Regulatory T-cell Depletion Alters the Tumor Microenvironment and Accelerates Pancreatic Carcinogenesis

Pancreatic ductal adenocarcinoma (PDA) is one of the deadliest human malignancies, with a 5-year overall survival rate of ∼9% (1). PDA and its most common precursor lesions, pancreatic intraepithelial neoplasia (PanIN), are characterized by an extensively fibrotic microenvironment, which includes abundant infiltrating immune cells. Success in targeting the pancreatic cancer microenvironment has been inconsistent, and immunotherapy approaches such as immune-checkpoint inhibition have largely failed (2, 3).

Oncogenic mutations are prevalent in pancreatic cancer (4–6) and are present with high frequency in precursor lesions (7). Mouse models genetically engineered to express oncogenic Kras in the pancreas—known as KC mice—develop PanIN lesions that progress to invasive cancer over time (8, 9). The formation of PanIN is accompanied by infiltration of suppressive immune cells (10), whereas cytotoxic CD8⁺ T cells are rare. Among these, myeloid cells are present in high abundance in the microenvironment, including both macrophages and myeloid-derived suppressor cells/immature myeloid cells (10–16).

This immune suppression is functionally important, as activation of a CD8⁺ T cell–mediated immune response is sufficient to block the onset of carcinogenesis (17–19) or induce tumor regression (11, 18, 20–22). Genetic elimination of CD4⁺ T cells results in the activation of CD8⁺ T cells and consequently prevents PanIN progression (17). Similarly, depletion of Th17 cells, a CD4⁺ T-cell subset, inhibits pancreatic carcinogenesis (23). Further, analysis of long-term survivors of pancreatic cancer reveals that antitumor T cells persist in their peripheral blood years after the initial diagnosis, further supporting the notion that activation of an antitumor immune response might be effective (24). However, our limited understanding of the mechanisms underlying immune suppression in pancreatic cancer limits our ability to translate these findings to the clinic.

Regulatory T cells (Treg), defined as CD4⁺CD25⁺FOXP3⁺ T cells, accumulate in mouse and human PanIN and pancreatic cancer (10, 17, 25). Moreover, Treg frequency positively correlates with tumor metastasis and poor prognosis in human patients with PDA (26, 27). Depletion of Tregs has been tested in a model of transplanted pancreatic cancer in mice, where it led to a CD8⁺ T cell–mediated antitumor immune response (21). However, the role of Tregs during the onset and progression of pancreatic cancer, as well as in spontaneous invasive tumors, had not been investigated, and is the focus of our study.

To study the relative abundance of Tregs, myeloid cells, and CD8⁺ T cells in human pancreatic cancer, we performed cytometry by time-of-flight (CyTOF) analysis. We included 3 surgical samples from Whipple resections and 2 endoscopic ultrasound–guided core needle biopsy samples. High-dimensional analysis using FlowSOM-viSNE (28, 29) revealed the presence of a CD45⁺CD3⁺CD4⁺CD25⁺ Treg population and several subsets of myeloid cells in each sample (Fig 1A; Supplementary Fig. S1A), whereas CD8⁺ T cells were rare (Fig. 1A; Supplementary Fig. S1A). To determine the localization of immune cells within the tissue, we have recently optimized the use of a fluorescent multiplex IHC system (Opal Multiplex IHC, PerkinElmer) for tumor immunophenotyping (30). Using this platform, we stained 39 human pancreatic cancer samples, 37 PanIN samples, and 52 adjacent normal samples or samples of chronic pancreatitis—a noncancerous inflammatory condition of the pancreas. Out of 52 chronic pancreatitis samples, only 14 had measurable Tregs (defined as CD3⁺ CD8⁻ FOXP3⁺). In contrast, Tregs were present in most of the tumor samples and often observed in close proximity to the tumor cells (Fig. 1B; Supplementary Fig. S1B); their abundance positively correlated with macrophages in the same tissue (Supplementary Fig. S1C) and—with two exceptions—with CD8⁺ T cells (Supplementary Fig. S1C). Interestingly, Tregs were abundantly present in the PanIN samples as well, possibly indicating a role during the early stages of carcinogenesis (Fig. 1B; Supplementary Fig. S1D and S1E). Similarly, in PanIN-bearing Kras^+/LSL-G12D;Ptf1a^+/Cre (KC) mice (8), we observed abundant Treg infiltration by FOXP3 immunostaining (Fig. 1C). To further characterize the Treg population, we performed single-cell sequencing analysis of spontaneous PanIN lesions in iKras* mice (31), and iKras*;p53* (32) pancreatic cancer cells orthotopically transplanted in syngeneic mice. In both experimental systems, we detected a cell population coexpressing Cd4, Ctla4, Foxp3, and Il2ra (Fig. 1D; Supplementary Fig. S1F). Similarly, in human samples (data not shown), FOXP3 was coexpressed with CTLA4 and IL2RA, and thus in bona fide Tregs (for a review of Treg classification, see ref. 33).

To study the effect of Treg depletion on the formation of PanIN, we generated Kras^+/LSL-G12D;Ptf1a^+/Cre;Foxp3^{tm3(DTR/GFP)Ayr} (KC;Foxp3^DTR) mice (Fig. 2A; refs. 8, 34). In these animals, Tregs can be depleted at will by administering diphtheria toxin (DT). During tissue homeostasis, Tregs play an important role in regulating the activity of the immune system and preventing autoimmune disease (34). Accordingly, depletion of Tregs for 3 weeks, starting at 4 to 5 weeks of age (Fig. 2B) in Foxp3^DTR mice, led to systemic inflammation shown by hypertrophy of the lymphoid organs, such as increased spleen to body weight ratio (Fig. 2C; Foxp3^DTR mice). In Foxp3^DTR pancreata, Treg depletion resulted in acinar cell loss (Fig. 2D) and pancreatitis with distinct histologic changes such as acinar–ductal metaplasia (ADM)—dedifferentiation of acinar cells to duct-like cells (Fig. 2D and E)—and elevated immune infiltration (see immunostaining for CD45 in Supplementary Fig. S2A) with characteristic presence of CD138⁺ plasma cells (Supplementary Fig. S2B). In KC;Foxp3^DTR mice, Treg depletion led to increased spleen to body weight ratio, as well as increased pancreas to body weight ratio, a measure of tumor burden (Fig. 2C). As expected, KC pancreata had sparse lesions, consistent with their young age (Fig. 2D, histopathologic quantification; Fig. 2E). In contrast, Treg depletion in KC;Foxp3^DTR mice presented with extensive ADM and PanIN [Fig. 2D–E; Supplementary Fig. S2A, periodic acid–Schiff (PAS) staining], as well as abundant immune cell infiltration (Supplementary Fig. S2A).

We reasoned that the pancreatitis caused by Treg depletion synergized with oncogenic Kras to promote PanIN formation as shown in other experimental models of pancreatitis (35–40). We then modified the experimental design to address the role of Tregs in preexisting PanIN. We administered caerulein (a cholecystokinin agonist) by 8 hourly intraperitoneal injections for 1 day to induce acute pancreatitis in 6-week-old KC and KC;Foxp3^DTR mice. KC;CD4^−/− mice were included as an additional control, as we have previously shown that lesions in CD4 ablated animals fail to progress because of productive antitumor immunity (17). The mice were allowed to develop PanIN lesions for 8 weeks after caerulein treatment and then received DT for 10 days or 3 weeks (Fig. 2F). The two cohorts showed a similar phenotype. Following this treatment, KC;Foxp3^DTR mice had increased pancreas to body weight ratio compared with either KC or KC;CD4^−/− mice (Fig. 2G). Histologic analysis of the pancreata showed acinar areas with interspersed lesions in KC and KC;CD4^−/− mice. In contrast, KC;Foxp3^DTR mice had widespread lesions that almost completely replaced the pancreas parenchyma (Fig. 2H–I; Supplementary Fig. S2C).

PanIN lesions, like pancreatic cancer, are surrounded by a fibrotic microenvironment. Accordingly, α-smooth muscle actin (SMA) staining of KC pancreata revealed characteristic areas of SMA⁺ cells surrounding the epithelial lesions (Fig. 3A and B). Intriguingly, lesions in KC;Foxp3^DTR pancreata lacked SMA expression, whether carcinogenesis was or was not accelerated by the induction of pancreatitis (Fig. 3A and B), whereas immunostaining for the fibroblast marker PDGFRβ appeared unchanged (Fig. 3B). We then counted PDGFRβ⁺ cells and SMA⁺ cells in KC and KC;Foxp3^DTR pancreata, and calculated their ratio. The data revealed no change in total fibroblast numbers, but a reduction in SMA⁺ fibroblasts, a finding that is consistent with reprogramming of the fibroblast population (Fig. 3B). Consistent with the immunostaining data, α-SMA (Acta2) mRNA expression was lower in KC;Foxp3^DTR compared with KC pancreata. Extracellular matrix (ECM) genes such as Col1 and Fn1, as well as genes related to fibroblast activation and ECM synthesis such as Tgfb1 and Ctgf were also downregulated in KC;Foxp3^DTR compared with KC pancreata (Fig. 3C). We identified the cellular source of Tgfb1 and putative responsive cells expressing the mRNAs encoding the TGFβ receptors by single-cell RNA sequencing (RNA-seq) of mouse PanIN and pancreatic cancer. Tgfb1 was expressed by both epithelial cells and T cells, including Tregs. Conversely, the 3 Tgfb receptors were expressed in the majority of fibroblasts, as well as a subset of epithelial cells (Fig. 3D). Supporting these findings, The Cancer Genome Atlas (TCGA) database (https://www.cbioportal.org/datasets) analysis revealed a positive correlation between a Treg signature and expression of ACTA2 and TGFB signaling genes in human tumors (Supplementary Fig. S1G–S1I). SMA-high fibroblasts define the “myofibroblastic CAF” (myCAF) population of cancer-associated fibroblasts, a phenotype that is driven by TGFβ (41, 42). Our data are consistent with the notion that Tregs are a source of TGFβ and drive myCAF differentiation of fibroblasts.

We then set out to determine the effect of Treg depletion on the immune microenvironment. To compare the immune infiltration among different groups, we devised a protocol that would result in a similar lesion load in the different genotypes, namely, KC, KC;CD4^−/−, and KC;Foxp3^DTR. In brief, adult mice received 8 hourly caerulein injections in a day. Eight weeks later, they were treated with DT for 1 week and then harvested for histology and flow cytometry (Fig. 4A; Supplementary Fig. S2D). PanIN lesions in mice lacking CD4⁺ T cells regress over time in a CD8⁺ T cell–mediated manner (17). Consistently, KC;CD4^−/− mice had increased CD8⁺ T cells and increased IFNγ-expressing CD8⁺ T cells compared with KC mice (Fig. 4B). In contrast, in KC;Foxp3^DTR pancreata, we observed only a modest increase in CD8⁺ T cells and no change in IFNγ⁺;CD8⁺ T cells compared with KC mice. In contrast, we observed an overall increase in immune cells (CD45⁺) and an increase in CD4⁺ T cells, as well as the expected loss of FOXP3⁺ cells. These data indicated that little or no productive immune response was elicited, notwithstanding the depletion of Tregs. By flow cytometry, total myeloid cells (CD11b⁺) and macrophages (CD11b⁺F4/80⁺; % of total cells) trended toward an increase, whereas immature myeloid cells/myeloid-derived suppressor cells (MDSC; CD11b⁺Ly6C⁺Ly6G⁺F4/80⁻) were elevated in KC;Foxp3^DTR and KC;CD4^−/− pancreata compared with KC pancreata (Fig. 4C). Immunostaining of KC and KC;Foxp3^DTR pancreas tissue showed an increase in CD8⁺ T cells in proximity to the lesions, but we also observed a comparable increase in F4/80⁺ macrophages in those areas (Supplementary Fig. S2E). Myeloid cells have a known immune-suppressive function (11, 13, 18); thus, we hypothesized that an increase in their number or suppressive potential might explain the difference in phenotype between Treg depletion and CD4⁺ T-cell elimination during pancreatic carcinogenesis.

To assess the changes in the myeloid compartment upon Treg depletion, we performed a comparative transcriptomic analysis of the myeloid cells (DAPI⁻EPCAM⁻CD45⁺CD11b⁺) sorted from KC and KC;Foxp3^DTR tumors (Fig. 4D). The RNA-seq results, including a principal component analysis (PCA) plot and a volcano plot illustrating differential gene expression in myeloid cells, are shown in Supplementary Fig. S3A and S3B. Interestingly, we detected an increase in Arg1, Chi3l3 (also known as Ym1), Retnlα, and Pdcd1lg2 expression in myeloid cells derived from KC;Foxp3^DTR pancreata, whereas Il1b was downregulated (Fig. 4E). Further, we observed an increase in expression of Arg1 and Chil3l3—markers of tumor-associated macrophages (TAM) with immunosuppressive function (43, 44)—in both fibroblasts and macrophages (Supplementary Fig. S2E). By flow cytometry, we observed that cell-surface PD-L1 protein expression was elevated in TAMs (CD45⁺CD11b⁺F4/80⁺), epithelial cells (CD45⁻EPCAM⁺), and fibroblasts (CD45⁻EPCAM⁻PDGFRα⁺) in KC;Foxp3^DTR compared with KC pancreata (Fig. 4F). Taken together, our data indicate that a compensatory immunosuppressive program is activated upon Treg depletion in mice bearing PanIN lesions.

The changes in expression of individual genes could be explained either by gene-expression changes across the entire myeloid population or by changes in the composition of cell types within the myeloid population. To evaluate these possibilities, we proceeded to immunophenotype our samples by CyTOF with 16 validated antibodies (Supplementary Table S1). viSNE analysis suggested there were at least 11 different populations in our data set (Fig. 5A and B; Supplementary Fig. S3C), and FlowSOM analysis with supervised hierarchical clustering identified 21 separate populations (Fig. 5C). Among these, we identified multiple myeloid subpopulations, including monocytic MDSCs (Ly6C⁺), granulocytic MDSCs (Ly6G⁺), infiltrating monocytes (CD11b⁺), and macrophage/dendritic cell subsets (F4/80^dim/+ and CD11c⁺, respectively). Interestingly, we found that both the monocyte/MDSC and macrophage subpopulations could be further divided based on the intracellular expression of arginase 1 (population 5 vs. 6 and populations 15–18 vs. 19–21). When we analyzed the frequency of each subpopulation in the KC versus KC;FoxP3^DTR mice, we found a consistent increase in the frequency of ARG1⁺ myeloid subsets after Treg depletion (Supplementary Fig. S3D). By flow cytometry, we found that the percentage of CD206⁻iNOS⁺ classically activated macrophages remained unchanged, whereas CD206⁺ARG1⁺ macrophages increased more than 8-fold upon Treg depletion (Fig. 5D), compared with KC control. More interestingly, we found a non–immune cell population (CD45⁻CD11b⁺)—possibly a subset of stromal fibroblasts—also expressed a higher level of ARG1 in KC;Foxp3^DTR pancreata upon Treg depletion (Fig. 5E). Therefore, Treg depletion resulted in a compensatory increase in a prevalently myeloid-driven immune-suppression program.

Treg depletion was recently reported to induce antitumor immune responses and inhibit tumor growth in a transplantation model of pancreatic cancer (21). We thus reasoned that either the stage of disease (onset vs. advanced disease) or different models (spontaneous vs. transplanted tumors) explained the divergent findings. To investigate this apparent inconsistency, we implanted a KPC (Ptf1aCre;LSL-Kras; Trp53^R172H/+; ref. 45) pancreatic cancer cell line (7940B; ref. 46) orthotopically into syngeneic C57BL/6 Foxp3^DTR mice and depleted Tregs 11 days later (Supplementary Fig. S4A). At harvest (day 20), we observed smaller tumors in the Foxp3^DTRcohort (Supplementary Fig. S4B) and an increase in the expression of CD8⁺ T-cell activation markers Ifny, Gzmb, and Prf1 (Supplementary Fig. S4C). Consistent with activation of antitumor immunity, concurrent depletion of CD8⁺ T cells partially rescued tumor growth (Supplementary Fig. S4B). Flow cytometry analysis revealed increased CD8⁺ T cells and myeloid cells, including macrophages, upon Treg depletion (Supplementary Fig. S4D). Immunostaining confirmed increased immune infiltration and apoptosis in Treg-depleted tumors (Supplementary Fig. S4E). At the same time, we observed fewer SMA-positive fibroblasts and more expression of the immunosuppressive factors Chi3l3 (notably in tumor epithelial cells) and Arg1 in Foxp3^DTR tumors (Supplementary Fig. S4E and S4F), which was consistent with our findings in the KC;Foxp3^DTR model. We then repeated a similar set of experiments in KPC;Foxp3^DTR (Ptf1aCre;LSL-Kras;Trp53^R172H/+; Foxp3^DTR) mice that develop spontaneous invasive pancreatic cancer. Upon detection of a tumor by ultrasound imaging, we administered DT to deplete Tregs (or used vehicle control) and followed tumor growth (Supplementary Fig. S4G). In this setting, we observed continued tumor growth even following Treg depletion. Upon harvesting and immunostaining, we observed an increase in both CD8⁺ T cells and myeloid cells (Supplementary Fig. S4G). Our data are consistent with Treg depletion failing to elicit a productive antitumor immune response in late-stage disease, likely due to compensatory myeloid-driven immune suppression.

To elucidate a potential causal link between the changes in the fibroblast and epithelial compartments and the shift in immune composition upon Treg depletion, we flow sorted epithelial cells (CD45⁻EPCAM⁺) and fibroblasts (CD45⁻EPCAM⁻PDGFRα⁺) from KC and KC;Foxp3^DTR pancreata (n = 3 for each group; Fig. 6A and B) and performed RNA-seq. We then analyzed the secretome transcripts in both compartments. Interestingly, overall gene expression, as well as expression of secreted factors, showed a distinct pattern in both epithelial cells and fibroblasts upon Treg depletion (Fig. 6C; Supplementary Fig. S5A and S5B). Among more than 400 differentially expressed secreted chemokines, cytokines, and immunosuppressive factors, we observed that the C–C motif chemokine ligands Ccl3, Ccl6, and Ccl8 (Fig. 6D) were among the top upregulated secreted factors in fibroblasts and/or epithelial cells upon Treg depletion, whereas interleukins such as Il1b were downregulated but to a lesser extent compared with what we saw earlier in the myeloid cell compartment (Fig. 4E). We validated these changes by qRT-PCR in 4 to 8 samples/genotype (Fig. 6E; Supplementary Fig. S5C). Finally, immunostaining confirmed upregulated expression of CCL8 in KC;Foxp3^DTR pancreata (Fig. 6F) compared with KC. Other notable changes in epithelial cells and fibroblasts were increases in immunosuppressive factors such as Arg1 and Chi3l3 as well as in the immune checkpoint molecules Cd274 (encoding PD-L1) and Pdcd1lg2 (Fig. 6E; Supplementary Fig. S5C).

CCL3, CCL6, and CCL8 bind a common receptor, CCR1, and they are known chemoattractants for myeloid cells (47). We identified 4 subsets of myeloid cells in PanINs and 3 in PDA by single-cell sequencing analysis (Supplementary Fig. S1F): myeloid 1: MDSCs (high expression of Itgam, Cd14, Fcgr3, S100a8, and Ly6g); myeloid 2: macrophages (high expression of Itgam, Cd68, Adgre1, and Mrc1); myeloid 3: dendritic cells (high expression of Itgax, H2-Eb1, Batf3, Itgae, and Clec9a); and myeloid 4: monocytes (high expression of Itgax and Ly6c2). Of note, myeloid 4 was not detected in tumors, but only in PanIN samples. Single-cell sequencing analysis of mouse PDA or PanINs confirmed that expression of Ccr1 was restricted to the myeloid compartment and included MDSCs and macrophages (Fig. 6G). We validated these findings in vitro by qRT-PCR from bone marrow–derived macrophages (M0) polarized to M1 with lipopolysaccharide (LPS), to M2 with IL4, and to tumor-educated macrophages (TEM) with pancreatic cancer cell conditioned media. Ccr1 expression was lowest in M1 macrophages and highest in TEMs (Fig. 6H). Our single-cell sequencing analysis also revealed that myeloid cells are the main source of Il1a both in PanINs and in pancreatic cancer, whereas the receptor Il1r is expressed in a subset of epithelial cells and in the vast majority of fibroblasts (Supplementary Fig. S5D). The expression of Il1a was not different in KC and KC;FoxP3^DTR myeloid cells, epithelial cells, and fibroblasts (Supplementary Fig. S5E). IL1α drives differentiation of “inflammatory CAFs” (iCAF), a proinflammatory subset of fibroblasts that also expresses elevated IL6 (41), and Il6 was similarly unchanged in our model (data not shown). Thus, although fibroblasts in KC;FoxP3^DTR become more inflammatory at least pertaining to recruitment of myeloid cells, they might not fit exactly in the iCAF subtype, consistent with the complexity of this cell type (42, 48).

Our data to this point demonstrated that Treg depletion resulted in an increase in several CCR1 ligands and a corresponding influx of immune-suppressive myeloid cells. To determine whether CCR1 activation is responsible for the increased myeloid infiltration and increased tumorigenesis, we utilized a commercial CCR1 inhibitor, BX471 (hereby termed CCR1i). We treated PanIN-bearing KC;Foxp3^DTR mice with CCR1i for a week and at the same time we depleted Tregs (Fig. 7A). At endpoint, KC;Foxp3^DTR mice that received DT and vehicle control showed enlarged pancreata and spleens compared with KC mice, an effect that was reversed by CCR1i treatment (Supplementary Fig. S6A). Further, histopathologic analysis revealed more acini and less ADM/PanIN (CK19⁺ cells) in KC;Foxp3^DTR mice treated with CCR1i (Fig. 7B; Supplementary Fig. S6B). Immunostaining for CD45/F4/80 showed decreased total immune infiltration and macrophages (Fig. 7B; Supplementary Fig. S6C), although the level of immune infiltration remained elevated compared with KC mice (data not shown). Thus, CCR1 inhibition bypassed Treg depletion–induced acceleration of carcinogenesis in KC mice driven by infiltration of immunosuppressive myeloid cells.

Reprogramming of fibroblasts and consequent increased infiltration of myeloid cells restored immune suppression upon Treg depletion. However, we were intrigued by the increase in total CD4⁺ T cells upon Treg depletion, and by the acceleration of carcinogenesis in contrast with genetic or antibody-driven CD4⁺ T-cell depletion (17, 23). We thus reasoned that Tregs might normally restrain pathogenic Th2 or Th17 responses, both of which have been associated with tumor progression in pancreatic cancer (23, 49). By flow cytometry, we observed an increase in CD3⁺CD8⁻CD4⁺IFNγ⁺ T cells in KC;Foxp3^DTR pancreata compared with both KC and KC;CD4^−/− mice (Supplementary Fig. S6D). Analysis of the cytokine profile in sorted T cells from KC;Foxp3^DTR pancreata revealed an increase in the Th2 cytokines Il4, Il10, and Il13 and a reduction in Th1 cytokines such as Il2 and Tnfa, as well as Il22 and Il17 in KC;Foxp3^DTR mice. Our data are thus consistent with a mixed Th1/Th2 response, but the cytokine profile suggests prevalence of a Th2 response (ref. 50; Supplementary Fig. S6E). We then depleted CD4⁺ T cells from wild-type, FoxP3^DTR, KC, and KC;Foxp3^DTR mice. Isotype control IgG was administered to the control cohorts, and DT was administered for consistency with previous experiments, as indicated (Fig. 7C). Although Treg depletion causes pancreatitis, CD4⁺ T-cell depletion had no effect on wild-type mice. Similarly, CD4 depletion did not accelerate carcinogenesis in young KC mice (prior to lesion formation), whereas Treg depletion promoted widespread metaplasia, PanIN formation, and accumulation of fibroinflammatory stroma (Fig. 7D; Supplementary Fig. S6F). Interestingly, CD4⁺ T-cell depletion prevented the formation of metaplasia and PanIN in KC;Foxp3^DTR mice, as well as ameliorated the pancreatitis in FoxP3^DTR mice (Fig. 7D). CD4⁺ T-cell depletion also prevented the extensive immune infiltration and T-cell infiltration elicited by Treg depletion (Supplementary Fig. S6G). Together, our data point to multiple roles for Tregs within the pancreatic cancer microenvironment, which include restraining pathogenic T-cell responses caused by other CD4⁺ T-cell subsets.

Pancreatic cancer precursor lesions, such as PanIN, are surrounded by a fibroinflammatory stroma that persists in late-stage disease (reviewed in ref. 4). Although immune cells are prevalent surrounding the lesions, they are mostly suppressive in nature (10). CD4⁺ T cells, and among those Tregs, are the most abundant T-cell population, whereas CD8⁺ T cells are rare. Depletion of CD4⁺ T cells inhibits PanIN formation (17, 23), although it is insufficient to activate antitumor immunity at later stages (51). Treg depletion is sufficient to induce antitumor immunity in melanoma (52). Further, Treg depletion in orthotopically transplanted tumors derived from the KPC model (expressing both mutant Kras and Trp53; ref. 45) in syngeneic C57BL/6 mice causes CD8⁺ T cell–mediated tumor regression (21), a finding that we reproduce here. Here, we generated KC;FoxP3^DTR and KPC;FoxP3^DTR mice, where Tregs can be depleted at will in the context of spontaneous carcinogenesis. In contrast with previous findings, we show that Treg depletion in mice before or after the development of PanIN caused accelerated neoplastic progression. In invasive tumors in KPC;FoxP3^DTR mice, Treg depletion failed to restrain tumor growth. Analysis of the tissue revealed that an increase in CD8⁺ T cells upon Treg depletion was offset by a compensatory increase in other CD4⁺ T cells and immunosuppressive myeloid cells, and by reprogramming of the fibroblast population, from predominantly SMA-expressing to largely SMA-low (Fig. 7E and F).

The role of fibroblasts in pancreatic cancer is complex, and different studies have provided contradictory results. Depletion of specific fibroblast subsets, such as the FAP⁺ cell, resulted in a reduction in tumor growth and improved response to immunotherapy (53). Conversely, depletion of SMA⁺ fibroblasts resulted in tumor promotion (54). A solution to this apparent contradiction came from the observation that heterogeneous fibroblast populations are present in the pancreatic cancer stroma (42, 48). The SMA-high myCAF population is predominantly driven by TGFβ and has been described as tumor restraining (42, 54). We thus measured TGFβ mRNA levels and found them reduced in the pancreas of Treg-depleted mice. Using single-cell sequencing, we determined that tumor epithelial cells, myeloid cells, and T cells, including Tregs, are the sources of TGFβ1 within the tissue. Conversely, epithelial cells and pancreatic fibroblasts express TGFβ receptors. TGFβ signaling in epithelial cells is known to be tumor-suppressive at early stages of carcinogenesis (55); thus, a direct effect of TGFβ reduction might partially explain the increase in carcinogenesis upon Treg depletion. We also observed fibroblast reprogramming upon Treg depletion, from tumor-restricting and SMA-high myCAFs to a tumor-promoting fate, providing in vivo evidence of a phenomenon described in tumor organoids by the Tuveson laboratory (41). Conversely, we did not observe changes in Il1a, a key driver of iCAF differentiation (41). We show that reprogrammed fibroblasts have increased secretion of chemokines that act as a chemoattractant for suppressive myeloid cells. In addition to an increase in myeloid cells, Treg depletion also resulted in elevated expression of immune-suppressive genes in fibroblasts, such as Arg1 and Cd274 (PD-L1). Clinically, fibroblast reprogramming using different approaches—such as vitamin D analogues or inhibition of focal adhesion kinase—as well as remodeling of the extracellular matrix are actively being tested for pancreatic cancer therapy (56–59).

To obtain mechanistic insight into the paradoxical Treg depletion–driven tumor progression, we sequenced RNA of flow-sorted fibroblasts from Treg-depleted KC pancreata, and uncovered increased expression of a number of chemokines, most notably Ccl3, Ccl6, and Ccl8. CCL3 (or MIP1α) and CCL8 (or MCP2) are chemotactic for many different immune cells, including macrophages and other myeloid cells (47). Ccl6 is a potent chemoattractant for macrophages and other immune cells in rodents (60). These 3 CCLs share a common chemokine receptor, CCR1 (47), expressed by myeloid cells and implicated in myeloid cell recruitment in rheumatoid arthritis (61) as well as myeloma and colon cancer (62, 63). CCR1 inhibition reduced myeloid infiltrates and reduced PanIN progression, supporting the notion that myeloid-suppressive cells are a key mediator of immune suppression in pancreatic cancer (11–13, 18). Treg depletion also unleashed pathologic T-cell responses that we attributed to a Th2-type response based on the cytokine profile of the tissue. Th2 cells, as well as Th17 cells, have been implicated in the promotion of pancreatic carcinogenesis (23, 49). Thus, several parallel mechanisms might explain the paradoxical promotion of carcinogenesis upon Treg depletion, and support the notion that Tregs might not be a key mediator of immune suppression of pancreatic cancer (64).

In summary, our work reveals a complex cellular cross-talk within the neoplastic pancreas that includes multiple mechanisms to induce immunosuppression. Currently, CCR2 and CXCR2 inhibition is being tested in clinical trials for pancreatic cancer, based on positive results in preclinical models (65–67). Based on our results, CCR1 blockade should be considered as an additional potential method to inhibit myeloid cell infiltration in the pancreas.

KC;Foxp3^DTR mice were generated by crossing KC (Ptf1a-Cre;LSL-Kras^G12D; ref. 8) with Foxp3^DTR mice (B6.129(Cg)-Foxp3^{tm3(DTR/GFP)Ayr}/J, Jackson Laboratory; ref. 34). KC;CD4^−/− mice were generated by crossing KC with CD4-deficient mice (B6.129S6-Cd4^tm1Knw/J, Jackson Laboratory).

For Treg depletion, Foxp3^DTR and KC;Foxp3^DTR mice were treated with DT (50 ng/g; Enzo Life Science) by intraperitoneal (i.p.) injection. DT injections were repeated according to the specific experimental design shown in figures. Control mice lacking Foxp3^DTR alleles received the same DT treatments. In KC, KC;Foxp3^DTR, and KC;CD4^−/− mice, mild acute pancreatitis was induced in 4- to 5-week-old mice by a series of 8 hourly i.p. injections of caerulein (Sigma-Aldrich, 75 μg/kg) over a 1-day period. For CCR1 inhibition, mice were subcutaneously dosed with the CCR1 antagonist BX471 (Sigma-Aldrich, 50 mg/kg) for 7 days at 12-hour intervals. DMSO was used as vehicle to dissolve BX471 at a concentration of 50 mg/mL.

To establish the orthotopic pancreatic cancer model, 1 × 10⁵ 7940B cells (C57BL/6J strain; ref. 46) derived from KPC tumor (Ptf1a-Cre;LSL-Kras^G12D; Trp53^flox/+) were injected into Foxp3^DTR mice of compatible genetic background. Cells were tested for Mycoplasma by the MycoAlertTM PLUS Mycoplasma Detection Kit (Lonza) and passages 15 to 20 were used for all experiments. For CD8⁺ T-cell depletion, anti-CD8 mAb (Bio X Cell clone 2.43; 200 μg/mouse) was injected i.p. twice per week. For CD4⁺ T-cell depletion, anti-CD4 mAb (Bio X Cell clone GK1.5; 200 μg/mouse) was injected i.p. at least every 3 days.

Mouse bone marrow cells were treated with L929 conditioned media for 5 days for macrophage differentiation; primary mouse pancreatic cancer cell iKras#1 (18) conditioned medium was used to further educate macrophages. In addition, LPS or IL4 was used to polarize differentiated macrophages as previously described (68).

Tissues were fixed overnight in 10% neutral-buffered formalin, embedded in paraffin, and sectioned. Hematoxylin and eosin (H&E), PAS, Gomori's trichrome, IHC, and immunofluorescence staining were performed on formalin-fixed, paraffin-embedded mouse pancreatic tissues as described previously (17). Antibodies used are listed in Supplementary Table S1. For immunofluorescence, Alexa Fluor (Invitrogen) secondary antibodies were used. Cell nuclei were counterstained with Prolong Diamond Antifade Mountant with DAPI (Invitrogen). In addition, TSA Plus Fluorescein system (PerkinElmer) was used when double immunofluorescence staining with primary antibodies raised in the same species. Images were taken using Olympus BX53F microscope, Olympus DP80 digital camera, and CellSens Standard software. The confocal images were acquired using Olympus IX-71 confocal microscope with FluoView FV500/IX software. Quantitative analysis for immunofluorescence staining was performed in at least 3 random nonoverlapping fields (200 × magnification) in each sample using ImageJ software to measure the percentage of positive area. To quantify the positive cell number, we used confocal images at 600× magnification. At least 3 samples per group were analyzed. Histopathologic quantification for H&E staining was performed by a pathologist (W. Yan) on deidentified images as previously described (69).

A formalin-fixed paraffin-embedded tissue microarray containing human PDA, PanIN, and chronic pancreatitis was acquired from the University of Michigan. Blocks were sectioned onto slides, creating a 5-μm thickness per core. The slides were then processed using the Opal 7 manual kit (PerkinElmer). Images were captured on the Mantra Quantitative Pathology Work Station (PerkinElmer) and analyzed using inForm Cell Analysis software (PerkinElmer), as previously described (30). For detailed methods, see Supplementary Methods (complex phenotype design for multispectral imaging analysis in Supplementary Table S2).

Single-cell suspensions of pancreata were prepared and stained with fluorescently conjugated antibodies as previously described (17). Flow-cytometric analysis was performed on a Cyan ADP analyzer (Beckman Coulter) and data were analyzed with FlowJo v10 software. FACS was performed using MoFlo Astrio (Beckman Coulter). Antibodies are listed in Supplementary Table S1.

Single-cell suspensions of mouse pancreata were prepared as previously described (17). Human patient tissues from core needle biopsy or surgery were immediately placed into DMEM supplemented with Y27632 (Rho-Kinase inhibitor) for transport to the laboratory. Whole blood was collected preoperatively into two 10-mL EDTA tubes and transported to the laboratory. Tissues were mechanically minced and enzymatically digested with collagenase P (1 mg/mL DMEM) and subsequently filtered through a 40-μmol/L mesh to obtain single cells. Up to 1 × 10⁷ cells were stained with Cell-ID cisplatin (1.67 μmol/L) for 5 minutes at room temperature, and then a Combining Fix and Perm-Sensitive Surface Epitopes and Nuclear Antigen Staining protocol was followed according to the manufacturer's instructions (Fluidigm) for mouse samples, and as previously described (70). Analysis was performed using the Premium CytoBank Software (cytobank.org). Additional details are provided in the Supplementary Methods.

Cell lysates of FACS-sorted cells were homogenized using QIAshredder, then total RNA samples were isolated using the RNeasy Plus Micro Kit (Qiagen), including an on-column DNase treatment using RNase-Free DNase Set (Qiagen) according to the manufacturer's instructions. RNA concentration and quality were determined, and strand mRNA libraries were prepared by the University of Michigan Sequencing Core. Libraries were sequenced using paired-end 50 cycle reads on a HiSeq 4000 (Illumina). Raw data are available at the NCBI's Gene Expression Omnibus database (GSE120395 and GSE128707).

The raw data were processed and analyzed by the University of Michigan Bioinformatics Core. The quality control was done using FastQC (version v0.11.3, http://www.bioinformatics.bbsrc.ac.uk/projects/fastqc/) for both pre- and post-alignment, and raw reads were aligned to the mouse reference genome of UCSC mm10 (http://genome.ucsc.edu/) using Bowtie2 (version 2.2.1) and TopHat (version 2.0.13) of the Tuxedo suite. Gene-expression quantitation and differential expression analysis were performed using HTSeq (version 0.6.1) and DESeq2 (version 1.14.1). Differentially expressed genes were defined by false discovery rate (FDR) of 0.05 and fold changes of 1.5 or more. For the secretome data, we used the intersection of MetazSecKB (curated secreted) and UniProt (secreted, reviewed, https://www.uniprot.org/) as of November 2017. Plots and heat maps were generated in R/Bioconductor.

An orthotopic pancreatic cancer model was established by injecting 5 × 10⁴ of iKras*F1 cells (FVB/N strain) derived from iKras*;p53* tumor (Ptf1aCre;TetO-Kras^G12D;R26^{rtTa-IRES-EGFP};Trp53^R172H/+; ref. 32) into wild-type mice of compatible genetic background. PanIN lesions were induced in iKras* (Ptf1aCre;TetO-Kras^G12D;R26^{rtTa-IRES-EGFP}) mice by caerulein and doxycycline as previously described (31). Single-cell suspensions of pancreatic tumors were derived as previously described (17). Dead cells were removed using the MACS Dead Cell Removal Kit (Miltenyi Biotec Inc.). Single-cell cDNA library was prepared and sequenced at the University of Michigan Sequencing Core using the 10x Genomics. Samples were run using paired-end 50 cycle reads on HiSeq 4000 (Illumina) to the depth of 100,000 reads.

The raw data were processed and analyzed by the University of Michigan DNA Sequencing Core. R package Seurat version 2.3.4 was used for single-cell RNA-seq data analysis similarly as previously described (71). Data were initially filtered to only include all cells with at least 200 genes and all genes in greater than 3 cells. Data were initially normalized using the NormalizeData function with a scale factor of 10,000 and the LogNormalize normalization method. Variable genes were identified using the FindVariableFeatures function. Data were assigned a cell-cycle score using the CellCycleScoring function and a cell-cycle difference was calculated by subtracting the S phase score from the G₂–M score. Data were scaled, centered, and batch corrected using linear regression on the counts, the cell-cycle score difference and run ID using the ScaleData function. PCA was run with the RunPCA function using the previously defined variable genes. Violin plots were then used to filter data according to user-defined criteria. Cell clusters were identified via the FindClusters function using a resolution of 1.2 for all samples, and t-distributed stochastic neighbor embedding (t-SNE) clustering algorithms were performed. A FindMarkers table was created, and clusters were defined by user-defined criteria. Raw data are available at the NCBI's Gene Expression Omnibus database (GSE 140628).

The human pancreatic adenocarcinoma RNA-seq data from TCGA were downloaded from cBioPortal (https://www.cbioportal.org/datasets). The tumor samples were ranked based on FOXP3 expression and assigned into FOXP3-low (n = 89) or FOXP3-high (n = 88) groups. The differentially expressed genes between the 2 groups were determined using limma bioconductor package with voom function in R software (https://www.r-project.org/).

Total RNA samples were isolated from frozen pancreatic tissues using the PureLink RNA Mini Kit (Invitrogen) or from cells using the RNeasy Plus Micro Kit (Qiagen). The High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems) was used to reverse-transcribe total RNA into cDNA. Quantitative PCR was prepared with 1X SYBR Green PCR Master Mix (Applied Biosystems) and various primers (primer sequences are listed in Supplementary Table S3). All primers were optimized for amplification under reaction conditions as follows: 95°C 10 minutes, followed by 40 cycles of 95°C 15 seconds and 60°C 1 minute. Melt curve analysis was performed for all samples after the completion of the amplification protocol. Cyclophilin A was used as the housekeeping gene-expression control.

GraphPad Prism 7 software was used for all statistical analyses. All data were presented as means ± standard error (SEM). Intergroup comparisons were performed using a two-tailed unpaired t test, and P < 0.05 was considered statistically significant. Pearson correlation coefficients were used to measure R and R². Comparison between genotypes using the individually defined immune subpopulations was corrected for multiple comparisons with the Benjamini–Hochberg method with an FDR of 0.1.

All animal studies were conducted in compliance with the guidelines of the Institutional Animal Care and Use Committee at the University of Michigan. Patient selection/sample procurement: Patients over the age of 18 referred for diagnostic endoscopic ultrasound of a pancreatic mass lesion suspected to be PDAC were consented according to IRB HUM00041280. Up to 2 extra passes were taken for research after biopsy obtained for clinical use. Surgical specimens were obtained from patients referred for Whipple or distal pancreatectomy according to IRB HUM000025339. Written informed consent forms were obtained from the patients, and the studies were conducted in accordance with recognized ethical guidelines. Human patient studies were approved by Institutional Review Boards of the University of Michigan Medical School.

M.A. Anderson is a consultant for Boston Scientific and Olympus, is on the Data Safety and Monitoring Committee of GlaxoSmithKline, is a consultant/advisory board member for ASGE, and has received other remuneration from Boehringer Ingelheim. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y. Zhang, M. Pasca di Magliano

Development of methodology: Y. Zhang, J. Lazarus, M.A. Anderson, F. Bednar

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Zhang, J. Lazarus, N.G. Steele, W. Yan, C.J. Halbrook, R.E. Menjivar, V.R. Sirihorachai, A. Velez-Delgado, K. Donahue, E.S. Carpenter, K.L. Brown, V. Irizarry-Negron, A.C. Nevison, A. Vinta, M.A. Anderson, C.A. Lyssiotis, T.L. Frankel

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Zhang, J. Lazarus, W. Yan, H.-J. Lee, Z.C. Nwosu, C.J. Halbrook, S.B. Kemp, V.R. Sirihorachai, A. Velez-Delgado, K. Donahue, K.L. Brown, A.C. Nevison, A. Vinta, T.L. Frankel, F. Bednar, M. Pasca di Magliano

Writing, review, and/or revision of the manuscript: Y. Zhang, J. Lazarus, N.G. Steele, H.-J. Lee, V.R. Sirihorachai, K. Donahue, M.A. Anderson, H.C. Crawford, C.A. Lyssiotis, T.L. Frankel, F. Bednar, M. Pasca di Magliano

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Zhang, H.C. Crawford, F. Bednar

Study supervision: Y. Zhang, M. Pasca di Magliano

We thank Dr. Gregory Beatty (University of Pennsylvania) for generously sharing the primary mouse pancreatic cancer cell line 7940B. The Ptf1a-Cre mouse was a generous gift from Dr. Chris Wright (Vanderbilt University). The CK19 antibody was obtained from the Iowa Developmental Hybridoma Bank. We thank Carlos Espinoza, Sion Yang, and Jeanine M. Ruggeri for technical support. Special thanks to Daniel Long for prompt histopathology services. This work was supported by the Cancer Moonshot Initiative (U01CA-224145) to M. Pasca di Magliano and H.C. Crawford. This project was also supported by NIH/NCI grants R01CA151588 and R01CA198074, the University of Michigan Cancer Center Support Grant (NCI P30CA046592), the American Cancer Society to M. Pasca di Magliano, and NCI-R50CA232985 to Y. Zhang. This project was also supported by 3-P30-CA-046592-28-S2 (Administrative Supplement to the Cancer Center Core Grant) to H.C. Crawford and M. Pasca di Magliano. F. Bednar was funded by the Association of Academic Surgery Joel Roslyn Award. T.L. Frankel was funded by K08CA201581. N.G. Steele is a recipient of the American Cancer Society Postdoctoral Award. Z.C. Nwosu was supported by the Michigan Postdoctoral Pioneer Program, University of Michigan Medical School. R.E. Menjivar was supported by T-32-GM007315; S.B. Kemp was supported by T32-GM113900; N.G. Steele, C.J. Halbrook, V.R. Sirihorachai, and K. Donahue were supported by T32-CA009676. A. Velez-Delgado was supported by the Rackham Merit Fellowship Program. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. This work was made possible by the Tissue and Molecular Pathology and Flow Cytometry Shared Resources at the Rogel Cancer Center and the University of Michigan DNA Sequencing Core. CyTOF samples were run at the University of Rochester Medical Center Flow Cytometry Shared Resource and at the Indiana University Simon Cancer Center Flow Cytometry Service.