Known variant vs. genome wide: which PRS is better for prostate cancer?

From the Darst Lab, Public Health Sciences Division

Prostate cancer is one of the most common forms of cancer, and it is the second leading cause of cancer deaths among men in the United States. Prostate cancer is also one of the most heritable cancers. Accurate predictive genomic markers for prostate cancer could help identify men who are at increased risk, and therefore may benefit from earlier or more frequent screening.

Dr. Burcu Darst, an assistant professor in the Public Health Sciences Division at Fred Hutch, studies the genetic factors for prostate cancer in diverse populations. Previously, Dr. Darst and colleagues had conducted a large multi-ancestry genomic-wide association study to identify 269 variants associated with increased prostate-cancer risk. Using these variants, they developed a risk score that was highly predictive of prostate-cancer risk across different populations. This polygenic risk score is called PRS269.

Despite the high predictive ability of PRS269, studies on other diseases with complex traits have shown that a different type of PRS that relies on more variants across the genome performs better. Unlike PRS269, which relies on the 269 known variants associated with prostate cancer risk, genome-wide PRS approaches include variants in the genome that do not reach genome-wide statistical significant thresholds (and would therefore be excluded from scores like PRS269). Genome-wide PRS approaches have been demonstrated to have superior predictive performance for complex traits including schizophrenia, type 2 diabetes, and breast and colorectal cancer.

In a new study published in the American Journal of Human Genetics, Dr. Darst led a study to compare the predictive ability of genome-wide polygenic risk scores for prostate cancer to the polygenic risk scores of established prostate-cancer risk variants (PRS269). The researchers found that the two models performed comparably.

The researchers started by training 6 genome-wide PRS models with summary statistics from the studies that were used for the construction of PRS269. These included data from European, African, East Asian, and Hispanic populations. The researchers then tested the genome-wide PRS models in men of African ancestry from the California Uganda Study, as well as in men of European ancestry from the UK Biobank. For additional validation, they applied the model to cases from the Million Veteran Program, which included men of European and African ancestry.

Diagram showing T1-3 stages of prostate cancer
Prostate cancer is one of the most heritable cancers. Accurate predictive risk scores for prostate cancer could help identify people who are at increased genetic risk, and therefore may benefit from earlier or more frequent cancer screening. Image courtesy of Cancer Research UK/Wikimedia Commons

Of the 6 genome-wide PRS models, an approach called PRS-CSx performed best in both the California Uganda Study and the UK Biobank testing datasets. However, in both of these datasets, PRS269 had comparable or even better predictive ability as PRS-CSx using the area under the curve measure of sensitivity and specificity. PRS269 was also associated with prostate-cancer odds ratios that were comparable to, or higher than, the odds ratios estimated with PRS-CSx. Likewise, in the validation Million Veteran Program study, PRS-CSx performed the best out of the 6 genome-wide PRS models. And, as with the testing datasets, PRS269 performed better or similarly to PRS-CSx with comparable or increased odds ratios.

The researchers have a few hypotheses as to why the PRS269 constructed from known variants performs comparably to genome-wide PRS for predicting prostate cancer risk. First, in studies of other complex-trait diseases where genome-wide PRS performed better, the known-variant PRS are often constructed from individuals with European ancestry. In contrast, PRS269 is based on a large, multi-ancestry genome-wide association study that included men from African, European, East Asian, and Hispanic populations. Additionally, the highly heritable nature of prostate cancer may explain the high predictive ability of the known-variant PRS.

Still, the researchers emphasize that the high performance of PRS269 does not mean it is perfect. They write in the paper that the “findings do not imply that the multi-ancestry PRS269 has reached optimal performance; to improve genetic risk prediction of prostate cancer, it will be important to increase the sample size of non-European ancestry men in the discovery GWAS, particularly the sample size of African ancestry men, where we and others have observed that the PRS has lower performance than in other populations.” The authors also stress the need to test whether the high predictive power of PRS269 observed in this study holds true for men of East Asian and Hispanic ancestry.

Ultimately, the improvement of predictive risk scores for prostate cancer means that those with high genetic risk can screen for prostate cancer earlier in life, and at more frequent intervals, which would allow for earlier detections of cancer to improve health outcomes.

This work was supported by the National Cancer Institute, Prostate Cancer Foundation, Andy Hill Cancer Research Endowment Distinguished Researchers Program, Fred Hutchinson Cancer Center/University of Washington SPORE Career Enhancement Program Award, and Million Veteran Program.

The Fred Hutch/University of Washington/Seattle Children’s Cancer Consortium member Drs. Burcu Darst and Catherine Tangen contributed to this work.

Darst BF, Shen J, Madduri RK, Rodriguez AA, Xiao Y, Sheng X, Saunders EJ, Dadaev T, Brook MN, Hoffmann TJ, Muir K, Wan P, Le Marchand L, Wilkens L, Wang Y, Schleutker J, MacInnis RJ, Cybulski C, Neal DE, Nordestgaard BG, Nielsen SF, Batra J, Clements JA, Cancer BioResource AP, Grönberg H, Pashayan N, Travis RC, Park JY, Albanes D, Weinstein S, Mucci LA, Hunter DJ, Penney KL, Tangen CM, Hamilton RJ, Parent MÉ, Stanford JL, Koutros S, Wolk A, Sørensen KD, Blot WJ, Yeboah ED, Mensah JE, Lu YJ, Schaid DJ, Thibodeau SN, West CM, Maier C, Kibel AS, Cancel-Tassin G, Menegaux F, John EM, Grindedal EM, Khaw KT, Ingles SA, Vega A, Rosenstein BS, Teixeira MR; NC-LA PCaP Investigators; Kogevinas M, Cannon-Albright L, Huff C, Multigner L, Kaneva R, Leach RJ, Brenner H, Hsing AW, Kittles RA, Murphy AB, Logothetis CJ, Neuhausen SL, Isaacs WB, Nemesure B, Hennis AJ, Carpten J, Pandha H, De Ruyck K, Xu J, Razack A, Teo SH; Canary PASS Investigators; Newcomb LF, Fowke JH, Neslund-Dudas C, Rybicki BA, Gamulin M, Usmani N, Claessens F, Gago-Dominguez M, Castelao JE, Townsend PA, Crawford DC, Petrovics G, Casey G, Roobol MJ, Hu JF, Berndt SI, Van Den Eeden SK, Easton DF, Chanock SJ, Cook MB, Wiklund F, Witte JS, Eeles RA, Kote-Jarai Z, Watya S, Gaziano JM, Justice AC, Conti DV, Haiman CA. 2023. Evaluating approaches for constructing polygenic risk scores for prostate cancer in men of African and European ancestry. Am J Hum Genet. 110:1-7.