SEARCH

SEARCH BY CITATION

Keywords:

  • conservation genetics;
  • genetic divergence;
  • genetic drift;
  • habitat fragmentation;
  • population size;
  • Ranunculus reptans

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Reduced genetic variation at marker loci in small populations has been well documented, whereas the relationship between quantitative genetic variation and population size has attracted little empirical investigation. Here we demonstrate that both neutral and quantitative genetic variation are reduced in small populations of a fragmented plant metapopulation, and that both drift and selective change are enhanced in small populations. Measures of neutral genetic differentiation (FST) and quantitative genetic differentiation (QST) in two traits were higher among small demes, and QST between small populations exceeded that expected from drift alone. This suggests that fragmented populations experience both enhanced genetic drift and divergent selection on phenotypic traits, and that drift affects variation in both neutral markers and quantitative traits. These results highlight the need to integrate natural selection into conservation genetic theory, and suggests that small populations may represent reservoirs of genetic variation adaptive within a wide range of environments.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

The fields of population and conservation genetics are particularly concerned with processes that cause small populations to suffer from genetic erosion. One such process is genetic drift, a change in allele frequencies caused by chance events. Drift acts especially strongly in small populations to cause fixation and loss of alleles, which reduces genetic diversity (Ellstrand & Elam, 1993; Frankham, 1996; Spielman et al., 2004). At sufficiently small population size, drift can even outweigh the force of selection, and as a consequence, it may lead to the loss of adaptive genetic variation and accumulation of deleterious mutations (Lande, 1988; Lynch & Gabriel, 1990; Lynch et al., 1995; Hedrick, 2001). Loss of neutral variation in small populations has been well documented by molecular markers (Frankham, 1996), and loss of quantitative genetic variation is often observed in laboratory studies of inbreeding (Van Buskirk & Willi, 2007). However, there is little evidence that variation in quantitative traits declines with population size in nature (Willi et al., 2006). Furthermore, in comparison with drift, natural selection on expressed traits has attracted little empirical attention from conservation geneticists, although population declines and habitat fragmentation potentially modify the direction and intensity of selection.

Fragmentation and decreasing patch size cause changes in environmental conditions such as microclimate, wind exposure, predation rate, and prevalence of diseases and parasites (Robinson et al., 1995; Laurance et al., 1998; Groppe et al., 2001; Cascante et al., 2002; Chalfoun et al., 2002; Allan et al., 2003). Conditions in fragments are sometimes associated with reduced offspring recruitment (Jules & Rathcke, 1999; Ward & Johnson, 2005) or enhanced mortality (Robinson et al., 1995; Laurance et al., 1998). These demographic impacts are likely to affect individuals nonrandomly with respect to genotype, imposing natural selection and shifting the genetic composition of fragmented populations. There are two possible consequences. One is that fragmentation might affect conditions in all habitat patches similarly, causing small populations to differ from large populations in a consistent direction. Warmer and drier conditions within forest fragments, for example, might favour similar adaptive responses in all populations occupying fragments. On the other hand, decreasing patch size could erode environmental conditions in different ways within different patches, leading to divergent selection among demes or different selection intensities. This possibility acknowledges that population sizes can be small for a wide variety of reasons (Brown, 1984), each of which potentially imposes a distinct selection regime. If small populations are often exposed to more divergent selection than larger ones, this would increase adaptive among-population genetic variation in declining species. Here we present an empirical test for the hypothesis that habitat fragmentation can lead to adaptive evolution.

An indirect estimate of the strength of drift and natural selection within populations is possible by comparing two kinds of between-population genetic variation: neutral marker (FST) and potentially adaptive, quantitative (QST) variation (Spitze, 1993; Merilä & Crnokrak, 2001; McKay & Latta, 2002). QST is the quantitative analogue of FST and therefore allows direct comparison between levels of population differentiation at marker loci and metric traits (Spitze, 1993; Lynch & Spitze, 1994). Similarity between estimates of QST and FST suggests that quantitative characters have been subjected to evolutionary forces similar to those targeting the neutral markers, primarily genetic drift. If QST deviates from FST, then selection has shaped differentiation.

This study investigated whether within-population genetic variance for quantitative traits depends on population size, and compared QST and FST values to assess the relative roles of drift and selection in divergence between populations of varying size. Our study organism was the creeping spearwort, Ranunculus reptans, occurring in fragmented populations at Lake Constance (Fig. 1). Shoreline development over the last century has reduced the number of extant populations and led to small population sizes. We sampled plants over the same spatial scale in 13 populations and assessed within- and between-population allozyme variation and genetic variation for morphological and life-history traits. If fragmentation causes consistent environmental differences between large and small populations, we should observe relatively high quantitative divergence among demes of very different size. Alternatively, if fragmentation increases the strength of divergent selection among small demes, we expect to find disproportionately high divergence in quantitative traits as local population size declines.

image

Figure 1.  Locations around Lake Constance of the Ranunculus reptans populations included within this study. The 13 local populations were distributed over three lake basins. Symbols represent population size: solid circles are the four largest populations, open triangles are medium-sized populations, and small open circles are the five smallest populations. There was no correlation between the size of pairs of populations (harmonic mean) and log-transformed geographic distance between them (rMANTEL = 0.09, P > 0.4) or lake basin (rMANTEL = 0.07, P > 0.5).

Download figure to PowerPoint

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Plant material and electrophoresis

Ranunculus reptans is a clonal, self-incompatible plant growing in a fragmented band on the shores of Lake Constance (Fig. 1). The persistence of these populations is associated with the regular occurrence of floods, because R. reptans is flood tolerant but a poor competitor. The species probably had a fragmented distribution even before strong human impact began during the 20th century, and little further shoreline development has occurred since about 1970. Our 13 study populations ranged in surface area from 40 to 2000 m2 in 2003; population sizes were estimated by multiplying surface area (m2) by a measure of density of rooting rosettes within R. reptans aggregations (Appendix S1). We estimated density by placing a 0.25 m2 frame haphazardly on the ground and counting the number of 5 cm by 5 cm squares that had at least one rosette with roots. The number of frame counts per population ranged from 11 to 93, depending on population size (Willi et al., 2005).

In spring 2002, we collected 163 plants from 13 populations. At each site, 14 individuals were collected at 5-m intervals along two transects separated by 5 m. The distance between the two transects was decreased to 4 m in four narrow populations. In five populations, the band of R. reptans along the shoreline was so short that we could only sample eight to 12 individuals. The number of sampled individuals was not correlated with population size (P > 0.3).

Neutral genetic population structure was estimated by cellulose acetate electrophoresis (Hebert & Beaton, 1993) of eight loci at seven enzyme systems: AAT-1 (EC 2.6.1.1), ACON-1 (EC 4.2.1.3), GPI-2 (EC 5.3.1.9), MDH-2 (EC 1.1.1.37), MDH-3, MPI (EC 5.3.1.8), 6-PGDH (EC 1.1.1.44), and SKD (EC 1.1.1.25). Allozyme banding patterns were interpreted at the allelic level using information on the number of isozymes and their quaternary structures from Wendel & Weeden (1989). Allele distribution within samples suggested a tetraploid-tetrasomic inheritance, because there was neither fixed heterozygosity nor recurring allele combinations, which would have indicated disomic inheritance (Gastony, 1990; Stehlik & Holderegger, 2000). This conclusion is supported by comparative cytological work on R. reptans (synonym: Ranunculus flammula var. filiformis) indicating a euploid, tetraploid chromosome number of (2n=) 32 (4x = 32) throughout mainland Europe and North America (Goepfert, 1974; Gibbs & Gornall, 1976).

Within-population genetic variation was measured using Nei's gene diversity, Hs (Nei, 1973) (Appendix S1). We estimated the degree of neutral genetic differentiation among populations using FST (Hardy & Vekemans, 2002), defined as FST = (qsqt)/(1–qt), where qs(qt) is the probability that two alleles from the same population (different populations) are identical (Excoffier, 2003) (matrix of pairwise FST in Appendix S2). FST can be biased in tetraploid species with tetrasomic inheritance, if selfing and double reduction affect identity probabilities of alleles (Ronfort et al., 1998). Two observations suggest that our estimates of FST are immune to this problem. First, selfing cannot affect identity probabilities in our case because R. reptans is gametophytically self-incompatible. Secondly, FIS values were not significantly larger than zero for all populations (one sample t-tests: all P > 0.1; range of population means for the eight allozyme loci -0.14 to 0.05).

Allozymes are valid markers for the study of neutral variation (Merilä & Crnokrak, 2001), and we observed no deviation from Hardy–Weinberg equilibrium in 11 of the 13 populations (Appendix S1). We also tested for evidence of selection on the allozymes by comparing observed FST values with those simulated under an island model (50 islands), using an infinite alleles mutational assumption, and with individuals and populations sampled according to the actual data (using fdist2; seeBeaumont & Nichols, 1996). The results gave no indication of selection: observed FST values for all loci were well within the 95% confidence intervals (Appendix S3).

Common garden field experiment

We estimated within- and between-population genetic variation, independent of environmental differences among the source locations, by rearing offspring of the field-collected parents in a common environment. Each genotype was crossed reciprocally with two other genotypes from the same population, resulting in 319 parental combinations (total of 638 reciprocal crosses: 13 populations × 12.2 field-collected plants on average × 4 crosses). We germinated the seeds indoors on a 3 : 1 mixture of horticulture soil and sand. After 6 weeks, we haphazardly chose one seedling per seed family for planting into a tub (10 cm × 10 cm × 11 cm) with a 1 : 2 mixture of soil and sand. Tubs were placed in random positions within outdoor beds under 50% shade cloth, and the experiment continued for 2 months. For each cross which produced at least one seedling (524 in total), we assessed eight quantitative traits: (i) average timing of germination (number of seedlings that had emerged after 3 weeks divided by number of seedlings that had emerged after 6 weeks), (ii) number of primary stolons growing from the rooted rosette with the highest number of leaves, (iii) number of leaves of this rosette, (iv) length of its longest leaf, (v) width of that longest leaf, (vi) ratio of the number of flowers and infructescenses to the number of rooted rosettes, as an estimate of allocation to sexual vs. vegetative reproduction, (vii) timing of flowering (number of infructescenses divided by the sum of flowers and infructescenses) and (viii) mean fresh biomass per rooted rosette. We log-transformed phenotypic measures except for proportions, which underwent an angular transformation.

Statistical analysis

Within-population additive genetic variances were calculated for each population and quantitative trait based on general linear models including siring male genotype (paternal half-sibs) and female genotype (maternal half-sibs). Additive genetic variance was four times the among-sire variance component (Lynch & Walsh, 1998). Heritability was additive genetic variance divided by the phenotypic variance.

We calculated differentiation of each quantitative trait between pairs of populations (QST) from variance components estimated by maximum likelihood (ML) with the varcomp procedure of sas (SAS Institute Inc., 1999). Total variance among plants for each trait and each possible pair of populations was partitioned among the random effects of population, siring male genotype nested within population, female genotype nested within population, and residual. Pooled results for all populations are presented in Appendix S4. Population differentiation was quantified by the following formula, which adjusts for tetraploidy:

  • image

where Vpop is the variance between populations, and Va is the additive genetic variance within populations. Va is multiplied by 4 because QST is based on a comparison of genotypes, whereas FST is based on comparisons of genes (Lynch & Spitze, 1994). To check whether the absence of negative variance components under ML estimation caused an estimation bias in QST, we also calculated variances from a general linear model.

Two-sided Mantel tests (Bonnet & Van de Peer, 2002) were used to study the relationship between pairwise FST and QST values, and to test if the two were related to geographic distance, lake basin (Fig. 1), and population size. Both simple and partial tests were used, with the latter performing permutations on raw values instead of residuals. Mantel tests indicate whether pairs of matrices differ from each other significantly. We began by testing the correlation between mean QST and FST to determine if quantitative traits showed the same pattern of divergence as the allozymes, which would indicate that quantitative traits are subject to drift. Next we tested the correlation between pairwise FST values and the harmonic mean of pairwise population sizes, with the prediction that genetic drift causes enhanced divergence when one or both populations of a pair is small. We used the harmonic mean in this analysis because the impact of drift scales with the harmonic mean of population sizes (Crow & Kimura, 1970). Finally, we used a partial Mantel test to check whether pairwise QST estimates were correlated with population size after correcting for FST values. A significant result would indicate that the strength of divergent selection depends on population size. Traits were somewhat correlated across all populations, so analysis of QST was done for each trait separately, but also for the mean over all traits.

The timing of germination had a strong maternal environmental basis (Appendix S4) and was therefore discarded from analyses of population divergence. The matrix of mean pairwise QST over the seven remaining traits is presented in Appendix S5. Allozyme and quantitative genetic variation were not correlated with distance from the target population to its nearest neighbouring population (P > 0.1; range 100−2200 m), so we did not consider population isolation further.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Measures of genetic diversity within individual R. reptans populations were positively associated with population size (Fig. 2). Log-transformed population size was correlated with both neutral genetic variation estimated from allozymes (Spearman correlation coefficient rs = 0.66, P < 0.05, n = 13 populations) and quantitative genetic variation calculated from the half-sib crossing design (mean h2 of eight quantitative traits: rs = 0.60, P < 0.05; mean Va: rs = 0.70, P < 0.01, Va standardized by [trait mean]2). Analysis of single traits revealed that two of the eight traits showed significant positive relationships between-population size and heritability: the number of stolons emerging from the primary rosette (rs = 0.66, P < 0.05) and the number of leaves of the primary rosette (rs = 0.60, P < 0.05). The corresponding relationship was positive but nonsignificant for three of the six remaining traits (rs = 0.30, 0.15, 0.14, −0.02, −0.04, −0.35), indicating that the overall result was caused by a combination of several characters. Neutral and mean additive genetic variation were weakly correlated with each other (rs = 0.50, P = 0.08). These results indicate that both kinds of within-population genetic variation are affected by small population size, and that neutral markers reflect to some extent the condition of additive genetic variation relevant for adaptive evolution.

image

Figure 2.  Relationships between population size and both gene diversity and mean additive genetic variance for eight phenotypic traits of 13 populations. Large populations of Ranunculus reptans had greater amounts of both neutral and quantitative genetic variation. Population size is surface area × plant density (estimated number of 5 × 5 cm cells with at least one rooted rosette, expressed in m2).

Download figure to PowerPoint

Estimates of between-population genetic variation revealed significant differentiation. The overall allozyme FST value of 0.0868 reflected significant marker divergence (P < 0.0001, one-sided test for deviation from zero based on 10 000 permutations of individuals and genes). Drift therefore acts to create moderate genetic structure in this metapopulation, under the assumption that allozymes are neutral and FST values reflect the magnitude of drift (Merilä & Crnokrak, 2001). Divergence among populations in seven morphological and life-history characters, or QST, averaged 0.2156 (P < 0.05, one-sided test based on 300 permutations of sires among populations). As reported by Fischer et al. (2000) for R. reptans, isolation-by-distance was not significant (both P > 0.1, geographical distance log-transformed). However, pairwise FST values, but not QST values, were lower between populations belonging to the same lake basin than between populations belonging to different lake basins (see Fig. 1, rMANTEL = 0.40, P = 0.0002 for FST; rMANTEL = 0.18, P > 0.1 for QST). Pairwise QST and FST were significantly positively correlated with each other (rMANTEL = 0.27, P < 0.05), so drift is likely to have contributed to the divergence of quantitative traits. Two separate traits showed a significant positive correlation with allozyme divergence: number of stolons emerging from the primary rosette (rMANTEL = 0.32, P < 0.05) and number of leaves of the primary rosette (rMANTEL = 0.29, P < 0.05). These are the same two traits that showed reduced heritability in small populations.

Divergence was greater among small populations than among large populations for both allozymes and quantitative traits. The correlation between FST and the harmonic mean of population sizes, for all possible pairs of populations, was rMANTEL = −0.36 (P < 0.001, partial Mantel test controlling for lake basin; Fig. 3a). This indicates that genetic drift acts more strongly in small populations to produce increased divergence. The corresponding relationship between average QST and mean population size was also significant (rMANTEL = −0.32, P < 0.01; Fig. 3a). Separate analyses of individual traits illustrated that the increase in QST among small population pairs arose primarily from two traits, the number of leaves of the primary rosette (rMANTEL = −0.33, P < 0.001) and width of the longest leaf (rMANTEL =−0.22, P < 0.05).

image

Figure 3.  Genetic differentiation between pairs of populations of Ranunculus reptans in relation to population size. (a) Genetic differentiation in both neutral markers (mean FST) and quantitative traits (mean QST, derived from maximum likelihood estimates of genetic variance components) depended on the harmonic mean of population sizes (measured as surface area × density). Small populations were more divergent phenotypically (QST) than at genetic markers (FST), indicating increased action of selection in small populations. (b) Between-population variances (Vpop) derived from a general linear model were higher for pairs of small populations, and additive within-population genetic variances (Va) were lower, leading to the pattern of QST values in panel a.

Download figure to PowerPoint

Average pairwise QST values were nearly twice as high as FST values (Fig. 3a; P < 0.0001; Monte Carlo simulation, 10 000 runs with the software PopTools; Hood, 2005) and higher differentiation in quantitative traits was maintained in small populations even after correcting for FST (partial correlation between QST and harmonic mean of population sizes, rMANTEL = −0.26, P < 0.05). This pattern was significant for two of the individual traits: the number of leaves of the primary rosette (rMANTEL = −0.27, P < 0.01) and the width of the longest leaf (rMANTEL = −0.23, P < 0.05). Three additional characters exhibited nonsignificant negative correlations (rMANTEL = −0.11, −0.07, −0.06, 0.00, +0.04), indicating that five traits combined to create the overall pattern. These results demonstrate that divergent selection in small populations increased quantitative population divergence more than expected due to drift alone. Figure 3b illustrates that this outcome was not influenced by our use of ML methods to estimate variance components: between-population variance declined with increasing population size (rMANTEL = −0.26, P < 0.01) whereas additive genetic variance within pairs of populations increased (rMANTEL = 0.36, P < 0.001). Thus, high QST values among pairs of small populations resulted from a combination of high between-population variance and low within-population variance. The results of Mantel tests were unchanged when we used residual values for pairwise QST and FST from analyses of variance including lake basin. Hence, the increase in quantitative divergence beyond the neutral expectation is not a consequence of isolation by distance.

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Our study of R. reptans at Lake Constance yielded several results with broad implications for evolutionary biology and conservation. First, we observed that small populations exhibited reduced within-population genetic variation in both allozymes and quantitative traits (Fig. 2). Both kinds of variation are subject to genetic drift and are predicted to decline in small populations (Kimura, 1955; Lynch & Hill, 1986). Empirical evidence for the impact of drift on neutral variation is very strong (Frankham, 1996; Spielman et al., 2004), but evidence for erosion of quantitative variation in natural populations has been scarce (Willi et al., 2006). The fact that R. reptans is tetraploid may influence our conclusions about the consequences of population size and drift for genetic variation. On the one hand, estimates of additive genetic variance are slightly inflated because the half-sib variance component contains part of the dominance variance in tetraploids (Kempthorne, 1969). But on the other hand, autotetraploids experience only about half the impact of genetic drift that diploids do (Ronfort et al., 1998), so if effects of drift on quantitative variation are detectable in tetraploids they should be even greater in diploid organisms.

Theory predicts that neutral and quantitative genetic variation should be correlated (Falconer & Mackay, 1996). Good support for this expectation comes from a recent Drosophila experiment comparing genetic variation in allozymes and two quantitative traits over a large range of population sizes (Gilligan et al., 2005). Our study found only a modest correlation between the two types of genetic variation, but many factors can weaken the relationship, including selection acting on quantitative traits (Houle, 1989; Reed & Frankham, 2001).

Our second main finding, that genetic differentiation in allozyme markers was correlated with that in quantitative traits, further indicates that drift is important in shaping quantitative traits. An earlier study of R. reptans found that three of the morphological and life-history traits we measured are under selection in the presence of competitors (proportion of flowering rosettes, number of branches per rosette and leaf length) (Fischer et al., 2004). Other traits included in our study, such as the timing of flowering and biomass of rooted rosettes, are likely to affect sexual or clonal fitness. Thus, it cannot be argued that the impact of drift on within-population genetic variation in quantitative traits occurred because these traits are simply neutral. Rather, drift was important in spite of the probable fitness consequences of phenotypic variation, and in the smallest populations drift may even counteract selection. One predicted consequence of strong genetic drift is an increase in fixed drift load, and indeed small populations of R. reptans are known to suffer from fixed drift load in female fertility (Willi et al., 2005).

The third key finding is that differentiation among small populations in quantitative traits was greater than that at neutral markers (Fig. 3). This means that drift alone cannot explain quantitative genetic differences among small populations. Our preferred interpretation is that some phenotypic divergence of small demes is due to drift, and this is indicated approximately by the magnitude of FST. The importance of drift is also indicated by patterns of within-population genetic variation, as noted above. But divergence beyond the neutral expectation is due to divergent selection, most likely resulting from the habitats occupied by small demes. After fragmentation, small populations often occupy marginal conditions and experience increased edge effects and ecological Allee effects (Soulé, 1986; Stephens & Sutherland, 1999), which can impose selection via changes in a variety of environmental conditions and species interactions (Laurance et al., 1998; Fagan et al., 1999; Ries et al., 2004). For example, a recent study of a woodland butterfly showed that lab-reared F2 offspring from a fragmented (agricultural) landscape differed in life-time fecundity from those in closed forest, depending on the rearing temperature (Karlsson & Van Dyck, 2005). This shows that altered biotic and abiotic conditions within fragmented habitats can cause selection and evoke adaptive responses.

In our study, enhanced quantitative genetic divergence among small demes suggests that each population experiences its own distinct selection regime. Quite a different result is expected if environmental conditions within each small population are similar (but different from conditions in large populations). In that case, small and large populations would differ in a consistent way, and maximum QST values would occur in pairs of populations of dissimilar size. Environmental conditions that could impose selection on these plants, and which vary around the shore of the Lake of Constance, include exposure to waves and steepness of the shore, density of interspecific competitors, and soil features such as gravel size and organic matter. Variation in these conditions could cause spatially divergent selection; indeed, previous studies of fine-scale differences within R. reptans populations have revealed adaptation to competition and flooding (Prati & Schmid, 2000; van Kleunen & Fischer, 2001; Lenssen et al., 2004). There is some indication that these conditions are especially distinct among the four smallest populations: two of these occur within a matrix of monocot species, one forms monospecific stands, and in the fourth the density of interspecific competitors is spatially variable.

Levels of genetic variation within and among populations can be influenced by gene flow in addition to drift and selection (Slatkin, 1973). Therefore, the reduced genetic variation that we observed within small populations, and enhanced divergence among them, could result in part from increased geographical isolation or interruption of dispersal among small demes. Small populations in our study area are not especially isolated, because there was no association between-population size and the distance to the nearest population. But it is possible that dispersal among small demes is limited, perhaps because pollinators are less likely to visit small populations or floating rosettes are less likely to strand in small populations. Reduced gene flow into small populations would have two consequences. First, the impact of drift would be accentuated, which would contribute to the high values of FST and QST observed in small populations. Secondly, the response to local selection (whether divergent or not) would be facilitated, which would inflate QST relative to FST in small populations. Thus, an alternative explanation of the pattern visible in Fig. 3a is that reduced dispersal increases the rate of adaptive evolution within small patches.

Three important insights for conservation emerge from this study. First, because the response to selection is proportional to the additive genetic variance of a trait (Falconer & Mackay, 1996), the low heritabilities for quantitative traits in small populations portend relatively low evolutionary responses to future environmental change. Secondly, our finding that genetic drift affects quantitative traits under selection implies that drift can influence or even counteract selection in small populations, so that genotypes are less well adapted to local conditions than are those in large populations. This reduces mean fitness in small populations below that caused by other genetic insults such as inbreeding (Willi & Fischer, 2005; Willi et al., 2005). Thirdly, we suggest that divergent selection is increased among small populations. At least in this instance, reduced population size is associated with increased divergent selection on phenotypic traits, in addition to its well-known link with drift.

It is sometimes recommended that isolated populations, identified by distinct phenotypes or neutral markers, should receive high priority for conservation (Ryder, 1986; Moritz, 1994; Crandall et al., 2000). Our results offer mixed support for this recommendation. On the one hand, we find that isolated populations that are also very small may be less closely adapted to local conditions, and less capable of responding to changing environments, than are large populations. Conservation efforts directed at such populations run the risk of protecting genetic variation favoured by idiosyncratic local selection regimes or produced by nonadaptive mechanisms. On the other hand, we find that small populations may be exposed to more diverse environmental conditions than large populations, and consequently may represent reservoirs of genetic variation adaptive within a potentially wide range of environments. In this case, conservation efforts targeted at small populations would protect individuals that are genetically and ecologically distinct and nonexchangeable.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

We thank B.R. Grant and M. Peintinger for discussions and advice. Thanks to S. Hoebee, R. Holderegger, B. Liebst and S. Röthlisberger for introduction to allozyme electrophoresis, and to D. Lang, R. Langenauer and E. Underwood for help with fieldwork. J. Bizozzero, E. Glaus, S. Käppeli, C. Kübler, D. Lang, S. Müller, S. Rahm, R. Scalone, V. Summa, R. Tuor, G. Vergnerie, A. Weidt, A. Willi, C. Willi, E. Willi and M. Zefferer gave technical assistance. Thanks to A. Hoffmann, D. Hosken, M. van Kleunen, M. Magrath and two anonymous reviewers for comments on the manuscript. We were supported by the Swiss National Science Foundation.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Appendix S1. Genetic variation at 8 allozyme loci of 13 populations of Ranunculus reptans. P, percentage of polymorphic loci; A, average number of alleles per locus; M, number of multilocus genotypes relative to the number of sampled plants; Ho and He mean observed and expected heterozygosity; F, mean fixation index; Hs, gene diversity; PS, population size (estimated number of 5 x 5 cm cells with at least one rooted rosette, expressed in m2). SD are given for A, Ho, He, and F; * P < 0.05 for paired t-tests of He and Ho for deviation from Hardy-Weinberg equilibrium, and one-sample t-tests for deviation of F from 0.

Appendix S2. Pairwise genetic distances FST (lower left triangle of the matrix) of 13 populations of Ranunculus reptans. Levels of significance are given in the upper right triangle of the matrix. Significance is based on 10 000 permutations (for individual and for gene), 2-sided testing over all loci, and Bonferroni-adjusted α' = 0.05 / 78 = 0.00064 ((*) P < 0.001, * P < 0.00064, ** P < 0.0001).

Appendix S3. The relationship between FST and heterozygosity for simulated neutral markers, under the following assumptions: island metapopulation model with 50 demes; 13 demes sampled, with an average of 12 individuals from each deme; infinite alleles mutation model; 20000 realizations. The simulation was implemented in program FDIST2 (Beaumont & Nichols, 1996). The solid line indicates the median FST and dashed lines depict the upper and lower 95% confidence interval. Filled circles are the observed values of FSTand heterozygosity for the eight allozyme loci. There is no evidence from these results that any of the loci is under homogenizing or divergent selection.

Appendix S4. Total variance attributed to population, paternal genotype nested within population, maternal genotype nested within population, and residual for eight phenotypic traits measured on offspring of within-population crosses of 13 Ranunculus reptans populations raised in a common garden (maximum-likelihood estimation implemented in VARCOMP procedure; SAS Institute Inc. 1999). Significance values are from a general linear model (Bonferroni-adjusted α = 0.0063, ? P < 0.05, ?? P < 0.01, * P < 0.0063, ** P < 0.001, *** P < 0.0001).

Appendix S5. Means of pairwise QST values based on seven phenotypic traits (lower left triangle of the matrix) of 13 populations of Ranunculus reptans and standard errors in the upper right triangle of the matrix.

FilenameFormatSizeDescription
JEB+1263sm.pdf121KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.