Disrupting upstream translation in mRNAs is associated with human disease

Abstract

Ribosome-profiling has uncovered pervasive translation in non-canonical open reading frames, however the biological significance of this phenomenon remains unclear. Using genetic variation from 71,702 human genomes, we assess patterns of selection in translated upstream open reading frames (uORFs) in 5'UTRs. We show that uORF variants introducing new stop codons, or strengthening existing stop codons, are under strong negative selection comparable to protein-coding missense variants. Using these variants, we map and validate gene-disease associations in two independent biobanks containing exome sequencing from 10,900 and 32,268 individuals, respectively, and elucidate their impact on protein expression in human cells. Our results suggest translation disrupting mechanisms relating uORF variation to reduced protein expression, and demonstrate that translation at uORFs is genetically constrained in 50% of human genes.

Fig. 1. Stop-introducing and stop-strengthening variants in translated uORFs are under strong negative selection.

a Examples of possible stop-gained (UTC upstream termination codon) or stop-strengthened variants in translated uORFs. b Points showing mutability-adjusted proportion of singletons (MAPS) scores for different classes of stop-introducing variants in translated uORFs. Gray, orange, and purple dashed lines represent MAPS scores for synonymous, missense, and predicted loss-of-function (pLOF) SNVs affecting canonical protein-coding sequences in gnomAD. (i) MAPS scores for non-uORF variants including all 5′UTR SNVs, stop gained variants in ncORFs in 3′UTRs, lncRNAs, and pseudogenes), and all 5′UTR stop gained variants. (ii) MAPS scores for all uORF SNVs and stop gained variants in uORFs show that uORF UTC variants are significantly enriched for singletons. This is also observed for UAA-creating, and stop-strengthening SNVs in translated uORFs. Error bars represent bootstrapped 90% confidence intervals. c Relative frequencies of trinucleotides used as uORF stop codons compared to untranslated regions of uORF-containing 5′UTRs, or all 5′UTRs shows uORFs are significantly enriched for weaker (UGA, UAG) stop codons and depleted of the UAA stop codons compared to control sequences. Mean proportion over n = 10,000 random samples with replacement are represented by columns. Error bars represent 95% bootstrapped confidence intervals. d Points representing the proportion of strongly conserved (phyloP >2) bases by phyloP scores from 100-way vertebrate alignments for uORF stop-creating, non-uORF stop-creating in uORF-containing UTRs, and non-uORF stop-creating in all UTR genomic positions. Error bars represent 90% bootstrapped confidence intervals.

Fig. 2. uORFs do not exhibit strong selective pressure to maintain amino acid identity.

a Points representing MAPS scores for single nucleotide variants within each ncORF category separated by predicted consequence (synonymous or missense) in each ORF. (i–iv) Allele frequencies for predicted missense SNVs are not significantly enriched for singletons than those for predicted synonymous SNVs. (v) MAPS scores are no different for long uORFs (>118 codons) compared to the rest (short). Gray, orange, and purple dashed lines represent MAPS scores for synonymous, missense, and predicted loss-of-function (pLOF) SNVs affecting canonical protein coding sequences in gnomAD. Error bars represent bootstrapped 90% confidence intervals. b Translated uORF variants ranked by predicted change to codon optimality using codon stability coefficient (CSC) scores from SLAM-seq (red = decreasing, blue = increasing). Gray dotted line denotes boundary separating optimality increasing versus decreasing SNVs. c Points representing MAPS scores for SNVs separated by predicted consequence on codon optimality shows heightened constraint against decreasing optimality variants, while variants increasing optimality are indistinguishable from all 5′UTR variants. Error bars represent bootstrapped 90% confidence intervals.

Fig. 3. Selective pressure to preserve uORF start codons.

a (i) Distribution of start codon usage for experimentally mapped translated uORFs and (ii) possible consequences of variants affecting uORF start codons. b Points representing (i) MAPS scores for start-disrupting SNVs are compared to uORF variants matched by trinucleotide mutation context. (ii) Start-disrupting SNVs for short (<20 codons) uORFs are under stronger negative selection compared to start-disrupting variants for long (≥20 codons) uORFs. Error bars represent bootstrapped 90% confidence intervals. c Points representing proportion of conserved bases for possible start codon disrupting positions in uORFs compared to all uORF SNVs, UTR-matched start-disrupting positions, and uORF-matched start-disrupting positions in translated uORFs. Start-disrupting genomic positions of short uORFs are more strongly conserved by phyloP scores compared to matched start-disrupting positions within uORFs. Error bars represent bootstrapped 90% confidence intervals.

Fig. 4. Phenome-wide association study (PheWAS) of predicted stop-strengthening variant in a translated uORF in PMVK.

PheWAS plot of translated uORF stop-strengthening variant in the 5′UTR of PMVK (N = 65 carriers) in the Penn Medicine BioBank. ICD-9 and ICD-10 Phecodes are organized and plotted by category on the X-axis. Each point represents a single Phecode. The height of each point corresponds to the −log₁₀(P value) of the association between the variant and Phecode using a logistic regression model adjusted for age, age, sex, and the first ten principal components (PCs) of genetic ancestry. The solid red line represents the threshold for Bonferroni-adjusted significance (P = 6.25e–6) and the red dashed line represents the FDR < 0.10 threshold (P = 1.25e-4) accounting for multiple comparisons. The direction of each arrowhead corresponds to increased risk (up) or decreased risk (down).

Fig. 5. Reporter gene assays for translated uORF stop-introducing and stop-strengthening variants.

Dual-luciferase reporter assay quantifies relative expression for uORFs with UTC and stop-strengthening variants associated with EHR phenotypes by PheWAS. Experimental 5′UTRs for a PMVK, b VPS53, and c BCL2L13 for uORF KO, stop-strengthened, or stop-introduced variants are shown. Bars represent co-transfected Firefly to Renilla Luciferase luminescence ratios normalized to associated wild-type 5′UTRs in HEK293T cells measured 48 h post-transfection. P values from one-sample t-test for each condition are displayed above each column. Error bars represent mean + SEM of at least three independent experiments. Raw luciferase measurements are included in the Source Data file.