Kidney Week

Abstract: TH-OR035

Improving Polygenic Risk Prediction of Kidney Function by Removing Biomarker-Specific Genetics

Session Information

• Genetics of Complex Kidney Traits
  November 06, 2025 | Location: Room 360A, Convention Center
  Abstract Time: 05:40 PM - 05:50 PM

Category: Genetic Diseases of the Kidneys

• 1202 Genetic Diseases of the Kidneys: Complex Kidney Traits

Authors

• Du, Jiawen, The University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, United States

Jiawen Du, MS

• Sun, Quan, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, United States

Quan Sun, PhD

• Zhou, Laura Y, Indiana University School of Medicine, Indianapolis, Indiana, United States

Laura Y Zhou, PhD

• Franceschini, Nora, The University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, United States

Nora Franceschini, MD, MPH

• Cox, Nancy Jean, Vanderbilt University Medical Center, Nashville, Tennessee, United States

Nancy Jean Cox, PhD

• Li, Yun, The University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, United States

Yun Li

• Shuey, Megan, Vanderbilt University Medical Center, Nashville, Tennessee, United States

Megan Shuey, PhD

Background

Current eGFR equations using creatinine and/or cystatin-C (eGFRcys, eGFRcr²⁰²¹, eGFRcr²⁰⁰⁹, eGFRcr-cys²⁰²¹, eGFRcr-cys²⁰¹²) show substantial variability. These discrepancies may arise from underlying biomarker-specific genetics that reflect variation in biomarkers’ basal levels independent of renal function. We hypothesize that using shared genetic architecture can remove biomarker-specific genetic predictors and improve genetic prediction of renal function.

Methods

We split UKB Europeans without kidney disease (N=441K) into training (N=350K) and testing (N=91K) for polygenic risk scores (PRS) development. Using LOGODetect we identified ~17.7K HapMap3 variants in 278 shared genetic regions between eGFRcr and eGFRcys in the training set. We constructed 3 PRSs using PRS-CS: (1) shared PRS, from shared variants (m=18K), (2) biomarker-specific PRS, from variants outside shared regions and excluding LD buddies (R² > 0.8) (m = 1M), and (3) benchmark PRS, from all variants (m=1.1M). These PRSs were evaluated for eGFR prediction in testing set and UKB participants of African ancestry (N=9K). Replications were performed in an independent cohort, BioVU.

Results

Variants from shared regions show up to 57% higher -log10 p-value correlation among the 5 eGFR GWAS. The shared PRS explains 60% more variance of eGFRcr. The variance explained was lower in eGFRcys (37.4% of benchmark PRS R²), likely due to greater extrarenal influences on this estimate. This was supported by including BMI as one of these influences. Specifically, the shared PRS R² for eGFRcys increased by 1.7%, while benchmark PRS R2 decreased by 4.3%; creatine-derived estimates were not impacted. In UKB AFR, the shared PRS derived from European yielded 49% higher R² increment for eGFRcr than the benchmark PRS, and incredibly, the shared PRS demonstrated equal power to benchmark PRS for eGFRcr-cys and 80% of its power for eGFRcys. Validation in BioVU confirmed that shared PRS outperformed the benchmark with 5.2% R² improvement to eGFRcr.

Conclusion

By including only variants from shared genetic regions, we demonstrated comparable or superior performance to genome-wide PRS in European and African ancestry cohorts. These results highlight the importance of considering biomarker-specific genetics and the potential to improve cross-population genetic prediction of renal function.

Funding

• Other NIH Support

Digital Object Identifier (DOI)

• doi: 10.1681/ASN.2025w23stb48