Understanding the role of Pax5 in development of taxane-resistant neuroendocrine like prostate cancers

Resistance to the current Androgen Receptor Signaling Inhibitor (ARSI) therapies has led to higher incidences of therapy-induced neuroendocrine-like prostate cancer (t-NEPC). This highly aggressive subtype with predominant small cell-like characteristics is resistant to taxane chemotherapies and has a dismal overall survival. t-NEPCs are mostly treated with platinum-based drugs with a combination of etoposide or taxane and have less selectivity and high systemic toxicity, which often limit their clinical potential. During t-NEPC transformation, adenocarcinomas lose their luminal features and adopt neuro-basal characteristics. Whether the adaptive neuronal characteristics of t-NEPC are responsible for such taxane resistance remains unknown. Pathway analysis from patient gene-expression databases indicates that t-NEPC upregulates various neuronal pathways associated with enhanced cellular networks. To identify transcription factor(s) (TF) that could be important for promoting the gene expression for neuronal characters in t-NEPC, we performed ATAC-Seq, acetylated-histone ChIP-seq, and RNA-seq in our NE-like cell line models and analyzed the promoters of transcriptionally active and significantly enriched neuroendocrine-like (NE-like) cancer-specific genes. Our results indicate that Pax5 could be an important transcription factor for neuronal gene expression and specific to t-NEPC. Pathway analysis revealed that Pax5 expression is involved in axonal guidance, neurotransmitter regulation, and neuronal adhesion, which are critical for strong cellular communications. Further results suggest that depletion of Pax5 disrupts cellular interaction in NE-like cells and reduces surface growth factor receptor activation, thereby, sensitizing them to taxane therapies. Moreover, t-NEPC specific hydroxymethylation of Pax5 promoter CpG islands favors Pbx1 binding to induce Pax5 expression. Based on our study, we concluded that continuous exposure to ARSI therapies leads to epigenetic modifications and Pax5 activation in t-NEPC, which promotes the expression of genes necessary to adopt taxane-resistant NE-like cancer. Thus, targeting the Pax5 axis can be beneficial for reverting their taxane sensitivity.


Introduction
Hormone sensitive recurrent or metastatic prostate cancer is preferentially treated with androgen deprivation therapies (ADT) alone or in combination with androgen receptor signaling axis inhibitors (ARSIs) such as abiraterone acetate or enzalutamide (1,2).However, over the time, most of these patients become resistant to ARSI therapies and progress to castration-resistant prostate cancer (CRPC) (1,3).
CRPC is highly heterogeneous but still predominantly adenocarcinoma in nature, which are generally, treated with rst line taxane-based chemotherapies following ARSI resistance (4-6).However, ~ 20% of these ARSI-resistant cases showed neuroendocrine like transformation [therapy-induced neuroendocrine cancer (t-NEPC) or neuroendocrine-like cancer (NE-like)], and often presented as small-cell like neuroendocrine characteristics (7)(8)(9)(10).Similar to de novo NEPC, t-NEPCs are highly aggressive, readily metastasize to visceral organs and share common features including expression of neuroendocrine speci c genes (10)(11)(12).However, t-NEPCs are resistant to taxane-based therapies and are preferentially treated with platinum-based drugs in combination with etoposides (13)(14)(15).Compared to well tolerated taxanes, these platinum-based therapies yield multiple systemic toxicities and fail to show much improvement on overall survival of t-NEPC patients (13,14).Currently, various clinical trials are ongoing to test the e cacy of targeted therapies; however, results are mostly inconclusive (16).In this background, identi cation of key molecular regulator/s or pathways responsible for taxane resistance of t-NEPC will be important for future therapeutic optimization.
Studies have shown that functional aberrations of tumor suppressors RB1 and/or TP53 are important for NE-like trans-differentiation; however, why losses of RB1 and TP53 induce t-NEPC differentiation, remains poorly understood (17,18).While studying the differential characteristics between CRPC-adenocarcinoma and t-NEPCs, we found that t-NEPCs selectively upregulate genes related to the neuronal pathways (19).Whether these neuronal phenotypes contribute to the taxane resistance of t-NEPC is unknown and demands clari cation.
To replicate the trans-differentiation processes, we previously generated and characterized various t-NEPC models from established CRPC cell lines (19).In this study, we have identi ed Pax5 as an important transcriptional regulator involved in various neuronal pathways.Pax5 expression is highly speci c to NE-like cancer, and depletion of Pax5 abrogates neuronal characteristics of t-NEPC.The present study highlights the importance of Pax5-mediated development of neuronal characteristics towards taxane resistance in t-NEPC.Furthermore, we discovered the importance of t-NEPC-speci c epigenetic modi cation-based transcriptional events in Pax5 expression.Overall, the current work emphasizes the role of Pax5 transcriptional signature as a crucial element in neuronal gene expression linked to therapy-resistant prostate cancer.

NE-like transformation adapts to a speci c gene signature
To understand how neuroendocrine prostate cancer functionally differs from adenocarcinoma, we analyzed published RNA-seq data derived from patient tissues (GSE126078, GSE66187, SU2C (20)).We used their pre-de ned classi cations, which incorporate an NE-score, marker expression, and pathology to compare patients consistent with t-NEPC to that of adenocarcinoma (details in materials and methods).
Next, differentially expressed genes were analyzed through Gene Set Enrichment Analysis (GSEA), which demonstrated that pathways related to secretion, synapse assembly, and neuronal signaling (Fig. S1A) are preferentially upregulated in the t-NEPC cases.
To investigate the signi cance of t-NEPC-speci c neuronal gene signatures, we have developed various cell lines, which can replicate neuroendocrine characteristics.Earlier we showed that these cell lines express neuroendocrine markers such as synaptophysin (Syp) and chromogranin (CHGA) and lack AR expression (Fig. S1C from (19).They are referred to as C4-2BER (by continuous exposure of adenocarcinoma cell line C4-2B to enzalutamide) and DKD (depleting RB1 along with TP53 from adenocarcinoma cell line LNCaP C4-2) respectively (19).We previously demonstrated that enzalutamide-resistant C4-2BER cells lose the expression of RB1 and TP53, which are the main factors behind the NElike transition (19).
To understand whether the cell lines we developed feature the t-NEPC like characteristics, we have initially carried out RNA-seq (in triplicate) between C4-2B and C4-2BER.Our results suggested that 6,632 genes are signi cantly upregulated and 1266 were downregulated in C4-2BER as compared to C4-2B (Fig. 1A).Further studies revealed that an increase in REST-repressed neuroendocrine gene signatures (Type I genes such as CHGA, SYP, SNAP25, CHRNB2 and SRRM4) and transcriptional regulators for t-NEPC differentiation (Type II gene sets such as Sox2, NKX2.1, POU3F2, etc.) (Fig. 1B, rst panel) is highly upregulated in C4-2BER as compared to its parental adenocarcinoma cell line C4-2B.These neuroendocrine speci c gene-expression were comparable with t-NEPC patient gene expression data sets reported earlier (1,18).Similar results were also observed when we compared the RNA-seq between C4-2 (adenocarcinoma cell line) and DKD (t-NEPC) (Fig. 1B, second panel) [we have reported the RNA-seq in our earlier publication (19)].Moreover, our derived neuroendocrine cell lines also express neuroendocrinerelated markers (19) such as beta III tubulin (19) and NeuN (Fig. S1B).On the contrary, AR expression (Fig. S1C) as well as AR driven gene signature such as expression of KLK3, or Nkx3.1 decreased in C4-2BER and DKD (representing t-NEPC) cells as compared to adenocarcinoma C4-2B and C4-2 cells respectively (Fig. 1C).Further, using RT-PCR, we validated that other AR regulated genes such as KLK2 and TMPRSS2 are also down-regulated in C4-2BER, DKD and NCI-H660 (a representative de novo neuroendocrine cell line from ATCC), and some of the features of t-NEPC have high similarity with this de novo neuroendocrine cancer cell line (Fig. S1D).These results indicated that our developed cell lines have acquired NE-like features.
To determine whether our NE-like cell lines exhibited neuronal features comparable to those seen in t-NEPC patients (Fig. S1A), we rst examined the common genes that were differentially regulated in both C4-2BER and DKD cell lines.We identi ed 4,560 common differentially expressed genes between C4-2B vs C4-2BER and C4-2 vs DKD respectively (Fig. 1D), which were used for pathway analysis using g-Pro ler (21), as well as gene ontology (GO) over representation analysis (ORA) and IPA (Fig. 1E, Fig. S1E-H).Interestingly, pathway analysis using gene expression pro le revealed that, similar to t-NEPC patients (Fig. S1A), C4-2BER and DKD show enrichment of neuronal pathways (Fig. 1E, Fig. S1E-H) related to neuronal adhesion, secretion, exocytosis, and axonal guidance.Overall, these ndings indicate that our NE-like cells acquired neuronal behavior, which could be important for their AR-independent growth and survival.
Morphologically, our developed NE-like cell lines showed neurite like protrusion while growing in 2D attachment culture [(Fig.S1C from Islam et al. 2022) and (19)].Interestingly, these derived cell lines grow partly in attachment and partly in suspension (similar to NCI-H660 or SCLC cell lines) and these suspended cells form strong cellular aggregates (Fig. S1I).Reports suggest that during small cell neuroendocrine transformation, cells form spherical aggregates and appear as crowded morphology (22)(23)(24).Whether such small cell characteristics are associated with enhanced therapy-resistance, has not been studied.Earlier, we showed that, similar to neuroendocrine like cancers, DKD and C4-2BER cells are highly resistant to taxane-based chemotherapies such as docetaxel as compared to its parental adenocarcinoma lines C4-2 and C4-2B (19).Here we predicted that neuronal adaptation is crucial for establishing strong cellular contact, thereby displaying a crowded phenotype which is associated with taxane resistance in clinically aggressive t-NEPC.Below, we will investigate how these t-NEPC cells developed neuronal-like traits and whether any key transcriptional events are associated with such morphological features.

Increased chromatin accessibility and histone acetylation induce transcriptional activation during NE-like transformation
Reports indicate that during NE-like transformation, adenocarcinoma cells undergo a series of chromatin modi cation and epigenetic alterations to enhance neuroendocrine-related gene expression (25,26).To understand whether our derived cell lines follows such characteristic changes in their DNA, we have carried out ATAC-Seq as well as histone ChIP-seq.Chromatin accessibility near promoter sites is an important component of transcriptional activity regulation in a cell (27).To understand whether differential upregulation of gene expression can be explained by altered chromatin accessibility, we performed assays for transposase-accessible chromatin with sequencing (ATAC-seq) in C4-2BER and compared it with C4-2B.As ARSI resistance is the major factor driving NE-like transformation (2, 28), we investigated chromatin alteration during NE-differentiation in C4-2BER cells using high throughput sequencing and compared them to adenocarcinoma C4-2B cells.We examined ATAC-seq signal near proximal promoters (+/-2kb of the transcription start sites [TSSs]) of upregulated genes de ned by pvalue < .05 and at least a 4-fold change in expression.Our ATAC-seq results display a marked increase in chromatin accessibility around upregulated genes' TSSs in C4-2BER compared to C4-2B (Fig. 2A).In contrast, we see slightly decreased accessibility for downregulated genes in C4-2BER (Fig. 2B).This pattern of altered accessibility explains the potential of C4-2BER to maintain differential transcriptional activity during NE-like differentiation from adenocarcinoma.For example, increased chromatin accessibility is evident across the Hox A locus corresponding to the increased expression of Hox A genes (Fig. 2C, Fig. S2A).We detect similar accessibility changes near the TSS of genes that encode various cell adhesion genes including neuronal adhesion proteins, such as NCAM1, VCAN, CD40 which coincides with their higher expression in C4-2BER (Fig. S2B-C).Interestingly, earlier studies have shown that overexpression of Hox genes or NCAM1 is linked with either a NE-like transformation or the development of AR signaling inhibitor-resistant prostate cancer (11,(29)(30)(31).
To further explore what drives the chromatin accessibility for these t-NEPC gene signatures, we examined histone acetylation levels, which are commonly used as marks of transcriptionally active chromatin (32).Our results indicate a global increase in acetylation of histone H3 at lysine 9, lysine 18 and lysine 27 in C4-2BER compared to its adenocarcinoma (Fig. 2D).Similarly, we detected increased H3K27Ac and H3K18Ac in DKD when compared to its parental line C4-2 (Fig. S2D).To evaluate the in uence of acetylated histones on genome-wide chromatin accessibility, we performed chromatin immunoprecipitation with sequencing (ChIP-Seq) for H3K27Ac and H3K18Ac in C4-2BER and C4-2B respectively.Consistent with activated chromatin, we found signi cant increases in H3K27ac and H3K18ac at C4-2BER accessible regions compared to parental C4-2B (Fig. 2E).To decipher the link between accessible chromatin sites and active histone acetylation marks, we determined the overlap of increased H3K18Ac and H3K27Ac ChIP-Seq peaks at C4-2BER ATAC-seq peaks.Indeed, we identi ed 7,016 accessible regions that coincide with increases in both H3K27Ac and H3K18Ac (Fig. 2F).
Examining our RNA-seq data, we found that these accessible regions correspond to increased expression of nearby genes such as CHGA, ASCL1 (Fig. 2G, S2E).Altogether, our ndings suggest that a chromatinbased activation of putative promoters can explain the gene expression signature speci c to t-NEPC compared to adenocarcinoma.

Pax5 expression increases following neuroendocrine transdifferentiation
Chromatin accessibility governs the ability of transcription factors to bind to their target loci, thereby controlling the transcriptional output of a cell.To explore the preferential binding of potential transcriptional regulators near the differentially exposed gene promoters in t-NEPC, we selected the 7,016 peaks (from Fig. 2F) that represent highly accessible promoters (from ATAC seq signal), with active histone marks H3K27ac and H3K18ac.These peaks correspond to promoters of transcriptionally active genes with increasing expression (from Fig. 2G) in C4-2BER.Using MEME-ChIP in combination with TOMTOM motif comparison tool, we identi ed an array of transcription factor motifs in these highly accessible promoters of differentially upregulated genes in C4-2BER (Table S1).Based on P-value signi cance score, we further narrowed down to the top 10 transcription factors (Table S1).Further, we examined the expression of these TFs among t-NEPC patient cohorts, which showed consistent upregulation of Pax5, ETV5, and KLF12 (Fig. S3A).Among these, Pax5, a lineage-speci c transcription factor, showed consistently high expression between our different cell line models (Fig. 3A, B) and has been associated with neuronal gene activation (33,34).Using RT-PCR, we have validated that compared to ETV5 or KLF12, Pax5 is highly expressed in DKD cells (Fig. 3C).Indeed, TOMTOM motif analysis of sites with increased chromatin accessibility (from ATAC-Seq) carrying H3K18ac and H3K27ac footprints within the TSS +/-1000bp in C4-2BER reveals a motif (P=1.30e-09)similar to the Pax5 recognition motif (Fig. 3D).We have tested Pax5 motif in our cell lines and found that its target genes such as TNC or DAB1, which are highly expressed in NE-like cells, have increased promoter chromatin accessibility and histone acetylation compared to adenocarcinoma cells (Fig. S3B, C).Together, these results suggest that increased accessibility and active chromatin marks at Pax5 binding sites near NE-speci c gene promoters is associated with t-NEPC transformation.
To explore Pax5 expression, we analyzed RNA and protein expression in the NE cell lines that we have developed.Our results show that Pax5 is preferentially expressed in DKD and C4-2BER as compared to adenocarcinoma counterpart C4-2 and C4-2B respectively (Fig. 4A, B, Fig. S4A, C). t-NEPC shares very similar characteristics and gene expression pro les with de novo neuroendocrine cancers (35,36).To understand whether Pax5 expression is speci c to t-NEPC or it appears during neuroendocrine differentiation, we have also analyzed the Pax5 expression in de novo small cell neuroendocrine NCI-H660 cell line.Similar to our NE cell lines, NCI H660 does not express AR and expresses all the classical NE-like markers (35).Our results indicate that NCI-H660 also expresses Pax5 (Fig. 4C, Fig. S4B).Further, using apalutamide (another ARSI) resistant NE-like C4-2B cells (known as C4-2BAR), we further validated that ARSI-resistant NE-like differentiation associates with high Pax5 expression (Fig. 4D).From the above evidence, our results indicates that Pax5 expression is mainly associated with neuroendocrine like transformation or more speci cally loss of adenocarcinoma characteristics in CRPC cells.
Earlier, Ku et.al. showed that mice develop metastatic prostate cancer by genetically knocking out prostate-speci c Trp53 and RB1 genes (37).Some of these metastatic loci showed neuroendocrine differentiation and often presented with low luminal keratin with high NE marker expression (Krt8low:Syp-high:AR-low). On the other hand, adenocarcinoma had higher luminal keratin and AR expression with low expression of NE-associated genes (represented as Krt8-high:Syp-low: AR-high).They studied the overall gene expression of various metastatic foci to identify the differential genetic signature associated with the development of metastatic neuroendocrine and adenocarcinoma (GSE90891).By reanalyzing these RNA-seq expression data, we found that Pax5 expression is higher in metastatic cancer with neuroendocrine differentiation (Fig. S4D).Moreover, we observed an increase in Pax5 expression in mouse-derived cell lines (Hi-Myc/PTEN / / Trp53 R172H/+ /Pb-Cre4 + mice) following depletion of RB1 (Fig. 4E), which is also concurrent with high CHGA expression.Therefore, these results suggest that depletion of RB1 in TP53 knockout background induces NE-like characteristics, which is associated with Pax5 expression.
Finally, we validated Pax5 expression in tissue microarrays (TMAs) derived from prostate cancer patient derived xenografts (PDX) (LuCaP series) and from metastatic CRPC (mCRPC) tissues obtained from Prostate Cancer Biorepository Network (PCBN).The detailed characterization of these PDXs have been described elsewhere (38).The LuCaP TMA from PDX contain 42 patient tissues (24 are from adenocarcinoma, 13 from CRPC, 4 patients are from t-NEPC and one AR null, NE null prostate cancer group often referred as double negative stage) in triplicate.We validated the nuclear expression of Pax5 (Fig. 4F), in LuCaP PDX with t-NEPC.LuCaP derived from CRPC-adenocarcinoma stained negative for Pax5 expression, whereas t-NEPC LuCaP PDXs (LuCaP 93, LuCaP 145.1 LuCaP 145.2, and LuCaP 173.1), expressed high Pax5 although LuCaP 173.1 showed lower expression compared to others.These results once again indicate that, overall, Pax5 expression is associated with NE-like transformation.
Similar to LuCaP models, we also analyzed Pax5 expression in the TMA derived from the metastatic PCa patients [Prostate Cancer Biorepository Network (PCBN].This TMA contains 70 visceral metastasis tissues from liver, lungs, lymph node and kidney as well as 51 bone metastatic cores from 45 metastatic CRPC (mCRPC) cases following rapid autopsy procedure.Of these, 13 patients showed t-NEPC transformation and the rest were classi ed as mCRPC adenocarcinoma.Our results indicate that 10 out of 13 t-NEPC patients stained positive for Pax5 expression (Fig. 4G).Again, Pax5 expression is not detected in metastatic CRPC (mCRPC) cases diagnosed with adenocarcinoma, but were present in those with t-NEPC, which supports that Pax5 expression is speci c to the NE-like lineage differentiation of PCa.
Earlier reports indicate that Pax5 is one of the most important transcription factors for AR-independent cell growth (12), further supporting our ndings.Pax5 is also expressed by in ltrating leukocytes surrounding the tumor (39).To con rm that high Pax5 expression in tumor tissue actually originates from the epithelial cells/NE-like cancer cells, we further analyzed single cell RNA-sequencing data published recently (GSE137829) (40).Our analysis validated that adenocarcinoma does not usually express Pax5, and cancer epithelial cells start expressing Pax5 only during NE-like transformation (Fig. S4E).Overall, our results validate that Pax5 is preferentially expressed during the AR-independent NE-like cancer progression.

Pax5 is involved in the gene expression pro le associated with neuronal pathways
To identify Pax5-related genes in NE-like cancer, we performed RNA-Seq under Pax5 depletion in C4-2BER and DKD cells.The resultant heat map indicated signi cantly differential genes between control and Pax5-depleted condition in both t-NEPC cell lines (Fig. 5A).GSEA pathway analysis of signi cant differentially regulated genes (from Fig. 5A) revealed that Pax5-regulated genes are involved in neuronal synapses and neuronal adhesion signatures in NE-like cells (Fig. 5B).Using qPCR in DKD and C4-2BER t-NEPC cell lines, we validated those various genes, such as NFASC, NrCAM, GRID1, SMARCA, etc., involved in neuronal pathways from Fig. 5B are downregulated upon depleting Pax5 expression (Fig. 5C-F).Further, to determine the occupancy of Pax5 within the promoters of these genes, we performed ChIP-qPCR, which revealed that Pax5 binds to the regulatory regions of those genes (Fig. 5G, S5A).Similar increase of gene expression has been reported in t-NEPC patients as compared to adenocarcinoma patients (1,41).Interestingly, ectopic expression of Pax5 alone in adenocarcinoma cells did not transduce these cells into NE-like phenotype (Fig. S5B-C).We reasoned that t-NEPC transformation from adenocarcinoma, under RB1/TP53 functional inactivation background, undergoes chromatin alterations to allow Pax5 binding to neuronal gene promoters.Thus, ectopic expression of Pax5 alone without any chromatin modi cation is not su cient to induce NE-like transformation.Overall, our results suggest that transcriptional accessibility enables Pax5 to upregulate neuronal adhesion signatures in t-NEPC.

Depletion of Pax5 affects cellular communication and increases therapeutic e cacy
While growing in 2D culture, NE-like cells present various neurite-like branching (Fig. S6A, B).These neurite-like branching are known to establish cell-cell interaction networks (42).Depleting Pax5 in NE-like C4-2BER cells reduced such neurite-like branching (Fig. S6A, B), thereby suggesting the importance of Pax5 in regulating cell-cell interaction in t-NEPC.Adhesion proteins play a vital role for such cellular interaction (43,44).Therefore, we predicted that adhesion proteins, especially Pax5-regulated neuronal adhesion signatures in t-NEPC cells might be important for maintaining cell-cell interaction.In neurons, neuronal adhesion proteins establish the communication between pre-and post-synaptic neurons or other surrounding cells to stabilize synaptic transmission by mechanically strengthening synapse formation (45,46).Therefore, by maintaining such trans-synaptic communication, these adhesion molecules play a vital role in healthy neuronal activity (47).Similar to synaptic communication, earlier, we showed that by maintaining exocytosis activity, NE-like cells communicate with each other (19) to enhance their therapy-resistance characteristics.Whether this cell-cell interaction is important for resisting therapeutic stress in t-NEPC, remains elusive.Hereby, we investigated whether Pax5-regulated adhesion function is important for enhancing therapy resistance in t-NEPC.To delineate the functional relevance of Pax5 expression in NE-like cancer, we studied cell-cell interaction pathway.Using a 3Dmatrigel culture system, we found that Pax5 depletion disperses the cellular cluster and inhibits cell-cell interaction in t-NEPC cells (Fig. 6A).
To understand whether this observed decrease in cellular interaction is due to loss of cellular adhesion properties, we studied cell-cell interaction of NE-like cells following depletion of Pax5.Herein, we stained for NCAM1 (CD56), a NE-speci c cell surface protein, as a read out of cell-cell interaction (48, 49).Alteration of NCAM1 surface localization will be a determinant for the extent of cell-cell interaction.We observed that depleting Pax5 disrupts the adhesion junctions (represented by NCAM1 in green) responsible for holding the cells in a cluster (Fig. 6B-F, Fig. S6C-F).Interestingly, depletion of Pax5 did not alter NCAM1 expression (Fig. 6D, Fig. S6G) but decreased the cell surface localization of NCAM1 (Fig. 6B-F, Fig. S6C-F).This indicates that Pax5 depletion decreases the cellular adhesion properties and thereby inhibits the cell-cell interaction in t-NEPC.Interestingly, ectopic expression of Pax5 following its depletion, rescues the cell surface distribution of NCAM1 and thereby restores the NE-like cell-cell interaction (Fig. 6B-F, Fig. S6C-F).Depletion of Pax5 is thus essential for inhibiting cell-cell interaction by decreasing cellcell contact adhesion in t-NEPC, which is re-established upon Pax5 over-expression (Fig. 6G).Using surface biotinylation assay followed by immunoblot, we additionally validated that depletion of Pax5 decreases the cell surface localization of NCAM1 without affecting its total protein levels (Fig. 6H).
Overall, our results imply that NE-like cells maintain cell-cell interaction through cell-cell contact adhesion, which is inhibited upon Pax5 depletion.
To understand whether such Pax5-induced cell-cell interaction evades therapeutic stress in NE-like cells, we have performed a cell death assay by exposing these NE-like cells to the 1st line chemotherapies.Here we hypothesized that decreasing cell-cell contact enhances the drug accessibilities.Earlier, we showed that these t-NEPC cells are highly resistant to the 1st line chemotherapy such as docetaxel (19).Our results showed that Pax5 depletion sensitized these aggressive NE-cells towards docetaxel (Fig. 6I, Fig. S6H, I).These results suggest that Pax5-mediated cell-cell interaction maintains the cellular survival axis through stabilizing cell-cell contact and therefore, altering such cell-cell interaction can restore the chemosensitivity of NE-like cells.
To understand why such cellular interaction enhances drug-resistance characteristic of t-NEPC, we therefore analyzed the downstream survival signaling axis.Our results indicate that depletion of Pax5 decreases AKT phosphorylation (Fig. 6J, S6J-K), which is an important intermediate to promote therapyresistance.AKT signaling is hyper-activated in neuroendocrine cancers (50-52), and we found the similar results in our cell lines, where we showed that neuroendocrine cells maintain a higher AKT activation as compared to adenocarcinoma cells (Fig. S6L).As AKT activation is mostly dependent on the growth factor receptor mediated signaling axis, therefore, we analyzed growth factors receptor localization in the surface.Here we hypothesized that by maintaining cell-cell interaction (maintained by adhesion proteins), NE-like cells retain the surface localization of growth factor receptors for a prolonged period, which can induce downstream signaling activity.To test that, we have stained the NE-like cells with growth factor receptors such as EGFR and MET.In neuroendocrine cancer, both MET and EGFR mediated signaling are highly activated, which is responsible for the adversity of the disease (53-56).Our results showed that Pax5 depletion decreases the cell surface localization of tyrosine kinase receptor EGFR and MET (Fig. 6K,  L).Cell surface localization of these growth factor receptors allows them to interact with their ligand, and hence, they promote downstream AKT signaling.This loss of growth factor receptors from the cell surface upon Pax5 depletion, decreases AKT phosphorylation and therefore, diminishes the survival signature of t-NEPC cells.Hereby, depletion of Pax5 not only affects cell-cell interaction, but also inhibits growth factor receptor mediated t-NEPC survival.

Pbx1 regulates Pax5 expression in NE-like prostate cancer
Pax5 expression is highly selective for NE-like cancer but not for prostate adenocarcinoma.This selective expression of Pax5 urges us to investigate its transcriptional regulatory mechanism.Therefore, we next investigated into the molecular events that lead to the preferential upregulation of Pax5 in NE-like cells.During NE-like differentiation, cells become independent of the AR signaling axis.Therefore, we questioned whether depletion of AR is su cient for preferential upregulation of Pax5 in NE-like cells or whether AR depletion is essential for additional modi cation, to induce Pax5 expression.Our results showed that immediate (5-7 days) inhibition of AR nuclear translocation by addition of AR-antagonist, enzalutamide, or apalutamide, did not induce Pax5 expression in the NE-like cells (Fig. S7A, S7B).This suggests that short-term depletion of AR activity is not su cient for Pax5 expression.Interestingly, while studying the Pax5 promoter regions, we found enhanced histone acetylation nearby the Pax5 promoter region of NE-like cells (C4-2BER) as compared to the adenocarcinoma cell line (Fig. 7A).Additionally, our in vitro ATAC-seq data validated that these loci are more accessible in NE-like cells as compared to adenocarcinoma (Fig. S7C).Overall, these data indicated that the Pax5 promoter is transcriptionally active in NE-like cells, and loss of functional AR is not an immediate driver of Pax5 expression.
To understand what drives Pax5 expression in NE-like cells, we analyzed the consensus TF-binding motifs at the transcriptionally active Pax5 promoter.Using the Biobase-transfac Gene regulation database and Transfac TF screening tool (Qiagen), we screened prospective transcription factor motifs within the Pax5 promoter (+500 to -2000bp) (Fig. S7D).Based on the scores of MATCH, we selected the top 10 transcription factors (Fig. S7E).After carefully comparing those top 10 TFs with t-NEPC patients' gene expression retrieved from GEO and SU2C databases, we found that Pbx1 shows a consistently higher overexpression speci cally in t-NEPC cohorts (Fig. S7F); however, other TFs such as MYB, Pax2 or HoxA3 are not exclusively over-expressed in all the patient databases as shown in Fig. S7F.Moreover, Pbx1 overexpression is consistent with upregulated Pax5 expression speci c to t-NEPC (Fig. S7F).In line with the above ndings in patient databases, Pbx1 is also upregulated in our NE-like cell lines (Fig. 7B, C).Indeed, the increased ATAC-seq signal in C4-2BER cells upstream of Pax5 overlaps with the Pbx1 motif (Fig. S7C), suggesting that activity of this Pbx1 binding may control Pax5 expression (Fig. S7F).
To investigate whether Pbx1 regulates Pax5 expression, we depleted Pbx1 from NE-like PCa cells and observed a decrease in Pax5 expression (Fig. 7D-F).Using ChIP-qPCR, we tested whether Pbx1 binds at the differentially accessible promoter region of ATAC-seq peaks in the NE-like cells.Our result showed that Pbx1 occupancy is highly enriched upstream of the Pax5 TSS (within -1400 to -1600bp) (Fig. 7G).These results indicate that Pbx1 acts as a putative regulator of Pax5 gene expression in NE-like cells.
Although PBX1 expression increases signi cantly from adenocarcinoma towards t-NEPC transformation, we questioned to understand what favors differential recruitment of PBX1 in Pax5 promoter in NE-like cells but not in adenocarcinoma.Interestingly, the Pbx1 binding site (ATAATTACT) falls within two wellconserved CpG islands within the Pax5 promoter.To determine whether these CpG islands have any potential effect on differential chromatin accessibility (as shown in ATAC-seq peaks analysis from Fig. 7A) between the adenocarcinoma and NE-like PCa, we investigated into the Pax5 promoter methylation status.To determine whether demethylation of CpG region favors chromatin accessibility at the Pax5 promoter, we analyzed the promoter methylation status of adeno and NE-like cells by performing EPIC Methylation Array between adenocarcinoma (C4-2) and NE-like cells (DKD).Our results indicate that in NE-like cells, the Pax5 promoter is heavily methylated compared to adenocarcinoma cells (Fig. 7H and Table S2) and therefore raised the question of how a methylated Pax5 promoter becomes accessible for Pbx1 binding.Recent studies have shown that hydroxymethylation can alter the chromatin compactness to a state that favors gene expression (57,58).However, bi-sul te treatment as performed in EPIC sequencing cannot distinguish between methylation and 5′-hydroxymethylation of cytosine (5hmC) (59).
Hydroxylation of 5-methyl cytosines is carried out by the Tet-family (Ten-Eleven Translocation) of enzymes (60).We therefore analyzed the Tet expression in the patient database and found that Tet2 expression is often upregulated in NE-like PCa (Fig. S7G).We measured protein expression of Tet2 in both adeno and NE-like PCa cell lines by western blotting and found that TET2 expression is increased in NElike cell lines compared to adenocarcinoma (Fig. 7I).To understand whether increased expression of TET2 correlates with hydroxymethylation at Pax5 promoter in NE-like PCa cells, we tested 5hmC level at Pax5 promoter by ChIP-qPCR using a 5hmC antibody.We found speci c enrichment of 5hmC peaks at Pax5 promoter CpG islands in NE-like cells (Fig. S7H).One of these 5hmC regions falls precisely overlaps with Pbx1 motif at Pax5 promoter (Fig. S7H, TSS -1400 to -1600bp).Interestingly, our results showed that the 5hmC footprint at the upstream Pbx1 binding site of Pax5 promoter is higher in NE-like cells than in adenocarcinoma (Fig. 7J).Overall, our results indicate that Pax5 promoter regions are differentially hydroxymethylated in NE-like cells, thereby, leading to chromatin relaxation and Pax5 expression.To validate that 5hmC can induce the Pbx1 binding at the promoter regions of the Pax5 promoter; we prevented hydroxymethylation by inhibiting Tet activity with a speci c Tet inhibitor, Bobcat339.We observed that inhibition of TET activity reduced Pbx1 binding at the 5hmC sites of the Pax5 gene (Fig. 7K), and reduced Pax5 expression (Fig. 7L).Alternatively, depleting Tet2 levels in NE-like DKD cells showed a similar reduction of Pax5 expression (Fig. S7I).Additionally, demethylation of the Pax5 promoter region (by treating with 5-azacytidine for a week) also decreases Pax5 expression (Fig. 7L); thereby indicating that hydroxymethylation of cytosine is pivotal in recruiting Pbx1 to induce Pax5 expression.Together, our results showed that the hydroxymethylation of cytosine within Pax5 promoter favors Pbx1 binding to initiate the Pax5 expression in NE-like PCa cells.
Overall, our results suggest that Pax5 expression in NE-like cancer is pivotal to maintaining the cellular interaction by strengthening the neuronal adhesion interaction in NE-like cells.Perturbation of such interaction renders the NE-like cells more sensitive to the existing 1st line chemotherapies (Fig. 7M).

Discussion
The use of new-generation anti-AR therapies may enhance the overall survival of CRPC cases; however, the resistant clones are highly aggressive and present in various clonal forms.Among these resistant populations, nearly 20% showed NE-like transformation with a distinct genetic signature compared to the adenocarcinoma sub-type (17).NE-like cancers are highly lethal and frequently responsible for the overspread of the cancers, which is di cult to manage.Although pathologically, these cancers are highly heterogeneous, they do follow some common genetic signatures (61, 62).Moreover, while the origin of therapy-induced trans-differentiation is still under investigation, recent reports indicate that, genetic signature of these cancers maintains distinct similarities with de novo neuroendocrine cancers, including small cell lung cancers (31).This suggests that a common mechanism underlines the clinical manifestation of NE-like PCa.In this study, by comparing chromatin accessibility, epigenetic signatures, and gene expression between adeno and NE-like cancers, we have identi ed Pax5 as a transcription factor important during NE-like transformation.
While analyzing the nature of gene expression associated with NE-like trans-differentiation processes, our results indicated that pathways associated with neuronal differentiation, axonogenesis, neuronal communications are selectively enriched in NE-like cancers, as reported elsewhere (12,62,63).Our results demonstrate that increased expression of neuronal genes in NE-like cancers is associated with increased marks of transcriptionally active chromatin, including accessibility and histone modi cations.While we focus on the chromatin activity status of promoters, it is likely that altered chromatin status at enhancer regions also contribute to t-NEPC transformation, and future work using Hi-C and HiChIP to understand enhancer-promoter interactions will likely reveal new insights in t-NEPC transformation.Currently, our results indicate that many differential genes have altered promoter chromatin status at Pax5 binding sites.Pax5 expression has been validated in various NE-cohorts, including PDX models.Further studies with metastatic TMA revealed that some of the neuroendocrine patients are negative for Pax5 expression; indicating either stage-speci c or clone-speci c expression of Pax5 associated with NE-like trans-differentiation.Similar to CRPC, Pax5 expression has been detected in neuroendocrine lung cancer as well as N-type neuroblastoma cells and reported to be associated with aggressive nature of cancers (64, 65).Interestingly, in neuroendocrine lung cancer, heterogeneous expression of Pax5 has been reported with the highest in the small cell (SCLC) subtype, followed by large cell and carcinoid(64).With the genetic similarities of SCLC and NE-like PCa (35,36), our results suggested similar heterogeneity in patient sample and thereby warrants more in-depth investigation with a larger cohort.The absence of Pax5 in any of the adenocarcinoma tissue indicates that Pax5 expression might be speci c transcriptional event in NE-subtype.However, IHC from metastatic TMA revealed that one core of LuCaP173.2A,which is represented as double negative prostate cancer (DNPC; AR negative with no expression of classical NE markers such as CHGA and SYP but express EZH2 and MYCN), also expresses a low level of Pax5.The report indicates that this LuCaP173.2with the serial passage starts expressing NE markers like SYP; therefore, specifying that such NE-differentiation might be a disease continuum from the double negative PCa (1).This LuCaP173.2thus serves as an intermediate of adeno and NE-like cancers (1).Therefore, detection of Pax5 in LuCaP 173.2A, further suggests that expression of Pax5 may be an early event of NE-like transformation following the loss of AR function.Recently, association of Pax5 has been predicted in AR-independent growth of CRPC (12).Therefore, Pax5 can be an early event of NE-like cancer and induce neuronal characteristics for the evasion of therapies.Moreover, another Pax5 subtype, known as Pax6, has been shown with prostate NE-like transformation (31).Overall studies suggest the importance of Pax-group of transcriptional regulators in the development of NE-like cancers.
While studying the underlying mechanism of Pax5 expression in NE-like cancer, we observed the involvement of Pbx1 in Pax5 transcription.Our results indicate that Pbx1 is selectively overexpressed in NE-like cancers, which was further validated in patient's gene expression data.Pbx1 expression has also been observed in xeno-transplanted neuroendocrine prostate cancer (66).Despite these connections, the functional importance of Pbx1 has not been investigated in t-NEPC.Our results presented here reveal the crucial importance of Pbx1 upregulation in prostate NE-like differentiation.Our ATAC-seq data shows NEspeci c chromatin accessibility at the Pbx1 binding site within the Pax5 promoter, along with hydroxymethylation of cytosine in CpG islands.It has been shown that high-5hmC is an adverse predictor for biochemical recurrence of ERG-negative prostate cancers(67) and can function as a prognostic marker for PCa development(68).Our study highlights the importance of examining 5hmC relative to gene expression in NE-like transformation.While we demonstrate 5hmC at the Pax5 promoter, future studies using genome-wide methods of measuring 5hmC will reveal the overlap between gene activation, chromatin accessibility, histone modi cation, and promoter 5hmC signatures in speci c subtypes of NElike trans-differentiation.
Our results indicate that Pax5 is involved in the transcription of speci c neuronal gene signatures in NElike cancers.As neuronal axis is one of the major axes for the prostate cancer growth and survival (following AR axis) (28, 69), inhibiting such axis can sensitize cancer cells to the therapy.This is especially important for NE-like cancer, which is highly dependent on the neuronal axis for its survival (70).In these regards, the identi cation of Pax5 as an important regulator for neuronal axis of PCa is highly signi cant.Depletion of Pax5 not only decreases some of the neuronal characteristics but also induces cells to respond to the existing therapies through modulating cell-cell contact.This neuronal adaptation is crucial for various cellular interaction pertaining to metastatic property of NE-like cells (71,72).Targeting NE-like PCa is challenging; therefore, identi cation of Pax5-mediated genes is necessary for the development of future therapeutic strategies.Recently, we published that NE-like cells are highly secretory in nature and this function is important for imparting broad chemotherapy-resistant to their micro-environment (19).Therefore, affecting such endocrine function of NE-like cells can be a key to inhibiting their aggressive behaviors.Our current results indicate that Pax5 controls the neuronal adhesion crucial for fundamental cellular processes such as, cellular interaction, cancer cell invasion, growth factor localization in the cell surface, etc. Thereby, depletion of Pax5 has a broad role to challenge cellular processes, which not only affects cellular communication but also sensitizes the cells towards therapies.
Overall, our results illustrate the functional importance of PBX1/Pax5 transcriptional axis in maintaining the NE-trans-differentiation process.This is signi cant in terms of understanding the disease etiology as well as screening or detection of neuroendocrine transformation.Understanding such heterogeneity has broad implications, especially in developing selective therapeutic strategies.In summary, our study provides a new avenue for screening NE-like prostate cancer.

Patient cohort for in-silico analysis:
RNA-Seq data of the following patient cohorts were selected from public domains for analysis.
GSE 126078: The study was conducted among metastatic castration resistant prostate cancer (mCRPC) patient specimens and patient-derived xenografts (PDX).The study identi ed subtypes with AR-low phenotype, amphicrine phenotype, double negative (AR-ve/NE-ve) phenotype and classical AR-ve/NE + ve phenotype through molecular pro ling.In the deposited RNA-seq fastq les, cases were already identi ed with the sub-classes.We used those subclasses for our analysis.
GSE 66187: The study was performed among 50 mCRPC patients and 24 LuCaP prostate cancer-derived PDX to characterize neuroendocrine (NE) phenotype among these mCRPC specimens.The study involved IHC staining criteria for androgen receptor (AR) and prostate-speci c antigen (PSA) expression, CHGA and SYP expression.Whole genome microarray, transcriptomics and IHC analysis were used to determine the NE phenotype.
GSE 137829: The study characterized the tumor-cell speci c diversity from 6 mCRPC patients through single-cell RNA-Seq (scRNA-Seq) analysis.4 of these mCRPC patients were identi ed as NE patients.SU2C-PCF: this is a multi-institutional prospective study, which involved a comprehensive genomic and transcriptomic pro ling among 429 patients.The study integrated the ndings of whole-exome, transcriptomics and histological analysis to provide a NE-like signature.All mCRPC patient cohorts who developed NE-like differentiation have received AR inhibitor treatments and represent treatment-refractory group.Based on the CHGA and SYP expression status, these entire cohorts have already subcategorized patients with adenocarcinoma and t-NEPC following treatment with second-generation AR signaling inhibition therapies and assigned with NE-score.The RNA-seq data was extracted from NCBI GEO using SRAtool Kit.The reads were then aligned to human reference genome (hg38 version) from UCSC genome browser and gene counts was quanti ed with HTSeq (v.0.9.1).Next, the raw counts were processed and normalized in DESeq2.Data processing was performed with the help of the Bioinformatics Core at UNMC.

Cell Culture:
Cells were cultured in RPMI 1640 with 10% FBS in presence of penicillin-streptomycin antibiotics (0.1%).Upon reaching con uency, these cells were washed with 1X DPBS and trypsinized with 0.25% (w/v) Trypsin-EDTA to detach the cells from the plate.The cells were collected in equal volumes of complete medium to neutralize the effect of trypsin and further centrifuged at 1000 g for 5 mins at room temperature.The pellet obtained from centrifugation was then resuspended in fresh complete media and plated in a T-75 ask and maintained at 37°C and 5% CO2 in a tissue culture incubator.C4-2BER cells were cultured under continuous presence of 10uM Enzalutamide (MDV 3100, Selleckchem, S1250).

RNA Extraction:
Total RNA was isolated by adding 1mL of TRIZOL Reagent (ThermoFisher Scienti c, CA) and allowed to stand for 5 minutes at room temperature.RNA isolation was performed with RNeasy Mini Kit (Qiagen, Germantown, MD) according to manufacturer's protocol.The RNA pellet was dissolved in UltraPureDNAse and RNAse free water (Life Technologies, 10977-015).The concentration and quality of the RNA were analyzed using Nanodrop Spectrophotometer.
Quantitative RT-PCR: and lysed with 26G.The lysate was collected after being centrifuged at 13,500 RPM for ve mins.The pellet was discarded, and the supernatant was used for protein analysis.Total protein estimation was carried out with Bradford reagent and the samples were prepared by the addition of SDS sample buffer containing β-mercaptoethanol and denatured at 95 ºC for ve minutes.The denatured samples were run on a precast 4-20% Mini-PROTEAN® TGX™ Gel (BioRad) and transferred onto a PVDF membrane (Life Technologies).The membrane was blocked in 3% bovine serum albumin (BSA) in 1X TBST (1X Tris Buffered Saline, 0.1% Tween-20) for at least 45 minutes.Primary antibody was diluted in blocking buffer and incubated with membrane overnight at 4ºC with continuous shaking at low speed.On the next day, the membrane was washed with 1X TBST for 3 times for 10 mins and incubated in appropriate dilution of secondary antibody conjugated with HRP for 1hr in 1X TBST with continuous shaking at low speed at room temperature.Following this, the membranes were washed in 1X TBST for 6 times for 10 minutes each wash to remove the excess secondary antibodies.The protein bands were detected using a combination dilution of SuperSignal™ West Femto Maximum Sensitivity Substrate and SuperSignal™ Pico Maximum Sensitivity Substrate captured on an X-ray lm.

Immunohistochemistry (IHC):
Immunohistochemical analyses were conducted on formalin-xed para n-embedded patient TMA as described before [Polavaram, 2021 #46].Antibodies used were provided in the table below.Tissue slides were scanned in UNMC Tissue Sciences Core Facility.Antigen retrieval was performed by heat induced epitope retrieval at 95°C using Dako antigen retrieval solution (pH 9).

Antibody (host) Dilution Manufacturer
Pax5 [EPR3730( 2)] (anti-Rabbit) 1:250 Abcam, ab109443 3D Matrigel culture and Immuno uorescence: Cells were transfected with siPax5 and Scramble as previously described.Cells were collected after trypsinization.For 3D Matrigel culture, 5X10 4 cells in 50µl were mixed gently with 50µl of ice-cold Matrigel Matrix Growth Factor Reduced Basement Membrane (Corning, #354230) and added onto 35mm uncoated glass bottom dishes (Mattek Corporation, P35G-1.5-14-C-GRID).The plates were incubated at 37ºC and 5%CO 2 incubator for 30 mins to solidify the Matrigel matrix.After 30-45 minutes, another 100 µl cold matrigel was added as a top-up layer and kept for solidi cation similarly.After 1-hour, complete media was added to cover the cells and incubated for 24-hrs at 37ºC and 5% CO 2 tissue culture incubator.
After 24-36 hrs incubation, cells were washed twice with 1X DPBS and xed with 4% paraformaldehyde for 15-20 mins.Cells were blocked in 0.2% saponin in 1% BSA prepared in 1X TBST and incubated overnight with respective primary antibodies diluted in blocking buffer.Cells were washed 2 times with 1X TBST for 5 mins each wash.Cells were incubated with HRP-conjugated anti-rabbit and anti-mouse secondary antibodies for 30 mins, respectively.Cells were next washed in 1X TBST and mounted with Vectashield mounting media containing DAPI (Vector Laboratories, H1200), and photomicrographs were captured using the confocal microscope.respectively.The graphical illustrations were made using GraphPad Prism 8 software.

5-Azacytidine Assay
Cells were seeded at a density of 0.1x10 6 on a 6-well plate.Cells were exposed to 5-azacytidine (at a dose of 0.1µM) and incubated for 6 days in tissue culture incubator.Cells were treated with 5uM Enzalutamide (AR inhibitor) to block the nuclear transport of AR.Cells were incubated in tissue culture incubator for 6 days.Cells were next collected and lysed.
Tet-activity inhibition: Cells were treated with Bobcat 339 at a dose of 50 uM (IC 50 = 33 uM for Tet1 and 73uM for Tet2) and incubated for 18-24 hrs in tissue culture incubator.After incubation, cells were subjected to ChIP assay using 5hmC and PBX1 antibodies respectively.
STAR was used to map RNA-seq paired-end reads to the hg38 human reference genome(76).StringTie was used to create count matrices and perform Transcript Per Million (TPM) normalization (77).DESeq2 was used to calculate differential gene expression and perform clustering of samples from each replicate.Genes with a padj value < 0.05 and a log2 fold change > = 1 or >= -1 were considered signi cantly differentially-expressed.EnhancedVolcano was used to generate a volcano plot comparing RNA expression between C4-2B and C4-2BER (https://github.com/kevinblighe/EnhancedVolcano).Gene ontology analysis of differentially expressed genes was performed using the web-based EnrichR gene-list enrichment analysis too(78)l.Gene Set Enrichment Analysis (GSEA) on the differentially expressed genes was performed using Webgestalt (79).The Canonical Pathways analyses were generated through the use of Qiagen Ingenuity Pathway Analysis(80). ATAC-Seq: viability and cell counting were performed using an automated cell counter (Biorad).Cells were centrifuged at 500g before proceeding with nuclei extraction.Nuclei were then isolated on an iodixanol gradient (The lysis was performed using 0.4% Igepal CA-630, 0.4% Tween 20 for 3 minutes) and counted after addition of trypan blue using automated cell counter (Biorad).50.000 counted nuclei were transferred to a new tube and were centrifuged at 500g before proceeding with the transposition reaction.Isolated nuclei were lysed and transposed for 30 minutes at 37°c using the prokaryotic Tn5 transposase system (Nextera DNA library kit, Illumina, FC-121-1030).Transposed DNA was then puri ed on Diapure columns (Diagenode, C03040001).After their ampli cation, the libraries were size selected and puri ed using Agencourt AMPure XP (Beckman Coulter) and quanti ed using Qubit dsDNA HS Assay Kit (Thermo Fisher Scienti c, Q32854).ATAC-Seq libraries were generated and then re-sequenced in paired-end mode 50 base pairs (PE50) on an Illumina NovaSeq 6000, running NovaSeq Control Software version 1.6.0.FastQC was used to determine the quality control of the sequenced reads.Nextera adaptors were trimmed from the reads using cutadapt -(https://cutadapt.readthedocs.io/en/stable/)a CTGTCTCTTATACACATCT -A CTGTCTCTTATACACATCT -m 20 -q 20 -pair-lter = both.Trimmed reads were aligned to the human reference genome (hg38 version) using Bowtie2 -X 2000 -no-mixed -nodiscordant(81).Reads were quality ltered based on mapQ > = 10 using samtools (82) and duplicate reads were removed using picard (https://broadinstitute.github.io/picard/).After deduplication, biological replicates were merged and peaks were called using macs2 -nomodel -shift 100 -extsize 200.Bigwig ATAC-seq signal tracks were created from the merged mapped reads using deeptools with BPM normalization(83, 84).

Differential promoter activity analysis:
BEDTools Intersect Bed was used to nd ATAC-seq peaks overlapping differential H3k27ac and H3k18ac peaks.ATAC-seq, H3K27ac ChIP-seq, and H3K18ac ChIP-seq signal was plotted across ATAC-seq peaks with differential or unchanged H3K28ac and H3K18ac ChIP-seq peaks(85).The ATAC-seq and ChIP-seq heatmaps were generated using deeptools.To determine differential transcription factor motif activity in C4-2BER, ATAC-seq peaks overlapping increased H3K27ac and H3K18ac peaks within − 2000 bp and + 500 bp of promoters was used for motif analysis.BEDTools getfasta was used to acquire the fasta sequence of ATAC-seq peaks with differential H3K18ac/H3K27ac and used as input for subsequent meme-ChIP and tomtom analysis (86-88).The human mononucleotide HOCOMOCO v11 motif collection was used for meme-ChIP and tomtom motif enrichment(89).Finally, the total list of enriched motifs from meme-ChIP and tomtom was further ltered for transcription factors upregulated in C4-2BER with a log2 fold change > 1 and a padj < 0.05.The Broad Institute Integrative Genomics Viewer was used to visualize ATAC-seq and ChIP-seq signal (90).

Transfac gene regulation database
The database (http://genexplain.com/transfac/)has been accessed and transcription factor selection for Pax5 promoter site was performed through MATCH tool as described (91).
ChIP-seq and ChIP-qPCR: assay was performed with Magnify ™ ChIP System (Invitrogen, Life Technologies,  according to manufacturer's guidelines.Cells plated on 100mm culture dish were washed twice with 1X DPS and scraped off.Cells were centrifuged at 1000 rpm for 5 mins at room temperature.Cells were then xed with 37% formaldehyde for DNA-protein crosslinking and incubated for 10 mins at room temperature.The crosslinking reaction was stopped by adding 1.25M glycine for 10 mins at room temperature.The cells were next centrifuged at 200g for 10 mins ta 4ºC.The pellet was further lysed for nuclear and cytoplasmic separation.The supernatant was discarded, and the pellet was lysed in hypotonic Buffer A (10mM HEPES pH 7.8, 10mM KCl, 2mM MgCl2, 0.1mM EDTA) with 0.5M DTT and 1% each of 0.1M PMSF, 10µg/µl Leupeptine, 10mg/ml Aprotinin protease inhibitors and Halt phosphatase inhibitor and incubated on ice for 17 mins.After incubation, the samples were vortexed mildly, and 10% NP-40 was added with a further incubation of 4-5 mins.This was further centrifuged at 1000 rpm for 5 mins at 4ºC.The supernatant containing the cytosolic contents was discarded and the nuclear pellet was further processed to yield quality DNA as per the manufacturer's guidelines.2-10ng of puri ed DNA was sent for sequencing.Sequencing was performed on an Illumina Novaseq 6000, running NovaSeq Control Software 1.6.0.Quality control of sequencing reads was performed using FastQC.Reads were aligned to the reference genome (GRCh38/hg38) obtained from the UCSC genome browser using bowtie2 version 2.3.4.1(81).Samples were ltered for regions blacklisted by the ENCODE project (92).Reads were quality ltered based on mapQ > = 10, and duplicate reads were removed using picard (https://broadinstitute.github.io/picard/).After deduplication, biological replicates were merged and peaks were called using macs2 (93).Differential ChIP-seq peaks were determined using manorm (94).
Bigwig ChIP-seq signal tracks were created from the merged mapped reads using deeptools with BPM normalization (83) For ChIP-qPCR analysis, we have develop the primers by analyzing the ChIP-Seq peaks of PBX1 (95)

Antibody
TMA for IHC analysis: LuCaP and mCRPC tumor microarrays were available commercially.These microarrays were obtained from the Prostate Cancer Biorepository Network (PCBN; LuCaP TMA number 90 A, B, C, D; mCRPC TMA number 92 A, B, C, and D).mCRPC TMA 92 contained tumor cores from multiple metastatic sites of 45mCRPC patients with known clinical diagnosis.A total of 15 out of 45 patients were clinically diagnosed with NE-like PCa.Pax5 expression in these TMA slides was evaluated by immunohistochemistry.For validation of Pax5 expression among t-NEPC patient cohorts, GSE126078, GSE 66187, Beltran 2016 and Stand Up to Cancer/Prostate Cancer Foundation) International Prostate Cancer Dream Team consortium, prospective clinical trial (identi er: NCT02432001) patient cohorts containing primary, mCRPC and NE-like PCa patient data were analyzed in-silico for Pax5 expression.
Cells were transfected with a scramble and Pax5 siRNA as mentioned earlier, and incubated for 24 hours at 37ºC and 5%CO2 tissue culture incubator.Cells were next treated with Docetaxel at various concentrations (2nM for C4-2BER and 10nM for DKD, respectively) and kept incubated for the next 18-24 hours.After incubation, cells were washed gently with 1XDPBS and incubated with PI (1:2000 dilution in PBS) for 10-15 minutes.Hoechst (1:2000 dilution in PBS) was used to stain the nucleus.Images were captured under 10X in UNMC confocal core facility.Surface biotinylation Assay:Cells were transfected with doxycycline inducible Pax5 shRNA for 48hrs.Control and shRNA transfected cells were carried out for labelling cell surface with biotin under cold conditions following the manufacturer's guidelines (Pierce Cell Surface Protein Isolation Kit, Thermo Scienti c #89881).The biotin-labelled proteins were eluted and analyzed by western blot.RNA-Seq:RNA Seq was performed for LNCaP C4-2, LNCaP C4-2 DKD, LNCaP C4-2B, LNCaP C4-2B-ER in triplicate condition.In addition we have carried out RNA-Seq following depletion of Pax5 by siRNA in DKD and C4-2BER cells.RNA from cells was isolated using the RNeasy Mini Kit (Qiagen, Germantown, MD) following the manufacturer's guidelines.The quality and integrity of RNA were con rmed using the Agilent Bioanalyzer.Paired end run of RNA libraries was carried out with Illumina NextSeq 500.Sequences were next aligned to the human reference genome (hg38 version) from the UCSC genome browser.Data analysis was performed with the help of the Bioinformatics Core at UNMC.Estimation of RNA abundance was carried out with feature Counts from the Sub-read package version 1.6.3.Downstream analyses were performed with the DESEQ2 R package version 1.18.1

Figure 4 Expression
Figure 4

Figure 6
Figure 6 Deamination of DNA is performed with the EZ-96 DNA Methylation Kit (Zymo Research) according to Illumina's guidelines.. Array Scan In nium Control BeadChips have been used which are equipped with a set of internal control probes.These control probes are used for identi cation of test samples with different data characteristics based on threshold parameters.These controls are also evaluated as per relative intensities.are The EPIC Array analysis has been done with GenomeStudio® Software 2011.1,Methylation Module v1.9 following the Illumina Methylation Module user guidelines (Controls The nium MethylationEPIC BeadChip assay, a genome-wide DNA methylation analysis technique has been performed with C4-2 and DKD cells.This array-based assay uses bisul te converted DNA and Illumina® technology to quantitatively detect the CpG island methylation level throughout the genome at a resolution of single nucleotide bases.