Reciprocal regulation of pancreatic ductal adenocarcinoma growth and molecular subtype by HNF4α and SIX1/4

Abstract

Objective:

Pancreatic ductal adenocarcinoma (PDAC) is an aggressive malignancy with a five-year survival of less than 5%. Transcriptomic analysis has identified two clinically relevant molecular subtypes of PDAC: Classical and Basal-like. The Classical subtype is characterized by a more favorable prognosis and better response to chemotherapy than the Basal-like subtype. The Classical subtype also expresses higher levels of lineage specifiers that regulate endodermal differentiation, including the nuclear receptor HNF4α. The objective of this study is to evaluate the role of HNF4α, SIX4 and SIX1 in regulating the growth and molecular subtype of PDAC.

Design:

We manipulate expression of HNF4α, SIX4 and SIX1 in multiple in vitro and in vivo PDAC models. We determine the consequences of manipulating these genes on PDAC growth, differentiation, and molecular subtype using functional assays, gene expression analysis and cross-species comparisons with human datasets.

Results:

We show that HNF4α restrains tumor growth and drives tumor cells toward an epithelial identity. Gene expression analysis of murine models and human tumors shows that HNF4α activates expression of genes associated with the Classical subtype. HNF4α also directly represses SIX4 and SIX1, two mesodermal/neuronal lineage specifiers expressed in the Basal-like subtype. Finally, SIX4 and SIX1 drive proliferation and regulate differentiation in HNF4α -negative PDAC.

Conclusion:

Our data show that HNF4α regulates the growth and molecular subtype of PDAC by multiple mechanisms, including activation of the Classical gene expression program and repression of SIX4 and SIX1, which may represent novel dependencies of the Basal-like subtype.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is an extremely aggressive malignancy with a five-year survival of less than 5% [1]. Several recent genome-wide analyses of PDAC have identified discrete molecular subtypes of this disease [2, 3, 4]. Most recently, the Cancer Genome Atlas Research Network reported an integrated genomic, transcriptomic and proteomic analysis of an additional 150 PDAC samples [5]. This integrative analysis found significant overlap between the different subtypes, ultimately suggesting that PDAC can fundamentally be classified into two major subtypes based on cancer cell autonomous properties: Classical and Basal-like. These subtypes appear to be clinically relevant because the Classical subtype is associated with a better overall survival and better response to first line chemotherapy than the Basal-like subtype [2, 4, 6].

One major difference between these two subtypes is the relative levels of endodermal lineage specifiers, i.e. transcription factors that regulate development and differentiation of endodermally-derived tissues. The Classical subtype expresses high levels of these genes, including HNF4α, GATA6, FOXA2, FOXA3, and others [2]. Histopathologic and gene expression analysis have shown that pancreatic intraepithelial neoplasia (PanIN, the most common non-invasive precursor lesion of PDAC) and well-differentiated PDAC exhibit a cellular identity that is distinct from the normal pancreas and is characterized by upregulation of transcripts highly expressed in the foregut. This suggests pancreatic neoplasia initially adopts a cellular identity resembling the tissue from which the pancreas originated during development [7, 8]. Moreover, these correlations raise the question of whether specific endodermal lineage specifiers are not merely markers of discrete subtypes, but might directly regulate molecular subtype, malignant potential and therapeutic response in PDAC.

The nuclear receptor superfamily member HNF4α is a master regulator of epithelial differentiation in multiple tissues, including the gastrointestinal (GI) tract [9, 10, 11, 12]. HNF4α also regulates a variety of biological processes that impact tumor progression, including proliferation [13] and metabolism [14]. Consistent with genomic studies mentioned above, HNF4α protein is detectable in human pancreatic neoplasia, and its levels correlate with tumor differentiation state [15]. Immunohistochemistry (IHC) on tissue sections has shown that PanIN lesions and well-differentiated PDAC express higher levels of HNF4α than the normal pancreas, whereas HNF4α is downregulated in poorly differentiated PDAC.

In previous work, we have shown that HNF4α plays a critical role in a mouse model of invasive mucinous adenocarcinoma of the lung [16], a subtype of lung cancer that expresses many of the same foregut markers observed in the Classical subtype of PDAC. Based on these data and the differential expression of HNF4α between the Classical and Basal-like subtypes of human PDAC, we sought to determine whether HNF4α also plays a functional role in this disease. We found that HNF4α is a major activator of the Classical gene expression program and restrains PDAC growth in multiple models of the disease. Although HNF4α loss is not sufficient for complete subtype switching, HNF4α represses the expression of SIX4 and SIX1, two mesodermal/neuronal lineage specifiers highly expressed in the Basal-like subtype of PDAC.

Results

Hnf4α deletion accelerates tumorigenesis in a mouse model of pancreatic ductal adenocarcinoma

To dissect the role of HNF4α in PDAC, we first used a mouse model of pancreatic neoplasia that closely mimics the human disease. This model employs the Pdx1-Cre transgene to activate expression of KRAS^G12D from its endogenous locus during pancreatic development, which thereby induces neoplasia that closely models human PanIN and PDAC [17, 18]. Greater than 90% of human PanINs and PDACs contain mutations that result in constitutive activation of the KRAS oncogene, which is believed to be the initiating event in most cases [19]. KRAS is a small GTPase that interacts with receptor tyrosine kinases at the plasma membrane to transduce growth factor-induced signals to several intracellular effectors [20].

Pdx1-Cre; Kras^LSL-G12D mice develop acinar to ductal metaplasia (ADM) and murine PanIN (mPanIN) lesions with complete penetrance, and these in situ lesions occasionally progress to invasive cancer in older mice. To accelerate cancer progression, a conditional allele of the p53 tumor suppressor is incorporated into the model. Excision of one allele of p53 (Pdx1-Cre; Kras^LSL-G12D; p53^F/+ mice) accelerates neoplastic progression to invasive and metastatic adenocarcinomas, which often develop in mice 3–6 months of age [17]. We initially evaluated a cohort of control mice (Pdx1-Cre; Kras^LSL-G12D; p53^F/+; Hnf4α^+/+) at 12 weeks of age (n=17, Figure 1A–C). As expected from published reports [15], HNF4α was robustly expressed in preinvasive lesions (ADM and mPanIN) arising in these mice, generally at higher levels than normal acinar cells (Figure 1A, left). Careful evaluation of HNF4α levels in the subset of mice harboring PDAC (n=7) showed that 3/7 tumors were diffusely (>90% of cells positive) HNF4α -positive. In contrast, one tumor contained a mix of HNF4α -positive and negative cells and 3/7 tumors were completely HNF4α -negative, despite the presence of adjacent HNF4α -positive mPanIN (Figure S1A). HNF4α -positive and mixed tumors exhibited a range of differentiation states (including well, moderately and poorly differentiated), whereas the three HNF4α -negative tumors were poorly differentiated. HNF4α expression in murine PDAC correlated with expression of the foregut marker Galectin 4 (Figure S1B). HNF4α -mixed tumors may represent the subclonal outgrowth of Basal-like/squamous cells in a background of Classical PDAC, which has recently been documented in the human disease [21]. Taken together, these data shown that HNF4α is expressed at higher levels in early pancreatic neoplasia than normal pancreas, but that HNF4α can also be stochastically downregulated during tumor progression.

Figure 1. Hnf4α deletion accelerates tumorigenesis in a mouse model of pancreatic ductal adenocarcinoma.

A. H&E and IHC for HNF4α on pancreatic neoplasia from Kras^LSL-G12D/+; p53^F/+; Pdx1-Cre; Hnf4α^F/F mice and Hnf4α^+/+ controls at 12 weeks of age. Scale bar: 100 microns. ADM: acinar to ductal metaplasia. mPanIN: Pancreatic intraepithelial neoplasia.

B-C. Percentage of Kras^LSL-G12D/+; p53^F/+; Pdx1-Cre; Hnf4α^F/F mice (n=21) and Hnf4α^+/+ controls (n=17) with macroscopic (B) and microscopic (C) PDAC at 12 weeks of age. *p<0.05, Chi-Square.

D. Survival curve of Kras^LSL-G12D/+; p53^F/+; Pdx1-Cre; Hnf4α^F/F mice (n=27) and Hnf4α^+/+ controls (n=38). p=0.0027, Log-rank.

Twelve isoforms of HNF4α have been identified to date [22], which arise as a result of transcription from two different promoters (P1 and P2) as well as alternative exon splicing. The P1 and P2 promoters are activated in a tissue-specific manner, and P1 and P2 isoforms have distinct roles in tumorigenesis [23, 24]. Using monoclonal antibodies specific for the P1 and P2 isoforms, we found that HNF4α -positive neoplasia predominantly expressed the P2 isoform in this model (Figure S1C). We also used these antibodies to evaluate isoform expression in a small panel of HNF4α -positive patient derived xenograft (PDX) models of human PDAC. A majority of these HNF4α -positive PDX models (4/7) only expressed the P2 isoform, as seen in the autochthonous mouse model (representative images in Figure S1C). However, the other three PDX models expressed both P1 and P2 isoforms. These data suggest that the autochthonous model recapitulates the biology of one subset of HNF4α -positive human PDAC, but the role of HNF4α in the dual-positive subset will require additional investigation.

In contrast to control mice, all neoplasia (ADM, mPanIN and PDAC) in Hnf4α^F/F mice at 12 weeks were HNF4α -negative (n=21, Figure 1A, right). We found no evidence of incomplete recombinants, which can be observed in the Cre-Lox system when there is selection for retention of a conditional allele. Although Hnf4α deletion had no obvious effect on the morphology or quantity of preinvasive lesions, it significantly increased the proportion of mice with either macroscopic (Figure 1B) or microscopic PDAC (Figure 1C) at 12 weeks of age. As expected, Hnf4α-deleted PDAC did not express Galectin 4 (Figure S1B).

In a separate survival analysis, we found that Hnf4α deletion significantly reduces survival in this model (median survival 113 days in Hnf4α^F/F mice vs. 147 days in Hnf4α^+/+ mice, Figure 1D). Consistent with our analysis of tumors in 12-week old mice, a subset of tumors in Hnf4α^+/+ mice exhibited stochastic HNF4α downregulation, whereas tumors in Hnf4α^F/F mice were consistently HNF4α -negative. Specifically, 9/20 tumors available for evaluation in Hnf4α^+/+ mice were HNF4α -mixed and 2/20 were HNF4α -negative (all of which were included in the overall survival analysis). Most tumors in the survival analysis harbored both moderately and poorly differentiated components, although a minority were entirely poorly differentiated (2/20 in Hnf4α^+/+ mice and 4/12 evaluated in Hnf4α^F/F mice). Almost all survival mice (18/20 Hnf4α^+/+ and 11/12 Hnf4α^F/F mice) harbored metastases to sites such as liver, kidney, peritoneal lymph nodes, diaphragm and lung.

In the absence of p53 mutations, Pdx1-Cre; Kras^LSL-G12D mice develop ADM and PanIN with high penetrance, but these lesions rarely progress to PDAC [17]. To determine whether Hnf4α deletion would be sufficient for PDAC development, we aged a cohort of Pdx1-Cre; Kras^LSL-G12D; Hnf4α^F/F mice (and Hnf4α^+/+ controls, n=4–8 mice/group at each timepoint) and analyzed their pancreata at 6 and 9 months of age. We observed ADM and PanIN in nearly all mice of both genotypes, but no evidence of PDAC, despite the lack of HNF4α expression in Hnf4α^F/F mice (Figure S1D).

Taken together, these data show that HNF4α restrains the growth of pancreatic neoplasia in this model system, predominantly at the stage of invasive PDAC. These results are also consistent with the biology of human PDAC, specifically the observation that Basal-like (HNF4α -low) tumors confer a worse prognosis, but that both HNF4α -positive and HNF4α -negative disease are highly lethal [2, 4, 6].

HNF4α reconstitution impairs PDAC growth and imposes epithelial differentiation in vitro and in vivo

To more precisely define the role of HNF4α in PDAC, we developed additional systems to manipulate its expression in established neoplasia. First, we generated two cell lines (HC800 and HC569) from PDAC arising in Pdx1-Cre; Kras^LSL-G12D; p53^F/+; Hnf4α^F/F mice using standard two-dimensional (2D) culture conditions. We then stably transduced these cell lines with lentivirus enabling doxycycline (dox) induction of the HNF4α 8 isoform, a major product of the P2 promoter. We used a P2 isoform for reconstitution studies because only HNF4α -P2 is expressed in the autochthonous model (Figure S1C). We then obtained single cell clones that exhibited optimal dox induction of HA-tagged HNF4α (Figure S2A) and subjected them to proliferation analysis. Exogenous HNF4α significantly inhibited the proliferation of both cell lines in vitro (Figure 2A) with no obvious induction of apoptosis as assessed by cleaved caspase-3 (Figure S2B). HC800 cells expressing exogenous HNF4α exhibited a more epithelial morphology in vitro (Figure S2C), suggesting that HNF4α reconstitution also modulated their differentiation state. Additionally, we stably transduced five different HNF4α -low/negative human PDAC cell lines, including a line derived from the xenograft PDX220 [25] with lentivirus encoding dox-inducible HNF4α 8 (Figure S2D). We observed that PDX220, Panc10.05, and BxPC3 cells displayed significantly reduced proliferation when HNF4α expression was induced compared to control cells, while MIA PaCa-2 and Panc1 proliferation was not affected (Figure 2A).

Figure 2. Restoration of HNF4α expression inhibits proliferation and induces epithelial differentiation (mesenchymal to epithelial transition) in PDAC.

A. Quantitation of proliferation in murine (HC800 and HC569) and human (PDX220, Panc10.05, BxPC3, MIA PaCa-2, and Panc1) PDAC cells stably expressing TRE-HNF4α 8 cells in the presence or absence of doxycycline (n=3 biological replicates, 2–3 independent experiments). Primary murine PDAC cell lines were single cell cloned. Data show growth fold change normalized to control condition. Graph represents mean +/− SEM. *p< 0.05; **p< 0.01; ***p< 0.001, Welch’s t test.

B-C. HC800, HC569, and PDX220 cells stably transduced with TRE-HNF4α 8 were injected subcutaneously into NSG mice. Mice were fed with doxycycline chow (red) or regular chow (blue) starting 1 week before transplantation. B. Flank tumor volume and mass are shown. C. Flank tumor growth was assessed by measuring mitoses per high power field (HPF). Data represented as mean ± SEM. *p< 0.05; **p< 0.01; ***p< 0.001 by Wilcoxon test (for tumor volume) or Mann-Whitney test (for tumor mass and mitoses).

D-F. HC800-TRE-HNF4α 8 cells were injected intraperitoneally into NSG mice. Mice received either doxycycline chow or control chow starting 1 week prior to cell injection (n=10 mice per group). Tumor cells were stably transduced with luciferase expression vector prior to injection. D. Quantitation of relative light units (RLU) in each group 20 days post flank injections. Graph shows fold change in RLU normalized to basal measurements. Data represented as mean ± SEM. *p<0.05 by t test. E. Survival analysis. ***p<0.001 by log-rank test. F. Representative H&E and IHC for HA-tagged HNF4α and target genes in intraperitoneal tumors from each group. Yellow line demarcates two fascicles of tumor cells oriented in opposite directions. Yellow arrows mark glandular differentiation. Scale bar: 100 microns.

To test the effect of HNF4α restoration in vivo, we first injected two murine (HC800 and HC569) and one human (PDX220) cell lines subcutaneously into mice fed control or dox-containing chow to induce HNF4α. HA-tagged HNF4α was robustly expressed in tumors of mice on dox chow (Figure S2E and S2H). Exogenous HNF4α significantly inhibited tumor growth of all three PDAC cell lines as assessed by both volume and mass (Figure 2B), and this was accompanied by a significantly lower proliferation rate than controls (Figure 2C). To better recapitulate the biology of PDAC, we also evaluated the growth of HC800 cells in the peritoneal space, which is one of the first sites of extra-pancreatic dissemination as PDAC progresses. We found that these cells grew readily in the peritoneum, forming numerous macroscopic nodules on the serosal surface of intraperitoneal organs. In a formal survival study, dox-mediated induction of HNF4α reduced tumor burden at 20 days post injection (Figure 2D) and significantly increased overall survival (Figure 2E).

In addition to restraining proliferation, HNF4α had a clear impact on differentiation state in vivo, shifting tumors toward a more well differentiated epithelial morphology (Figures 2F, S2E and S2H). This was particularly striking in the HC800 and PDX220 cell lines. Control HC800 tumors had a uniform morphology composed of spindled cells growing in a fascicular pattern that is characteristic of a sarcomatoid or quasi-mesenchymal differentiation state (Figure 2F, top row). IHC for the HA tag confirmed that exogenous HNF4α was not expressed in these tumors. In contrast, tumors growing in mice fed dox chow adopted the morphology of a moderately to poorly differentiated adenocarcinoma, with a glandular rather than fascicular growth pattern and cells that were round rather than spindled. These tumors expressed uniformly high levels of HA-tagged HNF4α, its target PK-L, and Galectin 4 (Figure 2F, bottom row). Of note, neither HC800 nor HC569 tumors expressed ΔNp63, consistent with their lack of squamous morphology (Figure S2F). In contrast, control PDX220 tumors were comprised of cells with two distinct morphologies: adenocarcinoma, which was predominantly HNF4α -positive/ ΔNp63-negative, and squamous carcinoma, which was predominantly HNF4α -negative/ ΔNp63-positive (Figure S2G). Importantly, we have not ruled out the possibility that rare dual-positive cells exist in control tumors. In contrast to controls, PDX220 tumors arising in dox-treated mice consisted entirely of adenocarcinoma cells that were better differentiated than control tumors (Figure S2H), showing that HNF4α not only suppressed squamous differentiation in this model but also modulated the differentiation state of the adenocarcinoma component. Taken together, these results show that restoring HNF4α can induce major changes in differentiation state and inhibit PDAC growth both in vitro and in vivo.

Hnf4α deletion in PDAC organoids alters differentiation and three dimensional growth pattern in vivo

Given that HNF4α is often stochastically downregulated during PDAC progression, we next developed a system that would enable us to inactivate endogenous Hnf4α in established murine pancreatic neoplasia. In mouse models of lung cancer, we have previously used a sequential recombinase strategy to delete genes in established tumors [16]. Merging this approach with pancreatic organoid culture systems [26], we established PDAC organoids by enzymatically dissociating pancreata from Kras^FSF-G12D/+; p53^Frt/Frt; Rosa26^FSF-CreERT2; Hnf4α^F/F mice. We then treated these cells in vitro with adenovirus expressing the FlpO recombinase, which activated transcription of oncogenic KRAS^G12D and inactivated p53, leading to the emergence of HNF4α -positive PDAC organoids (Figure 3A and S3A). FlpO also activated transcription of Cre^ERT2 from the Rosa26 locus. This enabled us to treat these cultures with 4-hydroxy-tamoxifen (4-OHT), thereby driving Cre^ERT2 into the nucleus where it excised the respective conditional alleles. IHC and qRT-PCR demonstrated loss of HNF4α as well as its target Pklr in four independently derived organoid cultures (Figure S3A–B).

Figure 3. Hnf4α deletion alters differentiation and three dimensional growth in murine organoid models of PDAC.

A. Schematic of organoid derivation from pancreata of Kras^FSF-G12D/+; p53^Frt/Frt; Rosa26^FSF-CreERT2; Hnf4α^F/F mice.

B-E. H&E and IHC analysis of a single cell suspension of SC1853 organoids injected subcutaneously into NSG mice. Mice were fed control (B-C) or tamoxifen (D-E) chow starting at 1 week prior to organoid implantation. Tumors were analyzed at 6 weeks post injection. B and D: scanning magnification, scale bar: 1 mm. C and E: High power images, scale bar: 100 microns.

We next asked whether loss of HNF4α would affect the growth and differentiation state of these organoid lines in vivo. We injected two lines (SC1853 and SC1693) subcutaneously into NOD/SCID-gamma chain deficient (NSG) mice. In mice treated with control chow, both lines formed macroscopic, fluid filled cysts (Figure 3B and S3C). These cysts were predominantly lined by a single layer of HNF4α -positive columnar epithelial cells (Figure 3C and S3D, top row). We also identified small areas of well-differentiated adenocarcinoma (Figure 3C and S3D, bottom row) within the walls of these cystic structures.

In contrast, tumors that grew in mice on tamoxifen chow exhibited a completely distinct three-dimensional growth pattern and differentiation state. These tumors consisted of a mixture of invasive adenocarcinoma and microscopic cystic structures (Figure 3D and S3E). Moreover, the adenocarcinoma component was less well differentiated than control tumors. Hnf4α-deleted tumors from both organoid lines exhibited areas of moderately differentiated (intermediate grade) HNF4α -negative adenocarcinoma (Figure 3E and S3F). Moreover, Hnf4α deletion in tumors of the SC1853 organoid was sufficient to induce areas of high grade, poorly differentiated adenocarcinoma (Figure 3E, bottom row).

HNF4α activates the Classical gene expression program in PDAC

Given its impact on growth and differentiation state in multiple models of PDAC, we sought to formally test the hypothesis that HNF4α regulates molecular subtype in PDAC. As expected from prior studies [2, 5], we observed a robust correlation (r = 0.71, p = 9.1e-13) between HNF4α expression and Classical score in high purity tumors from the TCGA-PAAD data (Figure 4A, left). In comparison, HNF4α levels and Basal-like score were weakly negatively correlated (r = 0.48, p = 1.4e-5) in this dataset (Figure 4A, right). We also evaluated the relationship between HNF4α levels and survival in high purity tumors (>30% cellularity) from two independent datasets [2, 5] by either splitting cases around median HNF4α expression or into upper and lower quartiles of HNF4α expression (Figure S4A). In the PACA-AU dataset, patients with tumors in the upper quartile of HNF4α expression survived significantly longer than in the lowest quartile (p=0.012), consistent with known survival differences between the Classical and Basal-like subtypes. Other analyses showed a similar (though non-significant) trend. Interestingly, in the TCGA dataset, a subset of patients with HNF4α-high tumors were still alive at the time of analysis, whereas none of the patients in the HNF4α-low cohort survived beyond the last recorded time point (2000 days).

Figure 4. HNF4α regulates PDAC molecular subtype.

A. Correlation between HNF4α expression in patients with Classical Score (r = 0.71, p= 9.1e-13) and Basal-like score (r = −0.48, p= 1.4e-5) in TCGA-PAAD data.

B. Scores were generated for Basal-like and Classical subtype expression from RNA-seq data. Each point represents one sample, and samples were grouped by their HNF4α status. Subtype scores were calculated using mouse homologs of the established Moffitt et al. signatures. P-values obtained via t-test (paired for cell lines). ***p<0.001, **p<0.01, *p<0.05.

C. Gene Set Enrichment Analysis was performed comparing the HNF4α -positive samples to the HNF4α -negative samples, and their respective Signal-to-Noise Ratios (SNR) are plotted. Red indicates that expression of the gene increases with HNF4α expression, while blue indicates expression decreases.

Next, we performed gene expression analysis on autochthonous murine PDAC tumors from Hnf4α^+/+ and Hnf4α^F/F mice (n=6 per genotype), using IHC to exclude any tumors from wild type mice that had stochastically downregulated HNF4α. We identified a total of 597 differentially expressed genes in this analysis (Supplemental Table 1), including known HNF4α targets in normal tissue such as Pklr and Apob. To determine whether Hnf4α deletion altered molecular subtype of these tumors, we assigned a Classical and Basal-like score to each dataset based on mouse homologues of the scores described in [4]. The Classical score of Hnf4α-deleted tumors was significantly lower than Hnf4α^+/+ tumors (Figure 4B–C, left panels). In contrast, the Basal-like score was not significantly different between the two groups, although there was a slight trend toward a higher score in the Hnf4α-deleted tumors. Evaluation of lineage specifiers that regulate endodermal differentiation (Figure S4B) showed that some appear to be HNF4α targets in this model (e.g., HNF1, FoxA3 and CDX2), whereas others are not significantly affected by Hnf4α deletion, suggesting they are situated adjacent to or upstream of HNF4α in the hierarchy of lineage specifiers specifying the Classical subtype of PDAC.

Using Illumina Correlation Engine, we performed pathway analysis on genes differentially expressed between HNF4α -positive and HNF4α -negative autochthonous tumors (Supplemental Table 2). GO terms and pathways associated with epithelial differentiation (e.g., “cell-cell adhesion”) and other known functions of HNF4α (e.g., “Maturity onset diabetes of the young” and various metabolic pathways) were significantly enriched in HNF4α -positive tumors. In contrast, GO terms associated with ectodermal/mesodermal differentiation (e.g., “regulation of neurogenesis” and “skeletal system development”) were significantly enriched in HNF4α -negative tumors. Consistent with these results, the Body Atlas analysis tool showed that the gene expression profile of HNF4α -positive tumors correlates highly with endodermally derived tissues, such as the GI tract, whereas HNF4α -negative tumors correlate with ectoderm/mesoderm derived tissues found in the central nervous system and cardiovascular system (Figure S4C and Supplemental Table 2).

To expand upon these observations and determine whether HNF4α can regulate PDAC identity in vitro in a cell-autonomous manner, we performed gene expression analysis on the previously described cell lines and organoids isogenic for HNF4α expression (Supplemental Tables 3 and 4). In both cell line systems, HNF4α -positive cells had a higher Classical score than HNF4α -negative cells (Figure 4B–C, middle and right hand panels). This effect was more pronounced in organoids than in 2D lines, where the difference was modest but still significant. In contrast, the Basal-like score did not significantly change with HNF4α -expression. Intersection of all three datasets identified a total of 49 genes that were upregulated by HNF4α in all model systems (Figure S4D and Supplemental Table 5). Very few HNF4α downregulated genes were consistently found in all model systems due primarily to lack of overlap with the organoid dataset. When we intersected in vivo tumors and 2D cell lines, excluding the organoid dataset, we identified 65 commonly repressed genes (Figure S4D and Supplemental Table 5).

Taken together, these data show that HNF4α is a critical regulator of cellular identity and molecular subtype in multiple models of PDAC. Specifically, HNF4α is required for activation of the complete Classical gene expression program. Although loss of HNF4α is not sufficient for full conversion to a Basal-like state, a subset of genes associated with the Basal-like subtype appear to be repressed by HNF4α

Analysis of HNF4α -chromatin interactions in PDAC

In order to identify direct HNF4α target genes and investigate the impact of HNF4α on chromatin accessibility, we performed ChIP-seq with an antibody that recognizes the HA-tagged HNF4α as well as ATAC-seq in HC800 and HC569 cells. For both assays, cells were treated with dox or control media for one week prior to chromatin analysis. We identified 18,419 and 15,593 reproducibly significant peaks by HA ChIP-seq in HC569 and HC800 dox-treated cells, respectively. We found that 84.4% of all sites were present in both cell lines, showing a strong degree of concordance in HNF4α binding (Figure 5A and S5A). As expected, motifs bound by HNF4α and HNF4α were the most significantly enriched in ChIP-seq peaks in both cell lines (Figure S5B). Examination of HNF4α binding distribution showed that 45% of peaks localized to promoters, whereas 13% were found in distal regions (Figure 5B). We corroborated HNF4α occupancy in relevant and well-known direct target genes such as Pklr, Lgals4, and Hnf1a (Figure 5C). To explore the connection between binding sites and gene expression, we intersected our ChIP-seq and RNA-seq datasets. We observed 69% of the differentially expressed genes displayed binding sites for HNF4α (Figure 5D). For those genes bound by HNF4α, we found that 61% were upregulated and 39% downregulated, indicating that HNF4α preferentially acts as a transcriptional activator in this context. Functional analysis of the differentially expressed genes bound by HNF4α using Ingenuity Pathway Analysis (IPA®) revealed a significant enrichment in molecular and cellular functions involved in lipid metabolism and transport, cell movement, and free radical scavenging (Supplemental tables 6 and 7).

Figure 5. Global analysis of HNF4α occupancy and chromatin accessibility landscape in PDAC cells.

A. Heatmaps displaying HNF4α binding sites in a 5-kb window around the ChIP-seq peak summits in HC569 and HC800 cells.

B. Pie chart displays the genomic distribution of HNF4α binding sites.

C. Representative browser track images of three known direct targets of HNF4α in HC569 (blue) and HC800 (red).

D. Overlap between RNA-seq and ChIP-seq data. Distribution of differentially expressed genes bound and not bound by HNF4α. Bound-UP: upregulated genes with HNF4α occupancy. Bound-DOWN: downregulated genes bound by HNF4α.

E. Heatmap shows signal at ATAC-seq regions with increase (red line) and decrease (blue line) chromatin accessibility upon dox-mediated HNF4α induction in HC800 (green) and HC569 (purple) cell lines.

F. Distribution of genomic features for ATAC-seq regions with increased and reduced chromatin accessibility.

G. Representative browser track images showing intensity of ATAC-seq signal at the indicated genes in dox-treated cells (blue: HC569, red: HC800) compared to controls (light blue: HC569, pink: HC800)

H. Overlap between RNA-seq and ATAC-seq data sets. Distribution of up- and downregulated genes between more and less accessible chromatin.

I. Venn diagram displays the overlap between HNF4α -bound sites and ATAC-seq peaks with increased or reduced genomic accessibility.

We further assessed the effect of HNF4α on global chromatin accessibility by performing ATAC-seq in both cell lines upon 1 week of treatment with doxycycline or vehicle. Principal component analysis showed that control (HNF4α -negative) HC569 and HC800 cells clustered separately, but HNF4α restoration caused similar changes in chromatin accessibility in both cell lines (Figure S5C). A total of 5,872 significant ATAC-seq peaks commonly found in both cell lines were identified, with 4,232 sites exhibiting increased accessibility and 1,640 sites displaying decreased accessibility in the presence of HNF4α (Figure 5E). Genomic distribution analysis of ATAC-seq peaks showed that most of the sites with significant changes in accessibility are found in distal regions and introns. Despite not being the most frequent genomic feature found in differentially accessible chromatin, promoters were overrepresented in closed genomic sites relative to chromatin with increased accessibility (Figure 5F). We observed increased signal of ATAC-seq peaks in HNF4α target genes that are transcriptionally activated in dox-treated samples (Figure 5G). On further inspection, we found that genomic regions with increased accessibility were enriched in motifs for binding of AP1 and nuclear receptor family members HNF4α /HNF4α /RAR, in addition to some repetitive motifs. In contrast, closed chromatin peaks lack HNF4α binding motifs but show significant enrichment for AP1 binding sites and some repetitive motifs (Figure S5D).

Intersection of ATAC-seq and RNA-seq data identified 743 differentially expressed genes (488 upregulated and 255 silenced genes) that displayed changes in chromatin accessibility upon HNF4α restoration. The vast majority of the upregulated genes correlated with genomic regions with increased accessibility (84%), while most of the repressed genes were associated with ATAC-seq peaks that had reduced genomic accessibility (55%), although a considerable percentage (31%) correlated with genomic areas that become more accessible (Figure 5H and S5E).

Additionally, intersection of HNF4α binding sites with ATAC-seq data demonstrated that less than 10% of the HNF4α bound sites displayed changes in chromatin accessibility and less than 20% of the ATAC-seq peaks with differential accessibility were occupied by HNF4α, suggesting that other transcription factors may be contributing to the HNF4α -dependent alterations in chromatin accessibility (Figure 5I and S5F). HNF4α bound chromatin with increased accessibility contained a large fraction of introns and distal regions, and the HNF4α binding site was the most highly enriched motif in these regions (E value=3.9e-¹⁴⁰, Figure S5G–H). In contrast, HNF4α bound regions with reduced accessibility contained a higher percentage of promoters and exhibited no enrichment for the HNF4α motif. Although we identified the HNF4α motif in HNF4α bound chromatin regions with no changes in accessibility, it did not rank as a top site (E value=1.4e-³⁸). Interestingly, the strongest HNF4α motifs were found in chromatin regions with increased accessibility, suggesting that these HNF4α bound sites are more likely to become open (Figure S5I).

Finally, intersection of ChIP-seq, ATAC-seq and RNA-seq datasets demonstrated a strong and statistically significant association between HNF4α bound regions with higher chromatin accessibility and upregulated genes compared to HNF4α bound sites with reduced accessibility and repressed genes (p <0.0001, 7.321 odds ratio, Fisher’s exact test) (Figure S5J).

These results demonstrate that most of the differentially expressed genes that define the transcriptional network regulated by HNF4α in PDAC constitute direct targets of this endodermal cell fate determinant. Additionally, our data suggest that most of the changes in chromatin accessibility seem to be indirect effects of HNF4α restoration and not caused by its direct binding.

HNF4α inhibits the expression of SIX4 and SIX1 in PDAC

Next, we sought to identify mechanisms by which loss of HNF4α might lead to de-repression of a subset of genes associated with the Basal-like subtype. Six4, a member of the sine oculis homeobox family, is the most differentially expressed transcription factor in Hnf4α-deleted autochthonous tumors compared with controls (Figure 6A). Three independent studies have shown that SIX4 is more highly expressed in Basal-like/Squamous human PDAC than Classical/Progenitor [2, 5, 27]. Evaluation of high purity tumors from the TCGA dataset showed that SIX4 expression positively correlated with the Basal-like subtype of human PDAC and negatively correlated with Classical Score and HNF4α expression (Figure S6A). The Six4 paralogue Six1 is also differentially expressed in Hnf4α-deleted autochthonous tumors. SIX1 levels positively correlate with SIX4 in human PDAC, but they do not significantly correlate with PDAC subtype (Figure S6B–C), suggesting that in human PDAC, other factors may have a stronger influence on SIX1 expression than HNF4α. Overall, it appears that a subset of SIX4-high human PDAC also express high levels of SIX1 (Figure S6C).

Figure 6. HNF4α inhibits the expression of mesodermal lineage specifiers SIX1 and SIX4.

A. SIX1 and SIX4 mRNA levels in indicated samples as assessed by RNA-seq. Left panel: 6 autochthonous PDAC samples per cohort (●). Middle panel: 2 biological replicates of murine PDAC cell line HC569 (●) and HC800 (▼). Right panel: 4 independent isogenic organoid lines, SC1640 (●), SC1693 (▼), SC1853 (▄) and SC1956 (✖).

B. Representative IHC for SIX1 from indicated PDAC models with different HNF4α expression levels. Autochthonous: Pdx1-Cre; Kras^LSL-G12D; p53^F/+ tumors from Hnf4α^+/+ (left) or Hnf4α^F/F (right) mice. HC800: intraperitoneally injected cell line with dox-inducible HNF4α. Organoid: derived from Kras^FSF-G12D/+; p53^Frt/Frt; Rosa26^FSF-CreERT; Hnf4α^F/F mice. Three dimensional cultures were treated in vitro with vehicle control (left) or 4-hydroxy-tamoxifen (right) to delete Hnf4α. Yellow arrows indicate SIX1-positive cells. Scale bar: 100 microns.

C-D. Representative immunoblot for SIX1 and SIX4 in doxycycline- treated murine (C) and human (D) PDAC cell lines.

E. Representative IHC for SIX4 from subcutaneous xenograft of PDX220 cells overexpressing (DOX) or not (CONTROL) HNF4α (left panel). Scale bar: 100 microns. Representative immunoblot for SIX1 and SIX4 in doxycycline- treated human PDX220 cell line (right panel).

F. Browser track images display signal intensity of ChIP-seq in 2D lines. Blue: HNF4α ChIP in HC569. Red: HNF4α ChIP in HC800.

SIX4 and SIX1 are located immediately adjacent to each other in both mouse and human genomes and are co-expressed in several tissues during development [28]. Both transcription factors bind the MEF3 DNA motif, and genetic knockout experiments have demonstrated their functional redundancy in the development of several tissues, including skeletal muscle, sensory neurons, gonads and kidney (reviewed in [29]). SIX1 has been reported to promote the growth of the mesenchymal/Basal-like human PDAC cell lines Panc1 and MIA PaCa-2 [30, 31]. However, there has been no systematic evaluation of SIX4 in PDAC, either alone or in combination with SIX1, despite their co-expression in human tumors and evidence of significant functional redundancy during development.

To further investigate the ability of HNF4α to regulate SIX4 and SIX1 expression, we evaluated RNA and protein levels in all of our mouse models. Analysis of RNA-seq data from individual tumors showed that Six1 and Six4 were not expressed in Hnf4α^+/+ autochthonous tumors (Figure 6A, left). Three out of six Hnf4α^F/F tumors exhibited de-repression of both genes, whereas the other three Hnf4α^F/F tumors had very low levels of Six1 and Six4 transcripts (similar to control tumors). Evaluation of SIX1 protein levels by IHC in an expanded panel of autochthonous tumors showed that SIX1 was largely undetectable in Hnf4α^+/+ autochthonous tumors (Figure 6B, top row). In contrast, a panel of Hnf4α^F/F tumors from 12-week old mice (n=14) exhibited a heterogenous pattern of SIX1 expression. SIX1 protein was detectable in 8 out of 14 Hnf4α^F/F tumors, whereas the other 6 tumors were negative. Within the 8 positive tumors, the percentage of SIX1-positive cells ranged from less than 5% to greater than 75% (Figure 6B, top row and data not shown).

We next evaluated the ability of HNF4α to regulate SIX4 and SIX1 in our other model systems. Exogenous HNF4α reduced Six1/4 mRNA expression in both Hnf4α-deleted murine cell lines (HC569 and HC800) (Figure 6A, middle panels), which led to decreased SIX4 and SIX1 protein levels in vitro and virtually undetectable SIX1 expression in vivo (Figure 6B, middle row, and 6C). Exogenous HNF4α also strongly decreased SIX4 and SIX1 protein levels in three human cell lines (BxPC3, Panc10.05 and PDX220, Figure 6D–E). RNA-seq analysis of organoids showed a non-significant trend toward increased Six4 levels after Hnf4α deletion (Figure 6A, right). Although Six1 did not score as a differentially expressed gene in organoids, we were able to detect differences at the protein level. Control murine PDAC organoids were uniformly SIX1-negative, but SIX1 protein was detectable in two of the four Hnf4α-deleted organoid lines (SC1640 and SC1853, Figure 6B, lower row). ChIP-seq revealed that HNF4α directly binds both Six1 and Six4 genes in proximity to their promoter region in both HC569 and HC800 cell lines. We also identified a significant peak in the intergenic region between Six1 and Six4 (Figure 6F).

Taken together, these data show that HNF4α represses SIX1 and SIX4, two lineage specifiers that drive mesodermal and neuronal development, likely through direct binding to their regulatory elements. However, loss of HNF4α expression is not sufficient for induction of SIX1 and SIX4 in all contexts, suggesting that other redundant mechanisms help repress SIX1/4 expression in the Classical subtype of PDAC.

SIX4 and SIX1 are required for optimal growth of HNF4α -negative PDAC in vitro and in vivo

Given that Six genes are frequently dysregulated in different cancers, we decided to further explore the oncogenic function of SIX4 and SIX1 in HNF4α -negative PDAC. We initially performed CRISPR-Cas9 genome editing to ablate Six1 and Six4 in HC569 and HC800 cells by using 2 independent single guide RNA sequences targeting Six1 (sgRNAs A and B) and 4 targeting Six4 (sgRNAs E-H). Following lentiviral-mediated delivery of the Cas9 and sgRNAs into the cell cultures, cells were subjected to selection with puromycin (selection marker in the plasmid bearing sgRNAs against Six1) or hygromycin (resistance cassette co-expressed with sgRNAs targeting Six4). To evaluate the efficiency of CRISPR-directed Six4 and Six1 knockouts in the bulk cell population, immunoblotting assays were carried out to determine protein expression. sgRNAs G and H showed the greatest degree of SIX4 protein reduction and were used for subsequent functional studies (Figure S7A, left panel). Both sgRNAs designed to disrupt Six1 were highly efficient, as confirmed by the virtually undetectable SIX1 expression (Figure S7A, right panel).

To first assess the impact of functionally disabled SIX4 and SIX1, individual cells were isolated by fluorescence-activated cell sorting (FACS) to obtain multiple single cell clones of each murine PDAC cell line (Figure S7B). In most cases, we tested at least one clone derived from each individual sgRNA in proliferation assays. In vitro proliferation of Six4 and Six1 knockout clones was remarkably impaired in both murine cell lines, ranging from 1% to 50% of the level of proliferation seen in control NT cells (Figure 7A, left panel).

Figure 7. SIX4 and SIX1 are each required for optimal growth of HNF4α -negative PDAC.

A. Quantitation of proliferation in human PDX200 and BxPC3 cell lines and single cell cloned HC569 and HC800 murine cells stably expressing Cas9 and the indicated sgRNA against Six1 (Six1 KO) or Six4 (Six4 KO). Data show growth fold change after 7 days of culture (n= 3 biological replicates, 2 independent experiments). Graph represents mean +/− SEM as fold change compared to the corresponding non-target control (NT). ***p<0.001, **p<0.01, *p<0.05, Ordinary one-way ANOVA, Bonferroni’s post-hoc test.

B-C. Single cell clones of Six4 knockout HC800 (B) and HC569 (C) cell lines were injected subcutaneously into the flank of NSG mice. Tumor volume was measured three times weekly starting 1 week post implantation (n=6 mice per group). Flank tumor volume and weight, and mitosis per high power field are shown. Data represented as mean ± SEM. p≤ 0.05, Wilcoxon test (for tumor growth). *p< 0.05; **p< 0.01; ***p< 0.001 by Mann-Whitney test (for tumor weight and mitosis).

D. Individual clonal populations of Six1 deleted-HC569 cell line were injected subcutaneously into the flank of NSG mice. Tumor volume was measured biweekly starting 1 week post implantation (n=5 mice per group). Flank tumor volume and weight, and mitosis per high power field are shown. Data represented as mean ± SEM. *p< 0.05; **p< 0.01; ***p< 0.001 by Wilcoxon test (for tumor growth) or Mann-Whitney test (for tumor weight and mitosis). Representative H&E and IHC for SIX1 are shown for each cohort. Scale bar: 100 microns.

Additionally, we evaluated the effect of Six4 ablation in the proliferation of PDX220 and BxPC3 human PDAC cells. Three out of four independent sgRNAs targeting human SIX4 were efficient in markedly disrupting SIX4 expression in PDX220 cell line, as well as the two tested in BxPC3 (Figure S7C). We found that parental PDX220 and BxPC3 cells did not tolerate single cell cloning, and therefore measured the proliferation of the bulk population after stable sgRNA transduction. Consistent with murine PDAC cells, both human cell lines are dependent on SIX4 for optimal proliferation (Figure 7A, right panel).

To further explore the effect of Six4 loss in vivo, we implanted Six4-mutated single cell clones of HC800 and HC569 cells into the flank of NSG mice. Both HC800 SIX4-negative clones formed tumors that were significantly smaller (as assessed by volume and mass) and less proliferative than their control counterparts (Figure 7B). HC569 Six4-knockout clone (H#6) also displayed significantly impaired growth in vivo (Figure 7C).

Targeted disruption of Six1 also impaired in vivo growth of both HC569 and HC800 cells. Subcutaneous tumors derived from Six1-negative clones (2 independent clones/sgRNAs per cell line) displayed sizes, weights and proliferation rates significantly lower than their control counterparts (Figure 7D and S7D). IHC confirmed the absence of SIX1 expression in tumors arising from both Six1-deleted HC569 clones, whereas SIX1 was readily detectable in control tumors (Figure 7D, right panel).

Taken together, these results show that SIX4 and SIX1 play an oncogenic role in HNF4α -negative PDAC. Proliferation of all tested HNF4α -negative murine and human PDAC cell lines is dependent on SIX4 expression. Both murine lines are also highly dependent on SIX1 for in vitro proliferation and for in vivo tumor grow, consistent with prior reports in other human cell lines [30, 31].

Concomitant SIX4 and SIX1 ablation hinders growth and alters cell identity in HNF4α -negative PDAC cells.

Given the partially overlapping functions of SIX4 and SIX1 during development [32], we sought to determine whether HNF4α -negative PDAC cells can tolerate concomitant loss of both SIX4 and SIX1. HC569 and HC800 murine cell lines were simultaneously transduced with lentiviruses encoding Cas9 and sgRNA A (targeting Six1) in combination with either sgRNAs G or H (recognizing Six4). Following 1 week of combined puromycin/hygromycin selection, cells were then single cell sorted to obtain clonally isolated lines. For both cell lines, approximately half of the individually sorted cells were able expand over a period of 6 to 9 weeks post-sorting. Careful evaluation of multiple clones for each cell line showed that only a very small fraction of clones bears allelic mutations that simultaneously impair SIX1 and SIX4 expression (data not shown and Figure S8A). We tested all three HC569 and all six HC800 confirmed double knockout (DKO) clones in proliferation assays. Four HC800 DKO clones showed strong proliferative impairment, while one clone (AH#3) displayed a more modest but still significant decline in proliferation compared to control NT cells. Interestingly, one HC800 DKO clone (AH#15) was able to grow at similar rate to control cells, despite showing undetectable levels of both SIX1 and SIX4 (Figure 8A and S8A, left panels). Simultaneous mutation of Six4 and Six1 in HC569 clones also hindered cell proliferation, with clones AG#1 and AH#2 displaying the most pronounced defect and AG#8 showing a more intermediate phenotype (Figure 8A, right panel). To further explore the consequences of SIX4 and SIX1 depletion, 3 individual DKO clones of each cell line were used for in vivo growth experiments. After subcutaneous implantation, tumors derived from DKO clones of each HC800 and HC569 cell line revealed a growth deficiency compared to SIX1/4-proficient tumors. Thus, tumors arising from DKO cells displayed smaller sizes, lower weights and a significant reduction in proliferation compared control tumors (Figure 8 B–C).

Figure 8. Concomitant SIX4 and SIX1 ablation hinders growth and alters cell identity in HNF4α -negative PDAC cells.

A. Quantitation of proliferation in Six1-Six4 double knockout single cell clones of HC800 (left panel) and HC569 (right panel) cells stably expressing Cas9 and the indicated sgRNA combinations. Change in cell number was measured after 7 days of culture. Graph represents mean +/− SEM as fold change in growth compared to NT control (n= 3 biological replicates, 2 independent experiments). ***p<0.001, **p<0.01, *p<0.05, Ordinary one-way ANOVA, Bonferroni’s post-hoc test.

B-C. Single cell clones of Six1-Six4 double knockout HC800 (B) and HC569 (C) cell lines were injected subcutaneously into the flank of NSG mice. Tumor volume was measured three times weekly starting 1 week post implantation (n=6 mice per group). Flank tumor volume and weight, and mitosis per high power field are shown. Data represented as mean ± SEM. p≤ 0.05, Wilcoxon test (for tumor growth). *p< 0.05; **p< 0.01; ***p< 0.001 by Mann-Whitney test (for tumor weight and mitosis).

D. Murine HC569 and HC800 cell lines were treated for 72hs with the specified concentrations of GSK-J4 and benzbromarone. Cell viability was determined by measuring cell number using IncuCyte Live-Cell Analysis Systems. Synergy scores were generated utilizing Combenefit Software. Mapped to d-r surface plots based on Loewe synergy model are shown. Experiments were carried out three times with similar results.

E. Table displays the top enriched terms in the Mouse Gene Atlas gene-set library selected by significance (terms with adjusted p< 0.05 are shown).

F. Heatmap displays the relative expression of the indicated cell fate determinants in murine control (NT) versus Six1-Six4 DKO HNF4α -null HC800 cells. Data show expression levels as Log2 of the RNA-seq normalized counts for each gene in three independent NT control samples and one or two biological replicates of the double knockout clones as specified.

G. Model depicting the role of HNF4α and SIX4/1 in proliferation and lineage specification in PDAC cells.

Both SIX4 and SIX1 physically interact with enzymatically active co-factors that mediate their transcriptional activity during development and in various disease states. SIX4 recruits the lysine demethylase UTX to muscle-specific genes during myogenesis, facilitating their activation [33, 34]. SIX1 activity depends in part on the EYA family of protein tyrosine phosphatases [35]. We therefore asked whether inhibitors of these enzymatic co-factors could be used to block the proliferation of HNF4α -negative PDAC. HC569 and HC800 cells were treated with increasing concentrations of GSK-J4 (a small molecule inhibitor of H3K27me3/me2-demethylases UTX/KDM6A and JMJD3/ KDM6B) [36] and benzbromarone (an active site inhibitor of EYA proteins [37] that is used clinically to treat gout). We then measured EC50 for each drug and assessed synergy/antagonism (Loewe Additivity method) using Combenefit Software [38]. We found that GSK-J4 treatment caused cytotoxicity in both cell lines (HC569 EC₅₀= 0.87 M, HC800 EC₅₀= 1.2 M, Figure S8B). Although no changes in cell number were observed when HC569 and HC800 were exposed to benzbromarone, we found synergistic anti-proliferative effects when both inhibitors were combined, particularly in HC569 cells (Figure 8D and S8C).

To identify genes coordinately regulated by SIX4 and SIX1 in PDAC, we performed RNA-seq on HC800 DKO clones (n=8) and control cells. Transcriptome analysis revealed a total of 279 differentially expressed genes (142 downregulated and 137 upregulated genes in control vs DKO cells (Figure S8D and Supplemental table 8)). There was no significant difference in molecular subtype between these groups, although DKO cells exhibited a trend toward a slightly lower Basal-like score (Figure S8E). Nevertheless, other pathway analyses revealed changes in cell identity in DKO cells. We found that the most significantly enriched terms in control SIX4/1-positive cells corresponded to osteoblast differentiation when compared to the Enrichr Mouse Gene Atlas datasets [39] (Figure 8E, S8F and Supplemental table 9). Analysis using the Body Atlas tool (Illumina Correlation Engine) showed that DKO cells display statistically significant negative correlation with gene sets associated with mesoderm-derived tissues, including myometrium, ovary and adipose tissue (Figure S8G and Supplemental table 10). Manual inspection further revealed that DKO cells express lower levels of multiple genes associated with mesodermal, mesenchymal and neuronal differentiation programs (Figure 8F).

Taken together, these results demonstrate that SIX4 and SIX1 function as oncogenic lineage specifiers in HNF4α -negative PDAC. Moreover, our data suggest that SIX4 and SIX1 may represent novel dependencies that might be exploited therapeutically in this particularly aggressive subtype of PDAC.

Discussion

Although two major molecular subtypes of human PDAC have been characterized clinically, the molecular regulators of their distinct biologic properties and clinical behaviors have not been fully identified. Here we show that the nuclear receptor HNF4α is a critical, non-redundant regulator of growth and molecular subtype in PDAC. Using complementary in vitro and in vivo model systems, we show that HNF4α restrains tumor growth and enforces an epithelial differentiation program by at least two distinct mechanisms. First, HNF4α directly activates multiple genes associated with the Classical subtype. Moreover, HNF4α activates other transcription factors (such as Hnf1a) that likely further amplify the transcriptional network governed by HNF4α. Second, HNF4α directly represses the expression of Six1 and Six4, genes encoding homeodomain transcription factors that are highly expressed in the Basal-like subtype of PDAC and normally drive non-endodermal differentiation programs, including mesoderm and neural crest. Importantly, we find that HNF4α -negative PDAC cells are dependent on SIX4, and its close paralogue SIX1, for in vitro proliferation and in vivo tumorigenesis, suggesting that SIX4/1 activity may be a novel subtype-specific vulnerability in HNF4α -negative PDAC. Recent work in a novel panel of human PDAC cell lines has also implicated HNF4α in the regulation of PDAC subtype and metabolism [40].

HNF4α is likely one of several endodermal lineage specifiers that regulate the Classical molecular subtype of PDAC (reviewed in [41]). For example, GATA6, FoxA1 and FoxA2 have been reported to promote epithelial differentiation and block EMT in human PDAC cell lines [42, 43]. Moreover, GATA6 binds and activates FOXA1 and FOXA2 [42], and FoxA1/2 activate HNF4α and other endodermal lineage specifiers in multiple contexts [16, 44]. Based on these data, a potential hierarchy of endodermal lineage specifiers that regulate the Classical subtype of PDAC begins to emerge. However, these studies did not formally compare gene expression changes induced by manipulation of each transcription factor with the Classical and Basal-like transcriptome profiles. It will likely be necessary to modulate the expression of these transcriptional regulators within the same physiologically relevant model system(s) to gain a comprehensive understanding of their individual impact on specification of the Classical subtype as well as their hierarchical regulatory relationships, which may involve extensive cross-talk and feedback loops as in normal tissue [45].

The ability of HNF4α to restrain tumor growth and overall progression of pancreatic tumors in our model systems is consistent with the better overall survival seen in the Classical subtype. Other endodermal cell fate determinants have been reported to restrain pancreatic cancer progression. For example, Ptf1a inhibits pancreatic tumorigenesis by not only preventing but also reversing tumor initiation [46]. Nevertheless, the effect of endodermal lineage specifiers on tumor growth is to some degree context dependent. For example, an in vitro CRISPR screen identified HNF4α (as well as its target HNF1 ) as dependencies in two out of nine human PDAC cell lines tested [47]. In this regard, dichotomous effects on PDAC growth have been reported for HNF1 [48, 49, 50, 51], PDX1 [52], FoxA1/2 [53, 54, 55, 56] and GATA6 [42, 57, 58]. It seems likely that even though the most malignant subset of PDAC downregulates the endodermal differentiation program, loss of this program may only confer a selective advantage in specific situations or stages of tumor progression. Furthermore, a dichotomous role for lineage specifiers in regulating malignant potential appears to be an emerging principle in cancer biology. For example, the pulmonary lineage specifier NKX2-1 can function as an oncogene in EGFR-mutant lung adenocarcinoma but restrains the growth of KRAS-mutant lung adenocarcinoma [59]. Moreover, HNF4α itself is required for initiation and growth of NKX2-1-deficient lung adenocarcinoma, a subtype of lung cancer with a GI-like differentiation program similar to the Classical subtype of PDAC [16].

It is intriguing that some human PDAC lines (Panc1 and MIA PaCa-2) did not respond to exogenous HNF4α. Interestingly, these two lines failed to respond to exogenous expression of PTF1A in a recent study [46]. This suggests that they may be too poorly differentiated to respond to re-introduction of a single endodermal lineage specifier. Mechanistically, FoxA1 and FoxA2 are required for full HNF4α activity in normal tissue [60, 61], and these two cell lines have lower levels of FOXA1 transcripts than the other human cell lines tested in this study based on publicly available data (depmap.org). Additional studies are clearly needed to identify the full complement of proteins required for HNF4α activity in PDAC.

The transcriptional network governing the Basal-like subtype of PDAC is even less well understood than the Classical subtype. The transcription factors ΔNp63 [62, 63] and Gli2 [64] have been reported to promote growth and activate a Basal-like gene expression program in human cell lines. Our work identifies the homeodomain transcription factors SIX4 and SIX1 as HNF4α -repressed drivers of proliferation and growth in the Basal-like subtype. HNF4α binds directly to the promoters of Six4 and Six1 (as well as a potential distal regulatory element located between them). SIX1 and SIX4 are partially redundant drivers of development in multiple tissues derived from the mesoderm and ectoderm [29]. Several SIX family members have been implicated as having oncogenic functions in various types of cancer [65, 66], including SIX1 in PDAC [30, 31]. However, SIX4 has not been previously shown to play a functional role in PDAC, despite its elevated levels in the Basal-like subtype. Here, we show for the first time that SIX4 plays an oncogenic role in PDAC. Simultaneous mutation of Six1 and Six4 further demonstrates a role for these lineage specifiers in both the proliferation and cellular identity of HNF4α –deficient PDAC. Although studies on SIX4 and SIX1 were conducted both in vitro and in vivo (subcutaneous tumors), it will be important to also study these genes in additional model systems as well as in more physiologically relevant settings (including orthotopic and metastatic models) to determine the influence of specific microenvironments on their ability to drive PDAC growth.

During the derivation of Six1/4 double knockout clones, we found that only a very minor proportion of the analyzed clones were actually null for both genes, while the vast majority showed disruption of either Six1 or Six4, or neither (Figure S8A and data not shown). Thus, we cannot rule out the possibility that acute, concomitant Six4 and Six1 inhibition might cause a more severe phenotype than we observed in the clones we were able to establish. However, the fact that one double null clone proliferated as well as control cells may provide an opportunity to identify mechanisms by which HNF4α –deficient PDAC cells might overcome loss of SIX4/SIX1 activity.

Although direct inhibitors of SIX4 and SIX1 have not been described, both physically interact with co-factors that are potentially druggable targets (UTX [33] and the EYA family [67, 68], respectively). Our observation that GSK-J4 can inhibit PDAC cell lines may seem unexpected given that one of its targets (UTX/KDM6A) has been reported to be a tumor suppressor that enforces endodermal differentiation and prevents squamous/mesenchymal differentiation in PDAC [50, 69]. We speculate that the effect of inhibiting UTX will be critically dependent on the transcription factors to which it binds, and that these are likely to be very different in the PDAC cell of origin (in which UTX was shown to be a tumor suppressor) vs. established HNF4α -null PDAC cells. Other considerations include the ability of GSK-J4 to inhibit other lysine demethylases, and the observation that the tumor suppressor function of UTX was independent of its enzymatic activity [69]. Nevertheless, our studies demonstrating synergy between two inhibitors of these enzymes provide an initial proof of principle that justifies further investigation into the mechanisms of SIX4 and SIX1 activity in PDAC as well as the development of more specific inhibitors.

In summary, we show that HNF4α, SIX4 and SIX1 are lineage specifiers that coordinately regulate the growth and molecular subtype of PDAC (Figure 8G). We anticipate that this work will further the field’s goal of developing therapeutic strategies targeting specific subtypes of this disease. Although both subtypes are highly lethal and require new therapies, Basal-like PDAC is even more refractory to standard chemotherapy than the Classical subtype [6]. We speculate that SIX4 and SIX1 may represent novel targetable vulnerabilities in this highly aggressive PDAC variant.

Material and Methods:

Genetically Engineered Mouse Models

Mice harboring Kras^LSL-G12D [70], Kras^FSF-G12D [71], p53^flox [72], Rosa26^LSL-tdTomato [73], Hnf4α^flox [14], Rosa-FSF-Cre^ERT2 [74], p53^frt [75], and Pdx1-Cre [17] alleles have been previously described. All animals were maintained on a mixed C57BL/6J x 129SvJ background. Animal studies were approved by the IACUC of the University of Utah and the University of California at San Francisco, and conducted in compliance with the Animal Welfare Act Regulations and other federal statutes relating to animals and experiments involving animals and adheres to the principles set forth in the Guide for the Care and Use of Laboratory Animals, National Research Council (PHS assurance registration number A-3031–01).

Tamoxifen and Doxycycline Administration

Mice were fed ad libitum food pellets supplemented with tamoxifen (500 mg/ kg; Envigo, Indianopolis, IN) or doxycycline hyclate (625 mg/kg; Envigo, Indianopolis, IN) in place of standard chow starting 1 week before performing engraftments and for the entire duration of the experiments.

Flank Tumor Transplantations

For subcutaneous allo- and xenografts experiments, were injected with 2.5 x10⁵ murine PDAC cells, 3.5 x10⁵ human PDX220 cells or 5 x10⁵ murine PDAC cells from organoid cultures were resuspended in 50 μL of culture media, then mixed 1:1 with Matrigel (Corning). Cells were injected subcutaneously into the left or right flank NSG recipient mice. Subcutaneous tumor dimensions were measured with calipers twice weekly for the duration of the experiment, n = 8–10 independent tumors per group. Tumor volume was calculated using the standard modified ellipsoid formula ½ (Length x Width2) formula. After completion of the experiment, tumors were removed, weighed, and fixed for histological analysis.

In Vivo Bioluminescence Imaging

Bioluminescence imaging was performed using an IVIS Spectrum In Vivo Imaging System (PerkinElmer). Mice bearing tumors of cell expressing firefly luciferase were injected with 200 μl/mouse of D-Luciferin (GoldBio) at 16.7 mg/ml. Optimal signals were obtained 10 min after subcutaneous injections of the D-Luciferin. Scanning was done with mice placed in prone position.

Patient derived Xenografts

Ethics

Tumor acquisition and experimental usage was approved by the University of Utah IRB (89989 and 10924) and the animal studies were approved by the University of Utah Institutional Animal Care and Use Committee (17–06008).

Tissue collection

Pancreatic adenocarcinoma tissue was aseptically collected through the Huntsman Cancer Institute Biorepository and Molecular Biology Core at the time of surgical resection, placed in sterile cold RPMI media (Hyclone, Fisher Scientific), and implanted within one hour of resection.

Tumor implantation

Resected tumor specimens were cut into smaller segments and mixed with Matrigel. Mice (NOD/SCID; female; 22–26g) were anesthetized with isoflurane, skin shaved and disinfected with alcohol and betadine, and a small incision was made in each of the four back quadrants of the mice. Tumor fragments were inserted into the subcutaneous pocket and incision closed with suture (6–0 Vicryl; Ethicon). When total tumor burden was approximately 2 cm in diameter the tumors were harvested and a portion fixed in 10% formalin.

Histology and Immunohistochemistry

All tissues were fixed in 10% formalin overnight and when necessary, lungs were perfused with formalin via the trachea. Organoids were first fixed in 10% formalin overnight and then mounted in HistoGel (Thermo Scientific, HG-4000–012). Mounted organoids and tissues were transferred to 70% ethanol, embedded in paraffin, and four-micrometer sections were cut. Immunohistochemistry (IHC) was performed manually on Sequenza slide staining racks (ThermoFisher Scientific, Waltham, MA). Sections were treated with Bloxall (Vector labs) followed by Horse serum 536 (Vector Labs, Burlingame, CA) or Rodent Block M (Biocare Medical, Pacheco, CA), primary antibody, and HRP-polymer-conjugated secondary antibody (anti-Rabbit, Goat and Rat from Vector Labs; anti-Mouse from Biocare. The slides were developed with Impact DAB (Vector) and counterstained with hematoxylin. Slides were stained with antibodies to BrdU (BU1/75, Abcam, Cambridge, MA) 1:400, HA tag (C29F4, CST) 1:400, Galectin 4 (AF2128, R&D Systems) 1:200, HNF4α (C11F12, CST) 1:500, PK-LR (EPR11093P, Abcam) 1:500, SIX1 (D5S2S, CST) 1:100, HNF4α P1 and P2 (Human HNF-4 alpha/NR2A1 MAb (Clone K9218) PP-K9218–00, Human HNF-4 alpha/NR2A1 MAb (Clone H6939) PP-H6939–00) 1:100. Pictures were taken on a Nikon Eclipse Ni-U microscope with a DS-Ri2 camera and NIS-Elements software. Mitoses quantitation and histological analyses were performed on hematoxylin and eosin-stained or IHC-stained slides using NIS-Elements software. All histopathologic analysis was performed by a board-certified anatomic pathologist (E.L.S.).

Derivation of Murine 2D Cell Lines and 3D Organoid Cultures

Cell lines and 3D cultures were created by enzymatic and mechanical dissociation of pancreatic tumors or normal pancreatic tissue harvested from Kras^LSL-G12D; p53^flox/+; Hnf4α^flox/flox; Pdx1-Cre and Kras^FSF-G12D; p53^fr/frt; Hnf4α^F/F; RosaCre^ERT2 mice, respectively. Two dimensional cultures were grown in DMEM/High Glucose medium (Gibco) supplemented with 10%FBS. For organoid derivation, single cell suspensions were transduced with an adenovirus expressing Flp recombinase for 2 hs at 37ºC, washed, and seeded in grow factor-reduced Matrigel (Corning) and grown in 50% L-WRN conditioned media [76]. HNF4α expression restoration was induced by supplementing the corresponding media with 1ug/ml doxycycline (Sigma-Aldrich, D9891) and Hnf4α genetic loss was carried out by adding 4-OH-Tamoxifen (Cayman Chemical, 68392–35-8) to the culture media. All cell lines were tested periodically for mycoplasma contamination. To maintain cell cultures mycoplasma free, all culture media were supplemented with 2.5 ug/ml Plasmocin.

Human Cell Lines and Cell Culture

Human HEK-293T cell line and BxPC3, MIA PaCa-2, Panc1, and Panc10.05 PDAC cells were obtained from ATCC. BxPC3 and Panc10.05 cells were cultured in RPMI-1640 medium (Gibco) supplemented with 10% FBS. MIA PaCa-2 and Panc1 cells were cultured in DMEM/High Glucose medium (Gibco) supplemented with 10%FBS. PDX220 cell lines were cultured in Advanced DMEM/F-12 medium (Gibco) supplemented with 10% FBS. HNF4α expression restoration was induced by supplementing the corresponding media with 1ug/ml doxycycline (Sigma-Aldrich, D9891). All cell lines were tested periodically for mycoplasma contamination. To maintain cell cultures free of mycoplasma, all culture media were supplemented with 2.5 ug/ml Plasmocin. Cell line authentication was carried out through STR analysis at the University of Utah DNA Sequencing Core.

Cell proliferation assay

We used a live cell imaging system (IncuCyte) to directly measure cell proliferation while monitoring morphological changes over time. Number of cells was tracked by detecting the tdTomato present in HC569 and HC800 cell lines or by transducing cells to express eGFP fused to the histone H2B, which labels cell nuclei.

Drug Treatments

Benzbromarone (Tocris, CAS number 3562–84-3) and GSK-J4 (Tocris, CAS number 1373423–53-0) were reconstituted in DMSO to a final concentration of 100 mM and used within 1 month. A total of 2,500 HC569 cells/well or 5,500 HC800 cells/well were seeded in 96-well plates and cultured overnight prior to Incucyte monitoring. Immediately after first scan to determine cell number, cultures were treated with serial dilutions of benzbromarone and/or GSK-J4, or vehicle DMSO as control and cell number was monitored daily for 72 h. Experiments were replicated three independent times.

Lentiviral Production

Lentivirus were produced by transfection of HEK-293T cells with TransIT-293 (Mirus Bio). Packaging vectors Δ8.9 (gag/pol) and VSV-G were used for lentiviral production. Supernatant was collected at 48, 60 and 72 hrs post-transfection, centrifuged, and filtered using 0.45um filter units.

cDNA Synthesis and qRT-PCR

Quantitative RT–PCR was performed on Trizol-extracted RNA using the iScript Reverse Transcription Supermix (BIO-RAD). qPCR reactions were performed using SsoAdvanced Universal Probes Supermix (BIO-RAD) and a CFX384 Touch Real-Time PCR Detection System (BIO-RAD). The following Taqman probes were used: Hnf4α (Mm01247712_m1) and Pklr (Mm00443090_m1). Transcript levels were normalized to Ppia (Mm02342430_g1) and quantitated by the ΔΔCt method.

Immunoblotting

Cells were lysed on ice for 20 minutes in RIPA buffer (50mM Tris HCl pH 7.4, 150 mM NaCl, 0.1 % (w/v) SDS, 0.5% (w/v) sodium deoxycholate, 1% (v/v) Triton X-100) plus complete protease phosphatase inhibitor cocktail (A32961, ThermosFiher Scientific). Cells were pelleted for 10 minutes at 4 ºC and protein concentration was quantitated with the Coomassie (Bradford) Protein Assay (ThermoFisher Scientific). A total of 50 ug of protein lysate was used for overexpression experiments and detection was performed through infrared visualization using Odyssey CLx Imaging System (LI-COR Biosciences). For detection of endogenous protein expression, 100 ug of protein lysate/lane was loaded and enhanced chemiluminescence system was required for visualization. Lysates were separated on Tris-Glycine (TGX) precast gels (Biorad) and transferred to nitrocellulose membranes (ThermoFisher Scientific). Membranes were probed overnight with antibodies to HA tag (C29F4, CST, 1:5000), HNF4α (C11F12, CST, 1:2000), ß-Tubulin (DSHB, 1:2000 when using infrared detection, 1:15,000 when using chemiluminescence), Vinculin (Abcam, 1:20000 when using infrared detection, 1:60,000 when using chemiluminescence), Six1 (D5S2S, CST, 1:1000 when using infrared detection, 1:250 when using chemiluminescence), Six4 (D-5, Santa Cruz Biotechnology, 1:1000 when using infrared detection, 1:100 when using chemoluminescence), and the appropriate species conjugated secondary antibodies (Infrared detection: ThermoFisher Scientific, 1:20000; Chemiluminescence: Jackson Immuno Research, 1:50,000).

CRISPR-Cas9 Genome Editing

SIX1 and SIX4 knock-out cell lines were generated by CRISPR-Cas9. LentiCRISPRv2 vectors, which express the single guide RNA (sgRNA), Cas9 and a selection cassette, were obtained from Addgene (ID: 52961 and 98291).

Cells were transduced and then subjected to corresponding puromycin and/or hygromycin selection. Two different sgRNA sequences were used to target murine Six1, four were tested for mouse and human Six4, and an inert non-targeting (NT) guide was used as a control. Each sgRNA was designed to target the first exons of each gene and those displaying the highest scores and lower number of predicted off targets were chosen (http://crispr.mit.edu/). sgRNA sequences are provided in Supplemental Methods.

Individual cells were FACS sorted and allowed to expand for a period of 6–9 weeks, after which time clonal expansions of Six1, Six4 or double knockouts grew. Confirmation of efficient deletion was carried out by detecting protein loss by Immunoblot for endogenous SIX1 or SIX4 and exogenously over-expressed SIX4, as its endogenous levels were essentially undetectable due to lack of antibody sensitivity.

Statistics

Statistical analyses were carried out using Prism 8 version 8.3.1. p-values were calculated using Mann-Whitney test, Chi-square test, Log-rank test, Wilcoxon test, Welch’s test, t-test, Fisher’s exact test or Ordinary one-way ANOVA followed by Bonferroni’s post hoc test, as accordingly stated for each experiment. RNA-Seq statistics are described above.

Details of RNA-seq, ChIP-seq and ATAC-seq methods are provided in Supplemental Methods.

Patient and Public Involvement Statement

Patients were not directly involved in this study. Patient Derived Xenograft (PDX) PDAC models were derived at the University of Utah.

Supplementary Material

Supp11. Figure S1, related to figure 1.

A. IHC demonstrating PDAC with mixed HNF4α expression (left) and complete stochastic HNF4α loss (center) arising in Kras^LSL-G12D/+; p53^F/+; Pdx1-Cre; Hnf4α^+/+ mice (12 weeks of age). Right: HNF4α -positive PanIN from pancreas with HNF4α -negative PDAC. Scale bar: 100 microns.

B. IHC for Galectin_4 in murine PDAC arising in Kras^LSL-G12D/+; p53^F/+; Pdx1-Cre mice of the indicated Hnf4α genotype. “HNF4α ” indicates whether the tumor is positive or negative for this protein by IHC.

C. IHC for P1 and P2 isoforms of HNF4α in murine pancreatic neoplasia (Kras^LSL-G12D/+; p53^F/+; Pdx1-Cre; Hnf4α^+/+ mice (12 weeks of age)) and in human patient derived xenografts. Scale bar: 100 microns.

D. IHC for HNF4α on mPanIN from Kras^LSL-G12D/+; p53^+/+; Pdx1-Cre; Hnf4α^F/F mice and Hnf4α^+/+ controls at 9 months of age. Scale bar: 100 microns.

Supp12. Figure S2, related to figure 2.

A-B. Immunoblot analysis for indicated proteins on two independent murine cell lines (HC800 and HC569) from PDAC arising in Kras^LSL-G12D/+; p53^F/+; Pdx1-Cre; Hnf4α^F/F mice. Cells were transduced with lentivirus encoding doxycycline-inducible HA-tagged HNF4α 8 isoform and then selected with blasticidin. Cells were treated with doxycycline for 2, 4 or 7 days prior to lysis.

C. Bright field images of HC800 cell lines treated with doxycycline or vehicle for 72hs. HC800 cell morphology changes from a spindle to a classic epithelial cobblestone growth pattern with HNF4α restoration.

D. Immunoblot analysis for indicated proteins on five independent human PDAC cell lines. Cells were transduced with lentivirus encoding doxycycline-inducible HA-tagged HNF4α 8 isoform and then selected with blasticidin. Cells were treated with doxycycline as indicated prior to lysis.

E. H&E and IHC for HA on subcutaneous tumors from HC800 and HC569 cells in the presence or absence of doxycycline treatment. Scale bar: 100 microns.

F. IHC for ΔNp63 on control subcutaneous tumors from HC800 and HC569 cells. Scale bar: 100 microns.

G. IHC for HNF4α and ΔNp63 on serial sections of PDX220 cells subcutaneously injected in NSG mice. Scale bar: 100 microns.

H. H&E and IHC for HNF4α, HA and ΔNp63 on subcutaneous tumors from PDX220 cells in the presence or absence of doxycycline treatment. Scale bar: 100 microns.

Supp13. Figure S3, related to figure 3.

A. Representative IHC for HNF4α in PDAC organoid cultures derived from pancreata of Kras^FSF-G12D/+; p53^Frt/Frt; Rosa26^FSF-CreERT2; Hnf4α^F/F mice. Cells were treated with ethanol (control) or 4-hydroxytamoxifen (4-OHT) to activate Cre^ERT2 and delete Hnf4α. Scale bar: 100 microns.

B. qRT-PCR in PDAC organoids cultures for indicated transcripts. Cells were treated with ethanol (Control) or 4-hydroxytamoxifen (4-OHT) to activate Cre^ERT2 and delete Hnf4α.

C-F. H&E and IHC analysis of SC1693 organoids injected subcutaneously into NSG mice. Mice were fed control (C-D) or tamoxifen (E-F) chow starting 1 week prior to flank injection. Tumors were analyzed at 6 weeks post injection. Left images: scanning magnification, scale bar: 1 mm. Right: High power images, scale bar: 100 microns.

Supp14. Figure S4, related to figure 4.

A. Survival analysis of patients with HNF4α high vs low PDAC in PACA-AU (left) and TCGA (right) cohorts. Top row: comparison of upper half vs. lower half (split around median HNF4α expression). Bottom row: comparison of upper vs. lower quartile. Analysis was limited to high purity tumors (>30% cellularity).

B. Heatmaps showing relative expression of endodermal lineage markers as measured by RNA-Seq in PDAC models. Samples sorted by HNF4α expression, rows clustered by Pearson.

C. Representative tissues exhibiting positive and negative correlations with HNF4α expression in autochthonous tumors (Body Atlas function, Illumina Correlation Engine).

D. Heatmaps showing relative expression of genes identified by DESeq2 to be consistently altered along with Hnf4α modification. Samples sorted by t statistic.

Supp15. Figure S5, related to figure 5.

A. Venn diagram shows the overlap between HNF4α ChIP-seq peaks identified in HC569 and HC800 cells.

B. Motif enrichment in HNF4α binding sites. STAMP logos and enrichment statistics are shown for both HC569 and HC800 cells.

C. Principal component analysis of ATAC-seq peaks within control and dox-treated HC569 and HC800 cells. Each dot represents an individual sample.

D. Significantly enriched transcription factor motifs in regions with increased and decreased chromatin accessibility.

E. Boxplot shows distribution of gene expression level among ATAC-seq regions with different chromatin accessibility. The group “both” refers to genes with at least one regulatory region nearby displaying increased chromatin accessibility and at least one that displays decreased accessibility.

F. Pie charts displaying distribution of HNF4α bound genomic regions with differential chromatin accessibility (upper panel) and distribution of differentially accessible chromatin regions bound or not to HNF4α (lower panel).

G. Pie charts depict distribution of different genomic features in HNF4α bound genomic regions with differential chromatin accessibility.

H. Motif enrichment in HNF4α -bound sites in chromatin with different accessibility. STAMP logos and enrichment statistics are shown.

I. Graph displays relative HNF4α binding site strength in chromatin regions with different accessibility. Fisher’s exact test, p 2.2e^-16.

J. Contingency test showing number of up- and downregulated genes with HNF4α bound sites with increased or decreased chromatin accessibility. Fisher’s exact test, p 0.0001, odd ratio= 7.321.

Supp16. Figure S6, related to Figure 6.

A. Correlation between SIX4 and HNF4α, Classical score and Basal-like score in human PDAC (high purity cases, TCGA).

B. Correlation between SIX1 and HNF4α, Classical score and Basal-like score in human PDAC (high purity cases, TCGA).

C. Correlation between SIX4 and SIX1 in human PDAC (high purity PDAC, TCGA).

Supp17. Figure S7, related to Figure 7.

A. CRISPR/Cas9 validation. Immunoblot for SIX4 in bulk population of HC800 cells overexpressing doxycycline inducible SIX4 and HNF4α. The 4 indicated sgRNA sequences recognizing Six4 were tested (left panel). Immunoblot for SIX1 in HC569 and HC800 bulk populations expressing Cas9 and either of the 2 indicated sgRNA sequences targeting Six1 (right panel).

B. CRISPR/Cas9 validation of single cell clones of HC800 and HC569 cell lines. Representative immunoblot for SIX1 and SIX4 in the indicated Six1 or Six4 knockout clones. Dotted lines demarcate different lanes on the same blot. Solid lines demarcate lanes from different blots.

C. Representative immunoblot of human PDX220 and BxPC3 cells stably expressing Cas9 and the indicated sgRNA against SIX4 or non-target control.

D. Single cell clones of Six1 knockout HC800 cell lines were injected subcutaneously into the flank of NSG mice. Tumor volume was measured three times weekly starting 1 week post implantation (n=6 mice per group). Flank tumor volume, weight and mitosis are shown. Data represented as mean ± SEM. p≤ 0.05, Wilcoxon test (for tumor growth). *p< 0.05; **p< 0.01; ***p< 0.001 by Mann-Whitney test (for tumor weight and mitosis).

Supp18. Figure S8, related to Figure 8.

A. CRISPR/Cas9 validation of single cell clones of HC800 and HC569 cells. Representative immunoblot for SIX1 and SIX4 in cells expressing Cas9 and the indicated combination of sgRNAs targeting both Six1 and Six4.

B. Dose-response curves of murine HC569 and HC800 PDAC cell lines to GSK-J4. Cell survival was determined utilizing IncuCyte Live-Cell Analysis Systems. The EC50 for each cell line is presented from dose-response curves using Combenefit Software.

C. Synergy was analyzed using Combenefit Software. Results show the Matrix synergy plots displaying the synergy/antagonism score for each drug combination (Loewe model). Statistical significance denoted as ***p<0.001, **p<0.01, *p<0.05.

D. Heatmap displaying gene expression levels in murine HNF4α -negative NT or double knockout HC800 cells. Three independent NT control samples and one or two independent replicates of each double knockout samples were subjected to RNA-seq as depicted.

E. Violin plots depicting basal-like and classical gene scores in NT controls and DKO clones (upper panel). Scores were calculated by taking the mean expression of the log2 expression of all genes in the subtype. Changes in subtype scores were found to be non-significant by Wilcoxon test (p>0.05). Gene Set Enrichment Analysis (GSEA) barcode plots (Gene rank, lower panel) showing gene-level changes within Basal-like and Classical signatures. GSEA was performed on collapsed NT vs DKO for overall trends in subtype variation.

F. Clustergram displays the top enriched terms in the Mouse Gene Atlas gene-set library selected based on combined scores. Columns= enriched terms. Rows= input genes (upregulated genes in each indicated mouse tissue). Red cells= indicate if a differentially expressed gene from the NT vs DKO RNA-seq data set is associated with a term. MEF= Mouse Embryonic Fibroblasts.

G. Body Atlas tool, Illumina Correlation Engine was used to query the HC800 DKO vs NT control dataset. Graph depicts the top 10 more strongly negatively correlated tissue datasets, which all correspond to non-endodermally derived tissues.

Summary Box.

What is already known about this subject?

Pancreatic ductal adenocarcinoma (PDAC) is a highly lethal disease.

PDAC has been classified into two major subtypes based on transcriptional profiling: Classical and Basal-like.

The Classical subtype confers a better prognosis and expresses regulators of endodermal differentiation such as HNF4α.

The Basal-like subtype confers a worse prognosis and expresses regulators of non-endodermal differentiation such as SIX4 and SIX1.

What are the new findings?

The mechanisms underlying PDAC subtype specification have not been fully characterized. Here we show that HNF4α, SIX4 and SIX1 are key lineage specifiers that coordinately regulate growth and molecular subtype of pancreatic ductal adenocarcinoma (PDAC).

HNF4α restrains PDAC progression, activates expression of genes associated with the Classical subtype, and directly represses expression of SIX4 and SIX1.

SIX4 and SIX1 drive proliferation and promote mesodermal/neuronal differentiation in HNF4α -negative PDAC.

How might it impact on clinical practice in the foreseeable future?

Characterizing molecular regulators of PDAC subtype is essential for identifying subtype-specific vulnerabilities and therapeutic strategies

HNF4α is a member of the nuclear receptor superfamily, a class of druggable transcription factors.

SIX4 and SIX1 both physically interact with druggable co-factors required for their activity.

Abbreviations:

ADM

Acinar to ductal metaplasia

ATAC

Assay for Transposase-Accessible Chromatin

ChIP

Chromatin immunoprecipitation

CRISPR

Clustered Regularly Interspaced Short Palindromic Repeats

DKO

Double knockout

EMT

Epithelial to mesenchymal transition

FACS

Fluorescence-activated cell sorting

GI

Gastrointestinal

HNF4α

Hepatocyte Nuclear Factor 4 Alpha

IHC

Immunohistochemistry

KRAS

Kirsten rat sarcoma 2 viral oncogene homolog

mPanIN

Murine pancreatic intraepithelial neoplasia

NSG

NOD/SCID-gamma chain deficient

PanIN

Pancreatic intraepithelial neoplasia

PDAC

Pancreatic ductal adenocarcinoma

PDX

Patient derived xenograft

sgRNA

single guide RNA

Data availability

Gene expression, ChIP-Seq and ATAC-seq data are accessible in the NCBI Gene Expression Omnibus (GEO) database under accession number GSE138145, GSE144750 and GSE138463.