Clonally-derived KRAS^G12D p53KO mouse embryonic fibroblasts (KP MEFs) and pancreatic KRAS^G12D p53^R172H (KPC) cells were previously described (16, 17). Human Hep3B and HEK293T cells were obtained from ATCC. All cells were grown in DMEM media supplemented with 5% FBS (Atlanta Biologicals) and 1x antibiotic/antimycotic (Gibco). For standard proliferation assays, cells were seeded into 6-well plates at a concentration of 4×10⁵ cells per well and counted cumulatively with a Coulter counter (Beckman) every three days for two weeks. Focus formation assays were performed as described (16, 18). Briefly, 10³ KP MEFs were co-cultured with 10⁵ p53KO feeder MEFs in 6 cm dishes. After two weeks, transformation efficiency was evaluated by manually counting macroscopic colonies. Transformed foci were stained with Giemsa (Sigma) for visualization. Stable KP MEF and KPC knockout cell lines were generated via selection in 2 μg/ml puromycin followed by single colony isolation. In vitro luciferase activity was measured using a Lumat model LB luminometer (Promega) and the Luciferase Reporter Gene Assay according to manufacturer’s instructions (Roche). Hep3B cells were stimulated with human IL-6 (Cell Signaling Technology) at a concentration of 20ng/ml for 24 hours.

Lentiviral (adapted from the pWPXL/pEF1a backbone) and pEGFP-N1 (Addgene) expression vectors encoding mutant STAT3 alleles were derived using site-directed mutagenesis. The final plasmids were sequence confirmed. The p3XGAS-Hsp70-Luc reporter plasmid was described (19). For CRISPR/Cas9-mediated knockouts, we used single guide RNAs (sgRNAs) for STAT3 (5’-gcagctggacacacgctacc-3’ or 5’-gtacagcgacagcttcccca-3’), SMAD4 (5′-ggtggcgttagactctgccg-3′), and TGFBR2 (5’-ccttgtagacctcggcgaag-3’) cloned into LentiCRISPRv2 puro (Addgene) (20). Lentiviruses were produced by transient transfection of HEK293T and collected according to standard protocols.

All animal studies were approved by the Institutional Animal Care and Use Committee at Stony Brook University. Male NU/J (nude) mice (5 weeks old) (The Jackson Laboratory) were inoculated orthotopically or subcutaneously with 10⁴ cells in 100 μl of Matrigel (Corning) diluted 1:7 with Opti-MEM (Corning). Orthotopic implantations into the pancreas were performed using standard procedures (21). Pancreatic tumor latency was determined through abdominal palpation. We defined subcutaneous tumor latencies as the period between implantation of tumorigenic cells into mice and the appearance of tumors 1 mm in diameter. The end point was a tumor diameter of 0.5 cm. Statistical analyses were performed using two-tailed Student’s t-test at the 95% confidence interval. P ≤ 0.05 was considered statistically significant. Tumor-initiating cell (TIC) frequency was determined by extreme limiting dilution assays and online ELDA software (http://bioinf.wehi.edu.au/software/elda/). The number of cells in each subcutaneous injection ranged from 10² to 10⁴. Mouse tumor tissue was harvested, fixed in 5 volumes of 4% paraformaldehyde for 48 hours, and processed via the Stony Brook University Histology Core. Paraffin-embedded formalin-fixed 5 μm sections were stained with hematoxylin and eosin for histology.

Western blotting was performed using antibodies against AKT1 (4691), P-AKT1 pS473 (4060), CDH1 (3195), P-ERK1/2 (4370), SMAD4 (46535), P-STAT3 pY705 (9131), VIM (5741) (all from Cell Signaling), ERK1/2 (05-157, Millipore), STAT3 (610190, BD) and TGFBR2 (sc-400, Santa Cruz). Whole cell extracts were prepared by lysing cells in buffer containing 10 mM Tris HCl, pH7.4, 150 mM NaCl, 1 mM EDTA, 10% glycerol, 1% Triton X100, 40 mM NaVO4, 0.1% SDS, and 1x protease inhibitors (Roche). Western blots were imaged using Image Studio software (LI-COR). Total cellular RNA was isolated using PureLink RNA (ThermoFisher) according to manufacturer’s specifications and phenol-extracted. Pancreatic tissues were incubated for 24 hours at 4°C in ≥5 volumes of RNAlater solution (ThermoFisher) to preserve RNA integrity. RNA sequencing and bioinformatics were performed by Novogene Corporation (http://en.novogene.com). STAT3 and SMAD4 knockout signature scores (STAT3KO_UP, STAT3KO_DN, SMAD4KO_UP, SMAD4KO_DN) were computed as the average of RNA expression values (fpkm) of the top up- or down-regulated genes in KPC cell lines. EMT and mouse Ras Dependency Index (RDI) scores were calculated using defined gene sets (22–24). For publicly available human datasets, RDI, KRAS_sig, RSK_sig, epithelial (EPI) and mesenchymal (MES) gene expression scores were calculated as the sum of RNA expression values (z-scores) using previously characterized gene modules (25, 26).

Statistical analysis was performed using two-tailed Student’s t-test, Fisher’s exact test or Wilcoxon test, as appropriate for the dataset. An FDR adjusted p-value (q-value) was calculated for multiple comparison correction. Individual mice and tumor cell lines were considered biological replicates. Statistical details for each experiment are denoted in the corresponding figures and figure legends. The micrographs (H&E) represent at least three independent experiments. All data are presented as mean ± SD. In box and whisker plots, the middle line is plotted at the median, the upper and lower hinges correspond to the first and third quartiles, and the ends of the whisker are set at 1.5 x IQR above the third quartile and 1.5 x IQR below the first quartile (IQR, interquartile range or difference between the 25th and 75th percentiles).

To assess the role of STAT3 in KRAS-driven tumorigenesis, we measured proliferation rates, contact inhibition, and tumor formation in mice. We have reported that p53-null mouse embryonic fibroblasts expressing endogenous mutant KRAS^G12D (termed KP MEFs) exhibit typical features of oncogenic transformation using quantitative and sensitive assays (16). CRISPR/Cas9-mediated gene editing was used to generate isogenic STAT3 knockouts in the KP MEFs, and gain of function (GOF) and loss of function (LOF) mutant STAT3 alleles were stably integrated into cells via lentiviral vectors (Figs. 1A, 1B). We used naturally occurring GOF mutants, Y640F, K658Y and D661Y, and a synthetic mutant STAT3C (A662C/N664C), which all render a persistently phosphorylated and thus hyperactive STAT3 pY705 (9, 27). LOF mutations targeted the STAT3 DNA binding (EE434-435AA and VVV461-463AAA) and transactivation domains (Y705F and S727A) (28) (Fig. S1A). All mutants exhibited relatively uniform expression levels that were ∼5-fold higher than endogenous STAT3 (Figs. 1B, S1B). As reported, STAT3 Y640F, K658Y and STAT3C displayed increased levels of Y705 phosphorylation. As neither STAT3 WT nor STAT3 GOFs exhibited robust phosphorylation on S727, we used a validated STAT3 mutant S727E that mimics the phosphorylation of S727 (29) (Fig. S1A).

Isogenic KP MEF cell lines harboring wild-type or mutant STAT3 were evaluated for growth, tumorigenesis, and pathway activation. Loss of STAT3 expression did not affect cell growth under standard culture conditions (Fig. S1C). Likewise, loss of STAT3 did not affect KRAS-induced transformation. This is indicated by the ability of STAT3 KO cells to grow in multilayers and form transformed foci to the same extent as controls (Fig. 1C). Similar results were obtained using LOF mutations in STAT3 DNA binding (EE434-435AA and VVV461-463AAA) and transactivation domains (Y705F and S727A). In contrast, STAT3 GOF mutations, Y640F, D661Y, and K658Y, impaired KRAS-induced focus formation (p<0.005 by two-tailed T test, Figs. 1C, 1D). Because the transduced STAT3 constructs co-expressed a GFP reporter, the formation of transformed foci was visualized by fluorescence microscopy (Fig. 1E). Through these real-time studies we discovered that only a small percentage of cells with hyperactive STAT3 Y640F had some ability to form transformed colonies, while the majority of cells remained contact inhibited. We noted that STAT3 S727E had no significant effect on cell transformation, while the Y640F/S727E double mutant displayed only a marginal increase in the number of transformed foci relative to STAT3 Y640F itself (Fig. 1C). Thus, the hyperactive Y640F mutation exerts a dominant influence over S727E in KRAS-transformed MEFs. As expected, lentiviral expression of wild-type or GOF STAT3 alleles, Y640F and D661Y, failed to transform primary p53KO MEFs or immortalized NIH 3T3 cells (Fig. S1D, data not shown), indicating that STAT3 does not display intrinsic oncogenicity of its own.

To investigate tumorigenic effects of STAT3 in vivo, subcutaneous implants of KP MEFs into nude mice were used. Tumors developed by STAT3 KO cells showed growth characteristics similar to those of STAT3 intact vector controls (Fig. 2A). In contrast, cells expressing the hyperactive STAT3 Y640F and, to a lesser extent, K658Y mutations were delayed in their ability to form tumors in mice. We used fluorescence-activated cell sorting for GFP to fractionate STAT3 Y640F MEFs into pools with low and high STAT3 pY705 expression (Fig. 2B). Following implantation in mice, cell populations with high STAT3 pY705 expression developed tumors more slowly compared to low expressing cells (p=0.005, Fig. 2C). Therefore, there is a dose-dependent ability of hyperactive STAT3 Y640F to limit tumorigenicity of KRAS-transformed MEFs. Limiting dilution assays in nude mice revealed that the frequency of tumor-initiating cells was reduced by approximately 9-fold in high STAT3 Y640F expressing cells compared to control cells (Fig. 2D).

To determine whether specific Y705 phosphorylation and DNA binding are required for STAT3 Y640F to exert its suppressive activity, we generated three double mutants: a phosphorylation defective STAT3 Y640F/Y705F, and two DNA-binding domain (DBD) mutants; STAT3 Y640F/R382W and Y640F/V463Δ (Fig. S1E). Both R382W and V463Δ are recurrent STAT3 mutations observed in humans (30). A STAT3-responsive luciferase reporter assay was used to confirm that the DBD mutants are impaired in their ability to induce STAT3-mediated gene transcription (Fig. S1F). Notably, the DBD mutants of STAT3 Y640F lost the ability to attenuate KRAS-mediated MEF transformation despite their continuous pY705 phosphorylation (Figs. 2E, S1E). The phosphorylation defective STAT3 Y640F/Y705F double mutant was likewise impaired. We conclude that STAT3 Y640F-mediated inhibition of tumor development and KRAS-induced MEF transformation is dependent on STAT3 phosphorylation at Y705, DNA binding, and gene-specific transactivation.

To elucidate the means by which hyperactive STAT3 suppresses KP MEF transformation, we evaluated pathway activity by Western blot analysis, and transcription profile by whole exome RNA sequencing (RNA-seq). Western blot analyses of control and STAT3 Y640F-expressing MEFs showed unperturbed RAS/MAPK and PI3K/AKT signaling (as assessed by phosphorylated ERK1/2 and AKT1), suggesting that STAT3 does not directly alter these downstream KRAS effectors (Fig. S1B). RNA-seq analysis showed that expression of STAT3 Y640F in MEFs resulted in the differential expression of approximately 290 genes (p<0.05) compared to control cells. Gene ontology (GO) classification of biological processes showed an enrichment of pathways consistent with the role of STAT3 as a mediator of immunity and inflammatory response (Fig. S1G). Biological processes attenuated by hyperactive STAT3 included differentiation and tissue development, and pathways mediated by TGF-β family signaling (12, 13, 31). We therefore tested whether inactivation of the TGF-β pathway could inhibit KRAS-induced MEF transformation. Indeed, ablation of the TGF-β signaling components TGFBR2 or SMAD4 in KP MEFs using CRISPR/Cas9-mediated gene editing nearly eliminated foci formation and effects of STAT3 (Fig. SH, S1I). The results support the premise that hyperactive STAT3 interferes with KRAS-induced transformation through suppression of the TGF-β pathway. This prompted us to investigate the functional interaction of STAT3 and TGF-β/SMAD4 in epithelial carcinogenesis.

A notable feature of KRAS mutant cancers, including those of the pancreas, colon, and lung, is that they tend to fall into two classes based on their canonical KRAS and TGF-β signaling; those that have a strong dependence on KRAS signaling (KRAS-dependent) or those that have less dependence on canonical KRAS signaling (KRAS-independent) (25, 26). KRAS-dependent tumors have been associated with an epithelial gene signature and morphology, whereas KRAS-independent tumors show enriched expression of mesenchymal genes. Since STAT3 and TGF-β have been shown to compete, cooperate, or antagonize each other in many other contexts (32–35), we investigated STAT3 as it relates to KRAS dependency. To that end, we used murine pancreatic ductal adenocarcinoma (PDAC)-derived cell lines bearing endogenous KRAS^G12D and TP53^R172H mutations (termed KPC) (17). CRISPR/Cas9 gene editing was used to ablate STAT3, SMAD4 or TGFBR2 expression in PDAC cells (Fig. S2A) and the behavior of these cells was tested alongside previously generated KRAS knockout cells (23). The cells were implanted orthotopically into the pancreata of nude mice, and animals were observed for latency of tumor formation and changes in tumor morphology.

Pancreatic tumors were detected within three weeks following implantation of 10⁴ parental control, STAT3 knockout (KO), SMAD4 KO, or TGFBR2 KO PDAC cells, and there was no statistical difference in tumor latency between the groups. However, there was considerable difference in tumor morphology. Tumors in the parental (intact) group were characterized by classical adenocarcinoma-like morphology with glandular structures (Fig. 3A). In comparison, KRAS KO tumors displayed a highly sarcomatoid morphology indicative of full EMT. Loss of STAT3 also induced a morphological change compatible with EMT, whereas overexpression of hyperactive STAT3 Y640F resulted in the sporadic co-occurrence of squamous and glandular differentiation. Loss of SMAD4 or TGFBR2 was associated with increased epithelial differentiation relative to controls. This is consistent with previous findings showing that the TGF-β pathway is a key regulator of tumor cell differentiation and malignant behavior, but not growth rate (36, 37).

We determined whether differences in tumor morphology were coordinate with changes in gene expression. Comparative RNA-seq analyses of parental KPC control and STAT3 KO cell lines revealed distinct transcriptional profiles that included more than 700 differentially expressed genes (Fig. 3B). The enriched genes in KPC parental cells included those corresponding to the major structural proteins in epithelial cells, such as cadherins, claudins, and tight junctions (Fig. 3C). In contrast, STAT3 KO cells were enriched in signatures of EMT and embryonic organ morphogenesis (Fig. 3D). The overall pattern of gene expression suggests that loss of STAT3 is associated with activation of partial rather than complete EMT, since epithelial markers (e.g. CDH1 and EpCAM) continue to be expressed, but mesenchymal markers (e.g. FN1 and various collagens) have been acquired (38, 39) (Figs. S2B). As TGF-β classically promotes EMT, we also focused on TGF-β family genes (40). Among these genes, STAT3 KO cells had a significant increase in TGFB1, TGFB3 and INHBA expression (Fig. S2C). The expression of EMT-activating transcription factors SNAI, TWIST and ZEB was not strongly affected, indicating that induction of EMT involves additional STAT3 dependent regulators. We did identify transcription factors, such as JUNB and SOX4, that associate with EMT (41). We computed EMT scores using gene expression values of epithelial (EPI) and mesenchymal (MES) genes. STAT3 and KRAS KO KPC cells displayed similar levels of EMT at the level of gene expression (Fig. S2D) (22). In contrast, SMAD4 KO cells displayed reduced expression of EMT-related genes, while genes involved in epithelial differentiation and RAS dependency were among the most upregulated (Fig. S2C).

Notably, comparative analysis between STAT3 and KRAS knockout KPC cells revealed approximately 250 STAT3 target genes (>30%) that were similarly up- or down-regulated (Pearson’s r=0.88, p<0.00001), suggesting that STAT3 loss partially phenocopies the effects of KRAS inactivation (Fig. 3E). GO classifications of the overlapping genes in KRAS and STAT3 knockouts included developmental processes and mesenchymal tissue remodeling (Figs. 3D, S2E). As a proof of concept for the functional connection of STAT3 to KRAS dependency, we used gene expression data to compute RAS dependency scores for parental and knockout cells. A mouse KRAS dependency signature (21 genes) was derived from single cell RNA-seq data of KRAS intact vs. KRAS knockout PDAC tumors and used for the analyses (Fig. S2F)(23). Results showed RAS dependency scores had a significant positive correlation with STAT3 expression and KRAS expression and a negative correlation with SMAD4 expression (Fig. 3E). Whole tumor RNA-seq analysis of STAT3 intact and knockout pancreatic tumors supported these findings (Fig. S2F). Together these data indicate that STAT3 is a genetic modifier that can regulate KRAS dependency and tumor development through counterposing EMT.

The apparent antagonism between STAT3 and induction of EMT was of particular interest, as TGF-β-induced EMT appears to confer adaptive resistance to KRAS inhibition (25, 42). To test the relative importance of the STAT3 and TGF-β pathways to tumor morphology and functionality, we generated STAT3/SMAD4 and STAT3/TGFBR2 double knockout (DKO) cell lines. DKO cell lines formed pancreatic tumors in mice, but their histological features were distinct from SMAD4 or TGFBR2 single knockouts as they produced mixed epithelial/mesenchymal morphologies with cells expressing both E-cadherin and vimentin (Figs. 4A, 4B). SMAD4 or TGFBR2 intact tumors displayed features of EMT in the absence of STAT3, whereas SMAD4 or TGFBR2 KO tumors displayed a well-differentiated epithelial phenotype only in the presence of STAT3 (Figs. 3A, 4A). The data reinforce the notion that functional antagonism of STAT3 and TGF-β/SMAD4 controls PDAC development and KRAS dependency (Fig. 4C).

We analyzed human PDAC databases to determine if our STAT3 and SMAD4 gene expression signatures could be projected onto human tumors. SMAD4 and STAT3 signature scores were computed from the top up- or down-regulated genes in KPC cell lines (Fig. S3A). When aligned with human PDAC samples from the TCGA cohort (stage I/II tumors), STAT3 and SMAD4-regulated gene signatures demonstrated significant statistical correspondence, supporting the selective antagonism of STAT3 and SMAD4 (Pearson’s r>0.5, p<0.00001) (Fig. 4D). As SMAD4 is frequently deleted in PDAC, samples expressing only wild-type SMAD4 were manually curated from the TCGA cohort (n=112). Tumors were classified as either epithelial (EPI) or mesenchymal (MES) using previously characterized gene sets (Figs. S3B). Results demonstrate that STAT3-regulated gene expression is closely associated with epithelial differentiation, while SMAD4-regulated gene expression is strongly associated with EMT (Pearson’s r>0.7) (Fig. S3C). The association with KRAS dependency status was also revealing. PDAC tumor samples from the TCGA cohort were grouped as KRAS-dependent/KRAS type or KRAS-independent/RSK type based on previously derived KRAS dependency signatures (Fig. S3D)(25, 26). KRAS-dependent tumors showed enriched expression of epithelial genes (EPI) and reduced expression of mesenchymal genes (MES), whereas the KRAS-independent samples displayed the inverse, as previously reported (24–26). Importantly, the KPC STAT3 knockout gene signature (i.e. upregulated genes) co-segregated with human KRAS-independent/mesenchymal PDAC tumors, while the SMAD4 knockout gene signature (i.e. upregulated genes) co-segregated with human KRAS-dependent/epithelial tumors (Figs. S3C). A similar trend was observed for the PanCuRx Translational Research Initiative (COMPASS, stage IV PDAC) cohort mainly composed of liver metastases, as the STAT3-reliant gene signature was also enriched in epithelial and KRAS dependent samples (r>0.7, p<0.00001) (Fig. 4E). Liver is the main site of PDAC metastases, and pancreatic cancer metastases commonly display a stabilized epithelial phenotype (43, 44). Overall, results demonstrate that STAT3-regulated gene expression is closely associated with epithelial differentiation and RAS dependency, while SMAD4-regulated gene expression is strongly associated with EMT and RAS independence. These findings underscore our basic premise that there exists an epistatic antagonism between STAT3 and SMAD4, highlighting a new role for STAT3 as a genetic modifier in KRAS mutant cancer.