Circulating Tumor Cell Genomic Evolution and Hormone Therapy Outcomes in Men with Metastatic Castration-Resistant Prostate Cancer

This article is featured in Highlights of This Issue, p. 933

Men with metastatic castration-resistant prostate cancer (mCRPC) face a range of life-prolonging treatment options, including novel androgen receptor (AR) inhibitors, sipuleucel-T, taxane chemotherapy, the radiopharmaceutical radium-223, and PARP inhibitors (1–5). However, despite the improved survival with AR inhibitors such as enzalutamide or abiraterone, most men with mCRPC will ultimately progress after 1 to 2 years of therapy, and around 20% of men do not respond to these treatments. Primary resistance to AR inhibitors has been linked to the detection of the AR splice variant AR-V7 in circulating tumor cells (CTC) in about 10% to 20% of patients, and we have recently validated two measures of CTC AR-V7 in a prospective multicenter trial called PROPHECY (3, 6, 7). However, many men with mCRPC lack CTC AR-V7 detection and alternative mechanisms of resistance, such as lineage plasticity and the neuroendocrine phenotype, TP53 or RB1 alterations, and alterations in AR itself may confer AR therapy resistance.

Comprehensive genomic studies of tumor biopsies in prostate cancer have revealed a complex mutational landscape with enormous interpatient and within-patient tumor heterogeneity (8, 9). This heterogeneity leads to a wide range of clinical outcomes and prognosis as the molecular genotype frequently is associated with clinical phenotypic events related to the timing of disease progression, metastasis, drug resistance, and tumor relapse. A major goal of precision oncology and clinical cancer research is to enable treatment selection through the identification of genomic alterations that match a patient with a therapy most likely to be beneficial. Given the challenges in metastatic biopsies in men with mCRPC, noninvasive methods may be preferred, and recent studies of CTCs and circulating tumor DNA (ctDNA) molecular alterations have demonstrated prognostic significance for a range of molecular alterations (8, 10–12). We have recently shown that CTCs can provide a unique source of biological material derived from cancer that may be missed in cell-free plasma DNA sequencing and can represent a patient-friendly, noninvasive, cost-effective real-time liquid biopsy marker for longitudinal monitoring, excluding background noise from healthy or normal cells, and can provide clinical utility as a biomarker of treatment response and outcomes (3, 12–16).

We recently demonstrated in the multicenter prospective PROPHECY clinical trial that men who have CTCs that express a ligand-binding domain truncated version of the AR, called AR-V7 variant, have a very poor response and a short time to progression or death when treated with the potent AR inhibitors. Here, we determined whether pre- or posttreatment genomic DNA alterations in CTCs from AR-V7-ve men with mCRPC could provide clinical utility in predicting clinical outcomes with these hormonal therapies.

In this prospective translational study (PROPHECY, NCT02269982), Institutional Review Board (IRB) approvals were obtained from Duke University IRB or the Weill Cornell Medical College (New York, NY) IRB (PROPHECY), along with written informed consent in accordance with Declaration of Helsinki ethical guidelines from all patients with mCRPC. All men had mCRPC with disease progression and were starting treatment with either abiraterone acetate with prednisone or enzalutamide. Eligibility and patient characteristics have been published previously (3), and disease progression at entry followed Prostate Cancer Clinical Trials Working Group (PCWG2) criteria (17). All patients had prostate adenocarcinoma, castrate levels of testosterone less than <50 ng/dL, and at least two poor risk factors for aggressive disease, including high serum lactate dehydrogenase, high alkaline phosphatase, radiographic progression, PSA doubling (<3 months), a high number of CellSearch CTCs (>=5 CTCs/7.5 mL whole blood), or prior enzalutamide or abiraterone therapy (3). In this analysis, we defined rapid progressors as a progression-free survival (PFS) of ≤3 months during treatment with abiraterone acetate (1000 mg daily) with prednisone or enzalutamide (160 mg daily), and extreme responders were defined as men who had PFS of >=6 months. We focused on those men with mCRPC who tested negative for the Epic AR-V7 nuclear protein assay, as these patients have a highly variable outcome, with some patients exhibiting long-term responses and some patients having primary treatment resistance (3).

CTCs prior to treatment and again at disease progression on abiraterone or enzalutamide were isolated using several methods in this study in separate blood tubes from men with mCRPC. To enumerate CTCs, we collected 7.5 mL whole blood in CellSave tubes, and CTCs were counted using the FDA-approved CellSearch CTC method based on EpCAM capture and characterization by lack of CD45 expression and positive pan-cytokeratin expression in a Clinical Laboratory Improvement Amendments (CLIA)-approved laboratory at Memorial Sloan Kettering Cancer Center (New York, NY).

CTC genomic studies were performed on Duke and Cornell patients with sufficient evaluable CTC DNA (Fig. 1 CONSORT diagram). For CTC genome-wide copy-number analysis, EpCAM-positive (EpCAM⁺) and CD45− CTCs were isolated from 7.5 mL mCRPC patient blood samples in independent tubes, as described in our previous study (14). Unspiked and spiked control EpCAM⁺ (T47D) cells into healthy volunteer blood were used for gating thresholds during FACS analysis (14). Independently, peripheral blood mononuclear cells (PBMC) were isolated from the same mCRPC patient's blood at the same timepoint to identify only somatic alterations in CTCs (14). Before hybridizing the circulating tumor cells DNA (CTC DNA) samples onto the array comparative genomic hybridization (aCGH) microarray chip (Agilent, 4 × 180K), all CTC's genomic DNA was amplified by using GenomePlex Single Cell Whole Genome Amplification Kits (Sigma, WGA4) to the yield required for cy3/cy5 labeling according to Agilent' aCGH manual (14). Furthermore, AR-V7 nuclear protein was detected in CTCs by using the Epic AR-V7 nuclear protein detection assays, as described previously (3).

For whole-genome copy-number alterations (CNA) analysis, two-color probe-based aCGH was performed using Agilent Oligonucleotide Array-based CGH for Genomic DNA Analysis (Sure Print G3 Human CGH array, 4 × 180K; ref. 14). The aCGH raw data were analyzed by using Agilent Cytogenomics Software default settings (≥3 probes, log₂ ratio for gain ≥ 0.25 and loss < −0.25), and all the CNAs were manually validated by two independent investigators (14, 18) and later, cross-referenced with publicly available prostate cancer datasets in cBioPortal (19).

For whole-exome sequencing (WES), initial CTC and white blood cell (WBC) DNA samples quality assessment, DNA library preparation, sequencing, and bioinformatics analysis were conducted at GENEWIZ, Inc. Details of the DNA Library Preparation, Quality Control, and HiSeq Sequencing are summarized in the Supplementary Materials and Methods, separately. We selected 11 men with mCRPC from our high-throughput genomic cohort of Duke and Cornell patients who had sufficient CTC and matched germline DNA at baseline and progression on abiraterone or enzalutamide for longitudinal WES and copy-number analysis for this analysis.

For WES analysis, Agilent software SureCall (v. 3.0.1.4) was used to analyze 150 bp paired–end FASTQ files from all the CTCs and WBC DNA samples as a paired analysis (tumor vs. matched WBC) under default parameters. Sequence reads were aligned to the reference genome (GRCh37/hg19) using the SureCall BWA–mem alignment tool following postalignment processing, included 100 bp region padding, removal of duplicates, and singletons (reads without a mate), followed by a normalization step, including GC and mappability corrections. Somatic variants were called using the SureCall SNPPET algorithm annotated with ClinSNP, ClinVar, Cosmic, gwasV1_ucsc, and NCBI. Only COSMIC validated variants were selected, and unknown variants were filtered out during WES analysis.

This analysis is considered an exploratory subset analysis of the overall PROPHECY trial, which had as its primary objective to determine the prognostic and predictive association of CTC AR-V7 detection by two methods (JHU mRNA and Epic Sciences nuclear protein) with PFS during treatment with abiraterone or enzalutamide (3). CTC genomic alterations were associated with outcomes using this case–control approach of sensitive versus resistant to determine the relevance of specific genomic alterations with primary resistance to these hormonal therapies. Secondarily, we examined longitudinal CTC samples for changes in genomic alterations over time to identify potential alterations associated with clonal selection and acquired resistance to abiraterone or enzalutamide. Because of the exploratory nature of the analysis, no P values are presented, and results are considered hypothesis generating.

A total of 120 men with progressive mCRPC starting treatment with abiraterone acetate or enzalutamide were enrolled in the PROPHECY study, and baseline characteristics of this cohort have been published previously (3). Of these 120 men, two did not start treatment, and five men who failed CTC extraction or CTC DNA quality testing were excluded from the downstream genomic analyses, leaving 113 evaluable men with mCRPC. Of these 113, 52 men consented to additional tubes of blood collection for our high-throughput sequencing cohort, where we enriched CTCs and paired PBMCs from 88 blood samples, including 52 baseline and 36 at disease progression, and performed whole-genome amplification (WGA4, Sigma). Of these 88 samples, 73 samples (45 baseline and 28 progression) had a sufficient amount of amplified genomic DNA for aCGH and WES analysis (Fig. 1 CONSORT diagram).

Overall, 48% (54/113) of mCRPC men at baseline had an unfavorable ≥ 5 CellSearch CTCs per 7.5 mL blood (range, 0–12,972 CTCs, median 4), and 27% (29/113) of mCRPC men had zero CTCs. Note that the 88 patient CTC samples were analyzed collectively for genomic analyses from individual men and timepoints and not single-cell sequenced, and each CTC sample represents a range of CTC numbers from <5 CTCs per blood sample to over 1,000 CTCs per sample, and a total of 5,198 CTCs were analyzed according to CellSearch CTCs criteria.

After whole-genome amplification of CTCs and paired PBMC germline DNA of these 52 men, 45 had sufficient pooled CTC DNA from single blood samples for a paired longitudinal analysis for CNAs at baseline (N = 25) and progression (N = 25). In addition, we classified patients into two groups defined by their outcomes to hormonal treatment, focusing on 37 men (23 sensitive and 14 resistant) who tested negative for the Epic Sciences CTC nuclear AR-V7 protein (3). Given our interest in exploring mechanisms of resistance to AR-targeted therapies in AR-V7–negative men with mCRPC, we further classified patients as sensitive or resistant to abiraterone or enzalutamide based on a PFS of > = 6 (n = 23) versus ≤3 months (n = 14), respectively. In this high-throughput cohort, 25 men had paired longitudinal CTC and PBMC collections for CNAs, and 11 men had paired CTCs, and PBMCs were analyzed for WES.

To associate genome-wide baseline and progression CTC CNAs with clinical outcomes during enzalutamide or abiraterone treatment, we first analyzed whole-genome CNAs using aCGH on 73 patients with pooled CTCs from the blood of 48 high-risk mCRPC men treated with enzalutamide or abiraterone acetate, including 45 baseline and 28 progression samples with matched germline reference genomic DNA (Fig. 1). Furthermore, we analyzed whole-genome CNAs data from CTCs DNA samples at baseline and progression and compared (baseline vs. progression CNAs). The top genes harboring genomic alterations in over 30% of baseline samples (N = 45) included the top gain in AR, IL3RA, KDM6A, MIR3690, or AKAP17A, and loss in ZFHX3, FGFR2, or BRCA1. At progression (N = 28), CTC DNA analysis demonstrated gains in AKAP17A, CD99, MXRA5, GYG2, or ARSF, and loss in NCOR1, ZFHX3, or RUNX2, are summarized in File S1. We next performed ingenuity pathway analysis among the above genomic altered genes at baseline and progression and found many common and distinct pathways associated with drug resistance and metastasis in mCRPC, respectively, which are summarized in Supplementary Table S1A and S1B. Similarly, we also compared the genomic alteration in sensitive (N = 23) and resistant baseline CTCs DNA at the genome level, and observed the copy gain in IL3RA, MUM1, AKAP17A, ZBED1, or CSF2RA, and copy loss in ZFHX3, BRCA1, or GRHL2, whereas, at resistant, demonstrated gain in JAK3, CSF2RA, CRLF2, EPOR, and PIK3R2, and loss in ZFHX3, PHLPP1, or BRCA1, all the genes are listed in Supplementary File S2. These data support the concept that CTC genomic alterations recapitulate the heterogeneity of common genomic alterations present in metastatic biopsies from men with mCRPC and suggest that CTC molecular profiling in such patients is feasible and could provide useful biologic information (14, 18).

Given the broad whole-genomic findings and heterogeneity between patients and over time, in this analysis, we focused on a panel of 60 recurrently altered and validated genes that have been previously associated with mCRPC pathogenesis, including those linked to androgen signaling, epigenetics, DNA damage and repair, osteomimicry, cell differentiation, metastasis, and drug resistance in mCRPC (8, 14, 20). We identified heterogeneity of somatic CNAs between men with mCRPC in this PROPHECY cohort. We analyzed blood samples from the 45 men for pooled CTCs DNA aCGH and identified typical genomic gain detected in AR (44%), KDM6A (40%), FOXA1 (36%), UGT2B17 (33%), and MYCN (31%), and loss in ZFHX3 (53%), FGFR2 (38%), BRCA1 (33%), ERG (31%), and PTEN (20%; Fig. 2A). Likewise, these commonly altered genes detected in baseline CTC DNA were also detected in 28 progression CTC DNA samples (Fig. 2B). Comprehensively, the 60-gene CNA status in CTC DNA at the baseline and progression are summarized in Supplementary Fig. S1A and S1B, respectively.

To validate these gains or losses identified in pooled CTC DNA, we analyzed publically available genomic data from 1,154 mCRPC tissue samples from 1,115 patients in the cBioPortal database (19). We validated many of these findings of recurrent losses or gains in mCRPC and the broad genomic heterogeneity in human metastatic tissue samples, providing some degree of cross-validation of our CTC genomic findings (Supplementary Fig. S2A and S2B).

To first study the clonal evolution and selection of genomic alterations associated with acquired treatment resistance to abiraterone or enzalutamide in men with mCRPC, we compared longitudinal CTC CNAs in 25 paired baseline and progression samples. We found a higher number of copy-number variants (CNV) in the progression samples compared with baseline (Fig. 3A), despite only a slightly higher number of CellSearch CTCs at progression as compared with 18 paired baseline (median 14 vs. median 9.5 CTCs per 7.5 mL whole blood; Fig. 3B). We also found that 56% (10/18) of patients had more CTCs at progression than their paired baseline cases (Fig. 3C).

Next, we compared 25 baseline and paired progression CTC CNA data to define novel CNAs associated with acquired or de novo resistance to abiraterone or enzalutamide. After progression, we detected a greater prevalence of CTCs harboring copy gain of ATM (44% at progression vs. 12% at baseline), FOXA1 (56% vs. 28%), UGT2B17 (52% vs. 28%), KDM6A (48% vs. 24%), CYP11B1 (44% vs. 20%), and MYC (24% vs. 4%; Fig. 3D). We also detected an enrichment for CTC copy loss of NCOR1 (64% vs. 20%), ZFHX3 (60% vs. 44%), RUNX2 (44% vs. 32%), PXN (36% vs. 20%) and ERG (36% vs. 20%), PTEN (32% vs. 24%), and RB1 (16% vs. 8%; Fig. 3E). In addition, the differential change (%) in the prevalence of copy gain or loss between progression and baseline CTCs DNA are shown (bottom; Fig. 3D and E). The CNAs of 60 genes in all 25 baseline and paired progression CTCs DNA samples together are summarized in Supplementary Fig. S3 as a heatmap.

We independently validated highly prevalent copy gain/loss genes that were identified at progression on abi/enza in the PROPHECY study cohort in a separate cohort of men with mCRPC starting radium-223 where all patients with mCRPC had progressed on prior therapy with enzalutamide or abiraterone (N = 14 cases; ref. 4). For example, in the radium-223 trial, we found gain in CYP11B1 (71%), AR (64%), FOXA1 (64%), and KDM6A (57%), and loss in ZFHX3 (86%), CHD1 (29%), and PHLPP1 (29%), which were in accordance with our PROPHECY longitudinal findings (Supplementary Fig. S4). Our data suggest that specific genomic alterations in CTCs are associated with clonal evolution or selection of cells contributing to hormonal therapy resistance in mCRPC men over time.

Given that many men with CTC AR-V7–negative mCRPC still have inadequate responses and rapid resistance to abiraterone or enzalutamide therapy, we next examined baseline CTC genomic alterations and associated these with clinical outcomes in Epic AR-V7–negative men (3). To understand the primary resistance mechanisms in CTCs, genomic alterations in 40 baseline AR-V7–negative (Epic AR-V7 nuclear protein assay) men's CTCs DNA were analyzed and compared. The patients with mCRPC benefiting from enza/abi therapy with ≥ 6 months PFS were considered as sensitive (N = 23), whereas patients with PFS ≤3 months were considered as resistant (N = 14). Patients with a PFS between 3 and 6 months were excluded from the analysis (N = 3) to enrich for outlier responses at either extreme of clinical benefit. We found that sensitive patients' PFS was broadly distributed with a PFS ranging from 6 to 25 months, whereas resistant patients' PFS had a tighter distribution by definition (Fig. 4A).

Next, we compared the number of CNVs calls at the genome level in sensitive and resistant patients and found that the resistant group had a trend of a higher number of CNVs calls than the sensitive group (Fig. 4B). However, we did not observe any difference in CellSearch CTCs counts between the sensitive and resistant groups, indicating that CTC number, while prognostic and a validated measure of disease burden (15), may alone not be a predictive marker of the benefits of AR inhibitor therapy (Fig. 4C). The median and range of CellSearch CTCs in resistant (N = 22) and sensitive (N = 14) groups were 5 (range, 0–194) and 1 (range, 0–193), respectively. We also compared the genomic copy gain and loss for 60 genes in the sensitive and resistant cases. We found that the resistant group harbored a higher number of genomic gains than the sensitive cases (median 20 vs. 14). However, we did not see any differences in the prevalence of large numbers of genomic losses between groups (Fig. 4D).

To examine specific CNAs associated with treatment response or primary resistance, we compared genomic CNAs between 23 sensitive and 14 resistant baseline CTCs DNA samples. We observed that men who had a clinical benefit to enza/abi (median PFS, 10 months) were more likely to have CTCs with genomic gains of ATM, NCOR2, HSD17B4, or PTEN, or loss of BRAF, ABL1, MYC, or NKX3-1. Conversely, men who did not respond to enza/abi treatment (median PFS, 2.6 months) had CTCs enriched for copy gain of BRCA2, APC, KDM5D, MYCN, CYP11B1, SPARC, and AR, and loss of CHD1, PHLPP1, PTEN, ERG, ZFHX3, and NCOR2. The prevalence of genomic copy gain/loss and the percent difference (delta) between resistant and sensitive CTCs DNA is shown in Fig. 4E and F as a superimposed bar chart and heatmap, respectively. Specifically, we found that the DNA damage repair genes ATM and BRCA2 were surprisingly differentially gained in resistant versus sensitive CTCs DNA, with BRCA2 gain enriched in resistant patients and ATM gain enriched in sensitive patients. We further validated the genomic gains of these genes in multiple independent metastatic prostate cancer datasets (N = 1,154 samples from 1,115 patients) in cBioPortal (Fig. 4G), suggesting that alterations in BRCA2 and ATM-mediated signaling may be inversely associated with outcomes during treatment with abiraterone or enzalutamide.

The preceding work focused mostly on CNAs, given their critical importance to mCRPC pathophysiology. However, mutations in oncogenes and tumor suppressors are also relevant genomic alterations associated with poor outcomes in mCRPC (8, 21, 22). To identify unique somatic variants that develop or undergo clonal selection in CTCs DNA overtime during abiraterone/enzalutamide treatment, we evaluated CTC exome mutations longitudinally in a subset of patients with paired samples. We performed WES in 11 Epic AR-V7–negative men at baseline and progression on enzalutamide/abiraterone and compared mutational calls against patient-matched germline DNA from leukocytes as a reference control and to rule out clonal hematopoiesis contamination of samples. The average read depth in analyzable target regions was over 87×, and all calls were annotated and functionally validated in COSMIC datasets. The raw data from the WES of CTCs and WBC DNA were analyzed as a paired analysis (CTCs vs. leukocytes) using Agilent SureCall-NGS software.

Focusing initially on the same 60 prostate cancer–associated and validated pathophysiologic genes from our CNA analysis, we detected high COSMIC-validated nonsynonymous pathogenic mutation frequencies in progressing patients' CTC DNA. In particular, we identified the enrichment of several pathogenic exome mutations at progression on abi/enza, including TP53, CDK12, KMT2D, BRAF, BRD4, and SPOP (Fig. 5A). For example, 45% (5/11) of progression cases harbored TP53 mutations, whereas TP53 mutations were detected in only 27% (3/11) of cases at baseline, as shown in Fig. 5B. Likewise, 36% (4/11) of progression samples had mutations in the CDK12 gene compared with 18% (2/11) of cases at baseline. Nonsynonymous pathogenic mutations in SPOP (patient's ID# 809-05, AA: D144G, AF: 0.573) and TP53 (patient's ID# 809-7, AA: T111K, AF: 0.994) were uniquely detected in progression CTCs DNA, but not in paired baseline or germline DNA, shown as a snapshot image in Integrative Genomics Viewer (IGV) browser (Fig. 5B). On the contrary, some of the genes had a high prevalence of pathogenic mutations detected at baseline CTCs DNA than paired progression CTCs DNA samples, including RB1, BRCA2, and PTEN (Fig. 5A). For all the genes, sequence read depth, position, genotype, COSMIC changes, genotype quality, and amino acid changes, etc., are summarized in File S3. In addition, to combine frequent CNAs and mutation in the same sample, we analyzed copy number and mutation data together for all 11 baseline and paired progression CTCs DNA samples and observed a higher somatic tumor mutation burden in progression CTCs samples compared with the baseline (Fig. 5C; Supplementary File S4).

To identify the mutations associated with enza/abi resistance, we compared WES data from 5 baseline sensitive and 5 resistant patients' CTCs DNA. Here, we found many genes with pathogenic mutations enriched in resistant samples, such as PTEN, TP53, CDK12, NRAS, or BRAF (Fig. 5D). For example, PTEN mutations detected in sensitive (1/5) and resistant (2/5) patients are shown as a lollipop graph generated in cBioPortal (Fig. 5E). For example, mutation detected uniquely in genes PTEN (patient's # 809-16, AA: D178G, nonsynonymous, missense, AF: 0.290) and CDK12 (patient's # 809–03, AA: E131K, nonsynonymous, missense, AF: 0.993) were visualized with sequenced read depth coverage by IGV genome browser, shown in Fig. 5E. We observed a high tumor mutation burden in resistant samples compared with sensitive cases (Fig. 5F, bottom, green color). Moreover, a high prevalence of common mutation and copy-number change was seen in resistant cases compared with sensitive cases shown in Fig. 5F. For all the sensitive and resistant samples, the read depth, COSMIC annotation, position, genotype, genotype quality, or amino acid changes are summarized in Supplementary File S5. Besides, all the mutations uniquely detected at baseline versus progression and sensitive versus resistant CTCs DNA (CTC vs. WBC) were visualized as a snapshot image showing reference and alternative alleles in the IGV browser are summarized in Supplementary Fig. S5 and S6, respectively. Our somatic pathogenic exome mutation analysis indicated that CTC DNA mutations were most likely associated with disease progression and clinical outcomes in patients with mCRPC treated with enza/abi.

In this prospective study, we analyzed whole-genomic CNAs and whole-exomic somatic mutations in CTCs in the context of mCRPC treatment resistance to identify potential predictive and resistance biomarkers of treatment outcomes with the potent AR inhibitors enzalutamide or abiraterone acetate. We found evidence for broad genomic heterogeneity in copy gains and losses using a liquid biopsy approach in AR-V7–negative men from the multicenter PROPHECY study, and more importantly, that some of these genomic alterations were associated with clinical outcomes with AR inhibitor therapy (3). While our data confirm prior work suggesting poor outcomes in the setting of PTEN loss, MYCN gain, AR gain, and TP53 mutations, we have identified novel associations with AR inhibitor sensitivity, including gains in ATM, NCOR2, and HSD17B4, and novel associations of primary resistance to AR inhibition with gains of BRCA2, APC, KDM5D, MYCN, CYP11B1, SPARC, and AR, and loss of CHD1, PHLPP1, PTEN, ERG, ZFHX3, and NCOR2. These findings, coupled with validation of this work in public datasets of mCRPC patient tissue samples, warrant further prospective validation and mechanistic studies.

We recently demonstrated that men with mCRPC in the PROPHECY trial who have CTCs that express AR-V7 and treated with enzalutamide or abiraterone have poor outcomes, including a lack of PSA or radiographic responses and short PFS or overall survival (3). Nevertheless, most men in this setting lack AR-V7 and additional genomic alterations are likely contributing to drug resistance (6). In this study, we show that the men who had the clinical benefit (PFS ≥ 6-months) to enzalutamide or abiraterone were more likely to have CTCs with genomic alterations in pathways implicated in DNA repair, steroid metabolism, PI3K, and WNT pathway signaling, chromatin and epigenetic gains linked to loss of CHD1 (23) and gain of KDM5D, and lineage plasticity genes including MYCN, FOXA1, and NKX3.1. Moreover, after disease progression on AR inhibitor treatment, we observed the clonal evolution of CTCs harboring genomic gain of ATM, FOXA1, UGT2B17, KDM6A, CYP11B1, and MYC, and acquired loss of NCOR1, ZFHX3, and ERG. While we cannot establish causality from this correlative work, collectively, these genomic alterations in CTCs suggest critical mechanisms of primary or acquired resistance to hormonal therapies in men with mCRPC. Furthermore, our somatic exome mutation analysis of CTCs DNA detected various COSMIC-validated nonsynonymous mutations in progressed or resistant AR-V7–negative patients with enrichment of pathogenic mutations, such as TP53, AKAP9, CDK12, KMT2D, BRD4, and SPOP. Thus, these genomic alterations in androgen signaling, WNT, lineage plasticity, epigenetic regulators, and DNA damage pathways support a variety of pathways that were associated with poor clinical outcomes in AR-V7–negative mCRPC men treated with enzalutamide or abiraterone.

In this study, we observed several genomic alterations that may be associated with tumor progression, drug resistance, lineage plasticity, and lethal disease tumor heterogeneity in mCRPC (24). AR overexpression and genomic amplification are the most common occurrence in enza/abi-treated patients with mCRPC, associated with poor clinical outcomes, where a large proportion of patients may develop primary or secondary resistance to hormonal therapy (14, 25). In this prospective study, we observed AR gain in 57% (8/14) of resistant and 39% (9/23) of sensitive baseline CTCs DNA samples, which might suggest the role of increases AR amplification/gain in developing primary resistant cancer. Remarkably, we did not observe any differences in AR gene alterations at the DNA level after disease progression [at progression 8/25 (32%), vs. baseline 11/25 (44%)], which reflects that therapy-resistant clones were present before hormonal therapy. Thus, the genomic alterations of CTCs could be indicative of clonal evolution to survive enza/abi treatment pressure, leading to the selection of resistant subclones while eliminating other CTC clones.

In the AR transcription complex, FOXA1, one of the key members for the AR transcriptional activity, is upregulated in advanced prostate cancer, associated with poor prognosis and disease progression, and correlated with PSA level and Gleason scores (26, 27). An in vitro study showed that the silencing of FOXA1 in LNCaP cells reduced cell proliferation and growth of prostate cancer (28). In accordance, we also observed that FOXA1 was one of the top genes with a high prevalence of copy gain detected in our CTCs DNA at baseline 7/25 (28%) and paired progression 14/25 (56%) in patients with mCRPC, overtime. Therefore, the increased prevalence of FOXA1 gain in mCRPC CTCs might be associated with acquired resistance, and AR reprogramming (27) requires further independent confirmations.

In our previous CTCs versus cfDNA study (18), we established that MYCN is uniquely copy gained in CTCs DNA, but not in paired cfDNA in the same patient reflected different CTCs biologics, associated with poor clinical outcomes, neuroendocrine-like prostate cancer phenotype, and resistance to androgen deprivation therapy in patient with mCRPC treated with enza/abi (18, 29, 30). We again observed MYCN gain in 43% (6/14) of resistant versus 22% (5/23) of sensitive baseline CTCs DNA. In contrast, we did not see any changes in copy number between the 25 baseline and paired progression CTCs DNA samples (28% vs. 28%). Here, our MYCN CNAs findings suggest that MYCN might be associated with the primary drug resistance and could be used as a prognostic biomarker for neuroendocrine prostate cancer (NEPC)-like CTCs during hormonal therapy, which requires further clinical validations and functional study (31). In addition, altered genes such as AURKA might also be associated with the primary resistance in the patients with mCRPC treated with abi/enza. However, on the other hand, AURKA gain may not be associated commonly with acquired resistance given its lack of enrichment in progression samples. Determination of AURKA's role in AR therapy resistance and neuroendocrine transformation will require further mechanistic and targeted therapy studies in selected patients (32).

In addition to genomic copy gain, there were many other genes with high prevalence in genomic copy loss, such as ZFHX3, BRCA1, PTEN, ERG, and FGFR2, which play an essential role in disease progression and primary resistance in mCRPC (8, 14, 33). For example, we found that ZFHX3, which acts as a tumor suppressor gene associated with Erk1–2/AKT signaling, development of prostatic intraepithelial neoplasia, and cell proliferation in prostate cancer (34), had a high prevalence of genomic loss in our 15 of 25 progressed (60%) CTCs DNA in comparison with the paired baseline 11/25 (44%) CTCs DNA. When we compared ZFHX3 loss between resistant and sensitive baseline CTCs DNA, we found a high prevalence of ZFHX3 genomic loss in resistant CTCs DNA (9/14, 64%), whereas 48% (11/23) was detected in sensitive baseline cases. The genomic loss of ZFHX3 in CTCs DNA could be associated with both primary and acquired resistance in patiengts with mCRPC treated with enza/abi, which needs further functional and clinical validations. Finally, these genomic alterations findings support that CTCs DNA genomic alterations may be the driving force in mCRPC progression and hormonal therapy failure.

In a prospective study, Zhang and colleagues showed that genomic loss of the CHD1 gene in metastatic prostate cancer promotes resistance to AR-targeted hormonal therapy through chromatin modifications, and associated with worst clinical outcomes in patients treated with enzalutamide may lead to the emergence of antiandrogen drug resistance in mCRPC (23). Here, we also observed CHD1 gene loss in 4 of 14 (29%) of the resistant versus 2 of 23 (9%) sensitive patients with mCRPC treated with enza/abi, which might be associated with hormonal therapy failure, requires further study.

It has previously been shown that lethal clones with clonal heterogeneity were distinguished from the primary tumor to metastases during disease progression in prostate cancer, longitudinally, where they observed many genetic alterations including pathogenic mutations in many vital genes such as APC, SPOP, PTEN, and TP53 in metastatic tissues, but not in primary tissues, (35). Similarly, we also observed a high prevalence in many genes, for example, SPOP, TP53, and BRD4, where somatic exonic non-synonymous pathogenic COSMIC validated mutations were detected in progressed CTCs DNA, but not in baseline CTCs DNA. For example, we observed a range of genomic alterations at progression, including new TP53 mutations in three cases, each of which had a PFS >6 months, indicating that this new mutation may be acquired during resistance. We observed two cases of persistent TP53 mutations at each timepoint, each of whom had PFS times less than 3 months, indicating primary resistance. The presence of these common genomic alterations and exomic mutations may indicate common clones associated with drug resistance and metastases in patients with mCRPC treated with enza/abi. Therefore, our findings suggest the emergence of progressive subclone with PTEN, TP53, SPOP, or BRD4 mutations within progression CTCs requires further validations in larger cohorts of patients with mCRPC (36).

The limitations of our current analysis first include the small size of our cohort that underwent whole-genomic or WES within the overall broader PROPHECY study, limiting our ability to perform statistical testing between the individual alterations with clinical outcomes. The ability to detect CTC genomic alterations is directly related to the number of CTCs evaluable at that time, and thus, genomic findings may be biased by disease burden and differences in sensitivity of whole-genomic assays at different timepoints due to changes in CTC numbers in response to therapy and at progression. Our work should be considered hypothesis generating and exploratory, necessitating prospective validation in larger cohorts. Second, it is possible that some of the identified genomic alterations may represent low-level passenger changes that, while common, may not be driving treatment outcomes. We focused the present study on those genomic alterations previously validated pathophysiologically and clinically in mCRPC or preclinical models and those alterations that were more common across patient samples, thus reducing the potential for bias related to misclassification during genomic amplification of CTC DNA, sequencing, and variant calling. We also excluded germline alterations using paired PBMCs to avoid contamination with genomic alterations associated with clonal hematopoiesis.

In conclusion, our prospective analysis showed the practical feasibility of longitudinally profiling whole-genome CNAs and exome mutation analysis of CTCs DNA. We identified an increased prevalence of common CNAs inclusive of genomic regions associated with critical pathways related to disease progression and treatment resistance in patients with mCRPC. These genomic alterations were associated with androgen signaling, tumor suppressors, WNT, NEPC, epigenetic regulators, and DNA damage and repair relevant to mCRPC biology (Fig. 6). The alterations we identified may provide insights into metastatic biology and provide future clinical applications for biomarker discovery in men with mCRPC in context to enza/abi treatment. The study of genomic alterations in CTCs provides putative clinically actionable information and tracking clonal evolution over time that could influence treatment options in mCRPC cases. Moreover, if validated, our results may suggest that CTC DNA biomarkers can identify and predict sensitivity and resistance to enzalutamide or abiraterone acetate treatment in men with mCRPC irrespective of AR-V7 status.

D.M. Nanus reports grants from Prostate Cancer Foundation and Department of Defense during the conduct of the study. D.J. George reports personal fees from AACR, Axess Oncology, Bayer, Capio Biosciences, Constellation Pharma, EMD Serono, Flatiron, Ipsen, Merck, Michael J Hennessey Assoc, Millennium Med Pub, Modra Pharma, Myovant Sciences, NCI Genitourinary, Nektar Thera, Physician Education Resource, Propella Therapeutics, UroGPO, UroToday, and Vizuri Health Sciences, LLC; grants and personal fees from Astellas, AstraZeneca, BMS, Exelixis, Janssen Pharma, and Sanofi; grants from Calithera, Novartis, and Pfizer outside the submitted work. A.J. Armstrong reports grants from Prostate Cancer Foundation, Movember, and NIH during the conduct of the study; grants and personal fees from Merck, BMS, Astellas, Dendreon, Bayer, and AstraZeneca; personal fees from Clovis; grants from Janssen and Genentech outside the submitted work. No disclosures were reported by the other authors.

S. Gupta: Conceptualization, resources, data curation, software, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. S. Halabi: Resources, formal analysis, supervision, funding acquisition, visualization, methodology, writing–review and editing. G. Kemeny: Resources, data curation, methodology. M. Anand: Resources, data curation. P. Giannakakou: Resources, supervision, funding acquisition, investigation, writing–review and editing. D.M. Nanus: Resources, supervision, funding acquisition, investigation, writing–review and editing. D.J. George: Conceptualization, resources, supervision, funding acquisition, investigation, project administration. S.G. Gregory: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, project administration, writing–review and editing. A.J. Armstrong: Conceptualization, resources, data curation, supervision, funding acquisition, validation, investigation, project administration, writing–review and editing.

We wish to thank the Prostate Cancer Foundation and Movember for their financial support of this Global Treatment Sciences Challenge Award for this multicenter study. A.J. Armstrong was supported by a Prostate Cancer Foundation https://doi.org/10.13039/100000892 grant and 1R01CA233585-01, and the DCI https://doi.org/10.13039/100014228 P30 CA014236 and Duke Cancer Institute (Durham, NC) shared resources for biostatistics, Flow Cytometry, and Sequencing and Genomic Technologies. We acknowledge support from the NIH/NCI and 1R01CA256157-01 (PI A.J. Armstrong and S. Halabi). We wish to thank the study coordinators at Weill Cornell (New York, NY), Duke University (Durham, NC), Johns Hopkins University (Baltimore, MD), University of Chicago (Chicago, IL), and Memorial Sloan Kettering Cancer Center (New York, NY). We wish to acknowledge the dedication of our patients to provide blood samples. We would like to acknowledge the assistance of the Duke Molecular Physiology Institute Molecular Genomics core facility for the generation of data for the article.

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.