Figure 4. Quantitative trait locus (QTL) and bulk segregant mapping reveals two QTL underlying coefficient of variation (CV) of germination time.

(A, B) Manhattan plots showing the QTL association results for each single nucleotide polymorphism (SNP) marker individually (black line) and for each marker when the Chr3 QTL SNP marker was added as a covariate (i.e. an additional variable) in the model (orange line). The orange line shows the variation in CV that is accounted for by each SNP across the genome when the variation that is explained by the Chr3 QTL SNP marker (orange point) is accounted for by adding it to the model as a covariate. The y-axis shows the p-values for the 1254 markers used, on a negative log₁₀ scale, such that higher peaks indicate a stronger association between the region of the genome and CV. The numbered panels represent the five chromosomes of Arabidopsis. The horizontal dashed line shows a 5% genome-wide threshold corrected for multiple testing (based on simulations in Kover et al., 2009). The vertical dashed line indicates the DOG1 gene. (A) is for the full set of 341 MAGIC lines that was phenotyped and (B) excludes the eight bimodal lines with very high CV. Figure 4—figure supplement 1 shows QTL mapping for mean and mode days to germination and percentage germination. Figure 4—figure supplement 2 shows estimated effects of accession haplotypes on CV, mode and percentage germination. (C) Mapping QTL by bulk-segregant analysis using whole-genome pooled sequencing of F2 pools from a Col-0 × No-0 cross. One early and two late germinating F2 pools were sequenced. The plot shows the No-0 allele frequency differences between pairs of pools indicated in the legend (Figure 4—figure supplement 3 shows details of pool selections; E1, "early pool"; L1, "late one pool"; L2, "late two pool"). The horizontal dashed lines indicate the 95% thresholds based on simulating the null hypothesis of random allele segregation, taking into account the size of the sampled pools and the sequencing depth at each site (Magwene et al., 2011; Takagi et al., 2013). Positive values above the top line indicate enrichment for No-0 alleles, while negative values below the bottom line indicate enrichment for Col-0 alleles. As predicted, late germinating pools were enriched for the No-0 haplotype in the region of the Chr5 QTL. Here, the peak of association overlaps with the DOG1 gene (dashed vertical line). Figure 4—figure supplement 4 shows germination phenotypes of F3 seeds from Col-0 x No-0 F2 plants that themselves germinated early or late.

Figure 4—figure supplement 1. Quantitative trait locus mapping for germination traits, with and without bimodal MAGIC lines.

(A) is for all 341 MAGIC lines that were phenotyped, and (B) is for 333 of these lines (the full set minus the eight bimodal lines with very high coefficient of variation [CV]). The y-axis shows the p-values for the 1254 markers used, on a negative log₁₀ scale. The numbered panels represent the five chromosomes of Arabidopsis, and the scan was performed for four germination traits: CV of germination time, mean germination time, mode germination time and percentage germination. The horizontal dashed line shows a 5% genome-wide threshold corrected for multiple testing (based on simulations in Kover et al., 2009).

Figure 4—figure supplement 2. Accession-specific quantitative trait locus (QTL) effects on coefficient of variation (CV), mode and percentage germination.

(A) Predicted accession effects at the two putative QTL on Chr3 and Chr5 (Figure 4). The effects of the 19 parental accession haplotypes were estimated by calculating the mean CV of MAGIC lines inferred to carry each particular haplotype. (B) Correlation between predicted QTL effects on percentage germination and CV. (C) Correlation between predicted QTL effects on mode days to germination and CV. In all panels, the mean effect of each parental accession’s QTL allele was estimated from the probabilistic assignment of each MAGIC line to that founder parent (Kover et al., 2009). Error bars in (A) show the 95% confidence intervals of these estimates (these were omitted from the other panels for clarity). All trait values were standardised, so that axis units represent the number of standard deviations away from the respective mean, with the horizontal and vertical dashed lines at zero highlighting the mean of the respective trait in the population. Pearson’s correlation, r, is indicated in each panel in B and C, with the 95% confidence interval in brackets.

Figure 4—figure supplement 3. Germination time distributions and DNA-seq pools of Col-0 × No-0 F2.

Eight batches of Col-0 × No-0 F2 seeds, each containing ~1100 seeds and collected from a different F1 parent plant, were sown on soil. Seeds from parental accessions Col-0 and No-0 were also included in the experiment, and, for these, batches of ~1100 seeds pooled from three parent plants were sown for each accession. The different F2 batches behaved similarly, so here we present an averaged (mean) distribution based on bulked data. The percentage germination each day is a percentage of all seeds that germinated (rather than of all seeds that were sown). (A) shows the full germination time distribution with pools used for DNA sequencing highlighted, and (B) shows days 10–60, with a different y-axis scale to show the late germinating seeds. The 'early pool' used for sequencing was composed of 152 individuals that germinated on day 4; the ‘late 1’ pool was composed of 321 individuals that germinated between days 31 and 39, and the ‘late 2’ pool was composed of 213 individuals that germinated between days 43 and 60. We reasoned that, since late germination is predominantly restricted to the more variable parent (No-0), late germinating F2 plants should be enriched for the No-0 accession at loci promoting high variability (including the Chr5 locus at ~20 Mb, where the No-0 haplotype is predicted to promote high CV). Figure 4—figure supplement 3—source data 1 contains the source data for (A) and (B).

Figure 4—figure supplement 4. Germination phenotypes of F3 seeds from Col-0 × No-0 F2 parent plants that themselves germinated early or late.

Coefficient of variation (CV) (i), mean (ii), mode (iii) and percentage germination (iv) for F3 seed batches collected from F2 plants that themselves germinated at different times (x-axis). Batches of seeds from F2 plants that germinated late (between days 30 and 60) had, on average, a significantly higher CV than seeds from plants that germinated early (on day 4) (the mean CVs of the two groups were 0.19 in the early group versus 0.27 in the late group, Wilcoxon rank-sum test W = 199, p-value=1.163e-05, n = 23 seed batches from early germinators, n = 47 seed batches from late germinators). Seeds of late germinating plants did not tend to have higher mean or mode germination times (the mean of mean germination times was 3.05 days in the early group versus 3 days in the late group; the mean of mode germination times was 3.00 in the early group versus 2.78 in the late group). Percentage germination shows a small but significant difference between the two groups of F3 seeds (mean percentage germination: 95.3 in the early group versus 87.67 in late group, Wilcoxon rank-sum test W = 754, p-value=0.00419, n = 23 seed batches from early germinators, n = 47 seed batches from late germinators). Each seed batch is from a separate F2 plant. Figure 4—figure supplement 4—source data 1 contains the source data for all panels.

Figure 4—figure supplement 5. Germination time distributions and abscisic acid (ABA) dose responses in quantitative trait locus candidate gene mutants.

(A, B) dog1-3 mutant in the Col-0 background, for seeds that were 5 days after harvest (DAH) (left panels) or 30 DAH (right panels). Histograms show germination time distributions for replicate seed batches from separate parent plants (shades of grey) of each genotype. The mean of the coefficient of variation of germination time for the different seed batches is displayed on the panels (along with the mean across batches of percentage germination, mode days to germination and mean days to germination). (B) shows an ABA dose response for the wild type and mutant, with three replicate seed batches (dots) for each genotype. The lines join the mean percentages of germination for each treatment for a particular line. (C, D) As for (A, B), but for a dog1 T-DNA mutant in the No-0 background. Here the No-0 control represents wild-type plants obtained from the segregating T-DNA population (see Materials and methods for details). The number of days of seed storage before sowing is shown above each panel. (E, F) As for (A, B), but for the ahg1-5 mutant in the Col-0 background. (G, H) As for (A, B), but for the anac060 mutant in the Col-0 background. In all experiments at least three separate replicate seed batches from separate parent plants were sown for each genotype, except in C, for 24 DAH, where only 2 replicates were used. In the histograms shown in (A) for 5 DAH and in (G) for 3 DAH, five and four separate experiments, respectively, were performed (each with three seed batches). In these cases, each shade of grey is the germination time distribution from a single experiment, with the behaviours of its three separate seed batches pooled together for plotting, treated as one sample. For all other histograms, one experiment was performed and shades of grey are separate replicate seed batches harvested and sown in that single experiment. Source data is provided in Figure 4—figure supplement 5—source data 1.