Abstract
Recent studies have thoroughly described genome-wide expression patterns defining molecular subtypes of pancreatic ductal adenocarcinoma (PDAC), with different prognostic and predictive implications. Although the reversible nature of key regulatory transcription circuits defining the two extreme PDAC subtype lineages “classical” and “basal-like” suggests that subtype states are not permanently encoded but underlie a certain degree of plasticity, pharmacologically actionable drivers of PDAC subtype identity remain elusive. Here, we characterized the mechanistic and functional implications of the histone methyltransferase enhancer of zeste homolog 2 (EZH2) in controlling PDAC plasticity, dedifferentiation, and molecular subtype identity. Utilization of transgenic PDAC models and human PDAC samples linked EZH2 activity to PDAC dedifferentiation and tumor progression. Combined RNA- and chromatin immunoprecipitation sequencing studies identified EZH2 as a pivotal suppressor of differentiation programs in PDAC and revealed EZH2-dependent transcriptional repression of the classical subtype defining transcription factor Gata6 as a mechanistic basis for EZH2-dependent PDAC progression. Importantly, genetic or pharmacologic depletion of EZH2 sufficiently increased GATA6 expression, thus inducing a gene signature shift in favor of a less aggressive and more therapy-susceptible, classical PDAC subtype state. Consistently, abrogation of GATA6 expression in EZH2-deficient PDAC cells counteracted the acquisition of classical gene signatures and rescued their invasive capacities, suggesting that GATA6 derepression is critical to overcome PDAC progression in the context of EZH2 inhibition. Together, our findings link the EZH2-GATA6 axis to PDAC subtype identity and uncover EZH2 inhibition as an appealing strategy to induce subtype-switching in favor of a less aggressive PDAC phenotype.
This study highlights the role of EZH2 in PDAC progression and molecular subtype identity and suggests EZH2 inhibition as a strategy to recalibrate GATA6 expression in favor of a less aggressive disease.
Introduction
Pancreatic ductal adenocarcinoma (PDAC) remains a major challenge in cancer medicine with a desperate need to develop better treatment strategies. In fact, despite remarkable efforts in translational research and drug development over the past few decades, the overall 5-year survival rate of less than 10% of patients has remained unchanged for almost 20 years (1). Major causes for the dismal disease outcome are the exceptionally aggressive tumor biology and the remarkable resistance to conventional antitumor treatments (2). Molecularly, PDAC is characterized by signature mutations in KRAS, TP53, CDKN2A, and SMAD4 that regularly co-occur with a set of additionally altered genes (3). Despite the indisputable contribution of these genetic alterations to PDAC development and progression (4–6), the heterogeneity and phenotypic characteristics of PDAC are not only determined by its genotype, but are significantly driven by epigenetic alterations (2, 7). Indeed, the epigenomic landscape in general, and environmental context-dependent chromatin modification in particular, critically shape tumor promotion and progression in the pancreas (7). Based on their dynamic nature, epigenetic changes particularly contribute to those tumor biological states that require a high degree of plasticity in order to guarantee tumor cell survival and to foster tumor progression. Consequently, epigenetic dysregulation in PDAC maintenance has been predominantly linked to highly dynamic oncogenic processes such as dedifferentiation (8), metastasis (9, 10), and therapy resistance (2, 8).
Although tumor cells regularly hijack epigenetic programs in favor of growth and survival advantages, the potentially reversible nature of epigenetic alterations also provides novel therapeutic options for cancer treatment (2, 9, 11). Indeed, therapeutic concepts targeting epigenetic regulators have already been implemented in clinical care of hematologic malignancies, and a plethora of clinical trials are evaluating epigenetic treatment strategies in solid tumors (2). This is for instance the case for targeting histone modifiers such as the histone methyltransferase enhancer of zeste homolog 2 (EZH2; e.g., NCT04179864, NCT03010982). EZH2 represents the catalytic component of the polycomb repressor complex 2 (PRC2) and targets the lysine 27 residue of histone 3 for trimethylation (H3K27me3; ref. 12), thus inducing transcriptional repression of target genes. In several tissues, alterations of PRC2 function and activity have been linked to tumor development and progression (12). In the pancreas, we and others have previously described important functions of EZH2 in acinar cell reprogramming and PDAC development (13, 14). Although the histone methyltransferase critically controls damage-induced pancreatic regeneration and protects acinar cells from malignant transformation (13, 14), aberrant EZH2 activity in advanced stages of pancreatic carcinogenesis drives precursor cell proliferation and progression toward invasive PDAC (14), thus characterizing EZH2 as a potent driver of PDAC development. Mechanistically, we found that EZH2-dependent promotion of pancreatic carcinogenesis involves transcriptional activation of Nfatc1 (14), an inflammatory oncogenic transcription factor (TF) with a significant contribution to PDAC development and progression (14, 15). However, the functional implications as well as the mechanistic basis of oncogenic EZH2 activity in PDAC maintenance remain largely elusive.
Here, we link EZH2 activity to increased tumor cell invasion and PDAC dedifferentiation and find that Ezh2 deficiency in a transgenic PDAC mouse model reduces PDAC incidence and formation of liver metastases. Genome-wide occupancy and expression analyses suggest EZH2 as a pivotal suppressor of transcriptional differentiation programs in PDAC and identify EZH2-dependent transcriptional repression of the lineage defining TF Gata6 (16) as a crucial mechanism in EZH2-driven PDAC progression. Importantly, EZH2 depletion and subsequent re-expression of GATA6 expression and activity are sufficient to induce a molecular PDAC subtype switch in favor of the less aggressive classical PDAC subtype. Together, our findings link the EZH2-GATA6 axis to PDAC subtype identity and reveal EZH2 inhibition (EZH2i) as a promising therapeutic strategy to overcome dedifferentiation and metastasis in PDAC.
Materials and Methods
Mouse strains and in vivo experiments
caNFATc1;KrasG12D mice were described previously (15). Ezh2flox/flox mice (13) were purchased from Charles River and crossed with KrasG12D- (5) or caNFATc1;KrasG12D mice to generate Ezh2fl/fl;KrasG12D- (14) or Ezh2fl/+;caNFATc1;KrasG12D animals, respectively. The p48Cre driver (5) was utilized for pancreas-specific excision of the floxed alleles or conditional expression of transgenes. Ezh2 deletion in the pancreas was confirmed by genotyping as described before (14). For survival studies, mice were followed until endpoint criteria were reached. For the generation of patient-derived xenograft (PDX) models, pieces of bulk primary PDAC tissue from patients who underwent tumor resection at the University Medical Center Goettingen (UMG; Goettingen, Germany) were subcutaneously transplanted in both flanks of NMRInu/nu mice, and tumors were allowed to grow until the tumor volume exceeded 1 cm³. For the isolation of PDX-derived primary PDAC (CDX) cells, PDX tumors with stable growth kinetics that were expanded at least over three generations in mice were subjected to harvesting and tumor dissociation utilizing the gentleMACS dissociator (Miltenyi Biotec) combined with enzymatic dissociation with help of a human tumor dissociation kit (Miltenyi Biotec). Upon dissociation, human tumor cells were positively selected utilizing a mouse cell depletion kit as per the manufacturers' instructions (Miltenyi Biotec) and were cultured on collagen type I–coated dishes (Millipore). After 5 to 6 passages on collagen-coated plates, cells were transferred to uncoated plates for further expansion and experimental approaches. For orthotopic PANC-1 models, 1 × 106 CRISPR/Cas9 ctrl or CRISPR/Cas9 EZH2 cells were mixed with 50% matrigel and were injected into the pancreatic tail of NMRInu/nu mice. Mice were monitored regarding general health symptoms and were sacrificed when endpoint criteria were reached. Subsequently, PDAC was extracted and subjected to IHC analysis (Supplementary Materials and Methods). All animal procedures were conducted in accordance with the protocols approved by the Institutional Animal Care and Use Committee (33.9–42502–04–14/1633; -15/2057, -17/2497 and -19/3085). The generation of PDX models has been approved by the ethical review board of the UMG (70112108).
Cells and culture conditions, EPZ6438 treatment
Primary PDAC cells derived from caNFATc1;KrasG12D (NKC) and KrasG12D;TP53R172H/+ (KPC) mice have been described previously (15). ENKC and KEC cells were isolated from tumor-bearing Ezh2fl/+;caNFATc1;KrasG12D- and Ezh2fl/fl;KrasG12D mice, respectively, as described before (15). Primary murine PDAC cells were cultured using DMEM containing 4.5 g/L D-glucose, l-glutamine supplemented with 10% FCS, and 1% nonessential amino acids. PANC-1 cells have been previously described (14) and were cultivated using DMEM containing 4.5 g/L D-glucose and l-glutamine supplemented with 10% FCS. CDX lines derived from the subcutaneous tumors of PDX mice were cultured in Keratinocyte-SFM:RPMI (in 3:1 ratio) media supplemented with 2% FCS, 1% PenStrep, bovine pituitary extract, and epidermal growth factor. EZH2 methyltransferase inhibition was performed by treatment of indicated concentrations of EPZ6438 (ChemieTek) for 72 hours. Control cells were treated with equal amounts of dimethyl sulfoxide. PANC-1 cells were obtained from the ATCC in 2004. All primary murine cells utilized in the study have been isolated from transgenic mice in the groups of Volker Ellenrieder, Albrecht Neesse, and Elisabeth Hessmann in the timespan from 2010 to 2018. Cell authentication by qPCR analyses was conducted by the Staatliche Gewerbeaufsichtsamt, ZUS AGG.
Generation of EZH2 shRNA and CRISPR/Cas9 clones and transient transfection
For shRNA-mediated EZH2 knockdown, NKC cells were subjected to lentiviral transduction, and transduced clones were subsequently selected with Puromycin (2 μg/μL). The generation of CRISPR/Cas9 EZH2 knockout cells was conducted as described before (14). Briefly, NKC, KPC, and PANC-1 cells were transfected with pSpCas9(BB)-2A-Puro PX459 vectors containing single guide RNA against murine or human EZH2 (Supplementary Table S1) with the help of lipofectamine 2000 (Invitrogen). Transfected cells were selected using Puromycin (2 μg/mL), and single clones were picked and allowed to expand in culture. siRNAs targeting EZH2 or Gata6 were purchased from Ambion. Plasmids expressing shRNA targeting Gata6 were purchased from SIGMA-Aldrich (MISSION shRNA). Cells were transfected with siRNA/shRNA using siLentFect (Biorad) or lipofectamine 2000, respectively. The HA-tagged EZH2 expression plasmid utilized for re-expression of EZH2 in CRISPR/Cas9 Ezh2 NKC clones was provided by addgene (#24230). Myctag-tagged EZH2 wt and EZH2 delset expression plasmids were a kind gift from Myles Brown, Dana-Faber-Cancer Institute, Boston. All expression plasmids were transfected using lipofectamine 2000.
Chromatin immunoprecipitation and chromatin immunoprecipitation sequencing
Chromatin immunoprecipitation (ChIP) analysis was performed as described previously (17). Briefly, cells were fixed using 1% formaldehyde in PBS for 20 minutes before the reaction was quenched by adding 1.25 mol/L glycine for 5 minutes. Antibodies utilized for immunoprecipitation (1 μg) are listed in Supplementary Table S2. Finally, qRT-PCR was conducted with primers depicted in Supplementary Table S3. ChIP analysis was performed with three biological and three technical replicates, each.
For ChIP sequencing (ChIP-seq), ChIP analysis was performed as described before (17) and with antibodies listed in Supplementary Table S2. Subsequently, DNA samples were sonicated in a Biorupter Pico (Diagenode) to obtain approximately 300 bp fragments, which were used for library preparation with the MicroPlex Library Preparation Kit (Diagenode) as per the manufacturer's protocol. Sequencing was performed using the HiSeq 2000 Illumina platform of the NIG. The quality of the ChIP-seq raw data files was assessed by running FASTQ quality check (FASTQC). The base quality score was above 28, confirming that the quality of the reads was good. The GC content per read matched the theoretical distribution. Subsequently, FastQ files were analyzed in the public server usegalaxy.org where the sequence reads were aligned to the mouse reference genome (mm9) using Bowtie2 (version 0.4) with default parameters. Peak calling was carried out by Model based Analysis of ChIP-seq (MACS2 version 2.1.0.20151222) on UseGalaxy (usegalaxy.org) server. The cutoff value for peak detection was set to 0.05. BigWig files were generated using BamCoverage (version 2.2.3) from deeptools and were visualized using integrative genomics viewer (IGV version 2.5.3). Further, BigWig and bed files were used to generate aggregate profiles and relative enrichment of ChIP regions over other genomic locations using CEAS tool on Galaxy cistrome. Differential binding analysis was performed to identify differentially occupied regions in shRNA EZH2- compared with shRNA control NKC cells using the Bioconductor R package Diffbind run on R version 3.6.1 according to the instruction manual. Furthermore, Genomic Regions Enrichment of Annotations Tool (GREAT) analysis was used to identify associated genes with regions identified by Diffbind analysis. Further, principal component analysis (PCA) for the H3K27me3 and H3K4me3 profiles was plotted in R. Heatmaps and average profiles for occupancy were generated using the computeMatrix and plotHeatmap tools on the UseGalaxy server, and peak center was set as the reference point mode. For analyses in PANC-1 cells, input and EZH2 ChIP-seq sequence reads were aligned against the human reference genome (hg38), and BigWig files were generated as described above. The input peaks were subtracted from the EZH2 binding peaks and visualized in IGV. Accession number for ChIP-seq data files is GSE153537.
RNA sequencing
ShRNA control and shRNA EZH2 NKC cells were cultured in triplicates, and RNA was isolated as described before (17). Note that 500 ng of total RNA was used to prepare libraries using the True seq (Illumina) RNA library preparation kit as per the manufacturers' instructions. cDNA library concentrations and fragment sizes were controlled by Qubit (Thermo Scientific, Q32854) and Bioanalyzer (Agilent 2100, 5067–4626), respectively, prior to sequencing in the NGS Integrative Genomics Core Unit (NIG) of the UMG. FastQ files were analyzed in the public server usegalaxy.org. The reads were aligned to the murine transcriptome mm9 using TopHat2 (version 2.1.0). Further, differential gene regulation and Fragment Per Kilobase Million (FPKM) values were obtained by Cuffdiff (version 2.2.1) and Cuffnorm (version 2.2.1.1), respectively. The read counts obtained by HTSeq (version 0.9.1) were utilized for PCA in R. Genes with FPKM of <0.2 were excluded from the analysis to reduce background signals. Gene ontology was performed using PANTHER GO Ontology database (Binomial test with Bonferroni correction), and pathways with FDR < 0.05 were considered as significant. Gene set enrichment analysis (GSEA) was performed using standard parameters (Signal2Noise metric for gene ranking). Accession number for RNA sequencing (RNA-seq) data files is GSE153491.
Tissue microarray analysis in human PDAC samples
Pancreatic cancer tissue for tissue microarray (TMA) analyses was derived from the Departments of Pathology and General, Visceral and Pediatrics Surgery of the University Medical Center Göttingen in accordance to the ethical regulations of the institute (70112108). TMAs were prepared from 53 patients with PDAC with three to four cores per patient. IHC reactions were performed as described previously (14) and utilizing an automated staining machine (OMNIS). Briefly, 2 μm tissue sections were incubated in EnVision Flex Target Retrival Solution, followed by incubation with primary antibodies against EZH2 (NCL-L-EZH2; Leica) diluted at 1:50 and GATA6 (NOVUS Biologicals) diluted at 1:500. Polymeric secondary antibodies coupled to horseradish peroxidase (EnVision Flex+; Dako) and 3,3′-diaminobenzidine tetra hydrochloride (Dako) were applied to visualize the sites of immunoprecipitations. Tissue samples were analyzed by light microscopy after counterstaining with Meyer's hematoxylin. TMAs were evaluated for nuclear EZH2 and GATA6 expression by an experienced scientist under supervision of a pathologist and were scored in accordance to the Immune Reactive Sore (IRS).
For further experimental procedures, please refer to Supplementary Materials and Methods and Supplementary Tables S4–S6.
Results
EZH2 drives PDAC progression by promoting PDAC cell plasticity, dedifferentiation, and metastasis
In order to elucidate the functional relevance of EZH2 in the development and maintenance of PDAC, we chose to employ a transgenic mouse model in which Nfatc1 expression is untangled from EZH2 regulation. In this model, Cre-mediated expression of HA-tagged NFATc1 occurs constitutively under the control of the Rosa26 promoter (caNFATc1;KrasG12D mice; ref. 15), thus prohibiting previously described (14) alterations of Nfact1 expression upon interference with EZH2 activity. As described before (15, 17), caNFATc1;KrasG12D mice give rise to all steps of pancreatic carcinogenesis with metaplastic lesions initially occurring as soon as 2 weeks after birth. Eight-week-old mice already displayed the full PanIN PDAC precursor spectrum, whereas nearly all 7-month-old animals suffered from PDAC (Supplementary Table S7). Importantly, although EZH2 expression was nearly absent in acinar cells in this model, we detected strong EZH2 expression in PDAC precursor lesions and, most notably, in invasive PDAC (Fig. 1A; Supplementary Fig. S1A). In order to explore the impact of EZH2 expression on PDAC development and progression in this experimental model, caNFATc1;KrasG12D mice were crossed with Ezh2fl/fl mice (13) to obtain Ezh2fl/fl;caNFATc1;KrasG12D animals. Surprisingly, mice with homozygous pancreatic Ezh2 loss died 5 to 10 days postnatally. Necropsy of Ezh2fl/fl;caNFATc1;KrasG12D mice revealed signs of liver inflammation, severe pancreatic atrophy, or complete absence of the pancreatic organ (Supplementary Fig. S1B), indicating that in the context of constitutively active NFATc1 the histone methyltransferase is required for the development of a morphologically and functionally intact pancreas. Given that this phenotype prohibited long-term evaluation of the model, we interbred Ezh2fl/fl;caNFATc1;KrasG12D mice with EZH2 wild-type animals to obtain Ezh2fl/+;caNFATc1;KrasG12D mice with heterozygous EZH2 expression. Haploinsufficient animals were viable, and young mice did not show signs of pancreatic developmental limitations. Ezh2fl/+;caNFATc1;KrasG12D mice that were allowed to age survived longer than caNFATc1;KrasG12D control mice (162 vs. 145 days, Supplementary Fig. S1C), although not to a significant extent. Comparable with caNFATc1;KrasG12D control littermates, the pancreas of 8-week-old Ezh2fl/+;caNFATc1;KrasG12D mice showed pancreatic architectural changes consistent with acinar-to-ductal metaplasia and PanIN lesions (Fig. 1A; Supplementary Fig. S1B; Supplementary Table S7). However, although caNFATc1;KrasG12D mice developed invasive PDAC with an incidence of nearly 100%, only 5 of 25 (20%) Ezh2fl/+;caNFATc1;KrasG12D mice that were allowed to age displayed PDAC (Fig. 1B). PDAC that did develop in Ezh2fl/+;caNFATc1;KrasG12D mice was over all more differentiated than the tumors observed in the caNFATc1;KrasG12D group (Fig. 1A), suggesting that EZH2 promotes dedifferentiation programs in established PDAC. Given that tumor dedifferentiation is associated with tumor cell invasion and metastasis, we thoroughly analyzed the incidence of metastases in tumor-bearing survival mice of both genotypes. Interestingly, we found a tendency toward a reduced incidence of liver metastases in Ezh2fl/+;caNFATc1;KrasG12D- compared with caNFATc1;KrasG12D mice (Fig. 1C and D), suggesting an involvement of EZH2 in driving PDAC progression and tumor spreading.
In order to further delineate the functional implications of EZH2 in PDAC maintenance, we isolated primary PDAC cells from EZH2-expressing caNFATc1;KrasG12D mice (further referred to as NKC cells; ref. 15) and depleted EZH2 utilizing an shRNA-based approach (Supplementary Fig. S1D). In line with the proposed tumor-promoting role of EZH2 in PDAC, shRNA-mediated EZH2 depletion reduced NKC cell proliferation and anchorage-independent growth (Fig. 2A and B) and reduced invasion and sphere formation (Fig. 2C and D).
Finally, we took advantage of a PDAC TMA with tumor samples from 53 patients and performed EZH2 IHC analysis. Approximately 75% of PDAC samples stained moderately to strongly positive for the histone methyltransferase (defined as IRS score >3; ref. 18), whereas EZH2 expression was either absent or weakly expressed in the epithelial part of healthy pancreatic tissue (Fig. 2E). Correlation of EZH2 expression with clinical parameters of the respective PDAC donor patients did not reveal an association of EZH2 expression with survival, tumor–node–metastasis stage, or tumor recurrence (Supplementary Table S8). However, although EZH2 expression in transformed tissue was detectable in both moderately and poorly differentiated PDAC (Fig. 2E and F), and in line with previous reports (8), EZH2 expression was significantly stronger in high-grade tumors (Fig. 2F). Hence, and consistent with the phenotypic characteristics of Ezh2fl/+;caNFATc1;KrasG12D mice, our findings in human PDAC suggest a critical involvement of the histone methyltransferase in PDAC dedifferentiation.
Together, our functional in vitro and in vivo studies in mice and human specimens indicate that EZH2 activity critically controls cell plasticity and dedifferentiation in PDAC, thus driving tumor progression in favor of a highly aggressive PDAC phenotype.
EZH2 controls gene signatures related to PDAC differentiation and subtype identity
In order to elucidate the mechanism underlying EZH2-dependent regulation of gene signatures involved in pancreatic plasticity and dedifferentiation, we subjected shRNA control and shRNA EZH2 NKC cells to RNA-seq–based transcriptome analysis. PCA and sample-to-sample distances confirmed the similarity of triplicates and indicated two different conditions (Supplementary Fig. S2A). Although gene signatures related to tumor cell proliferation, migration, and metastasis were enriched in shRNA control samples, pathways related to reduced growth or apoptosis were found to be enriched upon EZH2 depletion (Supplementary Fig. S2B). Consistently, gene set enrichment analysis (GSEA) revealed a positive enrichment of genes associated with a favorable prognosis in PDAC in shRNA EZH2 samples (Fig. 3A), thus further supporting an oncogenic function of the histone methyltransferase. Interestingly, and in accordance with our findings in transgenic mice and human PDAC, our transcriptome analysis further linked EZH2 depletion to induction of gene signatures associated with cellular differentiation (Supplementary Fig. S2B–S2D), arguing that the histone methyltransferase controls tumor cell plasticity by repressing transcription programs required for the maintenance of a differentiated state.
Given that our data linked EZH2 activity with an aggressive PDAC phenotype, dedifferentiation, and an unfavorable tumor prognosis, we next asked whether EZH2 controls gene signatures associated with a particular molecular PDAC subtype identity. Despite discrepancies in the specific definition and nomenclature of molecular PDAC subtypes, transcriptome and epigenome analyses consistently identified two major PDAC subtypes, which we further refer to as basal-like and classical subtypes (BL- and CL subtypes, respectively; refs. 3, 7, 19–22). While the BL subtype is associated with a higher tumor grade and poor survival, the CL subtype exhibits a more differentiated phenotype and exhibits a better prognosis (3, 19). In order to evaluate whether EZH2 is involved in the regulation of gene signatures defining CL versus BL PDAC subtypes, we intersected our RNA-seq data with publically available gene sets associated with CL and BL signatures (22). Interestingly, we found a positive enrichment of the CL gene signature in shEZH2 samples, whereas BL signatures were negatively enriched upon EZH2 depletion (Fig. 3B and C; Supplementary Fig. S2E and S2F). To confirm these findings, we utilized PDAC tissue from Ezh2fl/+;caNFATc1;KrasG12D mice and from an alternative genetically engineered mouse model (GEMM) characterized by homozygous Ezh2 deficiency in the presence of oncogenic Kras, but in the absence of caNFATc1 (Ezh2fl/fl;KrasG12D mice; ref. 14) and performed immunofluorescence staining of the classical and basal-like PDAC subtype markers HNF1 and KRT14 (7, 16), respectively. Importantly, both transgenic Ezh2-depleted models showed decreased KRT14 expression when compared with the respective control mice, whereas HNF1 was found to be upregulated in these tumors (Fig. 3D and E). Accordingly, Ezh2 depletion was associated with an enrichment of gene signatures controlled by the TFs HNF1 and HNF4 (Supplementary Fig. S2G and S2H) as well as with gene signatures linked to a lipogenic metabolic profile (Supplementary Fig. S2I), which has been reported to specifically occur in the CL molecular PDAC subtype (23).
Together, these data indicate that EZH2 expression critically influences subtype identity in PDAC and suggest that the tumor-promoting functions of EZH2 might result from transcriptional regulation of gene signatures associated with fostering a molecular PDAC subtype that has been linked to a dedifferentiated and aggressive PDAC phenotype (3, 7, 19, 22).
EZH2 transcriptionally represses GATA6 in favor of an aggressive PDAC phenotype
In order to identify EZH2 target genes involved in PDAC plasticity, we performed ChIP for EZH2 followed by high-throughput sequencing (ChIP-seq) analyses. In parallel, we also examined genome-wide occupancies of H3K27me3 and H3K4me3 in EZH2-proficient or -deficient NKC cells (Supplementary Fig. S3A). GREAT analysis revealed genome-wide EZH2 occupancy predominantly near the transcriptional start sites (TSS) of target genes (Supplementary Fig. S3B and S3C). Consistent with its function as a transcriptional repressor, EZH2-bound regions were enriched for the repressive H3K27me3 mark, which was decreased in EZH2-depleted cells (Fig. 4A and B). In contrast, trimethylation of H3K4 (H3K4me3), which marks transcriptionally active TSS regions, was increased at the same regions in the absence of EZH2 (Fig. 4C). In order to identify EZH2 target genes, we focused our analysis on genes showing (i) EZH2 occupancy in shRNA control samples, (ii) reduced H3K27me3 enrichment, and (iii) increased H3K4me3 occupancy upon EZH2 knockdown (Fig. 4D). Subsequently, we intersected these 965 EZH2 targets with genes for which RNA-seq analysis conducted in NKC cells revealed differential upregulation of expression in both EZH2 knockdown (shRNA EZH2 vs. shRNA control) and Ezh2 knockout (CRISPR/Cas9 Ezh2 vs. CRISPR/Cas9 control) samples (n = 47 genes; Fig. 4E; Supplementary Fig. S3D–S3F). In line with our phenotypic and functional data indicating a pivotal role of EZH2 in PDAC plasticity and dedifferentiation, gene ontology enrichment analysis linked this subset of putative direct EZH2 target genes to pathways associated with development and differentiation (Fig. 4F). Out of the 47 EZH2 targets, we selected a set of eleven genes with potential implication in PDAC biology (Supplementary Fig. S4A) and performed individual ChIP- and expression analyses (Supplementary Fig. S4B and S4C) to validate EZH2 occupancy, EZH2-dependent and TSS-specific H3K27me3 and H3K4me3 occupancies as well as increased gene expression upon EZH2 knockdown in independent experimental settings.
Importantly, one of the most abundantly regulated EZH2 targets was the TF Gata6 (Fig. 5A). GATA6 represents a pivotal regulator of endodermal lineage differentiation and pancreatic development (24, 25) and blocks pancreatic carcinogenesis and PDAC progression particularly by activation of transcription programs linked to epithelial differentiation (16, 26, 27). Moreover, previous reports have univocally characterized GATA6 as a central regulator of molecular PDAC subtype identity with high and low GATA6 expression characterizing CL and BL PDAC subtypes, respectively (3, 19, 28). EZH2 occupancy and EZH2-dependent regulation of H3K27- and H3K4 trimethylation specifically at the Gata6 TSS region could be confirmed in independent ChIP studies (Fig. 5B). Moreover, GATA6 expression in NKC cells was upregulated both at the mRNA and protein levels upon shRNA- (Fig. 5C; Supplementary Fig. S5A) or CRISPR/Cas9-mediated (Supplementary Fig. S5B and S5C) EZH2 depletion, whereas re-expression of wild-type EZH2 but not a catalytically inactive SET-domain mutated EZH2 construct, abrogated Gata6 expression in EZH2-depleted cells (Fig. 5C; Supplementary Fig. S5B). Similarly, inhibition of EZH2 methyltransferase activity with the highly specific small-molecular inhibitor EPZ6438/tazemetostat induced Gata6 transcription (Fig. 5D; Supplementary Fig. S5D), suggesting that blockade of EZH2 activity is sufficient to reinstall Gata6 expression. Further confirming Gata6 as a transcriptional EZH2 target, we also observed upregulation of Gata6 expression in primary PDAC cells and tissue derived from Ezh2fl/+;caNFATc1;KrasG12D mice when compared with the caNFATc1;KrasG12D model (Fig. 5E–G; Supplementary Fig. S5E). Importantly, EZH2-dependent repression of Gata6 transcription was not restricted to PDAC models characterized by constitutive NFATc1 activation, but was also confirmed upon CRISPR/Cas9-mediated (Supplementary Fig. S5F–S5I) or transgenic (Fig. 5F and G; Supplementary Fig. S5J) Ezh2 depletion in murine PDAC with endogenous NFATc1 expression. Based on our findings linking EZH2 activity to GATA6 repression, we asked whether EZH2 also interferes with transcription programs downstream of GATA6 and hence compared publically available GATA6 ChIP- and RNA-seq data obtained in PDAC cells (16) with our transcriptome data. Interestingly, genes that are occupied and activated by GATA6 (n = 58, Supplementary Fig. S6A) were found to be positively enriched in EZH2 knockdown samples (Fig. 5H), suggesting that EZH2 indeed controls the GATA6-dependent transcription programs in PDAC. Along with Gata6 induction, we detected increased E-cadherin expression in Ezh2fl/+;caNFATc1;KrasG12D versus caNFATc1;KrasG12D PDAC tissue (Supplementary Fig. S6B and S6C), which has been previously identified as a downstream target of GATA6 activity (16). Consistent with previous reports (3, 19, 28) and underpinning the significance of Gata6 repression for EZH2-dependent subtype determination and PDAC progression, Gata6 knockdown partially rescued increased CL target gene transcription observed upon EZH2 deficiency and restored the invasive potential of shRNA EZH2 NKC cells (Supplementary Fig. S6D–S6F).
Next, we asked whether EZH2-dependent GATA6 repression is also evident in human PDAC. Consistent with EZH2 ChIP-seq data revealing a strong enrichment of the methyltransferase at the GATA6 TSS region in human PANC-1 cells (Fig. 6A; ref. 29), CRISPR/Cas9-mediated EZH2 deletion in these cells significantly augmented GATA6 expression (Fig. 6B and C), and increased GATA6 levels were maintained in tumors formed upon orthotopic transplantation of these cells into mice (Fig. 6D; Supplementary Fig. S7A). Similarly, primary PDAC cells isolated from a PDX model upregulated GATA6 expression upon siRNA-mediated or pharmacologic EZH2 blockade (Supplementary Fig. S7B–S7F). To further scrutinize the significance of the EZH2-GATA6 axis in human PDAC specimens, we performed GATA6 RNAscope analysis (Fig. 6E) in the aforementioned PDAC TMA (Fig. 2E). Although a subset of PDAC samples showed coexpression of EZH2 and GATA6, overall GATA6 expression was significantly reduced in the cohort of EZH2high tumors (EZH2 IRS > 3, Fig. 6F), indicating that EZH2-dependent GATA6 repression is evident in a subgroup of human PDAC specimens. These findings were validated by IHC analysis for GATA6 (Supplementary Fig. S7G and S7H). Importantly, high GATA6 expression (Immunocytochemistry Score > 3) inversely correlated with tumor grade (Supplementary Fig. S7I) and reduced the propensity of local or distant tumor recurrence in EZH2high and EZH2low PDAC subtypes (Fig. 6G; Supplementary Fig. S7J). In line with previous reports (28), these data link GATA6 expression to a less aggressive PDAC course and suggest pharmacologic interference with EZH2 activity in the subset of EZH2high/GATA6low PDAC patients as a promising strategy to reinstall GATA6 expression in favor of a less aggressive tumor phenotype.
Discussion
Polycomb group proteins (PcG) including the histone methyltransferase EZH2 play pivotal roles in chromatin organization in development and differentiation and have fundamental implications in the transcriptional regulation of stem cell circuits and cancer progression (12). Accordingly, deregulation of EZH2 is frequently observed in a variety of cancer entities. For instance, increase of EZH2 expression occurs in breast and prostate cancer, in which its expression is linked to advanced tumor stages and a poor prognosis (30, 31). In accordance with its role in maintaining the self-renewal capacities of embryonic and adult stem cells (32–34), EZH2 overexpression in cancer has been associated with migration (35), invasion (36), the enrichment of cancer stem cells (37), and metastasis (38, 39). In line with these reports and previous studies in the pancreas (8, 14), our findings link EZH2 activity to PDAC dedifferentiation and plasticity and indicate that the methyltransferase promotes metastatic propensity in a transgenic mouse model of PDAC, thus characterizing EZH2 as a crucial driver of PDAC progression. Although genome-wide mapping of PcG target genes in the embryonic stem cell (ESC) genome has revealed more than 2,000 sites primarily encoding for developmental regulators (40), only little is known about direct PcG target genes and their EZH2-dependent transcriptional regulation in cancer, in general, and in PDAC, in particular. The herein described ChIP- and RNA-seq analyses provide insights into the mechanistic background of EZH2-dependent gene regulation in PDAC and link EZH2 activity to suppression of transcription programs fostering differentiation and a better outcome of patients with PDAC.
Low tumor grades, a relatively good disease prognosis, and improved therapeutic susceptibility of PDAC are unequivocally related to a subgroup of pancreatic tumors that is classified based on its molecular profile and referred to as the CL PDAC subtype (3, 7, 19, 28). In contrast to the BL PDAC subtype, which is associated with a dedifferentiated phenotype and the worst PDAC outcome (3, 19), the CL PDAC subtype is characterized by the expression and activity of lineage-defining TFs controlling endodermal cell identity (3, 7). Despite discrepancies in the definition of the molecular characteristics defining a certain PDAC subtype, the TF GATA6 is widely accepted as a robust biomarker for differentiating CL (GATA6high) and BL (GATA6low) PDAC (3, 7, 19, 28, 41–43). GATA6 drives endodermal lineage differentiation particularly via transcriptional activation of other lineage defining TFs (25), thus characterizing GATA6 as a hierarchical regulator of (endodermal) cell differentiation. Consistently, GATA6 maintains epithelial differentiation and suppresses mutant Kras-driven tumorigenesis in the mouse pancreas by activating differentiation-related transcription programs (16, 26, 27, 44). In established PDAC, GATA6 blocks dedifferentiation, abrogates epithelial–mesenchymal transition programs, and serves as a predictive marker for adjuvant chemotherapy response (16). Therefore, cumulative evidence characterizes GATA6 as a master regulator of the PDAC phenotype and a prerequisite for CL subtype identity.
Given the prognostic and therapy predictive implications of PDAC molecular subtypes (3, 19), reverting dedifferentiated BL PDAC into a less aggressive and more treatment-susceptible CL subtype represents an appealing strategy in PDAC therapy (45). Consistent with this idea, subtype-switching in PDAC has recently been reported for the subgroup of KDM6A-deficient BL PDAC in which inhibition of bromo- and extra-terminal domain proteins specifically reverts BL differentiation and restrains PDAC growth (10). Based on the critical involvement of GATA6 in CL PDAC subtype identity (3, 22, 25, 46) and PDAC differentiation (16), upregulation or stabilization of GATA6 expression constitutes an attractive strategy to foster CL subtype identity. However, the mechanisms controlling GATA6 expression in cancer, in general, and in PDAC, in particular, have remained mainly elusive. Nevertheless, reports from developmental biology provide evidence that GATA6 expression is regulated at the level of gene transcription. For instance, the SET- and MYND domain-containing protein 3 (SMYD3) targets GATA6 for H3K4 di- and trimethylation and thus promotes activation of the gene during embryonic development (47). On the other hand, GATA6 repression in ESCs is caused by increased activity of PcG members (48). Consistently, our findings link transcriptional Gata6 repression to EZH2-dependent suppression of gene signatures characterizing the CL PDAC subtype and suggests that abrogation of GATA6 expression is at least partially responsible for EZH2-driven PDAC dedifferentiation and progression. Consequently, our findings suggest pharmacologic interference with EZH2 activity as a promising strategy to induce subtype-switching in PDAC, thus fostering the acquisition of a more differentiated and less aggressive PDAC subtype with increased therapeutic susceptibility toward chemotherapeutic intervention (19, 28). However, our analyses in human samples also revealed a subgroup of patients with PDAC with coexpression of EZH2 and GATA6, suggesting the existence of EZH2-independent mechanisms controlling GATA6 expression. Indeed, genome-wide sequencing studies have revealed genetic amplification of GATA6 in approximately 8% of PDAC (1), possibly relieving the expression of the TF from epigenetic repression by EZH2. Consequently, the therapeutic efficacy of EZH2i in PDAC might be restricted to a subgroup of patients with PDAC characterized by (i) high EZH2 activity and (ii) absence of alternative mechanisms regulating GATA6 expression, thus emphasizing the significance of molecular stratification approaches prior to EZH2i in PDAC therapy.
Together, our findings provide insights into the functional and mechanistic implications of chromatin regulatory processes in PDAC progression and suggest stratification-based pharmacologic interference with the EZH2-GATA6 axis as a promising strategy to direct subtype switching in favor of a less aggressive and more therapy susceptible tumor disease.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Authors' Contributions
S. Patil: Formal analysis, validation, methodology. B. Steuber: Methodology. W. Kopp: Methodology. V. Kari: Methodology. L. Urbach: Methodology. X. Wang: Methodology. S. Küffer: Methodology. H. Bohnenberger: Methodology. D. Spyropoulou: Resources. Z. Zhang: Conceptualization. L. Versemann: Methodology. M.S. Bösherz: Investigation. M. Brunner: Investigation. J. Gaedcke: Resources. P. Ströbel: Resources. J.-S. Zhang: Resources. A. Neesse: Writing-review and editing. V. Ellenrieder: Resources. S.K. Singh: Writing-review and editing. S.A. Johnsen: Conceptualization, writing-review and editing. E. Hessmann: Conceptualization, resources, writing-original draft, writing-review and editing.
Acknowledgments
The authors sincerely thank Jessica Spitalieri, Kristina Reutlinger, Christin Kellner, Frederike Penz, and Dr. Sercan Mercan for outstanding technical assistance. Further we are thankful for the strong support by the employees of the Central Animal Facility of the UMG. The study was supported by the German Cancer Aid [70112108 and 70112505 (PiPAC) to E. Hessmann; 70113213 (Max Eder group) to A. Neesse; 70112999 (Max-Eder group) to S.K. Singh], China Scholarship Council (to X. Wang and Z. Zhang), the Wilhelm-Sander-Stiftung (2017.107.1 to E. Hessmann), the DFG (KFO 5002 to E. Hessmann and S.K. Singh), and the Volkswagen-Stiftung/Ministry for Culture and Science in Lower Saxony (MWK; 11-76251-12-3/16 to V. Ellenrieder).
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.