Poster #8: Karl Keat PhD

Biobank-scale genomic analysis of quantitative CT image-derived phenotypes links complex disease risk variants to organ traits

Karl Keat, PhD, University of Pennsylvania; David Zhang, PhD, University of Pennsylvania; Jeffrey Duda, PhD, University of Pennsylvania; Cameron Beeche, BS, University of Pennsylvania; Colleen Kripke, PT, DPT, MSA, University of Pennsylvania; Hersh Sagreiya, MD, University of Pennsylvania; Walter Witschey, PhD, University of Pennsylvania; Anurag Verma, PhD, University of Pennsylvania; Daniel J. Rader, MD, University of Pennsylvania

We leveraged clinical chest and abdomen CT scans in participants in the Penn Medicine Biobank, a large disease-agnostic medical biobank, and using TotalSegmentator generated IDPs from 45,972 CT studies in 22,876 individuals with available genotype data. We then performed GWAS on mean attenuation and organ volume stratified by genetic similarity to European and African reference populations, controlling for age, sex, and 6 PCs followed by fixed-effects meta-analysis. Our analysis uncovered a mix of known and novel loci influencing health and physiology. One key positive control was an association between a variant in PNPLA3 and liver attenuation (p=6.9e-17, B=-0.15), which is a major genetic factor for metabolic dysfunction-associated steatotic liver disease. A variant in SH2B3 that was associated with spleen volume (p=4.1e-21, B=0.09) replicated a similar signal in the UKB MRI study, validating this association across imaging modalities. LDSC between our GWAS and the UKB spleen GWAS found strong genetic correlation (rg=0.76, p=1.1e-10, se=0.12, h2=0.22). We found that this variant was strongly associated with lymphocyte counts in multiple independent studies. We also identified many associations that are novel but have strong biological support, such as between a variant in GDF5 and vertebral volume (p=1.5e-21, B=-0.08). In most cases, signals were unique to each organ, except for a locus within FTO, a known obesity marker, which was associated with heart, liver, and adrenal gland volume. We have conducted the largest multi-organ GWAS of image-derived quantitative traits derived from clinical imaging studies. Our results support the use of IDPs based on clinical imaging in a medical biobank to bridge mechanistic gaps between existing GWAS discoveries by linking them to imaging traits. Altogether, clinical imaging such as CT scans are a valuable tool to better understand the genetics of disease, progression and intermediate phenotypes not well captured by binary disease labels.