Genome-wide polygenic score with APOL1 risk genotypes predicts chronic kidney disease across major continental ancestries

Atlas Khan, Michael C. Turchin, Amit Patki, Vinodh Srinivasasainagendra, Ning Shang, Rajiv Nadukuru, Alana Jones, Edyta Małolepsza, Ozan Dikilitas, Iftikhar J. Kullo, Daniel J. Schaid, Elizabeth W. Karlson, Tian Ge, James B. Meigs, Jordan W. Smoller, Christoph Lange, David R. Crosslin, Gail P. Jarvik, Pavan K. Bhatraju, Jacklyn N. Hellwege, Paulette D. Chandler, Laura Rasmussen Torvik, Alex Fedotov, Cong Liu, Christopher Kachulis, Noura S. Abul‐Husn, Judy H. Cho, Iuliana Ionita‐Laza, Ali G. Gharavi, Wendy K. Chung, George Hripcsak, Chunhua Weng, Girish N. Nadkarni, Marguerite R. Irvin, Hemant K. Tiwari, Eimear E. Kenny, Nita A. Limdi, Krzysztof KirylukmedRxiv 2021 · DOI

Abstract

Chronic kidney disease (CKD) is a common complex condition associated with significant morbidity and mortality worldwide. Early detection is critical for effective prevention of disease progression. While polygenic prediction could enhance screening, current approaches have not been optimized for ancestrally diverse populations. We developed and validated a genome-wide polygenic score (GPS) for CKD (defined by eGFR <60 mL/min/1.73m²) by combining common variant association statistics with APOL1 risk genotypes. The score was optimized using UK Biobank data and validated across 15 independent cohorts encompassing European, African, Asian, and Hispanic/Latinx ancestries. Our results demonstrate reproducible performance, with the top 2% of the GPS distribution associated with a nearly 3-fold increased risk of CKD across all groups—a level of risk equivalent to a positive family history. This score provides a framework for clinically meaningful risk stratification across diverse global populations.

Introduction

Chronic Kidney Disease (CKD) represents a major global health burden, yet early detection remains a significant clinical challenge. While genetic factors are known to contribute substantially to CKD risk, most existing Polygenic Risk Scores (PRS) suffer from a "diversity gap." Because these scores are primarily trained on datasets of European ancestry, their predictive accuracy often diminishes when applied to other ancestral groups.

Furthermore, for individuals of African ancestry, CKD risk is heavily influenced by specific high-impact variants in the APOL1 gene. Standard genome-wide association studies (GWAS) often fail to capture these effects because the variants are rare in non-African populations. This study addresses these limitations by developing a "universal" genome-wide polygenic score (GPS) that integrates broad genomic signals with ancestry-specific risk factors to ensure equitable predictive power.

Methodology

The researchers developed a GPS that synthesizes two distinct genetic signals:

  1. Broad Polygenic Signal: Utilizing summary statistics from a massive GWAS of over 765,000 individuals, the score captures thousands of common variants associated with the estimated glomerular filtration rate (eGFR), a standard proxy for kidney function.
  2. High-Impact APOL1 Signal: The model specifically integrates the "recessive model" of APOL1 risk (carrying two risk alleles: G1G1, G2G2, or G1G2), which is a primary driver of kidney disease in African and admixed populations.

To ensure the score's utility across different groups, the authors implemented two post-hoc ancestry adjustment strategies using the 1000 Genomes Project as a reference. By adjusting for mean and variance, they established a single, standardized "high-risk" threshold applicable regardless of an individual's continental ancestry.

Data and Validation

The study's robustness is supported by an extensive array of genomic resources:

  • Optimization: The polygenic component was optimized using 70% of the UK Biobank (UKBB) European participants (~280,000 individuals), while APOL1 weighting was derived from UKBB African-ancestry participants.
  • Validation: The score was tested in 15 independent cohorts totaling over 170,000 individuals, including eMERGE-III, BioMe Biobank, and specialized studies such as REGARDS, HyperGEN, and GenHAT.

Key Insights

The Additive Nature of Risk

A critical finding of the study is that APOL1 risk and general polygenic risk are additive. The researchers found no significant interaction between the two (P=0.29), meaning they contribute to disease risk independently. Consequently, an individual may be at high risk due to their APOL1 status, a high polygenic background, or a combination of both.

Baseline Polygenic Differences

Even when excluding APOL1 variants, individuals of African ancestry exhibited a significantly higher average polygenic risk score for CKD compared to other groups. This suggests that the higher prevalence of CKD in African American populations is driven by both high-impact variants and a higher "background" of common risk alleles.

Clinical Actionability

The study focused on the top 2% of the GPS distribution as a high-risk tier. Across all ancestries, this group faced a ~3x increased risk of CKD. In clinical practice, this magnitude of risk is considered highly actionable, as it mirrors the risk level associated with a strong family history of kidney disease.

Results and Performance

The GPS demonstrated consistent and reproducible performance across all 15 testing cohorts. The disease Odds Ratio (OR) per standard deviation of the score was:

  • European: 1.49
  • African: 1.32
  • Asian: 1.59
  • Hispanic/Latinx: 1.42

For those in the top 2% of the GPS, the risk increase was stark: 3.09x for Europeans, 2.66x for Africans, 2.95x for Asians, and 4.27x for Hispanic/Latinx cohorts. Notably, in Hispanic/Latinx cohorts, the inclusion of APOL1 improved the score's performance, likely due to African ancestral segments within those admixed genomes.

Limitations and Future Directions

While this GPS represents a significant step toward equitable genomic medicine, some limitations remain. The underlying GWAS data is still 75% European, and larger non-European GWAS are necessary to further refine allelic weights. Additionally, the score relies on eGFR—a biomarker that can be influenced by non-genetic factors like diet and muscle mass—rather than biopsy-confirmed disease. Finally, prospective clinical trials are required to demonstrate that implementing this GPS in a clinical setting leads to improved patient outcomes and more effective early interventions.

Summary updated

Abstract

ABSTRACT Introduction Chronic kidney disease (CKD) is a common complex condition associated with significant morbidity and mortality in the US and worldwide. Early detection is critical for effective prevention of kidney disease progression. Polygenic prediction of CKD could enhance screening and prevention of kidney disease progression, but this approach has not been optimized for risk prediction in ancestrally diverse populations. Methods We developed and validated a genome-wide polygenic score (GPS) for CKD defined by estimated glomerular filtration rate (eGFR) &lt;60 mL/min/1.73m 2 using common variant association statistics from GWAS for eGFR combined with information on APOL1 risk genotypes. The score was designed to ensure transferability across major continental ancestries, genotyping platforms, imputation panels, and phenotyping strategies, and was tested following ClinGen guidelines. The polygenic component of the score was developed and optimized using 28,047 cases and 251,772 controls (70% of UK Biobank participants of European ancestry), while the weights for APOL1 effects were derived based on UK Biobank participants of African ancestry (967 cases and 6,191 controls). We tested the performance of the score in 15 independent testing cohorts, including 3 cohorts of European ancestry (total 23,364 cases and 117,883 controls), 6 cohorts of African ancestry (4,268 cases and 10,276 controls), 4 cohorts of Asian ancestry (1,030 cases and 9,896 controls), and 2 Hispanic/Latinx cohorts (1,492 cases and 2,984 controls). Results We demonstrated the risk score transferability with reproducible performance across all independent testing cohorts. In the meta-analyses, disease odds ratios per standard deviation of the score were estimated at 1.49 (95%CI: 1.47-1.50, P&lt;1.0E-300) for European, 1.32 (95%CI: 1.26-1.38, P=1.8E-33) for African, 1.59 (95%CI: 1.52-1.67, P=1.3E-30) for Asian, and 1.42 (95%CI: 1.33-1.51, P=4.1E-14) for Latinx cohorts. The top 2% cutoff of the GPS was associated with nearly 3-fold increased risk of CKD across all major ancestral groups, the degree of risk that is equivalent to a positive family history of kidney disease. In African-ancestry cohorts, APOL1 risk genotype and the polygenic risk components of the GPS had additive effects on the risk of CKD with no significant interactions. We also observed that individuals of African ancestry had a significantly higher polygenic risk score for CKD compared to other populations, even without accounting for APOL1 variants. Conclusions By combining APOL1 risk genotypes with the available GWAS for renal function, we designed, optimized, and validated a GPS predictive of CKD across four major continental ancestries. With the upper tail of the GPS distribution associated with disease risk equivalent to a positive family history, this score could be used for clinically meaningful risk stratification.
BibTeX
@misc{khan2021genome,
  author = {Atlas Khan and Michael C. Turchin and Amit Patki and Vinodh Srinivasasainagendra and Ning Shang and Rajiv Nadukuru and Alana Jones and Edyta Małolepsza and Ozan Dikilitas and Iftikhar J. Kullo and Daniel J. Schaid and Elizabeth W. Karlson and Tian Ge and James B. Meigs and Jordan W. Smoller and Christoph Lange and David R. Crosslin and Gail P. Jarvik and Pavan K. Bhatraju and Jacklyn N. Hellwege and Paulette D. Chandler and Laura Rasmussen Torvik and Alex Fedotov and Cong Liu and Christopher Kachulis and Noura S. Abul‐Husn and Judy H. Cho and Iuliana Ionita‐Laza and Ali G. Gharavi and Wendy K. Chung and George Hripcsak and Chunhua Weng and Girish N. Nadkarni and Marguerite R. Irvin and Hemant K. Tiwari and Eimear E. Kenny and Nita A. Limdi and Krzysztof Kiryluk},
  title = {{Genome-wide polygenic score with <i>APOL1</i> risk genotypes predicts chronic kidney disease across major continental ancestries}},
  year = {2021},
  doi = {10.1101/2021.10.25.21265398},
  howpublished = {medRxiv},
}
Figure