Predicting the impact of Lynch syndrome-causing missense mutations from structural calculations

Accurate methods to assess the pathogenicity of mutations are needed to fully leverage the possibilities of genome sequencing in diagnosis. Current data-driven and bioinformatics approaches are, however, limited by the large number of new variations found in each newly sequenced genome, and often do not provide direct mechanistic insight. Here we demonstrate, for the first time, that saturation mutagenesis, biophysical modeling and co-variation analysis, performed in silico, can predict the abundance, metabolic stability, and function of proteins inside living cells. As a model system, we selected the human mismatch repair protein, MSH2, where missense variants are known to cause the hereditary cancer predisposition disease, known as Lynch syndrome. We show that the majority of disease-causing MSH2 mutations give rise to folding defects and proteasome-dependent degradation rather than inherent loss of function, and accordingly our in silico modeling data accurately identifies disease-causing mutations and outperforms the traditionally used genetic disease predictors. Thus, in conclusion, in silico biophysical modeling should be considered for making genotype-phenotype predictions and for diagnosis of Lynch syndrome, and perhaps other hereditary diseases.
Author summary The protein quality control system targets misfolded proteins for degradation. So far it has not been possible from sequence or structural data to predict the biological stability of a misfolded protein, or the effect of mutations on intracellular protein levels. Here we demonstrate that in silico saturation mutagenesis and biophysical calculations of the structural stability of the human mismatch repair protein MSH2 correlate with cellular protein levels, turnover and function. Of 24 different MSH2 variants, some of which are linked to Lynch syndrome, a destabilization of as little as 3 kcal/mol is sufficient to cause rapid degradation via the ubiquitin-proteasome pathway. Thus, biophysical modeling can, to a large extent, predict the metabolic stability of proteins. We also show that the same biophysical calculations can be used to distinguish with high accuracy neutral sequence variation from pathogenic variants, and that the calculations outperform several traditionally used disease predictors. We therefore suggest the method to be of potential value for patient stratification in Lynch syndrome, and perhaps other hereditary diseases.