QTL mapping

Our dataset consisting of genotypes (B or W, corresponding to the BY and W303 parental background respectively) at 8505 loci (columns) of 258 segregants (rows) and their ploidy phenotype after evolution (binary data, haploid = 0, diploid = 1) was used as the input for QTL analysis using R/qtl v1.46-2 software as described below (following Broman and Sen 2009). Before QTL mapping, a battery of diagnostic probes, involving a test for segregation distortion of the markers and an analysis of anomalous genotyping similarity and number of crossover events for the segregants, were checked to avoid spurious mapping (see Supplementary Text S1 for details). This resulted in a clean dataset consisting of genotypes of 255 segregants at 8475 loci with their corresponding phenotypes, which then entered the following QTL mapping pipeline.

First, we computed LOD scores for all 8475 loci assuming the presence of a single QTL using standard interval mapping and the Haley–Knott regression method for a binary phenotype with LOD significance thresholds computed from 10000 permutations. Next, to find any potential interactions between multiple QTLs, we divided our data into predictor and test datasets. We chose 150 segregants arbitrarily to form a predictor dataset and subjected their genotype and phenotype data to a forward/backward stepwise search algorithm (stepwiseqtl) with LOD significance thresholds computed from 1000 permutations. Based on the LOD score profile of single-QTL analysis above (see Results) we restricted this search to chromosomes IV and XIV only, and the maximum number of QTLs allowed in a model was kept to 4. Subsequently, we fitted the predicted QTL model onto the remaining data consisting of 105 segregants (test dataset) using fitqtl followed by the refineqtl function.

Furthermore, to reveal any additional low-effect QTL for the autodiploidization phenotype, we rescanned the data using single-QTL analysis methods after regressing out the QTL with highest LOD score obtained above. Effect sizes of the two alleles of the QTL with statistically significant LOD score were estimated using the effectplot function.