Notch Signaling Regulates Immunosuppressive Tumor-Associated Macrophage Function in Pancreatic Cancer

A total of 90% of patients with pancreatic ductal adenocarcinoma (PDA) die within 5 years of their diagnosis (1). A key reason for this poor prognosis is resistance to existing therapies. This resistance is partly regulated by oncogenic KRAS signaling within the pancreatic tumor epithelium (2). Genetically engineered mouse models expressing oncogenic KRAS in the pancreas recapitulate the histologic progression of human PDA with formation of preneoplastic lesions including acinar-ductal metaplasia (ADM), pancreatic intraepithelial neoplasia (PanIN), and pancreatic adenocarcinomas with metastases (3–6). On the basis of observations from these models and human patients, it has become clear that epithelial oncogenic KRAS signaling promotes the formation of a complex immunosuppressive fibroinflammatory stroma that also contributes to treatment resistance (7).

Notch signaling is one of the core pathways dysregulated in PDA downstream of oncogenic KRAS (8). The core components of Notch signaling in mammals include the Notch transmembrane receptors (Notch1–4) and five membrane-bound ligands [Jagged1 (JAG1) and 2 (JAG2) and delta-like ligand 1, 3, and 4] (9). Cell-cell contact and ligand–receptor interactions between neighboring cells lead to a series of proteolytic events, culminating in the γ-secretase–mediated cleavage of the intracellular Notch domain, which is then released from the plasma membrane and translocated into the nucleus where it binds to the DNA-binding protein CBF1/Recombination Signal Binding Protein For Immunoglobulin Kappa J Region (RBPJκ). This complex also recruits the transcriptional coactivator mastermind-like (MAML) and leads to transcriptional activation of Notch target genes, including the Hes family of transcription factors (9). Notch epithelial signaling regulates pancreatic neoplastic progression in genetically engineered mouse models (10–14). Inhibitors targeting γ-secretase, a protease which activates the Notch proteolytic cascade, have been developed (15). Unfortunately, clinical trials based on these approaches have so far not yielded improved survival (16, 17). One reason for this lack of efficacy is incomplete understanding of how Notch regulates the PDA tumor microenvironment (TME).

The Notch pathway regulates multiple aspects of the tumor immune response, including T-cell differentiation and maturation and myeloid compartment functionality (18–20). In the context of the PDA TME, Notch inhibition via γ-secretase inhibition (GSI) leads to increased intratumoral hypoxia and sensitivity of the tumor epithelium to systemic therapy (21). In addition, Notch activation in vitro has been implicated in promoting M1-like, antitumor macrophage polarization (22, 23). Genetic approaches to activate or inhibit Notch signaling and transcriptional responses in the myeloid compartment of an autochthonous mouse model of pancreatic cancer demonstrate increased productive cytotoxic T-cell and macrophage antitumor responses after Notch stimulation (23). However, contrary to this observation, breast cancer models with Notch activation within tumor-associated macrophages (TAM) demonstrate protumorigenic, M2-like TAM polarization, T-cell inhibition, and a blunted antitumor immune response (24). The state of Notch signaling in the myeloid compartment of human patients with PDA remains unclear. In our work, we address the activation state of Notch within multiple compartments of primary human and mouse PDA and how the myeloid compartment and tumor immune response is shaped in the presence of pharmacologic Notch and immune checkpoint inhibition.

All animal studies were conducted in compliance with the guidelines of the Institutional Animal Care and Use Committee (IACUC) at the University of Michigan (Ann Arbor, MI). Patient selection/sample procurement: patients over the age of 18 referred for diagnostic endoscopic ultrasound of a pancreas mass lesion suspected of PDA were consented according to Institutional Review Board (IRB) HUM00041280. Up to two 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 IRBs of the University of Michigan Medical School. Details of the enrolled patient cohort have been published previously (25).

The original KC and KPC mice were sourced from our long-term internal breeding colonies on the C57BL/6J background. The relevant mouse strains containing the component alleles are available from Jackson Labs: LSL-KRAS^G12D (B6.129S4-Kras^tm4Tyj/J, Jackson Laboratory 008179), LSL-Trp53^R172H (129S-Trp53^tm2Tyj/J, Jackson Laboratory 008652), Ptf1a/p48-Cre (Ptf1a^tm1(cre)Hnak/RschJ, Jackson Laboratory 023329). CBF:H2B-Venus mice (26) were gifts from Dr. Sunny Wong, University of Michigan (Ann Arbor, MI). By using multiple CBF1 binding sites together with a subcellular-localized, genetically encoded fluorescent protein, H2B-Venus, the CBF:H2B-Venus transgenic strain of mice is capable of faithfully recapitulating Notch signaling at single-cell resolution. Venus mice were generated by crossing with C57BL/6J mice (Jackson Laboratory 000664). Male and female mice were included equally. LysMcre; DNMAML (LysMcre; Rosa26^LSL-DNMAML/+) mice were generated by crossing LysMcre (B6.129P2-Lyz2^{tm1(cre)Ifo/J}, Jackson Laboratory 004781) with ROSA26^DNMAMLf/+ (B6N.129-Gt(ROSA)26Sor^{tm1(MAML1)Wsp}/J, Jackson Laboratory 032613; refs. 27, 28). All genotypes of breeders and experimental animals were confirmed by genotyping PCR assays as detailed for each strain/allele within The Jackson Laboratory protocols. KC;CBF:H2B-Venus mice were aged 6, 8, or 12 weeks and KPC;CBF:H2B-Venus mice were aged 11 and 21 weeks (when pancreatic tumors developed). Pancreas, spleen, and duodenum were harvested for histology. All animal studies were conducted in compliance with the guidelines of IACUC at the University of Michigan (Ann Arbor, MI).

To establish the orthotopic pancreatic cancer model, 5 × 10⁴ of 7940B cells (29) and 1 × 10⁵ mT3-2D cells (30), both derived from KPC mouse tumors (Pdx1-Cre; LSL-Kras^G12D/+; LSL-Trp53^R172H/+) in C57BL/6J background, were injected into Venus or recipient C57BL/6J mice (6–8 weeks old). Tumor cells were tested for Mycoplasma contamination by MycoAlert PLUS Mycoplasma Detection Kit (Lonza), and passage 15–20 were used for all experiments. For each orthotopic injection, a left subcostal incision was made under anesthesia, the spleen and the tail of the pancreas was gently externalized, and tumor cells were injected under the capsule of the pancreas in a volume of 50 μL 50% serum-free cell culture media and 50% Matrigel (Corning 354234). Animals were monitored daily for infection and pain until the wound clips were removed between 7 and 10 days after surgery. Orthotopic tumors were allowed to grow for 1 week after implantation, and then mice were randomized into four groups with a minimum of 6 mice in each group. γ-secretase inhibitor (GSI) Crenigacestat (LY3039478, Selleckchem Chemicals) was given at 8 mg/kg by oral gavage every other day for Notch inhibition. Purified mouse anti-PD-1 (BioXcell #BE0033-2; clone J43) was used for in vivo PD-1 blockade at a dosage of 200 μg/mouse through intraperitoneal injection, repeated twice per week. Control mice received vehicle (2% DMSO + 30% PEG300 + 5% Tween 80 in ddH₂O) and control IgG (BioXcell #BE0091). Tumors were monitored twice a week by palpitation and harvested for histology before they reached 1.5 cm in diameter (19 days after implantation for 7940B, and 22 days for mT3-2D). Tumor weights were measured at harvesting.

All cells were cultured in Iscove's modified Dulbecco's medium (IMDM; Gibco) supplemented with 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco). The mouse pancreatic cancer cell line 7940B was used to generate our conditioned medium (CM). CM was filtered through a 0.2 μm filter before use. For in vitro TAM polarization, bone marrow–derived myeloid cells (BMDM) were generated from the femur and tibia of C57BL/6J mice. BMDMs were treated with CM (CM diluted 1:1 in fresh IMDM with 10% FBS) for 7 days for macrophage polarization. BMDMs were additionally cocultured with 7940B or mT3-2D cells in 6-well plates at 2:1 ratio and treated with Crenigacestat (LY3039478, Selleckchem Chemicals) at the concentration of 0.5 or 5 nmol/L. DMSO was added to the cocultures as control, and all cells were harvested 24 hours later for RNA extraction.

Hematoxylin and eosin, IHC, and immunofluorescent (IF) staining were performed on 5-μm-thick formalin-fixed, paraffin-embedded (FFPE) mouse or human pancreatic tissue sections. Antibodies used are listed in Supplementary Table S1. For IHC, fresh cut paraffin sections were rehydrated using sequential series of xylene, 100% ethanol, and 95% ethanol (two washes each). Slides were rinsed with water for 5 minutes, and then antigen retrieval was performed using Antigen Retrieval CITRA Plus (BioGenex) by heating in microwave on 100% power for 5 minutes and resting for another 3 minutes. Once cooled down, sections were blocked using 1% BSA (Sigma-Aldrich) in PBS for 30 minutes and then primary antibodies were used at the indicated dilutions. After primary antibody incubation at 4°C overnight, biotinylated secondary antibodies (Goat anti-Rabbit IgG, Vectorlabs) were used at a 1:300 dilution and applied to sections for 45 minutes at room temperature. Sections were then incubated for 30 minutes with ABC reagent from the Vectastain Elite ABC Kit (Peroxidase), followed by diaminobenzidine (Vectorlabs) staining. For immunofluorescence staining, after primary antibody incubation at 4°C overnight, Alexa Fluor secondary antibodies (Invitrogen, listed in Supplementary Table S1) were used, and then slides were mounted with Prolong Diamond Antifade Mountant with DAPI (Invitrogen). The Tyramide SuperBoost Kit (Invitrogen) was used for double IF staining with primary antibodies raised in the same host species. Images were taken with Olympus BX-51 microscope, Olympus DP71 digital camera, and DP Controller software. The IF images were acquired using the Olympus IX-71 confocal microscope and FluoView FV500/IX software. Quantitative analysis for IHC/IF staining was performed in three to five nonoverlapping fields for each sample using Fiji software to measure the percentage of positive area.

Single-cell suspensions of spleens and pancreatic tumors from CBF:H2B-Venus mice orthotopically implanted with PDA cells 7940B or mT3-2D were prepared. Briefly, spleens were crushed and passed through 40-μm cell strainers and treated with red blood cell lysis buffer (eBioscience) to eliminate erythrocytes. Pancreatic tumors were minced, and then incubated in 1 mg/mL collagenase (Sigma-Aldrich) in Hank's Balanced Salt Solution (Gibco) for 30 minutes at 37°C before passing through a series of 500-μm, 100-μm, and 40-μm cell strainers. Single-cell suspensions were blocked in 100% FCS (Gibco) for 30 minutes on ice, then stained with the fluorescently conjugated antibodies (BD Biosciences, BioLegend or eBioscience) listed in Supplementary Table S1 for 15 minutes on ice, protected from light. Flow cytometric analysis was performed on a Cyan ADP analyzer (Beckman Coulter), and data were analyzed with Summit 4.3 software. DAPI-negative single cells were gated for further characterization. Venus-negative (Venus⁻) or Venus-positive (Venus⁺) cells were measured in the following cell types: fibroblasts were identified as CD45⁻EpCAM⁻PDGFRα⁺; macrophages were CD45⁺CD11b⁺F4/80⁺; B cells were CD45⁺CD19⁺; CD4 T cells were CD45⁺CD3⁺CD4⁺CD8⁻; and CD8 T cells were CD45⁺CD3⁺CD8⁺CD4⁻. Cell sorting was performed using a MoFlo Astrio (Beckman Coulter). In addition, Venus⁻ or Venus⁺ macrophages (DAPI⁻CD45⁺CD11b⁺F4/80⁺) from orthotopic 7940B tumors in CBF:H2B-Venus mice were sorted and lysed in RLT buffer (Qiagen).

Total RNA was prepared using RNeasy (Qiagen), and reverse transcription was performed using the High-Capacity cDNA Reverse Transcription kit (Applied Biosystems) for flow-sorted macrophages and tumor-BMDM coculture samples. Samples (10 ng template cDNA) for qRT-PCR were prepared with 1X SYBR Green PCR Master Mix (Applied Biosystems) and various primers (primer sequences are listed in Supplementary Table S2). 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 using QuantStudio 6 Pro Real-Time PCR Systems (Applied Biosystems). Melt curve analysis was performed for all samples after the completion of the amplification protocol. Gene expression was normalized to Gapdh, a housekeeping gene, using the 2^−ΔCT method. Statistical significance was determined by two-tailed t tests.

Single-cell RNA sequencing (scRNA-seq) data of human pancreatic samples (16 PDA and three adjacent tissues) and mouse KPC tumor (one sample) have been described previously (25, 31). Raw human data are available at the NIH dbGaP database (phs002071.v1.p1), and processed data are available at NIH Gene Expression Omnibus (GEO) database (GSE155698; ref. 25). Raw KPC mouse data are available at the NCBI's GEO database (GSM6127792; ref. 31). Tuveson KPC data (four samples) are available at the GEO under the accession number GSE129455 (32). R Studio V3.5.1 and R package Seurat version V3.0 were used for downstream scRNA-seq data analysis as described previously (25, 33). Data were initially filtered to only include all cells with at least 200 genes and all genes in greater than three 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. Principal component analysis (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. FindMarkers table was created, and clusters were defined by user-defined criteria. To subcluster the myeloid population, we utilized markers described in recent work defining myeloid cells across a variety of tumors by scRNA-seq (34–36). We thus defined macrophages, neutrophils, monocytes, and dendritic cells (DC). General macrophage markers included CD68, MSR1, MRC1, CCR2, C1QC, IL1B, NLRP3, and MMP9. Classical macrophages were defined by expression of SIRPA, FCGR3A, CSF1R, LGALS3, and ADGRE1. CD207, TIMD4, VSIG4, and MARCO expression defined the resident macrophage population. Monocyte-derived macrophages were positive for CCR2, ITGAM, and CD14. Monocytes expressed CD14, CCR2, and ITGAM but were low/negative for general macrophage markers. Neutrophils were identified by expression of S100A8, S100A9, SELL, and FCGR3B. DCs included the conventional type 2 (cDC2) subpopulation expressing CCL17, CCR7, LAMP3, CLEC10A, and CD1C. Finally, the plasmacytoid (p)DC population was defined by the expression of IL3RA, NRP1, and MZB1.

GraphPad Prism 6 software was used for all statistical analysis. All data were presented as means ± SE (SEM). Intergroup comparisons were performed using two-tailed unpaired t test or one-way ANOVA, and P < 0.05 was considered statistically significant.

All sequencing data used within this article are publicly available on the GEO—GSM6127792 and GSE155698—and in the NIH database of Genotypes and Phenotypes (dbGaP) for the raw human sample sequences - phs002071.v1.p1.

To explore Notch signaling pathway component expression in the human PDA microenvironment, we analyzed our previously published scRNA-seq dataset of 16 human pancreatic cancer samples, as well as three adjacent benign or normal pancreata (25). In this dataset, we captured 13 different cell populations in both PDA and adjacent/normal pancreata (Supplementary Fig. S1A). Gene expression profiling identified Notch signaling pathway components across all cell clusters at various levels, except for acinar cells. In general, Notch signaling pathway gene expression was higher in PDA compared with adjacent/normal pancreata (Supplementary Fig. S1A). In particular, Notch receptors were highly expressed in cancer epithelial cells, endothelial cells, fibroblasts, and subsets of immune cells, including multiple myeloid populations (Supplementary Fig. S1A). The Notch pathway ligands, such as JAG1, were mainly expressed by epithelial cells, endothelial cells, and fibroblasts (Supplementary Fig. S1A). Notch pathway canonical target gene Hairy and Enhancer of Split-1 (HES1) expression was enriched in epithelial cells, endothelial cells, fibroblasts, myeloid cells, and mast cells (Supplementary Fig. S1A). Co-IF staining confirmed expression of HES1 in fibroblasts and macrophages (about 45% CD163⁺ macrophages are HES1⁺; Supplementary Fig. S1B), but not in T cells, in human PDA microenvironments (Supplementary Fig. S1C).

To further map the Notch signaling pathway within the immune system, we focused on gene profiling the immune cells only. We clustered immune cells into nine clusters (Fig. 1A; Supplementary Fig. S1D) containing multiple myeloid cell populations. The Notch target HES1 was expressed by mast cell, monocyte, and macrophage populations at relatively high levels compared with other immune cells (Fig. 1B). We further clustered myeloid cell populations based on a specific nomenclature for myeloid cell populations in scRNA-seq (refs. 34–36; Fig. 1C; Supplementary Fig. S1E) and found that resident macrophages, classical macrophages, and monocytes (mostly from PDA samples) had higher HES1 expression compared with monocyte-derived macrophages (mostly from adjacent/normal pancreata; Fig. 1D). Overall, HES1 expression was higher in tumor-associated myeloid cells compared with those from adjacent/normal pancreata (Fig. 1E). In addition, monocytes, neutrophils, and monocyte-derived macrophages had high expression of Notch receptors, primarily NOTCH 1 and 2. NOTCH 1 and 2 were upregulated in these myeloid cell subsets from pancreatic tumors compared with those from adjacent/normal pancreata (Fig. 1F), indicating that Notch signaling in the myeloid compartment may affect TME regulation in pancreatic cancer. The myeloid compartment also contained high expression of the proteolytic enzymes ADAM metallopeptidase domain 10 and 17 (ADAM10, ADAM17; Fig. 1F), which are involved in the Notch proteolytic cascade as well as downstream Notch signaling components (MAML1/2, RBPJ). This suggests the presence of a fully reconstituted Notch pathway that may play a role in shaping the myeloid compartment within the PDA microenvironment.

To address whether the existing mouse models of pancreatic cancer recapitulated our observations from human tumors, we used our previously published dataset where we performed a similar scRNA-seq analysis using the Ptf1a^Cre/+, Kras^LSL-G12D/+, Tp53^LSL-R172H/+ (KPC) mouse model (ref. 31; Fig. 2A). We observed high expression of Notch receptors, downstream signaling components, and Hes1 in cancer epithelial cells, endothelial cells, pericytes, fibroblasts, and myeloid cells, including DCs and macrophages (Fig. 2B). Co-IF staining confirmed abundant expression of HES1 in fibroblasts, macrophages, and endothelial cells in the KPC TME (Supplementary Fig. S2A). Our results support the presence of active Notch signaling within multiple compartments of the PDA TME, including macrophages.

To evaluate Notch signaling activity, we crossed a Notch signaling reporter mouse CBF:H2B-Venus (26) with two established mouse models of pancreatic cancer, KC (Ptf1a^Cre/+; Kras^LSL-G12D/+) and KPC, to generate KC;CBF:H2B-Venus and KPC;CBF:H2B-Venus mice. These models express the Venus fluorescent protein upon Notch signaling activation (Fig. 2C; Supplementary Fig. S2B). In the normal pancreas, Venus signaling was detected in ductal epithelial cells and was colocalized with HES1, cleaved NOTCH1, and NOTCH2 by IF staining (Supplementary Fig. S2C). At very early stages of pancreatic tumorigenesis in 6-week-old KC pancreata, where ADM predominates, we observed Venus-expressing macrophages and fibroblasts in the stroma (Fig. 2D) by co-IF staining of GFP (to detect Venus on FFPE sections) and F4/80 or SMA. Later, in 8- and 12-week-old KC and 11-week-old KPC mice we found PanIN, a precursor for invasive pancreatic cancer. More macrophages and fibroblasts were detected in the PanIN microenvironment expressed Venus, signifying active Notch signaling (Fig. 2E; Supplementary Fig. S2D). Similarly, co-IF staining showed Venus-expressing macrophages and fibroblasts in mouse PDA at 21 weeks in a KPC mouse (Fig. 2F). On the basis of these observations, Notch signaling activation in the TME starts at the earliest stages of carcinogenesis and persists through disease initiation and progression.

We next expanded our analysis to orthotopically implanted models using two syngeneic mouse PDA cell lines, 7940B and mT3-2D (both derived from Pdx1-Cre; Kras^LSL-G12D/+; Trp53^LSL-R172H/+ tumors), injected into C57BL/6J:CBF:H2B-Venus mice. We harvested the resulting tumors for flow cytometry analysis after 3 weeks (Fig. 3A). Venus expression was detected in multiple cell types from the TME, including fibroblasts, macrophages, B cells, and T cells (Fig. 3B–E). Macrophages comprised up to approximately 60% of all immune cells (Fig. 3C), and roughly half of the TAMs were Venus⁺, making them the largest Venus-expressing cell population within the TME (Fig. 3C and F). In contrast, fibroblasts constituted approximately 20% of total cells, and Venus-expressing fibroblasts were rare (Fig. 3F; Supplementary Fig. S3A). Corresponding spleen tissues from tumor-bearing mice were used as an internal control, and Venus expression was also found in splenic macrophages, B cells, and T cells (Supplementary Fig. S3B). To validate that Venus expression represented active Notch signaling, we performed co-IF staining on orthotopic PDA sections and determined that F4/80⁺GFP⁺ macrophages were also HES1-positive (Supplementary Fig. S3C). Thus, Notch pathway component and target gene expression characterizes macrophages infiltrating precursor lesions and advanced pancreatic cancer, and Notch activity is reflected by expression of the Venus reporter.

To functionally characterize the role of Notch signaling in TAMs, we used FACS to isolate Venus⁺ and Venus⁻ TAMs from the orthotopic 7940B tumors. The Venus⁺ and Venus⁻ TAM populations were similar in cell number (Fig. 4A), consistent with co-IF staining showing about half of the F4/80⁺ macrophages expressing Venus (Fig. 4B). We then performed qRT-PCR to molecularly characterize them (Fig. 4C). First, we confirmed that Hes1 expression, a measure of Notch signaling, was higher in Venus⁺ TAMs compared with Venus⁻ TAMs. We also found that Wnt signaling target lymphoid enhancer-binding factor 1 (Lef1) and several alternatively activated macrophage (M2-like) markers, including arginase 1 (Arg1), macrophage scavenger receptor 1 (Msr1), and mannose receptor C-type 1 (Mrc1), were elevated in Venus⁺ TAMs (Fig. 4C). Consistent with these findings, expression of immunosuppressive cytokines and chemokines such as Il10 and Tgfb1 was also higher in Venus⁺ TAMs. Finally, the classically activated macrophage (M1) marker nitric oxide synthase 2 (Nos2) was lower in Venus⁺ TAMs, and there were no significant differences in expression of Notch1, AXIS inhibition protein 2 (Axin2), Tnfα, M2 marker chitinase-like 3 (Chi3l3, also known as YM1; refs. 37, 38), immune checkpoint ligand Cd274, and several myeloid checkpoint genes including signal regulatory protein α (Sirpα), triggering receptor expressed on myeloid cells 2 (Trem2), sialic acid immunoglobulin-like lectin E (Siglece), stabilin 1 (Stab1), and neuropilin-1 (Nrp1) between the two TAM subsets (Fig. 4C; Supplementary Fig. S3D). Overall, the gene expression analysis demonstrates a positive correlation between Notch activation and M2-like polarization of TAMs.

To validate our findings, we further analyzed single-cell sequencing data obtained from the KPC model (31). We identified four subsets of macrophages (Fig. 4D; Supplementary Fig. S4A) and performed gene expression analysis for Notch pathway genes (Supplementary Fig. S4B). All four TAM subsets expressed Notch receptors at varying levels, with the highest expression noted for Notch1 and Notch2. The macrophage 4 population had the highest expression of Hes1, indicating high levels of Notch signaling. The same macrophage 4 population also expressed several M2 TAM markers including Arg1, macrophage receptor with collagenous structure (Marco), Cd274, and Chi3l3 (Fig. 4E; Supplementary Fig. S4C). We further analyzed scRNA-seq data from a publicly available KPC (Pdx1-Cre; Kras^LSL-G12D/+; Trp53^LSL-R172H/+) tumor dataset (Tuveson KPC; ref. 32; Fig. 4F; Supplementary Fig. S4D). In this dataset, the Hes1-high population also had the highest expression of Arg1 (Fig. 4G).

To gain a better understanding of the role of Notch signaling activation in the human TME, we compared the gene expression profiles between HES1-positive and HES1-negative myeloid cells from human patients. We discovered higher expression of complement genes C1QA and C1QB, alternatively activated macrophage markers CD163, MRC1, and MARCO, and an immunosuppressive molecule STAB1 in HES1-positive myeloid cells. We previously identified complement genes C1QA, C1QB, and TREM2 as part of a pancreatic tumor-specific signature in macrophages (39). HES1-positive myeloid cells also had higher expression of C-X-C motif chemokine ligand 2 (CXCL2) and IL1B, potentially linking Notch signaling activation in myeloid cells to proinflammatory responses in the TME of pancreatic cancer (Fig. 4H).

To elucidate a potential causal link between Notch activation and the immunosuppressive function of TAMs, we developed an in vitro coculture system in which BMDMs were differentiated and polarized to tumor-conditioned (TC) macrophages by cancer cell–conditioned media or direct coculture with PDA cells (Fig. 5A). Using BMDMs derived from C57BL/6J:CBF:H2B-Venus mice, we traced activation of Notch signaling by Venus expression. We observed Venus expression only in BMDMs cocultured with PDA cells directly, but not with cancer cell-conditioned media, supporting the notion that direct physical contact between the tumor cells and myeloid cells is needed for Notch activation as expected (Fig. 5B). The Venus-expressing BMDMs also expressed F4/80 and Arg1, markers of immunosuppressive macrophages (Fig. 5C). Moreover, when BMDMs cocultured with PDA cells were treated with the γ-secretase inhibitor Crenigacestat (40) to inhibit the proteolytic signaling downstream of Notch receptors, the expression of Hes1, Arg1, Chi3l3, Mrc1, Tgfβ1, and Il10 were downregulated compared with vehicle-treated cells (Fig. 5D and E). These results indicate direct tumor epithelial/macrophage contact in the PDA microenvironment, which activates macrophage Notch signaling and results in polarization of TAMs to an immunosuppressive phenotype.

Given our functional results implicating epithelial/macrophage cross-talk in Notch activation and immunosuppressive TAM polarization, we sought to investigate the therapeutic and immunomodulatory potential of Notch signaling inhibition in vivo, and to determine whether inhibition of Notch signaling would sensitize pancreatic cancer to immune checkpoint blockade. We implanted 7940B cells orthotopically into syngeneic C57BL/6J mice; a week later, tumor-bearing mice were treated with either GSI, anti–PD-1, or combination (Fig. 6A). At harvest (day 19), we observed smaller tumors in the combination treatment cohort compared with the controls or either treatment alone (Fig. 6B). Histologic analysis of the orthotopic tumors revealed an increase in tumor-infiltrating CD8⁺ T cells, along with increased expression of the T-cell activation marker granzyme B in the combination treatment cohort (Fig. 6C; quantification in Fig. 6E). We also observed increased cell apoptosis by cleaved caspase 3 IHC staining after GSI and anti–PD-1 combination treatment (Fig. 6D; quantification in Fig. 6F). Targeting Notch signaling alone (GSI group) also showed a trending increase of tumor-infiltrating CD8 T cells, although there was no change in granzyme B production or cell apoptosis compared with the control group. In contrast, the tumor-infiltrating macrophage number remained unchanged in all conditions (Supplementary Fig. S5A; quantification in Supplementary Fig. S5B).

We repeated the GSI and anti-PD-1 treatment with a second orthotopic model using mT3-2D cell line. Both GSI treatment alone and the combination treatment resulted in smaller tumors compared with either control or anti–PD-1 alone (Supplementary Fig. S5C). Consistent with our flow cytometry data (Fig. 3F), histologic analysis of the mT3-2D orthotopic tumors revealed low CD8⁺ T-cell density compared with the 7940B tumors, a similar heterogeneity in T-cell infiltration seen by others in both human and murine pancreatic tumors (41, 42). Here, we observed an increase in CD8⁺ T-cell infiltration in both GSI and combination treatment groups, although anti–PD-1 treatment alone led to a slight increase of tumor-infiltrating CD8⁺ T cells (Supplementary Fig. S5D). These findings are consistent with activation of a productive antitumor immune response upon combined treatment targeting Notch signaling and the PD-1 checkpoint. The results also support the notion that native Notch signaling promotes immunosuppressive TAM polarization and targeting Notch may be a useful addition to immunotherapy approaches in pancreatic cancer.

To target Notch specifically in myeloid cells, we used LysMcre; DNMAML (LysMcre; Rosa26^LSL-DNMAML/+) mice (Fig. 7A). The MAML transcriptional coactivator is an essential component of the complex that mediates Notch transcriptional activity in the nucleus; the dominant-negative form of MAML (DNMAML), conversely, blocks transcriptional activity downstream of all Notch receptors (27, 28). LysMcre; DNMAML mice express DNMAML upon Cre recombination, thus inactivating Notch signaling in myeloid cells. We implanted 7940B cells orthotopically into syngeneic LysMcre and LysMcre; DNMAML mice, both in C57BL/6J background, and tumors were harvested 3 weeks later (Fig. 7B). Myeloid Notch inhibition led to significantly smaller tumor weights in LysMcre; DNMAML mice (Fig. 7B). Histologic analysis revealed a trending increase in CD8⁺ T cells and granzyme B, a CD8⁺ T-cell activation marker and effector, (Fig. 7C and D) in LysMcre; DNMAML tumors compared with controls, suggesting cytotoxic effects of CD8⁺ T cells in LysMcre; DNMAML mice. We also observed a decrease in tumor-infiltrating macrophages in LysMcre; DNMAML tumors (Fig. 7E). Overall, our data suggest that specific inhibition of Notch activity in myeloid cells leads to decreased macrophage recruitment and increased CD8⁺ T-cell activation, which likely contribute to the impaired tumor growth observed in LysMcre; DNMAML mice.

Immunotherapy has so far not been effective as a treatment option for pancreatic cancer (43, 44). Specifically, targeting the PD-1 immune checkpoint with a single agent provides no benefit to patients; effective combination approaches are thus needed. The complex immunosuppressive nature of the PDA TME is the direct cause of this and reflects the interplay of many different cell types (7). TAMs play a key immunosuppressive role within this context (45–48).

Notch signaling has been previously implicated in myeloid development in bone marrow, as well as in local functional maturation in the tissue/TME (19, 49). In addition, tumor epithelial Notch signaling alters the tumor secretome to promote an immunosuppressive response in the surrounding microenvironment (50). A role for direct Notch signaling in myeloid cells has not been explored in depth in PDA despite evidence for it in other tumor types (18). Our paper demonstrates that Notch pathway components and active signaling are present in multiple cell populations in both human and murine tumors, including macrophages, T cells, fibroblasts, and the endothelium. We also use a noninvasive Notch pathway activity reporter to identify and characterize the phenotype and function of the myeloid cells in pancreatic tumors. Our data support the notion that Notch-active myeloid cells are primarily TAMs with a more immunosuppressive M2-like/alternatively activated phenotype marked by higher expression of immunosuppressive cytokines, arginase 1, and immune checkpoint molecules.

Contrary to our results, other investigators have found that Notch activation primarily drives an M1-like, tumor-suppressive phenotype in macrophages (22, 51). In addition, independent modulation of Notch signaling via Notch intracellular domain overexpression or deletion of Rbpj—a key transcriptional transducer of Notch signaling—in the myeloid compartment of an autochthonous mouse model of pancreatic neoplasia suggests that activation of Notch signaling in myeloid cells drives a stronger antitumor immune response (23). The discrepancy between these findings and our observations in myeloid cells with an intact, native level of signaling may be partly explained by the modular and combinatorial nature of the Notch signaling pathway (52–56). Final Notch signaling response is a product of integration of both cis and trans cellular signaling and inhibition. In addition to this, different Notch ligands can evoke distinct quantitative and qualitative cellular responses (56). Work published prior to our article often relied on a high level of activation or inhibition of Notch signaling through isolated ligand or Notch intracellular domain overexpression or complete genetic abrogation of the Notch transcriptional response. In addition, many of these interventions were present throughout myeloid development, migration, and local microenvironment differentiation, which may lead to signaling and functional myeloid states that would not normally be seen in an intact in vivo context. In comparison, our work utilized a noninvasive functional fluorescent reporter of Notch activity to identify cells with native levels of Notch signaling activation. This allowed us to isolate and molecularly and functionally characterize the myeloid cells with a productive Notch transcriptional response without disrupting any of the native activating and inhibitory receptor–ligand interactions. The immunosuppressive nature of these myeloid cells could be partially reversed by GSI, which blocks the proteolytic activation of Notch signaling. In addition, genetic ablation of the Notch transcriptional response specifically within the myeloid compartment inhibited orthotopic tumor growth, confirming the tumor-promoting role for myeloid Notch signaling in our models.

We also tested the hypothesis that Notch inhibition would further sensitize tumors to immunotherapy. Previous work has shown that both human and murine pancreatic tumors can be defined by the level of T-cell infiltration as T cell-high and -low (41, 42). Both of these states are modeled in our systems by the two syngeneic cell lines used in this article—7940B as the “T cell-high” and mT3-2D as the “T cell-low” model. The effect of Notch inhibition differs somewhat based on this context. In both systems, pharmacologic Notch inhibition leads to an increased influx of cytotoxic CD8⁺ T cells. In the T cell–high system, this occurs in conjunction with anti-PD-1 therapy and further augments an already high tumor T-cell infiltration, leading to higher apoptosis and slower tumor growth. In the T cell-low system, Notch inhibition alone leads to increased CD8⁺ T-cell influx that is not further augmented by anti-PD-1 therapy but is still associated with decreased tumor growth. The myeloid compartment serves as a signaling relay between tumor cells and infiltrating T cells to induce the T cell-low phenotype (41). We have also previously shown that myeloid cells can support tumor growth via direct trophic effects on the tumor epithelium and by inhibiting cytotoxic CD8⁺ T-cell function in pancreatic neoplasia (57). Further work in our genetically engineered systems will establish the contribution of a direct myeloid cell-tumor cell axis to the tumor-promoting function of myeloid Notch signaling. Overall, our data do implicate myeloid Notch as a context-dependent tumor-supportive and immunosuppressive signal in pancreatic cancer (Fig. 7F).

In addition to demonstrating direct Notch activation and function in the myeloid compartment, our work also implicates Notch in multiple other cell types in the PDA TME. The role of Notch in T-cell maturation and function is well documented but is less clear in the specific context of PDA (58, 59). Notch signaling has also been shown to regulate fibroblast phenotype and function in the TME and other inflammatory contexts (60, 61). In PDA, cancer-associated fibroblasts (CAF) can be parsed into multiple coexisting subtypes regulating the TME and immune response (32, 62–70). We and others have previously shown that dysregulation of Hedgehog signaling plays a key role in CAF function and immune polarization within the TME (62, 64, 71–73). How Notch integrates with Hedgehog and other signaling pathways in CAFs and how this affects the TME remains an active area of study. Similarly, Notch ligands are also highly expressed on tumor-associated endothelium (60). Myeloid cells recruited to the TME have to traverse the endothelium as the first barrier during their extravasation into the tumor and endothelial Notch ligands may serve as the first relevant Notch signal leading to myeloid polarization in the TME. Along these lines radiation-induced increased expression of the Notch ligand Jagged1 in lung endothelium leads to alternative M2-like polarization of myeloid cells as they are recruited into the lung parenchyma (74). Similar mechanisms are potentially at play in the PDA TME. In summary, our work implicates direct Notch signaling in immunosuppressive myeloid polarization in pancreatic cancer and suggests the presence of a complex Notch-specific network of interactions between various cell types regulating this polarization. Further understanding of Notch signaling and its role in pancreatic cancer may provide future alternative means to redeploy already-developed Notch-modulating drugs in combination chemoimmunotherapy regimens.

F. Bednar reports grants from NCI during the conduct of the study. No disclosures were reported by the other authors.

W. Yan: Data curation, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. R.E. Menjivar: Data curation, formal analysis, investigation, methodology. M.E. Bonilla: Data curation, formal analysis, investigation, methodology. N.G. Steele: Data curation, formal analysis, investigation, methodology. S.B. Kemp: Data curation, formal analysis, investigation, methodology. W. Du: Data curation, formal analysis, investigation, methodology. K.L. Donahue: Data curation, formal analysis, investigation, methodology. K.L. Brown: Data curation, formal analysis, investigation, methodology. E.S. Carpenter: Data curation, formal analysis, investigation, methodology. F.R. Avritt: Data curation, formal analysis, investigation, methodology. V.M. Irizarry-Negron: Conceptualization, resources, data curation, formal analysis, supervision, investigation, methodology, project administration. S. Yang: Data curation, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. W.R. Burns III: Conceptualization, resources, supervision, funding acquisition, methodology, project administration, writing–review and editing. Y. Zhang: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing. M. Pasca di Magliano: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, methodology, writing–review and editing. F. Bednar: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

M. Pasca di Magliano is supported by NIH/NCI grants R01CA271510, R01CA264843, U01CA224145, and U01CA268426, the University of Michigan Cancer Center Support Grant (NCI P30CA046592), the Pancreatic Cancer Action Network, and the American Cancer Society. F. Bednar is supported by grants from the Association of Academic Surgery, Association of VA Surgeons, American Surgical Association, and NCI R01CA271510. Y. Zhang was funded by NCI-R50CA232985. N.G. Steele was funded by the Cancer Biology Training Program T32-CA009676. N.G. Steele is also a recipient of the American Cancer Society Postdoctoral Award PF-19-096-01 and the Michigan Institute for Clinical and Healthy Research (MICHR) Postdoctoral Translational Scholar Program fellowship award. K.L. Donahue was funded by the Cancer Biology Training Program T32-CA009676 and NCI F31-CA265085-01A1. R.E. Menjivar was supported by the NIH Cellular and Molecular Biology Training Grant T32-GM007315, the Center for Organogenesis Training Program (NIH T32 HD007505) and NCI (F31-CA257533). S.B. Kemp was supported by NIH T32-GM113900 and NCI F31-CA247076. W. Du was supported by the University of Michigan Training Program in Organogenesis (NIH T32-HD007505). M.E. Bonilla was supported by Proteogenomics of Cancer Training Grant T32 CA140044-13.

We thank Dr. Howard Crawford and Daniel Long for histological services. This project was also supported by the Tissue and Molecular Pathology and Flow Cytometry Shared Resources at the Rogel Cancer Center and the University of Michigan Advanced Genomics Core.

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