Benefits of gene flow are mediated by individual variability in self‐compatibility in small isolated populations of an endemic plant species

Abstract Many rare and endemic species experience increased rates of self‐fertilization and mating among close relatives as a consequence of existing in small populations within isolated habitat patches. Variability in self‐compatibility among individuals within populations may reflect adaptation to local demography and genetic architecture, inbreeding, or drift. We use experimental hand‐pollinations under natural field conditions to assess the effects of gene flow in 21 populations of the central Appalachian endemic Trifolium virginicum that varied in population size and degree of isolation. We quantified the effects of distance from pollen source on pollination success and fruit set. Rates of self‐compatibility varied dramatically among maternal plants, ranging from 0% to 100%. This variation was unrelated to population size or degree of isolation. Nearly continuous variation in the success of selfing and near‐cross‐matings via hand pollination suggests that T. virginicum expresses pseudo‐self‐fertility, whereby plants carrying the same S‐allele mate successfully by altering the self‐incompatibility reaction. However, outcrossing among populations produced significantly higher fruit set than within populations, an indication of drift load. These results are consistent with strong selection acting to break down self‐incompatibility in these small populations and/or early‐acting inbreeding depression expressed upon selfing.

by increased isolation that eliminates gene flow among previously connected populations, increased selfing and mating among close relatives can reveal substantial genetic load that was masked in larger populations (Keller & Waller, 2002). The consequences of more frequent inbreeding within populations may be indicated by low fruit set or low-quality seed (early-acting inbreeding depression), as well as poor survival or reproduction of inbred progeny (late-acting inbreeding depression).
Benefits of reproductive assurance in small and isolated populations appear to outweigh the benefits of cross-pollination that are found in large populations (Delmas, Cheptou, Escarvage, & Pornon, 2014;Herlihy & Eckert, 2002;Holsinger, 2000;Igić, Bohs, & Kohn, 2006;Kalisz et al., 2004;Porcher & Lande, 2005). Shifts in the relative proportions of selfing and outcrossing within populations are the result of mating system evolution and the diversity of mating systems in plants is an indicator of the flexibility in responding to selection (Levin, 2012).
Evidence is increasing that the mating system itself may respond adaptively to small population size and fragmentation through breakdown of self-incompatibility (Busch, Joly, & Schoen, 2010;Karron et al., 2012;Stephenson, Good, & Vogler, 2000;Willi, 2009). Levin (1996) suggested this breakdown is often due to the action of modifier genes that alter the effectiveness of self-incompatibility alleles (i.e., pseudoself-fertility, PSF), a critical step in the evolution of self-fertility.
Gene flow among populations can alleviate inbreeding effects by introducing variation from relatively unrelated individuals that masks genetic load and restores compatible mating types (Cheptou & Donohue, 2011;Frankham, 2015;Spielman, Brook, & Frankham, 2004;Young & Pickup, 2010). The interaction between a species' dispersal ability and the distribution of habitat in a landscape determines patterns of gene flow under natural conditions. Long-term patterns of gene flow may be very different than current gene flow if habitat patches are smaller (fewer potential migrants) and are further apart (requiring longer dispersal distances) than they were under historical habitat distributions (Honnay & Jacquemyn, 2007).
Because mating system and gene flow are key to assessing the risks associated with small population size, the effects of crossing distance between individuals within and among populations have long been of interest to evolutionary biologists and conservation biologists (Edmands, 2007;Fenster & Sork, 1988;Frankham et al., 2011;Marsden, Engelhardt, & Neel, 2013;Weeks et al., 2011;Whitlock et al., 2013). Field studies linking the reproductive biology and mating system of species with an ecologically relevant scale of crossing distance can assist land managers in making informed decisions regarding the potential benefits associated with increased gene flow when conducting restoration activities or developing management plans (Marsden et al., 2013;Whitlock et al., 2013).
Thus, we sought to understand the effects of an artificial increase in gene flow in a species that exists in small and isolated populations, Kates Mountain Clover (Trifolium virginicum Small; Fabaceae), an endemic to the central Appalachian shale barrens. We investigated relationships between crossing distance and reproductive success across populations of different sizes and degrees of isolation. Trifolium virginicum is a perennial herbaceous plant species that is restricted to small habitat patches within the dominant woodland habitat on shale substrate. Shale barrens and T. virginicum are globally rare (NatureServe 2014). Within the shale barren region, however, T. virginicum has a relatively broad distribution, occurring in discrete barren patches within the Ridge and Valley Physiographic Province from southwestern Virginia and adjacent West Virginia, north through western Maryland and south-central Pennsylvania ( Figure 1).
As with many early successional habitats in eastern North America, shale barrens depend on periodic disturbance, such as wildfire, to retard succession to closed forest (Copeheaver, Fuhrman, Gellerstedt, & Gellerstedt, 2004;Foster et al., 2003;Norris & Sullivan, 2002;Tyndall, 2015). Lacking such disturbance, shale barren habitats have become restricted to small patches in which particularly harsh environmental conditions slow succession to woodland (Keener, 1983;Platt, 1951). Beyond forest succession, the shale barren region has experienced an increase in the number of potential barriers to gene flow via development and road construction (Copeheaver et al., 2004;Norris & Sullivan, 2002 Plants have a deep taproot and produce multiple 2-to 3-cm-diameter spherical flower heads on 4-to 15-cm-long peduncles that lie prostrate on the ground and elongate with age. Pollination is likely affected by one or more native bee species. Fruits are slender legumes (pods) containing 1-3 seeds. The perianth and pods are long-persistent; the small seeds (~2.2 to 2.7 mm diameter) are released in late summer upon disintegration of the inflorescence and lack obvious means for long-distance dispersal.
The genus Trifolium is known to possess gametophytic selfincompatibility (GSI; Lawrence, 1996). GSI is a widespread genetic system that enables hermaphroditic plants to avoid self-fertilization and mating with close relatives by rejection in the pistil of pollen carrying the same S-allele. Species in the genus Trifolium are known to have a large number of S-alleles (Casey et al., 2010;Lawrence, 1996).
Given the broad distribution of small, isolated barrens within the extensive forested matrix and lack of adaptation for long-distance seed dispersal, they likely represent relicts of a once more continuous distribution. If they are relicts, the extant patches have highly reduced population sizes and habitat areas that are far more isolated than they were prior to the last 100-160 years (Tyndall, 2015). This isolation of T. virginicum populations creates high risk of increased inbreeding, loss of genetic diversity, and loss of compatible mating types (S-alleles) via drift. These risks could be ameliorated by managing the habitat between barrens to establish larger openings and to enhance gene flow by linking currently isolated patches or by intentional supplementation of small populations that have experienced no increase in population size in >30 years. In this study, we seek to understand the risks and benefits of increasing gene flow in T. virginicum to inform management choices that range from maintaining the status quo (within population mating) to supplementing populations using pollen, plants, or seed from distant populations. We gain this understanding by performing experimental crosses using self-pollen (selfing), pollen from within sites (near-cross), and pollen from distant sites (far-cross).
We quantify the effect of these cross-types on reproduction (fruit set and seed weight) as measures of fitness, and examine variability in self-compatibility (from hand self-pollinations) among maternal plants and sites across a range of population sizes and degrees of isolation.
Because small plant populations are often pollen-limited (Knight et al., 2005), we characterize fruit set in open-pollinated plants under natural conditions and examine relationships with population size and degree of isolation. We calculate the index of self-incompatibility (ISI) as a species average and examine variation between type of outcross pollen (near or far) and variability among individual maternal plants. Thus, we go beyond the traditional usage of ISI in which species mating systems are categorized as self-compatible (ISI < 0.2), mixed-mating (0.2 < ISI > 0.8), or self-incompatible (0.8 < ISI) (see Raduski, Haney, & Igić, 2011). Additionally, we test for reproductive assurance in the absence of pollinators (autogamy). Our analyses focus on answering the following questions: 1. Are there relationships between fruit set in self-and in open-pollinated flowers with population size, or metrics quantifying the degree of population isolation? 2. Does cross-type influence fitness (fruit set) and how much variability is due to maternal plants and sites?
We expected to find evidence of GSI in T. virginicum and concomitantly we predicted that the success of self-pollination would be F I G U R E 1 Geographic distribution of Trifolium virginicum in the eastern United States (hatched), location of the State of Maryland (blue), and the study area in Allegany County, Maryland (star). Inset: detail of the study area including sampled populations (triangles), unsampled but extant populations (filled circles), and extirpated populations (stars) extremely low. However, preliminary data from five T. virginicum plants at three sites indicated great variation among maternal plants in ability to form fruit from hand self-pollinations. Thus, we sought to investigate this variability in selfing success with a larger sample. We predicted that success of self-pollination would vary with population size or degree of isolation. Specifically, we predicted that selfing success would increase with decreasing population size and increasing isolation due to strong selection for breakdown of self-incompatibility in putatively mate-limited populations experiencing little gene flow. In contrast, we predicted that fruit set in open-pollinated plants would increase with increasing population size and decreasing isolation due to higher probabilities of compatible mates, larger floral displays to pollinators, and shorter pollinator flight distances. Finally, we predicted that far-cross success would be consistently greater due to higher probabilities of delivering a novel S-allele as well as the effects of heterosis.
The possibility that T. virginicum may be pseudo-self-fertile (PSF, sensu Levin, 1996) became evident during our field studies. Genes conferring PSF have been demonstrated in the genus by Atwood (1942) and Townsend (1965Townsend ( , 1966Townsend ( , 1969. Trifolium repens and Trifolium pratense are known to exhibit breakdown in self-incompatibility due to PSF loci whose function may vary over time (Riday & Krohn, 2010;Yamada, Fukuoka, & Wakamatsu, 1989) or under high temperatures (Townsend, 1965). If PSF is a natural component of the mating system, we expected some selfing success, but with great variation across maternal plants (Levin, 1996). Variation in selfing success in early versus late flowers is also expected but because this experiment was not designed to test for PSF, we did not test for temporal variation.

| Site selection
We used graph theoretic analysis as implemented by the computer program Conefor 2.6 (Saura & Torné, 2009, 2012 to select sites that varied independently in size and connectivity. Because of the large amount of information that can be gained with few data inputs, graph theory is being increasingly applied to conservation problems (Calabrese & Fagan, 2004;Neel, Tumas, & Marsden, 2014;Pascual-Hortal & Saura, 2006). Graphs provide a spatially explicit representation of landscapes based on the distances at which habitat patches (termed nodes) are connected to one another into networks. We used a layer of all known populations of T. virginicum in Maryland as nodes.
We calculated pairwise geographic distance among patches in ArcMap 10.0 (ESRI 2011). Census size was used to represent habitat size and quality for each node. Census size of each population is based upon counts of individuals at peak flower when plants are most visible.
We selected two graph metrics (IIC connector and BC(IIC)) that provide meaningful and interpretable measures of connectivity (Bodin & Saura, 2010). IIC connector measures each patch's contribution to connectivity as a stepping stone through which other patches of the network are connected. BC(IIC) measures the degree to which a node sits among other nodes in a network based on the number of shortest paths for movement between all pairs of nodes that pass through that node (Baranyi, Saura, Podani, & Jordan, 2011;Bodin & Saura, 2010). We assess the importance of each node using forms of IIC connector and BC(IIC) that are calculated as the change in values between the full network and a network from which each focal node has been sequentially removed (called dIIC connector and dBC(IIC)). dIIC connector and dBC(IIC) are independent measures of connectivity (Bodin & Saura, 2010). The former measures the importance of a node based on how much connectivity would be reduced if the patch was lost. The latter measures the importance of a patch based on its centrality in an existing landscape.
To assess connectivity at multiple scales, we calculated patch importance values for dIIC connector and dBC(IIC) for interpatch distances from 150 m (the minimum distance between two patches) to 36,000 m in 100-m increments (measured from approximate population centroids). We graphically assessed the behavior across this range of distances and noted the distance(s) at which local maximum values (thresholds) were achieved. At these distances, the network of patches is particularly susceptible to changes in connectivity. We noted three such distances: 500, 1,000, and 1,850 m that we examined further.
We examined the rank of node importance values of dIIC connector and dBC(IIC) at the three distances for use. Sites were categorized as connected if both metrics were >0 at two of the three distances and isolated if both metrics were zero at two of the three distances.
Sites were subdivided into small (<50 plants), medium (≥50 and <100 plants), and large (≥100 plants) populations based on censuses by the Maryland Natural Heritage Program from 1984 to 2015. Sites were then evaluated for accessibility (e.g., not on private property) and appropriateness for study (e.g., supported >3 plants).
Twenty-one sites that covered the range of size and connectivity we sought were located in a 10 km × 15 km area that lies within the Green Ridge State Forest in Allegany County, Maryland ( Figure 1). We updated the census for these sites in 2013 (Table 1) (Table 1).

| Selection of maternal plants at sites
At each site, we selected maternal plants for pollination treatments (detailed in the next section) and deployed pollinator exclusion bags prior to flowering (Table 1) (Table 1). Finally, we chose 35 additional plants at 16 sites to test for autogamy (Table 1). These plants received no additional manipulation beyond bagging.

| Crossing design and procedures
On each experimental maternal plant, we established three treatments reflecting distance from pollen source. Implementing all treatments on the same plant controls for maternal genotype. The selfing treatment (S) used pollen from within the same inflorescence. The near-cross treatment (NX) used pollen from 20 to 130 anthers collected from 2 to 13 donors within the site that were ≥5 m from the maternal plant. Pollen for the FX treatment was collected 1-4 hr prior to application.
Anthers were mixed in a vial and kept on ice until they were applied to the mothers. In some cases, entire heads (with attached peduncle) were kept overnight, and anthers were harvested from freshly open flowers.
All crosses were performed under field conditions and sites were visited daily (weather permitting) between May 5 and May 15, 2013 to perform crosses. Pollination treatments were performed on flowers when the banner was expanded and reflexed exposing magentacolored nectar guides to pollinators. We had previously determined that the stigma was receptive in this period as assessed by testing for peroxidase activity using a 3% solution of hydrogen peroxide (Dafni, Kevan, & Husband, 2005). All flowers were emasculated using forceps.
Anthers were applied directly to the stigma in the self-treatment and outcross treatments used stamens with dehiscing anthers from the NX or FX anther pools as appropriate. One to two anthers were haphazardly selected from the appropriate tube for application to the stigma using forceps. Forceps were dipped in ethanol and flamed before moving to the next treatment.
We attempted to complete all treatments on one maternal plant (ranging from 35 to 60 min per plant) on a single day. If few flowers were receptive during a single visit, we returned the following day in

| Fruit set and number of seeds per pod in openpollinated flowers under natural conditions
We quantified natural fruit set from 62 open-pollinated heads from 62 maternal plants at 21 sites. We also quantified the average number of flowers per head and the number of seed per pod in 1,199 legumes containing at least one seed. All statistics reported as mean ± standard error (SE) or median and range of values. We explored relationships between the calculated values for connectivity metrics (dIIC connector , dBC(IIC)) at 500, 1,000, and 1,850 m, and percent fruit set in self-and open-pollination at each site (N = 21) using Spearman rank correlation analyses.

| Data analyses
We examined the effect of cross-type on probabilities of fruit set using a hierarchical generalized linear mixed model (GLMM; as implemented by PROC GLIMMIX, SAS v. 9.4). GLMMs are the appropriate tool for analyzing non-normal data with random effects (Bolker et al., 2008). We modeled fruit set as a binary outcome (failure = 0, success = 1) for each cross-type (self, near, far) resulting from individual hand-pollinated flowers (N = 766), nested within maternal plants (N = 46), which were nested within site (N = 15). We restricted our analysis to the 46 maternal plants with no missing data (all cross-types present) to control for maternal genotype. We specified a binomial distribution and a logarithmic link (logit) to transform the dichotomous outcome into a continuous variable (the log-odds). The logit transformation allows us to establish a linear relationship between our binary outcome variable (fruit set) and the predictor variables (cross-type as fixed effect and maternal plant and site as random effects). The logodds of fruit set and errors for cross-type, maternal plant and site was estimated using residual pseudo-likelihood (modeled in PROC GLIMMIX as the probability of fruit set failure). To account for the hierarchical nature of the data, log-odds is calculated using different intercepts for random subjects (site, maternal plant nested within site) to estimate variance parameters of subject and subject × cross-type interactions. The intercepts are the average log-odds of fruit set for the near-cross because this represents the most probable outcross event among maternal plants within sites. We estimate log-odds solutions for the fixed effects (cross-type = self, near, and far) predicting the probabilities of fruit set failure and test for significant differences in the probability of fruit set failure between cross-types.
For each random subject and subject × cross-type interaction, we report the variance parameter estimates, standard errors, and the percent of the total variance attributable to subjects by dividing the estimate by the sum of variances. We did not calculate log-likelihood ratio tests for significance of random effects because these methods have not been resolved for binary data (Bolker et al., 2008).
Because log-odds range from zero to positive infinity, we converted the log-odds of fruit set to predicted probabilities (expressed as a percent) of fruit set success and failure for fixed effects using the equation where e takes a value of approximately 2.72, ηij is the log-odds of failure, and φ is the probability of failure.
There was essentially no autonomous selfing based on fruit set in the 35 heads testing for autogamy (median = 0) (Table 2; Figure 2). There was no main effect or interaction of cross-type and site on seed weight (F 3, 45 = 1.08, p = 0.22) so seed weight was eliminated as a variable of interest.
The mean ISI value of 0.05 (N = 53) would place T. virginicum in the category of self-compatible species according to a survey of ISI values for >1,200 angiosperm taxa (Raduski et al., 2011). Observed values of ISI across maternal plants ranged from 1 to −4. This variation is obscured if only species-level averages are reported (as in Raduski et al., 2011;Schoen & Lloyd, 1992), and if negative ISI values are set to zero (as in Raduski et al., 2011). These negative values indicate when selfing outperforms outcrossed matings. The success of self-pollination varied dramatically among maternal plants finding instances of both full self-compatibility (100% fruit set) and apparent self-incompatibility (0% fruit set) (Table 2; Figure 3).
The mean ISI value indicated little difference in the overall success of self-versus outcross pollen in terms of fruit set when FX and NX, as outcross pollen, were combined. However, this result masks the contrasting ISI values for the two types of outcross. For the S/NX ratio, ISI of −0.12 (N = 43) indicates higher fruit set on average with self-pollen.   (Table 4).

| DISCUSSION
We Census sizes of all sampled populations are small enough to generate population bottlenecks in S-allele diversity (Thrall, Encinas-Viso, Hoebee, & Young, 2014;Young et al., 2012) and to yield high inbreeding (Leimu, Mutikainen, Koricheva, & Fischer, 2006). The benefits of gene flow as mediated by artificial pollination could act by adding Sallele diversity and by alleviating inbreeding depression (e.g., Frankham, 2015). The lack of relationship between self-success and population size would appear to contradict evidence of the importance of population size in mediating the effects of inbreeding (Angeloni et al., 2011;Leimu et al., 2006). However, given that all of our populations have <150 individuals and most have <50 (Table 1) be particularly long if there is strong selection against selfed progeny (Davies et al., 2015;Delmas et al., 2014;Young, Boyle, & Brown, 1996 T A B L E 3 Log-odds estimates and standard errors from a generalized linear mixed-model (GLMM) analysis predicting the probability of fruit set for fixed effects (cross-type). Significant (p < 0.05) contrasts with near-cross (intercept) are indicated by asterisks. Log-odds estimates were converted to predicted probabilities of fruit set failure (PP 0 ) in the last column. Probability of fruit set success (PP 1 ) = 1 − PP 0 Although we did find evidence for reproductive assurance, selfing success was significantly more variable than far-cross or openpollination, ranging from instances of full self-compatibility to complete failure indicating apparent rejection of self-pollen ( Figure 3).
Although error is a potential explanation, it is unlikely because the selfing treatment required little manipulation. This variability has two potential explanations: (1) random drift of alleles modifying the selfincompatibility reaction, that is, pseudo-self-fertility and (2)  We confirmed one characteristic of PSF identified by Levin (1996): higher fruit production with outcross pollen than self-pollen but a nearly continuous distribution of fruit set following self-pollination ( Figure 3). Additionally, two components of PSF identified by Levin (1996) (Table 4).
Alternatively, variability in the success of self-and near-crosspollination could be due to early-acting inbreeding depression because the former is the most extreme form of inbreeding and the latter likely due to mating by individuals that were already closely related.
According to theory, maintenance of a self-incompatible genetic system is linked with strong inbreeding depression (Gervais, Awad, Roze, Castric, & Billiard, 2014). A self-compatible mutant that arises within a self-incompatible population should spread rapidly as long as the costs of selfing do not exceed the 50% transmission advantage and the benefit of reproductive assurance. The severity of inbreeding depression will be affected by genetic relatedness among individuals and the efficacy of purging in these small populations, both primarily functions of population size (Young & Pickup, 2010). Purging of genetic load has been invoked to explain why small populations persist despite increased inbreeding (Garcia-Dorado 2015; Husband and Schemske 1996;Winn et al. 2011). However, the effectiveness of purging is dependent upon the distribution of deleterious alleles and severity of their effects. Recessive mutations with major fitness effects are more easily purged in an inbreeding population, whereas mutations of small effect may be fixed by drift (Carr & Dudash, 2003;Gervais et al., 2014) or sheltered by linkage to loci under frequency dependent selection such as the self-incompatibility locus (Glémin, Bataillon, Ronfort, Mignot, & Olivieri, 2001). The fitness consequences of increased inbreeding in our populations could theoretically be ameliorated by purging of genetic load and strong selection against inbred progeny (Crnokrak & Barrett, 2002;Fox, Scheibly, & Reed, 2010). However, the fraction of mutation load fixed by drift (drift load) is expressed even under random mating and may form a feedback loop where drift load lowers population size, which in turn enhances drift load (Carr & Dudash, 2003;Willi, Griffin, & Van Buskirk, 2013). A highlight of our study was finding a significant heterotic benefit of far-cross relative to near-cross, which is the best indicator of drift load in populations.
Fitness declines caused by drift load that is fixed within small and isolated populations cannot recover without increased gene flow (Willi et al., 2013).
We cannot distinguish between PSF and inbreeding depression with our data and doing so will be important for this species (and many other endemics) because of the implications for conservation.
If T. virginicum exhibits PSF, then this may be an evolutionary adaptive mechanism that allows persistence of small and isolated populations by ensuring some reproduction and thus the demographic integrity of these small populations. However, the benefits of reproductive assurance gained by breakdown in self-incompatibility may occur at a cost of reduced fitness in resulting progenies, and lifetime inbreeding depression may be particularly acute for long-lived perennial species (Alonso & Garcia-Sevilla, 2013;Delmas et al., 2014;Morgan, Schoen, & Bataillon, 1997). Given the current landscape of isolated patches, increasing the ratio of self-relative to outcross progeny may have long-term population-level-fitness consequences as genetic relatedness increases among remaining individuals and inbreeding reduces fruit and seed set.
Importantly, we found that reproductive assurance can occur in T. virginicum through pollinator-mediated selfing as determined analogously by hand-pollination in our study. However, we found no evidence that autogamous self-pollination is a reliable mechanism for reproductive assurance in absence of pollinators (Table 2; Figure 2); thus, pollinator abundance and floral attractiveness to pollinators in small populations will impact population persistence (Levin, 2012).
Current shale barren restoration and management strategies of selective tree removal and prescribed fire may theoretically restore available habitat for T. virginicum (see Tyndall, 2015). Although there are no data on T. virginicum dispersal distances, Matter, Kettle, Ghazoul, Hahn, and Pluess (2013) found that most dispersal events for the congener T. montanum in calcareous grassland fragments were <1 m and the maximum distance was 324 m. Thus, there is a high probability that T. virginicum seeds are dispersed near the maternal plant or into adjacent nonhabitat forest. Potential for colonizing newly created habitat patches will be limited unless they are extremely close to existing patches. Limitation in the colonizing ability of plant species is recognized as a major obstacle to habitat-based restoration (Donohue, Foster, & Motzkin, 2000). Even if historical disturbance regimes are reinstituted, many previously occupied patches may remain unoccupied and patches remain fragmented. An understanding of long-term gene flow patterns in T. virginicum represents an unmet need but is critical to understanding historical connectivity among populations. If historical patterns of gene flow are precluded by the current levels of isolation, then restoration goals should focus on re-establishing gene flow between sites rather than on only single-site management.
A reasonable hypothesis is that the historical landscape configuration facilitated gene flow, at least among proximal populations (e.g., those barrens occurring along the same ridgeline) and that changes in the landscape configuration and alteration in natural processes have populations (e.g., >100) is enough to conserve the evolutionary potential of these species. This management focus presumes that the larger barrens capture both general levels of genetic diversity and sufficient S-allele diversity. Given the small sizes of even the largest populations and habitat areas, this assumption may not be warranted. Additionally, the success of single-site management is measured by population increase, but the plants are long-lived and there may be a substantial lag time in discerning demographic trends, a characteristic of extinction debt (Hanski & Ovaskainen, 2002;Tilman, May, Lehman, & Nowak, 1994). Future work on T. virginicum focuses on assessing whether the heterosis observed after far-cross extends to lifetime fitness (germination to reproduction) of F1 progeny. Additionally, we need to assess genetic relatedness of individuals within sites to determine the extent to which inbreeding versus lack of S-allele diversity explains the differential success between near-and far-cross in this study. Finally, and critical for management, we need to elucidate long-term (historical) patterns of gene flow to develop appropriately scaled reserves.

DATA ARCHIVING STATEMENT
Locality and population metadata for rare species is confidential but is available upon legitimate request with submission of a data sharing agreement with the Maryland Natural Heritage Program, Annapolis, Maryland, USA. All other data are available in Supporting Information files (Appendix S1) or by request to the authors.