Massively parallel functional testing of MSH2 missense variants conferring Lynch syndrome risk

Abstract

The lack of functional evidence for the majority of missense variants limits their clinical interpretability and poses a key barrier to the broad utility of carrier screening. In Lynch syndrome (LS), one of the most highly prevalent cancer syndromes, nearly 90% of clinically observed missense variants are deemed "variants of uncertain significance" (VUS). To systematically resolve their functional status, we performed a massively parallel screen in human cells to identify loss-of-function missense variants in the key DNA mismatch repair factor MSH2. The resulting functional effect map is substantially complete, covering 94% of the 17,746 possible variants, and is highly concordant (96%) with existing functional data and expert clinicians' interpretations. The large majority (89%) of missense variants were functionally neutral, perhaps unexpectedly in light of its evolutionary conservation. These data provide ready-to-use functional evidence to resolve the ∼1,300 extant missense VUSs in MSH2 and may facilitate the prospective classification of newly discovered variants in the clinic.

Keywords: DNA mismatch repair; Lynch syndrome; MSH2; cancer; deep mutational scanning; genotype-phenotype; variants of uncertain significance.

Figure 1

Overview of MSH2 functional screen (A) Generating and testing MSH2 variant function in isogenic human cells. Cells with pathogenic loss-of-function variants (top) are resistant to 6-thioguanine (6-TG), while complementation by a benign functional variant (middle) restores 6-TG sensitivity. In a pooled screen (bottom), a mixed library of MSH2 missense variants is transduced, and 6-TG treatment selects for loss-of-function variants. (B) 6-TG resistance reflects MMR function in human cells. HAP1 MSH2 KO cells were transduced with either functional (wild-type) MSH2 cDNA (upper row) or loss-of-function missense variant p.Ala636Pro (bottom row), and grown for 11 days without (left) or with 6-TG (1 μM, right). (C) Readout using deep sequencing to quantify abundances of each MSH2 allele before and after 6-TG selection. The resulting loss-of-function (LoF) score is positive for deleterious variants and negative for functionally neutral ones.

Figure 2

Functional effect map of MSH2 missense variants (A and B) MSH2 secondary structure (cyan: alpha helices, pink: beta sheets) (A) and protein domains (B). (C) Heatmap of loss-of-function (LoF) scores after mutating each of 934 residues in MSH2 (rows) to each of 19 other possible amino acids (columns). Negative scores (shaded blue) indicate functionally neutral variants, while positive scores (red) indicate deleterious ones; yellow denotes wild-type amino acid; gray, no data. (D) Fractions of substitutions which are disruptive (LoF score > 0) by position. (E) Number of scored missense variants by position (max. possible: 19). (F) Evolutionary conservation (PSIC score) by position. (G) Distributions of LoF scores, shaded by variant class.

Figure 3

Concordance with existing functional and clinical classifications (A) Measured LoF scores (y axis) for published MSH2 missense variants (n = 295 reports covering 184 distinct variants), separated by previous characterization (x axis) in individual cellular or biochemical assays. (B) LoF scores for patient missense variants from ClinVar and InSight databases, by variant classification. (C) Distributions of LoF scores of MSH2 missense variants in the gnomAD database, for variants which do not appear in the homozygous state (gray) or those which do in ≥1 individual (orange). (D) LoF score versus gnomAD allele frequency for MSH2 missense variants; size of circle denotes number of heterozygous variants; orange squares mark missense variants observed in homozygous state in ≥1 individual. Gray region denotes singletons/doubletons (gnomAD allele count ≤2).

Figure 4

Comparison with bioinformatic predictions of pathogenicity (A) Precision-recall curves show LoF scores outperform all tested bioinformatic callers in predicting the functional status of the validation set of variants with known effect; area under the curve indicated for each. (B) Concordance of each bioinformatic predictor (Jaccard Index) with experimental LoF scores is plotted versus algorithm-specific score threshold for deleteriousness. Algorithms were most concordant with LoF scores among mutations with published functional data (red), followed by other mutations at the same residues as those mutations (orange) and mutations all other residues (gray). Dotted line and score indicate optimal threshold for each predictor on known the validation set.

Figure 5

Structural insights from deep mutational scanning data (A) MutSα crystal structure (PDB: 2O8E), with MSH6 surface in light gray and MSH2 shaded by the number of deleterious missense variants at each residue (length of bar in legend proportional to count of residues with denoted range of LoF mutations). (B) Extreme intolerance to substitution in MSH2 ATPase domain (upper) and connector domains (lower); selected residues or beta sheets shown with spheres or arrows, respectively. (C) LoF score heatmap (as in Figure 2C) showing mutational intolerance at Walker A motif of the ATPase domain (upper) and selected connector domain residues (lower) corresponding to (B).