Research Article

Selective trade-offs maintain alleles underpinning complex trait variation in plants

See allHide authors and affiliations

Science  03 Aug 2018:
Vol. 361, Issue 6401, pp. 475-478
DOI: 10.1126/science.aat5760

Fluctuating selection in nature

Natural environmental variation can lead to individuals within a species experiencing different selective parameters. Seep monkeyflower (Mimulus guttatus) populations are constrained by local moisture availability and the onset of summer drought. This results in a selective tradeoff between the amount of seed set, which is determined by plant size, and the timing of reproduction. Troth et al. sequenced and phenotyped 187 M. guttatus plants and identified genetic variants associated with plant and flower size and rapid flowering. In wild populations surveyed over 3 years, the magnitude of selection changed depending on the rainfall patterns. Thus, fluctuating selection may maintain genetic variation in this species.

Science, this issue p. 475


To understand evolutionary factors that maintain complex trait variation, we sequenced genomes from a single population of the plant Mimulus guttatus, identifying hundreds of nucleotide variants associated with morphological and life history traits. Alleles that delayed flowering also increased size at reproduction, which suggests pervasive antagonistic pleiotropy in this annual plant. The “large and slow” alleles, which were less common in small, rapidly flowering populations, became more abundant in populations with greater plant size. Furthermore, natural selection within the field population favored alternative alleles from year to year. Our results suggest that environmental fluctuations and selective trade-offs maintain polygenic trait variation within populations and also contribute to the geographic divergence in this wildflower species.

Understanding the selective forces that act on polygenic traits and the genetic basis of that variation will help us predict how native species may evolve in response to climatic changes. Surveys of individual genetic variants within and among populations allow us to test whether variation is caused by rare mutations that are eliminated by selection, or whether balancing selection is important in maintaining variation. One such complex trait, adult body size, is determined by genetic variants at dozens to perhaps thousands of loci (14). Larger body size is often correlated with greater fecundity (5). This suggests that variants that decrease size should be rare because of selection, but simple correlations between size and fecundity are often misleading (68). In fact, selection might maintain high genetic variation within populations as well as polygenic adaption among populations (9). Identifying genetic variants that contribute to phenotypic variation makes it possible to estimate population allele frequencies and to distinguish balancing selection from purifying selection (10). Genomic information can also identify the role of pleiotropy between different fitness components (11), how causal variants affect fitness in native environments (1214), and whether fluctuating selection due to environmental heterogeneity is preserving genetic variation (15, 16)—features essential to predicting population responses to environmental change.

In western North America, annual populations of Mimulus guttatus grow rapidly in the spring, racing to flower and set seed before they are killed by the summer drought. Size at reproduction is a highly variable, polygenic trait that is genetically correlated with time of first flower (8, 17). Flower size is a function of overall plant size at reproduction. Delayed flowering is correlated with large, highly fecund flowers, whereas rapid flowering is correlated with small flowers that set fewer seeds but are more likely to successfully reproduce before the summer drought. This antagonistic pleiotropy has been demonstrated for large genomic regions [quantitative trait loci (QTLs)] within the Iron Mountain, Oregon, USA (IM) population in both laboratory and field experiments (8, 18, 19). In contrast, perennial populations of M. guttatus that inhabit permanently wet habitats have delayed development and produce larger, more fecund flowers. Reciprocal transplant experiments demonstrate local adaptation and divergent selection on these life history traits between annual and perennial populations (20). Within IM, transplant experiments suggest that although large, late-flowering genotypes have increased pre-reproductive mortality, they can occasionally increase fecundity (via seed set) when flowering is permitted by a delayed onset of drought (18, 19). However, these prior studies were unable to distinguish the effects of closely linked loci on the genetic control of trait variation within or among populations.

Trait-associated polymorphisms exhibit antagonistic pleiotropy and characteristic allele frequencies

To identify DNA sequence polymorphisms that contribute to variation in the IM population, we generated 187 highly inbred lines and then randomly crossed pairs of inbred lines. We sequenced whole genomes of all 187 inbred lines, identified more than 10 million single-nucleotide polymorphisms (SNPs) and small insertions or deletions (indels), and inferred F1 genotypes. For both inbred lines and outbred F1 plants, we measured flower size, plant size, and days to flower under randomized and controlled greenhouse conditions. We used Fisher’s combined probability statistic to aggregate signal across inbred lines and outbred F1 plants for the same polymorphisms (21). This resulted in 45 distinct polymorphisms that were genome-wide significant for at least one trait (threshold determined by permutation, P < 0.05; tables S1 and S2). We thinned these to 24 distinct loci by collapsing closely linked and/or perfectly associated sites [linkage disequilibrium r2 = 1 in the line population; see (22) for treatment of interlocus associations]. These 24 polymorphisms exhibited antagonistic pleiotropy (Fig. 1A), with alleles that delay flowering also increasing plant size (flower size PC1 is an aggregate of floral dimensions). We also observed a relationship between phenotypic effect and population allele frequency (Fig. 1B). Nearly all minor alleles (those with a frequency of less than 0.5) increased trait values.

Fig. 1 Trait-associated alleles exhibit strong associations.

(A) The effect of the reference allele (base or indel) on days to flower is a strong positive predictor of its effects on flower size (blue) and plant height (green). (B) The effect of the minor allele at each locus on days to flower (black), height (green), and flower size (blue) is plotted against frequency in the lines. Here, effects are standardized by the standard deviation of the trait.

We also interrogated the inbred lines for structural polymorphisms (major deletions, duplications, or inversions) relative to the reference genome (23). Although there were far fewer of these (~120,000), a substantial number exhibited association with phenotypes (table S3). The general pattern mirrors the results from SNPs, where the minor allele at each locus almost invariably increased trait values (Fig. 1). There was only one significant structural polymorphism where the reference allele was less frequent than the alternative (a 663–base pair deletion on chromosome 9), and this was the only case where the reference allele increased plant or flower size. Together, SNPs and structural polymorphisms illustrate the pleiotropic trade-off between size at reproduction and reproductive timing.

Trait-associated polymorphisms respond to selection on flower size

We independently validated our trait-associated variants with artificial selection by selecting for increased and decreased flower size for nine generations in populations derived from more than 1000 plants collected from IM (17). This resulted in “High” and “Low” populations that differed in mean flower size by ~10 standard deviations. After sequencing pooled samples from the High and Low population [i.e., Poolseq (24)], we ascertained SNPs identified from our inbred line sequencing. Given that IM was the source of variation in this selection experiment, we predicted that alleles increasing flower size, as identified in the genome-wide association study (GWAS), should be inflated in frequency in the High population relative to the Low population (25). Because the allele frequencies are estimated with error due to finite read depth (median read depths are 71 and 76 at ascertained SNPs in Low and High, respectively), we expanded the trait-associated variant set to include 1567 SNPs and small indels with a trait association test of P < 10−5 in the greenhouse GWAS, hereafter the “10−5 set.” The 10−5 set undoubtedly includes some polymorphisms that do not affect flower size (false positives). However, this makes our tests conservative because neutral SNPs will not contribute to directional differences in allele frequency between populations.

The selected populations corroborated our greenhouse estimates in two important ways. First, minor alleles, which are typically associated with late flowering and large flower size (Fig. 1B), increased in frequencies in the High population relative to the Low (mean ΔminorHL = 0.012, t1340 = 4.73, P < 0.0001). Here, ΔminorHL is the difference in the frequency of the minor allele between High and Low across 1342 sites in the 10−5 set that are polymorphic in the combined High and Low population. At these sites, the increase in minor allele frequency substantially elevates the mean expected heterozygosity Embedded Image in the High population (mean = 0.1383, SE = 0.0034) relative to both the inbred line (mean = 0.1227, SE = 0.0029) and Low population (mean = 0.1172, SE = 0.0035). The elevated heterozygosity at flower size loci helps to resolve the paradoxical finding that selection increased not only the mean but also the genetic variance of flower size in the High population (17). Although directional selection should ultimately deplete genetic variation, short-term increases are possible if the direction of allelic effects is negatively associated with minor allele frequency (as in Fig. 1B). Many sites in the 10−5 set affect flower size only weakly, and if we restrict our focus on the subset of sites with P < 10−5 for flower size PC1, the divergence in allele frequency is greater (mean ΔminorHL = 0.021, t103 = 2.82, P = 0.006, n = 105).

We also tested whether the estimated phenotypic effect of the trait-associated alleles predicts the difference in allele frequency between High and Low populations. Here, ΔrefHL denotes the difference in reference base frequency (High – Low) to which effects are attributed. Using ΔrefHL as the response variable in a linear regression, we found that the GWAS-estimated effect of an allele on flower size PC1 is a significant positive predictor of ΔrefHL (F1,1340 = 9.69, P = 0.002), as is the effect on plant height (F1,1340 = 14.99, P < 0.0001). Complementary to the observed changes in minor alleles, the analysis based on phenotypic effects indicates that alleles estimated to affect flower size and plant size in the GWAS responded as expected to directional selection. This corroboration across independent experiments is not trivial. Large GWAS of Drosophila melanogaster have failed to replicate directional allelic effects of the same SNPs across populations (26), perhaps owing to pervasive genetic background effects (epistasis).

Minor alleles associated with large size and late flowering are more frequent in M. guttatus populations with larger plants

The polygenic adaptation hypothesis (9) predicts that the large, late-flowering alleles that segregate at lower frequencies within the IM population should be more frequent in M. guttatus populations that have longer growing seasons and larger average plant size. We estimated allele frequencies at trait-associated sites between simultaneously collected wild plants from IM and the nearby larger-flowered Quarry population. Each was assayed in a multiyear population resequencing experiment (27), and we combined samples from 2013 and 2014 within each population. We found that 1477 SNPs of the 10−5 set are polymorphic in the IM and Quarry (combined) data, with median read depths of 187 and 163 for IM and Quarry, respectively. These loci exhibited allele frequency divergence (Fig. 2), with minor alleles (associated with large size and late flowering in IM) more common in Quarry than in IM (mean ΔminorQ,IM = 0.099, SE = 0.005, t1475 = 18.3, P < 0.0001).

Fig. 2 SNPs and small indels with a trait association test of P < 10−5 in the greenhouse GWAS as ascertained in pooled samples from two natural populations, IM and Quarry.

(A) Density plots for the difference in frequency of the minor base (within the inbred line set) between Quarry and IM. Solid red curve, distribution for trait-associated sites; dashed black curve, distribution for all sites. (B) Allelic effect on flower size PC1 (in the greenhouse experiment) is a positive predictor of allele frequency divergence between Quarry and IM.

The comparison of ΔminorQ,IM values at trait-associated polymorphisms with the background distribution (Fig. 2A) reveals that trait-associated loci are more different between populations than the typical SNP. FST outlier tests (28) use this as the signal of local adaptation (FST is a measure of divergence among populations). Here, we see that the deviation between distributions is highly asymmetric, driven by a much higher frequency of sites with large positive ΔminorQ,IM among trait-associated sites relative to the genome as a whole (Fig. 2A). This pattern is reinforced by considering the estimated phenotypic effects of GWAS loci. The effect of the reference allele on flower size (PC1) is a strong positive predictor of ΔrefQ,IM (Fig. 2B; F1,1475 = 218, P < 0.0001). The more positive the effect of an allele on flower size in the IM GWAS, the larger the difference in this allele’s frequency in Quarry relative to IM. This is the signature of polygenic adaptation: Alleles that promote later flowering and greater size at flowering are at higher frequencies in slower-maturing, larger-flowered populations.

IM polymorphisms cause fitness variation and experience fluctuating selection in nature

The divergence between natural populations, in terms of both trait means and allele frequencies at trait-associated loci, is indicative of adaptation. To determine how selection acts on trait-associated loci within IM, we grew replicates of the same F1 lines used in the greenhouse GWAS experiments in their native habitat across 3 years (which corresponds to three full generations for these annual plants). We monitored plants for various flowering time and size traits, survival to seed set (viability), fecundity of survivors, and total seed count (a measure of lifetime female fitness) at the end of each season. Allelic effects were determined on each fitness component for loci with effects on flower size PC1 (10−5 set) in the greenhouse (Fig. 3). A midseason drought yielded high mortality in 2015, but 2014 and 2016 provide a clear contrast of selection regimes. Selection favored small, rapidly flowering alleles in 2014 primarily because of their greater survival to reproduction prior to the onset of summer drought. In contrast, late onset of drought in 2016 allowed increased survival of plants with a lower frequency of large, late-flowering alleles, and selection favored them as a result of their greater fecundity.

Fig. 3 Effects of minor frequency alleles on fitness components vary between years.

The estimated effects of the minor allele on (A) viability, (B) log(seed set of survivors), and (C) total seed set (dead included) is plotted against allele frequency for plants in 2014 (blue) and 2016 (green). These are loci from the 10−5 set that exhibited effects on flower size PC1.

As with the population resequencing data (e.g., Fig. 2), we performed analyses classifying polymorphisms by allele frequency and then according to phenotypic effect. For each polymorphism tested in the F1 plants, we calculated the mean effect of the minor allele as the difference in mean fitness between the heterozygote and the major homozygote. We ignored homozygotes for the minor allele here because very few were present in the field plants. We focused on SNPs with an effect on flower size PC1 from the greenhouse GWAS (those in Fig. 3). In 2014, the mean effect of the minor allele was –0.181 on viability and –6.12 on total seed (68 polymorphisms). In 2016, the corresponding values for viability and total seed were 0.045 and 21.3, respectively (73 polymorphisms). We tested whether these average effects were significantly different from zero by permuting F1 means for fitness components against F1 genome sequences within each year (22). The positive effect of minor alleles on fitness (total seed) in 2016 (Fig. 3C) is significant (P = 0.0008). The negative effect of minor alleles on viability in 2014 (Fig. 3A) is marginally nonsignificant (P = 0.11). We then tested for heterogeneity in selection by treating the change in mean effect (2016 versus 2014) as the test statistic. The change in viability effect (0.212) is marginally nonsignificant (P = 0.07), but the pronounced change in average minor allele effect on total seed (26.8) is highly significant (P = 0.004).

Considering the phenotypic effects, we regressed the estimated effects of alleles on a fitness component onto their respective estimated effects on flower size PC1 (Fig. 4). We included polymorphisms that have minimal effects on flower size because these loci provide the intermediate values for the predictor (allelic effect on flower size) in the regression. Using the same permutation procedure, we found that phenotypic effect estimates from the greenhouse GWAS are significant positive predictors of total seed in 2016 (P = 0.03; Fig. 4D). Although the apparently negative relationship in 2014 is not significant (Fig. 4B), the significant (P = 0.04) change in the slope (2016 versus 2014) demonstrates fluctuating selection across years.

Fig. 4 The effects of all trait-associated polymorphisms (P < 10−5 in the greenhouse GWAS) on survival and total seed set.

(A and B) Selection in 2014 (negative on flower size). (C and D) Selection in 2016 (positive on flower size). Viability is the response variable in (A) and (C), total seed in (B) and (D). The subset of polymorphisms that are significant for flower size PC1 are red. Here, effects are attributed to the reference allele at each polymorphism.

Evolutionary implications

The measurement of both inbred lines and outbred F1 plants in the greenhouse suggests that trait-associated loci have positively correlated effects on size (plant and flower) and reproductive age (Fig. 1A). In the IM population, early flowering and high fecundity both tend to increase fitness; the trade-off demonstrated by these alleles thus yields antagonistic pleiotropy. The correlations among size and life history traits previously demonstrated at the whole-genome level (8, 17) and at the QTL scale (18, 19, 29) are now mapped to sequence-level variants. The trade-off within IM extrapolates further to divergence in life history and morphology among annual and perennial populations throughout the species range.

We also find that alleles associated with large size and late flowering are nearly always less common in the IM population than the alternative alleles (Fig. 1B). This finding is consistent with the idea that selection often, but not always, favors plant genotypes that flower earlier, despite having lower seed set. Additionally, these same large, late-flowering alleles are more abundant in a neighboring population of M. guttatus with greater average plant size (Quarry versus IM in Fig. 2). The geographic comparison is straightforward because nearly all SNPs present in IM are also segregating in Quarry (27). Thus, divergence in allele frequency predicts the divergence in mean phenotype among populations; this is similar to the divergence of human adult height, where “tall” GWAS alleles are relatively more abundant in northern European than in southern European populations (30).

The divergence between Quarry and IM does not indicate how polymorphisms are maintained within IM. Antagonistic pleiotropy could maintain intrapopulation polymorphism, although a proper balance of positive and negative effects is required (11, 31). Alternatively, stabilizing selection might act on flowering time within IM; in that case, polymorphism would reflect a balance between selection and mutation and/or migration from large-flowered populations. However, the most simplistic migration or mutation models are not sufficient to explain polymorphism within IM. The minor alleles associated with late flowering and large size in IM are not genuinely rare when considered in relation to population size. A typical year at IM has at least 300,000 flowering adults, and thus an allele with a frequency of ~5% (typical of minor trait-associated alleles) is present in at least 30,000 copies (32). This is an excessively large number for mutation-selection balance, and migration-selection balance would require a large number of immigrants each generation to offset directional selection. However, high migration is inconsistent with the relatively high differentiation of allele frequencies among populations of M. guttatus. Across its geographic range, M. guttatus exhibits FST values near 0.5. Neighboring populations exhibit lower but still substantial differentiation: FST ≈ 0.1 at the scale of a few kilometers (33). Deleterious alleles can be inflated by a recent population bottleneck, but the high nucleotide diversity within IM argues against this possibility (34).

The 2016 field data (Figs. 3 and 4) provide the most direct evidence against directional selection favoring alleles that cause rapid flowering and small size within IM. In many years, a large fraction of plants never make it to flower (8, 18). However, 2016 had an atypically long growing season. Perhaps because of this, the average fitness of large, late-flowering alleles was significantly greater than that of their small, rapidly flowering alternatives. In the more typical 2014, selection favored alleles associated with early flowering and small flowers. The high-frequency alleles that decrease size within IM (Fig. 1B) may reflect the historical regularity of shorter seasons such as 2014. However, the 2016 data suggest that environmental fluctuations might preserve variation at the nucleotide and trait levels by allowing large, late-flowering alleles to persist at much higher frequencies than expected from migration- or mutation-selection balance. More generally, our synthesis of population genomic, phenotypic, and fitness data from the wild connects evolutionary processes responsible for both geographical variation and the persistence of high genetic variation within local populations of M. guttatus. The selective trade-offs evident for our trait-associated variants suggest that these alleles will respond to selection as climate change alters the initiation and duration of natural growing seasons.

Supplementary Materials

Materials and Methods

Tables S1 to S5

References (3557)

References and Notes

  1. See supplementary materials.
Acknowledgments: We thank C. Wessinger, W. Yuan, T. Mitchell-Olds, M. Rausher, R. Unckless, K. Brown, L. Flagel, G. Coop, A. Settlekowski, P. Monnahan, and S. Macdonald for input on the project and/or manuscript. We thank M. Zamora, R. Fitzpatrick, and C. Friesen (U.S. Forest Service) for assistance in data collection, script writing, and site access, respectively. Funding: Supported by NIH grant R01 GM073990 (J.K.K. and J.H.W.); NSF grants NPGI-IOS-1202778 (J.R.P.), IOS- 1354688 (J.H.W.), and DDIG-DEB-3320249 (J.H.W. and A.T.); and the NSF Graduate Research Fellowship Program (A.T.). Author contributions: A.T., J.H.W., and J.K.K. designed the experiments; A.T., J.R.P., R.S.K., and J.K.K. made the libraries and directed sequencing; A.T. and J.K.K. conducted the analyses; and A.T., J.H.W., and J.K.K. wrote the paper. Competing interests: Authors declare no competing interests. Data and materials availability: The sequence data for all the inbred lines are available at the NCBI Sequence Read Archive (SUB4115987, PRJNA344904) as accessions SAMN09387101 to SAMN09387290.

Stay Connected to Science


Navigate This Article