F2 yeast segregants datasets.

A) Reference panel and experiments from the barcoded bulk sequencing previously described in (29). The 99,995 F2 yeast segregants in the reference panel were derived from an F1 cross between a laboratory strain of yeast (BY) and a natural vineyard strain (RM). Thus, they only have 2 possible alleles at each of the 41,594 polymorphic sites. The lineage barcodes enabled fitness estimation from competition assays in 18 environments recapitulating the adaptation to temperature gradients, the ability to process different sources of carbon and the resistance to antifungal compounds. B) Pooled scRNA-seq dataset from a single batch. We performed scRNA-seq of the first batch of F2 segregants (n=4,489) to obtain genotypes that are similar to the reference panel and cell’s expression profiles. The F2 segregant barcodes are typically not expressed, so single cells have their own cell barcode, which is included in all amplified reads of the same droplet. Non-covered sites, sequencing errors and the presence of reads in the wrong library (index swapping) are corrected for using the HMM described in Figure S1.

scRNA-seq dataset features.

The doublets were defined as scRNA-seq barcodes with high coverage (>=third quartile) without significant genotype association to any F2 segregants and for which the 2 most similar F2 segregants do not share genotypic similarity (R2 < 0.1). The breadth of coverage is defined as the proportion of BY/RM polymorphic sites covered in a single cell.

Single-cell RNA-seq data recapitulate bulk DNA and RNA assays results.

A) Effect of the HMM on the relatedness between single-cell genotypes and their closest reference lineage. The single-cell original genotype represents the genotype of the cells before the correction with the HMM. The relatedness to the closest lineage in batch 1 has been measured with the adjusted R2. To control for genotype uncertainty, only the 13,069 barcodes with a significant lineage assignment (lineage-barcode genotype correlation FDR<0.05) and a reference lineage with lower uncertainty than the single cell HMM are selected, which represents 72.2% of the barcodes. We then rounded the genotypes to remove the uncertainty during the comparison. Wilcoxon signed test p-value is indicated above the violin plots. B) Narrow-sense heritability measured with scRNA-seq consensus genotypes and non-multiplexed DNA sequencing. Three datasets are compared through a t-test on the bootstrap narrow-sense heritabilities. The consensus genotypes of the batch 1 F2 segregants obtained from the associated scRNA-seq data are represented in blue, the batch 1 F2 segregant genotypes in the reference dataset (bulk sequencing) are represented in yellow and the whole reference dataset of F2 segregants is represented in grey. For each dataset, the narrow-sense heritability has been measured from 500 random subsamples of 1000 F2 segregants. The bars represent the mean narrow-sense heritability, and the error bar length represents the 95% CI. The results are illustrated for the 15 environmental conditions where a minority of batch 1 F2 segregants have missing fitness data. The 23C-30C represents the temperature for the competition assay in YPD media while the other phenotypes represent growth on YNB, molasses (mol), mannose (Mann) or raffinose (raff) and chemical resistance to copper sulfate (Cu), ethanol (eth), guanidinium chloride (gu), lithium acetate (Li), Sodium dodecyl sulfate (SDS) and suloctidil (suloc) (29).

Variance partitioning of the 30C phenotype from scRNA-seq.

We fitted F2 segregant fitness data to their cells’ consensus genotype and expression profiles using GTCA-GREML at different sample sizes. Because each segregant had multiple representatives/cells in the scRNA-seq data, the consensus profile of these lineages has been obtained from the cell with the highest transcript level. The bars represent the proportion of the fitness/phenotype variance across F2 segregants (whole rectangle area) explained by each model component. The size of the red genotype-exclusive component is obtained by subtracting the size of the shared component from the variance explained by the model in Equation 2. The red cell illustrates this effect as a mutation changes its fitness by modifying the structure of the transcript product. The size of the blue expression-exclusive component is obtained by subtracting the size of the shared component from the variance explained by the model in Equation 3. An epigenetic modification increasing the transcript level of the gene in the blue cell illustrates this effect. The purple area represents the effect of the interaction of mutation and expression on fitness variation observed across the segregants. In the example shown, a mutation in the promoter changes the fitness of the purple cell through a down-regulation of the gene expression. This component is obtained by subtracting the variance explained by the model in Equations 1 or 2 from the one in Equation 4, as there is a shared component of trait variation explained by genotype and expression variation. Finally, the black bar represents the model’s residual, which includes any phenomenon that causes fitness variation that we did not explicitly measure and include in the model. The size of the residuals is obtained by subtracting the sum of the size of all the defined components (red + blue + purple) from 1.

eQTL features underlying trait variation across the BY/RM segregants.

A) Mapping of the 30C QTL in the eQTL hotspots. We represent the hotspots of expression regulation as genomic windows (25 kb) to acknowledge the uncertainty around the real position of the eQTL due to linkage disequilibrium. We annotated the 5 top eQTL hotspots and the eQTL hotspots in which the top additive QTL identified by the BB-QTL mapping of the 30C phenotype are located. In these regions, we represented the most affected trans-regulated genes in red, the most affected cis-regulated gene in blue and the top QTL genes in black. The double quotation characters represent the absence of such genes in the associated region. We also represented the rank of the QTL in the set of 159 QTL of the 30C phenotype. B) Partitioning of the expression heritability or explained variance (R2) among cis– and trans-eQTL. Each pair of points connected by a line represents a gene. Green lines represent the genes that only have trans-eQTL and orange lines represent the genes that have both trans– and cis-eQTL. C) Comparison of the mean effect size between cis– and trans-eQTL. Each pair of points connected by a line represents a gene. The line color represents the ratio of the average effect size between cis– and trans-eQTL. The sample size of each eQTL category is represented in the x-axis. This is the number of trans-eQTL and cis-eQTL used for calculating the average effect sizes per gene not the number of points per distribution.