We have shown that KRAS–TP53 genomic coalteration is associated with immune-excluded microenvironments, chemoresistance, and poor survival in pancreatic ductal adenocarcinoma (PDAC) patients. By treating KRASTP53 cooperativity as a model for high-risk biology, we now identify cell-autonomous Cxcl1 as a key mediator of spatial T-cell restriction via interactions with CXCR2+ neutrophilic myeloid-derived suppressor cells in human PDAC using imaging mass cytometry. Silencing of cell-intrinsic Cxcl1 in LSL-KrasG12D/+;Trp53R172H/+;Pdx-1Cre/+(KPC) cells reprograms the trafficking and functional dynamics of neutrophils to overcome T-cell exclusion and controls tumor growth in a T cell–dependent manner. Mechanistically, neutrophil-derived TNF is a central regulator of this immunologic rewiring, instigating feed-forward Cxcl1 overproduction from tumor cells and cancer-associated fibroblasts (CAF), T-cell dysfunction, and inflammatory CAF polarization via transmembrane TNF–TNFR2 interactions. TNFR2 inhibition disrupts this circuitry and improves sensitivity to chemotherapy in vivo. Our results uncover cancer cell–neutrophil cross-talk in which context-dependent TNF signaling amplifies stromal inflammation and immune tolerance to promote therapeutic resistance in PDAC.

Significance:

By decoding connections between high-risk tumor genotypes, cell-autonomous inflammatory programs, and myeloid-enriched/T cell–excluded contexts, we identify a novel role for neutrophil-derived TNF in sustaining immunosuppression and stromal inflammation in pancreatic tumor microenvironments. This work offers a conceptual framework by which targeting context-dependent TNF signaling may overcome hallmarks of chemoresistance in pancreatic cancer.

This article is highlighted in the In This Issue feature, p. 1275

Pancreatic ductal adenocarcinoma (PDAC) is a lethal malignancy characterized by extreme therapeutic resistance (1, 2). The dominant contributors to therapeutic resistance in PDAC are an undruggable genomic landscape typified by co-occurring KRAS and TP53 mutations (3), tolerogenic signaling from myeloid cells (4, 5)—particularly neutrophilic/polymorphonuclear myeloid-derived suppressor cells (PMN-MDSC)—which promotes T-cell dysfunction and exclusion (6, 7), and proinflammatory polarization of cancer-associated fibroblasts (iCAF; ref. 8) that instigates stromal inflammation by elaborating soluble factors (e.g., IL6) that further accelerate myeloid chemotaxis and induce chemoresistant CAF–tumor cell IL6/STAT3 signaling (9). However, how these complex circuitries converge in the tumor microenvironment (TME) to mediate therapeutic resistance in PDAC is incompletely understood.

We have recently shown that KRAS–TP53 coalteration is associated with worse survival in patients with advanced PDAC (6). Moreover, KRAS–TP53 cooperativity encodes for cell and non–cell-autonomous transcriptional networks that drive innate immune cell–enriched and T cell–excluded immune microenvironments (6). Using an integrative molecular approach, we identified a KRAS–TP53 cooperative “immuno­regulatory” program—encompassing cancer cell–intrinsic Cxcl1 and myeloid-derived TNF, among others—associated with chemotherapy resistance and worse overall survival in PDAC patients enrolled in the COMPASS trial (10).

Here, through the lens of KRASTP53 cooperativity as a model for high-risk biology, we uncover how cancer cell–autonomous programs dictate PMN-MDSC activation and functional plasticity, which in turn regulates T-cell dysfunction and stromal inflammation in the PDAC TME. Specifically, cancer cell–autonomous Cxcl1—which is transcriptionally regulated by a Creb-dependent mechanism—imposes T-cell restriction via interactions with CXCR2+ PMN-MDSCs in spatially annotated human tumors. Silencing of tumor cell–intrinsic Cxcl1 overcomes CD8+ T-cell exclusion by inducing a fundamental reprogramming of PMN-MDSC function in vivo. Neutrophil-restricted TNF—via Cxcr2–Map3k8–Tnf signaling—emerges as a central driver of this MDSC reprogramming and TME remodeling, instigating feed-forward tumor cell/CAF-Cxcl1 overproduction, immune tolerance, and iCAF polarization via transmembrane TNF (tmTNF)–TNFR2 signaling. TNFR2 inhibition disrupts this circuitry to improve sensitivity to chemotherapy in vivo, implicating context-dependent tmTNF–TNFR2 signaling as a promising therapeutic target in this deadly disease.

Imaging Mass Cytometry Reveals Tumor Cell–Intrinsic Overexpression of CXCL1 in KRAS–TP53 Coaltered Human PDAC

We have previously shown that KRAS–TP53 genomic coalteration encodes cancer cell–autonomous transcriptional programs that orchestrate innate immune enrichment and T-cell exclusion in PDAC (6, 11). To identify key tumor cell–intrinsic factors mediating innate immune recruitment, we examined differentially expressed bulk transcriptomes from KRAS–TP53 coaltered (n = 23) versus KRAS-altered/TP53WT (n = 5) human PDAC cell lines from the Cancer Cell Line Encyclopedia (CCLE; Fig. 1A; ref. 12). Gene set enrichment analysis (GSEA) highlighted overexpression of several pathways related to granulocyte/neutrophil function in KRAS–TP53 coaltered PDAC (Fig. 1B; Supplementary Fig. S1A; Supplementary Table S1). Among transcripts conserved across these pathways, we observed significant upregulation of CXCL1—a secreted neutrophilic chemoattractant ligand—in KRAS–TP53 coaltered transcriptomes (Fig. 1B; Supplementary Fig. S1B). We corroborated these findings in preclinical genetic models, where Cxcl1 mRNA expression via RNA in situ hybridization (P = 0.0002; Fig. 1C; Supplementary Fig. S1C) and Cxcl1 immunostaining via IHC (Supplementary Fig. S1D) was substantially increased in the epithelial/ductal compartment of LSL-KrasG12D/+;Trp53R172H/+;Pdx-1Cre/+(KPC) compared with LSL-KrasG12D/+;Pdx-1Cre/+(KC) tumors.

Figure 1.

Cxcl1 is overexpressed in Ras–p53 cooperative PDAC and governs spatial exclusion of T cells in human tumors. A, Schematic of transcriptomic comparison between KRASTP53 coaltered (n = 23) and KRAS-altered/TP53WT (n = 5) human PDAC cell lines from the CCLE, with subsequent differential expression analysis (DEA) and gene set enrichment analysis (GSEA). B, Net plot visualizing the top 3 gene sets related to granulocyte/neutrophil function overexpressed in KRAS–TP53 coaltered PDAC tumor cell transcriptomes, with 5 transcripts conserved between these pathways (CXCL1 highlighted in dashed box) shown. C, H&E sections paired with immunostaining for pancytokeratin (PanCK), podoplanin (PDPN), and RNA in situ hybridization to detect Cxcl1 mRNA in representative sections from volume-matched tumors in genetic models LSL-KrasG12D/+; Trp53R172H/+; Pdx-1Cre/+ (KPC; 6-month-old) and LSL-KrasG12D/+;Pdx-1Cre/+ (KC; 12-month-old). Quantification of relative Cxcl1 mRNA expression in PanCK+ cells provided in Supplementary Fig. S1C. D, IMC comparing epithelial expression of CXCL1 in human KRAS–TP53 coaltered (n = 5) compared with KRAS-altered/TP53WT (n = 3) tumors at single-cell resolution, with an adjacent histogram showing quantification of an average number of PanCK+CXCL1+ cells per 5,000 single cells in each tumor section across groups. E, IMC image segmentation into PanCK+ tumor cells, αSMA+ fibroblasts, CD11b+ myeloid cells, CD3+ T cells in KRASTP53 coaltered (KRAS+TP53) versus KRAS-altered/TP53WT (KRAS) human PDAC sections, with adjacent quantification of mean intensity of CXCL1 expression in each cell type across groups. Cell populations were grouped according to any positive pixel intensity for the respective marker, with discrepancies and overlap between phenotypes reconciled manually. F, Uniform Manifold Approximation and Projection (UMAP) showing annotated clusters from single-cell RNA-seq (scRNA-seq) data in human PDAC patient samples (n = 16; top), with bubble plot showing relative CXCR2 expression between different clusters (bottom). G, Heat map showing tumor cell (donor) to granulocyte (recipient) ligand–receptor interactome in human scRNA-seq dataset using NicheNet algorithm, with CXCL1CXCR2 interaction highlighted (red box). H, Schematic representation of IMC workflow to provide spatially resolved single-cell phenotypes of human PDAC tumors. I and J, H&E, spatial phenotype map, and image segmentation of representative KRAS–TP53 coaltered human PDAC tumor section (I), with single-cell clustering in T-distributed stochastic neighborhood embedding (tSNE) maps (J) of 72,880 single cells in 8 predefined ROIs each from a unique patient sample, distributed into epithelial/tumor cell (PanCK+CXCL1+), stromal/fibroblast (αSMA+), endothelial (CD31+), myeloid (CD11b+), PMN-MDSC (CD15+CXCR2+), M2-like macrophage (CD68+CD163+), and CD8+ T-cell (CD3+CD8+) populations. K, Tissue heat maps showing expression of αSMA, PanCK, CXCL1, CD15, CXCR2, CD3, and CD8 in representative KRAS–TP53 coaltered human PDAC tumor section. L, Heat map depicting spatially resolved distances of single CXCR2+CD11b+CD15+ PMN-MDSC, CD11b+ myeloid cell, CD68+CD163+ M2-like macrophage, and CD3+CD8+ T cells from PanCK+ tumor cells in 8 ROIs from human PDAC sections.

Figure 1.

Cxcl1 is overexpressed in Ras–p53 cooperative PDAC and governs spatial exclusion of T cells in human tumors. A, Schematic of transcriptomic comparison between KRASTP53 coaltered (n = 23) and KRAS-altered/TP53WT (n = 5) human PDAC cell lines from the CCLE, with subsequent differential expression analysis (DEA) and gene set enrichment analysis (GSEA). B, Net plot visualizing the top 3 gene sets related to granulocyte/neutrophil function overexpressed in KRAS–TP53 coaltered PDAC tumor cell transcriptomes, with 5 transcripts conserved between these pathways (CXCL1 highlighted in dashed box) shown. C, H&E sections paired with immunostaining for pancytokeratin (PanCK), podoplanin (PDPN), and RNA in situ hybridization to detect Cxcl1 mRNA in representative sections from volume-matched tumors in genetic models LSL-KrasG12D/+; Trp53R172H/+; Pdx-1Cre/+ (KPC; 6-month-old) and LSL-KrasG12D/+;Pdx-1Cre/+ (KC; 12-month-old). Quantification of relative Cxcl1 mRNA expression in PanCK+ cells provided in Supplementary Fig. S1C. D, IMC comparing epithelial expression of CXCL1 in human KRAS–TP53 coaltered (n = 5) compared with KRAS-altered/TP53WT (n = 3) tumors at single-cell resolution, with an adjacent histogram showing quantification of an average number of PanCK+CXCL1+ cells per 5,000 single cells in each tumor section across groups. E, IMC image segmentation into PanCK+ tumor cells, αSMA+ fibroblasts, CD11b+ myeloid cells, CD3+ T cells in KRASTP53 coaltered (KRAS+TP53) versus KRAS-altered/TP53WT (KRAS) human PDAC sections, with adjacent quantification of mean intensity of CXCL1 expression in each cell type across groups. Cell populations were grouped according to any positive pixel intensity for the respective marker, with discrepancies and overlap between phenotypes reconciled manually. F, Uniform Manifold Approximation and Projection (UMAP) showing annotated clusters from single-cell RNA-seq (scRNA-seq) data in human PDAC patient samples (n = 16; top), with bubble plot showing relative CXCR2 expression between different clusters (bottom). G, Heat map showing tumor cell (donor) to granulocyte (recipient) ligand–receptor interactome in human scRNA-seq dataset using NicheNet algorithm, with CXCL1CXCR2 interaction highlighted (red box). H, Schematic representation of IMC workflow to provide spatially resolved single-cell phenotypes of human PDAC tumors. I and J, H&E, spatial phenotype map, and image segmentation of representative KRAS–TP53 coaltered human PDAC tumor section (I), with single-cell clustering in T-distributed stochastic neighborhood embedding (tSNE) maps (J) of 72,880 single cells in 8 predefined ROIs each from a unique patient sample, distributed into epithelial/tumor cell (PanCK+CXCL1+), stromal/fibroblast (αSMA+), endothelial (CD31+), myeloid (CD11b+), PMN-MDSC (CD15+CXCR2+), M2-like macrophage (CD68+CD163+), and CD8+ T-cell (CD3+CD8+) populations. K, Tissue heat maps showing expression of αSMA, PanCK, CXCL1, CD15, CXCR2, CD3, and CD8 in representative KRAS–TP53 coaltered human PDAC tumor section. L, Heat map depicting spatially resolved distances of single CXCR2+CD11b+CD15+ PMN-MDSC, CD11b+ myeloid cell, CD68+CD163+ M2-like macrophage, and CD3+CD8+ T cells from PanCK+ tumor cells in 8 ROIs from human PDAC sections.

Close modal

To validate and extend these findings in patient-derived PDAC tumors, we utilized imaging mass cytometry (IMC) to examine tissue-level CXCL1 expression at single-cell resolution in KRAS–TP53 coaltered (n = 5) versus KRAS-altered/TP53WT (n = 3) human tumors derived from patients with resectable or borderline resectable PDAC who underwent upfront surgical resection at our institution (Fig. 1D; Supplementary Table S2). Pathologist-selected regions of interest (ROI) from each tumor section probed with metal ion-conjugated antibodies for pan-cytokeratin (PanCK:epithelial), α-smooth muscle actin (α-SMA:fibroblast), CD11b (myeloid), CD3 (T-cell), and CXCL1 were laser-ablated, and atomized ions were acquired using time-of-flight mass cytometry (cyTOF). Image segmentation and quantification revealed a significantly higher proportion of PanCK+CXCL1+ epithelial/tumor cell islands in KRASTP53 coaltered versus KRAS-altered/TP53WT tumors (P = 0.0065; Fig. 1D). Mean CXCL1 expression was disproportionately higher in single PanCK+ tumor cells compared with αSMA+ fibroblasts, CD11b+ myeloid cells, or CD3+ T cells in KRASTP53 coaltered tumors (all P < 0.001; Fig. 1E). These findings were corroborated in two independent single-cell datasets, in which highest CXCL1 gene expression was observed in tumor/malignant cells compared with all other subclusters in single-nuclear RNA sequencing (snRNA-seq) data from 43 patients (13) and in single-cell RNA sequencing (scRNA-seq) data from 16 human PDAC tumors (ref. 6; Supplementary Fig. S1E and S1F). In IMC analysis, mean CXCL1 expression was also significantly increased in all other cellular compartments in KRASTP53 coaltered versus KRAS-altered/TP53WT tumors (Fig. 1E). We also observed >10-fold higher Cxcl1 secretion by human KRAS–TP53 coaltered (Panc02.03, Capan1, Mia-Paca2) versus KRAS-altered/TP53WT (Hs766t) cell lines in vitro (P < 0.0001; Supplementary Fig. S1G).

CD8+ T Cells Are Spatially Excluded from Cellular Neighborhoods Comprising Tumor Cell Cxcl1 and CXCR2+ PMN-MDSCs in Human PDAC

CXCL1 has been implicated as a critical regulator of T-cell exclusion in congenic KPC models (14) and is also a key component of the innate immunoregulatory transcriptional program overexpressed in KPC tumor cells via scRNA-seq in our recent study (6). To interrogate how tumor cell–intrinsic CXCL1 governs spatial relationships with innate and adaptive immune populations in human PDAC, we first sought to determine the localization of its cognate receptor CXCR2 within the TME. Via scRNA-seq and snRNA-seq, we identified that CXCR2 was near-exclusively expressed in intratumoral PMN-MDSCs from human PDAC samples (n = 16; ref. 15; Fig. 1F; Supplementary Fig. S2A and S2B; and n = 43; ref. 13; Supplementary Fig. S2C), as well as in KPC (Supplementary Fig. S2D) and Ptf1aCre/+;LSL-KrasG12D/+;Tgfbr2flox/flox (PKT; ref. 16; Supplementary Fig. S2E) genetic mouse models. Computational modeling of intercellular ligand–receptor interactomes (17) in human PDAC scRNA-seq data revealed strong predicted interaction between tumor cell–CXCL1 and neutrophil-CXCR2 (Fig. 1G). Single-cell deconvolution in the TCGA-PDAC dataset (n = 178) confirmed robust correlations between CXCL1 expression and enrichment in computationally inferred neutrophil (P < 0.0001), MDSC (P < 0.001), and macrophage (P = 0.002) populations (Supplementary Fig. S2F).

We next interrogated tissue-level spatial relationships between CXCL1-expressing tumor islands and immune populations in the human PDAC TME. Using IMC, we examined 72,880 cells from 8 predefined ROIs (each from a unique patient) in multicellular neighborhoods to quantify the expression of 11 phenotypic markers and spatial features of each cell (Fig. 1H and I). Image segmentation into single cells generated a phenotype map and subsequent clustering revealed distinct epithelial/tumor cell (PanCK+), stromal (αSMA+), endothelial (CD31+), myeloid (CD11b+), PMN-MDSC (CD15+CXCR2+), M2-like macrophage (CD68+CD163+), and CD8+ T-cell (CD3+CD8+) populations (Fig. 1J). Tissue heat maps confirmed the expression of CXCL1 predominantly in Pan­CK+ tumor cells compared with αSMA+ CAFs or CD11b+ myeloid cells, CXCR2 near exclusively in CD11b+CD15+ PMN-MDSCs, and CD8 in CD3+ T cells (Fig. 1K). Neighborhood analysis to quantify statistically significant interaction or avoidance between pairs of cellular phenotypes revealed strong spatial contiguity between PanCK+CXCL1+ tumor islands and CD11b+CD15+CXCR2+ PMN-MDSCs (median distance 7.3 ± 11.1 μm) and CD68+CD163+ M2-like macrophages (23.3 ± 20.0 μm). Importantly, CD3+CD8+ T cells were spatially excluded from contiguous PanCK+CXCL1+ and CXCR2+CD11b+CD15+ cellular communities (95.4 ± 63.2 μm, P < 0.0001; Fig. 1L).

KRAS and TP53 Mutations Cooperate to Transcriptionally Regulate CXCL1 via CREB Activation in Pancreatic Cancer Cells

To deconstruct individual elements of signaling networks driving the association between tumor cell–intrinsic Cxcl1 and T-cell exclusion in PDAC, we first investigated how KRAS and TP53 alterations cooperate to regulate cell–autonomous CXCL1 in tumor cells. To recapitulate KRAS–TP53 coalteration in isogenic human pancreatic epithelial systems, we transiently overexpressed mutant-TP53R175H or TP53WT cDNA constructs in KRAS-wild-type hPNE-KRASWT and KRASG12D-mutant hPNE-KRASG12D cells (18), and examined the dependence of Cxcl1 production on mutational status. Overexpression of KRASG12D, but not TP53R175H alone in hPNE-KRASWT cells, increased Cxcl1 secretion (P < 0.001). Furthermore, compared with TP53WT overexpression in hPNE-KRASG12D cells, KRASG12D and TP53R175H mutational cooperativity (Supplementary Fig. S3A) further augmented Cxcl1 production (P < 0.001; Fig. 2A).

Figure 2.

KRAS and TP53 mutations cooperate to transcriptionally regulate CXCL1 via CREB activation in pancreatic cancer cells. A, Schematic representing experimental constructs utilized to overexpress TP53WT or TP53R175H in isogenic hPNE-KRASWT or hPNE-KRASG12D pancreatic epithelial cells (top). Histogram showing Cxcl1 secretion from each hPNE cell system annotated by respective KRAS and/or TP53 mutational status (bottom). B, Bubble plot representing the top 10 transcription factors hyperphosphorylated in hPNE-KRASG12DTP53R175H compared with hPNE-KRASG12DTP53WT cells, with relative ratio of expression indicated on y-axis. C, Histograms representing relative fold change in CXCL1 gene expression (left) and secretion (in pg/mL; right) from hPNE-KRASG12DTP53WT and hPNE-KRASG12DTP53R175H in the absence or presence of Creb inhibitor 666-15 (CREBi 0.5 μmol/L for 24 hours, n = 3). D, Chromatin immunoprecipitation and sequencing (ChIP-seq) peak signals and heat maps of CREB regions in CREB and RNApol-II ChIP material (n = 2 biological replicates each) in KrasTrp53 cooperative KPC 6694c2 cells. E, ChIP-seq heat maps showing co-occupied CREB and RNApol-II peaks (N = 9,488), CREB-unique peaks (N = 6,799), RNApol-II unique peaks (N = 59,032), with adjacent callout box showing curated gene module implicated in inflammatory signaling and innate immune regulation. F, Integrative Genome Viewer (IGV) plot visualizing co-occupancy of peaks in CREB and RNApol-II ChIP-seq data at the transcriptional start site of Cxcl1 promoter. G, ChIP-qPCR of Cxcl1 and Cdh8 (negative control) from CREB and RNApol-II immunoprecipitated purified DNA in KPC 6694c2 cells. H and I,CXCL1 gene expression (left) and secretion (right) each in human MiaPaCa-2 cells (H) and human PDM-168 patient-derived organoids (I) in the absence or presence of CREBi 666-15 (0.5 μmol/L for 24 hours) or absence or presence of Creb siRNA (n = 3 each). J, Schematic of Creb inhibitor treatment of KPC orthotopic mice in vivo (top). Bar plots showing Cxcl1 gene expression via qPCR and protein levels via ELISA (in pg/mL) in whole-tumor lysates from vehicle-treated vs. Crebi-treated mice (n = 4–7 mice). K, Cxcl1 immunostaining with corresponding H&E staining in representative tumor sections from vehicle- vs. CREBi-treated mice (n = 4–5/group; scale bar, 50 μm).

Figure 2.

KRAS and TP53 mutations cooperate to transcriptionally regulate CXCL1 via CREB activation in pancreatic cancer cells. A, Schematic representing experimental constructs utilized to overexpress TP53WT or TP53R175H in isogenic hPNE-KRASWT or hPNE-KRASG12D pancreatic epithelial cells (top). Histogram showing Cxcl1 secretion from each hPNE cell system annotated by respective KRAS and/or TP53 mutational status (bottom). B, Bubble plot representing the top 10 transcription factors hyperphosphorylated in hPNE-KRASG12DTP53R175H compared with hPNE-KRASG12DTP53WT cells, with relative ratio of expression indicated on y-axis. C, Histograms representing relative fold change in CXCL1 gene expression (left) and secretion (in pg/mL; right) from hPNE-KRASG12DTP53WT and hPNE-KRASG12DTP53R175H in the absence or presence of Creb inhibitor 666-15 (CREBi 0.5 μmol/L for 24 hours, n = 3). D, Chromatin immunoprecipitation and sequencing (ChIP-seq) peak signals and heat maps of CREB regions in CREB and RNApol-II ChIP material (n = 2 biological replicates each) in KrasTrp53 cooperative KPC 6694c2 cells. E, ChIP-seq heat maps showing co-occupied CREB and RNApol-II peaks (N = 9,488), CREB-unique peaks (N = 6,799), RNApol-II unique peaks (N = 59,032), with adjacent callout box showing curated gene module implicated in inflammatory signaling and innate immune regulation. F, Integrative Genome Viewer (IGV) plot visualizing co-occupancy of peaks in CREB and RNApol-II ChIP-seq data at the transcriptional start site of Cxcl1 promoter. G, ChIP-qPCR of Cxcl1 and Cdh8 (negative control) from CREB and RNApol-II immunoprecipitated purified DNA in KPC 6694c2 cells. H and I,CXCL1 gene expression (left) and secretion (right) each in human MiaPaCa-2 cells (H) and human PDM-168 patient-derived organoids (I) in the absence or presence of CREBi 666-15 (0.5 μmol/L for 24 hours) or absence or presence of Creb siRNA (n = 3 each). J, Schematic of Creb inhibitor treatment of KPC orthotopic mice in vivo (top). Bar plots showing Cxcl1 gene expression via qPCR and protein levels via ELISA (in pg/mL) in whole-tumor lysates from vehicle-treated vs. Crebi-treated mice (n = 4–7 mice). K, Cxcl1 immunostaining with corresponding H&E staining in representative tumor sections from vehicle- vs. CREBi-treated mice (n = 4–5/group; scale bar, 50 μm).

Close modal

To determine signaling mechanisms regulating CXCL1 transcription downstream of KRAS–TP53 cooperativity, multiplex phospho-kinome arrays identified CREBSer133 as the top hyperphosphorylated transcription factor (P-adj < 0.01; Fig. 2B)—and top 10 hyperphosphorylated proteins overall (Supplementary Fig. S3B)—in hPNE-KRASG12D-TP53R175H compared with hPNE-KRASG12D-TP53WT cells. Interestingly, CREB1pSer133 has recently been implicated in hyperactivating prometastatic transcriptional networks in KRASTP53 cooperative PDAC via FOXA1-Wnt/β-catenin-mediated mechanism (19). Pharmacologic CREB inhibition with 666-15 (Supplementary Fig. S3C–S3E) in hPNE-KRASG12D-TP53R175H cells significantly reduced CXCL1 transcription (P = 0.009) and Cxcl1 secretion (P = 0.004; Fig. 2C). Moreover, 666-15–mediated reduction in Cxcl1 expression was validated in murine KPC tumor cells (Supplementary Fig. S3F).

To interrogate if Cxcl1 is a CREB target site, we next performed chromatin immunoprecipitation and sequencing (ChIP-seq) for CREB and RNApol-II—a mark associated with transcriptional start sites—in KrasG12D-Trp53R172H KPC6694c2 tumor cells. We observed 75,314 merged peaks between CREB and RNApol-II (Fig. 2D), with the majority of peaks present in genic/promoter regions close to transcription start sites (Supplementary Fig. S4A). A robust number of co-occupied and unique targets (∼9,488)—particularly genes encoding for proinflammatory mediators—were shared between CREB and RNApol-II (Fig. 2E; Supplementary Table S3). Notably, we observed significant cobinding of CREB and RNApol-II at the Cxcl1 promoter across biological replicates (Fig. 2E and F). Consistent with ChIP-seq findings, validation by ChIP-qPCR was performed in KPC 6694c2 cells, which showed co-occupancy of CREB and RNApol-II at Cxcl1 promoter, whereas no binding was observed for the negative control Cdh8 gene (Fig. 2G; Supplementary Fig. S4B), and in KPC K-8484 cells (refs. 20, 21; Supplementary Fig. S4C).

To corroborate the role of CREB in transcriptional regulation of CXCL1 in human KRASTP53 cooperative PDAC systems, we performed pharmacologic inhibition and RNA interference of CREB in KRASG12D-TP53G245S PDM-168 patient-derived organoid cells, and KRASG12C-TP53R248Wcoaltered MiaPaCa-2 cells. In both models, Cxcl1 expression and secretion were significantly reduced with CREB1-siRNA and 666-15 treatment (Fig. 2H and I). Finally, in vivo CREB inhibition using 666-15 in orthotopic KPC tumor-bearing mice resulted in a significant diminution in Cxcl1 expression and protein levels in whole-tumor lysates (P = 0.011; Fig. 2J), as well as a reduction in Cxcl1 immunostaining via IHC (Fig. 2K) and epithelial/ductal-specific Cxcl1 mRNA expression via RNA in situ hybridization (Supplementary Fig. S4D) in tumor sections. Altogether, these data highlight that cell-autonomous Ras and p53 mutations cooperate to regulate Cxcl1 via a CREB-dependent mechanism.

Genetic Silencing of Tumor Cell–Intrinsic Cxcl1 Overcomes T-cell Exclusion In Vivo

We next investigated how cell-intrinsic Cxcl1 mechanistically governs T-cell exclusion in vivo. We utilized KPC tumor cells (KPC 6694c2 backbone) in which Cxcl1 has been genetically silenced using CRISPR-Cas9 editing (ref. 14; Supplementary Fig. S5A) to generate syngeneic transplantation models with either control KPCEV (transduced with nontargeting sgRNA) or KPC-Cxcl1KO tumor cells in C57/BL6 mice. Coupled with reduced tumor growth kinetics in subcutaneous KPC-Cxcl1KO tumors (n = 5; Supplementary Fig. S5B), there was a dramatic reduction in primary tumor weights (n = 20, P < 0.0001; Fig. 3A) and metastatic outgrowth (Supplementary Fig. S5C and S5D), as well as significant improvement in overall survival (n = 15, median 48 vs. 21 days; P < 0.0001; Fig. 3B), in orthotopically injected KPC-Cxcl1KO compared with KPCEV tumor-bearing mice.

Figure 3.

Genetic silencing of tumor cell–intrinsic Cxcl1 overcomes T-cell exclusion and controls tumor growth in a CD8+ T cell–dependent manner in vivo.A, Violin plot representing difference in primary tumor weights in C57/BL6 mice orthotopically injected with KPCEV or KPC-Cxcl1KO tumor cells (n = 20/group; left), with representative images of tumors from each group showing phenotypic reproducibility (n = 5 each; right). B, Kaplan–Meier curves showing overall survival of KPC-Cxcl1KO and KPCEV orthotopically injected mice (n = 15; median 48 vs. 21 days). C, Bubble plot visualizing differentially upregulated pathways (using KEGG and Reactome knowledgebases) in 3-week whole-tumor transcriptomes from KPCEV compared with KPC-Cxcl1KO orthotopic tumors via RNA-seq (n = 3 biological replicates). D, Volcano plot depicting significantly enriched genes related to immune regulation in KPCEV (Cxcl1, Vegfa, Il6, and Csf2), and KPC-Cxcl1KO (Cxcl10, Gzmb, Cxcr3, Cxcl9, Cd96, Cd3d, Cd4, Ciita, and H2-Eb1). E, Pie charts showing relative proportions of immune-cell fractions of macrophages (Mϕ), PMN-MDSC, CD4+ T cells, and CD8+ T cells using CIBERSORT immune deconvolution from transcriptomes in KPCEV vs. KPC-Cxcl1KO tumors (n = 3 biological replicates per group). F and G, viSNE contour plots of flow-cytometric immunophenotyping in concatenated single-cell suspensions from KPCEV or KPC-Cxcl1KO orthotopic tumors (left), with adjacent violin plots (right) representing absolute cell counts of PMN-MDSC, CXCR2+ PMN-MDSC, monocytic MDSC (moMDSC), M2-like macrophage (F), CD4+ T cells and CD8+ T cells (G), and central memory T cells, effector memory T cells, degranulating CD8+ T cells (H) from each biological replicate (n = 6–8/group). I, Schematic of experimental design utilizing CD8+ T-cell neutralizing antibodies (CD8neuAb) in C57Bl/6 mice or CD8α−/− transgenic mice (B6.Cd8atm1Mak; CD8KO) for orthotopic injections. J, Representative ultrasound images from mice in each treatment group showing tumor growth dynamics in vivo. K and L, Tumor growth curves (K) and Kaplan–Meier survival estimates (L) from mice across 5 groups in T-cell depletion experiments (n = 10 mice/group), with median survival (MS) of each cohort indicated in parentheses.

Figure 3.

Genetic silencing of tumor cell–intrinsic Cxcl1 overcomes T-cell exclusion and controls tumor growth in a CD8+ T cell–dependent manner in vivo.A, Violin plot representing difference in primary tumor weights in C57/BL6 mice orthotopically injected with KPCEV or KPC-Cxcl1KO tumor cells (n = 20/group; left), with representative images of tumors from each group showing phenotypic reproducibility (n = 5 each; right). B, Kaplan–Meier curves showing overall survival of KPC-Cxcl1KO and KPCEV orthotopically injected mice (n = 15; median 48 vs. 21 days). C, Bubble plot visualizing differentially upregulated pathways (using KEGG and Reactome knowledgebases) in 3-week whole-tumor transcriptomes from KPCEV compared with KPC-Cxcl1KO orthotopic tumors via RNA-seq (n = 3 biological replicates). D, Volcano plot depicting significantly enriched genes related to immune regulation in KPCEV (Cxcl1, Vegfa, Il6, and Csf2), and KPC-Cxcl1KO (Cxcl10, Gzmb, Cxcr3, Cxcl9, Cd96, Cd3d, Cd4, Ciita, and H2-Eb1). E, Pie charts showing relative proportions of immune-cell fractions of macrophages (Mϕ), PMN-MDSC, CD4+ T cells, and CD8+ T cells using CIBERSORT immune deconvolution from transcriptomes in KPCEV vs. KPC-Cxcl1KO tumors (n = 3 biological replicates per group). F and G, viSNE contour plots of flow-cytometric immunophenotyping in concatenated single-cell suspensions from KPCEV or KPC-Cxcl1KO orthotopic tumors (left), with adjacent violin plots (right) representing absolute cell counts of PMN-MDSC, CXCR2+ PMN-MDSC, monocytic MDSC (moMDSC), M2-like macrophage (F), CD4+ T cells and CD8+ T cells (G), and central memory T cells, effector memory T cells, degranulating CD8+ T cells (H) from each biological replicate (n = 6–8/group). I, Schematic of experimental design utilizing CD8+ T-cell neutralizing antibodies (CD8neuAb) in C57Bl/6 mice or CD8α−/− transgenic mice (B6.Cd8atm1Mak; CD8KO) for orthotopic injections. J, Representative ultrasound images from mice in each treatment group showing tumor growth dynamics in vivo. K and L, Tumor growth curves (K) and Kaplan–Meier survival estimates (L) from mice across 5 groups in T-cell depletion experiments (n = 10 mice/group), with median survival (MS) of each cohort indicated in parentheses.

Close modal

To better understand the molecular basis for reduced tumor outgrowth in the Cxcl1-silenced model, we performed RNA-seq on 3-week KPCEV and KPC-Cxcl1KO orthotopic tumors (n = 3/group), and subsequent GSEA revealed effects predominantly related to immunoregulatory signaling (Fig. 3C; Supplementary Fig. S6A). KPCEV transcriptomes were expectedly enriched in genes and pathways related to innate immune recruitment/function (i.e., Il6, Csf2, neutrophil degranulation, etc.; Fig. 3C and D), whereas several pathways (Supplementary Fig. S6B) and genes regulating T-cell trafficking and effector activity (i.e., Cxcr3, Cxcl9, Cd96, and Gzmb) were upregulated in KPC-Cxcl1KO tumors (Fig. 3D). Additionally, we observed dysregulation of multiple oncogenic signaling pathways (e.g., TNF via NF-κB, JAK/STAT3, etc.) in KPC-Cxcl1KO tumors (Supplementary Fig. S6C).

Consistent with these gene/pathway-level results, quantification of immune-cell fractions from deconvoluted RNA-seq data revealed attenuation of macrophages and PMN-MDSCs, and significant enrichment in CD4+ and CD8+ T cells in KPC-Cxcl1KO tumors (Fig. 3E). Flow-cytometric immunophenotyping confirmed a significant reduction in tumor-infiltrating CD11b+F4/80LyGhiLy6Cdim PMN-MDSCs—specifically CXCR2+ PMN-MDSC—as well as F4/80+CD206+ M2-like macrophages and F4/80Ly6C+Ly6G monocytic-MDSCs in KPC-Cxcl1KO tumors (Fig. 3F). Notably, we observed a dramatic increase in intratumoral CD4+ and CD8+ T cells (Fig. 3G), as well as expansion in central memory, effector memory, and CD107+ degranulating CD8+ T cells, in KPC-Cxcl1KO tumors (Fig. 3H). This immunologic remodeling was recapitulated by pharmacologic inhibition using anti-Cxcl1–neutralizing antibodies in Ptf1acre/+;LSL-KrasG12D/+;Tgfbr2flox/flox (PKT) mice, confirming significant reductions in PMN-MDSC and M2-like macrophage trafficking as well as increased CD4+/CD8+ T-cell infiltration (Supplementary Fig. S5E).

Cell-Intrinsic Cxcl1 Controls Tumor Growth in a CD8+ T cell–Dependent Manner

To determine if antitumor effects of Cxcl1-silencing are CD8+ T-cell dependent, we performed KPC-Cxcl1KO orthotopic injections in transgenic CD8α−/− (B6.Cd8atm1Mak) mice or in C57/BL6 mice treated with CD8+ T cell–depleting antibodies (n = 10/group, Fig. 3I). Monitoring of in vivo tumor growth dynamics using real-time ultrasound (Fig. 3J) revealed significantly reduced latency of KPC-Cxcl1KO–mediated tumor control in transgenic CD8α−/− and anti-CD8–treated mice compared with vehicle-treated KPC-Cxcl1KO mice; notably, CD8+ T-cell depletion in KPCEV tumor-bearing mice had little impact on the aggressive tumor growth kinetics compared with vehicle-treated KPCEV mice (Fig. 3K). CD8+ T-cell depletion or CD8α silencing significantly shortened survival of KPC-Cxcl1KO mice compared with vehicle-treated KPC-Cxcl1KO controls (median 40 vs. 40 vs. 53 days, respectively; P < 0.0001; Fig. 3L). Therefore, silencing of cell-intrinsic Cxcl1 overcomes T-cell exclusion and restrains tumor growth in a CD8+ T cell–dependent manner in vivo.

Silencing of Tumor-Intrinsic Cxcl1 Reprograms Trafficking Dynamics and Immunosuppressive Potential of PMN-MDSCs

Given the striking reduction in myeloid infiltration into KPC-Cxcl1KO tumors, we interrogated how Cxcl1 silencing regulates real-time MDSC trafficking dynamics using a novel in vivo adoptive transfer system in which splenic PMN-MDSCs from KPCEV or KPC-Cxcl1KO tumor-bearing mice are labeled ex vivo with a sulfo-Cy5.5 maleimide probe—which has nanomolar affinity for surface thiol esters on PMN-MDSCs—and intravenously injected into KPCEV or KPC-Cxcl1KO subcutaneous tumor-bearing mice, with ensuing PMN-MDSC trafficking visualized using IVIS (n = 4/group, Fig. 4A). Following adoptive transfer, labeled PMN-MDSCs derived from KPCEV donor mice exhibited rapid migration to flank tumors in KPCEV recipients, but not to tumor sites in KPC-Cxcl1KO recipients (P = 0.002). Interestingly, PMN-MDSCs derived from KPC-Cxcl1KO donor mice did not traffic to recipient mice harboring Cxcl1-competent KPCEV tumors (P = 0.028; Fig. 4A; Supplementary Fig. S7A), suggesting reprogramming of PMN-MDSCs generated in Cxcl1-silenced tumor-bearing hosts that disrupts their migratory potential despite intact cognate chemokine gradients.

Figure 4.

Silencing of cancer cell–intrinsic Cxcl1 reprograms trafficking dynamics and immunosuppressive potential of tumor-infiltrating PMN-MDSCs. A, Experimental design of adoptive transfer experiments in which splenic neutrophils from donor mice are labeled with sulfur Cy5.5 maleimide, adoptively transferred into recipient flank tumor-bearing mice, and PMN-MDSC trafficking visualized using IVIS (left). Representative IVIS images visualizing trafficked adoptively transferred (A.T.) splenic MDSCs from KPCEV or KPC-Cxcl1KO tumor-bearing mice to subcutaneous KPCEV or KPC-Cxcl1KO tumor-bearing mice, as indicated in the legend (center), and adjacent quantification of total radiant efficiency (TRE) of trafficked Cy5.5-labeled PMN-MDSCs in each group via IVIS (n = 4/group, right). B, Schematic of experimental design (left) with adjacent heat map depicting relative fold change of Arg1, Mpo, and Ido gene expressions via qPCR in Ly6G+ cells isolated from bone marrow (BM), spleen, or tumor sites in KPCEV and KPC-Cxcl1KO orthotopic tumor-bearing mice (right). Gene expression in all other groups is relative to reference expression of genes in intratumoral Ly6G+ cells from KPCEV mice. C, Representative contour plots of arginase-1 (Arg-1) expression via flow cytometry in gated F4/80Ly6GhiLy6Cdim cells from BM, spleen, or tumor sites in KPCEV and KPC-Cxcl1KO orthotopic tumor-bearing mice (left), with an adjacent histogram showing arginase-1 mean fluorescence intensity (MFI) at respective sites in designated mice (n = 6/group). Arg-1 expression in intratumoral PMN-MDSCs is also quantified from KPC-Cxcl1KO mice treated with anti-CD8 neutralizing antibody (CD8neuAb). D, Arg-1 enzymatic activity via colorimetric assay in intratumoral PMN-MDSCs from KPCEV and KPC-Cxcl1KO mice (n = 2–3/group). E, Heat map depicting relative fold change of curated gene module from activated-PMN signature (22) via qPCR from RNA extracted from intratumoral Ly6G+ cells in KPCEV vs. KPC-Cxcl1KO tumors. F, Schematic of experimental design (left), with an adjacent histogram showing IFNγ release (in pg/mL) from CD3/CD28-stimulated T cells alone, or when cocultured (1:3 T-cell:MDSC ratio) in combination with intratumoral PMN-MDSCs from KPCEV and/or KPC-Cxcl1KO tumor-bearing mice (n = 4/group).

Figure 4.

Silencing of cancer cell–intrinsic Cxcl1 reprograms trafficking dynamics and immunosuppressive potential of tumor-infiltrating PMN-MDSCs. A, Experimental design of adoptive transfer experiments in which splenic neutrophils from donor mice are labeled with sulfur Cy5.5 maleimide, adoptively transferred into recipient flank tumor-bearing mice, and PMN-MDSC trafficking visualized using IVIS (left). Representative IVIS images visualizing trafficked adoptively transferred (A.T.) splenic MDSCs from KPCEV or KPC-Cxcl1KO tumor-bearing mice to subcutaneous KPCEV or KPC-Cxcl1KO tumor-bearing mice, as indicated in the legend (center), and adjacent quantification of total radiant efficiency (TRE) of trafficked Cy5.5-labeled PMN-MDSCs in each group via IVIS (n = 4/group, right). B, Schematic of experimental design (left) with adjacent heat map depicting relative fold change of Arg1, Mpo, and Ido gene expressions via qPCR in Ly6G+ cells isolated from bone marrow (BM), spleen, or tumor sites in KPCEV and KPC-Cxcl1KO orthotopic tumor-bearing mice (right). Gene expression in all other groups is relative to reference expression of genes in intratumoral Ly6G+ cells from KPCEV mice. C, Representative contour plots of arginase-1 (Arg-1) expression via flow cytometry in gated F4/80Ly6GhiLy6Cdim cells from BM, spleen, or tumor sites in KPCEV and KPC-Cxcl1KO orthotopic tumor-bearing mice (left), with an adjacent histogram showing arginase-1 mean fluorescence intensity (MFI) at respective sites in designated mice (n = 6/group). Arg-1 expression in intratumoral PMN-MDSCs is also quantified from KPC-Cxcl1KO mice treated with anti-CD8 neutralizing antibody (CD8neuAb). D, Arg-1 enzymatic activity via colorimetric assay in intratumoral PMN-MDSCs from KPCEV and KPC-Cxcl1KO mice (n = 2–3/group). E, Heat map depicting relative fold change of curated gene module from activated-PMN signature (22) via qPCR from RNA extracted from intratumoral Ly6G+ cells in KPCEV vs. KPC-Cxcl1KO tumors. F, Schematic of experimental design (left), with an adjacent histogram showing IFNγ release (in pg/mL) from CD3/CD28-stimulated T cells alone, or when cocultured (1:3 T-cell:MDSC ratio) in combination with intratumoral PMN-MDSCs from KPCEV and/or KPC-Cxcl1KO tumor-bearing mice (n = 4/group).

Close modal

To interrogate if MDSC-induced T-cell suppression is another core function affected by this reprogramming of PMN-MDSCs, we isolated F4/80Ly6Ghi neutrophils from bone marrow, spleens, and orthotopic tumors in KPCEV or KPC-Cxcl1KO mice (n = 3/group), and examined the relative expression of immunosuppressive markers. Tumor-derived PMN-MDSCs from KPC-Cxcl1KO mice demonstrated the most pronounced reduction in Arg1, Ido, and Mpo expression via qPCR (Fig. 4B) and Arg-1 expression via flow cytometry (n = 6/group; Fig. 4C), compared with their tumor-derived counterparts in KPCEV mice, which correlated strongly with differential Cxcl1 secretion gradients (i.e., ΔCxcl1) from lysates in corresponding tumors (P < 0.001; Supplementary Fig. S7B). Notably, reduction of Arg-1 expression in KPC-Cxcl1KO-derived PMN-MDSCs was unaffected by the neutralization of CD8+ T cells in these mice, which mitigates the discrepancy in tumor burden between KPCEV and KPC-Cxcl1KO tumors (refer to Fig. 3J). Colorimetry-based functional assays in intratumoral MDSCs confirmed significant reduction in arginase-1 activity in KPC-Cxcl1KO-infiltrating PMN-MDSCs (P = 0.0087; Fig. 4D; Supplementary Fig. S7C). Finally, compared with KPCEV-infiltrating PMN-MDSCs, transcriptomes of KPC-Cxcl1KO–infiltrating PMN-MDSCs demonstrated striking attenuation in a recently described Cd14hi “activated” gene module (22) associated with potent immunosuppressive activity (Fig. 4E).

In keeping with these findings, we observed rescue of IFNγ release from T cells cocultured with KPC-Cxcl1KO–derived PMN-MDSCs compared with expected T cell–suppressive effects of KPCEV-derived MDSCs (n = 4/group; Fig. 4F). These data indicate that cell-intrinsic Cxcl1 not only modulates trafficking dynamics of PMN-MDSCs but also governs their T cell–suppressive behavior upon arrival to the TME. This reprogramming of MDSC function may underlie the amelioration of T-cell exclusion and antitumor immunity in Cxcl1-silenced tumors.

Neutrophil-Intrinsic TNF Is a Central Regulator of MDSC Function via Cxcl1–CXCR2–MAPK Signaling

We next performed RNA-seq of sorted intratumoral PMN-MDSCs retrieved from KPCEV and KPC-Cxcl1KO orthotopic tumors to identify dominant neutrophil-intrinsic mechanisms that underlie the tumor-restraining and immune-potentiating effects of Cxcl1 silencing (Fig. 5A; Supplementary Fig. S8A). Differential gene expression (Supplementary Fig. S8B) and GSEA revealed strong enrichment in mitogen-activated protein kinase (MAPK) pathway signaling in MDSC-KPCEV relative to MDSC-KPC-Cxcl1KO transcriptomes (Fig. 5A), and Ingenuity Pathway Analysis nominated Tnf as the top upstream regulator of differentially expressed MDSC transcriptomes (Fig. 5B). As such, Tnf was upregulated among canonical MAPK signaling constituents differentially overexpressed in MDSC-KPCEV transcriptomes (Fig. 5C; Supplementary Fig. S8C and S8D).

Figure 5.

Neutrophil-intrinsic TNF is a central regulator of MDSC function via Cxcl1–CXCR2–MAPK signaling. A, Schematic of intratumoral-PMN-MDSC isolation and subsequent RNA-seq from KPCEV and KPC-Cxcl1KO orthotopic tumors 3 weeks post-injection (left). Bubble plot depicts strongest differentially upregulated signaling pathways (using KEGG and Reactome knowledgebases) in PMN-MDSC–infiltrating KPCEV relative to KPC-Cxcl1KO tumors (right), with adjusted P value indicated in legend. B, Histogram representing top 10 predicted upstream regulators of MDSC function comparing KPCEV vs. KPC-Cxcl1KO–derived PMN-MDSCs via Ingenuity Pathway Analysis (IPA), with −log(P value) indicated on the x-axis. C, Volcano plot showing all genes from curated MAPK signaling pathways (using KEGG database) relatively enriched in MDSC-KPCEV vs. MDSC-KPC-Cxcl1KO tumors, with Tnf highlighted. Data are plotted as log(fold change) against −log10P value. D, Putative mechanism of CXCR2–MAPK–TNF signaling cascade in PMN-MDSCs (top), with heat map visualizing relative reduction in Tnf expression upon treatment of J774M PMN-MDSCs with CXCR2 inhibitor AZD5069 (1 μmol/L), IKK inhibitor BAY-110782 (0.1 μmol/L), MAP3K8 inhibitor #871307-18-5 (250 nmol/L), and MEK inhibitor trametinib (250 nmol/L; bottom), with relative fold change indicated in legend. E,In vivo dosing scheme of trametinib treatment in orthotopic KPC mice (top), with violin plot and representative histogram plot from flow cytometry experiments showing TNF mean fluorescence intensity (MFI) in intratumoral CXCR2+ PMN-MDSCs compared between vehicle and MEKi-treated groups (n = 5/group). F, Dot plot showing relative TNF gene expression across cell clusters in scRNA-seq data from human PDAC patients (n = 16; ref. 15), PKT genetically engineered mouse model (GEMM; ref. 16), and KPC GEMM (29). G, Plot and adjacent histogram plots showing TNF mean fluorescence intensity (MFI) via flow cytometry in peripheral blood mononuclear cells (PBMC) retrieved from treatment-naive PDAC patients at the University of Miami (n = 57). H, Bubble plot highlighting top 5 downregulated oncogenic signaling pathways (KEGG) in KPC-Cxcl1KO compared with KPCEV whole-tumor transcriptomes, with TNF signaling bolded. I, Violin plots showing TNF levels (in pg/mL) using ELISA in whole-tumor lysates from KPCEV vs. KPC-Cxcl1KO (left), and vehicle-treated or CXCR2i AZD5069-treated orthotopic KPC mice (right; n = 5/group). J, Immunofluorescence using confocal microscopy showing Ly6G (red) and TNF (green) expression in polylysine coated cover slip-mounted intratumoral Ly6G+ cells from KPCEV, KPC-Cxcl1KO, KPCEV vehicle-treated, and KPCEV CXCR2i-treated tumor-bearing orthotopic mice.

Figure 5.

Neutrophil-intrinsic TNF is a central regulator of MDSC function via Cxcl1–CXCR2–MAPK signaling. A, Schematic of intratumoral-PMN-MDSC isolation and subsequent RNA-seq from KPCEV and KPC-Cxcl1KO orthotopic tumors 3 weeks post-injection (left). Bubble plot depicts strongest differentially upregulated signaling pathways (using KEGG and Reactome knowledgebases) in PMN-MDSC–infiltrating KPCEV relative to KPC-Cxcl1KO tumors (right), with adjusted P value indicated in legend. B, Histogram representing top 10 predicted upstream regulators of MDSC function comparing KPCEV vs. KPC-Cxcl1KO–derived PMN-MDSCs via Ingenuity Pathway Analysis (IPA), with −log(P value) indicated on the x-axis. C, Volcano plot showing all genes from curated MAPK signaling pathways (using KEGG database) relatively enriched in MDSC-KPCEV vs. MDSC-KPC-Cxcl1KO tumors, with Tnf highlighted. Data are plotted as log(fold change) against −log10P value. D, Putative mechanism of CXCR2–MAPK–TNF signaling cascade in PMN-MDSCs (top), with heat map visualizing relative reduction in Tnf expression upon treatment of J774M PMN-MDSCs with CXCR2 inhibitor AZD5069 (1 μmol/L), IKK inhibitor BAY-110782 (0.1 μmol/L), MAP3K8 inhibitor #871307-18-5 (250 nmol/L), and MEK inhibitor trametinib (250 nmol/L; bottom), with relative fold change indicated in legend. E,In vivo dosing scheme of trametinib treatment in orthotopic KPC mice (top), with violin plot and representative histogram plot from flow cytometry experiments showing TNF mean fluorescence intensity (MFI) in intratumoral CXCR2+ PMN-MDSCs compared between vehicle and MEKi-treated groups (n = 5/group). F, Dot plot showing relative TNF gene expression across cell clusters in scRNA-seq data from human PDAC patients (n = 16; ref. 15), PKT genetically engineered mouse model (GEMM; ref. 16), and KPC GEMM (29). G, Plot and adjacent histogram plots showing TNF mean fluorescence intensity (MFI) via flow cytometry in peripheral blood mononuclear cells (PBMC) retrieved from treatment-naive PDAC patients at the University of Miami (n = 57). H, Bubble plot highlighting top 5 downregulated oncogenic signaling pathways (KEGG) in KPC-Cxcl1KO compared with KPCEV whole-tumor transcriptomes, with TNF signaling bolded. I, Violin plots showing TNF levels (in pg/mL) using ELISA in whole-tumor lysates from KPCEV vs. KPC-Cxcl1KO (left), and vehicle-treated or CXCR2i AZD5069-treated orthotopic KPC mice (right; n = 5/group). J, Immunofluorescence using confocal microscopy showing Ly6G (red) and TNF (green) expression in polylysine coated cover slip-mounted intratumoral Ly6G+ cells from KPCEV, KPC-Cxcl1KO, KPCEV vehicle-treated, and KPCEV CXCR2i-treated tumor-bearing orthotopic mice.

Close modal

Next, we explored the signaling link between CXCR2 ligation, MAPK pathway activation, and Tnf production in PMN-MDSCs. scRNA-seq data from Kras–Trp53 Panc02 (23) and PKT (16) models revealed robust coexpression of Cxcr2 and Tnf in single-cell PMN-MDSC transcriptomes (Supplementary Fig. S9A and S9B), as well as differential enrichment of MAPK signaling in Tnfhi compared with Tnflo PMN-MDSC single-cell transcriptomes (Supplementary Fig. S9C and S9D). Constructing STRING knowledgebase-annotated signaling interactomes within Cxcr2–Tnf coexpressing transcriptomes revealed a putative Cxclx–CXCR2–Map3k8–Tnf signaling node (Supplementary Fig. S9E). Indeed, the IKK–Map3k8 complex has been previously shown to regulate innate inflammatory signaling via MEK/ERK activation downstream of G-protein–coupled receptor (e.g., PAR1) ligation (24). Preconditioning of murine CXCR2hi neutrophilic MDSC-like J774M cells (25) with recombinant Cxcl1, and treatment with CXCR2i AZD5069, IKKi BAY-110782 (26), MAP3K8i (27), and MEKi trametinib significantly reduced Tnf gene expression (>50% in all, P < 0.05; Fig. 5D). To validate the dependency of MDSC-intrinsic TNF on CXCR2–MAPK signaling in vivo, we observed disproportionate reduction in TNF expression in CXCR2hi (P = 0.04; Fig. 5E), but not CXCR2lo (Supplementary Fig. S9F), intratumoral PMN-MDSCs via flow cytometry in orthotopic KPC mice treated with MEKi trametinib (n = 5/group).

PMN-MDSCs Are the Dominant Source of TNF in PDAC

Given the unexpected emergence of MDSC-intrinsic TNF as a putative regulator of tumor-permissive MDSC function, we asked if PMN-MDSC–derived TNF was biologically relevant in human and murine PDAC. Examination of scRNA-seq data from PDAC patients (15), PKT (16), or KPC GEMM mice (28) revealed that, compared with all other cellular constituents, PMN-MDSC transcriptomes exhibited the highest TNF/Tnf expression (Fig. 5F). Moreover, using peripheral blood mononuclear cells retrieved from treatment-naive PDAC patients at the University of Miami (n = 57), we observed highest TNF expression in LinCD11b+CD14CD15+ PMN-MDSCs compared with all other circulating immune populations (i.e., CD4+/CD8+ T cells, NK, NKT, or B cells; Fig. 5G).

Disruption of Cxcl1–CXCR2 Engagement Abolishes MDSC-Restricted TNF Production to Attenuate Global TNF Signaling in the TME

Given these findings, we hypothesized that disrupting the Cxcl1–CXCR2 axis would attenuate tumor-wide TNF signaling despite augmenting antitumor T-cell immunity. KPC-Cxcl1KO tumor transcriptomes revealed robust downregulation of TNF signaling pathways compared with KPCEV transcriptomes (P-adj < 0.01; Fig. 5H; Supplementary Fig. S6C). More­over, we observed a significant reduction in TNF production from whole-tumor lysates in KPC-Cxcl1KO (P = 0.02, n = 4) and CXCR2i AZD5069-treated (P = 0.009, n = 5; Fig. 5I) orthotopic KPC mice. In KPC-Cxcl1KO and CXCR2i-treated mice, abolition of intratumoral neutrophil-restricted TNF expression via confocal microscopy (Fig. 5J) is associated with global reduction in intratumoral TNF signaling. Altogether, these data reveal a previously unrecognized paradox in which inhibiting Cxcl1–CXCR2 engagement invigorates T-cell activation despite dampening tumor-wide TNF signaling, predominantly via disruption of MDSC-restricted TNF.

PMN-MDSC–Derived TNF Sustains Tolerogenic Circuitries via tmTNF–TNFR2 Signaling

We have recently reported TNF as a regulator of innate immunoregulatory transcriptional networks in Ras-p53 PDAC (6). To build on these and current findings, we asked if such effects are driven by MDSC-derived TNF. Coculture of J774M PMN-MDSCs (Supplementary Fig. S10A) with KPC tumor cells or CAFs (Supplementary Fig. S10B) resulted in further >5-fold increase in Cxcl1 gene expression/secretion from same-well, but not transwell, cocultures (Supplementary Fig. S11A and S11B). This dependence of proinflammatory signaling on cell–cell contact between PMN-MDSCs and tumor cell/CAFs suggested a putative juxtacrine role for tmTNF–TNFR2—as opposed to paracrine soluble TNF (sTNF)–TNFR1—signaling in driving these effects. Indeed, although not constitutively expressed on CAF/tumor cells (Supplementary Fig. S11C), TNFR2 expression in CAFs was reciprocally upregulated whereas TNFR1 downregulated in a time-dependent manner upon coculture with J774M cells (Supplementary Fig. S11D).

Next, compared with baseline expression, cell-intrinsic Cxcl1 expression was upregulated >4-fold when coculturing KPC tumor cells (Fig. 6A) or CAFs (Fig. 6B) with intratumoral PMN-MDSCs derived from orthotopic KPC mice, but significantly reduced when intratumoral PMN-MDSCs were preconditioned with either etanercept (>70%) or CXCR2i AZD5069 (>60%) and utilized in ex vivo cocultures (P < 0.001). Notably, tumor cell/CAF-intrinsic Cxcl1 overexpression was unaffected by incubating MDSC-tumor cell or MDSC-CAF cocultures with selective soluble TNF inhibitor infliximab (ref. 29; Fig. 6A and B, respectively).

Figure 6.

PMN-MDSC–derived TNF sustains innate immunoregulatory and tolerogenic circuitry in the pancreatic tumor microenvironment. A and B, Experimental design of ex vivo cocultures of intratumoral PMN-MDSCs with KPC tumor cells or KPC cancer-associated fibroblasts (CAF; left). Histograms depicting relative Cxcl1 expression in KPC tumor cells (A) or KPC CAFs (B) alone, or cocultured with PMN-MDSCs with or without preconditioning with the CXCR2 inhibitor AZD5069 (1 μmol/L), TNFR2 inhibitor etanercept (20 μg/mL), or soluble TNF inhibitor infliximab (sTNFi; 20 μg/mL; right, n = 4/group). C,In vivo dosing scheme of etanercept in orthotopic KPC mice (top), with bar plot showing Cxcl1 production (in pg/mL) by ELISA in whole-tumor lysates from vehicle or etanercept (ETA)-treated mice. D, H&E and Cxcl1 immunostaining in matched tumor sections from the vehicle or ETA-treated mice (both 20×; scale bar, 50 μm), with the inset showing magnified region depicting epithelial-specific staining pattern. Adjacent bar plot shows quantification of %area Cxcl1 staining across biological replicates (n = 5 mice/group, 1 ROI/mouse). E, In the scRNA-seq dataset from PKT genetically engineered mice (15), Circos plot visualizing directionality of the TNF signaling pathway network from donor PMN-MDSC cluster to various cellular clusters (top), with adjoining Circos plot showing top donor ligands from PMN-MDSC (TNF highlighted in purple box) predicted to induce proinflammatory signaling genes/pathways (CXCL highlighted) in recipient tumor cell and CAF clusters. F, Representative contour plots and adjacent violin plots from flow cytometry experiments showing proportions (% of CD45+CD11b+) of PMN-MDSC and (% of CD45+CD11b+F4/80+) M2-like macrophages infiltrating vehicle or etanercept-treated KPC tumors (n = 9–10 mice/group). G, Experimental schematic (top), with representative contour plots showing the proportion of intratumoral IFNγ+ CD3+ T cells, and adjacent bar plot showing the quantification of IFNγ mean fluorescence intensity (MFI) within TCR-β+ T cells across biological replicates in vehicle- and etanercept-treated orthotopic KPC mice (n = 10 mice/group).

Figure 6.

PMN-MDSC–derived TNF sustains innate immunoregulatory and tolerogenic circuitry in the pancreatic tumor microenvironment. A and B, Experimental design of ex vivo cocultures of intratumoral PMN-MDSCs with KPC tumor cells or KPC cancer-associated fibroblasts (CAF; left). Histograms depicting relative Cxcl1 expression in KPC tumor cells (A) or KPC CAFs (B) alone, or cocultured with PMN-MDSCs with or without preconditioning with the CXCR2 inhibitor AZD5069 (1 μmol/L), TNFR2 inhibitor etanercept (20 μg/mL), or soluble TNF inhibitor infliximab (sTNFi; 20 μg/mL; right, n = 4/group). C,In vivo dosing scheme of etanercept in orthotopic KPC mice (top), with bar plot showing Cxcl1 production (in pg/mL) by ELISA in whole-tumor lysates from vehicle or etanercept (ETA)-treated mice. D, H&E and Cxcl1 immunostaining in matched tumor sections from the vehicle or ETA-treated mice (both 20×; scale bar, 50 μm), with the inset showing magnified region depicting epithelial-specific staining pattern. Adjacent bar plot shows quantification of %area Cxcl1 staining across biological replicates (n = 5 mice/group, 1 ROI/mouse). E, In the scRNA-seq dataset from PKT genetically engineered mice (15), Circos plot visualizing directionality of the TNF signaling pathway network from donor PMN-MDSC cluster to various cellular clusters (top), with adjoining Circos plot showing top donor ligands from PMN-MDSC (TNF highlighted in purple box) predicted to induce proinflammatory signaling genes/pathways (CXCL highlighted) in recipient tumor cell and CAF clusters. F, Representative contour plots and adjacent violin plots from flow cytometry experiments showing proportions (% of CD45+CD11b+) of PMN-MDSC and (% of CD45+CD11b+F4/80+) M2-like macrophages infiltrating vehicle or etanercept-treated KPC tumors (n = 9–10 mice/group). G, Experimental schematic (top), with representative contour plots showing the proportion of intratumoral IFNγ+ CD3+ T cells, and adjacent bar plot showing the quantification of IFNγ mean fluorescence intensity (MFI) within TCR-β+ T cells across biological replicates in vehicle- and etanercept-treated orthotopic KPC mice (n = 10 mice/group).

Close modal

Etanercept treatment—compared with vehicle—in orthotopic KPC mice significantly decreased tissue-level Cxcl1 production via ELISA in tumor lysates (Fig. 6C) as well as in epithelial/acinar and stromal compartments by IHC (Fig. 6D). Single-cell network analysis nominated TNF among dominant donor ligands from PMN-MDSCs, which induce proinflammatory signaling in tumor cell and CAF clusters—particularly expression of CXCLx genes—in PKT (Fig. 6E) and KPC (ref. 28; Supplementary Fig. S11E) tumor-derived scRNA-seq datasets.

Systemic TNFR2 Inhibition Remodels an Innate Immune-Enriched TME to Augment T-cell Activation

TNFR2i etanercept treatment in KPC tumor-bearing mice—compared with vehicle—significantly reduced neutrophil trafficking dynamics in adoptive transfer experiments (n = 4; Supplementary Fig. S11F) as well as intratumoral abundance of F4/80Ly6GhiLy6Cdim PMN-MDSCs and F4/80+MHC-IICD163+ M2-like macrophages via flow cytometry (Fig. 6F). Etanercept treatment concurrently augmented intratumoral trafficking of CD3+TCRβ+ T cells (Supplementary Fig. S11G).

We next sought to investigate the effects of tmTNF–TNFR2 inhibition on T-cell activation. Suppression of T-cell IFNγ release following coculture of CD3+ T cells with untreated intratumoral PMN-MDSCs was significantly rescued following coculture with etanercept-preconditioned PMN-MDSCs (n = 4/group, P < 0.001; Supplementary Fig. S11H). To corroborate these findings in vivo, intratumoral CD3+ T cells isolated from etanercept-treated KPC mice demonstrated significantly increased IFNγ+ expression compared with vehicle-treated mice (P = 0.0012; Fig. 6G). Taken together, these data uncover a role for PMN-MDSC–derived tmTNF–TNFR2 signaling in perpetuating feed-forward immunoregulatory circuitries that promote immune tolerance in the PDAC TME.

PMN-MDSC–Derived TNF Promotes Inflammatory CAF Polarization and IL6/STAT3 Signaling

We have recently described the contribution of tumor-associated neutrophils in instigating iCAF polarization and chemoresistant IL6/STAT3 cross-talk in PDAC (30). As such, intratumoral PMN-MDSCs derived from orthotopic KPC tumors induced expression of Il6 ∼40-fold in KPC CAFs (Fig. 7A) as well as other canonical iCAF markers C3, Clec3b, and Lif (Supplementary Fig. S11I), in ex vivo cocultures. This MDSC-induced iCAF gene expression was corroborated in human coculture systems using HL-60 neutrophils and patient-derived CAF PC-13 cells (ref. 31; Supplementary Fig. S11J). In interrogating if MDSC-derived TNF mediates such iCAF polarization, CAF-Il6 expression was significantly reduced when PMN-MDSCs were preconditioned with etanercept (67% Il6 reduction; P < 0.001) or CXCR2i (40% Il6 reduction; P < 0.001). Interestingly, CAF-intrinsic Il6 was induced further when MDSC-CAF cocultures were incubated with soluble-TNFi (Fig. 7A). Etanercept treatment of orthotopic KPC mice confirmed a significant reduction in IL6 levels in whole-tumor lysates (Fig. 7B), and flow-cytometric analysis demonstrated significant attenuation in iCAF (CD45CD31PDPN+Ly6C+MHC-II):myofibroblastic myCAF (CD45CD31PDPN+Ly6CMHC-II) ratios (P = 0.008; Fig. 7C). Etanercept treatment also generated striking remodeling of stromal organization—as evidenced by reduced acidic mucins (Alcian Blue) and fibrotic collagen deposition (Trichrome/Sirius Red; Fig. 7D)—as well as a substantial reduction in myofibroblast (PDPN+αSMA+) populations (Supplementary Fig. S11K).

Figure 7.

Systemic inhibition of TNFR2 mitigates stromal inflammation and sensitizes PDAC to chemotherapy. A, Experimental design of ex vivo cocultures of intratumoral PMN-MDSCs with KPC CAF (left), with histograms depicting relative fold change of Il6 gene expression via qPCR in KPC CAFs alone, or CAFs cocultured with PMN-MDSC with or without preconditioning with CXCR2i AZD5069 (1 μmol/L), TNFR2i etanercept (20 μg/mL), or soluble TNF inhibitor infliximab (sTNFi; 20 μg/mL; n = 4/group). B,In vivo dosing scheme of etanercept in orthotopic KPC mice (top), with the bar plot showing Il6 transcription (left) or protein levels by ELISA in whole-tumor lysates from vehicle- or etanercept-treated mice (n = 6–8 mice/group). C, Representative contour plots showing inflammatory CAF (iCAF; Ly6C+MHC-II), myofibroblastic CAF (myCAF; Ly6CMHC-II) and antigen-presenting CAF (apCAF; Ly6CMHC-II+) populations gated on CD45CD31PDPN+ cells via flow cytometry in vehicle- and etanercept-treated mice, and adjacent bar plot visualizing iCAF/myCAF ratio quantification across biological replicates (n = 6 mice/group). D, Representative H&E showing stromal–tumor ratio via H&E, Alcian Blue, Trichrome, and Sirius Red staining in tumor sections from vehicle- or etanercept-treated mice (scale bar, 50 μm), with adjacent bar plot visualizing respective quantifications across biological replicates (n = 5 mice/group, 1 ROI/mouse). E, Experimental design showing MDSC:CAF coculture groups—labeled 1 through 4—from which conditioned media were generated and incubated with KPC6694c2 tumor cells (top). Bar plot showing IL6 secretion via ELISA in KPC CAFs from experimental conditions 1–4 (bottom left, n = 3). Western blot for pStat3 (tyr-705), total Stat3, and β-actin protein levels in KPC6694c2 tumor cell lysates upon conditioning with media from coculture groups 1–4 (bottom center), with adjacent bar plot showing respective quantification of the pStat3/β-actin ratio (bottom right). ETA, etanercept; VEH, vehicle. F, pSTAT3 immunostaining in tumor sections from vehicle or ETA-treated mice (both 20×; scale bar, 50 μm), with the inset showing the magnified region depicting an epithelial-specific staining pattern. Adjacent bar plot shows the quantification of % area staining for pSTAT3 across biological replicates (n = 5 mice/group). G, Enrichment plot (left) and net plot (right) showing disproportionate downregulation of KEGG_JAK_STAT_SIGNALING_PATHWAY in KPC-Cxcl1KO compared with KPCEV tumor transcriptomes. H,In vivo treatment schedules (top), and Kaplan–Meier survival curves for each of the treatment groups (n = 10 mice/group): vehicle, etanercept (Eta) alone, gemcitabine + paclitaxel (Gem-Pac) alone, and Eta + Gem-Pac. Median survival (MS) of each group is indicated in parentheses.

Figure 7.

Systemic inhibition of TNFR2 mitigates stromal inflammation and sensitizes PDAC to chemotherapy. A, Experimental design of ex vivo cocultures of intratumoral PMN-MDSCs with KPC CAF (left), with histograms depicting relative fold change of Il6 gene expression via qPCR in KPC CAFs alone, or CAFs cocultured with PMN-MDSC with or without preconditioning with CXCR2i AZD5069 (1 μmol/L), TNFR2i etanercept (20 μg/mL), or soluble TNF inhibitor infliximab (sTNFi; 20 μg/mL; n = 4/group). B,In vivo dosing scheme of etanercept in orthotopic KPC mice (top), with the bar plot showing Il6 transcription (left) or protein levels by ELISA in whole-tumor lysates from vehicle- or etanercept-treated mice (n = 6–8 mice/group). C, Representative contour plots showing inflammatory CAF (iCAF; Ly6C+MHC-II), myofibroblastic CAF (myCAF; Ly6CMHC-II) and antigen-presenting CAF (apCAF; Ly6CMHC-II+) populations gated on CD45CD31PDPN+ cells via flow cytometry in vehicle- and etanercept-treated mice, and adjacent bar plot visualizing iCAF/myCAF ratio quantification across biological replicates (n = 6 mice/group). D, Representative H&E showing stromal–tumor ratio via H&E, Alcian Blue, Trichrome, and Sirius Red staining in tumor sections from vehicle- or etanercept-treated mice (scale bar, 50 μm), with adjacent bar plot visualizing respective quantifications across biological replicates (n = 5 mice/group, 1 ROI/mouse). E, Experimental design showing MDSC:CAF coculture groups—labeled 1 through 4—from which conditioned media were generated and incubated with KPC6694c2 tumor cells (top). Bar plot showing IL6 secretion via ELISA in KPC CAFs from experimental conditions 1–4 (bottom left, n = 3). Western blot for pStat3 (tyr-705), total Stat3, and β-actin protein levels in KPC6694c2 tumor cell lysates upon conditioning with media from coculture groups 1–4 (bottom center), with adjacent bar plot showing respective quantification of the pStat3/β-actin ratio (bottom right). ETA, etanercept; VEH, vehicle. F, pSTAT3 immunostaining in tumor sections from vehicle or ETA-treated mice (both 20×; scale bar, 50 μm), with the inset showing the magnified region depicting an epithelial-specific staining pattern. Adjacent bar plot shows the quantification of % area staining for pSTAT3 across biological replicates (n = 5 mice/group). G, Enrichment plot (left) and net plot (right) showing disproportionate downregulation of KEGG_JAK_STAT_SIGNALING_PATHWAY in KPC-Cxcl1KO compared with KPCEV tumor transcriptomes. H,In vivo treatment schedules (top), and Kaplan–Meier survival curves for each of the treatment groups (n = 10 mice/group): vehicle, etanercept (Eta) alone, gemcitabine + paclitaxel (Gem-Pac) alone, and Eta + Gem-Pac. Median survival (MS) of each group is indicated in parentheses.

Close modal

We have previously demonstrated that paracrine activation of tumor cell–STAT3 via CAF-IL6 is a key chemoresistant signaling node (32, 33). Induction of pSTAT3 in KPC6694c2 tumor cells when incubated with conditioned media (CM) from untreated MDSC:CAF cocultures was significantly reduced when incubated with CM from MDSC:CAF cocultures treated with either anti-IL6–neutralizing antibodies (P = 0.017) or when using PMN-MDSCs preconditioned with etanercept (P = 0.01; Fig. 7E). We also observed a significant reduction in epithelial pSTAT3 expression in etanercept-treated KPC tumors via IHC (P < 0.001; Fig. 7F). We then corroborated the relevance of the Cxcl1–CXCR2–TNF axis in regulating tumor-wide STAT3 signaling by showing downregulation of KEGG_IL6_JAK_STAT3_SIGNALING pathway in transcriptomes from KPC-Cxcl1KO—with its incident paucity of TNF-licensed PMN-MDSCs (see Fig. 5J)—compared with KPCEV tumors (Fig. 7G).

Systemic TNFR2 Inhibition Improves Sensitivity to Chemotherapy

As TNFR2i favorably remodels the immune microenvironment and mitigates stromal inflammation, we hypothesized that etanercept would improve chemosensitivity in vivo. In orthotopic KPC models, we observed no improvement in survival with etanercept monotherapy or gemcitabine + paclitaxel chemotherapy alone compared with vehicle treatment (P = ns). However, combining etanercept with gemcitabine + paclitaxel chemotherapy nearly doubled median survival (median 44 vs. 25 vs. 25 days, P < 0.0001; Fig. 7H) and reduced metastatic outgrowth (Supplementary Fig. S12A) compared with single treatments. There was no additional toxicity when TNFR2 inhibition was coadministered with chemotherapy, as determined by mouse weights (Supplementary Fig. S12B). Together, these results uncover a role for TNFR2 inhibition in mitigating stromal inflammation and CAF:tumor cell IL6/STAT3 signaling to sensitize PDAC to chemotherapy.

The present study illustrates a previously unrecognized cancer cell–neutrophil circuit that underlies the intersection between high-risk cancer genotypes (e.g., KRASTP53 cooperativity), tolerogenic immune contextures, and stromal inflammation that defines therapeutic resistance in PDAC. To first unravel the genotype–immunophenotype chasm, we used integrative molecular analysis and single-cell mass cytometry in spatially annotated human tumors to show for the first time that disproportionate overexpression of Cxcl1 in epithelial/tumor cell islands exemplifying such high-risk genomic features dictates spatial exclusion of CD8+ T cells in PDAC. These data not only provide in-human validation of the strong association between tumor cell–intrinsic Cxcl1 expression and T-cell exclusion in KrasTrp53 cooperative KPC murine models (6) but also contribute Cxcl1 to a growing compendium of secreted cancer cell–autonomous factors (e.g., Cxcl5, refs. 34, 35; IL8, ref. 35), which predominantly interact with innate immune populations to thwart antitumor adaptive immunity in solid cancers. In addition, building on recent descriptions of the role of CREB in hyperactivating prometastatic networks in KRASTP53 cooperative PDAC (19) as well as the conflation of KRASTP53 cooperative transcriptomes, p63-mediated squamous transdifferentiation, and inflammatory gene modules in PDAC tumor cells (6, 36), our data highlight CREB activation as a putative master transcriptional regulator of such inflammatory programming from cancer cells and raise the possibility that unifying cell-autonomous mechanisms may drive disparate hallmarks of its aggressive biology.

Although cancer cell–autonomous inflammatory programming is undoubtedly the incipient event in establishing a hostile TME, our data suggest that these tumor cell–derived signals galvanize neutrophils into “activated” functional states, in turn rendering these PMN-MDSCs a dominant hub of extracellular signaling that perpetuate tumor-permissive circuitry in PDAC. As such, these data establish a causal link between cancer cell–autonomous factors and phenotypic plasticity in PMN-MDSCs. The pronounced dampening of T cell–suppressive function as well as “activated” CD14hi signatures in intratumoral PMN-MDSCs derived from Cxcl1-silenced tumors lends support to the hypothesis that CD14hi tumor-resident PMN-MDSCs are likely sculpted by tumor-derived signaling gradients in situ and may not necessarily differentiate from obligate bone marrow–derived precursors (22). Moreover, the shared developmental ontogeny of T-cell-suppressive Cd14hi and Tnfhi neutrophilic states in our recently reported study using intratumoral MDSC single-cell transcriptomes (6) may explain why disrupting the Cxcl1–CXCR2 engagement invigorates adaptive antitumor immunity despite mitigating tumor-wide and neutrophil-restricted TNF signaling in PDAC. This intriguing paradox between heightened antitumor immunity and reduction in TNF signaling also offers contextual clues to help interpret recent data revealing reduced TNFα signaling as the only tumor-specific biomarker of response to gemcitabine, nab-paclitaxel, and nivolumab chemoimmunotherapy in advanced PDAC patients enrolled in the PRINCE trial (37). As such, our findings impel further investigations into understanding how chemo(immuno)therapy treatment trajectories impact the chronologic evolution of proinflammatory transcriptional programming (e.g., senescence-associated secretory phenotype; ref 38) in cancer cells and sculpt progressively tolerogenic TNFhi myeloid ecosystems in PDAC. These efforts may inform discovery of temporally sensitive biomarkers of response/resistance and therapeutically relevant targets at various stages of tumor progression.

The present data not only build on a growing body of evidence implicating tolerogenic and tumor-permissive TNF signaling in solid cancers (39–41) but also reveal novel insights into multiple aspects of TNF immunobiology. First, we were surprised to find that neutrophils are the major source of TNF in human—both in tumor tissue via scRNA-seq and circulation via flow cytometry in treatment-naive patients’ PBMCs—and murine PDAC. Although previous cancer-related studies investigating tolerogenic TNF functions have ascribed its cellular source to the stromal microenvironment rather than malignant cells (42), our data identify the precise cellular source of stromal TNF to recapitulate lessons learned from the infectious disease literature (43), where neutrophil-derived TNF is a dominant mediator of tissue injury and chronic inflammatory signaling. Second, we uncover the direct and indirect effects of neutrophil-derived TNF on T-cell function/activation, leveraging findings from chronic inflammatory models where TNF has been vilified as a master regulator of T-cell dysfunction and exhaustion (44). Although the majority of effects shown here—e.g., as observed in ex vivo coculture experiments with etanercept-primed PMN-MDSCs and T cells—are antigen-agnostic and may be related to activation-induced cell death and/or CD8+ regulatory polarization (CD8reg; refs. 45, 46), further investigation is warranted into whether neutrophil-derived tmTNF mediates antigen-restricted CD8+ T-cell tolerance indirectly via TNFR2hi CD4+ Treg interactions (47). Third, although TNFR2 is constitutively expressed in immune cells, we discover that neutrophilic cellular contact can induce reciprocal TNFR1-to-TNFR2 expression dynamics in PDAC cancer cells and CAFs. Ensuing neutrophil-derived tmTNF–TNFR2 juxtacrine interactions amplify feed-forward proinflammatory cytokine production—particularly Cxcl1—in tumor cells/CAFs (48). These data expose novel redundancies in TNF-driven cell and non–cell-autonomous signaling cascades that sustain immune dysfunction in the PDAC TME, and emphasize the need for thoughtful tmTNF–TNFR2 targeting strategies (49) to optimally unleash antitumor immunity in PDAC patients.

These data also establish—for the first time to our knowledge—an instructive role for neutrophil-derived tmTNF–TNFR2 signaling in regulating stromal inflammation, proinflammatory CAF-derived Cxcl1 and IL6 secretion, and chemoresistant CAF:tumor cell IL6/STAT3 signaling in PDAC. As such, TNFR2 signaling is enriched in tumor transcriptomes of advanced PDAC patients enrolled in the COMPASS trial (10) demonstrating chemoresistant (i.e., progressive disease) compared with chemosensitive (i.e., stable disease and/or partial response) disease (Supplementary Fig. S12C). Fascinatingly, the relationship between tmTNF–TNFR2 signaling and CAF-derived IL6 may offer a mechanistic explanation for the striking reduction in circulating IL6 levels observed in metastatic breast cancer patients treated with etanercept in a nonrandomized phase II trial nearly 2 decades ago (50). Our findings can also be reconciled with the canonical model of iCAF polarization in which CAF-derived IL6 (and Cxcl1, LIF, etc.) was predominantly driven by tumor cell–secreted IL1α using patient-derived tumor cell/CAF organoid models (51). In such experimental models, the lack of immune-cell reconstitution underestimates the contribution of innate immune cells in dictating fibroblast reprogramming. Although acknowledging the undoubted contribution of tumor cell–derived IL1α–IL1R1 signaling to iCAF skewness, the role of myeloid cell TNF-mediated effects on CAF plasticity described herein may explain the observation that silencing Il1a in tumor cells was insufficient to abolish iCAF polarization in vivo. Moreover, although recombinant TNF was sufficient to polarize iCAFs in ex vivo conditioning experiments, neutralization of soluble TNF did not impair induction of iCAF markers in pancreatic stellate cells cocultured with organoid-conditioned media (51). These latter observations can now also be explained by our data: tumor cells are not the dominant source of TNF in human or murine PDAC, and the failure to rescue inflammatory CAF polarization in experiments where soluble TNF is neutralized underestimates the preferential dependence of iCAF determination on tmTNF–TNFR2 versus sTNF–TNFR1 signaling—findings reflected in the dramatically divergent Il6 expression in CAFs cocultured with intratumoral PMN-MDSCs incubated with either TNFR2i or soluble TNFi. Notwithstanding, the incomplete suppression of CAF-Il6 expression following tmTNF–TNFR2 inhibition in these MDSC–CAF cocultures suggests additional mechanisms by which neutrophil-derived signaling governs iCAF polarization (e.g., PMN-MDSC-intrinsic IL1β; ref. 30). Although we cannot yet estimate the relative “dosages” of neutrophilic or tumor cell contributions to inflammatory programming and transcriptional heterogeneity in PDAC CAFs, our data emphasize the need for a multipronged strategy to restrain both proinflammatory TNFR2 and IL1 signaling to optimally subvert stromal inflammation and its tumor-permissive consequences in PDAC.

In summary, the present study “connects the dots” from high-risk cancer genotypes to unifying cell-autonomous mechanisms (i.e., CREB activation) regulating proinflammatory chemokines (i.e., Cxcl1) that dictate functional neutrophil plasticity to enforce T-cell exclusion in the PDAC TME, ultimately implicating an interwoven cancer cell–neutrophil Cxcl1–TNF circuit as a central regulator of stromal inflammation and T-cell dysfunction. Although we utilized KRASTP53 cooperativity—emblematic of canonical “high-risk” PDAC genomes (6)—as a springboard to uncover this Cxcl1–tmTNF–TNFR2 circuitry, the applicability of the proposed signaling axis in PKT mice and in heterogeneous patient-derived datasets underscore its broader relevance in PDAC biology. Moreover, our findings conceptually challenge the dated model of exclusively antitumor TNF signaling in which global TNF silencing incapacitates T cell–mediated immunologic control of tumors (52) and illustrates a more nuanced context-dependent influence of tmTNF–TNR2 signaling in mediating protumorigenic effects in PDAC. Indeed, ongoing studies are examining if conditional amplification of neutrophil-restricted tmTNF recapitulates stromal and T-cell dysfunction in the PDAC TME. Moreover, a shift in our heuristic framework toward a contextually sensitive tumor-permissive TNF signaling model in cancer may also pave the way for novel bioengineered TNF-targeting immunotherapy design—e.g., strategies to intercept the competitive advantage of tolerogenic CD14hiTNFhi PMN-MDSC developmental states in the TME, or neutrophil-directed nanotargeting approaches to interrupt the transmembrane quarantining of TNF. Leveraging such novel strategies to reprogram the dynamic immunoregulatory control exerted by MDSC-derived TNF in the PDAC TME is of paramount importance as they can mitigate inflammatory stromal signaling and T-cell dysfunction to overcome therapeutic resistance in patients with PDAC.

Transcriptomic Analysis in the CCLE Dataset

Genomic/transcriptomic sequencing data normalized in log2(RSEM+1) and clinical information (ver. 21Q4) were retrieved from CCLE (https://depmap.org/portal/ccle) and stratified into KRAS–TP53 coaltered (n = 23) and KRAS-altered/TP53WT (n = 5), as described previously (6). Comparative GSEA was performed using R package limma (ref. 53; Supplementary Table S1). P values were adjusted for multiple hypothesis testing based on the Benjamini–Hochberg procedure (54).

Spatial Annotation of Human PDAC Tumors via IMC

Eight patients with resectable or borderline resectable PDAC who underwent upfront surgery at our institution were identified, archived FFPE blocks from resection specimens were retrieved, and post hoc correlation with next-generation sequencing data from the medical record allowed genomic annotation into KRAS–TP53 coaltered (n = 5) versus KRAS-altered/TP53WT (n = 3). For detailed clinicogenomic data, see Supplementary Table S2. One section was stained with hematoxylin and eosin (H&E) to enable a board-certified GI pathologist (E.A. Montgomery) to select an ROI comprising tumor cells, fibroblasts, and immune cells from each unique tumor sample. A second section was stained with an IMC panel of 10 metal-conjugated antibodies and a cell intercalator (Supplementary Table S4), and the corresponding ROI was marked, and laser ablated using the Hyperion system (Fluidigm Inc.). Data were exported as MCD files and analyzed for single-cell segmentation, t-SNE clustering, and cellular neighborhood analysis using Visiopharm software. For details, refer to Supplementary Methods.

Human and Mouse scRNA-seq and Differential Gene Expression/Pathway Analysis

snRNA-seq data from 43 patients were obtained from the DUOS repository (ID 000139; ref. 13). scRNA-seq data from 16 human PDAC patients and from 4 KPC GEMM mice were retrieved from the NIH Gene Expression Omnibus database (GSE155698 and GSE129455, respectively), and computational inference of cluster-specific ligand–receptor interactions was performed using the NicheNet algorithm (17) and CellChat (55), as described in Supplementary Methods. scRNA-seq of treatment-naive tumors from 6.5-week-old PKT mice was performed as described previously (16) and detailed in Supplementary Methods.

Cell Lines, CRISPR/Cas9 Genetic Editing, and Plasmid Transfections

Parent LSL-KrasG12D/+;Trp53R172H/+;Pdx1Cre/+ (KPC-6694c2) cells were transduced with CRISPR vector lentiCRISPR v2 with stable Cas9 expression (Addgene plasmid #52961; ref. 56); nontargeting sgRNA infected cells designated as KPCEV), and deletion efficiency of Cxcl1 (KPC-Cxcl1KO) was confirmed as described previously (14). In orthotopic models, the immunogenicity of parental KPC-6694c2 and KPCEV cells did not differ significantly.

Tissue-resident cancer-associated fibroblasts (CAF) from 6-month-old LSL-KrasG12D/+;Trp53R172H/+;Pdx-1Cre/+ (KPC) mice were isolated as previously described (9), and limited characterization with qPCR confirmed exclusive expression of fibroblast lineage markers Pdgfra and Col1a1 compared with neutrophil-like J774M and KPC-6694c2 cells (Supplementary Fig. S10B). Human PDAC cell lines Mia-PaCa2 (CRL-1420), Panc02 (CRL-2553), Capan-1 (HTB-79), Hs766T (HTB-134), PDM-168 (HCM-BROD-0229-C25), and HL-60 (CCL-240) were purchased from ATCC. J774M cells were kindly gifted by Dr. Evanthia Torres/University of Southern California. Human hPNE-KRASWT and hPNE-KRASG12D cell lines were obtained from ATCC and maintained according to established guidelines. Mutant TP53R175H and/or TP53WT cDNA constructs were transiently overexpressed in hPNE-KRASWT or hPNE-KRASG12D cells as described previously (57). All cell lines were regularly tested using MycoAlert Mycoplasma Detection Kit (Lonza); for details, refer to Supplementary Methods.

Heterocellular Coculture Experiments

KPC CAFs or KPCEV tumor cells were cocultured with either ex vivo–isolated Ly6G+ cells and/or J774M cells and conditioning media and RNA from cocultured KPC CAFs or KPCEV tumor cells was collected. For details, refer to Supplementary Methods.

ChIP-seq and ChIP-qPCR

Preparation of cross-linked RNA-free chromatin, sonication, and immunoprecipitation protocols using Creb (Cell Signaling Technology, #9197S) and RNApol-II (Cell Signaling Technology, #14958) antibodies in KPC 6694c2 tumor cells are described in Supplementary Methods. DNA was quantified with Qubit HS kit (Invitrogen), library preparation was performed using NEB Next Ultra DNA Library Prep Kit for Illumina (New England Biolabs, #E7370) following manufacturer instructions, and sequenced on a NextSeq6000 (Illumina). Fastq files were processed and analyzed as described previously (58), and detailed in Supplementary Methods. ChIP-qPCR was performed in KPC 6694c2 and KPC K-8484 tumor cells. For details on ChIP-qPCR for validation of ChIP-seq data and specific primers used, refer to Supplementary Methods.

Animal Models

All animal experiments were performed in accordance with the NIH animal use guideline and protocol (21–057) approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Miami. C57BL/6 and Cd8atm1Mak(CD8α−/−) 6- to 8-week-old female mice were purchased from The Jackson Laboratory. For orthotopic injection models using KPCEV or KPC-Cxcl1KO cells in these mice, refer to Supplementary Methods. PKT mice were generated as previously described (59).

RNA-seq from Whole Tumors and Intratumoral PMN-MDSCs with Downstream Analysis

Purified RNA was obtained from orthotopically injected KPCEV or KPC-Cxcl1KO whole tumors (following 21-day growth in vivo), as well as intratumoral F4/80Ly6Ghi PMN-MDSCs (obtained after magnetic column separation from single-cell suspensions of whole tumors per manufacturer protocol (Miltenyi Biotec, #130-094-538) in biological triplicates using RNeasy Kit (Qiagen). RNA quality was assessed on a bioanalyzer using the RNA6000 Nano kit (5067-1511; Agilent Technologies); 0.2–1 μg of RNA with RIN>7 proceeded for library preparation using Illumina TruSeq kit. Ensuing libraries were sequenced using Illumina NovaSeq SP300. Reads from RNA-seq were mapped to Mus musculus genome (GRCm38) using STAR (v.2.5.0) aligner. Raw counts were generated based on Ensemble genes (GENCODE M13) with feature Counts (v.1.5.0). DE genes were identified using ESeq2 and determined by a threshold of FDR-corrected P < 0.05. GSEA/pathway analysis was performed using Canonical Pathway, KEGG, REACTOME, Hallmark, Gene Ontology, and Immunologic signature gene sets retrieved from MSigDB/ImmuneSigDB databases.

Isolation of Ly6G+ Cells from Tumors and Spleens

Single-cell suspensions from whole pancreas or spleens obtained from KPCEV or KPC-Cxcl1KO orthotopic tumor-bearing mice were subjected to magnetic column separation, as described previously (6).

Flow Cytometry

Single-cell suspensions from mouse tumors, spleens, and bone marrows from animals (as indicated), or PBMCs from chemotherapy-­naive PDAC patients (n = 57) were thawed, washed, incubated with FcR-blocking reagent (Miltenyi Biotec), and subsequently stained with fluorescently conjugated antibodies (Supplementary Table S5). Flow cytometry data acquisition was performed on Cytek Aurora and analyzed using FlowJo v.10 software. For details of all tissue processing and flow cytometry procedures, refer to Supplementary Methods.

MDSC Labeling with Sulfo-Cy5.5-Maleimide Dye and In Vivo Adoptive Transfer

Cell-surface thiol moieties on splenic Ly6G+ cells from KPCEV or KPC-Cxcl1KO mice were quantified using Ellman's reagent (Thermo Fisher #22582) per manufacturer protocol. Cells were washed with PBS and then resuspended at 1 × 106 cells/mL, incubated with 10 μg of sulfo-Cy5.5-maleimide dye (Lumiprobe, #17380) at 4°C for 30 minutes. To remove unreacted dye content, cells were washed 3× times with PBS, maintained at 4°C, and used for adoptive transfer through tail-vein injections in flank subcutaneous tumor-bearing animals. Twenty-four hours following injection, mice were temporarily anesthetized using isoflurane and imaged using the IVIS Spectrum optical imaging system (PerkinElmer).

T-cell Suppression Assays

Ly6G+ cells isolated from tumor-bearing mice were cocultured with stimulated T cells, and T-cell activity was measured via IFNγ secretion. T cells were isolated from spleens of tumor-naive C57BL/6 mice using a magnetic column PanT cell isolation kit as per the manufacturer's recommendation (Miltenyi Biotec, #130-095-130). T cells were cocultured with Ly6G+ cells at varying ratios (1:1–1:8 T-cell:Ly6G) and anti-CD3/CD28 beads (Thermo Fisher Scientific) at a bead-to-T-cell ratio of 1:1. After 48 hours, condition media were collected for IFNγ measurement by ELISA.

Confocal Microscopy in Intratumoral PMN-MDSCs

Ly6G+ cells were isolated from the orthotopic tumor as previously described (6, 16, 28). 5 × 105–1 × 106 cells were then seeded in a 24-well plate on Poly-L-lysine coated 12-mm coverslips (#354085, Corning) for 2 hours. Ly6G+ cells were then costained as per the immunofluorescence protocol described in Supplementary Methods.

RNA In Situ Hybridization

Pancreatic tissues were first processed for immunofluorescence analysis as previously described (6, 16, 28). Cxcl1 mRNA detection and amplification were performed on the same pancreatic tissue slides using HCR IHC + HCR RNA–in situ hybridization protocol (Mole­cular Instruments). For details, refer to Supplementary Methods.

Histopathologic Analysis, Western Blotting, QPCR, ELISA, and Cytokine Array

Histologic analysis, qPCR analysis, and Western blotting procedures were performed as described previously (60). Antibodies used for Western blotting are tabulated in Supplementary Table S6. Primers used for qPCR analysis are tabulated in Supplementary Table S7. For details, refer to Supplementary Methods.

Ethics Reporting

Archival tissue acquisition was performed in accordance with protocols approved by the Institutional Review Board of the University of Miami (considered exempt). All studies were conducted in accordance with ethical guidelines outlined in the Declaration of Helsinki.

Data Availability

Bulk RNA-seq data from KPCEV and KPC-Cxcl1KO whole tumors and sorted Ly6G+ intratumoral PMN-MDSCs isolated from KPCEV and KPC-Cxcl1KO tumors are available at BioProject #PRJNA926682 and #PRJNA926681, respectively. All other data generated in this study are available upon request from the corresponding author J. Datta ([email protected]) and/or first author A. Bianchi ([email protected]).

W.L. Hwang reports a patent for no. 63/069,035 pending and a patent for no. 63/346,670 pending. R. Brambilla reports grants from the NIH/National Institute of Neurological Disorders and Stroke and the Italian Multiple Sclerosis Foundation outside the submitted work. K. Komanduri reports other support from Aegle Therapeutics and personal fees from Avacta Therapeutics, Kite/Gilead, Janssen, Bristol Myers Squibb, Genentech/Roche, Cargo Therapeutics, and Iovance outside the submitted work. E. Toska reports grants from AstraZeneca outside the submitted work. B.Z. Stanger reports grants from Boehringer Ingelheim, Revolution Medicine, and iTeos Therapeutics outside the submitted work. D.I. Gabrilovich reports other support from AstraZeneca during the conduct of the study, as well as other support from AstraZeneca outside the submitted work. No disclosures were reported by the other authors.

A. Bianchi: Conceptualization, data curation, software, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. I. De Castro Silva: Data curation, software, formal analysis, investigation, methodology, writing–review and editing. N.U. Deshpande: Formal analysis, validation, investigation, methodology, writing–review and editing. S. Singh: Data curation, formal analysis, validation, investigation, methodology, writing–review and editing. S. Mehra: Data curation, validation, investigation, writing–review and editing. V.T. Garrido: Data curation, investigation, visualization, writing–review and editing. X. Guo: Software, formal analysis, visualization, methodology, writing–review and editing. L.A. Nivelo: Data curation, software, visualization, writing–review and editing. D.S. Kolonias: Software, formal analysis, validation, investigation, writing–review and editing. S.J. Saigh: Data curation, investigation, visualization, methodology, writing–review and editing. E. Wieder: Resources. C.I. Rafie: Data curation, investigation, writing–review and editing. A.R. Dosch: Data curation, validation, investigation, visualization, methodology, writing–review and editing. Z. Zhou: Data curation, investigation, writing–review and editing. O. Umland: Data curation, software, formal analysis, investigation, visualization, methodology, writing–review and editing. H. Amirian: Methodology. I.C. Ogobuiro: Data curation, investigation, visualization, writing–review and editing. J. Zhang: Data curation, formal analysis, validation, investigation, visualization, writing–review and editing. Y. Ban: Data curation, software, formal analysis, visualization, writing–review and editing. C. Shiau: Data curation. N.S. Nagathihalli: Resources, investigation, methodology, writing–review and editing. E.A. Montgomery: Data curation, software, formal analysis, methodology, writing–review and editing. W.L. Hwang: Data curation. R. Brambilla: Resources, data curation, validation, investigation, methodology, writing–review and editing. K. Komanduri: Resources, software, formal analysis, supervision, methodology, writing–review and editing. A.V. Villarino: Data curation, formal analysis, investigation, visualization, methodology, writing–review and editing. E. Toska: Data curation, software, formal analysis, investigation, visualization, methodology, writing–review and editing. B.Z. Stanger: Resources, supervision, investigation, methodology, writing–review and editing. D.I. Gabrilovich: Data curation, formal analysis, investigation, methodology, writing–review and editing. N.B. Merchant: Resources, supervision, funding acquisition, investigation, writing–review and editing. J. Datta: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This work was supported by a KL2 career development grant by the Miami Clinical and Translational Science Institute (CTSI) under NIH Award UL1TR002736, American College of Surgeons Franklin H. Martin Research Fellowship, Association for Academic Surgery Joel J. Roslyn Faculty Award, Society of Surgical Oncology Young Investigator Award, Elsa U. Pardee Foundation Award, and Pancreatic Cancer Action Network Career Development Award (to J. Datta); NIH R01 CA161976 (to N.B. Merchant). Research reported in this publication was supported by the NCI/NIH Award P30CA240139. We wish to thank Dr. Stephen Nimer for his mentorship and critical review of the manuscript, Dr. Wael El-Rifai for his generous mentorship and advice, Dr. Emiliano Cocco for assisting with research networking/collaborations, and Mr. Andrew Adams for technical assistance with ChIP-seq procedures. We wish to acknowledge the following Shared Resources at Sylvester Comprehensive Cancer Center for their able assistance of this project: Biospecimen (BSSR)—in particular Melinda Boone, Elena Cortizas, and Daniel Tran; Flow Cytometry (FSCR); Analytic Imaging (AISR); and Oncogenomics (OGSR)—in particular Dr. Sion Williams.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

Note: Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).

1.
Kleeff
J
,
Korc
M
,
Apte
M
,
La Vecchia
C
,
Johnson
CD
,
Biankin
AV
, et al
.
Pancreatic cancer
.
Nat Rev Dis Primers
2016
;
2
:
16022
.
2.
Yu
S
,
Zhang
C
,
Xie
KP
.
Therapeutic resistance of pancreatic cancer: roadmap to its reversal
.
Biochim Biophys Acta Rev Cancer
2021
;
1875
:
188461
.
3.
Maddalena
M
,
Mallel
G
,
Nataraj
NB
,
Shreberk-Shaked
M
,
Hassin
O
,
Mukherjee
S
, et al
.
TP53 missense mutations in PDAC are associated with enhanced fibrosis and an immunosuppressive microenvironment
.
Proc Natl Acad Sci U S A
2021
;
118
:
e2025631118
.
4.
Trovato
R
,
Fiore
A
,
Sartori
S
,
Cane
S
,
Giugno
R
,
Cascione
L
, et al
.
Immunosuppression by monocytic myeloid-derived suppressor cells in patients with pancreatic ductal carcinoma is orchestrated by STAT3
.
J Immunother Cancer
2019
;
7
:
255
.
5.
Banerjee
K
,
Kumar
S
,
Ross
KA
,
Gautam
S
,
Poelaert
B
,
Nasser
MW
, et al
.
Emerging trends in the immunotherapy of pancreatic cancer
.
Cancer Lett
2018
;
417
:
35
46
.
6.
Datta
J
,
Bianchi
A
,
De Castro Silva
I
,
Deshpande
NU
,
Cao
LL
,
Mehra
S
, et al
.
Distinct mechanisms of innate and adaptive immune regulation underlie poor oncologic outcomes associated with KRAS-TP53 co-alteration in pancreatic cancer
.
Oncogene
2022
;
41
:
3640
54
.
7.
Beatty
GL
,
Winograd
R
,
Evans
RA
,
Long
KB
,
Luque
SL
,
Lee
JW
, et al
.
Exclusion of T cells from pancreatic carcinomas in mice is regulated by Ly6C(low) F4/80(+) extratumoral macrophages
.
Gastroenterology
2015
;
149
:
201
10
.
8.
Ohlund
D
,
Handly-Santana
A
,
Biffi
G
,
Elyada
E
,
Almeida
AS
,
Ponz-Sarvise
M
, et al
.
Distinct populations of inflammatory fibroblasts and myofibroblasts in pancreatic cancer
.
J Exp Med
2017
;
214
:
579
96
.
9.
Dosch
AR
,
Singh
S
,
Dai
X
,
Mehra
S
,
Silva
IC
,
Bianchi
A
, et al
.
Targeting tumor-stromal IL6/STAT3 signaling through IL1 receptor inhibition in pancreatic cancer
.
Mol Cancer Ther
2021
;
20
:
2280
90
.
10.
Aung
KL
,
Fischer
SE
,
Denroche
RE
,
Jang
GH
,
Dodd
A
,
Creighton
S
, et al
.
Genomics-driven precision medicine for advanced pancreatic cancer: early results from the COMPASS trial
.
Clin Cancer Res
2018
;
24
:
1344
54
.
11.
Dosch
AR
,
Chatila
WK
,
Ban
Y
,
Bianchi
A
,
Deshpande
NU
,
Silva
IC
, et al
.
Ras-p53 genomic cooperativity as a model to investigate mechanisms of innate immune regulation in gastrointestinal cancers
.
Oncotarget
2021
;
12
:
2104
10
.
12.
Barretina
J
,
Caponigro
G
,
Stransky
N
,
Venkatesan
K
,
Margolin
AA
,
Kim
S
, et al
.
Addendum: the cancer cell line encyclopedia enables predictive modelling of anticancer drug sensitivity
.
Nature
2019
;
565
:
E5
E6
.
13.
Hwang
WL
,
Jagadeesh
KA
,
Guo
JA
,
Hoffman
HI
,
Yadollahpour
P
,
Reeves
JW
, et al
.
Single-nucleus and spatial transcriptome profiling of pancreatic cancer identifies multicellular dynamics associated with neoadjuvant treatment
.
Nat Genet
2022
;
54
:
1178
91
.
14.
Li
J
,
Byrne
KT
,
Yan
F
,
Yamazoe
T
,
Chen
Z
,
Baslan
T
, et al
.
Tumor cell-intrinsic factors underlie heterogeneity of immune cell infiltration and response to immunotherapy
.
Immunity
2018
;
49
:
178
93
.
15.
Steele
NG
,
Carpenter
ES
,
Kemp
SB
,
Sirihorachai
V
,
The
S
,
Delrosario
L
, et al
.
Multimodal mapping of the tumor and peripheral blood immune landscape in human pancreatic cancer
.
Nat Cancer
2020
;
1
:
1097
112
.
16.
Datta
J
,
Dai
X
,
Bianchi
A
,
De Castro Silva
I
,
Mehra
S
,
Garrido
VT
, et al
.
Combined MEK and STAT3 inhibition uncovers stromal plasticity by enriching for cancer-associated fibroblasts with mesenchymal stem cell-like features to overcome immunotherapy resistance in pancreatic cancer
.
Gastroenterology
2022
;
163
:
1593
612
.
17.
Browaeys
R
,
Saelens
W
,
Saeys
Y
.
NicheNet: modeling intercellular communication by linking ligands to target genes
.
Nat Methods
2020
;
17
:
159
62
.
18.
Lee
KM
,
Nguyen
C
,
Ulrich
AB
,
Pour
PM
,
Ouellette
MM
.
Immortalization with telomerase of the Nestin-positive cells of the human pancreas
.
Biochem Biophys Res Commun
2003
;
301
:
1038
44
.
19.
Kim
MP
,
Li
X
,
Deng
J
,
Zhang
Y
,
Dai
B
,
Allton
KL
, et al
.
Oncogenic KRAS recruits an expansive transcriptional network through mutant p53 to drive pancreatic cancer metastasis
.
Cancer Discov
2021
;
11
:
2094
111
.
20.
Totiger
TM
,
Srinivasan
S
,
Jala
VR
,
Lamichhane
P
,
Dosch
AR
,
Gaidarski
AA
3rd
, et al
.
Urolithin A, a novel natural compound to target PI3K/AKT/mTOR pathway in pancreatic cancer
.
Mol Cancer Ther
2019
;
18
:
301
11
.
21.
Messaggio
F
,
Mendonsa
AM
,
Castellanos
J
,
Nagathihalli
NS
,
Gorden
L
,
Merchant
NB
, et al
.
Adiponectin receptor agonists inhibit leptin induced pSTAT3 and in vivo pancreatic tumor growth
.
Oncotarget
2017
;
8
:
85378
91
.
22.
Veglia
F
,
Hashimoto
A
,
Dweep
H
,
Sanseviero
E
,
De Leo
A
,
Tcyganov
E
, et al
.
Analysis of classical neutrophils and polymorphonuclear myeloid-derived suppressor cells in cancer patients and tumor-bearing mice
.
J Exp Med
2021
;
218
:
e20201803
.
23.
Pan
Y
,
Lu
F
,
Fei
Q
,
Yu
X
,
Xiong
P
,
Yu
X
, et al
.
Single-cell RNA sequencing reveals compartmental remodeling of tumor-infiltrating immune cells induced by anti-CD47 targeting in pancreatic cancer
.
J Hematol Oncol
2019
;
12
:
124
.
24.
Hatziapostolou
M
,
Koukos
G
,
Polytarchou
C
,
Kottakis
F
,
Serebrennikova
O
,
Kuliopulos
A
, et al
.
Tumor progression locus 2 mediates signal-induced increases in cytoplasmic calcium and cell migration
.
Sci Signal
2011
;
4
:
ra55
.
25.
Sidiropoulos
DN
,
Rafie
CI
,
Jang
JK
,
Castanon
S
,
Baugh
AG
,
Gonzalez
E
, et al
.
Entinostat decreases immune suppression to promote antitumor responses in a HER2+ breast tumor microenvironment
.
Cancer Immunol Res
2022
;
10
:
656
69
.
26.
Lee
J
,
Rhee
MH
,
Kim
E
,
Cho
JY
.
BAY 11–7082 is a broad-spectrum inhibitor with anti-inflammatory activity against multiple targets
.
Mediators Inflamm
2012
;
2012
:
416036
.
27.
Gavrin
LK
,
Green
N
,
Hu
Y
,
Janz
K
,
Kaila
N
,
Li
HQ
, et al
.
Inhibition of Tpl2 kinase and TNF-alpha production with 1,7-naphthyridine-3-carbonitriles: synthesis and structure-activity relationships
.
Bioorg Med Chem Lett
2005
;
15
:
5288
92
.
28.
Elyada
E
,
Bolisetty
M
,
Laise
P
,
Flynn
WF
,
Courtois
ET
,
Burkhart
RA
, et al
.
Cross-species single-cell analysis of pancreatic ductal adenocarcinoma reveals antigen-presenting cancer-associated fibroblasts
.
Cancer Discov
2019
;
9
:
1102
23
.
29.
Mpofu
S
,
Fatima
F
,
Moots
RJ
.
Anti-TNF-alpha therapies: they are all the same (aren't they?)
.
Rheumatology (Oxford)
2005
;
44
:
271
3
.
30.
de Castro Silva
I
,
Bianchi
A
,
Deshpande
NU
,
Sharma
P
,
Mehra
S
,
Garrido
VT
, et al
.
Neutrophil-mediated fibroblast-tumor cell il-6/stat-3 signaling underlies the association between neutrophil-to-lymphocyte ratio dynamics and chemotherapy response in localized pancreatic cancer: a hybrid clinical-preclinical study
.
Elife
2022
;
11
:
e78921
.
31.
Nagathihalli
NS
,
Castellanos
JA
,
Lamichhane
P
,
Messaggio
F
,
Shi
C
,
Dai
X
, et al
.
Inverse correlation of STAT3 and MEK signaling mediates resistance to RAS pathway inhibition in pancreatic cancer
.
Cancer Res
2018
;
78
:
6235
46
.
32.
Nagaraj
NS
,
Smith
JJ
,
Revetta
F
,
Washington
MK
,
Merchant
NB
.
Targeted inhibition of SRC kinase signaling attenuates pancreatic tumorigenesis
.
Mol Cancer Ther
2010
;
9
:
2322
32
.
33.
Nagaraj
NS
,
Washington
MK
,
Merchant
NB
.
Combined Blockade of src kinase and epidermal growth factor receptor with gemcitabine overcomes STAT3-mediated resistance of inhibition of pancreatic tumor growth
.
Clin Cancer Res
2011
;
17
:
483
93
.
34.
Nywening
TM
,
Belt
BA
,
Cullinan
DR
,
Panni
RZ
,
Han
BJ
,
Sanford
DE
, et al
.
Targeting both tumour-associated CXCR2(+) neutrophils and CCR2(+) macrophages disrupts myeloid recruitment and improves chemotherapeutic responses in pancreatic ductal adenocarcinoma
.
Gut
2018
;
67
:
1112
23
.
35.
Lopez-Bujanda
ZA
,
Haffner
MC
,
Chaimowitz
MG
,
Chowdhury
N
,
Venturini
NJ
,
Patel
RA
, et al
.
Castration-mediated IL-8 promotes myeloid infiltration and prostate cancer progression
.
Nat Cancer
2021
;
2
:
803
18
.
36.
Somerville
TD
,
Biffi
G
,
Dassler-Plenker
J
,
Hur
SK
,
He
XY
,
Vance
KE
, et al
.
Squamous trans-differentiation of pancreatic cancer cells promotes stromal inflammation
.
Elife
2020
;
9
:
e53381
.
37.
Padron
LJ
,
Maurer
DM
,
O'Hara
MH
,
O'Reilly
EM
,
Wolff
RA
,
Wainberg
ZA
, et al
.
Sotigalimab and/or nivolumab with chemotherapy in first-line metastatic pancreatic cancer: clinical and immunologic analyses from the randomized phase 2 PRINCE trial
.
Nat Med
2022
;
28
:
1167
77
.
38.
Hu
X
,
Ghisolfi
L
,
Keates
AC
,
Zhang
J
,
Xiang
S
,
Lee
DK
, et al
.
Induction of cancer cell stemness by chemotherapy
.
Cell Cycle
2012
;
11
:
2691
8
.
39.
Arnott
CH
,
Scott
KA
,
Moore
RJ
,
Robinson
SC
,
Thompson
RG
,
Balkwill
FR
.
Expression of both TNF-alpha receptor subtypes is essential for optimal skin tumour development
.
Oncogene
2004
;
23
:
1902
10
.
40.
Moore
RJ
,
Owens
DM
,
Stamp
G
,
Arnott
C
,
Burke
F
,
East
N
, et al
.
Mice deficient in tumor necrosis factor-alpha are resistant to skin carcinogenesis
.
Nat Med
1999
;
5
:
828
31
.
41.
Torrey
H
,
Butterworth
J
,
Mera
T
,
Okubo
Y
,
Wang
L
,
Baum
D
, et al
.
Targeting TNFR2 with antagonistic antibodies inhibits proliferation of ovarian cancer cells and tumor-associated Tregs
.
Sci Signal
2017
;
10
:
eaaf8608
.
42.
Montfort
A
,
Colacios
C
,
Levade
T
,
Andrieu-Abadie
N
,
Meyer
N
,
Segui
B
.
The TNF paradox in cancer progression and immunotherapy
.
Front Immunol
2019
;
10
:
1818
.
43.
Cagnina
RE
,
Michels
KR
,
Bettina
AM
,
Burdick
MD
,
Scindia
Y
,
Zhang
Z
, et al
.
Neutrophil-derived tumor necrosis factor drives fungal acute lung injury in chronic granulomatous disease
.
J Infect Dis
2021
;
224
:
1225
35
.
44.
Baxter
AE
,
Kaufmann
DE
.
Tumor-necrosis factor is a master of T cell exhaustion
.
Nat Immunol
2016
;
17
:
476
8
.
45.
Ban
L
,
Zhang
J
,
Wang
L
,
Kuhtreiber
W
,
Burger
D
,
Faustman
DL
.
Selective death of autoreactive T cells in human diabetes by TNF or TNF receptor 2 agonism
.
Proc Natl Acad Sci U S A
2008
;
105
:
13644
9
.
46.
Bertrand
F
,
Montfort
A
,
Marcheteau
E
,
Imbert
C
,
Gilhodes
J
,
Filleron
T
, et al
.
TNFalpha blockade overcomes resistance to anti–PD-1 in experimental melanoma
.
Nat Commun
2017
;
8
:
2256
.
47.
Govindaraj
C
,
Scalzo-Inguanti
K
,
Madondo
M
,
Hallo
J
,
Flanagan
K
,
Quinn
M
, et al
.
Impaired Th1 immunity in ovarian cancer patients is mediated by TNFR2+ Tregs within the tumor microenvironment
.
Clin Immunol
2013
;
149
:
97
110
.
48.
Ben-Baruch
A
.
Tumor necrosis factor alpha: taking a personalized road in cancer therapy
.
Front Immunol
2022
;
13
:
903679
.
49.
Case
K
,
Tran
L
,
Yang
M
,
Zheng
H
,
Kuhtreiber
WM
,
Faustman
DL
.
TNFR2 blockade alone or in combination with PD-1 blockade shows therapeutic efficacy in murine cancer models
.
J Leukoc Biol
2020
;
107
:
981
91
.
50.
Madhusudan
S
,
Foster
M
,
Muthuramalingam
SR
,
Braybrooke
JP
,
Wilner
S
,
Kaur
K
, et al
.
A phase II study of etanercept (Enbrel), a tumor necrosis factor alpha inhibitor in patients with metastatic breast cancer
.
Clin Cancer Res
2004
;
10
:
6528
34
.
51.
Biffi
G
,
Oni
TE
,
Spielman
B
,
Hao
Y
,
Elyada
E
,
Park
Y
, et al
.
IL1-induced JAK/STAT signaling is antagonized by tgfbeta to shape CAF heterogeneity in pancreatic ductal adenocarcinoma
.
Cancer Discov
2019
;
9
:
282
301
.
52.
Baxevanis
CN
,
Voutsas
IF
,
Tsitsilonis
OE
,
Tsiatas
ML
,
Gritzapis
AD
,
Papamichail
M
.
Compromised anti-tumor responses in tumor necrosis factor-alpha knockout mice
.
Eur J Immunol
2000
;
30
:
1957
66
.
53.
Ritchie
ME
,
Phipson
B
,
Wu
D
,
Hu
Y
,
Law
CW
,
Shi
W
, et al
.
limma powers differential expression analyses for RNA-sequencing and microarray studies
.
Nucleic Acids Res
2015
;
43
:
e47
.
54.
Benjamini
Y
,
Hochberg
Y
.
Controlling the false discovery rate - a practical and powerful approach to multiple testing
.
J Roy Stat Soc B Met
1995
;
57
:
289
300
.
55.
Jin
S
,
Guerrero-Juarez
CF
,
Zhang
L
,
Chang
I
,
Ramos
R
,
Kuan
CH
, et al
.
Inference and analysis of cell–cell communication using CellChat
.
Nat Commun
2021
;
12
:
1088
.
56.
Sanjana
NE
,
Shalem
O
,
Zhang
F
.
Improved vectors and genome-wide libraries for CRISPR screening
.
Nat Methods
2014
;
11
:
783
4
.
57.
Chang
Z
,
Ju
H
,
Ling
J
,
Zhuang
Z
,
Li
Z
,
Wang
H
, et al
.
Cooperativity of oncogenic K-ras and downregulated p16/INK4A in human pancreatic tumorigenesis
.
PLoS One
2014
;
9
:
e101452
.
58.
Toska
E
,
Osmanbeyoglu
HU
,
Castel
P
,
Chan
C
,
Hendrickson
RC
,
Elkabets
M
, et al
.
PI3K pathway regulates ER-dependent transcription in breast cancer through the epigenetic regulator KMT2D
.
Science
2017
;
355
:
1324
30
.
59.
Chytil
A
,
Magnuson
MA
,
Wright
CV
,
Moses
HL
.
Conditional inactivation of the TGF-beta type II receptor using Cre:Lox
.
Genesis
2002
;
32
:
73
5
.
60.
Nagathihalli
NS
,
Castellanos
JA
,
Shi
C
,
Beesetty
Y
,
Reyzer
ML
,
Caprioli
R
, et al
.
Signal transducer and activator of transcription 3, mediated remodeling of the tumor microenvironment results in enhanced tumor drug delivery in a mouse model of pancreatic cancer
.
Gastroenterology
2015
;
149
:
1932
43
.