An early-onset specific polygenic risk score optimizes age-based risk estimate and stratification of prostate cancer: population-based cohort study

Abstract

Background: Early-onset prostate cancer (EOPC, ≤ 55 years) has a unique clinical entity harboring high genetic risk, but the majority of EOPC patients still substantial opportunity to be early-detected thus suffering an unfavorable prognosis. A refined understanding of age-based polygenic risk score (PRS) for prostate cancer (PCa) would be essential for personalized risk stratification.

Methods: We included 167,517 male participants [4882 cases including 205 EOPC and 4677 late-onset PCa (LOPC)] from UK Biobank. A General-, an EOPC- and an LOPC-PRS were derived from age-specific genome-wide association studies. Weighted Cox proportional hazard models were applied to estimate the risk of PCa associated with PRSs. The discriminatory capability of PRSs were validated using time-dependent receiver operating characteristic (ROC) curves with additional 4238 males from PLCO and TCGA. Phenome-wide association studies underlying Mendelian Randomization were conducted to discover EOPC linking phenotypes.

Results: The 269-PRS calculated via well-established risk variants was more strongly associated with risk of EOPC [hazard ratio (HR) = 2.35, 95% confidence interval (CI) 1.99-2.78] than LOPC (HR = 1.95, 95% CI 1.89-2.01; I² = 79%). EOPC-PRS was dramatically related to EOPC risk (HR = 4.70, 95% CI 3.98-5.54) but not to LOPC (HR = 0.98, 95% CI 0.96-1.01), while LOPC-PRS had similar risk estimates for EOPC and LOPC (I² = 0%). Particularly, EOPC-PRS performed optimal discriminatory capability for EOPC (area under the ROC = 0.613). Among the phenomic factors to PCa deposited in the platform of ProAP (Prostate cancer Age-based PheWAS; https://mulongdu.shinyapps.io/proap ), EOPC was preferentially associated with PCa family history while LOPC was prone to environmental and lifestyles exposures.

Conclusions: This study comprehensively profiled the distinct genetic and phenotypic architecture of EOPC. The EOPC-PRS may optimize risk estimate of PCa in young males, particularly those without family history, thus providing guidance for precision population stratification.

Keywords: Age-specific genome-wide association studies; Early-onset prostate cancer; Phenome-wide association studies; Polygenic risk score; UK biobank.

Fig. 1

Risk estimates for PCa associated with a 269-PRS in the General-, EO- and LO-population. A Weighted Cox proportional hazard models include all subjects regardless of the family history of PCa. B Weighted Cox proportional hazard models include participants without a family history of PCa. C Weighted Cox proportional hazard models include participants with a family history of PCa. Models were adjusted for age at assessment, BMI, smoking status, drinking status, assessment center and top 10 principal components. General-population covered all participants; EO-population was consisted of individuals with exit-age ≤ 55 years old; LO-population comprised individuals with exit-age > 55 years old. PCa prostate cancer, PRS polygenic risk score, HR hazard ratio, EO early-onset, LO late-onset, SD standard deviation

Fig. 2

Risk estimates for PCa associated with a General-PRS (A–C), an EOPC-PRS (D–F) and an LOPC-PRS (G–I) in the General-, EO- and LO-population stratified by family history. Weighted Cox proportional hazard models were adjusted for age at assessment, BMI, smoking status, drinking status, assessment center and top 10 principal components. General-population covered all participants; EO-population was consisted of individuals with exit-age ≤ 55 years old; LO-population comprised individuals with exit-age > 55 years old. PCa prostate cancer, PRS polygenic risk score, HR hazard ratio, EOPC early-onset prostate cancer, LOPC late-onset prostate cancer, SD standard deviation

Fig. 3

Risk estimates for PCa associated with a merged-General-PRS (A–C), a merged-EOPC-PRS (D–F) and a merged-LOPC-PRS (G–I) in the General-, EO- and LO-population stratified by family history. Weighted Cox proportional hazard models were adjusted for age at assessment, BMI, smoking status, drinking status, assessment center and top 10 principal components. General-population covered all participants; EO-population was consisted of individuals with exit-age ≤ 55 years old; LO-population comprised individuals with exit-age > 55 years old. PCa prostate cancer, PRS polygenic risk score, HR hazard ratio, EOPC early-onset prostate cancer, LOPC late-onset prostate cancer, SD standard deviation

Fig. 4

The area under the receiver operating characteristic (ROC) curve (AUC) evaluating the discriminatory accuracy for prediction of PRSs in a European ancestry population generated from the PLCO cohort and TCGA program. A–C Time-dependent ROC curves and AUCs from censored diagnosis data at 65-year (A), 70-year (cases diagnosed at 55 or younger removed) (B), and 85-year (cases diagnosed at 70 or younger removed) (C) of 269-PRS for prediction of PCa. D–I Time-dependent ROC curves and AUCs of General-PRS (D), EOPC- PRS (E), LOPC-PRS (F), merged-General-PRS (G), merged-EOPC-PRS (H) and merged-LOPC-PRS (I). PLCO Prostate, Lung, Colorectal and Ovarian; TCGA The Cancer Genome Atlas Program, EOPC early-onset prostate cancer, LOPC late-onset prostate cancer

Fig. 5

Phenome-wide association analysis of PCa, EOPC and LOPC through two-sample MR analyses based on UK Biobank GWAS summary data. A Venn diagram of high-confidence risk traits liked to PCa, EOPC and LOPC. B Scatter plots of EOPC specific (upper) and LOPC-specific (below) risk traits. C Venn diagram of high-confidence protective traits linked to PCa, EOPC and LOPC. D Scatter plots of EOPC specific (upper) and LOPC-specific (below) protective traits. EOPC early-onset prostate cancer, LOPC late-onset prostate cancer