An Immunosuppressive Dendritic Cell Subset Accumulates at Secondary Sites and Promotes Metastasis in Pancreatic Cancer

Pancreatic ductal adenocarcinoma (PDAC) is a particularly deadly form of cancer that often metastasizes to the liver or other organs before it is diagnosed, thereby eliminating the opportunity for surgical resection, which remains the only potentially curative treatment for this disease (1, 2). Indeed, only approximately 20% of patients present with resectable tumors and unresectable PDAC is largely unresponsive to existing therapies, contributing to the dismal 5-year survival rate of around 5% (1, 2). However, even with surgery or effective control of local disease, survival rates remain low due to distant recurrence or metastasis progression (3–6), underscoring the need for new strategies to prevent and treat metastasis in this cancer.

Elements of the immune system have been shown to both prevent and promote tumor progression across a range of malignancies (7–10), and certain tumor immune profiles (e.g., low Treg, high CD8 T-cell densities) are associated with favorable outcomes in human cancer (11, 12), including PDAC (13–15). In preclinical studies of PDAC, several myeloid and T-cell populations have been shown to exacerbate disease and others to suppress it (16–24). Fewer studies, however, have specifically investigated the immune response at metastatic sites and whether it can be manipulated to suppress tumor spread (25–28). Considering the poor performance of immunotherapies tested to date in PDAC (29, 30), a better understanding of the tumor-associated immune response is critical for the development of more effective treatments.

Genetically engineered mouse models (GEMM) have greatly enhanced our understanding of PDAC initiation and progression (31, 32), helping to identify critical tumor-intrinsic as well as -extrinsic factors. The latter include various immune cells and mediators that operate in different ways and stages to support disease progression (16–24). However, while some PDAC GEMMs eventually develop distant metastases, their highly variable rates and kinetics of metastasis preclude their use for a systematic study of the metastatic process. We recently developed an orthotopic mouse model of PDAC that resembles the human disease in its genetics, gross and microscopic appearance, and clinical manifestations (33). Most importantly, pancreatic tumors predictably metastasize to the liver in this immunocompetent model, permitting a systematic investigation of the metastasis-associated immune response. Recent studies have shown roles for different myeloid cell populations in metastatic seeding and outgrowth in various tumor models (7, 8, 34–36), but few have specifically investigated this process in PDAC (25–28). We therefore analyzed the livers of PDAC-bearing mice at early time points following tumor implantation to determine how the immune system affects metastasis development. Our results reveal a critical role for a phenotypically and functionally distinct population of dendritic cells (DC) in the spread of this intractable cancer.

The PDAC cell line, LMP, was established from liver metastases obtained from Pdx1-Cre; Kras^LSL-G12D/+; Trp53^LSL-R172H/+ mice on a B6/129 background, as described previously (33, 37). LMP cells stably labeled with tdTomato using a Sleeping Beauty transposon-based system were used for most experiments (38). PDA1-1 and PDA3-5 were established from primary tumors from Pdx1-Cre; Kras^LSL-G12D/+; Trp53^LSL-R172H/+ mice (33). The PDAC lines have been maintained in our laboratory since their establishment in 2008 and authenticated on the basis of mutant p53 expression by immunostaining. Panc02 and MC38 were obtained from the NCI DCTD tumor/cell line repository; and NIH/3T3, 4T1, B16-F10, CT26, and LLC from ATCC between 2008 and 2014. All cell lines were frozen down at early passages (<5), routinely tested for mycoplasma contamination (Lonza MycoAlert), and used within 5 passages after thawing. Cells were routinely cultured in DMEM (Gibco) supplemented with 10% heat-inactivated FCS, 2 mmol/L l-glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin.

Six- to 12-week-old B6129SF1/J (C57BL/6J x 129S1/SvImJ F1) mice were used as hosts in the LMP model unless otherwise indicated, and obtained from The Jackson Laboratory or bred in-house. C57BL/6J, Csf2rb KO, Foxp3^DTR, and B6.SJL-Ptprc^aPepc^b/BoyJ (CD45.1) mice were obtained from The Jackson Laboratory and used for generating bone marrow chimeras or crossed with 129S1/SvImJ mice to obtain tumor hosts. Mgl2^DTR mice on a C57BL/6 background (39) were crossed with 129S1/SvImJ mice to obtain tumor hosts. All procedures were approved by the Institutional Animal Care and Use Committee of Stanford University.

Orthotopic pancreatic tumors were established as previously described (33). Mice were injected in the pancreas with 2 × 10⁵ tdTomato-labeled LMP tumor cells suspended in growth factor-reduced Matrigel (BD/Corning) and used 3–4.5 weeks following tumor implantation unless otherwise indicated. Livers at this stage typically exhibited microscopic disease or small metastatic nodules. Normal livers were obtained from age-/sex-matched sham-operated or naïve mice. Details regarding tissue processing, cell isolation, and cell culture can be found in the Supplementary Materials and Methods. For experimental liver metastasis, mice were intrasplenically injected with 5 × 10⁵ tumor cells in PBS and analyzed at the indicated time points. C57BL/6J mice were used for studies with B16, LLC, MC38, and Panc02 cells. Unless otherwise indicated, metastatic burden was measured by fluorescence emission using an in vivo imaging system (Xenogen IVIS Lumina). Liver lobes were imaged on both sides using a DsRed filter set, and average total efficiency values, which correct for nonuniformity in illumination, were used to assess metastatic burden.

Cell suspensions were Fc-blocked (clone 93, BioLegend) prior to incubation with fluorescently conjugated antibodies and LIVE/DEAD fixable dead cell stains (Life Technologies) for 20 minutes on ice. Intracellular staining was performed using buffers for Foxp3 staining (eBioscience). Antibodies were obtained from BioLegend, eBioscience, and BD Biosciences (see Supplementary Materials). Data were acquired on a BD LSR II flow cytometer and analyzed using FlowJo. After gating on live CD45⁺ singlets, cell populations were defined as follows: PMN, CD11b⁺Gr1^hiCD11c⁻MHC-II⁻SSC^hi; inf-Mo, CD11b⁺Gr1^intCD11c⁻MHC-II⁻SSC^lo; CD11b⁺ DC, CD11b⁺CD11c^hiMHC-II^hi; CD11b⁻ DC, CD11b⁻CD11c^hiMHC-II^hi; KC/TAM, F4/80^hiCD11b^int; NK, NK1.1⁺CD3⁻; NKT, NK1.1⁺CD3⁺; CD4, NK1.1⁻CD3/CD90.2⁺CD4⁺; CD8, NK1.1⁻CD3/CD90.2⁺CD8α⁺; Treg, CD3/CD90.2⁺CD4⁺Foxp3⁺.

All statistical analyses were performed with GraphPad Prism. Unless otherwise indicated, two-tailed Student t tests were used to compare two groups, and one-way ANOVA with post hoc Tukey tests for multiple comparisons. Mann–Whitney U tests were used to compare non-normally distributed data where indicated, and medians are shown in the associated graphs. Values of P < 0.05 were considered significant.

Compared with tumor-naïve mice, CD11b⁺ myeloid cells began to accumulate in the liver of orthotopic pancreatic tumor (LMP)-bearing mice as early as 2 weeks following tumor implantation and further increased in frequency by 3.5 weeks (Fig. 1A and C), at which point mice predictably exhibit microscopic disease or small metastatic nodules in this model (33). Consistent with reports in other models (7, 8, 34, 35), the CD11b⁺ myeloid infiltrate included Gr1^hi (Ly6G⁺Ly6C^int) neutrophils (PMN) and Gr1^int (Ly6G⁻Ly6C^hi) inflammatory monocytes (inf-Mo; Fig. 1B and C; Supplementary Fig. S1A). These Gr1⁺ cell populations, sometimes referred to as granulocytic and monocytic myeloid-derived suppressor cells (MDSC), respectively, have been the focus of many previous studies of myeloid cells in metastasis (24–26, 34, 35). We also observed a surprising increase in Gr1⁻ (Ly6G⁻Ly6C^−/lo) myeloid cells that expressed high levels of the dendritic cell (DC) markers CD11c and MHC-II (Fig. 1B and C). A similar myeloid infiltrate was noted during the development of lung metastases in mice with subcutaneous PDAC tumors, indicating that this response occurs at multiple metastatic sites in pancreatic cancer (Supplementary Fig. S1B).

The abundant CD11b⁺CD11c^hiMHC-II^hi population was distinct from liver-resident Kupffer cells (KC) and/or tumor-associated macrophages (TAM), which expressed lower levels of CD11b and higher levels of F4/80 and other macrophage (Mφ) markers (Supplementary Fig. S1C and S1E). Indeed, similar to the CD11b⁻CD11c^hiMHC-II^hi population, CD11b⁺CD11c^hiMHC-II^hi cells in tumor-bearing mice expressed low levels of the Mφ markers F4/80 and CD64, and high levels of the DC marker CD24 (Fig. 1D; Supplementary Fig. S1D; refs. 40, 41). In contrast, F480^hiCD11b^int KCs/TAMs uniformly expressed CD64 but low levels of CD24 (Supplementary Fig. S1E). CD103 was expressed by a small proportion of CD11b⁺CD11c^hiMHC-II^hi cells but uniformly by the CD11b⁻CD11c^hiMHC-II^hi population (Fig. 1D; Supplementary Fig. S1D). These phenotypes are consistent with those described for CD11b⁺ and CD11b⁻CD103⁺ conventional DC subsets in normal (40) and tumor (41, 42) tissues, and distinct from those described for F4/80⁺CD24^loCD64⁺ Mφ/TAMs (41). Thus, we will henceforth refer to the CD11b^+/− CD11c^hiMHC-II^hi populations as CD11b⁺ and CD11b⁻ DCs.

The metastasis-associated myeloid subsets exhibited distinct microanatomical distributions. While CD11b⁺ cells accumulated throughout the liver of tumor-bearing mice (Supplementary Fig. S1F), micrometastases were typically bordered by dense networks of Gr1⁻CD11c⁺MHC-II⁺ cells (Fig. 1E; Supplementary Fig. S1G). In contrast, Gr1⁺CD11c⁻MHC-II⁻ cells were more evenly dispersed in tissue sections and less enriched in and around early metastases (Fig. 1E). To compare the myeloid subsets on a functional level, we analyzed their cytokine secretion profiles and found that CD11b⁺ DCs secreted the highest levels of most of the factors tested, including the protumoral mediators CCL2, CXCL1, CXCL2, IL6, and TNFα (Supplementary Fig. S1H; refs. 10, 43).

These observations prompted a more thorough analysis of the liver DC compartment during metastasis development. As expected, there was a marked increase in total CD11c^hiMHC-II^hi DCs in tumor-bearing mice, and the vast majority (∼80%) of these cells also expressed CD11b (Fig. 2A). This contrasts with normal liver DCs, which consist of roughly equal proportions of the CD11b⁺ and CD11b⁻ subsets (Fig. 2A). To assess potential functional differences among the DC populations, we analyzed the expression of various costimulatory/coinhibitory molecules and other maturation markers. DCs from tumor-bearing mice expressed lower levels of CD86 and higher levels of CD80, ICOSL, PD-L1, and PD-L2 relative to normal liver DCs (Fig. 2B). The selective expansion of CD11b⁺ DCs in tumor-bearing mice accounted for most of these changes, as these cells expressed less CD86 and more CD80, PD-L1, and PD-L2 relative to CD11b^– DCs (Fig. 2B and C). Both subsets expressed higher levels of ICOSL in tumor-bearing mice (Fig. 2B), while other maturation markers (4-1BBL, CCR7, CD40, RANK, OX40L) were minimally expressed under all conditions analyzed (data not shown). These data show that CD11b⁺ DCs richly associate with early metastases and express multiple immunosuppressive molecules and protumoral factors, suggesting that they may protect metastases from immune elimination and perform other protumoral functions. We thus decided to specifically study the ontogeny and function of this DC population.

A large fraction of metastasis-associated CD11b⁺ DCs also expressed the monocyte/macrophage marker CD115 (M-CSFR; Fig. 2B and C), suggesting that they could derive from monocytes. We tested this by adoptively transferring congenically marked (CD45.1) bone marrow monocytes from naïve mice into tumor-bearing mice. Prior to transfer, these cells expressed high levels of Gr1/Ly6C but not CD11c or MHC-II (Fig. 3A). However, when recovered from the liver after 5 days (Fig. 3B), more than 90% of the transferred cells expressed CD11c and MHC-II but not Gr1 (Fig. 3C and D), suggesting that Gr1⁺/Ly6C⁺ monocytes predominantly differentiate into CD11b⁺ DCs at metastatic sites. Like host CD11b⁺ DCs, the vast majority of donor-derived CD11b⁺CD11c⁺MHC-II⁺ cells expressed CD24, but not F4/80 or CD64 (Supplementary Fig. S2A and S2B). Similar results were obtained at earlier time points posttransfer (d3; data not shown), suggesting that monocytes directly traffic to and differentiate into CD11b⁺ DCs in the liver, although we cannot exclude the possibility of differentiation occurring elsewhere (e.g., primary tumor) prior to migration to the liver. We were unable to recover adequate numbers of cells for analysis from naïve mice (Fig. 3B), indicating that this recruitment and differentiation program is tumor-dependent. These results suggest that the large Gr1⁺/Ly6C⁺ monocyte population in tumor-bearing hosts is a major source of metastasis-associated CD11b⁺ DCs, although we cannot exclude the involvement of other potential precursors such as CD11c⁺Sirpα^int pre-cDCs, which have been shown to develop into CD11b⁺ DCs in other tumor models (42). In light of previous studies showing that Gr1⁺/Ly6C⁺ monocytes differentiate into more Mφ-like cells (F4/80⁺, CD64⁺) in other models (35, 36, 42), our results suggest that monocytes may exhibit a distinct differentiation pattern in PDAC.

To determine whether tumor-derived factors can drive monocytes to differentiate into CD11b⁺ DCs, we cultured bone marrow monocytes from naïve mice in the presence of tumor cell–conditioned medium (TCM) from the PDAC line (LMP) used throughout this study (33). LMP TCM treatment readily induced the formation of CD11c⁺MHC-II⁺ cells, while unconditioned or 3T3 fibroblast-conditioned media (3T3 CM) did not (Fig. 3E). Similar to metastasis-associated CD11b⁺ DCs, TCM-treated monocytes also expressed high levels of CD80, PD-L1, and PD-L2 (Fig. 3F). DC differentiation was even more efficient when monocytes were cocultured with LMP cells, with the majority of cells expressing CD11c and MHC-II after 2 days (Supplementary Fig. S2C). PDAC cells have been reported to produce large amounts of GM-CSF (16, 17), which is routinely used to generate monocyte-derived DCs. Consistent with this, we detected high levels of GM-CSF in LMP TCM, but not 3T3 CM (Supplementary Fig. S2D), and neutralizing GM-CSF inhibited DC differentiation and the upregulation of CD80, PD-L1, and PD-L2 in response to TCM treatment (Fig. 3G). To determine whether GM-CSF plays a role in CD11b⁺ DC development in vivo, we generated bone marrow chimeras with cells from WT (Csf2rb^+/+) or GM-CSFR KO (Csf2rb^−/−) mice, and used them as hosts in a model of experimental liver metastasis. Following intrasplenic injection of LMP cells, CD11b⁺ DCs accumulated in the liver of WT but not GM-CSFR KO bone marrow chimeras (Fig. 3H). Moreover, metastasis was markedly reduced in the KO bone marrow chimeras (Fig. 3I), reinforcing the pathophysiological function of GM-CSF in PDAC (16, 17).

These results collectively suggest that tumor-derived GM-CSF drives the accumulation of CD11b⁺ DCs with prometastatic functions. However, GM-CSF is a pleiotropic cytokine and has been shown to promote PDAC progression at the primary site through the expansion of Gr1⁺ myeloid cells (16, 17), which is consistent with our data showing trends toward reduced Gr1⁺ PMNs and inf-Mo in GM-CSFR KO bone marrow chimeras (Fig. 3H). We thus attempted to more specifically study the function of the metastasis-associated DCs in vitro and in vivo. To this end, we examined the effects of DCs from the liver of tumor-bearing mice (TLv-DC) on T-cell responses in conventional in vitro assays. Contrary to expectations, TLv-DCs more efficiently induced T-cell proliferation in response to polyclonal (Supplementary Fig. S3A) and antigen-specific (Supplementary Fig. S3B) stimuli, as well as in mixed lymphocyte reactions (data not shown), compared with normal liver DCs (NLv-DC). TLv-DCs stimulated more IFNγ and IL2 production under these conditions as well (Supplementary Fig. S3C). Despite these data suggesting that metastasis-associated DCs may be capable of inducing antitumor T-cell responses, this did not occur in vivo. Instead, the accumulation of CD11b⁺ DCs in the liver was paralleled by an increase in Tregs, with the frequency of Foxp3⁺ CD4 T cells nearly tripling in micrometastatic liver (Fig. 4A). This was accompanied by a poor CD8 T-cell response, leading to a doubling of the Treg:CD8 T-cell ratio (Fig. 4A). We also observed a profound enrichment of Tregs in the DC-rich areas bordering micrometastases, indicating that these cell populations interact in vivo (Fig. 4B). We detected a corresponding increase in Ki67⁺ Tregs (Fig. 4A) and colocalization of phosphorylated histone H3 and Foxp3 in DC-rich perimetastatic tissues (Supplementary Fig. S4A), suggesting that CD11b⁺ DCs may stimulate Treg proliferation in situ.

Consistent with the Treg-biased response observed in vivo, culturing syngeneic CD4 T cells with TLv-DCs in the absence of additional stimuli led to a doubling of Foxp3⁺ Tregs compared with T cells cultured with NLv-DCs or IL2 alone (Fig. 4C). These results were largely attributable to the selective proliferation of Tregs in coculture, with 70%–80% of Tregs undergoing cell division (Fig. 4C and D). Interestingly, Treg division rates were similar whether T cells were isolated from naïve or tumor-bearing mice, indicating that the T-cell response does not depend upon prior tumor exposure (Fig. 4E). In contrast, TLv-DCs poorly stimulated the proliferation of Foxp3⁻ CD4 and CD8 T cells from tumor-bearing mice (Fig. 4F). Treg differentiation from naïve CD4 T cells or purified Foxp3⁻ CD4 T cells (from Foxp3^GFP mice) was not induced under the same conditions (data not shown), suggesting that TLv-DCs selectively expand preexisting Tregs. Correspondingly, neutralizing TGFβ, which is critical for the development of induced Tregs (44), did not inhibit the Treg expansion but instead slightly enhanced it (Supplementary Fig. S4B), and more than 80% of Tregs in the liver of tumor-bearing mice expressed Helios (Supplementary Fig. S4C), a putative marker of natural or thymic Tregs (45).

Extending these results to more physiological settings, we observed spontaneous Treg proliferation when total nonparenchymal cells (NPC) from the livers of tumor-bearing mice (TLv-NPC) were cultured ex vivo in the absence of other stimuli (Fig. 4G; Supplementary Fig. S4D). In contrast, Tregs in cultures from naïve mice (NLv-NPC) exhibited poor survival and minimal proliferation (Supplementary Fig. S4D). Treg proliferation was markedly reduced when TLv-NPCs were depleted of either CD11b⁺ or CD11c⁺ cells, confirming a role for metastasis-associated CD11b⁺ DCs in this process (Fig. 4G). Furthermore, footpad injection of TLv-DCs induced an expansion of Tregs in draining compared with nondraining popliteal lymph nodes, demonstrating that these cells can expand Tregs in vivo (Fig. 4H).

We next attempted to clarify the molecular mechanisms involved in the Treg expansion mediated by metastasis-associated DCs. Blocking MHC-II largely abolished Treg proliferation (Supplementary Fig. S4E), as well as the low level of proliferation among Foxp3⁻ CD4 T cells, indicating a requirement for the presentation of tumor-derived or endogenous self-peptides. Tregs can proliferate in the absence of TCR-mediated signaling if provided with IL2 and appropriate costimulation (46). Consistent with this, supplementing cultures with IL2 partially rescued Treg proliferation under MHC-II blockade (Supplementary Fig. S4F). The costimulatory requirements for Treg proliferation were determined by blocking molecules highly expressed by CD11b⁺ DCs. Blocking CD80 significantly inhibited the Treg expansion (Supplementary Fig. S4G), and blocking ICOSL produced similar trends (data not shown). Surprisingly, antagonizing the PD-1 pathway with antibodies against PD-1, PD-L1, or PD-L2 failed to inhibit the Treg expansion in autologous pan-T-cell-DC cocultures from tumor-bearing mice and, instead, slightly enhanced it (Supplementary Fig. S4H). However, blocking PD-1 signaling in this context also roughly doubled CD8 T-cell division rates (Supplementary Fig. S4I), suggesting that this pathway can still be targeted to improve CD8 T-cell responses. Thus, metastasis-associated CD11b⁺ DCs preferentially stimulate Treg proliferation through the presentation of appropriate peptides and costimulatory molecules while suppressing CD8 T-cell responses through the expression of PD-1 ligands.

These findings suggest that CD11b⁺ DCs may facilitate metastasis by both expanding Tregs and suppressing circulating tumor cells. To determine how the Treg response at secondary sites affects metastasis development, we used Foxp3^DTR mice as hosts in the orthotopic tumor model, allowing pancreatic tumors to grow for 3 weeks before depleting Tregs for 2 weeks by weekly diphtheria toxin (DT) injections. Remarkably, the development of macroscopic liver metastases was nearly completely inhibited by Treg depletion despite the delayed depletion strategy (Fig. 4I). There was a trend toward smaller primary tumors in Treg-depleted mice (Supplementary Fig. S4J), which could account for some of the reduction in metastatic burden. However, Treg depletion also markedly reduced metastasis following intrasplenic tumor cell injection (Supplementary Fig. S4K). Although we cannot exclude a role for the system-wide immunostimulatory effects of Treg depletion in the results (47), these studies suggest that the Treg response supported by CD11b⁺ DCs at secondary sites facilitates metastasis development.

While seeking a way to specifically study metastasis-associated CD11b⁺ DCs in vivo, we discovered that most of these cells also expressed the C-type lectin MGL2 (CD301b; Fig. 5A), which was recently shown to be expressed by CD11b⁺ DC populations in multiple tissues (39, 48), including tumors (41). Importantly, MGL2 expression was almost completely restricted to CD11b⁺ DCs in the livers of tumor-bearing mice, with over 95% of MGL2⁺ cells also expressing CD11b, CD11c, and MHC-II (Fig. 5B and C). Similar to the total CD11b⁺ DC population, MGL2⁺ cells also expressed high levels of CD24, but not CD64 or F4/80 (Fig. 5B; Supplementary Fig. S5A). Furthermore, both spontaneous and experimental liver metastases were associated with dense networks of MGL2⁺CD11c⁺ cells (Fig. 5D; Supplementary Fig. S5B). CD45.1⁺ monocytes transferred into tumor-bearing mice acquired high levels of MGL2 expression, with over 70% of donor-derived CD11b⁺ DCs expressing MGL2 (Supplementary Fig. S5C). Moreover, TCM treatment of bone marrow monocytes induced the formation of MGL2⁺CD11b⁺ DCs in a GM-CSF–dependent manner (Supplementary Fig. S5D), suggesting that tumor-derived GM-CSF also drives the development of these cells. To determine whether MGL2⁺CD11b⁺ DCs contribute to metastasis in other models, we analyzed the liver 2 weeks after intrasplenic injection of B16, LLC, MC38, or LMP cells. Although all lines formed metastases, only LMP cells induced an accumulation of MGL2⁺CD11b⁺ DCs (Fig. 5E). Consistent with a role for tumor-derived GM-CSF in the development of the DCs, the other tumor lines produced low levels of GM-CSF compared with PDAC cells in vitro (Fig. 5F). Interestingly, like LMP, MC38 also induced an accumulation of CD11b⁺CD11c^hiMHC-II^hi cells (Supplementary Fig. S5E), but these cells expressed much lower levels of MGL2 compared with those elicited by LMP (Supplementary Fig. S5F), suggesting that alternative factors drive the development of the CD11b⁺ DC (or DC-like cells) in the MC38 model. On the basis of a recent study (42), these cells may differ in other ways from the CD11b⁺ DC population in our model, including higher expression of the monocyte/Mφ markers Ly6C and CD64.

To study the functional role of MGL2⁺CD11b⁺ DCs in PDAC metastasis, we used mice that express DTR under the control of the endogenous Mgl2 promoter as tumor hosts (Mgl2^DTR; ref. 39). We examined the effects of MGL2⁺ cell depletion in the experimental metastasis model, as MGL2⁺CD11b⁺ DCs also accumulated at the primary site in the orthotopic model, albeit at lower frequencies compared with the liver (Supplementary Fig. S5G). DT treatment effectively depleted MGL2⁺ DCs in tumor-bearing mice, reducing their frequency to <1% of liver NPCs (Fig. 5G). MGL2⁺ cell depletion beginning 10 days after tumor injection led to a marked reduction in metastatic burden (Fig. 5H), indicating an overall protumoral function for this DC population. Given their immunosuppressive features, we hypothesized that MGL2⁺ DC depletion reduces metastasis by removing restraints on antitumor immune responses. We tested this by examining the effects of cell depletion on the immune response to established metastases, allowing metastases to develop for 20 days before treating with DT or PBS for 5 days. Although this short treatment course at late-stage disease did not yield consistent changes in the frequencies of major lymphocyte subsets (Supplementary Fig. S5H), it stimulated the activation and proliferation of multiple populations, with NK, NKT, and CD8 T cells all showing significant increases in Ki67 expression (Fig. 5I). The percentages of CD69⁺ NK cells and granzyme B⁺ CD8 T cells also tended to increase in DT-treated mice (Supplementary Fig. S5I, J). Remarkably, MGL2⁺ cell depletion also reduced the frequency of Foxp3⁺ Tregs by approximately 40% (Fig. 5J) and nearly doubled the CD8:Treg ratio (Fig. 5K).

These studies suggest that CD11b⁺ DCs promote metastasis by supporting Treg responses and suppressing various immune effector cells, but the molecular mechanisms responsible for their immunosuppressive effects remain unclear. Considering their expression of both PD-1 ligands (Fig. 2B and C) and evidence that these molecules suppress T-cell responses in vitro (Supplementary Fig. S4H), we wondered whether CD11b⁺ DCs utilize this particular pathway to regulate immune responses in vivo. Although less studied than PD-1 and PD-L1, we focused on the role of PD-L2, as its expression was also restricted to CD11b⁺ DCs in the liver of tumor-bearing mice (Figs. 2B and C and 6A and B). Furthermore, PD-L2 staining was largely restricted to CD11c⁺ cells associated with metastases, whereas PD-L1 was also expressed in normal liver tissues (Fig. 6C).

We first tested the effects of PD-L2–blocking antibodies in the orthotopic model, allowing tumors to grow for 2 weeks before beginning antibody treatment for 3 weeks. Blocking PD-L2 did not affect primary tumor growth (Supplementary Fig. S6A), but significantly reduced liver metastasis (Fig. 6D). Remarkably, many αPD-L2–treated mice with large primary tumors exhibited minimal metastatic disease, eliminating the correlation between primary tumor weight and metastatic burden observed in control mice (Supplementary Fig. S6B). To determine whether these effects depend upon stimulation of tumor immunity, we analyzed immune cells in the liver after 10 days of treatment. Blocking PD-L2 did not significantly alter the frequencies of most major lymphocyte populations (Fig. 6E), including Foxp3⁺ Tregs (Fig. 6F), which is consistent with our in vitro studies. However, CD8 T cells tended to increase in αPD-L2–treated mice (Fig. 6E), resulting in a significant rise in the CD8:CD4 T-cell ratio (Fig. 6G). Blocking PD-L2 increased the proliferation of CD8 T cells, but not NK or CD4 T cells (Fig. 6H), further indicating selective activation of a CD8 T-cell response. We obtained similar results in the experimental metastasis model, wherein mice began antibody treatment 1 week following tumor injection (Fig. 6I). The trend toward reduced metastatic burden with αPD-L2 treatment was abolished in mice concomitantly treated with CD8 T-cell–depleting antibodies (Fig. 6I), suggesting that the effects of PD-L2 blockade depend upon induction of CD8 T-cell–mediated tumor immunity. Altogether, these studies suggest that CD11b⁺ DCs protect developing metastases from immune attack by expanding Tregs and suppressing CD8 T cells through a pathway involving PD-L2.

Extending these findings, MGL2⁺ DCs accumulated in and around liver metastases formed following intrasplenic injection of Panc02 cells (Fig. 7A) and in primary tumor tissues from a PDAC GEMM (Pdx1-Cre; Kras^LSL-G12D/+; Cdkn2a^−/−; Fig. 7B). Moreover, we observed large numbers of CD11c⁺HLA-DR⁺ cells associated with human PDAC liver metastases (Fig. 7C), particularly at the leading edges, where PD-L2⁺ stromal cells were also abundant (Supplementary Fig. S7A). In these same areas, we observed an accumulation of stromal cells expressing MGL (Supplementary Fig. S7B), the human homolog of murine MGL1/2.

Further supporting our studies in the mouse model, gene expression levels of MGL (CLEC10A) and PD-L2 (PDCD1LG2) correlate with each other (Fig. 7D) and with CD11b (ITGAM), CD11c (ITGAX), HLA-DR (HLA-DRA), and the human DC marker DC-SIGN (CD209) in human PDAC tissues (Fig. 7E). Foxp3 (FOXP3) expression is also highly correlated with both MGL and PD-L2 (Fig. 7E), suggesting that analogous DCs may support Treg responses in human PDAC. Consistent with the development of CD11b⁺ DCs in the mouse model, expression levels of MGL/PD-L2 correlate with monocyte markers (CCR2, CD14, CSF1R), PD-L1 (CD274), GM-CSF (CSF2) and its receptor (CSFR2A/B), and the transcription factor IRF4 (IRF4), which was shown to regulate the development of MGL2⁺PD-L2⁺CD11b⁺ DCs in normal (48) and tumor (41) tissues. Altogether, these data suggest that immunosuppressive DCs derived from monocytes are involved in PDAC progression in multiple mouse models as well as the human disease.

The prognosis for PDAC patients remains dismal (1, 2), largely due to presentation with inoperable locally advanced or metastatic disease as well as high rates of recurrence following surgery, especially at distant sites (3–5). Recent studies have begun to clarify how the immune system affects PDAC progression at the primary site (16–24). However, fewer have investigated the immune response associated with metastasis development at secondary sites. When studied, Mφ or Gr1⁺ myeloid cells have been emphasized in this process (24–28), similar to reports in other models (7, 8, 34, 35). While also finding that these cells accumulate at metastatic sites, we were surprised to find that early metastases are enriched with a population of CD11b⁺ DCs. However, rather than stimulating immunosurveillance mechanisms that suppress tumor spread, these antigen-presenting cells facilitate metastasis by generating an immunosuppressive microenvironment that is enriched with Tregs and by inhibiting CD8 T-cell responses (Supplementary Fig. S8A).

These findings were enabled by the highly restricted expression of both MGL2 and PD-L2 to CD11b⁺ DCs in the metastatic microenvironment. Indeed, we show here for the first time that PDAC metastasis is reduced by depleting a specific DC subset on the basis of MGL2 expression or by blocking PD-L2 (Supplementary Fig. S8B and S8C). Consistent with our studies, coexpression of MGL2/CD301b and PD-L2 has been found to characterize IRF4-dependent CD11b⁺ DCs in both normal (39, 48) and tumor (41) tissues. The effects observed upon MGL2⁺ cell depletion or PD-L2 blockade support our in vitro studies showing that CD11b⁺ DCs preferentially induce Treg proliferation and impair CD8 T-cell responses. Although our results suggest that multiple mechanisms may contribute to the antimetastatic effects of MGL2⁺ cell depletion, our studies with PD-L2–blocking antibodies indicate that the DCs utilize this specific pathway to suppress CD8 T-cell–mediated tumor immunity. We also show that these DCs strongly stimulate Treg responses, but whether there exist mechanisms that can be targeted to selectively disrupt this interaction remains to be determined. The NK/NKT-cell activation observed following MGL2⁺ cell depletion suggests that CD11b⁺ DCs suppress these lymphocyte populations through pathways that remain to be identified as well. Finally, while our results collectively suggest that CD11b⁺ DCs facilitate metastasis by directly suppressing immune responses in the metastatic microenvironment, we cannot exclude a role in our studies for MGL2⁺ and/or PD-L2⁺ DCs that reside in lymph nodes or other tissues apart from the metastatic site (39, 48).

Poor outcomes are associated with high Treg and low CD8 T-cell infiltration in many cancers (11, 12), including PDAC (13–15). Our studies suggest that a specific DC subset supports both arms of this unfavorable immune response at metastatic sites, highlighting the therapeutic potential of targeting such cells to inhibit PDAC progression. Although our studies focus on the role of CD11b⁺ DC at secondary sites, these cells also infiltrate primary tumors and thus may play a role in PDAC initiation and progression at the primary site. Our results reinforce the pathophysiological function of tumor-derived GM-CSF in PDAC (16, 17) and potentially other GM-CSF–producing cancers (49), finding that it drives Gr1⁺/Ly6C⁺ monocytes to differentiate into CD11b⁺ DCs that express PD-1 ligands and expand Tregs. Consistent with these findings, expression levels of MGL and PD-L2 correlate with DC markers, Foxp3, and GM-CSF and its receptor in human PDAC specimens. Recent studies in other models have shown that Gr1⁺/Ly6C⁺ monocytes predominantly differentiate into Mφ-like cells that express CD64 and/or F4/80 at primary and metastatic sites (35, 36, 42). However, our studies show that Gr1⁺/Ly6C⁺ monocytes exhibit a distinct differentiation pattern in PDAC metastasis, developing into cells that express low levels of the monocyte/Mφ markers CD64, F4/80, and Gr1/Ly6C, and high levels of the DC markers CD11c, MHC-II, and CD24. We suspect that these differences are driven by the cytokine milieu encountered by infiltrating monocytes, with the relative levels of M-CSF, GM-CSF, and potentially other factors (e.g., Flt3L) determining whether monocytes develop into cells more closely resembling Mφ or DCs (50). Supporting this, MGL2⁺CD11b⁺ DCs accumulate in metastatic liver following injection of GM-CSF–producing PDAC cells but not tumor lines characterized by low GM-CSF production (B16, LLC, and MC38). Furthermore, PDAC supernatants stimulate DC differentiation and expression of both MGL2 and PD-L2 by monocytes in a GM-CSF–dependent manner. Interestingly, MC38 also induces an infiltration of CD11b⁺ DC-like cells, but these cells express low levels of MGL2 compared with PDAC-associated CD11b⁺ DCs, suggesting that alternative factors drive their development. On the basis of a recent study of the tumor-associated DC compartment (42), we suspect that these cells share a monocytic lineage but undergo a distinct differentiation program, resulting in a more monocyte/Mφ-like phenotype characterized by Ly6C and CD64 expression. Viewed in this context, our studies underscore the diversity of monocyte differentiation patterns in cancer as well as the critical role of the tumor microenvironment in this process.

Our findings are consistent with a recent study that partitioned the tumor myeloid compartment into discrete DC and macrophage populations, and showed an essential role for CD11b⁻ (CD103⁺) DCs in CD8 T-cell–mediated tumor immunity (41). A CD11b⁺ DC population dependent on GM-CSF signaling was identified in the same study, but its effects on tumor immunity and disease progression were not determined. Aligning with our own studies, the CD11b⁺ DC population expanded in response to tumoral GM-CSF production and uniquely expressed both MGL2 and PD-L2. Leveraging the selective expression of these molecules, we show here that CD11b⁺ DCs represent the functional antithesis of tumor immunity–inducing CD11b⁻ DCs, as they both directly suppress CD8 T cells and expand Tregs in the metastatic microenvironment. Correspondingly, depleting or blocking the activity of the CD11b⁺ DC subset stimulates antitumor immune responses and suppresses metastasis development. The drastic shift toward CD11b⁺ DCs that occurs at secondary sites in our model is consistent with this functional dichotomy and may be essential for the evasion of immunosurveillance mechanisms that would block metastasis formation. The antitumor T-cell response induced by CD11b⁺ DC depletion suggests that this approach could also make tumors more responsive to other immunotherapies, including T-cell–directed checkpoint blockers and agonistic antibodies. However, the optimal molecular target for depletion of this DC subset in human cancer remains unclear, as human MGL is also expressed in a number of normal tissues (proteinatlas.org). More broadly, our findings highlight the heterogeneity of the tumor-associated myeloid response and suggest that myeloid cell–directed immunotherapies need to be tailored to different cancers. Ultimately, effective management of the tumor myeloid compartment may require multiple interventions, with efforts aimed at both expanding immunostimulatory cell populations (e.g., CD103⁺ DCs) and depleting or modulating immunosuppressive ones (e.g., CD11b⁺ DCs, TAMs, and/or MDSCs). Our studies contribute to this latter goal and suggest multiple ways of targeting an abundant, immunosuppressive DC population for therapeutic effect in pancreatic cancer.

No potential conflicts of interest were disclosed.

Conception and design: J.A. Kenkel, W.W. Tseng

Development of methodology: J.A. Kenkel, W.W. Tseng, M.G. Davidson, E.S. Seeley

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.A. Kenkel, W.W. Tseng, M.G. Davidson, L. Tolentino, O. Choi, N.E. Reticker-Flynn

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.A. Kenkel, E.S. Seeley, D. Winer, E.G. Engleman

Writing, review, and/or revision of the manuscript: J.A. Kenkel, D. Winer, N.E. Reticker-Flynn, E.G. Engleman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Tolentino, O. Choi, N. Bhattacharya

Study supervision: J.A. Kenkel, E.G. Engleman

We are grateful to Dr. Michael Bachmann for assistance with tumor cell labeling, Drs. Brendan Visser and Moe Jalali for help in obtaining human specimens, and Dr. Akiko Iwasaki for providing the Mgl2^DTR mice. We also thank past and present members of the Engleman lab for their assistance and feedback, especially Xianne Penny, Michael Alonso, Lei Shen, Claudia Benike, Tyler Prestwood, Yaron Carmi, and Robert Yuan.

This work was supported by NIH grants R01 CA196657, U01 CA141468, and U54 CA209971 to E.G. Engleman; T32 AI007290 to J.A. Kenkel and M.G. Davidson; and F32 CA189408 to N.E. Reticker-Flynn.

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.