PC346C (p25) was generated in our lab as described previously (25, 26) and maintained in prostate growth medium (PGM) consisting of DMEM-F12, supplemented with 2% FCS (both from Lonza, Breda, Netherlands), 1% insulin-transferrin-selenium (Gibco), 100 ng/ml fibronectin (Harbor Bio-Products, Tebu-bio, Netherlands), 20 μg/ml fetuin (ICN Biomedicals, Zoetermeer, Netherlands), 0.01% BSA, 10 ng/ml epidermal growth factor, 50 ng/ml cholera toxin, 0.1 mM phosphoethanolamine, 0.6 ng/ml triiodothyronine, 500 ng/ml dexamethasone (all from Sigma-Aldrich), 0.1 nM R1881 (Sigma), and penicillin/streptomycin antibiotics (100 U/ml penicillin, 100 μg/ml streptomycin; Cambrex BioWhittaker). VCaP (p30) (20) (kindly provided by Dr. K.J. Pienta, Baltimore, MD, USA) and DuCaP (p52) (21) (kindly provided by Dr. J.A. Schalken, Nijmegen, Netherlands) were maintained in RPMI1640 (Lonza), supplemented with 10% FCS and penicillin/streptomycin antibiotics. LAPC4 (p49) (22) was kindly provided by Organon (Oss, Netherlands) and maintained in Iscove's Modified Dulbecco's Medium (IMDM) (Lonza), supplemented with 7.5% FCS and 10 nM R1881. Cell line authentication was performed by short tandem repeat analysis by the Promega PowerPlex 16 kit.

The ADT condition consisted of each parental cell line’s respective medium, substituting FCS with dextran-coated charcoal-stripped FCS (DCC) and removal of R1881. For ARTA conditions, BIC (kindly provided by Organon), FLU (Shering Plough Research Institute), or RD162 (Axxon Medchem, Netherlands) was added to the medium at a final concentration of 1 μM. Cell culture medium was refreshed twice per week, and at 80%, confluency cells were trypsinized and passaged. Once time between passaging approximated that of parental cell lines grown in standard androgen-supplemented medium, clones were considered resistant.

To test androgen responsiveness, cells were plated in steroid-stripped medium at 10,000 cells per well in 100-μl medium in 96-well culture plates with 8 replicates per condition. After overnight attachment, medium with R1881 and BIC, FLU, or RD162 was added with ethanol (for R1881) and dimethyl sulfoxide (DMSO) (for ARTA) as vehicle controls. Based on preliminary androgen response curves, we selected 0, 10^-11, 10^-10, and 10^-9 M R1881 for VCaP, DuCaP, and PC346C as androgen concentration range for proliferation assays. For LAPC4, we selected 0, 10^-10, 10^-9, and 10^-8 M. After a 9-day incubation period at 37°C and 5% O₂, cell numbers were estimated using the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay as described previously (40).

All proliferation assay results were summarized using the average MTT signal intensity for each condition relative to the MTT signal intensity at day 0. Basal growth was assessed in the absence of steroid and with vehicle control. Growth induction by increasing concentrations of R1881 was assessed relative to vehicle control. The effect of ARTA was determined relative to vehicle control at the specified amount of R1881. To group growth responses, values were log₂ transformed, clustered using Cluster (version 3.0) (41) using complete linkage based on Euclidian distance, and visualized by Java TreeView (version 1.16r4) (42).

For qPCR, cells harvested at passaging were stored as cell pellets at -80°C until further processing. RNA was isolated using RNA-Bee (TEL-TEST, Inc., Friendswood, USA). Reverse transcriptase and qPCR runs were performed as described previously (43) using an ABI Prism 7900 Sequence Detection System under standard conditions. cDNA (20 ng) was amplified in SYBR Green PCR Master Mix (Applied Biosystems, Foster City, USA) or TaqMan Universal Master Mix (Applied Biosystems). PCR efficiency was verified by cDNA dilution curves and exceeding 90% for all assays. Gene expression was calculated as fold expression over housekeeping genes glyceraldehyde 3-phosphate dehydrogenase (GAPDH) and porphobilinogen deaminase (PBGD).

Relative gene expression levels were expressed as ΔCt relative to the geometric mean of housekeeping genes GAPD and PBGD. GraphPad Prism (version 9) was used to group and visualize -ΔCt values. For cluster analyses, -ΔCt values were median centered per gene and clustered using Cluster using complete linkage with Spearman Rank Correlation (version 3.0) and visualized by Java TreeView (version 1.16r4). For statistical analyses and clustering of samples, undetectable gene expression (Ct >40) was normalized to a –ΔCt of -20.

All statistical analyses were carried out using SPSS version 26 (IBM, Armonk, NY, USA). The χ² test was used to test the null hypothesis that overall take rates between cell lines, between ADT and ARTA, and between ADT and ARTA within each cell line were equal. For the proliferation assay analysis, a regular two-way ANOVA was used to test the null hypothesis that fold growth rates between parental and CRPC cells and between hormonal/ARTA conditions were equal. For comparison of mRNA expression levels between cell lines and culture conditions, the ΔCt values were used as input to maintain a normal distribution on a log2 scale, using a regular one-way ANOVA with a Tukey-corrected post-hoc test to test the null hypothesis that these were equal between parental cell lines and/or ADT and specified ARTA. For all tests, a p < 0.05 was considered statistically significant. To assess at what time CRPC cell growth had stabilized, 5-knot smooth splines were fitted through passage times, defined as time in days until next passage, on the y-axis and passage number on the x-axis using GraphPad Prism (version 9).

PCa cell lines VCaP, DuCaP, PC346C, and LAPC4 were long-term cultured in 5–10 independent replicates in androgen-depleted medium alone (ADT) or with ARTA BIC, FLU, or RD162. The emergence of CRPC sublines (Figure 1A) varied substantially between cell line origin and per selection condition. Overall, PC346C was more prone to generate a CRPC cell line with an overall success rate of 97% vs. 69%, 42%, and 42% for VCaP, DuCaP, and LAPC4, respectively (p < 0.0001).

Strikingly, the addition of FLU or BIC to the medium at 1 μM significantly increased the success rate of CRPC subline generation compared to castrate medium (ADT) alone (p < 0.0001). However, this effect was cell line-dependent, with PC346C showing no preference for any selective pressure, while VCaP, DuCaP, and LAPC4 showed enhanced CRPC survival in the presence of BIC or flutamide compared to either ADT or the more potent AR antagonist RD162 (Figure 1A).

The median passage number (range) for successful establishment of CRPC sublines with similar growth rates as the parental line was 21 (14–41) for VCaP, 24 (21–27) for DuCaP, 27 (23–47) for PC346C, and 20 (18–27) for LAPC4 (Figure 1B). Median total selection time was 635, 587, 508, and 589 days for VCaP, DuCaP, PC346C, and LAPC4, respectively. Median number (range) of passages to definitive failure, defined as the inability to grow after passaging, was 3 (0–9) for VCaP, 16 (0–22) for DuCaP, 2 for PC346C, and 3 (0–6) for LAPC4. Median (range) days to definitive failure were 208 (131–635) for VCaP, 404 (390–440) for DuCaP, 77 for PC346C, and 215 (56–247) for LAPC4. Of note, DuCaP sublines took significantly longer to demonstrate failure due to contamination with murine stromal cells, which is in line with a previous report (41). This made it harder to identify which flasks no longer contained viable epithelial cells (Supplementary Figure S1).

To test at what passage of CRPC subline growth rate stabilized, indicative of full adaptation to its selection medium, we plotted the average time between passages of the selection cultures against the passage number. Smooth splines fitted through these data revealed that growth rates did not stabilize before the 10th passage from initiation of treatment (Figure 1C; Supplementary Figure S2).

We subsequently tested a random set of the generated CRPC cell lines for its response to ADT and ARTA. Clustering proliferation assay data revealed reproducible and cell line-specific differences in androgen response. For VCaP, DuCaP, LAPC4, and to a lesser extent PC346C, the greatest distance in AR growth response patterns was observed between parental and CRPC sublines (Figure 2A). CRPC sublines in all 4 models clustered together, with no differences in androgen response between sublines that became resistant to either ADT or ARTA with BIC, FLU, or RD162. As resistant sublines from the same cell line revealed a similar androgen response, we subsequently combined proliferation assay results of the parental and CRPC cell lines for an overall response visualization (Figure 2B).

Apart from all CRPC sublines showing higher proliferation rates in castrate conditions, we found marked differences in AR response for CRPC sublines from each model. The growth of VCaP CRPC sublines was stimulated at lower levels of R1881 compared to parental cells (10^-11 vs. 10^-10 M) and inhibited at 10^-9 M R1881, with further growth induction with AR antagonists at 1 µM at 10^-10 R1881 and up, implying hypersensitivity to the AR stimulus. Parental and CRPC DuCaP sublines showed growth induction with 1 µM ARTA with an optimal growth stimulus at 10^-11 M. Interestingly, higher levels of R1881 (10^-9 M) that were growth stimulatory for parental DuCaP showed growth inhibition in the CRPC clones. This demonstrates an increased AR hypersensitivity in CRPC sublines. In PC346C, androgens continued to induce cell growth in CRPC sublines with a similar optimum concentration of 10^-10 M R1881 as observed in the parental lines. Incubation with ARTA reduced androgen-induced cell growth both in parental and CRPC sublines. However, the increased basal proliferation of CRPC sublines (compared to parental samples) in castrate conditions was virtually unaffected by ARTA, implying that castration-resistant growth was independent of AR.

LAPC4 clones revealed the largest difference between parental and CRPC clones in androgen response and response to ARTA. Parental cells could only be stimulated to grow at 10^-10 and 10^-9 M R1881, and this stimulus could be completely inhibited by 1 μM ARTA for the 10^-10 M R1881. In contrast, the CRPC lines showed no significant growth difference between the various conditions and had no response to additional androgens or antiandrogen therapy, suggesting that LAPC4 CRPC clones all developed an AR-independent phenotype, not anymore relying on AR signaling for cell growth.

Importantly, and in contrast to our previous findings in PC346C cells treated with FLU (29), we did not observe any growth induction by the specific antiandrogens used for selection in any of the CRPC clones tested, suggesting the absence of mutations in the ligand-binding domain (LDB) of the AR that would turn these antagonists into agonists.

We subsequently tested human mRNA expression of 15 genes commonly associated with castration resistance as compared to the human GAPDH and PBGD housekeeping genes (Table 1). We first compared expression levels between individual samples from different passages of parental cell lines to further elucidate a priori presence of resistance mechanisms (Figure 3A) and account for natural variance in expression levels over time. We found that, in general, the expression of these markers was similar between different passages/samples from the same cell line. Taking into account that every ΔCt (y-axis) means a 2-fold difference, differential expression between the 4 cell lines was found for AR, FKBP5, AR-V7, AKR1C3, SRD5A1, GR, SGK1, SNAI1, and WNT5A. As has been reported previously (44), the expression of AR was highest in VCaP and DuCaP. The expression of PSA was comparable for all parental cell lines. The expression of AR-V7 was highest in VCaP and lowest in LAPC4. The steroidogenic enzymes CYP17A1 and AKR1C3 were highest in VCaP, with CYP17A1 being undetectable in the other models. SRD5A1 was highest in PC346C and LAPC4. GR expression was lowest in DuCaP. GR-target gene SGK1 expression was highest in PC346C and LAPC4. All models expressed similar levels of MYC and no IL6. The expression of the EMT marker SNAI1 was highest in PC346C. The expression of WNT5A was only detectable in PC346C and LAPC4. Clustering expression levels of these markers showed a clear discrimination between the four cell lines, with the closest relation between VCaP and DuCaP, which are derived from different metastases from the same patient (Figure 3B).

Next, we analyzed the differential expression of these 15 genes between CRPC derivatives and parental samples grouped per original cell line to elucidate resistance mechanisms per cell line.

In DuCaP and PC34C, CRPC derivatives showed increased expression of AR, while the expression of AR-V7 was enhanced in CRPC sublines of VCaP, DuCaP, and PC346C compared to parental expression levels (Supplementary Figures S3–S5). In line with the loss of AR responsiveness, the expression of AR, AR-V7, and AR target genes PSA and FKBP was lower in CRPC sublines of LAPC4 (Supplementary Figure S6). Of note, when the expression of AR-V7 was normalized to the expression of AR wt, we found an elevated expression of AR-V7 in both VCaP and PC346C CRPC sublines resistant to BIC and FLU in case of VCaP (Supplementary Figure S7). In DuCaP, AR-V7 levels did not increase relative to AR, suggesting that the differential expression of AR variants upon ADT/ARTA may occur but not in every tumor per se.

Steroidogenic enzymes CYP17A1, AKR1C3, and SRD5A1 were differentially expressed between CRPC derivatives and parental samples in VCaP and DuCaP but not in PC346C or LAPC4. In contrast, the expression of GR increased in CRPC sublines of PC346C and LAPC4 but not in VCaP or DuCaP. Markers for WNT signaling remained undetectable in VCaP and DuCaP while showing increased expression in CRPC derivatives of PC346C and LAPC4.

To elucidate whether specific types of ARTA would generate a single phenotype, we clustered the expression levels of the 15 resistance-associated genes of CRPC sublines and parental samples (Figure 4). We found that overall, parental samples grouped together in VCaP, DuCaP, and LAPC4 but mixed with CRPC samples in PC346C. In VCaP, DuCaP, and LAPC, no obvious ARTA-specific grouping was observed, while in PC346C, the greatest distance was found between FLU-resistant sublines and the other samples.