Fig. 3: Function scores capture fitness effects secondary to splicing alterations and impairment of protein function.

a, Targeted RNA-sequencing of VHL mRNA from edited cells was performed to calculate RNA scores for n = 1,626 exonic SNVs. The relationship between RNA scores and function scores reveals most LoF variants are expressed at normal levels in mRNA, and only low RNA scores reliably predict LoF at the cellular level (Extended Data Fig. 4c). Two synonymous variants shown independently to disrupt splicing are indicated^23,57, as well as c.462A>C, the splice region variant with the lowest RNA score. b,c, The maximum SpliceAI score for each SNV is plotted against RNA scores for exonic SNVs (Pearson’s R = −0.70; b) and function scores for intronic SNVs (R = −0.90; c). d, The proportion of missense variants scoring as depleted by SGE (q < 0.01) is displayed for each residue exposure label. e, FoldX ΔΔG predictions were higher for n = 242 depleted missense SNVs (median = 3.63) than for n = 697 nondepleted SNVs (median 0.70) (boxplot: center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range). f–h, The average score of missense variants at each amino acid position is shown in color on the VHL structure (PDB: 1LM8)⁵⁸. Residues highly intolerant to missense variation include S111, H115, W117 and W88 at the HIF1A recognition site (g), as well as L158 and C162 at the ELOC interface (h).