Report

A Map of Local Adaptation in Arabidopsis thaliana

See allHide authors and affiliations

Science  07 Oct 2011:
Vol. 334, Issue 6052, pp. 86-89
DOI: 10.1126/science.1209271

Abstract

Local adaptation is critical for species persistence in the face of rapid environmental change, but its genetic basis is not well understood. Growing the model plant Arabidopsis thaliana in field experiments in four sites across the species’ native range, we identified candidate loci for local adaptation from a genome-wide association study of lifetime fitness in geographically diverse accessions. Fitness-associated loci exhibited both geographic and climatic signatures of local adaptation. Relative to genomic controls, high-fitness alleles were generally distributed closer to the site where they increased fitness, occupying specific and distinct climate spaces. Independent loci with different molecular functions contributed most strongly to fitness variation in each site. Independent local adaptation by distinct genetic mechanisms may facilitate a flexible evolutionary response to changing environment across a species range.

Adaptation to local environments has been observed experimentally in many organisms (1) and may critically limit a given species’ capacity to evolve in the face of rapid environmental change (24). However, the molecular basis of local adaptation remains largely unexplored (5, 6). Understanding the genetic mechanisms of adaptation requires understanding the genetic basis of fitness variation within and across natural environments (7, 8). Although genome scans for signatures of past selection have identified candidate loci showing high levels of environmental differentiation (9, 10), few studies have directly connected fitness variation measured in the natural environment of the species to the corresponding molecular variation (11, 12). Determining the extent of local adaptation requires identification of the loci associated with individual fitness in different natural environments, as well as characterization of the distribution pattern of adaptive variants, their environment specificity, and the type of genes involved.

To identify loci associated with fitness in the annual plant Arabidopsis thaliana, we grew a geographically diverse set of ecotypes (inbred lines derived from natural populations) across their native range, in replicated common garden experiments in four European field sites (fig. S1). Sites in Oulu (Finland) and Valencia (Spain) spanned the species climate range limits from Nordic to Mediterranean environments; sites in Halle (Germany) and Norwich (UK) represented continental and oceanic climates at similar mid-range latitude (13). Mean survival and lifetime fruit (silique) production differed markedly among ecotypes within each planting site, indicating heritable variation among source populations in viability and fecundity (table S1). We carried out a genome-wide association study (GWAS) for survival and silique number using 213,248 single-nucleotide polymorphisms (SNPs) in a mixed-model approach to eliminate confounding due to genomic background (14, 15). For each fitness trait in each of the four field sites, we defined a set of associated SNPs corresponding to the 0.05% of the SNPs that explained the most variance (around 100 per GWAS; figs. S2 and S3). Individual SNPs explained a substantial amount of variation in lifetime fitness, with R2 in GWAS models ranging on average from 9% for the SNP set associated with survival in England to 24% for the SNP set associated with survival in Finland (fig. S1).

We tested whether alleles associated with high fitness in a given site were more locally abundant than genomic controls, as expected if they contributed to the local adaptation of that population (16). Indeed, the geographic centroids of the alleles associated with higher survival in England and Spain and silique number in Germany, England, and Spain were significantly closer to the planting sites in Germany, England, and Spain, respectively, relative to genomic controls; this constitutes evidence of local adaptation for specific loci (Fig. 1). Similar analyses excluding low-frequency polymorphisms provided similar results, demonstrating that the result was not biased as a result of the presence of rare alleles (table S2). In contrast, we found no evidence that the alleles conferring high fitness in Finland were locally abundant. However, our ability to detect locally adapted alleles near the species’ northern and southern range limits may have been limited by sample size in Finland (16) as well as by the sparse sample of local ecotypes available for our sites in Finland and Spain (fig. S1).

Fig. 1

Locations of the centroids of alleles associated with increased silique number for the 0.05% most strongly associated SNPs in Spain or Finland (left) and in England or Germany (right). Large squares represent the location of the planting site; small squares represent the centroid of the allele increasing the fitness in the planting site of similar color. The higher-fitness alleles in a particular environment are distributed significantly closer to that location than are genomic controls, as tested by 10,000 random draws.

We next examined whether local alleles associated with fitness are associated with specific climatic factors, as expected if climate variation has strongly contributed to adaptation. We used maximum entropy models [MaxEnt (17)] to infer the contributions of 11 bioclimatic variables to the distribution of each SNP allele (table S3) (16) and tested whether fitness-associated alleles occupy specific climate spaces relative to genomic controls. We then used a random resampling procedure to compare climate specificity scores for each fitness-associated SNP for each climate variable to genomic controls. Excluding rare alleles (frequency <5%; table S4) had little effect on the results. The distribution of alleles associated with survival was particularly limited by temperature variables (fig. S4 and table S3). In some cases, the alleles associated with high fitness exhibited greater climate specificity (e.g., temperature of the coolest quarter for survival in Germany, Finland, and Spain; table S4). However, more often alleles associated with low fitness exhibited greater climate specialization (e.g., to precipitation in the coolest quarter for survival and to temperature in the warmest quarter for silique number) (Fig. 2 and table S3). Such alleles may be effectively neutral within the specific climate conditions where they occur, but deleterious when transplanted to field sites with different climates.

Fig. 2

Influence of temperature and precipitation on the distribution of alleles linked to fitness variation. (A) Variation in temperature during the warmest quarter of the year. (B) Variation in precipitation during the coolest quarter of the year. (C to F) Box plots of the difference in climatic contribution between pairs of SNP alleles significantly associated with fitness in Germany, England, Finland, and Spain. Negative values indicate when the deleterious alleles are more correlated with the climate variable; positive values indicate when the beneficial alleles are more correlated with the climate variable. Light and dark shades represent the 25% and 50% quantile statistics of genome-wide neutral expectation obtained through permutation (16). *P < 0.004, **P < 0.0001. All test statistics are reported in table S2.

We also investigated whether locally adaptive alleles tended to be rare or common relative to the overall genome. If selection favors specific alleles in restricted geographic regions, we expect those variants to be common only in those regions and otherwise rare. On the other hand, if local adaptation acts largely by culling locally deleterious alleles that are neutral elsewhere in the range, then we might expect high-fitness alleles to occur at relatively high frequency. We therefore tested whether the global frequency of the alleles associated with high fitness was different relative to genomic controls. Alleles associated with increased silique number were significantly rarer than genomic controls (table S5), which suggests that their advantage was local for all sites except Finland (where no evidence of local adaptation was found). In contrast, alleles improving survival were high-frequency alleles, which suggests that viability selection keeps deleterious alleles rare, confining their genetic load to environments where they have no effects (i.e., are neutral; fig. S5). We also tested whether fitness-associated SNPs were involved in recent selective sweeps by computing the integrated extended haplotype homozygosity (iEH) around each SNP (18, 19), both in the global sample and in the local population for each field site (16). With the exception of survival in Germany and marginally in Finland, we found no evidence that loci associated with fitness had significantly greater iEH than genomic controls (table S5), which suggests that no recent selective sweeps have occurred at these loci. Thus, environment-specific fitness may depend more on standing variation than on recent positive selection at specific loci, as recently observed in aspen (20).

We examined the genetic architecture of local adaptation by determining whether local selection in different environments more commonly acted upon alternate alleles of the same loci, or if entirely different loci were found among different environments. Individual SNP additive effects for survival were weakly negatively correlated between Finland and England or Spain, which suggests cross-environment tradeoffs at specific loci (fig. S6) (16). In contrast, SNP effects on survival were weakly positively correlated between England and Spain, which suggests that they share a similar genetic basis for fitness. However, all the observed correlations across sites were relatively weak, and loci associated with fruit production in England showed little to no effect on fitness in either Finland or Spain; such findings suggest an environment-specific genetic basis. Overall, among the 797 top SNPs, only 12 SNPs were associated with fitness in more than one environment; loci with major genetic effects were largely independent across sites. This result suggests that local adaptation relies on different loci in each environment. Such independent local adaptation by different genetic mechanisms may facilitate flexible evolutionary response to changing environments across the species range.

We sought insight into the molecular functions of the genes contributing to local adaptation by testing whether the candidate genes present in 10,000–base pair (10-kbp) windows surrounding the associated SNPs were enriched for a particular subset of the Gene Ontology (GO slim) annotations relative to 1000 genomic permutations (16). The loci associated with fitness displayed significant enrichment for specific molecular functions in all sites (fig. S7). However, no molecular function was significantly over- or underrepresented in all sets of candidate genes (table S6). Typically, only transcription factor and protein binding activities showed significantly different enrichment for the same trait in more than one site. Protein binding was overrepresented in genes associated with survival in Germany and underrepresented in England, whereas transcription factor activity was underrepresented in both England and Spain. Overall, those molecular functions that were significantly over- or underrepresented relative to the whole genome always differed among field sites. This result suggests that local selection acts not only on different target loci but also on different molecular processes in different environments across the species climate range.

To identify potential candidate genes for local adaptation, we searched for fitness-associated SNPs in the top 0.05% (with minor allele frequency >8% to exclude rare deleterious mutations) that also appeared in the top 0.5% of allele climate differentiation and in the top 5% of global iEH (Table 1). This screen identified four SNPs within four candidate genes, and linked within 10 kbp to eight additional candidate genes. The three most interesting candidate genes with the associated SNP situated in the coding region were the LAC1 gene (AT1G18140), involved in the response to necrotrophs (21), associated with silique number in Germany; CHR8 (AT2G18770), involved in DNA repair after viral infection (22), associated with survival in Germany; and SAG21 (AT4G02380), involved in water stress tolerance (23, 24) and associated with survival in Finland. Note that for both CHR8 and SAG21, the allele showing iEH evidence of a selective sweep corresponds to the minor and geographically restricted allele. The CHR8 allele associated with low survival in Germany is established in the species northern range, and the SAG21 allele associated with low survival in Finland is restricted to the southern part of the species range (Fig. 3). Recent positive selection may have favored these alleles locally, but their geographic range may be restricted by negative pleiotropic effects on fitness in other environments, such as our field sites.

Table 1

List of candidate genes associated with fitness in field condition.

View this table:
Fig. 3

Geographic distribution probability of the survival-associated alleles located within the SAG21 gene (left) and the CHR8 gene (right). Probabilities were calculated with MaxEnt models as described (16). For both genes, the minor allele is distributed at the species range margin following a particular climate space and shows signs of recent positive selection. They correspond to the best candidate genes for local adaptation reported in this study.

This study provides robust molecular evidence for broad scale local adaptation in A. thaliana. By investigating SNPs linked to loci experiencing real-time selection in different natural environments, we found that the genetic basis of fitness differs dramatically across sites. Moreover, alleles associated with high fitness within sites tend to be local alleles linked to particular climatic factors, providing evidence of local adaptation in A. thaliana at the scale of the European continent. Our finding that loci associated with fitness in natural environments display geographic and climatic signatures of adaptation complements recent observations of candidate loci associated with climate (10). GWAS of fitness traits measured in multiple natural environments, combined with geographic and climatic analyses, constitutes a powerful approach to identify environment-specific candidate genes for local adaptation.

Supporting Online Material

www.sciencemag.org/cgi/content/full/334/6052/86/DC1

Materials and Methods

Figs. S1 to S7

Tables S1 to S6

References

References and Notes

  1. See supporting material on Science Online.
  2. Acknowledgments: We thank M. Blázquez, C. Dean, M. Hoffmann, H. Kuittinen, and O. Savolainen for hosting the experiments; J. Anderson, D. Eaton, J. F. Egan, C. Lopez-Gallego, L. J. Martin, C. D. Muir, R. Petipas, J. Roe, and R. N. Schaeffer for managing the field experiments; Brown undergraduates for fruit counts; and J. Bergelson, A. Hancock, and J. Mossman for comments on the manuscript. Supported by NSF grant EF-0425759 and an Alexander von Humboldt Research Award. Supplemental data for the phenotypes and the associated SNPs are available at http://dx.doi.org/10.5061/dryad.37f9t. Genotype data for the SNPs used in this study are available at http://regmap.uchicago.edu.
View Abstract

Navigate This Article