Genome-wide association study of prostate-specific antigen levels in 392,522 men identifies new loci and improves cross-ancestry prediction

Thomas J. Hoffmann, Rebecca E. Graff, Ravi Madduri, Alex A Rodriguez, Clinton L. Cario, Karen Feng, Yu Jiang, Anqi Wang, Robert J. Klein, Brandon L. Pierce, Scott Eggener, Tong Lin, William J. Blot, Jirong Long, Louisa B. Goss, Burcu F. Darst, Timothy R. Rebbeck, Joseph Lachance, Caroline Andrews, Akindele O. Adebiyi, Ben Adusei, Oseremen I Aisuodionoe-Shadrach, P. Fernandez, Mohamed Jalloh, Rohini Janivara, Wenlong Carl Chen, James E. Mensah, Ilir Agalliu, Sonja I. Berndt, John P. Shelley, Kerry Schaffer, Mitchell J. Machiela, Neal D Freedman, Wen‐Yi Huang, Shengchao A. Li, Phyllis J. Goodman, Cathee Till, Ian M. Thompson, Hans Lilja, Dilrini K. Ranatunga, Joseph C. Presti, Stephen K. Van Den Eeden, Stephen J. Chanock, Jonathan D. Mosley, David V. Conti, Christopher A. Haiman, Amy C. Justice, Linda Kachuri, John S. WittemedRxiv 2023 · DOI

TL;DR

Researchers conducted the largest multi-ancestry genome-wide association study (GWAS) of prostate-specific antigen (PSA) levels to date, involving over 392,000 men. By identifying 447 genetic variants—nearly 300 of which are novel—the team developed a polygenic risk score (PRS) that significantly improves the accuracy of PSA-based prostate cancer screening across diverse populations, potentially reducing unnecessary biopsies.

What Problem Does This Paper Address?

Prostate-specific antigen (PSA) is the primary biomarker used to screen for prostate cancer. However, it is a "noisy" signal. PSA levels are influenced by many non-cancerous factors, including age, prostate volume, inflammation, and an individual's baseline genetic makeup.

This noise leads to a major clinical challenge: many men with naturally high PSA levels undergo invasive, unnecessary biopsies (overdiagnosis), while others with naturally low levels might have their cancer missed. Furthermore, most existing genetic studies on PSA have focused on men of European ancestry, leaving a significant performance gap in polygenic risk scores (PRSs) for men of African, Hispanic/Latino, and Asian descent. This paper aims to "de-noise" PSA levels by accounting for genetic variation across a diverse global population.

Methods

The study followed a multi-stage GWAS pipeline:

  1. Discovery: A multi-ancestry analysis of log-transformed PSA levels in ~297,000 men using mJAM (multi-ancestry Joint Association Mapping), which utilizes population-specific linkage disequilibrium (LD) reference panels to account for different genetic architectures.
  2. Replication: Findings were tested in an independent cohort of ~96,000 men.
  3. Meta-analysis: The teams combined discovery and replication data to identify the final set of associated variants.
  4. PRS Construction: They used PRS-CSx, a Bayesian method that integrates GWAS summary statistics from multiple ancestries to improve cross-population prediction.
  5. Clinical Modeling: The researchers calculated "genetically adjusted PSA" by dividing a person's measured PSA by their genetic predisposition factor. They then evaluated how this adjustment would have changed biopsy recommendations in real-world cohorts.

Datasets

The study leveraged several massive datasets:

  • Million Veteran Program (MVP): Provided 96.5% of the discovery cohort.
  • Replication Cohorts: Included data from previous studies (Kachuri et al. 2023).
  • Validation Cohorts: Out-of-sample evaluation was performed using the GERA cohort, the SELECT and PCPT clinical trials, and the All of Us (AOU) research program.
  • Ancestry Breakdown: The total sample included 297,166 men of European ancestry, 61,745 of African ancestry, 26,644 Hispanic/Latino men, and 6,967 men of Asian ancestry.

Key Insights and Ideas

  • Diversity is a Performance Multiplier: By including a higher proportion of underrepresented populations, the researchers more than doubled the predictive power of PRSs for men of African ancestry compared to previous models.
  • The "De-noising" Effect: Genetically adjusting PSA levels helps distinguish between men who have high PSA because of their genes versus those who have high PSA because of aggressive cancer.
  • Age Matters: The study found that the variance in PSA explained by genetics decreases as men get older. This suggests that environmental factors or cumulative prostate health issues (like BPH) begin to outweigh germline genetics in older age groups.
  • Functional Links: Many of the identified variants were linked to genes involved in hormone signaling (like the androgen receptor AR) and even male fertility (like CFTR and LMTK2), providing new biological leads for prostate physiology.

Key Results

  • Variant Discovery: Identified 447 independent variants associated with PSA levels, 295 of which were previously unknown.
  • Variance Explained: The new PRS explained a substantial portion of PSA variation: 16.9% in European, 18.6% in Hispanic/Latino, 15.3% in Asian, and 9.5% in African ancestries.
  • Biopsy Reclassification: In the GERA cohort, using genetically adjusted PSA resulted in a 7.4% net improvement in reclassifying healthy European controls (meaning they would have avoided an unnecessary biopsy).
  • Aggressive Cancer Detection: Higher genetic scores for PSA were inversely associated with aggressive disease (Gleason score ≥7). This confirms that men with high "genetic PSA" are often biopsied for the wrong reasons, while those with low "genetic PSA" might have aggressive cancers that stay "hidden" below standard clinical thresholds.

Limitations and Future Work

  • African Ancestry Gap: While prediction improved significantly for men of African ancestry, the variance explained (9.5%) still lags behind other groups, likely due to the smaller discovery sample size and higher genetic diversity in this population.
  • Missing Clinical Data: Many cohorts lacked data on prostate volume and Benign Prostatic Hyperplasia (BPH), both of which are major non-genetic drivers of PSA levels.
  • Screening Bias: The authors acknowledge that GWAS of PSA can be influenced by "screening bias"—men with higher genetic PSA are more likely to be screened and diagnosed, which can complicate the relationship between PSA variants and cancer risk.
  • Future Directions: The authors suggest that future work should focus on even larger diverse cohorts and the development of PRSs that specifically target different constituents of total PSA (like free vs. bound PSA).

Summary updated

Abstract

We conducted a multi-ancestry genome-wide association study of prostate-specific antigen (PSA) levels in 296,754 men (211,342 European ancestry; 58,236 African ancestry; 23,546 Hispanic/Latino; 3,630 Asian ancestry; 96.5% of participants were from the Million Veteran Program). We identified 318 independent genome-wide significant (p≤5e-8) variants, 184 of which were novel. Most demonstrated evidence of replication in an independent cohort (n=95,768). Meta-analyzing discovery and replication (n=392,522) identified 447 variants, of which a further 111 were novel. Out-of-sample variance in PSA explained by our genome-wide polygenic risk scores ranged from 11.6%-16.6% in European ancestry, 5.5%-9.5% in African ancestry, 13.5%-18.2% in Hispanic/Latino, and 8.6%-15.3% in Asian ancestry, and decreased with increasing age. Mid-life genetically-adjusted PSA levels were more strongly associated with overall and aggressive prostate cancer than unadjusted PSA. Our study highlights how including proportionally more participants from underrepresented populations improves genetic prediction of PSA levels, offering potential to personalize prostate cancer screening.
BibTeX
@misc{hoffmann2023genome,
  author = {Thomas J. Hoffmann and Rebecca E. Graff and Ravi Madduri and Alex A Rodriguez and Clinton L. Cario and Karen Feng and Yu Jiang and Anqi Wang and Robert J. Klein and Brandon L. Pierce and Scott Eggener and Tong Lin and William J. Blot and Jirong Long and Louisa B. Goss and Burcu F. Darst and Timothy R. Rebbeck and Joseph Lachance and Caroline Andrews and Akindele O. Adebiyi and Ben Adusei and Oseremen I Aisuodionoe-Shadrach and P. Fernandez and Mohamed Jalloh and Rohini Janivara and Wenlong Carl Chen and James E. Mensah and Ilir Agalliu and Sonja I. Berndt and John P. Shelley and Kerry Schaffer and Mitchell J. Machiela and Neal D Freedman and Wen‐Yi Huang and Shengchao A. Li and Phyllis J. Goodman and Cathee Till and Ian M. Thompson and Hans Lilja and Dilrini K. Ranatunga and Joseph C. Presti and Stephen K. Van Den Eeden and Stephen J. Chanock and Jonathan D. Mosley and David V. Conti and Christopher A. Haiman and Amy C. Justice and Linda Kachuri and John S. Witte},
  title = {{Genome-wide association study of prostate-specific antigen levels in 392,522 men identifies new loci and improves cross-ancestry prediction}},
  year = {2023},
  doi = {10.1101/2023.10.27.23297676},
  howpublished = {medRxiv},
}
Figure