To evaluate the roles of Notch ligands DLL1 and Jag2 expression on DCs in the regulation of T-cell-mediated anti-tumor immunity, we generated mice with CD11c-lineage-specific deletion of their genes. Mice with hetero- or homozygous allele deletion of Dll1 or Jag2 appeared normal in gross morphology with respect to their wild type littermates with floxed alleles, DLL1 ^flox/flox or Jag2 ^flox/flox . A representative mRNA analysis of the respective Notch ligands in flow-sorted CD11⁺ DC populations from wild type and genetically modified mice is shown (Fig. 1a). Transcripts for the tested Notch ligand mRNAs were absent in CD11c⁺ cells but present in CD11c⁻ splenic cells or whole splenocyte populations from mice with homozygous deletion of the ligands.Fig1Fig. 1

Genetic ablation of Dll1 in CD11c⁺ cells in mice accelerates tumor growth with decreased survival. a Deletion of Notch ligand genes Dll1 and Jag2 in CD11c⁺ cells was confirmed by RT-PCR performed with RNA isolated from CD11c⁺ or CD11c⁻ cells from splenocytes of genetically modified and wild-type mice. Lewis lung carcinoma (LLC) tumor growth (b) and log-rank survival curves (c) for mice with hetero or homozygous deletion of Dll1 in CD11c⁺ cells and wild-type littermates. d Pancreatic MT5 tumor growth in CD11c⁺ cell-specific Dll1 ^−/− and wild-type mice. e LLC tumor growth in CD11c⁺ cell-specific Jag2 ^−/− and wild-type littermates. Mean ± SEM, 8–10 mice per group; *, p < 0.05; **, p < 0.01

To test whether genetic ablation of specific Notch ligands in DCs affected cytokine secretion patterns, we evaluated IFN-γ and IL-4 production in tumor-infiltrating T-cells by ELISPOT assay following restimulation with CD3/CD28 antibody activator beads or with LLC tumor antigenic MHC class-I-restricted peptide MUT1 loaded on autologous splenocytes. We found that in mice with hetero and homozygous deletion of Dll1 in DCs, the numbers of IFN-γ-producing MUT1-specific lymphocytes were markedly decreased in the tumor, whereas the numbers of tumor-infiltrating IL-4-producing cells were not altered (Fig. 2a-c). Similarly, a reduction of IFN-γ-secreting lymphocytes was observed in the tumor-draining lymph nodes (LN) (Fig. 2d). These observations provided the explanation for the observed differences in tumor growth rates by pointing to the crucial role of DC-expressed DLL1 for the induction of anti-tumor cytotoxic T-cell responses.Fig2Fig. 2

CD11c-lineage-specific ablation of Notch ligands alters cytokine production. IFN-γ and IL-4-producing cells were enumerated by ELISPOT assay among LLC tumor-infiltrating lymphocytes (TIL) from mice with CD11c lineage-specific deletion of Dll1 and wild-type littermates following re-stimulation with anti-CD3/CD28 beads (a, b) or with LLC tumor antigenic peptide MUT1 (FEQNTAQP) loaded on autologous splenocytes for 48 h (c). d Evaluation of IFN-γ-producing cells in a pool of tumor-draining lymph node cells from the same mice following re-stimulation with anti-CD3/CD28 beads. e Evaluation of IFN-γ and IL-4-producing cells among tumor-infiltrating lymphocytes from Jag2 ^−/− or wild-type littermate mice following re-stimulation with anti-CD3/CD28 beads. Mean ± SEM, 5 mice per group; *, p < 0.05, **, p < 0.01

We performed extensive immunophenotyping of myeloid and lymphoid populations infiltrating the tumors and in spleens of mice with CD11c⁺ lineage-specific hetero or homozygous deletion of Dll1 and their wild-type littermates on day 17–18 after LLC tumor establishment. The deletion of either one or two alleles of Dll1 resulted in a moderate increase in total cell counts of tissue-resident CD11b⁺CD11c⁺ DC populations in the tumor or spleen compared to wild-type littermates, but ability of DCs to undergo maturation or infiltrate the tumor was not affected (Fig. 3a-c). Numbers of CD11b⁺CD11c⁺ DCs expressing maturation markers MHCII, CD40, CD80, CD86 and CD209 also did not change (Fig. 3a, b, d). This is consistent with the hypothesis that the observed alterations in anti-tumor T-cell responses is due to the absence of DLL1 expression on DCs. Other obvious changes in the myeloid compartment included increased numbers of CD11b⁺Gr-1⁺ cells in Dll1-ablated mice. Further characterization showed that both Ly6C⁺ monocytic and Ly6G⁺ granulocytic populations of CD11b⁺Gr-1⁺ cells were significantly higher in the tumors of DC-Dll1-ablated mice (Fig. 3a, d). In the lymphoid organs of spleen and LN also, increased numbers of CD11b⁺Gr-1⁺Ly6G⁺ granulocytic cells were observed (Fig. 3b). On the other hand, a decline was noted in the proportions of CD68⁺MHCII⁺F4/80⁺ M1 and CD68⁺MHCII⁺CD86⁺CD206⁺ M2 macrophages in the tumor-infiltrate and spleen of Dll1-ablated mice (Fig. 3a, b).Fig3Fig. 3

CD11c-lineage-specific Dll1-ablated tumor-bearing mice exhibit no change in dendritic cells but increase CD11b⁺Gr1⁺ cell proportions. Myeloid populations were evaluated by flow cytometry on day 17–18 after LLC tumor initiation in Dll1 knockout and wild-type littermate mice. Percentage of indicated populations are shown in the tumor infiltrate (TIL) (a) and in a pool of splenocyte and LN cells (b). c Total cell yields in the splenocytes and tumor single cell suspensions. d Representative FACS plots for CD11b versus CD11c, Ly6C or Ly6G staining (c). Mean ± SEM, 5–7 mice per group; *, p < 0.05; **, p < 0.01

Activation of Notch receptor proteolytic cleavage and signaling requires a context-dependent multivalent interaction between Notch receptors and ligands, whereas soluble monovalent forms of ligands are known to inhibit Notch signaling [24, 29]. We engineered a monovalent soluble DLL1 construct (sDLL1) comprising one DSL and two N-terminal EGF repeat domains, and compared its effects with multivalent clustered DLL1, a complex formed by DLL1-IgG Fc fusion protein, biotinylated anti-Fc antibodies and avidin that selectively triggers DLL1-Notch signaling [24]. The monomeric sDLL1 construct inhibited Notch signaling triggered by multivalent clustered DLL1 as manifested by the dose-dependent decrease in expression of the Notch downstream target Hes1 mRNA in treated murine 3 T3 fibroblast cells (Fig. 5a). Thus, sDLL1 acts as a competitive inhibitor of multivalent  DLL1-triggered signaling.Fig5Fig. 5

Monomeric soluble DLL1 or Dll1-ablated dendritic cells restrict Notch signaling and impair T-cell cytotoxic responses. (a) Expression of Notch downstream target Hes1 mRNA was assessed by qRT-PCR in 3 T3 cells treated with clustered DLL1 in the presence of soluble DLL1 (sDLL1) construct at indicated concentrations for 16 h. b, c T-cell proliferation was measured after co-incubating allogeneic T-cells labeled with Cell Tracer Violet fluorescent dye with bone marrow-derived Dll1 ^−/− or wild-type DC in the presence of soluble anti-CD3 for 5 days. In some T-cell cultures with wild-type DC, soluble DLL1 construct was added at the indicated concentrations. Representative Cell Tracer Violet dye dilution profile is shown (b). d Tumor volume was measured in LLC tumor-bearing mice treated with sDLL1 construct 1 mg/kg body weight, i.p. every 2 days for 20 days. e IFN-γ producing tumor-infiltrating cells from these mice were enumerated by ELISPOT assay on day 18 after LLC tumor initiation. Mean ± SEM, 8 mice per group; *, p < 0.05; **, p < 0.005. f, g C57BL/6 mice were transplanted with BALB/c heart allografts on day 0 and treated with sDLL1 construct (1 mg/kg) i.p. on days − 3, − 1, 1, 3, 5 and 7. f Heart allografted C57BL/6 mice log-rank survival. g IFN-γ ELISPOT assay on recipient CD8⁺T cells isolated after heart allograft and re-stimulated with mitomycin C-treated donor spleen cells in the presence of recipient C57BL/6 splenocytes. h Percentage of FoxP3⁺ cells among CD4⁺ splenocytes after heart allograft. Mean ± SEM, 4–8 mice per group; *, p < 0.05

Our results showing differential cytokine patterns in mice with DC-specific deletion of Dll1 and Jag2 ligands suggested that Notch ligands had differential effects on the induction of immune responses. We constructed a monovalent soluble JAG1 (sJAG1) construct comprising total five N-terminal domains (MNNL, DSL and 3 EGF repeats) of mouse JAG1, and evaluated the significance of JAG1-mediated Notch signaling in anti-tumor responses. LLC tumor-bearing mice were treated with sJAG1 at a dose of 1 mg/kg body weight or vehicle control, i.p. every 2 days. The treatment with sJAG1 resulted in significant reduction of tumor growth and improved survival of animals (Fig. 6c, d). Moreover, these effects were associated with the decreased numbers of CD4⁺FoxP3⁺ Treg cells. We also noted significantly reduced accumulation of tumor-infiltrating CD11b⁺Gr1⁺ cells (40.1 ± 8.3% vs. 11.0 ± 3.8% among CD45⁺ cells for control and sJAG1-treated groups, respectively), and increased IFN-γ production by lymphoid cells (Fig. 6f). These data suggest an important role of JAG1-mediated Notch signaling in the induction of anti-tumor T-cell responses.

We also assessed whether the engineered Notch ligand DLL1 and JAG1 constructs modulate the differentiation of memory T-cells in vitro in a T:DC stimulation co-culture. Results show that inhibition with soluble JAG1 or stimulation with clustered DLL1 constructs increased the frequency of CD8⁺T-central memory cells concomitant with a decrease in the frequency of CD8⁺T-effector memory cells (Fig. 7a and Additional file 1: Figure S1). The decrease in the frequency of CD8⁺Tem cells was not due to exhaustion as both constructs significantly decreased the expression of checkpoint molecule PD-1 (about 3-fold by sJAG1 and 2.5-fold by clustered DLL1) in CD8⁺Tem cells, but not in CD8⁺Tcm, in a dose-dependent manner (Fig. 7a and Additional file 2: Figure S2). The expression of CTLA-4 was negligible and unchanged following treatments with the constructs.Fig7Fig. 7

Dendritic cell Jagged expression correlates with PD-1 expression on T-effector-memory cells. a Purified T cells were stimulated in vitro in a T:DC (3:1) stimulation co-culture with allogeneic bone marrow-derived dendritic cells in the presence of CD3/CD28 beads (1 μg/mL) for four days with or without treatment with clustered DLL1 (1.5 μg/mL) or monovalent soluble JAG1 (20 μg/mL) constructs. Expression of CD62L, CD44, CTLA-4 and PD-1 was assessed on gated populations as indicated by flow cytometry. Dot plots from a representative experiment out of two independent experiments with duplicates are shown. b-c Lung tumor single cell suspensions from 10 patients were evaluated for the expression of NOTCH ligands on tissue-resident CD11b⁺CD11c^high dendritic cells and PD-1 and NOTCH receptors in populations of T cells by flow cytometry. NOTCH ligands in CD11b⁺CD11c^high cells were compared to PD-1 positivity of Tem and Tcm cells (b) or to NOTCH receptor positive T cell subsets by Pearson’s correlation (c.) All p-values were corrected using the Benjamani- Hochberg procedure; n = 8; * p < 0.05. Color code indicates the strength of correlation. d Scheme summarizing available data on the regulation of T cell responses by Notch ligands

We sought to determine if there is a relationship between the expression of NOTCH ligands on antigen presenting cells and T-cell phenotype in human lung cancers. We profiled the expression of NOTCH ligands, NOTCH receptors and PD-1 on various subsets of myeloid and lymphoid cells in tumor-infiltrating immune cells in primary lung cancers. The analysis revealed a highly significant correlation between the proportion of JAG1 or JAG2-expressing tissue-resident CD11b⁺CD11c^high DCs and the numbers of PD-1-expressing T-effector-memory (Tem) CD8⁺CCR7⁻CD45RA⁻ and T-terminal-effector (Temra) CD8⁺CCR7⁻CD45RA⁺ cells, with JAG1 demonstrating the strongest correlation (p = 0.0005) (Fig. 7b and Additional file 3: Figure S3). Correlations between DC-expressed DLL1 (p = 0.007) or DLL4 (p = 0.01) and PD-1 on Tem subsets were also observed; however, significance of these correlations was substantially less than that for JAG1 (Fig. 7b). Correlations between CD11b⁺CD11c^high cells expressing JAG ligands and PD-1-expressing CD8⁺CCR7⁺CD45RA⁻ Tcm were marginal, with no significant correlation between PD-1 in Tcm and Delta-like ligands (Fig. 7b).

In contrast to PD-1, numbers of Tem cells expressing NOTCH receptors highly significantly correlated with the proportion of DCs expressing DLL1 or DLL4 (Fig. 7c). While statistically significant correlation of Tem-expressed NOTCH2 and NOTCH3 was observed with DC-expressed JAG1 and JAG2, it was less pronounced. No statistically significant correlations were identified for Tcm cells, except for NOTCH4 receptor with DC-expressed JAG1, DLL1 and DLL4 (Fig. 7c). There was no correlation between NOTCH ligand expression on DCs and expression of PD-1 or NOTCH receptors in the populations of CD8⁺CCR7⁺CD45RA⁺ naïve CD8 T cells or naïve, effector or memory CD4 T-cells (Additional file 3: Figure S3). The above results imply that the interactions between select DC-expressed Notch ligands and Notch receptors in T-cells present a key point of regulation for T-cell-mediated immunity by modulation of T-cell differentiation and functionality in human lung tumors.

Murine Lewis lung carcinoma (LLC) and 3 T3 cell lines were obtained from the American Type Culture Collection (Manassas, VA). Murine MT5 pancreatic cells were a kind gift from Dr. Tuveson (Cold Spring Harbor Laboratory, Cold Spring Harbor, NY). Low-passage (less than 10) cultures were used for the experiments. All active cell cultures were checked for mycoplasma routinely using a commercial PCR test. Purity and identity of cells were also confirmed by flow cytometry (FACS) analysis with antibodies to appropriate markers.

Male and female C57BL/6 and BALB/c mice (7 to 8-week-old) used at equal numbers were purchased from The Jackson Laboratory (Bar Harbor, MN).

C57BL/6 mice with floxed alleles for the Dll1 gene were received from Dr. J. Lewis (Cancer Research UK, London, UK); Jag2 gene-targeted floxed mice were kindly provided by Dr. T. Gridley (Maine Medical Center, Scarborough, MN). Generation of Dll1 ^flox/flox and Jag2 ^flox/flox conditional knockout mice and genotyping of floxed and deleted alleles have been described previously [45–47]. B6.Cg-Tg(Itgax-cre)1-1Reiz/J mice expressing Cre recombinase under the CD11c (integrin-αX; CD11c-Cre) promoter were purchased from The Jackson Laboratory. The animals were housed in pathogen-free units.

We generated mice bearing deletion of Dll1 or Jag2 in CD11c⁺ cells by mating syngeneic B6.Cg-Tg(Itgax-cre)1-1Reiz/J mice expressing Cre-recombinase under CD11c promoter and DLL1 ^flox/flox or Jag2 ^flox/flox mice and then by crossing their progeny. In the resultant mice, CD11c⁺ cells with hetero- or homozygous allele deletion had genotype Dll1 ^flox/− Cre ^+/− , Jag2 ^flox/− Cre ^+/− , Dll1 ^−/− Cre ^+/− or Jag2 ^−/− Cre ^+/− , respectively. Their littermates with “floxed” alleles but without Cre recombinase transgene served as respective controls in all animal experiments. The allele deletion was confirmed by genotyping and by the assessment of Notch ligand mRNA expression in flow-sorted CD11c⁺DC populations from the spleen by PCR and RT-PCR using genomic DNA and RNA samples, respectively, with sets of primers specific for floxed and deleted alleles and for ligand mRNA described previously [24, 45–47] (Fig. 1a).

RT-PCR was utilized to confirm deletion of Notch ligand genes in CD11c⁺ cells. CD11c⁺ cells were isolated from splenocytes by flow sorting, as described below. RNA was extracted with RNeasy Mini kit and possible genomic DNA contamination was removed by on-column DNase digestion using the RNase-free DNase kit (Qiagen; Valencia, CA). cDNA was synthesized using SuperScript III Reverse Transcriptase kit (Invitrogen, Grand Island, NY) and used in PCR reactions with gene-specific primers, described previously [24]. Amplification of endogenous β-actin was used as an internal control.

We engineered a DLL1 construct where part of the soluble extracellular domain of mouse DLL1 protein comprising DSL, EGF1 and EGF2 domains with TEV and 6-His sequences was expressed in E. coli and isolated using Ni-column (Bio-Rad, Hercules, CA). The preparation was 90% pure as assessed by polyacrylamide gel electrophoresis with Coomassie R-250 staining. Mouse JAG1 construct comprising MNNL, and DSL followed by three EGF domains of JAG1 was produced in E. coli by MyBiosource, Inc. (San Diego, CA). The inhibitory activity of these reagents was confirmed by their ability to decrease Hes1 expression in response to Notch stimulation with the respective ligand in cell culture assay. These constructs are referred to in the text and figures as soluble DLL1 (sDLL1) and soluble JAG1 (sJAG1), respectively.

To induce tumors, mice were inoculated subcutaneously (s.c.) in the flank with 0.25 × 10⁶ LLC or 10⁶ MT5 cells as described previously [25, 30, 48]. Tumor volume was measured with calipers. For survival experiments, mice were observed until they reached exclusion criteria as determined by the IACUC protocol. To evaluate immunological correlates, mice were euthanized on days 17–18 and 14–15 for LLC and MT5 models, respectively.

To inhibit ligand-specific Notch signaling, tumor-bearing mice received sDLL1 or sJAG1 at a dose of 1 mg/kg body weight (25 μg per injection) of the protein in 100 μl of vehicle (Saline, 5% glycerol) intraperitoneally (i.p.) every other day. The control groups received 100 μl of saline vehicle instead of the ligand. Multivalent form of DLL1 (clustered DLL1) was utilized to stimulate DLL1-mediated Notch activation in vivo at a dose of 0.2 mg/kg body weight (5 μg per injection) of DLL1-Fc fusion protein i.p. every other day, as described previously [24, 25]. All treatments started at day 3–4 after tumor injection.

Single cell suspensions of tumors from mice were prepared using Miltenyi Biotec (Auburn, CA) Tissue dissociation kit and Gentle MACS instrument according to the manufacturer’s recommendations. Lymphocytes were than enriched by Lympholyte M (Cedarlane, Burlington, Canada) gradient centrifugation and used to quantify the cytokine producing cells. LLC cells express a defined antigenic peptide MUT1 (spontaneously mutated connexin 37), FEQNTAQP allowing evaluation of antigen-specific CD8⁺T-cell responses [49]. Briefly, 5 × 10⁵ cells per well from the tumor cell suspensions or 2 × 10⁵ cells per well from tumor-draining LN were restimulated with 10 μM of MUT1 or irrelevant control peptide loaded on autologous mitomycin-C treated splenic cells for 48 h and IFN-γ or IL-4-producing cells were evaluated by dual ELISPOT assay (CTL, Shaker Heights, OH) according to the manufacturer’s protocol. Peptides were synthesized by the American Peptide Company, Inc. (Sunnyvale, CA). Alternatively, gradient centrifugation-enriched cells (1.5 × 10⁵ cells per well) were stimulated with Dynabeads Mouse T-Activator CD3 and CD28 antibodies coupled to beads (Life Technologies, Carlsbad, CA), as recommended by the manufacturer, and IFN-γ or IL-4-producing cells were enumerated by ELISPOT assay. Parts of tumor single cell suspension, splenocytes and LN cell population from the same mice were used for immunophenotyping of cells by FACS (see below).

The effect of Notch ligand gene knockout on T-cell stimulatory activity of DCs was evaluated in allogeneic mixed lymphocyte reaction (MLR). DCs were generated from bone marrow of wild type or knockout animals in the presence of GM-CSF and IL-4, as described earlier [30]. T-cells from allogeneic mice isolated by negative selection using T-cell isolation columns (R&D Systems, Minneapolis, MN) were labeled with Cell Tracer Violet dye (ThermoFisher Sci., Grand Island, NY) and incubated for 5 days with bone-marrow derived DC in the presence of a soluble anti-CD3. Dye dilution in proliferating T-cells was measured by flow cytometry.

Wild type C57BL/6 mice were transplanted with MHC-mismatched heterotopic BALB/c heart allografts as previously described [50]. Mice with transplants were treated with sDLL1 construct (1 mg/kg body weight, i.p.) or vehicle control (Saline, 5% glycerol) on days − 3, − 1, 1, 3, 5 and 7 relative to the day of transplantation. Heart allograft survival was evaluated daily by palpation, and rejection was confirmed by laparotomy. IFN-γ ELISPOT assay was performed on recipient splenic CD8⁺ T-cells isolated at the time of graft rejection and restimulated with mitomycin C treated donor BALB/c spleen cells in the presence of self-antigen presenting cells.

Ten freshly resected de-identified lung cancer samples were obtained under an informed consent from an unselected patient population in terms of tumor types and stages, age, sex and ethnicity. Tumor tissue single cell suspensions were prepared using the Tissue dissociation kit and Gentle MACS instrument from Miltenyi Biotec according to the manufacturer’s recommendations. Cells were live frozen until the analysis. Analysis of tumor- infiltrating immune cells was performed by flow cytometry using lineage-specific antibodies and antibodies to PD-1, NOTCH receptors and ligands. After gating for T-cell and myeloid subsets, cell populations were further gated by PD-1, Notch ligand, or Notch receptor positivity. Populations of PD-1-positive, Notch ligand-positive, or Notch receptor-positive cells were compared using Pearson’s correlation in R using the packages “Hmisc” and “corrplot”.

Fluorochrome-labeled cell surface or intracellular protein specific antibodies were obtained from BD Biosciences, Pharmingen and eBioscience, Inc. (San Diego, CA). For staining of cell-surface markers, cells were incubated with the antibodies for 20 min on ice. For intracellular FoxP3 staining, cells were first stained for lineage-specific markers, then permeabilized for 20 min with BD fixation/permeabilization kit and incubated with fluorochrome-labeled FoxP3-specific antibody. Matched fluorochrome-conjugated isotype IgG controls were used as staining controls. Flow sorting of CD11c⁺ cells from splenocytes of wild type mice or animals with Notch gene deletion was performed using Aria IIu cell sorter (BD Immunocytometry). Nonviable cells were excluded using 7-amino actinomycin D staining. Antigen negativity was defined as having the same fluorescent intensity as the isotype control. FACS data were acquired using a FACS LSR II (BD Immunocytometry) or a Guava EasyCyte HT (Millipore) instrument and analyzed with FlowJo software (Tree Star, Ashland, OR).

Data were analyzed using the GraphPad Prism 4.0 software (GraphPad Software Inc., San Diego, CA) and presented as mean ± SEM. Comparisons between treatment and control groups were performed using one-way ANOVA followed by Dunnett’s post-tests. Comparisons between two groups were performed using two-tailed unpaired t tests. Survival curves were compared using Mantel-Haenszel log rank test. The p-values for multiple comparisons in human sample analyses were adjusted using the Benjamani-Hochberg procedure. Differences were considered statistically significant when p-values < 0.05.