Im, C.

The generalizability of general-population GWAS hits in childhood cancer survivors: An examination of complex anthropometric and cardiometabolic phenotypes in the St. Jude Lifetime Cohort Study (SJLIFE) and the Childhood Cancer Survivor Study (CCSS)
Im, C., Wang, Z., Qin, N., Qiu, W., Howell, C.R., Sapkota, Y., Moon, W., Chemaitilly, W., Gibson, T.M., Mulrooney, D.A., Ness, K.K., Wilson, C.L., Zhang, J., Armstrong, G.T., Bhatia, S., Hudson, M.H., Robison, L.L., Yasui, Y.

Large-scale (N>10,000) meta-analyses of genome-wide association studies (meta-GWAS) have identified robust associations (P<5x10-8) between single nucleotide polymorphisms (SNPs) and many complex traits/diseases (i.e., "meta-GWAS hits"). With mounting interest in translating established meta-GWAS hits in diverse clinical/research settings, evaluating their generalizability in target populations is crucial. Childhood cancer survivors differ from "general population" GWAS mega-cohorts: survivors are at greater risk for serious health conditions earlier in life, in part due to exposures to treatments. Here we evaluate the generalizability of 1,376 robust meta-GWAS hits for 12 anthropometric/cardiometabolic phenotypes in long-term (≥5-year) survivors.

We compiled meta-GWAS hits from 46 studies (published before 11/20/2017). We used linear/logistic regression models to replicate meta-GWAS hits in St. Jude Lifetime Cohort Study (SJLIFE) survivors (N=2,231), followed by a secondary analysis of five comparable phenotypes in Childhood Cancer Survivor Study (CCSS) survivors (N=4,212). With replication power estimates, we tested whether observed replication frequencies were greater/less than expected using Poisson generalized estimating equations. Experimental DNA methylation data for genomic sites linked to meta-GWAS SNPs were assessed for associations with chemotherapies and radiotherapies in 236 SJLIFE survivors.

In SJLIFE, we anticipated to replicate ~279 meta-GWAS hits. We observed a median replication rate of 13.2% (IQR=11.4-20.2%) and significantly fewer replications than expected overall (observed-to-expected ratio=0.68, P=2.4x10-9). Stronger replication depletions relative to expected were seen across all five comparable phenotypes in CCSS (observed-to-expected ratio=0.53, P=1.1x10-16). Greater meta-GWAS hit replication depletions were seen in survivors exposed to cancer therapies when treatments had larger contributions to phenotype variation. Using complementary DNA methylation data, we discovered treatment-methylation patterns at genomic sites linked to GWAS hits may disrupt established genetic signals in survivors.

Because of strong associations between treatments and cardiometabolic phenotypes, not all general-population meta-GWAS hits apply in survivors. Treatment-induced epigenetic alterations are a potential explanation for observed replication depletions in survivors.