Prostate cancer is the second leading cause of male cancer deaths. Castration-resistant prostate cancer (CRPC) is a lethal stage of the disease that emerges when endocrine therapies are no longer effective at suppressing activity of the androgen receptor (AR) transcription factor. The purpose of this study was to identify genomic mechanisms that contribute to the development and progression of CRPC.
We used whole-genome and targeted DNA-sequencing approaches to identify mechanisms underlying CRPC in an aggregate cohort of 272 prostate cancer patients. We analyzed structural rearrangements at the genome-wide level and carried out a detailed structural rearrangement analysis of the AR locus. We used genome engineering to perform experimental modeling of AR gene rearrangements and long-read RNA sequencing to analyze effects on expression of AR and truncated AR variants (AR-V).
AR was among the most frequently rearranged genes in CRPC tumors. AR gene rearrangements promoted expression of diverse AR-V species. AR gene rearrangements occurring in the context of AR amplification correlated with AR overexpression. Cell lines with experimentally derived AR gene rearrangements displayed high expression of tumor-specific AR-Vs and were resistant to endocrine therapies, including the AR antagonist enzalutamide.
AR gene rearrangements are an important mechanism of resistance to endocrine therapies in CRPC.
Castration-resistant prostate cancer (CRPC) is the lethal manifestation of the disease that occurs when tumor cells have developed resistance to therapies that inhibit transcriptional activity of the androgen receptor (AR). This study establishes AR as one of the most frequently rearranged genes in CRPC metastases. We find two main patterns of AR rearrangement: those concurrent with AR amplification and those occurring as the only detectable AR alteration. We demonstrate that the latter can cause enzalutamide resistance by promoting expression of constitutively active AR variants (AR-V). AR-Vs resulting from AR rearrangements resemble AR-V7, a biomarker of endocrine therapy resistance in CRPC. However, these AR-Vs are molecularly distinct from AR-V7 and undetectable by AR-V7 assays. Therefore, AR rearrangements could capture a broad spectrum of patients who are unlikely to receive benefit from endocrine therapies that target AR.
Androgen receptor (AR) is a ligand-activated transcription factor that regulates expression of genes critical for homeostasis of the normal prostate, as well as growth and progression of prostate cancer. Accordingly, AR is the key target for systemic therapy of prostate cancer patients (1, 2). Therapeutic inhibition of AR is achieved by blocking production of the androgen ligands that bind and activate AR (testosterone and dihydrotestosterone) or by competitive antagonists that bind and repress AR. Although these therapies extend patient survival, long-term efficacy is limited by inevitable evolution and progression to an aggressive disease manifestation termed castration-resistant prostate cancer (CRPC; ref. 3).
Evolutionary changes in CRPC cells include AR gene amplification that provides transcriptional hypersensitivity of AR to castrate levels of androgens, mutations in the AR gene that broaden the spectrum of activating ligands for AR, and splicing alterations that promote expression of AR variant 7 (AR-V7; ref. 4). AR-V7 can function as a constitutively active, ligand-independent transcription factor (5, 6), but preferentially heterodimerizes with full-length AR to transcriptionally activate canonical AR targets and also transcriptionally repress tumor suppressor genes (7–9). Collectively, CRPC cells with these AR alterations have been shown to remain dependent on full-length AR. This knowledge has driven the clinical development and regulatory approvals of many second-generation drugs that provide a more complete blockade of androgen production or bind the AR ligand binding domain with higher affinity (10–13). Drugs in this second-generation category include the androgen synthesis inhibitor abiraterone and the AR antagonist enzalutamide.
Despite the widespread clinical use of abiraterone and enzalutamide, CRPC remains a uniformly fatal disease. Clinical data such as rising PSA (an AR-regulated gene) are often used as a surrogate for resistance to these AR-targeted therapies, supporting the concept that CRPC remains AR dependent in the majority of patients (14). To understand additional mechanisms that promote AR activity in the context of clinical CRPC, we analyzed whole-genome and targeted DNA-sequencing (DNA-seq) data from an aggregate cohort of 272 prostate cancer patients with variable exposure to abiraterone and enzalutamide. By assessing genomic structural rearrangements that would disrupt the linear architecture of coding genes, we identified AR as one of the most frequently rearranged genes in CRPC patients' tumor genomes. We investigated the mechanisms by which these diverse AR gene rearrangement events can promote CRPC progression.
Materials and Methods
Prostate cancer tissues
A cohort of 41 patients with metastatic CRPC had tumor tissue collected from metastatic lesions by biopsy or surgery at The Institute of Cancer Research and Royal Marsden Hospital (ICR/RMH) between 2009 and 2015 (26 patients), or warm autopsy under the aegis of the Prostate Cancer Donor Program at the University of Washington as described (15 patients; ref. 15). All patients treated at the ICR/RMH had provided written informed consent and were enrolled in institutional protocols approved by the Royal Marsden NHS Foundation Trust Hospital (London, United Kingdom) ethics review committee (reference 04/Q0801/60). For the 26 patients treated at ICR/RMH, a single tissue sample was obtained from 22 CRPC patients, 2 tissue samples from 2 independent metastatic sites were obtained from 3 CRPC patients, and 3 tissue samples from 3 independent metastatic sites were obtained from 1 CRPC patient. For 5 of these 41 CRPC patients, archival prostatectomy tissue was available from their prior surgeries at ICR/RMH. For the 15 rapid autopsy subjects, 2 tissue samples were collected from 2 independent metastatic sites per patient as described (15). A cohort of 101 patients with CPRC that had a metastatic biopsy collected through the Stand Up 2 Cancer/Prostate Cancer Foundation-funded West Coast Prostate Cancer Dream Team project has been described (16). A cohort of 130 patients who had localized prostate cancer tissue collected by surgery has been described (17). Tissue details including enzalutamide/abiraterone treatment status of CRPC patients prior to or during tissue collection are described in Supplementary Data 1. All studies were conducted in accordance with the Declaration of Helsinki.
Whole-genome DNA-seq data from 101 metastatic biopsies collected from the 101 patient CRPC cohort were obtained from ref. 16. Whole-genome DNA sequencing data from 130 prostatectomy specimens were obtained from the International Cancer Genome Consortium (ICGC) study PRAD-CA (17). AR-targeted DNA sequencing data from 30 metastatic tissues collected by autopsy of 15 CRPC patients were obtained from dbGaP phs001223.v1.p1 (15). For the 41 tissue specimens obtained from 26 patients at ICR/RMH, DNA was isolated and submitted to the University of Minnesota Genomics Center for DNA-seq library preparation and hybrid capture with a custom SureSelect (Agilent) bait library as described (15). Post-capture sequencing libraries were pooled and sequenced with an Illumina HiSeq 2500 using 2 × 150 bp settings and rapid run chemistry.
DNA-seq data analysis
DNA rearrangements (structural variants), somatic mutations, and copy-number variants were called from whole-genome sequencing data in the 101 patient CRPC cohort and 130 patient prostatectomy cohort as described (16, 17). DNA somatic mutations and copy-number variants were called from AR-targeted sequencing data in the 41 patient CRPC cohort as described (15). AR rearrangements (structural variants) were considered true positives based on identification by two independent structural variant callers (LUMPY and Delly), quantification of variant allele fraction using SHEAR, and containing at least 10 supporting split reads and 10 supporting paired-end reads as described (15). BAM files of mapped reads were visualized in Integrative Genomics Viewer (IGV; ref. 18).
PCR primers were designed to specific genomic regions flanking breakpoints of AR gene rearrangements. PCR primers are listed in Supplementary Table S1. Genomic DNA from CRPC tumors was subjected to whole-genome amplification (WGA) using the REPLI-g Amplification kit (Qiagen) and purified using a QIAquick Nucleotide Removal Kit (Qiagen). WGA DNA or DNA isolated directly from cell lines was used for PCR with specific primer pairs using AccuStart II PCR SuperMix (Quanta Biosciences) as per the manufacturer's recommendations.
LNCaP cells were obtained from ATCC (#CRL-1740). The R1-AD1 cell line (CWR-R1, androgen-dependent 1, referred to as “deletion-negative clone 1” in the original publication; ref. 19) is a subline derived from single-cell cloning of the CWR-R1 cell line. R1-AD1 cells are androgen responsive and contain a structurally normal copy of the AR gene (19, 20). Cells were maintained in RPMI-1640 medium supplemented with 10% fetal bovine serum (FBS) in a 5% CO2 incubator at 37°C. Cell line authentication and mycoplasma monitoring are described in the Supplementary Methods section.
The lentiGuide-Puro plasmid was a gift from Feng Zhang (Addgene plasmid # 52963). The pLenti-Cas9-GFP plasmid was a gift from David Sabatini (Addgene plasmid # 86145). Pairs of synthetic oligonucleotides (Supplementary Table S2) encoding gRNA sequences were annealed, phosphorylated, and ligated with lentiGuide-Puro vector that had been digested with BsmBI to remove a ∼2 kb filler.
R1-AD1 and LNCaP cells were cotransfected with 3 μg pLenti-Cas9-GFP plasmid and 1.5 μg each of two separate gRNA-containing lentiGuide-Puro plasmids by electroporation using an ECM 830 Square Wave Electroporator (BTX). DNA mixtures were incubated with approximately 4 × 106 cells in 400μL of tissue culture medium in a 4 mm gap-width electroporation cuvette (BTX) for 5 minutes at room temperature, and subjected to 250 V, 30 ms (R1-AD1) or 200 V, 2 × 10 ms (LNCaP) electrical pulses prior to a 15-minute recovery at room temperature and plating.
Clonal enrichment assays with CRISPR/Cas9-transfected cells
Electroporated R1-AD1 and LNCaP cells were treated with 6 μg/mL puromycin for 4 days, and allowed to recover in puromycin-free medium for 2 days. Cells were trypsinized, and an aliquot was stored at −20°C to represent a baseline (day 0) sample. Trypsinized cells were reseeded on 6-well plates (“2D conditions”) at 500,000 cells/well in regular growth medium containing 10 μmol/L enzalutamide or 0.1% v/v DMSO as vehicle control, or medium containing charcoal-stripped (steroid-depleted) serum. Alternatively, cells were seeded for soft-agar growth assays (“3D conditions”) by mixing 10,000 cells with 1 mL of 0.35% agar in phenol red free growth medium and plating on 6-well plates that had been prefilled with 1 mL of 0.7% agar in growth medium containing 10 μmol/L enzalutamide or 0.1% v/v DMSO as vehicle control, or medium containing charcoal-stripped (steroid-depleted) serum. Cells grown in 2D on plates were re-fed 1 to 2 times per week and harvested after 7 or 14 days of culture. Cells grown in 3D in soft agar were re-fed weekly and harvested after 28 days of culture. Genomic DNA was isolated from plated cells using a NucleoSpin Tissue kit (Takara). Cells grown in 3D in soft agar were collected by heating the agar with colonies at 90°C for 1 minute, pelleting cells 11,000 × g for 1 minute, removing the agar while it was still in liquid form, and washing the pellet 3 times with 90°C PBS followed by pelleting at 11,000 × g to remove any residual agar. Genomic DNA was isolated from these pelleted colonies using a NucleoSpin Tissue kit.
Quantitative PCR analysis of CRISPR/Cas9-transfected cells
Quantitative PCR was performed using PerfeCTa SYBR Green fastMix (Quanta Biosciences), with 80 ng gDNA used as template and the same primers used for endpoint PCR detection of tumor-specific AR gene rearrangements (Supplementary Table S1). PCR reactions were analyzed using a CFX Connect Real-Time System (BioRad). Fold change in mRNA expression levels was calculated by the comparative Ct method, using the formula 2−(ΔΔCt) where AR exon 1 PCR was used as calibrator using primers (forward: 5′-TGGATGGATAGCTACTCCGG; reverse-5′- TTTACCCTGCTGAGCTCTCC). All experiments were performed as 3 separate biological replicates, each performed in technical duplicate (n = 6). Two-tailed unpaired t tests were used to calculate P values.
Preparation of cell lysates and Western blot analysis
At experimental endpoints, cells were harvested in 1× Laemmli buffer. Alternatively, insoluble nuclear fractions were prepared from cells as described (21) and boiled in 1× Laemmli buffer. Lysates in Laemmli buffer were subjected to Western blot as described (22) using primary antibodies (AR SP107, Sigma; AR-441, Santa Cruz; Actin C4, Santa Cruz; Histone H3 ab32356, abcam) diluted 1:1,000 and secondary antibodies diluted 1:10,000.
Total RNA (50 ng) was used for AR 3′RACE reactions using a 5′/3′RACE kit, second generation (Roche) as described (23). The final AR 3′RACE products were purified using a QIAquick PCR Purification Kit (Qiagen).
PacBio isoform sequencing
AR 3′RACE products were submitted to the University of Minnesota Genomics Center and converted to barcoded SMRTbell libraries using the PacBio Barcoded Adapters for Multiplex Single Molecule Real-Time (SMRT) Sequencing protocol as per the manufacturer's recommendations (Pacific Biosciences). Barcoded SMRTbell libraries were pooled and prepared for diffusion loading on a Pacific Biosciences Sequel and sequenced using Sequel 2.5 chemistry (R1-X-11 cells) or Sequel 3.0 chemistry (clinical specimens).
Analysis of AR 3′RACE/PacBio sequencing data
Circular consensus reads requiring a minimum of three full passes were generated from raw unaligned BAM files, then demultiplexed using Pacific Bioscience's lima software (v.1.9; https://github.com/PacificBiosciences/barcoding) and adapters that included the barcode plus the 5′ and 3′ RACE primers as input. IsoSeq 3.1 was used to generate high-quality reads for alignment to hg19 with minimap2 (24). Pacific Bioscience's Cupcake ToFU supporting scripts (https://github.com/Magdoll/cDNA_Cupcake/wiki/Cupcake-ToFU) were used to collapse reads into transcripts, identify fusion transcripts, and quantify the abundance of both. Non-fusion transcripts were summarized across samples using TAMA (https://github.com/GenomeRIK/tama/wiki). Fusion transcripts were summarized across samples using a Cupcake ToFU script. The exon structure of resulting summarized transcripts, and heat maps and plots were generated in R version 3.5.0 using custom scripts.
Generation of the R1-X-11 cell line
R1-AD1 cells were transfected with 3 μg pLenti-Cas9-GFP and 1.5 μg of each gRNA and then selected with 6 μg/mL puromycin for 4 days. Cells were plated by limiting dilution in 96-well plates in RPMI-1640 medium supplemented with 10% charcoal-stripped (steroid-depleted) serum. Wells containing single colonies were visualized by microscopy after 2 to 3 weeks and transferred to 48-well plates by trypsinization. Genomic DNA from individual clones was screened by PCR primers targeting the break junction to detect the translocation (chrX forward: 5′-CCTCTGATGCTTGGTTTTCC; chr11 reverse: 5′-GCCCAAACTTACCCAAGCTA) and PCR primers flanking the chromosome X gRNA target site in AR intron 3 (chrX forward: 5′-TGTAACAGCACCAACAGGCA-3; chrX reverse: 5′-GTACCCTTGGAAAGTGCGGA-3). The cell line R1-X-11 was identified based on a positive PCR signal for the translocation event, and negative PCR signal for retention of the wild-type intron 3 segment.
Total RNA was extracted using the ReliaPrep RNA Cell Miniprep system (Promega), and 1 μg total RNA was used for cDNA synthesis using cDNA qScript SuperMix (Quanta Biosciences). cDNA (1 μL) was used for PCR detection of a splice junction between AR exon 3 and fusion exon 89 in chromosome 11 using primers AR exon 3 forward and chr11 exon f89 reverse (Supplementary Table S1). PCR products were analyzed by agarose gel electrophoresis and Sanger sequencing. For quantitative PCR, gene expression of FASN, FKBP5, and ABCC4 was measured using primers listed in Supplementary Table S1 and GAPDH as calibrator using the relative quantitation calculation 2−ΔΔCt.
Cells were transfected with 40 pmol/L of siRNA targeting the fusion exon 89 in chromosome 11 using electroporation as described (20). siRNA-1 had sense sequence: 5′-AAACAGAGCUCUAUCAACAUU and antisense sequence 5′-UGUUGAUAGAGCUCUGUUUUU. siRNA-2 had sense sequence: 5′-UUACCUAUCUGGAGGGUCAUU and antisense sequence: 5′-UGACCCUCCAGAUAGGUAAUU.
Cells were seeded at initial densities of 1 × 104 cells/well in 48 well plates. Seeded cells were cultured in RPMI-1640 medium containing 10% FBS supplemented with 0.1% v/v DMSO (vehicle control) or 30 μmol/L enzalutamide. Crystal violet assays were performed 24 hours after plating to represent day 0. Additional wells were stained after 2, 4, or 6 days of growth.
The AR gene is frequently disrupted by rearrangements in metastatic CRPC
Structural rearrangements that accumulate in intragenic regions of cancer cells have the potential to alter or interrupt the normal linear organization of exons that encode individual genes (Supplementary Fig. S1A). To identify genes that are frequently disrupted by these types of genomic rearrangements in advanced prostate cancer, we analyzed whole-genome DNA-seq data from 101 prospectively collected metastatic tumor biopsies (Supplementary Data 1; ref. 16). For inclusion, a gene was required to have at least 1 genomic rearrangement breakpoint located within its gene body. We found that 3,377 genes were recurrently affected by a genomic rearrangement breakpoint in at least 2 CRPC samples (Supplementary Fig. S1B) and 27 genes were recurrently affected in at least 15 CRPC samples (Supplementary Fig. S1C). Inspection of these frequently rearranged genes revealed that many were large (>1 Mb) and therefore might have a high chance of accumulating passenger rearrangements from genomic instability (Supplementary Fig. S1C). We corrected for this by normalizing gene rearrangement frequency by gene size (Supplementary Fig. S1D and Supplementary Data 2). This normalization highlighted genes known to be affected at high frequency by structural rearrangements in prostate cancer, including oncogenic TMPRSS2:ERG fusions (25), and tumor suppressors PTEN and TP53 (16).
Normalized data also highlighted a high frequency of structural rearrangements in the AR gene (23/101 tumors) (Supplementary Fig. S1D). AR gene rearrangements were observed in a previous targeted DNA-seq study of rapid autopsy subjects that died of CRPC (15). However, AR gene rearrangements have not been described in unbiased DNA-seq studies of CRPC genomes because these prior studies utilized whole-exome sequencing (26–28), which is blind to structural rearrangements with breakpoints in intronic or intergenic regions. Therefore, the relevance of AR gene rearrangements to clinical CRPC is not well established. To investigate AR gene rearrangements, we first evaluated their presence in the context of other known AR gene alterations including somatic point mutations as well as amplification of the AR gene body and/or amplification of an enhancer located upstream of AR (Fig. 1A). In a cohort of 130 localized prostate cancers analyzed by whole-genome DNA-seq (17), no AR alterations were detected (Fig. 1B). Conversely, the AR gene displayed at least one alteration in 88 of 101 metastatic tumors (Fig. 1B; Supplementary Data 3). AR gene rearrangements were concurrent with AR gene and upstream enhancer amplification in 20 of 101 metastatic tumors, but also occurred as the only detectable alteration in 3 of 101 metastatic tumors.
AR genomic structural rearrangements are clonally heterogeneous and occur in patients after exposure to first- and second-generation AR-targeted therapies
Mapping AR gene rearrangement breakpoints revealed a heterogeneous landscape of deletion, inversion, tandem duplication, and translocation events (Fig. 1C; Supplementary Table S3). In the 3 tumors where an AR gene rearrangement was the only AR alteration detected, the deletion and duplication events were evident from visual inspection of read coverage plots (Fig. 1D). Tumors that were positive for AR gene rearrangements had a higher burden of genomic inversions as well as cumulative burden of all genomic structural variations (deletions, inversions, and tandem duplications) compared with tumors that were negative for AR gene rearrangements (Supplementary Fig. S2). AR gene rearrangements were independent of alterations in other prostate cancer driver genes, with the exception of PTEN where positive enrichment was observed (odds ratio = 3.63; Supplementary Fig. S3). We also observed a slightly higher frequency of AR gene rearrangements in tumors from patients who had been treated with one or both of the potent second-generation AR signaling inhibitors abiraterone or enzalutamide compared with patients who were naïve to these agents and had only been treated with first-generation endocrine therapies (Fig. 2A and B; Supplementary Data 1).
A slightly higher frequency of AR gene rearrangements in abiraterone and/or enzalutamide-treated patients was also observed in a separate cohort of metastatic tumors from 41 patients where we analyzed the AR gene by deep targeted DNA-seq (hereafter referred to as the AR-targeted cohort; Fig. 2C and D; Supplementary Data 1 and 4; Supplementary Tables S4 and S5). Patients with AR gene rearrangements were observed at a higher frequency overall in this AR-targeted cohort (16/41, 39% of patients) compared with the whole-genome DNA-seq cohort (23/101, 23% of patients). This higher frequency of AR rearrangement–positive patients could be due to higher detection sensitivity from the greater sequencing depth provided by AR-targeted sequencing, or higher probability of sampling an AR rearrangement-positive lesion since 19 patients in the 41-patient cohort had more than one metastatic site analyzed (Supplementary Data 1). The AR gene rearrangement events in the AR-targeted cohort were highly heterogeneous, similar to the 101-patient whole-genome DNA-seq cohort (Fig. 2E). All AR gene structural rearrangements discovered in the AR-targeted cohort were validated orthogonally using PCR and Sanger sequencing (Supplementary Fig. S4; ref. 15).
For 5 of the AR gene rearrangement–positive patients in the AR-targeted cohort, archival prostatectomy tissue was available to enable longitudinal assessments of AR gene rearrangements using PCR (Fig. 2F). Consistent with the general observation that AR gene rearrangements are specific to CRPC-stage tumors (Fig. 2B), there was no evidence for preexistence of cell clones harboring these specific AR gene rearrangements in the patients' primary tumors (Fig. 2G–K).
AR genomic rearrangements provide a clonal growth advantage to cells under conditions of AR-targeted therapy
The heterogeneous landscape of rearrangement type and intra-AR breakpoint location across the patients who were positive for AR gene rearrangements in these 2 CRPC cohorts argues against a deterministic model wherein AR-targeted endocrine therapies induce specific types of recurrent AR gene rearrangements at susceptible genomic sites. Further, previous studies showing that AR-targeted endocrine therapies inhibit DNA repair by nonhomologous end joining (NHEJ; ref. 29–31) is incongruent with a deterministic model because signatures of NHEJ repair occurred at every AR gene rearrangement breakpoint (Supplementary Fig. S4). Rather, the origin of AR gene rearrangements is more consistent with a clonal evolution model wherein AR-targeted therapies provide selective pressure for outgrowth of rare subclones that harbor rearranged AR gene architectures.
To test this clonal evolution model, we used the CRISPR/Cas9 system to induce rearrangements in the AR gene in prostate cancer cell lines by targeting two DNA double-strand breaks to genomic locations that corresponded to the breakpoints observed for AR gene rearrangements in patients C9_a, V5300, and V5301 (Fig. 3A and B). We used R1-AD1 (32) and LNCaP prostate cancer cell lines for these experiments because they display androgen-sensitive growth and harbor one copy (R1-AD1) or two copies (LNCaP) of the AR gene on the X chromosome. We performed targeted PCR with genomic DNA to identify and track the growth of CRISPR/Cas9-engineered clones over time under different hormonal conditions that model AR-targeted therapies. Guide RNAs (gRNA) that targeted CRISPR/Cas9-mediated DNA double-strand breaks to AR intron 3 and chromosome 11 (chr11) induced translocation events in a subset of cells, which reflected the translocation observed in a metastasis from patient C9_a (Fig. 3C). When transfected cells were grown in two-dimensional culture for 7 or 14 days, or three-dimensional culture for 28 days, cells harboring the AR-chr11 translocation displayed clonal enrichment relative to the overall cell population as measured by quantitative PCR, only under conditions that modeled AR-targeted endocrine therapy (castration or enzalutamide; Fig. 3D–F). When we evaluated clonal evolution of cells that were engineered to harbor the two separate deletions observed in patient V5300 (Fig. 3G–J; Supplementary Fig. S5) or the two separate deletions observed in patient V5301 (Fig. 3K–N; Supplementary Fig. S6), we observed similar relative growth advantages of cells harboring targeted AR rearrangements under conditions of AR-targeted therapy.
Tumors harboring AR gene rearrangements without other AR alterations express abundant AR variant transcripts that are molecularly distinct from AR-V7
To understand the mechanisms by which AR gene rearrangements could promote a clonal growth advantage under conditions that modeled AR-targeted therapies, we examined the effects of these AR gene rearrangements on AR mRNA expression. For this, we performed 3′ rapid amplification of cDNA ends (RACE) with RNA isolated from these tumors using a forward primer anchored in AR exon 1. The 3′RACE reactions were subjected to long-read SMRT isoform sequencing (Iso-Seq) to simultaneously identify and quantify complete, intact AR mRNA transcripts that were expressed in these tumors. In a metastasis from patient C9_a, full-length AR mRNA was a major species (Fig. 4A and B; Supplementary Table S6). We also detected several mRNA fragments that likely originated from mis-priming of the oligo(dT) 3′RACE reverse primer at A-rich sequences in exons 7 and 2 of AR mRNA. Interestingly, we also detected expression of a minor AR variant (AR-V) mRNA species, which consisted of contiguously spliced AR exons 1 to 3 with a 3′ terminal exon derived from chr11. Presumably, this fusion transcript originated from the AR-chr11 translocation in this metastasis.
In a previous study, we found that metastatic tumors in patient C6_a expressed high mRNA and protein levels of ARv567es (also referred to as AR-V12), which arises from splicing of AR exons 1–4 and exon 8 (15). ARv567es/AR-V12 is a constitutively active form of the AR that lacks the C-terminal ligand binding domain and can drive CRPC phenotypes in experimental models (32, 33). In patient V5300, AR 3′RACE/SMRT-Iso-Seq identified full-length AR mRNA as the major isoform, as well as high expression of ARv567e/V12 (Fig. 4C and D). In patient V5301, ARv567es/AR-V12 was the major isoform (Fig. 4E and F). To ask whether these patterns of altered AR expression were also reflected in CRISPR/Cas9 clonal evolution models, we performed Western blot analysis of lysates from cells that had been transfected with Cas9 and gRNAs designed to model one of the deletions in tumor V5300 and for which enzalutamide selection drove high clonal enrichment of deletion-positive cells (Supplementary Fig. S5D). In this model, 1 and 2 weeks of culture under the condition of enzalutamide selection promoted high expression of AR-Vs, predominantly an AR-V species with a molecular weight that was consistent with ARv567es/AR-V12 (Fig. 4G).
Patient V4002 represented another patient in the 41-patient AR-targeted cohort that harbored an AR gene rearrangement as the only detectable AR alteration (Fig. 2E) at a high clonal enrichment (37%) in a lymph node metastasis (Supplementary Data 4). This structural rearrangement was an ∼60 Mb tandem duplication within the 153 Mb X chromosome, which would not be amenable for rapid modeling using CRISPR/Cas9. In this patient's metastasis, AR 3′RACE/SMRT-Iso-Seq revealed that full-length AR mRNA was a major species, with high expression of minor species that contained contiguously spliced AR exons 1 to 3 and 3′ terminal exons derived from the 5′ end of the 60 Mb tandem duplication segment (Fig. 4H and I). Interestingly, AR-V7 and AR-V9 were also expressed in this metastasis, but at levels that were much lower than the novel AR fusion transcripts. Collectively, these data demonstrate that AR gene rearrangements occurring in the absence of other known AR gene alterations promote expression of diverse AR variant mRNA species that display splicing alterations after AR exons 1 to 3. Notably, these transcripts and their predicted protein products are molecularly distinct from AR-V7 (Fig. 4J).
Tumors harboring AR gene rearrangements concurrent with AR amplification express high levels of AR and a diversity of minor AR mRNA species
To address whether this relationship between AR gene rearrangements and high-level AR variant expression extended to tumors that concurrently displayed AR gene amplification, we performed AR 3′RACE/SMRT-Iso-Seq with 9 additional tumor tissues from 6 patients in the 41-patient AR-targeted cohort (Supplementary Fig. S7). In tumors that harbored AR gene rearrangements concurrent with AR gene amplification, expression of AR-V4, AR-V7, and AR-V9 was frequent, and these AR-V species were often coexpressed. Additional novel AR-V species were identified, but their fractional contributions to overall AR expression were similar to or less than AR-V7, AR-V4, and AR-V9. An exception was a novel AR-V transcript that contained contiguously spliced AR exons 1 to 3 and a novel 3′ terminal exon that contained cryptic exon 1 (CE1) but did not utilize annotated CE1 splice sites. This AR-V transcript was the most abundant AR-V species in a tumor from patient V5246, accounting for approximately 10% of overall AR expression (Supplementary Fig. S7).
It is challenging to reconstruct the architectures of complex rearranged and amplified genomic loci such as AR in heterogeneous tumors from short-read DNA-seq data. Therefore, we were unable to predict whether and how the AR expression patterns observed by AR 3′RACE/SMRT-Iso-Seq could be explained by underlying AR gene structures in these patients with amplified and rearranged AR (Supplementary Fig. S7). Adding further complexity, whole-genome DNA-seq data from the 101-patient cohort indicated that these amplified and rearranged AR gene architectures always occurred in the context of amplification of the upstream AR enhancer (Fig. 1B). Given that AR mRNA levels are highest in tumors harboring amplification of the AR gene body and upstream enhancer (16), we asked whether the presence of AR gene rearrangements concurrent with AR gene body and enhancer amplification affected AR expression levels. In the 101-patient whole-genome DNA-seq cohort, AR expression was higher in tumors that harbored AR gene rearrangements and amplification compared with tumors that harbored AR amplification alone (Supplementary Fig. S8). Thus, tumors harboring complex amplified and rearranged AR gene structures express a diversity of AR mRNA species, and also express high levels of AR overall.
Prostate cancer cells with engineered AR gene rearrangements express AR variants that drive resistance to AR-targeted therapies
To further refine the mechanisms by which AR gene rearrangements contribute to prostate cancer progression, we focused on patient C9_a. A metastatic tumor in patient C9_a harbored a chromosomal translocation fusing AR intron 3 to a region of chromosome 11, which provided a clonal growth advantage when modeled with CRISPR/Cas9 (Fig. 3C–F). We performed single-cell cloning and PCR screening of CRISPR/Cas9-transfected R1-AD1 cells to isolate a cell line clone that harbored this AR-chr11 translocation event (Fig. 5A–C). Western blot analysis of this clone (named R1-X-11) demonstrated a loss of full-length AR protein expression and gained expression of a truncated AR protein species (Fig. 5D). Analysis of AR mRNA transcripts expressed in R1-X-11 cells using AR 3′RACE/PacBio (Fig. 5E and F) and RT-PCR (Fig. 5G and H) demonstrated utilization of identical splice junctions and expression of the identical AR-V fusion transcript that was discovered in patient C9_a. Two separate siRNAs designed to target the fusion exon f89 from chromosome 11 knocked down expression of the truncated AR protein species in R1-X-11 cells, confirming this truncated protein was a novel AR-V originating from translation of an AR/chromosome 11 fusion transcript (Fig. 5I).
R1-X-11 cells displayed enzalutamide-resistant growth, whereas the parental R1-AD1 cell line was enzalutamide sensitive (Fig. 6A). Further, expression of AR target genes FASN, FKBP5, and ABCC4 was insensitive to enzalutamide in R1-X-11 cells, but inhibited by enzalutamide in R1-AD1 cells (Fig. 6B). To explore the mechanism for enzalutamide-insensitive cell growth and expression of AR target genes in R1-X-11 cells, we investigated whether the novel truncated AR-V protein was functioning as a transcription factor. In R1-AD1 cells grown in androgen-replete medium, binding of full-length AR to chromatin was inhibited by enzalutamide (Fig. 6C), which is a known mechanism of enzalutamide action (34). However, chromatin binding of the AR-V species expressed in R1-X-11 cells was insensitive to enzalutamide (Fig. 6C). Knockdown of the AR-V species in R1-X-11 cells using 2 separate siRNAs targeting the fusion exon f89 from chromosome 11 inhibited expression of FASN, FKBP5, and ABCC4, confirming transcriptional activity (Fig. 6D). Importantly, growth of R1-X-11 cells was reduced by the siRNAs targeting this AR-V, whereas growth of parental R1-AD1 cells was unaffected by these siRNAs (Fig. 6E). These data from a genetically engineered cell line model demonstrate a causal role for a translocation fusing AR intron 3 to chromosome 11 in a metastatic tumor, and an enzalutamide-resistant growth phenotype driven by an enzalutamide-insensitive AR-V transcription factor lacking the AR ligand binding domain.
This investigation of an aggregate cohort of 272 prostate cancer patients has demonstrated that AR gene rearrangements are a common feature of CRPC-stage tumors. Compared with other genes, AR is among the most frequently rearranged genes in the CRPC genome, with AR gene rearrangements detectable in 23% of CRPC patients analyzed by whole-genome DNA-seq and 39% of CRPC patients analyzed by deep AR-targeted DNA-seq. The highest frequencies of AR gene rearrangements occurred in patients treated with potent second-generation AR inhibitors abiraterone and enzalutamide. AR gene rearrangements were not detectable in primary prostate cancer, even in primary tumors from patients who ultimately progressed to CRPC and displayed AR gene rearrangements in their metastatic tumors. Overall, these findings support a model wherein AR gene rearrangements emerge during treatment with first-line hormonal therapies and become more frequent in prevalence with subsequent lines of AR-targeted therapy. In this regard, AR gene rearrangements display similar evolutionary trajectories to AR gene amplification and mutation events, which are well-defined mechanisms of resistance first reported in CRPC patients over 2 decades ago (35, 36).
The first AR gene rearrangement reported in prostate cancer was discovered in the CRPC 22Rv1 cell line, consisting of a 35 kb tandem duplication encompassing AR exon 3 (37). Subsequent studies with the CWR-R1 cell line, which originated from the same patient as 22Rv1, demonstrated a 50 kb deletion within AR intron 1 (19). Additional AR gene rearrangements were characterized in several patient-derived xenograft models, which consisted of deletions or inversions impacting the genomic segment containing AR exons 5–7 (32, 38). Investigation of a 15 patient rapid autopsy cohort documented the first AR gene rearrangements in CRPC tissues (15). Across these prior studies with CRPC models and specimens, as well as the current study, over 90 different AR gene rearrangements have been defined, consisting of deletions, duplications, inversions, and translocations, with breakpoints distributed along the length of the AR gene body. No specific AR gene rearrangement events have been recurrent in terms of rearrangement type and exact breakpoint coordinates.
This extreme heterogeneity of AR gene rearrangements across patients presents a challenge for clinical utility of testing for these events. This heterogeneity also poses a challenge for distinguishing whether these are driver or passenger events in prostate cancer progression. To address this, in the current study, we grouped AR gene rearrangements into two distinct classes. The first class of AR gene rearrangements occurred in tumors that lacked evidence for any other AR gene alteration. This class of AR gene rearrangements occurred in 8 of the aggregate 142 CRPC patients (5.6%) analyzed in this study. Analysis of representative AR gene rearrangements from this class using short-term CRISPR/Cas9 modeling, long-read AR isoform sequencing of tumors, and interrogation of an experimentally derived cell line demonstrated that these events drive high expression of tumor-unique AR variant species that are required for cell growth. The second, more frequent, class of AR gene rearrangements were concurrent with AR gene amplification and were detectable in 31 of the aggregate 142 CRPC patients (21.8%) analyzed in this study. The exact architectures of these AR gene rearrangements, and the number of AR gene copies affected by them, are difficult to decipher due to challenges inherent in inferring structures of amplified and rearranged loci in heterogeneous tumors from short-read DNA-seq data. Nevertheless, analysis of these tumors using long-read AR isoform sequencing demonstrated a diversity of AR isoforms expressed in these tumors, including frequent detection of AR-V7. Additionally, the AR levels in these tumors with amplification and rearrangement of the AR gene were the highest among all other groups studied. Understanding whether causal relationships underlie these observations will require a more precise knowledge of the structures of amplified/rearranged AR gene architectures than can be ascertained from short-read DNA-seq data.
The finding of novel tumor-specific AR-Vs expressed as major AR species exclusively in CRPC tumors contrasts with the broad expression of AR-V7 as a minor AR species that is observed across primary prostate cancer as well as CRPC patients (26, 39, 40). Detection of AR-V7 mRNA or protein in patient circulating tumor cells correlates with clinical variables such as overall survival (41, 42). Therefore, a positive signal for AR-V7 in circulating tumor cells is being used as a predictive biomarker of resistance to AR-targeted therapies in CRPC patients (43). Our long-read AR isoform analysis of tumors that harbored AR gene rearrangements but no other identifiable AR gene alterations revealed that most were AR-V7 negative, but positive for high expression of novel tumor-specific AR-Vs. The unique 3′ ends of these AR-V mRNA species, and the unique COOH-termini of their corresponding proteins, would not be detectable by the primer sets used to detect the AR exon3/CE3 splice junction, or antibodies used to detect the novel C-terminus of AR-V7 protein (Fig. 4J). As a result, current clinical assays for AR-V7 detection will miss these tumors where resistance to AR-targeted therapies is driven by the alternative AR-Vs that arise from underlying AR gene rearrangements. Given that the breakpoints from AR gene rearrangements can be detected in circulating tumor DNA from plasma of CRPC patients (44, 45), prospective studies are warranted to test the utility of AR gene rearrangements as biomarkers, especially for cases where no other AR alterations are detectable. In more complex cases where AR gene rearrangements and amplification are concurrent, knowledge that these tumors express a diversity of AR species and also have the highest levels of AR expression overall could be important for biomarker development and clinical trial design.
In summary, our study has documented widespread AR gene rearrangements in a large cohort of CRPC-stage tumors, and elucidated important roles for diverse AR gene rearrangement events as drivers of AR-V dependence and resistance to AR-targeted therapies. Prospective monitoring of patients for the presence or emergence of AR gene rearrangements is likely to have clinical utility.
Disclosure of Potential Conflicts of Interest
T.H. Hwang is an employee/paid consultant for MEDICALIP, reports receiving commercial research grants from AITRICS, and holds ownership interest (including patents) in Peloton. F.Y. Feng is an employee/paid consultant for Janssen, Sanofi, Astellas, Bayer, Celgene, EMD Serono, Ferring, PFS Genomics, SerImmune, and Nutcracker Therapeutics. J.S. de Bono is an employee/paid consultant for The Institute of Cancer Research, and reports receiving other commercial research support from AstraZeneca, Astellas, Janssen, Sanofi Aventis, Seattle Genetics, Bayer, Merck Sharp & Dhome (MSD), Merck Serono, Shattuck, Boehringer Ingelheim, BioExcel, Eisai, Roche, and Daiichi Sankyo. No potential conflicts of interest were disclosed by the other authors.
Conception and design: Y. Li, J.S. de Bono, S.M. Dehm
Development of methodology: Y. Li, C. Passow, B. Auch, J.S. de Bono, S.M. Dehm
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Li, Y. Ho, S. Carreira, D.N. Rodrigues, C. Bertan, D.A. Quigley, M. Fraser, S.R. Plymate, F.Y. Feng, J.S. de Bono
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Li, R. Yang, C.M. Henzler, Y. Ho, T.H. Hwang, D.A. Quigley, H.X. Dang, M. Fraser, C.A. Maher, J.S. de Bono, S.M. Dehm
Writing, review, and/or revision of the manuscript: Y. Li, R. Yang, C.M. Henzler, C. Passow, D.N. Rodrigues, D.A. Quigley, S.R. Plymate, C.A. Maher, F.Y. Feng, J.S. de Bono, S.M. Dehm
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.N. Rodrigues, C. Morrissey, M. Fraser, J.S. de Bono
Study supervision: S.M. Dehm
This study was supported by the Movember Foundation/Prostate Cancer Foundation Challenge Award (S.M. Dehm), NIH R01CA174777 (S.M. Dehm), and U.S. Department of Defense Prostate Cancer Research Program Transformative Impact Award W81XWH-13-2-0093 (to S.R. Plymate and S.M. Dehm). The University of Washington Prostate Cancer Donor Rapid Autopsy Program was supported by the Department of Defense Prostate Cancer Biorepository Network (PCBN; W81XWH-14-2-0183), the Pacific Northwest Prostate Cancer SPORE (P50CA97186), the PO1 NIH grant (PO1 CA163227), and the Institute for Prostate Cancer Research (IPCR). We thank the patients and their families. We acknowledge Celestia Higano, Evan Yu, Elahe Mostaghel, Heather Cheng, Pete Nelson, Bruce Montgomery, Mike Schweizer, Daniel Lin, Eva Corey, Funda Vakar-Lopez, Lawrence True, and the rapid autopsy teams for their contributions to the University of Washington Medical Center Prostate Cancer Donor Rapid Autopsy Program.
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.