Prioritizing variants of uncertain significance for reclassification using a rule-based algorithm in inherited retinal dystrophies

Abstract

Inherited retinal dystrophies (IRD) are a highly heterogeneous group of rare diseases with a molecular diagnostic rate of >50%. Reclassification of variants of uncertain significance (VUS) poses a challenge for IRD diagnosis. We collected 668 IRD cases analyzed by our geneticists using two different clinical exome-sequencing tests. We identified 114 unsolved cases pending reclassification of 125 VUS and studied their genomic, functional, and laboratory-specific features, comparing them to pathogenic and likely pathogenic variants from the same cohort (N = 390). While the clinical exome used did not show differences in diagnostic rate, the more IRD-experienced geneticist reported more VUS (p = 4.07e-04). Significantly fewer VUS were reported in recessive cases (p = 2.14e-04) compared to other inheritance patterns, and of all the genes analyzed, ABCA4 and IMPG2 had the lowest and highest VUS frequencies, respectively (p = 3.89e-04, p = 6.93e-03). Moreover, few frameshift and stop-gain variants were found to be informed VUS (p = 6.73e-08 and p = 2.93e-06). Last, we applied five pathogenicity predictors and found there is a significant proof of deleteriousness when all score for pathogenicity in missense variants. Altogether, these results provided input for a set of rules that correctly reclassified ~70% of VUS as pathogenic in validation datasets. Disease- and setting-specific features influence VUS reporting. Comparison with pathogenic and likely pathogenic variants can prioritize VUS more likely to be reclassified as causal.

Fig. 1. Diagnosis, inheritance, and gene distribution.

VUS reporting frequency for IRD (a) and non-IRD cases (b). For IRD, the inheritance pattern was compared for solved and partially solved cases versus VUS cases (c). Four VUS cases have not been included in the plot since an inheritance pattern has not been established for them. (d) VUS (number appearing beside VUS diagnoses for each gene) and pathogenic/likely pathogenic variants were compared on IRD genes.

Fig. 2. NGS platform, gene panels, and analyst impact.

Diagnostic status in IRD using TSO (a) and CES clinical exome panels (b). Correlation between IRD panels size used and number of VUS/ pathogenic-likely pathogenic variants found (c). Diagnosis status by analyst, 23 cases were analyzed by two analysts at the same time and were not considered in this analysis (d).

Fig. 3. Gene size and genomic region distribution.

Correlation between gene size and number of pathogenic/likely pathogenic and VUS variants found (a) and genomic region distribution of pathogenic/likely pathogenic variants (b) and VUS (c).

Fig. 4. Pathogenicity and variant consequence.

Behavior of pathogenicity predictors for VUS variants (a) and pathogenic/likely pathogenic (b) variants, and variant consequence distribution for pathogenic and VUS (c) variants.

Fig. 5. Rule-based algorithm.

Applied to VUS reclassified as pathogenic/likely pathogenic dataset (a), VUS in recessive solved cases (b), VUS in homozygous solved cases (c), and applied to VUS dataset (d).