Introduction

Most DNA methylation marks are erased before or soon after conception, however methylation marks are known to be effectively inherited from parents to offspring in rare cases, e.g. near MLH1 and MSH2 [1]. Such heritable methylation marks can be caused by genetic variation, in which case the relevant genetic loci are called methylation quantitative trait loci (mQTLs), or due to non-genetic causes, in which case the methylation mark is called an epimutation [2].

Some epimutations, and most methylation marks that are strongly associated with an mQTL, are systemic (i.e. they affect all tissues) [3], which makes them easily detectable in blood and enables an assessment in blood to predict cancer risk in other tissues. Epimutations can mimic germline pathogenic variation, and their contribution to cancer predisposition may have been underestimated because previous studies have mainly used candidate-gene approaches, though some genome-wide searches for heritable methylation marks have more recently been conducted, e.g. [3, 4].

One of the most important risk factors for prostate cancer is having a family history of the disease but less than half of the familial risk for prostate cancer is explained by the currently identified genetic risk factors. This is despite modern genomic studies of prostate cancer risk being based on tens of thousands of cases, suggesting that many heritable risk factors for prostate cancer might exist that are not genetic. We therefore conducted a systematic, genome-wide search for heritable methylation marks associated with prostate cancer risk, and assessed the extent to which their variation was explained by common genetic variants.

Methods

Our study was based on previously reported methods [3, 5, 6] and was conducted in two phases, a family-based phase and a population-based phase (see Supplementary Methods for more details).

The family-based phase was based on 25 multiple-case prostate cancer families drawn from the Australian Prostate Cancer Family Study (APCFS​) [7]. Peripheral blood DNA methylation was measured for 133 of the 469 family members using the Infinium methylation EPIC array. To identify heritable DNA methylation marks (whether epimutations or mQTLs), we calculated a measure of heritability, \(\varDelta l\), for each methylation mark, with high values of \(\varDelta l\) corresponding to Mendelian patterns of inheritance within the families [3]. Methylation marks on sex chromosomes or within 10 base pairs of a known SNP were excluded.

Because we were interested in causes of familial prostate cancer, we selected only the 1,000 most heritable methylation marks. For each of these, we calculated the probability that each family member carries a hypothetical genetic variant causing aberrant methylation at the mark, based on M-values and family structure but not ages or affected statuses. We tested these carrier probabilities for association with prostate cancer using Cox proportional hazards survival models. We accounted for multiple testing (for 1,000 tests) using the Bonferroni p-value threshold of 0.05/1000. Risk estimates from the family-based phase are biased by ascertainment so are not presented, though p-values are valid because the test statistic is not affected by ascertainment under the null hypothesis.

The population-based phase was based on unrelated individuals recruited irrespective of family history to the Melbourne Collaborative Cohort Study (MCCS) [8, 9]. Peripheral blood DNA methylation was measured in 869 incident cases (including 430 aggressive cases) and matched controls (matched on year of birth, year of blood draw, country of birth, and sample type) using the HM450 array, as described previously [8, 9]. This data was used to further investigate the marks from the family-based phase that are heritable, associated with prostate cancer risk, and common to the EPIC and HM450 arrays (the two arrays used in the two phases). These marks were tested for association with prostate cancer using conditional logistic regression adjusted for body-mass index, tobacco smoking, alcohol consumption, age at blood draw and estimated blood cell composition. A genome-wide search for mQTLs was also conducted for these methylation marks, using 4,307 unrelated MCCS participants genotyped on the OncoArray-500 K BeadChip [6]. Sites near VTRNA2-1 have bimodal distributions so, as a sensitivity analysis, we also dichotomised the methylation values of these sites and estimated their associations with prostate cancer using conditional logistic regression, as in the main analyses, above.

Results

The 1,000 most heritable methylation marks from the family-based phase are listed in Supplementary Table 1. Of these 1,000 methylation marks, 41 were associated with prostate cancer risk at the Bonferroni-corrected significance level (Table 1).

Table 1 The 41 heritable methylation marks associated with prostate cancer risk from the family-based phase and, for 25 of these marks that were measured in the population-based phase, their association with prostate cancer risk (overall and aggressive) in the general population
Full size table

Of the 41 methylation marks from the family-based phase, 25 were included on the HM450 array and so had been measured in the population-based phase. These 25 marks were tested for association with prostate cancer, and nominally significant associations (p < 0.05) with aggressive prostate cancer were found for all 9 marks near VTRNA2-1 (Table 1), with most remaining nominally significant after dichotomising (Supplementary Table 2), (as previously reported, based on the same datasets [5]). A genome-wide search for mQTLs showed that the marks in the VTRNA2-1 region had either no mQTLs or few and weak mQTLs, while most of the other methylation marks were associated with a substantial number of mQTLs and a large proportion of their variance was explained by a single SNP (Table 1).

Discussion

Our study has identified 41 methylation marks associated with familial prostate cancer, and 9 of these marks (near VTRNA2-1) also have nominally significant associations with aggressive prostate cancer risk in the general population. Note that we would not expect all 41 methylation marks to be associated with risk in the population, e.g. BRCA1 is usually not detected by genome-wide association studies. Also, the magnitude of risk in the population-based and familial settings could differ greatly, due to rare mQTLs or epimutations causing changes in methylation that are many times the population standard deviation.

Nine of the 41 heritable methylation marks associated with prostate cancer risk are in the imprinted VTRNA2-1 region, with a loss of imprinting in this region consistent with a Mendelian pattern of inheritance; see also [3]. Imprinted regions are often associated with tissue growth, and a loss of imprinting can be linked to tumorigenesis, as is well-described for the H19/IGF2 region. VTRNA2-1 has tumour suppressor gene properties, as it regulates cell growth via inhibition of protein kinase RNA-activated (PKR). Down regulation of VTRNA2-1 in a variety of tumours and cancer cell lines has been well documented and associated with promoter CpG hypermethylation. As we found previously using the same datasets, VTRNA2-1 methylation marks are associated with aggressive prostate cancer in the population [5] and are largely independent of the underlying genetic sequence [6]. We and others have described the VTRNA2-1 locus as a metastable epiallele because the loss of imprinting in this region occurs systemically, can be modulated by the periconceptional environment and persists through adulthood [10].

Seven of the heritable methylation marks associated with prostate cancer risk are located at peptidase M20 domain containing 1 (PM20D1), a known methylation and expression quantitative trait locus associated with risk of Alzheimer’s disease. These and the other annotated and unannotated CpGs identified in this study require further research to understand the biological explanation for their association with heritable prostate cancer risk. Non-genetic causes of heritability could not be investigated in the current study, but it is possible that familial environmental and lifestyle factors play a role in determining DNA methylation at these loci.

Despite having a modest sample size, we were able to identify many heritable, cancer-associated methylation marks. Excluding non-heritable methylation marks before testing for association with cancer excludes many marks that cannot cause familial cancer, while retaining any that can. Our method of excluding non-heritable marks before testing for association with disease is therefore a very powerful way of enriching the candidate set of methylation marks for those that could cause familial disease. Further discussion of the methodology can be found in Joo et al., [3].

Heritable methylation marks can mimic the effects of genetic variants, so identifying them is similar in many ways to finding genetic loci that are associated with cancer. As for genetic loci, these marks can implicate new biological mechanisms and therefore shed light on the processes of prostate cancer initiation and progression. These methylation marks could also be used in risk-prediction algorithms to give more precise estimates of a person’s risk of prostate cancer, and so provide more tailored screening.

Our study has several strengths, including its use of an innovative method to identify heritable methylation marks, its method of enriching the candidate set of methylation marks for those that could cause familial disease, and its use of a cohort study to further investigate the findings from the family-based phase. The main weakness of the study was that some of the marks from the family-based phase could not be investigated in the population-based phase, due to the use of different arrays in the two phases. Our modest sample size is also a weakness, though this is offset by the enrichment step described above.

In summary, our study has identified 41 heritable methylation marks that are associated with prostate cancer risk in the context of multiple-case families, with 9 of these marks near VTRNA2-1 likely to be associated with aggressive prostate cancer in the general population.