MUC1-C Dictates JUN and BAF-Mediated Chromatin Remodeling at Enhancer Signatures in Cancer Stem Cells Open Access

This article is featured in Highlights of This Issue, p. 499

The MUC1 gene evolved in mammals to protect epithelia from exposure to the external environment (1). The transmembrane MUC1-C subunit is expressed at the apical borders of epithelial cells and is activated by loss of homeostasis associated with infections, inflammation, and physical damage (1). Activated MUC1-C homodimers promote proliferative, inflammatory, and remodeling responses associated with wound healing (1). Prolonged activation of MUC1-C as in chronic inflammation with repetitive cycles of damage and repair contributes to oncogenesis (1, 2). In association with this capacity, MUC1-C is aberrantly expressed in carcinomas, localizes to the nucleus, and drives multiple hallmarks of the cancer cell, including induction of the epithelial-to-mesenchymal transition (EMT) and epigenetic reprogramming (1, 3, 4). MUC1-C also induces DNA methyltransferases (DNMT) and DNA methylation of tumor suppressor genes (TSG; ref. 5). In addition, MUC1-C activates the polycomb repressive complex PRC2, interacts with EZH2, and promotes H3K27 trimethylation in suppressing gene expression (6). The effects of MUC1-C on transcriptomes of cancer cells are not limited to gene repression in that MUC1-C interacts directly with transcription factors, such as (i) MYC in driving dedifferentiation (2, 7, 8) and (ii) E2F1 in activating components of the embryonic stem cell BAF (esBAF) and polybromo BAF (PBAF) chromatin remodeling complexes (9, 10). Whereas MUC1-C controls the activity of these transcription factors, it is not known how MUC1-C contributes to their transactivation functions.

The importance of MUC1-C–induced epigenetic reprogramming and gene regulation has been further supported by the involvement of MUC1-C in driving dedifferentiation in cancer progression (1, 2, 7, 8). MUC1-C promotes the progression of castration-resistant prostate cancer (CRPC) to the highly aggressive form of neuroendocrine prostate cancer (NEPC) by inducing NE dedifferentiation, self-renewal, and tumorigenicity (8). In addition, MUC1-C drives reprogramming and dedifferentiation of triple-negative breast cancer (TNBC) cells (7). These findings have supported involvement of MUC1-C as an effector of the cancer stem cell (CSC) state. Chromatin remodeling is critical to attain the pluripotent state and lineage plasticity (11, 12). However, there has been no evidence linking MUC1-C–induced dedifferentiation with alterations in chromatin accessibility. This work reveals an unrecognized role for MUC1-C in driving global chromatin remodeling in association with the activation of genes that contribute to CSC progression. Our results demonstrate that MUC1-C induces changes in chromatin accessibility at proximal enhancer–like and distal enhancer–like signatures of stemness-associated genes that are mediated by JUN/AP-1 and ARID1A/BAF. There is no previous evidence linking MUC1-C to (i) JUN or ARID1A at enhancer-like signatures, or (ii) their regulation of chromatin accessibility.

Human DU-145 CRPC cells (ATCC) were cultured in RPMI1640 medium (Thermo Fisher Scientific) containing 10% heat-inactivated FBS (GEMINI Bio-Products). Human BT-549 TNBC cells (ATCC) were cultured in RPMI1640 medium (Thermo Fisher Scientific) containing 10% heat-inactivated FBS (GEMINI Bio-Products), 100 μg/mL streptomycin, 100 U/mL penicillin, and 10 μg/mL insulin. Authentication of the cells was performed by short tandem repeat (STR) analysis. Cells were monitored for Mycoplasma contamination using the MycoAlert Mycoplasma Detection Kit (Lonza).

MUC1shRNA (MISSION shRNA TRCN0000122938; Sigma) or a control scrambled shRNA (CshRNA; Sigma) was inserted into the pLKO-tet-puro vector (Addgene). JUNshRNA (sc-29223-SH, Santa Cruz Biotechnology), ARID1AshRNA (MISSION shRNA TRCN0000059090), NOTCH1shRNA (MISSION shRNA TRCN0000003362), or a control scrambled shRNA (CshRNA; Sigma) was inserted in the pLKO-puro vector (Addgene RRID:Addgene_32017). The viral vectors were produced in 293T cells (KCB catalog no. KCB 200744YJ, RRID:CVCL_0063; ref. 8). Cells transduced with the vectors were selected for growth in 1–3 μg/mL puromycin. Cells expressing the tet-inducible vectors were treated with 0.1% DMSO as the vehicle control or doxycycline (Millipore Sigma).

Total cellular RNA was isolated using TRIzol reagent (Thermo Fisher Scientific). cDNAs were synthesized and amplified as described previously (8). Primers used for qRT-PCR are listed in Supplementary Table S1.

Total lysates prepared from subconfluent cells were immunoblotted with anti-MUC1-C (HM-1630-P1ABX, 1:400 dilution; Thermo Fisher Scientific), anti-JUN (3742, 1:1,000; Cell Signaling Technology), anti-ARID1A (12354, 1:1,000; Cell Signaling Technology), anti-NOTCH1 (3608, 1:1,000; Cell Signaling Technology), anti-β-actin (A5441, 1:100,000; Sigma), and anti-GAPDH (5174, 1:5,000; Cell Signaling Technology).

Assay for transposase-accessible chromatin using sequencing (ATAC-seq) libraries were generated from three biologically independent replicates per condition. Library preparation and quality control were performed as described previously (13). Briefly, 50,000 cells were subjected to tagmentation reactions with 2.5 μL Tn5 Transposase (Illumina), purified with Qiagen MinElute PCR Purification Kit (Qiagen) and PCR amplified for 8–9 cycles. Quality of ATAC-seq libraries was confirmed with Bioanalyzer prior to sequencing.

A customized analysis pipeline was applied to process the sequencing data of ATAC-seq libraries (https://github.com/macs3-project/genomics-analysis-pipelines). The data from each replicate were aligned to the human genome (version GRCh38) with bwa-mem (https://arxiv.org/abs/1303.3997v2), evaluated by customized quality control scripts, and cleaned by Picard (https://broadinstitute.github.io/picard/, RRID:SCR_006525) tools to mark and remove the duplicated reads. Detection of the accessible regions for each replicate was performed using MACS2 (14) with q-value cutoff 0.05 in paired-end mode. To combine all replicates, all peaks from replicates were first merged, then only the consensus genomic regions detected as accessible region in at least three replicates regardless the condition were kept, and the regions were split into nonoverlapping 100-bps bins. The UCSC toolkit bigWigAverageOverBed was used to extract the count of fragments in each bin for each replicate and to make a count-table for the differential analysis.

DNase1 chromatin accessibility assays were performed on chromatin isolated as described previously (15, 16). Aliquots of chromatin were left untreated or digested with 3 U/100 μL DNase I (Promega) for 7 minutes at room temperature. Reactions were stopped by addition of 10 mmol/L EDTA and 2 mmol/L EGTA and incubated at 65°C for 10 minutes. DNA was then treated with RNase (Ambion, Thermo Fisher Scientific) for 30 minutes at 37°C and Proteinase K (Invitrogen, Thermo Fisher Scientific) for 2 hours at 65°C. DNA was purified and amplified by qPCR using primers for the NOTCH1, EGR1, and LY6E genes (Supplementary Table S2). qPCR results were analyzed according to the formula 100/2^{Ct(DNase I) −Ct(no DNase I)}. The data were normalized to input DNA without DNase I treatment.

The count-table containing fragment counts in each 100 bps bin within consensus accessible regions for each replicate was loaded into DESeq2 (RRID:SCR_015687; ref. 17). Then the Wald test for the negative binominal general linear model coefficients was used to call the differential accessible regions (DAR). A fold-change cutoff of 2 and FDR cutoff 1e−5 was applied for DU-145 cells, and FDR cutoff 1e-2 for BT-549 cells. The more stringent cutoff was used for DU-145 to control the total number of differential accessible regions called from both cells to around 10,000. The DARs were further categorized into DARs with increased and decreased accessibilities in response to doxycycline treatment.

The ChIPseeker package (18) was used to determine the genomic distribution of opening and closing DARs. The “annotatePeak” function was used to find the percentage of DARs located in either promoters, UTRs, exons, introns, downstream, or distal intergenic regions, while defining the promoter regions as 3,000 bps up and downstream from transcription start sites (TSS). Genes with TSSs within 100 kbps of the DARs were identified and then enriched using “enrichGO” function in ClusterProfiler (19) with a Benjamini–Hochberg adjusted FDR cutoff of 0.05, minimum group size of 5, and maximum group size of 500. The Giggle (20) tool was used to query the DARs against the collection of the human ENCODE Candidate Cis-Regulatory Elements (ENCODE cCREs; ref. 21) derived from 706 DNase-seq experiments. The query results contained log2 odds-ratios (ORs) and P values from Fisher exact test based on the overlap of DARs and either of the five categories of cCREs, including the pELS, dELS, and PLS. Giggle was also used to search against all publicly available ChIP-seq datasets. The target datasets are the top 1,000 peaks from every one of the 13,976 ChIP-seq datasets collected in the Cistrome DB (22). In addition, the giggle score (combination of P value and OR) was used to represent the significance of the overlap in Supplementary Fig. S2.

The BETA (RRID:SCR_005396; ref. 23) tool was used to search for enriched motifs in the DARs near the up- or downregulated genes under doxycycline treatment. A P value cutoff of 0.05 in the Kolmogorov–Smirnov test was applied to find the list of DEGs close to the DARs. The enrichGO function in clusterProfiler (RRID:SCR_016884; ref. 19) tool was used to identify enriched GO terms in the four categories of genes – upregulated DEGs close to opening DARs, upregulated DEGs close to closing DARs, downregulated DEGs close to opening DARs, and downregulated DEGs close to closing DARs.

Total RNA from cells cultured in triplicate was isolated using RNeasy Plus Mini Kit (Qiagen). TruSeq Stranded mRNA (Illumina) was used for library preparation. RNA-seq analysis was performed as described previously (7, 8, 24). The AP-1 Q4 and Q6 gene signatures were obtained from the Molecular Signature Database (MSigDB).

Chromatin immunoprecipitation (ChIP) was performed on cells crosslinked with 1% formaldehyde for 5 minutes at 37°C, quenched with 2 mol/L glycine and washed with PBS, and then sonicated in a Covaris E220 sonicator to generate 300–600 bp DNA fragments. Immunoprecipitation was performed using a control IgG (Santa Cruz Biotechnology) and antibodies against MUC1-C (Cell Signaling Technology, catalog no. 16564, RRID:AB_2798765), JUN (Abcam, catalog no. ab32137, RRID:AB_731608), ARID1A (Cell Signaling Technology, catalog no. 12354, RRID: AB_263710), PBRM1 (Cell Signaling Technology, catalog no. E6N2K), EP300 (Cell Signaling Technology, catalog no. D2X6N), H3K27ac (Abcam, catalog no. ab4729, RRID:AB_2118291), H3K4me1 (Abcam, catalog no. ab8895, RRID:AB_306847), and H3K4me3 (Abcam, catalog no. ab8580; RRID:AB_306649). Precipitated DNAs were detected by PCR using primers listed in Supplementary Table S3. Quantitation was performed on immunoprecipitated DNA using SYBR-green and the CFX384 Real-Time PCR Machine (Bio-Rad). Data are reported as fold enrichment relative to IgG (8).

pMIEG-c-Jun (Addgene plasmid #40348 RRID: Addgene_21915-plko tet puro) was used to construct and purify full-length (FL) GST-JUN (aa 1–331) and to generate the (i) GST-JUN (aa 1–194) and (ii) GST-JUN (aa 195–331) fragments. GST-MUC1-CD (FL; aa 1–72), GST-MUC1-CD (aa 1–45), GST-MUC1-CD (aa 46–72), and GST-MUC1-CD (CQC→AQA) were prepared as described previously (7). GST-JUN and GST-MUC1-CD proteins were cleaved with thrombin to remove GST. Purified JUN was incubated with GST or GST-MUC1-CD proteins bound to glutathione beads, and the adsorbates were analyzed by immunoblotting with anti-JUN. Purified MUC1-CD was incubated with GST or GST-JUN proteins bound to glutathione beads, and the adsorbates were analyzed by immunoblotting with anti–MUC1-CD CD1 antibody (7).

Single-cell suspensions were cultured in MammoCult Human MediumKit (StemCell Technologies) at a density of 5,000 cells per well of a 6-well ultralow attachment culture plate (Corning) for 10 days. Mammospheres with a diameter >50 μm were counted under an inverted microscope in triplicate wells.

Each experiment was performed at least three times. Data are expressed as the mean ± SD. All statistical analyses for qRT-PCR and ChIP studies were performed by the unpaired Mann–Whitney U test or Student t test to determine differences between means of groups. A P value of < 0.05 denoted by an asterisk (*) was considered statistically significant.

The RNA-seq and ATAC-seq data have been deposited in the GEO database under accession codes GSE139335, GSE164141, and GSE180599.

MUC1-C drives dedifferentiation and self-renewal of DU-145 CRPC cells (8) and BT-549 TNBC cells (7). However, the mechanisms responsible for these findings remain unknown. Accordingly, we selected these models to investigate the potential involvement of MUC1-C–induced chromatin remodeling in association with driving the CSC state. ATAC-seq was therefore performed on DU-145 and BT-549 cells with MUC1-C silencing (Supplementary Fig. S1A; refs. 7, 8) to assess the effects of MUC1-C on chromatin accessibility. The ATAC-seq libraries had distributions of smaller fragment lengths representing internucleosomal open chromatin regions and larger fragments representing areas spanning nucleosomes (Supplementary Fig. S1B). Read density distribution mapped most of the ATAC-seq peaks to within 1 kb of the TSS, indicating positioning at proximal regions (Supplementary Fig. S1C). Principal components analysis (PCA) across different replicates and doxycycline treatment conditions further showed that PCA sorted according to the amount of data variability and that the samples are clustered by experimental condition (Supplementary Fig. S1D).

In analyzing the DU-145 cell ATAC-seq samples by DESeq2, we found that silencing MUC1-C results in 12,752 DARs. Of the MUC1-C–regulated DARs, 6,584 and 6,168 were identified with increases and decreases, respectively, in accessibility (Fig. 1A). In BT-549 cell samples, MUC1-C silencing was associated with 8,223 DARs of which 4,730 exhibited increases and 3,493 decreases in accessibility (Fig. 1B). We found that the MUC1-C–induced DARs (all, closing and opening) are widely located over the entire genomes of DU-145 (Fig. 1C; Supplementary Fig. S1E) and BT-549 (Fig. 1D; Supplementary Fig. S1F) cells. In addition, DARs were located at proximal regions within 1–3 kb from the annotated TSS and distal intergenic regions within 50–500 kb from a TSS in DU-145 (Fig. 1E, left) and BT-549 (Fig. 1E, right) cells.

Comparison of MUC1-C–induced closing and opening DARs identified 44 and 132, respectively, that are common in DU-145 and BT-549 cells (Fig. 2A). Analysis of the DARs with decreased accessibility for localization to nearest genes and for gene ontology (GO) with ClusterProfiler (19) demonstrated associations with GO BIOLOGICAL PROCESSES of (i) cell polarity, axonogenesis, and actin filament organization in DU-145 cells (Fig. 2B) and (ii) regulation of the type I IFN pathway and response to viral infection in BT-549 cells (Fig. 2C). Of the MUC1-C–induced DARs with increases in accessibility, we identified (i) Hippo and transmembrane receptor serine/threonine kinase signaling pathways in DU-145 cells (Fig. 2D), and (ii) regulation of cellular responses to growth factor stimulus and TGFβ in BT-549 cells (Fig. 2E). We then utilized the Giggle (19) tool to look for potential transcription factor–binding sites or cistromes within the DARs against over 13,000 publicly available human transcription factor ChIP-seq datasets collected in the Cistrome DB (22). We found colocalization of experimentally detected binding sites of JUN and FOS with (i) opening DARs in BT-549 and DU-145 cells, and (ii) closing DARs in BT-549 cells (Supplementary Fig. S2), which was of interest in that there is no known association between MUC1-C and AP-1. NKX2-binding sites were also identified in closing DARs in DU-145 cells (Supplementary Fig. S2).

RNA-seq demonstrated that MUC1-C silencing in DU-145 cells is associated with 882 upregulated and 1,159 downregulated genes (1.5-fold, FDR<0.05; Fig. 3A, left), and in BT-549 cells with 2,465 upregulated and 3,562 downregulated genes (1.5-fold, FDR<0.05; Fig. 3A, right). In DU-145 cells, comparisons of opening DARs with DEGs identified 144 (108+36) upregulated genes and 84 (64+20) downregulated genes (Fig. 3B, left). Closing DARs were associated with 92 (56+36) upregulated and 118 (98+20) downregulated genes (Fig. 3B, left). Comparisons of opening DARs with DEGs in BT-549 cells further identified 238 (223+15) upregulated genes and 146 (122+24) downregulated genes (Fig. 3B, right). Moreover, closing DARs were associated with 52 (37+15) upregulated and 302 (278+24) downregulated genes (Fig. 3B, right). These findings were extended by demonstrating that MUC1-C–induced DARs and corresponding DEGs significantly correlate with GO BIOLOGICAL PROCESSES associated with stress response and inflammatory signaling pathways (Supplementary Figs. S3A–S3D). In extending this analysis to identify specific TFs that activate these pathways, we found that MUC1-C–induced DARs and corresponding DEGs are located at promoter-like signatures (PLS), proximal enhancer–like signatures (pELS) within 2 kb of a TSS and distal enhancer–like signatures (dELS) that fall more than 2 kb from the nearest TSS (ref. 21; Fig. 3C and D). Consistent with our analysis of ChIP-seq datasets in the Cistrome DB (ref. 22; Supplementary Fig. S2), Binding and Expression Target Analysis (BETA; ref. 23) demonstrated that MUC1-C–induced DARs associated with DEGs are enriched for FOS, JUN, and NEF2-binding motifs (Fig. 3E; Supplementary Fig. S3E, left and right), recognized by members of the activator protein 1 (AP-1) family of leucine zipper TFs.

Little is known about the involvement of MUC1-C in the regulation of JUN or other AP-1 family members. Accordingly, we focused on JUN and found that MUC1-C has no apparent effect on JUN expression (Supplementary Fig. S4A and S4B). We therefore probed for an interaction between the MUC1-C and JUN proteins. MUC1-C consists of a 58 aa extracellular domain, a 28 aa transmembrane domain, and a 72 aa cytoplasmic domain (Fig. 4A). The MUC1-C cytoplasmic domain (MUC1-CD; 1–72 aa; Fig. 4A) binds directly to certain TFs, such as MYC, and regulates expression of their target genes (7). Along these lines, we found that GST-MUC1-CD (1–72), but not GST, binds directly to full-length JUN (1–331) (Fig. 4B). Further analysis demonstrated that MUC1-CD (1–45) and not MUC1-CD(46–72) binds to JUN (1–331; Fig. 4C). We also found that MUC1-CD binds predominantly to JUN (1–194), which contains the transactivation domain and to a lesser extent with the JUN (195–331) region that includes the DNA-binding domain (Fig. 4D). MUC1-CD includes a CQC motif that is necessary for MUC1-C homodimerization and interactions with certain TFs (1). Mutation of MUC1-CD CQC motif to AQA abrogated binding to JUN in support of dependence on the Cys residues for the interaction (Fig. 4E). In further support of a MUC1-C–JUN interaction, GSEA analysis of the RNA-seq datasets from DU-145 (Fig. 4F; Supplementary Fig. S4C) and BT-549 (Fig. 4G; Supplementary Fig. S4D) cells demonstrated that MUC1-C silencing significantly associates with the regulation of AP-1 target genes. In addition, we found that MUC1-C regulates common sets of AP-1 target genes in DU-145 and BT-549 cells (Supplementary Fig. S4E and S4F), supporting a functional interaction between MUC1-C and JUN in dictating gene expression that is independent of cell context. AP-1 members can recruit BAF to regulate chromatin accessibility (25, 26). ARID1A is an AT-rich interactive domain containing subunit that maintains the BAF complex, particularly at enhancers (27). Along these lines, we found that silencing MUC1 and ARID1A is also associated with the regulation of overlapping sets of AP-1 target genes (Supplementary Fig. S4G and S4H), indicating that ARID1A/BAF may contribute to the regulation of these signatures.

MUC1-C drives NE dedifferentiation of CRPC/NEPC cells by activating the ARID1A/BAF complex and inducing NOTCH1 expression (8, 9). MUC1-C also activates the ARID1A/BAF complex in TNBC cells (9) and, as reported for MUC1-C (4, 7), NOTCH1 plays a role in the progression of TNBC cells (28, 29). In concert with direct binding of MUC1-C and JUN, we detected occupancy of MUC1-C/JUN complexes on the NOTCH1 gene in a region (-1575 to -1590 bp upstream to the TSS) that contains an AP-1–binding motif (Fig. 5A; Supplementary Fig. S5A). We also identified occupancy of ARID1A and not that of the PBAF PBRM1 subunit (Fig. 5A). Significantly, silencing MUC1-C suppressed JUN and ARID1A occupancy (Fig. 5B). In addition, (i) silencing JUN (Supplementary Fig. S5B) decreased occupancy of MUC1-C and ARID1A (Fig. 5C) and (ii) silencing ARID1A (Supplementary Fig. S5C) was associated with decreases in MUC1-C and JUN occupancy (Fig. 5D). These findings thus supported the identification of a MUC1-C/JUN/ARID1A complex on a NOTCH1 pELS that is dependent on the presence of each component. Consistent with a pELS (21), we detected EP300 and H3K27ac, as well as relatively higher levels of H3K4me3 compared with H3K4me1 signals, all of which were decreased by MUC1-C silencing (Fig. 5E). In addition, silencing MUC1-C decreased chromatin accessibility as evidenced by the IGV snapshot (30) and by direct nuclease analysis of that region in DU-145 (Fig. 5F, left and right) and BT-549 (Fig. 5G, left and right) cells. As previously shown in DU-145 cells (8, 10), silencing MUC1-C in BT-549 cells downregulated NOTCH1 expression (Supplementary Fig. S5D). In addition, silencing JUN and ARID1A suppressed NOTCH1 expression in DU-145 and BT-549 cells (Supplementary Fig. S5E and S5F). NOTCH1 is necessary for self-renewal of DU-145 cells (9). By extension, silencing NOTCH1 in BT-549 cells was similarly associated with inhibition of tumorsphere formation (Supplementary Fig. S5G and S5H), indicating that MUC1-C–induced regulation of NOTCH1 chromatin accessibility and expression contributes to the CSC state in different types of carcinoma cells.

Among the common sets of MUC1-C–induced AP-1 target genes, we identified EGR1, which is associated with promoting CRPC and TNBC CSC states (31–33). Having demonstrated that MUC1-C regulates chromatin accessibility of the NOTCH1 pELS, EGR1 was of interest in that putative AP-1–binding elements were identified in a (i) potential pELS (-1985 to -1974 bp) within 2 kb from the TSS, and (ii) dELS (+6350 to +6361) at over 6 kb from the TSS (Fig. 6A). We found that MUC1-C, JUN, and ARID1A occupy the EGR1 pELSs and dELSs in DU-145 (Fig. 6B, left and right) and BT-549 (Supplementary Fig. S6A, left and right) cells. Moreover, as identified in the NOTCH1 studies, silencing MUC1-C (Fig. 6C, left and right; Supplementary Fig. S6B, left and right), JUN (Fig. 6D, left and right; Supplementary Fig. S6C, left and right) or ARID1A (Fig. 6E, left and right; Supplementary Fig. S6D, left and right) decreased occupancy the MUC1-C/JUN/ARID1A complex on the EGR1 pELSs and dELSs. In addition, we found that silencing MUC1-C in DU-145 cells decreased (i) EP300 occupancy, and (ii) H3K27ac and H3K4me3 signals on the EGR1 pELS and dELS (Fig. 6F, left and right). In BT-549 cells, similar results were obtained for the EGR1 pELS; however, there was no enrichment of H3K27ac or H3K4me3 at the EGR1 dELS (Supplementary Fig. S6E, left and right), suggesting that the EGR1 dELS may be an inactive site. Nonetheless, chromatin accessibility of the EGR1 pELS and dELS was decreased by silencing MUC1-C in DU-145 (Fig. 6G and H) and BT-549 (Supplementary Fig. S6F and S6G) cells. Consistent with these results, silencing MUC1-C, JUN, and ARID1A also resulted in suppression of EGR1 expression in DU-145 and BT-549 cells (Supplementary Figs. S6H–S6J).

To further investigate involvement of MUC1-C in integrating activation of enhancer regions, we studied the human LY6E gene, which encodes a protein related to mouse stem cell factor 1 (Sca-1) and is linked to CRPC and TNBC progression (34–37). Putative AP-1–binding motifs were identified in potential pELS (-699 to -689 bp), PLS (+35 to +47 bp), and dELS (+8231 to +8243 bp), which were of interest in that they exhibited distinct patterns of chromatin accessibility in DU-145 and BT-549 cells (Fig. 7A). In this regard, whereas peaks of chromatin accessible regions were detectable in the PLS in both cell types, the pELS region was more prominent in DU-145 cells and the dELS region was substantially broader and more pronounced in BT-549 cells (Fig. 7A). We found that the pELS and dELS, but not PLS, were significantly occupied by MUC1-C, JUN, and ARID1A in DU-145 cells (Fig. 7B). In BT-549 cells, we detected significant occupancy of MUC1-C, JUN, and ARID1A in the PLS and dELS, but not in the pELS (Supplementary Fig. S7A), conceivably due to the relatively low level of chromatin accessibility in that region. Silencing MUC1-C in DU-145 cells decreased JUN and ARID1A occupancy in the LY6E pELS and dELS, but not significantly in the PLS (Fig. 7C). Accordingly, we focused on the LY6E pELS and dELS and found that JUN (Fig. 7D) and ARID1A (Fig. 7E) are necessary for occupancy of MUC1-C/JUN/ARID1A complexes. Moreover, in BT-549 cells, we found that MUC1-C (Supplementary Fig. S7B), JUN (Supplementary Fig. S7C), and ARID1A (Supplementary Fig. S7D) are necessary for occupancy of MUC1-C/JUN/ARID1A complexes on the PLS and dELS. Consistent with these results, MUC1-C induced increases in EP300, H3K27ac, and H3K4me3 were detectable in the (i) pELS and dELS in DU-145 cells (Fig. 7F), and (ii) in the PLS in BT-549 cells (Supplementary Fig. S7E). In addition, silencing MUC1-C was associated with decreases in chromatin accessibility and nuclease sensitivity in the pELS and dELS in DU-145 cells (Fig. 7G and H) and in the PLS and dELS in BT-549 cells (Supplementary Figs. S7F and S7G). By extension, MUC1-C, JUN, and ARID1A were each necessary for LY6E expression in DU-145 and BT-549 cells (Supplementary Figs. S7H–S7J). These results indicate that MUC1-C–mediated activation the LY6E pELS, PLS, and dELS is differentially regulated depending on cell context.

Epigenetic regulatory mechanisms, including ATP-dependent chromatin remodeling, are essential for maintaining tissue homeostasis and the wound healing response (38). The MUC1-C protein promotes activation of inflammatory, proliferative, and repair pathways found in wound healing and cancer (1). MUC1-C induces changes in DNA methylation patterns and histone modifications in cancer cells (1); however, there has been no evidence that MUC1-C dictates chromatin remodeling. This work demonstrates that MUC1-C regulates global remodeling of chromatin accessibility in DU-145 CRPC and BT-549 TNBC cells. We selected these cells based on findings that MUC1-C drives their dedifferentiation and progression (7, 8). These models were also selected to compare effects of MUC1-C on CSCs derived from different tumor types. In this regard, we found that MUC1-C governs similar patterns of global chromatin remodeling in DU-145 and BT-549 cells; that is, induction of DARs across their entire genomes. Subsets of MUC1-C–driven opening and closing DARs were common to both cell types and were distributed at proximal promoter and distal intergenic regions. In addition, most DARs were significantly associated with induction or repression of genes. Parallels in the CRPC and TNBC CSC models were extended by identifying widespread enrichment of binding motifs in publicly available cistromes for the JUN, FOS, and NFE2 TFs and other members of the AP-1 family. AP-1 is a dimeric complex that includes the JUN, FOS, ATF and MAF proteins that form diverse homodimers and heterodimers in regulating oncogenic and tumor-suppressive activities (39, 40). Given these complexities, we focused on JUN with the understanding that MUC1-C may interact with other AP-1 family members. Indeed, little had been known to date about interactions between MUC1-C and AP-1 TFs; in this regard, our findings identified a previously unrecognized MUC1-C–driven program of chromatin regulation.

MUC1-C induces expression of the Yamanaka OCT4, SOX2, KLF4, and MYC (OSKM) pioneer transcription factors in the dedifferentiation of CRPC and TNBC cells (7, 8). Among the OSKM factors, MUC1-C binds directly to MYC in the regulation of MYC target genes (7, 8). The OSKM factors induce dynamic changes in chromatin states during induction of pluripotency and interact with JUN and other AP-1 family members in regulating accessibility (41, 42). We found that MUC1-C binds directly to JUN and forms a nuclear complex with JUN in driving gene expression. In addition, analysis of DU-145 and BT-549 RNA-seq datasets demonstrated that silencing MUC1-C is significantly associated with the activation and repression of common sets of AP-1 target genes. The parallels in the DU-145 and BT-549 cell models were further extended by the demonstration that MUC1-C and JUN are necessary for activation the NOTCH1 gene, which is of importance for the progression of CSCs (9, 28, 29). MUC1-C/JUN complexes were detectable on a NOTCH1 pELS, invoking the potential recruitment of the ARID1A/BAF chromain remodeling complex. Along these lines, MUC1-C activates the ARID1A component of BAF in the dedifferentiation of DU-145 and BT-549 cells (9). In addition, AP-1 members can cooperate with cell type–specific TFs in the recruitment of BAF to regulate chromatin accessibility (25, 26). In support of a model in which MUC1-C interacts with JUN with the recruitment of ARID1A to the NOTCH1 pELS, MUC1-C was necessary for (i) occupancy of JUN and ARID1A complexes, (ii) increases in H3K27ac and H3K4me3 levels, and (iii) opening of chromatin accessibility. The importance of JUN and BAF in this setting was further supported by the dependence on ARID1A for NOTCH1 activation (9) and the findings that MUC1-C and ARID1A regulate common sets of AP-1 target genes.

Our studies of the NOTCH1 pELS revealed an unexpected role for MUC1-C in chromatin remodeling and provided the basis for extending these investigations to dELSs. Along these lines, AP-1 can operate both proximally to the TSS and at distal enhancer sites (40, 43). Given the localization MUC1-C–induced DARs to distal intergenic regions, we studied EGR1 and LY6E as other stemness-associated genes with potential pELSs and dELSs. As found for the NOTCH1 pELS, we detected MUC1-C/JUN/ARID1A complexes occupying the EGR1 pELS and dELS in DU-145 and BT-549 cells. In addition, silencing MUC1-C, JUN, or ARID1A disrupted occupancy of these complexes, in support of their interdependency for inducing gene activation. MUC1-C also induced increases in chromatin accessibility and DNase sensitivity at the EGR1 pELS and dELS in both DU-145 and BT-549 cells. Notably, MUC1-C increased H3K27ac and H3K4me3 signals at the EGR1 pELS and dELS in DU-145 cells and, in contrast, only at the EGR1 pELS in BT-549 cells, highlighting differences in EGR1 regulation. As further evidence for cell context–dependent regulation, studies of the LY6E gene identified pELSs and dELSs in both DU-145 and BT-549 cells that were distinct in terms of MUC1-C–induced (i) patterns of JUN and ARID1A occupancy, and (ii) intensities of H3K27ac and H3K4me3 signals. MUC1-C was also necessary for driving chromatin accessibility of the pELS and dELS in both cell types. A common finding in our studies of the NOTCH1, EGR1, and LY6E genes is that, in addition to activating H3K27ac signals, MUC1-C was associated with increases in H3K4me3 as compared with H3K4me1. H3K4me3 is a characteristic signal of active promoters and is also detectable at active enhancers bound by RNA Pol II (44, 45). Our work shows that MUC1-C–induced activation of NOTCH1, EGR1, and LY6E is associated with predominant H3K4me3 signals at enhancer-like signatures. Further studies will be needed to address the mechanistic basis for these findings and whether they are related to MUC1-C–induced regulation of H3K4 methyltransferases or demethylases that contribute to aberrantly increased transcriptional activity.

The identification of MUC1-C–induced EGR1 and LY6E stemness-associated genes informs new avenues for investigation of how they may contribute to the CSC state, which will be a focus of our subsequent studies. Additional studies will also be needed to address the diversity of MUC1-C–driven AP-1 target genes, particularly in the context of lineage-specific TFs, which can dictate cell type–specific recruitment of BAF. Along these lines, the observations that silencing MUC1-C and ARID1A is associated with the regulation of common sets of AP-1 target genes indicate that MUC1-C may broadly induce global remodeling of chromatin through AP-1 and the esBAF complex in driving stemness (9). CSCs have the capacity for self-renewal and differentiation. As such, it will be of interest to determine whether changes in CSC chromatin structure are incorporated in their differentiated progeny. Finally, these results do not exclude the possibility that MUC1-C–induced changes in chromatin accessibility are restricted to the BAF complex. Consistent with this notion, MUC1-C activates the NuRD nucleosome and PBAF chromatin remodeling complexes in CRPC and TNBC cells (7, 10), which could further contribute to the effects of MUC1-C on global chromatin architecture.

D. Kufe reports personal fees and other support from REATA Pharmaceuticals, personal fees from CanBas, other support from Genus Oncology, and other support from Hillstream Biopharma outside the submitted work. No disclosures were reported by the other authors.

A. Bhattacharya: Conceptualization, investigation, methodology, writing–review and editing, bioinformatic analysis. A. Fushimi: Bioinformatics analysis. N. Yamashita: Investigation. M. Hagiwara: Investigation. Y. Morimoto: Investigation. H. Rajabi: Investigation. M.D. Long: Bioinformatics analysis. M. Abdulla: Investigation. R. Ahmad: Investigation. K. Street: Bioinformatics analysis. S. Liu: Funding acquisition, bioinformatics analysis. T. Liu: Methodology, writing–review and editing, bioinformatics analysis. D. Kufe: Conceptualization, funding acquisition, writing–original draft, writing–review and editing.

Research reported in this publication was supported by the NCI of the NIH under grant numbers CA97098, CA166480, and CA233084 (to D. Kufe), and CA232979 and OD024973 (to S. Liu).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.