Combinatorial Immunotherapy with Agonistic CD40 Activates Dendritic Cells to Express IL12 and Overcomes PD-1 Resistance

Over the past decade, as checkpoint inhibitors have entered the clinic, melanoma has become a model for successful immunotherapy treatment. Standard-of-care therapy with anti–programmed death protein-1 (PD-1) and anti-cytotoxic T lymphocyte–associated antigen-4 (CTLA-4) antibodies (Ab) has transformed many metastatic melanomas from a rapidly progressing, fatal illness into treatable disease (1). Despite these advances, treatment resistance remains a clinical challenge in approximately 50% of patients. Overall 5-year survival for treatment-naïve patients with metastatic melanoma treated with nivolumab (anti–PD-1) is 44% (2). Addition of ipilimumab (anti–CTLA-4) increases 5-year survival rates to 52% (2). Alternative treatment options are limited. Chemotherapy does not confer a survival benefit (3, 4), and IL2 treatment causes severe toxicity with minimal response rates (5). Although targeted therapy with BRAF and MEK inhibition is effective in some patients, responses are typically not durable (6). These statistics reveal a critical need for novel strategies to overcome treatment resistance.

Checkpoint inhibitor resistance can develop in many ways (7). A lack of cytotoxic T cells or an abundance of exhausted or regulatory T cells (Treg) in the tumor microenvironment (TME) can hinder checkpoint blockade (8–10). Myeloid cells in the TME also play a crucial role in setting the immunological status of the tumor. Tumor-associated macrophages (TAM) are multifunctional cells that support tumor growth by promoting angiogenesis, invasion, metastasis, and immunosuppression (11). Increased presence of TAMs is a negative prognostic marker in many cancers, including melanoma, and may reduce the efficacy of checkpoint blockade (12, 13). Unlike TAMs, dendritic cells (DC) are a rare population in the tumor. As the main antigen-presenting cells (APC), they act as a bridge between innate and adaptive immunity, coordinating effective antitumor responses through T-cell priming. Because of this fundamental role in T-cell activation, new strategies are being developed to stimulate DCs, including tumor antigen vaccines (14, 15). Thus, simultaneously targeting innate immune cells and T cells in the TME is a promising strategy for overcoming checkpoint inhibitor resistance.

One promising treatment target is CD40, a costimulatory receptor highly expressed on APCs (16). CD40 can be engaged and activated through agonistic Abs. As a therapeutic agent, CD40 agonist (CD40ag) has a broad range of effects on different cell types in the TME. Upon CD40ag treatment, macrophages display increased nitric oxide production and expression of proinflammatory cytokines, including TNFα, in many tumor models such as B16 melanoma (17, 18). Furthermore, CD40 signaling is critical for DC licensing and antigen presentation (19). CD40ag treatment has been associated with increased CD103⁺ migratory conventional type 1 DCs (cDC1) in tumor-draining lymph nodes and increased intratumoral CTLs in a pancreatic cancer model, as well as increased activation of intratumoral DCs in the MC38 colorectal cancer model (18, 20). Although CD40ag monotherapy has been shown to slow tumor growth in Braf^V600E/Pten^−/− murine melanomas independently of T cells (21), it also enhances T cell–dependent anticancer immune responses (22, 23). CD40ag has also been shown to increase the efficacy of chemotherapy in mesothelioma and pancreatic cancer models (24–26), of TLR7 agonism in a malignant mesothelioma model (27), and of checkpoint blockade in models of colon (28), pancreatic (29), and mammary (30) carcinoma, although not all of these combinations were curative.

Another emerging strategy to reprogram the TME is colony-stimulating factor-1 receptor (CSF1R) blockade. CSF1R activation by cognate ligand macrophage CSF (M-CSF/CSF1) induces macrophage production, survival, and differentiation (31). In the context of malignancy, CSF1R⁺ macrophages display a protumorigenic phenotype with increased PD-L1 surface expression and TGFβ secretion (32). CSF1R expression correlates with poor prognosis in melanoma, breast, ovarian, and pancreatic cancers (12, 33, 34). CSF1R blockade reduces the frequency of tumor-promoting TAM subsets (35) and can increase expression of inflammatory markers on remaining macrophages, promoting lymphocyte infiltration into tumors (31). Combining CSF1R blockade with checkpoint inhibition slowed tumor growth in a pancreatic cancer model and induced complete regression in two models of melanoma (12, 35).

CSF1R blockade has been demonstrated to augment the action of CD40ag and significantly improve survival in various tumor models (36–38). Our group has previously shown that combining CD40ag and CSF1R blockade induces a population of proinflammatory TAMs and depletes TAM subsets associated with angiogenesis and invasion (38). Given that both CD40ag and CSF1R blockade independently enhance checkpoint inhibition, we hypothesized that combining CD40ag and CSF1R blockade with PD-1 blockade would sufficiently boost immune activity in the TME to overcome checkpoint inhibitor resistance. We tested the combination of CD40ag, CSF1R blockade, and PD-1 inhibition (i.e., TTx) in YUMMER1.7, an anti–PD-1–resistant murine melanoma model. Our work shows that combining myeloid- and T cell–targeting treatments induces complete tumor regression and results in durable long-term survival. Furthermore, we demonstrated that immediately after administration, TTx was primarily a CD40ag-driven treatment. Using a novel single-cell cytokine secretion assay, we functionally profiled DCs to identify a subset of IL12-producing DCs as the principal effectors of TTx treatment in this early phase. This work paves the way for the clinical translation of myeloid-targeting strategies in checkpoint inhibitor-resistant melanoma.

YUMMER1.7 cells were derived from YUMM1.7 cells, which were generated from a cutaneous mouse melanoma containing the alleles Braf^V600E, Pten^−/−, Cdkn2a^−/− (39). Irradiation of YUMM1.7 included three rounds of 1,500 J/m² UVB (3 W for 500 seconds) when cells were 50% to 70% confluent. Cells were given time to recover and proliferate before being replated and before the next UV treatment. After the final UV treatment, a single cell was clonally expanded. RENCA cells were obtained from the ATCC. YUMMER1.7 stably expressing GFP and luciferase was generating using retroviral transduction, as previously described (40, 41). All cell lines were maintained in DMEM/F-12 (Thermo Fisher Scientific, 11320–033), including l-glutamine and 2.438 g/L sodium bicarbonate, and supplemented with 1× nonessential amino acids (Thermo Fisher Scientific, 11140050), 10% FBS (Thermo Fisher Scientific, 16140–071), and 1× penicillin and streptomycin (Thermo Fisher Scientific, 15140–122). All cells were cultured at 37°C, 5% CO₂, and kept at low passage (less than 10 passages) and showed negative test results for Mycoplasma contamination by Mycoplasma testing within the last year.

All animal studies were conducted in accordance with protocols approved by the Yale University Institutional Animal Care and Use Committee before initiation of experiments. For subcutaneous tumors, C57BL6/J mice (purchased from The Jackson Laboratory) were injected with a total of 0.5×10⁶ YUMMER1.7 cells (Braf^V600E/ Pten^−/−Cdkn2a^−/−; a UV-irradiated melanoma cell line previously described (40), RRID:CVCL_A2AX). BALB/c mice were injected subcutaneously with a total of 0.5×10⁶ RENCA cells (RRID:CVCL_2174). Mice were treated with 50-μg anti-CD40 (FGK4.5, BioXCell, RRID:AB_1107601), 200-μg anti-CSF1R (m CSF1R.2 m.G1- D265A, Bristol Myers Squibb), and 200-μg anti–PD-1 (RMP1–14, BioXCell, RRID:AB_10949053), along with corresponding isotype control rat IgG2a at 8 mg/kg (2A3, BioXCell, RRID:AB_1107769), every 3 days intraperitoneally for a total of five treatments, beginning 7 days after tumor inoculation. Tumors were measured by calipers at 3- to 4-day intervals, and the endpoint was determined when tumors reached 1 cm³ per Yale University Institutional Animal Care and Use Committee approved protocols.

For the metastatic tumor model, C57BL6/J mice were administered with a total of 0.5×10⁶ YUMMER1.7 cells labeled with EGFP and firefly luciferase via intracardiac injections. In brief, mice were anesthetized with 100 mg/kg ketamine and 10 mg/kg xylazine, shaved, and disinfected on the ventral thorax. Ketamine and xylazine were acquired and prepared in compliance with Yale University Policy on the Use of Controlled Substances in Research. EGFP⁺ luciferase⁺ 0.5×10⁶ YUMMER1.7 cells were filtered through a 40-mm mesh filter and then injected into the left ventricle. Immediately after injection, mice were administered 100 μL of 15-μg/mL D-luciferin (PerkinElmer) intravenously and visualized using the IVIS imaging system. Temperature was maintained with heating pads, and mice were monitored every 15 minutes until recovery from anesthesia. Following metastasis induction, mice were monitored every other day and sacrificed upon exhibiting initial signs of distress.

For depletion experiments, mice were treated by intraperitoneal injection with Abs against CD8a (2.43, BioXCell, RRID:AB_1125541), CD4 (GK1.5, BioXCell, RRID:AB_1107636), or IFNγ (R4–6A2, BioXCell, RRID:AB_1107692 or XMG1.2, BioXCell, RRID:AB_1107694), along with corresponding isotype control rat IgG2a (2A3, BioXCell, RRID:AB_1107769), beginning 1 to 3 days before therapy initiation and continuing twice weekly for a minimum of 3 weeks. All depleting Abs were used at 8 mg/kg per treatment.

For the IL12 mRNA experiments, mice inoculated with YUMMER1.7 tumors were treated with a mouse (m)IL12 mRNA (Moderna) at 5 μg and/or 200-μg anti–PD-L1 (Moderna) 7 days after tumor implantation, along with corresponding isotype control rat IgG2a (2A3, BioXCell, RRID:AB_1107769) for the latter therapy at 8 mg/kg per treatment. The mRNAs used in these studies (both IL12 and non-translating control, also administered at 5 μg) were lipid nanoparticle (LNP)-encapsulated. mRNA was synthesized and LNP formulated as previously described (42, 43).

At least 3 separate samples for untreated and anti–PD-1+anti-CSF1R+anti-CD40-treated YUMMER1.7 tumors were collected 24 to 72 hours after treatment (with treatment occurring at 7 days after tumor inoculation as previously described) and formalin-fixed, paraffin-embedded by the Yale Pathology department Histology core, according to standard protocols. 5-μm-thick sections, cut in a Leica cryostat, were stained by either H&E or IHC using Cell Signaling Technology Abs against CD45 (D3F8Q), CD3 (D4V8L), CD8 (D4W2Z), FOXP3 (D6O8R), and F4/80 (D2S9R). Samples were analyzed in a Leica SP5 confocal microscope at Yale CCMI Imaging Core. Staining was quantified by counting stained cells per mm². At least three separate fields of view per tumor were quantified for analysis.

Whole-blood was collected from healthy (n = 5), untreated YUMMER1.7-bearing (n = 8), and treated YUMMER1.7-bearing (n = 21) mice 7 to 9 days after tumor inoculation (24 hours after treatment if applicable). Treatments included anti–PD-1 monotherapy, anti-CSF1R monotherapy, anti-CD40 monotherapy, each pairwise combination of these three therapies, and all three therapies together (n = 3 per condition). Samples were collected into tubes containing 5 mmol/L EDTA (AmericanBio) in PBS and centrifuged at 1,000 x g twice at 4°C. Plasma was collected and frozen at −80°C until further processing. Samples were sent to Eve Technologies Corp. for multiplexed protein quantification using the Mouse Cytokine/Chemokine 44-plex Discovery Assay Array (Eve Technologies, catalog no. HD65). Eve Technologies then calculated the requisite standard curves and returned quantified protein concentration measurements (pg/mL). Out of range values were assigned the highest or lowest standard curve value per Eve Technologies instructions. Multiplex secretion samples were visualized and hierarchically clustered using the clustermap function from the Seaborn module in Python. The non-normalized data were then embedded in two dimensions using principal component analysis (PCA), as implemented in the multivariate statsmodels module in Python.

Systemic cytokine secretion measurements from a patient group were sourced from Weiss and colleagues (44). The study described previously in Weiss and colleagues was approved by the Yale University Institutional Review Board and was conducted in accordance with ethical guidelines as outlined by the U.S. Common Rule and with an assurance filed with and approved by the U.S. Department of Health and Human Services.

In its entirety, this dataset includes patients with biopsy-proven melanoma, non–small cell lung cancer, or renal cell carcinoma who had previously progressed on anti–PD-(L)1. Further details regarding written consent, inclusion/exclusion criteria, and other patient characteristics can be found in the original publication. The original study design included six therapy cohorts. Cohorts 1, 3, and 5 received cabiralizumab (anti-CSF1R) at a fixed dose (4 mg/kg) in combination with APX005M (anti-CD40) at escalating doses (0.03, 0.1, and 0.3 mg/kg). Cohorts 2, 4, and 6 received the same therapies as cohorts 1, 3, and 5, respectively, in addition to nivolumab (anti–PD-1) at a fixed dose (4 mg/kg). For our analyses, we limited the dataset to patients with melanoma only, which included patients in cohorts 2 (n = 2), 3 (n = 2), 4 (n = 2), and 6 (n = 5).

Patient-systemic cytokine measurements were taken from serial plasma samples that were collected from patients before and 4 and 24 hours after two cycles of therapy (therapy cohorts described above). Samples were processed and sent to Eve Technologies for multiplexed protein quantification as described in “Mouse systemic cytokine profiling and analysis” above. Similar to the mouse data, the patient data from the first cycle of therapy were embedded in two dimensions using PCA as implemented in the multivariate statsmodels module in Python. In addition, pairwise correlations between cytokines and chemokines measured from patients were calculated using the pearsonr function from the scipy.stats package in Python and visualized using custom Python scripts.

Tissues from untreated and anti–PD-1+anti-CSF1R+anti-CD40-treated YUMMER1.7 tumors (at 8 days after tumor inoculation and 24 hours after treatment, if applicable) used for FACS, single-cell RNA sequencing (scRNA-seq), single-cell secretion profiling, and intracellular cytokine staining were minced in RPMI (Gibco) with 2% FBS (Thermo Fisher Scientific), and then incubated with agitation at 37°C, with 1 mg/mL collagenase IV (C5138, Sigma-Aldrich) and 0.1 mg/mL DNAse I (10104159001, Roche) for 30 minutes. Samples were placed on ice before filtering through a 70-μm filter to remove undigested pieces. Red blood cells were lysed with ammonium–chloride–potassium lysis buffer (Lonza), and single-cell suspensions were washed and resuspended in 2% FBS. Cells were then counted and ready for further analysis. For scRNA-seq, one tumor was used per sample. For single-cell cytokine secretion studies, eight tumors were pooled per sample and resuspended in RPMI with 20% FBS. Tumors that displayed expected growth (i.e., tumors that measured 100 to 200 mm³ at 7 days after tumor inoculation) were selected for the aforementioned experiments.

For scRNA-seq, samples were sorted into three populations: CD45⁺CD3⁺ (T cells), CD45⁺CD3⁻ (non-T immune cells), CD45⁻CD3⁻ (tumor and stroma). The populations were then recombined in a 2:1:1 ratio and submitted for sequencing (described below). For single-cell secretion profiling, samples were sorted into three populations: CD45⁺CD64⁺F4/80⁺ (macrophages), CD45⁺CD103⁺ (migratory cDC1), and CD45⁺MHCII⁺CD11b⁺CD11c⁺Ly6G⁻CD64⁻F4/80⁻ (other DCs). The populations then underwent single-cell secretion profiling (described below).

Cell staining was performed using the following Abs: CD45 (I3/3.2), CD3 (17A2), CD19 (6D5), MHCII (M5/ 114.15.2), CD64 (X54–5/7.1), CD11b (M1/70), Ly6G (1A8), F4/80 (BM8), CD11c (N418), purchased from BioLegend. Abs for the TCRβ chain (H57–597) and CD103 (M290) were purchased from BD Biosciences, and Live/Dead Aqua was purchased from Invitrogen. Cell sorting was performed in a FACS Aria, and cells were kept at 4°C while sorted and collected in RPMI supplemented with 20% FBS (Thermo Fisher Scientific).

A total of 10,000 cells from each condition (TTx-treated and control) were processed and pre-sorted as indicated above and loaded onto the 10X Chromium System. Library preparation with VDJ barcoding was conducted by the Yale Center for Genome Analysis (YCGA) using the Chromium Single Cell 5′ v1 Reagent Kit (10X Genomics) according to the manufacturer's instructions [see: (https://medicine.yale.edu/keck/ycga/sequencing/10x/singcellsequencing/)]. Briefly, single cells were isolated in 1 nL of gel beads in emulsion using the GemCode technology. Cell barcoding, lysis, and reverse transcription of mRNA occurred within each gel bead in emulsion. cDNA libraries were then generated using next-generation sequencing PCR amplification with the 5-prime chemistry. cDNA quality control was performed with an Agilent TapeStation. Both scRNA-seq and single-cell T-cell receptor sequencing (scTCR-seq) libraries were generated for each sample, and both libraries underwent paired-end sequencing on the HiSeq 4000 platform (Illumina) by YCGA.

scRNA-seq reads were aligned to the reference genome mm10 using the cellranger count pipeline from the CellRanger software v3.1.0 (10X Genomics) to generate a cell-by-gene matrix for each library. Downstream analysis was performed using the Seurat package in R and the Scanpy package in Python. Cells were filtered for quality control to avoid doublets and dead cells. Cells with total gene counts less than 200 or greater that 2,500 or with mitochondrial gene counts exceeding 5% were removed from the dataset for downstream analyses. Dimensionality reduction and visualization were completed using the Scanpy implementation of Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) in Python.

We adapted the cell-type annotation pipeline from Kumar and colleagues (45) to use a simple feed-forward neural network to label our sequencing data by broad cell type, as was done in Wasko and colleagues (46). Differentially expressed genes (DEG) between cell type populations were identified using the rank_genes_groups function from the Scanpy module in Python with default parameters (log₂FC > 1.5 and P_adj < 0.01). We then performed enrichment analysis on the top 100 DEGs for each group using g:Profiler (47). Gene Ontology (GO) biological processes, Reactome (REAC), KEGG, and WikiPathways (WP) were used as reference databases. Enrichment P values were adjusted for false discovery using the Benjamini–Hochberg procedure.

NicheNet is a powerful algorithm that makes inferences about ligand binding at the cell surface from patterns in expression of target genes in receiving cells (48). To generate hypotheses about the potential ligands driving changes in gene expression observed in DCs, macrophages, and each T-cell subset, we identified target genes as genes significantly upregulated in these cell populations after TTx compared with control (log₂FC > 0.25 and P_adj < 0.05). NicheNet then identified likely ligands driving gene expression patterns by correlating observed expression with expected regulatory potential between ligands and target genes from the prior model. NicheNet is available as an open-source software package in R.

Myeloid cells from the scRNA-seq data, as identified during cell-type annotation (see above), were also scored by average expression of genes associated with cDC1 and cDC2 subsets, and the novel mregDC (mature DCs enriched in immunoregulatory molecules) state (49). Scoring was accomplished using the score_genes function from the Scanpy module in Python. The cDC1 score summarized expression of Xcr1, Clec9a, Cadm1, and Naaa. The cDC2 score summarized expression of Itgam, Cd209a, Sirpa, and H2-DMb2. The mregDC score summarized expression of Cd80, Cd40, Cd83, Relb, Cd274, Pdcd1lg2, Cd200, Fas, Socs1, Socs2, Aldh1a2, Ccr7, Fscn1, Il4ra, Il4i1, Myo1g, Cxcl16, Adam8, Icam1, Marcks, and Marcksl1.

scTCR-seq reads were processed using the cellranger vdj pipeline from CellRanger v3.1.0 (10X Genomics, reference genome mm10). TCR contigs called by cellranger vdj with high confidence were retained for downstream analyses. Downstream analyses were performed using the Scirpy extension for Scanpy in Python (50). Extracted clonotypes were analyzed using custom scripts written in Python.

The single-cell secretion profiling experiments were performed as previously described (38). Abs for each target were selected from commercially available pairs validated for ELISA (Supplementary Table S1). Briefly, captured Abs were flow-patterned onto epoxysilane-coated glass slides (Super-Chip; Thermo Fisher Scientific). The polydimethylsiloxane microchips containing nanowell and the Ab-barcoded glass slides were blocked using complete RPMI + 10% FBS. As indicated in “FACS,” samples were FACS-sorted into three populations: CD45⁺CD64⁺F4/80⁺ (macrophages, ∼100K cells), CD45⁺CD103⁺ (migratory cDC1), CD45⁺MHCII⁺CD11b⁺CD11c⁺Ly6G⁻CD64⁻F4/80⁻ (other DCs). The latter two populations were combined in the single-cell cytokine secretion assay into a single DC population (∼20K cells). Cell staining was performed using the following Abs (from R&D Systems unless otherwise noted): Chi3l3 mouse (DY2446), MMP-9 mouse (DY6718), IL10 mouse (DY417), CCL17 mouse (DY529), CCL22 mouse (DY439), CCL2 mouse (DY479), CCL5 mouse (DY478), IFNγ mouse (DY485), CXCL1 mouse (DY453), CCL3 mouse (DY450), IL6 mouse (DY406), TNFα mouse (88–7324–88, Thermo Fisher Scientific), IL12p40 mouse (555165, BD Biosciences).

Following the FACS described above, cells were resuspended in complete RPMI with 20% FBS and supplemented with 125 nmol/L of live cell marker (Calcein AM; Thermo Fisher Scientific) for live-cell detection. After pipetting cells into the microchip wells, the microchip was covered with the Ab-barcode slide to seal the wells, and cells were incubated for 12 hours at 37°C. The device was then imaged with an automated inverted microscope (Axio Observer; Zeiss) to record well position and cell locations. The device was then disassembled to perform the sandwich immunoassay. The glass slide was incubated with a mixture of detection Abs (Supplementary Table S1) for 2 hours at room temperature, followed by an incubation at 37°C with 20 μg/mL streptavidin-APC (eBioscience) for 30 minutes, rinsed with PBS and deionized water, and scanned with a Genepix 4200A scanner (Molecular Devices). In this study, 543 untreated macrophages, 368 treated macrophages, 299 untreated DCs, and 285 treated DCs were isolated as single cells and their readouts retained for downstream analyses.

Device images were analyzed using a custom script in MATLAB (MathWorks, RRID:SCR_001622) described previously (38). Following automatic well and live-cell detection, and signal image registration by the software, the results were manually curated. The software then automatically extracted the signal intensity from each Ab barcode for the nanowell arrays. The signal across the chip for each individual Ab was normalized by subtracting a moving Gaussian curve fitted to the local zero-cell well intensity levels. A secretion threshold for each Ab was then set at the 99th percentile of all normalized zero-cell wells. Finally, the data were transformed using the inverse hyperbolic sine with a cofactor set at 0.8× secretion threshold.

Of the isolated single cells, 227 untreated macrophages, 180 treated macrophages, 105 untreated DCs, and 140 treated DCs were found to secrete at least one cytokine or chemokine on the panel above threshold. These single-cell secretion measurements were then visualized using the Seaborn module in Python. The data were embedded in two-dimensional space with UMAP. To investigate functional heterogeneity, cluster analysis of the single-cell secretion data was accomplished with PhenoGraph, which is an unsupervised, graph-based clustering method developed to identify subpopulations in high-dimensional single-cell data (51). Extracted clusters were analyzed using custom scripts written in Python as indicated in “Code availability” below. UMAP and PhenoGraph (RRID:SCR_016919) are both available as publicly available software packages in Python.

Single-cell suspensions from untreated and anti–PD-1+anti-CSF1R+anti-CD40-treated YUMMER1.7 tumors were obtained as detailed before (“Tumor processing”), and then were incubated with 5ug/mL Brefeldin A (420601, BioLegend) in RPMI complete media, in 6-well plates at 37°C for 4 hours. Cells were then washed and processed for immunostaining. In brief, cells were stained in 2% FBS PBS for extracellular antigens, fixed and permeabilized with CytoFix/CytoPerm (554714, BD Biosciences), and then stained for intracellular cytokines. Cells were analyzed in an LSRFortessa (Becton Dickinson), and data were processed in FlowJo (TreeStar, Inc., RRID:SCR_008520). The following Abs were used for flow cytometry panels: CD45 (30-F11), CD11b (M1/70), Ly6G (1A8), NK1.1 (PK136), CD64 (X54–5/7.1), CD4 (RM4–5), CD8a (53–6.7), CD25 (PC61), FoxP3 (MF-14), PD-1 (29F.1A12), IL12/23 p40 (C15.6), IFNγ (XMG1.2), TNF (TN3–19.12), from BioLegend, and FcBlock anti-Mo CD16/CD32 (93), and Live/Dead eFluor506 from eBiosciences. The following populations were assessed: CD4⁺FOXP3⁻ T cells (Live⁺CD45⁺CD11b⁻NK1–1⁻CD4⁺FOXP3⁻) and myeloid cells (Live⁺CD45⁺CD11b⁺Ly6G⁻).

Data were presented as mean ± SEM unless otherwise specified. Statistical analysis was generally performed by two-sided, unpaired Student t test and the Benjamini–Hochberg method of correction for pairwise multiple comparisons, or as specified in the figure legends. Normal and equal distribution of variances was assumed. Values were considered significant at P < 0.05. All analyses were performed using custom python scripts. For single-cell measurements, statistics were calculated using the bootstrapping procedure to construct confidence intervals associated with sampling error in single-cell data. Bootstrapping analysis was performed using the bootstrapped module in Python with default parameters. Statistical differences in measurements were decided by nonoverlapping bootstrapped error bars of the mean.

The code used to generate the figures presented in this study is available in the following public GitHub repository: [https://github.com/miller-jensen-lab/krykbaeva-ttx]. In addition, the custom MATLAB software for extracting and processing single-cell secretion data is available in the following public GitHub repository: [https://github.com/miller-jensen-lab/Single-Cell-Analysis].

The scRNA-seq data generated in this study are publicly available in Gene Expression Omnibus (GEO) at GSE230004. The scTCR-seq data generated in this study are publicly available in GEO at GSE236303. All other data supporting the findings of this study are available from the co-corresponding authors upon reasonable request.

We sought to address tumor resistance to checkpoint therapy by supplementing PD-1 inhibition with CSF1R inhibition (CSF1Ri) and CD40 activation (i.e., combinatorial therapy or TTx). We studied subcutaneous YUMMER1.7 tumors, a murine melanoma model driven by BRAF^V600E, CDKN2A^−/−, PTEN^−/−, UV-irradiated and with partial sensitivity to anti–PD-1 treatment (40), in immunocompetent C57BL/6 mice. Visible tumor formation occurred within seven days. Combinatorial therapy was initiated on day 7, and complete regression (if it occurred) was confirmed within 14 days after therapy initiation (Fig. 1A and B, Supplementary Fig. S1A). TTx produced durable responses without recurrence in 79% (11/14) of mice with cutaneous YUMMER1.7 tumors (Fig. 1B and C). There was a significant survival advantage of TTx compared with the monotherapies and doublet combinations, exhibiting more than additive effects (Fig. 1C). A similar response profile was seen with RENCA tumors, a renal cell carcinoma model less immunogenic than YUMMER1.7, which was not otherwise responsive to anti–PD-1 monotherapy (Fig. 1D; Supplementary Fig. S1B). Furthermore, in the YUMMER1.7 metastatic model, generated through intracardiac injection, we observed an additional approximately 40% overall survival (OS) in mice treated with TTx compared with anti–PD-1 alone (Fig. 1E).

To determine the kinetics of response to TTx, we examined histologic sections of YUMMER1.7 tumors harvested 24 to 48 hours after treatment initiation. Compared with untreated controls, treated tumors exhibited significant necrosis and reduced mitotic activity within 48 hours, as well as a slight increase in intratumoral CD3⁺ T cells within 24 hours (Fig. 1F and G). Furthermore, using flow cytometry, we observed increased intratumoral myeloid infiltrate (Fig. 1H) 24 hours post-TTx. Given the rapid response induced by TTx, we focused on changes at this early time point.

To understand how TTx activated immune pathways, we conducted systemic cytokine profiling of blood samples from treated and control YUMMER1.7-bearing mice. We observed a robust increase in inflammatory cytokines and chemokines 24 hours after treatment (Fig. 2A). This included IL6 and TNFα, which are associated with effective antitumor immune responses; CXCL9 and CCL5, which have been shown to promote CD8⁺ and CD4⁺ T-cell migration to the TME (52, 53); and IFNγ and IL12p40, which are critical for TAM repolarization, CTL recruitment and activation, and generation of antitumor Th1 CD4⁺ T cells (Fig. 2A and B). Taken together, TTx-treated mice exhibited a proinflammatory systemic response, with activation of cytokine signals that target innate and adaptive immune cell populations in the TME.

To resolve the contribution of each treatment component to the observed induction of cytokines, we embedded the multiplexed secretion data in two dimensions using PCA (Fig. 2C). PC1 stratified samples by CD40ag inclusion, implicating CD40ag as the primary driver of variability in serum cytokine secretion in this early-stage of treatment. Inspecting the top coefficients along PC1 identified cytokines and chemokines associated with a CD40ag-induced serum profile (Fig. 2D), including TNFα, IFNγ, IL12, CCL5, and IL6. This discovery is consistent with previous studies showing that CD40ag activates proinflammatory cytokines and chemokines in murine melanoma models (38).

Although PC2 did not correlate with treatment (Supplementary Fig. S2A), PC3 further separated CD40ag-inclusive samples by whether CSF1R blockade was also added (Fig. 2C, orange vs. red). These clusters correlated with treatment efficacy, as CD40ag+CSF1Ri±anti–PD-1 treatments yielded better survival outcomes than CD40ag±anti–PD-1 treatments (Fig. 1C). To understand the added benefit of CSF1Ri with CD40ag, we examined fold changes in serum secretion over control of these two treatments separately and in combination (±anti–PD-1). Whereas CSF1Ri alone did not induce secretion of CD40ag-induced proinflammatory factors, such as IFNγ and IL12, CSF1Ri+CD40ag treatment did have a synergistic effect on their secretion (Fig. 2E). This synergy was also observed for many chemokines and growth factors (Supplementary Fig. S2B). These results indicated that CD40ag was largely responsible for activation of inflammatory signaling, whereas the addition of CSF1R blockade served to amplify this signaling.

Next, we investigated whether these treatment-induced changes in systemic cytokine secretion were also observed in humans. We analyzed blood samples taken from patients in a Phase I clinical trial (NCT03502330) of APX005M (CD40ag) and cabiralizumab (CSF1Ri) with or without nivolumab (anti–PD-1), whose disease had previously progressed on anti–PD-(L)1 (44). We included data from 11 patients with melanoma enrolled in four of the six original cohorts in our analyses (Supplementary Table S2). Patient plasma samples exhibited a cytokine surge 24 hours after the first cycle of treatment, with a similar profile of induced cytokines. TNFα and IL12p40 were both significantly upregulated after treatment, whereas IFNγ and IL6 exhibited an upward trend that mirrored what was observed in mice (Fig. 3A).

To understand the relationship between these cytokines and treatment, we took advantage of the clinical trial design, in which CD40ag was dose-escalated (0.03, 0.1, or 0.3 mg/kg i.v.) with a fixed dose of CSF1Ri±anti–PD-1. We embedded patient samples from cycle 1 in two dimensions using PCA, analogously to the mouse data. PC1 revealed interpatient differences as the primary source of variability in patient serum secretion (Supplementary Fig. S2C). The variability explained by PC2 could be attributed to time point (Supplementary Fig. S2D) and/or escalating CD40ag dose (Fig. 3B), the latter suggesting that CD40ag drives the proinflammatory effects of TTx in humans similar to mice. 10 of the top 20 loadings along PC2 overlapped with the top 20 PC loadings associated with CD40ag treatment in mice (compared with Fig. 2C), including TNFα, IFNγ, IL12p40, and IL6 (Fig. 3C). Pairwise correlation of the cytokines and chemokines measured across patients revealed that some clustered together similarly to CD40ag-treated mice (Fig. 3D), such as IL12p40, TNFα, and CCL22. This evidence suggests that these cytokine clusters form co-occurring networks that represent downstream effector pathways of TTx that are dependent on CD40ag.

We took advantage of the timeline of serum secretion collection in the clinical trial to understand potential mechanisms and additional effects of repeated therapy administration. Comparison of secretion across patients before cycle 1 versus 2 indicated that serum levels returned to baseline between treatments (Supplementary Fig. S3A). However, most measured cytokines and chemokines exhibited smaller fold changes between pretreatment and 24 hours post-APX005M after cycle 2 compared with cycle 1 (Supplementary Fig. S3B), which could indicate an attenuated response to TTx with subsequent doses. Altogether, these results indicated that CD40ag treatment, in combination with anti–PD-1 and CSF1R blockade, increased systemic proinflammatory cytokine and chemokine levels similarly in mice and humans, suggesting that the mouse model recapitulates key elements of the treatment response observed in patients.

To assess changes in the TME after TTx at greater resolution, we collected untreated and TTx-treated tumors 24 hours after therapy (day 8 post-YUMMER1.7 injection) and prepared cells for scRNA-seq. Projection of the scRNA-seq data into two-dimensional space showed that this dataset captured multiple cell populations (Fig. 4A), with evident treatment-specific clusters. To classify single cells by cell type, we adapted an approach (45) for supervised cell-type annotation (see Materials and Methods). Our method robustly classified 10 cell types in the scRNA-seq data, including immune populations of T-cell subsets, macrophages and DCs, and non-immune populations of fibroblasts and tumor cells (Fig. 4B). Patterns in canonical marker expression confirmed specificity to the assigned cell-type clusters (Fig. 4C).

To understand which cell types responded to TTx, we inspected expression of therapy targets Pdcd1 (PD-1), Csf1r, and Cd40. The greatest Pdcd1 expression was observed in CD4⁺ T cells, whereas Csf1r and Cd40 were most highly expressed in macrophages and DCs, respectively (Fig. 4D). We then calculated the average expression of transcripts encoding for the CD40ag-associated cytokine/chemokine serum signature observed in the early response to treatment (i.e., serum response signature, Fig. 2A), linking the transcriptional data to the systemic secretomic profiles. Inspection of this serum-response signature across single cells identified the highest intensity among TTx-treated macrophages, DCs, and neutrophils (Fig. 4E), suggesting that myeloid cells were the primary responders to TTx in the tumor 24 hours after therapy. This result is consistent with receptor expression and with other studies that have reported that CD40ag targets TAMs and DCs in melanoma (21, 54).

We then inspected the TAMs, DCs, and neutrophils to further evaluate their potential contributions to the observed antitumor response after TTx. Macrophages were separated into two clusters based on treatment status, whereas neutrophils and DCs were each represented by one cluster primarily made up of treated cells (Fig. 4F). To further understand the early transcriptional differences between TTx-responding populations, we next identified DEGs for macrophages and DCs relative to the remaining myeloid cells (Supplementary Table S3; top 30). We looked for enrichment of these DEGs for GO biological processes and pathways in the Reactome, KEGG, and WP databases (Fig. 4G). This analysis implicated DCs as facilitators of T-cell recruitment and activation after TTx. DC DEGs were also enriched for IL12 production, which has been shown to induce activation and IFNγ production in CD4⁺ and CD8⁺ T cells (55). DAP12 signaling in DCs has also been demonstrated to induce functional antitumor responses in the B16 mouse melanoma model through T-cell stimulation (56). In contrast, genes upregulated by TAMs were enriched for hypoxia-inducible factor 1 signaling, IFNγ-induced gene expression, and production of proinflammatory cytokines TNF and IL6, the former of which functions in enhancing macrophage-mediated tumor killing and driving activation and proliferation of CD4⁺ and CD8⁺ T cells (57, 58).

Focusing on the DC response in the TTx-treated TME, we looked for evidence of conventional DC subset representation, as well as the recently identified “mature DC enriched in immunoregulatory molecules” (mregDC) state. To do this, we calculated a DC cluster score using marker genes for cDC1s, cDC2s, and mregDCs as previously defined (49). A clear enrichment for the mregDC program was observed, with little to no expression of genes associated with the cDC1 or cDC2 subsets (Supplementary Fig. S4). Overall, these results demonstrate distinct roles for DCs and macrophages in activation and inflammatory repolarization, respectively, 24 hours post-TTx.

Our scRNA-seq results revealed that DCs and macrophages upregulated different extracellular signaling programs in response to TTx, with TAMs exhibiting higher expression of genes enriched for TNF production, whereas DCs had greater expression of Il12b and Il6. Direct visualization of Ccl5, Il12b, Tnf, and Il6 expression across the control and TTx-treated TAM and DC populations confirmed TTx-specific upregulation of Il12b and Il6 expression in DCs (Fig. 4H, orange). In contrast, TAMs produced modest levels of Tnf and Il6 transcripts regardless of treatment (Fig. 4H, blue). In the case of Ccl5, DCs expressed relatively high levels regardless of treatment, whereas TAMs exhibited upregulation with treatment.

Characterization of the cytokine milieu at the level of protein translation provides a more accurate readout of functional change than RNA transcription. Therefore, to explore whether the observed changes in transcription translated into distinct secretion functionality, we used a microwell assay to measure the multiplexed secretion of a panel of 15 cytokines and chemokines from single cells (59). We profiled macrophages and DCs, as these were the major target populations identified for CSF1R blockade and CD40ag, respectively (Fig. 4D). We measured cytokines and chemokines upregulated by TTx and differentially expressed by DCs and macrophages, along with anti-inflammatory–associated products (e.g., Chi3l3 and MMP9) that we previously observed to be secreted by melanoma-associated myeloid cells (38). We analyzed YUMMER1.7 tumors either untreated or treated with TTx on day 7. Tumors were processed 24 hours after treatment and pooled before FACS to ensure adequate DC representation (Fig. 5A). CD45⁺CD64⁺F4/80⁺ cells were designated as TAMs. Two separate populations, CD45⁺CD103⁺ and CD45⁺MHCII⁺CD11b⁺CD11c⁺F4/80⁻CD64⁻Ly6G⁻, were combined into a single DC designation. Sorted populations were then cultured in microwells overnight without further stimulation. We found that, across conditions, most cells did not secrete any proteins in the panel above background levels (Supplementary Fig. S5), suggesting minimal nonspecific cell activation induced by the isolation procedure. Cells secreting at least one or more proteins (i.e., “responding cells”) were isolated for further analysis. We found DCs to be the primary IL12p40-secreting population, from which the responding fraction significantly increased from 18% to 29% after TTx (Fig. 5B, top), as well as the average secretion level per cell. Moreover, the fraction of CCL5-secreting DCs significantly increased by 8.6-fold upon treatment, as did the average secretion per cell. Interestingly, CCL22 secretion showed the same trend, with a significant 2.7-fold increase induced after treatment, further suggesting activation of the mregDC program in DCs (ref. 49; Supplementary Fig. S4). Together, CCL5 and IL12 form a DC-based cytokine network that is likely the main facilitator of T-cell immunity in response to TTx.

Treatment did not cause any statistically significant changes in TAM secretion for any of the measured cytokines and chemokines on the panel (Fig. 5B, bottom). However, TAMs secreted marked amounts of TNF, CCL3, and CCL5 regardless of treatment, in partial agreement with the transcript data. Interestingly, in addition to these proinflammatory factors, TAMs secreted high levels of Chi3l3 and MMP9, consistent with our previous observations that TAMs display mixed pro- and anti-inflammatory–associated functions (38).

To identify subpopulations of TAMs and DCs based on single-cell secretion, we clustered “responding” cells using an unsupervised, graph-based clustering method (51), and visualized cell types, treatments, and clusters in UMAP space (Fig. 5C). Four functional clusters emerged (Fig. 5D). The most abundant cluster, comprising nearly half of all cells, contained low-secreting DCs and macrophages (‘Quiescent’), which was expected given that the cells were not treated ex vivo. The other three clusters were predominantly TNF⁺ TAMs, Chi3l3⁺MMP9⁺ TAMs, and IL12p40⁺ DCs, in order of prevalence (Fig. 5D). The separation of the TNF⁺ and Chi3l3⁺MMP9⁺ TAM clusters in two-dimensional space was independent of treatment and suggested that functionally distinct TAM subsets were present before treatment. Importantly, TTx-treated DCs accounted for a majority of the IL12p40⁺ cluster (Fig. 5C). IL12p40⁺ DCs also cosecreted CCL5, CCL22, and CCL17 (Fig. 5D), consistent with mregDC program activation. Taken together, these results show that TTx induced secretion of T cell–recruiting and -activating factors IL12, CCL5, and CCL22 in a subset of DCs in the TME.

In addition, single-cell cytokine secretion profiling highlighted differences between cytokine transcription and translation. In particular, we observed a mismatch between TNF transcription and secretion (Fig. 4H), with minimal gene expression and high levels of secretion in both control and TTx-treated TAMs, at 24 hours after treatment. The opposite trend was noted for IL6 in TTx-treated DCs. This result highlights the importance of functional profiling at the level of protein translation.

Besides the rapid changes we characterized in myeloid cells targeted by TTx, we also sought to understand the impact of treatment on T cells in the TME. Notably, although TTx induced a successful antitumor response in approximately 85% (6/7) of mice, depletion of both CD4⁺ and CD8⁺ T cells completely abrogated the effect of therapy (Fig. 6A), demonstrating that T cells are necessary for effective antitumor immunity with TTx. Considering specific T-cell subsets, depleting CD4⁺ or CD8⁺ T cells separately reduced responses to TTx, with long-term OS in 29% (2/7) and 43% (3/7) of mice, respectively, 60 days after tumor initiation (Fig. 6A). These results show that both T-cell lineages are required for the TTx-induced antitumor response.

Because TTx induced increased expression of T cell–attracting chemokines (Figs. 2A and 4E), we hypothesized that treatment could induce T-cell migration into the TME. IHC staining showed more CTLs and fewer FOXP3⁺ Tregs in TTx-treated tumors compared with controls (Fig. 6B and C). This distribution of T cells may indicate a shift in the tumor toward a more activated cytotoxic state. scRNA-seq confirmed that T-cell gene expression in treated tumors was distinctly different from that in controls (Fig. 6D). Expression of cytotoxic effector genes Prf1, Gzma, and Gzmb was increased in CD4⁺ and CD8⁺ T cells in treated tumors (Fig. 6E), further supporting enhanced cytotoxicity in the tumor in the acute treatment response. In contrast, Tox was expressed similarly across CD4⁺ and CD8⁺ T cells in control and treated tumors (Supplementary Fig. S6A), suggesting that the acute response to TTx captured by the scRNA-seq may have been too early to see T-cell exhaustion. Furthermore, Mki67 expression, encoding for proliferative marker Ki67, was not observably different between control and treated tumors (Fig. 6E), indicating similar rates of T-cell proliferation. However, there was a small difference in representation of T-cell clonality in the treated samples (Supplementary Fig. S6B and S6C). Most notably, Ifng was expressed mainly by CD4⁺ non-Treg T cells in TTx-treated tumors (Fig. 6E). Protein-level measurement by intracellular cytokine staining (ICS) confirmed upregulation of IFNγ in CD4⁺FOXP3⁻ T cells 24 hours post-TTx compared with control (∼2.3-fold increase, Fig. 6F). This trend was also maintained 3 days after therapy. ICS further confirmed previously reported expression of IL12 and TNF in myeloid cells sorted from control and TTx-treated tumors (Supplementary Fig. S7). Blocking IFNγ also fully abrogated the effect of TTx (Fig. 6G), similar to depleting T cells, implying a central role for CD4⁺FOXP3⁻ T cells in the TTx-induced antitumor response. Taken together, these results suggest that TTx promotes recruitment of activated, tumor antigen–specific T cells into the TME that are required for the observed antitumor immune response.

To link patterns in DC, macrophage, and T-cell gene expression to extracellular signals in the TME and understand TTx-induced crosstalk between them, we inferred cell–cell communication from scRNA-seq using NicheNet. NicheNet integrates prior knowledge of signaling and gene regulatory networks to identify likely ligand–target gene interactions (48). We defined target genes as genes differentially expressed in DCs, macrophages, and T-cell subsets after TTx relative to control. Inspection of predicted regulatory potential values from NicheNet's prior model linked Cd40lg with a majority of DC and TAM target genes (Fig. 7A and B), supporting a mechanism in which changes in DC and TAM gene expressions result directly from CD40 agonism. Interestingly, NicheNet also linked Ifng to DC and TAM target genes, suggesting that feedback from CD4⁺FOXP3⁻ T cells might influence DCs and TAMs within 24 hours post-TTx. Tnf also had potential regulatory links to TAM target genes (Fig. 7B), suggesting a role for autocrine signaling. These results are consistent with a mechanism by which DCs and TAMs are directly stimulated by CD40ag, and indirectly receive feedback from CD4⁺FOXP3⁻ T cells via IFNγ and, for TAMs only, from autocrine signaling via TNF within 24 hours after treatment.

Given that TTx was ultimately T cell–dependent (Fig. 6A), we repeated NicheNet analysis for T-cell subsets to explore signaling indirectly downstream of therapy. We narrowed the search to myeloid-specific ligands. This analysis linked Il12b and Tnf to a majority of the CD4⁺FOXP3⁻ and CD8⁺ T-cell target genes (Fig. 7C and D). Ifng was a predicted target gene in CD4⁺ T cells, particularly in response to IL12, but not in CD8⁺ T cells. NicheNet analysis for Tregs did not link Il12b or Tnf to Treg target genes (Supplementary Fig. S8). Combined with our previous analyses of single-cell secretion (Fig. 5) and gene set enrichment (Fig. 4) of myeloid cells, these results strongly support IL12 and TNF signaling from DCs and macrophages, respectively, as drivers of gene expression in CD4⁺FOXP3⁻ and CD8⁺ T-cell subsets, and that the resulting CD4⁺FOXP3⁻ T cell–specific IFNγ signaling further activates DCs and TAMs.

Specifically, our data suggest that within 24 hours, TTx establishes a positive feedback loop in which CD40ag-stimulated IL12 secretion by DCs activates IFNγ secretion by CD4⁺FOXP3⁻ T cells, and IFNγ then signals back to DCs and TAMs, reinforcing their activation (Fig. 7E). The inclusion of CSF1Ri likely supported this loop by indirectly amplifying IL12 and IFNγ signaling (Fig. 2E). A similar IL12-dependent IFNγ-positive feedback loop has been implicated in other tumor immunotherapy responses (54, 60, 61). Thus, we reasoned that a strong IL12 signal alone might be sufficient to induce tumor regression. To test this, we injected immunocompetent C57BL/6 mice subcutaneously with YUMMER1.7 and allowed tumors to develop for 7 days. Mice with established tumors were treated by intratumoral injection with an LNP-encapsulated mouse (m)IL12 mRNA therapeutic (43), anti–PD-L1, or IL12 mRNA+anti–PD-L1. Treatment with IL12 mRNA resulted in complete tumor regression, with 100% of mIL12-treated mice experiencing durable responses for up to 40 days after treatment (Fig. 7F). Treatment with IL12 mRNA was effective even in the absence anti–PD-L1, although individual tumor growth curves showed that mIL12-treated tumors regressed slightly faster in combination with anti–PD-L1 (Supplementary Fig. S9). These data demonstrate that stimulation of IL12 signaling in the TME is sufficient to reverse tumor growth and promote an effective antitumor response, which supports our conclusion that CD40ag-induced IL12 secretion by DCs drives antitumor immunity in the TTx-treated TME.

Checkpoint therapy resistance is a major clinical challenge in cancer immunotherapy. A common factor in many resistance pathways is the contribution of the surrounding TME. Myeloid cells in particular have many roles in the TME that skew the immune microenvironment toward a protumorigenic or antitumorigenic state (11). Here, we explored the combination of myeloid-targeted CD40 agonist and CSF1R blockade, in combination with PD-1 inhibition, in intrinsically anti–PD-1–resistant models. We found that combinatorial therapy, driven largely by CD40ag, remodels the TME and stimulates T cell–based immunity by activating an IL12-secreting DC subset.

The immediate responding cells of TTx treatment were primarily DCs. CD40ag induced a subset of DCs to secrete IL12, likely stimulating IFNγ production in T cells. IL12 alone was sufficient to induce tumor regression in mice treated with mIL12 mRNA therapy, suggesting that increased DC-derived IL12 signaling in the TME upon TTx enhanced the antitumor response. IL12 and IFNγ further form a feed-forward loop that classically drives antitumor immunity by polarizing CD4⁺ T cells to a Th1 phenotype and activating tumor antigen–specific CD8⁺ T cells (62). CD40ag-treated DCs also produced CCL5, which aids in T-cell recruitment and, along with IL12, establishes the link between the innate and adaptive effects of TTx. IL12 is important for DC maturation, migration to lymph nodes, and antigen presentation (62). CCL5 specifically mediates Th1 migration into the tumor (53).

Our evidence suggests that TTx induces an mregDC phenotype within the key IL12-secreting DC subset. mregDCs are defined by expression of both maturation and immunoregulatory markers upon antigen uptake (49). Secretion of CCL17 and CCL22, which is characteristic of the mregDC program, was observed in DCs after TTx. Il12b expression was specific to treated DCs in the scRNA-seq data, which were enriched for the mregDC gene signature and lacked enrichment for conventional DC subset signatures. Other studies in the literature corroborate the role of IL12-producing mregDCs in the induction of response to CD40ag (20, 63). IL4 blockade has been shown to enhance IL12 production in mregDCs and could potentially augment efficacy of TTx (49).

Although intratumoral TAM cytokine secretion was not significantly affected by TTx, TAMs produced substantial amounts of TNF, CCL3, and CCL5, all of which are associated with a proinflammatory macrophage response. TAMs displayed upregulation of genes involved in macrophage activation, glycolysis, and IFNγ-induced patterns of expression, further indicative of a proinflammatory phenotype. TAMs also secreted marked amounts of Chi3l3 and MMP9, which are considered immunosuppressive/protumorigenic factors. The observation that tumor-associated myeloid cells produce both pro- and anti-inflammatory factors in the TME is consistent with previous observations of CD40ag+CSF1Ri-treated TAMs (38). However, our previous study lumped DCs and macrophages into one myeloid population. By extending the use of our single-cell secretion assay and separating DCs and macrophages, we were able to show that IL12p40 was produced almost exclusively by DCs, and that this population was distinct from the TNF-producing TAM subset. Overall, our work highlights the biological insights gained from functionally profiling immune cells isolated from the TME.

T cells are the final effectors of TTx and are necessary for treatment efficacy. Our results suggest distinct roles for CD4⁺ and CD8⁺ T cells with TTx. CD4⁺ T cells were the primary IFNγ-producing cell type, aiding coordination of the immune response. Depletion experiments indicated that CD4⁺ and CD8⁺ T cells were both required for treatment efficacy.

TTx leverages both innate and adaptive branches of the immune system to overcome PD-1 resistance. CD40ag is the main driver of treatment via DC stimulation and early cytokine secretion. Treatment combinations that include CD40ag display a survival advantage over those that do not, suggesting strong antitumor activity. CSF1R blockade was ineffective as a monotherapy, demonstrating a notable limitation of this therapy in its inability to remodel the TME in isolation. However, CSF1R blockade did augment the action of CD40ag, as treatment with CSF1Ri+CD40ag (±anti–PD-1) improved tumor clearance in our preclinical models compared with treatment with CD40ag alone. CSF1R blockade has been shown to preferentially deplete immunosuppressive subsets of TAMs (31, 35). It is therefore possible that this CSF1Ri-driven depletion releases a break on the CD40ag-driven proinflammatory response, resulting in the amplification of CD40ag-induced signals, including IL12, IFNγ, and CCL5, which we observed in serum secretion with combinatorial treatment. TTx was also significantly more effective than CD40ag+CSF1Ri, implicating a separate role for PD-1 inhibition in the triple combination. Anti–PD-1 has been shown to assist in successful priming of T cells via immediate blockade of negative signals in a pancreatic cancer model (29). It may be acting in a similar fashion on both newly primed T cells, as well as exhausted T cells, in the context of TTx. Taken together, although CD40ag is the primary driver of the proinflammatory response observed 24 hours post-TTx, blockade of CSF1R and PD-1 help by tuning the TME to respond to proinflammatory reprogramming more effectively.

A Phase I trial (NCT03502330) with CD40ag, CSF1R blockade, and anti–PD-1 was conducted in biopsy-proven patients with melanoma who progressed on anti–PD-(L)1 (44). Patients experienced a proinflammatory cytokine surge that peaked 24 hours post-APX005M, analogously to TTx-treated mice. Although clinical efficacy was limited, these results suggest that TTx acts analogously in humans and validate its translational potential. Minimal clinical responses may be due to suboptimal CSF1Ri dosing and excessive myeloid depletion. Moreover, patients received anti–PD-(L)1 before study recruitment. Acquired resistance to anti–PD-(L)1 may have made additional pathways of immune escape available that were not corrected by proinflammatory myeloid reprogramming. Crucially, TTx acts by repolarizing the TME and recruiting activated T cells into the tumor. It would be most effective in TAM-saturated tumors that exclude T cells from the TME. Patient inclusion criteria were based on clinical status and therapeutic history, not tumor composition. It is possible that patient tumors did not match this TME profile. In future studies, it will be important to select patients based on TME characteristics as well as clinical status.

Our results support DC activation and IL12 secretion as primary mechanisms of CD40ag, CSF1Ri, and anti–PD-1, with TAM repolarization as an important secondary effect, 24 hours after initial therapy administration. Comparison of trends in patient serum secretion between cycles 1 and 2 of TTx suggested that subsequent doses of therapy induced a similar mechanism, although the response may have been attenuated. We also focused on DCs and TAMs, as they were identified as the major targets of treatment in YUMMER1.7 tumors. However, there may be other myeloid subsets in the TME that either contribute to repolarization or use additional mechanisms to stimulate antitumor immunity upon treatment. CD40 is a pleiotropic cytokine and exerts its effects on many cell types within the TME (64). TNF is upregulated systemically, but we did not find evidence of significantly increased TNF secretion by either TAMs or DCs with treatment. It is possible that TNF is secreted by other cells with CD40 expression such as neutrophils or monocytes. Furthermore, although we found very few B cells in the TME, there may be rare B cells that are also targeted by CD40 agonism. In future studies, it will be important to elucidate the roles of these cell populations with TTx.

A.F. Alexander reports grants from National Science Foundation during the conduct of the study. M.K. McGeary reports grants from NCI during the conduct of the study. V. Muthusamy reports grants and nonfinancial support from Astra Zeneca during the conduct of the study; grants and nonfinancial support from Enlivex Therapeutics and Cybrexa Therapeutics, and grants from Stradefy Biosciences outside the submitted work. N. Luheshi reports personal fees from AstraZeneca during the conduct of the study; as well as reports a patent for methods of use of mRNAs encoding IL12 pending. S. Weiss reports other support from Apexigen and Bristol Myers Squibb, and grants from NCI K12CA215110 and P50 CA196530 during the conduct of the study; as well as personal fees from Lyell Immunopharma outside the submitted work. S.M. Kaech reports personal fees and other support from Affini-T Therapeutics and EvolveImmune Therapeutics, and personal fees from Arvinas and Pfizer outside the submitted work. H.M. Kluger reports grants from Apexigen and grants and personal fees from Bristol Myers Squibb during the conduct of the study; grants and personal fees from Merck, personal fees from Iovance, nonfinancial support from Celldex, as well as personal fees from Clinigen, Shionogi, Chemocentryx, Calithera, GigaGen, Signatera, GI Reviewers, and Pliant Therapeutics outside the submitted work. K. Miller-Jensen reports grants from NIH/National Cancer Institute during the conduct of the study. M. Bosenberg reports grants from AstraZeneca during the conduct of the study. No disclosures were reported by the other authors.

I. Krykbaeva: Conceptualization, formal analysis, investigation, visualization, methodology, writing–original draft. K. Bridges: Conceptualization, data curation, software, formal analysis, visualization, writing–original draft, writing–review and editing. W. Damsky: Conceptualization, investigation. G.A. Pizzurro: Investigation, writing–review and editing. A.F. Alexander: Investigation. M.K. McGeary: Data curation, software, formal analysis. K. Park: Investigation. V. Muthusamy: Investigation. J. Eyles: Investigation. N. Luheshi: Conceptualization, investigation. N. Turner: Investigation. S.A. Weiss: Investigation. K. Olino: Conceptualization, investigation. S.M. Kaech: Writing–original draft. H.M. Kluger: Conceptualization, resources, supervision, writing–original draft. K. Miller-Jensen: Conceptualization, resources, data curation, software, supervision, funding acquisition, visualization, project administration, writing–review and editing. M. Bosenberg: Conceptualization, resources, supervision, funding acquisition, visualization, project administration, writing–review and editing.

This work was supported by the NIH/NCI Ruth L. Kirschstein F30CA228444 (to I. Krykbaeva) and the NIH/NCI U01CA238728 award (to K. Miller-Jensen and M. Bosenberg). The authors thank all members of the Bosenberg and Miller-Jensen laboratories for insightful discussions and experimental advice. The anti–CSF1R-blocking antibody was provided by Bristol Myers Squibb. The LNP-encapsulated mouse (m)IL12 mRNA and anti–PD-L1–blocking antibodies were provided by Moderna.

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.