Homeobox B13 G84E Mutation and Prostate Cancer Risk

Background The homeobox B13 (HOXB13) G84E mutation has been recommended for use in genetic counselling for prostate cancer (PCa), but the magnitude of PCa risk conferred by this mutation is uncertain. Objective To obtain precise risk estimates for mutation carriers and information on how these vary by family history and other factors. Design, setting, and participants Two-fold: a systematic review and meta-analysis of published risk estimates, and a kin-cohort study comprising pedigree data on 11 983 PCa patients enrolled during 1993–2014 from 189 UK hospitals and who had been genotyped for HOXB13 G84E. Outcome measurements and statistical analysis Relative and absolute PCa risks. Complex segregation analysis with ascertainment adjustment to derive age-specific risks applicable to the population, and to investigate how these vary by family history and birth cohort. Results and limitations A meta-analysis of case-control studies revealed significant heterogeneity between reported relative risks (RRs; range: 0.95–33.0, p < 0.001) and differences by case selection (p = 0.007). Based on case-control studies unselected for PCa family history, the pooled RR estimate was 3.43 (95% confidence interval [CI] 2.78–4.23). In the kin-cohort study, PCa risk for mutation carriers varied by family history (p < 0.001). There was a suggestion that RRs decrease with age, but this was not significant (p = 0.068). We found higher RR estimates for men from more recent birth cohorts (p = 0.004): 3.09 (95% CI 2.03–4.71) for men born in 1929 or earlier and 5.96 (95% CI 4.01–8.88) for men born in 1930 or later. The absolute PCa risk by age 85 for a male HOXB13 G84E carrier varied from 60% for those with no PCa family history to 98% for those with two relatives diagnosed at young ages, compared with an average risk of 15% for noncarriers. Limitations include the reliance on self-reported cancer family history. Conclusions PCa risks for HOXB13 G84E mutation carriers are heterogeneous. Counselling should not be based on average risk estimates but on age-specific absolute risk estimates tailored to individual mutation carriers’ family history and birth cohort. Patient summary Men who carry a hereditary mutation in the homeobox B13 (HOXB13) gene have a higher than average risk for developing prostate cancer. In our study, we examined a large number of families of men with prostate cancer recruited across UK hospitals, to assess what other factors may contribute to this risk and to assess whether we could create a precise model to help in predicting a man's prostate cancer risk. We found that the risk of developing prostate cancer in men who carry this genetic mutation is also affected by a family history of prostate cancer and their year of birth. This information can be used to assess more personalised prostate cancer risks to men who carry HOXB13 mutations and hence better counsel them on more personalised risk management options, such as tailoring prostate cancer screening frequency.


Introduction
The homeobox B13 (HOXB13) gene is involved in prostate development [1], and in vitro results have suggested that its transcription factor is involved in prostate cancer (PCa) cell growth through androgen receptor interaction, regulation by FOXA1, and other pathways [2]. The HOXB13 missense mutation G84E is a founder mutation in Nordic populations [3], with reported carrier frequencies of 0.2-1.4% [4][5][6][7], and is carried by 0.1-0.5% in other Western European populations [4,8]. Mutation carrier frequencies are lower in Southern European populations [4,9], and the variant is very rare in African and Asian ancestry populations [4,9,10]. HOXB13 G84E is associated with PCa risk, but reported relative risks (RRs) have shown considerable heterogeneity and often wide confidence intervals (CIs) [3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22]. Most risk estimates come from case-control studies, but because G84E mutations are rare, the small number of mutations in controls can lead to imprecision. Moreover, estimates may be biased if participants are not randomly recruited from cases unselected for age at diagnosis or family history, and if population-matched controls are not available. In contrast, kin-cohort or family-based studies, in which affected individuals are screened for the mutation and data on relatives are used to estimate cancer risks, enable observation of a larger number of mutation carriers, and often provide greater precision and unbiased estimates provided analyses are adjusted for ascertainment [23][24][25].
Genetic counselling for men at an elevated PCa risk has relied predominantly on family history, ethnicity [26], and BRCA1 or BRCA2 mutation carrier status [27]. Recently, a consensus conference recommended additionally testing for HOXB13 mutations [28]. However, to provide individualised counselling to mutation carriers, it is important to have valid and precise age-specific cancer risk estimates and information on how other factors including PCa family history modify these risks.
The aim of the present analysis was two-fold. First, we performed a systematic review and meta-analysis of the PCa risk for HOXB13 G84E carriers based on case selection and study ascertainment criteria. Previous meta-analyses [29,30] combined RR estimates from unselected and high-risk cases, resulting in pooled estimates that may not be widely applicable. Second, using family data from the largest kin-cohort PCa study to date in which participants were genotyped for HOXB13 G84E, we estimated relative and absolute PCa risks for mutation carriers, and assessed how PCa risks vary by family history, birth cohort, and age. We used the results to obtain clinically relevant absolute risk estimates by various PCa family history configurations, applicable to mutation carriers identified in different contexts, for example, in clinical genetics or through population-based screening programmes.

2.
Patients and methods

Systematic review and meta-analysis
We performed a systematic review and meta-analysis to synthesise the available evidence on HOXB13 G84E mutation and PCa risk. Details are given in the Supplementary material (systematic review and metaanalysis We included families of probands genotyped for the HOXB13 G84E mutation. To ensure consistency with sequential ascertainment rules [31] to obtain unbiased risk estimates, the analysis was based on systematically collected data from the proband, first-degree relatives (FDRs), and second-degree relatives (SDRs).
A previous case-control study has reported on PCa risks for G84E carriers using the UKGPCS case probands compared with healthy controls [8]. In the present kin-cohort study, we used data on the relatives of the probands and complex segregation analysis, and therefore this represents an independent dataset and analysis.
All participants provided written informed consent. The study was approved by the local medical research and ethics committees.

Genotyping
Genotyping was conducted using the Infinium OncoArray-500 K BeadChip (Illumina, San Diego, CA, USA), comprising single nucleotide polymorphisms (SNPs) for genome-wide coverage and custom SNP content selected across multiple consortia based on suspected associations with one of five common cancers [32]. The HOXB13 G84E SNP call rate was 99.99% [33]. Mutation frequencies in the probands were consistent with Hardy-Weinberg equilibrium proportions (exact test,

Statistical analysis
We followed male family members from age 35

UKGPCS families
The proband had been genotyped for HOXB13 G84E in 11 983 of 15 670 (76%) eligible families (Fig. 2). Table 1 summarises the probands' and family members' characteristics. One hundred and eighty-three probands (1.5%) carried the mutation, two of whom were homozygous carriers; the proportion of mutation carriers was highest in the familybased PRS arm (2.6%) and lowest in the population-based PRM arm (1.1%). In total, 45% of mutation carriers had at least one relative who had developed PCa compared with 30% of noncarriers; these differences were most apparent in the young-onset PRY arm. Most participants were of European ancestry regardless of carrier status. Age at diagnosis was available for 62% of FDRs and 31% of SDRs with PCa; we imputed missing ages.  Table 2). However, multiplicative, dominant and general models of inheritance provided similar fit (Supplementary Table 3), most likely due to the low number of homozygous carriers. When a familial polygenic component was included to allow for the residual familial effects not explained by HOXB13 G84E mutations, the model fit improved (Table 2). This model included calendar-periodand cohort-specific incidences that capture the changing PCa incidence over time. A model in which incidences from a single calendar period (2015) were instead assumed to apply to all family members had a worse fit (AIC = 83 526.5; Table 2; Supplementary Table 4). Thus, we chose the multiplicative polygenic model, with calendar-period-and cohort-specific incidences as the main model for all subsequent analyses. Under this model, the average per-allele RR was 3.86 (95% CI 2.16-6.88), with a risk allele frequency of 0.20% (95% CI 0.11-0.36%) and a polygenic SD of 2.72 (95% CI 2.64-2.80). In this model, the polygenic component is assumed to act multiplicatively with HOXB13 G84E. We tested this assumption by fitting a model that did not allow for this multiplicative effect (ie, by assuming a polygenic SD of 0 for mutation carriers), which had a significantly worse fit (p < 0.001). Fitting a model with separate polygenic SDs, one for mutation carriers and one for noncarriers, provided no significant evidence that the magnitude of the polygenic SD differs between mutation carriers and noncarriers (p = 0.3; Table 2;  Supplementary Table 4).

Model fitting and PCa risks
We fitted models where the RR for mutation carriers was allowed to vary with age. None of these improved the model fit significantly, but point estimates indicated higher RRs at younger ages (Supplementary Table 5). The best-fitting model with age-specific RRs was a model that allowed the RR to vary continuously with age (p = 0.068), with estimated RRs of 5.07 at age 50 that decreased to 3.70 at age 70 ( Table 2). A model with separate RRs for mutation carriers in the seven assumed birth cohorts showed higher RR estimates for men born more recently and fitted significantly better than the main polygenic multiplicative model where a single RR was assumed to apply to all mutation carriers (p = 0.014; Supplementary Table 5). However, the best fitting model with cohort-specific RRs, as determined by AIC, was a model that included an RR parameter for men born in 1929 or earlier and a separate RR parameter for men born in 1930 or later (AIC = 40609.6, p = 0.004 compared with the main polygenic multiplicative model; Table 2; Supplementary Table 5). Finally, a model that allowed for both age-and cohort-specific RRs did not have improved fit compared with the model with cohort-specific RRs (p = 0.7; Table 2). Thus, the model with birth-cohort-specific RRs was the most parsimonious; in this model, the estimated per-allele RR was 3.09 (95% CI 2.03-4.71) for men born in 1929 or earlier and 5.96 (95% CI 4.01-8.88) for men born in 1930 or later, with a risk allele frequency of 0.14% (95% CI 0.09-0.21%) and a polygenic SD of 2.72 (95% CI 2.65-2.80). Table 3 and Fig. 3 show the predicted age-specific risks of developing PCa for a HOXB13 G84E mutation carrier born in 1960 or later, based on the most parsimonious model and under different assumptions about PCa family history. The [ ( F i g . _ 1 ) T D $ F I G ] Fig. 1 -Forest plot of previous estimates of the relative risk of prostate cancer for HOXB13 G84E mutation carriers, by study design and case selection [3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22]. CI = confidence interval; RR = relative risk. 7 5 ( 2 0 1 9 ) 8 3 4 -8 4 5 average predicted PCa risk by age 85 is 62% (95% CI 47-76%) for HOXB13 G84E mutation carriers, compared with 15% for noncarriers. For a mutation carrier with an affected father, the corresponding risk estimate ranges from 69% to 92% depending on the father's age at diagnosis, and for a man with two affected FDRs, the risk estimate ranges from 70% to 98%. The predicted average risks for mutation carriers born prior to 1960 are shown in Supplementary Figure 4.
The Supplementary material (sensitivity analyses) shows the results of our sensitivity analyses. Restriction to FDRs produced similar results as for the main analysis (Supplementary Table 6), as did alternative age imputation schemes (Supplementary Table 7). Censoring all family members with missing ages at diagnosis at age 0 resulted in lower point estimates but similar results with respect to the higher RR estimates for men born in 1930 and later (Supplementary Table 7). Splitting the data by the study arm yielded somewhat higher RR estimates when the model was fitted using only the PRY compared with the PRM families (Supplementary Table 6).

Discussion
The use of genetic information is becoming increasingly important in urological practice, both to estimate risk and to target treatments. It is therefore crucial to have precise risk estimates for the known PCa susceptibility variants, including mutations in HOXB13. We have performed a systematic The results suggest that the PCa family history should be taken into account in the genetic counselling process of HOXB13 G84E mutation carriers and that a single set of penetrance estimates would not be applicable to all mutation carriers. We have presented absolute risk estimates both applicable to the average mutation carrier in the UK population and tailored to men with family history configurations typically seen in family clinics.
We used calendar-period-and birth-cohort-specific population PCa incidences that account for the rising population incidences over time. This may be particularly important given the recent more widespread use of prostate-specific antigen (PSA) testing. Despite this, RR estimates for mutation carriers were higher in more recent birth cohorts, over and above the general rise in population incidences. For example, the predicted PCa risks by age 85 are 19%, 54%, and 62% for mutation carriers born in 1909 or earlier, during 1930-1939, and in 1960 or later, respectively. This could reflect a true birth cohort effect, resulting from changes in environmental or lifestyle factors over time. Alternatively, it may also result from the possibility that men with affected relatives may be more likely to request a PSA test and that the increased availability of PSA testing thus might have resulted in clustering of PCa diagnoses in more recent generations. One way to evaluate this potential source of bias would be to assess the association with the risk of aggressive or fatal PCa only. Unfortunately, PCa in relatives was reported by the probands and data on tumour aggressiveness in relatives were not available to us. We note, however, that previous studies revealed at most minor differences in tumour aggressiveness between carriers and noncarriers of HOXB13 G84E mutation [5][6][7][8][9][10][11][12]14,16,20,21] and that familial RRs of PCa estimated on the basis of the UKGPCS population-based arm were in line with the estimates from other large epidemiological studies [40]. Previous studies reported higher RRs at younger ages [3,5,6,10]. We found some evidence of decreasing RRs with age, but a model with both age-and cohort-specific RRs did not fit significantly better than a model with only cohortspecific RRs and estimated an adjusted per-year-of-age RR of 1.00 (95% CI 0.98-1.01). This could however be due to lack of power, and our results suggest that it is difficult to distinguish between decreasing age-specific RRs and increasing risks with more recent birth cohort because the effects are confounded. Only one other study investigated variation in RRs by birth cohort but did not observe significant differences [13]; however, it was based on only 19 families with mutations.
The meta-analysis of previous studies revealed significant differences in reported RRs by case selection, in particular, when case selection depended on family history. This is consistent with the results of the present analysis a Population-based arm: men diagnosed or treated at the Royal Marsden NHS Foundation Trust at any age. b Family-based arm: men from families with at least two prostate cancer cases, one of whom were diagnosed at age 65 or earlier, or three family members diagnosed at any age. c Young-onset arm: men diagnosed at 60 or earlier. d Families in the family-ascertained PRS arm in which the proband had prostate cancer above the censoring endpoint of age 85 were retained in the study, in line with the cohort's ascertainment criteria that allow entry of families where at least three members were diagnosed at any age. e Fulfilled inclusion criteria for the PRS arm by having third-degree or more distant relatives with prostate cancer. f Men with prostate cancer at or above age 85 were censored at age 85. Therefore, they are not included in the total number of men with prostate cancer elsewhere in this table.  g The table shows   where we found that PCa risks for mutation carriers vary by PCa family history and suggest that model-based estimates, which consider family history similar to those presented here, may be more appropriate for counselling purposes. Furthermore, among unselected case-control studies, we found heterogeneity between published RR estimates as well as indications of funnel plot asymmetry. This may reflect differences in study design, selection criteria, adjustment variables used in the analysis, and/or publication bias [41], and caution is required in the use of the resulting pooled estimate.
Based on the most parsimonious model, approximately one in 360 individuals carries the HOXB13 G84E mutation in the UK. This is consistent with previous UK and Western European population estimates [4,8,9]. Based on this mutation frequency estimate and the RR estimate for men born in 1930 or later and assuming a familial RR in FDRs of 2.5 [42], HOXB13 G84E accounts for approximately 3.6% of the excess familial risk of PCa [43].
Family-based studies can produce high precision risk estimates, due to the aggregation of likely mutation carriers. However, because individuals are ascertained through affected family members and thus generally are at a higher than average risk of disease, adjustment for the ascertainment is needed to avoid biased estimates [24]. Among the previous family-based studies on HOXB13 G84E and PCa risk [3,10,13,19], only two adjusted for the ascertainment procedure or the relatedness between subjects [13,19]. Here, we adjusted for ascertainment using the ascertainment-assumption-free approach, which gives unbiased estimates provided that all information related to the ascertainment is available, however at the cost of somewhat reduced precision [37].
Strengths of our study include the large sample size and the use of the kin-cohort study design, allowing the use of cancer history information in relatives of mutation carriers. The dataset included both families ascertained through population-based PCa cases and additional families enriched for a young age at diagnosis or a family history of PCa. This provided information on mutation carriers with a wide range of ages at diagnosis and family history configurations, which enabled us to model the variation in risks by family history and other characteristics. Our modelbased estimates would thus be applicable to not only mutation carriers identified in family clinics, but also carriers identified through population-based mutation screening programmes. Limitations include the reliance on self-reported cancer family history, which can be inaccurate, particularly for more distant relatives [44]. We evaluated the impact of including information on SDRs on the results by refitting the model using only FDRs, and the estimates remained similar. Furthermore, since men were unaware of their mutation status at study entry, no differential reporting of family history by mutation status should be expected. We imputed missing ages at diagnosis in relatives using the age distribution in the population-based families, which may approximate the age-at-diagnosis variation of the general population. While imputation adds uncertainty, alternative imputation schemes based on external population incidence information or those that instead allowed studyarm-specific imputations produced similar results. When all relatives with unknown ages were censored, the estimated RRs and polygenic SD were lower and likely underestimated. However, the differences in RR estimate sizes between the birth cohorts remained, indicating that these findings are not driven by the assumed imputation scheme. In subgroup analyses, the families ascertained through a young case generally showed higher RR estimates compared with the population-based families, which may reflect a residual bias even after the ascertainment adjustment. Confidence intervals were overlapping and apparent subgroup differences could be due to chance, but we cannot exclude the possibility that imperfect ascertainment adjustment may have resulted in somewhat overestimated RRs. Finally, although a multiplicative model showed best fit, the low number of homozygous mutation carriers complicates distinguishing multiplicative from dominant or general models of inheritance.

Conclusions
We have shown that the risk of PCa for HOXB13 G84E mutation carriers varies by PCa family history and by birth cohort. The family-history-and birth-cohort-specific risks may be useful in the counselling of mutation carriers. The current estimates should be incorporated into comprehensive risk prediction models, which also consider other known genetic predisposition variants including low-risk common susceptibility alleles identified through genomewide association studies, to enable tailored clinical risk prediction of this highly polygenic disease.