Purpose: The transition of prostate adenocarcinoma to a predominantly androgen receptor (AR) signaling independent phenotype can occur in the later stages of the disease and is associated with low AR expression +/− the development of small-cell or neuroendocrine tumor characteristics. As metastatic tumor biopsies are not always feasible and are difficult to repeat, we sought to evaluate noninvasive methods to identify patients transitioning toward a neuroendocrine phenotype (NEPC).

Experimental Design: We prospectively studied a metastatic tumor biopsy, serum biomarkers, and circulating tumor cells (CTC, Epic Sciences) from patients with castration-resistant prostate cancer (CRPC) including those with pure or mixed NEPC histology present on biopsy. CTCs labeled with the patient's clinical status were used to learn features that discriminate NEPC patients, which was then applied to an independent cohort.

Results: Twenty-seven patients with CRPC including 12 NEPC and 5 with atypical clinical features suggestive of NEPC transition were studied. CTCs from NEPC patients demonstrated frequent clusters, low or absent AR expression, lower cytokeratin expression, and smaller morphology relative to typical CRPC. A multivariate analysis of protein and morphologic variables enabled distinguishing CTCs of NEPC from CRPC. This CTC classifier was applied to an independent prospective cohort of 159 metastatic CRPC patients and identified in 17/159 (10.7%) of cases, enriched in patients with high CTC burden (P < 0.01) and visceral metastases (P = 0.04).

Conclusions: CTCs from patients with NEPC have unique morphologic characteristics, which were also identified in a subset of CRPC patients with aggressive clinical features potentially undergoing NEPC transition. Clin Cancer Res; 22(6); 1510–9. ©2015 AACR.

Translational Relevance

NEPC is an aggressive variant of prostate cancer that most commonly arises in later stages of prostate cancer as a mechanism of treatment resistance. Diagnosis of NEPC typically relies on metastatic tumor biopsy as serum markers are unreliable. We show that the CTC populations from patients with NEPC characterized with the Epic platform demonstrate a unique morphology, lower AR expression, and lower cytokeratin expression compared with CTCs from other patients with castration-resistant prostate cancer (CRPC). The presence of NEPC-like CTC subpopulations occurs in approximately 10% of unselected patients with CRPC and is associated with aggressive clinical features. CTCs may provide utility for early diagnosis of NEPC-associated acquired resistance and warrants larger clinical studies.

Neuroendocrine prostate cancer (NEPC) is an aggressive, androgen receptor (AR) independent subtype of prostate cancer that most commonly becomes manifest in the later stages of castration-resistant prostate cancer (CRPC) and is associated with treatment resistance (1–5). The diagnosis of NEPC remains challenging and currently relies on a combination of pathologic and clinical features suggestive of AR signaling independence. Before NEPC develops, metastatic tumor biopsies often show mixed features with both adenocarcinoma and neuroendocrine carcinoma cells present. There are no reliable serum markers to consistently diagnose patients transforming to the NEPC phenotype and the incidence of circulating tumor cells (CTC) in these patients is unknown. Detection of NEPC has clinical implications, as NEPC patients would not be expected to respond well to currently approved AR-targeted therapies for CRPC and may be better served by therapies specifically directed to NEPC.

CTCs provide the potential for noninvasive, real-time molecular characterization of cancer in patients with metastatic disease. To date, the only FDA-cleared test for CTC detection and enumeration is the CellSearch technology, based on immunomagnetic enrichment of CTCs expressing the epithelial cell adhesion molecule (EpCAM). Several other platforms have recently been developed to improve sensitivity of CTC detection, most of which include enrichment and/or other physical selection methods (6, 7). There is mounting evidence that nontraditional populations of CTCs also exist, including EpCAM/cytokeratin (CK)-negative CTCs (8) and/or cells smaller in size than traditional CTCs, some even smaller than neighboring white blood cells (6, 9). The Epic Sciences platform is a non–selection-based platform that characterizes all nucleated cells and identifies CTCs based on a multiparametric digital pathology process identifying abnormal cells among the normal white blood cells utilizing protein expression and cell morphology (10–12). This technique has demonstrated the ability to identify distinct CTC populations including traditional (CK+, CD45), apoptotic, CK-negative, and CTC clusters (12, 13). We aimed to characterize CTCs from patients with CRPC and NEPC utilizing the Epic platform and correlate results with patient-matched tumor biopsy and clinical features.

CTC collection

Under Institutional Review Board approved protocols at the Weill Cornell Medicine and Memorial Sloan Kettering Cancer Center, patients with metastatic CRPC including those with pure or mixed NEPC were prospectively enrolled. NEPC was defined by the presence of either a pure or mixed small-cell high-grade neuroendocrine carcinoma histology in a metastatic tumor biopsy and confirmed by at least 20% positive immunohistochemical (IHC) staining for the neuroendocrine marker chromogranin and/or synaptophysin. CRPC was defined clinically, with or without a metastatic biopsy confirming prostate adenocarcinoma. CRPC patients were subclassified as atypical CRPC if a biopsy showed adenocarcinoma and the patient had clinical features suggestive of an AR-independent transition, which included radiographic progression in the setting of a low serum prostate specific antigen (PSA) <1 ng/mL, visceral progression in the absence of PSA progression [defined by the Prostate Cancer Working Group 2 criteria (14)] and/or elevated serum chromogranin A > 3× upper limit of normal.

Clinical demographics including prior therapies, sites of metastases, PSA, serum neuroendocrine marker levels, and CTC number (CellSearch) were collected. Blood (10 mL) from each subject was shipped to Epic Sciences within 48 hours and processed immediately on arrival. Red blood cells were lysed, approximately 3 million nucleated blood cells dispensed onto 10 to 16 glass slides as previously described (10–13) and stored at −80°C.

CTC identification

Two slides from each patient were evaluated by immunofluorescence (IF) (refs. 12, 13, 15; Fig. 1A) using antibodies targeting cytokeratins (CK), CD45, AR, and 4′,6-diamidino-2-phenylindole (DAPI) counterstain. Slides were imaged using a platform that captures all 3 million cells per slide in less than 15 minutes, and analyzed by a proprietary software that characterizes each cell by parameters including cell size, shape, nuclear area, presence of macronucleoli, CK and AR expression, uniformity, and cellular localization. CTC candidates were identified in an interactive report, reviewed by trained technicians. CK+/CD45 cells with intact, DAPI+ nuclei exhibiting tumor-associated morphologies were classified as traditional CTCs. CTCs with nontraditional characteristics were recorded, such as CK/CD45 cells with morphologic distinction and/or AR positivity, CK+/CD45 small cells, CTC clusters, CTCs with multiple marconucleoli, and apoptotic CTCs (with nuclear or cytoplasmic fragmentation).

Figure 1.

A, Epic platform workflow starting from a single blood tube to the identification and characterization of all nucleated cells. Steps include (1) blood lysed, nucleated cells from blood sample placed onto slides; (2) slides stored in −80°C biorepository; (3) slides stained with CK, CD45, DAPI, and AR; (4) slides scanned; (5) multiparametric digital pathology algorithms run; (6) software and human reader confirmation of CTCs and quantitation of biomarker expression; B, observed CK expression, and C, AR expression of each CTC subtype: traditional CTC (blue), clustered CTCs (green), apoptotic CTCs (red), and CK− CTCs (purple) from each patient sample organized by their clinical diagnosis from biopsy.

Figure 1.

A, Epic platform workflow starting from a single blood tube to the identification and characterization of all nucleated cells. Steps include (1) blood lysed, nucleated cells from blood sample placed onto slides; (2) slides stored in −80°C biorepository; (3) slides stained with CK, CD45, DAPI, and AR; (4) slides scanned; (5) multiparametric digital pathology algorithms run; (6) software and human reader confirmation of CTCs and quantitation of biomarker expression; B, observed CK expression, and C, AR expression of each CTC subtype: traditional CTC (blue), clustered CTCs (green), apoptotic CTCs (red), and CK− CTCs (purple) from each patient sample organized by their clinical diagnosis from biopsy.

Close modal

Pathologic evaluation

Patient-matched metastatic tumor biopsies were reviewed by two anatomic genitourinary pathologists and classified as adenocarcinoma or NEPC based on presence of either pure or mixed small-cell high-grade neuroendocrine carcinoma histology in a metastatic tumor biopsy and confirmed by at least 20% positive IHC staining for the neuroendocrine marker chromogranin and/or synaptophysin (16). IHC was quantified on scale of 0 to 3 and positive IHC was defined as any staining intensity seen in target cells above background. To assess AURKA amplification, we used a locus-specific probe plus reference probe fluorescence in situ hybridization (FISH) assay, as previously described (17).

Statistical analysis

CTC morphologic/molecular data and clinical information were compiled into patient datasets (NEPC, CRPC, and atypical CRPC) using KNIME, where cytokeratin expression, AR expression, presence of clusters, and various nuclear and cytoplasmic morphologic features were analyzed with single cell resolution (Supplementary Table S1). Kernel density estimates (KDE) of each CTC characteristic were performed to provide univariate distributions across each aggregate subtype. Patient samples were analyzed for frequency of cell types at calculated cell counts per milliliter of blood, and univariate distributions of CTC biomarkers were compared at the patient level for each diagnostic category. Supervised learning was performed using the Random Forest classifier algorithm (R package “randomForest”) built with 1,001 decision trees and configured to provide a probability output (18).

Leave-one-out cross-validation

To evaluate the robustness of the Random Forest classifier, leave-one-out cross-validation was performed; CTCs from patients with atypical CRPC were removed from analysis, and CTCs from NEPC were labeled NEPC+ and CRPC were labeled NEPC. Leave-one-out cross-validation at the blood sample level with the dataset partitioned into training and test sets is shown in Supplementary Figs. S1–S3. For each blood tube, CTCs from every other sample were used to train a classifier, and CTCs from the blood tube being evaluated were held out as a test set. CTCs from the test set were analyzed by the trained classifier, where the output is an estimated probability of class membership to NEPC+ and NEPC for each CTC belonging to the held-out sample. This cycle was repeated iteratively for each sample, and the classifier output was collected at the end of each iteration. The criteria for patient-level class membership was established as at least three CTCs with a p(NEPC) score greater than 0.95.

Atypical CRPC and contemporary cohort analysis

A classifier was first trained on NEPC and CRPC samples, without atypical CRPC samples. This classifier was then used to classify the atypical CRPC sample CTCs, as well as CTCs from a 159 patient validation cohort. In the validation cohort, the same criteria for patient positivity [at least three CTCs with p(NEPC) greater than 0.95] was applied to generate patient-level predictions from the classifier's single-cell output. KDE curves were used to plot the distribution of NEPC+ class membership values for individual CTCs for each patient.

CTCs from 27 patients with metastatic prostate cancer were evaluated. The patients identified either pathologically as NEPC (n = 12) or clinically as atypical CRPC (n = 5) as defined above demonstrated a higher frequency of liver metastases and lower PSA compared with other CRPC patients (Table 1, Supplementary Table S2). Overall, bone metastases were present in 24/27 (88.9%) of patients, and liver metastases were present in 8/12 (66.7%) of NEPC and 5/15 (33.3%) of CRPC of whom 4 had atypical clinical features (Supplementary Table S3). Median serum PSA level was 1.9 ng/mL in NEPC, 2.8 ng/mL in atypical CRPC, and 53.4 ng/mL in other CRPC patients. Serum neuroendocrine marker levels varied considerably within the NEPC subgroup and were also elevated in cases of CRPC (Supplementary Table 2).

Table 1.

Clinical data including diagnosis, sites of metastases, prior therapies, biopsy site, pathology, and IHC results

Metastatic biopsy
Sites of metastasesImmunohistochemistry
Patient #DiagnosisLILULNBOADPLPPINumber of therapies for CRPCPrior therapiesBiopsy sitePathologySYPCgAAR
NEPC     ADT Lymph node Neuroendocrine carcinoma NA 
NEPC    ADT, carboplatin-etoposide Liver, TURP Poorly differentiated carcinoma w/neuroendocrine differentiation − − 
NEPC      ADT Prostate High grade prostate cancer with extensive neuroendocrine differentiation NA 
NEPC      ADT, cisplatin-etoposide Lymph node Neuroendocrine carcinoma NA 
NEPC      ADT, abiraterone Lymph node High grade prostate cancer with neuroendocrine differentiation NA 
10 NEPC     ADT, cisplatin-etoposide Bone Small-cell carcinoma − 
12 NEPC    ADT, abiraterone Liver Small-cell carcinoma − − 
13 NEPC      ADT, cisplatin-docetaxel Brain Small-cell carcinoma − 
14 NEPC        Prostate Prostate carcinoma with overlap features of small-cell carcinoma − 
24 NEPC     ADT, carboplatin-etoposide Liver Small-cell carcinoma NA 
16 NEPC     Lymph node Small-cell carcinoma − − 
26 NEPC      ADT, cisplatin-etoposide, topotecan Liver Small-cell carcinoma NA NA NA 
atypical CRPC     ADT TURP Poorly differentiated adenocarcinoma − − 
atypical CRPC      ADT, docetaxel, abiraterone Lymph node Poorly differentiated adenocarcinoma − − 
atypical CRPC      ADT, docetaxel, enzalutamide, cabazitaxel Liver Poorly differentiated Adenocarcinoma − − 
atypical CRPC     ADT, cisplatin-etoposide, abiraterone Pleura Poorly differentiated Adenocarcinoma − − NA 
11 atypical CRPC     ADT, docetaxel, abiraterone Liver Poorly differentiated adenocarcinoma − − 
17 CRPC     ADT, docetaxel, abiraterone Liver Poorly differentiated adenocarcinoma − − 
18 CRPC       ADT, enzalutamide, radium-223, J591 Bone Poorly differentiated adenocarcinoma NA NA NA 
19 CRPC      ADT, sipuleucel-T, abiraterone Not performed NA NA NA NA 
20 CRPC      ADT, sipuleucel-T Not performed NA NA NA NA 
21 CRPC      ADT, sipuleucel-T, J591, docetaxel, abiraterone Not performed NA NA NA NA 
22 CRPC      ADT, abiraterone, enzalutamide, J591, docetaxel, carboplatin-paclitaxel Not performed NA NA NA NA 
23 CRPC      ADT, J591, docetaxel-lenalidamide, ipilumimab, sipieucel-T, abiraterone, cabozantinib Bone Poorly differentiated adenocarcinoma NA NA NA 
25 CRPC      ADT, abiraterone TURP Poorly differentiated adenocarcinoma NA − NA 
27 CRPC       ADT, abiraterone Not performed NA NA NA NA 
28 CRPC     ADT, docetaxel, abiraterone+GDC0068/placebo Pleura Poorly differentiated adenocarcinoma − − NA 
Metastatic biopsy
Sites of metastasesImmunohistochemistry
Patient #DiagnosisLILULNBOADPLPPINumber of therapies for CRPCPrior therapiesBiopsy sitePathologySYPCgAAR
NEPC     ADT Lymph node Neuroendocrine carcinoma NA 
NEPC    ADT, carboplatin-etoposide Liver, TURP Poorly differentiated carcinoma w/neuroendocrine differentiation − − 
NEPC      ADT Prostate High grade prostate cancer with extensive neuroendocrine differentiation NA 
NEPC      ADT, cisplatin-etoposide Lymph node Neuroendocrine carcinoma NA 
NEPC      ADT, abiraterone Lymph node High grade prostate cancer with neuroendocrine differentiation NA 
10 NEPC     ADT, cisplatin-etoposide Bone Small-cell carcinoma − 
12 NEPC    ADT, abiraterone Liver Small-cell carcinoma − − 
13 NEPC      ADT, cisplatin-docetaxel Brain Small-cell carcinoma − 
14 NEPC        Prostate Prostate carcinoma with overlap features of small-cell carcinoma − 
24 NEPC     ADT, carboplatin-etoposide Liver Small-cell carcinoma NA 
16 NEPC     Lymph node Small-cell carcinoma − − 
26 NEPC      ADT, cisplatin-etoposide, topotecan Liver Small-cell carcinoma NA NA NA 
atypical CRPC     ADT TURP Poorly differentiated adenocarcinoma − − 
atypical CRPC      ADT, docetaxel, abiraterone Lymph node Poorly differentiated adenocarcinoma − − 
atypical CRPC      ADT, docetaxel, enzalutamide, cabazitaxel Liver Poorly differentiated Adenocarcinoma − − 
atypical CRPC     ADT, cisplatin-etoposide, abiraterone Pleura Poorly differentiated Adenocarcinoma − − NA 
11 atypical CRPC     ADT, docetaxel, abiraterone Liver Poorly differentiated adenocarcinoma − − 
17 CRPC     ADT, docetaxel, abiraterone Liver Poorly differentiated adenocarcinoma − − 
18 CRPC       ADT, enzalutamide, radium-223, J591 Bone Poorly differentiated adenocarcinoma NA NA NA 
19 CRPC      ADT, sipuleucel-T, abiraterone Not performed NA NA NA NA 
20 CRPC      ADT, sipuleucel-T Not performed NA NA NA NA 
21 CRPC      ADT, sipuleucel-T, J591, docetaxel, abiraterone Not performed NA NA NA NA 
22 CRPC      ADT, abiraterone, enzalutamide, J591, docetaxel, carboplatin-paclitaxel Not performed NA NA NA NA 
23 CRPC      ADT, J591, docetaxel-lenalidamide, ipilumimab, sipieucel-T, abiraterone, cabozantinib Bone Poorly differentiated adenocarcinoma NA NA NA 
25 CRPC      ADT, abiraterone TURP Poorly differentiated adenocarcinoma NA − NA 
27 CRPC       ADT, abiraterone Not performed NA NA NA NA 
28 CRPC     ADT, docetaxel, abiraterone+GDC0068/placebo Pleura Poorly differentiated adenocarcinoma − − NA 

Abbreviations: AD, adrenal; ADT, androgen deprivation therapy; AR, androgen receptor; BO, bone; CgA, chromogranin A; LI, liver; LN, lymph node; LU, lung; N/A, not performed; PL, pleural; PPI, pelvic peritoneal implants; SYP, synaptophysin; TURP, transurethral resection of prostate.

CTCs in NEPC versus CRPC

Enumeration of CTCs using both the CellSearch and Epic platforms was performed. Of note, 6/13 evaluated NEPC and atypical CRPC patients had CellSearch CTC count of less than 5 CTC/7.5 mL (range 0–384, with 5 of these 13 patients having a CellSearch CTC count of 0). In contrast, all 17 NEPC and atypical CRPC patients had CTCs greater than or equal to 5 CTC/7.5 mL using the Epic platform. Further characterization of the detected CTCs revealed heterogeneity of CK and AR expression in both NEPC and CRPC, with a significantly greater proportion of CK-negative and AR-negative CTCs in NEPC compared with CRPC (Figs. 1 and 2; Supplementary Table S4). CTCs in NEPC patients overall had lower AR expression, higher cytoplasmic circularity, and higher nuclear to cytoplasmic ratio. The prevalence of CK-negative CTC subpopulations in NEPC patients is potentially consistent with epithelial–mesenchymal transition (EMT; refs. 19, 20).

Figure 2.

A, Kernel density estimate (KDE) curves for cytokeratin expression (left) and androgen receptor expression (right) for CTCs aggregated from all NEPC (red), CRPC (blue), and atypical CRPC (green) patient samples. B, Representative CTC images from patients with NEPC and CRPC, shown with the classifier output, which is the estimated probability of the CTC's class membership as NEPC+.

Figure 2.

A, Kernel density estimate (KDE) curves for cytokeratin expression (left) and androgen receptor expression (right) for CTCs aggregated from all NEPC (red), CRPC (blue), and atypical CRPC (green) patient samples. B, Representative CTC images from patients with NEPC and CRPC, shown with the classifier output, which is the estimated probability of the CTC's class membership as NEPC+.

Close modal

Within the NEPC subgroup, there was a greater proportion of small-cell CTCs in patients with metastatic biopsy confirming small-cell carcinoma highlighting phenotypic similarities between tumor and CTCs. CTCs were tested by IF for the presence of the neuroendocrine marker CD56 (Fig. 3A). Of those samples with matched with metastatic biopsy showing neuroendocrine features and detectable CTCs, 7/12 (58%) had at least one CD56+ CTC whereas 0/8 (0%) non-neuroendocrine samples with detectable CTCs had ≥1 CD56+ CTC. Of patient samples with small-cell carcinoma pathology on tumor biopsy, 5/7 (71%) had ≥1 CD56+ CTC. A confusion matrix demonstrates high specificity for small-cell NEPC patients, demonstrating concordance to tumor tissue (Supplementary Table S5). Additional molecular characterization of these CTCs using FISH for AURKA, a gene commonly amplified in NEPC (21), showed concordance with matched metastatic biopsies in selected cases (an example is highlighted in Supplementary Fig. S4) but was not present in all cases or all cells in positive cases.

Figure 3.

A, representative images of CTCs from CRPC and NEPC patients evaluated by IF for CD56 expression. All CTCs evaluated from CRPC patients were CD56 negative. Heterogeneous expression of CD56 was observed within and among NEPC patient samples. CTCs including CK+/CD56 (a–d, and f), CK/CD56+ (e), CK+/CD56+ (g and h), small CK+/CD56+ (g), and CK+/CD56+ clusters (i) were identified in patient samples. a, patient 21, CD56 negative; b, patient 23, small CD56 negative; c, patient 23, CD56 negative; d, patient 28, CD56 negative, NEPC; e, patient 24, small CK− CD56 positive; f, patient 12, CD56 negative; g, patient 12, CD56 positive; h, patient 10, CD56 positive; i, patient 26, cluster CD56 positive. B, Kernel density estimate (KDE) curves are shown for CTC morphologic features ranging from nuclear/cytoplasmic ratio (top left), nuclear area (top center), maximum cell area (right top), nuclear circularity (bottom left), cytoplasmic circularity (bottom center), and nuclear entropy (bottom right) for CTCs aggregated from all NEPC (red), CRPC (blue), and atypical CRPC (green) patient samples.

Figure 3.

A, representative images of CTCs from CRPC and NEPC patients evaluated by IF for CD56 expression. All CTCs evaluated from CRPC patients were CD56 negative. Heterogeneous expression of CD56 was observed within and among NEPC patient samples. CTCs including CK+/CD56 (a–d, and f), CK/CD56+ (e), CK+/CD56+ (g and h), small CK+/CD56+ (g), and CK+/CD56+ clusters (i) were identified in patient samples. a, patient 21, CD56 negative; b, patient 23, small CD56 negative; c, patient 23, CD56 negative; d, patient 28, CD56 negative, NEPC; e, patient 24, small CK− CD56 positive; f, patient 12, CD56 negative; g, patient 12, CD56 positive; h, patient 10, CD56 positive; i, patient 26, cluster CD56 positive. B, Kernel density estimate (KDE) curves are shown for CTC morphologic features ranging from nuclear/cytoplasmic ratio (top left), nuclear area (top center), maximum cell area (right top), nuclear circularity (bottom left), cytoplasmic circularity (bottom center), and nuclear entropy (bottom right) for CTCs aggregated from all NEPC (red), CRPC (blue), and atypical CRPC (green) patient samples.

Close modal

Based on the observed differences in CTCs between groups, we sought to identify CTC characteristics specific to NEPC. Cell-level features were utilized to train Random Forest cell-level classifiers for both leave-one-out cross-validation and for the classification of CTCs in the test cohort, shown in Supplementary Table S1. KDE analysis of the patient groups' CTCs in aggregate revealed significant differences in CK, AR, and morphologic characteristics between patients with NEPC and CRPC (Figs. 2 and 3B).

Identification of NEPC CTCs

To demonstrate the diagnostic potential of CTC characteristics in distinguishing NEPC, the observed differences between NEPC and CRPC were used to train a Random Forest classifier. Results from leave-one-out cross-validation of NEPC and CRPC samples are shown in Supplementary Figs. S5 and S6, where the output from the classifier is a p(NEPC+) value and a p(NEPC) value for each CTC, corresponding to the estimated probability of the cell's class membership as NEPC+ and NEPC.

From the density curve in Supplementary Fig. S4A, the samples from patients with NEPC demonstrated a spike in the curves near the high end of the p(NEPC+) spectrum, with many curves peaking near a p(NEPC+) score of 95%. In Supplementary Fig. S4B, the number of CTCs/mL with a p(NEPC+)*** score greater than or equal to 95% is presented as a bar chart for each patient sample, where each column is colored by the actual clinical diagnosis that the classifier is trying to predict.

Obtaining positive signals at the CTC level from samples that the classifier does not encounter during training demonstrates the classifier's ability to detect NEPC from CRPC in a robust manner that mitigates the risk of overfitting. These conditions simulate the environment that the classifier would face in practice, in the sense that any future blood sample sent in for NEPC analysis is presented to the algorithm as a series of CTCs that it has not encountered during training, and the classifier then estimates the probability of class membership for each CTC from the new sample.

Atypical CRPC

The clinical significance of patients with castration-resistant adenocarcinoma that develop progressive disease in the setting of low serum PSA <1 ng/mL, visceral metastases in the absence of PSA progression, or elevated serum chromogranin is not well established. One hypothesis is that these tumors are less androgen responsive and may be in transition toward an AR low or NEPC phenotype and/or demonstrate intratumoral heterogeneity with both adenocarcinoma and NEPC present within or between metastases. We applied the NEPC classification model trained to distinguish NEPC versus CRPC CTCs to the five atypical CRPC patients and found that atypical CRPC is associated with an increase in heterogeneity of CRPC cells and a higher burden of NEPC-like cells compared with CRPC patients (Supplementary Fig. S4, Supplementary Table S6).

Patient case studies

Atypical CRPC patient 6, for example, harbored CTCs of various morphologies with a predominance of NEPC+ CTCs (Fig. 4). Patient 6 is a 64-year-old man who presented with metastatic hormone naive prostate cancer, developed clinical progression within 6 months on primary hormonal therapy, had short duration of response to subsequent abiraterone, radium-223, and docetaxel, and developed progressive bone metastases and new liver metastases in the setting of a non-rising PSA. Despite his bone biopsy at progression showing adenocarcinoma without neuroendocrine features (Fig. 4A), his clinical history and CTC characteristics obtained at the time of bone biopsy supported AR independence.

Figure 4.

Metastatic biopsy showing morphologic characteristics and IHC for synaptophysin (SYP) and androgen receptor (AR) of metastatic biopsies from patients 6 (A) and 12 (B). Atypical CRPC patient 6 tumor was characterized as poorly differentiated adenocarcinoma, synaptophysin negative, AR positive; NEPC patient 12 tumor was characterized as small-cell carcinoma, synaptophysin positive, AR negative. C, Patient 6 distribution based on NEPC probability class memberships. D, Patient 12 distribution based on NEPC probability class membership.

Figure 4.

Metastatic biopsy showing morphologic characteristics and IHC for synaptophysin (SYP) and androgen receptor (AR) of metastatic biopsies from patients 6 (A) and 12 (B). Atypical CRPC patient 6 tumor was characterized as poorly differentiated adenocarcinoma, synaptophysin negative, AR positive; NEPC patient 12 tumor was characterized as small-cell carcinoma, synaptophysin positive, AR negative. C, Patient 6 distribution based on NEPC probability class memberships. D, Patient 12 distribution based on NEPC probability class membership.

Close modal

Another example of how CTCs can be used to understand disease heterogeneity is illustrated in the case of patient 12, a 68-year-old gentleman with CRPC who had a bone biopsy at the time of castration resistance for research, which showed prostate adenocarcinoma. He was treated with abiraterone and prednisone. Despite PSA stability, follow-up imaging at 3 months on abiraterone revealed new liver and lung metastases and his serum chromogranin was markedly elevated at 17,340 ng/mL (ULN 95 ng/mL). Liver biopsy was consistent with NEPC (small-cell carcinoma; Fig. 4B). Similar to patient 6, CTCs at time of liver biopsy showed heterogeneous CTC populations including both NEPC and CRPC cell characteristics, suggesting intrapatient heterogeneity. These cases support CTCs as potentially useful in capturing tumor heterogeneity that might not be assessed on metastatic biopsy.

Validation cohort

We evaluated baseline CTCs from 159 CRPC patients prospectively enrolled in an independent patient cohort at MSKCC for the presence of NEPC+ CTCs (Fig. 5A). NEPC+ CTC subpopulations were identified in 17 of 159 (10.7%) cases. A significantly higher proportion of CRPC patients with visceral metastases harbored NEPC+ CTCs compared with those that were NEPC (35% versus 15%, respectively; P = 0.04). Patients with NEPC+ CTCs also had an overall higher CTC burden (median CTC count 64.6 versus 4.2; P < 0.01). To address whether the CTC classifier was a reflection of an overall higher CTC count, linearity was assessed with a Pearson coefficient showing a weak relationship between frequency of NEPC CTCs and total cell count (Supplementary Fig. S6). Representative images of NEPC+ CTC characteristics observed in the validation cohort are shown in Fig. 5B.

Figure 5.

A, clinical data of validation cohort (n = 159) including sites of metastases, age, serum PSA, and CTC count. NS, not statistically significant at 0.05 level. P values from two-sided tests comparing NEPC+ to NEPC are based on the Fisher exact test for categorical parameters and Wilcoxon rank sum for continuous parameters. B, representative images showing CTC characteristics from patients in validation cohort that are classifier positive.

Figure 5.

A, clinical data of validation cohort (n = 159) including sites of metastases, age, serum PSA, and CTC count. NS, not statistically significant at 0.05 level. P values from two-sided tests comparing NEPC+ to NEPC are based on the Fisher exact test for categorical parameters and Wilcoxon rank sum for continuous parameters. B, representative images showing CTC characteristics from patients in validation cohort that are classifier positive.

Close modal

Histologic and molecular subtyping of cancer often influences clinical decision making, and tissue confirmation is typically required at cancer diagnosis before treatment recommendations are offered. Prostate cancer is the most common cancer in men in the United States and Europe (22, 23), and in nearly all cases diagnostic biopsies reveal adenocarcinoma upon initial diagnosis. Prostate adenocarcinomas are characterized by AR expression and activation, and therefore hormonal therapies targeting the AR are the mainstay of systemic therapy (24). Small-cell neuroendocrine carcinoma of the prostate is a rare histologic subtype at diagnosis, representing less than 1% of all new prostate cancer diagnoses (25). However, in a subset of patients with metastatic prostate adenocarcinoma treated with AR-targeted therapies, prostate adenocarcinomas can develop histologic transformation toward a predominantly neuroendocrine carcinoma likely as a mechanism of acquired resistance (1–5). The NEPC phenotype is associated with aggressive disease, frequent visceral metastases, and low or absent AR expression on metastatic tumor biopsy (4). In this setting, patients are often offered platinum-based chemotherapy with regimens similar to small-cell neuroendocrine carcinoma of the lung (26, 27). Therefore, identification of advanced prostate cancer patients that have acquired NEPC has potential clinical implications.

However, the diagnosis of NEPC can be complex as there is a spectrum of morphologies seen in advanced prostate cancer with AR-positive adenocarcinoma and AR-negative small-cell carcinoma representing the extreme phenotypes. Metastatic biopsies often reveal mixed features of both adenocarcinoma and neuroendocrine carcinoma, with variable AR or neuroendocrine marker protein expression (16). The clinical significance of mixed tumors is less clear and treatment decisions are often individualized based on a combination of pathologic and clinical features. Furthermore, for patients with an atypical clinical presentation such as rapid radiographic progression in setting of a low or modestly elevated PSA, platinum-based therapies are sometimes considered even in the absence of neuroendocrine morphology on biopsy. Another challenge in the diagnosis of NEPC is that metastatic biopsies are not always feasible for patients suffering from advanced prostate cancer, may carry additional risks for the patient including complications from biopsy procedure or delay of initiation of appropriate systemic therapy, and does not always capture disease heterogeneity. Therefore, a noninvasive marker to detect NEPC progression and simultaneously capture intrapatient heterogeneity is an unmet need.

We found that CTCs from metastatic prostate cancer patients are often phenotypically heterogeneous. CTCs from patients with pathologically confirmed NEPC were predominantly of smaller size compared with other CRPC patients and demonstrated lower AR expression and abnormal nuclear and cytoplasmic features. There was also a higher prevalence of low cytokeratin expressing CTCs in NEPC, possibly related to EMT changes that can occur during metastatic transit and treatment resistance (28, 29). When applied to an independent cohort, we found that up to 10% of CRPC patients also harbored similar NEPC+ CTC subpopulations and their presence was associated with aggressive clinical features (i.e., visceral metastases and high CTC burden). These data support the possible detection of circulating neuroendocrine cancer cells in patients with metastatic CRPC; however, the lack of definitive markers and mixed cellular subpopulations observed reinforce the biologic and clinical complexity underlying disease progression and NEPC transformation. Differences in nuclear size may also be contributed by the presence of visceral metastases as has been recently observed by Chen and colleagues (30). What remains unclear are the dynamics by which these CTCs arise and how the classifier performs as a predictive or prognostic biomarker. Serial monitoring of CTCs in larger cohorts could help elucidate how consistently the classifier emerges during the course of therapy and during the CRPC-to-NEPC transition. Future studies including single-cell sequencing of CTCs will also be important to molecularly characterize these heterogeneous populations and may improve our understanding of this complex resistance phenotype.

In this proof of principle study, we demonstrate that CTCs from patients with NEPC have distinct characteristics. The results presented here indicate the feasibility of analyzing CTCs using the Epic platform and support the development of further studies to validate the clinical utility of CTCs for the early detection of patients transforming toward NEPC and the prognostic and potential predictive impact of CTC characteristics in predicting response to AR-directed therapies in CRPC.

H.I. Scher is a consultant/advisory board member for BIND Therapeutics, Exelexis, and Janssen, and reports receiving commercial research grants from BIND Therapeutics, Epic Sciences, Exelexis, and Janssen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H. Beltran, D. Marrinucci, R. Dittamore, H.I. Scher

Development of methodology: A. Jendrisak, M. Landers, D. Marrinucci, R. Dittamore, H.I. Scher

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Beltran, M. Landers, J. Louw, R. Krupa, D.M. Nanus, S.T. Tagawa, R. Dittamore

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Beltran, A. Jendrisak, M. Landers, J.M. Mosquera, M. Kossai, R. Dittamore, H.I. Scher

Writing, review, and/or revision of the manuscript: H. Beltran, A. Jendrisak, M. Landers, J.M. Mosquera, J. Louw, R. Krupa, R.P. Graf, D.M. Nanus, S.T. Tagawa, D. Marrinucci, R. Dittamore, H.I. Scher

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Beltran, A. Jendrisak, J.M. Mosquera, J. Louw, N. Schreiber

Study supervision: H. Beltran, R. Dittamore

Department of Defense W81XWH-13-1-0275 (to H. Beltran), and Damon Runyon-Gordon Family Clinical Investigator Award CI-67-13 (to H. Beltran).

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.

1.
Aggarwal
R
,
Zhang
T
,
Small
EJ
,
Armstrong
AJ
. 
Neuroendocrine prostate cancer: subtypes, biology, and clinical outcomes
.
J Natl Compr Canc Netw
2014
;
12
:
719
26
.
2.
Tagawa
ST
. 
Neuroendocrine prostate cancer after hormonal therapy: knowing is half the battle
.
J Clin Oncol
2014
;
32
:
3360
4
.
3.
Beltran
H
,
Tagawa
ST
,
Park
K
,
MacDonald
T
,
Milowsky
MI
,
Mosquera
JM
, et al
Challenges in recognizing treatment-related neuroendocrine prostate cancer
.
J Clin Oncol
2012
;
30
:
e386
9
.
4.
Beltran
H
,
Tomlins
S
,
Aparicio
A
,
Arora
V
,
Rickman
D
,
Ayala
G
, et al
Aggressive variants of castration-resistant prostate cancer
.
Clin Cancer Res
2014
;
20
:
2846
50
.
5.
Aparicio
A
,
Tzelepi
V
. 
Neuroendocrine (Small-Cell) carcinomas: why they teach us essential lessons about prostate cancer
.
Oncology (Williston Park)
2014
;
28
:
831
8
.
6.
Ozkumur
E
,
Shah
AM
,
Ciciliano
JC
,
Emmink
BL
,
Miyamoto
DT
,
Brachtel
E
, et al
Inertial focusing for tumor antigen-dependent and -independent sorting of rare circulating tumor cells
.
Sci Transl Med
2013
;
5
:
179ra47
.
7.
Yap
TA
,
Lorente
D
,
Omlin
A
,
Olmos
D
,
de Bono
JS
. 
Circulating tumor cells: a multifunctional biomarker
.
Clin Cancer Res
2014
;
20
:
2553
68
.
8.
Yu
M
,
Bardia
A
,
Wittner
BS
,
Stott
SL
,
Smas
ME
,
Ting
DT
, et al
Circulating breast tumor cells exhibit dynamic changes in epithelial and mesenchymal composition
.
Science
2013
;
339
:
580
4
.
9.
Phillips
KG
,
Kuhn
P
,
McCarty
OJ
. 
Physical biology in cancer. 2. The physical biology of circulating tumor cells
.
Am J Physiol Cell Physiol
2014
;
306
:
C80
8
.
10.
Hsieh
HB
,
Marrinucci
D
,
Bethel
K
,
Curry
DN
,
Humphrey
M
,
Krivacic
RT
, et al
High speed detection of circulating tumor cells
.
Biosens Bioelectron
2006
;
21
:
1893
9
.
11.
Marrinucci
D
,
Bethel
K
,
Bruce
RH
,
Curry
DN
,
Hsieh
B
,
Humphrey
M
, et al
Case study of the morphologic variation of circulating tumor cells
.
Hum Pathol
2007
;
38
:
514
9
.
12.
Marrinucci
D
,
Bethel
K
,
Kolatkar
A
,
Luttgen
MS
,
Malchiodi
M
,
Baehring
F
, et al
Fluid biopsy in patients with metastatic prostate, pancreatic and breast cancers
.
Phys Biol
2012
;
9
:
016003
.
13.
Werner
SL
,
Graf
RP
,
Landers
ML
,
Valenta
DT
,
Schroeder
M
,
Greene
SB
, et al
Analytical validation and capabilities of the epic CTC platform: enrichment-free circulating tumour cell detection and characterization
.
J Circulating Biomarkers
2015
;
4
:
1
13
.
14.
Scher
HI
,
Halabi
S
,
Tannock
I
,
Morris
M
,
Sternberg
CN
,
Carducci
MA
, et al
Design and end points of clinical trials for patients with progressive prostate cancer and castrate levels of testosterone: recommendations of the Prostate Cancer Clinical Trials Working Group
.
J Clin Oncol
2008
;
26
:
1148
59
.
15.
Marrinucci
D
,
Bethel
K
,
Lazar
D
,
Fisher
J
,
Huynh
E
,
Clark
P
, et al
Cytomorphology of circulating colorectal tumor cells: a small case series
.
J Oncol
2010
;
2010
:
861341
.
16.
Epstein
JI
,
Amin
MB
,
Beltran
H
,
Lotan
TL
,
Mosquera
JM
,
Reuter
VE
, et al
Proposed morphologic classification of prostate cancer with neuroendocrine differentiation
.
Am J Surg Pathol
2014
;
38
:
756
67
.
17.
Mosquera
JM
,
Beltran
H
,
Park
K
,
MacDonald
TY
,
Robinson
BD
,
Tagawa
ST
, et al
Concurrent AURKA and MYCN gene amplifications are harbingers of lethal treatment-related neuroendocrine prostate cancer
.
Neoplasia
2013
;
15
:
1
10
.
18.
Breiman
L
. 
Random forests
.
Mach Learn
2001
;
45
:
5
32
.
19.
Tam
WL
,
Weinberg
RA
. 
The epigenetics of epithelial-mesenchymal plasticity in cancer
.
Nat Med
2013
;
19
:
1438
49
.
20.
Scheel
C
,
Weinberg
RA
. 
Cancer stem cells and epithelial-mesenchymal transition: concepts and molecular links
.
Semin Cancer Biol
2012
;
22
:
396
403
.
21.
Beltran
H
,
Rickman
DS
,
Park
K
,
Chae
SS
,
Sboner
A
,
MacDonald
TY
, et al
Molecular characterization of neuroendocrine prostate cancer and identification of new drug targets
.
Cancer Discov
2011
;
1
:
487
95
.
22.
Siegel
RL
,
Miller
KD
,
Jemal
A
. 
Cancer statistics, 2015
.
CA Cancer J Clin
2015
;
65
:
5
29
.
23.
Malvezzi
M
,
Bertuccio
P
,
Levi
F
,
La Vecchia
C
,
Negri
E
. 
European cancer mortality predictions for the year 2014
.
Ann Oncol
2014
;
25
:
1650
6
.
24.
Cooper
CS
,
Eeles
R
,
Wedge
DC
,
Van Loo
P
,
Gundem
G
,
Alexandrov
LB
, et al
Corrigendum: analysis of the genetic phylogeny of multifocal prostate cancer identifies multiple independent clonal expansions in neoplastic and morphologically normal prostate tissue
.
Nat Genet
2015
;
47
:
689
.
25.
Wang
W
,
Epstein
JI
. 
Small cell carcinoma of the prostate. A morphologic and immunohistochemical study of 95 cases
.
Am J Surg Pathol
2008
;
32
:
65
71
.
26.
Papandreou
CN
,
Daliani
DD
,
Thall
PF
,
Tu
SM
,
Wang
X
,
Reyes
A
, et al
Results of a phase II study with doxorubicin, etoposide, and cisplatin in patients with fully characterized small-cell carcinoma of the prostate
.
J Clin Oncol
2002
;
20
:
3072
80
.
27.
Aparicio
AM
,
Harzstark
AL
,
Corn
PG
,
Wen
S
,
Araujo
JC
,
Tu
SM
, et al
Platinum-based chemotherapy for variant castrate-resistant prostate cancer
.
Clin Cancer Res
2013
;
19
:
3621
30
.
28.
Armstrong
AJ
,
Marengo
MS
,
Oltean
S
,
Kemeny
G
,
Bitting
RL
,
Turnbull
JD
, et al
Circulating tumor cells from patients with advanced prostate and breast cancer display both epithelial and mesenchymal markers
.
Mol Cancer Res
2011
;
9
:
997
1007
.
29.
Tanaka
H
,
Kono
E
,
Tran
CP
,
Miyazaki
H
,
Yamashiro
J
,
Shimomura
T
, et al
Monoclonal antibody targeting of N-cadherin inhibits prostate cancer growth, metastasis and castration resistance
.
Nat Med
2010
;
16
:
1414
20
.
30.
Chen
J
,
Ho
H
,
Lichterman
J
,
Lu
Y
,
Zhang
Y
,
Garcia
MA
, et al
Subclassification of prostate cancer circulating tumor cells by nuclear size reveals very small nuclear circulating tumor cells in patients with visceral metastases
.
Cancer
2015
;
121
:
3240
51
.

Supplementary data