Abstract
Chronic kidney disease (CKD) is a multifactorial condition driven by diverse etiologies that lead to a gradual loss of kidney function. Although genome-wide association studies (GWAS) have identified numerous genetic loci linked to CKD, a large portion of its genetic basis remains unexplained. This knowledge gap may partly arise from the reliance on single biomarkers, such as estimated glomerular filtration rate (eGFR), to assess kidney function. To address this limitation, we developed and applied a novel multi-phenotype approach, combinatorial Principal Component Analysis (cPCA), to better understand the complex genetic architecture of CKD. Using UK Biobank dataset (n = 337,112), we analyzed 21 CKD-related phenotypes, generating over 2 million composite phenotypes (CPs) through cPCA. Nearly 50,000 of these CPs demonstrated significantly higher classification power for clinical CKD compared to individual biomarkers. The top-ranked CP-a combination of albumin, cystatin C, eGFR, gamma-glutamyltransferase, HbA1c, low-density lipoprotein, and microalbuminuria, achieved an AUC of 0.878 (95% CI: 0.873-0.882), significantly outperforming eGFR alone (AUC: 0.830, 95% CI: 0.825-0.835). Genetic association analysis of the ~ 50,000 high-performing CPs identified all major eGFR-associated loci, except for the SH2B3 locus rs3184504, a loss-of-function variant, which was uniquely identified in CPs (p = 3.1[Formula: see text]10-56) but not in eGFR within the same sample size. In addition, SH2B3 locus showed strong evidence of colocalization with eGFR, supporting its role in kidney function. These results highlight the power of the multi-phenotype cPCA approach in understanding the genetic basis of CKD, with potential applications to other complex diseases.