High-purity CTC RNA sequencing identifies prostate cancer lineage phenotypes prognostic for clinical outcomes

Abstract

The development of treatment resistance remains universal for patients with metastatic prostate cancer, driven by AR alterations and lineage state transitions. Identifying the evolution of lineage transitions in treatment resistance has been limited by the challenges of collecting serial tissue biopsies on treatment, which can be overcome using blood-based liquid biopsies. Utilizing a novel circulating tumor cell (CTC) isolation approach, we collected 273 CTC samples from 117 patients with metastatic prostate cancer for RNA sequencing. 146 samples from 70 patients had tumor purity comparable to tissue biopsies. We identified four CTC transcriptional phenotypes, mirroring lineage states identified in tissue. Patients with a luminal-B-like CTC phenotype defined by persistent AR signaling and high proliferation, as well as those with a neuroendocrine CTC phenotype, had significantly shorter survival than patients with luminal-A-like and low proliferation phenotypes. In a prospective substudy, pre-treatment CTC luminal-B-like phenotype was associated with early progression on 177Lu-PSMA-617.

Figure 1.

RNA-seq of high-purity CTCs isolated from patients with metastatic prostate cancer reveals four transcriptional phenotypes. A, Tumor/CTC purity was assessed using the ESTIMATE algorithm in prostate cancer cell line samples (n = 12, green), CTC samples processed with standard-stringency (n = 41, blue) or novel high-stringency (n = 211, orange) method; healthy donor blood samples processed analogous to novel high-stringency CTC samples (n = 12, pink) and a published dataset of mCRPC tissue biopsies separated by biopsy site (n = 629, gray). ESTIMATE score is plotted for each sample. B, ESTIMATE tumor fraction (left) and CIBERSORTx immune content (right) scores for standard-stringency (n = 41, blue) or high-stringency (n = 211, orange) CTC methods are plotted and compared with prostate cancer cell lines (n = 12, green). C, Assessment of the number of samples meeting high-purity threshold for standard-stringency vs. high-stringency methods. D, Consensus k-means clustering of single-sample pathway z-scores of 146 high-purity CTC samples from 70 patients with metastatic prostate cancer identifies four distinct clusters. CRPC, castration-resistant prostate cancer.

Figure 2.

Gene expression of epithelial and prostate adenocarcinoma–associated genes across CTC phenotypes. A, Schematic of the CTC transcriptional phenotypes. B and C, Epithelial keratin (KRT8 and KRT18) gene expression across phenotypes; for patients with multiple CTC samples, the highest purity sample is included (Low_CTC, n = 30; Low_prolif, n = 12; LumA, n = 24; LumB, n = 31; and NE, n = 3). D,KLK3 (PSA), (E) FOLH1 (PSMA), and (F) MKI67 (Ki-67) gene expression across phenotypes; for patients with multiple CTC samples, the highest purity sample is included (Low_CTC, n = 30; Low_prolif, n = 12; LumA, n = 24; LumB, n = 31; and NE, n = 3).

Figure 3.

Luminal-B CTC phenotype is associated with poor prognosis and adverse clinical features. A, Serum PSA (ng/mL) at the time of CTC collection in the survival analysis subset (Low_CTC, n = 63; Low_prolif, n = 12; LumA, n = 21; LumB, n = 18; and NE, n = 3). B, Proportion of samples in each CTC transcriptional phenotype for each disease category in the survival analysis subset (mCSPC, n = 20; mCRPC, n = 93; and NEPC, n = 4) is shown. C, Kaplan–Meier plot of the OS from the time of the first CTC collection for each patient (n = 117) by CTC transcriptional phenotype at the first CTC collection. Median survival, and HR and log-rank P value relative to low CTC burden (Low_CTC) phenotype are shown in the inset. Risk table is shown below. D, Multivariate analysis of the CTC phenotype clustersgroups demonstrates that LumB remains a poor prognostic factor after adjusting for adverse clinical features. E, Distribution of CTC transcriptional phenotypes by metastatic sites involved at the time of the first CTC collection including lymph node only (LN only, n = 7), bone ± lymph node (bone ± LN, n = 64), at least one non-liver soft-tissue site (non-liver visceral, n = 14), and at least one liver site (liver, n = 20).

Figure 4.

Classification of mCRPC tissue biopsy samples and low-purity CTC samples into the high-purity CTC phenotypes recapitulates survival differences. A, Classification of mCRPC tissue biopsy samples (n = 203) with a k-nearest neighbor (KNN) classifier trained on the high-purity CTC sample phenotypes recapitulates the differences in survival outcome seen between the four CTC phenotypes. B, Low CTC burden (Low_CTC) samples with at least minimal epithelial gene expression defined as an expression sum of four epithelial markers (EpCAM, KRT8, KRT18, and KRT19) of >10 transcripts per million (n = 30) were reclassified using the same KNN classifier as in A. Survival differences between the reclassified Low_CTC samples recapitulates the differences in survival outcome seen between the four CTC phenotypes in the high-purity CTC samples.

Figure 5.

Luminal-B CTC phenotype and persistent PSMA expression are associated with poor response to ¹⁷⁷Lu–PSMA-617. A, Clinical benefit rate (the best radiographic response of stable disease, partial response, or complete response) for ¹⁷⁷Lu–PSMA-617 in patients with favorable pretreatment CTC phenotypes (Low_CTC, LP, and LumA) vs. pretreatment LumB phenotype. B, Proportion of pretreatment LumB vs. favorable CTC phenotypes for patients with and without ¹⁷⁷Lu–PSMA-617 early progression (disease progression within the first three cycles/18 weeks of treatment). C, Baseline CTC FOLH1 (PSMA) gene expression for LumB (n = 8) vs. favorable CTC phenotypes (n = 29). Kaplan–Meier plot for (D) radiographic progression-free survival (rPFS) and (E) OS illustrates decreased rPFS and OS after starting ¹⁷⁷Lu–PSMA-617 for patients with the pretreatment LumB CTC phenotype. F, Longitudinal evaluation of consensus AR/luminal and neuroendocrine pathway scores during ¹⁷⁷Lu–PSMA-617 treatment. Patient 818 (left) represents an example of a patient with early progression on ¹⁷⁷Lu–PSMA-617 who demonstrated a rapid increase in NE score and decrease in AR score on treatment. Patient 978 (right) illustrates a patient with a longer duration of response who also shows an increased NE score and decreased AR score at the time of disease progression.

Figure 6.

Cell-surface target expression is variable across CTC phenotypes. A, Heatmap of gene expression of cell-surface targets ordered by clustering in Fig. 1D. B–J, Cell-surface target expression by CTC phenotype for targets associated with prostate adenocarcinoma (KLK2, STEAP1, and STEAP2), pan-cancer cell-surface targets expressed in prostate cancer including TACSTD2 (TROP2), ERBB2 (HER2), and tumor immune checkpoint cell-surface protein CD276 (B7H3), as well as targets associated with prostate neuroendocrine differentiation (DLL3, CEACAM5, and SSTR2). For patients with multiple CTC samples, the highest purity sample is included (Low_CTC, n = 30; Low_prolif, n = 12; LumA, n = 24; LumB, n = 31; and NE, n = 3). Low_CTC samples are included for reference but not included in statistical comparisons because they are expected to have low/no expression due to low CTC content.