Rare Variation in TET2 Is Associated with Clinically Relevant Prostate Carcinoma in African Americans

This article is featured in Highlights of This Issue, p. 1447

Prostate cancer is a large medical problem in the developed world. More than 180,000 new cases are diagnosed, and 26,000 men die from the disease each year in the United States alone (1). The discrepancy between the number of men diagnosed with the disease and the number of men who die of the disease is due in part to diagnosis of indolent and clinically unimportant prostate cancer in many men. Most men will develop histologic prostate cancer at some point in their lifetime (2), but only a subset will be diagnosed with clinically relevant disease. Many of the tumors detected by modern PSA-based screening prove to be clinically indolent. Yet prostate cancer still accounts for 9% of all cancer deaths in men in the United States. Although improvements in screening and treatment have led to a 3.6% drop in mortality per year since 1994 (1), many men still develop metastatic disease, which is uniformly fatal.

Identifying men who are at increased risk of clinically relevant disease is therefore one of the most important challenges in prostate cancer management. Unfortunately, whether the most common screening tool (PSA) effectively makes this discrimination is a matter of debate. Two large randomized trials have only added to the PSA screening controversy. While the ERSPC trial demonstrated a benefit to PSA screening, the PLCO trial did not (3, 4). Even in the ERSPC trial, over 1,000 patients needed to be screened, biopsied, and treated to save one life. It has been estimated that 50% or more of patients diagnosed by PSA screening have clinically indolent tumors (5, 6) and therefore do not benefit from screening.

A growing body of evidence suggests that the risk for aggressive prostate cancer also has a strong heritable component. Prostate cancer in general appears to have a particularly strong genetic component, with upward of 58% of disease risk attributed to genetic factors (7). Family history and ethnicity are core components of risk stratification for prostate cancer screening. The male offspring of patients who die of prostate cancer have 4 times the risk of presenting with aggressive pathology and 2 times the risk of prostate cancer death than the general population (8–10). Analysis of germline DNA may allow improved screening by identifying individuals predisposed to clinically relevant disease.

Genome wide association studies (GWAS) have identified and replicated multiple loci that influence the risk of prostate cancer (11–13). Importantly, however, most of the prostate cancer cases in these studies were diagnosed with less aggressive tumors. While retrospective analyses of these datasets (e.g., meta-analysis) have reported some loci associated with disease aggressiveness (14–18), the associations have not been independently replicated, nor do they account for a large fraction of the heritability of aggressive disease.

A fundamental weakness of GWAS efforts is that they typically do not interrogate rare or private variants (19). Surveys of well-known cancer genes in other malignancies have demonstrated the important contributions of rare variants (20–23) and current estimates suggest that rare variants account for as much as 42% of prostate cancer risk (24). Importantly, such rare variants tend to have larger effect sizes than common variants identified by GWAS, and therefore could potentially guide screening, prophylaxis, and treatment (25).

Some of the greatest disparities in cancer are in prostate cancer. African American men have the highest risk of disease, with incidence rates of 208.7 per 100,000 compared with 123.0 per 100,000 in European Americans. Mortality rates demonstrate similar trends when comparing men with African to European ancestry; prostate cancer mortality rates in these populations are 47.2 and 19.9 per 100,000, respectively (1). African American men tend to present with more advanced prostate cancer and have a worse prognosis than European American men, particularly in younger age groups (26). This disparity remains even after adjusting for access to health care, tumor grade, tumor stage, and other important clinical variables (26, 27).

Observations at known prostate cancer risk loci suggest that some of the risk and mortality disparity may arise from differences in the underlying genetic architecture. For example, while 8q24 has been identified as a risk factor for prostate cancer in men of both European and African ancestry, the risk alleles differ between ancestries (28–30). Other studies have demonstrated that variants in close proximity to risk variants in European American populations are strongly associated with risk in African American populations (31). A comprehensive study of genetic variation underlying aggressive prostate cancer risk must therefore include individuals from both ancestry groups.

Here, we describe a targeted sequencing study in aggressive prostate cancer patients and disease-free controls in African Americans and European Americans. The first stage examined the whole exome and then the second stage examined 800 promising candidates.

The samples in this study were collected at two institutions: Washington University in St. Louis, Missouri (WU), and Johns Hopkins Medical Institutions in Baltimore, Maryland (JHMI). WU samples comprised 272 cases (150 European Americans and 122 African Americans) and 300 controls (150 European Americans and 150 African Americans) recruited from urology and oncology clinics in the St. Louis metropolitan area. JHMI samples comprised 384 cases (305 European Americans and 79 African Americans) and 463 controls (305 European Americans and 158 African Americans) recruited from urology and oncology clinics in the Baltimore metropolitan area.

We included only cases with aggressive prostate cancer, which we defined as follows: evidence of metastatic disease (either pathologic or radiologic), a PSA of greater than 50 ng/mL, or Gleason grade of 8 to 10. No patients fit the criteria for familial prostate cancer (defined as 3 or more first-degree relatives with prostate cancer or 2 two relatives affected before age 55). None had a history of a familial tumor syndrome.

Ancestry-matched controls were recruited from both sites from individuals deemed to have a minimal risk of developing clinically relevant prostate cancer. Men had no personal or family history of prostate cancer, PSA < 4.0 ng/mL, and a benign digital rectal examination. WU controls were over the age of 75 years or 65 years for the European Americans and African Americans, respectively. The goal was to select controls with low risk of ever developing clinically relevant prostate cancer. JHMI controls had identical inclusion criteria except that they were age-matched to the cases.

All study subjects in all two cohorts provided informed consent under a protocol approved by Human Research Protection Offices. Genomic DNA was prepared from peripheral blood samples.

We used two strategies to select genes for targeted sequencing. First, we selected genes already implicated in prostate cancer from several sources (Fig. 1A), including genetic studies of prostate cancer susceptibility (GWAS loci), sequencing studies of prostate tumors (COSMIC mutation database), curated databases of cancer genes (Ingenuity and the Sanger Cancer Gene Consensus), and two pathway analysis tools (GeneGo and Acumenta). In total, these sources nominated 392 unique candidate genes. Although there was some redundancy among gene sets, most genes (318, or 79%) were nominated by a single source. Even the two largest sets of previously implicated genes—from GeneGo and Acumenta pathway analysis of prostate cancer—showed surprisingly little overlap: of 230 genes, only 36 (15.7%) were nominated by both pathway analyses (Fig. 1B).

To include potential novel prostate cancer susceptibility genes, we performed discovery whole-exome sequencing on a subset of individuals in our study cohort. Our goal was to identify genes with a differential burden of rare coding variants between cases and controls, independent of prior implication in prostate cancer. We therefore generated exome sequencing data for the WU population: 572 individuals: 272 cases (150 European Americans and 122 African Americans) with aggressive disease and 300 race-matched controls (150 European Americans, 150 African Americans).

We analyzed populations separately and computed the burden of rare (MAF < 0.05), deleterious variants among cases and controls on a gene-by-gene basis (Supplementary Fig. S2). We selected 474 genes that were enriched or depleted for rare variants in cases relative to controls at a nominal level of significance (P < 0.05). Of these, 66 were among the genes already compiled from previous sources (Fig. 1C). The full set thus comprised 800 genes (392 known, 408 potential novel; see Supplementary Table S1) which are the focus of our targeted sequencing and analysis.

Exome capture was performed on genomic DNA from 272 cases and 300 controls (compromising the entire WU population) using the Nimblegen SeqCap EZ v2.0 exome kit according to the manufacturer's protocols (Roche). The remaining samples (compromising the entire JHMI population) underwent targeted capture with a custom-designed Nimblegen reagent for the exons of the 800 selected genes (total space: 6.76 Mbp; design available upon request).

Exome and custom capture libraries underwent paired-end sequencing (2 × 101 bp) on Illumina HiSeq2000 instruments according to the manufacturer's protocols (Illumina). On average, 17.2 Gbp and 0.48 Gbp of raw sequence data were generated for each exome and custom capture library, respectively. On average, 87% of target bases were covered at ≥20× and 93% were covered at ≥10× in the exome data. Fifteen samples (3 European American cases, one African American case, and 11 African American controls) failed to meet our threshold for minimum coverage (>70% of bases covered ≥20×) and were removed prior to analysis.

All samples that underwent exome sequencing were genotyped using Affymetrix 600K SNP arrays according to the manufacturer's protocols. We performed multidimensional scaling (MDS) clustering of autosomal SNPs using PLINK v1.07 to obtain the first 10 dimensions (–mds-cluster 10). MDS plots of the first two dimensions readily distinguished African American from European American samples (Supplementary Fig. S1).

Data were aligned with BWA-MEM v0.7.10 with quality trimming (−q 8) to remove low-quality bases at the ends of reads to the GRCh37-lite reference sequence. Data from individual runs were merged (if necessary) with Picard (http://picard.sourceforge.net) v1.113. Duplicates were marked using Picard MarkDuplicates. Reads marked as duplicates were retained in the alignment file (BAM file) but not used for subsequent analysis.

Single-nucleotide polymorphisms were called using Samtools v1.16 (samtools pileup –cv) with default settings and VarScan v2.3.6. Variant calls from both tools were filtered to remove false positives as previously described, and then merged (unique-union) using joinx v1.3 (https://github.com/genome/joinx). Conflicts where both callers predicted a variant at a position, but with differing genotypes, were resolved by taking the Samtools prediction. Small insertions and deletions (indels) were called using VarScan v2.3.6 and filtered using a different set of parameters.

MDS clustering of SNP array data was performed using PLINK v1.07 and specifying 10 dimensions (−k 10). In the discovery phase, single variant tests and SKAT aggregation tests were performed using a pipeline of Perl scripts and R libraries. Following the replication phase, MDS clustering was performed for all samples using the common (MAF > 1%) variant calls in the shared target region set. For the combined analysis (discovery + replication samples, 800 genes), the SKAT-O aggregation tests were performed using the EPACTS framework. Fisher exact tests (FET) were conducted using a customized Perl script and the fet() function of the R package.

All aggregation tests utilized only variants that were rare (defined as MAF < 5% in the population set) and either truncating (frameshift, essential splice site, and nonsense) or missense and predicted to be deleterious (by at least one of Polyphen, SIFT, or Condel) as annotated by Variant Effect Predictor (VEP) release 74. The analysis of rare truncating mutations, however, only included variants annotated as nonsense (SNVs only), essential splice site (SNVs/indels), or frameshift (indels only).

The goal of this study was to identify rare variation associated with clinically relevant prostate cancer in two ethnic groups: African Americans and European Americans. We recruited 652 patients with aggressive prostate cancer and 752 healthy age-matched controls from the urology and oncology clinics at two institutions (Table 1). Because whole-exome sequencing of 1,400 samples was not feasible, we instead sought to identify a subset of genes in which rare variants might affect risk of aggressive prostate cancer.

We combined both exome and targeted sequencing data for the 800 selected genes, which yielded a total sample size of 497 African Americans (200 cases and 297 controls) and 907 European Americans (452 cases and 455 controls). To identify rare variants associated with aggressive prostate cancer, we performed two gene-level aggregation tests, FET and SKAT-O. Only rare (MAF < 0.05) deleterious variants were included, and populations were analyzed separately and combined (Fig. 2).

To correct for multiple testing, we established a threshold of 3.125 × 10⁻⁵ (0.05/1,600 tests) for study-wide significance. The only gene to meet that threshold was TET2 in African Americans, in which 24.35% of cases harbored a rare coding variant compared with 9.61% of controls (FET P = 1.84 × 10⁻⁵, OR = 3.0; SKAT-O P = 2.74 × 10⁻⁵). Rare variants unique to cases clustered in at least two regions of the gene (Fig. 3A), suggesting that rare variation in specific protein domains drives this association.

In addition, we identified an additional 56 genes for whom there was enrichment of rare variants among either cases or controls in one of our gene-based tests which was suggestive of an association (P < 0.01). Eight demonstrated suggestive evidence of enrichment in both statistical tests (Table 2).

Of the additional 4 genes identified in African American cases and controls, 3 demonstrated enrichment of rare variants in the cases (CEP63, TRBV7-7, and NUBP2) and one demonstrated enrichment of rare variants in the controls (SPRR3). No gene reached our study-wide significance threshold in European Americans. However, 2 genes demonstrated suggestive evidence of association using both statistical tests. Both ZSWIM2 and PARP2 variants were more common in controls than in cases suggesting a protective effect.

We also analyzed European and African American samples together, using individual ancestry as a covariate (SKAT-O only). TET2 and indeed virtually all of the promising genes showed weaker association in the combined set, suggesting that the associations (if true) are specific to only one ancestry. However, an exception was PARP2. The combined analysis yielded stronger evidence of association for PARP2 by both tests (FET P = 7.06 × 10⁻⁴, OR = 0.3; SKAT-O P = 1.88 × 10⁻³), though it still did not reach our threshold for study-wide significance.

A key advantage of sequencing studies over GWAS is the ability to identify rare truncating mutations (nonsense, splice site, and frameshift) that are poorly interrogated by most SNP arrays. Only 226 (∼28%) of the genes surveyed here harbored a rare truncating variant in at least one sample. Of those, five exhibited an excess of rare truncating variants in cases relative to controls (Table 3) FET P < 0.05. Although none of these achieved the threshold for study-wide significance, the presence of three well-studied cancer predisposition genes (BRCA1, BRCA2, and MSH6) lends support to the growing body of evidence that rare loss-of-function variants in the DNA repair pathway may contribute to inherited prostate cancer risk.

In this study, we performed targeted sequencing of 800 genes in men from two ancestry groups (African Americans and European Americans) to identify rare variation associated with clinically relevant prostate cancer. Genes were chosen for analysis on the basis of previously implicated gene (known set), implicated pathway (pathway set) or whole-exome sequencing (discovery set). Most of the prior genetic studies either included benign as well as clinically relevant tumors (11–13) or were retrospective analyses (14–18) whose reported associations have not been independently replicated. Importantly, all of these studies relied on GWAS approaches, which generally do not interrogate rare genetic variation.

The expectation that rare variants could predispose individuals to prostate cancer is consistent with well-established models of susceptibility for other cancer types. In breast cancer, for example, up to 24% of patients carry germline loss-of-function variants BRCA1/2 or one of ten other genes involved in genome stability and DNA repair (32). While the variants are independently rare, they confer an increased risk that can be as high as 10-fold, and the cumulative effect from multiple genes is even larger (33, 34).

Several studies have demonstrated that rare variants in specific genes play a role in the hereditary predisposition to prostate cancer. Ewing and colleagues demonstrated that the G84E variant of HOXB13 was associated with a more than 20-fold increased risk of prostate carcinoma (35). Subsequent studies have demonstrated a more modest 3- to 4-fold increased risk; however, this demonstrates that a single rare variant found in less than 1% of the general population can account for an appreciable increased risk of disease (20). Mononen and colleagues examined a single rare SNP in the hormone-binding region of AR (R726L) which altered the transactivation specificity of the receptor. In a Finnish population, the rare leucine allele confers a 6-fold increase in risk for prostate cancer (36). A recent multi-ancestry study estimates that 42% of prostate cancer genetic risk is due to variants with frequencies between 0.1% and 1% (24).

We demonstrated that rare variation in TET2 is associated with aggressive prostate cancer in African Americans. TET2 variants appear to cluster in at least two domains (Fig. 3A) and only demonstrated evidence of association in African Americans. TET2 is expressed in many tissues including prostate (37). It is thought to be a tumor suppressor on the basis of ∼15% of myeloid cancers harboring somatic mutations (38). Prior work has identified variants in TET2 in GWAS studies (39). Eeles and colleagues examined 1,854 prostate cancer cases and in 1,894 controls and identified 7 novel variants associated with prostate cancer risk, one of which was in close proximity to TET2. Eeles and colleagues found that the TET2 variant was protective in European American, which is not consistent with it being a tumor suppressor gene. Interestingly, in contrast, the same variant was associated with nonstatistically increased risk in the relatively smaller African American subgroup. It is plausible that TET2 could play a role in both ethnicities but through different mechanisms of action or the genetic architecture differs between African Americans and European Americans. The fact that variants increase the risk of aggressive prostate cancer is supported by findings that somatic mutations in TET2 may contribute to metastatic potential in prostate cancer (40). At least some of the TET2 variants identified in cases would be predicted to be function altering, e.g., C1273F affecting coordination of Zn++ (41).

Our study also highlighted some genes with suggestive evidence of association that did not reach study-wide significance, most notably those in the DNA repair pathway. BRCA1, BRCA2, MSH6, and PARP2 were all implicated, predominantly in the European American, raising the possibility that these are important in this population.

Our observation that ∼3% of aggressive prostate cancer cases harbored rare truncating mutations in BRCA1 or BRCA2 adds to the growing body of literature implicating them as bona fide prostate cancer predisposition genes. Prior work has identified single base pair variants (as opposed to frameshift indels) in BRCA1 and BRCA2 as risk alleles for prostate cancer in patients in familial cohorts. Our results implicate truncating variants. The only frameshift variant identified in the control population was at the 3′ end of the gene and therefore less likely to alter function (Fig. 3B). Comparison of known carriers of BRCA1 and BRCA2 mutations has demonstrated a 4- and 5-fold increased risk of prostate cancer, respectively (21, 22) and in particular to an increased risk of aggressive disease (23, 42–44). Of greatest interest, men with BRCA variants undergoing curative therapy for localized prostate cancer demonstrated decreased metastasis-free survival and overall survival compared with men without germline variants (23). A recent sequencing study of 150 metastatic prostate tumors reported BRCA1/BRCA2 somatic alterations in 13.3% of cases (45). The observation that 90% of patients with germline alteration in BRCA2 exhibited biallelic loss in metastatic deposits (45) supports a causal role for loss-of-function mutations in this gene. Our study represents an important addition to this prior work, in that we examined patients without a family history of cancer syndromes.

Less is known about the role of other candidate DNA repair genes in prostate cancer. Analysis of Lynch families revealed that they had a 5-fold increased risk of prostate cancer; however, only 2 of the 11 men diagnosed had MSH6 mutations and the syndrome was not associated with aggressive disease. PARP2 variants clustered in a known regulatory domain (Fig. 3C). A recent familial study of patients with prostate cancer also identified PARP2 as a candidate susceptibility gene and reported a variant (A283G) in the same regulatory domain (46). That study, along with the report that pharmacologic PARP inhibition is therapeutic for advanced prostate cancer (47), lends further support to the idea that disruption of the PARP-associated DNA repair pathway may protect against aggressive disease.

A possible concern about our study is that we examined aggressive prostate cancer patients and disease-free controls, but not patients with indolent or low-risk disease. Although it would be interesting to examine rare genetic variation across the full spectrum of patients, we excluded those with indolent disease for two reasons. First, patients with indolent prostate cancer are clinically similar to patients without disease because both groups have excellent survival rates when given close observation (48). Second, we excluded patients with indolent tumors because the clinical course remains unknown: without intervention, they might very well have progressed to aggressive disease (case status). However, it is important to recognize that because we did not examine indolent disease, the variation identified could be associated with general risk of prostate cancer, not specific to aggressive disease. A study examining these variants in patients with indolent disease will be needed to prove the association is specific for aggressive prostate cancer.

Sample size is often the main limitation of genetic studies, due to either cost or the availability of samples. In our target sequencing cohort, we decided to focus on 800 genes (rather than the entire exome) so that we could sequence all available samples while controlling costs. Yet even our entire cohort (652 cases and 752 controls) is modest in size. Unlike indolent prostate cancer, aggressive, potentially lethal prostate cancer has a relatively low incidence; ascertaining thousands of samples remains a significant challenge. Even so, to our knowledge, our study of 800 genes in ∼1,400 samples represents the largest germline sequencing study of aggressive prostate cancer to date. As the cost of sequencing continues to fall, more ambitious studies that use exome or whole-genome sequencing in larger populations may become feasible.

We applied targeted sequencing of 800 genes to in over 1,400 individuals and found that rare variation in TET2 is associated with aggressive prostate cancer in African Americans. We also identified several genes with promising (if not statistically significant) evidence of association, including several members of DNA repair pathways. If confirmed in clinically relevant populations, these targets have the potential to identify a population for targeted screening, prophylaxis, and treatment.

I.B. Borecki is Head Analytical Genetics at Regeneron. No potential conflicts of interest were disclosed by the other authors.

Conception and design: D.C. Koboldt, D.E. Larson, R.S. Fulton, I.B. Borecki, R.K. Wilson, E.R. Mardis, A.S. Kibel

Development of methodology: D.C. Koboldt, K.L. Kanchi, R.S. Fulton, A. Kraja, R.K. Wilson, A.S. Kibel

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R.S. Fulton, W.B. Isaacs, R.K. Wilson, E.R. Mardis, A.S. Kibel

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.C. Koboldt, K.L. Kanchi, D.E. Larson, W.B. Isaacs, I.B. Borecki, L. Jia, R.K. Wilson, E.R. Mardis, A.S. Kibel

Writing, review, and/or revision of the manuscript: D.C. Koboldt, K.L. Kanchi, W.B. Isaacs, A. Kraja, R.K. Wilson, A.S. Kibel

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B. Gui, D.E. Larson, R.K. Wilson, A.S. Kibel

Study supervision: R.K. Wilson, E.R. Mardis

A.S. Kibel was supported by the Anthony DeNovi Fund, the Donald C. McGraw Foundation, and the St. Louis Men's Group Against Cancer. W.B. Isaacs was supported by William Gerrard, Mario Duhon, John and Jennifer Chalsty and the Patrick C. Walsh Cancer Research Fund. The remaining authors were supported by the National Human Genome Research Institute (grant #5U54HG003079, P.I. R.K. Wilson).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.