Figures and data
Experimental setup.
(A) Location of the six field sites and origin of the 200 accessions. Note that sites NA and SR were used in both types of experiments. (B) Heatmap showing kinship between the accessions, hierarchically clustered by similarity. Marginal colors indicate membership in one of four genetic groups—the same colors are used for the sampling locations in panel A. Numbers are FST estimates between groups. (C) Schematic of the experiments. In the common-garden experiments, which were replicated over two consecutive seasons, established seedlings were transplanted into the field in fall (using trays with holes in the bottom so that plants would root in native soil), and overwinter survival and fecundity assessed the following spring. In the evolution experiments, seeds were sown directly on the ground in equal numbers in early fall 2011 and surviving descendants sequenced before flowering in 2013.
The distribution of accession survival probabilities across sites and years.
Numbers in plots give total mortality in each experiment.
The joint distribution of accession survival probabilities between the three experiments with significant mortality.
The curves were fitted using the S1 accessions only.
Overwinter survival was affected by herbivory.
Left: Slug damage in the SR 2011-12 experiment affected groups differently (Kruskal-Wallis test: p < 0.01 for all comparisons). Right: Average slug damage for an accession decreased its probability of survival (ANOVA: p = 1.6 × 10–23).
Quantile-quantile plots of p-values for overwinter survival GWAS against the expected uniform distribution.
Left plot without correction for structure; right plot including a standard mixed-linear model kinship correction. Only SNPs with Minor Allele Frequency (MAF) greater than 5% were included to avoid outlier affects. Experiments with significant mortality exhibit genome-wide inflation of significance, while remaining experiments demonstrate that tails are inflated even for noise phenotypes.
Slug damage was associated with glucosinolate profiles.
Left: Slug damage in the SR 2011-12 experiment affected the glucosinolate profiles defined in Katz et al. (2021) differently (data for 122 of our accessions; Kruskal-Wallis test: p < 5 × 10–13). Right: The difference in mortality between groups (cf. Figure 4) is associated with glucosinolates profiles, with the putatively protective “allyl” phenotype present in 100% of B and N accessions, 79% of S1 accessions, but only 22% of S2 accessions.
Variance-partitioning (ANOVA) from full model of fecundity.
The top bar zooms in on the explained variance from the complete partitioning in the bottom bar.
The loadings of the first three PCs from a PCA of the fecundity BLUPs on the eight experiments.
The distribution of fecundity BLUPs for each experiment by group.
The distribution of fecundity BLUPs for each group by year and site (cf. Figure 9).
GWAS of rosette purpleness in SU 2011-12 identifying Production of Anthocyanin Pigment 2, (PAP2, AT1G66390), located on chromosome 1 between 24,763,941 and 24,765,541 bp.
A. Genomen-wide Manhattan plot B. Zoom-in on the 100 kb window around PAP2, represented by the orange bar.
Quantile-quantile plots for GWAS fecundity data.
Left plot without correction for structure; right plot including a standard mixed-linear model kinship correction. The extreme confounding for SR 12 reflects the unusually large difference between S1/S2 and B/N in this experiment (Figure 9). Only SNPs with MAF greater than 5% were included to avoid outlier affects.
Scatter-plots comparing estimated accession frequencies in the four selection experiments.
Accessions that are potential natives in a site are indicated by color and the two obvious outliers (Bar1 in NB, and Vår2-6 in ST) are highlighted. Figure 13—figure supplement 1. Same as Figure 13, but with the two obvious outliers removed to show remaining data better.
The distribution of the estimated relative fitness by genetic group (same data as in Figure 13 but rescaled by dividing by the expected frequency in the absence of selection).
The distribution of seed size (data from Clauw et al., 2022) by genetic group, and, on the right, mean fitness (across sites, from Figure 14) as a function of seed size.
Mean fitness as functions of various phenotypes related to seedling establishment.
A. Root length, from Slovak et al. (2020). B. Same as A, but corrected for seed size, which was positively correlated with early root growth. C. Hypocotyl elongation. D. Primary seed dormancy, from Kerdaffrec et al. (2016).
Seed size distribution for 246 additional accessions sampled from 9 sites in southern Sweden in 2017 (see Methods).
The top 6 sites were located on beaches and the bottom 3 were located inland. The experimental B accessions hail from the top four sites.
Left: results of selection scan for chromosome 1. Right: scatter plot of p-values from selection scan (all chromosomes) vs. seed size GWAS.
Accessions plotted in the simplex space generated by the admixture proportions.
Sowing and installation date for common-garden experiments.
Blocks A–C were sown with two-day intervals.
Workflow for classifying selection experiment samples.
Isolation-by-distance in Swedish A. thaliana.
On the left, sample locations and the decay in pairwise identity as a function of distance using the data of Platt et al. (2010a); on the right, the same plots for the new samples from Skåne (Figure 17). The blue dots show the sites used in the evolution experiments.
Top: the distribution of ςp in the four selection experiments. Middle: the variance of ςp as a function of p0(1 – p0). Bottom: the distribution of ςp scaled by its standard deviation. The curves are PDFs of normal distributions with the observed mean and variance 1.
