about three million years ago, but the two species can interbreed to produce viable hybrid offspring (Bunch et al., 2006;Young & Manville, 1960). Domestic and bighorn sheep have the same number of chromosomes and are expected to have high genomic synteny (Poissant et al., 2010). An estimated 24,000 SNPs in the HD Ovine array are informative for Rocky Mountain bighorn sheep, and the domestic sheep reference genome enables mapping SNPs to chromosomes (Kohn et al., 2006;Miller et al., 2015).  Team, 2017;RStudio Team, 2015). For kinship calculations, we filtered SNPs using a minor allele frequency of less than 0.0001 to remove monomorphic and extremely rare markers, and we removed markers with poor performance by requiring a SNP call rate greater than 0.99 (de Cara et al., 2013;Huisman et al., 2016). The dataset used for the kinship analysis did not require additional filtering of low frequency SNPs prior to analysis based on KING software guidelines (Manichaikul et al., 2010). To infer population structure and ancestry, we further filtered the dataset using a minor allele frequency of 0.01 and a Hardy-Weinberg equilibrium p-value of less than .00001 (Huisman et al., 2016). We used linkage disequilibrium (LD) pruning to remove nonindependent SNPs that informed the presence of nearby variants, using a window size of 100 SNPs, window increment of 25 SNPs, and an LD statistic of r 2 (Huisman et al., 2016).
Code used for filtering and analyses is in Appendix S1.

| Population structure, ancestry, and kinship analyses
Because we did not know the degree of genetic similarity among  Pritchard et al., 2000;Raj et al., 2014). This analysis uses clustering to estimate the number of distinct genetic groups and employed a probabilistic analysis to assign individuals to one or more genetic populations without reference to the sampling locations. The STRUCTURE approach assumes that there are a certain number of genetic populations that contain random mating and that these distinct groups have different allele frequencies. Thus, the fast-Structure model can provide useful information regarding if animals moved in a translocation successfully bred with the resident population, as global ancestry results can detect hybrids of multiple genetic populations. However, the approach has lower accuracy with uneven sampling and assumes random mating, which is frequently inaccurate for wild populations (Alexander et al., 2009;Frankham et al., 2017;Puechmaille, 2016). In addition, while STRUCTURE models can evaluate admixture, they lack a temporal assessment of fragmentation .
Thus, we also completed a nested multidimensional scaling (MDS) analysis using KING v2.1.4 (Manichaikul et al., 2010). In contrast to fastStructure, MDS can identify clusters of populations without assumptions regarding Hardy-Weinberg equilibrium, random mating, or the cause of population structure (i.e., isolation-by-distance). This is because MDS does not assign individuals to a genetic population before or after the analysis. Instead, MDS uses an unsupervised approach to reduce the dimensionality of the genomic dataset to an interpretable plot, which allows for visualization of patterns of genomic variation and identification of separation among individual samples to address research questions regarding genetic similarity between individuals and populations. We used a nested approach to understand substructure across multiple levels of organization.
To determine the lineage of reintroduced populations in comparison to their hypothesized founding source, we estimated a bifurcating tree that described differentiation among populations. Because we did not expect most of the evaluated bighorn sheep populations to have dispersal of breeding animals between them, we assigned population identity of individuals based on sampling location. To model differentiation due to genetic drift among these populations defined by geography while still accounting for gene flow, we estimated a maximum likelihood bifurcating tree of populations using Treemix v1.13 (Pickrell & Pritchard, 2012). Treemix uses allele frequencies from each population and Gaussian approximation for genetic drift to estimate a tree that represents each population on a branch (Pickrell & Pritchard, 2012). The amount of genetic drift that occurred among populations is represented by a drift parameter calculated by the Treemix software (Pickrell & Pritchard, 2012). Thus, we could use this analysis to determine the lineage of reintroduced populations, in comparison to their hypothesized founding source.
Possible admixture (gene flow) events between branches of the tree were evaluated using a stepwise likelihood approach, where the software searched the tree for the optimal location of each translocation or dispersal event (Pickrell & Pritchard, 2012). To further specifically evaluate gene flow between predefined populations, we conducted a three-population test (Pickrell & Pritchard, 2012;Reich et al., 2009). A negative value of the f 3 statistic produced by the three-population test suggests that a specified population is admixed (Pickrell & Pritchard, 2012;Reich et al., 2009). Thus, we used these analyses to specifically identify previously unknown admixture events due to dispersal and confirm genetic contribution of augmentation events.
Finally, we estimated mean kinship between populations to evaluate genetic differences in reintroduced herds from their documented founding source. This information can also inform future augmentation decisions by identifying potential source and recipient populations based on their level of genetic similarity (Ballou & Lacy, 1995;Frankham et al., 2017). Kinship, also called coancestry, represents the probability that two randomly sampled alleles from two individuals are identical by descent (Manichaikul et al., 2010).
Mean kinship calculated between populations serves as a measure of population similarity, with higher values interpreted as populations that are more related . Thus, genetic differentiation between populations is represented by one minus mean kinship . We estimated mean kinship between populations using KING v2.1.4 (Manichaikul et al., 2010). To assess how characteristics of translocations influenced the current kinship between reintroduced populations and their founding source, we evaluated six herd attributes that may affect divergence of reintroduced populations using boxplots.
These attributes included founder population size, number of generations since herd establishment, number of augmentations from the founding source, number of source populations, number of augmentations from other sources, and level of connectivity with neighboring herds. See Methods S1 for details regarding described analyses.

| Genomic dataset and quality control
We genotyped 541 bighorn sheep samples using the HD Ovine array, resulting in a dataset composed of 606,006 SNPs. Using a sample call rate of 0.85, we filtered 30 samples from the dataset and subsequently evaluated 511 samples from 17 different populations. We met our sample size goal of at least 20-25 bighorn sheep for fourteen Rocky Mountain bighorn sheep populations, excluding Highlands and Galton (Table 1). We included Highlands and Galton populations in the MDS and fastStructure analyses, where overall sample size was less likely to affect the results, but excluded these two low sample size populations from the Treemix analysis and interpreted their mean kinship results with caution (Highlands n = 17; Galton n = 5). After filtering, 33,289 SNPs were used for kinship estimates; 6,155 SNPs were used for the remaining analyses (Results S1). The dataset used for the kinship analysis did not require additional filtering of low frequency SNPs prior to analysis based on KING software guidelines (Manichaikul et al., 2010).

| Population structure
To evaluate population structure, we estimated the number of ge- Standard deviation from the mean is in parentheses. Smaller values indicate lower mean kinship. Sample size for each population is next to each population name on the y-axis; populations with fewer than 20 genotypes are labeled in red   Creek formed its own cluster, and in the K = 9 analysis, Lost Creek formed its own cluster, which may be due to past augmentation from sources not included in this study ( Figure S1; Table S2).
We detected past augmentations between genetic groups, where the recipient population contained partial ancestry from the cluster of the source population and the populations were geographically distant, such that movement of breeding animals between the areas was unlikely. We detected the following translocation events, indicating that translocated individuals survived and reproduced successfully at the release site: Wild Horse Island to Taylor Hilgard, Lost Creek to Taylor Hilgard, Wild Horse Island to Tendoys, and Castle Reef to Tendoys (Figure 2; Figure S1). The success of an augmentation from Castle Reef to Spanish Peaks was unclear, as we detected genetic ancestry from the cluster including Castle Reef in fewer than five individuals in Spanish Peaks ( Figure S1). We also detected augmentations from populations that were not directly sampled but were also augmentation sources to herds in our study. The Whiskey

| Population tree and genetic contributions from augmentations
We used Treemix to evaluate the lineages of populations defined by geography using a bifurcating tree that accounted for gene flow.
The population tree generated by Treemix was consistent with fast-Structure, as the nodes on the tree generally grouped together populations found in the same fastStructure clusters (Figure 4). Sierra Nevada was defined as the outgroup, and detailed information about Treemix model selection can be found in Results S1. Geographically The Treemix model identified four augmentation events between specific populations (Figure 4, orange lines; Table 2). All four identified gene flow events represented known augmentations where shared ancestry between populations was identified by fast-Structure (Tables 2 and S2). The direction of the plotted augmentation event was the least stable feature of the Treemix analysis, and we reversed the direction of augmentation events identified 3rd and 4th, as the direction was known and unassisted gene flow was unlikely due to geographic separation of the identified areas (Tables 2   and S2). Translocation weight estimated the proportion of alleles contributed by the source population, assuming admixture occurred in one generation (Pickrell & Pritchard, 2012 (Table S3). In addition, the three-population test suggested unassisted gene flow (natural movement of breeding animals) between Stillwater and the Beartooth-Absaroka, consistent with Figure 3a.

| Comparing translocation history and genomic analyses
To identify which reintroduction and augmentation efforts made a genetic contribution to the recipient population, we synthesized genomic evidence from fastStructure, MDS, and Treemix for 24 different translocation events where both populations were included in the study (

| Genetic similarity
We estimated mean kinship between bighorn sheep populations to evaluate genetic similarity and inform future augmentation efforts based on mean kinship (Table 1)  Note: Translocation weight can range from 0 to 0.5 and estimated the proportion of alleles contributed by the source population.

TA B L E 2 Translocation events
identified by Treemix based on the population tree (Figure 4) TA B L E 3 Evidence for all known translocations between herds in this study detected using MDS (multidimensional scaling), fastStructure, and Treemix analyses Note: An asterisk after the total number of bighorn sheep indicates that biologists suspected the translocation failed and did not contribute to the receiving population. For the MDS analysis (Figure 3), a translocation detection was designated as "possible" if both herds were found in the same group that was evaluated in a subset MDS analysis and "not detected" if herds were not evaluated together in a subset MDS analysis. For fastStructure (Figure 2), a translocation was considered "detected" if a common cluster was found for more than five animals in either herd, "possible" if a common cluster was found for less than five animals in either herd or there was an alternative source herd within the same cluster as the herd providing animals, and "not detected" if a common cluster was not found in both herds. For Treemix (Figure 4), a translocation was considered "detected" if identified as a migration event by the software, "possible" if herds were plotted on nearby branches in the tree, and "not detected" if herds were not plotted on nearby branches in the tree.
We estimated mean kinship between reintroduced herds and their founding source to evaluate six attributes that could affect population evolution since reintroduction ( Figure S4). These attri-

| D ISCUSS I ON
We examined the genomic consequences of bighorn sheep restoration to enhance understanding of evolutionary processes such as gene flow from genetic connectivity or genetic drift from isolation that affect population genetics and to inform genetic management of fragmented populations through future translocation strategies.
First, our genetic structure results indicated genetic connectivity due to natural dispersal of individuals within large, spatially struc-

| Natural genetic connectivity detected only in large populations
Genetic structure and connectivity across native populations prior to fragmentation due to human activities can serve as a baseline goal for bighorn sheep restoration in other areas. Glacier and Beartooth-Absaroka served as examples of large, continuous native populations, and our results suggested genetic connectivity within these populations and that gene flow within these herds was influenced by geographic distance and at least one natural barrier. Genetic differences between the north and south portions of Glacier suggested partial fragmentation due to a large, central lake that may serve as a barrier to extensive gene flow (Figures 2 and 3). In addition, our results suggested isolation-by-distance within Stillwater and Beartooth-Absaroka, as results for Stillwater, Yellowstone, northern hunt units 1-3, and southern hunt units 5 and 22 suggested small genetic differences within a larger population (Figures 2 and 3). Love Stowell et al. (2020) (Boyce et al., 1999;Rubin et al., 1998), rams often disperse during breeding season to genetically connect ewe groups (Geist, 1971). Our genotyping approach accounted for male genetic contributions. In addition, for most populations we sampled animals across a broad distribution via helicopter search and capture. Thus, we expect that our results represent genetic isolation among examined herds, rather than genetic subgroups within a larger population. We detected two pairs of native herds that had past genetic connectivity but were likely not connected at the time of sampling, in-

| Translocation, not recolonization, started all populations thought to be reintroduced
To evaluate evolution of newly founded populations and inform reintroduction planning, we compared herds thought to be reintroduced with their suspected founding source. Other studies suggested that bighorn sheep reintroduction efforts may be more successful when founders are sourced from either matching environmental conditions, due to greater recruitment when ecotypes are matched, or native populations, due to typically higher levels of genetic diversity (Bleich et al., 2018;Fitzsimmons et al., 1997;.
We evaluated eight populations that originated from reintroductions where a founding source was in our study. Castle Reef was the source of six out of eight reintroduced herds, and the remaining two were started by reintroduced populations initially founded by Castle Reef animals that were translocated (i.e., moved by managers to an formerly occupied area), rather than recolonization (i.e., founded by natural dispersal from nearby populations). This result was suggested by each of our three main analyses. In the fastStructure analysis, six of the reintroduced herds shared the same genetic cluster as Castle Reef (Figure 2). In the MDS analysis, eight reintroduced herds clustered with Castle Reef (Figure 3c). In the population tree, the same reintroduced herds that had sufficient sample size to be included  (Malaney et al., 2015;Rominger et al., 2004;Wiedmann & Sargeant, 2014). For example, translocated bighorn sheep can adjust the timing of parturition to match the environmental characteristics of a new area (Whiting et al., 2011(Whiting et al., , 2012. Local adaptation in bighorn sheep herds is still possible (Wiedmann & Sargeant, 2014), given that native populations found in different ecological regions were genetically differentiated, but genetic distinctiveness could also be explained by genetic drift (Figure 2). Future research to evaluate the possibility for local adaptation would involve assessing patterns of correlation between individual genotypes and environmental characteristics across space, for which there are many analytical approaches (Balkenhol et al., 2017;Rellstab et al., 2015;Selmoni et al., 2019). If ecologically matched native herds have low genetic diversity or are not available for a reintroduction, multiple sources could be used, as maximizing genetic diversity might increase adaptive potential in a new environment (Broadhurst et al., 2008;Olson et al., 2013;White et al., 2018).

| Genetic divergence of reintroduced populations from their founding source was mainly influenced by augmentations
After selecting source population(s) for reintroduction, geneticists recommend using a large number of founders to replicate the genetic profile of the founding source, but recommendations can range from 20 to 50 animals and depend on the species (Jamieson & Lacy, 2012;Taylor & Jamieson, 2008;Weeks et al., 2011). It is useful to evaluate how much reintroduced populations genetically diverged from their founding source, to assess whether the released number of founders successfully represented the source's genetic profile, and to determine whether genetic drift occurred, which is the main process by which small populations lose genetic variation, as in reintroduced populations of Alpine ibex (Capra ibex ibex; Biebach & Keller, 2009;Lacy, 1987;Templeton, 2006). Thus, greater genetic divergence from the source population may indicate a loss of genetic variation in the reintroduced herd due to chance. We expected

| Augmentations of two males were less successful than larger, mixed sex groups
Augmentations provide an opportunity to assess the impact of spe- Identifying which animals will contribute genetically to a recipient population includes considerations such as adaptation, number, sex, age, and disease status. To synthesize results across the fastStructure, MDS, and Treemix analyses, we defined a translocation event as detected only if we found evidence for its genetic contribution in the intended recipient population in at least two out of three analyses (  (Geist, 1971) and potentially depart the augmentation destination, which may result in no genetic contribution to the intended recipient population. In addition, because bighorn sheep are a polygynous species, a small number of dominant rams may competitively exclude translocated males due to female mate preference for residents or poor condition after transport/release, suggesting that translocating a greater proportion of females may be more effective for augmentation (Mulder et al., 2017;Sigg et al., 2005). After an augmentation of females, translocated ewe groups may socially segregate from the resident population following release, which was observed in Taylor  Augmentations are often promoted for genetic rescue (Hogg et al., 2006;Tallmon et al., 2004;Whiteley et al., 2015). To increase effectiveness of a genetic rescue, source and recipient populations should have been previously connected but recently isolated to allow differentiation over multiple generations in the past 500 generations (Allendorf & Luikart, 2009;Falconer et al., 1996;Frankham et al., 2017). Many types of information should be evaluated to determine optimal sources for augmentation. Our results can provide guidance on selecting sources for genetic rescue augmentations by combining information on genetic differentiation and mean kinship.
One possible approach, if managers want to maintain populations that are currently differentiated, would be to identify possible sources for future augmentations within clusters of populations in MDS or fastStructure analyses (Figures 2 and 3). Within identified clusters, managers can select a source population that has low mean kinship with the intended recipient herd (Table 1). Minimizing mean kinship between source and recipient populations within identified clusters would be an approach to retain genetic diversity and minimize inbreeding at the population level while still considering the possibility for local adaptation (Ballou & Lacy, 1995;Frankham et al., 2017). For example, augmentations could be implemented within the cluster associated with Castle Reef (Figure 2), and mean kinship minimized between source and recipient by consulting a mean kinship table associated with that cluster (Table S4).

| CON CLUS I ON S AND MANAG EMENT IMPLIC ATIONS
Our

ACK N OWLED G M ENTS
We thank the wildlife agency staff who contributed to this work.

CO N FLI C T O F I NTE R E S T
None declared.

DATA AVA I L A B I L I T Y S TAT E M E N T
Genotypes are posted on Figshare, at http://doi.org/10.6084/ m9.figsh are.10394711. Code is in Appendix S1.