Integrating genetic predictors into breast cancer risk stratification among female survivors of childhood cancer

Background: Childhood cancer survivors are at high risk for developing breast cancer (BC). Existing BC risk stratification tools primarily consider childhood cancer treatment risk factors, but do not assess genetic predictors. The St. Jude Survivorship Portal (https://survivorship.stjude.cloud) is a new open-access online data resource that supports deeper investigations of late effects genetic susceptibility. In this study, we evaluated data available in this portal, specifically common and rare genetic variation, for their relative contributions to subsequent BC risk stratification. Methods: Five-year female survivors from 2 cohort studies (Childhood Cancer Survivor Study; St. Jude Lifetime Cohort) were analyzed. Common variants were assessed using 99 published general population sporadic BC polygenic risk scores (PRSs), computed using whole-genome sequencing (WGS) or imputed array-based genotype data. Hazard ratios (HRs) were estimated with Cox regression using age as the time scale and adjusted for genetic ancestry, batch, and chest radiotherapy (RT) and anthracycline doses. PRS-RT interactions assessed whether PRSs modified risks conferred by chest RT dose. Rare pathogenic/likely pathogenic (P/LP) variants in BC susceptibility genes examined in general population sequencing studies (BRCA1, BRCA2, CHEK2, PALB2, RAD51C, RAD51D, TP53, ATM) were evaluated. Results: In this multi-ancestry cohort (N=5388; 88% European, 8% African, 4% Asian ancestry), the median attained age was 38y (IQR 30-46) and 63% were treated with chest RT. 319 survivors developed BC. Overall, 90.1% of sporadic BC PRSs were nominally replicated (P<0.05). No PRS modified chest RT-related BC risk. Notably, the most studied BC PRS (313 variants; evaluated in >70 studies) did not show the strongest risk association in survivors (HR per SD=1.17, 95% CI=1.04-1.31). Instead, PRSs with more variants genome-wide demonstrated effect sizes similar to some treatment predictors (e.g., “best” PRS: ~6.4 million variants, HR per SD=1.71, 95% CI=1.43-2.05; P=4.2x10-9). BC PRSs with the strongest associations (top 25%) were 4.7-times more likely to include variants mapped to multiple (2) DNA repair pathways (P=0.018). Among the 3292 survivors with WGS, 61 (1.8%) carried rare P/LP variants in BC risk genes. Carrying BC P/LP variants was suggestively associated with BC risk (HR=2.40, P=0.055). While the best BC PRS had greater risk stratification potential (top vs bottom 20%: HR=5.94, 95% CI=2.70-13.07, P=8.6x10-6), most (4 out of 6) BC-affected survivors with BC P/LP variants were not captured by this PRS cutoff. Conclusions: Overall, our results suggest specific published sporadic BC PRSs could be considered in future updates to childhood cancer survivorship care guidelines and in subsequent BC risk prediction models. Further study of targeted BC pathogenic risk variant screening is needed in this population.