BCL-2 is selectively and consistently induced by castration in patient PCa and xenograft models
We first examined a transcriptomic dataset71 (supplementary Table 1) for BCL-2 mRNA levels in 3 epithelial cell (CD45-EpCAM+) populations of the normal human prostate, i.e., basal (B; CD49fhiCD38lo), luminal (L; CD49floCD26+CD38hi) and luminal progenitor (LP; CD49floCD26+CD38lo) cells. We found that BCL-2 mRNA was expressed at high levels in LP and basal cells but had lower expression in mature luminal cells (supplementary Fig. 1a). We then examined BCL-2 levels in 422 untreated primary tumors in TCGA-PRAD, which displayed an increasing trend that correlated with tumor grade (i.e., combined Gleason Score, GS) as supported by Jonckhere-Terpstra (J-T) trend test (Fig. 1a). We subsequently investigated the mRNA levels of all 5 prosurvival BCL-2 family members, i.e., BCL-2, BCL-xL, MCL-1, BCL-W and A1/BFL-1, in 4 treated patient datasets including 3 datasets of PCa patients subjected to neoadjuvant ADT (nADT; short-term (~ 2–6 months) ADT prior to prostatectomy)72,73,74 and 1 dataset of mCRPC patients who failed long-term ADT/Enza.23 As shown in Fig. 1b, BCL-2 was the only member that was commonly and consistently induced in the 3 nADT datasets and showed a trend of upregulation in Enza-resistant PCa. Finally, we interrogated a recent scRNA-seq dataset75 consisting of FACS-purified human benign prostate epithelial cell subpopulations, 2 primary PCa (Pri-PCa), and 3 CRPC including CRPC1 and CRPC NE in the prostate and one mCRPC (Fig. 1c). We re-mapped this dataset76 and used AMACR to separate PCa cells from benign epithelial cells, which clustered the 24,142 high-quality cells to 9 major cell populations including 4 benign and 5 PCa cell populations (Fig. 1c, left; supplementary Fig. 1b, c). Interestingly, the 2 Pri-PCa was characterized by high levels of KLK3 and ERG but lack of KRT5 and KRT14 expression whereas the 3 CRPC samples all lacked KRT5, KRT14 and ERG expression and 2 of the 3 CRPC (i.e., CRPC1 and mCRPC) showed high KLK3 expression (supplementary Fig. 1d–g). Importantly, BCL-2 mRNA was enriched in proximal luminal (stem), basal and castration-resistant epithelial cells and was expressed at higher levels in 3 CRPC compared to 2 Pri-PCa (Fig. 1c). In contrast, BFL-1 was barely detected and BCL-xL, BCL-W and MCL1 showed similar levels in CRPC and Pri-PCa (supplementary Fig. 1h).
Fig. 1The alternative text for this image may have been generated using AI.
Selective upregulation of BCL-2 by castration (ADT) and antiandrogens. a BCL-2 mRNA levels are increased in high-grade PCa and correlate with tumor grade. Data were extracted from TCGA_PRAD consisting of PCa with increasing grade (GS, Gleason Score; n indicated below). *p < 0.05 (Student’s t-test). J-T trend test showed that BCL-2 mRNA levels correlated with increasing tumor grade (*p < 0.05). b BCL-2 mRNA (among 5 members) was upregulated in patients’ PCa treated with nADT (the first 3 datasets) and showed a trend of upregulation in the Alumkal pre-/post-Enza cohort (i.e., dataset 4). Note that A1/BFL-1 was not detected in datasets 3 and 4. c BCL-2 mRNA is elevated in CRPC compared to Pri-PCa in a scRNA-seq dataset.75 Shown on the left are the 9 major cell populations we recently re-mapped out of the total of 24,142 cells.76 Shown on the right is the BCL-2 mRNA expression in the 9 cell populations. Note the higher BCL-2 levels in the 3 CRPC (CRPC1, mCRPC and CRPC NE) samples than in the 2 Pri-PCa (Pri-PCa1 and Pri-PCa2). d, e BCL-2 family mRNA (d) and protein (e) levels in AD/AI (CRPC) xenograft models. d BCL-2 mRNA is commonly increased in 6 CRPC datasets (n indicated in parentheses). Datasets 1 and 2 were comparisons of LNCaP clones with intact AR vs. AR knockout (ARKO) either cultured in vitro (dataset 1) or propagated in castrated NSG mice (dataset 2). Datasets 3–6 were from our earlier publication (PMID and GSE# indicated). e AD and AI tumors serially passaged (p; indicated in parentheses) were used in WB analysis of the proteins indicated. 20 μg protein was loaded for each lane and actin was used as loading control. f Quantification of the normalized protein levels of AR, ARv7, PSA, FKBP5, GR, and the 5 BCL-2 members (BCL-2, BCL-xL, MCL-1, BCL-W, and BFL-1/A1) in our 4 paired AD/AI xenograft models. Each circle (data point) represents the target protein band from an independent WB experiment. Protein abundance was normalized to housekeeping proteins (GAPDH and/or actin) to enable direct comparison of absolute expression levels across models and conditions. Group differences were evaluated using unpaired two-tailed Student’s t-tests, with statistical parameters reported according to GraphPad Prism (*p < 0.05; **p < 0.01; ***p < 0.001)
Our laboratory has developed 4 castration-resistant or androgen-independent (AI; also called primary (1°) CRPC) xenograft models, i.e., LNCaP, LAPC9, VCaP, and LAPC4, by serially passaging the parent androgen-dependent (AD) tumors in castrated mice20 (supplementary Fig. 2a). As in our previous findings,20 IHC characterization of the AR using the AR 441 monoclonal antibody (supplementary Table 2), which was raised against the aa 299-315 in the N-ter domain of AR and recognizes the full-length AR as well as all C-ter truncated AR variants (ARvs),20 revealed the LNCaP-AI tumors to be (nuclear) AR+/hi and LAPC9-AI tumors AR-/lo whereas LAPC4-AI and VCaP-AI tumors largely ARcyto (with ~10% AR in the nucleus;20 supplementary Fig. 2b). These AI models, when compared to the corresponding AD tumors, showed model-specific changes in AR, ARv7, AR signaling activity, and glucocorticoid receptor (GR) (supplementary Figs. 2c–h, 3, see below).20 Consistent with our early studies,20 the VCaP-AI was the only model that expressed ARv7 (supplementary Figs. 2d, f, g; 3b). We systematically analyzed the mRNA and protein levels of BCL-2 family members in these AD/AI tumors and related models (Fig. 1d–f; supplementary Figs. 2c–h, 3). We observed that BCL-2 mRNA was commonly and exclusively upregulated in castration-resistant (primary CRPC), castration/Enza-resistant (2° CRPC) and AR-knockout (ARKO) LNCaP models as well as in AR-/lo LAPC9 CRPC20 (Fig. 1d). At the protein level, BCL-2 protein was upregulated in 3 of the 4 AI models except VCaP while the other 4 prosurvival BCL-2 family proteins showed model-dependent changes (Fig. 1e ; supplementary Figs. 2c–h, 3), as supported by quantitative analysis of western blotting (WB) results from 10 independent experiments with multiple AD/AI tumors for each model (n = 2–15 tumors) (Fig. 1f). Briefly, BCL-xL was upregulated in VCaP-AI (the only model that expressed ARv7), MCL-1 was downregulated in LAPC9-AI and LNCaP-AI, and BCL-W was reduced in LNCaP-AI whereas BFL1 did not change in any of the 4 AI models (Fig. 1e, f; supplementary Fig. 3).
In summary, these data from both PCa models and patient specimens indicate that castration (ADT and/or ADT/Enza) commonly and selectively upregulates BCL-2.
Treatment-naive primary prostate tumors are populated mostly by AR+BCL-2- PCa cells
The above observations suggest a reciprocal relationship between AR signaling and BCL-2 expression. To confirm and extend these findings, we employed Vectra-based quantitative multiplex immunofluorescence (qmIF) to assess PCa cells expressing BCL-2 and/or AR in regular FFPE (formalin-fixed and paraffin-embedded) sections as well as TMA (tissue microarray) and whole-mount (WM) sections from benign prostate (n = 123), treatment-naïve primary PCa (n = 125) and treatment-failed CRPC (n = 25) (Fig. 2; supplementary Figs. 4–8, supplementary Table 3). We utilized cytokeratin (CK) staining to demarcate the epithelial compartment. Consistent with earlier reports27,38,41,44 and mRNA data (Fig. 1c; supplementary Fig. 1a), BCL-2 protein was expressed in the basal cell layer while AR was detected in luminal layer in benign glands (Fig. 2a; supplementary Figs. 4c, 5a), and BCL-2+ cells and AR+ cells displayed a strong inverse correlation in benign prostate (R = −0.74, p = 0.00159; Fig. 2e, top). In untreated Pri-PCa, there was a striking expansion of AR+ and AR+BCL-2- cell populations compared to benign tissues (Fig. 2b, f; supplementary Figs. 4d, 5b, 8c) and the AR+ and BCL-2+ PCa cells still exhibited an inverse correlation (R = −0.626, p = 0.0014; Fig. 2e, middle). Consistent with decreases in BCL-2+ PCa cells, the BCL-2 mRNA levels were reduced in Pri-PCa compared to benign tissues (supplementary Fig. 8a, b).
Fig. 2The alternative text for this image may have been generated using AI.
qmIF analysis reveals predominant AR+BCL-2- cells in untreated primary PCa and markedly increased (AR+ or AR-) BCL-2+ cells in CRPC. a In benign prostatic glands (HPCa14N), BCL-2+ cells are mainly in the basal cell layer whereas AR+ cells in luminal layer. Cytokeratin (CK) staining was used to mark epithelial compartment. Note both AR and BCL-2 were also expressed in stromal cells (scale bar, 50 μm). Magnified images of individual stains from the boxed region in the whole-mount (WM) image (top) were presented below (scale bar, 20 μm for all 4 lower panels). b Primary PCa is characterized by significantly increased AR+BCL-2- cells. Shown above are two WM images of HPCa31T (scale bars, 800 μm) and down below zoom-in images of individual or merged markers (scale bars, 80 μm). c, d Increased cellular heterogeneity and markedly expanded (AR+ or AR-) BCL-2+ cell population in CRPC. c WM low-magnification image showing AR/BCL-2 expression (scale bar, 300 μm). d Zoom-in images of the boxed area in c showing individual markers (scale bar, 50 μm). e Relationship between (CK+) AR-expressing and/or BCL-2-expressing cells in benign tissue (top), primary tumor (middle) and CRPC (bottom). Regression line, and Pearson R and P values are indicated. f Box plots summarizing the relative % of 4 PCa cell subtypes in WM images analyzed in benign tissues, primary tumors and CRPC. Each dot in the box plots represents a CK+ ROI. P values were determined by repeated measures two-way ANOVA with Bonferroni multiple comparison test
Cell subtype analysis revealed that among all CK+ ROIs (Region of Interest, 1 mm2) in benign prostate, the AR-BCL-2+ population represented the majority followed by AR+BCL-2- cells, with low representation of AR+BCL-2+ (double positive) and AR-BCL-2- (double-negative) cells (Fig. 2f; supplementary Fig. 8d–f). In Pri-PCa, the AR+BCL-2- cells became the predominant cell population followed by AR-BCL-2- cells with low representation of the double-positive and AR-BCL-2+ cell subtypes (Fig. 2f; supplementary Fig. 8d–f).
Castration drives PCa cell heterogeneity and upregulates BCL-2+ PCa cells in patient CRPC
We analyzed a total of 25 CRPC including 20 in TMA and 5 WM sections (supplementary Fig. 4b) and found that the reciprocal relationship between BCL-2+ and AR+ PCa cells observed in Pri-PCa was lost in CRPC (Fig. 2e). Relative to Pri-PCa, CRPC showed reduced fraction of AR+ but significantly increased fraction of BCL-2+ PCa cells (supplementary Fig. 8c). Notably, we observed increased PCa cell heterogeneity: AR-BCL-2+ cells, typical of the LAPC9-AI cells (Fig. 1e, f; supplementary Fig. 2b–d), constituted ~50% of the 4 cell-type (AR-BCL-2+, AR+BCL-2-, AR-BCL-2- and AR+BCL-2+) subpopulations amongst (CK+) CRPC cells (Fig. 2d, f; supplementary Figs. 5c–d, 6c, 7, 8d–f). We also observed significant increases in AR+BCL-2+ PCa cells in CRPC (Fig. 2d, f; supplementary Figs. 6b, 8d–f), which were characteristic of LNCaP 1o/2o CRPC cells (Fig. 1e, f). Aggregating cell density (cell count/mm2) in WM sections (supplementary Fig. 8e) or in all specimens (supplementary Fig. 8f) and compared to Pri-PCa, we observed that the AR+BCL-2- PCa cells represented the majority in primary tumors whereas in CRPC the AR-BCL-2+ cells became the predominant cell population accompanied by increased AR+BCL-2+ and AR-BCL-2- cells (Fig. 2f; supplementary Fig. 8d). Of interest, in patient CRPC, while the majority of AR+ PCa cells showed nuclear AR, we also observed increased LAPC4-AI-like ARcytoBCL-2+ (supplementary Fig. 7b, d) and VCaP-AI-like ARcytoBCL-2- (supplementary Fig. 7d, e) cells.
Castration-induced dynamic changes in AR+/-BCL-2+/- PCa cell subpopulations in xenograft tumors recapitulate the changes observed in patient CRPC
Next, we employed both Vectra-based qmIF and Imaging Mass Cytometry (IMC) platforms to quantitatively assess the dynamics of AR+/-BCL-2+/- cells in the 4 PCa AD/AI xenograft models developed in our lab (Fig. 3; supplementary Figs. 9–12), which showed model-related alterations in AR, ARv7, AR activity (PSA and FKBP5), and GR (Fig. 1e–f; supplementary Figs. 2 and 3). Re-analysis of individual AI tumors under therapy20 also revealed distinct Enza responses in 1o CRPC associated with the AR heterogeneity: AR+/hi LNCaP-AI responded to Enza for ~4 weeks (supplementary Fig. 9a) and AR-/lo LAPC9-AI were de novo refractory to Enza (supplementary Fig. 9b) whereas ARcyto LAPC4-AI and VCaP-AI responded to Enza with protracted latencies prior to emergence of castration/Enza-resistant 2° CRPC (supplementary Fig. 9c, d).
Fig. 3The alternative text for this image may have been generated using AI.
Vectra and IMC analysis reveals castration-induced dynamic changes in AR+/-BCL-2+/- cell types in 4 xenograft CRPC models. a–d tSNE plots of IMC-derived single-cell imaging data. Shown are the individual and merged tSNE plots from LNCaP-AD/AI (a), LAPC9-AD/AI (b), LAPC4-AD/AI (c), and VCaP-AD/AI (d) xenografts. Columns represent merged AD and AI cell clusters (first column), individual AD and AI cell populations and contours (second and third columns), and AR and BCL-2 expression overlaid on tSNE maps (fourth and fifth columns). Blue and red contours indicate AD and AI cell subpopulations, respectively. e LAPC4-AD tumors are populated mostly by AR+BCL-2- PCa cells. Shown on top are qmIF WM images (scale bar, 800 μm) and at the bottom zoom-in images (scale bar, 80 μm for all panels). f LAPC4-AI (1° CRPC) tumors are populated by ARcytoBCL-2+ PCa cells. Shown on top are qmIF WM images (scale bar, 800 μm) and at the bottom zoom-in images illustrating cytoplasmic AR+ LAPC4-AI cells with upregulated BCL-2 (scale bar, 80 μm for all panels). g IMC images of AR and BCL-2 in LAPC4-AD and LAPC4-AI tumors (scale bar, 200 μm) with representative zoom-in images of AR and BCL-2 shown below (scale bar, 100 μm for all panels)
In the LNCaP-AD/AI system (Fig. 3a; supplementary Fig. 10), qmIF analysis of WM LNCaP-AD tumors revealed that both AR- and AR+ cells expressed little BCL-2 (supplementary Fig. 10b). Most cells in LNCaP-AI (1° CRPC) tumors turned AR+/hi with slightly increased BCL-2 (supplementary Fig. 10c). Sensitive IMC-generated single-cell profiling with tSNE-based cell clustering corroborated AR and BCL-2 upregulation in LNCaP-AI cells (Fig. 3a; supplementary Fig. 10e). Strikingly, most PCa cells in castration/Enza-resistant 2° LNCaP-CRPC were AR+/hiBCL-2+/hi (supplementary Fig. 10d).
The LAPC9-AD tumors resembled LNCaP-AD tumors in containing both AR-BCL-2- and AR+BCL-2- cells (supplementary Fig. 11a). In contrast, LAPC9-AI tumors, unlike LNCaP-AI tumors, apparently evolved into AR-BCL-2+/hi (supplementary Fig. 11b). IMC and tSNE-anchored cell subpopulation analyses also revealed reduced AR expression concomitant with elevated BCL-2, resulting in substantially increased AR-BCL-2+ cells in LAPC9-AI tumors (Fig. 3b; supplementary Fig. 11c).
Most LAPC4-AD tumor cells exhibited an AR+BCL-2- phenotype on qmIF analysis (Fig. 3e) and castration reprogrammed LAPC4-AI cells to the ARcytoBCL-2+ phenotype (Fig. 3f). Single-cell analysis by IMC confirmed the qmIF-described ARcytoBCL-2+ phenotype in LAPC4-AI tumors (Fig. 3c, g). Finally, the VCaP AD/AI model was an exception, where most VCaP-AD tumor cells were AR+BCL-2+ while VCaP-AI tumor cells became ARcytoBCL-2-/lo (supplementary Fig. 12a, b). IMC analysis validated the AR+BCL-2+ phenotype of VCaP-AD cells and demonstrated cytoplasmic AR redistribution and reduced BCL-2 in VCaP-AI tumors (Fig. 3d; supplementary Fig. 12c).
In summary, 3 of the 4 AD xenograft models (LNCaP, LAPC9 and LAPC4) recapitulated the major AR+BCL-2- cellular phenotype observed in human Pri-PCa whereas VCaP-AD tumors, intriguingly, displayed an AR+BCL-2+ phenotype. Consistent with the diverse AR+/-BCL-2+/- cell subtypes observed in patient CRPC, the 4 xenograft CRPC models manifested variegated cellular subpopulations: the LNCaP (1° and 2°) CRPC predominantly AR+BCL-2+, LAPC9-CRPC AR-BCL-2+, LAPC4-CRPC ARcytoBCL-2+ and VCaP-CRPC predominantly ARcytoBCL-2- phenotype, respectively.
Developing castration/Enza-resistant LAPC4 culture models to recapitulate castration-induced ARcytoBCL-2+ phenotype
Our previous analysis of 195 CRPC specimens revealed that as much as 40% of the CRPC had the mixed cytoplasmic/nuclear AR (i.e., largely ARcyto phenotype with ~10–15% nuclear AR).20 qmIF studies herein also identified prominent ARcyto CRPC cells, which often expressed high levels of BCL-2 and presented an ARcytoBCL-2+ phenotype (e.g., supplementary Fig. 7b, d), much like the ARcytoBCL-2+ LAPC4-CRPC in vivo (Fig. 3f, g). To further study the ARcytoBCL-2+ CRPC, we established the castration-resistant and dual castration- and Enza-resistant LAPC4 cell models (Fig. 4; supplementary Fig. 13). To this end, we cultured regular LAPC4-AD cells long-term (4 months) in charcoal dextran stripped serum (CDSS) medium to develop castration-resistant LAPC4 (LAPC4-CR or LAPC4-AI) cells, which were subsequently exposed to either 20 μM or 100 μM Enza-containing CDSS medium for 1 month, resulting in the LAPC4-Enza(20)-R or LAPC4-Enza(100)-R models (supplementary Fig. 13a). These CDSS and CDSS/Enza selected LAPC4 sublines displayed more elongated and mesenchymal morphology (supplementary Fig. 13b) and proliferated more slowly (supplementary Fig. 13c) than LAPC4-AD cells. Moreover, although LAPC4-AD cells were inhibited by both 20 μM and 100 μM of Enza (Fig. 4a), the LAPC4-CR and LAPC4-Enza(20)-R cells were sensitive only to 100 μM of Enza (Fig. 4b, c). LAPC4-Enza(100)-R cells were resistant to both 20 μM and 100 μM of Enza (Fig. 4d). WB analysis revealed a progressive redistribution of AR from nucleus to cytoplasm and increase in BCL-2 expression (Fig. 4e, f). Of interest, we observed a bi-phasic change in total AR protein levels in CDSS-cultured LAPC4 cells, which plateaued at 96 h then decreased afterwards (Fig. 4e, f, the WCL panels, left). RT-qPCR analysis revealed a time-dependent increase of AR mRNA levels within the 48–96 h of culturing LAPC4 cells in CDSS (Fig. 4g), suggesting that the early AR protein increase in CDSS-cultured LAPC4 cells (i.e., up to 96 h) was driven by castration-induced AR mRNA expression. However, castrated LAPC4 cells maintained steady high levels of AR mRNA levels from 96 h up to 3 weeks and only showed significant reduction of AR expression at 4 weeks (Fig. 4g), suggesting that the decreases in AR protein levels at ≥1 week in LAPC4-CDSS cells (Fig. 4e, f, left WCL panels) were due to post-transcriptional mechanisms.
Fig. 4The alternative text for this image may have been generated using AI.
Enzalutamide resistance is associated with AR cytoplasmic relocalization and BCL-2 upregulation in the LAPC4 model. a–d Cell proliferation assays showing relative cell numbers over time in LAPC4-AD (a), LAPC4-CR (b), and two Enza-resistant sublines—LAPC4-Enza(20)-R (c) and LAPC4-Enza(100)-R (d). Shown are the mean ± SEM (p < 0.001, two-way ANOVA). e. Immunoblot analysis of LAPC4-AD cells cultured in CDSS over time shows dynamic changes in AR and BCL-2 expression using whole-cell lysates (WCL), and cytosolic and nuclear fractions. Lamin-B1 and actin serve as loading controls. f Densitometric quantification of immunoblots in (e) showing fold change (FC) of AR and BCL-2 relative to loading controls over the time course in WCL (left) and cytosolic (middle) and nuclear fractions (right). g Biphasic changes of AR mRNA levels in LAPC4 cells cultured in CDSS. *p < 0.05 when comparing 48 h and 96 h vs. 0 h; #p < 0.05 when comparing 4w vs. 3w. h IF staining of LAPC4-AD, LAPC4-CR, and Enza-resistant sublines (20 μM and 100 μM) showing AR (red), BCL-2 (yellow), mitochondria (green), and nuclei (DAPI, blue). Images reveal increased cytoplasmic AR and mitochondrial BCL-2 in resistant sublines. Scale bars, 20 μm. i Quantification of AR and BCL-2 expression patterns from the immunofluorescence data in (h), categorized into four groups: AR⁺BCL-2⁻, ARcytoBCL-2⁺, AR⁻BCL-2⁺, and AR⁻BCL-2⁻
Regardless, despite increased AR at both mRNA and protein levels within 48–96 h, the CDSS cultured LAPC4 cells showed immediate (i.e., within 48–96 h) AR protein redistribution from the nucleus to the cytosol (Fig. 4e, f). Multiplex IF imaging analysis also showed a shift from nuclear AR+ and low BCL-2 in LAPC4-AD to an ARcytoBCL-2+/hi phenotype in resistant sublines, especially in LAPC4-Enza(100)-R cells in which BCL-2 co-localized with the Mito-tracker (Fig. 4h; supplementary Fig. 13d). The LAPC4-Enza(20)-R cells displayed an intermediate phenotype (supplementary Fig. 13d), suggesting an ongoing transition. Quantification of cell subtypes demonstrated that in LAPC4-AD cells, the AR+BCL-2- phenotype predominated while the resistant sublines showed a prominent expansion in the ARcytoBCL-2+ subpopulation (Fig. 4i).
BCL-2 mRNA levels consistently anti-correlated with the AR activity
The preceding qmIF and IMC studies of BCL-2 and AR protein expression at the single-cell levels in patient CRPC and xenograft models revealed a reciprocal relationship between the two, suggesting that AR may transcriptionally represses BCL-2. In support, the mRNA levels of BCL-2, but not other BCL-2 family members, were consistently correlated, inversely, with the canonical AR signaling activity or AR-A74 in multiple clinical datasets of PCa patients who received nADT or failed long-term Enza treatment as well as in our AD/AI models (Fig. 5a–c; supplementary Figs. 14 and 15). Briefly, in 3 cohorts of PCa patients who underwent nADT,72,73,77 BCL-2 mRNA levels exhibited a perfect inverse correlation with the AR-A: AR-A was high and BCL-2 was low in pre-nADT PCa whereas AR-A was reduced but BCL-2 levels were upregulated in patient-matched post-nADT samples (Fig. 5a; supplementary Fig. 14a–c). We also observed an inverse correlation between BCL-2 mRNA levels and AR-A in an in-house RNA-seq dataset of 25 PCa patients treated for greater than 2 months with nADT compared with 25 stage/grade matched control cases (i.e., no nADT) (Fig. 5b; supplementary Fig. 14d). Among the other 4 BCL-2 family members, BCL-2L1 (encoding BCL-xL) and BCL-2L2 (encoding BCL-W) mRNA levels were each positively correlated with AR-A in 2 of the 4 nADT cohorts (supplementary Fig. 14a–d). In contrast, MCL1 mRNA levels anti-correlated with AR-A in 3 (i.e., Rajan, Sharma, and Long) of the 4 nADT datasets and BCL2A1 (encoding A1/BFL1) showed an inverse correlation with AR-A only in the Long cohort (supplementary Fig. 14a–d). Analysis of the Westbrook dataset consisting of 25 matched pre-/post-Enza mCRPC23 similarly revealed an anti-correlation between BCL-2 mRNA and AR-A but a positive correlation between BCL-2L2 and AR-A (supplementary Fig. 14e).
Fig. 5The alternative text for this image may have been generated using AI.
AR directly represses BCL-2 transcription in PCa. a Inverse correlation between BCL-2 mRNA levels and AR activity in 3 nADT cohorts. b Upregulation of BCL-2 mRNA and inverse correlation between BCL-2 mRNA levels and AR activity in the matched Roswell Park nADT cohort. c Inverse correlation between BCL-2 and AR mRNA levels in the 40 CRPC specimens in the Tang F et al. dataset.78 d–f LNCaP cells castrated in vitro time-dependently downregulate AR and AR signaling but upregulate BCL-2. Shown are WB of AR, PSA and BCL-2 protein levels in LNCaP cells cultured in CDSS-containing medium for the time intervals indicated (d), densitometric quantification normalized to GAPDH (e), and changes in BCL-2 mRNA levels measured by RT-qPCR (f; n = 3, mean ± SD; *p < 0.05). g LNCaP cells cultured in CDSS were exposed to DHT (1 nM) for the time intervals indicated, and BCL-2 and TBP (TATA-binding protein; housekeeping gene) transcripts were quantified by qPCR. BCL-2 expression was normalized to TBP and reported as fold change (FC) relative to 0 h (set to 1). Data are presented as mean ± SEM (n = 3). *p < 0.05 (Mann–Whitney test). h AR binds to multiple ARBS in BCL-2 genomic locus. Shown are the 7 potential ARBS in LNCaP-AD (top) that are lost in LNCaP-AI (bottom) cells (Neg: genomic region used as negative control in ChIP-qPCR). Shown on right are AR peaks and canonical AR and FOXA motifs. i Zoom-in representation of ARBS1-3 in LNCaP-AD (top) vs. LNCaP-AI (bottom) cells. The sizes of the respective qPCR amplicons for ARBS1-3 are indicated by a yellow horizontal line. j ChIP–qPCR analysis of AR binding at the ARBS1–3 in the indicated 3 AD/AI xenograft models. Shown are the mean ± S.D of 3 independent experiments (*p < 0.05; **p < 0.01; ***p < 0.001; Student’s t-test). k BCL-2 mRNA levels were elevated in the 3 AI models accompanying the loss of AR binding to ARBS. Total RNA extracted from the same AD/AI tumors as in j were employed in RT-qPCR analysis of BCL-2 mRNA levels. Shown are the results from 3 independent AD/AI tumors for each model. *p < 0.05 and **p < 0.01 (Mann–Whitney test)
We further analyzed a dataset78 of 40 patient and organoid CRPC samples classified by the ATAC-seq (Assay for Transposase-Accessible Chromatin using sequencing) and RNA-seq analyses as CRPC_AR (AR high), CRPC_SCL (stem-cell like; AR low), CRPC_Wnt (AR negative), and CRPC_NE (AR negative) subtypes (Fig. 5c; supplementary Fig. 15a). We observed a moderate anti-correlation between BCL-2 mRNA levels and AR-A and between AR vs. BCL-2 mRNA levels, with the CRPC_AR subtype having the highest AR and lowest BCL-2 while CRPC_Wnt and CRPC_NE the lowest AR but highest BCL-2 (Fig. 5c; supplementary Fig. 15a). Interestingly, among the 13 CRPC_SCL samples, 7 co-segregated with the CRPC_AR subtype and 6 with the CRPC_Wnt/CRPC_NE (Fig. 5c), suggesting that the chromatin accessibility-stratified CRPC_SCL subtype is heterogeneous with ‘bifurcated’ ARhiBCL-2lo and ARloBCL-2hi profiles. In this dataset, BCL-2L1 (BCL-xL) mRNA levels positively correlated with AR-A (but not AR mRNA levels) whereas BCL-2L2 (BCL-W) levels positively correlated with both AR mRNA levels and AR-A (supplementary Fig. 15a).
Finally, we observed a similar striking anti-correlation between BCL-2 mRNA levels and AR-A in both AR+/hi LNCaP-AI and AR-/lo LAPC9-AI models, in which BCL-2 mRNA was upregulated when AR-A became suppressed and BCL-2 levels strongly anti-correlated with the AR-A (supplementary Fig. 15b–e).
AR directly binds to the BCL-2 genomic locus to repress its transcription
We subsequently conducted molecular studies to demonstrate that AR directly represses BCL-2 transcription and ARPI relieves this inhibition leading to BCL-2 upregulation (Fig. 5d–k; supplementary Fig. 16). Experimental castration of LNCaP cells in CDSS time-dependently downregulated AR protein and inhibited AR activity (i.e., reducing PSA) over time (Fig. 5d, e) but induced BCL-2 mRNA and protein (Fig. 5e, f). In contrast, re-stimulation of CDSS-cultured LNCaP cells with DHT reduced BCL-2 mRNA levels (Fig. 5g). We analyzed the AR chromatin immunoprecipitation sequencing (ChIP-seq) data79 from LNCaP cells under regular (AD; GSM699631) and castrated (AI; GSM699630) conditions (Fig. 5h). We observed, in LNCaP-AD cells, multiple AR-binding sites (ARBS) across the BCL-2 locus (Chr18: 63.12–63.32 Mb), some of which were enriched for AR and/or FOXA motifs (Fig. 5h, top). We identified seven prominent ARBS peaks (ARBS1–7), with the ARBS1 representing the main peak located at the BCL-2 promoter region (Fig. 5h). Strikingly, in LNCaP-AI cells, AR binding was lost at all sites except for a faint residual ARBS1 peak (Fig. 5h). Consistently, most ARBS, especially ARBS1, were reduced or lost in multiple castration-resistant LNCaP sublines (supplementary Fig. 16a). We also observed loss of ARBS1 and other ARBSs in castration-resistant 22Rv1 cells in comparison to parent CWR22 (data not shown).
We performed ChIP-qPCR experiments using primers specific for ARBS1–ARBS3 (Fig. 5i; see Methods and supplementary Table 4) in our paired LNCaP-AD/AI, LAPC4-AD/AI and LAPC9-AD/AI xenograft tumors in which all 3 AI models upregulated BCL-2 (Fig. 1e, f). We found that strikingly, AR binding to ARBS1 in BCL-2 locus was lost in all 3 distinct AI (i.e., AR+/hi LNCaP-AI, ARcyto LAPC4-AI and AR-/lo LAPC9-AI) models whereas AR binding to ARBS2 and ARBS3 was either lost or reduced (Fig. 5j). These results also corroborated ARBS1 as the major cis-regulatory region of BCL-2 by AR. Accompanying the loss of AR binding to ARBS1 (and other ARBSs), RT-qPCR analysis using the same paired AD/AI tumors revealed concurrent BCL-2 mRNA upregulation, especially in AR-/lo LAPC9-AI tumors (Fig. 5k).
Together, these data suggest that in most CRPC subtypes, AR directly binds and represses the BCL-2 locus in a ligand-dependent manner and inhibition of AR signaling would block AR occupancy at key regulatory regions such as ARBS1 allowing for transcriptional activation of BCL-2. To provide further support for this mechanism in patient tumors, we interrogated an integrated dataset (GSE130408)80 of AR ChIP-seq in patient Pri-PCa and CRPC and observed globally reduced AR enrichment across the BCL-2 genomic region in CRPC compared to Pri-PCa (supplementary Fig. 16b–g). We identified a total of 15 AR-binding peaks (except for ARBS6; supplementary Fig. 16c–e) in patient Pri-PCa and CRPC samples all of which except ARBS11 and 12 were reduced in CRPC (supplementary Fig. 16c, g) as confirmed by side-by-side Pri-PCa vs. CRPC comparisons (supplementary Fig. 16f).
Increased chromatin accessibility in the BCL-2 genomic region in AR-/loBCL-2+ CRPC
We leveraged the aforementioned (Fig. 5c) ATAC-/RNA-seq dataset of 40 patient and organoid CRPC samples78 to determine the potential differences in chromatin accessibility in the BCL-2 locus in AR+ vs. AR-/lo CRPC (supplementary Fig. 17). We found that in CRPC_AR tumors, chromatin accessibility across the BCL-2 genomic region remained low (supplementary Fig. 17a), which coincided with low BCL-2 transcript levels (supplementary Fig. 17b). In contrast, the AR-/lo CRPC_Wnt and CRPC_NE exhibited increased chromatin accessibility (supplementary Fig. 17a), accompanied by elevated BCL-2 mRNA expression (supplementary Fig. 17b). Quantitative analysis of ATAC-seq peak intensities across CRPC subtypes revealed a strong positive correlation between BCL-2 mRNA levels and chromatin accessibility, with the AR-/lo CRPC_NE and CRPC_Wnt exhibiting the highest values for both (supplementary Fig. 17c). Furthermore, BCL-2 mRNA levels were found to inversely correlate with AR activity (supplementary Fig. 17d). These findings reveal increased chromatin accessibility surrounding the BCL-2 genomic locus in AR-/loBCL-2+ subtype of CRPC and suggest that loss of AR expression and signaling might also promote BCL-2 expression via modulating the chromatin landscape.
Involvement of AR, ARv7 and GR signaling in driving the BCL-2-/lo phenotype in VCaP-AI
Among the 4 AD/AI models we studied, the VCaP-AD model had the highest ‘baseline’ BCL-2 protein levels (supplementary Fig. 18a) and VCaP-AI was the only model that expressed ARv7 (supplementary Fig. 2 d,f,g, 3b)20 and displayed the opposite BCL-2-/lo phenotype (e.g., Fig. 1e, f; supplementary Figs. 2b, h, 3b, c, 12). Like LNCaP-AI and LAPC4-AI, the VCaP-AI tumors showed elevated GR expression (Fig. 1f (note the evident increase in GR protein levels in most samples despite not reaching statistical significance as a group); supplementary Fig. 2c, f–h). Of clinical relevance, VCaP-AI-like cells can be observed in patient CRPC (supplementary Fig. 7d, e). The VCaP model, which constitutively has AR genomic amplifications and TMPRSS2-ERG fusion and induces ARv7, de novo steroidogenesis and reactivation of AR signaling when progressing to castration resistance, has been used extensively in studying AR signaling and transcriptomic changes during CRPC development.81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99
We interrogated and re-analyzed multiple VCaP datasets81,82,83,84,85,89,93,94,95,96 and performed new studies aiming to shed lights on the BCL-2-/lo phenotype in VCaP-AI (supplementary Fig. 18b–o). Analysis of AR ChIP-seq datasets from several groups revealed, surprisingly, minimal AR occupancy in the BCL-2 genomic region including ARBS1 in regularly cultured VCaP (VCaP-AD) cells, and this lack of AR binding was not affected by ERG knockdown (supplementary Fig. 18c, dataset c1). On the other hand, in VCaP cells cultured in CDSS for 48 h or 4 days, AR binding to several ARBSs, especially ARBS1, was enhanced by DHT or R1881 (supplementary Fig. 18c, datasets c1-c3) but the DHT effect was blunted by Enza pre-treatment (supplementary Fig. 18c, dataset c3). Similarly, DHT stimulated AR binding to ARBS1 in VCaP cells cultured in CDSS for 48 h but not in VCaP cells pretreated with Enza for 3 weeks (VCaP_Enza-3w; supplementary Fig. 18c, dataset c4). Finally, DHT induced prominent AR binding to the BCL-2 ARBS1 in VCaP cells treated with Enza either short-term (72 h) or long-term (8 weeks in 16 μM Enza; i.e., VCaP16) but not in the same cells with the presence of Enza (supplementary Fig. 18c, dataset c5). These AR ChIP-seq data suggest that the AR protein in VCaP-AD cells was not binding to the BCL-2 ARBSs, but AR was ligand-responsive and functional. In support, re-analysis of the RNA-seq data in Helminen dataset94 (supplementary Fig. 18d–g) revealed DHT-stimulated AR-A in VCaP cells, which was dampened in VCaP_Enza-3w cells (supplementary Fig. 18d). Strikingly, consistent with our data showing reduced BCL-2 protein in VCaP-AI, BCL-2 mRNA was decreased in VCaP_Enza-3w cells (supplementary Fig. 17e).
We investigated potential involvement of ARv7 in mediating the BCL-2-/lo phenotype in VCaP-AI (supplementary Fig. 18h–m). In mCRPC_SU2C dataset, the ARv7 mRNA levels positively correlated with AR mRNA levels and the (total) AR-A, like in other datasets (supplementary Figs. 14 and 15), also inversely correlated with BCL-2 mRNA levels (supplementary Fig. 18h, top right panel). Of interest, although AR mRNA levels, as expected, strongly correlated with the AR-A, the ARv7 mRNA levels also exhibited moderate but statistically significant positive correlations with AR-A and negative correlations with BCL-2 mRNA levels (supplementary Fig. 18h). These results suggest that in ARPI-failed mCRPC, signaling from both full-length AR (AR-FL) and ARv7 contributes to the total AR-A and BCL-2 repression. In VCaP-CRPC, ARv7 may interact with AR-FL to activate canonical AR target genes and ARv7 may also possess autonomous chromatin-binding and transcriptional capabilities that can sustain AR functional output when AR-FL is lost or its activity inhibited.86,87,88,89,90,91,92,93,96,97,98,99 Moreover, although ARv7 and AR-FL genomic binding is inter-dependent and co-localized, ARv7 may interact with FOXA1 to preferentially mediate transcriptional repression.89 We found that in long-term Enza-treated VCaP16 cells, ARv7 ChIP-seq revealed a subtle but notable increase in ARBS1 binding upon DHT stimulation (supplementary Fig. 18j). Re-analyzing GSE252841 by Poluben et al.,96 we observed that siRNAs targeting both AR Exon 1 (siEX1, targeting both AR-FL and ARv7) and Exon 7 (siEX7, targeting only AR-FL), but not siRNAs targeting ARv7 (siV7), reduced the AR transcript levels (supplementary Fig. 18i, left). However, KLK3, the functional readout of AR activity, was reduced by both siV7 and siEX1 but not by AR-FL-specific siEX7 (supplementary Fig. 18i, middle). Strikingly, siV7 induced BCL-2 in VCaP16 cells (supplementary 18i, right), implicating ARv7 in repressing BCL-2 in VCaP-AI cells. In support, we exposed our own VCaP-AI cells (derived by culturing VCaP-AD cells in CDSS for 1 week) to an ARv7-selective PROTAC degrader,100 which decreased ARv7 but increased BCL-2 mRNA levels (supplementary Fig. 18k). Notably, the ARv7 PROTAC completely degraded the ARv7 and dramatically upregulated the BCL-2 protein in our VCaP-AI cells (supplementary Fig. 18l). Interestingly, the siAR20 only slightly reduced ARv7 but also significantly upregulated BCL-2 (supplementary Fig. 18l). Finally, we found that in VCaP16 cells where AR-FL was degraded by an AR PROTAC degrader (ARD),96 BCL-2 was repressed rather than induced (supplementary Fig. 18m). These results (supplementary Fig. 18h–m) together indicate that in VCaP-AI cells, ARv7 is a major driver of AR activity and the primary regulator to repress BCL-2 gene expression (especially when AR-FL is lost).
Finally, we analyzed potential roles of GR in regulating BCL-2 in VCaP-AD/AI systems. We observed that the GR agonist dexamethasone (DEX) increased AR-A in both VCaP and VCaP_Enza-3w cells (supplementary Fig. 18f), consistent with the knowledge that GR, as one of the nuclear steroid hormone receptors sharing DNA-binding sequences with AR, can drive AR transcriptional programs and mediate the so-called ‘bypass’ resistance mechanisms in some PCa settings.101 In VCaP-AD cells, DEX induced lower AR-A than DHT (supplementary Fig. 18d, compare condition 6 vs. 2). However, in castration-resistant VCaP_Enza-3w cells, DEX induced a more pronounced increase in AR-A than DHT (supplementary Fig. 18d, compare conditions 8 vs. 4). Interestingly, DEX upregulated BCL-2 mRNA in VCaP but not VCaP_Enza-3w cells (supplementary Fig. 18g). Unexpectedly, GR ChIP-seq data revealed little GR binding to the BCL-2 locus (supplementary Fig. 18n). As GR might gain access to many of its genomic targets via interacting with FOXA1,84 we examined FOXA1 ChIP-seq data and found that FOXA1 bound to several ARBSs (around ARBS4-7) of the BCL-2 region in VCaP cells cultured in CDSS for 24 h or 48 h but not in VCaP_Enza-3w cells (supplementary Fig. 18o). Notably, in CDSS-cultured (but not Enza-3w) VCaP cells, DHT, but not DEX, significantly increased FOXA1 binding to the BCL-2 ARBS1 (supplementary Fig. 18o).
Our results, taken together, suggest the following potential mechanisms for BCL-2 regulation in the VCaP-AD/AI models (supplementary Fig. 18p, q). In the majority of Pri-PCa cells, which have the (AR+)BCL-2- phenotype (Fig. 2b, f; supplementary Figs. 4d, 5b), FOXA1 is in equilibrium with AR and ‘pioneers’ the chromatin to facilitate AR co-occupancy at FKHD (Forkhead Domain) and ARE (Androgen Response Element) sites102 to repress BCL-2 transcription (supplementary Fig. 18q, scenario q1). In VCaP-AD cells, which have an AR+BCL-2+ phenotype, the major ARBS1 of the BCL-2 locus has FOXA motif (Fig. 5h) but lacks ERG motif (data not shown; see “Methods”). Therefore, there lacks appreciable AR binding to ARBS1 (which is not impacted by ERG knockdown; supplementary Fig. 18c) leading to a lack of AR-mediated BCL-2 repression. Meanwhile, FOXA1 binds to FKHD at the BCL-2 ARBS1 and promotes GR loading and GR-mediated BCL-2 transcription (supplementary Fig. 18g, left; supplementary Fig. 18p, left; supplementary Fig. 18q, scenario q2) leading to high baseline BCL-2 expression (supplementary Fig. 18a). In VCaP-AI, characterized by intracrine androgen production, ARv7 induction, and reactivation of AR signaling (supplementary Fig. 18p, right), several intertwined mechanisms may lead to BCL-2 repression and the BCL-2-/lo phenotype. Primarily, ARv7, induced de novo and being largely nuclear (due to its ligand-independent nuclear import),91,97 may interact with FOXA1 and co-occupy the ARBS1 (e.g., supplementary Fig. 18j, VCaP16_DHT-4 h; supplementary Fig. 18o) where ARv7 functions as a corepressor (CoR)89 to suppress BCL-2 transcription (supplementary Fig. 18q, scenario q3, left). Meanwhile, AR-FL in VCaP-AI cells becomes mostly cytosolic, but as ARv7 levels rise, ARv7 forms heterodimers with AR-FL, pulling AR-FL back into the nucleus and stabilizing it on chromatin.89,96,97,98,99 Together, the ARv7/AR-FL complex, promoted by intracrine androgens, co-occupies the ARBS1 (and other ARBSs) of the BCL-2 locus (e.g., supplementary Fig. 18c, DHT conditions) leading to BCL-2 transcriptional repression (supplementary Fig. 18q, scenario q3). Concurrently, ARv7 may also cooperate with AR-FL as a co-activator to promote canonical AR signaling leading to increased AR and PSA expression (Fig. 1f) and the ARhiPSAhi phenotype of VCaP-AI tumors (supplementary Fig. 18q, scenario q3, right).
The ARcytoBCL2+ (LAPC4-AI) subtype of CRPC is susceptible to BCL-2 inhibition
Hereafter, we extended our study to address whether BCL-2 plays a functional role in CRPC progression and may represent a therapeutic vulnerability in diverse subtypes of CRPC (Figs. 6 and 7; supplementary Figs. 19 and 20). We started with the LAPC4-AI model that represented the ARcytoBCL-2+ CRPC subtype, as shown by castration-induced ARcyto phenotype in vivo (Fig. 3f, g; supplementary Fig. 2b) and in vitro (Fig. 4; supplementary Fig. 13) and whose clinical relevance is supported by presence of ARcytoBCL-2+ cells in patient CRPC (e.g., supplementary Fig. 7b, d). WB revealed that the LAPC4 1° CRPC and, particularly, the 2° CRPC upregulated AR and GR as well as BCL-2 (Figs. 1e, f and 6a, b; supplementary Fig. 2c, e). We first conducted drug sensitivity assays in organoids derived from LAPC4-AD and LAPC4-AI tumors (Fig. 6c–g; supplementary Fig. 19), which manifested the AR+BCL-2- and ARcytoBCL-2+ phenotypes, respectively (Fig. 3e–g). Under optimized assay conditions (supplementary Fig. 19a–c), the LAPC4-AD organoids were more sensitive to Enza than LAPC4-AI organoids (Fig. 6c, d; IC50 43 μM vs. 95 μM, respectively; p = 0.0286, Mann–Whitney U test). In contrast, the BCL-2 inhibitor (BCL-2i) ABT-199 elicited selective toxicity to LAPC4-AI organoids (Fig. 6e, f; IC50 ~ 10 μM) but barely showed any inhibitory effect on LAPC4-AD (Fig. 6e, f; IC50 not reached) that lacked BCL-2 expression. Also, combination of Enza and ABT-199 synergistically inhibited the LAPC4-AI (Fig. 6g) but not LAPC4-AD (supplementary Fig. 19d) tumor organoids. Interestingly, RU486, a GR antagonist, although showing overall similar toxicities against LAPC4-AD and LAPC4-AI organoids when used alone (supplementary Fig. 19e, f), exhibited synergistic inhibitory effects on LAPC4-AI organoids when combined with Enza (supplementary Fig. 19g).
Fig. 6The alternative text for this image may have been generated using AI.
BCL-2 induced in the ARcyto LAPC4-CRPC represents a therapeutic vulnerability. a WB analysis of representative regulators in AR signaling and castration resistance in LAPC4-AD and its derived CRPC xenograft tumors, including first-generation (1° CRPC; lanes 3–6) and second-generation (2° CRPC; lanes 7–10) tumors. PC3 and parental LAPC4-AD cells (lanes 1–2) serve as controls. AR, GR and BCL-2 are highlighted in red. (Note part of this panel was re-organized and presented in supplementary Fig. 2e to provide an integrated view of all 4 AD/AI models). b Quantification of AR, GR, and BCL-2 protein levels from panel a, normalized to β-actin and shown as fold-change relative to LAPC4-AD (mean ± SD, n = 4 tumors/group; p-values were calculated using a two-tailed Student’s t-test). c–f Dose–response curves showing reduced sensitivity to Enza but increased response to ABT-199 in LAPC4-AI compared to LAPC4-AD in organoids assays, with corresponding IC₅₀ values indicated (c, e). Cell viability was measured by Resazurin. Data represent the mean ± SD (n = 3; *p < 0.05, **p < 0.01; Student’s t-test). g Loewe synergy analysis evaluating the combined effects of Enza and ABT-199 on LAPC4-AI cells. A 3D synergy surface plot depicts the interaction across increasing concentrations of both agents
Fig. 7The alternative text for this image may have been generated using AI.
BCL-2 represents a functional therapeutic target across distinct CRPC subtypes. a, b Experimental schema (a) and tumor growth curves of LAPC4-AI xenografts under indicated treatments (n = 8 mice/group) (b). Presented are the tumor volumes normalized to the mean tumor volumes at the beginning of treatment, i.e., week 4 (mean ± SEM; **p < 0.001, ***p < 0.0001, and ****p < 0.00001; two-way ANOVA). c–f Therapeutic studies in the progressive LNCaP-AD/LNCaP-AI models (c). In vivo tumor growth of LNCaP-AD (d) and LNCaP-AI (e) xenografts treated with vehicle (control), Enza, ABT-199, or Enza + ABT-199 (Combo). The results showed that LNCaP-AD tumors were sensitive to Enza but not ABT-199 (d) while LNCaP-AI tumors are resistant to Enza but sensitive to ABT-199 and the combination (e). Tumor volume was normalized to baseline (n = 10 per group; mean ± SEM; p-values determined by two-way ANOVA). Shown in (f) is an independent therapeutic study showing significant inhibition of tumor incidence (top) and weight (bottom) in LNCaP-AI tumors treated with Enza and ABT-199 (ABT) combination compared to Enza monotherapy (tumor incidence and endpoint weight were determined by Fisher’s exact test and unpaired Student’s t-test, respectively). g–k Therapeutic studies in the LAPC9-AI model. Shown are tumor growth curves of LAPC9-AI xenografts treated with ABT-199 (h; note the systemic toxicities of AT-101), endpoint tumor weight (i) and images (j), and an independent ABT-199 monotherapy study in LAPC9-AI xenografts (n = 8 mice/group; p-value determined by Student’s t-test)
We then performed in vivo therapeutic studies in castrated male NOD/SCID mice bearing LAPC4-AI tumors (Fig. 7a; supplementary Fig. 20a, b). The results revealed that although Enza modestly inhibited LAPC4-AI, ABT-199 exhibited strong single-agent tumor-inhibitory effects on ARcytoBCL-2+ LAPC4-AI tumors whereas the ABT-199/Enza combination demonstrated slightly better tumor-controlling effects than ABT-199 alone (Fig. 7b; supplementary Fig. 20c, d).
The AR+/hiBCL-2+ (LNCaP-AI) subtype of CRPC is similarly sensitive to BCL-2 inhibition
Next, we examined the sensitivity of AR+/hiBCL-2+ subtype of CRPC to ABT-199 using the progressive LNCaP-AD → 1° CRPC (AI) → 2° CRPC (castration/Enza-resistant) models (Fig. 7c; supplementary Figs. 2f, g, 10b–e). We first established castration- and castration/Enza-resistant LNCaP cell models that showed similar dynamic changes in AR and BCL-2 to those in the in vivo models (supplementary Fig. 20e–g). Briefly, cells were cultured in CDSS-containing media for 8 weeks to generate castration-resistant LNCaP-CR cells, which were then cultured in CDSS medium containing Enza (20 µM) for 2 weeks to derive LNCaP-C/Enza-R cells (supplementary Fig. 20e). Immunoblot analysis showed a reduction in AR protein in LNCaP-CR cells, followed by AR re-expression in LNCaP-C/Enza-R cells; in contrast, BCL-2 expression progressively increased from the LNCaP-CR to LNCaP-C/Enza-R state (supplementary Fig. 20f, g). LNCaP-C/Enza-R cells were treated with increasing doses of ABT-199 (supplementary Fig. 20h) and viability assays demonstrated higher sensitivity in LNCaP-C/Enz-R cells to BCL-2 inhibition across all doses compared to LNCaP-AD cells (supplementary Fig. 20i).
In vivo therapeutic studies also revealed strikingly different responses based on androgen dependency: the AR+BCL-2- LNCaP-AD tumors responded to Enza but not ABT-199 and the combination treatment resulted in similar tumor-controlling effects to Enza alone (Fig. 7d) while the AR+/hiBCL-2+ LNCaP-AI tumors responded well to ABT-199 and the combination treatment resulted in more pronounced tumor-inhibitory effects than ABT-199 alone (Fig. 7e). In an independent therapeutic study, the ABT-199/Enza combination more prominently inhibited both the incidence and endpoint weights of the LNCaP-AI tumors compared to Enza alone (Fig. 7f).
The AR-/loBCL-2+ (LAPC9-AI) subtype of CRPC is also vulnerable to BCL-2 inhibition
Finally, we asked whether the AR-/loBCL-2+ LAPC9-AI (Fig. 7g; supplementary Fig. 11), which was refractory to Enza20 (supplementary Fig. 9b), may also be sensitive to BCL-2 inhibition. To this end, we treated castrated male NOD/SCID mice bearing LAPC9-AI tumors (supplementary Fig. 20j, k) with ABT-199 or as a control, AT-101, an early-generation BCL-2/BCL-xL inhibitor derived from gossypol.57,58 We observed that ABT-199 significantly inhibited the growth of LAPC9-AI tumors (Fig. 7h–j) without apparent toxicity (supplementary Fig. 20l). An independent therapeutic study verified the inhibitory effect of ABT-199 on LAPC9-AI (Fig. 7k). AT-101 also inhibited LAPC9-AI growth (Fig. 7h) but it manifested significant systemic toxicity in mice such that the treatment had to be terminated early (supplementary Fig. 20l).
Correlative studies in a phase Ib clinical trial link BCL-2 inhibition to treatment response
These preclinical studies revealed the efficacy of BCL-2i ABT-199, either alone or together with Enza, in inhibiting the 3 subtypes (i.e., AR+/hi, AR-/lo, and ARcyto) of BCL-2+ CRPC. These results, coupled with our early observations,20 led us to conduct a Phase Ib clinical trial70 evaluating Enza plus venetoclax (ABT-199) in 10 patients with mCRPC (NCT03751436), and our correlative studies suggested potential therapeutic benefit of this combination in a subset of mCRPC patients (Fig. 8; supplementary Fig. 21).
Fig. 8The alternative text for this image may have been generated using AI.
Correlative studies of a Phase Ib trial validates BCL-2 as a target in PCa patients. a Schematic workflow. b Tabulated summary of the trial patients’ information. c One-dimensional ddPCR fluorescence plots demonstrating expression profiles of the indicated target genes across treatment timepoints in the responder 1-03. d Quantitative longitudinal analysis of BCL-2 transcript levels (total copies) in patient 01-03. e PSA response stratified by molecular responder vs. non-responder groups. f Log-transformed correlation of baseline CTC burden (Log10 EpCAM expression) with PSA response
Longitudinal peripheral blood samples were collected from patients for isolation and transcriptomic profiling of circulating tumor cells (CTCs) using a standardized workflow incorporating microfluidic (Parsortix) enrichment, RNA amplification, and digital droplet PCR (ddPCR) (Fig. 8a, b). This approach enabled high-sensitivity quantification of dynamic changes in BCL-2 mRNA levels, AR expression, AR pathway activity (KLK3, TMPRSS2), and potential compensatory or ‘bypass’ survival signaling (e.g., NR3C1, i.e., GR) (Fig. 8c; supplementary Fig. 21a, b). CTC collections were also characterized by ddPCR for TMPRSS2-ERG fusion (type III) and for relative CTC abundance using the epithelial marker EPCAM (Fig. 8c; supplementary Fig. 21a, b). These CTC molecular profiles were then integrated with serum PSA changes, CTC burden, and prior ARPI exposure to identify potential biomarkers of clinical benefit and resistance (Fig. 8b–f; supplementary Fig. 21).
Amongst the 10 trial patients, we observed 3 potential responders, i.e., 1-03, 2-07 and 2-01, all of whom underwent multiple cycles of combination treatment (Fig. 8b, marked in red).70 For example, CTC collections from patient 1-03, who was on treatment the longest (35 cycles and 980 days; Fig. 8b), were characterized as TMPRSS2-ERG fusion negative and showed treatment-related reductions in BCL-2 and AR, AR activity (KLK3 and TMPRASS2) and overall CTC burden (EPCAM) (Fig. 8c). Correspondingly, this patient displayed treatment-related decreases in BCL-2 copy numbers in CTCs (Fig. 8d) and the most significant drop in serum PSA levels (Fig. 8e; supplementary Fig. 21c). Two additional patients, 02-01 and 02-07, whose CTCs were characterized as TMPRSS2-ERG positive (supplementary Fig. 21b), were treated for 9 and 12 cycles, respectively (Fig. 8b) and exhibited some molecular and clinical responses. Late in treatment cycles and towards end of treatment (EOT), the CTCs in both patients showed decreases in BCL-2 and AR expression, AR activity (KLK3 and TMPRSS2), and overall abundance as evidenced by reduced EPCAM and the TMPRSS2-ERG fusion (supplementary Fig. 21b). Correspondingly, these two patients also had PSA response (Fig. 8e). Nevertheless, earlier treatment discontinuation in both patients was followed by faster PSA rebound (supplementary Fig. 21c) compared to patient 1-03.
In contrast to the above 3 responders, PSA dynamics categorized most other patients as non-responders70 (Fig. 8b, e; supplementary Fig. 21c). These non-responders (e.g., 1-05, 3-07, 3-02, 3-04) were generally treated with the Enza/venetoclax combination for only 1–3 cycles (Fig. 8b) and exhibited persistent PSA rise (Fig. 8e; supplementary Fig. 21c). At the molecular level, although the CTCs from the non-responders did exhibit reduced AR, KLK3 and TMPRSS2-ERG mRNA levels at EOT, their CTCs showed minimal baseline (pre-Tx) expression of BCL-2 that did not appreciably change on therapy (supplementary Fig. 21a). Intriguingly, unlike in the 3 responders, the NR3C1 expression levels in the CTCs of non-responders increased at EOT (compare supplementary Fig. 21a vs. supplementary Fig. 21b and Fig. 8c).
Finally, to assess how CTC levels might be related to treatment response, we performed a correlation analysis between EPCAM-based CTC levels and PSA values across all available timepoints from all patients. This longitudinal patient-integrated analysis allowed us to track how CTC abundance evolved in relation to PSA dynamics (i.e., treatment response) over time. The results revealed a strong negative correlation (r = –0.76, p = 0.001) between the two (Fig. 8f), suggesting that decreased CTC abundance is closely associated with PSA response during treatment.
