Pre-existing cell subpopulations in primary prostate cancer tumors display surface fingerprints of docetaxel-resistant cells

Docetaxel-resistant (DOC) DU145 and PC3 PCa cell lines were generated as previously reported [17]. The docetaxel-sensitive PC346C and PC339 spheroid lines and their docetaxel-resistant derivatives, PC346C DOC and PC339 DOC were established from PCa patient-derived xenografts (PDXs) and propagated as described previously [18, 19]. All cell lines were maintained as defined in Supplementary Materials and Methods. Cells were routinely tested for mycoplasma contamination and authenticated using an AmpFLSTR Identifiler Plus PCR Amplification Kit (TFS, Czech Republic) to verify their origin.

The protocol published previously [20], also detailed in Supplementary Materials and Methods describes antibody-based cell surface screening and spectral flow cytometry procedures. Briefly, for high-throughput cell surface screening, cells were expanded, barcoded with CellTrace Violet and/or CellTrace DDAO (Far Red) amine-reactive fluorescent dyes, and subjected to the staining of 332 surface antigens using the LEGENDScreen Human Cell Screening PE Kit (cat. no. 700001; Biolegend, San Diego, CA, USA). Data were acquired on a FACSVerse (Becton Dickinson (BD), Franklin Lakes, NJ, USA), accessed with a Universal Loader. For multicolor spectral flow cytometry analysis, single-cell suspensions from PDXs and patient samples were subjected to red blood cells lysis and stained using a cocktail of fluorochrome-conjugated primary antibodies. Human cells within in vivo PDX models were selected based on anti-human CD298 positivity [21]. In patient samples, anti-human CD45, CD31, CD90 antibodies were used for the exclusion of leukocytes, endothelial-like cells, and stromal cells [22, 23]. Samples were analyzed using a SONY SP6800 Spectral Cell Analyzer (SONY, Japan). Acquired FCS files were exported and analyzed using FlowJo software (v10.0.7; BD). In all (spectral) flow cytometric experiments, dead cells were excluded from the analysis based on their positivity to LIVE/DEAD Fixable Dead Cell Stains (various dyes; Invitrogen, TFS). Cell aggregates and debris were excluded from the analysis based on a dual-parameter dot plot displaying the pulse ratio (signal height/y-axis vs. signal area/x-axis). A representative gating strategy is shown in Supplementary Fig. S1A. Dilution, clonality, fluorochrome information, and catalog numbers of the antibodies used for spectral flow cytometric analyses are provided in the Supplementary Materials and Methods and Supplementary Table S4. All antibodies were titrated before use or used as the manufacturer recommended.

PDX spheroids were harvested, dissociated, and inoculated subcutaneously into the right flank of six-week-old male severe combined immuno-deficient (SCID) hairless outbred (SHO) (Crl:SHO-Prkdc^scidHr^hr) mice obtained from Charles River Laboratories (Wilmington, MA, USA). Each cohort (animals inoculated with a particular PDX model) contained four mice, corresponding to a total of 16 mice. Mice were euthanized with CO₂ three weeks after inoculation when tumors reached 1 cm³ in volume, as determined by caliper measurements. Anesthesia was not administered during the experiment. Tumors were surgically excised, enzymatically dissociated using digestion medium, and stained. All European Union Animal Welfare lines (EU Directive 2010/63/EU for animal experiments) were followed. A detailed description of in vivo xenografts and tumor dissociation is provided in the Supplementary Materials and Methods. The animal experiments were approved by the Ethical Committee of IBP CAS and REKOZ, Academy of Sciences of the Czech Republic (AVCR 65/2016), supervised by the local ethical committee, and performed by certified individuals (SD and KS). Fresh PCa tumor samples (50–150 mg), evaluated by licensed pathologists, were obtained from the University Hospital Olomouc from patients undergoing robotic prostatectomy. All human tissue samples were obtained with the approval of the University Hospital Olomouc Ethical Committee (Ref. no. 83/19) from donors who provided written informed consent. Tissue samples were minced, enzymatically digested, and stained as described in Supplementary Materials and Methods.

To isolate extracellular vesicles (EVs), cells were expanded, harvested with 0.05% trypsin/EDTA, washed twice in PBS, and plated onto new plates into a vesicle-free medium. Supernatants were collected 72 h later. EVs were isolated by differential ultracentrifugation coupled with the sucrose cushion flotation step. Cryo-electron microscopy (cryo-EM) was used for EVs visualization. The size and concentration of the EVs were assessed using a multi-angled dynamic light scattering technique. The presence of EVs was validated using SDS-PAGE and western blotting. All procedure steps are described in detail in a previously published protocol [24] and in the Supplementary Materials and Methods.

EV suspensions were solubilized and lysed in a solubilization buffer. The lysates were processed by filter-aided sample preparation. The resulting peptides were extracted into LC-MS vials, concentrated, and subjected to LC-MS analysis. Data were acquired in a data-independent acquisition mode (DIA). For a detailed sample preparation and LC-MS analysis description, see Supplementary Material and Methods. Four biological replicates were analyzed, with two being excluded due to quality control issues.

As for the Antibody-based cell surface screening, cell lines were deconvolved based on the signal of the fluorescent barcode as described previously [25]. Both the medians of fluorescence and the percentage of positivity for the PE channel were exported and analyzed. For Spectral flow cytometry analyses, spectral overlaps were calculated and compensated based on spectral unmixing algorithms using the SONY SP6800 Software (SONY, Japan). All data, including multidimensional data from single-cell analyses and visualizations, were analyzed using FlowJo (v10.0.7; BD). Multiparametric data were processed using multidimensional reduction algorithms, including tSNE, UMAP, TriMap, FlowSOM, PhenoGraph, and X-shift. Statistical analyses were performed using Prism (v5, GraphPad, La Jolla, CA, USA). Clinical data sets (accession numbers GSE193898, GSE54460, and TCGA-PRAD) were retrieved via https://​portal.​gdc.​cancer.​gov/​projects/​TCGA-PRAD, GEO (NCBI), Oncomine (TFS), and the Genomic Data Commons Data Portal (NCI). Kaplan–Meier plots were assessed via customized analysis in the DriverDBv3 cancer omics database [26], and the median was set as a cutoff. The Biocarta Pathway (Harmonizome) was used for the correlation analysis of the z-score of sets of proteins employed within a particular pathway [27]. Heat maps were generated in Prism (v9, GraphPad, La Jolla, CA, USA). As for Mass spectrometry analysis of EVs, reported protein intensities were further processed using the software container environment (https://​github.​com/​OmicsWorkflows), using the DIA-NN_PGs_LFQ_general_0.8 workflow which is available upon request. The workflow included removal of decoy his and contaminant protein groups. Only proteins identified at least on one proteotypic peptide were retained. Ratios between respective conditions were computed using MaxLFQ intensities, in case either numerator or denominator are missing, they were replaced by minimal value and reported separately. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE [28] partner repository with the dataset identifier PXD050073.

To date, no experimental approach has been utilized to identify a specific surface fingerprint associated with docetaxel resistance. In this study, we adapted a previously introduced high-throughput flow cytometry screening platform for cell surfaceomes using a commercially available LEGENDScreen kit and fluorescent barcoding (Figs. 1A and S1A) [25]. Analysis of the surface molecule signature of two well-described docetaxel-resistant cell lines DU145 DOC and PC3 DOC revealed that out of 332 surface markers (Supplementary Tables S1 and S2), 81 antigens were expressed on the surface of at least one cell model (Supplementary Fig. S1B), 14 antigens were noticeably upregulated in both docetaxel-resistant cell lines, and only one antigen, EpCAM, was upregulated in both docetaxel-sensitive parental cell lines (Fig. 1B and C). From the list of surface antigens upregulated consistently in both docetaxel-resistant cell lines, the 12 most robustly upregulated antigens, CD9, CD44, CD59, CD63, CD70, CD71, CD81, CD95, CD97, CD166, CD201 and SSEA-4 (Fig. 1C) were, along with EpCAM, selected for further profiling (Supplementary Fig. S1C). Data analysis of the above-mentioned deregulated antigens showed docetaxel-resistant and docetaxel-sensitive cell lines clustering, highlighting similarities in the phenotypic features attributable to acquired docetaxel resistance (Fig. 1D).

Extracellular vesicles (EVs) have recently been discovered to be critical modulators of resistance to cancer therapy [29]. Indeed, it has been summarized that EVs can function as sequesters of anticancer drugs, reducing their effectivity at target sites and thus serve as biomarkers for anticancer therapy response [30]. Here, we first isolated EVs from in vitro docetaxel-resistant and docetaxel-sensitive DU145 and PC3 cells and performed quality control to determine the presence, size, and mass of EVs. We employed cryo-EM (Fig. 1A and Supplementary Fig. S1A), dynamic light scattering (DLS) (Supplementary Fig. S2B), and western blot analysis of the proteins typically enriched in EVs (EV markers) markers CD9, FLOT1, and HSP70 as well as proteins reduced in EV isolates (EV negative markers) golgin-97 and Acetyl-α-Tubulin (Supplementary Fig. S2C) Next, we performed proteomic profiling of whole EVs and identified 21 proteins that were substantially enriched in both docetaxel-resistant models (Fig. 1E and Supplementary Fig. S2D). According to this analysis, two antigens, CD95 (Fas) and CD70, previously shown to be elevated on the surface of both docetaxel-resistant models, were also significantly upregulated in EVs isolated from both docetaxel-resistant models compared to docetaxel-sensitive models (Fig. 1E). In addition, EpCAM, CD9, CD44, CD63, and CD97 showed consistent trends of deregulation in both models (Supplementary Fig. S2E) with regard to the initial surface antigen screen (Fig. 1D). Other surface antigens displayed either no change or a trend in only one of the two docetaxel-resistant models (Supplementary Fig. S2D). These pilot data are intended to provide an initial, never published, overview of the composition of EVs originating from docetaxel-resistant prostate cancer cell lines and complement cellular surface profiling results. Overall, these data suggest that the surface antigen fingerprint unique to docetaxel-resistant cells is, to a large degree, reflected on the surface of EVs and thus has the potential to serve as a predictive parameter of docetaxel resistance.

Although our in vitro data portray a panel of surface antigens consistently deregulated in both docetaxel-resistant cell lines, these models lack lineage plasticity, a phenomenon that occurs within the primary tumor and contributes to intratumoral heterogeneity, EMT, diverse responses to therapy and taxane resistance [13, 31, 32]. Therefore, we validated surface markers identified by the in vitro screen in vivo, using preclinical docetaxel-resistant androgen-sensitive PDX PC346C and PC339 models [18, 19]. Positive selection of human cells from resected primary mouse tumors was achieved using human CD298 staining. Multicolor spectral flow cytometry assay followed by multidimensional data analysis using tSNE revealed partial heterogeneity with regard to selected surface antigens in both the PC346C and PC346C DOC PDX as well as PC339 and PC339 DOC models; nonetheless, docetaxel-resistant and docetaxel-sensitive cells clustered separately (Fig. 2A and Supplementary Fig. S3A). In addition to CD71 (level not detected) and EpCAM (level not changed), all antigens were expressed and were significantly altered in at least one of the two docetaxel-resistant PDXs as presented by fold change of median fluorescence intensity (Fig. 2B) as well as % positivity (Supplementary Fig. S3B). While the CD9 antigen was significantly reduced, CD44, CD59, CD63, CD81, CD95, CD97, CD201, and SSEA-4 antigens were commonly upregulated in both docetaxel-resistant PDXs as compared to their respective sensitive counterparts (Fig. 2A and B). CD70 and CD166, the two antigens that were linked to docetaxel resistance in vitro, were not considered for follow-up analysis due to inconsistent surface expression between both PDX pairs (Fig. 2B). Collectively, we established a 9-molecule surface fingerprint shared between the most relevant and available docetaxel-resistant models in vitro and in vivo.

In the follow-up investigation, we profiled a cohort of eight untreated primary PCa patient samples using the previously determined 9-molecule surface fingerprint of docetaxel-resistant cells combined with the epithelial marker EpCAM (Supplementary Table S3). The specimens were collected by radical prostatectomy, dissociated using a well-established protocol, and stained as described in the Supplementary Material and Methods. We used human CD45, CD31, and CD90 staining to discriminate between leukocytes, endothelial cells, and cancer-associated stromal cells, respectively, and investigated the expression of selected antigens in primary tumor epithelial cells only. Advanced, multidimensional data analysis of patient samples revealed heterogeneity in the expression of all antigens, except for CD81 and CD97, which were not expressed in any subpopulation of the patient samples and thus further considered as not clinically relevant (Fig. 3A). Unsupervised FlowSOM clustering algorithms uncovered a cell subpopulation (Pop3) that was characterized by high expression of CD44, CD59, CD95, and SSEA-4 and low levels of EpCAM and CD9, reflecting the surface signature of previously examined docetaxel-resistant PDX models, PC346C DOC and PC339 DOC (Fig. 3B–D and Supplementary Fig. S4A). In addition, our data revealed a cell subpopulation (Pop1) with the opposite phenotype, expressing low CD44, CD59, CD95, and SSEA-4, and high EpCAM and CD9 levels, resembling the surface antigen expression profiles shared across all previously inspected docetaxel-sensitive cell lines and PDX models (Fig. 3B–D and Supplementary Fig. S4A). Due to very low to absent expression of CD63 and CD201 within these two patient sample subpopulations (Fig. 3A and D), we disregarded these antigens from further investigations. In light of previous results, a 6-molecule surface fingerprint containing EpCAM, CD9, CD44, CD59, CD95, and SSEA-4 was examined in the Pop1 and Pop3 PCa patient sample subpopulations to prove its potential as a comprehensive fingerprint for prediction of docetaxel resistance or sensitivity in PCa patients (Fig. 3E and Supplementary Fig. S4B). The presence of Pop3 subpopulation, based on the 6-molecule surface fingerprint, was further validated to exist using multiple clustering and projection algorithms enabling multidimensional data reduction (Supplementary Fig. S5). Importantly, we could show that the Pop3 subpopulation, reflecting the docetaxel-resistant phenotype, was present in 5 out of 8 patient samples as overt or rare subpopulation ranging from 1.3% (PCa3) to 55.7% (PCa6), illustrating patient-to-patient variability (Fig. 3F). In conclusion, based on preclinical models of docetaxel resistance and primary docetaxel-naïve PCa patient samples, we propose a 6-molecule surface fingerprint composed of EpCAM, CD9, CD44, CD59, CD95, and SSEA-4 as a candidate set of surface antigens that may potentially predict the response to docetaxel in PCa patients before therapy.

Considering previous data, CD95 and SSEA-4 were some of the most differentially expressed surface antigens compared to docetaxel-resistant and docetaxel-sensitive cell lines and PDXs and heterogeneously expressed in primary PCa patient samples. Therefore, we aimed to examine whether their elevated levels alone would reflect docetaxel resistance and poor disease outcomes in patients with PCa. Here, we provide an illustrative set of primary PCa samples that demonstrates substantial heterogeneity in CD95 and SSEA4 expression (Fig. 4A and B and Supplementary Fig. S6). This phenomena is accompanied by heterogeneous N-cadherin and consistent E-cadherin expression, which is commonly used for EMT assessment in PCa [33], within similar areas of tissue from two PCa patient samples previously used for surface profiling (Fig. 4A and B). Furthermore, we retrieved data from a publicly available RNA-seq database, which included 11 paired pre- and post-docetaxel-based PCa tumor samples. Our data analysis showed a significant increase in CD95 expression post-therapy in all 11 patients (Fig. 4C). In addition, the expression of two fundamental enzymes involved in glycosphingolipid SSEA-4 synthesis, ST3GAL2 and ST3GAL1 (Supplementary Fig. S7A), was significantly upregulated in all post-therapy samples compared to that in pre-therapy samples (Fig. 4D). Notably, the other antigens from the 6-molecule fingerprint, EpCAM, CD9, CD44, and CD59, showed a similar trend in expression change as that previously observed in preclinical models of docetaxel resistance and primary PCa patient samples (Supplementary Fig. S7B). These results together corroborate the clinical relevance of these surface antigens in docetaxel-resistant PCa.

CD95/CD95L pathway modulation and activation has been associated with the increased killing of PCa cells and enhanced sensitivity to chemotherapy-induced apoptosis [34]. Therefore, we tested whether the upregulation of CD95 in docetaxel-resistant PCa cells is associated with increased sensitivity to CD95L. We analyzed cell viability using the CyQuant assay in docetaxel-resistant and -sensitive PC3 cells treated with increasing concentrations of CD95L and showed that docetaxel-resistant PC3 cells exhibit significantly higher sensitivity to CD95L than their sensitive counterparts (Supplementary Fig. S7C).

Given our data, previously published studies on PCa cells indicating EMT transition as a function of docetaxel resistance [17, 35, 36] and the link between EMT and increased expression of CD95 and SSEA-4 [37, 38], we computed the EMT score reflecting the extent to which cells display epithelial or mesenchymal phenotype on a scale from −1 (Epi) to 1 (Mes), in all pre- and post-docetaxel-based therapy samples using a two-sample Kolmogorov–Smirnov test [39]. This revealed an EMT switch and gain of mesenchymal features post-docetaxel-based therapy in all 11 matched patient samples (Fig. 4E). Furthermore, data from publicly available datasets showed that high expression of CD95 (Fig. 4F), as well as ST3GAL2, ST3GAL1, and B3GALT5 (essential for synthesis of SSEA-4 precursor, SSEA-3) (Fig. 4G and Supplementary Fig. S7A), correlated with an unfavorable prognosis determined by survival probability, particularly in a cohort of pre-docetaxel patients with progressive disease (Fig. 4H). Moreover, ST3GAL2 expression was also positively correlated with the z-score of a gene set associated with resistance to anti-microtubule agents, notably in a cohort of docetaxel-naïve patients with high-grade disease (GS ≥ 8) (Supplementary Fig. S7D and S7E). Likewise, analysis of TCGA data of primary untreated patients demonstrated a positive correlation between ST3GAL2 and CD95 (Fig. 4I and Supplementary Fig. S7F) and a negative correlation between CD95 and ST3GAL2 with EpCAM and CD9 expression along with a positive correlation with CD44 and CD59 expression (Fig. 4J and K and Supplementary Fig. S7G and S7H). Notably, in general, stronger correlations were observed in patients with high-grade disease (GS ≥ 8), emphasizing the importance of these markers in advanced PCa (Fig. 4I–K).

These analyses endorse further clinical relevance of the 6-molecule surface fingerprint as a potential predictive tool for docetaxel resistance in patients with PCa.