A STAT3/integrin axis accelerates pancreatic cancer initiation and progression

Abstract

The signal transducer and activator of transcription 3 (STAT3) pathway drives pancreatic ductal adenocarcinoma (PDAC) progression by coordinating cellular responses to stress and inflammation. We perform ChIP-seq on hypoxia- or oncostatin-M-treated PDAC cells to identify sites at which phospho-STAT3 binds to regulate the expression of genes linked to poor survival. A top hit among these is ITGB3, which we show promotes PDAC initiation and progression. Single-cell transcriptomics reveal that ITGB3 expression is enriched in PDAC cells experiencing oxidative stress due to chemotherapy. Moreover, high ITGB3 expression positively correlates with STAT3 signaling, hypoxia, and the basal subtype. Mechanistically, chromatin accessibility at ITGB3 enhancers controls STAT3's ability to induce ITGB3 expression, illuminating a plastic regulatory mechanism modulating STAT3 activity. Leveraging this insight, we identify additional STAT3 target genes regulated similarly to ITGB3 to establish an 18-gene signature involved in adaptive responses and able to stratify survival outcomes. Collectively, these findings highlight a novel opportunity to stratify PDAC subpopulations for STAT3-targeted therapies.

Keywords: CP: Cancer; STAT3; cellular stress; cytokine; enhancer; gene signature; hypoxia; inflammation; integrin; pancreatic cancer; tumor initiation.

Figure 1.. Integrin β3 is a STAT3-regulated stress/inflammatory response gene

(A) Schematic (left) depicts parallel ChIP-seq experiments to identify genomic regions to which STAT3-pY705 binds in response to the inflammatory cytokine OSM (10 ng/mL) and hypoxia (1% O₂) in FG pancreatic cancer cells. Graph (right) shows 106 genes associated with peaks common to both hypoxia and OSM. Hazard ratio (HR); previously unknown STAT3 target (purple); previously known STAT3 target (gray). (B) KM plot shows the probability of relapse-free survival (RFS) in patients with PDAC with high vs. low ITGB3 expression. (C) ChIP-seq peaks for pSTAT3 at enhancer regions upstream (GH17J047239) and downstream (GH17J047305) of the ITGB3 transcription start site (TSS) on chromosome 17 (ch17) in response to OSM (red) or hypoxia (teal) vs. control (black). Panels (right) show enlargements of the two enhancers. (D) ChIP-PCR for pSTAT3 and H3K27ac at upstream and downstream enhancers for ITGB3. FG and SUIT2 cells were untreated or stimulated with OSM (10 ng/mL) for 72 h. n = 3 independent experiments per cell line per condition. Graph displays the mean ± SEM. p values reflect Student’s t test vs. untreated. (E) Representative immunoblots for FG and SUIT2 cells treated with vehicle, IL-6-sIL-6R, LIF, or OSM for 72 h. Results represent three independent experiments. (F) Flow cytometry for integrin αvβ3 expression in FG and SUIT2 cells treated with vehicle control/normoxia, IL-6-sIL-6R, LIF, OSM, or hypoxia for 72 h. Graph displays the mean ± SEM for the percent of αvβ3+ cells per group for n = 3 independent experiments. ns, not significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 by Student’s t test. See also Figure S1 and Table S1.

Figure 2.. ITGB3 is enriched following chemotherapy and linked to a mesenchymal phenotype

(A) Schematic (panel 1) and UMAP plots of scRNA-seq of n = 6 naive (gray) and n = 6 chemotherapy-treated (red) patient tumors (panel 2). Cells were annotated by cell type (panel 3) to subset the epithelial cancer cells. The top 10% ITGB3-High-expressing (red) and bottom 10% ITGB3-Low-expressing (black) expressing cells (panel 4) were assessed for transcriptional differences. (B) DEGs between ITGB3-High and ITGB3-Low tumor cells (left). Significance score represents negative log10 probability of a gene not being differentially expressed. Enrichment score represents log fold change. Labeled genes are STAT3-regulated genes derived from the ChIP sequencing in Figure 1. Violin plot (right) shows the distribution of STAT3 gene set scores for ITGB3-Low and ITGB3-High tumor cells. p value represents Mann-Whitney U statistic. (C) Plot of the distribution of ITGB3-High and ITGB3-Low cells between naive and treated patients. Fold change (FC) represents treated vs. naive. (D) Gene set enrichment analysis of DEGs between ITGB3-High and ITGB3-Low tumor cells. (E) DEGs between ITGB3-High and ITGB3-Low tumor cells (left). Genes represent classical, quasi-mesenchymal, and basal gene signatures. Violin plots (right) show the distribution of classical, quasi-mesenchymal, and basal gene set scores for ITGB3-Low and ITGB3-High tumor cells. p value represents Mann-Whitney U statistic.

Figure 3.. STAT3 is necessary for ITGB3 expression, which promotes STAT3-mediated tumor initiation/progression

(A) FG cells treated with hypoxia, OSM, or respective controls (72 h). Immunoblots show STAT3 (Y705) and integrin β3 for cells with CRISPR KO of STAT3 or pharmacological inhibition of STAT3 upstream regulators using inhibitors. (B) Schematic of KTC vs. KTC-Stat3KO mice. Immunofluorescence staining shows DAPI (blue), integrin β3 (green), and keratin 19 (Krt19, orange) for the representative pancreatic tumor tissue. Scale bar represents 50 μm. Graph (top) depicts the mean ± SEM for the ratio of β3 to Krt19 staining for n = 3 tumors per group. Graph (bottom) depicts the degree of differentiation in the pancreas, quantified as the mean ± SEM for the ratio of carboxypeptidase A1 (Cpa1) to Krt19 for n = 3 tumors per group. p values represent Student’s t test. Additional images and Cpa1 staining are shown in Figure S2D. (C) In vivo limiting dilution experiments were performed to assay the tumor-initiating capacities of control FG pancreatic cancer cells (STAT3WT + EV) relative to FG CRISPR STAT3KO with empty vector (STAT3KO + EV) or with ectopic expression of integrin β3 (STAT3KO+β3). Immunocompromised nu/nu mice were injected subcutaneously with varying cell densities. Graphs depict tumor take vs. time for n = 12–20 tumors per group. See also Figure S2.

Figure 4.. Ability to gain integrin β3 enhances tumor initiation

(A) Limiting dilution tumor initiation assay of control FG pancreatic cancer cells (FG + EV) vs. FG CRISPR β3KO cells with empty vector (FG β3KO + EV) or ectopic β3 (FG β3KO+β3). Immunocompromised nu/nu mice were injected subcutaneously with 1 × 10⁶, 2.5 × 10⁵, or 1 × 10⁵ cells per injection. Graphs depict tumor take vs. time for n = 12–16 tumors per group. (B) Schematic depicts 6-week-old tamoxifen-inducible KPC (i-KPC; Kras^LSL−G12D/Tp53^flox/flox/Pdx1-Cre^ER) mice crossed with Itgb3^flox/flox mice to establish tamoxifen-inducible i-KPC-β3KO (Kras^LSL−G12D/Tp53^flox/flox/Itgb3^flox/flox/Pdx1-Cre^ER) mice. (C) Probability of survival vs. time following tamoxifen injection for n = 14 mice per group. The HR and p value reflect KM analysis. (D) Serial sections of early-stage (60-day) pancreatic tissues (n = 5 i-KPC, n = 6 i-KPC-β3KO) were stained using hematoxylin and eosin (H&E) and quantified for the total tumor area (left graph; mean ± SEM). Late-stage tumors (n = 14 i-KPC, n = 14 i-KPC-β3KO) were measured by tumor weight (right graph; mean ± SEM). Tumor area is outlined in red. Representative images for each group are shown. Scale bars represent 500 μm (top) and 50 μm (bottom). (E) Early-stage and late-stage tumors were stained for integrin β3. Positive staining is graphed as the mean ± SEM percent of the total tumor area on whole-tissue sections. p values represent one-way ANOVA with Tukey’s multiple comparisons. Scale bars represent 500 μm (top) and 50 μm (bottom). See also Figure S3 and Table S2.

Figure 5.. ITGB3 expression promotes progression from a classical to basal subtype in advanced tumors

(A) Representative immunohistochemistry demonstrating co-localization of β3 (purple) with Krt19 (brown) in early-stage i-KPC tumors. Scale bars represent 25 μm. (B) Representative Krt19 immunostaining of end-stage tumors. Scale bars represent 500 μm (top) and 50 μm (bottom). Krt19 expression classified by blinded observers into poorly, moderately, and well-differentiated categories. Graph depicts the mean ± SEM for the percent tissue area for each category of differentiation for 10 i-KPC tumors and 13 i-KPC-β3KO tumors. p values represent two-way ANOVA with Šidák’s multiple comparisons. (C) RNA quantified using RT-qPCR on flash-frozen late-stage tumors (n = 6 i-KPC, n = 6 i-KPC-β3KO). Heatmap (left), upregulation (red), and downregulation (blue) depicting classical, quasi-mesenchymal, and basal gene sets plotted as the Z score (per gene) of log2 FC. Graph (right) depicts summarized gene set score for subtypes calculated as the mean of the Z scores for genes in each subtype. ns, not significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 by Student’s t test (heatmap; left) and unpaired t test with single pooled variance (graph; right).

Figure 6.. Chromatin accessibility for STAT3 binding dictates the ability to express integrin β3

(A) PDAC lines were treated with vehicle or 10 ng/mL OSM for either 30 min or 48 h. Immunoblots for integrin β3 and STAT3-pY705. (B) Depiction of ATAC-seq peaks at the GH17J047239 and GH17J047305 enhancers for PDAC lines. Inducible cells, green; non-inducible cells, blue. p value indicates significance following multiple testing correction (adjusted p) of the log2 fold difference for the mean peak area of reads in each enhancer region computed for n = 3 cell lines per group (inducible vs. non-inducible). (C) Model whereby chromatin accessibility at ITGB3 enhancers controls STAT3 binding to induce β3 expression. Vorinostat promotes chromatin opening to enable STAT3 binding. (D) Inducible vs. non-inducible cells were untreated or treated with OSM (10 ng/mL) and vorinostat (5 μM) alone or in combination and then immunoblotted for β3, STAT3-pY705, total STAT3, H3K9ac, and β-actin. Representative of n = 3 independent experiments per cell line per condition. (E) A non-inducible cell (ASPC1) was untreated or treated with hypoxia (1% oxygen) and vorinostat (5 μM) alone or in combination (left). Alternatively, cells were untreated or treated with OSM (10 ng/mL) and vorinostat (5 μM) alone or in combination (right). pSTAT3 binding at enhancers for ITGB3 was measured using ChIP-PCR, plotted as the mean ± SEM log2 FC relative to untreated for n = 3 independent experiments per condition. ns, not significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 by ordinary one-way ANOVA with Tukey’s multiple comparisons. (F) Volcano plot illustrates DEGs between five inducible and five non-inducible lines. Significant DEGs (yellow) have log2 FC > 1 or < −1 and adjusted p value < 0.05. Genes comprising the β3-inducible signature (green) are defined as upregulated with normalized gene count standard deviations of <2. (G) CCLE PDAC lines were assigned β3-inducible signature scores and converted to Z scores and ranked. Dark green and dark blue bars represent cell lines with known inducible and non-inducible (respectively) phenotypes. Light green and light blue bars represent cell lines predicted to have inducible or non-inducible phenotypes, respectively. See also Figures S4 and S5.

Figure 7.. Subset of STAT3 targets (STRESS gene set) reflects the critical role of STAT3 during tumor initiation and progression of PDAC

(A) Schematic summarizes steps used to generate the 18-gene STRESS signature. (B) Graph shows the number of genomic read pile-ups for STAT3-pY705 binding at enhancers. (C) Correlation matrix depicts the co-expression Pearson correlation coefficient (according to the color map) and corresponding p value (size of each circle) for the 18 genes included in the STRESS signature for patients in The Cancer Genome Atlas (TCGA) PAAD dataset. Genes indicated in purple are those that are likely driven by STAT3 and whose high expressions are negatively correlated with survival as a consequence. Genes in gold are those that are likely repressed following STAT3 binding to enhancers and whose high expressions are positively correlated with survival. (D) Upper panel shows unbiased clustering of the 18-genes STRESS signature for TCGA-PAAD dataset. Lower panel depicts value of the computed STRESS signature score. (E) KM plot of RFS in patients with PDAC according to the STRESS or STRESS-Up signatures compared with Collisson and Moffitt signatures. See also Figures S6 and S7 and Table S3.