Loss of KDM6A activates super-enhancers to induce gender-specific squamous-like pancreatic cancer and confers sensitivity to BET inhibitors

SUMMARY

KDM6A, an X chromosome encoded histone demethylase and member of the COMPASS-like complex, is frequently mutated in a broad spectrum of malignancies and contributes to oncogenesis with poorly characterized mechanisms. We found that KDM6A loss induced squamous-like, metastatic pancreatic cancer selectively in females through deregulation of the COMPASS-like complex and aberrant activation of super-enhancers regulating ΔNp63, MYC, and RUNX3 oncogenes. This subtype of tumor developed in males had concomitant loss of UTY and KDM6A, suggesting overlapping roles, and points to largely demethylase independent tumor suppressor functions. We also demonstrate that KDM6A deficient pancreatic cancer is selectively sensitive to BET inhibitors, which reversed squamous differentiation and restrained tumor growth in vivo, highlighting a therapeutic niche for patient tailored therapies.

Graphical abstract

INTRODUCTION

Pancreatic Ductal Adenocarcinoma (PDA) is a lethal malignancy driven by active mutations in KRAS and inactivation of TP53, CDKN2A, and SMAD4 (Reichert and Rustgi, 2011; Ying et al., 2016). Oncogenic KRAS promotes the development of histologically distinct precursor lesions known as Pancreatic Intraepithelial Neoplasia (PanIN) that progress to metastatic cancer in the context of additional genetic and epigenetic changes. Recent sequencing of pancreatic cancer genomes revealed four molecular subtypes—Squamous-like, Aberrantly Differentiated Endocrine Exocrine, Pancreatic Progenitor, and Immunogenic—defined by distinct transcriptional and epigenetic states. These molecular subtypes correlate with histopathological manifestation and carry prognostic value, with squamous-like being the most aggressive (Bailey et al., 2016). In this classification, the squamous-like subtype encompasses the previously described quasi-mesenchymal (QM) subtype (Collisson et al., 2011), as they share common gene expression signatures characterized by downregulation of transcription factors that drive endodermal commitment (SOX17, PDX1, and GATA6), and activation of MYC and p63 transcriptional programs (Bailey et al., 2016; Collisson et al., 2011). Thus, identifying the molecular drivers and vulnerabilities of each subtype may pave the way for the development of targeted, patient-tailored therapies for pancreatic cancer.

KDM6A (also known as UTX—ubiquitously transcribed X chromosome tetratricopeptide repeat protein) is a Jumonji C (JmjC) domain histone H3K27me3 demethylase and a member of the KDM6 family that includes UTY (encoded by the Y chromosome), and KDM6B (encoded by an autosomal gene) (Hong et al., 2007; Lee et al., 2007). KDM6B also catalyzes the demethylation of H3K27me3, whereas UTY lacks demethylase activity due to the substitution of critical amino acids within the JmjC domain (Walport et al., 2014). KDM6A antagonizes Polycomb Repressive Complex 2 (PRC2) mediated H3K27 tri-methylation catalyzed by the methyltransferase EZH2 to regulate developmental pathways (Agger et al., 2007; Lan et al., 2007; Welstead et al., 2012). However, deletion of Kdm6a in mice demonstrated that several functions in development were largely independent of the demethylase activity (Shpargel et al., 2012; Shpargel et al., 2014). KDM6A is also an integral component of the COMPASS (COMplex of Proteins ASsociated with Set1)-like complex, which—besides core proteins WDR5, RBBP5, DPY30, and ASH2L—contains KMT2C or KMT2D methyltransferases, that mono-methylate H3K4 to delimit enhancer chromatin (hereafter referred to as COMPASS) (Cho et al., 2007; Hu et al., 2013; Piunti and Shilatifard, 2016). H3K4me1-defined enhancers become active upon H3K27 acetylation (H3K27ac) by CBP/p300 (Creyghton et al., 2010; Pasini et al., 2010). Clusters of H3K27ac-marked enhancers form super-enhancers (SE), which tend to regulate genes that orchestrate cell fate and lineage commitment, and frequently become hijacked during oncogenic transformation (Hnisz et al., 2013; Zhang et al., 2016). KDM6A, KMT2C, and KMT2D are among the most frequently mutated epigenetic regulators in cancer, including pancreatic cancer (Bailey et al., 2016; Biankin et al., 2012; Hoadley et al., 2014; van Haaften et al., 2009; Waddell et al., 2015; Witkiewicz et al., 2015). Although evidence from sequencing human cancer genomes has linked KDM6A to oncogenesis, its role and therapeutic targetability in pancreatic cancer remains unknown.

RESULTS

Mutations and deletions of the KDM6A locus are frequent in squamous-like pancreatic cancer

Interrogation of the International Cancer Genome Consortium, The Cancer Genome Atlas (TCGA-PAAD, 2017), and COSMIC databases revealed that pancreatic cancer genomes frequently carry mutually exclusive mutations and genomic deletions in loci encoding components of COMPASS (KDM6A, UTY, KMT2C, and KMT2D) and CBP/p300 (CREBBP and EP300, Figures S1A–B). Integration of copy number and gene expression changes in over 1,000 human cancer cell lines from The Cancer Cell Line Encyclopedia (Barretina et al., 2012) showed selective copy number losses and downregulation of KDM6A in pancreatic cancer (Figure S1C). Western blotting in a panel of 17 human pancreatic cancer cell lines confirmed the absence of KDM6A in MIAPACA, HPAC, L3.6PL, PK1, and PK9 compared to immortal, but non-transformed, HPDE and HPNE cell lines (Figure S1D). Mutations or loss of KDM6A expression in the PACA-AU and TCGA-PAAD cohorts defined a subset of pancreatic tumors characterized by upregulation of TP63—a marker of squamous differentiation which inversely correlated with overall survival (Figures 1A–C). Pancreatic tumors with squamous differentiation (TP63 high) in male patients, in addition to KDM6A loss or mutations, also harbored mutations in or reduced expression of UTY (Figure 1D), and a positive correlation was observed in the expression of these genes (Figure S1E). At the molecular level, downregulation of UTY was either linked to aberrant methylation of a CpG island in its promoter (Figure 1E) or deletions of the Yq11 cytogenetic band encoding UTY (Table S1). Along the same lines, loss of the entire Y chromosome has been reported to occur in about 40% of male pancreatic cancers (Bardi et al., 1993; Gorunova et al., 1998; Griffin et al., 2007; Kowalski et al., 2007). Patients with defective KDM6A or UTY exhibit shorter survival, and mutations in these tumor suppressors were absent in long-term survivors (Figure 1F).

Figure 1. Mutations and genomic losses of KDM6A in poorly differentiated and squamous-like pancreatic cancer.

(A) Scatter dot plots showing the z-score normalized expression of KDM6A and TP63 in the Immunogenic (IMG), Pancreatic Progenitor (PP), Aberrantly Differentiated Endocrine Exocrine (ADEX), and Squamous (SQ) subtypes of pancreatic cancer in the Australian Pancreatic Cancer Cohort (PACA_AU).

(B) Scatter dot plot showing the z-score normalized expression of TP63 in the TCGA-PAAD cohort in samples stratified based on KDM6A copy number variations (CNV). Red circles and squares denote samples carrying deep deletions and mutations of KDM6A, respectively.

(C) Kaplan-Meier plot showing the overall survival of patients in the TCGA-PAAD cohort stratified based on TP63 expression. Median survival is shown in brackets. n, number of patients. HR, Hazard Ratio; CI, Confidence Interval.

(D) Scatter dot plot showing UTY expression in tumors in male patients stratified based on TP63 expression. Tumors carrying mutations or genomic deletions of UTY (blue), KDM6A (red) or both (green) loci are highlighted.

(E) Schematic showing the genomic locus of UTY and the position of cg04448376 probe that detects the methylation of a CpG island in its promoter (top). Scatter plot showing the correlation between UTY expression and methylation of the CpG island (bottom).

(F) Patients in the TCGA-PAAD cohort were stratified based on the presence (red) or absence (black) of mutations or genomic deletions of KDM6A and UTY in their tumors. Median survival is shown in brackets. n, number of patients. HR, Hazard Ratio; CI, Confidence Interval.

(G) RT-PCR (top) and western blots (bottom) showing the expression of KDM6A and p63 in the indicated cell lines. LOY: Loss of Y chromosome.

(H, I) Left: Representative images of IHC for KDM6A in TMA representing different stages of pancreatic cancer progression (H) and histology (I). Right: contingency tables and stacked bar graphs showing the staining intensity of KDM6A. Significance was determined by a chi-square test; L, Liver; n.s., non-significant. Scale bar 200 µm (100 µm inset in (H)).

Data in (A), (B), and (D) are presented as mean ± SEM. **, p < 0.01 and ***, p < 0.001 as determined by one-way ANOVA and followed by Tukey’s multiple comparison test.

See also Figure S1 and Table S1.

Western blotting confirmed the upregulation of p63 in KDM6A null (MIAPACA, L3.6PL, PK1, and PK9) cell lines (Figures 1G and S1F). Besides KDM6A, human pancreatic cancer cell lines carrying mutations in other COMPASS and CBP/p300 complex members (BXPC3) also presented with high expression of TP63 (Figures 1G and S1G). HPAC—which carries a mutation predicted to alter splicing to produce a JmjC-truncated protein—retained the ability to repress TP63 expression (Figures 1G and S1G), suggesting that the regulation of TP63 by KDM6A may be independent of the catalytic activity. Likewise, only a small fraction of KDM6A mutations mapped within the JmjC domain, while the majority are predicted to cause structural alterations producing either out-of-frame transcripts or truncated proteins that undergo degradation (Figures S1B and S1H). Thus, given that KDM6A and UTY—which lacks demethylase activity—are integral components of the COMPASS complex (Figures S1I–J), their tumor suppressor roles may relate to the regulation of enhancer chromatin.

To confirm these findings, using a KDM6A antibody validated for immunohistochemistry (IHC, Figure S1K), we stained two pancreatic cancer tissue microarrays (TMA) comprised of specimens (a) representing PanIN progression to metastatic disease (Figure 1H) and (b) stratified based on grade and histology (Figure 1I). KDM6A is expressed in islets and displays a weak mosaic staining in the exocrine pancreas, mainly in acinar cells (Figures 1H–I). A similar pattern was observed in murine pancreata (Figure S1K). Expression of KDM6A was strongly upregulated in PanIN as well as in some cases of well-differentiated PDA. In contrast, KDM6A was absent in poorly-differentiated tumors and metastases (p < 0.001, Figure 1H). KDM6A expression was undetectable in squamous (0/4) and metastatic (0/6) pancreatic cancer, and a statistically significant inverse correlation was observed between tumor grade and KDM6A staining (Figure 1I).

Kdm6a restrains Kras^G12D driven pancreatic cancer in a gender-specific manner

To study the impact of Kdm6a loss in pancreatic cancer, we crossed Kdm6a^fl/fl (Welstead et al., 2012) mice with Pdx1^Cre (Hingorani et al., 2003) and Ptf1α^Cre (Kawaguchi et al., 2002) transgenic strains to ablate Kdm6a expression in the pancreas (Figures S2A–B and S1K). No changes were observed in the global levels of H3K27me3, H3K27ac, H3K4me1, EZH2, other COMPASS complex members, nor in the Snf5/Smarcb1 tumor suppressor which loss was associated with the transition to QM-squamous histology (Genovese et al., 2017) (Figure S2B and data not shown). Pdx1^Cre;Kdm6a^null and Ptf1α^Cre;Kdm6a^null mice of either sex were born at the expected gender and genotype ratios and exhibited normal pancreatic histology.

Targeted activation of Kras^G12D in murine pancreata induces PanIN which rarely progress to invasive pancreatic cancer (Hingorani et al., 2003). We found that concomitant loss of Kdm6a in Pdx1^Cre;Kras^G12D and Ptf1α^Cre;Kras^G12D cohorts caused aggressive tumors selectively in females which developed palpable abdominal masses, jaundice, and succumbed to the disease no later than twenty weeks of age (Figure 2A). Heterozygous females and null males (with intact Uty) presented with slower disease kinetics, suggesting that KDM6A and UTY harbor similar tumor suppressor functions independent of the JmjC domain. Serial sectioning and histological analyses of tumors isolated from the Ptf1α^Cre;Kras^G12D cohort demonstrated that deletion of Kdm6a accelerated disease progression characterized by loss of acinar architecture (amylase IHC) and widespread malignant lesions as early as six weeks of age in females (Figures 2B–C). Unlike the majority of males that developed well-differentiated PDA, tumors from females lacked mucin production (Alcian blue), showed extensive desmoplasia (Sirius red), and presented with histologic features of squamous differentiation (7/12 mice) such as large, irregular cells in “whorls” (Figures 2B and S2C–D). Less frequently we observed tumors with QM differentiation or mixed features (4/12 mice). IHC for p63 confirmed the squamous differentiation in female Kdm6a null tumors (Figure 2D). One out of twelve knockout females developed a mucinous cystic neoplasm (MCN) characterized by the formation of large multilocular cysts lined by tall, mucin-producing cells (Figures S2C–D). Likewise, MCN in humans frequently harbor mutations in COMPASS members (Naveed et al., 2014; Rokutan et al., 2016). Only Kdm6a null females presented with invasion in neighboring organs, liver metastases, and bloody ascites (5/12 mice, Figures S2C and S2E). Cell lines established from female Kdm6a null tumors exhibited increased proliferation in vitro and colonized the lungs of immunodeficient mice upon tail vein injection—confirming their metastatic potential (Figures S2F–G). Similar kinetics and squamous differentiation were also observed in the Pdx1^Cre;Kras^G12D;Kdm6a^fl/fl cohort (Figures S2H–J). Careful histological examination of female Ptf1α^Cre;Kdm6a^null pancreata over 40 weeks of age, revealed the presence of metaplastic foci in all mice (4/4)—in the absence of Kras^G12D—further suggesting that Kdm6a safeguards pancreatic cell identity, and its loss can predispose to malignant transformation (Figure S2K).

Figure 2. A gender-specific tumor suppressor role for Kdm6a.

(A) Kaplan-Meier plots showing the survival of Kdm6a mutant mice in Pdx1^Cre;Kras^G12D (left) and Ptf1α^Cre;Kras^G12D (right) cohorts. Asterisks denote animals in control arms that were not terminally ill, but were euthanized because they presented with enlarged abdomen that would compromise their ability to move and feed. Median survival is shown in brackets. n, number of mice.

(B) H&E staining of murine pancreata isolated from male and female mice of the indicated genotypes and ages from the Ptf1α^Cre;Kras^G12D cohort. Arrows point to low-grade lesions. Scale bar 200 µm.

(C, D) Tumors of the indicated gender and genotype from the Ptf1α^Cre;Kras^G12D cohort were stained for Amylase, Alcian Blue, and Sirius Red in (C), and p63 in (D). Scale bar 200 µm.

See also Figure S2.

We found that two male Kdm6a null tumors (2/9) presented with decreased expression of Uty (62% and >90%; Figure S2L). Serial sectioning of these tumors revealed areas with mixed histology, encompassing well-differentiated PDA, MCN, and areas of squamous differentiation, the latter confirmed by p63 staining (Figure S2M). Thus, concomitant loss of Uty and Kdm6a induced squamous-like tumors in male mice. Unlike human, the murine Uty locus does not harbor a CpG island in its promoter, and we failed to detect changes in DNA and H3K27 methylation that could explain its silencing (data not shown). Instead, we detected genomic deletion of several genes within the YqA1 cytogenetic band including Uty (Figure S2N). Interestingly, smoking inversely correlated with the expression of UTY and KDM6A in the TCGA-PAAD cohort (Figure S2O), suggesting that it may favor the development of squamous-like pancreatic cancer.

Loss of Kdm6a induces gene expression changes that favor squamous and quasi-mesenchymal differentiation independent of H3K27me3

To delineate the molecular pathways dependent on Kdm6a, we obtained the transcriptomes of cell lines established from 12–16 week old Ptf1α^Cre;Kras^G12D wild-type and Kdm6a null pancreata. Pearson’s rank correlation and principal component analysis revealed strong changes in gene expression selectively in females (803 differentially expressed genes in females versus 188 in males, Figures 3A and S3A). Consistent with the squamous histology in female tumors, Ingenuity Pathway (IPA) and Gene Set Enrichment (GSEA) analyses of the differentially expressed genes revealed activation of transcriptional programs regulating pathways—such as epithelial to mesenchymal transition (EMT), cell motility, response to hypoxia, cell cycle, proliferation, EGF/MAPK, pro-inflammatory, and p53-dependent signaling—previously found to be enriched in the “C2-squamous-like” subtype identified through a multiplatform analysis of 12 cancer types in TCGA (Figures 3B and S3B–C). No clear gene signature was imposed by loss of Kdm6a in males. Consistently, COMPASS members (KMT2C, KMT2D, and KDM6A) are mutated in over 40% of the Pan-Cancer “C2-squamous-like” samples and conferred poor prognosis regardless of the originating tissue, suggesting an important role in safeguarding cell identity (Hoadley et al., 2014).

Figure 3. Loss of Kdm6a activates gene expression programs favoring squamous and quasi-mesenchymal differentiation.

(A) Heatmap showing the pairwise comparison (Pearson’s rank correlation) of gene expression profiles from indicated genotypes of the Ptf1α^Cre;Kras^G12D cohort.

(B) IPA of differentially expressed genes in female Ptf1α^Cre;Kras^G12D wild-type and Kdm6a null tumors. The x axis corresponds to the raw binomial p values.

(C) Bar graphs showing the relative expression (mean ± SD) of the indicated transcripts (left) and cell morphology (right) of cell lines established from female pancreata of the indicated genotypes from the Ptf1α^Cre;Kras^G12D cohort. Squamous-like (SQ) and QM cell lines formed irregular islets comprised of relatively large polygonal and small fibroblast-looking cells, respectively, as opposed to classical PDA cell lines driven by Kras^G12D which formed well-defined islets made of flat round cells. The cell lines are color coded based on the expression of markers of SQ and QM differentiation. Scale bar 200 µm.

(D) Western blots showing the expression of ΔNp63 isoform in Kdm6a null females and a male cell line with concomitant loss of Uty.

(E) Composite heatmap showing the genome-wide distribution and signal intensity of H3K4me3 and H3K27me3 in Ptf1α^Cre;Kras^G12D wild-type and Kdm6a null pancreatic cancer cell lines. Each horizontal line represents the normalized signal intensity for a gene over its transcription start site (TSS). A ± 10 kb window is shown for each TSS. The grayscale bar shows the normalized signal intensity (RPKM, reads per kb per million mapped reads). Right: heatmap shows the genome-wide binding of KDM6A in Ptf1α^Cre;Kras^G12D female. The color scale bar shows the relative binding intensity. Regions were sorted in ascending order of H3K27me3.

(F) Read density profiles of H3K4me3 and H3K27me3 over the TSS ± 10 kb. The y axis shows the mean RPKM.

(G) IPA of genes with de novo bivalent promoters in Kdm6a knockout female. The x axis corresponds to the raw binomial p value.

See also Figure S3.

Human and murine TP63 loci employ two distinct promoters to produce TAp63 and ΔNp63 isoforms (Figure S3D). A molecular hallmark of the “C2-squamous-like” subtype is strong upregulation of the ΔNp63 isoform which has been linked to squamous identity (Hoadley et al., 2014; Keyes et al., 2011). In our microarray data, although the full-length TAp63 transcript was not differentially expressed, probe level analysis across the exon junctions showed that female, but not male, Kdm6a null tumors exhibited a more than three-fold upregulation of probe 17324492 (PSR/Junction ID) which specifically detects the ΔNp63 isoform (Figure S3D). Indeed, qRT-PCR and western blots confirmed the upregulation of ΔNp63 in cell lines established from squamous pancreatic carcinomas, while upregulation of Zeb1 and Vim was detected in cell lines established from QM tumors (Figures 3C–D). A male Kdm6a null tumor cell line, with loss of Uty, also upregulated ΔΝp63 (Figure 3D). Last, Kdm6a null tumors decreased the expression of Sox17 and Pdx1 that drive endodermal and pancreatic cell fate confirming the loss of cell identity.

To systematically study the epigenetic alterations dependent on KDM6A, and causatively link them to changes in gene expression, we employed ChIP-seq to map the distribution of active-H3K4me3 and repressive-H3K27me3 modifications. Peak calling and genome-wide stratification of the normalized signal intensity over the transcription start site (TSS) revealed an increase of H3K27me3 in a subset of genes in Kdm6a null females, with less pronounced changes in males (Figures 3E–F). No major alterations were observed in the signal intensity and distribution of H3K4me3 in either gender. Several genes with gains in H3K27me3 were already marked with H3K4me3, causing transition to a bivalent state, mainly in knockout females (Figure S3E). IPA of the de novo bivalent genes in null females revealed an enrichment for pathways regulating cell motility through Rho small GTPase and Phospholipase C pathways as well as IL-6 and JAK/STAT signaling (Figure 3G). However, less than 10% of differentially expressed genes showed alterations in H3K27me3 or de novo bivalency, suggesting the changes in gene expression induced by Kdm6a loss may be independent of the demethylase activity.

Loss of Kdm6a activates super-enhancers regulating genes that drive squamous differentiation and metastasis

To explain the changes in gene expression, we examined alterations in enhancer chromatin through mapping the distribution of H3K4me1 and H3K27ac that mark active SE. K-means clustering and genome-wide pairwise comparison revealed that loss of Kdm6a decommissioned a group of SE in both genders and selectively activated a distinct group in females (Figure 4A). These changes correlated with increased KMT2D occupancy and H3K4me1 signal over SE in knockout cells (Figures 4A and S4A)—suggesting that KDM6A may govern the localization of the COMPASS complex. Although we detected a weak increase in knockout females, SE were largely void of H3K27me3 compared to adjacent genomic regions, suggesting the demethylase activity of KDM6A is not responsible for the extensive rewiring of SE (Figure S4A). Besides changes in H3K4me1/H3K27ac, we also detected an increase in the size of SE—especially in knockout females (Figure S4B)—further suggesting that KDM6A regulates COMPASS complex-mediated SE delimitation.

Figure 4. Loss of Kdm6a activates super-enhancers linked to genes encoding transcription factors that drive squamous and quasi-mesenchymal differentiation.

(A) Composite heatmap showing the distribution and signal intensity of H3K4me1, KMT2D, KDM6A, and H3K27ac over SE in Ptf1α^Cre;Kras^G12D wild-type and Kdm6a null cell lines. K-means clustering (K = 6) was performed based on the H3K27ac signal that defines active SE. Each horizontal line represents the normalized signal intensity over a SE. A ± 100 kb window is shown for each SE. The grayscale bar shows the normalized signal intensity (RPKM, reads per kb per million mapped reads). Right: color heatmap showing the normalized H3K27ac signal and highlights SE that are selectively rewired in knockout females.

(B) Venn diagram showing the overlap of genes associated with active SE in the indicated cell lines from the Ptf1α^Cre;Kras^G12D cohort.

(C) Canonical pathway and upstream regulator IPA of genes associated with de novo SE in both genders. The x axis corresponds to the raw binomial p value.

(D) Rank ordering of SE based on the normalized H3K27ac signal intensity. The positions of SE linked to Myc, Tp63, and Runx3 as well as the SE that ranks first in each cell line are indicated.

(E, F) IHC (E) and western blot (F) showing MYC levels in murine tumors of the indicated gender and genotype of the Ptf1α^Cre;Kras^G12D cohort. Scale bar 200 µm, inset 50 µm (bottom).

(G) 1 × 10⁶ L3.6PL and MIAPACA cells were electroporated with constructs expressing wild-type or mutant (H1146A/E1148A) KDM6A, or UTY, or with empty vector. The bulk of the cells were plated for western blotting (left) whereas 5 × 10⁴ cells were plated for proliferation assays in triplicate (right). The bar graph shows the number of cells (mean ± SD) after six days. Representative images of colonies stained with crystal violet are shown (bottom). *, p < 0.05 and **, p < 0.01 as determined by two-tailed unpaired Student's t-test for each condition compared to the control.

(H) L3.6PL and MIAPACA were infected with lentiviruses expressing short hairpins targeting p63 and MYC. Left: Western blots showing the knockdown efficiency. Right: 5 × 10⁴ cells were plated in triplicate and passaged every three days. The line graph showing the cumulative number of cells (mean ± SD) for the duration of the experiment.

See also Figure S4.

IPA of genes linked to de novo active SE in Kdm6a null cells—identified in Figure 4B—revealed enrichment for pathways regulating Axon guidance, EMT, Integrin, STAT3, TGF-β signaling, among others, in both genders (Figure 4C). However, only female knockout cells were predicted to activate gene networks regulated by p63, ZEB1, RUNX3, and MYC—transcription factors that drive squamous, QM differentiation, and metastasis—as well as BRD4, which functions as a global reader of enhancer chromatin (Figure 4C). Ranking the strength based on the H3K27ac signal revealed that only female knockout cells simultaneously activated SE controlling Tp63, Myc, and Runx3 (Figures 4D and S4C). Unlike SE linked to Tp63 and Myc, we did not detect biding of either KMT2D or KMD6A on the SE associated to Runx3, suggesting an indirect effect. Consistently, we detected a strong increase of H3K27ac in SE linked to TP63, MYC, and RUNX3, selectively in human KDM6A mutant cell lines (Figure S4D, data from (Diaferia et al., 2016; Mishra et al., 2017)). Besides upregulation of Tp63, activation of the aforementioned SE in knockout tumors caused a strong upregulation of MYC (Figures 4E–F). Although the epigenetic alterations of the Runx3 locus were relatively weak and accompanied by a two-fold upregulation in the Runx3 transcript, an over 100-fold increase was observed in cell lines established from primary tumors with metastatic potential (Figure S4E). Only L3.6PL—a female squamous pancreatic cancer cell line established from liver metastases of L3.3 cells after successive cycles of selections in nude mice (Bruns et al., 1999)—showed strong upregulation of RUNX3 (Figure S4F). IHC confirmed upregulation of Runx3 in primary tumors from mice with metastatic disease (Figure S4G), consistent with recent findings that RUNX3 is a major driver of metastasis in pancreatic cancer (Whittle et al., 2015).

Only a small number of genes simultaneously acquired bivalency and association with active SE (Figure S4H), suggesting that changes in H3K27me3 and H3K27ac induced by Kdm6a loss occur independently and impact distinct gene sets. Transient expression of either wild-type or demethylase deficient KDM6A, or UTY, downregulated p63 and MYC in L3.6PL and MIAPACA and inhibited their growth (Figure 4G), confirming that the tumor suppressor effect of KDM6A is largely demethylase independent. No changes in the levels of RUNX3 were observed. Knockdown of endogenous TP63 and MYC inhibited cell proliferation confirming that their upregulation upon loss of KDM6A is causatively linked to the oncogenic properties of these cells. Knockdown of MYC potently downregulated p63 and RUNX3 levels (Figure 4H), whereas MYC overexpression adversely impacted survival and has been causatively linked to the development of adenosquamous pancreatic cancer in mice and humans (Figure S4I) (Lin et al., 2013; Witkiewicz et al., 2015). We conclude that changes in the SE landscape due to deregulation of COMPASS resulting from loss of KDM6A is an etiologic factor driving squamous-like pancreatic cancer.

Loss of KDM6A sensitizes pancreatic cancer to bromodomain and extra-terminal (BET) domain inhibitors

We screened 17 human pancreatic cancer cell lines for their sensitivity towards 78 small molecule inhibitors that target epigenetic writers, erasers, readers, and transcriptional regulators. Considering that the IC₅₀ value of those compounds ranges from nM to µM, we administered two different concentrations. Unsupervised hierarchical clustering revealed three groups in response to treatment (Figure 5A). The first was comprised of pan-histone deacetylase inhibitors (i.e. SAHA, M344, LMK235), which abrogated growth in all cell lines tested. The second group contained BET (i.e. JQ1, iBET151, and Bromosporine) and G9α/GLP histone methyltransferases (i.e. UNC0638, BIX01294, UNC0646) inhibitors. This group also contained GSK-J4, a compound originally identified as a pan-KDM6 inhibitor, with the lowest IC₅₀ towards KDM6B (Kruidenier et al., 2012). However, it was later shown that GSK-J4 also inhibits KDM5 family members (Heinemann et al., 2014), thereby the effect seen in our screen may be unrelated to KDM6B and was not further pursued. The third group contained compounds that did not elicit an effect, including inhibitors of EZH2 (GSK126 and GSK343), which tri-methylates H3K27 and antagonizes KDM6A.

Figure 5. KDM6A deficient pancreatic cancer cell lines are sensitive to BET inhibitors.

(A) Heatmap showing the sensitivity of 17 human pancreatic cancer cell lines treated for 72 hr with 0.5 and 5 µM of compounds. KDM6A null (red) and EP300/KMT2D mutant (green) cell lines are highlighted. Euclidean distance was used for row and column clustering. The color bar shows the normalized effect of compounds on cell viability (blue = inhibit and red = stimulate) compared to cells treated with DMSO.

(B) Estimation of JQ1 (top) and iBET-151 (bottom) IC₅₀ values for human pancreatic cancer cell lines treated in triplicate with a range of concentrations [0.3125 to 20 µM]. Data points showing cell viability (mean ± SEM) from duplicates. KDM6A null (MIAPACA and L3.6PL), EP300 (BXPC3), and KMT2D (HUPT4) mutant cell lines are highlighted.

(C) Estimation of JQ1 IC₅₀ values for murine cell lines. The average IC₅₀ values (mean ± SD) are shown from two independent cell line preparations for each genotype.

See also Figure S5.

Considering that COMPASS/CBP/p300 mutant cell lines were selectively sensitive to BET inhibitors, we performed dose-response experiments with JQ1 (Filippakopoulos et al., 2010). We found that MIAPACA (IC₅₀ ~670 nM) and L3.6PL (IC₅₀ ~780 nM) were at least five-fold more sensitive compared to KDM6A wild-type pancreatic cancer cell lines (Figure 5B). Similar results were obtained with iBET-151 (Dawson et al., 2011) which also preferentially inhibited MIAPACA and L3.6PL proliferation (IC₅₀ ~2 µM and 1.2 µM, respectively). Cells lines with mutations in EP300 and KMT2D were also sensitive to BET inhibitors (Figure 5B). Consistently, murine Kdm6a knockout cell lines were sensitive to BET inhibitors compared to wild-type or Trp53 null cell lines (Figure 5C). Meta-analysis of large-scale drug screens (Basu et al., 2013; Iorio et al., 2016) confirmed that sensitivity to BET inhibitors is a general phenomenon in various COMPASS/CBP/p300 mutant cancers (Figure S5).

JQ1 blocks squamous differentiation and restrains Kdm6a null pancreatic cancer in vivo

To dissect the molecular mechanisms by which BET inhibitors sensitize KDM6A deficient cell lines, we obtained the gene expression profiles of MIAPACA and L3.6PL treated with JQ1. IPA of differentially expressed genes revealed enrichment of several metabolic pathways directly regulated by MYC, such as pyrimidine biosynthesis, a well-established target of BET inhibitors (Delmore et al., 2011; Lane and Fan, 2015), as well as pro-apoptotic and p53-dependent pathways. Upstream regulator analysis predicted a strong inhibition of BRD4, the bona fide target of JQ1, and likely compensatory activation of MAPK signaling (Figure 6A). JQ1 potently decreased MYC protein levels in human and murine pancreatic cancer cell lines carrying mutations in COMPASS members (Figures S6A–B), consistent with previous reports (Delmore et al., 2011). Besides MYC, JQ1 also potently decreased the expression of genes associated with de novo SE in Kdm6a null cells, such as p63 (Figures 6B–C data from (Rathert et al., 2015)) and RUNX3 (Figure S6C). ChIP assay showed a decrease in BRD4 binding on SE regulating ΔNp63 and RUNX3 in L3.6PL cells (Figures 6D and S6D). In the case of ΔNp63 that drives squamous differentiation, Chromatin Conformation Capture (3C; see STAR Methods) revealed a long range interaction between the SE and promoter of ΔNp63 in L3.6PL. Strikingly, JQ1 administration not only evicted BRD4 from the SE, but also disrupted the long range interaction with the ΔNp63 promoter, explaining the silencing of this isoform (Figures 6E and S6E).

Figure 6. JQ1 disrupts long range interaction between the ΔNp63 promoter and super-enhancer to reverse squamous differentiation.

(A) Canonical pathways (top) and upstream regulator (bottom) IPA of differentially expressed genes in MIAPACA and L3.6PL cells after treatment with 500 nM of JQ1 for 24 hr. Top: the x axis corresponds to the raw binomial p values. Bottom: the x axis shows normalized z-scores > 2 and p overlap values < 0.01.

(B) qRT-PCR (left) and western blots (right) of the indicated cell lines treated with 500 nM JQ1 for 24 hr. Bar graphs showing the relative expression (mean ± SD) of TP63 isoforms from two independent experiments.

(C) Expression of TP63 in HUPT4 (KMT2D mutant) and MIAPACA (KDM6A null) treated with 200 nM JQ1 for 2 and 24 hr. Raw data from GSE63782.

(D) BRD4 ChIP in L3.6PL cells treated with 500 nM JQ1 for 24 hr. The bar graph shows the relative enrichment (mean ± SD) over the control (IgG) in the SE and promoter of ΔNp63 from two independent experiments.

(E) Chromosome Conformation Capture (3C) in L3.6PL cells treated with either vehicle or 500 nM JQ1 for 24 hr. The H3K27ac track (GSE90110) shows the enrichment over the TPRG1-TP63 locus, and shaded boxes highlight the SE (pink) and promoter (yellow); the HindIII fragments tested as regions of interest (red stripes) are depicted above the H3K27ac track (top). The graph shows the contact frequency between the SE within the TPRG1 locus and the ΔNp63 promoter. Data points in the contact matrix show the mean of duplicates ± SD from two independent 3C library preparations. The schematic (right) shows disruption of the interaction between the SE and ΔNp63 promoter upon JQ1 treatment.

See also Figure S6.

To address whether BET inhibitors are effective in vivo, pairs of sex matched Kdm6a null littermate mice were treated with JQ1 for two weeks. Unlike vehicle treated mice, we found that JQ1 administration led to significantly smaller tumors displaying well-differentiated features, reminiscent of classical PDA (Figures 7A and S7A). A similar effect was also observed in male knockout tumors (Figure 7A). This was linked to a decreased number of Ki-67 positive proliferating cells within the neoplastic lesions in both genders upon JQ1 treatment (Figure 7B). No changes in cleaved caspase 3 (apoptotic marker) were detected (data not shown). IHC also confirmed downregulation of MYC and p63 consistent with the well differentiated histology displayed by JQ1 treated tumors (Figures 7C–D). Additional changes included a reduction in α-smooth muscle actin (αSMA) positive cancer associated fibroblasts which have been shown to secrete pro-inflammatory cytokines and contribute to the desmoplastic reaction (Kalluri, 2016; Ohlund et al., 2017), suggesting the therapeutic effect of JQ1 may also involve remodeling of the tumor microenvironment (Figure S7B). Consistent with in vitro results, JQ1 inhibited the growth of KDM6A null L3.6PL and MIAPACA cells upon transplantation in immunodeficient mice (Figure S7C).

Figure 7. JQ1 restrains Kdm6a null pancreatic cancer and reverses squamous differentiation in vivo.

(A) Littermate pairs of male (n = 4) and female (n = 6) Ptf1α^Cre;Kras^C12D;Kdm6a^null mice were treated with JQ1 (50 mg/kg) or vehicle control starting at three weeks of age, every three days for five doses. Scale bar 500 µm. Right: Scatter dot plots show the average tumor area (mean ± SEM). Each dot represents a mouse.

(B) IHC for Ki-67 in Ptf1α^Cre;Kras^C12D;Kdm6a^null mice treated with JQ1. The scatter dot plot shows the number of Ki-67 positive cells (mean ± SEM) in vehicle and JQ1 treated female knockout mice. Three different tumor areas were analyzed from two independent littermate pairs (black and gray dots). Scale bar 100 µm.

(C, D) IHC for MYC (C) and p63 (D) in JQ1-treated Ptf1α^Cre;Kras^G12D;Kdm6a^null mice. Scale bar 100 µm in (C) and 200 µm (100 µm inset) in (D).

(E) Loss of KDM6A in the context of the KRAS oncogene induces squamous-like pancreatic cancer through activation of SE that regulate the ΔNp63, MYC, and RUNX3 oncogenes. Treatment with JQ1 disrupts long range SE-promoter interactions to restore pancreatic cell identity.

See also Figure S7.

DISCUSSION

Pancreatic cancer exhibits a simple pattern of genetic alterations, and comparative genomic analyses of primary tumors and metastases in the same patients revealed limited heterogeneity as driver mutations were shared by all subclones (Makohon-Moore et al., 2017). Thus, metastatic disease may be driven by epigenetic mechanisms (McDonald et al., 2017; Roe et al., 2017; Tzatsos et al., 2013). By interrogating next-generation sequencing data in several pancreatic cancer cohorts and human cell lines we found recurrent mutations in COMPASS/CBP/p300 members that delimit active SE. Hijacking of SE is emerging as a recurrent theme in cancer and contributes to aberrant transcriptional programs, loss of cell identity, and metastasis (Hnisz et al., 2013; Piunti and Shilatifard, 2016; Zhang et al., 2016).

We found that mutations or downregulation of KDM6A and other COMPASS members are enriched in a subset of TCGA-PAAD cases characterized by a strong upregulation of TP63 and poor prognosis. Staining of both human and murine pancreatic cancer specimens revealed that KDM6A is downregulated upon the progression of PanIN to poorly-differentiated and metastatic tumors, suggesting that it functions as a barrier to Kras-driven malignant transformation. It is worth mentioning that even in the absence of an oncogenic insult, aged Kdm6a null females developed metaplastic foci. Although these lesions are benign and did not progress to PanIN or cancer, patients with germline mutations of KDM6A and genes encoding other COMPASS complex members found in congenital disorders, such as Kabuki syndrome (Tumino et al., 2010), have a higher risk in developing a spectrum of malignancies, and possibly pancreatic cancer.

Conditional deletion of Kdm6a in two different mouse models of Kras-driven pancreatic cancer accelerated PanIN progression and induced aggressive squamous-like tumors selectively in females, providing a genetic proof-of-principle linking this subtype to COMPASS complex deregulation. Heterozygous Kdm6a females and Kdm6a knockout males with intact Uty were protected, suggesting dose-dependent and largely overlapping tumor suppressor functions of KDM6A and UTY. Although the majority of Kdm6a null males retained Uty and developed well-differentiated PDA, spontaneous deletion of Uty triggered squamous differentiation, a phenomenon that occurs in human pancreatic tumors which frequently encompasses the entire Y chromosome. Although we cannot formally attribute this effect exclusively to Uty—considering that additional putative tumor suppressor genes were also deleted (such as Usp9y, the homolog of tumor suppressor Usp9x (Perez-Mancera et al., 2012))—our data suggest that Uty is a tumor suppressor and restrains Kras-driven transformation in males. At the molecular level, downregulation of UTY may result either from genomic instability stemming from the loss of KDM6A (Hofstetter et al., 2016), or smoking which has been linked to mosaic loss of chromosome Y (Dumanski et al., 2015; Wright et al., 2017).

Based on the demethylase activity towards H3K27me3, we hypothesized that KDM6A null tumors would be sensitive to inhibition of EZH2 as is the case in urothelial bladder cancer (Ler et al., 2017). Surprisingly, we found that the tumor-suppressor role of KDM6A in pancreatic cancer was largely independent of its catalytic activity, as EZH2 inhibitors did not alter the growth of Kdm6a null cells, and ectopic expression of demethylase-deficient KDM6A and UTY repressed p63 and MYC, and restrained the growth of KDM6A null cells. Although we detected increases in the H3K27me3 signal in bona fide Polycomb targets and genome-wide expansion of bivalent promoters, knowledge-based pathway analyses of gene expression changes did not reveal enrichment in Polycomb-regulated pathways. Although unexplored in cancer, demethylase independent roles for KDM6A have been previously described in stem cell homeostasis and developmental processes (Shpargel et al., 2012; Shpargel et al., 2014), emphasizing its role in regulating cell identity.

We found that loss of KDM6A rewired enhancer chromatin and activated SE regulating ΔNp63, MYC, and RUNX3. ΔNp63 and MYC drive squamous differentiation and are molecular markers of poor prognosis (Hoadley et al., 2014; Keyes et al., 2011; Lin et al., 2013; Witkiewicz et al., 2015), whereas RUNX3 is a recently identified driver of metastasis (Whittle et al., 2015). Unlike a previous report which showed synergy of JQ1 with either HDAC inhibitors or gemcitabine in Kras^G12D;Trp53 null mice (Mazur et al., 2015), our data showed that JQ1 alone sufficed to elicit a therapeutic effect in Kras^G12D;Kdm6a null mice and COMPASS deficient PDA cell lines. JQ1 administration not only reversed MYC upregulation, but also potently downregulated p63 by interfering with BRD4 binding and disrupted the long range interactions to restore cell identity (model in Figure 7E). Besides changes in the neoplastic cells, JQ1 reduced the number of αSMA expressing cancer associated fibroblasts and limited the desmoplastic reaction. Although on-going studies aim to characterize the growth factors and cytokines that organize the desmoplastic milieu, we observed that loss of Kdm6a activated gene signatures linked to IL-6 signaling, a key pro-inflammatory pathway implicated in triggering desmoplasia (Kalluri, 2016; Ohlund et al., 2017). Along the same lines, upregulation of RUNX3 may also contribute to desmoplastic and metastatic niche formation by regulating the secretion of cytokines (Whittle et al., 2015).

It is becoming evident that SE reprogramming is a critical step in metastatic pancreatic cancer including the squamous-like (current work) and the pancreatic progenitor subtypes (Roe et al., 2017). Although these subtypes employ different patterns of SE reprogramming that trigger the activation of distinct pathways, they may align at the molecular level with a common sensitivity to BET inhibitors. Thus, BET inhibitors could be a promising adjuvant therapy in treating squamous-like and metastatic pancreatic cancer by restoring cell identity and sensitizing tumors to current therapies.

CONTACT FOR REAGENT AND RESOURSE SHARING

Further information and requests for reagents should be directed to the Lead Contact, Alexandros Tzatsos (atzatsos@gwu.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Animal studies

Mouse experiments were conducted under protocols A292, A293, and A308 that were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of the George Washington University, Washington DC. To generate the experimental cohorts, we crossed Kdm6a^fl/fl mice (Welstead et al., 2012) with Pdx1^Cre (Hingorani et al., 2003), Ptf1α^Cre (Kawaguchi et al., 2002), and LSL-Kras^G12D/+ (Tuveson et al., 2004) transgenic strains to generate compound mutant mice with targeted deletion of Kdm6a and expression of Kras^G12D in the pancreas. Genotyping primers are provided in Table S2, provided as an excel file. Mice were bred on a mixed C57BL/6;129/Sv background. Gender was considered an independent variable and was factored in the study design and analysis. For xenograft experiments and tail vein injections employing human and murine pancreatic cancer cell lines, we used the NOD.SCID-Il2rg^−/− (NSG) immunodeficient strain. The identity of the transplanted cell lines, cell number, duration of the experiment, and details in assessing tumor growth in xenografts are provided in the corresponding figure legends. Tumor volumes were assessed with a digital caliper, and the volume was calculated using the formula: tumor volume = ½ (length × width²). An unpaired, two-tailed Student’s t-test was used to evaluate the differences in the tumor volume and a p < 0.05 was considered statistically significant. Treatment of mice with JQ1 occurred intra-peritoneally at a dose of 50 mg/kg every three days starting at three weeks of age. JQ1 was solubilized in DMSO, and prepared in 5% (2-hydroxypropyl)-β-cyclodextrin/PBS for injections. In all experiments, littermate pairs of both genders were used to assess the effect of JQ1 on tumor growth.

Human and murine pancreatic cancer cell lines

The human pancreatic cancer cell lines used in this study were previously described (Tzatsos et al., 2013), and bought from ATCC or kindly provided by colleagues as described in the STAR key resources table. Cell lines were maintained in DMEM supplemented with 10% (vol/vol) cosmic calf serum, 4 mM L-glutamine, 1 mM sodium pyruvate, and 1% (vol/vol) penicillin/streptomycin. E6/E7- and TERT-immortalized, but non-transformed, Human Pancreatic Ductal Epithelial (HPDE) and Human Pancreatic Nestin-Expressing (HPNE) cells, respectively, were maintained in serum-free defined Keratinocyte-SFM (Furukawa et al., 1996; Lee et al., 2003; Tzatsos et al., 2013).

Murine cell lines were prepared by mincing pancreas or pancreatic tumors in G solution [PBS/0.09% glucose, 1% Pen/Strep]. Floating tissue was aspirated away, and washed with G solution until all floating tissue was removed. Minced tissue was resuspended in 10 ml of 1.5 mg/ml collagenase IV in complete DMEM, digested at 37°C for 40 min. Two vol umes of G solution were added, and cells were poured through a 40 µm strainer. The strainer was inverted in a new tube and rinsed with G solution to collect the cells. After the tissue settled, the G solution was removed and replaced, twice. Cells were washed twice with G solution and centrifuged at 1,000 rpm for 1 min at 4°C with no brake. Tissue was resuspended in 2 ml Trypsin/EDTA and incubated at room temperature for 5 min, washed with complete DMEM and centrifuged as above. Cells were plated on collagen-coated plates and maintained in media formulated to sustain the growth of murine pancreas cells (DMEM/F12 without phenol red, 5 mg/ml D-Glucose, 1.22 mg/ml Nicotinamide, 100 ng/ml Cholera toxin, 5 ml/L Insulin-Transferrin-Selenium Plus, 1% Pen/Strep, 0.1 mg/ml soybean trypsin inhibitor, 20 ng/ml EGF, 5% Nu-Serum IV culture supplement, 25 µg/ml bovine pituitary extract, 5 nM 3,3,5-tri-iodo-L-thyronine, 1 µM dexamethasone, and a broad spectrum of antibiotics). Two to three weeks after plating, collagen was removed and cells were expanded in 75% pancreatic media/25% complete DMEM. Early passage cell lines were used for all in vitro experiments. For proliferation assays, 5 × 10⁴ cells were plated in duplicate in twelve-well plates and passaged at a 1:2 or 1:3 ratio; cells were counted with a BioRad TC20 Cell Counter. The sex of all cell lines are indicated in the figures by standard sex symbols, highlighted in blue for males and red for females.

METHOD DETAILS

Tumor histology

Human tissues

We used commercially available anonymized and de-identified human tissue microarrays (TMA). According to The George Washington University Institutional Review Board and based on the guidelines from the Office of Human Research Protection the conducted research meets the criteria for exemption #4 (45 CFR 46.101(b) Categories of Exempt Human Subjects Research) and does not constitute human research. TMA were either bought from the Cooperative Human Tissue Network (CHTN, which is funded by the National Cancer Institute; www.chtn.org/) or from Biomax (www.biomax.us/). The CHTN_PancProg1 TMA contains a series of normal, pre-malignant (low- and high-grade PanIN), pancreatic carcinoma, and metastatic specimens. The Biomax TMA contained samples from normal and clinically annotated pancreatic cancer patients. Redundant patient samples among the Biomax TMA employed were excluded from the analysis. Scoring for KDM6A staining was based on the intensity as “absent, weak, or strong” and determined with the Zen lite 2012 software. The chi-square test was used to determine whether there was a significant difference between the expected and observed frequencies in one or more categories.

Murine tissues

Pancreata were formalin fixed and paraffin embedded. 4 µm sections were deparaffinized in two changes of xylenes, rehydrated sequentially in ethanol.

Immunohistochemistry (IHC)

Slides were washed in IHC wash buffer [0.5% Triton X-100 in PBS], rinsed in water, and submerged in citrate buffer (pH 6.0) to unmask antigens using the Retriever 2100 (Aptum) according to manufacturer’s protocol. The slides were allowed to cool, washed three times, incubated with 3% H₂O₂ to block endogenous peroxidase activity, washed three times, and blocked with 10% normal horse serum in IHC wash buffer for one hr. Slides were incubated with primary antibodies overnight at 4°C. The next day, slides were washed three times, and incubated with biotinylated secondary antibodies for one hr at room temperature. Specimens were washed three times, treated with ABC reagent for 30 min, followed by three washes. Slides were developed with the DAB substrate kit, inactivated in water, counterstained with hematoxylin, dehydrated sequentially in ethanol then cleared slides in two changes of xylenes, and mounted with DPX. For RUNX3 IHC, the M.O.M kit was used according to manufacturer’s protocol.

Alcian Blue staining

Slides were stained with Alcian Blue solution pH 2.5 [1% Alcian Blue/3% Acetic Acid] for 30 min. Stained slides were then rinsed thoroughly in water, counterstained with Nuclear Fast Red for 5 min, rinsed in water, dehydrated sequentially in ethanol, cleared in two changes of xylenes, and mounted with DPX.

Sirius Red staining

Slides were placed flat in a humid chamber, 200–300 µl of Sirius Red/Fast green stain was added to cover tissues, and stained for 30 min. Slides were thoroughly rinsed in water and briefly dehydrated through sequential alcohols, cleared briefly in xylenes, and mounted with DPX.

Tissue morphometry

All slides were photographed with a Zeiss AxioLab.A1 and AxioCam ICc5 camera and analyzed with the Zen lite 2012 software as indicated in the figure legends. To quantify the tumor area in pancreata of JQ1 treated mice, H&E stained sections were photographed at 5× magnification, the tumor area was delimited with the “spline contour” tool of the Zen lite 2012 and calculated by the software. Data are presented as mean tumor area (mm²) ± SEM.

Mapping mutations on KDM6A protein

The ProteinPaint tool (Zhou et al., 2016) was employed to map and display different types of KDM6A mutations in pancreatic and other cancers.

Western Blotting and Immunoprecipitation

Cells were washed in ice-cold PBS and solubilized in lysis buffer [25 mM Tris-HCl pH 7.6, 200 mM NaCl, 1 mM EDTA, 1% Igepal CA-630, 0.1% Na-Deoxycholate, 0.1% SDS supplemented with a mixture of protease inhibitors]. The lysates were briefly sonicated and centrifuged for 10 min at 13,000 rpm. The supernatant was analyzed by western blotting. For immunoprecipitation of COMPASS complex, HEK293T cells were transfected with Flag-tagged constructs for: KMD6A, KDM6A-MUT, UTY, WDR5, DPY-30, RBBP5, and BRG1 (see Key Reagents Table). Nuclear extract was prepared by hypotonic disruption of the plasma membrane with buffer A [20 mM HEPES, 10 mM KCl, 1 mM MgCl₂], spun down at 4000 rpm for 5 min; nuclei were resuspended in buffer C [20 mM HEPES, 480 mM KCl, 0.1% Igepal CA-630, 0.1% SDS], and incubated on ice for 20 min, briefly sonicated, and cleared by centrifugation at 13,000 rpm for 10 min. Soluble nuclear fraction was collected, and 40 µl of anti-FLAG beads were added and incubated overnight at 4°C, with gentle agitation. Immunoprecipitated materials were washed three times, eluted with 2× loading buffer at 95°C for 5 min, and analyzed by western blotting. For ectopic expression of KDM6A and UTY, L3.6PL and MIAPACA were electroporated with the corresponding plasmids using the Neon Transfection System (>60 and >80% efficiency, respectively), plated, and transfected with the same plasmids using Lipofectamine 3000 after three days. To detect endogenous KMT2D, protein lysates have been separated in 6% SDS-PAGE at 4°C for 4–5 hr. Althoug h the molecular weight of KMT2D is ~600 kDa (593 kDa for human and 600 kDa for mouse), endogenous KMT2D migrates with an apparent molecular weight of ~350–400 kDa. Figure S1J shows that (a) short hairpins against the endogenous transcript decreased KMT2D protein level, and (b) endogenous KMT2D co-immunoprecipitates with COMPASS complex members such as RBBP5, WDR5, and KDM6A. Thus, KMT2D antibody (ABE1867_Millipore) specifically recognizes the endogenous full length protein.

Gene expression analysis

Total mRNA was isolated from cell lines using an RNeasy spin column kit according to the manufacturer’s protocol. The mRNA was quantitated using a BioSpectrometer (Eppendorf). Sample preparation was performed in-house and according to manufacturer’s protocol. Briefly, 100 ng of mRNA was used as input for the PrimeView 3.0 (3’ IVT human array) and MoGene 2.0 ST (mouse exon array) microarrays. Synthesis, labeling, and purification of cRNA (ss-cDNA synthesis and purification for MoGene 2.0 ST), and fractionation were carried out according to manufacturers’ instructions. Hybridization was performed using a GeneChip Hybridization Oven 640 overnight at 45°C. Microarray washing and staining was performed on a GeneChip Fluidics Station 450 and were scanned using the GeneChip Scanner 3000 7G, commanded by AGCC software. Probe level analysis including background subtraction and quantile normalization took placed with the Robust Multi Array Average Algorithm (RMA) using the Affymetrix Expression Console Software 1.3. Differentially expressed genes (p < 0.05 and fold change > 2 for the PrimeView and > 1.4 for the MoGene 2.0 ST exon array) were determined using the Transcriptome Analysis Console v3.0. Raw and processed Affymetrix data have been deposited in the Gene Expression Omnibus repository under accession number GSE98067.

Ingenuity Pathway (IPA) and Gene Set Enrichment (GSEA) analyses

IPA Software is a knowledge-based approach for interpreting genome-wide expression profiles through a curated database (Ingenuity Knowledge Database, IKD). The experimental data sets were used to query the IPA and to compose a set of interactive networks taking into consideration canonical pathways, relevant biological interactions, as well as cellular and disease processes. The significance of the association between the experimental data sets and the IKD concepts ware measured in two ways: (a) a ratio of the number of molecules from the data set that map to a particular IKD concept, and (b) a Fisher’s exact test was used to estimate the statistical significance of overlap between the two gene sets. To predict alterations in the activity of transcription factors that can explain the observed gene expression changes caused by loss of Kdm6a, we considered transcription factors with z-scores > 2 and a p overlap value < 0.01. The regulation z-score represents a statistical metric of the direction of the transcriptional changes caused by Kdm6a loss compared to the changes regulated by the transcription factor of interest in the IKD. The positive or negative sign of z-score indicates activation or repression, respectively, of a particular transcription factor. Similarly, GSEA software was used to determine whether an a priori defined set of genes in the Molecular Signatures Database (MSigDB) shows statistically significant, concordant differences with changes caused by deletion of Kdm6a. Analyses were performed with default parameters (“scoring_scheme: weighted”, “metric: Signal2Noise”, and “nperm:1000”).

Chromatin Immunoprecipitation (ChIP-seq)

10⁷ cells were cross-linked in 1% formaldehyde for 20 min, then quenched with glycine [final concentration 125 mM] for 5 min. Cells were washed with PBS and then lysed in ice-cold 0.5% Igepal CA-630 in PBS supplemented with protease inhibitors for 10 min on ice. Nuclei were pelleted, resuspended in micrococcal nuclease (MNase) buffer, and digested with 1 µl MNase for 3 min at 37°C to fragment chromatin, ly sed with SDS lysis buffer [50 mM Tris-HCl (pH 8.0), 1% SDS, 150 mM NaCl, and 5 mM EDTA], and briefly sonicated. Cells were centrifuged 14,000 ×g for 10 min at 4°C, and ten volumes of dilution buffer [16.7 mM Tris (pH 8.0), 167 mM NaCl, 0.01% SDS, 1.1% Triton X-100, and 1.2 mM EDTA] was added to the soluble fraction. ChIP was set-up on protein A agarose beads with one of the following antibodies from Cell Signaling (H3K4me1, H3K4me3, H3K27me3, H3K27ac, KDM6A/UTX, BRD4) or Millipore-Sigma (MLL4/KMT2D). Binding occurred overnight at 4°C. The ChIP was washed twice for 10 min each with: low salt buffer [20 mM Tris-HCl (pH 8.0), 150 mM NaCl, 0.1% SDS, 1% Triton X-100, and 2 mM EDTA], high salt buffer [20 mM Tris-HCl (pH 8.0), 500 mM NaCl, 0.1% SDS, 1% Triton X-100, and 2 mM EDTA], LiCl buffer [10 mM Tris-HCl (pH 8.0), 0.25 M LiCl, 1% Igepal CA630, 1% sodium deoxycholate, and 1 mM EDTA], and TE [10 mM Tris pH 7.6, 1 mM EDTA]. DNA-protein cross-links were eluted with elution buffer [0.1 M NaHCO₃, 1% SDS] at room temperature for 30 min. Reverse cross-linking was performed overnight at 65°C by increasing the concentration of NaCl. Proteins were digested with proteinase-K for one hr at 42°C, and the ChIP was cleaned with PCR Purify Kit. For ChIP-seq experiments, a small aliquot was validated in-house before sequencing. Libraries were prepared with the NEBNext Ultra II kit, and sequencing took place on an Illumina Hi-Seq instrument as previously described (Andricovich et al., 2016). For BRD4 ChIP-qPCR, L3.6PL cells were treated with 500 nM JQ1 overnight; enrichment was detected with Luna Universal qPCR master mix, and quantified using a Bio-Rad CFX Connect real-time PCR detection system.

ChIP-seq Antibody Validation for KDM6A

We employed murine Kdm6a knockout cell lines to assess the following commercial antibodies against KDM6A in ChIP experiments: Cell Signaling (33510), MilliporeSigma (ABE1865), and Bethyl (A302-374A). Based on initial ChIP-qPCR results determining specificity for each antibody and relative enrichment on key targets, we proceeded with the Cell Signaling and MilliporeSigma antibodies and sequenced KDM6A-ChIP libraries generated from female wild-type and Kdm6a null cell lines. Data analysis revealed that only the Cell Signaling antibody presented with unique peaks (FDR < 10⁻⁴) in the wild-type when compared to knockout cells which served as control. The figure below shows screen shots of KDM6A peaks across a 31 mb region around the Myc locus on chromosome 15, as well as across the entire chromosome 4.

ChIP-Seq Data Analysis

Quality control and read alignment: FastQC (Version 0.10.1) was used to examine the read quality. Reads were aligned to the reference mm9 genome using Bowtie aligner (Version 1.1.1, (Langmead et al., 2009)), with default parameters. Only uniquely aligned reads were kept. Peak Calling: SICER (Version 1.1 (Zang et al., 2009)) was used to identify the chromatin domains enriched for histone modification marks. Redundant reads, which might be the result of PCR artifacts, were filtered before peak calling. Window size=200 bp and gap size=200 bp was used for H3K27ac and H3K4me3, and window size=200 bp and gap size=600 bp for diffusive signal H3K27me3. FDR threshold was set to 1e-8 for all histone modifications, and 1e-4 for all transcription factors. Peak annotation: We used RefSeq transcripts from the UCSC genome browser to link the identified peaks to genes. For each gene, a ± 2 kb window spanning the Transcription Start Site (TSS) was used to address whether it overlaps with a peak. If there was an overlap, this gene was considered to be enriched by the corresponding histone modification or transcription factor. Genome-wide comparison and K-means clustering of signal intensity as well as plotting the read density profile over the TSS and SE took place with EaSeq (Lerdrup et al., 2016). Island filtered reads from SICER were used as input. Super-enhancer analysis: Calling of SE took place with ROSE [Rank Ordering of Super-Enhancers; (Whyte et al., 2013)] using H3K27ac ChIP-Seq signal as input. H3K27ac peaks identified in previous steps were transferred to GFF format and used as input. All other parameters were set to default. SE identified by the ROSE with length < 2 kb were filtered out. The GREAT (version 3.0.0) online tool was used to identify genes linked to SE based on the association rule: Basal+extension: 1 kb upstream, 1 kb downstream, 1000 kb max extension. Raw and processed data from ChIP-seq experiments have been deposited in the Gene Expression Omnibus repository under accession numbers GSE98568 and GSE104739.

Small molecule inhibitor screen

A custom-made library of inhibitors, based on the Tocriscreen Epigenetic Toolbox, was used in this screen. Cell lines were plated at a density of 3–5 × 10⁴ cells per well in 96-well plates. The next day, plates were treated with either 0.5 or 5 µM of the library. Three days later, cell viability was assessed with the CellTiter-Glo 2.0 Cell Viability Assay. The plates were read using a GloMax luminometer, and light signal was normalized within (DMSO-treated cells) and across plates (empty well) to adjust for experimental variations. For clustering and visualization of the normalized reads, we employed the ClustVis software that builds on several R-packages (Metsalu and Vilo, 2015). The IC₅₀ values for JQ1 and iBET-151 inhibitors were calculated by treating cell lines with eight concentrations of each compound (2-fold dilution series), in triplicate, ranging from 0.3125 to 20 µM. Curve-fitting and IC₅₀ values were calculated using Prism5.

Chromosome Conformation Capture (3C) experiments

L3.6PL cells were treated with DMSO or 500 nM JQ overnight. Cells were fixed and nuclei were prepared as in ChIP experiments. Nuclei were resuspended in 500 µl 1.2× CutSmart buffer with 14 µl 10% SDS, and permeabilized at 37°C for 1 hr. SDS was sequestered by the addition of 50 µl 20% Triton X-100, and incubated at 37°C for 1 hr. Next, 5–20 µl “undigested” was reserved, and 400 U of HindIII was added to the remaining sample and digested overnight at 37°C with end-over-end rotation. The second day, 5–20 µl of “digested” material was reserved, and 40 µl of 20% SDS was added to remaining sample to inactivate HindIII by incubating at 65°C for 25 min. The samples were transferred to 15 ml conical tubes and diluted with the following 1.15× ligation buffer recipe: 352 µl 10× T4 ligase buffer, 2.71 ml water, and 187.5 µl 20% Triton X-100. Samples were incubated at 37°C for 1 hr. Next, 5000 U T4 ligase was added, and ligation took place with gentle end-over-end rotation at 16°C for 4 hr, and then 45 min at room temperature. Reverse crosslinking took place by the addition of 300 µg proteinase K at 65°C, overnight. On day three, 30 0 µg RNase-A was added, and samples were placed at 37°C for one hr. To extract DNA, 4 ml of phenol-chloroform was added, vortexed for a full min, and centrifuged at 2,200 × g for 15 min. The aqueous phase was collected in a new 50 ml tube and diluted with an equal volume of water (4 ml) and with 800 µl of 2 M sodium acetate pH 5.6; next, 20 ml of ethanol was added, samples were inverted 10 times, and placed at −80°C for 1–4 hr to precipitate the D NA. The samples were centrifuged at 2,200 × g for 45 min at 4°C and washed with 70% ethanol. The 3C libraries were then allowed to dry briefly, without letting the pellet become dull. The libraries were resuspended in 100–600 µl TE. The digestion efficiency, as well as the quality and quantity of 3C libraries, were assessed before downstream analyses. The Q5 Taq polymerase was used for PCR reactions using the following protocol: 98°C 30 sec, 35 cycles [98°C 10 sec, 70°C 15 sec, 72°C 10 s ec], 72°C 2 min. Reactions were run on 2% agarose gels and analyzed using the ImageLab software. Bands were excised, then DNA was extracted and sequenced to confirm specificity of primers and loop identity. Data points plotted in the contact matrix are the averages of duplicates ± SD from two independent library preparations. Primers were designed using a unidirectional strategy (Naumova et al., 2012) with the help of the my5C primer design tool to select suitable HindIII fragments.

QUANTIFICATION AND STASTICAL ANALYSIS

Data are presented as mean ± SEM unless otherwise indicated in the figure legends. Significance was analyzed using two-tailed Student’s t-test. A p < 0.05 was considered statistically significant. To compare more than two experimental groups, one-way ANOVA was used. A chi-square test was used to assess the significance of the KDM6A status and clinical features in the analysis of pancreatic cancer TMA. In the Kaplan-Meier plots the Log-rank (Mantel-Cox) test was used to estimate the median survival to compare different groups. In our analysis of the TCGA-PAAD cohort we plotted the “Overall Survival” using raw data from https://portal.gdc.cancer.gov/ or linked portals http://www.cbioportal.org/ and http://xena.ucsc.edu/ downloaded in November 2017. Patients without follow-up data and status “Alive” were censored (censored event = 0; death event = 1). For these patients the “Overall Survival” value coincides with the “Days To Last Follow Up”. In the analysis of the TCGA-PAAD cohort, patients TCGA-HZ-7289, TCGA-FB-A7DR, TCGA-HZ-8638, TCGA-3A-A9IR, TCGA-3A-A9IO, TCGA-3A-A9IL, TCGA-IB-7654, TCGA-2L-AAQM, TCGA-3A-A9IJ, TCGA-3A-A9IN, TCGA-3A-A9IS, TCGA-2J-AABP, TCGA-3A-A9IV were excluded because according to the pathology reports the tumors were of neuroendocrine origin which is a distinct malignancy from PDA. Similarly, patient TCGA-HZ-8637-01 was excluded because although was found to carry deep deletion of KDM6A, RNA-seq analysis revealed that it expresses abundant KDM6A transcript questioning the purity of this specimen. Statistical analysis took place with Prism5 (GraphPad).

DATA AND SOFTWARE AVAILABILITY

Murine expression profiles and exon analyses

Raw and processed Affymetrix data from primary murine pancreatic tumor cell lines have been deposited in the Gene Expression Omnibus repository under accession number GSE98067.

Murine ChIP-seq data sets

Raw and processed data from H3K4me3, H3K27me3, H3K27ac, H3K4me1, KMT2D, and KDM6A ChIP-seq experiments have been deposited in the Gene Expression Omnibus repository under accession number GSE98568 and GSE104739.

Human expression profiles

Raw and processed Affymetrix data from L3.6PL and MIAPACA cell lines treated with DMSO or JQ1 have been deposited in the Gene Expression Omnibus repository under accession number GSE98067.

ADDITIONAL RESOURCES

For data mining we used The NCI's Genomic Data Commons portal (https://portal.gdc.cancer.gov/), cBioportal (http://www.cbioportal.org/), the Xena browser (https://xenabrowser.net/), the International Cancer Genome Consortium (http://icgc.org/), Cellosaurus (http://web.expasy.org/cellosaurus/), COSMIC (http://cancer.sanger.ac.uk/cosmic), and the Cancer Cell Line Encynclopedia (https://portals.broadinstitute.org/ccle).

Supplementary Material

2

Table S1: Chromosomal abnormalities in band Yq11 involving UTY, Related to Figure 1.

3

Table S2: Complete list of primers, Related to STAR Methods.

SIGNIFICANCE.

Pancreatic cancer is a lethal malignancy with a dismal prognosis. Sequencing of pancreatic cancer genomes revealed four molecular subtypes—Squamous-like, Aberrantly Differentiated Endocrine Exocrine, Pancreatic Progenitor, and Immunogenic—defined by distinct transcriptional and epigenetic states. The squamous-like subtype harbors the worst prognosis and accounts for 20–30% of cases. We found that loss of KDM6A, an X chromosome encoded tumor suppressor, holds prognostic value as females develop aggressive squamous-like cancer and exhibit dismal outcomes. This subtype of tumor developed in males also lose UTY, the Y chromosome encoded KDM6A homolog that lacks demethylase activity. We identified that BET inhibitors restored pancreatic identity and restricted tumor growth of Kdm6a null tumors, suggesting a therapeutic option for squamous-like pancreatic cancer.

HIGHLIGHTS.

• Loss of KDM6A induces squamous-like and metastatic pancreatic cancer in females.

• Squamous-like pancreatic tumors concomitantly lose KDM6A and UTY in males.

• Loss of KDM6A deregulates the COMPASS-like complex and enhancer chromatin.

• KDM6A mutant pancreatic cancer is sensitive to BET inhibitors.

Andricovich et al. show that KDM6A loss, via aberrant activation of super-enhancers regulating several oncogenes, induces squamous-like, metastatic pancreatic cancer in females. In males, both KDM6A and UTY loss is required to induce squamous-like tumors. Importantly, KDM6A-deficient pancreatic cancer is sensitive to BET inhibitors.