Exome Sequencing of African-American Prostate Cancer Reveals Loss-of-Function ERF Mutations

Extensive prior work has explored socioeconomic contributions to prostate cancer disparities; however, our knowledge of the extent to which molecular and genetic mechanisms may also contribute to prostate cancer disparities has been limited (1–6). The notion that somatic genetic factors may influence tumor biology differentially across distinct ancestral backgrounds is exemplified by high EGFR mutation rates in patients of Asian ancestry with non–small cell lung adenocarcinoma (up to 50% of patients), compared with patients of European ancestry (10%–15% of cases; ref. 7). Large-scale genomic characterization studies are predominated by tumor samples from patients of European ancestry (8). Although these studies are of immense value, their limited racial and ethnic diversity may preclude the detection of genomic events and patterns that are unique or enriched in underrepresented groups. For example, large-scale studies such as The Cancer Genome Atlas (TCGA) have examined the genomic landscape of primary prostate cancer and have been confined mainly to men of European ancestry (81.1%, 270/333; refs. 9–11). In particular, African-American (AA) men, who have a 1.4-fold higher incidence and 2.4-fold higher mortality rate from prostate cancer compared with non-Hispanic whites, have been underrepresented in most systematic studies of prostate cancer performed to date (10–12).

We hypothesized that differences in mutational events in AA prostate cancers may in part underlie these disparities in outcomes. We also reasoned that the power to discover novel cancer genes might increase through inclusion of diverse ancestral backgrounds in large-scale cancer genome studies. To test these hypotheses, we performed whole-exome sequencing on a discovery set of 102 localized primary prostate tumors and matched normal controls from a cohort of AA men and performed targeted sequencing on an extension set of 90 primary prostate tumors.

We focused on intermediate- and high-risk prostate cancers corresponding to Gleason grades 7 and higher, or pathologic stages pT2a–pT3c (Supplementary Tables S1 and S2). Exome sequencing identified 3,059 somatic mutations, corresponding to a median of 7 silent and 23 nonsilent mutations per tumor (range, 0–19 silent; 4–47 nonsilent). The median mutation rate for this cohort was approximately 0.83 mutations/Mb (range, 0.11–1.75), similar to mutation rates in exome sequencing cohorts of primary prostate cancer drawn predominantly from men of European ancestry (10, 11).

Overall, the majority of tumors from AA men do not harbor recurrent mutations in known cancer driver genes. Instead, most AA prostate cancers harbor somatic copy-number alterations (SCNA) that are characteristic of those seen in other published cohorts (Fig. 1 and Supplementary Fig. S1A; refs. 11, 13). However, the overall frequencies of SCNAs appear to be lower in this AA prostate cancer (AAPC) cohort. Comparing the frequency of SCNAs in primary prostate cancer, we found that of 19 loci that undergo recurrent copy-number changes, 16 loci were altered at a lower frequency in the AAPC cohort compared with the TCGA cohort (Supplementary Fig. S1B, Fisher exact test; P < 0.05). Adjusting for a lower threshold for SCNAs in the AAPC cohort (see Supplementary Methods), we found that 10 of the 19 loci were still significantly different in frequencies between the two cohorts (Supplementary Fig. S2). Given that the TCGA dataset comprises higher frequencies of Gleason 8 and higher tumors in comparison with the AAPC cohort, these copy-number differences were less pronounced when stratified by Gleason score yet persisted at certain loci (Supplementary Table S3 and Supplementary Fig. S2). For example, overall, PTEN deletions were more common in the TCGA cohort (32%) compared with the AAPC cohort (6%), consistent with previous reports, and this difference remained when examining Gleason 7 and Gleason 8 (or higher) tumors (Supplementary Fig. S2; ref. 2). We note that focal copy-number gains at 17q25.3, a locus containing the gene FASN and a significant amplification by GISTIC analysis in the AAPC cohort, occur in 12.7% of AAPC tumors, in contrast with 1% in the TCGA dataset (P = 0.0001, Fisher exact test; Supplementary Figs. S1B and S3; refs. 14, 15). The mean fraction of the copy number–altered genome in the AAPC cohort was lower in comparison with the TCGA cohort (7.94% vs. 15.6%; Wilcoxon rank sum test, P = 0.0025; Supplementary Fig. S4). When we stratified by Gleason grades, the mean fraction of copy number–altered genome did not differ significantly between Gleason 7 and lower tumors but did differ between the AAPC and TCGA cohorts in Gleason 8 and higher tumors (12.8% vs. 25.7%, Wilcoxon rank sum test, P = 0.017; Supplementary Fig. S4).

In the AAPC cohort, a focused germline analysis for pathogenic mutations in genes in the DNA repair pathway revealed 4 patients with germline mutations in BRCA1 (with a concomitant hemizygous BRCA1 deletion), CHEK2, and ATM for an overall prevalence of 3.9% (Supplementary Table S4; ref. 16).

In the analysis of somatic mutations, we identified three genes (SPOP, ERF, and FOXA1) wherein recurrent base mutations reached statistical significance in the discovery cohort (FDR q < 0.1; Fig. 1; Supplementary Table S5; ref. 17). Of the significantly mutated genes, SPOP and FOXA1 have previously been identified as drivers in primary prostate cancer; however, ERF has not been implicated in this setting (10, 11). ERF is a member of the ETS transcription factor family and therefore was of interest given the prominent role that ETS transcription factor rearrangements play in prostate cancer (18). Of the five nonsynonymous ERF mutations present in our discovery cohort, three were loss-of-function events (R183*, K91fs, and R218*) and another, which occurred within the ETS DNA-binding domain (Y89C), was predicted to be a damaging event by Polyphen-2 analysis (19). In order to determine whether ERF mutations led to decreased ERF expression, we tested whether ERF mutants in prostate tumors from the AAPC cohort showed loss of ERF mRNA expression. We used RNA hybridization (RNAish) to show that ERF mutants from the AAPC cohort were associated with a significant loss of ERF mRNA expression (Fig. 2A–C and Supplementary Fig. S5).

Using GISTIC analysis to examine significant SCNAs, we also noted that a focal deletion occurred at chr19q13.2 harboring a number of genes including ERF and a known tumor suppressor, CIC (Supplementary Table S6; ref. 14). This peak represented three hemizygous copy-number losses (∼3%) of ERF in our exome discovery cohort (Fig. 2D). These 3 patients had tumors with higher risk features: Gleason 8, pT3b; Gleason 8, pT3b with PSA of 42.3; and Gleason 7, pT2c with PSA of 12.8; 2 of these patients had biochemical recurrences. We verified copy-number loss of ERF in an AAPC tumor using FISH (Fig. 2E and F). To extend the finding of ERF copy-number loss, we assessed three prostate cancer cohorts for ERF deletion by FISH analysis. Only 1 of 105 cases of localized prostate cancer in AAs in a previously published cohort demonstrated hemizygous ERF deletion (2). None of 33 cases of localized prostate cancer in the predominantly white Early Detection Research Network cohort showed deletion in ERF. This evaluation included all tumor nodules in the same prostate gland, when multiple foci were present. We also interrogated 82 cases of advanced castration-resistant prostate cancer (CRPC), four of which harbored ERF hemizygous deletions (∼5%; ref. 20). These cases were part of the Precision Medicine Clinical Trial at Weill Cornell, and an updated analysis of the whole-exome sequencing data from this cohort shows 8 of 175 cases (∼5%) harbored genomic alterations of ERF, including 6 cases with deletion (3 hemizygous and 3 homozygous), 1 case with hemizygous deletion and concomitant H31P missense mutation, and 1 additional case with a G299 frameshift deletion (Supplementary Table S7; ref. 21).

To determine whether ERF might represent a recurrently mutated gene in primary AAPC, we expanded the dataset by including an additional 90 prostate cancer samples (AAPC extension cohort; Supplementary Table S8) and performed targeted hybrid capture sequencing for 41 known or putative prostate cancer genes (Supplementary Table S9). Tumor–normal pairs from this additional AA prostate cancer cohort were sequenced at high coverage (mean target coverage: tumor 335x, normal 348x; Supplementary Methods). In the extension cohort, we identified five additional nonsynonymous ERF mutations, three of which were predicted loss-of-function frameshift mutations (Fig. 3A and Supplementary Table S10). In total, the prevalence of ERF mutations in the discovery and extension AA primary prostate cancer cohorts was 5.2% (10/192). We validated by Fluidigm array 7 of 7 of the ERF mutations that we were able to evaluate (Supplementary Fig. S6). Thus, ERF is recurrently mutated in primary prostate cancer in AA men.

Although ERF was found to be a significantly mutated gene in the AAPC cohort, it did not reach statistical significance in the TCGA cohort (n = 333; Fig. 3B; ref. 11). However, taking into account deletions as well as mutations in ERF, the frequency of somatic alterations in ERF is comparable between the AAPC cohort and other primary prostate cancer cohorts (5% vs. 3%), suggesting that loss-of-function by mutation or deletion may be mechanisms to dysregulate ERF. Overall, combining publicly available exome or whole-genome sequencing datasets from primary prostate cancer revealed that ERF was altered in approximately 3% of primary prostate cancer cases either by mutation (0.76%; 5/661) or by homozygous deletion (2.2%; 15/661; Supplementary Figs. S7 and S8A; Supplementary Table S11; refs. 9–11). Therefore, ERF is recurrently mutated or deleted in primary prostate cancer.

Although ERF was not previously recognized as a recurrently mutated gene in primary prostate cancer, lethal CRPC cohorts showed missense or loss-of-function ERF mutations in approximately 3% of CRPC tumors (Supplementary Figs. S7 and S8B–S8C; refs. 22, 23). An updated analysis of genomic data from a cohort of CRPC cases shows that ERF is recurrently mutated at a frequency of approximately 3% (8/269) and undergoes copy-number loss (hemizygous or homozygous) at a frequency of approximately 17.5% (47/269; Supplementary Fig. S8C; ref. 22). We also examined prostate cancer cell line data from the Cancer Cell Line Encyclopedia (CCLE) and Catalogue of Somatic Mutations in Cancer (COSMIC) databases and found that ERF is mutated in one of six prostate cancer cell lines (DU-145; p.A132S; refs. 24, 25).

We next asked whether alterations in ERF might be associated with more aggressive disease. ERF copy-number loss was associated with a number of aggressive pathologic features, including higher Gleason grade (Supplementary Fig. S9; P = 0.0035), higher pathologic T stage (P = 0.00696), and residual tumor (P = 0.0435; www.firebrowse.org). Of the 5 patients with ERF mutations (5/492) in the TCGA dataset, 4 had Gleason 8 or higher tumors (3 with Gleason 9), and 2 of these patients experienced biochemical recurrences. Among the 8 patients with ERF mutations or deletions in the AAPC exome cohort, 3 experienced biochemical recurrences (vs. 16 of 73 ERF wild-type) and 4 of 8 were pT3 (vs. 31 of 94 ERF wild-type). Moreover, a germline analysis of ERF in the AAPC cohort revealed an ERF coding variant (S295I) in a patient with Gleason 9 prostate cancer that experienced a biochemical recurrence (Supplementary Table S12). Overall, these data raise the possibility that ERF mutations and deletions may be linked to more aggressive forms of prostate cancer.

ERF was first characterized as an ETS and RAS tumor-suppressor protein with a transcriptional repressor function (26, 27). In addition to prostate cancer, mutations in ERF occur in other tumor types at similar frequencies: stomach adenocarcinoma (∼4%), colorectal adenocarcinoma (∼4%), and Ewing sarcoma (∼3%), which is also notably driven by a common ETS rearrangement, EWS–FLI (28–30). We asked whether ERF might function as a tumor suppressor in prostate cancer cells. Using lentiviral shRNAs, we knocked down ERF in the PC3 prostate cancer cell line and demonstrated a significant increase in anchorage-independent growth of prostate cancer cells (Fig. 3C; Supplementary Fig. S10). ERF knockdown also increased invasion of PC3 cells and increased mouse tumor xenograft growth (Supplementary Fig. S11A–S11C). Overexpression of ERF reduced colony growth, whereas a mutant ERF harboring a mutation (Y89C) identified in the AAPC cohort diminished this effect (Supplementary Fig. S12A–S12C). Furthermore, ERF knockdown in another prostate cancer cell line (LNCaP) and immortalized prostate epithelial cell line (RWPE-1) showed increased growth proliferation but no significant increase in invasion (Supplementary Fig. S13A–S13H). Overexpression of wild-type ERF in the DU-145 cell line harboring a mutation in ERF had no effect on proliferation in a focus formation assay but led to a reduction in colony growth in a low-attachment assay and decreased invasion (Supplementary Fig. S14A–S14D). These results were consistent with a possible tumor-suppressor role for ERF in prostate cancer.

Given its known role as a repressor of the ETS transcription factor family, we hypothesized that loss of ERF might activate a transcriptional program that resembles the output of ETS transcriptional activators such as ERG or ETV1 (31, 32). To test this hypothesis, we performed lentiviral shRNA knockdown of ERF in an immortalized prostate epithelial cell line (LHS-AR) and two prostate cancer cell lines (VCaP and LNCaP) that each harbor oncogenic ERG (VCaP) or ETV1 (LNCaP) rearrangements (Supplementary Fig. S10; ref. 33). We then generated transcriptome data (RNA sequencing) to derive a gene signature of ERF knockdown from the top 100 genes upregulated, according to the difference of means, when ERF is knocked down in comparison with control (see Methods). The ERF knockdown (KD)_UP signature from the LHS-AR cells was correlated with ETV1 and ERG signatures across the CCLE and was correlated with ETV1′s target expression across the TCGA and CRPC tumor datasets (Fig. 4A; Supplementary Fig. S15). We then generated a combined ERF KD_UP signature from VCaP and LNCaP cell lines, hypothesizing that in the context of ETS activation that ERF loss might augment an ETS oncogenic transcriptional program. We found that this signature correlated with ERG pathway expression across TCGA and CRPC tumor datasets (Fig. 4B). Supporting the idea that ERF loss or dysfunction may function similarly to ERG activation, ERF deletions and mutations were mutually exclusive of ERG rearrangements in the AAPC exome cohort (Fig. 1). Furthermore, mutations and homozygous deletions of ERF were mutually exclusive of ERG fusion events in the published TCGA dataset (Supplementary Fig. S16; Fisher exact test: P < 0.05). In addition, overexpression of wild-type ERF in DU-145, an ERF mutant cell line, diminished the ERF KD signatures and ERG signature (Supplementary Fig. S17A).

We next asked whether the ERF KD signature could be associated with a more aggressive phenotype in prostate cancer. We tested whether the ERF KD gene signature was also correlated with features of aggressive prostate cancer in the TCGA dataset and found that higher Gleason scores correlated with the ERF KD gene signature (Supplementary Fig. S17B).

In an unbiased pathway analysis of cell lines from the CCLE, we found an androgen receptor signaling (AR) signature (NELSON_RESPONSE_TO_ANDROGEN_UP) as the top correlated signature with respect to the ERF KD signature_UP (Supplementary Fig. S18; ref. 34). We projected this ERF KD signature into the RNA-sequencing datasets generated from tumor samples in cohorts of primary prostate cancer (TCGA) and CRPC (11, 22). We found that AR signatures were correlated with the ERF KD signature_UP in the TCGA and CRPC datasets (Fig. 4C; ref. 35). These data suggest that AR signatures are correlated with the transcriptional program of ERF knockdown and that loss of ERF is associated with a transcriptional program that can mimic ETS activation and may impinge on androgen signaling. To test the hypothesis that ERF loss may promote androgen signaling, we used the CRISPR/Cas9 system to target the ERF coding sequence in prostate cancer cells (36). We showed that loss of ERF is associated with an increase in androgen-dependent growth (Fig. 4D).

We also investigated other genes that were recurrently mutated but did not reach statistical significance in the AAPC discovery cohort. YBX1 (K81T, R244*, N76S) was mutated 3 times and has been implicated as an oncogene in prostate cancer but had not been previously identified as recurrently mutated in any prostate cancer cohorts (37). We identified missense mutations in the steroid hydroxylase CYP11B1, as well as other cytochrome P450 family members that occurred in a total of approximately 14% of samples in the discovery cohort (Supplementary Table S13).

Among known cancer genes, we observed that PIK3CA, which is recurrently mutated at a frequency of approximately 3% (20/667) in primary prostate cancers, was not mutated in the AAPC discovery or extension cohorts (Fisher exact test,P = 0.0115; Fig. 1; refs. 10, 11, 30). Previous analyses also suggest that PTEN is less commonly deleted in cohorts of men of African ancestry (2, 38). Our data suggest that alteration of the PI3K signaling pathway through either PTEN deletion or PIK3CA mutation is a less common event in AAPC. Interestingly, we identified missense mutations in FOXA1 (F254V and H247L) in the AAPC cohort that occurred at the same residues only in men of African ancestry in the TCGA dataset (F254V and H247Y), raising the possibility of somatic mutations in prostate cancer that may be associated with ancestry (39).

Finally, to test the addition of tumors from AA men to a large cohort of primary prostate cancer largely from men of European ancestry, we performed a combined analysis of the AAPC discovery (n = 102) and TCGA (n = 457) cohorts which nominated several new significantly mutated genes, including SMARCA1 and ZFHX3 (Supplementary Fig. S19; ref. 17).

Prostate cancer sequencing studies have been comprised primarily of men of European ancestry, and there have been few large studies focused on men of African ancestry. Here, we use exome sequencing to identify genomic features of primary prostate cancer in AA men. We report the identification of recurrent mutations in ERF, a prostate cancer gene that had not been previously appreciated. Although analysis of this cohort identified ERF as a significantly mutated gene in prostate cancer, we note that ERF was mutated and located within a focal deletion in the TCGA cohort, composed primarily of men of European ancestry. Focal deletions at chr19q13.2 encompass ERF and CIC, a tumor suppressor that has been shown to regulate the ETS factors ETV1, ETV4, and ETV5 (40). Our exome sequencing study implicates ERF as a potential target of these deletions. Still, we cannot exclude the possibility that deletion of both ERF and CIC contributes to prostate cancer. Therefore, our data in conjunction with TCGA data suggest that ERF can be altered through mutation or deletion.

Dysregulation of ETS transcription factors plays a major role in prostate carcinogenesis, and several studies have reported a lower prevalence of TMPRSS2–ERG rearrangements in prostate tumors of AA men (2, 41). Here, we present evidence of a mechanism of affecting ETS transcriptional output through an ETS transcriptional repressor. Furthermore, alteration of ERF may also be associated with more aggressive prostate cancers.

We observed a number of recurrent SCNAs that differed between the primary AAPC and TCGA cohorts. The limitations of this comparison with TCGA include differences in technologies (exome sequencing vs. SNP arrays) and stringencies for copy-number detection. Still, our results, in conjunction with other studies that have examined specific copy-number alterations, suggest that overall the SCNA landscape may be distinct in primary prostate cancer in AA men. In addition, we find that alterations of the PI3K signaling pathway through deletion of PTEN or mutation of PIK3CA are uncommon in primary AAPC, suggesting that distinct patterns of genomic alterations may occur in this cohort with implications for precision medicine.

Our results suggest that increasing the ancestral diversity of study populations for cancer genomic characterization may help increase the discovery potential of these studies, which we believe to date have not included sufficiently large numbers of men of African ancestry. Given the relatively lower mutation rate of prostate cancer, larger cohorts of AA patients with prostate cancer may be required to identify recurrently mutated genes that may contribute to prostate carcinogenesis or to aggressive prostate cancer features in this population. These studies will inform whether alterations in these genes may be enriched in certain ancestral groups. Recent studies have implicated prostate tumor location, differential gene expression, and somatic genomic events such as LSAMP deletions in prostate cancers in AA men (2–4, 42–44). Our study suggests that there are still unexplained reasons for the aggressive nature of prostate cancer in AA men, which is only partially explained by the genomic studies to date. Additional studies focused on metastatic CRPC samples from AA men, and development of methodologies to integrate analyses of somatic and germline data may improve our understanding of the nature of aggressive prostate cancer in these patients (23, 45–47). Our results suggest that inclusion of sufficient numbers of patients of African ancestry in cancer genomic studies may enable the discovery of new cancer genes and inform the inclusion of diverse populations toward precision cancer medicine (48).

The discovery cohort comprised specimens from Weill Cornell Medicine (WCM) and Karmanos Cancer Center (KCC; Supplementary Table S2). The extension cohort comprised samples from KCC, Johns Hopkins University/Prostate Cancer Biorepository Network (PCBN), and Roswell Park Cancer Institute (RPCI; Supplementary Table S8). All cohort samples were from patients self-identified as African-American. Archival pathology specimens were obtained retrospectively from four different institutions under Institutional Review Board (IRB) protocols: WCM (IRB #1007011157), RPCI (IRB #BDR-035413), Johns Hopkins University (JHU; IRB # NA_00048544), and KCC (IRB # 044812MP4E). Hematoxylin and eosin–stained slides were reviewed by study pathologists at WCM (J.M. Mosquera, B.D. Robinson, F. Khani, and M.A. Rubin), RPCI (G. Azabdaftari), JHU (A.M. De Marzo), and KCC. Annotated slides containing tumor and benign tissue were used for somatic and germline DNA. Clinical and pathologic data are summarized in Supplementary Tables S2 and S8. Data on ERG rearrangement, PTEN deletion, SPOP mutation, and SPINK1 expression were available for the WCM cohort, as previously published (2).

To assess ERF deletion in tissues, we developed a dual-color FISH assay consisting of a locus-specific probe (W12-2967N22) plus reference probe spanning a stable region of the chromosome (RP11-46I12). All clones were tested on normal metaphase spreads of comparative genomic hybridization target slides as previously described (9, 21). ERF deletion was defined as the presence of 0 or 1 copy on average per nucleus compared with two reference signals. At least 100 nuclei were evaluated per tissue section using a fluorescence microscope (Olympus BX51; Olympus Optical).

This single-color chromogenic detection assay uses pairs of specially designed oligonucleotide probes that, through sequence-specific hybridization, recognize both the specific target ERF RNA sequence and the signal amplification system (see ERF oligonucleotide list, Supplementary Table S14; Affymetrix, Inc.). Based on unique coordinates on chromosome 19, unique target probe oligonucleotides were designed (see ERF map sequence, Supplementary Fig. S20). The latter is designed to hybridize in tandem to the target RNA. Cross-hybridization to other sequences is minimized by screening against the entire human RNA sequence database. The signal amplification system consists of the preamplifier, amplifier, and enzyme-conjugated label probe, which assemble into a tree-like complex through sequential hybridization. Signal amplification occurs at target sites bound by probe pairs only. Nonspecific off-target binding by single probes does not result in signal amplification. All steps of ERF RNAish staining of the slides were performed manually (49). Briefly, formalin-fixed, paraffin-embedded (FFPE) unstained tissue sections (5 μm) were mounted on positively charged microscopic glass slides, deparaffinized in xylene, and dehydrated through a series of alcohols. The dehydrated sections were then treated, and sequential hybridization of probe and amplifiers was performed according to Affymetrix protocols. The rehydrated sections were treated with 3% hydrogen peroxide at room temperature for 10 minutes to block endogenous peroxidase. Sections were then boiled in 1 × citric buffer (10 nmol/L Nacitrate, pH 6.0) for 15 minutes and incubated with protease (2.5 mg/mL; Sigma Aldrich) at 40°C for 30 minutes. The slides were hybridized sequentially with target probes (20 nmol/L) in hybridization buffer A [6 × saline sodium citrate (SSC) buffer (1 × SSC is 0.15 mol/L NaCl and 0.015 mol/L Na-citrate), 25% formamide, 0.2% lithium dodecyl sulfate (LDS), and blocking reagents] at 40°C for 2 hours, signal preamplifier in hybridization buffer B (20% formamide, 5 × SSC, 0.3% LDS, 10% dextran sulfate, and blocking reagents) at 40°C for 30 minutes, amplifier in hybridization buffer B at 40°C for 30 minutes, and horseradish peroxidase– or alkaline phosphatase–labeled probes in hybridization buffer C (5 × SSC, 0.3% LDS, and blocking reagents) at 40°C for 15 minutes. Hybridization signals were detected under brightfield microscope as red colorimetric staining followed by counterstaining with hematoxylin. Signals were granular and discrete red dots corresponding to individual RNA targets.

RNAscope SpotStudio Software from Definiens, Inc. was utilized for image analysis of ERF RNA in situ expression at a single-cell resolution.

For our whole-exome sequencing discovery cohort, we collected treatment-naïve radical prostatectomy specimens collected from two primary sites: New York City (WCM) and Detroit (Karmanos Cancer Institute/Wayne State University). All patients were self-reported AA. All tissue DNA was extracted from FFPE tissues. We sequenced 128 matched tumor–normal pairs, and after quality control for contamination and low tumor purity (Supplementary Table S15), we analyzed 102 matched pairs (see Supplementary Methods). The baits for exon capture targeted 98.2% of genes in the Consensus CDS (CCDS) database. A mean target coverage depth of 100× per sample was achieved, with 80% of targets covered at a depth of ≥20×.

We used Mutect and the Indelocator (http://www.broadinstitute.org/cancer/cga/indelocator) for calling single-nucleotide changes and insertion/deletions (Supplementary Methods). We used an FFPE filter to remove mutations likely from FFPE artifacts. We also used the GATK HaplotypeCaller to find germline variants within a specific gene.

The analysis of germline variants focused on variants identified among 20 genes that are associated with autosomal-dominant cancer-predisposition syndromes. These genes were chosen for their crucial role in the maintenance of DNA integrity. Pathogenicity of germline variants was determined according to the most recent guidelines published jointly by the American College of Medical Genetics and Genomics and the Association for Molecular Pathology. Germline variants were evaluated against published literature and publicly available databases such as ClinVar and variant-specific databases. In addition, population-based frequency databases including 1000 Genomes and the Exome Aggregation Consortium were examined. Only pathogenic and likely pathogenic variants with high or moderate penetrance were reported in this study. Low penetrance variants were excluded.

For validation, we used the Fluidigm Access Array microfluidic device. PCR products were barcoded, pooled, and subjected to Illumina sequencing on a MiSeq instrument.

LNCaP and PC3 cells were cultured in RPMI (Gibco) with 10% FBS. VCAP cells were cultured in DMEM with 10% FBS. Cells stably expressing shRNAS were generated by seeding cells at 3 × 10⁵ cells per well of a 6-well followed by transduction with lentivirus expressing given shRNAs. shRNAs in pLKO vectors were obtained from the Genetic Perturbation Platform at the Broad Institute: shERF.1—shRNA TRCN0000349615: CCTGGTGTCTTCCGAGTCTAT; shERF.2—shRNA TRCN0000273970: CCACACCCAAAGCGTCTACAA; shRNA TRCN0000013908: CCTGTCTCTGTGGGTTTCTAA; shRNA TRCN0000013911: GAGGTGACTGACATCAGTGAT. Selection was performed subsequently with RPMI media with 10% FBS containing puromycin (2 μg/mL). Cell lines were kindly provided by Dr. William C. Hahn during 2014 to 2016 and were authenticated annually by DDC Medical or by the Molecular Diagnostics Laboratory at the Dana-Farber Cancer Institute using short tandem repeat profiling. The LHS-AR cell line was provided by Dr. William C. Hahn (33).

Cell lysates were prepared by lysing cells in 1% NP-40 with protease inhibitors (Roche) and phosphatase inhibitors (Calbiochem). Lysates were fractionated by SDS-polyacrylamide gel electrophoresis and transferred to nitrocellulose membranes using the iBlot system (Life Technologies). Immunoblotting was performed using LI-COR reagents (Odyssey Blocking Buffer and IRDye 800CW and IRDye 680RD secondary antibodies) according to the manufacturer's instructions (LI-COR Biosciences). Fluorescence detection was performed using an Odyssey CLx Infrared Imaging System, and quantitation was performed using Image Studio software (LI-COR). Anti-ERF antibody was purchased from Abcam (ab61108). Antibody for vinculin was purchased from Sigma (#V9131).

Total RNA was isolated using RNAeasy (Qiagen). cDNA was prepared using approximately 1 μg of RNA and the Superscript III Kit (Life Technologies). ERF transcript levels were quantified using SYBR green (Applied Biosystems) and measured using Quantstudio 6. The following primers were used for ERF:

• ERF_1_FP  5′ GCA AGC CCC AGA TGA ATT ACG 3′

• ERF_1_RP  5′ CCC CTT GGT CTT GTG CAG AA 3′

Total RNA was isolated using RNAeasy (Qiagen), then processed with the NEBNext PolyA mRNA Magnetic Isolation Module (NEB, E7490), and then further processed with the NEBNext Ultra Directional RNA Library Prep Kit (NEB, E7420S). Libraries for RNA sequencing were then sequenced on a NextSeq (Illumina).

To generate the ERF signatures, we analyzed the RPKM RNA-sequencing profiles of ERF shRNA knockouts in LNCaP and VCaP cell lines (see ERF shRNA experiments above). We independently ranked genes according to the difference of means between the shERF-infected LNCaP and control samples. The same procedure was performed for the VCaP cell line. To obtain a consensus shERF signature, we computed the overlap between the top 400 differentially expressed (upregulated and downregulated) genes in LNCaP and VCaP cell lines. This overlap resulted in 65 upregulated genes and 61 downregulated genes. These two gene sets are provided as part of the Supplementary Materials and will also be made publicly available as part of the C6 subcollection of the Molecular Signatures Database (MSigDB) in a future release (Supplementary Table S16; ref. 50).

PC3 cells were transduced with lentivirus generated with pLKO-puro-shRNA plasmids. Twenty-four hours after infection, cells were selected with 2 μg/mL puromycin. After 48 hours of selection, cells were split for qRT-PCR assays and for passaging in normal serum-containing media for soft-agar assays. A total of 1 × 10⁴ cells were suspended in 1 mL of 0.33% select agar in RPMI/FBS and plated on a bottom layer of 0.5% select agar in 6-well plates. Each cell line was analyzed in triplicate. Colonies were photographed after 11 days and quantified using CellProfiler.

Single-guide RNA guide sequences targeting the ERF coding region were cloned into the plentiCRISPR vector (pxpr001; ref. 36). sg4 ERF: CAC CGG GGT ACA TCG GGC TCA GCG T sg5 ERF: CAC CGG ATC CCC GCG CCC GAC CAC C; control guide sg5 GFP: GAAGTTCGAGGGCGACACCC. Lentivirus was produced in 293T cells and then LNCaP cells were lentivirally transduced followed by puromycin (2 μg/mL) selection. Western blot analysis was used to confirm ERF knockdown. LNCaP cells were seeded at a density of 1 × 10⁴ cells in triplicate in 12-well plates in RPMI media supplemented with 10% charcoal-stripped serum (CSS; Gibco #12676). The following day cells were treated with R1881 synthetic androgen in RPMI/CSS media, and media were changed every 3 to 4 days. After 14 days, cells were fixed with 4% formaldehyde and stained with 0.5% crystal violet solution. Cells were photographed using a Leica microscope and imaging software. Quantification of crystal violet uptake for each sample was performed by destaining cells with 10% acetic acid and measurement of absorbance at 595 nm using a SpectraMax 190 instrument.

All animal experiments were approved by the Dana-Farber Cancer Institute Institutional Animal Care and Use Committee and were performed in accordance with institutional and national guidelines.

The accession number for sequencing files is dbGAP: phs000945.

L.A. Garraway reports receiving commercial research grants from Astellas, BMS, Merck, and Novartis, has ownership interest (including patents) in Foundation Medicine, and is a consultant/advisory board member for Boehringer Ingelheim, Foundation Medicine, Novartis, Third Rock, and Warp Drive. No potential conflicts of interest were disclosed by the other authors.

Conception and design: F.W. Huang, J.M. Mosquera, J.R. Osborne, E.M. Van Allen, M.A. Rubin, L.A. Garraway

Development of methodology: F.W. Huang, J.M. Mosquera, J. Chimene-Weiss, J.R. Osborne, J.W. Kim, M.A. Rubin, L.A. Garraway

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F.W. Huang, J.M. Mosquera, M. Baco, B. Rabasha, S. Bahl, B.D. Robinson, F. Khani, J. Chimene-Weiss, G. Azabdaftari, A. Woloszynska-Read, A.M. De Marzo, S. Gabriel, M.A. Rubin, I.J. Powell, L.A. Garraway

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F.W. Huang, J.M. Mosquera, A. Garofalo, C. Oh, M. Baco, A. Amin-Mansour, S.A. Mullane, S. Aldubayan, E. Kim, M. Hofree, A. Romanel, F. Demichelis, E.M. Van Allen, J. Mesirov, P. Tamayo, M.A. Rubin, L.A. Garraway

Writing, review, and/or revision of the manuscript: F.W. Huang, J.M. Mosquera, A. Garofalo, M. Baco, A. Amin-Mansour, S.A. Mullane, B.D. Robinson, E. Kim, J.R. Osborne, A. Woloszynska-Read, E.M. Van Allen, J. Mesirov, M.A. Rubin, I.J. Powell, L.A. Garraway

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F.W. Huang, S. Bahl, B.D. Robinson, B. Karir, K. Sfanos, I.J. Powell, L.A. Garraway

Study supervision: F.W. Huang, J.M. Mosquera, S. Gabriel, M.A. Rubin, L.A. Garraway

We thank David Takeda, Rick Wilson, Ginevra Botta, Cory Johannessen, Paz Polak, and Manaswi Gupta for helpful discussions. We thank the Broad Genomics Platform. We thank Susan Bolton from Wayne State University; Himisha Beltran from WCM Division of Hematology and Medical Oncology and Englander Institute for Precision Medicine; Andrea Sboner from WCM Pathology and Laboratory Medicine, Institute for Computational Biomedicine, and Englander Institute for Precision Medicine; and Kyung Park, Peyman Tavassoli, Zohal Noorzad, and Jaclyn Croyle from WCM Pathology and Laboratory Medicine. The authors also thank Elena Pop and Elizabeth Brese from Roswell Park, and Helen Fedor from Johns Hopkins University.

This work was supported by the NCI U01 CA162148 (L.A. Garraway), the Department of Defense Prostate Cancer Research Program Physician Research Training Award W81XWH-14-1-0514 (F.W. Huang), Prostate Cancer Foundation Young Investigator Award (F.W. Huang), ASCO Young Investigator Award (F.W. Huang), the US NIH R01 CA116337 (M.A. Rubin), 5U01 CA111275-09 (J.M. Mosquera and M.A. Rubin), U54 HG003067 (S.Gabriel), R01CA154480 (J. Mesirov and P. Tamayo), R01CA121941 (J. Mesirov and P. Tamayo), U01CA176058 (J. Mesirov and P. Tamayo), R01CA109467 (J. Mesirov and P. Tamayo), U54 CA137788 (J.R. Osborne), the Starr Cancer Consortium (M.A. Rubin), and by a Stand Up To Cancer–Prostate Cancer Foundation Prostate Dream Team Translational Research Grant (Grant Number: SU2C-AACR-DT0712; M.A. Rubin and L.A. Garraway). Stand Up To Cancer is a program of the Entertainment Industry Foundation. Research grants are administered by the American Association for Cancer Research, the Scientific Partner of SU2C. This work was also supported in part by the Translational Research Program at WCM Pathology and Laboratory Medicine. This work was in part supported by the Department of Defense Prostate Cancer Research Program, DOD Award No. W81XWH-10-2-0056 and W81XWH-10-2-0046 PCRP Prostate Cancer Biorepository Network (PCBN). This work was supported by NCI grant P30CA016056 and the Pathology Network and Clinical Data Network Shared Resources at Roswell Park Cancer Institute.

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.