Platelet-Coated Circulating Tumor Cells Are a Predictive Biomarker in Patients with Metastatic Castrate-Resistant Prostate Cancer

Advanced prostate cancer presents in certain cases with clinical characteristics of a distinct subtype of the disease known as aggressive-variant prostate cancer (AVPC; refs. 1–3). This subtype is characterized by an aggressive clinical course with poor prognosis and more frequently arises following hormonal therapy in patients with metastatic castration-resistant prostate cancer (mCRPC; refs. 1–3). Studies have shown that while AVPC responds poorly to hormonal therapies, it is vulnerable to chemotherapy, although the response is short-lived (1, 4). Our current knowledge about this aggressive variant is incomplete and guidelines on specific treatment recommendations have not been established. To overcome this challenge, diagnostic and predictive biomarkers are urgently needed to stratify patients with mCRPC for this subtype who could benefit from alternative therapeutic interventions.

Although several clinicopathologic criteria of AVPC (AVPC-C) were initially described to facilitate its recognition (1), clinical manifestations are difficult to identify, necessitating molecular biomarkers. So far, genomic characterization of solid biopsies from primary and metastatic tumor tissues has revealed that AVPC can be characterized by a molecular signature (aggressive variant prostate cancer molecular signature; AVPC-MS) composed of combined defects in at least two of the three tumor suppressor genes PTEN, RB1, and TP53 (2,5). Importantly, the presence of this molecular signature may predict AVPC vulnerability to taxane-platinum combination therapies relative to taxane alone (4). However, a missing aspect in the clinical management of mCRPC is the minimally invasive assessment of changes in the tumor to identify patients with AVPC in a timely manner at the correct time point for the appropriate therapeutic decision-making. This aspect is ideally suited for a liquid biopsy approach.

Liquid biopsy approaches offer a potential source of circulating tumor cells (CTC) amongst a spectrum of analytes that can be serially obtained during the course of the disease and treatment interventions (6–8). This may overcome critical limitations of solid biopsies in dissecting the intratumor heterogeneity and its evolution in an individual patient over time (9,10). Successfully used in clinical management of late-stage prostate cancer (11,12), the direct imaging liquid biopsy high-definition single-cell assay (HCSCA; ref. 13) is used here in an advanced implementation for the quantitative separation of multiple CTC subtypes. In addition to describing phenotypic, morphometric, and organizational features, the HDSCA allows for parallel downstream single-cell genomic and targeted proteomic characterization via low-pass whole-genome sequencing (6) and imaging mass cytometry (14,15), respectively. It has been used in both peripheral blood (PB) and bone marrow aspirates (BMA) to discern features most significant to the dynamics of metastatic progression in advanced prostate cancer (15,16), and to portray genomic instability as a distinctive feature in patients with AVPC in comparison with patients with mCRPC that are AVPC-negative (17).

Herein, we report on the use of the advanced capabilities of the third-generation HDSCA (HDSCA3.0) in separating a wider spectrum of CTC subtypes and other disease-related cells from patients with mCRPC. Its application in mCRPC identifies platelet-coated CTCs (pc.CTC) as a predictive biomarker for improved response to combination versus single chemotherapy.

PB and BMA samples were collected from participants immediately starting treatment on trial NCT01505868 entitled “Study of cabazitaxel with or without carboplatin in patients with metastatic castration-resistant prostate cancer”. The trial aimed to test the hypothesis that carboplatin improves the efficacy of cabazitaxel in men with advanced prostate cancer with the additional intention of evaluating the effect of aggressive variant features on response and outcome. This was a prospective, randomized, open-label, phase-I (n = 9 patients) and II (n = 160 patients) study at the University of Texas MD Anderson Cancer Center and Barbara Ann Karmanos Cancer Institute (4). All patients were required to have castration-resistant disease progression, an Eastern Cooperative Oncology Group (ECOG) performance status between 0 and 2, and adequate organ function. For phase II, patients were stratified by factors including ECOG performance status, previous docetaxel treatment, response to docetaxel among those who received it, and the presence of at least one of the seven AVPC-C criteria (1). The study was approved by the corresponding institutional review boards and was conducted in accordance with ethical principles founded in the Declaration of Helsinki. All patients gave written informed consent.

For this study, we evaluated a subset of 79 patients based on the availability of liquid biopsy samples (PB and BMA) for analysis. All patients had at least 1 sample type collected and 48 had paired samples. A total of 68 PB and 60 BMA samples were analyzed. PB samples from 11 normal blood donors (NBD) from our repository, and defined as individuals with no known malignancy, were used as negative controls.

PB (7.5 mL) and BMA (7.5 mL) samples were collected in 10-mL collection tubes (Cell-free DNA BCT, Streck) at the clinical sites and sent to the Convergent Science Institute in Cancer (CSI-Cancer) at the University of Southern California, for sample processing within 24 hours as previously described (13,16, 17). In brief, upon receipt, samples were subjected to erythrocyte lysis in isotonic ammonium chloride solution and the entire nucleated cell population was plated as a monolayer onto custom cell adhesion glass slides (Marienfeld) with approximately 3 million cells per slide. The cells were then incubated in 7% BSA, dried, and stored at −80°C for long-term storage before use.

Slides were stained with the use of an IntelliPATH FLX autostainer (Biocare Medical LLC) in batches of 50 [46 patient slides (2 per patient), plus 2 NBD slides and 2 NBD slides spiked with LnCAP cells (ATCC CRL-1740) as negative and positive quality controls, respectively]. All steps were performed at room temperature. Slides were thawed for 1 hour and cells were then fixed with 2% neutral buffered formalin solution (VWR) for 20 minutes. Nonspecific binding sites were blocked with 10% goat serum (Millipore) for 20 minutes. Slides were subsequently incubated with a conjugate containing 2.5 μg/mL of a mouse IgG1 anti-human CD31:Alexa Fluor 647 mAb (clone: WM59, MCA1738A647, BioRad) and 100 μg/mL of a goat antimouse IgG monoclonal Fab fragments (115–007–003, Jackson ImmunoResearch) for 4 hours (Fig. 1A). After incubation with CD31-Fabs, cells were permeabilized using 100% cold methanol for 5 minutes. They were then incubated with an antibody cocktail consisting of mouse IgG1/IgG2a anti-human cytokeratin (CK) 1, 4, 5, 6, 8, 10, 13, 18, and 19 (clones: C-11, PCK-26, CY-90, KS-1A3, M20, A53-B/A2, C2562, Sigma), mouse IgG1 anti-human CK 19 (clone: RCK108, GA61561–2, Dako), mouse antihuman CD45:Alexa Fluor 647 (clone: F10–89–4, MCA87A647, AbD Serotec), and rabbit IgG antihuman vimentin (VIM): Alexa Fluor 488 (clone: D21H3, 9854BC, Cell Signaling Technology) for 2 hours. Slides were then incubated with Alexa Fluor 555 goat anti-mouse IgG1 antibody (A21127, Invitrogen) and counterstained with 4′,6-diamidino-2-phenylindole (DAPI; D1306, Thermo Fisher Scientific) for 1 hour. Finally, slides were mounted with a glycerol-based aqueous mounting media before adding coverslips to maintain cell integrity. Meanwhile, platelet confirmation was performed with manual staining with VIM replaced by rabbit IgG anti-human CD61 (clone: SJ19–09; MA5–32077, Thermo Fisher Scientific) on the NBD sample spiked with SKRB3 cells (ATCC HTB-30) followed by Alexa Fluor 488 goat anti-rabbit IgG antibody (A11034, Invitrogen).

The cells on the slides were imaged as previously reported (13). Briefly, the slides were scanned using automated high-throughput fluorescence scanning microscopy at 10x objective magnification and producing 2,304 frame images per fluorescence channel per slide (Fig. 1B). Exposure times and gain for Alexa Fluor 555 (CK), Alexa Fluor 488 (VIM or CD61), Alexa Fluor 647 (CD45/CD31), and DAPI (DNA) fluorescence channels were automatically set up by the scanning microscope set to yield the same background intensity level across slides for normalization purposes.

The HDSCA methodology is based on cell imaging data analysis to identify several rare cells in the blood as opposed to cell enrichment approaches that can only identify a specific cell phenotype. For this reason, we first detect and classify rare cells amongst all nucleated cells to then identify CTC subtypes within them based on morphologic features. Rare cell candidates are detected using Outlier Clustering Unsupervised Learning Automated Report (OCULAR), a custom pipeline based on image processing, dimensionality reduction, and unsupervised clustering methods (Fig. 1C). The algorithm first uses the “EBImage” R package (ref. 18; EBImage_4.12.2) to segment the images of every event on the slide, separating DAPI⁺ (cells) and DAPI⁻ events. OCULAR then performs feature extraction for each cell with the “computeFeatures” function from the “EBImage” package, collecting 761 quantitative cellular and nuclear parameters. Principal component analysis (PCA) of these parameters followed by a hierarchical clustering on principal components (using the top 350 components) are then applied to identify rare cells and common cells from all DAPI⁺ events. Because PCA and hierarchical clustering are static methods and introduce no stochasticity, all results are repeatable.

Rare cells are further classified into 8 distinct groups based on their fluorescence signal intensities for Alexa Fluor 555 (CK), Alexa Fluor 488 (VIM), and Alexa Fluor 647 (CD45/CD31) channels as shown in Fig. 1C. Importantly, OCULAR is independent of specific biomarkers in the fluorescence channels and is compatible with other fluorescence protocols including previous versions of the HDSCA (13). Further manual curation of the classified rare cells is performed to remove artifacts and noninteresting cells providing the cleaned rare cell groups and their cell counts. Subsequently, morphologic parameters are used to distinguish 4 CTC subtypes amongst the 4 CK⁺ rare cell groups and details were described in the results.

We followed previously described methods for rare cell relocation, reimaging, isolation, next-generation sequencing (NGS), and copy-number alteration (CNA) analysis (6,15, 19). In brief, slides were transferred to a Nikon 80i microscope, cells of interest were relocated using registered coordinates, and 40× images were captured. Subsequently, individual cells were extracted from slides using a robotic micromanipulator system followed by single-cell whole-genome amplification (WGA; Sigma-Aldrich; catalog no. WGA4). Libraries were constructed using the DNA Ultra Library Prep Kit (New England Biolabs; catalog no. E7370) and sequenced using Illumina HiSeq at USC Genomics Core or Fulgent Genetics. The reads uniquely mapped to the human genome were used to reconstruct the copy-number profile of each individual cell. Only cells with total reads above 30,000 per cell, total alignment rate above 50%, nonsignificant noise, and no apoptosis-induced alterations were included in the analysis. The bin-based copy-number gains or losses was used to estimate gene-level CNAs in PTEN, RB1, and TP53 for AVPC-MS assessment.

The clinical characteristics and outcomes, including progression-free survival (PFS) and overall survival (OS), were previously reported (4) and subsequently updated for further correlation analysis. PFS was calculated from the date of specimen collection to the date of first occurrence of progression or last follow-up used for censorship. OS was calculated from the date of specimen collection to the date of death or last follow-up used for censorship. For the Kaplan–Meier analyses, via the “survival” and “survminer” R package (versions 3.1.7 and 0.4.8), two-sample and multiple comparison log-rank tests and univariate models were used to measure observed survival differences between groups. The “Complex Heatmap” R package (version 2.1.0) was used to generate summary heatmaps of CTC subgroup presence and AVPC-MS status. In the tables, categorical features were listed as numbers of cases and their percentages and numeric features were presented as median values and their ranges. Association analyses were performed via a Student t test (two groups) for parametric data, Mann–Whitney U test (two groups) for nonparametric data, and χ² test for categorical data.

In this study 68 PB and 60 BMA samples from 79 patients with mCRPC and PB samples from 11 NBDs were evaluated. Patients' clinical characteristics at time of draw are summarized in Table 1. Cell-based liquid biopsies have previously demonstrated the existence of multiple cell subtypes of the tumor microenvironment in the PB ranging from epithelial to mesenchymal to endothelial staining patterns and displaying ranges in both size and shape (20–23). Consequently, we adapted both our immunofluorescence (IF) assay and analytical methods to optimize the ability to identify and characterize a wide range of these subtypes in a single workflow. HDSCA3.0 uses a cocktail of fluorescent antibodies to detect cells of epithelial, endothelial, mesenchymal, and immune origin. Starting with the previously validated cocktail of anti-CK and anti-CD45 antibodies (13,16), we added two antibodies raised against CD31 and VIM (Fig. 1A) to detect events with endothelial or mesenchymal characteristics in addition to epithelial cells. Scanning microscopy-based cell-image acquisition (Fig. 1B) along with data collection and analysis were conducted using unsupervised clustering implemented in an inhouse data analysis workflow (OCULAR). This methodology uses the IF-imaging data to identify rare cells and classify them into 4 CK⁺ and 4 CK⁻ cell groups according to the four possible combinations with VIM and/or CD45/CD31 (Fig. 1C). Importantly, CK⁺ cell groups include the cells captured by previous versions of HDSCA where CK was the primary positive criterion for identifying CTCs in addition to morphologic parameters. In general, and consistent with prior reports, the enumeration of total rare cells per mL of volume was significantly greater in the BMA compared with the PB (median: 4,096.8 vs.75.0 rare cells/mL), as well as total 4 CK⁺ groups (1,846.0 vs. 41.7 cells/mL) and the “CK⁺ only” group (261.3 vs. 9.7 cells/mL) and all 8 rare cell groups were represented in both sample types (Fig. 2A).

Subsequent to identification, single-cell genome-wide CNA analysis provides a tool to confirm if the cell is genomically normal or rearranged. and if at least two cells share two or more CNAs with common breakpoints within the same patient, they are further recognized as clonal cells, reflecting a proliferating lineage (Fig. 2B, left). This CNA analysis was then performed in all cell groups from 65 sequenced patients and cells with clonal alterations were found in all 4 CK⁺ groups. Nonclonal genomic rearrangements were found in a small portion of cells in all groups, except for the “(CD45/CD31)⁺ only” group (Fig. 2B, right). Among the 701 sequenced cells from the four CK⁺ groups, cells from the “CK⁺ only” group had the highest frequency of clonal cells (304/352, 86.3%). The frequencies of clonal cells in “double positive CK|VIM”, “double positive CK|(CD45/CD31)”, and “triple positive CK|VIM|(CD45/CD31)” groups were 11/49 (22.4%), 20/57 (35.1%), and 2/243 (0.8%), respectively (Fig. 2B, right). These results suggest the presence of intragroup genomic heterogeneity as well as intergroup differences in frequency of clonal alterations.

In addition to IF intensity, we utilized morphologic features to perform the biological categorization of rare cell groups. A trained analyst further curated the detected rare cells from those four CK⁺ groups and identified CTC subtypes from them (Fig. 3A): epithelial-like CTCs (epi.CTCs) were classified as cells that are CK positive and CD45/CD31-negative identified from the “CK⁺ only” group, with distinct appearing nucleus by DAPI morphology as described previously (13). Epi.CTCs expressing VIM were classified as mesenchymal-like CTCs (mes.CTC) from the “double positive CK|VIM” group. In addition, epi.CTCs and mes.CTCs presenting a punctuated pattern in the Alexa Fluor 647 (CD45/CD31) corresponding to CD31 signal from platelets were tracked separately as platelet-coated (pc) epi.CTC (pc.epi.CTC) and mes.CTC (pc.mes.CTC), respectively, from the “double positive CK|(CD45/CD31)” and “triple positive CK|VIM|(CD45/CD31)” groups. These punctuated particles were also observed by the IF staining with the platelet-specific biomarker CD61 in the spiked NBD sample (Supplementary Fig. S1).

Next, we applied the newly identified CTC subtypes into the same group of 79 patients with mCRPC and 11 NBDs (Fig. 3B and C; Supplementary Fig. S2). Overall, CTCs were detected in 77 of 79 (97.5%) patients with median enumeration 6.0 cells per mL in the PB and 231.2 cells per mL in the BMA. Regarding the CTC subtypes, epi.CTC was the dominant subtype, detected in 77 of 79 (97.5%) patients, while a significant reduction of frequency was observed in other three subtypes, 50/79 (63.3%) with mes.CTC, 23/79 (29.1%) with pc.epi.CTC and 11/79 (13.9%) with pc.mes.CTC. As for the difference between PB and BMAs, the presence frequency of mes.CTC (71.7% vs. 26.5%), pc.epi.CTC (35.0% vs. 2.9%) and pc.mes.CTC (18.3% vs. 0.0%) were higher in the BMA than the PB samples. Also, we combined pc.epi.CTC and pc.mes.CTC and found 24 of 79 (30.4%) patients that had at least 1 pc.CTC in the PB and/or BMA. In the PB samples from 11 NBDs, only 2 of them were detected with low numbers of CTCs, one with 1 epi.CTC and the other one with 2 mes.CTCs and none of them had pc.epi.CTC or pc.mes.CTC.

In addition to clonality assessment, single-cell CNA analysis was also utilized to determine copy-number changes for individual gene regions, particularly the three tumor-suppressor genes, PTEN, RB1, and TP53 (Fig. 4A). The molecular signature of AVPC in single CTC (AVPC-MS) is defined as losses of at least two of those three tumor-suppressor genes as previously described (17). CTCs from PB and/or BMA samples were individually sequenced across the 65 patients. A patient qualified as AVPC-MS if at least one CTC matched the AVPC-MS criteria. Of these patients, 25 (38.5%) were AVPC-MS positive (Fig. 4B): 12 of them harbored all three gene losses while the other 13 patients had two gene losses within which the combination of PTEN and RB1 (7/13, 53.8%) was more frequent than the other two combinations. As for individual gene analysis, the RB1 frequency (28/65, 43.1%) was higher than PTEN (22/65, 33.8%) and TP53 (20/65, 30.8%) and a high percentage of homozygous loss was observed in PTEN (12/22, 54.5%) while only a few or none existed in RB1 (4/28, 14.3%) or TP53 (0/20, 0.0%). To further assess the sensitivity in different sample types, we compared 46 PB with 48 BMA samples and did not observe significant differences either for the AVPC-MS detection or single-gene CNA analysis.

Next we analyzed the concordance with previously-published AVPC-C, AVPC-MS in primary tumor by IHC (AVPC-MS-IHC) and AVPC-MS in circulating tumor DNA (ctDNA) by NGS (AVPC-MS-ctDNA; ref. 4) and found that AVPC-MS in CTCs and ctDNA shared the highest concordance (21/33, 63.6%) compared with the concordances with AVPC-C (35/65, 53.8%) and AVPC-MS-IHC (15/24, 62.5%) (Supplementary Table S1). Consistent with previous clinical-correlation analysis of AVPC-C, AVPC-MS-IHC, and AVPC-MS-ctDNA (4), patients with AVPC-MS in CTCs showed more aggressive phenotypes, including significantly higher prostate-specific antigen (PSA), lactate dehydrogenase (LDH) and bone-specific alkaline phosphatase (BAP), compared with negative counterparts (Supplementary Table S1) as well as shorter PFS (4.7 vs. 6.0 months), OS (14.6 vs. 21.2 months), (Supplementary Fig. S3A) and greater treatment response from combination versus cabazitaxel (6.4 vs. 3.0 months), despite absence of statistical significance (Supplementary Fig. S3B and S3C).

To establish the clinical utility of AVPC-MS in CTCs, we examined the value of pc.CTC in combination with the AVPC-MS as prognostic and predictive biomarkers in mCRPC (Fig. 5). Patients with pc.CTC and AVPC-MS–positive status had a median PFS of 1.7 months versus 5.8 months in patients with pc.CTC but AVPC-MS negative (P = 0.08) and 6.0 months in those without pc.CTC (P < 0.01). Patients with pc.CTC and AVPC-MS positive had a median OS of 8.2 months versus 27.3 months (P < 0.01) and 19.9 months (P < 0.001) for the same groups (Fig. 5A). Patients with pc.CTC and AVPC-MS positive had a median PFS of 2.3 months when treated with cabazitaxel versus 1.8 months when treated with the combination [HR = 0.85, 95% confidence interval (CI) = 0.19–3.87, P = 0.84] and a median OS of 10.6 months versus 8.1 months (HR = 1.83, 95%CI = 0.32–10.38, P = 0.49; Fig. 5B). Patients with pc.CTC but AVPC-MS negative had a median PFS of 3.7 months when treated with cabazitaxel versus 7.8 months when treated with the combination (HR = 0.36, 95% CI = 0.11–1.22, P = 0.10) and a median OS of 20.7 months versus 41.3 months (HR = 0.17, 95% CI = 0.03–0.88, P = 0.03; Fig. 5C). Survival analysis of patients without pc.CTCs are shown in (Supplementary Fig. S4).

The purpose of this study was to expand the available predictive biomarkers for directing treatment of mCRPC using an enhanced version of the HDSCA liquid biopsy. Noninvasive methods for patient stratification are critical for molecular characterization in clinical scenarios such as AVPC where the disease subtype emerges, advances, or changes over the course of treatment. The commercialized version of the HDSCA has previously demonstrated clinical utility and been recently approved for reimbursement for the evaluation of patients with advanced prostate cancer as the Oncotype DX AR-V7 Nucleus Detect test (11,12). We have now extended the capabilities of the HDSCA to identify a broader spectrum of rare cells in PB and BMA samples from patients with progressive mCRPC. This version of the assay combines an automated rare cell detection system (OCULAR) that sorts rare cell populations based on different combinations of CK, VIM, and CD45/CD31 into 8 distinct rare cell groups. These groups include various combinations of epithelial, endothelial, and mesenchymal markers that are significantly more abundant in BMA than PB samples and certain groups, such as “CK⁺ only”, are significantly enriched in patients with cancer compared with NBDs. Additional morphologic characterization was then used to subdivide the 8 staining groups and to interpret their clinical and biologic importance. This detailed analysis led to the identification, among others, of platelet-coated cells that are the focus of the clinical results we report here.

The addition of single-cell genome-wide copy-number profiling reveals that the rare-cell groups comprise a mixture of clonally rearranged cells, presumably arising from the tumor lineage, along with cells that lack clonal CNAs. As expected, clonal cells occur with highest frequency in the pure epithelial population (“CK⁺ only” group), but also represent a significant fraction of cells in the other 3 CK⁺ groups (CK with VIM and/or CD45/CD31), potentially indicating progression of cancer cells toward mesenchymal or endothelial-like phenotypic states. Meanwhile, a significant percentage of CK⁺ groups in circulation carrying genomes absent of CNAs suggests additional benefit of utilizing single-cell genomics for diagnostic accuracy. Those rare cells lacking CNAs in both the CK⁺ and CK⁻ rare cell groups will likely require methylomic and transcriptomic approaches to confirm their tissue origins and cell types (24). Finally, a proportion of rare cells show nonclonal CNAs that appear randomly distributed throughout the genome with low frequency (usually 1–5 alterations per genome), and their genetic mechanism and potential biological function are still unclear.

The combination of CD31 with CD45 into the same channel technically solves the problem of 5 markers into the HDSCA 4-channel system and utilizes the imaging features to separate different phenotypes. This enables the observation of the platelet-CTC interaction due to the robust expression of CD31 in the platelets, megakaryocytes, or endothelial cells (25). The further separation between CD31 and CD45 relies on subcellular location (e.g., membrane vs. cytoplasm), combination with other markers (e.g., VIM) and morphologic features such as cell shape. Of particular note for this study were cells in which the CD31 signal appeared to be localized to the perimeter of the cell with punctuated pattern, and which we identify as putative pc.CTCs. This identification, echoed by CD61 IF staining in the spiked NBD sample, requires future multiple platelet-biomarkers confirmation with patient samples using the imaging mass cytometry. Although the role of platelet association with CTCs has not been established clinically (26,27), studies in model systems have suggested multiple roles for platelet-coated CTCs in cancer progression, from stimulating the epithelial–mesenchymal transition (EMT) and extravasation from blood vessels, to protection from immune surveillance as well as physical shear stress and extended survival in circulation (28–33). Several molecular mechanisms were postulated from in vitro and in vivo experiments, including the activity of CD97, an adhesion G protein–coupled receptor, as the adaptor protein to bridge platelet and tumor cells to stimulate bidirectional signaling including ATP-release–induced endothelial disruption and Rho-activation-induced cell migration, eventually leading to cancer metastasis (34). Another study discovered that Hsp47, a chaperone facilitating collagen secretion and deposition, could be genetically amplified and highly expressed in CTCs to enhance the platelet attachment (35). Furthermore, the potential impact of platelets has been inferred mainly from studies on the hematogenous dissemination of CTCs. In this study, the first to observe them in a clinical liquid biopsy, we found that the pc.CTCs were more frequently detected in BMA rather than PB samples, suggesting a potential role in bone metastases of mCRPC. Given these pivotal roles of platelets as a potential prognostic and predictive biomarker, we further evaluated this CTC subtype in combination with the AVPC-MS status and compared it with the AVPC-MS biomarker alone.

Our group has previously reported that the presence of AVPC-C and AVPC-MS in primary tumor and ctDNA is associated with clinically meaningful improvements in both PFS and OS when treated with the carboplatin/cabazitaxel combination therapy (4), and that AVPC-MS in CTCs is associated with poor prognosis (17). We therefore sought to evaluate whether AVPC-MS discerned by CTCs from PB and/or BMA could predict improvements in survival with the combination carboplatin/cabazitaxel relative to cabazitaxel alone. We therefore investigated whether the presence of pc.CTC, in combination with AVPC-MS in CTCs could predict improved performance be associated with improvements in PFS and OS with the carboplatin/cabazitaxel combination therapy relative to cabazitaxel alone. Strikingly, our results show that patients with pc.CTC and AVPC-MS had the shortest OS (8.2 months). Contrary to our previously reported data using AVPC-MS in IHC and/or ctDNA alone that showed improvements in PFS and OS with the combination therapy, we found no benefit in PFS or OS in the patients with this combined signature. Interestingly, we showed that patients with pc.CTC but without AVPC-MS had the longest OS (27.3 months) and present significant improvement in OS when treated with the combination therapy (41.3 months). Regardless of the AVPC-MS status, patients with no detectable pc.CTCs showed no significant differences in survival and a modest improvement of PFS and OS was observed with the combination. Thus, to our knowledge, we are the first to include the presence of pc.CTCs and AVPC-MS in CTCs as a combined predictive and prognostic biomarker for the stratification of patients that could benefit from the addition of carboplatin to cabazitaxel regime or present unfavorable prognosis in advanced prostate cancer. Taken together, for any given patient a potential decision tree hypothesis can be generated: the addition of platinum to taxane would be suggested to the patients with pc.CTC⁺/AVPC-MS⁻, while taxane only would be applied for the pc.CTC⁺/AVPC-MS⁺ and the pc.CTC⁻ groups (Supplementary Fig. S5).

Meanwhile, there are several caveats which concern the validity of results. The original trial had enrolled 169 patients while liquid biopsy samples were only collected from 79 patients for retrospective analysis. Among those patients, 24 of them (30.4%) were detected with pc.CTCs, 25 of 65 (38.5%) sequenced patients were positive for AVPC-MS, and 8 of them (12.3%) were observed with the dual presence of pc.CTCs and AVPC-MS. Thus, considering the small cohort size, low incidence of positivity, and retrospective analysis, the validation of initial observation and its generated hypothesis will, of course, require additional validation and expansion studies. We additionally did further research into the incidence of pc.CTCs and found one study that reported 18 out of 61(29.5%) participants with platelets on CTCs (27), similar to our results, which echoes the necessity of larger cohort studies.

In conclusion, we have utilized a next generation of the HDSCA liquid biopsy for the detection, categorization, and genomic characterization of circulating rare cell populations. We identified the presence of a specific CTC subtype, platelet-coated CTCs, and the tumor suppressor gene molecular signature related to AVPC, as a combined prognostic and predictive biomarker to taxane-platinum combination therapy. The presence of this signature could extend the therapeutic stratification of patients with advanced prostate cancer in addition to previously identified AVPC biomarkers.

S. Chai reports grants from Prostate Cancer Foundation Award; and grants from David and Janet Polak Foundation Fellowship in Convergent Science during the conduct of the study; in addition, S. Chai has a patent for USC Serial No.: 62/914,763 issued to University of Southern California. B. Ormseth reports a patent for US Serial No. 62/914,763 issued to University of Southern California. K. Cunningham reports other support from Epic Sciences outside the submitted work. A. Kolatkar reports personal fees from Epic Sciences outside the submitted work; in addition, A. Kolatkar has a patent for US10613089B2 issued. J. Hicks reports personal fees from Epic Sciences outside the submitted work. P. Kuhn reports grants from Prostate Cancer Foundation Award, Breast Cancer Research Foundation; nonfinancial support from NCI's USC Norris Comprehensive Cancer Center (CORE) Support; other support from Vicky Joseph Research Fund, Vassiliadis Research Fund, Susan Pekarovics during the conduct of the study; other support from Epic Sciences outside the submitted work; in addition, P. Kuhn has a patent for Systems, Methods and Assays for Outlier Clustering Unsupervised Learning Automated Report (OCULAR) issued to US Serial No. 62/914,763. No disclosures were reported by the other authors.

S. Chai: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. N. Matsumoto: Software, methodology, analytical pipeline development. R. Storgard: Data curation, investigation, classification and enumeration. C.-C. Peng: Data curation, investigation. A. Aparicio: Resources, funding acquisition, writing–review and editing, sample acquisition. B. Ormseth: Investigation, methodology, assay development. K. Rappard: Investigation, methodology, assay development. K. Cunningham: Investigation, methodology, assay development. A. Kolatkar: Resources, software, methodology. R. Nevarez: Software, methodology. K.-H. Tu: Investigation, methodology, single-cell genomics. C.-J. Hsu: Investigation, methodology, single-cell genomics. P. Malihi: Conceptualization. P. Corn: Resources. A. Zurita: Resources. J. Hicks: Conceptualization, data curation, supervision, funding acquisition, project administration, writing–review and editing. P. Kuhn: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing. C. Ruiz-Velasco: Conceptualization, data curation, formal analysis, supervision, methodology, writing–original draft, project administration, writing–review and editing.

The authors would like to thank all the patients who participated in this study, Dr. Jeremy Mason, Dr. Stephanie Shishido, Dr. Rishvanth Prabakar, and Nikki Higa for critical reviewing, comments, and statistical advice, Xiomara Villaseñor for scheduling staining and scanning, Libere Ndacayisaba for project discussions, reimaging, and picking, Aidan Plant for enumerating mes.CTCs and reimaging, Annie Amacker for curating NBD reports, Elvia Nuñez and Allison Welsh for project/financial management, Lisa Welter, Nikki Higa, Jiyoun Seo, Liya Xu, Sonia Maryam Setayesh, Drahomír Kolenčík, Carlisle Maney, Dean Tessone, Sean Solomon, Daniel Liu, Eric Yang, Ali Hashemi, Amanda Hmelar, Olivia Hart, Mihir Kumar, Arushi Agrawal, Isabelle Vu, Sachin Narayan, and Ayeshna Desai for reimaging and picking, and BioRender.com with which we created the visual overview. This work is supported fully or partially by the Prostate Cancer Foundation Award 17CHAL01 (A. Aparicio, P. Kuhn, J. Hicks, and S. Chai), Breast Cancer Research Foundation Award BCRF-20-089 (J. Hicks and P. Kuhn), NCI's USC Norris Comprehensive Cancer Center (CORE) Support 5P30CA014089-40 (P. Kuhn), the Solon Scott III Prostate Cancer Research Fund (P. Corn), David and Janet Polak Foundation Fellowship in Convergent Science (S. Chai and P. Malihi), USC-Taiwan Postdoctoral Scholars Program (C.C. Peng), USC Provost Undergraduate Research Fellowship (R. Storgard), Vicky Joseph Research Fund, Vassiliadis Research Fund, and Susan Pekarovics.

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.