Author for correspondence: Christopher G. Eckert Tel: +1 613 533 6158 Fax: +1 613 533 6617 Email: firstname.lastname@example.org
• Geographically peripheral populations are expected to exhibit lower genetic diversity and higher differentiation than central populations because of their smaller size and greater spatial isolation. In plants, a shift from sexual to clonal asexual reproduction may further reduce diversity and increase differentiation.
• Here, these predictions were tested by assaying 36 inter-simple sequence repeat (ISSR) polymorphisms in 21 populations of the woody, clonal plant Vaccinium stamineum in eastern North America, from the range center to its northern limit where it has ‘threatened’ status. Populations decline in frequency, but not size or sexual reproductive output, across the range.
• Within-population diversity did not decline towards range margins. Modest genetic differentiation among populations increased slightly towards range margins and in small populations with high clonal propagation and low seed production, although none of these trends was significant. Low seed production and high clonal propagation were not associated with large-scale clonal spread.
• By combining demographic and genetic data, this study determined that increased population isolation, rather than reduced population size, can account for the weak increase in genetic differentiation at range margins.
Determining the ecological and historical factors that influence the genetic structure of natural populations has long been a major goal of evolutionary biology and, more recently, the burgeoning field of conservation genetics. It is likely that some of the most important of these factors vary across species’ ranges, resulting in significant geographical variation in population genetic diversity and differentiation. For instance, it is generally expected that a species is most abundant at the center of its geographic distribution, where individual survival and reproductive success are highest, and becomes progressively rarer further from this environmental optimum (reviewed in Sagarin & Gaines, 2002). Thus, populations should be less frequent, smaller, less productive, and less stable demographically towards range limits, thereby reducing effective population size and allowing the erosion of genetic diversity via genetic drift. Greater spatial isolation of peripheral populations may reduce gene flow, further decreasing genetic diversity and increasing differentiation (reviewed in Vucetich & Waite, 2003).
In plants, population genetic structure can also be strongly influenced by variation in reproductive mode (Hamrick & Godt, 1990). For the many plant species that combine sexual reproduction via seed with asexual clonal reproduction through vegetative multiplication, the relative importance of sexual reproduction may decline towards the geographical range limits as a result of marginal ecological conditions or genetic effects rendering sexual reproduction unsuccessful (Eckert, 2002). Reduced recombination may limit genotypic diversity within peripheral populations (Dorken & Eckert, 2001; Billingham et al., 2003). Moreover, the restricted dispersal of clonal propagules compared with sexually produced seed may reduce gene flow, further reducing diversity within and increasing differentiation among peripheral populations (Eckert, 2002). These expectations have rarely been tested with large-scale genetic surveys.
The genetic structure of populations may also be influenced by historical range expansion and contraction, especially for north-temperate species that have recolonized deglaciated areas (Hewitt, 1996; Pamilo & Savolainen, 1999). Post-glacial recolonization is generally expected to reduce diversity and increase differentiation through recurrent founder effect and genetic drift (Cwynar & MacDonald, 1987; Allen et al., 1996).
Lower contemporary effective population size and gene flow combined with greater clonal reproduction, overlaid on a history of post-glacial range expansion, may all contribute to lower diversity and higher differentiation of northern peripheral populations compared with southerly central ones. These expected consequences of geographic peripherality have been evaluated for a wide range of species (reviewed in Eckert et al., 2008). The most common approach simply compares a sample of peripheral to central populations. This provides little information on the form of the relation between population genetic structure and peripherality. In addition, most studies assume that peripheral populations are smaller and more isolated than central populations without verifying geographic variation in demographic variables. Without this information, it is usually not possible to determine whether population diversity and differentiation closely covary with contemporary population size, isolation and/or reproductive mode, or whether geographic variation in genetic structure is likely the result of historical processes (Vucetich & Waite, 2003).
Here, we investigate geographic variation in the genetic structure of the perennial shrub Vaccinium stamineum L. (Ericaceae, deerberry). The species reproduces clonally through rhizomes and sexually through white, buzz-pollinated flowers that yield multi-seeded berries, likely dispersed via ingestion by birds (Vander Kloet, 1988). It occurs in the understory of dry, open, rocky woods throughout eastern North America, from central Florida and west to Texas, and north to southern Ontario (Fig. 1). Based on observations of striking morphological variation, early taxonomy divided V. stamineum into 22 species (Baker, 1970; but see Vander Kloet, 1988). Populations vary in characters such as leaf size, shape and pubescence. In some populations, plants are small and prostrate, while in others, plants are small trees.
In a companion study (Yakimowski & Eckert, 2007), we showed that populations of V. stamineum exhibit a steady, roughly linear decline in frequency from the range center towards northern range margins, and a moderate decline in size. We also demonstrated exceptional variation in sexual reproduction among populations unrelated to geographic position. The mean number of seeds produced per ramet varied 0-1106 among populations, and 46% of populations failed to produce any seed, owing to low flower production and fruit set. This variation provides an excellent opportunity to test the predicted effects of demography and reproductive mode on population genetic structure.
Our work on V. stamineum is also motivated by the species being designated by the Canadian government as ‘threatened’ at the northern periphery of its range. The number of known populations in Canada (currently four) has decreased by > 60% over the last 10–20 yr as a result of habitat destruction (White & Oldham, 2000), although there are no estimates of population extinction from elsewhere in the range for comparison. The conservation value of these geographically peripheral populations is debatable (Hunter & Hutchinson, 1994; Lesica & Allendorf, 1995; Pamilo & Savolainen, 1999). On one hand, environmental extremes at range limits may cause rapid diversification via natural selection and, ultimately, speciation. Peripheral populations may also play an important role in the response of species to changing climates; or they may be prone to extinction, holding little evolutionary potential as a result of low genetic variability and making them difficult to conserve. This study addresses the need to evaluate the genetic structure of threatened northern peripheral populations of V. stamineum in a broader geographical context. Hence, we focus our sampling effort on the northern half of the species’ range.
Specifically, we ask:
• Does genetic diversity within populations decline from the geographic center of abundance towards northern range limits?
• Does within-population diversity increase with contemporary population size?
• Given the marked morphological variation in this species, are populations genetically differentiated and do they become more differentiated towards the northern range limits?
• To what extent do genotypes spread clonally within populations, and does clonal spread correlate with the wide variation in sexual reproductive output?
Materials and Methods
Population sampling strategy and demographic characteristics of populations
In a previous study, we located 51 populations of Vaccinium stamineum L. from the center of the distribution in Virginia to the northern limit in Massachusetts, Ontario and Pennsylvania. For this study, we chose a representative, geographically stratified subsample of 21 populations evenly spread out across the northern half of the geographic range (Fig. 1). For all populations, we calculated the distance to the range center (DRC) and distance to the closest range margin (DRM). Because of the irregularly shaped geographic range of V.stamineum, we defined the range center as the geometric center of the longest south-west to north-east dimension defined by our sampling transect: Chestnut Mountain, Pulaski, Virginia (37.057°N 80.839°W), which we used as the range center for calculation of DRC. Range limits were interpolated from the most geographically peripheral species locations based on extensive field surveys and herbarium records from five major national and regional herbaria (BH, CAN, EC, QC, UNC). Eastern limits were not included because they are imposed by the Atlantic Ocean. All populations were closer to the northern or western limits than to the southern limit. Distance was calculated as the shortest distance between locations measured along the surface of a sphere (great circle distance). Among the 21 populations sampled, DRC and DRM correlated negatively, though only moderately (r = −0.61, P = 0.0033), and hence we used both measures because they capture slightly different aspects of centrality/peripherality (measures are further discussed in Yakimowski & Eckert, 2007).
Plants of V. stamineum consist of aerial stems arising from a clonally spreading underground rhizome. We defined ramet as a primary stem and its associated branches arising directly from the rhizome. Ramets are patchily distributed within populations, presumably because of clonal spread. Hence we defined the patch as a cluster of ramets, all within 1 m of one another and spatially separated from other clusters. We defined a population as a group of patches separated from other groups by at least 1 km, usually much more. For each sampled population, we estimated the number of patches by direct count or by quantifying patch density in three representative 50 × 10 m transects and then multiplying patch density by the total area of the population estimated as the area of a polygon joining the exterior patches. We also estimated the number of ramets in each of 20 patches randomly selected from throughout each population. We used patches per population as the best estimates of population size. Although, we can calculate ramets per population as the product of patches per population and ramets per patch, patches per population is likely a better estimate of the number of genets per population and this interpretation is supported by estimates of within-patch clones presented later.
Sexual reproductive output was measured as the population mean number of seeds produced per ramet, estimated from three component variables: the number of infructescences on each of three randomly chosen ramets for each of 20 patches per population; the number of mature fruit on each of three infructescences randomly chosen from among these three ramets sampled for each patch; and the number of viable seeds in each of 10 fruits per population (each from a different patch). Population size (patches per population), sexual reproduction (seeds per ramet) and clonal propagation (ramets per patch) were all log-normally distributed, so we used log10-transformed values of these parameters in all analyses.
Inter-simple sequence repeat (ISSR) assays
To quantify geographic variation in population-level diversity and differentiation, we sampled leaves from one ramet in randomly chosen patches within each of 21 populations, and flash-dried the material in silica gel. On average, 22 patches were sampled per population (range = 12–25). For five of the six populations in which we sampled < 20 patches, the populations themselves were small (range = 12−30). We also tested whether ramets within patches tended to be from the same clonal genotype compared with ramets from different patches, by sampling leaves from pairs of ramets within each of 10 randomly chosen patches for each of two northern (ONGI and NYIC) and two central (VAML and VABH) populations.
We extracted DNA from each sample of leaves using a modified CTAB protocol. For each sample, 20 mg of dried leaf was placed in a 2 ml microtube with two 0.64 cm stainless steel beads, and ground for 30 s with a Mini-Beadbeater 8™ (Biospec Products, Bartlesville, OK, USA). We added 600 µl of 2 × CTAB buffer (100 mµ Tris-HCl, pH 8, 1.4 m NaCl, 20 mm EDTA, 2% hexadecyltrimethyl-ammonium bromide, 0.2% 2-β-mercaptoethanol; pH = 7.5) and 60 µl of 20% PVP (polyvinylpyrrolidone, 111 MW) to the ground tissue, and then incubated for 45 min at 65°C. DNA was isolated using chloroform : isoamyl alcohol (24 : 1) separation (22 000 g for 10 min), with overnight precipitation in isopropanol at −20°C, after which the DNA was pelleted (6000 g for 10 min), cleaned twice in 70% ethanol, and resuspended in 100 µl of TE buffer (10 mm Tris-EDTA, pH 7.5). The DNA concentration was estimated from duplicate assays on a plate spectrophotometer.
We screened for ISSR polymorphisms using one sample from each of five broadly distributed populations assayed for 93 primers (UBC Set #9 plus all primers from the ISSR Resource Website http://www.biosci.ohio-state.edu/~awolfe/ISSR.html) in duplicate PCR reactions. ISSR polymorphisms are anonymous, dominant markers that are expected to provide good information on genome-wide variation in accordance with other dominant markers as well as co-dominant allozyme polymorphisms. Although ISSRs tend to produce somewhat higher estimates of within-population diversity and slightly lower estimates of population differentiation than more widely used dominant markers (RAPD and AFLP, Nybom, 2004), this does not complicate our interpretations because all samples were screened for the same set of ISSR polymorphisms, allowing unbiased comparisons among populations.
Thirty-five ISSR primers did not consistently produce PCR products and 14 produced monomorphic profiles. From the 44 primers that yielded at least one polymorphic DNA fragment (band), we selected two primers, (CT)8–TG and (CA)6–RG, that produced many highly repeatable polymorphic bands across a broad range of samples. We performed PCR amplification using these two primers in 15 µl PCR reactions containing 20 ng template DNA, 1 × PCR buffer (10 mm Tris HCl, 1.5 mm MgCl2, 50 mm KCl; pH 8.3), 200 µm of each dNTP, 0.67 µm primer, and 0.5 units Taq polymerase (Roche Diagnostics, Basel, Switzerland) run in an Applied Biosystems GeneAmp 9700 thermal cycler. We added MgCl2 to 2.0 mm for primer (CA)6–RG. We used a touchdown program to increase the repeatability of amplification: 90 s denaturation at 94°C; 10 cycles of 40 s denaturation at 94°C, 45 s annealing at 54°C (−1°C each cycle), and 90 s extension at 72°C; followed by 31 cycles of 40 s denaturation at 94°C, 45 s annealing at 44°C, and 90 s extension at 72°C, with a final extension for 5 min.
We resolved PCR products in 9 × 20 cm 1.5% agarose gels stained with ethidium bromide run for 3–4 h at 87 V. We digitally photographed gels under UV light and measured fragment sizes against a 100 base-pair DNA ladder with Gene ImagIR (v. 4.03, Li-Cor Biotechnology Division, Lincoln, NE, USA), allowing us to consistently discern fragments differing by as little as 30 bp. We scored fragments as present or absent, and different sized fragments were treated as independent markers. We replicated PCR reactions for 10 samples per population plus within-patch samples for both primers in different runs to identify ISSR polymorphisms that exhibit 100% repeatability across those 250 samples. Single PCR reactions were then run for the remaining 235 samples.
ISSR data analyses
Genetic diversity within populations was measured as the proportion of loci polymorphic (PP) and the expected heterozygosity (He), calculated for dominant markers following Lynch & Milligan (1994). This calculation of He assumes that marker genotype frequencies are at Hardy–Weinberg equilibrium, which is a fair assumption given that V. stamineum is likely obligately outcrossing owing to very strong self-sterility (Hokanson & Hancock, 2000; S. P. Vander Kloet, unpublished) and exhibits very high genotypic diversity indicative of sexual reproduction (see the Results section). We tested whether within-population genetic diversity declined towards range margins using Pearson correlations between He and PP and both DRC and DRM with one-tailed tests of significance. PP and He correlated positively but not strongly among populations (r = +0.56, P = 0.0086). Analyzing the first principal component of the correlation between these two somewhat redundant measures of genetic diversity (following Garner et al., 2004; PC1 eigenvalue = 1.56, variance explained = 77.9%) yielded similar results. Hence we present univariate analyses to facilitate comparisons with other studies (reviewed in Eckert et al., 2008).
We tested for overall genetic differentiation among populations using analysis of molecular variance (AMOVA, Excoffier et al., 1992). We measured genetic distance between pairs of populations using an analogue of GST for binary dominant polymorphisms (ΦPT), calculated as the variance between populations relative to the total variance (Peakall & Smouse, 2006). We then tested for isolation-by-distance among populations by correlating ΦPT with the great circle surface distance between populations using a Mantel's test (1000 permutations, GenAlEx v. 6, Peakall & Smouse, 2006).
We tested the prediction that population genetic divergence should increase towards range limits using the mean ΦPT between each population and all other populations as a measure of genetic divergence, and then testing for a positive correlation of genetic distance, linearized following Rousset (1997) as ΦPT/(1 − ΦPT), with DRC and a negative correlation with DRM. Because pairwise values of ΦPT among populations are not independent from one another, it follows that mean ΦPT values are not statistically independent. Hence, we tested the significance of each correlation (r) by permuting the matrix of pairwise ΦPT to generate a null distribution of r. The one-tailed permutation test type I error (Pper) is the proportion of null r values greater than the observed r. We used the same approach to test the expectations that population divergence should correlate negatively with population size (patches per population) and sexual reproduction (seeds per ramet), and positively with clonal propagation (ramets per patch). These three variables were log10-transformed to improve normality.
We quantified population-level clonal structure by determining what proportion of ramets sampled from discrete patches within each population (one ramet per patch) were distinct multilocus genotypes, summarized as genotypic richness, R = (G − 1)/(n − 1), where G is the number of distinct genotypes and n is the number of ramets sampled (Dorken & Eckert, 2001). Because all ramets sampled were genetically distinct for 16 of the 21 populations (see later), we could not use a correlation approach as described earlier to test for associations between R and measures of geographic position (DRC and DRM) or population size and reproductive outputs. Instead we used one-tailed randomization tests (1000 permutations of the data per test) to determine whether the populations with R < 1 ranked particularly highly for DRC and clonal reproduction, or particularly lowly for DRM and sexual reproduction.
For the four populations in which we sampled pairs of ramets within patches, we tested for patch-level clonal structure in two ways. First, we determined whether ramets sampled from the same patch expressed identical genotypes more often than ramets sampled from different patches. We then tested whether ramets sampled from the same patch (n = 10 pairs per population) were more closely related than expected by chance, by calculating the genetic distance between them (d = 1 − proportion of ISSR bands shared between ramets) and comparing this with a distribution of mean d from 1000 sets of 10 pairs chosen randomly from different patches. The significance of the comparison is represented by Pd, the proportion of resampled sets with mean d less than d observed for ramets within patches. Although a sample of two northern and two southern populations does not provide a powerful test for geographic variation in patch-level clonal structure, we performed a preliminary comparison between northern and central populations. We contrasted the proportion of ramet pairs sampled within patches that expressed the same multilocus genotype (d = 0) using a 2 × 2 contingency table chi-squared test, and the mean d within patches using analysis of variance with a planned orthogonal contrast between northern and central populations.
Variation in population size and sexual reproductive output
Population size (patches per population), sexual reproduction (seeds per ramet) and clonal propagation (ramets per patch) varied substantially among the 21 populations sampled (Table 1). However, only ramets per patch covaried with geographic position: increasing with DRC and decreasing with increasing DRM, as expected. Population size did not correlate with either sexual reproduction (r = +0.15, P = 0.51) or clonal propagation (r = +0.02, P = 0.93). The expected negative correlation between sexual reproduction and clonal propagation was weak and not significant (r = −0.14, 1-tailed P = 0.54).
Table 1. Geographic variation in population size, sexual reproduction and clonal propagation among 21 populations of Vaccinium stamineum sampled across the northern half of the geographic range in eastern North America
All statistics are based on population means (CV, coefficient of variation). Pearson correlations between each variable and distance to range center (DRC) and distance to range margin (DRM) are presented. All population variables were log10-transformed for correlation analysis. One-tailed tests of significance for the correlations are denoted as: *, P < 0.05; **, P < 0.01; P > 0.2 for all others.
Using two ISSR primers, we resolved 49 bands, of which 38 were clear and 100% repeatable. Thirty-six of these were polymorphic (14 ranging 500 to 1200 bp using primer [CT]8–TG, 22 ranging 400 to 2000 bp using [CA]6–RG). The proportion of loci polymorphic (PP) ranged between 0.39 and 0.69 (mean ± SE = 0.58 ± 0.01), and did not exhibit the expected decrease towards range margins (Fig. 2a, Table 2). The population with the lowest PP (MAMT) was a northern peripheral population. However, populations with high PP were distributed throughout the range. Estimates of expected heterozygosity (He) ranged between 0.15 and 0.22 (mean ± SE = 0.18 ± 0.01) and tended to decrease towards range margins, but did not correlate significantly with DRC or DRM (Fig. 2b, Table 2). The four populations we sampled from the very northern limit of the range, PAPI, NYWI, ONGI, and MAMT (see Fig. 1), ranked third, eighth, 10th and 18th, respectively, for He. We only detected one private band (i.e. a band detected in a single population), which occurred in central population TNOC. None of the correlations between diversity and measures of population size, sexual reproduction or clonal propagation were significant (Table 2).
Table 2. Correlations between measures of genetic diversity and differentiation and population geographic position (DRC and DRM), size (patches per population), sexual reproduction (seeds per ramet) and clonal propagation (ramets per patch) among 21 populations of Vaccinium stamineum sampled across the northern half of the geographic range in eastern North America
Among-population correlation between genetic parameter and:
All analyses used Pearson correlations with standard one-tailed tests of significance, except those involving mean ΦPT, which used one-tailed permutation tests. Population size, sexual reproduction and clonal reproduction were log10-transformed for analysis. None of these correlations was significant at P < 0.05. *P < 0.1. Correlation coefficients with signs in contrast to predictions are in parentheses.
AMOVA (Table 3) detected significant differentiation among populations (P < 0.001), with 11% of the total genetic variation partitioned among populations (mean ΦPT = 0.115). Pairwise genetic distance, measured as ΦPT/ (1 − ΦPT), correlated positively with the geographic distance between populations (r = +0.33, Mantel's test P < 0.001). Our measure of population genetic divergence, the mean ΦPT between each population and all other populations, correlated negatively but weakly with both measures of within-population diversity, although only the correlation with PP was near significant (PP r = −0.35, one-tailed permutation test Pper = 0.07; He r = −0.12, Pper = 0.32). Although mean ΦPT correlated positively with DRC and negatively with DRM, neither correlation was significant (Fig. 3a, Table 2). The four northern peripheral populations ranked second, third, fifth and 16th, respectively, for ΦPT. Also as predicted, mean ΦPT correlated negatively with population size and sexual reproduction and positively with clonal propagation, although, again, none of these correlations was significant (Fig. 3b–d, Table 2).
Table 3. Analysis of molecular variance (AMOVA) for 445 Vaccinium stamineum ramets from 21 populations sampled across the northern half of the geographic range in eastern North America
Source of variation
Clonal genetic structure within populations
There was little evidence of large-scale clonal structure within populations. Among 445 ramets sampled from 21 populations, we detected 434 distinct genotypes. All ramets sampled per population were genetically distinct (R = 1) for all but five populations (NYAT and PAPI, R = 0.95; VADC, R = 0.89; MAMT and NCBR2, R = 0.84). Four of these five populations exhibited very low mean seed production (< one seed per ramet), ranking in the bottom quartile of the distribution of seeds per ramet. However, the ranking of these five populations was not significantly lower than expected by chance (randomization test one-tailed P = 0.095). Four of these five populations also ranked in the top half of the distribution of clonal propagation (stems per patch), but these rankings were not significantly higher than expected by chance (P = 0.15). There was a tendency for populations with R < 1 to be farther from the range center and closer to range margins but neither was significant (DRC, P = 0.22; DRM, P = 0.36). They also tended to exhibit lower within-population diversity and higher among-population differentiation, although only the trend for PP was significant (PP, P = 0.023; He, P = 0.11; mean ΦPT, P = 0.22).
We detected significant clonal structure within patches. Overall, 19 of 40 pairs of ramets sampled within patches expressed identical ISSR genotypes, and this never occurred when ramets were sampled from different patches within these four populations. Mean genetic distance (d) between ramets from the same patch was lower than d between ramets from different patches for all populations (all Pd < 0.001). Excluding identical pairs, d within patches was still lower than d between patches for three of four populations (NYIC, Pd = 0.09; ONGI, Pd < 0.001; VAML, Pd < 0.001; VABH, Pd < 0.001). The proportion of within-patch pairs expressing identical genotypes was not higher for northern (0.40 overall; ONGI = 0.5, NYIC = 0.3) than central (0.55 overall; VABH = 0.5, VAML = 0.6) populations (2 × 2 contingency table, likelihood ratio χ2 = 0.91, df = 1, P = 0.34). There was heterogeneity among populations in d between ramets within patches (one-way ANOVA: r2 = 0.27, F3,36 = 4.5, P = 0.01). However, this resulted from higher mean d for one northern population (NYIC d = 0.142) than the others (ONGI, d = 0.031; VAML, d = 0.036; VABH, d = 0.053), and the contrast between northern vs central populations was not significant (F1,36 = 2.9, P = 0.10).
The main goal of this study was to test the general prediction that geographically peripheral populations are less genetically diverse and more differentiated than central populations because of geographic variation in population size, spatial isolation, reproductive mode and/or post-glacial history. Overall, our results suggest that V. stamineum, in the northern half of its geographic range, exhibits some of the expected geographic trends in genetic structure but the patterns were weak, at best. We tested 10 correlations involving within-population genetic diversity (Table 2) but only four were in the predicted direction, and all were very weak (all | r | < 0.2). By contrast, all five correlations involving genetic differentiation were in the predicted direction, although, again, none was significant individually (Table 2), and a binomial test for the likelihood of all five correlations in the predicted direction was not quite significant (P = 0.062). We found little evidence for population-level clonal spread in V. stamineum. For 16 populations, all ramets we sampled were genetically distinct (R = 1). The five populations with R < 1 tended to be located somewhat further from the range center, closer to range margins and to exhibit lower sexual reproduction and higher clonal propagation, but none of these associations was significant.
Although some taxonomists have viewed tremendous morphological variation within V. stamineum as evolutionary differentiation (Baker, 1970), the populations we sampled were only weakly differentiated (overall ΦPT = 0.115) compared with other widespread, temperate species that combine clonal reproduction with sexual reproduction through outcrossing and endozoochorous seed dispersal (mean GST ≥ 0.200 based on allozymes; Hamrick & Godt, 1990).
Geographic covariation between genetics, demography and history?
The general prediction that geographically peripheral populations should exhibit lower diversity and higher differentiation is based on the ‘abundant center model’, which proposes that population size decreases and the spatial isolation of populations increases from the range center towards the range limits (Sagarin & Gaines, 2002). Eckert et al. (2008) exhaustively reviewed published studies that had tested these predictions for a total of 67 plant species. Of 81 studies, 64% supported the prediction of lower within-population genetic diversity towards range margins. Similarly, 70% of 37 studies supported the prediction of higher diversity among peripheral compared with central populations. Although the degree of support for both predictions is significantly higher than expected by chance, data from a substantial proportion of species failed to support either prediction. Most of the studies testing for reduced diversity and increased differentiation towards range margins have used a categorical sampling strategy, while relatively few studies (32%) have used the continuous sampling strategy employed here, and, of these, far fewer (47%) found evidence for reduced diversity towards range edges. It is not clear why these two sampling strategies have yielded different results (Eckert et al., 2008), but it is clear that the results of this study are in line with those from other studies using a similar population sampling strategy. Moreover, differences in diversity or differentiation between central vs marginal populations, even when they appeared to occur, were rarely very large. This raises an important question: why is the expected relationship between geographic position and genetic structure not more common?
Perhaps the main reason that peripheral populations are often not genetically depauperate or highly differentiated is that these predictions rely on an underlying geographic distribution which conforms to the abundant center model. By contrast, most geographic surveys of habitat occupancy and local abundance have failed to support this model of geographic distribution (Sagarin & Gaines, 2002). Unfortunately, few population-genetic studies (< 20%) have quantified geographic variation in population demographic parameters thought to cause variation in genetic structure (reviewed in Eckert et al., 2008). For instance, of the 81 studies on plants, only 26% estimated population size; only 6% estimated some aspect of population isolation (Lönn & Prentice, 2002; Jump et al., 2003; Van Rossum & Prentice, 2004); and even fewer estimated geographic variation in sexual or clonal reproduction (Dorken & Eckert, 2001; Jump et al., 2003). Among the 21 studies that did test for the assumed geographic trend in population size, there was no significant difference in support for the prediction of reduced diversity towards range limits between those that detected a decline in population size (69%) and those that did not (75%), but the sample of studies is still too small for any rigorous meta-analysis (Eckert et al., 2008).
Our survey of 51 V. stamineum populations across the northern half of the range indicates that this species conforms to the abundant center model with respect to declining population frequency and increased isolation, but not population size. Our best estimate of population size, patches per population, correlated moderately with DRM (r = +0.43) but only weakly and not significantly with DRC (r = −0.26; Yakimowski & Eckert, 2007). However, the correlation with DRM was not significant among the 21 populations subsampled here, suggesting the trend is weak and sensitive to sampling. Patches per population did not correlate with geographic position or either measure of within-population diversity. This is consistent with other studies, described earlier, where a decrease in genetic diversity is less likely when populations do not decline in size towards range limits. Thus, our demographic data suggest that the weak geographical variation in population differentiation we observed can likely be attributed to increased population isolation towards northern range margins but probably not to variation in population size. Alternatively, post-glacial colonization can also contribute to geographic variation in population differentiation within north temperate species (Allen et al., 1996; Van Rossum & Prentice, 2004). In the case of V. stamineum, however, the increase in population isolation towards the range margins would seem sufficient to account for the weak geographic trend in genetic differentiation. Although post-glacial processes cannot be discounted, there is no justification for invoking them (Vucetich & Waite, 2003).
Geographic variation in clonal reproduction?
Vaccinium stamineum exhibits considerable among-population variation in sexual reproductive output and clonal propagation, and the latter increases significantly towards range margins, even among the smaller sample of 21 populations studied here. Five of the 21 populations we sampled produced no seed at all, and another three produced fewer than one seed per ramet, on average. In some species, striking reproductive variation is reflected in a reduction in the genotypic diversity of geographically marginal populations (Dorken & Eckert, 2001; Billingham et al., 2003). Peripheral populations of V. macrocarpon, for example, exhibit lower genotypic diversity than central populations (Stewart & Excoffier, 1996). By contrast, 16 of the 21 populations of V. stamineum that we sampled were genotypically diverse (R = 1), including four of the eight populations with very low seed production (seeds per ramet < 1). Our results are consistent with the only other genetic study of V. stamineum (Kreher et al., 2000), which found that almost all ramets sampled from different patches within a South Carolina population expressed unique RAPD genotypes.
The populations with some potential indication of clonal spread (R < 1) tended to rank lowly for sexual reproduction and highly for clonal propagation, as predicted, but this trend was not significant. Similarly, there was a weak and nonsignificant tendency for these populations to be located further from the range center and closer to a range limit than populations with R = 1. High genotypic diversity in populations of long-lived perennial plants with low average seed production can be maintained by sporadic seed production and/or seedling recruitment (McLellan et al., 1997). This may be the case in V. stamineum, as extensive population studies revealed substantial between-year variation in seed production (Yakimowski & Eckert, 2007).
Although almost all ramets sampled from different patches within populations of V. stamineum were genotypically distinct, ramets sampled from within patches were genetically identical in c. 50% of patches sampled. When ramets from the same patch were distinct, their genotypes were, on average, more similar than expected by chance. This is consistent with patch-scale clonal spread and limited seed dispersal. Restricted clonal spread was also detected using genetic markers for a South Carolina population of V. stamineum (Kreher et al., 2000) and within populations of V. myrtillus (Albert et al., 2003). Such fine-scale genetic structure has been documented for a wide variety of plant species, and may become more pronounced in low-density, geographically peripheral populations where ‘shadows’ around individuals within which seedlings and clonal progeny are recruited overlap to a lesser extent (reviewed in Gapare & Aitken, 2005). Further study of this requires more within-patch sampling in more populations from both the center and periphery of the range. So far our modest sampling within patches has not detected higher within-patch genetic similarity in two northern populations compared with two central populations of V. stamineum.
Implications for conservation of geographically peripheral populations
A major goal of this study was to evaluate the genetics of threatened peripheral populations of V. stamineum in a geographic context. Much debate about the conservation of peripheral populations revolves around two questions (Lesica & Allendorf, 1995): is the small size and spatial isolation of peripheral populations associated with low genetic diversity that might limit short-term persistence and long-term evolutionary potential; or might peripheral populations, as a result of selection under ecologically marginal conditions, be adaptively diverged in ways that contribute to the long-term evolutionary potential of the species? Here, we will discuss how our previous study of population demography, reproductive output, dispersal potential and offspring fitness combines with the present analysis of genetic diversity and differentiation to inform some aspects of these larger conservation-oriented questions.
First, demographic and genetic data exhibit no evidence of reduced viability of northern peripheral populations. Increased rarity of V. stamineum towards its northern limit is not associated with reduced diversity of ISSR markers (Fig. 2). Moreover, previous studies have shown that seed production in peripheral populations is also high (Yakimowski & Eckert, 2007). Despite the increased isolation and sometimes smaller size of northern peripheral populations, clonal spread is not rampant, and sexual reproduction and genetic diversity are not low. From a conservation perspective, this implies that although these populations are rare, they should remain viable and persist, especially if protected from human disturbance.
In addition to northern peripheral populations being relatively viable, there is some evidence that these populations are also adapted to more northern climatic conditions. Glasshouse experiments have shown that, compared with central populations, plants in northern peripheral populations produce large, highly viable seeds that germinate and develop more quickly, which is likely adaptive in more seasonal northern environments (Yakimowski & Eckert, 2007). This pattern of seed size evolution can likely be attributed to selection imposed by a climatic gradient rather than genetic drift because ΦPT estimated with neutral genetic markers is weak overall and only increases very weakly towards northern range margins, yet the correlation between seed size and latitude is strong and highly significant (r = +0.79, P < 0.0001). We also showed that plants in the most northern peripheral populations of the Thousand Islands provision fruit with proportionately more pulp than populations to the south (Yakimowski & Eckert, 2007). Given that pulp represents allocation to seed dispersal, probably by frugivorous birds, high pulp allocation may be the product of selection during rapid cycles of population extinction and recolonization at range limits (see also Darling et al., in press). Increased germination, seedling growth and allocation to seed dispersal observed in the northern range limit might also counter any effect of lower population density and higher spatial isolation on population-level and patch-level genetic structure of peripheral populations. Although the expected negative consequences of geographic peripherality sometimes occur in this clonal, woody shrub, the evolution of key reproductive traits may facilitate the persistence of populations under marginal conditions. All this is consistent with northern populations possessing significant evolutionary potential, and provides evidence that these geographically peripheral populations of V. stamineum may be of significant conservation value.
We thank Jessica Montague and Jeremy Brown for help in the field; Heather Stefanison and Karen Samis for help in the laboratory; Chris Burns (Ontario Ministry of Natural Resources) and especially Jeff Leggo (Parks Canada) for logistic support and arranging funding; Sam Vander Kloet and Adele Crowder for information and helpful suggestions; Spencer Barrett, Steve Lougheed and Ruth Shaw for comments on the manuscript; Queen's University Advisory Research Council, Parks Canada, and Ontario Ministry of Natural Resources for research grants to CGE; and the Natural Sciences and Engineering Research Council of Canada for a postgraduate scholarship to SBY and a Discovery Grant to CGE.