Introduction:
The PS4 criterion, a variant’s clinical statistical association with a condition, is a crucial component of assessing pathogenicity, but has historically been difficult to obtain. This is particularly true in a screening setting,  where the patient does not yet have symptoms. With the advent of large clinicogenomic databases, it is now feasible to maintain public resources showing the PS4 evidence for variants in existing data, which will help to standardize and improve the clinical validity of variant interpretation.

Methods:
We utilize data from three large clinicogenomic datasets–the Helix Research Network, UK Biobank, and All of Us programs–to generate statistical evidence for all variants in a set of genes with support for use in a screening context: ATM, BRCA1, BRCA2, CHEK2, and PALB2 for breast and ovarian cancer, in MLH1, MSH2, MSH6, and PMS2 for colon and endometrial cancer; and in APOB, PCSK9 and LDLR for hypercholesterolemia. We use the established cutoff for support for pathogenicity for HBOC of OR>4 with the lower 95%CI>2, or the upper 95%CI<2 for support for benign. We then built similar appropriate cutoffs for for Lynch syndrome and familial hypercholesterolemia based on the associations seen for known pathogenic variants for these conditions: for Lynch syndrome, an OR >10 with 95%CI >4 for pathogenic, or OR<2 and 95%CI<4 for benign; and for LDL levels accounting for statin use, age, sex, and BMI as a proxy for FH, normalized effect size>0.5 with 95%CI>0.25, with variants with their upper 95%CI<0.25 being classified as benign. We additionally incorporate appropriate carrier count cutoffs to be powered and reduce false positives for pathogenic or benign classification.

Results:
We present results from 59,661 rare (MAF<1%) variants in these genes: 4% of which are LoF, 31% other coding, and 65% synonymous, intronic, or otherwise in the gene region. Of these variants, 2,054 (3.4%) had a sufficient carrier count to be powered for the analysis. We find that 4.5% of these 2,054 variants have sufficient evidence to qualify as supportive of being pathogenic (60% of LoFs, 5.7% of other missense, and 0.5% of other). We find that 59% have support for being benign (31% of LoFs, 57% of other missense, and 62% of other), with the remainder having insufficient evidence either way. Of note, the LoF variants with evidence for being benign were either in known gain of function genes (APOB, PCSK9) or were in the final exon of the gene. Although only 3% of the variants had enough evidence in support of classification given the current available sample sizes, they make up a larger proportion of the carriers due to their higher frequency: 57% of all the individuals with rare variants in these genes can be considered to have a benign variant, and 0.3% a pathogenic one (for those with coding or LoF variants, these percentages are 53% of the individuals having a benign variant and 0.9% having a pathogenic one). These results make a major dent in the population screening workload, as they provide evidence for the large majority of variants in these genes to be benign and thus rule them out for further consideration. We make them available as a public resource.

Conclusion:
Classifying variants for CDCT1 conditions can be improved and standardized by the use of a common PS4 evidence resource, which we provide.