Targeting T cell checkpoints 41BB and LAG3 and myeloid cell CXCR1/2 results in anti-tumor immunity and durable response in pancreatic cancer

Abstract

Pancreatic ductal adenocarcinoma (PDAC) is considered ‘non-immunogenic’ with trials demonstrating its recalcitrance to PD1 and CTLA4 immune checkpoint therapies (ICTs). Here, we sought to systematically characterize the mechanisms underlying de novo ICT resistance and identify effective therapeutic options for PDAC. We report that agonist 41BB and antagonist LAG3 ICT alone and in combination, increased survival and anti-tumor immunity, characterized by modulating T cell subsets with anti-tumor activity, increased T cell clonality and diversification, decreased immunosuppressive myeloid cells and increased antigen presentation/decreased immunosuppressive capability of myeloid cells. Translational analyses confirmed the presence of 41BB and LAG3 in human PDAC. Since single and dual ICTs were not curative, T cell-activating ICTs were combined with a CXCR1/2 inhibitor, targeting immunosuppressive myeloid cells. Triple therapy resulted in durable complete responses. Given similar profiles in human PDAC and availability of these agents for clinical testing, our findings provide a testable hypothesis for this lethal disease.

Editor summary:

Gulhati et al. demonstrate therapeutic efficacy for combinatorial administration of 41BB agonist, LAG3 antagonist and a CXCR1/2 inhibitor in murine pancreatic cancer models, resulting in a remodeled tumor microenvironment.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is among the most lethal of human cancers, with a 5-year overall survival (OS) rate of 11% [1]. Given rising incidence and minimal change in mortality rates, PDAC is expected to become the second leading cause of cancer deaths by 2030 [2, 3]. The mainstay of treatment for metastatic PDAC is chemotherapy with gemcitabine- or fluorouracil-based regimens; however, chemotherapy benefit is often modest and transient [2]. While ICT has transformed the treatment and survival for numerous advanced cancers, PDAC remains recalcitrant to numerous ICT agents and combinations, including anti-PD1/PD-L1, anti-CTLA4 and combined anti-PD1+anti-CTLA4 [4, 5, 6, 7].

The lack of response to ICT has been attributed to immunosuppressive conditions in the tumor immune microenvironment (TIME), including prominent myeloid cell infiltration, as well as scarcity and dysfunction of CD8⁺ T cells among others [2, 4]. Following chronic antigen exposure in the TIME, CD8⁺ T cells differentiate into a dysfunctional state, characterized by loss of proliferative capacity and effector functions, as well as expression of inhibitory immune checkpoints, suggesting that these checkpoints may mediate CD8⁺ T cell exhaustion [2, 4]. The functional effects of targeting these immune checkpoints on dysfunctional/exhausted CD8⁺ T cells in PDAC is currently unknown. Several preclinical and early phase clinical trials have demonstrated signals of activity with immunotherapy combinations, encouraging further investigation [2, 4]. It is tempting to speculate that rational combinatorial treatments targeting non-redundant mechanisms of immune resistance may enhance the efficacy of ICT in PDAC.

Here, immune and single cell RNA sequencing (scRNA-seq) profiling of a murine PDAC model defined the TIME under various immune therapy perturbations. At baseline, the TIME was dominated by CXCR2-expressing myeloid derived suppressor cells (MDSCs) and tumor associated macrophages (TAMs), as well as CD4⁺ Tregs with high CTLA4 and OX40 expression and exhausted CD8⁺ T cells with high PD-1, LAG3, 41BB and TIM3 expression. We hypothesized that de novo resistance of murine and human PDAC to anti-PD1/PD-L1 and/or anti-CTLA4 may relate to alternative immune checkpoints and/or cooperative immune suppressive mechanisms across fibrotic stroma, MDSCs and/or TAMs. To that end, single and combined ICT and targeted therapy, coupled with immune profiling, was used to identify an immunotherapy combination regimen capable of re-invigorating anti-tumor immunity in the TIME leading to disease eradication in orthotopic tumors as well as prolonged survival with durable remissions in autochthonous tumors. We validated the presence of these targets in human PDAC.

Results

Myeloid cells predominate the iKRAS tumor microenvironment.

The inducible oncogenic KRAS mouse model (p48Cre; tetO_LSL-Kras^G12D; ROSA_rtTA; p53^L/+) designated “iKRAS” recapitulates the hallmark features of human PDAC including resistance to all standard therapies used to date [2, 8]. For multi-arm drug testing, several independently derived iKRAS cell lines were used to generate large cohorts with orthotopic PDAC tumors in syngeneic immunocompetent mice. Similar to autochthonous models, these orthotopic tumors grow rapidly to large volumes of ~1000mm³, demonstrate avid fluorodeoxyglucose (FDG) uptake, and can be detected using PET/CT, MRI and bioluminescence (Extended Data Fig 1a, 1b). All mice succumb to PDAC after 3 to 8 weeks (Extended Data Fig 1b). Mirroring human PDAC, the histological features of orthotopic and autochthonous murine iKRAS tumors include glandular tumor structures with moderately differentiated cells, significant desmoplasia with abundant collagen and high stromal expression of smooth muscle actin (SMA) and vimentin, as well as local invasion into surrounding lymph nodes and organs such as the duodenum (Extended Data Fig 1c–f).

To comprehensively audit the constellation of tumor infiltrating immune cells, we performed time-of-flight mass cytometry (CyTOF), which confirmed a significant increase in CD45⁺-infiltrating immune cells in established iKRAS tumors (4 weeks after initial detection on imaging, tumor volume ~1000mm³) (Fig 1a) consistent with human PDAC tumors (Extended Data Fig 1f) [9, 10]. Spanning-tree progression analysis of density-normalized events (SPADE)-derived tree [11] (Fig 1b) revealed the complexity of the PDAC TIME with cancer cells (EpCAM⁺CD45^-), non-immune TME cells (EpCAM^-CD45^-), and infiltrating immune subpopulations (EpCAM^-CD45⁺). Autochthonous versus orthotopic iKRAS PDAC tumors possessed similar composition and relative proportions of various immune cells (Fig 1c). Within the CD45⁺-infiltrating immune cells, MDSCs (CD45⁺CD11b⁺Gr1⁺) and TAMs (CD45⁺CD11b⁺Gr1⁻F4/80⁺), comprised a significant proportion of the immune population. The majority of MDSCs within the iKRAS PDAC TIME are neutrophilic/granulocytic in nature (Extended Data Fig 1g). In co-culture assays, these intratumoral CD11b⁺Gr1⁺ MDSCs suppressed anti-CD3 and anti-CD28 antibody-induced T cell proliferation and activation (IFNγ production) (Fig 1d, 1e), validating that CD11b⁺Gr1⁺ cells are indeed functional MDSCs [12]. Intratumoral MDSCs (characterized by S100A9 and Arginase-1 expression [12]) and TAMs (characterized by F4/80 expression) were found directly adjacent to cancer cells in iKRAS tumors (Extended Data Fig 1h, 1i), findings that mirror human tumor immune profiles. With respect to CD3⁺ T cells, CyTOF revealed tumor-infiltrating CD4⁺ and CD8⁺ T cells (Fig 1c). Using flow cytometry, the majority of intratumoral CD4⁺ and CD8⁺ T cells displayed an effector memory (CD44^hiCD62L^lo) phenotype (Fig 1f) and, among the CD4⁺ T cells, were a small proportion of FoxP3⁺ T_regs (Extended Data Fig 1j). IHC analyses demonstrated similar proportions and distributions of CD8⁺ T cells and S100A9⁺ MDSCs and their adjacency to cancer cells in autochthonous and orthotopic iKRAS tumors (Extended Data Fig 1h). These murine findings aligned with those in human PDAC specimens (Extended Data Fig 1k), which also included CD8⁺ T cells with memory and cytotoxic phenotypes (Fig 2a) [16].

Figure 1.

Prominent infiltration of myeloid immunosuppressive cells in iKRAS tumors. A. Quantification of tumor infiltrating CD45⁺ cells in syngeneic iKRAS tumors assessed by CyTOF at 4 weeks after initial tumor detection (n=10 samples/group). Two-sided Student’s t-test. B. SPADE tree derived from CyTOF analysis of whole-tumor cell population from syngeneic iKRAS PDAC tumors (n=10 tumors). Live single cells were used to construct the tree. Cell populations were identified as pancreatic ductal adenocarcinoma (PDAC) cells (EpCAM⁺CD45^-), non-immune TME cells (EpCAM^-CD45^-), CD4 or CD8 T cells (CD45⁺CD3⁺TCRβ⁺), B cells (CD45⁺B220⁺CD19⁺), Natural killer (NK) cells (CD45⁺NK1.1⁺), dendritic cells (CD45⁺CD11c⁺), MDSCs (CD45⁺CD11b⁺Gr1⁺) and macrophages (CD45⁺CD11b⁺Gr1^-F4/80⁺). C. CyTOF analysis of tumors from syngeneic and autochthonous iKRAS PDAC tumors with equivalent tumor volume (~1000mm³) (n=10 samples/group). D. Representative CFSE flow-cytometry histograms (left) showing the effect on in vitro T cell proliferation by MDSCs isolated from iKRAS tumors, and summarized result (right). Unstimulated T cells were used as negative control. Position of CFSE peaks can be used to denote the T cell division times. High and low proliferation were defined as T cell division ≥2 and ≤1, respectively (n=3 biological replicates). E. Effect on IFN-γ secretion from CD8⁺ T cells by MDSCs isolated from iKRAS tumors, measured by ELISA (n=3 biological replicates). Two-sided Student’s t-test. F. Quantification of tumor infiltrating CD4⁺ and CD8⁺ T cells in iKRAS tumors (n=3 biological replicates), assessed by flow cytometry and analyzed by FlowJo. Cell populations were identified as naive (CD44^lowCD62L^high), central memory (CD44^highCD62L^high), and effector memory (CD44^highCD62L^low). Data in A, E and F are presented as mean ± s.e.m.

Figure 2.

Prominent infiltration of myeloid immunosuppressive cells in human PDAC tumors. A. Representatives multiplex immunofluorescence images of human PDAC tissues on FFPE slides stained with the indicated proteins. Each experiment was replicated twice with similar results. B. Representatives multiplex immunofluorescence images of human PDAC tissues on FFPE slides stained with the indicated proteins. Each experiment was replicated twice with similar results. C. CIBERSORTx quantification of immune cell subsets in human PDAC samples; TCGA (n=178 patients) and ICGC-AU (n=92 patients). D. SPADE tree derived from CyTOF analysis of whole-tumor cell population from human PDAC samples (n=5 patients). Live single cells were used to construct the tree. See Supplementary Table 1 for clinicopathologic and demographic information about patients; Supplementary Table 2 for antibodies. Cell populations were identified as pancreatic ductal adenocarcinoma (PDAC) cells (EpCAM⁺CD45^-), non-immune TME cells (EpCAM^-CD45^-), CD4 (CD45⁺CD3⁺CD4⁺) or CD8 T cells (CD45⁺CD3⁺CD8⁺), B cells (CD45⁺CD19⁺), Natural killer (NK) cells (CD45⁺CD161⁺CD56⁺), Dendritic cells (CD45⁺CD33⁺HLA-DR⁺CD14^-CD15^-CD16^-CD11c⁺), MDSCs (CD45⁺CD33⁺HLA-DR^-CD11b⁺CD14^-CD15⁺ [neutrophilic/granulocytic] or CD45⁺CD33⁺HLA-DR^-CD11b⁺CD14⁺CD15^- [monocytic]) and macrophages (CD45⁺CD33⁺HLA-DR⁺CD14⁺CD15^-CD16^-CD11c⁺). E. CyTOF analysis of human PDAC tumors (n=5 patients). See Supplementary Table 1 for clinicopathologic and demographic information about patients; Supplementary Table 2 for antibodies. Red indicates high level of indicated marker expression; blue indicates no marker expression.

Validating the fidelity of our murine models, IHC analysis of treatment-naïve human PDAC tissues confirmed higher CD11b⁺ myeloid cells, including CD68⁺ macrophages and CD15⁺ neutrophils/granulocytes (Extended Data Fig 2a). Multiplex IF demonstrated CD33⁺CD11b⁺CD66b⁺ neutrophils/granulocytes and CD33⁺CD14⁺CD68⁺ TAMs in treatment naïve human PDAC tissues, consistent with prior studies of PDAC patients, and similar to iKRAS tumors (Fig 2b) [9, 10, 14, 15]. A 39-gene MDSC signature [16] and unsupervised clustering categorized 178 TCGA primary PDAC tumors into MDSC-high (n=114), MDSC-medium (n=54), and MDSC-low (n=10) subgroups, revealing that 94% tumors had either MDSC-high or -medium signatures (Extended Data Fig 2b). CIBERSORTx analysis of immune cell subsets in PDAC TCGA and ICGC-AU cohorts to enumerate fractions of immune cell subsets [17], revealed macrophages/monocytes as the predominant immune cell type (Fig 2c). While CIBERSORTx cannot deconvolute MDSCs from macrophages and other myeloid cells, this analysis showed that the predominant macrophages/monocytes population displayed an immunosuppressive M2-like macrophage signature (Extended Data Fig 2c). Correspondingly, IHC analysis of human PDAC tumors showed increased CD163⁺ M2 macrophages (Extended Data Fig 2d).

Humans and mice possess two TAM subtypes (Spp1⁺ and C1q⁺) with distinct origins and functions [18]. Using TAM subtype gene signatures [18], higher Spp1⁺ TAM frequency, but not C1q⁺, correlated with significantly lower OS in the PDAC TCGA cohort (Extended Data Fig 2e). Given that neoadjuvant chemotherapy/radiation can reshape the PDAC TIME [15], a validated CyTOF panel assessed myeloid cell representation in fresh PDAC specimens from patients who completed neoadjuvant chemotherapy and/or radiation (n=5) (Supplementary Table 1 and Supplementary Table 2) [19]. MDSCs and macrophages were confirmed to be the major immune cell sub-populations by SPADE analysis (Fig 2d); and, similar to iKRAS tumors, the majority of MDSCs within the human PDAC TIME are neutrophilic/granulocytic (Fig 2e). In summary, two major populations of immunosuppressive myeloid cells, neutrophils/granulocytes and macrophages, are present in treatment naïve and post chemotherapy/radiation treated human PDAC specimens. In addition, we detected intratumoral memory and cytotoxic CD8⁺ T cells, although these T cells become progressively exhausted/dysfunctional as described previously [10, 15, 20–22].

The dysfunctional phenotype of T cells in iKRAS tumors.

To delineate the immune composition and heterogeneity of iKRAS tumors, single-cell RNA sequencing (scRNA-seq) was performed on live CD45⁺ immune cells sorted from tumors harvested at four weeks after initial detection on imaging (tumor volume ~1000mm³). A total of 4,080 sorted individual immune cells from 3 iKRAS tumors were sequenced to an average depth of 50K reads per cell (Extended Data Fig 3a). Dimensional reduction analysis (UMAP) and clustering applied to the expression data revealed that live CD45⁺ immune cells clustered into several subgroups with similar fractions (Fig 3a–b) as those identified by CyTOF analysis of orthotopic and autochthonous iKRAS tumors (Fig 1c). Specifically, signature genes and known functional markers identified neutrophils/granulocytes (S100A8, S100A9, Gos2 expression), macrophages/monocytes (Mafb, C1qa, C1qc, Apoe, Lgmn, Spp1 expression), B cells (Cd79 expression), T cells (Cd3 expression), NK cells (Klr, Ncr1 expression) and dendritic cells (Fscn1, Ccl22 expression) (Extended Data Fig 3b, 3c). The myeloid compartment, comprising neutrophils/granulocytes and macrophage/monocytes, were the predominant immune cells in the iKRAS TIME as demonstrated by both CyTOF (Fig 1c) and scRNA-seq (Fig 3a, 3b; Extended Data Fig 3b, 3c), consistent with human PDAC [9, 14, 15] (Fig 2d).

Figure 3.

Heterogeneity of myeloid cells in iKRAS PDAC tumors identified by single cell gene expression profiling (n=3 tumors). A. UMAP projection of immune cell clusters and B. proportion of immune cell subtypes. C. UMAP projection of myeloid cell clusters and D. proportion of myeloid cell subtypes. E. Cell cycle scoring for five myeloid cell clusters. F. UMAP projection of dendritic cell clusters and G. proportion of dendritic cell subtypes.

Myeloid cells, including neutrophils/granulocytes and macrophages/monocytes, exhibit subtle differences in their cell states, existing along a continuum rather than discrete phenotypic states [12, 18, 23–25]. There is a high degree of plasticity within the myeloid population in PDAC as well as significant phenotypic heterogeneity between mouse and human myeloid cells [9, 10, 14, 15, 26–27]. To evaluate the intrinsic myeloid cell heterogeneity in iKRAS tumors, we applied clustering and identified five myeloid cell clusters (Myeloid_c1–5) with differential expression of signature genes and known functional markers (Fig 3c, 3d; Extended Data Fig 3d, 3e; Supplementary Table 3). Both myeloid_c2 and myeloid_c3 clusters demonstrate Cxcr2 and Ly6g expression consistent with prior studies describing CXCR2 expression on neutrophils/granulocytes or neutrophilic/granulocytic MDSCs [28, 29]. All myeloid cell clusters exhibited low replicative potential based on cell cycle scoring genes (Fig 3e). Classical dendritic cells (cDCs), which are critical for antigen-priming, T cell activation and ICT responsiveness [2, 4], were present in iKRAS tumors (Fig 3f, 3g; Extended Data Fig 3f; Supplementary Table 3).

To further characterize intratumoral T cell populations, we performed unsupervised clustering and identified six clusters including two clusters of CD4⁺ (CD4_c1, CD4_c2) and four clusters of CD8⁺ (CD8_c1, CD8_c2, CD8_c3, CD8_c4) T cells (Fig 4a, 4b), which clearly aligned with subsets in human PDAC [10, 11, 29] and other tumors [30]. The two clusters of CD4⁺ T cells included naïve/central memory CD4⁺ T cells (CD4_c1; Ccr7, Sell [encoding CD62L], Lef1 expression) and Tregs (CD4_c2; Foxp3 expression as well as Ctla4, Tnfrsf4 [OX-40] expression) (Extended Data Fig 4a, 4b). CD8⁺ T cell clusters included naïve/central memory CD8⁺ T cells (CD8_c1; Ccr7, Sell [encoding CD62L], Lef1 expression), two separate clusters (CD8_c2 and CD8_c3) with expression of cytotoxic genes (Nkg7 and Gzmb), although one of these clusters displayed higher expression of T cell exhaustion markers including Pdcd1, Lag3 and Havcr2 consistent with exhausted CD8⁺ T cells (CD8_c3), and a small cluster of highly replicating CD8⁺ T cells (CD8_c4; high Ki-67, Stmn1 expression as well as expression of the anti-apoptotic gene Birc5 [31, 32]) (Extended Data Fig 4a, 4b). We validated the high replicative potential of the latter cluster using cell cycle scoring genes (Extended Data Fig 4c). Notably, a subset of T cells in the PDAC TIME were naïve/central memory based on scRNA-seq (Fig 4a, 4b; Extended Data Fig 4a, 4b), consistent with our prior findings using flow cytometry (Fig 1f). Both clusters of naïve/central memory T cells (CD4_c1 and CD8_c1) expressed TCF7 (encoding the transcription factor TCF1), which has been associated with a progenitor or stem-like state, ICT response and improved outcomes [33, 34] (Extended Data Fig 4b).

Figure 4.

Dysfunctional phenotype of T cells in iKRAS PDAC tumors identified by single cell gene expression profiling (n=3 tumors). A. UMAP projection of T cell clusters and B. proportion of T cell subtypes. C. Ordering of CD8⁺ T cells along pseudotime in a two-dimensional state-space defined by Monocle2. Each point corresponds to a single cell, and each color represents a CD8⁺ T cell cluster. D. Heatmap of immune checkpoint expression on various clusters of CD4⁺ and CD8⁺ T cells. E. Volcano plot showing differentially expressed genes between naïve/central memory CD8⁺ T cells and exhausted CD8⁺ T cells (left), CD8⁺ T cells and exhausted CD8⁺ T cells (middle), and naïve/central memory CD4⁺ T cells and Tregs (right).

We further interrogated the developmental trajectory of CD8⁺ T cells within the PDAC TME using Monocle2 [35]. Clusters of CD8⁺ T cells formed a linear structure, which when rooted with naïve/central memory CD8⁺ T cells, was followed by non-exhausted cytotoxic CD8⁺ T cells and ended with exhausted CD8⁺ T cells (Fig 4c). Thus, exhausted T cells were highly enriched at the late period of pseudotime, a pattern consistent with the CD8⁺ T cell state transition from naïve/central memory to activated/non-exhausted to exhausted. Exhausted T cells expressed high levels of Granzyme B (GzmB) with low levels of Granzyme K (GzmK) (Extended Data Fig 4d). Expression of activating (Tnfrsf9 and Tnfrsf4) and inhibitory (Pdcd1, Lag3, Ctla4 and Havcr2) immune checkpoints was noted on the exhausted CD8⁺ T cell cluster but not in the naïve/central memory or intermediate non-exhausted/activated CD8⁺ T cell states (Fig 4c–e; Extended Data Fig 4d), raising the possibility that these molecules may mediate the exhausted state of CD8⁺ T cells in the PDAC TIME. We found preferential enrichment of CD4⁺ Tregs with high Ctla4 and Tnfrsf4 expression as well as exhausted CD8⁺ T cells with high Pdcd1, Lag3, Tnfrsf9 and Havcr2 expression amongst the differentiated T cell population in PDAC (Fig 4d, 4e). Flow cytometry validated expression of these immune checkpoint molecules on CD4⁺ and CD8⁺ T cells (Extended Data Fig 4e). Since exhausted T cells and Tregs are targets for cancer immunotherapies [2, 4], we focused our analyses on the role of hitherto uncharacterized immune checkpoints, where the consequences of targeting each checkpoint in PDAC is not known.

Efficacy of agonist 41BB and antagonist LAG3 ICT in iKRAS tumors.

The presence of intra-tumoral CD8⁺ T cells in iKRAS models (Fig 1c, Extended Data Fig 1h), together with improved OS associated with increased CD8⁺ T cell infiltration and their proximity to cancer cells in human PDAC [13, 22, 36, 37], and the lack of efficacy of PD-1/CTLA-4 in clinical trials [4–7], prompted us to consider the presence of a poised immune microenvironment that can be activated by targeting alternate immune checkpoints. To test this hypothesis, immunocompetent C57BL/6 mice with orthotopic iKRAS tumors were treated with agonist and antagonist ICT antibodies targeting the aforementioned checkpoints expressed on differentiated T cells in iKRAS tumors (Extended Data Fig 5a). Mice with MRI-documented PDAC of equivalent size were treated with a single high-dose of gemcitabine (100mg/kg) [38] and subsequently randomized to receive single or combination ICT treatments for 4 weeks before endpoint analysis (Extended Data Fig 5a). Gemcitabine, a standard chemotherapy treatment, was administered to provoke tumor cell death and release neoantigens as well as decrease MDSC/Treg accumulation and activity in murine PDAC tumors as reported previously [3, 39, 40].

Consistent with human clinical trials, antagonist PD1 and CTLA4 antibodies had no effect on tumor growth or OS (Fig 5a, 5b) [5, 6, 7]. Although a prevailing view holds that this poor response relates to poor infiltration of effector T cells, the above immune profiles clearly demonstrate tumor infiltrating CD4⁺ and CD8⁺ T cells with high Ctla4 and Pdcd1 expression. In exploring alternate immune checkpoints, we detected increased Lag3, Tnfrsf9 (41BB) and Havcr2 (TIM3) expression in response to anti-PD1 or anti-CTLA4 monotherapy relative to control antibody treatment (Extended Data Fig 5b). Conversely, we noted decreased Tnfrsf4 (OX-40) expression in response to anti-PD1 and anti-CTLA4 monotherapy. Given these immune checkpoint profiles, tumor-bearing mice were treated with agonist (41BB and OX40) or antagonist (LAG3 and TIM3) antibodies as monotherapy. Strikingly, impaired PDAC progression and increased OS were observed with agonist 41BB and antagonist LAG3 antibodies (Fig 5a, 5b). Combined agonist 41BB and antagonist LAG3 antibody treatment, which was well tolerated during the 4-week treatment period, produced significantly increased survival relative to either monotherapy (Fig 5c, 5d). However, all dual ICT-treated mice eventually succumbed. These unexpected murine findings prompted analysis of 41BB and LAG3 expression in treatment naïve human PDAC specimens (n=54 patients) (Fig 5e). Using multiplex IF, 81% of PDAC patient specimens demonstrated 41BB⁺ T cells, while 93% demonstrated LAG3⁺ T cells. Notably, both 41BB and LAG3 are elevated on tumor infiltrating T cells compared to circulating T cells in human PDAC patients [22].

Figure 5.

Efficacy of immune checkpoint therapy (ICT) and treatment effects on immune microenvironment. A. Tumor volume after 4 weeks of treatment with control or anti-PD1 or anti-CTLA4 or anti-41BB or anti-TIM3 or anti-OX40 or anti-LAG3 antibody in mice bearing established (tumor volume ~250mm³ prior to treatment initiation) orthotopic iKRAS tumors (n=13–14 mice/group). Two-sided Student’s t-test. B. Overall survival of mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control or anti-PD1 or anti-CTLA4 or anti-41BB or anti-TIM3 or anti-OX40 or anti-LAG3 antibody (n=13–14 mice/group). C. Tumor volume after 4 weeks of treatment with control or anti-41BB or anti-LAG3 or anti-41BB+anti-LAG3 antibodies in mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) (n=13 mice/group). Two-sided Student’s t-test. D. Overall survival of mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control or anti-41BB or anti-LAG3 or anti-41BB+anti-LAG3 antibodies (n=13 mice/group). E. Representative spectral composite image of immunofluorescence staining in human PDAC samples with the indicated proteins (left). Each experiment was replicated twice with similar results. Quantification of proportion of human PDAC samples with positive and negative staining for indicated proteins (right; n=54 patients). F. Quantification of change in proportion of immune cell subtypes in single-cell sequencing analysis of established iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) after treatment with anti-PD1, anti-CTLA4, anti-41BB or anti-LAG3 antibody for 4 weeks (n=3 tumors/group). Mixed effect model. Statistical differences in B and D were identified by Kaplan-Meier with log-rank test. Data in A, C and F are presented as mean ± s.e.m.

Furthermore, analysis of 41BB and LAG3 expression in scRNA-seq of 40 treatment naïve human PDAC patients from two datasets validated expression of 41BB and LAG3 on T cells in human PDAC (Extended Data Fig 5c) [10, 27]. Since co-expression of multiple co-inhibitory receptors results in dysfunctional/exhausted phenotype of T cells [41], we examined co-inhibitory receptor expression of 41BB and LAG3 on tumor infiltrating CD8⁺ T cells in iKRAS tumors. We found that <0.5% of CD8⁺ T cells examined co-expressed 41BB and LAG3 suggesting that 41BB and LAG3 axes may represent non-redundant mechanisms of T cell exhaustion, which further supports strategies that co-target these checkpoints. Thus, activation of T cell activity and anti-tumor activity with agonist 41BB and/or antagonist LAG3 antibodies in iKRAS PDAC tumors, coupled with comparable target expression in human PDAC, portends translational relevance.

To evaluate dynamic changes in the TIME with the various ICT agents, scRNA-seq was used to evaluate various immune cell subpopulations and their transcriptional changes in iKRAS tumors following a 4-week treatment period with effective (agonist 41BB and antagonist LAG3) and ineffective (antagonist PD1 and CTLA4) agents compared to control (n=3 per treatment group) (Extended Data Fig 5a). Single cell analysis of sorted CD45⁺ immune cells yielded data on a total of 72,440 cells with an average depth of 50K reads per cell (Extended Data Fig 5d). To define the subpopulation structure, we computationally pooled data from the various treatment groups and subsequently identified transcriptional clusters consisting of individual cell types (Extended Data Fig 5e–f, 6a). Dimensional reduction analysis (UMAP) revealed that immune cells clustered into similar subtypes of immune cells as untreated tumors (Fig 3a, 3b; Extended Data Fig 3b–c; Extended Data Fig 5e–f, 6a). Compared to untreated tumors (Fig 3a–d, 4a–b; Extended Data Fig 3b–3e; Extended Data Fig 4a–b), there were several T cell (Extended Data Fig 6b, 6c), neutrophil/granulocyte (Extended Data Fig 6d, 7a) and macrophage/monocyte (Extended Data Fig 7b, 7c) clusters identified, which were unique to treated tumors. A total of five CD8⁺ and four CD4⁺ T cell clusters were identified including naïve/central memory CD4⁺ (T_c5) and CD8⁺ (T_c4), exhausted CD8⁺ (T_c1), effector CD4⁺ (T_c2, T_c6) and CD8⁺ (T_c3, T_c8), T_regs (T_c7), T_h17 (T_c9) and replicating CD8⁺ (T_c10) (Extended Data Fig 6b, 6c). Myeloid cells were classified as either neutrophils/granulocytes (characterized by expression of S100A8, S100A9, Gos2, which are also highly expressed in neutrophils/granulocytes in human PDAC [9, 27]) or macrophages/monocytes (characterized by expression of Apoe [associated with non-inflammatory, immunosuppressive macrophages)], Spp1, Lyz2 [expressed by classical monocytes (10)], C1qa/c [associated with tissue resident macrophages (10)], Arg1 [suggestive of immunosuppressive potential (14)]), consistent with clustering in untreated iKRAS tumors (Extended Data Fig 5e–f, 6a; Fig 3c, 3d; Extended Data Fig 3d–e). Further interrogation of myeloid cell heterogeneity revealed a continuum of states which resolved into five neutrophil/granulocyte clusters (N_c1–5; Extended Data Fig 6d, 7a; Supplementary Table 3) and five macrophage/monocyte clusters (M_c1–5; Extended Data Fig 7b, 7c; Supplementary Table 3), consistent with recently described subsets in other human and murine tumors [18, 24]. N_c3 was characterized by Cxcr2 expression, which is implicated in MDSC/neutrophil migration into PDAC [28, 29].

Treatment with ineffective antagonist PD1 and CTLA4 did not significantly impact the T cell infiltrates, whereas agonist 41BB antibody treatment resulted in T cell expansion, predominated by non-exhausted cytotoxic CD8⁺ T cells expressing high Ccl5, high GzmK and low GzmB (Fig 5f; Extended Data Fig 7d, 8a). 41BB treated T cells were enriched in cluster T_c3, compared to control treated T cells which were enriched in cluster T_c1 (Fig 6a). 41BB treatment elicited a shift in expression from GzmB high to GzmK high T cells (Fig 6b). GzmK expressing pre-dysfunctional effector memory T cells have a less exhausted phenotype compared to GzmB expressing dysfunctional exhausted T cells [30, 34, 42]. Comparing gene signatures of CD8⁺ T cell inhibition [43], terminal differentiation [25], progenitor exhaustion [33] and terminal exhaustion [34] in T cells from clusters T_c1 and T_c3 (Fig 6c), we found that expression of gene signatures of inhibition, terminal differentiation and terminal exhaustion were significantly enriched in GzmB expressing cells (T_c1) compared to GzmK expressing cells (T_c3). Meanwhile, GzmK expressing cells (T_c3) were significantly enriched in expression of progenitor exhausted gene signature (associated with ICT response [33]) compared to GzmB expressing cells (T_c1). Pathway analysis revealed upregulation of intracellular signaling, cytokine production, proliferation and cytolytic activity in T cells from 41BB treated iKRAS tumors compared to control (Extended Data Fig 8b), consistent with prior studies [44–46].

Figure 6.

Effects of immune checkpoint therapy (ICT) treatment on immune microenvironment. A. Fraction of top clonotypes in each T cell cluster among control and anti-41BB antibody-treated mice (n=3 mice/group). B. Relative expression levels of Granzyme K (Gzmk) and Granzyme B (Gzmb) among T cells in control and anti-41BB antibody-treated mice (n=3 mice/group). (*p<0.05 two-sided unpaired Wilcox test) C. Relative expression of gene signatures of T cell inhibition [43], terminal differentiation [25], progenitor exhaustion [33] and terminal exhaustion [28] in T cells from clusters T_c1 and T_c3. (n=1762 cells [T_c1], 1809 cells [T_c3]) (*p<0.05 two-sided unpaired Wilcox test) D. Fraction of overlapping T cell receptor CDR3 sequences between mice after 4 weeks of treatment with control or anti-PD1 or anti-CTLA4 or anti-41BB or anti-LAG3 antibody in mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) (n=3 mice/group). E. Fraction of CD3⁺CD4^-CD8^-NK1.1^- T cells among CD3⁺ T cells after 4 weeks of treatment with control or anti-PD1 or anti-CTLA4 or anti-41BB or anti-LAG3 antibody in mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) (n=3 mice/group). Mixed effect model. F. Relative expression of IL17a among subtypes of CD3⁺ T cells. G. Violin plots showing expression of β2m, H2-Aa, Cd74 and H2-Ab1 among myeloid cells from control, anti-PD1, anti-CTLA4, anti-41BB and anti-LAG3 antibody-treated tumors (n=3 tumors/group) (*p<0.05 two-sided unpaired Wilcox test). Data in E are presented as mean ± s.e.m.

T cell clonality is lower in PDAC compared to CRC and melanoma consistent with the relative paucity of coding mutations [47–49]. While T cells in human PDAC display minimal clonal expansion, higher clonality scores trend toward increased T cell receptor (TCR) signaling and effector phenotype [22], and higher number of expanded T cell clones correlates with improved OS in PDAC patients receiving ICT [50]. Therefore, in iKRAS tumors, the TCR repertoire was evaluated using scTCR-seq matched with transcriptome data to assess changes in phenotype, clonality, and TCR sequences after ICT treatment. 41BB treatment resulted in CD8⁺ cell clonotype expansion, comprised mainly of non-exhausted cytotoxic T cells (cluster T_c3), while cells from the control group were enriched in clusters T_c4 and T_c5 (Extended Data Fig 8a, 8c). Cells from the most expanded TCRs were almost exclusively in cluster T_c3. None of the other ICT agents tested (PD1, CTLA4, LAG3) impacted the clonality of T cells. Next, we evaluated TCR diversity which decreases with progressive exhaustion of T cells [43] and tracks with poor outcomes [51]. TCR diversification is associated with improved therapeutic benefit from ICT treatment [50, 52]. Upon evaluation of the overlap in TCR CDR3 sequences from T cells, anti-PD1 and anti-CTLA4 treated mice harbored significant overlap amongst TCRs between mice within their treatment group, similar to the control treatment group (Fig 6d; Extended Data Fig 8d). In striking contrast, agonist 41BB and antagonist LAG3 antibody treatments elicited complete loss of TCR overlap, consistent with TCR diversification.

While LAG3 treatment did not increase T cell infiltration into iKRAS tumors (Fig 5f; Extended Data Fig 7d), treatment doubled the fraction of CD3⁺CD4^-CD8^-NK1.1^- T cells, which are known targets of ICT and have anti-tumor effects in PDAC and other tumor types [53–55] (Fig 6e). These CD3⁺CD4^-CD8^- cells are characterized by high IL-17 expression and accounted for ~75% of the total IL17⁺ immune cells in iKRAS tumors (Fig 6f) [53]. They display higher CCR7 expression compared to CD4⁺ and CD8⁺ T cells consistent with prior studies [53] (Extended Data Fig 8e). Higher proportion of CD3⁺CD4^-CD8^- T cells display CCR7 expression compared to CD4⁺ and CD8⁺ T cells, while lower proportion of CD3⁺CD4^-CD8^- T cells display IL2RB expression compared to CD4⁺ and CD8⁺ T cells as described previously [53] (Extended Data Fig 8f). These CD3⁺CD4^-CD8^- cells express CCR2, which facilitates their recruitment to the PDAC TIME (Extended Data Fig 8g) [53]. They express adhesion ligand JAML (Amica1), cytotoxic marker CD107a (Lamp1), STAT1, TGFβ, TNF and have minimal IL10 expression consistent with their immunogenic functions (Extended Data Fig 8g). They also express Ccl5, which can reprogram myeloid cells towards anti-tumor immunity (Extended Data Fig 8g) [53].

ICT can remodel both T cell and myeloid compartments in the TIME [23]. Correspondingly, effective ICT agents (antagonist LAG3 and agonist 41BB) decreased immunosuppressive neutrophils/granulocytes whereas ineffective ICT agents (antagonist PD1 and CTLA4) resulted in an increase (Fig 5f; Extended Data Fig 7d). LAG3 treatment also stimulated myeloid cell reprogramming including upregulation of antigen presentation genes (H2-Ab1, H2-Aa, CD74, B2m; Fig 6g) and downregulation of M2-associated transcription factors (Stat6, Socs3, IL1b) (Extended Data Fig 8h). 41BB treatment reprogrammed myeloid cells by increasing antigen presentation gene expression (H2-Ab1, H2-aa, CD74, B2m; Fig 6g), upregulating genes related to T cell chemoattraction (Cxcl10) and IFN signaling (Stat1) (Extended Data Fig 8i), and downregulating genes/transcription factors driving M2-polarization (CD206, IL10 and Socs3) (Extended Data Fig 8i).

Treatment with both effective and ineffective ICT agents resulted in a modest increase in DCs compared to control treatment (Fig 5f; Extended Data Fig 7d). Treatment with LAG3 resulted in increased B cells, while PD1 and CTLA4 treatment resulted in decreased B cells, consistent with prior studies demonstrating that B cell infiltration is associated with better prognosis in PDAC [56] and may promote immunotherapy response [57]. In summary, 41BB and LAG3 treatment promote anti-tumor immunity by modulating T cell subsets with anti-tumor activity, increasing T cell clonality and diversification, decreasing immunosuppressive myeloid cells and increasing antigen presentation/decreasing immunosuppressive capability of myeloid cells. However, this dual ICT combination was not sufficient to induce durable complete elimination of established tumors (Fig 5c, 5d), suggesting the need to target additional immune suppressive mechanisms in the iKRAS TIME.

Targeting CXCR2 expressed on MDSCs in iKRAS tumors.

Increased frequency of MDSCs in the bone marrow, circulation and TIME correlates with advanced disease stage and poor survival in PDAC patients [14, 15, 58, 59]. In human PDAC, unsupervised clustering analysis of the 39-gene MDSC signature [16] categorized 178 TCGA primary PDAC tumors into MDSC-high (n=114), -medium (n=54), and -low (n=10) groups (Extended Data Fig 2b), revealing that MDSC-high patients showed significantly lower OS compared to MDSC-low patients (Extended Data Fig 9a). These clinical correlations, coupled with abundant MDSCs in iKRAS PDAC tumors (Fig 1c–e; Extended Data Fig 1g, 1h), prompted exploration of the impact of MDSC neutralization in iKRAS PDAC tumor progression. Using a well-characterized anti-Gr1 neutralizing antibody [16], treatment of mice with established iKRAS tumors depleted MDSCs (decreased S100A9) (Extended Data Fig 9b), increased intra-tumoral CD8⁺ T cells (Extended Data Fig 9b), impaired tumor progression and increased OS, although all mice eventually succumbed (Fig 7a, Extended Data Fig 9c).

Figure 7.

Efficacy of targeted therapy directed against Cxcr1/2 and treatment effects on immune microenvironment. A. Overall survival of mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control or anti-Gr1 neutralizing antibody for 4 weeks (n=10 mice/group). B. Overall survival of mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control or SX-682 for 4 weeks (n=10 mice/group). C. Quantification of total CD45⁺ immune cells, CD4⁺ and CD8⁺ T cells, granulocytic- and monocytic-MDSCs, in established iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control, SX-682 or combination (anti-LAG3+anti-41BB+SX-682) for 4 weeks assessed by flow cytometry and analyzed by FlowJo (n=3 biological replicates). Two-sided Student’s t-test. D. Relative expression of Ifng and Tnf on T cells in scRNA-seq analysis of iKRAS tumors following treatment with control, SX-682 or combination (anti-LAG3+anti-41BB+SX-682) (n=3 tumors/group). (*p<0.05 two-sided unpaired Wilcox test) E. Multiple-testing corrected 95% binomial confidence intervals on the probability of a cell in each treatment group containing a TCR CD3R sequence which overlaps that of another cluster. (*p<0.05) F. Overall survival of mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control, SX-682 or SX-682 with CD8 T cell depleting antibody (n=10 mice/group). Statistical differences in A, B and F were identified by Kaplan-Meier with log-rank test. Data in C and D are presented as mean ± s.e.m.

CXCR2 is highly expressed in the myeloid_c2 and myeloid_c3 clusters of untreated iKRAS tumors (Extended Data Fig 3d–e) and in sorted granulocytic/neutrophilic MDSCs (Extended Data Fig 9d), mirroring treatment-naïve human PDAC specimens where CXCR2 is co-expressed with CD15 on CD11b⁺ myeloid cells (Extended Data Fig 9e). Similarly, scRNA-seq of 16 treatment naïve human PDAC patients showed CXCR2 expression in granulocytes/neutrophils (Extended Data Fig 9f; [10]). By comparison, CSF1R, CCR2 and TREM2, are predominantly expressed on the macrophages/monocytes in human PDAC and iKRAS tumors (Extended Data Fig 7c, 9f). These CXCR2 expression profiles coupled with its key role in MDSC recruitment to tumors [60], prompted evaluation of a clinical-stage Cxcr1/2 inhibitor, SX-682. Treatment with SX-682 significantly decreased migration of MDSCs isolated from iKRAS tumors towards conditioned medium in vitro (Extended Data Fig 9g) and inhibited tumor growth and increased OS of tumor-bearing iKRAS mice, although all the mice eventually succumbed (Fig 7b, Extended Data Fig 9h). Flow cytometry and confirmatory IHC analysis of iKRAS tumors following 4 weeks of SX-682 treatment showed reduced intratumoral CXCR2⁺ granulocytic MDSCs and increased CD8⁺ T cell infiltrate (Fig 7c; Extended Data Fig 9i–l), and showed modestly increased DC infiltration and no significant change in TAM infiltration (Extended Data Fig 9j). SX-682 treatment resulted in significant increase in the effector memory phenotype of both CD4⁺ and CD8⁺ T cells (Extended Data Fig 9i). Consistent with prior studies [59, 61], reduced CXCR2⁺ neutrophilic/granulocytic MDSCs was associated with a reciprocal increase in monocytic MDSCs by flow cytometry (Fig 7c) and scRNA-seq showing increased cluster M_c2 (S100a4, S100a6, Ly6c2, Ccr2 expressing monocytic MDSCs) (Extended Data Fig 9m). SX-682 treatment resulted in increased TNF expression by T cells, but no change in IFNγ expression (Fig 7d), as well as increased TCR diversification (Fig 7e). We performed CD8⁺ T cell depletion in the SX-682 trial showing loss of tumor inhibition and survival benefit, indicating that the therapeutic effects of SX-682 are mediated by CD8⁺ T cells (Fig 7f; Extended Data Fig 9n). These findings are consistent with prior studies, which demonstrated a critical role for Cxcr2 signaling in the myeloid compartment, in mediating PDAC tumorigenesis, antitumor immunity and chemotherapeutic response [29, 61]. Together, these findings suggest that, similar to targeting exhausted T cells with ICT agents (41BB and LAG3), targeting myeloid cells alone is of transient benefit.

Combination immunotherapy renders iKRAS tumors curable.

The above findings prompted combined ICT and CXCR2 inhibitor therapy, producing complete regression of established orthotopic iKRAS tumors in all mice (Fig 8a, Extended Data Fig 10a). The response was durable with 90% mice still alive at 18 months after discontinuation of treatment without evidence of relapse (Fig 8a). From a mechanistic standpoint, flow cytometric and IHC analyses showed that the triple combination of 41BB agonist+LAG3 antagonist+SX-682 resulted in near complete depletion of intratumoral CXCR2⁺ granulocytic/neutrophilic MDSCs and associated marked increase in CD8⁺ and CD4⁺ T cell infiltrates (Fig 7c; Extended Data Fig 9i–l). Correspondingly, scRNA-seq showed increased clusters T_c2 (high GzmK, high GzmB, high Ccl5 activated CD4⁺ T cells) and T_c3 (high GzmK, low GzmB, high Ccl5, activated CD8⁺ T cells) and decreased naïve/central memory CD4⁺ (cluster T_c5) and CD8⁺ (cluster T_c4) T cells (Fig 8b). Triple therapy resulted in higher TNF and IFNγ expression by T cells (Fig 7d) and increased effector memory phenotype of infiltrating CD4⁺ and CD8⁺ T cells (Extended Data Fig 9i). scRNA-seq analysis of T cells validated that triple therapy induced increased effector, memory signatures compared to control (Fig 8c) [62]. There was a modest increase in DC infiltration (Extended Data Fig 9j) and no significant change in TAM infiltration (Extended Data Fig 9j) consistent with the finding that CXCR1/2 inhibition selectively targets MDSCs. We noted that depletion of CXCR2⁺ neutrophilic/granulocytic MDSCs results in compensatory increase in monocytic MDSCs, similar to SX-682 monotherapy treatment (Fig 7c). scRNA-seq confirmed significant increase in cluster M_c2 fraction (S100a4, S100a6, Ly6c2, Ccr2 expressing monocytic MDSCs) (Extended Data Fig 9m). The increase in monocytic MDSCs after combination treatment did not impact CD4⁺/CD8⁺ T cell infiltration/activity or therapeutic efficacy of the combination regimen given durable, complete responses up to 18 months after discontinuation of combination treatment (Fig 8a; Extended Data Fig 10a), suggesting that monocytic MDSCs are unable to fully substitute for the immune suppressive activity of neutrophilic/granulocytic MDSCs in the iKRAS model. Treatment with the combination also resulted in TCR diversification, consistent with the effects noted with 41BB agonist, LAG3 antagonist and SX-682 monotherapies (Fig 7e). In contrast, SX-682+anti-PD1+anti-CTLA4 treatment resulted in modest decrease in tumor size along with increased survival, although none of the mice cleared their tumors following therapy and there were no durable responses noted (Fig 8d, Extended Data Fig 10b). These findings highlight the profound anti-tumor activity of the combination of 41BB agonist+LAG3 antagonist+SX-682, revealing that specific checkpoint combinations synergize with CXCR2 inhibition in the iKRAS model. Moreover, triple therapy was well tolerated with no treatment-related deaths during the 4-week treatment period and 9/10 mice survived for >18 months after treatment discontinuation (Fig 8a). Transient elevation in liver function tests (AST/ALT) was noted, consistent with the previously described hepatotoxicity related to 41BB antibodies (Extended Data Fig 10c) [63, 64].

Figure 8.

Efficacy of ICT in combination with targeted therapy directed against Cxcr1/2 and treatment effects on immune microenvironment. A. Overall survival of mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control or anti-LAG3+anti-41BB or anti-LAG3+anti-41BB+SX-682 for 4 weeks (n=10 mice/group). B. Changes in the fraction of cells in clusters T_c2, T_c3, T_c4 and T_c5 as a proportion of total T cells in scRNA-seq analysis of iKRAS tumors following treatment with control or combination (anti-LAG3+anti-41BB+SX-682) for 4 weeks (n=3 tumors/group). Mixed effect model. C. Relative expression of effector, memory, naïve and exhausted signatures [62] in T cells from scRNA-seq analysis of iKRAS tumors after combination treatment (anti-LAG3+anti-41BB+SX-682) for 4 weeks compared to control (n=3 tumors/group). (*p<0.05 two-sided unpaired Wilcox test) D. Overall survival of mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control or anti-PD1+anti-CTLA4 antibodies or SX-682 or anti-PD1+anti-CTLA4+SX-682 for 4 weeks (n=10 mice/group). E. Overall survival of mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) that were cured (survival >6 months after treatment discontinuation) by the combination in A. (anti-LAG3+anti-41BB+SX-682) re-challenged with secondary tumors (n=5 mice/group). Treatment naïve mice were animals who had never been exposed to iKRAS cells previously. F. Overall survival of mice bearing established autochthonous iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control or anti-PD1+anti-CTLA4 antibodies or anti-LAG3+anti-41BB antibodies or SX-682 or anti-LAG3+anti-41BB+SX-682 for 4 weeks (n=10 mice/group). Animals in the “LAG3+41BB+SX-682 Extended” treatment group received extended treatment with the combination regimen for 6 months or until death. Statistical differences in A, D, E and F were identified by Kaplan-Meier with log-rank test. Data in B is presented as mean ± s.e.m.

The effectiveness of triple therapy was further evidenced by tumor re-challenge studies, where 100% of the cured mice showed tumor rejection, consistent with a memory T cell response (Fig 8e). Moreover, given that tumor cells with identical tumor-initiating genetic alterations may elicit variable immune infiltrates and differential responses to immunotherapy [65], the curative efficacy of triple therapy was confirmed in additional independently derived iKRAS cell lines which exhibit different patterns of immune cell infiltration, specifically lower T cells and higher myeloid cells (Extended Data Fig 10d–f). It is possible that the orthotopic model may not fully recapitulate the complex fibroblastic stroma that is characteristic of the autochthonous model or human PDAC. Therefore, we tested the efficacy of triple therapy in established autochthonous tumors in the iKRAS model, using the same dose and schedule as the orthotopic iKRAS PDAC studies (Extended Data Fig 10g). Consistent with findings from the orthotopic model and human clinical trials [7], the combination of antagonist PD1 and CTLA4 antibodies had no appreciable effects, whereas triple therapy increased survival of iKRAS GEMM mice (Fig 8f). We also evaluated the effects of extended dosing (i.e. continuous dosing beyond 28 days) with the triple therapy regimen to evaluate whether it could result in durable remissions (Extended Data Fig 10g). We found that 2/10 mice had a durable response lasting >6 months with extended dosing with the triple therapy regimen (Fig 8f). Upon necropsy, these surviving mice had no evidence of primary tumor in the pancreas or metastases in the liver or lung, consistent with disease eradication.

Discussion

In this study, high-dimensional immune profiling of human and mouse PDAC was used to guide the development of an effective combination immunotherapy regimen, leading to unprecedented complete, durable responses and markedly improved survival in the treatment resistant iKRAS PDAC model. In contrast to anti-PD1 and anti-CTLA4, agonist 41BB and antagonist LAG3 treatment reprogrammed the TIME towards anti-tumor immunity with increased T cell subsets with anti-tumor effects, increased T cell clonality and diversification, decreased immunosuppressive myeloid cells and increased antigen presentation/decreased immunosuppressive capability of remaining myeloid cells. The addition of therapy targeting CXCR1/2 on neutrophilic/granulocytic MDSCs revealed that effective immune treatment is possible in PDAC but requires neutralization of distinct immunosuppressive mechanisms. Although we demonstrate reprogramming of MDSCs with SX-682 in mice, whether the same mechanisms govern myeloid cell migration into the TIME in human PDAC remains unknown.

As with all preclinical model systems and human bio-specimen correlations, prospective clinical trials will be needed to substantiate the hypothesis generated from this work. Along these lines, it is notable that expression of both 41BB and LAG3 is elevated on tumor infiltrating T cells compared to circulating T cells in human PDAC [25], consistent with the validated expression of both 41BB and LAG3 on T cells in human PDAC. Moreover, we found 81% of PDAC patients have 41BB expressing T cells, while 93% of PDAC patients have LAG3 expressing T cells, suggesting that these targets may be relevant for a meaningful fraction of PDAC patients. In this context, it is noteworthy that the majority of ongoing immunotherapy trials in PDAC employ PD1 and/or CTLA4 as the ICT backbone, including trials combining ICT with CXCR2 antagonists [4].

Recent studies implicate the CD155/TIGIT axis in mediating immune-evasion in PDAC, and human trials are ongoing to evaluate ICT antibodies targeting TIGIT [37]. Our findings suggest that ICT antibodies targeting 41BB and LAG3 also hold the potential to elicit meaningful responses in PDAC patients. More specifically, the tumor shrinkage/radiographic response in all orthotopic or autochthonous tumors treated with triple therapy points to a window of opportunity trial for surgical resection in patients with previously unresectable primary tumors due to involvement of nearby blood vessels, LNs or organs (such as duodenum). Moreover, the cures noted in orthotopic tumors and significant improvement in overall survival and durable remissions noted in autochthonous iKRAS tumors indicates that this lethal cancer can be rendered vulnerable to combination immunotherapy.

Methods

Transgenic and syngeneic mouse studies

All animal work performed in this study was approved by MD Anderson and Rutgers Institutional Animal Care and Use Committee. All animals were maintained in pathogen-free conditions and cared for in accordance with the International Association for Assessment and Accreditation of Laboratory Animal Care policies and certification. iKRAS (p48Cre; tetO_LSL-Kras^G12D; ROSA_rtTA; p53^L/+) genetically engineered mice were described previously and were backcrossed to the C57BL/6 background for more than eight generations to achieve a pure B6 mouse to generate syngeneic cell lines [8]. Female or male mice were administered doxy water (dox 2mg/ml, sucrose 40mg/ml) starting at 4 weeks of age to activate transgenic KRAS^G12D expression. For orthotopic pancreas transplantation, C57BL/6 female or male mice aged 5–7 weeks (Jackson Labs) were anaesthetized using ketamine/xylazine. An incision was made in the left abdomen and the pancreas was gently pulled out along with the spleen. iKRAS cells were slowly injected into the tail of the pancreas using a Hamilton syringe. Cells (5 × 10⁵ in 5μl) mixed with 5μl Matrigel were injected. Analgesic was administered after surgery along with temperature-controlled post-surgical monitoring. Doxycycline was provided to the animals in the form of doxy water (dox 2mg/ml, sucrose 40 mg/ml) starting on the day of tumor cell injection. Animals were imaged (IVIS Spectrum, PerkinElmer and Bruker ICON MRI) 10 days after surgery to assess successful implantation of tumors. Only tumors of similar volume (~250mm³) were used for treatment studies. Animals underwent MRI imaging to monitor the progression of the tumors. Mice with autochthonous PDAC tumors were euthanized for tumor collection once tumor volume was approximately the same as orthotopic tumors (~1000mm³). Owing to the retroperitoneal location of PDAC tumors, we used signs of lethargy, reduced mobility and morbidity, rather than maximal tumor size, as a protocol-enforced endpoint.

Non-invasive mouse imaging

For MRI imaging with Bruker ICON, animals were anesthetized with 1–3% isoflurane and placed on the ICON animal bed. The MRI coil was secured into position over the animals and the entire bed assembly was placed into the Bruker ICON MRI bore. Rapid acquisition with relaxation enhancement (RARE) T2-weighted images were acquired in coronal, sagittal and axial planes. Coronal and sagittal T2 parameters were as follows: TE-18ms, TR-2197ms, slice thickness of 1mm, with a 1.25mm slice gap. Axial T2 parameters were as follows: TE-14.2ms, TR-1464ms, slice thickness 1.25mm with a 1.5mm slice gap. After imaging was completed, the animals were allowed to recover under a heating lamp until fully conscious. MRI images were loaded into ImageJ to manually demarcate the contour of the pancreas and calculate the total volume. Bioluminescence imaging with IVIS Spectrum (Perkin Elmer) was performed by intraperitoneal injection of 1.5mg of D-luciferin (PerkinElmer). The Living Image 4.7 software (PerkinElmer) was used for analysis of images post-acquisition. PET was performed using a Bruker Albira PET/CT scanner 1 hour after injection of ~150μCi of ¹⁸FDG. Respiratory rate was monitored with a BIOPAC physiological monitoring system used to gate the CT.

CyTOF analysis of mouse and human tumors

Tumor cells were isolated from iKRAS tumors using the Mouse Tumor Dissociation Kit (Miltenyi Biotec). A total of five PDAC patients undergoing pancreaticoduodenectomy were recruited at MD Anderson Cancer Center through informed written consent following IRB approval. All patients consented to participation with publication of de-identified data. Pancreatic tissue was delivered to the laboratory on ice after surgical resection in DMEM in a 15mL conical tube. Tumor cells were isolated from human PDAC tumors using the Human Tumor Dissociation Kit (Miltenyi Biotec). Single cells were isolated from tumors using standard protocol and as described previously [16, 19]. All isolated cells were depleted of erythrocytes by hypotonic lysis. For CyTOF analysis, cells were blocked for FcγR for 10 min and incubated with CyTOF antibody cocktail mix (see Supplementary Table 2 for list of antibodies) for 30 min at room temperature. Cells were washed once and incubated with Cell-ID Cisplatin (Fluidigm) at 2.5μM for 2 min for viability staining. Cells were fixed with MaxPar Fix and Perm Buffer containing Cell-ID Intercalator-Ir (Fluidigm) at 0.125 μM and incubated at 4^oC overnight to stain the nuclei. The samples were analyzed with CyTOF instrument (Fluidigm) in the Flow Cytometry and Cellular Imaging Core Facility at MD Anderson Cancer Center. Data were analyzed with FlowJo (Tree Star) and SPADE software.

Flow cytometry analysis

Single cells were obtained as described above for CyTOF. To assess cell viability, cells were incubated with Ghost dye violet (Tonbo Biosciences) for 15 minutes in dark and then stained with indicated antibodies for 30 minutes on ice prior to FACS analysis. Fluorochrome-conjugated antibody information is listed in the reporting summary. For FOXP3 staining, cells were fixed and permeabilized (eBioscience FOXP3/Transcription Factor Staining Buffer Set) and stained with FOXP3. All samples were acquired with the FACSAria Fusion sorter (Becton Dickinson) and analyzed with FlowJo software (Tree Star).

T cell suppression and MDSC migration assay

MDSCs were isolated from the spleens of iKRAS mice using a Mouse MDSC Isolation Kit (Miltenyi Biotec) and plated in RPMI1640 supplemented with 10% FBS and antibiotics. MDSCs (1×10⁵ cells/well) were seeded in the top chamber of the transwell (Corning). Conditioned media from cultured iKRAS cells were collected and added to the bottom layer of the transwell. After 4h incubation, cells that had completely migrated to the bottom chamber were counted. A T cell suppression assay was performed as described previously [16, 19] using equal numbers of MACS-sorted MDSCs and CFSE (Invitrogen)-labeled MACS-sorted (Miltenyi Biotec) CD8⁺ T cells in anti-CD3- and anti-CD28-coated 96 well plates at an MDSC/T cell ratio of 0:1, 1:1, and 4:1 with 5.0×10⁵ MDSCs. MDSCs were isolated from iKRAS tumors and CD8⁺ T cells were isolated from the spleen of C57BL/6 mice (Jackson Labs). CFSE intensity was quantified 72 hours later with peaks identified by a BD LSRFortressa Cell Analyzer. CFSE peaks indicated the division times. Division times 0–2 and 3–4 were defined as low proliferation and high proliferation, respectively.

Multiplex Immunofluorescence Staining

Multiplex immunofluorescence staining was performed as described and validated previously [66]. Briefly, four micrometer-thick formalin fixed, paraffin embedded tissue sections were stained using panels containing the antibodies outlined in the reporting summary. All the markers were stained in sequence using their respective fluorophores in the Opal 7 kit (Akoya Biosciences/PerkinElmer). The stained slides were scanned using the multispectral microscope, Vectra 3.0.3 imaging system (Akoya Biosciences/PerkinElmer PerkinElmer), under fluorescence conditions at low magnification (10x). After scanned in low magnification, each core was scanned at high magnification (20x) and analyzed by a pathologist using InForm 2.4.0 image analysis software (Akoya Biosciences/PerkinElmer PerkinElmer). Marker co-localization was used to identify specific cell phenotypes (CD33⁺CD11b⁺CD66b⁺; CD33⁺CD14⁺CD68⁺; CD3⁺41BB⁺; CD3⁺LAG3⁺; CD3⁺CD8⁺CD45RO⁺; CD3⁺CD8⁺GzmB⁺). All data were consolidated using R studio 3.5.3 (Phenopter 0.2.2 packet, Akoya Biosciences/PerkinElmer) and SAS 7.1 Enterprise.

Cell lines

iKRAS (p48Cre; tetO_LSL-Kras^G12D; ROSA_rtTA; p53^L/+) syngeneic cell lines have been described previously [8]. Cells were maintained in culture in RPMI medium supplemented with 10% FBS and doxycycline. All cell lines were tested for Mycoplasma and found to be negative within 3 months of performing experiments.

Immunohistochemistry

Human PDAC samples were obtained from MD Anderson’s Tissue Biobank. Human studies were approved by MD Anderson’s Institutional Review Board (IRB), and prior informed consent was obtained from all subjects under IRB protocol LAB05–0854. Mouse tissues were fixed in 10% formalin overnight and embedded in paraffin. Immunohistochemical staining was performed as described previously [16]. Slides were scanned using Aperio AT2 slide scanner (Leica Biosystems). Images were visualized and a pathologist selected regions of interest (tumor or normal pancreas); necrotic areas and areas with artifact were excluded from the analysis. Primary antibodies for mouse and human tissue staining are listed in the reporting summary.

Immune checkpoint therapy, chemotherapy and targeted therapy

For in vivo pharmacological inhibition, gemcitabine (Selleck Chemicals) was dosed at 100mg/kg IP. SX-682 (Syntrix Pharmaceuticals) was dosed PO ad libitum (formulated concentration 714mg/kg feed); therapeutic plasma levels (range 0.5–10 μg/mL) were confirmed with this feed using LC/MS-MS. For ICT and Gr1/CD8-neutralizing antibody treatment, anti-PD1 (clone RMP1–14, BioXCell, BE0146), anti-CTLA4 (clone 9H10, BioXCell, BE0131), anti-TIM3 (clone RMT3–23, BioXCell, BE0115), anti-OX40 (clone OX-86, BioXCell, BE0031), anti-41BB (clone LOB12.3, BioXCell, BE0169), anti-LAG3 (clone C9B7W, BioXCell, BE0174), anti-CD8 (clone 2.43, BioXCell, BE0061) and anti-Gr1 (clone RB6–8C5, BioXCell, BE0075), antibodies (or their respective isotype IgG controls) were intraperitoneally administered at 200μg per injection three times per week. The duration of treatment was 4 weeks before endpoint analysis and survival analysis unless otherwise indicated.

Computational analysis of human PDAC TCGA, ICGC and single-cell RNA sequencing data

For TCGA GSEA analysis, the TCGA PDAC mRNA dataset, gene mutations and clinical survival was downloaded from TCGA website. For ICGC GSEA analysis, the ICGC PDAC AU mRNA dataset, gene mutations and clinical survival was downloaded from the ICGC data portal. For analysis of human PDAC data, we utilized a 39 gene human MDSC signature, which was described previously [16]. The gene expression data of 178 TCGA PDAC samples were clustered using the 39 MDSC genes into MDSC-high, MDSC-low, and MDSC-medium (distance between pairs of samples was measured by Manhattan distance, and clustering was then performed using complete-linkage hierarchical clustering). Similarly normalized gene expression data from PDAC TCGA (178 samples) or ICGC-AU (92 samples) were used to infer the relative proportions of infiltrating immune cells using the CIBERSORTx algorithm which was described previously [17]. Estimated fractions of each immune cell subset were related to survival using univariate Cox regression. Two human PDAC single-cell RNA sequencing cohorts were used from Peng et al. [27] and Steele et al. [10]. For Peng et al. cohort, original cell type annotations of single-cell RNA clusters were used. For Steele et al. cohort, data were processed and clustered according to the R scripts from the original paper. UMAP clusters were further annotated using rSuperCT algorithm [67]. Expression of selected marker genes were compared among different cell types. All data processing and analysis were implemented in R 4.0.5 environment and Seurat package version 4.0.1.

Single-cell RNA sequencing, transcriptomic and T cell receptor analysis of mouse tumors

Flow cytometry to isolate live, CD45⁺ cells was performed using standard protocol as noted above on FACSAria Fusion sorter (Becton Dickinson) and analyzed with FlowJo software (Tree Star). Live, CD45⁺ cells were processed with the 10X Genomics Chromium platform, with the 5’ V(D)J solution chemistry per the manufacturer’s protocol. T cell receptor sequences were enriched with the primers listed below, and the resulting libraries were pooled and sequenced 150bp PE on the Illumina Miseq with the V2 300cycle kit. Single cell transcriptome libraries were pooled and sequenced 26bp-91bp PE on the Novaseq S2 to a targeted depth of 100,000 reads per cell. Raw sequencing data were processed through the 10X Genomics cellranger pipeline version 2.1.0 and then analyzed in R. Cells were detected using the DropletUtils package [68] with an FDR of 0.01, and barcode swapped counts were removed using swappedDrops. Supernatant RNA contamination was filtered using the package SoupX [69]. Data were then processed using the Seurat package [70]. Cells with a mitochondrial gene percentage greater than 15% were filtered, and samples were corrected for batch effects by aligning the first 35 canonical correlations using the MultiCCA function. Cells were clustered with SNN. Differences in cluster fractions were assessed by the significance of treatment as a fixed effect in a binomial mixture model (glmer in the lme4 R package) with replicate included as a random effect. Pseudotime analysis was performed using the Monocle2 package per the recommended workflow [35]. To determine the lineage of each individual T cell in the PDAC tumors after the various ICT treatments, we designed primers for the mouse α and β TCR locus and performed targeted PCR on the 10X genomics single-cell 5’ cDNA product (Supplementary Table 4). From the TCR product library, we assembled the full-length TCR α and β sequences. Raw TCR sequencing data were processed through the 10X Genomics cellranger vdj pipeline version 2.1.0. Clonotypes where only an α or β chain were detected, but exactly matched arCDR3 nucleotide sequence from an α-β paired clonotype were combined into the paired clonotype for further analysis.

Statistics and reproducibility

Continuous measurements were compared pairwise using a two-sample Welch Student’s t-test. Data are presented as mean ± s.d. unless indicated otherwise. Measures expressed as percentages were transformed (to improve normality) via a logit transformation prior to using Student’s t-tests. Survival outcomes were compared using log-rank tests and Kaplan-Meier survival curves. In the case of multiple pairwise comparisons, Benjamini-Hochberg adjusted p-values were reported [71]. Figs 2a, 2b and 5e show representative images from a total of n=54 patients.

Transgenic and syngeneic mouse experiments were randomized and investigators were blinded to allocation during experiments and outcome assessment. No statistical method was used to predetermine sample size. The animal cohort sizes for the study were estimated based on previous experience using similar mouse models that showed significance. No animals or data points were excluded from the analyses. Data distribution was assumed to be normal but this was not formally tested.

Changes in average relative expression or expression of gene signature scores in scRNA-seq analyses was performed by two-sided unpaired Wilcox test. For single cell populations, differences in cluster fractions were assessed by the significance of treatment as a fixed effect in a binomial mixture model (glmer in the lme4 R package) with sample replicate included as a random effect.

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Extended Data

Extended Data Fig. 1. Prominent infiltration of myeloid immunosuppressive cells in iKRAS tumors.

A. PDAC tumor development in syngeneic mouse model with representative images of tumor detected by bioluminescence, PET/CT and MRI at indicated timepoints. B. Tumor volume measured by MRI at indicated timepoints (top) and Kaplan-Meier curve depicting overall survival (bottom) for untreated iKRAS tumor bearing mice (n=10 mice). C. Representative images of normal pancreas, orthotopic and autochthonous (GEMM) iKRAS tumors with H&E, Masson Trichrome, smooth muscle actin (SMA) and vimentin staining. Scale bars: 100μm. D. Representative H&E images of iKRAS tumors invading into adjacent lymph nodes (left=4x magnification, right=20x magnification). E. Representative coronal and axial MRI images of iKRAS tumor invading into duodenum. F. Representative images of normal pancreas and human PDAC tumors with SMA, Vimentin and CD45 staining. Scale bars: 100μm. Red arrow indicates positively stained cells. G. Percentage of granulocytic (CD45⁺CD11b⁺Ly6G⁺Ly6C^-) and monocytic MDSCs (CD45⁺CD11b⁺Ly6G^-Ly6C⁺) within syngeneic iKRAS tumors (n=10 tumors) assessed by CyTOF at 4 weeks after initial tumor detection. Two-sided Student’s t-test. H. Representative images (bottom) of normal pancreas, orthotopic and autochthonous (GEMM) iKRAS tumors with indicated staining. n=6 biological replicates. Scale bars: 100μm. The bar graph (top) shows quantification of each cell type as analyzed by IHC. Two-sided Student’s t-test. I. Representative images of normal pancreas and orthotopic iKRAS tumors with indicated staining. Scale bars: 100μm. J. Percentage of Treg (CD45⁺CD3⁺TCRβ⁺CD4⁺FoxP3⁺) among CD4⁺ T cells within syngeneic iKRAS tumors (n=10 tumors) assessed by CyTOF at 4 weeks after initial tumor detection. Two-sided Student’s t-test. K. Representative images of normal pancreas and human PDAC tumors with indicated staining. Scale bars: 100μm. Red arrow indicates positively stained cells. Data in G, H and J are presented as mean ± s.e.m.

Extended Data Fig. 2. Prominent infiltration of myeloid immunosuppressive cells in human PDAC tumors.

A. Representative images of normal pancreas and human PDAC tumors with indicated staining. Scale bars: 100μm. Red arrow indicates positively stained cells. B. Clustering of human TCGA PDAC samples (n=178 patients) into MDSC-high, MDSC-low, and MDSC-medium groups using a 39-gene MDSC signature [16]. C. CIBERSORTx quantification of monocyte/macrophage subset fraction in human PDAC samples; TCGA (n=178 patients) and ICGC-AU (n=92 patients). D. Representative images of normal pancreas and human PDAC tumors with indicated staining. Scale bars: 100μm. Red arrow indicates positively stained cells. E. Kaplan-Meier plot depicting overall survival of TCGA PDAC patients (n=178 patients) grouped by the gene expression signatures of C1q⁺ TAM (top) and Spp1⁺ TAM (bottom).

Extended Data Fig. 3. Heterogeneity of myeloid cells in iKRAS PDAC tumors identified by single cell gene expression profiling.

A. UMAP of all live CD45⁺ cells used for scRNA-seq analysis of untreated iKRAS tumors (n=4,080 cells). B. Representative genes and functional markers used for identification of immune cell clusters. C. Heatmap of six immune cell clusters with unique signature genes. D. Representative genes and functional markers used for identification of myeloid cell clusters. E. Heatmap of myeloid cell clusters with unique signature genes. F. Representative genes and functional markers used for identification of dendritic cell clusters.

Extended Data Fig. 4. Dysfunctional phenotype of T cells in iKRAS PDAC tumors identified by single cell gene expression profiling.

A. Representative genes and functional markers used for identification of T cell clusters. B. Heatmap of two CD4⁺ and four CD8⁺ T cell clusters with unique signature genes. C. Cell cycle scoring for two CD4⁺ and four CD8⁺ T cell clusters. D. Relative expression of select genes in CD8⁺ T cells as a function of pseudotime from Monocle2 inferred trajectory. Each point corresponds to a single cell, colored by CD8⁺ T cell cluster. Lines represent average expression at that location in the trajectory. E. Quantification of immune checkpoint expression on infiltrating CD4⁺ and CD8⁺ T cells in iKRAS tumors (n=3 biological replicates), assessed by flow cytometry and analyzed by FlowJo.

Extended Data Fig. 5. Efficacy of immune checkpoint therapy (ICT) and treatment effects on immune microenvironment.

A. Treatment schedule and monitoring procedures for preclinical trials to evaluate effect of ICT on iKRAS PDAC bearing mice. B. Heatmap of immune checkpoint expression on T cells after 4-week treatment with control, anti-PD1 or anti-CTLA4 antibody (n=3 mice/ group). C. UMAP demonstrating cell types in single-cell RNA sequencing of human PDAC samples from Peng et al. [27] and Steele et al. [10] (left), and expression of LAG3 and 41BB (TNFRSF9) on T cells (right). D. UMAP of all live CD45⁺ cells used for scRNA-seq analysis of iKRAS tumors treated with control, anti-PD1, anti-CTLA4, anti-41BB, anti-LAG3, SX-682 or combination (anti-LAG3+anti-41BB+SX-682) treatment (n=3 mice/group). E. UMAP projection of immune cell clusters (top) and cells with TCR detected (bottom). F. Violin plots displaying relative expression of representative genes and functional markers used for identification of immune cell clusters.

Extended Data Fig. 6. Efficacy of immune checkpoint therapy (ICT) and treatment effects on immune microenvironment.

A. Heatmap of six immune cell clusters with unique signature genes. B. UMAP projection of T cell clusters (top) and violin plots displaying relative expression of representative genes and functional markers used for identification of T cell clusters (bottom). C. Heatmap of ten T cell clusters with unique signature genes. D. UMAP projection of neutrophil/granulocyte clusters (top) and violin plots displaying relative expression of representative genes and functional markers used for identification of neutrophil/granulocyte clusters (bottom).

Extended Data Fig. 7. Efficacy of immune checkpoint therapy (ICT) and treatment effects on immune microenvironment.

A. Heatmap of five neutrophil/granulocyte clusters with unique signature genes. B. UMAP projection of monocyte/macrophage clusters (top) and violin plots displaying relative expression of representative genes and functional markers used for identification of monocyte/macrophage clusters (bottom). C. Heatmap of five monocyte/macrophage clusters with unique signature genes. D. Proportion of immune cell subtypes in single-cell sequencing analysis of established iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control, anti-PD1, anti-CTLA4, anti-41BB or anti-LAG3 antibody for 4 weeks (n=3 tumors/group).

Extended Data Fig. 8. Effects of immune checkpoint therapy (ICT) treatment on immune microenvironment.

A Proportion of T cell subtypes in scRNA-seq analysis of established iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control, anti-PD1, anti-CTLA4, anti-41BB or anti-LAG3 antibody for 4 weeks (n=3 tumors/group). B. Top gene ontologies from GSEA of differential expression in T cells from anti-41BB and control antibody treated mice (n=3 mice/group). C. Circos plots of T cell receptor clonotype frequencies and expression states of CD8⁺ T cells in iKRAS tumors after treatment with control (left) and anti-41BB antibody (right) for 4 weeks. Outer histogram is the frequency of each clonotype. Inner bars show the fraction of cells of particular clonotype in each expression state (colors correspond to the clusters in Extended Data Fig 8a). Inner dendrograms are the hierarchical clustering of gene expression centroids for each clonotype. D. Multiple-testing corrected 95% binomial confidence intervals on the probability of a cell in each treatment group containing a TCR CD3R sequence which overlaps that of another cluster. (*p<0.05) E. Violin plots showing CCR7 expression in CD4⁺, CD8⁺ and CD4^-CD8^- T cells. (*p<0.05 two-sided unpaired Wilcox test) F. Proportion of CD4⁺, CD8⁺ and CD4^-CD8^- T cells with expression of CCR7 (left) and IL2RB (right). G. Expression of genes and functional markers on CD3⁺CD4^-CD8^- T cells. H. Violin plots showing expression of Stat6, Socs3 and Il1b among myeloid cells from control and anti-LAG3 antibody-treated tumors (n=3 mice/group). (*p<0.05 two-sided unpaired Wilcox test) I. Violin plots showing expression of Cxcl10, Stat1, Il10, Mrc1 and Socs3 among myeloid cells from control and anti-41BB antibody-treated tumors (n=3 mice/group). (*p<0.05 two-sided unpaired Wilcox test)

Extended Data Fig. 9. Efficacy of targeted therapy directed against Cxcr1/2 and treatment effects on immune microenvironment.

A. Kaplan-Meier plot depicting overall survival differences between patients with MDSC-high vs. MDSC-low signatures based on clustering of human TCGA PDAC samples (n=178 patients) shown in Extended Data Fig 2b. B. Representative images (left) of established iKRAS tumors treated with control and anti-Gr1 neutralizing antibody for 4 weeks with indicated staining. Scale bars: 100μm. The bar graph (right) shows quantification of each cell type as analyzed by IHC. n=6 biological replicates. Two-sided Student’s t-test. C. Tumor volume after 4 weeks of treatment with control or anti-Gr1 neutralizing antibody in mice bearing established (tumor volume ~250mm³ prior to treatment initiation) orthotopic iKRAS tumors (n=10 mice/group). Two-sided Student’s t-test. D. Expression of Cxcr2 on granulocytic MDSCs in untreated iKRAS tumors, assessed by flow cytometry and analyzed by FlowJo (n=3 tumors). E. Representative images of human PDAC tumors with indicated staining. Scale bars: 100μm. Red arrow indicates positively stained cells in the same area of a core specimen. F. UMAP demonstrating cell types in single-cell RNA sequencing of human PDAC samples from Steele et al. [10] with expression of CXCR1 and CXCR2 on granulocytes/neutrophils, and expression of CSF1R, CCR2 and TREM2 on monocytes/macrophages. G. Migration of MDSCs toward conditioned medium from iKRAS tumor cells treated with control or SX-682 (n=3 biological replicates). Student’s t-test. H. Tumor volume after 4 weeks of treatment with control or SX-682 in mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) (n=10 mice/group). Two-sided Student’s t-test. I. Stratification of infiltrating CD4⁺ and CD8⁺ T cells as naive (CD44^lowCD62L^high), central memory (CD44^highCD62L^high), and effector memory (CD44^highCD62L^low), in established iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control, SX-682 or combination (anti-LAG3+anti-41BB+SX-682) for 4 weeks assessed by flow cytometry and analyzed by FlowJo (n=3 biological replicates). Two-sided Student’s t-test. J. Quantification of total tumor associated macrophages (TAM) and dendritic cells (DC) in established iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control, SX-682 or combination (anti-LAG3+anti-41BB+SX-682) for 4 weeks assessed by flow cytometry and analyzed by FlowJo (n=3 biological replicates). Two-sided Student’s t-test. K. Expression of Cxcr2 on myeloid cells in established iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control, SX-682 or combination (anti-LAG3+anti-41BB+SX-682) for 4 weeks assessed by flow cytometry and analyzed by FlowJo (n=3 biological replicates). L. Representative images (left) of control, SX-682 or combination (anti-LAG3+anti-41BB+SX-682) treated iKRAS tumors with indicated staining. Scale bars: 100μm. The bar graphs (right) show quantification of each cell type as analyzed by IHC. n=6 biological replicates. Two-sided Student’s t-test. M. Quantification of change in the proportion of cells in cluster M_c2 as a proportion of total monocyte/macrophage cells in scRNA-seq analysis of iKRAS tumors following treatment with control, SX-682 or combination (anti-LAG3+anti-41BB+SX-682) for 4 weeks (n=3 mice/group). (*p<0.05 mixed effect model) N. Tumor volume of mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control, SX-682 or SX-682 with CD8 T cell depleting antibody (n=10 mice/group). Two-sided Student’s t-test. Data in D, G, I, J and M are presented as mean ± s.e.m.

Extended Data Fig. 10. Efficacy of ICT in combination with targeted therapy directed against Cxcr1/2 and treatment effects on immune microenvironment.

A. Tumor volume of mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control or anti-LAG3+anti-41BB antibodies or combination (anti-LAG3+anti-41BB+SX-682) for 4 weeks (n=10 mice/group). Two-sided Student’s t-test. B. Tumor volume of mice bearing established orthotopic iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control or anti-PD1+anti-CTLA4 antibodies or SX-682 or anti-PD1+anti-CTLA4+SX-682 for 4 weeks (n=10 mice/group). Two-sided Student’s t-test. C. Body weight of mice (top left), before (pre-treatment), during (2 weeks) and after (4 weeks) treatment with control, SX-682 or combination (anti-LAG3+anti-41BB+SX-682) for 4 weeks (n=4 biological replicates). Mouse toxicity tests including creatinine, blood urea nitrogen (BUN), aspartate aminotransferase (AST), alanine aminotransferase (ALT), total bilirubin, and alkaline phosphatase in the indicated treatment groups (n=4 biological replicates). Representative images of H&E staining (middle) of the lung, heart, liver, kidney and spleen in the indicated treatment groups (n=4 mice/group). Inset (bottom) shows representative H&E staining of liver tissues in the indicated treatment groups at higher magnification. D. CyTOF analysis of tumors from syngeneic iKRAS 2 and iKRAS 3 tumor bearing mice with equivalent tumor volume (~1000mm³) (n=10 tumors/group). E. Quantification of tumor infiltrating CD45⁺ cells in syngeneic iKRAS 2 and iKRAS 3 tumors with equivalent tumor volume (~1000mm³) assessed by CyTOF (n=10 tumors/group). F. Overall survival of mice bearing established orthotopic iKRAS 2 and iKRAS 3 tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control or anti-LAG3+anti-41BB+SX-682 for 4 weeks (n=10 mice/group). Statistical differences were identified by Kaplan-Meier with log-rank test. G. Treatment schedule and monitoring procedures for preclinical trial to evaluate overall survival of mice bearing established autochthonous iKRAS tumors (tumor volume ~250mm³ prior to treatment initiation) treated with control or anti-PD1+anti-CTLA4 antibodies or anti-LAG3+anti-41BB antibodies or SX-682 or combination (anti-LAG3+anti-41BB+SX-682) (n=10 mice/group). Animals in the “extended” treatment group received treatment with the combination regimen for 6 months or until death. Data in E are presented as mean ± s.e.m.

Data Availability Statement

Source data for main figures and extended data figures are provided in the online version of this paper. Murine single-cell RNA sequencing and TCR sequencing data supporting the findings of this study have been deposited on Sequence Read Archive (SRA) under BioProject accession code PRJNA496487. Human PDAC genomic data were derived from the TCGA Research Network [http://cancergenome.nih.gov] and ICGC Research Network [https://dcc.icgc.org]. Human PDAC single-cell RNA sequencing data were derived from [CRA001160, https://ngdc.cncb.ac.cn/gsa/browse/CRA001160] and [GSE155698, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE155698]. All other data are available from the corresponding authors upon reasonable request.