New insights into the pathogenicity of non-synonymous variants through multi-level analysis

Abstract

Precise classification of non-synonymous single nucleotide variants (SNVs) is a fundamental goal of clinical genetics. Next-generation sequencing technology is effective for establishing the basis of genetic diseases. However, identification of variants that are causal for genetic diseases remains a challenge. We analyzed human non-synonymous SNVs from a multilevel perspective to characterize pathogenicity. We showed that computational tools, though each having its own strength and weakness, tend to be overly dependent on the degree of conservation. For the mutations at non-degenerate sites, the amino acid sites of pathogenic substitutions show a distinct distribution in the classes of protein domains compared with the sites of benign substitutions. Overlooked disease susceptibility of genes explains in part the failures of computational tools. The more pathogenic sites observed, the more likely the gene is expressed in a high abundance or in a high tissue-specific manner, and have a high node degree of protein-protein interaction. The destroyed functions due to some false-negative mutations may arise because of a reprieve from the epigenetic repressed state which shouldn't happen in multiple biological conditions, instead of the defective protein. Our work adds more to our knowledge of non-synonymous SNVs' pathogenicity, thus will benefit the field of clinical genetics.

Conflict of interest statement

The authors declare no competing interests.

Figures

Figure 1

Overreliance on the degree of conservation in pathogenicity predictions. (a) Percentage of highly conserved positions of false-positive (FP) variants. (b,c) Percentage of lowly conserved positions of variants with false negative (FN) predictions for pathogenic variants annotated by ClinVar (b) and for DM variants annotated by HGMD (c). The dashed lines show the proportion of positions with high (a) or low (b,c) conservation score over all positions of benign (a) or pathogenic/DM (b,c) variants. The vertebrate phastCons score cutoff for high or low level conservation is set at 0.5. The observed excess for the positions are evaluated by p-values based on Pearson’s chi-squared test with respect to the proportion of all annotated positions of benign/pathogenic/DM variants. The significances are indicated as * for p < 0.05, ** for p < 10−5 and *** for p < 10−10.

Figure 2

The maximum (a) and coefficient of variation (b) of prediction scores assigned to the four types of nucleotides at the non-degenerated sites corresponding to the four groups of variants annotated by ClinVar and DM variants annotated by HGMD. Wilcoxon tests were used to test the significance of the differences between groups of variants. Significant differences were observed between pathogenic variants and benign variants for all the computational tools.

Figure 3

Characterization of pathogenicity at the gene level. Cumulative probability distributions of the maximum expression level (a) and the tissue specificity of expression (b) among the 53 human normal tissues examined, and the ranked protein-protein interaction network degree (c) for all genes, ClinVar genes and HGMD genes. (d,e,f,g,h,i) Analysis on ClinVar genes and HGMD genes corresponding to the number of pathogenic variants found in the gene. Cumulative probability distributions of the maximum expression level (d,g), the tissue specificity of expression (e,h) and the ranked protein-protein interaction network degree (f,i) for genes carrying at least one sites of pathogenic variants (n > 0) and genes carrying more than 20 sites of pathogenic variants. ClinVar genes are defined as genes that contain pathogenic variant(s) annotated by ClinVar database, and HGMD genes are defined as genes that contain DM variant(s) annotated by HGMD database. Kolmogorov-Smirnov tests were used to test the significance of the differences between gene groups. P-values from pairwise comparisons are shown.

Figure 4

Percentage of cell types in PolyComb state for the non-synonymous sites. Sites of pathogenic variants annotated by ClinVar and DM variants annotated by HGMD are frequently found in repressed PolyComb state (a) as well as in weak repressed PolyComb state. (b) For most computational models analyzed in this study, sites of variants with false-negative predictions (red) are more frequently found in repressed or weak repressed PolyComb state compared to the sites of variants with true-positive predictions (blue) for pathogenic variants annotated by ClinVar (c) and DM variants annotated by HGMD. (d) Wilcoxon tests were used to test the significance of the differences. P-values are shown.