BRCA1 and BRCA2 pathogenic variants and prostate cancer risk: systematic review and meta-analysis

Background BRCA1 and BRCA2 pathogenic variants (PVs) are associated with prostate cancer (PCa) risk, but a wide range of relative risks (RRs) has been reported. Methods We systematically searched PubMed, Embase, MEDLINE and Cochrane Library in June 2021 for studies that estimated PCa RRs for male BRCA1/2 carriers, with no time or language restrictions. The literature search identified 27 studies (BRCA1: n = 20, BRCA2: n = 21). Results The heterogeneity between the published estimates was high (BRCA1: I2 = 30%, BRCA2: I2 = 83%); this could partly be explained by selection for age, family history or aggressive disease, and study-level differences in ethnicity composition, use of historical controls, and location of PVs within BRCA2. The pooled RRs were 2.08 (95% CI 1.38–3.12) for Ashkenazi Jewish BRCA2 carriers, 4.35 (95% CI 3.50–5.41) for non-Ashkenazi European ancestry BRCA2 carriers, and 1.18 (95% CI 0.95–1.47) for BRCA1 carriers. At ages <65 years, the RRs were 7.14 (95% CI 5.33–9.56) for non-Ashkenazi European ancestry BRCA2 and 1.78 (95% CI 1.09–2.91) for BRCA1 carriers. Conclusions These PCa risk estimates will assist in guiding clinical management. The study-level subgroup analyses indicate that risks may be modified by age and ethnicity, and for BRCA2 carriers by PV location within the gene, which may guide future risk-estimation studies.


INTRODUCTION
Pathogenic variants (PVs) in BRCA1 and BRCA2 are associated with prostate cancer (PCa) risk, but a wide range of relative risk (RR) estimates has been reported . A systematic review and metaanalysis on PCa risks for men with germline BRCA1/2 PVs (henceforth, "BRCA1/2 carriers") was published in 2019, and estimated pooled RRs of 1.35 (95% CI 1.03-1.76) for BRCA1 and 2.64 (95% CI 2.03-3.47) for BRCA2 carriers [27]. However, that meta-analysis did not consider variation in the RRs by age, PCa family history, ethnicity or PV location despite evidence of variation by these factors [1-8, 10-12, 14-17, 23, 28-33], and did not include two subsequent studies that reported prospective RR estimates for BRCA1/2 carriers: the IMPACT screening trial [20] and the EMBRACE study [23].

Study aims
This systematic review and meta-analysis aimed to synthesise the available evidence on the RRs of PCa for male BRCA1 and BRCA2 carriers, overall and by age groups, and to explore potential explanatory factors for the variation in the reported estimates by study-level covariates. Secondarily, we aimed to estimate RRs of PCa applicable to BRCA1/2 carriers with a PCa family history, and RRs of aggressive PCa.

METHODS
We sought to identify all available estimates of the RRs of PCa for BRCA1/2 carriers, based on valid study designs [34]. On June 19, 2021, the first author (TN) searched PubMed, Embase, MEDLINE and Cochrane Library with no time or language restrictions. The search query is available in the Supplementary Material. The first author removed duplicates, conference abstracts and publications that did not report original data, and screened the remaining publications based on their titles and abstracts to identify those potentially relevant. The first author thereafter screened these articles in their entirety. We contacted the authors of five articles to ask for clarifications.
We included case-control, prospective cohort and family-based retrospective cohort studies [34] that estimated the RR and 95% CI of diagnosed PCa (regardless of histopathology) for carriers of rare PVs in BRCA1 and/or BRCA2 compared to the general population or to noncarriers, or studies where RRs and/or CIs were not reported but the study provided sufficient information to allow calculation of the missing measures. Whenever available, we used estimates adjusted for age and/ or ancestry as reported in the publications. PVs were defined as any deleterious variants as determined by the study investigators or in a clinical setting to be clinically actionable based on established clinical guidelines. Studies that only reported on PVs in the two genes together, without providing separate risk estimates for BRCA1 and BRCA2 PVs, were not included. We did not include retrospective cohort studies that recruited PV carriers in clinical settings and assessed association with previous cancer diagnoses, because of the likely ascertainment bias associated with such study designs; [34] nor cross-sectional studies that compared frequencies of prevalent PCa between PV carriers and noncarriers, because prevalence ratios are unbiased RR estimates only under strong assumptions about the population incidence [35]. When data from the same study had been published more than once, we only included the most recent publication.

Statistical analysis
We used the DerSimonian-Laird method for the between-study variance [36] and derived pooled estimates according to both fixed-effects and random-effects models. To assess heterogeneity between RR estimates, we used the DerSimonian-Laird heterogeneity of effects chi-square test and reported the corresponding I 2 statistic [37]. We assessed whether the study estimates varied by covariate moderators using nested chi-square tests for categorical moderators or meta-regression for quantitative moderators [38]. To assess potential publication bias, we used funnel plots and tested for funnel plot asymmetry using the rank correlation test [39]. To assess the impact of individual studies on the results, we performed leave-one-out sensitivity analyses by omitting one of the included studies at a time and refitting the models.
For the meta-analysis by age groups, we initially considered all reported estimates by age at diagnosis with no restriction on age cutpoints considered, and also specifically those that used an age cutpoint of 65 years. For the meta-analysis of aggressive PCa, we considered studies that had exclusively or preferentially included participants with aggressive PCa, or studies that reported aggressive PCa-specific RRs, with PCa aggressiveness as defined by the study authors. In addition, when no RR of aggressive PCa had been reported but sufficient data were available within a study (e.g. Gleason score frequencies for PCa cases by PV status), we estimated the RR of Gleason score ≥7 PCa. We explored whether the variability between the estimates could be explained by the following study-level covariates (defined in Supplementary Table S1): study design; the majority ethnic ancestry of the study participants; age-adjustment approach; participant, case or control selection; use of historical or external controls; and the proportion of observed BRCA2 PVs that were located within the wide definition ovarian cancer cluster region (OCCR) [8,16,23,29,30,32,40]. We performed the meta-analyses using R software [41], with the meta [42] and metafor [38] packages.
The reported RR estimates showed a high degree of variability, particularly those for BRCA2 carriers (BRCA1: I 2 = 30%, BRCA2: I 2 = 83%; Figs. 2 and 3). The funnel plots indicated both high and low RR estimates as outliers and that smaller BRCA2 studies generally reported lower RR estimates than larger studies. However, there was no statistically significant funnel plot asymmetry (Supplementary Figs. S1 and S2).
The RR estimates from studies that selected participants for PCa diagnosis at a young age, PCa family history or aggressive PCa were higher than estimates from studies in unselected participants (BRCA1: test for subgroup differences, P = 0.056, BRCA2: test for subgroup differences, P < 0.001; Supplementary Table S2). We restricted the main meta-analysis to studies unselected for age at PCa diagnosis, PCa family history or aggressive PCa, but separately analysed these subgroups. Table 3 summarises the pooled RR estimates from the further restrictions, subgroup analyses and adjustments made in the meta-analysis.

BRCA1
Studies on BRCA1 carriers that relied on historical controls reported higher RR estimates than other studies (test for subgroup differences, P = 0.044; Supplementary Table S3). BRCA1: studies without historical controls. Restricted to studies of BRCA1 carriers that did not use historical controls, the heterogeneity between estimates was low (I 2 = 8%; Supplementary Figs. S3 and S4; Supplementary Table S4). A leave-one-out analysis identified the prospective EMBRACE study [23] as a high outlier (P = 0.013; Supplementary Table S5). The EMBRACE study reported a screening-bias-corrected estimate; [23] Table 3 shows the pooled  RR when this estimate was used instead (Supplementary Figs. S3  and S4 and Table 3).

BRCA2
BRCA2 studies in Ashkenazi Jewish men reported lower RR estimates than studies in other populations (test for subgroup differences, P = 0.011). The RR estimates were lower in studies where ≥50% of the reported PVs were located in the OCCR (test for subgroup differences, P = 0.002; Supplementary Table S3).
BRCA2: prostate cancer risk by ethnicity. Table 3 shows pooled RR estimates based on studies in Ashkenazi Jewish populations (Supplementary Figs. S5 and S6), where the heterogeneity between estimates was low (I 2 = 0%; Supplementary Tables S6  and S7).
For studies of BRCA2 carriers in non-Ashkenazi European ancestry populations (Supplementary Figs. S5 and S6), the heterogeneity between estimates was high (I 2 = 66%). A leaveone-out analysis identified three outliers (Supplementary Table S7): a UK family-based retrospective cohort study (P = 0.010) [16], the IMPACT screening trial (P = 0.013) [20], and a Dutch kin-cohort study (P = 0.017) [8]. Table 3 shows pooled RR estimates after excluding these studies. Notably, the main estimate from the EMBRACE study [23] was not an outlier among the estimates for BRCA2 carriers (P = 0.6), and if instead a screening-effect-adjusted estimate was used, the RR estimate was an outlier and significantly lower than the other estimates (P = 0.025).
BRCA2: prostate cancer risk by pathogenic variant location. Table 3 shows pooled RR estimates in studies split by OCCR proportion, before and after exclusion of the IMPACT study [20] which was a low outlier among studies with <50% OCCR PVs (P = 0.002; Supplementary Figs. S7 and S8; Supplementary Tables S8 and S9), and after restriction to the available OCCR-or non-OCCR-specific estimates.
Prostate cancer risk by age group Supplementary Figs. S10 and S11 show all reported RR estimates by the age cutpoints used to define age groups. Restricted to RR estimates by age groups younger or older than 65 years, the RRs were heterogeneous for both BRCA1 (age <65 years I 2 = 47%, age ≥65 years I 2 = 65%; Supplementary Figs. S12 and S13) and BRCA2 carriers (age <65 years I 2 = 63%, age ≥65 years I 2 = 0%; Supplementary Figs. S14 and S15).
BRCA1. The age-specific estimates from a large international kincohort study [5] were somewhat lower at age≥65 years than estimates from other studies (age <65 years P = 0.4, age ≥65 years P = 0.019; Supplementary Tables S10 and S11). However, we could not identify any likely methodological explanation for this outlying estimate and therefore retained the study. The age-specific RR estimates from one case-control study in Ashkenazi Jewish men [11] were somewhat lower at younger ages and somewhat higher at older ages than estimates from other studies (age <65 years P = 0.073, age ≥65 years P = 0.15; Supplementary Table S11) and the RR estimates from the EMBRACE study [23] were somewhat higher than estimates from other studies at both younger and older ages (age <65 years P = 0.14, age ≥65 years P = 0.11; Supplementary Table S11), but these differences were not significant. Table 3 shows the results when excluding the study in Ashkenazi men, including screening-effect-adjusted estimates from EMBRACE, or restricting to studies that did not rely on external population frequency estimates.
BRCA2. The RR estimate for younger BRCA2 carriers from one study of Ashkenazi Jewish men [11] was a low outlier (age <65 years P = 0.005, age ≥65 years P = 0.5; Supplementary Fig. S14; Supplementary Tables S12 and S13). Table 3 shows pooled RR estimates by age group before and after excluding this study.
Prostate cancer risk by family history of prostate cancer The pooled RR estimate for BRCA1 carriers with PCa family history was 2.79 (95% CI 1. 33-5.88; I 2 = 0%). Only one study reported a RR specifically for BRCA2 carriers with a family history, of 7.31 (95% CI 3.40-15.7).

DISCUSSION
A wide range of PCa RR estimates have been reported for BRCA1 and BRCA2 carriers. The results of this meta-analysis suggest that the heterogeneity may in part be explained by selection for age, family history or aggressive disease, and study-level differences in   When available, the table includes adjusted odds ratio estimates (as indicated in the "age-adjustment" field). Otherwise, the reported unadjusted odds ratio estimates by each study are included. For studies that did not report odds ratios, unadjusted odds ratio estimates calculated from the frequencies of case and control PV carriers were used in the meta-analysis (not shown in this descriptive table but included in the forest plots). b

Reported on both BRCA1
and BRCA2, but is not included in the BRCA1 meta-analysis due to observing no BRCA1 PVs in the cases which hence did not enable estimation of a 95% CI for the RR. c The main analysis in this study compared cases with aggressive prostate cancer to a combined comparison group comprising cases with non-aggressive prostate cancer and unaffected men. The meta-analysis includes the supplementary analysis of cases with aggressive prostate cancer versus unaffected men. the age and ethnic ancestry composition of the study participants, the reliance of some studies on historical controls, and the proportion of the studied BRCA2 carriers who have PVs within the OCCR. The pooled RR estimates indicate that male BRCA2 carriers are at higher than population risk of PCa at all ages, whereas BRCA1 carriers may be at somewhat increased risk with the increased risk restricted to younger ages. Based on the most restrictive inclusion criteria considered, the overall random-effects RR estimates were 2.08 (95% CI 1.38-3.12) for Ashkenazi Jewish BRCA2 carriers and 4.35 (95% CI 3.50-5.41) for non-Ashkenazi European ancestry BRCA2 carriers. This heterogeneity in BRCA2 PCa risks by ethnicity indicates the need for further research to explore ethnicity-specific risk estimates for male BRCA2 carriers. The reported RRs for African and Asian ancestry BRCA2 carriers were similar to those for non-Ashkenazi European ancestry men, but this was based on a small number of studies and should be interpreted with caution. However, even if the RRs are similar, this would translate to different absolute risks for BRCA2 carriers by ethnicity, because the baseline population risks differ between ethnic groups [43,44]. For BRCA1 carriers, there was no significant difference in reported RRs by ethnicity and the overall RR was estimated to be 1.18 (95% CI 0.95-1.46). For both BRCA1 and BRCA2 carriers, the reported RRs were higher at younger ages. Based on the most restrictive inclusion criteria, the estimated age-specific RRs applicable to non-Ashkenazi European ancestry men were 7.14 (95% CI 5. 33-9.56 The reported overall RR estimates for BRCA2 carriers were lower from studies where a majority of the BRCA2 PVs were located in the OCCR (pooled RR = 2.30, 95% CI 1.74-3.06). The metaregression showed a trend towards decreasing RRs with increasing study-level proportions of PVs located in the BRCA2 OCCR, consistent with the observations that carriers of BRCA2 PVs within the OCCR have a lower risk of PCa than other BRCA2 PV carriers [8,23,[29][30][31][32]. The Ashkenazi BRCA2 studies reported exclusively on the Ashkenazi founder PV c.5946delT that is located in the OCCR, and the RRs from these studies (pooled RR = 2.08, 95% CI 1.38-3.12) were comparable with the RRs reported from studies in non-Ashkenazi European ancestry populations where the majority of participants had PVs located in the OCCR (pooled RR = 2.53, 95% CI 1.71-3.75). Hence, as has previously been suggested [11], it is possible that the lower PCa risks observed for Ashkenazi BRCA2 carriers [3,4,6,7,11,12,30,33,45] is explained by risk variation by the location of PVs within the BRCA2 gene.
By contrast, there was no significant variation in the reported overall BRCA1 RR estimates by the ethnic ancestry of the study participants. The studies in Ashkenazi Jewish men reported exclusively on the two Ashkenazi founder PVs c.68_69delAG and/or c.5266dupC. A lack of variation in the PCa risk by specific founder PVs is consistent with previous findings of a lack of significant variation by the location of PVs within BRCA1 [31]. Moreover, the reported RR estimates were higher from two studies that compared Israeli PCa patients to controls from previous studies of US Ashkenazi individuals [4,6]. The use of cases and controls from different settings and time periods make the studies susceptible to bias from population stratification, and place-and time-specific differences in e.g. opportunistic screening rates. Only one study in Ashkenazi Jewish BRCA1 carriers had reported agespecific RR estimates [11], and these were somewhat lower for younger carriers and somewhat higher for older carriers compared to estimates from studies in non-Ashkenazi European ancestry populations. This study was however limited by the use of a selfselected sample and ascertainment bias may be likely. Hence, the finding may not be inconsistent with the finding of no significant differences by ethnicity in the meta-analysis of overall RR estimates for BRCA1 carriers.  Fig. 3 Forest plots of overall BRCA2 RR estimates. a All initially considered studies; b after restriction to studies unselected for age at diagnosis, family history or aggressive disease.   The RR estimates from the EMBRACE study were identified as high outliers among the BRCA1 but not the BRCA2 estimates. The EMBRACE study was limited by potential confounding by screening effects [23]. BRCA2 PVs are associated with a more aggressive PCa phenotype than BRCA1 PVs [11,12,20,23,46], and the results may hence reflect that BRCA2 carriers are more likely than BRCA1 carriers to have clinically significant PCa which is diagnosed regardless of screening. When we instead included BRCA1 RR estimates from a sensitivity analysis that adjusted for potential screening effects, these RR estimates were consistent with those reported in other studies. The IMPACT screening trial reported an RR estimate for BRCA2 carriers that was significantly lower than estimates from other studies. Enhanced screening makes early diagnoses of indolent tumours likely in the trial arms. Hence, bias towards the null may be expected compared to the risk for the average BRCA1/2 carrier in the population, if overdiagnosis rates are similar in the carriers and non-carriers.
One case-control study included only cases with a family history of PCa and an unselected control group, and did not adjust for this family history-based ascertainment [25]. This is likely to lead to higher RR estimates compared to RRs based on case-control studies of unselected cases, because of likely enrichment of PCa PVs in subjects from PCa families. Although such designs may provide valid tests of association, they can lead to biased RR estimates [47]. Two family-based retrospective cohort studies in relatives of breast or ovarian cancer cases reported estimates that were significantly higher [16] or lower [8] than estimates from other studies. Assuming that no other shared genetic and familial risk factors besides BRCA1/2 PVs exist between PCa, breast and ovarian cancer, such ascertainment should in principle not introduce ascertainment bias. However, given the excess breast cancer risk in relatives of PCa cases [48] and the established associations between BRCA1/2 PVs and PCa, it cannot be ruled out that testing for BRCA1/2 PVs in individuals with breast cancer may in some instances have been influenced by the presence of PCa cases in the family. If so, failing to adjust for the PCa events that determined the ascertainment would bias the resulting PCa RRs away from the null. One study included biopsy-negative individuals as controls [18], one study used controls who had other cancers [24] and two studies used controls identified in healthcare settings [21,22]. Such control selection might bias the corresponding RR estimates if the PV frequency among the controls differs systematically from the population. However, the meta-analysis did not suggest significant differences between these estimates and estimates from other studies.
The systematic review and meta-analysis has a number of strengths. Since the most recent previous systematic review and meta-analysis [27], seven studies [20][21][22][23][24][25][26] have been published, including two prospective studies [20,23] and studies in African [21] and Asian [22] ancestry populations. By incorporating these studies, we update the available evidence. Furthermore, our metaanalysis expanded on previous meta-analyses by exploring variability in risks, which identified several possible explanatory factors for the heterogeneity between studies. We have provided estimates that synthesise all available data on the RRs of PCa for male BRCA1 and BRCA2 carriers.
The systematic review and meta-analysis also has limitations. Publication bias and selective reporting of significant outcomes within studies may bias meta-analysis estimates [49]. Because only a subset of the studies reported RRs by age, family history and PV location, and RRs of aggressive PCa, such bias cannot be ruled out. Funnel plots for the age-specific estimates showed no clear asymmetry, indicating that selective reporting is less likely. Another limitation is the potential overlap between the participants of different studies. As noted above some studies used the same historical controls, and the BRCA1/2 carrier participants partially overlapped between EMBRACE [23] and IMPACT [20]. This invalidates the assumption that the RRs are estimated based on independent samples, which may bias the pooled RR estimates and underestimate the width of the associated CIs. The metaanalysis of BRCA2 OCCR PVs was limited by a lack of separate estimates of the risks associated with OCCR and non-OCCR PVs. The analysis predominantly relied on study-level data on the proportion of reported PVs that were located within the OCCR. For some studies, this proportion was based on the family-level rather than the individual-level PV distribution. However, despite these limitations, the resulting RR estimate (pooled RR = 2.30, 95% CI 1.74-3.06) was consistent with the eight separate estimates reported for OCCR PVs (pooled RR = 2.10, 95% 1.55-2.86). Risk variation by the OCCR was however not present when split by age group. This might be due to the use of the study-level proportion of OCCR PV carriers, which may be a poor proxy for the proportion of OCCR PV carriers within age-stratified subgroups of the study participants. These study-level subgroup analyses are hypothesisgenerating and larger studies are needed to estimate the agespecific risk associated with specific subgroups of BRCA1/2 carriers based on individual-level data, e.g. by ethnic ancestry and PV location. Finally, the literature search and review was performed by a single reviewer rather than several reviewers, and although the review assessed sources of study-specific bias, it did not use a standardised rating scale.

Conclusion
This meta-analysis has identified several potential effect modifiers that may guide future studies, and has provided pooled RR estimates, overall and by age group, of the risk of PCa for male BRCA1 and BRCA2 carriers that incorporate the current accumulated evidence. These risk estimates will be informative for the genetic counselling of male BRCA1 and BRCA2 carriers.

DATA AVAILABILITY
The literature review and meta-analysis datasets generated and analysed during the current study are available from the corresponding author on reasonable request.