Germline genetic risk and prostate cancer–specific mortality in a population-based incident cohort

Abstract

Background

Multiple germline variants are associated with prostate cancer (PCa) susceptibility and aggressive features; however, their value for predicting prostate cancer–specific mortality (PCSM) at the time of diagnosis remains uncertain, particularly among men with clinically localized disease.

Methods

We evaluated associations between germline risk factors and PCSM among 14,644 men with incident PCa in the UK Biobank. Associations between PCSM and reported germline risk factors, including 11 genes recommended by the National Comprehensive Cancer Network (NCCN), nine other candidate genes, two common variants [KLK3 (I179T) and HSD3B1 (1245 A > C)], and three polygenic risk scores (PRSs), were tested using Fine–Gray competing-risk models accounting for non–PCa mortality.

Results

During a median follow-up of 6.73 years after diagnosis, 1581 men (10.8%) died from PCa. PVs in three NCCN-recommended DNA damage–repair genes (BRCA2, MSH6, PALB2) were individually associated with increased PCSM and were defined as Tier-1. Aggregated PVs in seven additional NCCN-recommended DDR genes (Tier-2) and the KLK3 I179T variant were also independently associated with PCSM. In contrast, HOXB13, nine other candidate genes, HSD3B1, and all three PRSs were not associated with cumulative PCSM risk. Overall, 14.45% of men carried Tier-1 or Tier-2 PVs and/or KLK3 I179T and experienced significantly earlier PCa-specific mortality. Importantly, associations remained significant among men without metastatic disease at diagnosis.

Conclusions

Reported inherited susceptibility to PCa does not uniformly confer risk of lethal progression. Germline PVs in NCCN-recommended DDR genes and the KLK3 I179T variant identify a subset of men at elevated risk of PCSM, including those with apparently localized disease. These findings support consideration of selected germline markers for prognostic risk stratification at diagnosis to inform individualized management decisions.

Introduction

Prostate cancer (PCa) is the second leading cause of cancer-related death among men, despite most cases being diagnosed at a localized stage with favorable short-term prognosis [1, 2]. Although established clinicopathologic factors, such as PSA level, Gleason grade group, and clinical T stage, are widely used for risk stratification and recommended by National Comprehensive Cancer Network (NCCN) guidelines, a subset of men with apparently low- or intermediate-risk disease nonetheless progress to metastatic disease and ultimately die from PCa [1, 3–5].

Importantly, treatment decisions for localized PCa require careful balance. While definitive local therapy may reduce the risk of progression in higher-risk disease, treatments such as androgen deprivation therapy (ADT), often administered with radiation therapy, have been associated with increased risks of secondary primary cancers and cardiometabolic complications, contributing substantially to long-term mortality [6, 7]. Thus, accurately predicting the risk of prostate cancer–specific mortality (PCSM), particularly at the time of diagnosis and among men with localized disease, is critical for guiding treatment decisions that minimize overtreatment while reducing both PCSM and treatment-related mortality. More precise tools to identify men at elevated risk of lethal progression remain an important unmet clinical need.

Germline genetic variants represent a potentially valuable source of prognostic information [8]. Rare pathogenic variants (PVs) in multiple DNA damage–repair (DDR) genes have been associated with aggressive PCa phenotypes, including high-grade and metastatic disease [9–19]. Germline testing for selected genes is currently recommended by NCCN guidelines for subsets of PCa patients[4]. However, most supporting evidence derives from comparisons of PCa cases with unaffected controls or from selected metastatic or hereditary cohorts. Far fewer studies have evaluated associations specifically with PCSM in population-based cohorts of incident PCa patients [19]. Aside from BRCA2, evidence linking many other genes to lethal disease remains inconsistent.

In addition to rare monogenic variants, two common missense variants (KLK3 I179T and HSD3B1 1245 A > C) have been associated with PCSM in prior studies [20–23]. However, neither is currently incorporated into standard germline testing panels. Furthermore, polygenic risk scores (PRSs), which aggregate the effects of hundreds to thousands of common variants associated with PCa susceptibility, have been shown to predict PCa risk and, in some settings, aggressive disease [24–27]. Whether PRSs provide prognostic information for PCSM among men already diagnosed with PCa remains uncertain.

We therefore conducted a comprehensive evaluation of previously reported germline genetic risk factors, including NCCN-recommended genes, additional candidate DDR genes, two common variants, and three representative PRSs, in relation to PCSM within the UK Biobank (UKB), a large population-based cohort with long-term follow-up. By applying competing-risk methods to incident PCa cases, we aimed to clarify which germline risk factors are independently associated with PCSM and to assess their potential utility for risk stratification at the time of diagnosis.

Methods

Details are provided in Supplementary Methods. Briefly, Participants were men with incident PCa from the UKB, a population-based cohort of individuals aged 40–69 years at recruitment [28]. PCa diagnoses were identified through linkage to national cancer registry records (ICD-10 code C61). Men diagnosed with PCa prior to recruitment were excluded. Metastatic disease at diagnosis was defined using ICD-10 codes C77–C79 within three months of diagnosis. PCSM was ascertained through national death registry records.

We evaluated pathogenic or likely pathogenic (P/LP) variants in 11 NCCN-recommended genes and nine additional candidate DDR genes reported to be associated with aggressive PCa (Supplementary Table 1) [1, 4, 9–19]. Variants were classified according to ACMG/AMP criteria [29–31]. Two common missense variants, KLK3 I179T (rs17632542) and HSD3B1 1245 A > C (rs1047303), were also analyzed [20–23].

Three previously published PRSs for PCa were calculated using individual-level genotype data (Supplementary Table 2) [24–26, 32]. To minimize overlap with KLK3, variants within 1 Mb of KLK3 were excluded from PRS construction. PRSs were standardized to one standard deviation (SD) for primary analyses and categorized into prespecified risk strata for secondary analyses.

Associations between germline genetic risk factors and PCSM were evaluated using age as the underlying time scale with delayed entry at age at diagnosis. Fine–Gray subdistribution hazard models were used for primary inference to account for competing non-PCa mortality. Cause-specific Cox proportional hazards models were conducted as complementary analyses.

All models were adjusted for genetic ancestry using the first four principal components. Bonferroni correction was applied within each class of genetic risk factors to control for multiple comparisons.

Model calibration was assessed by comparing predicted and observed cumulative incidence of PCSM by age 75 years. Risk stratification was illustrated using cumulative incidence curves for carriers and non-carriers. Discriminative performance was quantified using time-dependent C-indices, with differences between models evaluated using bootstrap resampling (1000 replicates). Stratified analyses were performed by metastatic status at diagnosis and ancestry group.

Results

Incident PCa patients and PCSM

Among 218,964 men without PCa at recruitment, 14,644 (6.71%) were diagnosed with incident PCa after a median follow-up of 15.66 years (interquartile range [IQR], 14.8–16.41). Most cases (95.1%) had no evidence of metastatic disease at diagnosis. During a median follow-up of 6.73 years (IQR, 3.96–10.42) after PCa diagnosis, 1581 men (10.8%) died from PCa and 1385 (9.46%) died from other causes (Table 1).

Table 1.

Characteristics of prostate cancer patients included in the UK Biobank (UKB).

Characteristic	Incident prostate cancer in UKBa
Patients, N (%)	14,644 (100)
Genetic ancestry, N (%)
European (NFE, FIN, ASJ)b	13,946 (95.23)
Non-European (remaining)	698 (4.77)
Age diagnosis, years, median (IQR)	69.61 (65.38-73.86)
Evidence for metastatic disease at diagnosis,c N (%)
No	13,926 (95.1)
Yes	718 (4.9)
Follow-up time for PCSM, years, median (IQR)	6.73 (3.96-10.42)
PCSM, N (%)	1581 (10.8)
Age at PCSM, years, median (IQR)	75.35 (71.3-79.38)
Death from other causes, N (%)	1385 (9.46)

^a. Subjects with missing genotype data and with age dx before recruitment were excluded.

^b. Subjects with combined probability of >0.75 for FIN (Finnish), NFE (non-Finnish European), and ASJ (Ashkenazi Jewish).

^c. Metastatic disease with 3 months of prostate cancer diagnosis.

PCSM prostate cancer-specific mortality, IQR interquartile range.

Association of individual germline risk factors with PCSM

Using Fine–Gray competing-risk models (Supplementary Table 3), three NCCN-recommended DDR genes (BRCA2, MSH6, PALB2) were significantly associated with PCSM after Bonferroni correction (P < 0.0045). Hazard ratios (HRs) were approximately 3.4–3.5 for each gene. These genes were subsequently grouped as Tier-1 DDR genes.

The remaining seven NCCN-recommended DDR genes (ATM, BRCA1, CHEK2, MLH1, MSH2, PMS2, TP53) showed elevated but individually non-significant associations with PCSM after correction. Given their shared biological function and guideline inclusion, these were grouped as Tier-2 DDR genes.

Among nine additional candidate DDR genes (Tier-3), none met the Bonferroni-corrected threshold. A nominal association was observed for MRE11A (HR 3.40; 95% CI 1.39–8.34; P = 0.01), but this did not withstand multiple testing correction.

For common variants, KLK3 I179T was significantly associated with increased PCSM (HR 1.26; 95% CI 1.09–1.45; Bonferroni-adjusted P < 0.025), whereas HSD3B1 1245 A > C was not associated with PCSM. None of the three PRSs, modeled as continuous variables, was significantly associated with PCSM. Results were similar when excluding variants within the KLK3 region from PRS construction.

Multivariable and sensitivity analyses

In multivariable Fine–Gray models including Tier-1 genes, Tier-2 genes, and KLK3 I179T simultaneously, each remained independently associated with PCSM (all P < 0.05; Table 2). HRs (95% CI) were 3.50 (2.60–4.71) for Tier-1 genes, 1.51 (1.17–1.95) for Tier-2 genes, and 1.25 (1.08–1.44) for KLK3 I179T.

Table 2.

Association of implicated genetic risk factors with PCSM (multivariable analysis).

Genes	Subjects, N. (%)	PCSM, N. (per 100,000 person-year)	HR (95% CI)a	p-valuea
NCCN-DDR Tier-1 genes*
Non-carriers	14,475/14,644 (98.85)	1531 (1448.55)	1.00 (ref)
Carriers	169/14,644 (1.15)	50 (4699.57)	3.50 (2.60-4.71)	2.20E-16
NCCN-DDR Tier-2 genes**
Non-carriers	14,230/14,644 (97.17)	1518 (1461.68)	1.00 (ref)
Carriers	414/14,644 (2.83)	63 (2170.58)	1.51 (1.17-1.95)	1.60E-03
KLK3-I179T
Non-carriers	13,044/14,644 (89.07)	1368 (1428.54)	1.00 (ref)
Carriers	1600/14,644 (10.93)	213 (1937.54)	1.25 (1.08-1.44)	2.70E-03

*Tier 1 genes: BRCA2, MSH6, PALB2.

**Tier 2 genes: ATM, BRCA1, CHEK2, MLH1, MSH2, PMS2, TP53.

^a. Fine-Gray models accounting for competing risks (death) using age as the time scale, with left-truncation at age of diagnosis, and adjusted for top 10 PCs,

PCSM prostate cancer-specific mortality, IQR interquartile range.

As a complementary analysis, Cause-specific Cox models yielded similar results for monogenic genes and KLK3 I179T (Supplementary Table 3). In contrast, different from the Fine–Gray models, two of the three PRSs (excluding SNPs within the KLK3 region) were significantly associated with reduced cause-specific hazards of PCSM.

Risk stratification and discrimination of PCSM

Four genetic models were evaluated for risk stratification: Tier-1 genes only, Tier-2 genes only, KLK3 I179T only, and their combination. Calibration plots demonstrated reasonable agreement between predicted and observed cumulative incidence of PCSM by age 75 across risk quartiles (Supplementary Fig. 1a–d).

Cumulative incidence curves showed meaningful separation between carriers and non-carriers for all models (Fig. 1a–d). The largest separation was observed for Tier-1 genes, where carriers (1.14% of patients) reached the cohort-wide median cumulative incidence of PCSM 7.74 years earlier than non-carriers. The combined genetic model identified a larger proportion of patients at elevated risk (14.45%) and showed a median difference of 3.2 years between carriers and non-carriers. Patients carrying both Tier-1 variants and KLK3 I179T experienced the earliest PCSM, with a median difference of 10.45 years compared with non-carriers (Fig. 1e).

Fig. 1. Cumulative incidence of prostate cancer–specific mortality (PCSM) according to germline genetic risk factors.

Cumulative incidence curves for PCSM estimated using Fine–Gray competing-risk models accounting for death from other causes. Panels show comparisons between carriers and non-carriers for a NCCN Tier-1 DDR genes (BRCA2, MSH6, PALB2), b NCCN Tier-2 DDR genes (ATM, BRCA1, CHEK2, MLH1, MSH2, PMS2, TP53), c KLK3 I179T, d the combined genetic model (Tier-1, Tier-2, and KLK3 I179T), and e carriers of both Tier-1 pathogenic variants and KLK3 I179T compared with non-carriers. The horizontal dotted line indicates the cohort-wide median cumulative incidence of PCSM (11.95%), used as a visual reference to illustrate differences in PCSM-free survival between groups. Median differences in time to reach this incidence threshold are described in the text.

Discriminative performance was modest but statistically improved for the combined model compared with individual genetic models (all P < 0.01). Time-dependent C-indices (95% CI) were 0.696 (0.68–0.713) for the combined model, 0.691 (0.675–0.708) for Tier-1 only, 0.687 (0.671–0.704) for Tier-2 only, and 0.686 (0.67–0.703) for KLK3 I179T only (Table 3).

Table 3.

Discrimination performance of genetic models for PCSM.

Models	C-index (95% CI)a	P-valueb
NCCN-DDR Tier-1 genes only*	0.691 (0.675-0.708)	0.01
NCCN-DDR Tier-2 genes only**	0.687 (0.671-0.704)	1.98E-03
KLK3-I179T only	0.686 (0.67-0.703)	3.27E-04
Combined***	0.696 (0.68-0.713)	Ref.

*DDR Tier 1 genes: BRCA2, MSH6, PALB2.

**DDR Tier 2 genes: ATM, BRCA1, CHEK2, MLH1, MSH2, PMS2, TP53.

***DDR-Tier 1 + DDR-Tier 2 + KLK3 I179T.

^a. Time-dependent C-indices.

^b. Compared to the reference group (Ref.) using bootstrap resampling with 1000 replicates.

PCSM prostate cancer-specific mortality.

Stratified analyses

When restricted to men without metastatic disease at diagnosis (95.1% of cases), associations for Tier-1 genes, Tier-2 genes, and KLK3 I179T remained directionally consistent and statistically significant (Supplementary Table 4). Among men with metastatic disease at diagnosis, effect estimates were generally stronger but not consistently statistically significant due to smaller sample size.

Results were also consistent among men of European ancestry (95.2% of cases) (Supplementary Table 5). Analyses among non-European participants were underpowered for reliable evaluation of individual germline risk factors.

Discussion

In this large population-based cohort analyzed using competing-risk methods, we found that only a subset of previously reported germline risk factors for aggressive PCa was independently associated with PCSM. PVs in three Tier-1 NCCN-recommended DDR genes (BRCA2, MSH6, PALB2) were strongly associated with PCSM, with approximately threefold increased risk. In contrast, seven additional Tier-2 NCCN-recommended DDR genes were associated with PCSM only when evaluated collectively, suggesting more moderate effects that are individually underpowered to detect. The common KLK3 I179T variant was also independently associated with PCSM, whereas HSD3B1 and all three evaluated PRSs were not associated with cumulative incidence of PCSM in competing-risk models. These findings demonstrate that germline susceptibility to PCa does not uniformly translate into risk of lethal disease and underscore the importance of population-based, incident case analyses when evaluating prognostic genetic markers. Our approach was intentionally clinically oriented rather than discovery-driven, prioritizing evaluation of established germline risk factors already used or recommended in clinical practice. Although genome-wide approaches may identify novel prognostic variants, they currently lack sufficient event numbers and validation datasets for robust evaluation of PCSM.

A clinically important aspect of our findings is that these associations were observed among men initially diagnosed with localized PCa, who comprised the majority of incident cases. Treatment decisions in localized disease require balancing the risk of progression against potential harms of therapy. Although definitive treatment may reduce risk of lethal progression in high-risk disease [1, 3, 4], PCa therapies have been associated with cardiometabolic complications and secondary malignancies that may contribute substantially to long-term mortality [6, 7]. Thus, identifying men at elevated inherited risk of lethal progression at the time of diagnosis could support more individualized treatment strategies, potentially reducing both undertreatment of aggressive disease and overtreatment of indolent tumors.

Our primary endpoint was PCSM to specifically capture prostate cancer–related lethality, as all-cause mortality may dilute associations driven by tumor biology due to competing non–cancer deaths. Evaluation of overall mortality, particularly in the context of treatment-related harms, represents a complementary but distinct clinical question. Because germline variants are present at or before diagnosis and remain stable over time, they represent a uniquely early and durable source of prognostic information.

The magnitude of risk separation observed in our study further supports potential clinical relevance. Carriers of Tier-1 PVs reached the cohort-wide median cumulative incidence of PCSM nearly eight years earlier than non-carriers, and individuals carrying both Tier-1 variants and KLK3 I179T exhibited an even larger separation. Although these differences are descriptive and not causal estimates, they provide a clinically interpretable context for urologists. For context, widely used tumor-based prognostic biomarkers such as Decipher have been shown to stratify metastatic or PCa-specific mortality risk over similar multi-year time horizons [33]. Importantly, expanding beyond Tier-1 genes to include Tier-2 genes and KLK3 substantially increased the proportion of patients classified as high genetic risk, enhancing potential applicability in clinical settings.

Our findings also clarify the role of PRSs in prognostic assessment. Despite consistent associations of PRSs with PCa susceptibility [24–27], none was associated with cumulative incidence of PCSM in Fine–Gray models. In cause-specific Cox analyses, some PRSs were associated with reduced instantaneous hazards of PCSM, likely reflecting earlier age at diagnosis rather than reduced lethality. These results suggest that currently available PRSs primarily shift the timing of PCa diagnosis rather than materially influencing disease-specific mortality after diagnosis. This distinction highlights the importance of using competing-risk frameworks when evaluating prognostic biomarkers in aging populations with substantial non–PCa mortality.

An important methodological consideration is the grouping of DDR genes into Tier-1 and Tier-2 categories. Tier-1 genes were defined by statistically robust individual associations with PCSM, whereas Tier-2 genes were grouped based on shared biological function in homologous recombination and mismatch repair pathways and their inclusion in NCCN-recommended germline testing panels. This pathway-level aggregation reflects a burden-testing approach commonly applied to rare variants and was intended to address limited statistical power for individually infrequent PVs rather than to create data-driven groupings.

Gene-specific findings were largely consistent with prior literature for BRCA2 but differed for several other genes [11–19]. Although ATM and CHEK2 have been implicated in aggressive PCa in selected cohorts [9, 13–16, 18, 19], their associations with PCSM in this population-based setting were attenuated and not statistically significant after correction for multiple testing. Notably, effect estimates were directionally consistent with prior reports, suggesting that limited statistical power for rare variants may contribute to discrepancies. Conversely, MSH6 and PALB2 demonstrated significant associations with PCSM in our cohort, supporting a broader role for mismatch repair and homologous recombination pathways in lethal progression. These findings emphasize the value of large, unselected cohorts with incident cases and competing-risk analysis to reduce biases related to case selection and outcome definition.

The association between KLK3 I179T and PCSM is notable given its reproducibility across large cohorts and the absence of its inclusion in current guideline-recommended germline panels [20, 21]. In addition to its well-established associations with PSA levels and PCa detection, functional studies suggest that this missense variant alters PSA enzymatic activity, supporting a biologically plausible role in tumor progression rather than detection bias alone.

In contrast, HSD3B1 was not associated with PCSM in our cohort. Prior studies suggest its adverse prognostic impact may be most pronounced in men receiving ADT [21–23, 34–39], which was not systematically captured in this population-based dataset.

Finally, although MRE11A demonstrated only nominal association, its established role in DNA double-strand break repair and prior links to metastatic PCa suggest it warrants further investigation in larger, harmonized cohorts [40].

Several limitations merit consideration. First, detailed clinicopathologic variables at diagnosis were incompletely captured in the UKB, limiting evaluation of incremental prognostic value beyond tumor characteristics. Nevertheless, our focus on PCSM, a robust and clinically meaningful endpoint, and the use of age as the underlying time scale mitigate some concerns related to unmeasured baseline heterogeneity. Second, PCSM was ascertained using death registry records based on ICD-coded death certificates, which may be subject to some misclassification. However, cause-specific mortality derived from national registries is generally considered reliable for major disease categories such as cancer, and the use of competing-risk methods helps mitigate potential bias. Third, the cohort was predominantly of European ancestry, and analyses among minority populations were underpowered. Future studies in more diverse cohorts will be essential to confirm the generalizability of our findings and to address disparities in PCa outcomes. Forth, despite the large overall sample, statistical power remained limited for very rare PVs with moderate effect sizes. Longer follow-up and pooled analyses across diverse cohorts will be necessary to refine risk estimates and evaluate ancestry-specific effects.

In conclusion, in this large population-based cohort, a subset of germline genetic risk factors, particularly rare PVs in key DDR genes and the common KLK3 I179T variant, was significantly associated with PCSM. These findings demonstrate that not all genetic risk factors for PCa susceptibility convey prognostic value and highlight the potential role of selected germline markers in refining risk stratification and guiding treatment decision-making at the time of diagnosis.

Author contributions

Jianfeng Xu and Jim Lu conceived and designed the study. Jianfeng Xu contributed to data acquisition and cohort curation. L Lu, Z Shi, T Huy, and J Wei performed the statistical analyses. Jianfeng Xu, J Lu and Julian Xu drafted the manuscript. L Lu, Julian Xu, Z Shi, V Engelmann, H Tran, J Wei, A Ashworth, B Cornell, C Pieczonka, WB Isaacs, SL Zheng, BT Helfand, J Luo, Jianfeng Xu, and J Lu all contributed to interpretation of the data, critically revised the manuscript for important intellectual content, and approved the final version.

Funding

This research received no specific grant from any funding agency in the public or not-for-profit sectors.

Data availability

The UK Biobank data used in this study are available to approved researchers upon application to the UK Biobank (https://www.ukbiobank.ac.uk/). Derived data and analytic materials supporting the findings of this study are available from the corresponding author upon reasonable request, subject to applicable UK Biobank policies and data-use agreements.

Competing interests

Endeavor Health has agreements with GenomicMD and GoPath Diagnostics for genetic tests of polygenic risk scores. J. Xu serves a scientific advisory board member for GoPath Diagnostics and GenomicMD.

Ethics approval and consent to participate

This research has been conducted using the UK Biobank Resource under Application Number 50295. All methods were performed in accordance with the relevant guidelines and regulations. The UK Biobank study received ethical approval from the North West Multi-centre Research Ethics Committee (reference 11/NW/0382), and all UK Biobank participants provided informed consent.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41391-026-01140-5.