Cell-Autonomous Cxcl1 Sustains Tolerogenic Circuitries and Stromal Inflammation via Neutrophil-Derived TNF in Pancreatic Cancer
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 KRAS–TP53 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-Kras;Trp53;Pdx-1(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.
Collaborative work led by the Datta lab at the University of Miami. I contributed to the computational analysis.
Pancreatic ductal adenocarcinoma resists almost everything, and three things are usually blamed: an undruggable genome dominated by co-occurring KRAS and TP53 mutations, myeloid cells that suppress the T-cell response, and inflammatory fibroblasts that make the stroma hostile.
Those are usually studied separately. This paper argues they are one circuit, and finds the wire connecting them.
Starting from a spatial observation
Tumors carrying both KRAS and TP53 alterations have worse survival and immune-excluded microenvironments. Looking for what the cancer cells themselves do to cause that, CXCL1 — a secreted neutrophil chemoattractant — stood out as overexpressed in the co-altered tumors, in cell line transcriptomes, in single-cell and single-nucleus data from patients, and in mouse models.
Its receptor, CXCR2, sits almost exclusively on neutrophilic myeloid-derived suppressor cells.
Imaging mass cytometry on human tumors makes the consequence visible. Across roughly 73,000 cells, CXCL1-expressing tumor islands and CXCR2+ MDSCs sit directly against each other — a median of about 7 μm apart. CD8+ T cells are excluded from those neighborhoods, sitting some 95 μm away. The tumor recruits a cell type that keeps T cells out, and you can see the arrangement in tissue.
Upstream and downstream
Upstream, KRAS and TP53 mutations cooperate to drive CXCL1 through CREB. Phospho-kinome arrays put CREB at Ser133 as the top hyperphosphorylated transcription factor in co-altered cells; a CREB inhibitor cut CXCL1 transcription and secretion; and ChIP-seq showed CREB and RNA Pol-II co-bound at the Cxcl1 promoter.
Downstream, CRISPR deletion of Cxcl1 in tumor cells reduced tumor weight and metastasis and extended median survival from 21 days to 48. In mice lacking CD8+ T cells, that benefit disappeared — so the effect runs through adaptive immunity rather than through tumor cell growth.
The MDSCs are reprogrammed, not just unrecruited
The obvious reading is that deleting the chemokine removes the gradient and the MDSCs stay away. An adoptive transfer experiment says otherwise.
Labelled MDSCs from control tumor-bearing mice trafficked to control tumors and not to Cxcl1-deleted ones, as expected. But MDSCs taken from Cxcl1-deleted tumor-bearing mice failed to traffic even into tumors with intact CXCL1.
Their migratory capacity is altered, independent of the gradient in front of them. Their suppressive machinery is too: lower Arg1, Ido and Mpo, lower arginase activity, and a loss of the "activated" transcriptional module. T cells co-cultured with them recovered IFN-γ release.
TNF, from an unexpected source
RNA sequencing of MDSCs sorted from tumors nominated TNF as the top upstream regulator, acting through a CXCR2–MAP3K8–TNF node. Blocking any step in that chain cut TNF expression by more than half.
The source is the surprise. In human single-cell data and in patient blood, the highest TNF expression was in neutrophilic MDSCs, above every other cell type. TNF's tolerogenic role in cancer has generally been attributed to the stroma.
And the signal is delivered by contact. Co-culture drove more than a fivefold increase in CXCL1 only when the cells touched — separate them in a transwell and it vanished. Membrane-bound TNF signalling through TNFR2 is what matters here, which is why a soluble-TNF inhibitor did nothing while a TNFR2-capable one worked.
That closes a feed-forward loop: tumor CXCL1 recruits MDSCs, MDSC membrane TNF drives more CXCL1 from tumor cells and fibroblasts, which recruits more MDSCs.
Where the paradox sits
Blocking the CXCL1–CXCR2 axis reduces TNF signalling across the whole tumor and improves antitumor immunity. TNF is supposed to be inflammatory and antitumor. Here the MDSC-restricted pool of it is doing the opposite, which is worth holding onto: the same cytokine reads differently depending on which cell makes it and how it is presented.
The same axis drives the fibroblasts. MDSCs induced roughly 40-fold Il6 in cancer-associated fibroblasts; TNFR2 inhibition cut that by two-thirds, shifted the fibroblast population away from the inflammatory phenotype, reduced collagen deposition, and lowered STAT3 activation in the tumor cells.
The therapeutic result
TNFR2 inhibition alone did not extend survival. Gemcitabine and paclitaxel alone did not extend survival. Together they nearly doubled median survival, from 25 days to 44, with less metastasis and no added toxicity.
The circuit is not a growth driver you can hit directly. It is the thing that makes chemotherapy fail, and removing it lets the chemotherapy work.
Caveats
The functional work is in genetically engineered mouse models; the human data is descriptive — expression, spatial arrangement and correlation. Etanercept blocks more than tmTNF–TNFR2, so the in vivo results are less specific than the co-culture experiments that identified the mechanism. And no chemotherapy trial in PDAC has yet tested this combination.