While genetic analyses of common variants for body mass index (BMI) have leveraged huge sample sizes, relatively few studies have examined rare non-coding variation in relation to BMI. Through whole-genome sequencing (WGS) in TOPMed we have a unique opportunity to examine BMI genetics in a large multi-ethnic sample of deeply sequenced individuals.

Analyses were performed in a sample of 45,159 individuals from 20 cohorts and 6 ancestry groups. Age-, study- and PC-adjusted BMI residuals were created within ancestry and sex strata, then rank-normal transformed and rescaled by strata standard deviation. Pooled residuals were analyzed with linear mixed models including a variance component for empirical kinship plus separate residual variance components for each sex-ancestry group.

Single-variant analyses were performed using MMAP on all variants (N=87,683,863) with a minor allele count of >= 5. Analyses identified single variants associations at six loci that surpassed a suggestive significance threshold of p<1e-8: in addition to known loci at FTO, MC4R and SEC16B, three novel rare variant associations were identified (ARHGAP24, rs546474633, MAF 0.11%, B=-4.3, p=8.4E-9; ISPD, rs757499854, MAF=0.02%, B=10.8 kg/m2, p=3.7E-9 and FRMD5, rs897162855, MAF=0.01%, B=16.6 kg/m2, p=9.6E-9). Complete WGS coverage also allows a better understanding of the genetic architecture of previously identified loci. For example, we identified an independent rare variant signal ( rs62106258 , MAF=3.2%, B=-0.77 kg/m2, P=3.5E-8) in the known intergenic region upstream of TMEM18 (rs2867125, B= 0.33 kg/m2, MAF=15, P=1.4E-5).

Rare variants (minor allele frequency < 1%) were aggregated into gene-based units utilizing strategies that infer the target gene for enhancer regions along with variants that fell within the FANTOM5 promoter regions and nonsynonymous coding variants. Enhancer regions and their inferred target gene were identified using GeneHancer and EnhancerAtlas neuronal data sets. Gene-based tests were analyzed using SKAT in the GENESIS R package. No genes surpassed a Bonferroni correction for multiple testing.

These findings require further investigation and we are seeking performing replication using the next set of samples (50K) soon to be released by TOPMed.