CSF1/CSF1R Blockade Reprograms Tumor-Infiltrating Macrophages and Improves Response to T-cell Checkpoint Immunotherapy in Pancreatic Cancer Models

Pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal human malignancies. Current therapies are ineffective at treating late stage disease. The few durable responses to therapy seen in patients with PDAC are often associated with significant cytotoxic lymphocyte (CTL) infiltration into tumor tissue, suggesting that effective immunotherapy would hold promise to improve patient outcome (1, 2). However, attempts to use immunotherapeutics as single agents have achieved only limited clinical success (3, 4). Although multiple factors can contribute to the resistance of PDAC to immunotherapies, one dominant player is the presence of a suppressive immune microenvironment. Critical drivers of this immunosuppressive microenvironment include tumor-associated macrophages (TAM), monocytic myeloid-derived suppressor cells (Mo-MDSC), and granulocytic MDSCs (G-MDSC). These leukocytes can also promote tumor cell proliferation, confer resistance to cytotoxic stress, and facilitate metastatic dissemination (5, 6). Therefore, high numbers of tumor-infiltrating myeloid cells often correlate with early local or metastatic relapse, leading to poor survival in patients with pancreatic cancer (7–9). Therapeutics that can reprogram these myeloid responses might overcome immunosuppression to enhance responses to immunotherapy. Previous work by our group and others demonstrated that combining cytotoxic chemotherapy with the blockade of colony-stimulating factor 1 receptor (CSF1R), which is prominently expressed by monocytes, Mo-MDSCs, and macrophages, results in improved antitumor T-cell responses (10–12). These data suggest that CSF1R blockade could be effective at alleviating local tumor-induced immune suppression and bolstering the response to immunotherapy.

In this report, we investigate the mechanisms by which inhibition of CSF1R signaling alleviates immune suppression. We demonstrate that CSF1/CSF1R blockade not only decreases the number of TAMs, but also reprograms remaining TAMs to support antigen presentation and bolster T-cell activation within the tumor microenvironment. This in-turn leads to reduced immune suppression and elevated interferon responses, which restrain tumor progression. However, in response to reduced immune suppression, programmed death 1 ligand 1 (PDL1) is upregulated on tumor cells and cytotoxic T lymphocyte antigen 4 (CTLA4) on T cells. These checkpoint molecules limit the potential of CSF1R inhibition to stimulate antitumor immunity. Both programmed cell death protein 1 (PD1) and CTLA4 antagonists demonstrate limited ability to restrain PDAC growth in this mouse model, similar to reported efficacy as single agents in patients with PDAC (3, 4). However, CSF1R blockade overcomes these limitations to achieve regression in even well-established tumors. These data suggest that reprogramming myeloid cell responses via CSF1/CSF1R blockade could improve the efficacy of checkpoint-based immunotherapeutics.

Tissue microarray (TMA) studies were conducted on surgically resected PDAC specimens from 60 patients diagnosed in the Department of Pathology at Washington University (St. Louis, MO). Patients underwent pancreaticoduodenectomy followed by adjuvant chemotherapy. Fifty-nine of the 60 patients did not receive neoadjuvant therapy. To assemble TMAs, clearly defined areas of tumor tissue were demarcated and two biopsies (1.0-mm diameter) were taken from each donor block. The Washington University School of Medicine ethics committee approved this study. Fully automated image acquisition was performed using an Aperio ScanScope XT Slide Scanner system with a ×20 objective (Aperio Technologies) to capture whole-slide digital images. Fluorescent staining analysis was performed using MetaMorph software.

Tissues were fixed in 10% formalin, embedded in paraffin, and dehydrated in 70% ethanol. Of note, 5-μm-thick sections were deparaffinized in xylene, rehydrated in graded ethanol, and subjected to antigen retrieval by steam heating in Citra antigen retrieval solution (BioGenex). CSF1 was stained with clone 2D10 at 1:100 (Thermo) and detected using indirect immunofluorescence.

KC cells were derived from PDAC tumor tissue obtained from p48-CRE/LSL-KRas/p53^flox/flox mice (backcrossed C57/B6, n = 6 by speed congenic) by our laboratory. Kras-INK (KI) cells were obtained from Dr. Hanahan's laboratory (13, 14). All cell lines were negative for MAP and mycoplasma. Subsets of these cells were labeled with a polycistronic click beetle red luciferase-mCherry reporter.

Syngeneic orthotopic PDAC tumors were established by surgical implantation, as previously described (15). Briefly, we injected 200,000 cells in 50 μL Matrigel (BD Biosciences) into each mouse's pancreas. Cohorts of mice were randomized into different treatment groups by either bioluminescence imaging on day 12 or gross palpation of the pancreas. Mice were treated with 50 mg/kg gemcitabine (GEM; Hospira) by intravenous (i.v.) injection into the right retro-orbital sinus every 4 to 5 days. Preclinical studies were conducted with 10 to 15 10-week-old female mice per group. Tumor burden was measured by establishing gross wet weight of the pancreas/tumor and comparing it with that of 5 parallel mice sacrificed at the beginning of treatment. All studies involving animals were approved by the Washington University School of Medicine Institutional Animal Studies Committee.

CSF1 neutralizing antibody (clone 5A1, BioXCell) was administered via intraperitoneal (i.p.) injection every 4 to 5 days, with the first injection containing 1 mg and subsequent injections 0.5 mg. CSF1R inhibitors (CSF1Ri) were provided by Plexxikon Inc. PLX3397 is a selective bispecific inhibitor for c-Fms and the c-Kit receptor tyrosine kinases (12, 16–17). GW2580 has been described in detail previously (18). Both GW2580 and PLX3397 were administered at 800 mg/kg in chow. CTLA4 and PD1 antagonists (clones UC10-4F10 and RMP1-14, BioXCell) were given every 4 to 5 days at 250 and 200 μg/dose, respectively.

Single-cell suspensions were prepared from dissected pancreatic tumors by manual mincing using a scalpel, followed by enzymatic digestion with 3.0 mg/mL collagenase A (Roche) and DNase I (Sigma) for 30 minutes at 37°C with constant stirring. Digestion mixtures were quenched by 10% FBS, and filtered through 40-μm nylon strainers (Fisher Scientific). Cells were incubated for 10 minutes at 4°C with rat anti-mouse CD16/CD32 mAb (eBioscience) at 1:200 dilution. Cells were washed twice in PBS/BSA and incubated for 20 minutes with 100 μL of fluorophore-conjugated anti-mouse antibodies [CD3e (145-2C11), CD4 (GK1.5), CD8a (53-6.7), CD11b (M1/70), CD11c (N418), CD19 (MB19-1), Ly6C (HK1.4), CD45 (30-F11), CD115 (AFS98), F4/80 (BM8), MHCII (M5/114.15.2), FoxP3 (FJK-16s), CD44 (IM7), CD69 (H1.2F3), PD1 (J43), PDL1 (MIH5), PDL2 (122), CTLA4 (UC10-4B9), IgG2α/κ (eBR2a)], (all from eBioscience) and/or Ly6G (1A8, BioLegend), and CD206 (MR5D3, AbDSerotec) using the manufacturers' recommended concentrations. Data acquisition was performed on the LSR-II system (BD Biosciences) and FlowJo software version 9.2 (Tree Star) was used for analysis.

Additional details are in the Supplementary Data.

Previously, we reported that inhibition of CSF1/CSF1R signaling could improve the efficacy of chemotherapy in murine PDAC models by enhancing chemotherapy-induced antitumor immunity (11). However, the mechanisms by which inhibition of CSF1/CSF1R signaling regulates antitumor immunity are not well understood. To determine the cellular sources of CSF1 and CSF1R in human pancreatic cancer patients, we analyzed TMAs constructed from 77 cases of invasive PDAC and four samples of normal pancreatic tissue. IHC staining showed that CSF1 is frequently, but not exclusively, expressed by malignant PDAC cells (Fig. 1A). In addition, tumors frequently had elevated expression of CSF1 compared with normal tissue. PDAC cells in 70% of tumor specimens exhibited moderate to high levels of CSF1 expression (Fig. 1A–C). In contrast, CSF1R was frequently detected in the tumor stroma, whereas only approximately 10% of the tumors examined had CSF1R expression in the epithelial compartment (Fig. 1A and D). These observations are consistent with other reports (19, 20) and suggest that PDAC tumor cells frequently produce high levels of CSF1.

To understand the impact of CSF1R signaling on the tumor microenvironment, we compared the gene expression profile of PDAC tumor tissue following treatment with either CSF1R inhibitors (CSF1Ri) or vehicle. Toward this end, we orthotopically implanted KI PDAC tumor cells into syngeneic mice. This cell line produces high levels of CSF1 but does not express CSF1R (11). Starting on day 14 postimplantation, we treated mice with either vehicle or the CSF1R tyrosine kinase inhibitor, PLX3397. Additional details on PLX3397 can be found in the Materials and Methods section and published elsewhere (16, 17, 18, 21). Eight days of CSF1Ri treatment resulted in a significant reduction in the number of tumor-infiltrating CD11b⁺Ly6G⁻Ly6C^LoF4/80^HiMHCII⁺ macrophages and CD11b⁺Ly6G⁻Ly6C^Hi monocytes/Mo-MDSCs, but not CD11b⁺Ly6G⁺Ly6C⁺MHCII^Low G-MDSCs (Fig. 2A and Supplementary Fig. S1). Microarray analyses of whole-tumor tissue mRNA expression revealed 204 downregulated and 158 upregulated genes following CSF1Ri treatment (Fig. 2B and Supplementary Table S1). As expected, expression of genes indicative of macrophage infiltration, including Cd68, Mrc1, Msr1, and Csf1r, was decreased in CSF1Ri-treated tumors (Fig. 2D). The list of downregulated genes was enriched for molecules involved in “inflammatory responses, chemotaxis, myeloid leukocyte-mediated immunity, and proteolysis,” consistent with the decreased number of infiltrating macrophages (Fig. 2C and D). The list of upregulated genes was enriched for molecules involved in “antigen presentation, allograft rejection, interferon responses, and T_H1 immunity” (Fig. 2C). This is consistent with the idea that CSF1R blockade can overcome immune suppression. Corresponding to these altered pathways, genes indicative of CTL responses (Ifng, Cd3e, Cd8a, and Prf1), T-cell recruitment (Cxcl10, Ccl3, and Ccl4), and IFN responses (e.g., Ifng, Stat1, Irf1, and Irf9) were upregulated (Fig. 2E). Array results were also validated by qRT-PCR on a second set of samples (Fig. 2F). To determine the impact of these alterations, we applied these gene lists to existing gene expression datasets from patients with PDAC (22) and found that the core elements of the downregulated gene list were indicative of poor clinical outcomes (Fig. 2G). Taken together, these results suggest that: (1) inhibition of CSF1R signaling in the stromal compartment decreases myeloid responses and reprograms the tumor microenvironment to support T-cell–mediated antitumor immunity and (2) these changes could improve patient outcomes.

To determine how inhibition of CSF1/CSF1R signaling impacts myeloid responses, we treated tumor-bearing mice with CSF1 neutralizing antibodies for 6, 12, 24, or 48 hours or 8 days and analyzed tumor-infiltrating myeloid cell composition and cell death at these time points. Within the first 6 hours of αCSF1 treatment, total TAM numbers began to decrease. By 8 days, TAM numbers had decreased by approximately 60% (Fig. 3B). TAMs are a heterogeneous population of macrophages with diverse biologic activities (23–27). Although classical activation of macrophages can restrain cancer development, alternative activation often plays a protumorigenic role (28, 29). Distinct surface markers have been used to distinguish between classically and alternatively activated macrophages. Murine PDAC tumors contain a distinct subset of CD206^Hi TAMs (Fig. 3A and Supplementary Fig. S1), and their counterparts in human pancreatic cancer have been associated with poor clinical outcomes (7). Quantification of CD206^Hi and CD206^Low TAM subsets revealed that αCSF1 treatment for 8 days led to a >90% depletion of CD206^Hi TAMs, whereas CD206^Low TAMs decreased by only approximately 45% (Fig. 3C and D). Similar results were seen following CSF1Ri treatment (Fig. 3G). The loss of CD206^Hi TAMs could result from either preferential killing of this TAM subset or altered CD206 expression. To distinguish between these possibilities, we analyzed the kinetics of macrophage cell death. We found that in PDAC tumors, CD206^Hi TAMs experienced significantly higher levels of cell death following αCSF1 treatment than CD206^Low TAMs (Fig. 3D and E). These data suggest that CD206^Hi TAMs are more sensitive to the CSF1R signal blockade. Consistent with this differential sensitivity, we found that CD206^Hi TAMs express higher levels of CSF1R (Fig. 3F). In addition, although total Mo-MDSCs (CD11b⁺/Ly6G⁻/Ly6C⁺) did not demonstrate decreased infiltration until after 8 days of αCSF1 treatment, CD206^Hi Mo-MDSCs were markedly reduced as early as 12 hours after CSF1 neutralization (Supplementary Fig. S2A). In contrast, the number of CD206^Low Mo-MDSCs, CD11b⁺/Ly6G⁺/Ly6C⁻/MHCII⁺ mature granulocytes, and CD11b⁺/Ly6G⁺/Ly6C⁺ G-MDSCs remained unaffected until after 8 days of CSF1/CSF1R blockade (Supplementary Fig. S2B). Taken together, these data suggest that the blockade of CSF1/CSF1R signaling preferentially, but not exclusively, depletes CD206^Hi TAMs and CD206^Hi Mo-MDSCs in pancreatic tumors.

Despite extensive loss of macrophages and Mo-MDSCs, 40% to 50% of TAMs remain after αCSF1 or CSF1Ri treatment. To determine whether CSF1 blockade reprograms the remaining macrophages to support antitumor activities, we FACS sorted TAMs from 8-day vehicle or αCSF1-treated mice bearing established KI tumors and compared their gene expression profiles. TAMs from αCSF1-treated tumors displayed reduced expression of immunosuppressive molecules, including Pdcd1lg2, Il10, Arg1, Tgfb1, and Ccl22. In contrast, antitumor immunity genes, such as Il12a, Ifna, Ifnb1, Ifng, Cxcl10, and Nos2, were upregulated (Fig. 3H). We also observed markedly increased surface expression of MHCII after CSF1 or CSF1R inhibition (Fig. 3I). Taken together, these data suggest that the CSF1/CSF1R blockade reprograms remaining TAMs to support antitumor IFN responses and T-cell activities.

On the basis of the observed differences in cytokine profiles among TAMs, we predicted that CSF1/CSF1R blockade might also alter the ability of macrophages to suppress T-cell functions. To address this hypothesis, we assessed the immunosuppressive activity and antigen presentation capacity of macrophages in PDAC tumors from mice following CSF1 blockade. Consistent with the reduced expression of immunosuppressive factors (Fig. 3H), we found that FACS-sorted TAMs from 8-day αCSF1-treated mice had significantly reduced ability to block CD8⁺ T-cell activation in ex vivo assays (Fig. 4A). These data suggest that the TAMs that remain after CSF1 blockade have reduced immunosuppressive activity.

We also analyzed how CSF1 blockade might impact the number and function of antigen-presenting cells (APC) in the tumor microenvironment. To identify potential APCs in PDAC tumors, we orthotopically implanted mCherry-labeled KI tumor cells. This model allowed us to identify potential APCs by their uptake of tumor antigens, based on their mCherry fluorescence (Fig. 4B; ref. 30). We were able to detect tumor-derived mCherry signal in granulocytes, monocytes, TAMs, and dendritic cells (DC; Fig. 4B). The highest levels of mCherry uptake were observed in TAMs and a subset of CD11b^low/−/Ly6G/C⁻/CD19⁻/CD11c⁺/MHCII⁺ cells, presumably lymphoid-like DCs (LyDC). CSF1/CSF1R blockade did not affect mCherry uptake. Interestingly, unlike in TAMs, CSF1/CSF1R blockade significantly increased the number of tumor-infiltrating LyDCs and their surface expression of MHCII (Fig. 4C and Supplementary Fig. S2C–S2E). Because of the high level of tumor antigen uptake by TAMs and LyDCs, we tested the ability of these two cell types to present antigen to naïve CD8⁺ T cells and stimulate their proliferation. We isolated TAMs and LyDCs from orthotopic KC tumors obtained from mice treated with either vehicle or αCSF1 for 8 days. These leukocytes were then loaded with SIINFEKL peptide and assessed for their ability to activate OT1 T cells. Although macrophages and LyDCs isolated from vehicle-treated tumors had very limited ability to activate T cells, αCSF1 treatment significantly enhanced the capacity of these two cell types to induce CD8⁺ T-cell proliferation (Fig. 4D). Taken together, these data suggest that CSF1 blockade alleviates immunosuppressive activities and enhances APC potential in both TAMs and tumor-infiltrating LyDCs.

To further understand how the blockade of CSF1/CSF1R signaling might reprogram the tumor microenvironment to regulate tumor progression, we assessed alterations in tumor-infiltrating T lymphocytes and tumor growth following CSF1 or CSF1R blockade in established murine PDAC tumors. Mice bearing established (12 days, ∼1 cm) orthotropic KI or PAN02 tumors were treated with αCSF1 IgGs or CSF1Ri. Tumor progression was modestly reduced by αCSF1 or CSF1Ri treatment as a single agent (Fig. 5A–C). This reduction in tumor growth correlated with increases in CD3⁺CD8⁺ CTLs and CD3⁺CD4⁺ effectors T cells, decreases in CD4⁺ Foxp3⁺ T regulatory cells (T_Regs), and significantly improved effector-to-T_Reg ratios (Fig. 5D–E). Although the majority of tumor-infiltrating CD8⁺ CTLs had a CD69⁺, CD44⁺, and CD62L⁻ activated phenotype, CSF1R blockade led to a modest increase in both the number of CD69⁺ CD8⁺ T cells (65%–76%) and the level of CD44 expression (Fig. 5F). The observed increase in T-cell numbers and enhancement of activation status correspond to our results from gene expression profiling in Fig. 2.

Although the CSF1/CSF1R blockade enhanced T-cell infiltration, we hypothesized that antitumor immunity might be limited via the engagement of T-cell checkpoints. We found that approximately 70% of activated CTLs had a high level of PD1 expression, which was unaffected by CSF1R blockade. In contrast, CTLA4 expression on CD8⁺ CTLs was significantly upregulated by CSF1R inhibition (Fig. 5F). Along these lines, our array analysis (Fig. 2) showed that Cd274 (PDL1) was significantly upregulated following CSF1R blockade. We verified these results using qRT-PCR, and found that both Cd274 and Ctla4, but not Pdcd1lg2 (PDL2), are upregulated in tumor tissues following CSF1 or CSF1R blockade (Fig. 6A and B). These data suggest that although CSF1 blockade reprograms the tumor microenvironment to enhance effector T cell infiltration, engagement of T cell checkpoints is also enhanced.

To determine the cellular sources of these molecules, we analyzed PDL1, PDL2, and PD1 expression on tumor cells and tumor-infiltrating myeloid cells from vehicle- or CSF1Ri-treated mice. We found that TAMs expressed high levels of PD1, PDL1, and PDL2, but consistent with a decreased immunosuppressive capacity, tumor-infiltrating macrophages from CSF1Ri-treated mice had markedly decreased PDL2 and PD1 expression (Fig. 6C and 6F). CSF1Ri treatment also decreased the total number of PD1- and PDL2-positive TAMs (Fig. 6D and F). Similar effects were also seen with αCSF1 treatment (data not shown). Neither Mo-MDSCs nor G-MDSCs expressed significant levels of PDL2. Although CSF1R blockade did not alter PD1 or PDL1 expression in G-MDSCs, PDL1 expression was modestly elevated in Mo-MDSCs following CSF1Ri treatment.

Expression of PDL1, PD1, and PDL2 has been reported on human PDAC tumor cells, potentially allowing them to evade immune surveillance by suppressing T-cell function. To determine whether CSF1R blockade affects the expression of these molecules on PDAC cells, we used mCherry-expressing KI or KC cells to identify tumor cells in vivo. We found that both KI and KC cells express PDL1 at modest levels in vivo, but neither cell line expresses PDL2 or PD1 (Fig. 6C and 6F, and not shown). However, following CSF1 or CSF1R blockade, the number of PDL1⁺ tumor cells and overall expression level of PDL1 were markedly upregulated on PDAC tumor cells (Fig. 6C and 6E). These observations correspond with the increased mRNA levels of Cd274 identified by array analysis and qRT-PCR validation (Fig. 2 and 6A). Taken together, these results suggest that although CSF1/CSF1R blockade reprograms macrophage responses to bolster CTL responses, this reprogramming also leads to upregulation of PDL1 on tumor cells and CTLA4 on T cells. These checkpoints will likely limit the efficacy of observed antitumor immune responses.

On the basis of the above data, we hypothesized that CSF1 or CSF1R blockade could enhance PDAC responses to PD1- and/or CTLA4-antagonist–based immunotherapy. To assess this hypothesis, we treated mice bearing established KI tumors with αPD1 or αCTLA4 with or without CSF1Ri in combination with gemcitabine. PD1 and CTLA4 antagonists in combination with gemcitabine had only limited efficacy at blunting the progression of established tumors (Fig. 7A and B). In contrast, the combination of CSF1R blockade with either PD1 or CTLA4 antagonists reduced tumor progression by more than 90%. Because combined PD1 and CTLA4 antagonist therapy is being tested clinically for the treatment of both melanoma and PDAC, we also tried this combined therapeutic approach. In the absence of chemotherapy, even combined αPD1/αCTLA4 treatment only limited tumor progression by approximately 50%. However, the addition of CSF1R blockade to αPD1/αCTLA4 treatment completely blocked tumor progression and even regressed established tumors by 15% (Fig. 7C). When CSF1 blockade was combined with αPD1/αCTLA4 and gemcitabine treatment, we observed complete tumor regression in 30% of animals and an average tumor regression of approximately 85% (Fig. 7D). Similar results were seen in orthotopic KC tumors, and when the less potent CSF1R inhibitor, GW2850, was used (Fig. 7B and Supplementary Fig. S3A and S3B). Analysis of T-cell responses following combined therapy with αCSF1 and αPD1/αCTLA4 antagonists demonstrated increased CD8⁺ CTL and CD4⁺ effector T-cell infiltration and decreased CD4⁺ Foxp3⁺ T_Reg numbers (Fig. 7E). In addition, the number of TAMs, Mo-MDSCs, and G-MDSCs decreased following this combined therapeutic regimen (Fig. 7F).

To determine whether alterations in tumor burden in CSF1Ri treatment mice were due to increased T-cell responses, we conducted CD4 and CD8 T depletion studies and found that CSF1R blockade no longer improved checkpoint-based therapy (Fig. 7G). Taken together, these results suggest that CSF1/CSF1R blockade improve checkpoint immunotherapy by enhancing CD4⁺ and CD8⁺ T-cell activities.

In this report, we show that blockade of CSF1/CSF1R signaling in pancreatic tumors depletes CD206^Hi TAMs and reprograms remaining macrophages to support antitumor immunity. The blockade alone modestly enhances antitumor IFN responses, promotes CTL infiltration, and slows tumor progression. However, the therapeutic effect is limited by the induction of T-cell checkpoint molecules, including PDL1 on tumor cells and CTLA4 on T cells. Addition of the CSF1/CSF1R blockade markedly improved the efficacy of αPD1 and αCTLA4 checkpoint immunotherapy and led to the regression of even well-established PDAC tumors. These data suggest that CSF1/CSF1R signaling may be an effective therapeutic target to reprogram the immunosuppressive microenvironment of human PDAC tumors and enhance the efficacy of immunotherapy.

Recent data from several groups suggest that inhibition of CSF1R signaling alters the immunologic responses of tumor-infiltrating macrophages in several cancer types (10–12, 31–33). Mok and colleagues targeted CSF1R signaling using the compound PLX3397 in a murine melanoma model; PLX3397 treatment depleted >80% of TAMs, leaving behind a small population of MHCII^Hi macrophages (10). These effects led to increased efficacy of adoptively transferred T-cell–based therapies. These data agree with our report here. In addition, recent work by Pyonteck and colleagues has shown that blockade of CSF1R signaling, using the small-molecule inhibitor BLZ945, significantly blunts murine glioma tumor growth by reprogramming macrophage responses (31). In contrast with pancreas, melanoma, and breast models, macrophage numbers in these murine glioma studies were not reduced. Instead, TAM survival was sustained by tumor-derived factors. However, in glioma, CSF1R blockade impairs the tumor-promoting functions of TAMs and regresses established tumors. Taken together, these results suggest that CSF1/CSF1R signaling can regulate both the number and the function of TAMs, but these activities may be highly dependent on tumor-type/tissue-specific factors.

One possible mechanism by which CSF1Ri reprograms the remaining TAMs is that CSF1R signaling may promote tumor-promoting macrophage phenotypes, while its blockade polarizes TAMs into the antitumor phenotype. In a study by Fleetwood and colleagues, macrophages cultured in CSF1 or CSF2 demonstrated different cytokine profiles and transcription activity (34). For example, in response to lipopolysaccharide, CSF2-derived macrophages preferentially produce IL6, IL12, and TNFα, whereas CSF1-derived macrophages produce IL10 and CCL-2, but not IL12. These data suggest that the exact cytokine milieu differentially program macrophages to play diverse roles. Intriguingly PDAC tumors can also produce high levels of CSF2 (35, 36), which could reprogram TAMs toward DC-like phenotypes when unopposed by CSF1R signaling.

Alternative to TAMs being reprogrammed by CSF1Ri, another possible mechanism is that CSF1R signaling blockade selects for a subset of tumor-restraining macrophages that are insensitive to the CSF signal kills-off a subset of TAMs that have a protumor phenotype. In many physiologic and pathologic settings, including cancers, macrophages are composed of heterogeneous subsets of populations with distinct functions (23). These subsets may depend on different factors for their survival, proliferation, and effector functions. Selection pressure due to CSF1 signal blockade may have enriched for subsets of antitumor macrophages in PDAC tissue that are less dependent on CSF1 signaling for their survival. Our analysis of cell death in CD206^HiMHCII^Low versus CD206^LoMHCII^Hi TAM sensitivity to αCSF1 IgG supports this hypothesis (Fig. 3B). Although both CD206^Hi and CD206^Low TAM populations had detectable cell death upon CSF1 neutralization, the CD206^Hi populations were preferentially depleted. The CD206^Hi TAM subset had significantly higher CSF1R expression levels, suggesting that this population may be more dependent on the CSF1 signal. Taken together, the heterogeneity of macrophages within the tumor tissue suggests that subsets of TAMs can be targeted to modulate the tumor microenvironment and enhance tumor elimination.

CD206 is expressed in many subsets of myeloid cells other than macrophages, including immature dendritic cells and monocytes (37). Whether CD206 expression is correlated to differential activation status in these cell types is not known. Interestingly, Tie2⁺ monocytes almost uniformly express CD206 (38). It remains to be seen whether the loss of CD206^Hi tumor-infiltrating monocytes upon αCSF1 treatment (Supplementary Fig. S2A) involves the Tie2⁺ monocytes and/or affects tumor vasculature.

Although CSF1/CSF1R blockade enhances the antitumor activity of myeloid cells and T-cell responses, its efficacy can be blunted by upregulation of immune checkpoint molecules, especially PDL1. Although tumor intrinsic pathways have been reported to drive PDL1 expression in tumor cells (4), multiple lines of evidence suggest that PDL1 expression by epithelial tumors is an adaptive response to IFN signaling from tumor stroma. Several groups have reported that IFNγ and IFNα directly lead to the upregulation of PDL1 (39–42). Consistent with these studies, in vitro treatment with recombinant IFNγ markedly upregulated PDL1 expression in our PDAC cell lines (not shown). Given the elevated expression of IFNs and IFN response genes in CSF1Ri-treated PDAC tumor tissue, we reason that CSF1Ri-mediated IFN production might drive the upregulation of PDL1 in PDAC cells, an inherent limitation of this therapy.

Even though T-cell checkpoint inhibitors alone have achieved impressive clinical benefits in some other cancers, particularly melanoma (43, 44), their application in pancreatic cancer as single agents has had limited efficacy (3). This is potentially due to the immunosuppressive microenvironment of PDAC tissue, which could be alleviated by therapeutic strategies that reprogram dominant myeloid responses to allow for effective checkpoint therapy.

D.C. Linehan received commercial research support from Washington University/Pfizer Biomedical Collaborative. No potential conflicts of interest were disclosed by other authors.

Conception and design: Y. Zhu, D.C. Linehan, D.G. DeNardo

Development of methodology: Y. Zhu, D.G. DeNardo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Zhu, B.L. Knolhoff, M.A. Meyer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Zhu, J. Luo, D.C. Linehan, D.G. DeNardo

Writing, review, and/or revision of the manuscript: Y. Zhu, M.A. Meyer, B.L. West, J. Luo, A. Wang-Gillam, S.P. Goedegebuure, D.C. Linehan, D.G. DeNardo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Zhu, T.M. Nywening, D.G. DeNardo

Study supervision: D.G. DeNardo

Other (provided agents): A. Wang-Gillam

This work was supported by the Genome Technology Access Center, which is partially supported by NCI Cancer Center Support grant #P30 CA91842 and ICTS/CTSA grant #UL1RR024992. D.G. DeNardo acknowledges the generous support from a Lustgarten Innovation Award, an Edward Mallinckrodt Jr. Award, The Cancer Research Foundation, and a Siteman Cancer Center Career Development Award and R01 CA177670-01. D.C. Linehan acknowledges the Siteman Cancer Frontier Fund and NCI R01 CA168863-01. M.A. Meyer acknowledges funding from the Siteman Cancer Center Cancer Biology Pathway. A. Wang-Gillam acknowledges Washington University Clinical and Translational Grant KL2TR000450. T.M. Nywening acknowledges NCI grant T32 CA 009621.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.