Blocking IL1 Beta Promotes Tumor Regression and Remodeling of the Myeloid Compartment in a Renal Cell Carcinoma Model: Multidimensional Analyses

Renal cell carcinoma (RCC) is the eight most common cancer in the United States, with an estimated 68,000 deaths attributable to this disease in 2020 (1). Numerous studies showed that RCC is unresponsive to chemotherapy, yet relatively responsive to immunotherapies, including high-dose IL2 (2). On the basis of these clinical data, metastatic RCC represents a fertile ground to translate combinatorial approaches involving immune checkpoint inhibition. To that end, there are now three FDA-approved regimens combining PD-1- or PD-L1–targeted therapeutics with additional agents for first-line treatment of patients with metastatic kidney cancer (3).

The current standard-of-care treatment for first-line metastatic RCC involves either combination PD-1/CTLA-4 blockade (4) or combining an anti-PD-1/PD-L1 therapy with a VEGF-targeted tyrosine kinase inhibitor (TKI), such as axitinib (5, 6). While these combinatorial approaches show improved response rates and overall survival relative to treating with a single-agent TKI (sunitinib), approximately 40%–60% of patients do not respond to first-line combination therapy. Even more striking, close to 50% of patients with metastatic RCC progress to the extent that they are unable to move on to a second-line treatment, highlighting the notion that maximizing efficacy in the first-line setting may improve overall survival (7).

One potential reason for the failure of current combination approaches is the presence of immunosuppressive cell populations within the tumor microenvironment (TME), including myeloid-derived suppressor cells (MDSC), M2-like tumor-associated macrophages (TAM), and regulatory T cells (Treg), that prevent effective immune-mediated tumor rejection (8). In human RCC, increased MDSC infiltration within the TME is associated with increased expression of a number of cytokines, including IL1β, IL8, and CXCL5, and overall poor prognosis. Hence, targeting myeloid cells may represent a potential strategy to overcome tumor tolerance (9).

One potential target to manipulate innate immune cells in RCC is IL1β (10, 11). Intratumoral macrophages, monocytes, tumor cells, and tumor vasculature are all proposed sources of IL1β, which has numerous protumorigenic properties, including inflammasome activation, promotion of tumor angiogenesis, and recruitment of immunosuppressive cells, like polymorphonuclear MDSCs (PMN-MDSC; ref. 12, 13). Early studies with IL1-knockout mice showed that tumor growth was nearly completely abrogated in mice lacking IL1β, suggesting that at early stages of tumor growth, IL1β mediates sterile inflammation to promote tumor invasiveness (14). In preclinical models of breast cancer, IL1β blockade decreased macrophage-mediated tumor immunosuppression, and combining anti-IL1β with anti-PD-1 resulted in near complete tumor eradication (15). In humanized breast cancer models, IL1 was implicated in driving tumor-promoting inflammatory changes that were reversed with TGFβ or IL1 blockade (16). Similarly, transfection of CT26 tumors with IL1β led to a 50-fold increase in PMN-MDSCs, suggesting that IL1 promotes immunosuppressive cell infiltration (17). In human xenograft models of kidney cancer, blocking the IL1 receptor axis resulted in reduced protumorigenic TAM and decreased in vivo tumor growth (18). In the same study, gene expression analyses showed that IL1β expression correlated with myelomonocytic markers and advanced tumor stage, suggesting that IL1β may promote RCC tumor progression. Of note, the most common treatment for RCC, VEGF tyrosine kinase inhibition, has not been extensively modeled in combination with IL1β blockade.

Recent clinical data from a large, double-blinded placebo-controlled trial evaluating the efficacy of IL1β blockade in secondary cardiac prevention provide further support for testing IL1β blockade in cancer (19). Canakinumab Anti-Inflammatory Thrombosis Outcomes Study (CANTOS) was a randomized trial testing the ability of the IL1β-blocking antibody, canakinumab, to prevent recurrent vascular events in patients following a myocardial infarction (19). The study met its primary cardiac endpoint, and a prespecified secondary endpoint of CANTOS quantified lung cancer incidence and mortality (20). These analyses showed that the total cancer incidence was strikingly lower in patients receiving canakinumab, with an estimated incidence per 100 person-years of 0.64 in the placebo group in comparison with 0.55, 0.40, and 0.31 in the canakinumab-treated groups at doses of 50, 150, and 300 mg, respectively (P_trend = 0.0007 across all treated patients as compared with placebo). Lung cancer mortality was also decreased in patients treated with the 300 mg dose of canakinumab in comparison with patients treated with placebo (HR, 0.49). These preclinical and emerging clinical data support targeting IL1β in cancer, but a paucity of data exist regarding the immune changes mediated by IL1β blockade in combination with other therapeutics; that is, to date, a single study evaluated the additive effects of combination PD-1/IL1β blockade (15), and no such data exist in an RCC model. We hypothesized that combining IL1β blockade with agents active in human RCC, anti-PD-1 or the multi-targeted TKI, cabozantinib, would lead to delayed tumor outgrowth and changes in the myeloid compartment of the TME in a murine model of RCC.

Here, we analyzed the immunologic effects of anti-IL1β on intratumoral immune cell subsets and show that while anti-PD-1, cabozantinib, or anti-IL1β monotherapy exerts antitumor effects in established tumors, combination therapy with anti-PD-1 or cabozantinib plus anti-IL1β led to a more significant reduction in tumor growth. This enhancement in antitumor response was associated with decreased intratumoral PMN-MDSCs and skewing of TAMs toward an M1-like phenotype. Deep immune profiling of treated tumors using multiparameter flow cytometry showed that combination treatment with cabozantinib plus anti-IL1β has pronounced effects on intratumoral myeloid populations, suggesting that this combination may act through a relatively T-cell–independent mechanism. Together these data support the hypothesis that IL1β may function as an important mediator of intratumoral immunosuppression and suggest that anti-IL1β–based combination regimens may have activity in patients with RCC.

The murine RCC line (RENCA) was obtained from the ATCC. Cells were maintained in RPMI medium supplemented with 10% FCS and penicillin/streptomycin with nonessential amino acids and l-glutamine (RPMI complete). Cells were Mycoplasma tested prior to implantation by using PCR by ATCC.

Female BALB/C mice (6–8 weeks old) were purchased from The Jackson Laboratory. All mice were housed in microisolator cages and treated in accordance with NIH and American Association of Laboratory Animal Care Regulations. All mouse procedures and experiments for this study were approved by the Columbia University Medical Center Institutional Animal Care and Use Committee Regulations (New York, NY). Ten to 15 mice per treatment group were used in tumor outgrowth studies; prior data in this model suggest these numbers provide a 90% power and a 5% significance level in terms of detecting differences in tumor volume.

On day 0, mice were injected subcutaneously with 5 × 10⁵ RENCA cells in the right flank. On day +12, animals with well-established tumors measuring 25–50 mm² were treated with antibody therapeutics every 3 days × two doses as indicated. Treatments were given as single agents or in combination, with the following regimens for each drug: anti-PD-1 (BioXCell, catalog no., #BE0146, Clone RMP1-14) 200 μg every 72 hours by intraperitoneal injection, Anti-IL1β (BioXCell, catalog no., #BE0246, Clone B122) 200 μg every 72 hours by intraperitoneal injection, and Cabozantinib (Selleckchem, catalog no., #S1119, XL184, dissolved in 2% DMSO + 30% PEG + 5% Tween80 + ddH₂0) at 3 mg/kg/mouse every 24 hours by oral gavage. Control antibodies included polyclonal Armenian hamster IgG (BioXCell, catalog no., #BE0091) at 200 μg every 72 hours by intraperitoneal injection or rat IgG2a isotype control at 200 μg every 72 hours by intraperitoneal injection. Vehicle-treated mice in the cabozantinib group received 20 μL of dilution buffer without cabozantinib. To minimize the number of mice utilized, the anti-IL1β- and vehicle-treated mice were used as controls for experiments with both anti-PD-1 (Figs. 1 and 2) and cabozantinib (Figs. 3 and 4). On day +18 mice were sacrificed and spleen, tumor draining lymph node, and tumor were isolated as described previously (21, 22). Experiments in Figs. 1 and 3 were replicated at least three times. To minimize the number of mice used, vehicle and anti-IL1β mice were used as controls performed simultaneously for the data in Figs. 1 and 3.

Tumors were harvested and dissociated using the Miltenyi Murine Tumor Dissociation Kit (catalog no., #130-096-730) per the manufacturer’s protocol with a Gentlemacs Dissociator (Miltenyi Biotec). Single-cell suspensions were washed with PBS and incubated with ACK lysis buffer (3 cc for 3 minutes) and quenched with 45 cc PBS. To limit instrument and reagent use, 25% of each tumor lysate was used for subsequent flow cytometry experiments. Cells were incubated with Fc-block (BD Biosciences, catalog no., #553141) and then stained with Near-IR Live dead and T-cell or myeloid makers. For a myeloid cell panel, cells were stained with anti-Ly6G (BV421 anti-Ly6G, Clone 1A8, BD Biosciences, catalog no., #562737), anti-CD3 (BV786 anti-CD3, Clone 17A2, BD Biosciences, catalog no., #564010), anti-CD45 (BV510 anti-CD45, Clone 30-F11, BD Biosciences, catalog no., #563891), anti-MHC II (PE anti-MHC II, Clone M5/114.15.2, eBioscience, catalog no., #12-5321-82), anti-F4/80 (PE/Cy7, Clone BM8, BioLegend, catalog no., #123114), anti-Ly6C (APC anti-Ly6C, Clone HK1.4, BioLegend, catalog no., #128016), and anti-CD11b (AF700 anti-CD11b, Clone M1/70, BioLegend, catalog no., #101222). For a T-cell panel, cells were stained with anti-CD45 (BV510 anti-CD45, Clone 30-F11, BD Biosciences, catalog no., #563891), anti-CD4 (BV650 anti-CD4, Clone RM4-5, BioLegend, catalog no., #00555), anti-CD8 (BV711 anti-CD8, Clone 53-6.7, BioLegend, catalog no., #100759), anti-LAG-3 (PE eBioscience, Clone C9B7W, catalog no., #12-2231-83), anti-PD-1 [PE/Cy7 anti-CD279 (anti-PD-1), BioLegend, Clone 29F.1A12, catalog no., #135216], anti-TIM-3 [APC anti-CD366 (TIM-3), BioLegend, Clone RMT3-23, catalog no., #119706], and anti-CD11b (AF700 anti-CD11b, Clone M1/70, BioLegend, catalog no., #101222). For intracellular staining, cells were fixed and permeabilized using the Invitrogen eBioscience Foxp3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific, catalog no., #00-5523-00) per the manufacturer’s protocol. For a T-cell panel, single-cell suspensions were stained with PE anti-FOXP3 (Clone FJK-16s, eBioscience/Thermo Fisher Scientific) and AlexaFluor647 anti-CTLA-4 (Clone L3D10, BioLegend) per the manufacturer’s protocol.

Tumors were harvested, dissociated, and washed for flow cytometry experiments. Cells were stained with Near-IR Live/Dead (Thermo Fisher Scientific, catalog no., #L10119) and BV650 anti-CD45 (BioLegend, catalog no., #103151, Clone 30-F11) and CD45⁺ cells isolated by FACS using a BD Influx Instrument (BD Biosciences). RNA was isolated using Qiagen RNAeasy Mini Kit (catalog no., #74134, Qiagen). RNA sequencing (RNA-seq) was performed in the Genome Analysis Core at Mayo Clinic (Rochester, MN). Sequence reads were aligned to the mouse reference genome (23).

Tumors were excised, snap-frozen in liquid nitrogen, resuspended in 500 μL MSD TRIS Lysis Buffer (Mesoscale Discover, catalog no., #R60TX-3), and pulverized using stainless Steel Beads (Qiagen, catalog no., #69989) in a Tissue-Lyser II (Qiagen) set at 2 × 30 seconds at 20 Hz. Samples were centrifuged at 14,000 rpm for 10 minutes at 4°C. Supernatant was transferred and stored at −80°C in aliquots. Protein was quantified using the Bradford assay, and 250 μg of total protein was used in triplicate for quantification of cytokines using the Mesoscale MESO QuickPlex SQ 120 Instrument (Mesoscale). For these experiments, the U-Plex 10-Plex kit was used; this includes the cytokines: IFNγ, IL1β, IL2, IL4, IL5, IL6, IL10, IL12p70, KC/GRO, and TNFα. To conserve animals, the vehicle- and anti-IL1β–treated sorted immune cells were from the same animals in the experiments detailed in Figs. 2 and 4.

Single-cell suspensions were generated as for flow cytometry using the Miltenyi GentleMACS per the manufacturer’s instructions from tumors treated with vehicle, cabozantinib, or cabozantinib + anti-IL1β (n = 3/treatment group). As described above, CD45⁺ cells were isolated by FACS and loaded as a gel emulsion into the Chromium 10x platform per the manufacturer’s instructions (10x Genomics). Unique molecular identifier (UMI) counts per gene for each cell were obtained on a sample-by-sample basis using the 10X cell ranger pipeline, and then merged for downstream analysis using Seurat v3. Cells were quality control filtered to exclude those with coverage of less than 500 genes, and those where mitochondrial genes representing more than 5% of total UMI count. UMI counts for high-quality cells were then log-normalized and scaled. Clustering was performed using the Louvain algorithm, with top genes per cluster identified by combined likelihood ratio test for single-cell gene expression (24). For ease of cluster phenotyping, the top gene sets were separately filtered to include only statistically significant cluster-specific immune-relevant genes enumerated by the I/O 360 NanoString Platform (NanoString Technologies; ref. 25).

Single-cell suspensions were stained with extracellular antibodies as shown in Supplementary Table S2. Cells were then washed, fixed, and permeabilized using the Invitrogen eBioscience Foxp3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific, catalog no., #00-5523-00) per the manufacturer’s protocol. Cells were next stained with antibodies to intracellular markers. Single-stain controls were performed using UltraComp eBeads (Thermo Fisher Scientific, catalog no., #01-2222-41) and samples were unmixed using CyTEK Aurora Software (Cytek Corporation). Sample files were cleaned, downsampled (5 × 10⁴ live CD45⁺ cells/sample), and concatenated using FlowJo Software (v10, Becton Dickinson). Nonlinear dimensionality reduction was performed on concatenated samples using either the t-distributed Stochastic Neighbor Embedding (tSNE) implementation within FlowJo v10 or visual interactive SNE (viSNE) within Cytobank (relevant parameters: 5,000 iterations, k = 50). Unbiased clustering was performed using the PhenoGraph and FlowSOM Plug-ins in FlowJo, and further downstream data processing, heatmap generation, and statistical analyses were performed in FlowJo, Excel 2016, and GraphPad Prism 8.

To evaluate the antitumor activity of anti-IL1β, we used late, well-established subcutaneous RENCA tumors, as RENCA represents one of the few syngeneic RCC models available. On day +12 after implantation, when tumors were palpable at approximately 25–50 mm³, animals were treated with anti-IL1β alone or in combination with anti-PD-1 (Fig. 1A). Of note, the RENCA tumor model is generally moderately resistant to anti-PD-1 monotherapy (26). We found that anti-IL1β monotherapy significantly delayed tumor growth, and that combining anti-IL1β with anti-PD-1 further delayed tumor progression (Fig. 1B and C).

To understand the immunologic effects on both lymphocytes and myeloid cells in anti-IL1β–treated tumors, we quantified immune cell subsets by flow cytometry. Immunophenotyping showed that blocking IL1β did not significantly alter the percentages of CD8, CD4, or Treg infiltrates in the TME (Fig. 1D). To analyze the phenotype of these T-cell populations, we used high-dimensional spectral cytometry; those data are shown in Fig. 6 and Supplementary Fig. S3, and are discussed below.

MDSCs constitute a major component of the immunosuppressive TME and are increased in number in established RENCA tumors (27). In contrast to its effects on T cells, anti-IL1β significantly decreased infiltration of granulocytic cells as defined by expression of CD11b, Ly6G^hi, and Ly6C^int; this was also noted in the context of anti-IL1β plus anti-PD-1 combination treatment (Fig. 1E). These data suggest that IL1β blockade may block the expansion or recruitment of immunosuppressive PMN-MDSCs to the TME. However, it should be noted that these markers cannot definitively differentiate between neutrophils and PMN-MDSCs (28, 29). In contrast, monocytic-MDSCs within the TME, as defined as CD11b⁺ Ly6C^hi Ly6G⁻ and MHC-II⁺, were relatively unchanged by monotherapy or combination anti-IL1β plus anti-PD-1 treatment. Quantification of TAMs showed that IL1β blockade skewed polarization toward an M1-like phenotype as broadly defined by class II MHC expression. Furthermore, anti-IL1β plus anti-PD-1 combination therapy increased M1-like TAM infiltration relative to either monotherapy. To more deeply analyze the phenotype of these myeloid populations, we used high-dimensional spectral cytometry; those data are shown and discussed in Fig. 6 below. Taken together, these data suggest that IL1β blockade modulates PMN-MDSCs and promotes M1 TAM prevalence in the TME, and that these effects are potentiated in the presence of anti-PD-1.

On the basis of the above results, we hypothesized that IL1β blockade might drive distinct gene expression profiles within myeloid and lymphoid cell populations. To test that hypothesis, we isolated CD45⁺ tumor-infiltrating cells using FACS, and performed gene expression profiling by RNA-seq on those sorted populations. We found that anti-IL1β monotherapy led to decreased expression of several cytokine genes, including IL1β and IL6 (Fig. 2A). Although there is no direct murine homolog to human IL8, Cxcl1, Cxcl2, Cxcl12, MIP-1alpha, and KC/GRO are murine surrogates with homology to human CXCL8, the gene encoding the IL8 protein, and have been proposed as potential surrogates of murine IL8 (30–33). Treatment with anti-IL1β led to decreased gene expression of Cxcl2, Cxcl12, and KC/GRO (Fig. 2A), suggesting that IL1β may lay upstream of these surrogates of IL8; that is, IL1β expression may drive CXCL8 expression, ultimately resulting in PMN-MDSC recruitment to the TME (34). Conversely, treatment with anti-PD-1 monotherapy was associated with relative upregulation of Cxcl2, Cxcl12, and IL6. To evaluate potential changes in genes involved in angiogenesis, we performed gene set enrichment analysis (GSEA) using a well-described angiogenesis gene signature from the gene ontology database (refs. 35, 36; Supplementary Fig. S1). These analyses showed that anti-IL1β blockade, alone or in combination with anti-PD-1, demonstrated statistically significant decreases angiogenesis pathway transcripts.

To validate treatment-associated changes in cytokine levels at the protein level, we used the Meso Scale Discovery (MSD) multiplex cytokine detection platform on tumor lysates from mice treated as in Fig 2A, and found that combination treatment with anti-IL1β plus anti-PD-1 decreased intratumoral IL1β protein expression (Fig. 2B), consistent with the message-level data. In addition, anti-IL1β plus anti-PD-1 combination treatment decreased intratumoral levels of KC/GRO, a potential homolog of human IL8, and was associated with decreased IL10 levels. There was a trend toward increased intratumoral IFNγ and TNFα with anti-IL1β plus anti-PD-1 combination therapy, but these changes did not reach statistical significance. Taken together, these data also suggest that IL1β may be an upstream driver of multiple suppressive cytokines in RCC.

Given the well-established efficacy of VEGF-TKI treatment in RCC (37), we tested whether combining anti-IL1β with a VEGF-TKI would have additive or potentially synergistic antitumor effects, and whether these effects were analogous to those observed in immunotherapy combination studies above. For these studies we chose cabozantinib, which is commonly used in second-line treatment of RCC. Our choice of cabozantinib was further driven by its activity as a relatively broad-spectrum TKI, with activity against c-MET, RET, KIT, AXL, and FLT3 in addition to VEGFR2 (38). The treatment scheme for these studies is shown in Fig. 3A. We found that both anti-IL1β and cabozantinib delayed tumor outgrowth, with further inhibition noted in animals treated with anti-IL1β plus cabozantinib combination therapy (Fig. 3A and B). We also tested cabozantinib, anti-IL1β, and combination cabozantinib/anti-IL1β treatment in the MC38 colorectal cancer model, and did not observe statistically significant changes in tumor outgrowth in this model, likely reflecting differences in the baseline myeloid infiltrate in this murine model relative to RENCA (Supplementary Fig. S9A and S9B).

We next quantified key immune cell populations in the TME by using flow cytometry. While no statistically significant changes were noted in CD8, CD4, or Treg infiltration across treatment groups, a nonsignificant trend toward decreased intratumoral Tregs was observed with either cabozantinib or anti-IL1β plus cabozantinib combination treatment (Fig. 3D). Unlike anti-IL1β, treatment with cabozantinib did not alter intratumoral PMN-MDSC infiltration (Fig. 3E), however, the reduction in PMN-MDSCs mediated by anti-IL1β monotherapy was maintained in the presence of cabozantinib, and was similar to that noted with combination anti-IL1β plus anti-PD-1 therapy (Supplementary Fig. S2C). Quantification of monocytic-MDSCs showed no differences in intratumoral predominance across treatment groups. Combining cabozantinib with anti-IL1β increased M1-like TAMs within the TME; importantly this effect was more pronounced than observed with anti-IL1β plus anti-PD-1 (Supplementary Fig. S2D). Deep immune profiling was also performed in the MC38 treatment model, there we found less profound changes in the myeloid and lymphoid compartment of the TME with cabozantinib and anti-IL1β therapy (Supplementary Fig. S9C–S9G). Taken together, these data support the notion that in addition to anti-PD-1, anti-IL1β also has antitumor activity in combination with cabozantinib in the RENCA model, and that the anti-IL1β plus cabozantinib combination may preferentially modulate TAMs in tumor models with a higher degree of myeloid infiltration.

Using the methods outlined in Fig. 2, we performed RNA-seq of sorted CD45⁺ cells to more broadly understand the immune effects of anti-IL1β plus cabozantinib combination treatment. The effects of anti-ILβ therapy were similar to those observed previously (Fig. 2A); while cabozantinib monotherapy appeared to increase message-level expression of numerous cytokines, including IL1β, an opposite effect to that seen with anti-PD-1. Expression of several other cytokines appeared to be increased at the message level by cabozantinib, these included IL6, IL10, CXCL1, CXCL2 (a potential murine IL8 homolog), and Cxcr2, a gene encoding the receptor for IL8 in humans (Fig. 4A). Conversely, cabozantinib monotherapy decreased message-level expression level of Ly6C, which is associated with PMN-MDSCs. Interestingly, the proimmunogenic effects of anti-IL1B dominated in the combination group of anti-IL1β plus cabozantinib treatment, where we observed decreased expression of IL1β, IL6, Cxcl1, and CXCL2. GSEA to assess the relative angiogeneic effects of cabozantinib and anti-IL1β did not show statistically significant differences in key angiogenesis genes (Supplementary Fig. S1E).

We further interrogated these effects at the protein level, using the MSD discovery platform, as in Fig. 2B. At the protein level, cabozantinib treatment did not significantly increase IL1β intratumoral protein expression (Fig. 4B), suggesting a possible disconnect between message and protein levels. However, anti-IL1β plus cabozantinib significantly decreased IL1β levels, as well as levels of the IL8 homolog, KC/GRO. Consistent with the data in Fig. 2B, IFNγ levels were not significantly increased by IL1β monotherapy; and the increased IFNγ noted with anti-IL1β plus anti-PD-1 was not observed with anti-IL1β plus cabozantinib. Taken together, these data suggest that anti-IL1β has significant antitumor effects in several combination regimens, and that those effects may be driven by disparate mechanisms.

To develop a more comprehensive understanding of the broader effects of these agents on the TME, we performed single-cell RNA-seq on CD45⁺ cells sorted from RENCA tumors treated with vehicle, cabozantinib, and cabozantinib plus anti-IL1β. Sorted cells were sequenced to an average read depth of >1,200 transcripts per cell with the aim of defining cellular phenotypes. Clustering using Seurat V3.0 identified 13 distinct clusters (Fig. 5A; Supplementary Figs. S4–S6), of these clusters 1, 4, 8, and 12 demonstrated significant changes in prevalence between cabozantinib versus the anti-IL1β plus cabozantinib groups (Fig. 5B). Evaluation of the most highly upregulated genes in clusters 1–13 identified two clusters corresponding to T cells (cluster 6 and 8) and 11 clusters corresponding to cells of the myeloid lineage (Fig. 5C). Clusters 0–5 had relatively high expression of ADGRE consistent with intratumoral TAM. Clusters 9–13 defined granulocytic, monocytic, natural killer cells, and B-cell populations as indicated. Inspection of the most highly expressed genes within each cluster suggested that clusters 1, 4, and 12 correlate with intratumoral TAMs, and cluster 8 with T cells (Supplementary Fig. S6). To more deeply interrogate expression changes within each cluster, the top 15 genes from clusters 1, 4, 8, and 12 were determined (Fig. 5C; Supplementary Fig. S7). These data show that treatment with cabozantinib resulted in upregulation of complement genes, including C1qa and C1qb, and the macrophage-specific gene, TREM2, in clusters 1 and 4. Within cluster 12, the combination of cabozantinib and anti-IL1β showed decreased expression of Cxcl1, Cxcl2, and IL1β, consistent with bulk RNA-seq of sorted immune cells (Fig. 4). Taken together, these data show that the combination of cabozantinib and anti-IL1β promotes myeloid remodeling across multiple intratumoral populations.

To better understand which immune cell populations in the TME are modulated anti-IL1β monotherapy and combination therapies (and how they correlate with treatment effects), we optimized a 24-parameter flow cytometry panel to facilitate simultaneous myeloid and T-cell profiling (Supplementary Table S2). Using this panel, we analyzed treated tumors, and then performed unbiased dimensionality reduction using the tSNE algorithm (Fig. 6A). These data broadly defined myeloid and lymphoid populations (Fig. 6B and F). We next performed a second, dependent tSNE analysis of each compartment along with unbiased clustering using the PhenoGraph and FlowSOM algorithms. These data revealed 21 distinct myeloid cell clusters (Fig. 6C) and 21 lymphoid clusters (Fig. 6G) across treatment groups (Supplementary Table S1). The relative mean fluorescence intensity (MFI) for each protein marker for myeloid (Fig. 6C) and lymphoid markers (Fig. 6G) defines each specific cluster. We next normalized the frequency of each cluster to its prevalence in vehicle-treated tumors, to determine which cell populations were affected by each treatment in both the myeloid (Fig. 6D) and lymphoid compartment (Fig. 6H). To identify statistically significant changes between cluster frequency and treatment group, two-way ANOVA of selected treatment comparisons was performed (Fig. 6E).

We first sought to identify anti-IL1β–driven effects on intratumoral myeloid cells (Fig. 6D). Treatment with anti-IL1β was associated with a decrease in Ly6G⁺, Ly6C⁻, CD11b dim cells (clusters 1 and 3), consistent with a granulocytic or PMN-MDSC lineage. Anti-IL1β therapy also increased M1-like TAMs, defined as MHC I high, iNOS high cells (clusters 5 and 15). Similarly, treatment with anti-PD-1 resulted in a decrease in PMN-MDSCs (clusters 1 and 3). Cabozantinib monotherapy decreased cluster 1 PMN-MDSCs and led to a >2-fold log increase in M1-like TAMs in clusters 10, 17, and 20 (Fig. 6D). These data support that anti-IL1β, anti-PD-1, and cabozantinib exert significant changes within the myeloid compartment, and that cabozantinib monotherapy induced intratumoral M1-like TAMs.

We next performed similar immunophenotyping for combination therapy with anti-IL1β + anti-PD-1 relative to vehicle alone (Fig. 6E). With combination treatment, statistically significant decreases in PMN-MDSCs in clusters 1, 3, and 6 (P < 0.05; Fig. 6E, third row) and increases in the prevalence of M1-like TAMs (clusters 10, 14, and 15) were noted by two-way ANOVA (P < 0.05; Fig. 6E, third row). The phenotypic changes within the macrophage and granulocytic compartment were more pronounced with combination therapy than with either anti-IL1β or anti-PD-1 monotherapy alone.

Perhaps most striking was the degree to which combination anti-IL1β and cabozantinib increased M1-like TAM predominance within the TME, that is, combination anti-IL1β plus cabozantinib significantly increased M1-like TAMs (clusters 10 and 14; P < 0.001 and P < 0.01, respectively) and decreased M2-like TAMs (MHC II low, F4/80 DIM, cluster 18; P < 0.01) relative to cabozantinib therapy alone (Fig. 6E). Statistical analysis of each cluster frequency between all treatment groups using two-way ANOVA with Tukey correction for multiple comparisons confirmed that observed changes in PMN-MDSCs (cluster 1) and M1-like TAMs (cluster 10) in the combination cabozantinib and anti-IL1β group were significantly different as relative to vehicle-treated tumors and monotherapy-treated controls (Fig. 6E).

On the basis of the observed phenotypic changes in the myeloid compartment, we also performed phenotyping of T lymphocytes from tumors treated with anti-IL1β, anti-PD-1, or cabozantinib monotherapy (Fig. 6G, H, and E, at right). Deeper immune profiling of the lymphocyte compartment identified less pronounced phenotypic changes within CD4 and CD8 immune cell subsets as relative to the myeloid compartment, but identified additional differences that were not apparent in prior flow analysis using more basic panels (Figs. 1 and 3). We observed increases in CD8⁺, CCR7⁺, CD62L-low cells (cluster 20) and CD4⁺, CCR7⁺, CD62L-low T cells (cluster 21) with anti-IL1β, anti-PD-1, and cabozantinib monotherapy relative to vehicle, respectively.

We then performed similar phenotypic analyses with combination anti-PD-1 and anti-IL1β–treated tumors. Similar to observations with anti-PD-1 monotherapy and anti-IL1β monotherapy, increases in the frequency of cluster 20 and 21 T cells were noted. Analysis of combination cabozantinib and anti-IL1β therapy showed a statistically significant increase in cluster 21 CD4⁺, CCR7⁺, CD62L⁺ T cells, suggesting that combination cabozantinib and anti-IL1β therapy promoted the most significant CD4⁺ effector T-cell response in this model. Similar analyses from lymphocytes within the tumor draining lymph node showed an increase in naïve T cells within the tumor draining lymph node with combination cabozantinib and anti-IL1β (CD62L high, CD44 low, clusters 7 and 25; Supplementary Fig. S3). Together these data support that anti-IL1β monotherapy exerts relatively minor changes within the T-cell compartment.

Taken together, these profiling data support the conclusion that anti-IL1β promotes M1-like antitumor TAMs (clusters 10 and 18), and decreases PMN-MDSCs within the TME (cluster 1). While these effects were potentiated to some effect in combination with anti-PD-1, combination treatment with cabozantinib and anti-IL1β therapy promoted more robust M1-like TAMs within the TME. Overall, in the RENCA model, the antitumor effects of cabozantinib + anti-IL1β correlate with effects on MDSCs and TAMs, along with an increase in CD4+ CCR7⁺ effector T cells.

Chronic inflammation is a recognized promoter of tumor invasiveness and progression, dating back to early observations from Virchow in the 1800s (39). Dampening chronic inflammation while leaving T-cell functionality intact is a formidable challenge in an era where immune checkpoint blockade is a pillar of cancer treatment for several tumor types (40). We found that anti-IL1β delayed tumor outgrowth in an established RCC model. The antitumor effects of IL1β blockade appear to be additive with immune checkpoint blockade using anti-PD-1 in the RENCA model, which is relatively resistant to anti-PD-1 therapy. These data are broadly consistent with results in immune-responsive tumor models evaluating IL1β blockade (15). In combination with anti-PD-1, the immunogenic effects of IL1β blockade appear to primarily affect the myeloid compartment, suggesting that IL1β and anti-PD-1 might have nonredundant and complementary mechanisms of action, and that targeting IL1β might be a candidate approach to overcome adaptive immune resistance.

Clinically, a majority of patients with metastatic RCC receive anti-PD-1 or anti-PD-L1 immunotherapy in combination with either anti-CTLA-4 or the VEGF-TKI, axitinib, as first-line therapy. We showed that the myeloid effects of the VEGF-TKI, cabozantinib, are augmented by the addition of anti-IL1β through targeting of both immunosuppressive MDSCs and reprogramming of TAMs. Combining cabozantinib with anti-IL1β decreased tumor growth, with an associated decrease in PMN-MDSC infiltration. Immunophenotypic analysis by flow cytometry and single-cell RNA-seq showed that the immunogenic effects of combination cabozantinib and anti-IL1β treatment in this tumor model are also associated with significant changes in the myeloid compartment, manifest as a decrease in intratumoral PMN-MDSCs. Furthermore, cabozantinib in combination with anti-IL1β decreased tumor outgrowth and led a significant increase in M1-like TAMs within the TME. To our knowledge, this is the first study to illustrate potential synergy between anti-IL1β blockade and a VEGF-targeted TKI.

We also found that the intratumoral cytokine profile within the TME is shifted with anti-IL1β blockade. Systemic therapy with anti-IL1β decreased transcript level expression of several cytokines, including IL6, CXCL2, CXCL12, and KC/GRO. These transcriptional profiles have consequences for the intratumoral cytokine milieu, as anti-IL1β alone decreased protein levels of IL6 and homologs of IL8 (KC/GRO). These data are consistent with work from several groups illustrating that IL1β promotes increased IL6 expression, but to our knowledge this is one of the first reports to suggest that IL1 is an upstream mediator of homologues of IL8 (41, 42).

Furthermore, combination therapy with either anti-PD-1 or cabozantinib augmented the immunogenic effects of IL1β blockade on intratumoral cytokines. The addition of anti-PD-1 to anti-IL1β further decreased transcript level expression of several downstream intratumoral cytokines that are upregulated with anti-PD-1 monotherapy, including IL10 and TNFα, however, decreases of in IL10 and TNFα were not reflected at the protein level due to variation in protein quantification. Similar effects on intratumoral cytokine profiles were observed with combination cabozantinib and anti-IL1β with statistically significant decreases in IL6 and the IL8 homolog, KC/GRO. Gene expression analysis was also performed to assess the relative effects of IL1β on angiogenesis based on the known contribution of IL1β to the vascular endothelium (Supplementary Fig. S1). Treatment with anti-IL1β alone or in combination with anti-PD-1 led to statistically significant decreases in key angiogenesis genes within immune cells (Supplementary Fig. S1A and S1C), consistent with observations from protein-level cytokine quantification within the TME. In contrast, combination cabozantinib and anti-IL1β did not significantly alter the angiogenesis gene signature relative to cabozantinib monotherapy (P = 0.19). In terms of the leading edge genes (Fig. 1B and D), comparison of anti-IL1β relative to vehicle and anti-IL1β/anti-PD-1 versus anti-PD-1 showed downregulation of several angiogenic genes. These data suggest that IL1β may also dampen angiogenic effects as monotherapy or when combined with immune checkpoint blockade. Overall, these data show that anti-IL1β has differential effects on angiogenesis genes that depend on the context for rationale combination therapy, and support the hypothesis that IL1β blockade has immunogenic effects in combination with cabozantinib. This observation suggests a potential mechanism for additive antitumor effects with either immune checkpoint blockade or a TKI. To this end, several downstream targets of IL1β, including IL6 (43, 44), IL8 (45), and TNFα (46), are being targeted in clinical trials with the goal of augmenting the anti-PD-1 immunotherapy response. Together, these findings support the notion that IL1β may be an upstream regulator of adaptive myeloid resistance, and that IL1β blockade may have the potential to reverse immunosuppressive cytokine programs that prevent successful antitumor immunity.

We also studied IL1β-mediated changes within the myeloid cell compartment in the RENCA model by performing deep immune profiling of sorted populations by both single-cell RNA-seq and flow cytometry. Using single-cell RNA-seq we delineated 13 unique clusters of immune cells. Treatment with cabozantinib or cabozantinib and anti-IL1β showed changes within the frequency of three distinct myeloid clusters (clusters 1, 4, and 12) consistent with earlier flow cytometry data, suggesting that both treatments remodel the myeloid compartment. Using spectral flow cytometry, we identified 21 distinct clusters of myeloid and lymphoid cells. These analyses also identified a potentially important population of CD4⁺, CCR7⁺, CD62-low effector T cells, which were most prevalent within the TME following cabozantinib plus anti-IL1β combination therapy. The presence of CCR7⁺ T cells within the TME is consistent with skewing of the T-cell compartment toward central memory phenotypes, which in some contexts are associated with antitumor response to immune checkpoint blockade (47, 48). Furthermore, and consistent with recent data from a murine model of combined CTLA-4/PD-1 blockade (49), as well as deep phenotyping of the human RCC TME (50), our immune cell clustering data show that the populations in the RENCA TME are more complex than those previously defined using markers of canonical immune subsets.

Overall, our results are consistent with prior observations that anti-IL1β has antitumor properties, and that IL1β blockade is potentially additive or synergistic with immune checkpoint blockade (15). Consistent with observations from earlier studies in which overexpression of IL1β increased MDSC infiltration within mammary carcinomas (51), we observed decreased intratumoral PMN-MDSCs following anti-IL1β treatment. In other studies, evaluating immune checkpoint blockade with anti-IL1β, tumor eradication was, in part, dependent on CD8 cells, as CD8 T-cell depletion rendered IL1β knockout ineffective (15). While we found less pronounced differences in CD8 T cells, this discrepancy is likely model dependent because other studies utilized the 4T1 tumor model, which is fairly responsive to anti-PD-1 therapy. Consistent with this, CD8 T-cell depletion in the RENCA tumor model had little effect on the efficacy of combination cabozantinib and anti-IL1β therapy (Supplementary Fig. S8). Our data also support the hypothesis that targeting IL1β in established tumors results in antitumor M1-like TAM infiltrates and a decrease in intratumoral PMN-MDSCs. It remains an open question whether anti-IL1β decreases migration of PMN-MDSCs to the TME, or whether the decreased prevalence noted by us and others reflects decreased local polarization.

Several studies have attempted to address the duality of IL1β (52) in promoting tumor progression in certain models, whereas in other models, IL1β expression for cancer is required for tumor eradication (52–54). When administered at supraphysiologic levels, IL1β has the potential to induce both T_H1 and T_H17 responses with antitumor properties (52). Indeed, tumor regression is observed with exogenous injection of IL1 in some tumor models (47). In contrast, NLRP3 inflammasome production of IL1 is correlated with an increased tumor burden and tumor progression in colorectal cancer (55). This finding is supported by other data illustrating that IL1β signaling in epithelial cells drives stemness and tumor progression (56). For instance, in prostate cancer models, transient upregulation of IL1 promotes prostatic proliferative inflammatory atrophy and leads to aggressive prostate adenocarcinoma (57). The commonality in all of the above studies is that IL1β appears to interact predominantly with the myeloid compartment, with less pronounced effects on T cells, consistent with the findings described here (58).

In early-tumor treatment models, the addition of anti-IL1β to anti-PD-1 resulted in additive effects on tumor outgrowth and improved T-cell–mediated tumor killing (15). One strength of our work is that we studied later, well-established tumors, and found that myeloid compartment remodeling is achievable. This suggests that even earlier treatment might be preferable; in that regard an ongoing clinical study is evaluating combination therapy with spartalizumab (anti-PD-1) plus canakinumab (anti-IL1β) prior to nephrectomy for patients with high-risk RCC with the goal of decreasing intratumoral MDSCs and decreasing the risk of relapse (NCT04028245). Consistent with this hypothesis, a query of IL1β gene expression through The Cancer Genome Atlas illustrates that high IL1β RNA expression is associated with worse overall survival (Supplementary Fig. S10). One potentially unique implication of our data is that IL1β might be upstream of IL8 homologs, this is important because elevated IL8 levels are associated with lack of response to anti-PD-1 (59). Clinical studies blocking IL8 in combination with anti-PD-1 are ongoing (NCT03400332, NCT03689699, and NCT04123379; refs. 45, 60).

A potential limitation of our study is that RENCA as tumor model lacks the mutations commonly associated with clear-cell RCC including VHL loss, PBRM-1, BAP-1, and SETD2. Despite the shortcomings of RENCA as a genetic model of RCC, this tumor model contains a robust myeloid cell infiltrate, which may reflect more aggressive human RCC phenotypes. For instance, transcriptomic profiling of human RCC from the COMPARZ trial clearly showed worse overall survival for patients with angiogenesis^lo, macrophage^hi gene signatures, which was largely independent of mutational status in VHL, PBRM1, and BAP1 (61). Hence, the dense myeloid infiltrate observed in RENCA may be similarly reflective of immunosuppressive TMEs associated with treatment refractory RCCs. To assess the relative contribution of combination cabozantinib and anti-IL1β on tumor growth in a tumor type with less TAM infiltrate, we also evaluated this combination in the MC38 colorectal cancer model (Supplementary Fig. S9). We did not note statistically significant changes in tumor outgrowth with cabozantinib, anti-IL1β, or combination treatment in this model. These data suggest that combination of anti-IL1β with a VEGF-targeted TKI is not likely to be a universal immune combination therapy in human RCC, and may require a level of patient selection. Therapeutics that rely on myeloid remodeling may be central to future RCC treatment selection, particularly as novel biomarkers are tested that reflect the relative degree of macrophage infiltration.

Finally, as IL1β is also known to promote angiogenesis (12, 62), it was interesting to find that treating established tumors with anti-IL1β plus cabozantinib led to a marked reduction in tumor growth and increased M1-TAM infiltration. In summary, our data inform future clinical trial development, by supporting the notion that anti-IL1β might have activity in RCC, either in combination with anti-PD-1 or in combination with the TKI, cabozantinib.

D.H. Aggen reports other from Boehringer Ingelheim outside the submitted work, as well as a patent for University of Illinois issued, licensed, and with royalties paid from AbbVie. W. Mao reports other from Kite Pharma outside the submitted work. J.E. Hawley reports grants from ASCO and NIH, other from Regeneron Pharmaceuticals, and nonfinancial support and other from Dendreon Pharmaceuticals outside the submitted work. M.C. Dallos reports personal fees from Clovis and Bayer outside the submitted work. C.G. Drake reports personal fees from AZ Medimmune, Bayer, BMS, Compugen, Ferring, F-Star, Genocea, Harpoon, Janssen, Kleo, Merck, Merck-Serono, Pfizer, Pierre Fabre, Roche/Genentech, Shattuck Labs, Tizona, UroGen, and Werewolf outside the submitted work, as well as a patent for BMS licensed to Johns Hopkins University. No disclosures were reported by the other authors.

D.H. Aggen: Conceptualization, data curation, software, formal analysis, supervision, investigation, writing-original draft, writing-review and editing. C.R. Ager: Conceptualization, data curation, investigation, writing-review and editing. A.Z. Obradovic: Data curation, formal analysis, methodology. N. Chowdhury: Investigation, methodology. A. Ghasemzadeh: Conceptualization, investigation, methodology. W. Mao: Conceptualization, investigation, methodology. M.G. Chaimowitz: Investigation. Z.A. Lopez-Bujanda: Conceptualization, methodology, writing-review and editing. C.S. Spina: Conceptualization, writing-review and editing. J.E. Hawley: Writing-review and editing. M.C. Dallos: Investigation, writing-review and editing. C. Zhang: Formal analysis, writing-review and editing. V. Wang: Conceptualization, investigation, writing-review and editing. H. Li: Formal analysis, writing-review and editing. X.V. Guo: Data curation, methodology, project administration. C.G. Drake: Conceptualization, resources, supervision, investigation, writing-original draft, writing-review and editing.

These studies were supported by the ASCO Conquer Cancer Foundation Young Investigator Award; NIH grants R01 CA127153, 1P50CA58236-15, and P30CA006973; and CUMC institutional funds to C.G. Drake. The authors acknowledge Drs. Siu-Hong Ho and Lu Caisheng of the Flow Cytometry Core Facility at the Herbert Irving Comprehensive Cancer Center.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.