HNF4A and GATA6 Loss Reveals Therapeutically Actionable Subtypes in Pancreatic Cancer

Abstract

Pancreatic ductal adenocarcinoma (PDAC) can be divided into transcriptomic subtypes with two broad lineages referred to as classical (pancreatic) and squamous. We find that these two subtypes are driven by distinct metabolic phenotypes. Loss of genes that drive endodermal lineage specification, HNF4A and GATA6, switch metabolic profiles from classical (pancreatic) to predominantly squamous, with glycogen synthase kinase 3 beta (GSK3β) a key regulator of glycolysis. Pharmacological inhibition of GSK3β results in selective sensitivity in the squamous subtype; however, a subset of these squamous patient-derived cell lines (PDCLs) acquires rapid drug tolerance. Using chromatin accessibility maps, we demonstrate that the squamous subtype can be further classified using chromatin accessibility to predict responsiveness and tolerance to GSK3β inhibitors. Our findings demonstrate that distinct patterns of chromatin accessibility can be used to identify patient subgroups that are indistinguishable by gene expression profiles, highlighting the utility of chromatin-based biomarkers for patient selection in the treatment of PDAC.

Keywords: GATA6; GSK3B; HNF4A; PDAC subtypes; chromatin landscapes; intronic and distal promoters; metabolic targeting; therapeutic tolerance.

Figure 1.. Metabolic Differences Between Squamous and Classical (Pancreatic) PDCLs

(A) Heatmap of pathways and molecular processes involved in cancer metabolism showing enrichment of transcripts in pathways important in mTOR signaling and glycolysis in the squamous subtype. PDCLs were ranked from most classical (pancreatic) (orange) to most squamous (blue), using gene expression or pathway activity, and grouped into metabolic processes. PDCL ID is listed below the heatmap. (B) The same signature from (A) was applied to the RNA-seq cohort of bulk tumor from Bailey et al. (2016). Subtype classification is depicted by annotated colors on the top row. The immunogenic subtype has a transcriptional signature associated with immune infiltrate and shares transcriptional networks associated the classical (pancreatic) subtype (Bailey et al., 2016). ADEX, aberrantly differentiated endocrine eXocrine subtype defined by transcriptional networks important for pancreatic differentiation. (C) Heatmaps of key genes involved in glycolysis-gluconeogenesis and triglyceride biosynthesis. Genes are ranked by most differentially expressed between classical (pancreatic) (orange) and squamous (blue) subtypes of PDCLs, with color saturation proportional to degree of either classical or squamous enrichment, which is compared with and on the whole recapitulated in bulk tumors. (D) Relative lactate release and glucose consumption from squamous (TKCC-10 and TKCC-26) and classical (pancreatic) (TKCC-22, Mayo 5289, and Mayo 4636) PDCLs were determined by colorimetric analysis. Raw values were normalized to cell counts. (E) Glycolysis activity profile of squamous and classical (pancreatic) PDCLs using Agilent Seahorse XF Glycolysis Stress Test. (F) Agilent Seahorse XF Cell Mito Stress Test profiles of squamous and classical PDCLs. (G) Left: ECAR values for cells treated as in (E) corrected for non-glycolytic acidification. Right: OCR values for cells treated as in (F) corrected for oxygen consumption resultant from processes other than mitochondrial respiration. Boxplots are annotated using one-way ANOVA. Error bars represent mean ± SD. Independent experiments are shown, n = >6. ***p ≤ 0.001, ****p ≤ 0.0001. (H) Left: untargeted metabolomic analysis of indicated PDCLs. Right: metabolite pathway enrichment analysis of significantly altered metabolites between classical and squamous PDCLs. See also Figures S1 and S2 and Tables S1, S2, S3, and S4.

Figure 2.. Subtype-Specific Differences in Endodermal TF Expression

(A) Heatmap showing differential expression of regulatory genes central to pancreatic endodermal cell fate determination. Note loss of pancreatic transcripts HNF4A and GATA6 in the squamous subtype indicated by RNA-seq analysis. (B) Immunoblots of endodermal cell fate determining transcription factors across a selection of PDCLs representative of both classical (pancreatic) and squamous subtypes. 20 μg of the same protein lysate was probed with stated antibodies on different blots. Actin panel is a representative loading control (HNF1A loading shown). (C) Plots showing regulation of gene expression by methylation. Methylation of HNF4A (left) or GATA6 (right) is associated with the concordant downregulation of the indicated gene expression. Pearson correlation and adjusted p values are provided for each gene methylation comparison. Boxplot colors designate class: squamous (blue) and classical (pancreatic) orange. (D) Schematic representation of where the selected classical (pancreatic) PDCLs rank in terms of subtype. Expression of genes involved in endodermal cell fate was used to rank subtype. See also Figure S2 and Table S1.

Figure 3.. HNF4A Loss in Classical (Pancreatic) PDCLs Drives a Switch toward a Squamous-Associated Metabolic Profile

(A) Venn diagram showing the number of common and unique genes differentially expressed (p ≥ 0.05, fold change ≥ 2) after either HNF4A or GATA6 knockdown in the classical (pancreatic) Mayo 5289 PDCL. (B) ECAR in classical (pancreatic) PDCLs following siRNA-mediated knockdown of HNF4A. Boxplots are annotated using one-way ANOVA, mean ± SD. Technical replicates are shown, n ≥ 6. For all graphs: **p ≤ 0.01. (C) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis of significantly altered pathways identified after HNF4A knockdown in Mayo 5289 PDCL. Adjusted p value for each annotation is represented by color scale. Gene ratio is represented by dot size. Enriched terms and pathways were identified as significant at an adjusted p value ≤ 0.05 and FDR ≤ 0.05. (D) Comparison of molecular pathways identified in bulk tumor and PDCLs RNA-seq analysis with significant gene changes following HNF4A knockdown. (E) Right: Mayo 5289 PDCLs treated with two independent HNF4A siRNA oligos for 72 h were immunoblotted with indicated antibodies. Left: transient or stable HNF4A knockdown in PacaDD137 and Mayo 4636 PDCLs, respectively. Actin panel is a representative loading control (HNF4A loading shown). (F) Stable HNF4A knockdown in Mayo 5289 PDCL immunoblotted with PI3K signaling proteins identified from RNA-seq analysis. Actin panel is a representative loading control (HNF4A loading shown). (G) A selection of PDCLs ranked from classical (pancreatic) to squamous immunoblotted with indicated antibodies. Actin panel is a representative loading control (DEPTOR loading shown). For all blots in (E)–(G), 20 mg of the same protein lysate was probed with stated antibodies on different blots. (H) Correlation graph demonstrating a negative correlation of HNF4A expression with glycolysis pathway expression from bulk tumor samples described by Bailey et al. (2016) (left) and in PDCLs (right). See also Figure S3 and Table S5.

Figure 4.. A Subset of Squamous PDCLs Acquires GSK3β Drug Tolerance after Chronic Suppression of Glycolysis

(A and B) Schematic of experimental setup (A) and dose-response curves (mean ± SD) (B) for classical (pancreatic) and squamous PDCLs treated with TDZD-8 (GSK3βi) or tideglusib (GSK3βi) for 72 h. Independent experiments are shown, n ≥ 3. DMSO-treated cells were set to 100%. (C and D) Experimental setup (C) and (top) representative Glyco Stress Test curves for (D) classical (pancreatic) or (bottom) squamous PDCLs. (E) ECAR values (mean ± SD) after treatment with TDZD-8 or tideglusib for 4 h in classical (progenitor) and squamous PDCLs. Technical replicates are shown, n ≥ 5. (F and G) Schematic of experimental setup (F) and comparison of IC₅₀ values (mean ± SD) (G) after either 72- or 144-h treatment with either TDZD-8 (GSK3βi) or tideglusib (GSK3βi) in PDCLs. Unpaired t test. Independent experiments are shown, n = 3. (H and I) Schematic of experimental setup (H) and ECAR values (I) after 144-h treatment with either TDZD-8 (GSK3βi) or tideglusib (GSK3βi) in GSK3βi-tolerant squamous (TKCC-15, TKCC06, TKCC-18, and TKCC-26) PDCLs. Technical replicates are shown, n = 8. For all graphs: *p < 0.05; **p % 0.01; ***p % 0.001; ****p < 0.0001. Figure legend colors designate class: classical (pancreatic) = orange/brown; squamous = blue. See also Figures S4 and S5 and Table S6.

Figure 5.. PDAC PDCLs Express WNT Ligands

(A) Heatmap showing mRNA expression of indicated WNT ligands in PDAC subtypes determined by RNA-seq analysis. (B) Left: boxplots showing a significant association of WNT7A, WNT7B, and WNT10A expression in the squamous subtype from RNA-seq analysis of bulk tumor samples from Bailey et al. (2016). Kruskal-Wallis test. Right: boxplots showing WNT expression in the PDCLs. Wilcoxon test. (c) Kaplan-Meier plots showing overall survival based on data reported by Bailey et al. (2016). Tumor samples were stratified based on WNT7A (left), WNT7B (center), or WNT10A (right) expression. Blue shading represents patients with low WNT7A, WNT7B, or WNT10A expression, respectively. Yellow shading represents patients with high WNT7A, WNT7B, or WNT10A expression, respectively. Log rank p value. (D) Western blot for indicated targets in squamous PDCLs TKCC-26 and TKCC-18 after 24 h GSK3βi (tideglusib or TDZD-8) ± PORCN (LGK-974). GSK3α/β (CHIR99021) was used as a positive control. See also Figure S6.

Figure 6.. ATAC-Seq and Transcriptomic Analysis Revealed a Uniquely Accessible WNT Gene Program in Squamous PDCLs that Are Tolerant to GSK3B Inhibition

(A) Western blot (WB) for either HNF4A or GATA6 in representative PDCLs of the classical (pancreatic) or squamous subtype. 20 μg of the same protein lysate was probed with stated antibodies on different blots. Actin panel is a representative loading control (HNF4A loading shown). (Above) Oncoplot showing somatic mutations in genes involved in chromatin regulation. Green = structural variant (SV); purple = single-nucleotide variant (SNV) or indel. (B) Venn diagram showing the number of common and unique annotated gene peaks in PDCLs grouped by response to GSK3βi. GSK3βi resistant = PacaDD137, TKCC-22, Mayo 5289; GSK3βi tolerant = TKCC-26, TKCC-06, TKCC-15, TKCC-18; GSK3βi sensitive = TKCC-09, TKCC-10, TKCC-2.1. (C) ATAC-seq density plots of accessible genes in 10 PDCLs representative of the classical (pancreatic) or squamous subtypes. (D) ATAC-seq genomic tracks for WNT7A. Highlighted regions show subtype-specific genomic peaks. PDCLs are grouped based on response to GSK3β inhibitor. (E) Chart showing the genomic distribution of ATAC-seq peaks in squamous PDCLs that are sub-grouped based on response to GSK3βi. (F) KEGG pathway enrichment analysis of enriched pathways accessible in GSK3β-tolerant squamous PDCLs found at intronic and distal promoter sites. (G) WNT7A expression in squamous PDCLs treated with GSK3b (TDZD-8) for 144 h. For all graphs: **p < 0.01; ***p < 0.0001. See also Figures S6 and S7 and Table S7.

Figure 7.. Porcupine Inhibition Overcomes WNT-Driven Acquired Resistance to GSK3β Inhibition

(A) RNAscope hybridization for Wnt7a in PDAC GEMM KPC, KC Ptenfl/+, and KPC + LGK974 (porcupine inhibitor). Nuclear counterstaining is with hematoxylin. The scale bar represents 100 μm. (B) Quantification of samples described in (A) using HALO software. (C) Correlation graph demonstrating a positive correlation of PI3K-AKT activation with WNT signaling in the squamous subtype from bulk tumor samples described by Bailey et al. (2016). (D) Schematic of experimental setup. (E) Indicated PDCLs treated with either GSK3βi (tideglusib or TDZD-8), AMPKi/ULKi (SBI), or Porcupine-I (LGK974) alone or in combination for 144 h before cell number analysis. (F) GSK3β-sensitive squamous PDCLs (TKCC-10, TKCC-2.1, and TKCC-09) were treated with GSK3β(TDZD-8) for 144 h. Note that these cells remain sensitive to GSK3β(TDZD-8)-targeted therapy. **p ≤ 0.01; ***p ≤ 0.001. See also Table S7.