Extensive and coordinated control of allele-specific expression by both transcription and translation in Candida albicans

Abstract

Though sequence differences between alleles are often limited to a few polymorphisms, these differences can cause large and widespread allelic variation at the expression level. Such allele-specific expression (ASE) has been extensively explored at the level of transcription but not translation. Here we measured ASE in the diploid yeast Candida albicans at both the transcriptional and translational levels using RNA-seq and ribosome profiling, respectively. Since C. albicans is an obligate diploid, our analysis isolates ASE arising from cis elements in a natural, nonhybrid organism, where allelic effects reflect evolutionary forces. Importantly, we find that ASE arising from translation is of a similar magnitude as transcriptional ASE, both in terms of the number of genes affected and the magnitude of the bias. We further observe coordination between ASE at the levels of transcription and translation for single genes. Specifically, reinforcing relationships--where transcription and translation favor the same allele--are more frequent than expected by chance, consistent with selective pressure tuning ASE at multiple regulatory steps. Finally, we parameterize alleles based on a range of properties and find that SNP location and predicted mRNA-structure stability are associated with translational ASE in cis. Since this analysis probes more than 4000 allelic pairs spanning a broad range of variations, our data provide a genome-wide view into the relative impact of cis elements that regulate translation.

Figure 1.

Sensitively detecting ASE at translational level with ribosome profiling. (A) Schematic of the approach. For a given gene with two SNPs, transcripts from the B allele may be more abundant, whereas translation favors the A allele, as indicated by increased density of ribosomes, shown in green. These biases are revealed by RNA-seq and ribosome profiling, respectively. Allele-specific reads are summed across all SNPs in the gene, and translational efficiency (“TE”) is calculated from the mRNA and footprint (“FP”) levels. (B,C) Signal is consistent across many SNPs. There are 17 distinct SNP windows in orf19.169/CHO2, and the majority indicates a translational bias toward the B allele, but roughly equal transcript levels (B), with little error across SNPs (C); error bars, ±SEM. (D) The sum of allele-specific reads (red and blue bar) matches the level of nonspecific reads that do not include SNPs (gray bar) for orf19.169; error bars, ±SEM. (E) Across all genes, the fraction of SNP-containing reads corresponds strongly to the fraction of gene length comprised of SNP-containing windows.

Figure 2.

A total of 4.2% of genes show translational allelic bias. (A) Schematic of the bootstrapping procedure. (Top) For a mock gene containing a single SNP, ∼30 consecutive positions contain allele-specific information, and three scenarios for read distributions are shown: (#1) shows reproducible bias toward allele A in blue; (#2) shows how a single position could suggest a bias that is not supported by other positions, and (#3) shows a consistently reported lack of bias. (Middle) For each of 5000 iterations, 30 positions are selected randomly and with replacement, and a TE value is calculated from the mRNA and FP reads from those positions. (Bottom) The results are tabulated into a histogram, where the mean and standard deviation of the bootstrap distribution reflect the magnitude and confidence, respectively, of allele-specific bias in TE. (B) Scatterplot and accompanying histograms (top and right) showing the bootstrap means and standard deviations for the 3285 genes with at least five reads for mRNA_A, mRNA_B, FP_A, and FP_B (shading indicates the metric with the fewest read counts, as shown in the legend at bottom). Blue-rimmed circles indicate genes that pass the 5% FDR threshold.

Figure 3.

ASE at translational level is as strong as at transcriptional level. (A–C) Scatterplots of the respective allelic levels of mRNA (A), FP (B), and TE (C). (D) Histograms of mRNA_ALFD, FP_ALFD, and TE_ALFD, where gray dotted lines indicate the two-standard-deviation boundary of error distributions from biological replicates (Supplemental Fig. S1). (E–G) Top panels show 2D heatmap PDFs of observed data from A–C, and bottom panels depict predicted data from simulation; cool and warm colors indicate lowly and highly populated bins, respectively. (H) Simulated data in E–G was used to plot histograms of mRNA_ALFD, FP_ALFD, and TE_ALFD.

Figure 4.

Biases in transcription and translation are often coordinated, with interactions favoring compensation over reinforcing. (A,B) Specific examples of compensatory (A) and reinforcing (B) interactions between transcription and translation. Genes with compensatory interactions have higher allelic difference at the mRNA level than at the FP level, whereas those with reinforcing interactions differ more at the FP level than at the mRNA level. (C) Scatterplot of mRNA_ALFD and TE_ALFD levels, where genes from A and B are indicated as darkened circles, and shaded regions indicate compensatory and reinforcing interactions, as indicated. The purple and green regions’ curved portions reflect the two-standard-deviation spread of the data along the y = x and y = −x lines, respectively, and straight segments are based on a heuristic chosen to ensure that both the mRNA_ALFD and TE_ALFD values are nonzero (see Methods). (D,E) PDFs indicating the number of reinforcing (D) and compensatory (E) genes from permuted data; arrows indicate the number of genes from the observed data.

Figure 5.

mRNA structure stability near start codon and SNP positioning near termini correlate with translational ASE bias. (A) Genes with high TE_ALFD are shaded in blue and red, and those with low TE_ALFD are shaded gray. (B) Allelic disparities in codon bias are not different among the gene sets in A. (C) Scatterplot of TE_ALFD versus the difference in predicted folding energy of the 60-nt window surrounding the start codon for all genes with at least one SNP in the window. Shading indicates regions that are at least one standard deviation away from zero on each axis, with purple regions representing the expected relationship between structure stability and TE, and gray indicating the unexpected relationship. (D) Pie chart quantifying the number of genes falling in each region demarcated in C. (E) PDF of SNP density as a function of position for genes with high (green) or low (gray) TE_ALFD (see Methods); the dashed line shows the uniform distribution. (F,G) PDFs indicating the sum-squared deviation from the uniform distribution from permutation analyses for genes with high (F) and low (G) TE_ALFD (see Methods); observed values indicated by black arrows.