Reshaping the Tumor Microenvironment of KRAS^G12D Pancreatic Ductal Adenocarcinoma with Combined SOS1 and MEK Inhibition for Improved Immunotherapy Response Open Access

KRAS mutations are one of the most prevalent oncogenic alterations in cancer, occurring in roughly 30% of all cancer types (1). In pancreatic ductal adenocarcinoma (PDAC), nearly 90% of all patients harbor a KRAS mutation (2) and these mutations are implicated in the initiation, maintenance, and progression of pancreatic tumors (3). Thus, the KRAS signaling network represents a major target for therapeutic intervention in PDAC; there is no targeted therapy approved in this indication.

Recent efforts to target KRAS have been reinvigorated by the development of KRAS^G12C mutant-specific inhibitors, Sotorasib and Adagrasib (4–6). Although these inhibitors have shown encouraging clinical results in non–small cell lung cancer thus far, nearly half of the patients do not experience significant tumor shrinkage and nearly all patients that initially showed a response eventually progress due to the emergence of resistance and evasion mechanisms (7, 8). Moreover, KRAS^G12C mutations occur in only 1% of patients with PDAC, limiting the patient population that would benefit from these targeted therapies (9, 10). KRAS^G12D represents 35% of KRAS mutations in PDAC (10) and a selective KRAS^G12D inhibitor (MRTX1133) has started phase I/II trials for non–small cell lung cancer, colorectal, and pancreatic adenocarcinoma. MRTX1133 showed promising antitumor effects in preclinical PDAC mouse tumor models, but primary refractory and posttreatment relapses were observed (11). Other indirect approaches to block KRAS signaling involve targeting downstream MAPK signaling, such as BRAF, ERK, and MEK inhibitors. But similar to newer mutant-specific inhibitors, responses are often not durable, and resistance is common (12).

The tumor microenvironment (TME) of PDAC is highly complex and composed of malignant cancer, stromal, and immune cells preclinically best covered in syngeneic mouse models or genetically engineered mouse models. The cross-talk in the TME between cancer-associated fibroblasts (CAF), myeloid cells, T cells, and tumor cells is important in development of acquired resistance and suppression of antitumor immunity (13). Single-cell RNA sequencing (scRNA-seq) has revealed the heterogeneity of CAFs and identified multiple subsets of CAF populations, with the most abundant being the myofibroblastic and inflammatory CAFs (myCAFs and iCAFs; ref. 14). Both populations have been shown to play a role in immune cell suppression and tumor progression (15–19). In addition to CAFs, KRAS mutations have been reported to regulate the TME via driving immune-suppressive myeloid cell populations, and causing dendritic cell paucity, both impairing antitumor immunity (20–22). Therefore, it is highly likely that a combination approach with KRAS and immuno-oncology (I/O) to target both tumor proliferation and antitumor immunity will be needed to circumvent KRAS-targeted resistance.

To recapitulate the heterogeneity of the TME in patients with PDAC, we utilized KPCY (Kras^LSL-G12D/+; Trp53^LSL-R172H/+; Pdx1-Cre; Rosa26^YFP/YFP) tumor clones previously derived from an autochthonous mouse model of PDAC (22). The KPCY tumors are stromal rich, and clones were derived that result in both T cell “hot” and “cold” TMEs. The combination of SOS1 inhibitor (SOS1i)+MEK inhibitor (MEKi) has been previously shown to exert a strong antiproliferative effect on KRAS-driven tumors but has not yet been investigated in immune-competent mice harboring tumors with immune-suppressive TMEs (23). Using the KPCY tumor models, we show that SOS1i+MEKi are capable of effectively controlling tumor growth but tumors quickly relapse upon therapy removal. scRNA-seq revealed that SOS1i+MEKi increase M2 macrophage composition and results in decreased dendritic cell quality in the TME. Addition of CD40 agonist prolonged survival and increased efficacy in combination with SOSi+MEKi. Finally, addition of anti-PD-1/anti-CTLA-4 was needed for long-term durable responses.

Through rationale I/O combinations of anti-PD-1, anti-CTLA-4, and agonist-CD40, we were able to revert M2 macrophage polarization, increase dendritic cell maturation, and activate T cells. This combination results in complete tumor regressions with immune memory responses. Our findings provide strong rationale that the state of myeloid cells significantly influences I/O responses and highlights the need for combining KRAS cancer-targeted therapy with myeloid activating I/O to enhance and prolong antitumor effects.

Murine KPCY PDAC cell lines [2838c3 (TH2), 6694c2 (TC1), 6556c3 (TC2), 6419c5 (TC3), 6422c1 (TC4), 2699c4 (TC5)] were obtained from Dr. Stanger at University of Pennsylvania (Philadelphia, PA). Cells were cultured in DMEM (Corning #10-017-CV) supplemented with 10% heat-inactivated FBS (Genesee Scientific #25-514H), 1% GlutaMAX (Gibco #35050061), and 1% penicillin/streptomycin (Gibco #15140-122). Cell lines were routinely tested for Mycoplasma through Charles River Research Animal Diagnostic Services.

Female C57BL/6J wild-type and CD45.1 mice were purchased from Jackson Laboratories. Mice were maintained in pathogen-free facilities at the Boehringer Ingelheim animal facility. All animal studies were performed in accordance with the Boehringer Ingelheim Institutional Animal Care and Use Committee. For tumor growth and survival experiments, 3.0 × 10⁵ cells in 100 µL sterile DMEM were injected into the right flank of 6 to 8 weeks old female C57BL/6 mice. For concurrent and bilateral tumor inoculation, 3.0 × 10⁵ TC3 cells and 3.0 × 10⁵ TC5 cells were inoculated on the left and right flanks, respectively. For adoptive transfer, T cells were isolated from CD45.1 female mice. T cells were isolated using the MACS Pan T cell isolation kit II (catalog no.: 130-095-130) and stained with Cell Trace Violet (C34557). Ten million cells were injected intravenously one day prior to Vehicle or SOS1i + MEKi treatment.

Mice were randomized when tumors reached 40–100 mm³ and tumor volume, (length × width² × π)/6, was measured 3x weekly. Tumor volumes of 800 mm³ were used as an endpoint for survival analysis. Tumor regressions and waterfall plots were calculated benchmarking to the tumor size at randomization. Tumor growth inhibition (TGI) was calculated using the formula 100 − [(Tumor volume of treated − Initial tumor volume treated)/(mean final tumor volume control − mean initial tumor volume control) × 100].

Clinical grade SOS1i (BI 1701963) and MEKi (BI 3011441) compounds were formulated in Gelucire Solvent [1% Tween-80(v/v) + 15% Gelucire 44/14(v/v) + 84% DI H₂O(v/v)] and dosed at a final concentration of 50 and 0.3 mg/kg, respectively through oral gavage twice daily. For Fig. 2B–D, SOS1i (BI-3406; ref. 23) and trametinib (GSK1120212) were used by dissolving in 0.5% Natrosol + DMSO at 50 and 0.1 mg/kg, respectively. aPD-1(29F1A12) and aCTLA-4 (9D9) were dosed at 200 µg i.p. 2x weekly for three doses, and aCD40 (FGK45) was dosed at 200 µg i.p. for a single dose. IgG2b isotype control and IgG1 isotype controls were used. Antibodies were purchased from BioXCell.

Cells were seeded at 500 cells per well in a 384-well plate with reduced serum media (2% FCS) for proliferation and seeded at 180,000 cells per well in 12-well plates for Western blots. Treatments were applied the following day using the HP D300e Digital Dispenser. After 2 hours of treatment, protein lysates were prepared using MSD Tris Lysis Buffer #R60TX-3 complemented with Thermo Fisher Scientific Halt Protease & Phosphatase Inhibitor Cocktail (#1861284). Protein separation and immunodetection by Jess (ProteinSimple by Bio-Techne). Primary antibodies were diluted as follows: pERK1/2 (#4370, Cell Signaling Technology) 1:50; ERK1/2 (#9102, Cell Signaling Technology) 1:50; pMEK1/2 (#9121, Cell Signaling Technology) 1:10; MEK1/2 (#9122, Cell Signaling Technology) 1:50. Sample loading was normalized between all samples detecting the total protein amount in each sample using the ProteinSimple kit #DM-TP01. A total of 96 hours treatment was used for proliferation measurements using the CellTiter-Glo Luminescent Cell Viability Assay (Promega). Data were analyzed and plotted with GraphPad PRISM.

Tumors were harvested and processed on the gentleMACS Octo Dissociator. Digestion cocktail contained 800 µg/mL Dispase II, 100 µg/mL DNase I, and 400 µg/mL Collagenase P. Cell viability was assessed with Zombie Yellow Fixable viability dye. Surface marker staining was performed in the presence of anti-mu CD16/CD32 (BioLegend Tru Stain FcX). For intracellular staining, cells were fixed and permeabilized with eBiosicence Intracellular Fixation and Permeabilization buffer. Count Bright absolute counting beads (Invitrogen) were used for quantification. Flow cytometric analysis was performed on a BD LSR Fortessa Cell Analyzer. Antibodies used in flow analysis are described in the Key Resources Table (Supplementary Fig. S9).

Formalin-fixed (24 hours) paraffin-embedded tissue samples were immunostained using the BOND IHC polymer refine detection kit (DS9800). Sections were cut and transferred to the BOND RX (Leica). All subsequent steps [e.g., deparaffinization, antigen retrieval (ER2), staining, and development] were performed by the default 3,3-diaminobenzidine (DAB) IHC protocol. For CD8⁺ quantification, samples were stained with anti-CD8 antibody (CST98941). Slides were scanned and visualized using an Aperio AT2. Stained cells were counted in the total tissue size (mm²) and quantified with HALO 2.1 (Indica labs).

For single-cell sequencing, three groups of 7 mice were inoculated with TC3, TC4, and TC5, respectively. In each group, 3 mice were treated with vehicle and 4 mice were treated with SOS1i+MEKi when tumors reached 40–100 mm³. After 10 days of treatment, tumors were harvested, and single-cell suspensions were isolated from tumors as described in digestion above. Single-cell sequencing library preparation was performed using the 10x Genomics Chromium Controller and Chromium Next GEM Single Cell 3′ Kit v3.1 Chemistry Library, Gel Bead, and Chip Kits (10x Genomics) according to the manufacturer's protocols and instructions. A total of 10,000 cells were targeted per library and processed in parallel. Any samples with a proportion >10% of cDNA 150-400 bp, were subjected to an additional 0.6xSpri cleanup before proceeding with the remainder of library preparation. Libraries produced were checked for quality control using the Agilent Tapestation, quantified using the KAPA library Quantification Kit, diluted as appropriate, then sequenced using a NovaSeq6000 with paired-end sequencing and dual indexing as appropriate for the library preparation. Dual index libraries were sequenced using cycling conditions 28, 10, 10, and 90 for Read 1, i7 index, i5 index, and Read 2, respectively. Commercial kits are listed in the Key Resources Table (Supplementary Fig. S9).

Fastq files were used as an input to CellRanger (version 7.0.0) to generate count matrices, with mm10 as the mouse genome reference and default parameters. Downstream analysis was done using Scanpy (version 1.9.3, python version 3.8.2). Cells were filtered on the basis of the following criteria: 500 < number of unique molecular identifier (UMI) counts < 150,000, 200 < number of genes expressed < 10,000, percentage of mitochondria genes < 0.2. Then doublet removal was done using scrublet (version 0.2.3) with expected_doublet_rate = number of cells/1,000*0.008 for each sample. Cell count was normalized using scanpy.pp.normalize_total with a scaling factor of 10,000 and then log1p transformed using scanpy.pp.log1p. Highly variable genes were determined using scanpy.pp.highly_variable_genes with n_top_genes = 4,000. Principle component analysis was done using scanpy.tl.pca with n_comps = 50. Neighborhood graph was calculated using scanpy.pp.neighbors with n_neighbors = 15. Uniform Manifold Approximation and Projection (UMAP) topology was determined using scanpy.tl.umap with default parameters. Cell clustering was done using scanpy.tl.leiden with resolution = 1. Major cell types were manually identified on the basis of common marker genes: Epithelial cell (Krt7, Krt8, Krt18, Krt19, Epcam), Endothelial cell (Plvap, Cd34, Vwf), Fibroblast (Dcn, Col1a1, Col3a1), Pericyte (Rgs5, Acta2), Myeloid cell (Cd14, Itgam, Cd68), T and natural killer (NK) cell (Cd3d, Cd3e, Cd3g, Nkg7), B cell (Cd79a, Ly6d, Ms4a1), Neutrophil (S100a8, S100a9, G0s2), Mast cell (Tpsab1, Kit, Cpa3). Subclustering analysis for fibroblast, myeloid, and T/NK cells were done subsequently to identify fine-grained subpopulations for these three major cell types. Specifically, highly variable genes (n = 4,000) were reselected within each cell type. Principle components (n_comps = 15, 50, 50 for fibroblast, myeloid and T/NK cell, respectively) and neighborhood graphs (n_neighbors = 15) were recalculated. UMAP projections and Leiden clustering (resolution = 0.05, 0.4, 1.2 for fibroblast, myeloid, and T/NK cell, respectively) were done within each cell type. Then subpopulation cell types were manually identified using published marker genes (Fig. 4C). Of note, sample 167 (TC5 treated with vehicle) had significantly lower tumor epithelial cell proportion (<50% out of all cells), thus was removed for all downstream analysis, resulting in 153,678 total cells, 10,610 fibroblast, 22,818 myeloid cells and 7,956 T/NK cells.

Differential gene expression analysis was done using scanpy.tl.rank_genes_groups (method = wilcoxon and P value adjusted by Benjamini–Hochberg) and pathway overrepresentation analysis was done using enrichr API (gene_sets = MSigDB_Hallmark_2020) accessed through gseapy (version 1.0.5). For identifying pathways enriched in tumor epithelial cells from different cell clones, vehicle-treated epithelial cells were extracted. Cells from each clone were compared with the rest of the two clones and differentially expressed genes (DEG) were determined by adjusted P value < 0.05 and |log fold change|>1. For identifying shared DEG in response to SOSi+MEKi treatment within tumor cells, all epithelial cells were extracted and DEG for each clone were first determined by comparing TC3 SOSi+MEKi versus TC3 Vehicle, TC4 SOSi+MEKi versus TC4 Vehicle and TC5 SOSi+MEKi versus TC5 Vehicle with adjusted P value < 0.05 and |log fold change|>0.8. Shared upregulated and downregulated genes were identified and subject to pathway analysis.

To evaluate the potential for cell–cell interactions between cell types of interest, we used CellChat (version 1.6.1, R version 4.1.2) to infer cell-cell communications with a curated mouse ligand-receptor database. We ran two sets of cell-cell communication analyses: one for all major cell types to understand cross-talk between tumor and non-tumor cells; the other for subpopulations of CAF, myeloid, T/NK cells only to understand cross-talks among TME. For each analysis, CellChat was run in each cell clone and treatment condition, namely TC3 Vehicle, TC3 SOSi+MEKi, TC4 Vehicle, TC4 SOSi+MEKi, TC5 Vehicle, TC5 SOSi+MEKi separately with population.size = TRUE to normalize against cell proportion in each cell group during the probability calculation and other default parameters. Results were then merged into one cellchat object for comparisons for each analysis.

scRNA-seq data generated in this study are publicly available in Gene Expression Omnibus at GSE264527.

To understand the in vitro efficacy of a MEKi plus a SOS1i combination on KPCY tumor cells, we performed proliferation assays in two-dimensional monolayer conditions using two cell lines derived from subcutaneous growing tumors as published previously (22); TH2 has been shown to promote high T cell infiltration (hot) into the TME in vivo, and TC4 has been characterized by low T cell infiltration (cold) in vivo (22). The cell lines were treated with 0.8 to 1,000 nmol/L of MEKi (trametinib or tram), 0.004 to 10 µmol/L of SOS1i (BI 1701963), or the combination of the two inhibitors. Antiproliferative effects and cell viability was assessed after 96 hours. MEKi alone potently inhibited the growth of TH2 and TC4 cells with an IC₅₀ of 5.6 nmol/L and 7 nmol/L, respectively (Fig. 1A). In comparison with MEKi as a single agent, addition of increasing doses of SOS1i further decreased cell viability in both cell lines and reduced IC₅₀ values by 1.25- and 4-fold.

The impact of the drug treatments on MEK and ERK phosphorylation was also evaluated. TH2 and TC4 cells were treated with SOS1i (200, 1,000, or 1,400 nmol/L), MEKi (0.1 or 1 nmol/L) or the combination of both drugs for 2 hours. High pERK activity for both cell lines was observed at baseline (Fig. 1B). MEKi treatment alone resulted in a concentration-dependent pERK reduction in both cell lines. SOS1i treatment alone reduced pERK levels by approximately 50% in TH2 cells, and by 25%–40% in TC4 cells (Fig. 1B; Supplementary Fig. S1). Combined SOS1i and MEKi treatment resulted in further reduction of pERK, compared with MEKi or SOS1i treatments alone. This synergistic effect of the combination was particularly evident when SOS1i was combined with 0.1 nmol/L of MEKi, which resulted in near complete loss of pERK as compared with 40%–56% loss seen with 0.1 nmol/L MEKi alone. As expected, neither concentration of MEKi alone resulted in decreased MEK phosphorylation, likely due to the rebound of MEK activity caused by decreased ERK phosphorylation. However, in TH2 and, to a lesser extent, in TC4 cells, the addition of SOS1i resulted a concentration-dependent decrease in pMEK (Fig. 1B), suggesting that SOS1 inhibition prevents the upregulation pMEK by inhibiting the reactivation of an ERK-mediated negative feedback loop.

To evaluate the effects of SOS1i+MEKi therapy in vivo, we treated C57B/L6 mice bearing established subcutaneous KPCY tumors with either a combination of SOS1i+MEKi or vehicle for 5 or 10 days (Fig. 2A). KPCY TC1–5 (T cell cold) and KPCY TH2 (T cell hot) tumors were monitored for tumor growth. At the end of the study (∼12 hours after last dose), histology was assessed for treatment-related changes. All six KPCY clones exhibited dramatic reduction in tumor growth during treatment with SOS1i+MEKi combination with a statistically significant TGI ranging from approximately 73%–110% after 5 days of twice/day dosing (Fig. 2B). Similar TGI was seen in a sustained treatment model for 18 days where the control tumors reach an average of 870 mm³ while SOS1i+MEKi–treated tumors were only 340 mm³. However, upon discontinuing SOS1i+MEKi treatment, tumors quickly relapsed and matched control tumor volumes within 5 days suggesting SOS1i+MEKi treatment does not result in sustained control of tumor cell proliferation (Supplementary Fig. S2A).

Next, we assessed the CD8⁺ T cell density in the TME using IHC to understand how SOS1i+MEKi affects the immune compartment. In KPCY TC4 and TC5, increased CD8⁺ T cell infiltration was observed after 10 days of treatment (Fig. 2C and D). This was further validated by flow cytometry (Supplementary Fig. S2B and S2C). TC1, TC2, TC3, and TH2 did not show an increase in CD8⁺ T cells. We then assessed monotherapy versus combination of SOS1i+MEKi to understand what was driving the TGI and T cell infiltration in KPCY TC5. Comparing with vehicle control, SOS1i alone or MEKi alone led to only TGI of 38.62% and 11.35%, respectively. In contrast, the combination of SOS1i+MEKi resulted in statistically significant TGI of 74.51% (Fig. 2E). Similarly, only the SOS1i+MEKi combination, but neither of the two monotherapies, induced statistically significant increase in CD8⁺ T cells in the TME (Fig. 2F).

Finally, two additional in vivo studies were performed to determine the trafficking of CD8⁺ T cells in response to SOS1i+MEKi treatment. To determine whether the SOS1i+MEKi treatment-induced CD8⁺ T cell increase was systemic or local, concurrent bilateral tumor inoculations were utilized. KPCY TC5 cells, the clone showing increased intratumoral CD8⁺ T cells after therapy, were subcutaneously implanted on the right flank, and KPCY TC3, the clone with no T cell increase after therapy, was implanted on the left flank (Supplementary Fig. S2D). Mice bearing established tumors were treated either with vehicle or SOS1i+MEKi combination. Following treatment, the CD8⁺ T cell composition was assessed in both tumors by flow cytometry. SOS1i+MEKi treatment led to a statistically significant increase of CD8⁺ T cells in TC5 tumors as observed earlier. The increase of T cells in TC5 tumors was not impacted by the presence of the TC3 tumor on the opposing flank of the same mouse. Likewise, TC3 tumors were not impacted by TC5 tumor on the opposing flank and remained uninfiltrated by T cells after SOS1i+MEKi (Fig. 2G). This suggested that T cell infiltration was induced only locally following SOS1i+MEKi treatment and unique to the specific TME.

To differentiate T cell infiltration versus local proliferation in response to SOS1i+MEKi treatment, KPCY TC5 tumor-bearing mice were adoptively transferred with naïve T cells from CD45.1 mice labeled with cell trace violet (CTV) one day prior to starting therapy (Supplementary Fig. S3A). Following SOS1i+MEKi treatment, a greater accumulation of CD45.1⁺CTV⁺ undivided cells were detected in the tumors compared with vehicle-treated tumors although not statistically significant. Similar amounts of CD45.1⁺CTV⁻ divided cells were detected, suggesting that SOS1i+MEKi affected T cell infiltration, not T cell proliferation. Distribution of CD45.1⁺ dividing and nondividing cells was not affected by SOS1i+MEKi treatment in draining lymph nodes, blood, and spleens of all groups. This further suggested that SOS1i+MEKi effects are localized to the TME as shown previously (Supplementary Fig. S3B). Overall, the combined SOS1i+MEKi treatment results in TGI in KPCY tumor models and furthermore affects T cell recruitment in a subset of KPCY clones.

To investigate the cellular and molecular mechanisms underlying SOS1i+MEKi-induced intratumoral CD8 T cell infiltration, we performed scRNA-seq on KPCY TC3, TC4, and TC5, after treatment with vehicle or the combination of SOS1i+MEKi (Fig. 3A). After quality control, we obtained 153,678 high-quality cells for downstream analysis. UMAP and clustering identified nine major cell types (Fig. 3B). Epithelial/tumor cells identified by expression of Krt, Epcam comprised the largest population in the TME. In addition to epithelial/tumor cells, nonmalignant TME cell groups were identified comprised of immune cells including T cells (Cd3), B cells (Cd19), myeloid cells (Cd68), neutrophils (S100a8, S100a9), mast cells (Tpsab1, Cpa3), stromal cells including endothelial cells (Plvap), and CAFs (Dcn, Col1a1; Fig. 3B). Tumor cells from TC3, TC4, and TC5 were transcriptionally distinctive, while nonmalignant cell types from TC3, TC4, and TC5 origin are mixed indicating their global transcriptional similarity (Supplementary Fig. S4A and S4B). Differential expressed gene and pathway analysis of vehicle-treated tumor epithelial cells revealed the intrinsic heterogeneity of tumor cells: TC3 tumor cells were enriched in KRAS signaling, IL6/JAK/STAT signaling, and IFN responses; TC4 tumor cells were enriched in epithelial mesenchymal transition and hypoxia pathways; while TC5 tumor cells were enriched in cell proliferation pathways, TNFα signaling, IL2/STAT5 signaling, and Wnt-β catenin signaling pathways (Supplementary Fig. S4C; Supplementary Tables S1–S3). Consistent with TGI in vivo (Fig. 2), SOS1i+MEKi reduced the proportion of tumor cells in all three KPCY lines compared with vehicle treatment (Fig. 3C).

To identify tumor-intrinsic genes and pathways that are modulated in response to SOS1i+MEKi, we first compared SOS1i+MEKi tumor cells with vehicle-treated tumor cells for TC3, TC4, and TC5 separately (Supplementary Fig. S4C; Supplementary Tables S1–S3) and then extracted the shared 165 upregulated and 37 downregulated genes across all three clones. Pathway overrepresentation analysis was performed on the basis of the shared gene lists against hallmark gene sets. The results demonstrated strong decrease in cell cycle pathways, consistent with reported function of KRAS inhibition (23, 24), while IFNγ and IFNα responses, inflammatory response, and TNFα signaling were upregulated (Fig. 3D). Interestingly, Cd274, or PD-L1 was upregulated after treatment in all clones (Supplementary Fig. S4D).

TC5 is classified as warm due to the presence of higher numbers of CD8⁺ T cells at baseline in tumors. Upon SOS1i+MEKi treatment, a significant amount of CD8⁺ T cells infiltrated into the TME. To understand tumor cell–specific factors that could regulate this infiltration compared with the other clones, we focused on chemokines and cytokines that significantly increased in TC5 tumors and failed to increase in nonresponsive TC3 tumors. Three factors were identified: Cxcl9, Cxcl10, and Ccl3. Upon SOS1i+MEKi, these chemokines increased further in TC5 (Fig. 3E). Cxcl9 and Cxcl10 are broadly associated with CD8⁺ T cell infiltration and may account for the further increase in CD8⁺ T cells in TC5 (25–28).

To gain a better understanding of SOS1i+MEKi-induced TME modulation, we clustered and annotated subpopulations of CAF, myeloid, and T/NK cells, resulting in three CAF subpopulations: apCAF (Cd74), iCAF (Col14a1, Clec3b, Has1, Il6), and myCAF (Lrrc15, Acta2, Tagln); five myeloid subpopulations: macrophage (Cd14), cDC1 (Clec9a, Xcr1), cDC2 (Cd209a), mRegDC (Ccr7, Fscn1), and pDC (Siglech); six T- and NK-cell subpopulations: CD4T (Cd4), CD8T (Cd8a, Cd8b1), NK (Ncr1), dividing T (Mki67), γδ T (Tgd; Trdc, Blk), Treg (Foxp3; Fig. 4A; Supplementary Fig. S5A). We measured the frequencies of subpopulations per sample and the resulting bar plots revealed a differential distribution of diverse cell subsets in different KPCY TME, as well as their varying response to the SOS1i+MEKi treatment (Fig. 4B). SOS1i+MEKi increased the proportion of iCAF while decreasing the proportion of myCAF across all clones. In the myeloid cell compartment, macrophage proportions were increased across all clones while cDC1, mRegDCs, and pDCs had no change, and cDC2s were increased in KPCY TC3 and TC5 after SOS1i+MEKi.

To infer intercellular communication between cellular subsets, we performed subset composition correlation analysis that revealed the presence of two groups of cell subsets with highly correlated frequencies across the TME. Notably, iCAF, macrophage, and CD8⁺ T cells were found to be enriched in the TME following SOS1i+MEKi treatment. Interestingly, dendritic cell populations, cDC1, and cDC2 were also correlated (Fig. 4C). We then utilized CellChat (29) to further examine the expression of genes encoding ligands and receptors that can form pairs to potentiate cross-talk between cell subsets. We observed that CAF subsets harbored the highest number of ligand and receptor (L-R) pairs along with macrophages and dendritic cells (cDC1, cDC2, and pDC) also harboring L-R pairs with CAF subsets (Fig. 4D). This prompted us to further investigate how SOS1i+MEKi might influence the interactions between CAF subsets, macrophages, and dendritic cells.

To understand how SOS1i+MEKi influence CAF interactions with macrophages and dendritic cells, we calculated the L-R interaction strength that changed after treatment. Only interactions between iCAF and macrophages were notably changed after treatment (Fig. 4E; Supplementary Fig. S5B). To determine iCAF interactions with macrophages, we calculated genes and pathways that were differentially changed centered around iCAF as a ligand sender and macrophages as receptor receivers (Supplementary Fig. S5C). We found that CCL7–CCR1/2 interactions were increased, known to play a role in myeloid cell recruitment (30, 31) and CCR2 was highly upregulated in macrophages upon SOS1i+MEKi (Supplementary Figs. S5C, S5D, and S6A).

To understand how SOS1i+MEKi treatment and the subsequent interaction with iCAFs impacts macrophages, we performed differential gene expression analysis. Pathway overrepresentation analysis on shared upregulated or downregulated genes by SOS1i+MEKi across clones revealed an increase in NFκb and IL6/STAT signaling, pathways associated with skewing macrophages toward an M2 phenotype (Fig. 4F; Supplementary Tables S4–S6). IL6 was shown to be upregulated in iCAFs after SOS1i+MEKi (Supplementary Fig. S6B). DEGs for macrophage polarization (Mrc1, Tgm2, Chil3, Arg1, Ccl22, Cd163, Stat6) confirmed that after SOS1i+MEKi treatment, macrophages within the TME were skewed toward an M2 phenotype (Fig. 4G; Supplementary Fig. S6C). No change in interaction strength was found between iCAFs and dendritic cells (Fig. 4E). However, upon investigation of dendritic cell maturation markers, CD80 and CD40 were modulated depending on the KPCY clone. CD86 was also downregulated in all tumor models after treatment and further suggests a blunted dendritic cell maturation (Supplementary Fig. S6D). Therefore, SOS1i+MEKi treatment results in an immunosuppressive TME with increased iCAFs, polarized M2 macrophages, and impaired dendritic cell function.

Given that SOS1i+MEKi treatment can recruit intratumoral CD8⁺ T cells into cold KPCY clones but also result in immunosuppressive skewing of myeloid compartments, we hypothesized that SOS1i+MEKi treatment may sensitize KPCY tumors to T cell rejuvenating immunotherapy with myeloid activation. To test this hypothesis, we inoculated C57BL/6 mice with TC4, a “cold” KPCY clone that has been previously shown to be unresponsive to a combination treatment including standard-of-care chemotherapy plus immunotherapy [gemcitabine, nab-paclitaxel, anti-CD40 agonist (aCD40), anti-CTLA-4 (aCTLA-4), and anti-PD-1 (aPD-1)] (22). When tumors reached a size of 40–100 mm³, tumors were randomized and treated for 10 days with SOS1i+MEKi plus aPD-1, aCTLA-4, and aCD40 (Fig. 5A; ref. 32). TGI, tumor regression and complete response rate were calculated after completion of all treatments for immediate responses (Study Day 23) and also 38 days after withdrawal of SOS1i+MEKi treatment (Study Day 60) for durability of response. On day 23, SOS1i+MEKi had a TGI of 75% and only 2 of 10 mice showed tumor regression (Fig. 5B–D; Supplementary Fig. S7A). Checkpoint inhibitors, aCTLA-4 and aPD-1, have minimal effect in tumor growth control. The combination of aCTLA-4-4 and aPD-1 with SOS1+MEKi was able to significantly delay tumor growth and increase survival. However, the addition of myeloid activator aCD40 with SOS1i+MEKi treatment significantly increased TGI to 99% and increased early regression rates in 8 of 11 mice. While SOS1i+MEKi+aCD40 significantly increased overall survival (Fig. 5E), there were no complete responses on study. We then added checkpoint blockade (aPD-1, aCTLA-4) to SOS1i+MEKi+aCD40. Initially, these results looked similar to the SOS1i+MEKi+aCD40 at Study Day 23, but at the end of the study 4 of 10 mice were tumor free when treated with SOS1i+MEKi+aPD-1+aCTLA-4+aCD40 (Fig. 5B–E). Of note, weight loss was observed in aPD-1+aCTLA-4+aCD40, SOS1i+MEKi+aCD40, and SOS1i+MEKi+aPD-1+aCTLA-4+aCD40, respectively. This was observed 2 days after dosing of anti-CD40, suggesting toxicity; these mice rebounded in weight over the next several days (Supplementary Fig. S7B).

To interrogate the memory T cell response, we rechallenged the remaining complete responder mice with KPCY TC4, the same tumor cell clone. Age-matched, naïve mice had rapid tumor growth. In contrast, in the complete responder mice previously treated with combination of SOS1i+MEKi+aPD-1+aCTLA-4+aCD40, 2 of 3 mice (66%) rejected the newly implanted tumor clone indicating T cell memory (Fig. 5F; Supplementary Fig. S7C).

Our data indicate that SOS1i+MEKi results in TGI, M2 myeloid cell expansion, and increased immature DCs. Targeting these myeloid populations for activation through CD40 is an essential and unique step to licensing the tumor immune response during KRAS inhibition, enhances preclinical response to SOS1i+MEKi, and can induce checkpoint blockade–mediated complete response.

Here, we studied the combination of SOS1 and MEK inhibitors in a library of p53-deficient, Kras^G12D-driven KPCY clones with distinct transcriptional and T cell infiltration profiles. We demonstrate that combination of SOS1i+MEKi not only controls the tumor growth of murine Kras-p53–mutant PDAC tumors but also rewires intercellular cross-talk within the TME. The TME changes upon KRAS-targeted therapy rendering tumors vulnerable to an immunotherapy combination regimen that resulted in tumor clearance and immune memory.

It is well known that PDAC tumor cells are addicted to the KRAS oncogene, making the MAPK pathway an attractive target (24, 33). Prior research in the field has relied on MEK inhibition for treatment in KRAS-driven tumors and has assessed its role in overall antitumor immunity (34). MEKi as a single agent fails to induce a durable response (35, 36). However, multiple preclinical studies combined MEKi with other chemotherapeutic/immunotherapeutic agents, such as anti-PD-1 or anti-PD-L1, to modulate the TME, increase CD8 Teff cells, reduce MDSCs, and reduce tumor proliferation and overall tumor burden (36–39). Clinically, some BRAF^V600E-driven melanoma and advanced cutaneous melanoma demonstrate promising outcomes with immunotherapy (40, 41), yet others (phase II MSS mCRC study) show minor partial and no complete responses (42), underlying the need to explore additional combinations to enhance the efficacy of MEK inhibition.

Previous in vitro and in vivo studies in immune-compromised mice demonstrated that SOS1i reduced the formation of GTP-loaded RAS, limited cellular proliferation of human tumors driven by various KRAS mutations, and enhanced MEK inhibition in KRAS-dependent cancers (23). Consistent with these findings, we observed sensitivity to combined SOS1i+MEKi treatment in murine KPCY cells in vitro and TGI in vivo. To understand the full impact of SOS1i+MEKi in immune-competent mice, we utilized scRNA-seq and noted downregulation of G₂–M, E2F-regulated cell cycle, and proliferation pathways resulting in decreased tumor cell proportions in vivo. However, Hofmann and colleagues (23) and the current study following treatment with SOS1i+MEKi combination did not observe long-term tumor regressions in all tumor models. Therefore, we turned our attention to microenvironmental changes after SOS1i+MEKi treatment to deepen and lengthen the effects in combination with immunotherapy.

It is well appreciated that the type of TME plays a pivotal role in response to immunotherapy with the presence of CD8⁺ T cells a major indicator (43). While all KPCY clones showed in vivo TGI in response to SOS1i+MEKi, varying degrees of CD8⁺ T cell infiltration into these tumors were seen after treatment. This suggests differing tumor-intrinsic pathways are important for tumor growth control and local recruitment of CD8⁺ T cells, such as CXCL9/10 upregulation in TC5. Upon withdrawal of SOS1i+MEKi, tumors immediately relapsed regardless of T cell infiltration, indicating that even in clones with infiltrated T cells, T cells were either not fully functional in killing tumor cells or a potential immunosuppressive environment had led to a dysfunctional T cell compartment. Kemp and colleagues noted a similar phenomenon using a specific KRAS inhibitor upon which T cells infiltrate but do not determine compound efficacy (11). Further evidence of immunosuppression after SOS1i+MEKi inhibition was indicated by increased PD-L1 on tumor cells and downregulation of CD86 on dendritic cells. Stratification of patients by tumor cell expression of PD-L1 and TME type can inform and tailor combination modalities (43) and we therefore explored clones with significant CD8⁺ T cell infiltration in response to checkpoint blockade. Anti-PD-1 therapy is largely thought to be nonresponsive in patients with PDAC and other solid tumors (44). Despite a TME favorable for aPD-1 therapy with increased CD8⁺ T cells and upregulation of PD-L1 on tumor cells after SOS1i+MEKi therapy, aPD-1+aCTLA-4+SOS1i+MEKi was still ineffective in decreasing tumor growth and increasing survival.

A multitude of cell types within the microenvironment could explain the prolonged suppression T cell responses. Cellular composition changes after SOS1i+MEKi revealed an increase in fibroblasts, specifically iCAFs, and an increase in predicted cell–cell interactions between iCAFs and macrophages in all KPCY clones. Elevated levels of CCL7 and IL6 observed after SOS1i+MEKi by iCAFs is known to drive CCR2+ macrophages into tumors and induce an M2 protumorigenic macrophage phenotype (31) through IL6/JAK/STAT3 signaling (45, 46). The presence of CCR2+ M2 polarized macrophages has been shown to enable tumor recurrence (30) and clinically, inhibition of CCR2+ macrophages increases CD8⁺ T cells (47). Thus, the iCAF-induced increase in CCR2+ polarized suppressive macrophages after SOS1i+MEKi will need to be addressed for successful therapy. In addition to fibroblast and macrophage changes, SOS1i+MEKi treatment altered dendritic cells toward an immature phenotype, losing their ability to act as professional antigen-presenting cells and fully activate T cell responses (48). These CD86 reduced dendritic cells may also explain the ineffectiveness of aPD-1 therapy as it has been shown to be required for efficacy (49).Overall, this combination of fibroblast, myeloid, and dendritic cell rewiring after treatment creates an immunosuppressive environment not conducive to prolonged T cell activation and function.

To achieve immune activation and enhance and maintain efficacy of SOS1i+MEKi treatment, we utilized our scRNA-seq insights as a roadmap for immunotherapy combinations. We applied SOS1i+MEKi alongside checkpoint immunotherapy combinations of agonistic aCD40, aPD-1, and aCTLA-4 with the rationale to: (i) reverse macrophage polarization toward an M1 phenotype, (ii) restore dendritic cell maturation and antigen presentation by utilizing agonistic aCD40, a myeloid activator (32, 50, 51), and (iii) maintain and further activate T cells inhibited by PD-L1 upregulation on tumor cells. We applied this triple immunotherapy combination to a cold tumor, normally unresponsive to this regimen (22). Only when SOS1i+MEKi was combined with aCD40, aPD-1, and aCTLA-4, complete responses with robust immune memory were observed (Supplementary Fig. S8). Interestingly, aCD40 agonist alone alongside of SOS1i+MEKi was able to improve survival but was unable to cause complete regressions. Previous studies have seen similar effects through combination of MEK inhibition and aCD40. Yet, these studies required multiple doses of aCD40 for durable responses (52). Clinical trials with agonist CD40 antibody (Selicrelumab) in patients with resectable PDAC showed similar phenotypes to our study: a shift in macrophage polarization from M2 to M1, increase in T cell proliferation and PD-1 upregulation (53). This antibody and others (APX005M Sotigalimab) have been associated with modest tumor responses with several treatment-related adverse events such as elevated liver aspartate aminotransferase/alanine aminotransferase, lymphopenia, thromboembolic events, and cytokine release syndrome (53, 54). Given the known toxicities associated with aCD40 agonism, we achieved long-term responses here with a single dose followed by three doses of T cell–targeted therapies to produce complete tumor immune responses. This result gives light to the notion that activating only T cells or only myeloid cells is not enough to drive durable antitumor immunity and suggests that agonistic aCD40 plays an indispensable and non-redundant “licensing” role in enabling anti-PD-1/anti-CTLA-4 to exert their antitumor efficacy.

Importantly, unlike KRAS mutant-specific inhibitors, SOS1 inhibition and MEK inhibition in our study is not limited to targeting KRAS pathway only in tumor cells. It is highly likely that the microenvironmental changes observed in fibroblasts, myeloid cells, and dendritic cells may result from SOS1i+MEKi direct effects in addition to the changes to the tumor cells. However, recent studies with KRAS mutant-specific inhibitors note similar TME changes suggesting that the majority of effects observed with our pan KRAS inhibitors are likely driven by on-target tumor cell-intrinsic effects (55). Kemp and colleagues showed an upregulation of fibroblasts with KRAS^G12D inhibition while Mahadevan and colleagues also showed an upregulation of inflammatory iCAFs after KRAS^G12D inhibition in late stages of tumor growth (11, 56). Interestingly, these studies, including ours, contrast with the genetic iKRAS model, in which removal of KRAS activation led to a loss of αSMA⁺ fibroblasts (24, 57, 58). However, studies utilizing specific KRAS^G12D inhibitors were performed in late-stage tumors, whereas the iKRAS studies were performed at early intraepithelial neoplasia stages suggesting that tumor stage may impact the phenotype.

Our finding that KRAS-driven mechanisms maintain cellular growth and immunosuppression of both adaptive and innate immunity, highlights future therapeutic strategies that should consider a combination of KRAS cancer-targeting therapy with strategies targeting both T cells and myeloid cells. However, the toxicity associated with myeloid cell activators, such as aCD40, should be considered and carefully managed or modified for TME-specific delivery. Modalities incorporating dual myeloid and T cell targeting could be considered for further therapeutic development to best impart the full potential of KRAS-targeting agents.

Y. Xia is currently an employee of Boehringer Ingelheim. B. Flynn is currently an employee of Boehringer Ingelheim. M. Hinkel reports grants from Austrian Research Promotion Agency (FFG) during the conduct of the study; personal fees from Boehringer-Ingelheim outside the submitted work. S. Martinez-Morilla reports other from Boehringer Ingelheim Pharmaceuticals Inc during the conduct of the study; other from Boehringer Ingelheim Pharmaceuticals Inc outside the submitted work. K. Kostyrko reports grants from Austrian Research Promotion Agency (FFG) during the conduct of the study; personal fees from Boehringer-Ingelheim outside the submitted work. M.H. Hofmann reports grants from Austrian Research Promotion Agency (FFG) during the conduct of the study; personal fees from Boehringer-Ingelheim outside the submitted work; and he is listed as inventor on several patent applications for SOS1 inhibitors. A.S. Kashyap reports personal fees from Boehringer Ingelheim during the conduct of the study; personal fees from BOEHRINGER INGELHEIM outside the submitted work. No disclosures were reported by the other authors.

R.J. Norgard: Conceptualization, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. P. Budhani: Conceptualization, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. S.A. O'Brien: Conceptualization, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. Y. Xia: Conceptualization, data curation, formal analysis, investigation, writing-original draft, writing-review and editing. J.N. Egan: Data curation, investigation, visualization, methodology. B. Flynn: Data curation, investigation, visualization, methodology. J.R. Tagore: Data curation, visualization. J. Seco: Conceptualization, data curation, investigation, visualization. G.W. Peet: Data curation, investigation. A. Mikucka: Data curation, formal analysis, investigation, visualization. R. Wasti: Resources, data curation, software, methodology. L.-C. Chan: Data curation, investigation. M. Hinkel: Data curation, formal analysis, investigation, visualization. S. Martinez-Morilla: Data curation, formal analysis. J. Pignatelli: Data curation, formal analysis, supervision. F. Trapani: Data curation, formal analysis, supervision. E. Corse: Conceptualization, resources, supervision. D. Feng: Formal analysis, supervision. K. Kostyrko: Data curation, formal analysis, supervision, investigation, visualization, methodology, writing-original draft, writing-review and editing. M.H. Hofmann: Conceptualization, formal analysis, supervision, visualization, methodology, writing-original draft, writing-review and editing. K. Liu: Conceptualization, formal analysis, supervision, visualization, methodology, writing-original draft. A.S. Kashyap: Conceptualization, formal analysis, supervision, methodology, writing-original draft, project administration, writing-review and editing.

This study was supported and funded by Boehringer Ingelheim. We acknowledge and thank Theresa Goletz, Xavier Salce, Yarong Wang, Suzanne Segal, Mark Love, and the rest of the Cancer Immunology and Immune Modulation and Translational Medicine and Clinical Pharmacology department for their support and contributions to this project. We acknowledge Astrid Jeschko and Gabriela Melo-Zainzinger for their strong support in compound formulation and their transfer of knowledge to run the in vivo experiments. We thank Mohanapriya Kamalakannan, Jennifer Mele, and Kamey Philavanh and the entire Boehringer Ingelheim Cytometry Core. We thank the Analytical Biosciences Limited team (N. Rahmatpour, C. Kang, and S. Wu) for their help in scRNA-seq. We thank Ben Z. Stanger for KPCY cell lines and insights into maintaining them. Images were created using BioRender.com.