Enlibio,Your biological research partner
Xinqidi Biotech Co.,Ltd,Wuhan,China 2008-2022
R&D 14th year

ASCL1 activates neuronal stem cell-like lineage programming through remodeling of the chromatin land

Issuing time:2022-04-28 13:51

Abstract

Treatment with androgen receptor pathway inhibitors (ARPIs) in prostate cancer leads to the emergence of resistant tumors characterized by lineage plasticity and differentiation toward neuroendocrine lineage. Here, we find that ARPIs induce a rapid epigenetic alteration mediated by large-scale chromatin remodeling to support activation of stem/neuronal transcriptional programs. We identify the proneuronal transcription factor ASCL1 motif to be enriched in hyper-accessible regions. ASCL1 acts as a driver of the lineage plastic, neuronal transcriptional program to support treatment resistance and neuroendocrine phenotype. Targeting ASCL1 switches the neuroendocrine lineage back to the luminal epithelial state. This effect is modulated by disruption of the polycomb repressive complex-2 through UHRF1/AMPK axis and change the chromatin architecture in favor of luminal phenotype. Our study provides insights into the epigenetic alterations induced by ARPIs, governed by ASCL1, provides a proof of principle of targeting ASCL1 to reverse neuroendocrine phenotype, support luminal conversion and re-addiction to ARPIs.

Introduction

Potent androgen receptor (AR) pathway inhibitors (ARPIs), such as enzalutamide (ENZ) and abiraterone (Abi), have increased patient survival with advanced prostate cancer disease;1,2 however, resistance ultimately occurs. In particular, a subset of tumors shed their luminal identity and dependency on the canonical AR signaling. These variants exhibit lineage plasticity and neuroendocrine differentiation3,4,5,6,7,8 and are referred to as treatment-induced neuroendocrine prostate cancer (t-NEPC). While de novo NEPC is rare9,10, development of treatment-induced NEPC accounts for 20% of advance, treatment-refractory castration-resistant prostate cancer (CRPC)3,5,11. NEPC is characterized by loss of canonical AR signaling and expression of neuronal lineage markers, such as chromogranin (CHGA) and synaptophysin (SYP), distinct small cell morphology4,7,12, along with a stem cell transcriptional program13. Apart from alterations in RB1 and TP53, which have been associated with lineage plastic phenotypes14, CRPC and NEPC share relatively similar genomic landscape15,16. The evolution of CRPC to NEPC is accompanied by extensive transcriptional reprogramming12,15, suggesting that the emergence of a neuroendocrine phenotype may be driven predominantly by epigenetic dysregulation. The heterogeneous nature of prostate cancer17,18 provides possibility of multiple drivers for the transition of the lineage under the pressure of current therapeutic strategies. It is still unclear, mechanistically, how tumors govern variation in response to treatment and how they define alternative cell fate post ARPI.

In this work, we investigate the epigenetic landscape of CRPC after ENZ treatment by profiling global chromatin accessibility to uncover the earliest factors that drive cellular plasticity and commitment to the neuroendocrine lineage. We find that the DNA binding motif for the neuronal lineage-guiding transcription factor ASCL1 becomes hyper-accessible following ENZ treatment and ASCL1 is required for ENZ-induced lineage plasticity. Loss of ASCL1 expression alters the epigenetic programming in t-NEPC by disrupting the polycomb repressive complex 2 (PRC2) and reducing the EZH2 chromatin bound to support the lineage reversal to a luminal AR-driven state. This effect is attributed to an increase of p-EZH2-T311 through UHRF1/AMPK axis.

Results

An epigenetic plasticity emerges in response to hormone therapy

Despite similar genetic profiles, the conversion from CRPC to NEPC post ARPIs is accompanied by extensive transcriptional re-wiring12,15. This suggests that reprogramming of the chromatin landscape may play a central role in this lineage plasticity. To explore this premise, we interrogated changes in the chromatin landscape of CRPC cells post ENZ treatment by performing assay for transposase-accessible chromatin using sequencing (ATACseq) (Fig. 1a). We observed that acute ENZ treatment (3 days) led to altered chromatin accessibility with 2,595 regions (Fig. 1b, region II) gaining accessibility at 3 days post-treatment compared to 10 days and non-treated CRPC. By 10 days post-treatment, widespread changes in chromatin accessibility were observed with 25,538 newly hyper-accessible regions (Fig. 1b, region III) when compared to 3 days and non-treated CRPC, while 2,694 accessible regions (Fig. 1b, region I) were accessible in non-treated CRPC only (Fig. 1b, c). A slight bias toward increased accessibility of promoter regions following treatment of ENZ was observed (Fig. 1d). To further investigate the significance of regions affected by ENZ treatment (opened or closed specifically in response to ENZ), we integrated ATACseq with RNAseq from matched treatment. We defined activated as newly accessible and expressed genes (described as 50 kb distance to accessible peak center with expression of log2Fold Change>1, <-1) in response to ENZ. While, genes that lost accessibility and expression following ENZ treatment were defined as repressed. We found that ENZ redirected the chromatin accessibility from canonical “AR-driven” transcriptional program in CRPC to positively regulated pathways involved in cell plasticity. Importantly, we observed that pathways involved in stem cells were highly enriched compared to the neuronal pathways at 3 days while at 10 days post-treatment, the neuronal pathways become more enriched. As expected, repression of canonical AR transcriptional program was observed at 3 days, with further repression at 10 days post-treatment (Fig. 1e and Supplementary Fig. 1a, b).

Fig. 1: Hormone therapy triggers epigenetic plasticity.
figure 1

a Schematic representation of the study outline. b Heatmap indicating accessibility in 16DCRPC with enzalutamide (ENZ) treatment for 0, 3, or 10 days (n = 3 biologically independent samples). cNumber of ATACseq peak (top) average signal profile from unique 16DCRPC 10 day ENZ-treated regions (bottom). d Genomic annotation of all accessible peaks presented as percentage of all peaks. e Gene set enrichment analysis (GSEA) shows transcriptional response to ENZ treatment associated with gain/loss accessibility in 16DCRPC cells represented as: early repressed; late repressed; early activated; late activated. Dotted line represent false discovery rate (FDR) of 0.05, p < 0.05, statistical analysis was performed using a hypergeometric test. See also Supplementary Fig. 1. f Transcription factor (TF) binding motifs surrounding accessible chromatin in unique vs. shared regions, ranked based on differential p-value. Each dot represents a motif. Statistical analysis was performed using a cumulative hypergeometric test. g Genomic annotation for genes mapped to ASCL1 motif in 16DCRPC 10 days ENZ-treated, presented as percentage of all peaks. See also Supplementary Fig. 1. h Pathways associated to genes mapped to ASCL1 motif in 16DCRPC 10 days ENZ-treated. Dotted line represents FDR = 0.05, p < 0.05, statistical analysis was performed using a hypergeometric test. See also Supplementary Fig. 1. Source data are provided as a Source Data file.

To identify potential regulators of this large-scale epigenetic reprogramming in response to ARPI, we performed transcription factor (TF) motif analysis within a 50 bp window surrounding ATACseq peaks and discovered that both androgen-response element and glucocorticoid response element are the most enriched motifs in 16DCRPC, while consensus binding sequence for neuronal lineage TFs (ASCL1, Olig2, NeuroD1, and NeuroG2) were enriched in ENZ-treated CRPC (Fig. 1f). Ranking motifs by p-value, we found the DNA binding motif for the pro-neural TF ASCL1 to be disproportionally enriched in hyper-accessible chromatin regions post ENZ treatment (Fig. 1f, Supplementary Data 2). Specifically, ASCL1 was the most favorable TF, as their rank went from 126th place in non-treated cells to 8th after 3 days and 4th after 10 days of ENZ treatment. ASCL1 motif is highly accessible in NE-like state GEM model14 compared to adenocarcinoma (Supplementary Fig. 1c). Analysis of genomic distribution of ASCL1 motif in unique accessible regions of 10 days ENZ-treated CRPC (Supplementary Fig. 1d and Supplementary Data 3), revealed a bias towards enhancer regions (intronic and intergenic) (Fig. 1g and Supplementary Fig. 1e). In particular, regions associated with DNA-binding motif of ASCL1 were enriched for genes associated with both stem and neuronal lineage programing (Fig. 1h and Supplementary Fig. 1f). These data suggest that ENZ induces a well-organized chromatin dynamic that function to unlock lineage plasticity that may support treatment resistance.

Enzalutamide-mediated chromatin remodeling supports neuroendocrine differentiation

In order to delineate whether alterations in chromatin landscape mediated by ENZ treatment support a neuronal phenotype, we compared the chromatin accessibility profile (ATACseq) of ENZ-treated CRPC (10 days) with the t-NEPC model (42DENZR) and de novo NEPC cell line (NCI-H660). We found that a number of genomic loci accessible in ENZ-treated CRPC (10 days) remains accessible in NEPC cell lines; ~75% overlap with 42DENZR and ~40% with NCI-H660 (region I) (Fig. 2a and Supplementary Fig. 2a). This ENZ-induced chromatin remodeling in CRPC was found to be accessible in NEPC cell lines. For instance, analysis of the genomic loci of neuroendocrine genes CHGA and NCAM1 and stem cell gene SOX2 revealed accessibility of promoter regions of these genes in CRPC cells after ENZ treatment and in NEPC cell lines (Supplementary Fig. 2b). In addition, we observed a further opening of chromatin in NEPC cell lines, with distinct accessible regions (~28,800 peaks shared between NEPC cell lines) (region II) (Fig. 2a and Supplementary Fig. 2a). Genes associated with stem and neuronal transcriptional network were enriched within 16DCRPC 10 days treated with ENZ, 42DENZR, and NCI-H660 shared regions and NEPC cell lines (Fig. 2b and Supplementary Fig. 2c) with ASCL1 binding motifs enriched surrounding accessible peaks (Fig. 2c and Supplementary Data 4).

Fig. 2: Enzalutamide-induced epigenetic plasticity leads to neuronal differentiation.
figure 2

a Heatmap showing chromatin accessibility comparing 16DCRPC 10 days ENZ-treated unique accessible regions to neuroendocrine prostate cancer (NEPC) (left) average signal profile (right). bGSEA in accessible shared regions (top) and accessible in NEPC (bottom). Dotted line represent p = 0.05, statistical analysis was performed using a hypergeometric test. See also Supplementary Fig. 2. c TF binding motifs surrounding accessible chromatin in shared accessible region (left) accessibility in NEPC (right) vs. unique to 16DCRPC 10 days ENZ-treated, ranked based on differential p-value. Each dot represents a motif. Statistical analysis was performed using a cumulative hypergeometric test. d ASCL1 mRNA expression normalized to GAPDH (n = 2 biologically independent samples). e ASCL1 score reported as log2FPKM mean ± SD. (16DCRPC n = 3, 16DCRPC 3d ENZ n = 1, 16DCRPC 10d ENZ n = 1, 42DENZR and NCI-H660 n = 3 biologically independent samples. 42DENZR and NCI-H660 p < 0.0001; two-tailed p-value). f ASCL1 expression presented as normalized read counts in serial sections from naïve and neoadjuvant androgen-deprivation therapy (ADT)/TAX-treated prostate tumors21. Each dot represent individual patient. Data shown as mean ± SD, with significance assessed using a two-tailed unpaired t-test, p = 0.0135. g ASCL1 expression reported as log2TPM in adenocarcinoma (Adeno) and NEPC12, 15 cohorts. Violin plots show median (middle solid line), quartiles as dotted lines and interquartile range, each dot represent a patient, with significance assessed using a two-tailed unpaired t-test. (Beltran et al. Adeno n = 30 and NEPC n = 19; and Labrecque et al. Adeno n = 76 and NEPC n = 22 patients) (left). Volcano plot shows expression and activity of ASCL1 in Beltran and Labrecque cohorts12, 15. Each dot represents a gene, with “ASCL1 signature” (blue) and “ASCL1” (red) highlighted. Dotted line represent p = 0.05, statistical analysis was performed using a hypergeometric test (right). Source data are provided as a Source Data file.

In agreement with changes in the chromatin architecture, ASCL1 expression rapidly increased upon ENZ treatment of adenocarcinoma cell lines in a time-dependent manner (Fig. 2d), in LNCaP cell lines following androgen deprivation (Supplementary Fig. 2e), and in patient derived xenografts (PDX) models post castration19 (Supplementary Fig. 2f). The expression of ASCL1 was maintained in the NEPC cell lines 42DENZR and NCI-H660 (Fig. 2d) and was correlated with increase of its binding to neuronal genes (Supplementary Fig. 2g) and transcriptional activity, by using an ASCL1 score20 (Fig. 2e). Similar to ENZ treatment, siRNA-mediated silencing of AR in CRPC yielded increased expression of ASCL1 (Supplementary Fig. 2h). Attesting to human relevance, patients treated with 4-6 months neoadjuvant hormone therapy21 exhibit high expression of ASCL1 when compared to the treatment naïve group (Fig. 2f), which correlated with high expression of SYP (Supplementary Fig. 2i) supporting our in vitro data that ASCL1 upregulation is an early event following suppression of AR signaling in prostate cancer. Moreover, ASCL1 expression was found to be increased in NEPC patient tumors12,15 (Fig. 2g) and in neuroendocrine LuCaP PDX models compared to adenocarcinoma12 (Supplementary Fig. 2j). Increased of ASCL1 expression and activity was also observed in a GEM model14 following prostate-specific deletion of RB1 and TP53 that develop NE-like state from adenocarcinoma (Supplementary Fig. 2k). Based on these results, ASCL1 might be an important factor in establishing the NE lineage identity in prostate cancer. Our results are in alignment with previous studies, showing direct reprograming of induced pluripotent stem cells (iPS) to functional neurons by ectopic ASCL1 expression22,23.

ASCL1 is required to establish the neuronal and stem cell-like lineage

To further explore the biological function of this transcription factor, ASCL1 was overexpressed in 16DCRPC and RNAseq was performed, first we validated the upregulation of ASCL1 expression and activity (Supplementary Fig. 3a) and found that ASCL1 induced stem and neuronal programs (Fig. 3a). Importantly, ASCL1 was sufficient to increase NE and CSC markers (Fig. 3b, left panel), which were enhanced with ENZ treatment (Supplementary Fig. 3c), similar results were observed in lung adenocarcinoma (Supplementary Fig. 3d). ASCL1 induces NCAM1 and CD44 hybrid cell population and aldehyde dehydrogenase (ALDH) activity (Fig. 3b, right panel). These data suggest that ASCL1 alone is sufficient to induce cell plasticity and neuroendocrine phenotype. To determine whether ASCL1 is required for the development of neuronal and stem cell-like phenotype, knockdown of ASCL1 in 16DCRPC prevented the ENZ-induced upregulation of CSC and NE markers (Fig. 3c) as well as neuronal-like morphology (Supplementary Fig. 3e), and prevented ENZ-mediated upregulation of NCAM1 and CD44 (Supplementary Fig. 3f). Similar results were observed in C4-2 cell line (Supplementary Fig. 3g). These data suggest that ASCL1 is a requirement for neuronal and stem cell differentiation and functions to bias the cell fate towards neuronal and stem cell-like lineage, similar to what has been previously reported in pericyte to neurons re-programing24.

Fig. 3: ASCL1 is a potent regulator of neuronal stem cell-like phenotype in prostate cancer.
figure 3

a GSEA in 16DCRPC cells over-expressing (OE) ASCL1. Dotted line represents p = 0.05, statistical analysis was performed using a hypergeometric test (16DCRPC CTL n = 3 and ASCL1 OE n = 1 biologically independent samples). b Neuronal and plasticity genes mRNA expression reported as mean of replicates normalized to GAPDH (n = 2 biologically independent samples) (left) NCAM1 and CD44 expression reported as mean of replicates (n = 2 biologically independent samples) (right top) ALDH activity reported as mean of replicates (n = 2 biologically independent samples) (right bottom) in CRPC over-expressing ASCL1. See also Supplementary Fig. 3. c 16DCRPC shASCL1 treated with ENZ reported as mean of replicates (n = 2 biologically independent samples) (left) Neuronal-like morphology reported as mean ± SD with significance evaluated at endpoint, (n = 3 biologically independent samples; two-tailed unpaired t-test) (right). d GSEA in 42DENZR shASCL1 and NCI-H660 siASCL1. Dotted line represents p = 0.05, statistical analysis was performed using a hypergeometric test. e Neuronal and plasticity genes mRNA expression normalized to GAPDH (left) NCAM1 and CD44 expression (right top) ALDH activity (right bottom) in NEPC shASCL1 reported as mean ± SD (42DENZR p-value of ASCL1 < 0.000001, CHGA = 0.000009, SYP = 0.006, NSE = 0.007, NCAM1 = 0.001, SOX2 = 0.000003, NANOG = 0.00001, OCT4 = 0.00004 and NCI-H660 p-value of ASCL1 < 0.000001, CHGA = 0.000002, SYP = 0.001, NSE < 0.000001, NCAM1 = 0.000004, SOX2 = 0.000006, NANOG = 0.006 and OCT4 = 0.01; two-tailed unpaired t-test; n = 3 biologically independent samples). See also Supplementary Fig. 3. f Cell cycle phases comparing 42DENZR shASCL1 vs shCTL reported as mean in percentage (p-value of G0/G1 = 0.04, S = 0.006 and G2/M = 0.3; two-tailed unpaired t-test, n = 3 biologically independent samples). g Proliferation of 42DENZR (left) NCI-H660 (right) following shASCL1 reported as mean ± SD, with significance evaluated at the end point (p-value of 42DENZR = 0.003 and NCI-H660 = 0.019; two-tailed unpaired t-test, n = 3 biologically independent samples). See also Supplementary Fig. 3. h 42DENZR tumor size reported as gram (gr) (left) tumor intake shown as percentage of mice developed tumors post injection (right) in shASCL1 vs shCTL reported as mean ± SEM, with significance evaluated at the end point, p = 0.0056, shASCL1 n = 5 and shControl n = 6 biologically independent animal; two-tailed unpaired t-test. See also Supplementary Fig. 3. Source data are provided as a Source Data file.

To evaluate the importance of ASCL1 in maintaining the neuronal phenotype and the plastic state, ASCL1 was silenced in 42DENZR and NCI-H660 cell lines. Gene set enrichment analysis revealed that loss of ASCL1 expression suppress pathways related to proliferation, stemness and neuronal development (Fig. 3d). Specifically, ASCL1 knockdown downregulates both CSC and NE genes expression following ASCL1 silencing using shRNA (Fig. 3e), siRNA or CRISPR in 42DENZR and NCI-H660 (Supplementary Fig. 3h), surface markers NCAM1 and CD44, ALDH activity (Figs. 3e) and 3D spheroids as a measure for functional properties of stemness (Supplementary Fig. 3i). These observations were further validated by western blot (Supplementary Fig. 3j). Moreover, ASCL1 knockdown in 42DENZR resulted in ~6% increase in G0/G1 population (Fig. 3f), and a decrease in cell proliferation capacity in vitro in NEPC cell lines 42DENZR and NCI-H660 using shRNA (Fig. 3g), CRISPR (Supplementary Fig. 3k), or siRNA (Supplementary Fig. 3l). These data are in-agreement with GSEA in Fig. 3d showing a decrease in “dream targets” and “E2F targets” pathways. This reduction in proliferation rate was not due to an increase in apoptosis (Supplementary Fig. 3m). The reduced proliferation was translated in vivo, where 42DENZR xenografts bearing knockdown of ASCL1 grew at slower rate compared to the control group. In addition, we found that ASCL1 was required for tumor initiation measured by tumor intake ratio (Fig. 3h). Expression of NE genes was significantly lower in tumors with ASCL1 knockdown compare to control (Supplementary Fig. 3n).

ASCL1 cistrome is enriched for stem cell and neuronal targets

To begin decipher the role of ASCL1 in lineage programming, we mapped genome-wide occupancy of ASCL1, using chromatin immunoprecipitation sequencing (ChIPseq) in NEPC cell lines. We identified 18,659 and 36,031 regions bound by ASCL1 in 42DENZR and NCI-H660 cell lines, respectively (Fig. 4a). As expected, ASCL1 binding was predominately centered on its canonical E-box binding motif (Fig. 4b). ASCL1 bound regions corresponded largely to enhancer (intronic and intergenic) regions (Fig. 4c), consistent with previous reports in glioblastoma and normal neurons25,26.We identified 3,205 ASCL1 bound genes common between the two cell lines and as expected, pathway analysis identified ASCL1 bound genes to be involved in stem and neuronal programming in NEPC cell lines (Fig. 4d). Visualization of genomic loci from ChIPseq demonstrated direct regulation of both CSC genes including SOX2 (reported to be regulated by ASCL1, in small cell lung cancer (SCLC)27), NANOG and OCT4(encode by POU5F1); and NE genes including CHGA, ENO2 (NSE), NCAM1, DLL1 (known ASCL1 target28) (Fig. 4e). Further supporting the direct regulation of CSC and NE genes by ASCL1, the expression of these genes was downregulated upon ASCL1 knockdown (Fig. 3e, Supplementary Figs. 3h and 4a). Accordingly, we sought to investigate the distinct ASCL1 cistrome and its association with NEPC programing. We integrated ASCL1 cistrome data with matched RNAseq. We identified enhancer regions bound by ASCL1 that lost their corresponding gene expression following ASCL1 knockdown, subsequent pathway analysis revealed loss of stem and neuronal programming following knockdown of ASCL1 within these NEPC enhancer regions (Fig. 4f). Notably, these enhancers were upregulated after overexpression of ASCL1 in 16DCRPC (Supplementary Fig. 4b). These findings suggest that ASCL1 cistrome may function to unlock the lineage plasticity and further confirming the association between ASCL1 and the development of stem cell and neuronal phenotype.

Fig. 4: ASCL1 cistrome is enriched for PRC2 targets.
figure 4

a Heatmap of ASCL1 binding intensity in NEPC presented as fold change over input, with each horizontal line represent a 3 kb locus. 18,659 and 36,031 peaks were called in 42DENZR and NCI-H660, respectively. b Motif enrichment from ASCL1 ChIPseq in NEPC. Statistical analysis was performed using a cumulative hypergeometric test. c Genomic annotation of ASCL1 binding location in NEPC presented as percentage of total peaks. d Venn diagram showing overlapped genes in 42DENZR and NCI-H660 (top) GSEA of ASCL1 bound genes common between 42DENZR and NCI-H660. Dotted line represents p = 0.05, statistical analysis was performed using a hypergeometric test. e Visualization of genomic loci of NE and CSC markers using IGV showing relative occupancy of ASCL1 over input. f Pie chart showing the percentage of ASCL1 peaks at enhancer regions (defined as intergenic and intronic) in 42DENZR and NCI-H660 cell lines (left) bar chart showing the percentage of up- or down-regulated genes annotated to enhancers in ASCL1 knockdown (right). g Pathways associated to down-regulated enhancers in 42DENZR shASCL1 and NCI-H660 siASCL1. Dotted line represents FDR = 0.05, p < 0.05, statistical analysis was performed using a hypergeometric test. See also Supplementary Fig. 4. Source data are provided as a Source Data file.

ASCL1 is required for EZH2 cistrome reprogramming

We observed that ASCL1 expression is heterogeneous in NEPC with two clusters (ASCL1-high and ASCL1-low) (Fig. 2g). Pathway analysis comparing ASCL1-high vs. ASCL1-low revealed that NEPC patients with high ASCL1 expression are enriched with pathways regulating plasticity and EZH2 activity (Fig. 5a). These data were supported by a strong positive correlation between ASCL1 and EZH2 expression (Supplementary Fig. 5a) and activity (Fig. 5b and Supplementary Fig. 5b, c) (R2 = 0.3756 in Beltran dataset15 and R2 = 0.5240 in Labrecque dataset12, p-value < 0.05). ENZ induced ASCL1 expression in CRPC concomitantly with increase EZH2 activity (Supplementary Fig. 5c, d). Interestingly, ASCL1 cistrome was enriched with PRC2 targets (Fig. 4d) and loss of ASCL1 decreased EZH2 activity in NEPC cell lines (Fig. 4f, g). Conversely, over-expression of ASCL1 in CRPC led to an increase in EZH2 activity, as measured by enrichment of pathways associated with histone 3 lysine 27 tri-methylation (H3K27me3), a surrogate measure of EZH2 activity (Fig. 5c and Supplementary Fig. 5e). These data were further validated at the protein level showing that ASCL1 overexpression increased H3K27me3, and conversely, knockdown of ASCL1 abrogates this histone mark (Fig. 5c; Supplementary Fig. 5f). This alteration of EZH2 activity by ASCL1 was not unique to prostate cancer, but seems like a common effect of ASCL1 as observed in lung adenocarcinoma (Supplementary Fig. 5c) and SCLC (Supplementary Fig. 5h). Expression of EZH2 or the PRC2 subunits SUZ12 and EED were not altered by ASCL1 overexpression or knockdown (Fig. 5c), suggesting that ASCL1 alters EZH2 activity. Overlaying ASCL1 and H3K27me3 ChIPseq revealed a 40% overlap which were lost after ASCL1 knockdown (Fig. 5d), were the number of H3K27me3 peaks went from 43,622 in control to 10,336 in knockdown ASCL1 (Supplementary Fig. 5i). Regions co-bound by ASCL1 and H3K27me3 were enriched with ASCL1 motif (Supplementary Fig. 5j). Of note, within same region DNA binding motif of TF OCT was enriched. This region was enriched for PRC2 targets as well as basal/luminal phenotype (Fig. 5g). Visualization of genomic loci of TMPRSS2and ALDH1A3, genes correlating with luminal phenotype in prostate cancer29,30, showed co-occupation of ASCL1 and H3k27me3 on promoter and gene body (Supplementary Fig. 5k).

Fig. 5: ASCL1 knockdown pheno-copies EZH2 inhibition.
figure 5

a ASCL1 expression reported as log2TPM mean ± SD in Beltran and Labrecque datasets12, 15 (left) GSEA in ASCL1-high vs ASCL1-low NEPC. Dotted line represents p = 0.05, statistical analysis was performed using a hypergeometric test (right). b EZH2 and ASCL1 score correlation reported as z-score, with each dot represents a patient tumor, p < 0.05, two-tailed unpaired t-test. See also Supplementary Fig. 5. c Pathways gain/loss in 16DCRPC over-expressing ASCL1. Dotted line represents p = 0.05, statistical analysis was performed using a hypergeometric test. d Western blot shows expression of polycomb repressive complex 2 (PRC2) members and its activity as measured by H3K27me3 expression in castration-resistance prostate cancer (CRPC) over-expressing ASCL1 (left) and NEPC following knockdown of ASCL1 (shASCL1) (right), with actin and H3 expression used as loading controls (n = 3 biologically independent samples). e ChIPseq showing ASCL1 and H3K27me3 binding intensity reported as fold change over input. f Pathways associated with ASCL1-H3K27me3 co-bound regions. Dotted line represents p = 0.05, statistical analysis was performed using a hypergeometric test. See also Supplementary Fig. 5. g ChIPseq showing EZH2 binding intensity reported as fold change over input. h Western blot shows H3K27me3 in 42DENZR shASCL1 and EZH2 inhibition (EZH2i), with actin and H3 expression used as loading controls (n = 3 biologically independent samples). i PRC2 complex measured by co-immunoprecipitation in 42DENZR shCTL and shASCL1 (n = 3 biologically independent samples). j Visualization of nuclear localization of EZH2 in control vs cytoplasmic localization in shASCL1 or EZH2 inhibition (EZH2i), “EZH2” (green), “DAPI” (purple), scale bar=10 μm (n = 3 biologically independent samples). k UHRF1 expression and pAMPK-T172 in CRPC over-expressing ASCL1 (left) NEPC shASCL1 (right), actin was used as loading control (n = 3 biologically independent samples). l Correlation between expression of UHRF1 and ASCL1 in prostate and small cell lung cancer cell lines. Data reported as log2FPKM, with each dot represent one cell line. m Expression of UHRF1 in prostate cancer patients reported as log2FPKM mean ± SD, Adeno n = 155 and NEPC n = 5 patients, p < 0.0001; two-tailed unpaired t-test. Source data are provided as a Source Data file.

To evaluate how loss of ASCL1 regulates the EZH2 cistrome, we performed ChIPseq for EZH2 in 42DENZR cell line following knockdown of ASCL1. Unexpectedly, comparative analysis between control and ASCL1 silencing condition revealed a significant loss of EZH2 binding to chromatin (36,139 peaks in control vs 293 in ASCL1 knockdown) (Fig. 5h). We found that ASCL1 knockdown decreased H3K27me3 to comparable level as targeting EZH2 (EZH2i) enzymatic activity using GSK126 (Fig. 5i). This loss of H3K27me3 was due to the disruption of PRC2, where EED was pulled down at considerably lower rate after ASCL1 silencing compared to control (Fig. 5j) similar to inhibiting EZH2 with GSK12631. ASCL1 knockdown induces EZH2 accumulation in the cytoplasm (Fig. 5k) and its phosphorylation on threonine 311 (pEZH2-T311) (Fig. 5k), known to play an important role in EZH2 function and localization32. EZH2 phosphorylation on T311 has been reported to be regulated by AMPK, we found that ASCL1 knockdown increased AMPK activity as measured by its phosphorylation on T172 (Fig. 5k). Interestingly, AMPK activity has been recently shown to be controlled by the nuclear factor UHRF1 (a key epigenetic regulator that bridges DNA methylation and chromatin modification33) in diverse cell models34. We found that ASCL1 positively regulates UHRF1 expression (Fig. 5k and Supplementary Fig. 5l) by binding upstream of its promoter in prostate35 and small cell lung cancer20 (Supplementary Fig. 5n). Expression of UHRF1 and ASCL1 significantly correlates in NEPC patients12, in prostate and small cell lung cancer cell lines as well as GEM model14 (Fig. 5l and Supplementary Fig. 5m). UHRF1 expression is significantly higher in NEPC tumors compared to adenocarcinoma16 (Fig. 5m). Together we identified a pathway by which ASCL1 regulates UHRF1, the gate keeper of AMPK, to regulate EZH2 methyl-transferase activity.

Loss of ASCL1 initiates a lineage switch from neuronal to luminal

ASCL1 knockdown in 42DENZR led to a change in the transcriptome similar to those observed in 16DCRPC (Fig. 6a) with increase in the expression of luminal genes in 42DENZR and NCI-H660 following knockdown of ASCL1 (Fig. 6b). Of note, these luminal genes are co-bound by ASCL1 and EZH2 and are methylated at promoter region in 42DENZR cell line (Supplementary Fig. 6a). These data support the notion that ASCL1 knockdown re-activates the luminal lineage and the conversion of lineage to an AR-dependent state. These data are supported with analysis of the chromatin landscape using ATACseq, where widespread chromatin remodeling was observed after loss of ASCL1 in 42DENZR (73,603 peaks in control vs 24,016 peaks in ASCL1 knockdown) (Fig. 6c, d). In agreement, quantification of chromatin condensation showed an increase in chromatin compacting after knockdown of ASCL1 in 42DENZR (Supplementary Fig. 6b). Comparative analysis of accessibility between 42DENZR control and ASCL1 knockdown showed that ASCL1 knockdown change the resembling to CRPC. In support, accessible regions shared between 42DENZR control, ASCL1 knockdown and 16DCRPC (Fig. 6c, region II) were significantly less accessible in 42DENZR ASCL1 knockdown and CRPC (Fig. 6c, d). Regions unique to 42DENZR (Fig. 6c, region I) were slightly more accessible in ENZ-treated CRPC (Supplementary Fig. 6c). Motif enrichment analysis identified loss of accessibility at number of DNA-binding motif of neuronal TF including ASCL1, while TF KLF was among the highly accessible TF following ASCL1 silencing (Fig. 6g). Interestingly, regions that lost accessibility (Fig. 6c, region I) were associated with enhancer regions regulating stem and neuronal programing, while conversely, regions that remained accessible (shared between 42DENZR control and ASCL1 knockdown) (Fig. 6c, region II) were equally mapped to promoter and enhancer regions enriched for various biological function such as housekeeping, proliferation and canonical AR signaling (Fig. 6f and g). These data support that ASCL1 is crucial in maintaining the chromatin architecture that is required for stem/neuronal phenotype. Of significance, H3K27me3 ChIPseq from the matching cell lines, identified shared regions carrying H3K27me3 histone mark only in 42DENZR and absent in 16DCRPC and 42DENZR ASCL1 knockdown (Supplementary Fig. 6d), corresponding to increase protein level of H3K27me3 in 42DENZR compare to 16DCRPC (Supplementary Fig. 6e). Interestingly, we observed a different pattern of histone methylation between these regions (Supplementary Fig. 6f). Knockdown of ASCL1 pheno-copies EZH2 inhibition leading to a decrease of NE pathways and re-activation of pathways implicated in canonical AR signaling (Fig. 6h). This lineage reversal was measured by increased of AR binding at KLK3 (coding prostate-specific antigen (PSA)) enhancer resulting in re-expression of PSA (Fig. 6i and Supplementary Fig. 6g) with concomitant loss of expression in NE markers (Supplementary Fig. 6h), similar to EZH2 inhibition31. Together, we have shown that the large-scale chromatin remodeling induced in t-NEPC was lost following knockdown of ASCL1 leading to a switch in the lineage toward a luminal one and support the notion that the neuronal phenotype induced by ENZ can be reversed by targeting ASCL1, at least in a lineage plastic state.

Fig. 6: Loss of ASCL1 re-activates the luminal lineage.
figure 6

a Global transcriptome profiling of the indicated cell lines presented as principal component analysis (PCA) (n = 3 biologically independent samples). b Expression of luminal markers in NEPC following ASCL1 knockdown reported as fold change over control (Data reported as mean of replicates. 42DENZR shCTL and shASCL1 n = 3 and NCI-H660 CTL and siASCL1 n = 1). c Heatmap of accessible region in 42DENZR control, shASCL1 and 16DCRPC represented as 3 kb window around the peak center. See also Supplementary Fig. 6. d 73,603 accessible regions identified in 42DENZRcontrol vs 24,016 in shASCL1 (left) reported as average accessibility signal profile (right). e Ranked transcription factor motif comparing regions remained accessible in shASCL1 vs. regions unique to shControl, ranked based on differential p-value. Statistical analysis was performed using a hypergeometric test. f Genomic annotation shown as percentage of all peaks. g Pathways associated with shared accessible promoters (top) unique lost enhancers (bottom). Dotted line represent p = 0.05, statistical analysis was performed using a hypergeometric test. h GSEA of pathways up- or down-regulated in 42DENZR shASCL1 or EZH2 inhibition (EZH2i). Dotted line represents p = 0.05, statistical analysis was performed using a hypergeometric test. See also Supplementary Fig. 6. i Androgen receptor (AR) ChIP-PCR shows AR binding to prostate specific antigen (PSA) enhancer region (n = 2 biologically independent samples) (right) mRNA expression of PSA normalized to GAPDH (n = 2 biologically independent samples) (middle) and western blot shows protein expression of PSA, with actin as loading control (right) (n = 3 biologically independent samples). Source data are provided as a Source Data file.

Discussion

The implementation of next-generation androgen receptor pathway inhibitors such as abiraterone and enzalutamide have increased the survival of patients with metastatic castrated resistant prostate cancer1,2. These agents maintain the ability to blunt AR signaling. However, prolonged AR pathway inhibition can alter the archetypal course of the disease, leading to histological dedifferentiation and alterations in cell lineages including the aggressive treatment induced neuroendocrine phenotype3,15,36. Importantly, these therapies are now used earlier in clinical management for patients with aggressive localized prostate cancer37,38. While follow-up on these trials are still limited, it is reasonable to speculate that the use of potent ARPIs may lead to higher incident of treatment induced NEPC. Therefore, the need for new therapies centered on targeting lineage plasticity and neuroendocrine differentiation is paramount.

In this study, we dissected the early epigenetic and transcriptional events regulating the trans-differentiation of CRPC to NEPC in response to ARPIs. Analysis of the CRPC transcriptome and chromatin architecture following ENZ treatment revealed an acute luminal-to-neuroendocrine lineage switch. Particularly, suppression of canonical AR signaling was concomitant with synchronized activation of stemness and neuronal lineage programs. We identified ASCL1 as one of the top enriched motifs in CRPC following ARPIs, which was also observed in NEPC cells, in GEM models with MycN overexpression in the context of PTEN and RB1 deletion39 and in luminal prostate cells transformed to neuroendocrine with Myc, Akt and Bcl2 in the context of RB1 and TP53 deletion40. These data suggest that determination of alternative cell fates is decided at the chromatin level early during the evolution of CRPC to NEPC and emphasizes the power of ARPIs in driving this process.

We report here that remodeling of the chromatin by ARPIs was coupled with increase expression and activity of ASCL1. High levels of ASCL1 were observed in subset of NEPC patients (Fig. 2f, g; and Supplementary Fig. 2i, j)41, SCLC42 and GBM43. ASCL1-low NEPC can be driven by other neuronal transcription factors such as NeuroD1, YAP1 or POU2F3 similar to what was observed in SCLC39,44,45,46Building on the observation that ASCL1 induces rapid neurogenesis during normal neuron development47,48, neuronal differential in glioblastoma stem cell25 and neuroendocrine differentiation in lung cancer49 and induces NE markers in prostate cancer50,51, we found that ASCL1 induces neuroendocrine phenotype by directly regulating neuronal and stem cell programs. Significantly, we identified that ASCL1 binds to PRC2 targets and regulates EZH2 activity. Mechanistically, ASCL1 through direct transcriptional regulation of UHRF1, the AMPK gatekeeper, mediates AMPK inactivation independently of AMPK upstream kinases models34. UHRF1 directly binding to AMPK and recruits the phosphatase PP2A complex to trigger AMPK T172 de-phosphorylation34; hence, stabilizes the PRC2 complex and increases H3K27 tri-methylation. Conversely, ASCL1 knockdown inhibits EZH2 activity. These resulted to a shift in the chromatin landscape back to a CRPC-like state and allowed the conversion of the neuroendocrine to luminal lineage. These data suggest that ASCL1 and EZH2 may represent a molecular conduit that contributes to lineage plasticity and treatment resistance. In support of this concept, high levels of EZH2 were reported in NEPC14,52 and its expression was required for the acquisition of lineage plasticity and neuroendocrine differentiation post ARPIs31.

ASCL1 plays a central role in promoting and maintaining neuronal stem cell fate. We found that loss of ASCL1 switches the cell lineage to a luminal state by modulating genome-wide chromatin remodeling. In recent years, a number of clinical studies have focused on targeting epigenetic factors in prostate cancer, including EZH2, in combination with hormone therapy53. Our work provides basis for targeting transcription factor ASCL1 in the lineage plastic neuroendocrine-like tumors, to induce similar downstream effect as envisage for targeting EZH2 and provides a proof of principle that in highly plastic AR positive t-NEPC seen in the clinic3 patients may benefit from targeting ASCL1 or EZH2 to reverse the neuroendocrine phenotype to an alternative lineage and re-addict tumors to ARPIs.

In closing, we report a role for pro-neuronal transcription factor ASCL1 in modulating the chromatin dynamics to support a plastic lineage by orchestrating early chromatin events and regulatory networks that determine a neuronal stem cell-like lineage commitment. In the treatment-resistant, high plasticity state inhibition of ASCL1 reverses the lineage switch to epithelial-luminal, providing a potential for targeting these highly aggressive tumors. Similar to NEPC, a subset of glioblastoma and small cell lung cancers are defined by elevated expression of ASCL1. This work provides much-needed insight into ASCL1 function and dependency that together nominates ASCL1 as a bona fide clinical target.


Article classification: Biological abstract
Share to:
Tel:+86-027-87610298
Tel:+86-027-87610297
Add:Room A11-329, 1st Floor, No.1, SBI Venture Street, Optics Valley, East Lake
New Technology Development Zone, Wuhan, China.
Certificate NO.:U18Q28010569R0S