Categories
Prostate cancer

Ectopic JAK–STAT activation enables the transition to a stem-like and multilineage state conferring AR-targeted therapy resistance

The JAK–STAT pathway is altered concomitantly with TP53, RB1 and SOX2

To investigate the mechanisms of lineage plasticity in TP53/RB1-deficient mCRPC with SOX2 upregulation, we first inquired which transcriptional programs were altered concomitantly with both the loss of TP53 and RB1 and the upregulation of SOX2. By leveraging a series of LNCaP/AR cell lines we have previously– generated5, we profiled transcriptomic changes induced by TP53/RB1 deficiency and overexpression of SOX2 in four cell lines (control non-targeting short hairpin RNA (shNT), shTP53/RB1, shTP53/RB1/SOX2 and SOX2 overexpression (SOX2-OE)) before exposure to the AR therapy drug enzalutamide (Enz)25. As expected, these genetic modifications led to global transcriptomic changes, and gene set enrichment analysis (GSEA) revealed significantly altered pathways (Fig. 1a and Supplementary Tables 16), including the duality of specific pathways, where they demonstrated upregulation in TP53/RB1 double-knockdown (shTP53/RB1) and SOX2-OE cells and, by contrast, downregulation in TP53/RB1/SOX2 triple-knockdown cells (Fig. 1a). To further decipher which of these transcriptional changes specifically contribute to AR therapy resistance, we investigated signaling pathways enriched following treatment with Enz compared to vehicle (Extended Data Fig. 1a and Supplementary Tables 16). Notably, the JAK–STAT signaling pathway was the sole cancer-related pathway that was concomitantly altered with TP53/RB1 loss and SOX2 upregulation (Extended Data Fig. 1b) and was also consistently upregulated in the sgTP53/RB1 Enz-resistant cells (Extended Data Fig. 1c–g). Interestingly, the JAK–STAT pathway was not significantly altered in shNT cells treated with Enz compared to cells treated with vehicle, suggesting that the JAK–STAT pathway has a specific role in the context of TP53/RB1 deficiency (Extended Data Fig. 1a).

Fig. 1: JAK–STAT signaling is required for Enz resistance in TP53/RB1-deficient mCRPC.
figure 1

a, Heat map representing the significantly changed signaling pathways in LNCaP/AR cell lines transduced with annotated shRNAs based on GSEA analysis. Three comparisons are presented. Reads from n = 3 independently treated cell cultures in each group were used for analysis. Signaling pathways concomitantly altered with TP53/RB1 loss and SOX2 upregulation are labeled with a red bracket. b, Relative gene expression of canonical genes activated in the JAK–STAT signaling pathway in LNCaP/AR cells transduced with Cas9 and annotated guide RNAs; n = 3 independently treated cell cultures. P values were calculated using a two-way ANOVA with a Bonferroni multiple-comparison test. c, Relative cell numbers of LNCaP/AR cells transduced with Cas9 and annotated CRISPR guide RNAs. Cells were treated with 10 µM Enz for 8 d, and cell numbers (viability) were measured using a CellTiter-Glo assay, with all values normalized to the sgTP53/RB1 group; n = 3 independently treated cell cultures. P values were calculated by one-way ANOVA with a Bonferroni multiple-comparison test; RLU, relative light units. d, Relative cell numbers of LNCaP/AR cells transduced with Cas9 and annotated CRISPR guide RNAs. Cells were treated with 10 µM Enz for 8 d, and cell numbers (viability) were measured using a CellTiter-Glo assay, with all values normalized to the sgTP53/RB1 group; n = 3 independently treated cell cultures. P values were calculated by one-way ANOVA with a Bonferroni multiple-comparison test; NS, not significant. e, Tumor growth curve of xenografted LNCaP/AR cells transduced with Cas9 and annotated guide RNAs in castrated mice. Cas denotes castration 2 weeks before grafting. Enz denotes Enz treatment at 10 mg kg–1 from day 1 of grafting; n = number of independent xenografted tumors in each group (two tumors per mouse); sgNT, n = 8 tumors; sgTP53/RB1, n = 12 tumors; sgTP53/RB1/JAK1, n = 8 tumors; sgTP53/RB1/STAT1, n = 12 tumors. P values were calculated by two-way ANOVA with a Bonferroni multiple-comparison test. f, IHC staining of JAK–STAT proteins on annotated xenografted tumor slides showing representative images of n = 2 independent tumors.

Source data

JAK–STAT signaling regulates various biological processes, such as embryonic development, immune response, inflammation, cell fate decision, differentiation and hematopoiesis26,27. Notably, numerous lines of evidence implicate JAK–STAT signaling in the regulation of stem cell self-renewal and multilineage differentiation28. The consequence of JAK–STAT activation on tumorigenesis is complicated and considered a ‘double-edged sword’. On one hand, JAK–STAT signaling promotes antitumor immune surveillance and therapy-induced cell death and is associated with a favorable clinical outcome in various cancers29,30. On the other hand, constitutive activation of JAK–STAT signaling has been correlated with poor clinical outcomes in hematological malignancies and many solid tumors, including PCa31,32,33,34,35,36,37,38,39,40,41,42. In addition, JAK–STAT activation promotes epithelial-to-mesenchymal transition (EMT), invasion and metastasis of PCa43,44,45,46,47, further indicating its role in regulating PCa lineage transition. Thus, the observed ectopic upregulation of JAK–STAT signaling in the TP53/RB1-deficient and SOX2-OE PCa cells raises the intriguing possibility that it may play a crucial role in acquiring lineage plasticity-driven AR therapy resistance.

JAK–STAT signaling is required for lineage plasticity and resistance

To examine the role of JAK–STAT signaling in Enz resistance, we first surveyed a series of PCa cell lines and determined the protein levels of TP53, RB1 and JAK1. Here, we observed a substantial accumulation of JAK1 in all three Enz-resistant cell lines (DU145, PC3 and H660; Extended Data Fig. 2a), which are all characterized by TP53/RB1 deficiency (deletion/mutation), compared to in Enz-sensitive cell lines (LNCaP/AR, CWR22Pc, MDA-PCa-2b, VCaP and CWR22Rv). To further dissect the role of JAK–STAT signaling, we generated a stable sgTP53/RB1 clone by knocking out TP53 and RB1 in LNCaP/AR cells with CRISPR guides cis linked with red fluorescent protein (RFP). These sgTP53/RB1 cells proliferated significantly quicker after exposure to Enz than sgNT cells expressing green fluorescent protein (GFP; Extended Data Fig. 2b,c and Supplementary Fig. 1). sgTP53/RB1 cells displayed clear lineage plasticity, as they express significantly decreased levels of luminal lineage genes and increased levels of non-luminal lineage genes (Extended Data Fig. 2d). We also observed significant upregulation in the expression of canonical JAK–STAT signaling genes in sgTP53/RB1 cells, which was comparable to the levels of JAK–STAT signaling genes induced by SOX2 OE (Fig. 1b). Interestingly, only double knockout (KO) of TP53/RB1, but not individual KO of either TP53 or RB1, led to significant JAK–STAT activation and lineage plasticity (Extended Data Fig. 2e,f), suggesting that TP53 and RB1 cooperatively suppress ectopic JAK–STAT activation.

To determine whether sustained JAK–STAT signaling is required to maintain resistance, we knocked out key JAK–STAT signaling genes in sgTP53/RB1 cells and observed that only KO of JAK1 and STAT1 significantly blunted resistant growth of sgTP53/RB1 cells (Fig. 1c and Extended Data Fig. 3a–c). However, these results did not preclude the possibility that different JAK and STAT proteins may function within a cooperative network to regulate AR-targeted therapy resistance. Therefore, we knocked out various combinations of JAK and STAT proteins in the TP53/RB1 double-KO cells and observed that KO of JAK1 and JAK2 had a significantly more profound effect on inhibiting Enz-resistant growth of PCa cells than KO of JAK1 alone, suggesting a cooperative function of JAK2 and JAK1 in conferring Enz resistance (Fig. 1d). Similarly, KO of STAT1 and STAT3 had a significantly more profound effect on inhibiting Enz-resistant growth than KO of STAT1 alone (Fig. 1d), demonstrating how STAT3 and STAT1 function cooperatively to regulate resistance. These results were further validated in an additional Enz-sensitive PCa cell line, CWR22Pc (Extended Data Fig. 3d,e). Interestingly, KO of JAKSTAT genes in wild-type sgNT cells or in sgTP53/RB1-KO cells treated with vehicle did not influence tumor cell survival (Extended Data Fig. 3f,g), suggesting a specific role of JAK–STAT signaling in lineage plasticity-driven AR therapy resistance. These findings were validated in vivo in castrated severe combined immunodeficient (SCID) mice treated with Enz, where the depletion of JAK1 and STAT1 largely resensitized sgTP53/RB1 xenografted tumors to Enz (Fig. 1e,f).

To determine the connection between JAK–STAT signaling and lineage plasticity, we examined the expression of canonical lineage markers in sgTP53/RB1/JAK1 cells, which have suppressed JAK–STAT signaling (Extended Data Fig. 4a,b). We observed that JAK1 depletion largely attenuated the downregulation of AR signaling and the expression of luminal lineage genes (Fig. 2a,b) and upregulation of the expression of stem-like, basal, EMT and NE-like marker genes (Fig. 2c–e), which reinforces its crucial role in the acquisition of non-luminal and multilineage transcriptional programs. Immunofluorescence (IF) staining validated this transition from an exclusively AR-driven luminal lineage to an AR-independent, multilineage state after TP53/RB1 depletion (Extended Data Fig. 4c), which was largely reversed following JAK1 KO (Extended Data Fig. 4c). JAK1 KO also reversed the increased migratory and invasive abilities of sgTP53/RB1 cells (Fig. 2f–i), supporting the necessity of JAK–STAT signaling in the maintenance of an EMT lineage program. Furthermore, JAK1 or STAT1 KO also reversed the enhanced prostasphere formation of sgTP53/RB1 cells (Fig. 2j,k), which corroborates the role of JAK–STAT signaling in promoting a stem-like state.

Fig. 2: JAK1 KO stagnates the lineage transition to a stem-like and multilineage state.
figure 2

ae, Relative expression of canonical AR target genes and lineage marker genes in LNCaP/AR cells transduced with Cas9 and annotated guide RNAs; n = 3 independently treated cell cultures. P values were calculated by two-way ANOVA with a Bonferroni multiple-comparison test. f, Representative images of an LNCaP/AR cell transwell migration assay of three independent treated cell cultures. g, Quantification of the migrated cell numbers of nine representative images taken from three independent treated cell cultures for each of the cell lines. P values were calculated by one-way ANOVA with a Bonferroni multiple-comparison test. h, Representative images of an LNCaP/AR cell invasion assay of three independent treated cell cultures. i, Quantification of the numbers of invading cells of nine representative images taken from three independent treated cell cultures for each of the cell lines. P values were calculated by one-way ANOVA with a Bonferroni multiple-comparison test. j, Representative images of an LNCaP/AR cell prostasphere formation assay of three independent treated cell cultures. k, Quantification of the prostaspheres formed from three independent treated cell cultures for each of the cell lines. P values were calculated by one-way ANOVA with a Bonferroni multiple-comparison test. Unless otherwise noted, data are represented as mean ± s.e.m.

Source data

JAK–STAT activation correlates with poor clinical outcomes

Given the prominent role of JAK–STAT signaling in promoting EMT and AR therapy resistance in our preclinical model, we examined the impact of JAK–STAT upregulation in various clinically relevant models and scenarios. We performed immunohistochemistry (IHC) staining of key JAK–STAT proteins in a collection of deidentified human PCa samples and matched benign prostate tissues and validated the substantial augmentation of JAK–STAT signaling in human PCa samples, especially CRPC samples, compared to matched benign tissue (Fig. 3a). Consistent with the IHC results, human PCa tumor samples exhibited a significant enhancement in the expression of JAK1 and STAT1 compared to that observed in benign tissues (Fig. 3b,c). We then treated seven independent human-derived explants (PDE) and observed an upregulation of JAK1 and STAT1 following Enz treatment (Fig. 3d–f)48,49, further demonstrating their role in mediating AR therapy resistance. Next, we investigated two human PCa cohorts (The Cancer Genome Atlas (TCGA) and SU2C) and hypothesized that reduced sensitivity to AR-targeted therapy would correlate with a higher frequency of copy number variations of JAK–STAT genes in mCRPC tumors than in hormone-sensitive primary tumors50,51,52. Indeed, the frequencies of copy number amplifications and somatic mutations in JAK–STAT signaling genes were significantly higher in mCRPC (SU2C) than in hormone-naive PCa (TCGA; Extended Data Fig. 5a,b). Finally, we examined both the pathological characteristics and the expression of JAK–STAT signaling genes in the TCGA cohort and discovered that individuals with regional lymph node metastasis (N1) or high-grade tumors (Gleason score of ≥8) had significantly higher JAK–STAT signaling gene expression than individuals lacking regional lymph node metastasis (N0) or with low-grade tumors (Gleason score of ≤7; Extended Data Fig. 5c,d).

Fig. 3: Ectopic JAK–STAT activation correlates with poor clinical outcomes.
figure 3

a, IHC staining of annotated JAK–STAT proteins on benign prostate tissues or PCa samples; n = 2 independent tumors in each group. b,c, Relative expression of JAK1 (b) and STAT1 (c) in benign prostate tissues or PCa samples. The center line indicates the median, the box limits indicate upper and lower quartiles and the whiskers indicate maximum and minimum values. P values were calculated by a two-sided Mann–Whitney test; n = 10 benign prostate samples; n = 11 PCa tumors. d, Schematic figure representing the generation and examination of the PDE model. The figure was created with BioRender.com; Veh, vehicle. e, Relative expression of JAK1 in a series of PDEs treated with vehicle (DMSO) or Enz (10 µM) for 24 h. f, Relative expression of STAT1 in a series of PDEs treated with vehicle (DMSO) or Enz (10 µM) for 24 h. For e and f, n = 7 independent PDEs, and data show mean ± s.e.m. P values were calculated by two-sided t-test. g, Principal-component analysis (PCA) plots of human CRPC biopsy samples; participant 1, n = 2,691 cells, CRPC-adeno; participant 5, n = 2,123 cells, CRPC-NE. For each sample, single-cell transcriptomic profiles are colored by the expression (log2 CPM) of selected genes representing canonical signaling pathways and lineage-related transcriptional programs. The schematic figure was created with BioRender.com. hl, Violin plots representing the expression scores of canonical JAK–STAT signaling, AR signaling and lineage marker genes in subclones with high versus low TP53/RB1 expression in both participants 1 and 5. The center line indicates the median, upper and lower lines indicate upper and lower quartiles and violin limits indicate maximum and minimum values; TP53/RB1-high: participant 1 n = 2,215 cells and participant 5 n = 1,796 cells; TP53/RB1-low: participant 1 n = 476 cells and participant 5 n = 327 cells. P values were calculated by two-sided Mann–Whitney test.

Source data

To determine whether JAK–STAT signaling is specifically upregulated in human PCa with reduced TP53/RB1 expression, we performed transcriptomic analysis of an existing human CRPC scRNA-seq dataset24. Among the six individuals of this cohort, we identified two major clusters of PCa cell subpopulations expressing either high or low levels of both TP53 and RB1 in participant 1 (CRPC-adeno) and participant 5 (CRPC-NE; Fig. 3g). Transcriptomic analysis revealed increased expression of JAK–STAT signaling genes, such as JAK1, STAT1 and IL6ST, in the TP53/RB1-low subpopulation compared to in the TP53/RB1-high subpopulation in both individuals (Fig. 3g,h). Strikingly, the TP53/RB1-low subpopulations displayed substantially higher expression of stem-like (TACSTD2, ATXN1, KRT4 and CD55) and EMT (VIM, SNAI2 and CDH11) gene and lower AR target (KLK3, PTGER4 and ACSL3) gene (Fig. 3g,i–k), which is consistent with the role of JAK–STAT signaling in promoting the transition from an AR-dependent state to an AR-independent, multilineage and stem-like state. Interestingly, an increase in NE-like lineage in the TP53/RB1-low cells was only observed in participant 1 (CRPC-adeno) but not in participant 5 (CRPC-NE; Fig. 3g,l). These data indicate that JAK–STAT may be dispensable for tumor cells exclusively expressing NE-like lineage. To further validate whether ectopic JAK–STAT is required for resistance in human PCa, we surveyed a series of three-dimensional (3D)-cultured human-derived organoid (PDO) models (Extended Data Fig. 6a)53,54,55 and observed ectopic upregulation of JAK–STAT signaling genes in PDOs with TP53/RB1 deficiency (Extended Data Fig. 6b). Among those PDOs, MSKPCa8 and MSKPCa9 belong to a subclass defined by increased stem-like, EMT-like and interferon response-related transcriptional programs54,55. Strikingly, JAK–STAT signaling inhibition by the JAK1 inhibitor filgotinib (Filg) largely resensitized these Enz-resistant PDOs (Extended Data Fig. 6c,d), supporting the crucial role of JAK–STAT in mediating AR therapy resistance.

JAK1 inhibition reverses lineage plasticity and resistance

Identification of JAK–STAT signaling as a crucial executor of lineage plasticity-driven resistance raises the hope that appropriate therapeutic approaches targeting this pathway could overcome AR-targeted therapy resistance. Indeed, in vitro cell viability assays demonstrated that combination treatment of Filg and Enz significantly inhibited the growth of Enz-resistant sgTP53/RB1 LNCaP/AR cells (Fig. 4a). Dose–response measurements (half-maximum inhibitory concentration (IC50)) validated that sgTP53/RB1 cells exhibit less sensitivity to Enz than sgNT cells (Extended Data Fig. 7a), while the sgTP53/RB1 cells are more susceptible to Filg than sgNT cells (Extended Data Fig. 7b). These results were again validated in CWR22Pc cells, where Filg significantly inhibited the growth of Enz-resistant cells and attenuated the upregulation of non-luminal lineage programs (Extended Data Fig. 7c,d). Furthermore, Filg impaired the growth of DU145 and PC3 cells, two Enz-resistant PCa cell lines expressing ectopic levels of JAK1 (Extended Data Fig. 7e,f). These in vitro results are further supported by in vivo xenograft experiments, as the combination treatment of Enz and Filg stagnated the growth of Enz-resistant sgTP53/RB1 tumors and induced more tumor regression than either drug alone (Fig. 4b).

Fig. 4: JAK1 inhibitor restores Enz sensitivity.
figure 4

a, Relative cell number of LNCaP/AR cells transduced with Cas9 and annotated CRISPR guide RNAs and treated with annotated treatments in CSS medium and normalized to the vehicle group; Enz, 10 μM Enz; Filg, 5 μM Filg; Enz + Filg, combination of Enz and Filg; vehicle, DMSO treatment with equal volume as Enz. Cells were treated for 8 d, and cell numbers were measured by a CellTiter-Glo assay. b, Waterfall plot displaying changes in tumor size of xenografted LNCaP/AR-sgTP53/RB1 cells after 2 weeks of treatments. All animals were treated with Enz at 10 mg kg–1 orally 1 d after grafting. Beginning from week 3 of xenografting, animals were randomized into three groups and treated with Enz only at 10 mg kg–1 orally, Filg only at 20 mg kg–1 orally twice daily or a combination of Enz plus Filg; n = the number of independent xenografted tumors in each group (two tumors per mouse); Enz, n = 10 tumors; Filg, n = 10 tumors; Enz + Filg, n = 10 tumors. P values were calculated by one-way ANOVA with a Bonferroni multiple-comparison test. c, IF staining of the Trp53loxP/loxPRb1loxP/loxP + empty (Trp53/RB1-WT) and Trp53loxP/loxPRb1loxP/loxP + Cre (Trp53/RB1-KO) organoids in 3D with annotated antibodies; representative images of n = 2 independent treated cell cultures are shown. d, Brightfield images of annotated organoids treated with DMSO (vehicle), 1 μM Enz, 5 μM Filg or Enz and Filg (Enz + Filg) for 6 d; representative images of n = 3 independent treated cell cultures are shown. e, Relative cell numbers of annotated organoids treated with annotated treatments for 6 d normalized to the vehicle group. Treatments are the same as described in d. f, Percentage of organoids that display lumen or hyperplasia morphology. Treatments are the same as described in d. g, Relative expression of JAKSTAT and lineage marker genes in organoids treated with the treatments annotated in d. h, IF staining of the annotated organoids with antibodies targeting the proteins encoded by AR target genes and lineage marker genes; representative images of n = 2 independent treated cell cultures are shown. Unless otherwise noted, n = 3 independent treated cell cultures, and data represent mean ± s.e.m. P values were calculated by two-way ANOVA with a Bonferroni multiple-comparison test.

Source data

To further explore the effect of JAK1 inhibition in a genetically defined model, we used the previously established mouse prostate organoids derived from Trp53loxP/loxPRb1loxP/loxP mice, followed by infection with Cre or empty lentivirus5. In contrast to the typical lumen structure, which the Trp53loxP/loxPRb1loxP/loxP + empty (Trp53/Rb1-wildtype (WT)) organoids formed in 3D culture, Trp53loxP/loxPRb1loxP/loxP + Cre (Trp53/Rb1-KO) organoids displayed a hyperplastic morphology, where the organoid cells formed a solid ball with protrusive structures invading the surrounding Matrigel (Fig. 4c,d). The Trp53/Rb1-KO organoids expressed significantly elevated levels of JAK–STAT proteins compared to Trp53/Rb1-WT organoids (Fig. 4c and Extended Data Fig. 8a). Although these Trp53/Rb1-KO organoids were significantly more resistant to Enz than Trp53/Rb1-WT controls (Fig. 4d,e), they responded well to the combination of Enz and Filg (Fig. 4d,e). Remarkably, we also observed that a substantial number of Trp53/Rb1-KO organoids reestablished a classic lumen-like structure when treated with Filg (Fig. 4d,f), indicating that JAK1 inhibition impairs the acquisition of non-luminal programs and restores the luminal program. Consistent with this hypothesis, the percentage of lumen-like organoids in the Trp53/Rb1-KO group significantly receded when treated with Enz and Filg (Fig. 4d,f), suggesting that Enz sensitivity was restored in those lumen-like organoids. The reversal of the lineage plasticity within Filg-treated organoids is supported by quantitative PCR (qPCR) results and IF staining, which demonstrated attenuated downregulation of AR and luminal gene expression and upregulation of non-luminal gene expression (Fig. 4g,h and Extended Data Fig. 8b).

As JAK1/JAK2 and STAT1/STAT3 may cooperatively mediate lineage plasticity and resistance (Fig. 1d), we examined the inhibitory effects of various pharmaceutical inhibitors targeting different JAK and STAT proteins, including ruxolitinib (JAK1/JAK2 inhibitor), fludarabine (STAT1 inhibitor) and niclosamide (STAT3 inhibitor). Interestingly, the dual JAK1/JAK2 inhibitor ruxolitinib had a greater inhibitory effect on TP53/RB1-KO cells than Filg (Extended Data Fig. 8c). Similarly, combined administration of fludarabine and niclosamide achieved a more profound inhibitory effect on Enz-resistant growth than fludarabine or niclosamide alone (Extended Data Fig. 8c), supporting the cooperative roles of both JAK1/JAK2 and STAT1/STAT3. To further examine whether JAK–STAT signaling mediates lineage plasticity-driven resistance in a broader fashion, we surveyed a series of xenograft-derived, Enz-resistant cell lines with CHD1 loss, which display clear lineage plasticity12, and identified three cell lines with ectopic JAK–STAT signaling (Extended Data Fig. 8d). JAK–STAT inhibition through both Filg and ruxolitinib largely resensitized xenograft-derived resistant cells to Enz (Extended Data Fig. 8e–g), suggesting that PCa cells may hijack JAK–STAT signaling as a general avenue to promote lineage plasticity and resistance.

SOX2 promotes JAK–STAT signaling in a positive feedback fashion

We next sought to reveal the mechanism through which JAK–STAT signaling is upregulated. Interestingly, SOX2 KO in the TP53/RB1-deficient cells impaired the upregulation of JAK–STAT signaling genes (Fig. 1b), indicating a critical role of SOX2 in activation of JAK–STAT signaling. SOX2 chromatin immunoprecipitation (ChIP)–qPCR analysis supports this hypothesis by demonstrating a significant augmentation of SOX2 binding at JAKSTAT gene loci in cells with TP53/RB1 KO or ectopic SOX2 expression (Fig. 5a–d). Consistent with these SOX2 ChIP–qPCR results, an increase in histone 3 lysine 27 (H3K27) acetylation (H3K27ac) and a decrease in H3K27 trimethylation (H3K27me3) at the JAK1 gene locus following TP53/RB1 KO or SOX2 OE were also identified, indicating a transcriptional upregulation of JAK1 by SOX2 (Extended Data Fig. 9a,b). This hypothesis was further supported by analyzing an existing SOX2 ChIP–seq dataset generated from another mCRPC cell line with ectopic SOX2 expression56, CWR-R1, which demonstrated PCa-specific SOX2 binding sites in JAKSTAT genes compared to canonical SOX2 binding sites in the embryonic stem cell line WA01 (Extended Data Fig. 9c,d). To explore whether JAK and STAT are required for SOX2-promoted lineage plasticity and resistance, we knocked out JAK1 and STAT1 in the SOX2-OE cells and observed significantly impaired resistant growth of those cells, as shown in cell proliferation assays (Fig. 5e) and CellTiter-Glo viability assays (Fig. 5f). Furthermore, JAK1 and STAT1 KO in the SOX2-OE cells largely attenuated the acquisition of lineage plasticity (Fig. 5g). JAK1 inhibition by Filg significantly resensitized SOX2-OE cells to Enz (Extended Data Fig. 9e) and attenuated the acquisition of lineage plasticity in these cells (Extended Data Fig. 9f).

Fig. 5: SOX2 enables JAK–STAT activation in a positive feedback fashion.
figure 5

ad, SOX2 ChIP–qPCR of JAK1 (a,c) and STAT1 (b,d) genomic loci in LNCaP/AR cells transduced with annotated CRISPR guide RNAs or overexpressing constructs. e, Relative cell numbers of LNCaP/AR cells transduced with annotated constructs and treated with Enz or vehicle, normalized to the vehicle group; Enz, 10 μM Enz; vehicle, DMSO treatment with equal volume as Enz. Cells were treated for 6 d, and cell numbers were measured by cell proliferation assay. f, Relative cell number fold change of LNCaP/AR cells transduced with annotated constructs. Data are normalized to the SOX2-OE + sgNT group; Enz, 10 μM Enz treatment for 8 d. Cell numbers were measured by a CellTiter-Glo assay. P values were calculated by one-way ANOVA with a Bonferroni multiple-comparison test. g, Relative expression of canonical lineage marker genes in LNCaP/AR SOX2-OE cells transduced with annotated constructs. h, Relative expression of canonical lineage marker genes in LNCaP/AR cells transduced with JAK1 or STAT1 cDNA constructs. SOX2 expression is highlighted in red. i, Relative expression of SOX2 in LNCaP/AR cells transduced with annotated guide RNAs. j, Relative expression of SOX2 in LNCaP/AR cells treated with 5 µM Filg or 5 µM ruxolitinib (Rux) or DMSO for 8 d. P values in i and j were calculated by one-way ANOVA with a Bonferroni multiple-comparison test. k, Relative gene expression levels of canonical JAK–STAT signaling and lineage marker genes in the inducible shTP53/RB1 LNCaP/AR cells treated with Dox for various lengths of time. Data are normalized to 0 h. Unless otherwise noted, n = 3 independent treated cell cultures, and data represent mean ± s.e.m. P values were calculated by two-way ANOVA with a Bonferroni multiple-comparison test.

Source data

To reveal whether JAK–STAT activation is sufficient to promote lineage plasticity, we overexpressed JAK1 and STAT1 (JAK1-OE and STAT1-OE) in LNCaP/AR cells and observed significantly upregulated expression of stem-like, EMT, basal and NE-like marker genes (Fig. 5h). Notably, the observed upregulation of SOX2 in JAK1-OE and STAT1-OE cells (Fig. 5h) suggests positive feedback regulation between SOX2 and JAK–STAT activation. Consistent with this feedback model, JAK1 inhibition through either CRISPR-mediated KO or Filg treatment in sgTP53/RB1 cells led to a ~30% reduction of SOX2 expression (Fig. 5i,j). Furthermore, combination of KO or pharmaceutical inhibition of various JAK and STAT proteins led to a more profound downregulation of SOX2 expression (Fig. 5i,j), suggesting that various JAK and STAT proteins cooperatively regulate SOX2 in a similar feedback fashion. Finally, to further decipher the dynamic of this SOX2- and JAK–STAT-regulated lineage plasticity, we used an inducible shRNA-transduced LNCaP/AR model, where doxycycline (Dox)-inducible TP53/RB1 knockdown led to upregulation of JAK–STAT signaling genes as soon as 12 h following Dox administration (Fig. 5k). Remarkably, stem-like and EMT-like programs were spontaneously upregulated with JAK–STAT signaling as soon as 12 h after Dox induction, while NE-like programs were not upregulated until 24 h after Dox administration (Fig. 5k). Furthermore, although stem-like and EMT-like programs were simultaneously reversed to wild-type levels following the downregulation of JAK–STAT signaling after Dox removal, NE-like programs were not fully restored (Fig. 5k), suggesting that NE-like programs were retained in a subset of cells. These results may suggest that JAK–STAT signaling is required for therapy resistance of stem-like and multilineage cells rather than cells exclusively expressing NE-like lineage.

Single-cell transcriptomics reveal lineage heterogeneity

To examine the role of JAK–STAT in heterogeneous cell subpopulations, we performed scRNA-seq and transcriptomic analysis using the series of LNCaP/AR cell lines treated with Enz or vehicle. As expected, clustering of the sequenced cells was primarily driven by genetic and treatment perturbations (Fig. 6a–c). Interestingly, the majority of both the sgNT and sgTP53/RB1/JAK1 cells were clearly separated by different treatments (Fig. 6a,c), while sgTP53/RB1 cells did not display a similar separation (Fig. 6b). These data support the observation that a majority of the sgTP53/RB1 cells exhibit Enz resistance. Because AR antagonists can promote PCa cell cycle arrest57, we performed cell cycle prediction analysis and observed a dramatically increased cell cycle arrest occurring in the sgNT cells treated with Enz (Fig. 6a,d). By contrast, Enz treatment did not increase the population of sgTP53/RB1 cells in G1 phase, suggesting that the majority of sgTP53/RB1 cells are resistant to Enz (Fig. 6b,d). Remarkably, JAK1 KO substantially increased the percentage of cells entering G1 after Enz treatment compared to that observed in the vehicle-treated group (Fig. 6c,d). These data validate the specific role of JAK–STAT in mediating AR-targeted therapy resistance. To further assess the dynamics of resistance, we investigated whether AR signaling was restored in resistant subclones. Not surprisingly, the sgNT + vehicle group consisted of the greatest number of cells expressing canonical AR score genes (Supplementary Table 7), and inhibition of their expression was subsequently verified after Enz exposure (Extended Data Fig. 10a–f). By contrast, both sgTP53/RB1 vehicle and sgTP53/RB1 Enz groups lacked expression of AR genes, supporting the emergence of AR-independent transcriptional programs (Extended Data Fig. 10a–f). The expression of AR targets was largely reestablished in many cells belonging to the sgTP53/RB1/JAK1 + vehicle group (two-thirds of AR score genes; Supplementary Table 7) compared to that observed in the sgTP53/RB1 + vehicle group (Extended Data Fig. 10a–f). These data suggest a partial restoration of AR signaling and AR dependency among the sgTP53/RB1/JAK1 cells.

Fig. 6: JAK–STAT is required for AR therapy resistance of heterogenous subclones.
figure 6

ac, UMAP plots of single-cell transcriptomic profiles of LNCaP/AR cells transduced by annotated CRISPR guide RNAs and treated with vehicle (DMSO) or 10 µM Enz for 5 d; sgNT (Veh, n = 14,268 cells; Enz, n = 15,149 cells; a); sgTP53/RB1 (Veh, n = 12,267 cells; Enz, n = 9,850 cells; b); sgTP53/RB1/JAK1 (Veh, n = 25,200 cells; Enz, n = 11,096 cells; c). Cells on the left are colored according to sample origin, while cells on the right are colored by predicted cell cycle phase. d, Bar plot presenting the percent distribution of single cells in different cell cycle phases in each sample. The numbers of cells (n) are the same as in ac. P values were calculated by two-sided Fisher’s exact test. e, Single-cell profile of LNCaP/AR cells based on clustering. A UMAP plot of single cells colored by unsupervised clustering of six subsets is presented; cluster 0 (C0), n = 26,944 cells; C1, n = 15,994 cells; C2, n = 14,029 cells; C3, n = 14,278 cells; C4, n = 10,025 cells; C5, n = 6,560 cells. f, Single-cell profile of LNCaP/AR cells based on subclustering. A UMAP plot of single cells colored by unsupervised clustering of 13 subclusters is presented; C0, n = 26,944 cells; C1, n = 15,994 cells; C2-1, n = 9,513 cells; C2-2, n = 2,402 cells; C2-3, n = 2,114 cells; C3-1, n = 2,578 cells; C3-2, n = 6,079 cells; C3-3, n = 5,621 cells; C4-1, n = 3,680 cells; C4-2, n = 3,459 cells; C4-3, n = 2,886 cells; C5-1, n = 4,775 cells; C5-2, n = 1,785 cells. g, Single-cell profile of LNCaP/AR cells transduced with annotated CRISPR guide RNAs and treated with vehicle or Enz. A UMAP plot of single cells colored by samples is represented. The area and number of clusters in e are highlighted with colored circles. h, Single-cell profile of LNCaP/AR cells based on cell cycle states. A UMAP plot of single cells colored by cell cycle prediction is presented. The area and number of clusters in e are highlighted with colored circles. i, Bar plot presenting the percent distribution of single cells in different cell cycle phases in each of the six clusters. The number of cells (n) in each sample is the same as in e. P values were calculated by two-sided Fisher’s exact test.

Source data

To characterize lineage-specific tumor heterogeneity in resistant PCa cells, we performed unsupervised graph clustering (uniform manifold approximation and projection (UMAP))58 and identified six distinct cell subsets labeled as clusters 0–5, with further partitioning to 13 subclusters (Fig. 6e,f). Consistent with transcriptomic changes caused by TP53/RB1/JAK1 modification, five of the six clusters (clusters 0–4) predominantly overlapped with the clusters identified by genetic and treatment perturbations (Fig. 6g), while cluster 5 is a mixture of a small fraction of cells from five groups (Fig. 6e–g). To examine the cell proliferation state of these clusters, we overlapped the transcriptomic-based clustering with cell cycle prediction (Fig. 6h). Interestingly, cells within clusters 0, 1, 3 and 5 remain proliferative (termed the ‘winner’ clusters; Fig. 6i), whereas cluster 2 contains a much higher percentage of cells in cell cycle arrest (termed the ‘loser’ cluster; Fig. 6i). Lastly, cells within cluster 4 express elevated levels of cell cycle phase heterogeneity (Fig. 6h), a finding that will be expounded on later.

JAK–STAT signaling is required for stem-like and multilineage clones

We next probed the well-established AR score and five lineage-specific gene signatures (Supplementary Table 7)5,24,59,60,61and analyzed the expression of genes (z score) comprising these signatures across all clusters and samples (Fig. 7a–c). In congruence with the luminal epithelial cell lineage of LNCaP/AR cells, cluster 2 and cluster 3, which consist of cells originating from the sgNT groups, represent the two clusters expressing the highest level of luminal genes (Fig. 7a–d). Most of cluster 2 cells, while retaining their luminal lineage, displayed loss of AR signaling gene expression and entered cell cycle arrest following Enz administration (Fig. 7a–e). Notably, the most substantial proportions of clusters 0 and 1, consisting primarily of cells originating from the sgTP53/RB1 groups, expressed the lowest levels of the luminal gene signature and relatively high levels of non-luminal and multilineage gene signatures (Fig. 7a–i). Surprisingly, clusters 0 and 1 also contained a proportion of cells from the sgTP53/RB1/JAK1 + vehicle group, which maintained non-luminal programs (Fig. 7b–i), supporting the hypothesis that JAK–STAT inhibition does not impair the survival of those subclones in the absence of Enz (Figs. 6i and 7b,c). However, Enz dramatically diminished the survival of sgTP53/RB1/JAK1 subclones and the expression of stem-like and multilineage programs, suggesting that JAK–STAT inactivation restored AR dependency and impaired lineage plasticity (Fig. 7b,c). This hypothesis is further supported by restored AR signaling in sgTP53/RB1/JAK1 subclones (Extended Data Fig. 10a–f). Interestingly, JAK1 KO did not substantially impair the resistance of subclones only expressing an NE-like lineage program (Fig. 7b,c,i), indicating that JAK–STAT signaling is specifically required for the transition to a stem-like and multilineage state rather than the transition to an exclusive NE-like state.

Fig. 7: JAK–STAT is required for stem-like and multilineage subclones.
figure 7

a, Heat map representing the lineage scores of canonical lineage marker gene signatures in cell clusters. Winner clusters (without increased cell cycle arrest) are highlighted in green, and the loser cluster (with increased cell cycle arrest) is highlighted in red. b, Radar plot representing the lineage scores and distribution of different cell clusters. c, Radar plot representing the lineage scores and distribution of different samples. In ac, lineage scores were scaled from 0 to 1 across all clusters. d, UMAP plot of single-cell transcriptomic profiles colored by luminal gene signature score (z score) for each cell (dot). e, UMAP plot of single-cell transcriptomic profiles colored by AR gene signature score (z score) for each cell (dot). f, UMAP plot of single-cell transcriptomic profiles colored by EMT gene signature score (z score) for each cell (dot). g, UMAP plot of single-cell transcriptomic profiles colored by stem cell-like gene signature score (z score) for each cell (dot). h, UMAP plot of single-cell transcriptomic profiles colored by basal gene signature score (z score) for each cell (dot). i, UMAP plot of single-cell transcriptomic profiles colored by NE-like gene signature score (z score) for each cell (dot). In di, distribution areas of each cluster are labeled in color circles. The color density of each cell is scaled by the color bar. For all data, the numbers (n) of cells in each sample and cluster are the same as in Fig. 6, and lineage scores were scaled from 0 to 1 across all cells.

To decipher the dynamics of lineage plasticity, we performed pseudotime reconstructing trajectory analysis (Fig. 8a–c). We started with the transcriptional landscape of the only loser cluster, cluster 2, and tracked the changes of cell proliferation and lineage states. The increased pseudotime correlated with cell fitness, as reflected by an increased percentage of cells with active cell cycle and proliferation (Fig. 8c,d). Because clusters 2 and 3 predominantly contain wild-type sgNT cells (Fig. 7e,f), Enz treatment caused a substantial decrease of both cell fitness and pseudotime of the luminal and AR-dependent cells in those two clusters (Fig. 8c,d). Genetic perturbation of TP53/RB1 KO (clusters 0 and 1) led to the transition to a multilineage and stem-like state, which confers an increase in cell fitness and pseudotime (Fig. 8a–d,f–h). Interestingly, JAK1 KO did not immediately impair fitness nor reduce pseudotime of multilineage subclones but rather restored AR signaling in those cells (Fig. 8e). Indeed, Enz substantially impaired the fitness of those JAK1 KO subclones and led to a decrease in pseudotime (Fig. 8c,d), supporting the hypothesis that JAK–STAT inhibition restored AR dependency of those cells. Notably, the subclones only expressing NE-like lineage maintained both high fitness and pseudotime (Fig. 8h), suggesting that JAK–STAT signaling is inessential for those subclones.

Fig. 8: Dynamics of lineage plasticity driven by ectopic JAK–STAT activation.
figure 8

a, UMAP plots represent the reconstructive trajectory of single cells in each of the samples. b, UMAP plots represent the reconstructive trajectory of single cells in each of the subclusters. c, UMAP plots represent the pseudotime reconstructive trajectory of single cells. Color intensity represents the pseudotime estimation of each single cell. Arrows and the dotted line represent the direction of pseudotime flow. d, UMAP plots represent the S phase score per cell in each single cell within the pseudotime reconstructive trajectory. eh, UMAP plots represent the AR signaling and lineage scores per cell in each single cell within the pseudotime reconstructive trajectory. i, UMAP plots represent the pseudotime reconstructive trajectory of single cells of cluster 4. Color intensity represents the pseudotime estimation of each single cell. Arrows and the dotted line represent the direction of pseudotime flow. jl, UMAP plots represent lineage scores per cell in each single cell within the pseudotime reconstructive trajectory of single cells of cluster 4. m, Schematic figure illustrating that SOX2 ectopically activates JAK–STAT signaling, which enables the transition of mCRPC to a stem-like and multilineage state. Figure created with BioRender.com. The numbers (n) of cells in each sample and cluster are the same as in Fig. 6.

We continued to explore the lineage heterogeneity of the subclusters of cluster 4 (Fig. 6f and Extended Data Fig. 10g), which contain cells originating from the sgTP53/RB1/JAK1 + Enz group (Fig. 6e,g). The three subclusters of cluster 4 expressed diverse levels of the JAK–STAT signaling genes (Extended Data Fig. 10i–r), presumably because JAK–STAT signaling was not fully deactivated in a proportion of JAK1-KO cells due to compensatory signaling driven by JAK2 (Extended Data Fig. 10j). Cluster 4-3 contained the ‘outlier’ cells, which partially maintain JAK–STAT signaling, likely driven by JAK2 (Figs. 6f and 8i and Extended Data Fig. 10j). Remarkably, the cells within cluster 4-3 maintained expression of multilineage programs as well as the highest level of cell fitness, regardless of treatment conditions (Fig. 8i–k and Extended Data Fig. 10h). The other two subclusters of cluster 4 demonstrated two contrasting fates following deactivation of JAK–STAT signaling. Cluster 4-1 cells, which lose the multilineage and stem-like programs, restored the exclusive expression of the luminal program (Fig. 8j). Thus, cells of this subcluster were highly responsive to Enz (Extended Data Fig. 10h), which caused a substantial diminishment in cell fitness (Fig. 8i). By contrast, the cells of cluster 4-2, which exclusively express NE-like lineage programs, maintained cell fitness even in the absence of JAK–STAT signaling (Fig. 8l and Extended Data Fig. 10h), supporting the hypothesis that JAK–STAT signaling is not required for the cells fully differentiated to an NE-like state. The juxtaposition between different subclusters of cluster 4 further supports the crucial role of ectopic JAK–STAT signaling in maintaining AR therapy resistance of stem-like and multilineage subclones rather than subclones exclusively expressing an NE-like lineage (Fig. 8m).

Leave a Reply