Single nucleotide variants in transcription factors associate more tightly with phenotype than with gene expression

Abstract

Mapping the polymorphisms responsible for variation in gene expression, known as Expression Quantitative Trait Loci (eQTL), is a common strategy for investigating the molecular basis of disease. Despite numerous eQTL studies, the relationship between the explanatory power of variants on gene expression versus their power to explain ultimate phenotypes remains to be clarified. We addressed this question using four naturally occurring Quantitative Trait Nucleotides (QTN) in three transcription factors that affect sporulation efficiency in wild strains of the yeast, Saccharomyces cerevisiae. We compared the ability of these QTN to explain the variation in both gene expression and sporulation efficiency. We find that the amount of gene expression variation explained by the sporulation QTN is not predictive of the amount of phenotypic variation explained. The QTN are responsible for 98% of the phenotypic variation in our strains but the median gene expression variation explained is only 49%. The alleles that are responsible for most of the variation in sporulation efficiency do not explain most of the variation in gene expression. The balance between the main effects and gene-gene interactions on gene expression variation is not the same as on sporulation efficiency. Finally, we show that nucleotide variants in the same transcription factor explain the expression variation of different sets of target genes depending on whether the variant alters the level or activity of the transcription factor. Our results suggest that a subset of gene expression changes may be more predictive of ultimate phenotypes than the number of genes affected or the total fraction of variation in gene expression variation explained by causative variants, and that the downstream phenotype is buffered against variation in the gene expression network.

Figure 1. Histogram of R-squared values obtained for the linear models describing the effect of genotype on the expression of individual genes.

The R-squared values obtained are on the x-axis and the numbers of gene expression models with the particular R-squared values are on the y-axis. A) Histogram of the R-squared values for all 5792 genes in the S. cerevisiae genome. B) Histogram of the R-squared values for the 289 significant gene expression models (inset). Significant models have an unadjusted model p-value≤0.006.

Figure 2. Fraction (%) of sporulation and gene expression variance explained by main (pink) and interaction effects (cyan) of all four sporulation QTN together.

Only the 289 ORFs with significant gene expression models are shown. The ORFs are ordered by fraction of total variance explained in the full model. Each column represents the amount of variation in gene expression explained for a given ORF. The last column represents the fraction of sporulation efficiency variance explained by the QTN. Only the significant ANOVA factors in both the sporulation efficiency and gene expression models were considered to calculate the fraction of variance explained by main and interaction effects (f-statistic p-value<0.1).

Figure 3. Fraction (%) of gene expression variance explained by main (pink) and interaction effects (cyan) of each of the four sporulation QTN.

The QTN effect on the 289 ORFs with significant gene expression models is shown. The ORFs are ordered by fraction of total variance explained in the full model. Plot includes only those models in which the fraction of gene expression variance explained by the particular QTN is greater than zero. Each column represents the amount of variation in gene expression explained for a given ORF. Only the significant ANOVA factors (f-statistic p-value<0.1) for each QTN were considered.

Figure 4. Histogram of total fraction (%) of gene expression variance explained by each QTN.

For each QTN, total fraction of gene expression variance explained (x-axis) is calculated by the sum of the significant main and interaction terms. The number of significant gene expression models with the given fraction is plotted on the y-axis. Only the significant ANOVA factors (f-statistic p-value<0.1) for each QTN were considered. The black line represents the fraction of the variation in sporulation efficiency that is explained by the given QTN (also listed in each figure).

Figure 5. Scatter plot comparing the total fraction of variance explained by the Ime1nc and Ime1c alleles.

a) Fraction of variance explained by main and interaction effects. b) Fraction of variance explained by main effects alone. For each QTN, fraction of expression variance explained is calculated by using the significant main and interaction terms of the QTN (f-statistic p-value<0.1). Results are shown for the 289 ORFs with significant gene expression models.