Cabozantinib Enhances Anti-PD1 Activity and Elicits a Neutrophil-Based Immune Response in Hepatocellular Carcinoma

Hepatocellular carcinoma (HCC) is the most common form of primary liver cancer, which is the third leading cause of cancer-related death and a major health problem globally (1, 2). Around 50%–60% of patients will be ultimately exposed to systemic therapies, where the combination of atezolizumab and bevacizumab has become the standard of care in first line. In the phase III IMbrave150 trial, this combination demonstrated significantly better overall survival (OS) and progression-free survival (PFS) versus sorafenib (median OS 19.2 vs. 13.4 months; refs. 3, 4), the multikinase inhibitor which had been the previous standard of care for advanced HCC for the past 12 years (5). This marks the first combination therapy and first treatment regimen involving an immunotherapy to induce a greater survival benefit for patients with HCC compared with existing therapies, although objective responses (OR) were restricted to 30% of patients (3). Single-agent immune checkpoint inhibitors (ICI)—that is, nivolumab and pembrolizumab—had also shown promising results over the past decade, with median OS of 13–16 months reported in patients with advanced HCC (6, 7). However, no ICI monotherapy significantly extended OS in phase III trials. In addition, recent data have revealed that nonviral HCC, particularly nonalcoholic steatohepatitis (NASH)-related HCC, might be less responsive to immunotherapy due to the intratumor accumulation of a population of dysfunctional activated CD8⁺PD1⁺ T cells (8). Thus, there is an unmet need to better understand the complex mechanisms underlying antitumor immunity and explaining response and resistance to immunotherapies, to favor the development of more efficacious therapeutic strategies with immunotherapies (9), including combination therapies.

Given that immunotherapeutic strategies show better results among tumors with an inflammatory microenvironment (10), efforts are being made to enhance the antitumor activity of ICIs by promoting the infiltration or reactivation of immune cells in tumors which lack an inflamed profile. One approach to this is via the combination of ICIs with antiangiogenic agents, in light of the reported immunosuppressive role of proangiogenic molecules (mainly VEGF, angiopoietins, hepatocyte growth factor and the platelet-derived growth factor family; refs. 11, 12). Encouraging results have been obtained for combinations of ICIs plus antiangiogenic therapies in preclinical studies (13, 14), as well as in clinical trials across different cancer types including HCC (3, 15, 16).

Cabozantinib, a tyrosine kinase inhibitor (TKI), proved its efficacy in a phase III trial with patients with advanced HCC who had progressed on sorafenib, and extended median OS to 10.2 months from 8.0 months with placebo (17). Among other kinases, cabozantinib targets VEGFR, AXL, and MET, key factors in pathways promoting angiogenesis, proliferation, and epithelial-to-mesenchymal transition (18, 19). Therefore, the combination of cabozantinib and ICIs may lead to enhanced clinical benefits over monotherapies in HCC, and in fact, the combination of nivolumab + cabozantinib conveyed a median PFS of 5.5 months in patients with advanced HCC, with median OS not reached (20). The phase III study assessing the combination of atezolizumab + cabozantinib (COSMIC-312; ref. 21) reported in the first interim analysis a significant improvement in PFS versus sorafenib (HR: 0.63; 99% CI, 0.44–0.91; P = 0.0012), and a nonsignificant trend in terms of OS impact (22). Studies in experimental models of prostate cancer have also suggested that cabozantinib may have immunomodulatory activity (23); however, this has not yet been explored in HCC.

Our study explored the antitumoral, mechanistic, and immunomodulatory effects of cabozantinib alone and in combination with anti-PD1 in two preclinical HCC mouse models. The combination of cabozantinib with anti-PD1 (i) was associated with greater efficacy than any single treatment alone, (ii) enhanced neutrophil recruitment and neutrophil activation profiles, and (iii) decreased CD8⁺PD1⁺ T and regulatory T cell (Treg) infiltration in the tumor. Finally, human HCC data revealed that tumors with neutrophil-based enrichment are linked with better molecular and clinical features. Therefore, the combination of cabozantinib and anti-PD1 has the potential to bring together the activation of adaptive and innate immune responses against the tumor in patients with HCC.

Immunocompetent murine models of HCC were generated by subcutaneously implanting 5 × 10⁶ Hepa1-6 cells (model 1) or 5 × 10⁶ Hep 53.4 cells (model 2) suspended in Matrigel (50% volume for volume) into the right flank of male 5-week-old C57BL/6 mice (model 1: n = 80, model 2: n = 40). When tumors reached a volume of 200–300 mm³, animals were randomly assigned to receive: (i) placebo: Hamster IgG 10 mg/kg i.p. every 3 days for a total of five doses; (ii) cabozantinib: 30 mg/kg orally daily until the end of the study; (iii) anti-PD1: Hamster anti-murine PD1 mAb J43 (BioXCell) intraperitoneal at 10 mg/kg every 3 days for a total of five doses; or (iv) combination of cabozantinib with anti-PD1. Tumor volume measurements were taken every 2–3 days using bilateral callipers, and tumor response rates were calculated as described in the Supplementary Materials and Methods. In model 1, 6 mice per group were randomly selected and culled on day 14 of treatment for pharmacodynamic analyses. The remaining 14 mice per group were monitored until their tumor volume surpassed 1,500 mm³ or reached day 32 of treatment (Supplementary Fig. S1). Model 2 was used as validation and all animals were culled at day 14. Animal experiments were approved by the corresponding review board. As a further independent model, we also analyzed the RNA sequencing (RNA-seq) data from paired human colorectal cancer patient-derived xenograft (PDX) models treated with cabozantinib (30 mg/kg daily orally) or vehicle (24).

Peripheral blood samples were collected by cardiac puncture under deep terminal anesthesia and subjected to red blood cell lysis. Tumor tissues were minced with forceps and scissors, followed by chemical and mechanical disaggregation. The resulting single-cell suspensions were incubated with antibody conjugates to identify specific populations of myeloid (dendritic cells, macrophages, and neutrophils) and lymphoid (leukocytes, CD4⁺ T cells, CD8⁺ T cells, Tregs, PD1-expressing CD8⁺ T cells, memory T cells, and B cells) immune cells, and stained with a viability dye (Supplementary Tables S1–S3) and in the presence of a Fc receptor blocking reagent. Samples were analyzed on a 5-laser Fortessa flow cytometer (BD Biosciences) using BD FACSDiva software v8.0.

Gene expression profiling of mouse tumor samples was performed using a whole-genome microarray platform (GeneChip Mouse Genome 430 2.0 Array, Affymetrix; GSE174770). Mouse expression data were retrieved, quality checked, and processed through RMA using the R package affy (v.1.62.0). After quality filtering, probe-gene correlates were identified with the R package htmg430pmprobe (v.2.18.0). Subsequently, the mouse gene expression matrix was humanized using the mouse-human orthologs data available in ENSEMBL (retrieved using BioMart) and Mouse Genome Informatics (MGI, informatics.jax.org). Molecular Signature Database gene sets (MSigDB, broadinstitute.org/msigdb) and previously reported gene signatures representing different states of inflammation, HCC subclasses, or distinct immune cell populations (25–28) were tested using enrichment tools implemented in GenePattern [ref. 29; gene set enrichment analysis (GSEA) and single-sample GSEA (ssGSEA)]. In addition, the GenePattern module Comparative Marker Selection (CMS) was used to identify differential gene expression between treated and untreated tumors (FDR < 0.05, FC ≥ 1.5), and the Enrichr tool (30) was used to evaluate enrichment in specific pathways and biological functions among the differentially expressed genes (DEG; Supplementary Tables S5 and S6). We considered as neutrophil-related genes those DEGs classified within biological functions linked to neutrophil activity. A signature reflecting the specific molecular effect of combination therapy (CAP-100) was retrieved by selecting the upregulated genes in the combination arm versus placebo and deducting those genes commonly upregulated between the combination and the monotherapies.

We analyzed the gene expression data of 167 paired HCC tumor and tumor-adjacent liver obtained by our group (25, 31) from untreated, surgically resected fresh-frozen samples (cohort 1: Heptromic, GSE63898). Supplementary Table S7 depicts the clinical features of the subset of 167 patients, and a complete description of the full cohort can be found elsewhere (25, 31). In addition, we used a second cohort of 57 untreated patients with HCC with paired expression data from the tumor and tumor-adjacent liver (cohort 2: Supplementary Table S7, GSE174570). Samples from both cohorts were collected through the International HCC Genomic Consortium under the corresponding Institutional Review Board approval for each center. We also used the survival and expression data of 319 patients with HCC publicly available from The Cancer Genome Atlas (TCGA; ref. 32).

RNA extractions from fresh-frozen samples were performed as reported previously (31), and RNA profiling was conducted using the Human Genome U219 Array Plate (Affymetrix). The processing of transcriptome data (normalization, background correction, and filtering) was carried out as described previously (31). Prediction of mRNA-based signature positivity was performed with GenePattern's nearest template prediction (NTP) module, and gene expression signature enrichment as well as differential gene expression analysis were carried out as with the mouse tumor expression data. K-medoids clustering was used for the unsupervised classification of samples according to the enrichment in gene expression signatures (Supplementary Fig. S9). The optimal number of clusters was determined as the k with the maximum silhouette width.

Statistical analyses were performed using R (version 3.6.1). Specifically, nonparametric tests were used for the comparison of distribution of continuous variables (Wilcoxon or Kruskal–Wallis tests). For the assessment of correlations between two continuous variables, we used Spearman rank correlation coefficient test. Univariate analysis of survival was conducted using the log-rank test, while a multivariate Cox proportional hazards model was conducted to identify independent predictors of survival. In the event of a multiple comparison testing, Benjamini–Hochberg correction was applied unless specified otherwise. In general, adjusted P values <0.05 (two-sided) were considered to be statistically significant and are denoted as follows (unless stated otherwise in specific analyses): *, P < 0.05; **, P < 0.01; ***, P < 0.001. NS indicates not significant, P > 0.05.

Data generated or analyzed by the authors is available in Gene Expression Omnibus at GSE174770, GSE60939, GSE63898, and GSE174570.

Full details of the Materials and Methods are given in the Supplementary Data.

We used subcutaneous murine models of HCC to assess the antitumor efficacy of cabozantinib, anti-PD1, and combination (Supplementary Fig. S1), and evaluated tumor response rate according to RECIST adapted for murine models. In the Hepa1-6 model, the combination of cabozantinib and anti-PD1 most rapidly induced an OR out of all treatment arms, with statistically significant improvements versus cabozantinib alone or anti-PD1 monotherapy, as well as placebo (median time to OR: 12.6 days for combination vs. 20.1 days for cabozantinib monotherapy, with time to OR not reached in the anti-PD1 and placebo arms; P < 0.05; Fig. 1A; Supplementary Fig. S2A). Of note, cabozantinib and anti-PD1 monotherapies also significantly accelerated time to achieve an OR compared with placebo (P < 0.05). In addition, the combination treatment induced the highest OR and complete response rates (CRR; 70% ORR and 15% CRR), followed by cabozantinib (35% ORR and 5% CRR) and anti-PD1 (25% ORR and 15% CRR), and all treatment-induced ORRs were significantly higher as compared with placebo (0% ORR, P < 0.05; Fig. 1B). All treatments also delayed tumor growth versus placebo (P < 0.05), and no adverse clinical signs or loss in starting body weight were detected throughout the study (Supplementary Fig. S2B and S2C). These findings were also recapitulated in the Hep53.4 model (Supplementary Fig. S2D–S2F).

Excised HCC tumors were then evaluated at the molecular and histopathologic level. First, examination of the tumor sections revealed that cabozantinib and combination treatments were associated with significant reductions in viable tumor volume and increases in the percentage of tumor necrosis (Fig. 1C). On the other hand, and confirming the antiangiogenic activity of cabozantinib (18, 19, 24), IHC detection of endothelial cells by CD31 staining revealed that only placebo and anti-PD1 tumors contained marked or moderate levels of vascular endothelial cells, and cabozantinib-treated tumors (monotherapy and combination) were associated with significantly reduced areas of CD31⁺ staining compared with placebo (P < 0.05; Fig. 1D). In addition, only cabozantinib and combination-treated tumors displayed an absence of vessels encapsulating tumor clusters (VETC; Fig. 1D).

Analysis of the tumor whole-genome expression profile revealed the impact of therapies in reducing proliferation and angiogenesis. Regarding proliferation, using the NTP and ssGSEA modules of GenePattern, we observed that placebo-treated tumors were associated with the Chiang proliferation (26) and the Boyault G3 (28) subclasses of HCC, and also displayed increased expression of a wide array of proliferation and DNA repair signatures (Fig. 1E). Furthermore, administration of cabozantinib (monotherapy or combination) significantly impacted angiogenesis by reducing VEGF and endothelial cell signaling and promoting enrichment in molecular signatures of hypoxia (Fig. 1E). Combination was the only treatment associated with a significant reduction in proliferation and DNA repair signatures as compared with placebo, with a similar but nonsignificant trend also observed in cabozantinib monotherapy–treated tumors. Taken together, these data demonstrate that cabozantinib treatment induced antiproliferative effects in the tumors and these were more marked when administered in combination with anti-PD1.

Immune population analysis using flow cytometry, IHC, and transcriptomic data revealed that cabozantinib administration promoted the recruitment of neutrophils in the tumor tissue. Specifically, through flow cytometry, we observed median 2.5- and 1.8-fold increases in the prevalence of tumor-infiltrating neutrophils in cabozantinib and combination treatment arms versus placebo (P < 0.05 vs. placebo and anti-PD1; Fig. 2A). Furthermore, in combination-treated animals, the increased proportion of neutrophils coincided with high ESTIMATE immune cell infiltrate scores (33), suggesting that neutrophils were a prevalent immune subtype in these higher infiltrated tumors (Supplementary Fig. S2G). We further assessed neutrophil intratumor infiltration and activation by IHC using a panel of three markers: Ly6G, which is expressed by neutrophils and granulocytes; myeloperoxidase (MPO), which is expressed by active polymorphonuclear cells including neutrophils; and neutrophil elastase (NE), which is expressed in the lysosomal granules of active neutrophils. Overall, we observed a significantly elevated percentage of stained cells in tumors from the cabozantinib (median 4% for Ly6G, 12% for MPO, 8% for NE) and combination (8.5% for Ly6G, 15% for MPO, 4% for NE) arms versus placebo (1.5% for Ly6G, 4.2% for MPO, 0.5% for NE; P < 0.05; Fig. 2B). These data suggest that active neutrophils are significantly increased in the intratumoral regions of mice treated with combination and cabozantinib, whereas no changes in peritumor neutrophil infiltration were observed across treatment groups (Supplementary Fig. S4A and S4B). These findings were also recapitulated in the Hep53.4 model (Fig. 2B), and when compared with Hep53.4 tumor samples treated with the antiangiogenic lenvatinib ± anti-PD1 also generated by our lab (34), this effect was revealed to be cabozantinib specific (Supplementary Fig. S2H).

To further investigate this observation, we extracted DEGs (FDR < 0.05, fold change > 1.5) among the treatment groups through CMS analysis of tumor transcriptomic data. Combination treatment had the greatest impact reshaping the molecular profile of the tumor, because it was associated with the greatest number of DEGs compared with placebo at 157 genes, followed by anti-PD1 with 86 DEGs, while only seven genes were significantly upregulated by cabozantinib. The 100 genes exclusively upregulated in the combination arm were used to generate the cabozantinib plus anti-PD1 100-gene signature (CAP-100), capturing the specific molecular effect of the combination (Supplementary Table S5). Gene Ontology (GO) enrichment analysis revealed that CAP-100 was significantly associated with 12 biological function GO terms, of which six (50%) were linked to neutrophil activity (P < 0.05; Fig. 2C; Supplementary Table S6). The antitumor neutrophil phenotype in the combination arm was further supported by higher enrichments in gene expression signatures of the N1 neutrophil profile (35), neutrophil activity, and maturation (late neutrotime; ref. 36; Fig. 2D). In parallel, the seven genes upregulated by cabozantinib monotherapy were associated with the chemotaxis and migration of leukocytes, fundamentally neutrophils (5/12 GO terms; 42%), and with innate immunity responses (4/12 GO terms, 34%; Fig. 2C; Supplementary Table S6). On the other hand, the 86 genes upregulated by anti-PD1 monotherapy were linked to broader proinflammatory signaling cascades (17/47 GO terms; 36%) and chemotaxis of a wide variety of leukocyte subpopulations (12/27 GO terms, 26%).

We then validated the molecular effect of cabozantinib using RNA-seq data from colorectal cancer tumors in a PDX model (ref. 24; Supplementary Fig. S3A). GO enrichment analysis of the 244 genes upregulated by cabozantinib versus placebo revealed a positive regulation of neutrophil chemotaxis (Supplementary Fig. S3B; Fig. 2E), and GSEA identified an enrichment in neutrophil-, inflammation-, hypoxia-, and apoptosis-related pathways (Supplementary Fig. S3C).

To further investigate the immunomodulatory roles of cabozantinib and its combination with anti-PD1, we assessed several infiltrating lymphoid and myeloid immune populations along with the tumor transcriptomic profiles. Regarding antitumor immunity features, anti-PD1 and combination induced a higher enrichment in stromal and immune microenvironment components based on expression data (Supplementary Fig. S4D). Flow cytometry analysis revealed a specific reduction in the proportions of the exhausted CD8⁺PD1⁺ T-cell subpopulation induced by cabozantinib and combination (mean 7% and 6% of CD45⁺ cells; P < 0.05 vs. 25% in placebo and 35% in anti-PD1; Fig. 3A), even though no significant differences in the IHC counts of infiltrating CD8⁺ and CD4⁺ T cells were observed (Supplementary Fig. S4E and S4F). In addition, the location of CD8 positivity was mostly intratumoral in combination and anti-PD1 tumors (means of 73% and 60%, respectively; P < 0.05 vs. 26% in placebo; Fig. 3C), suggesting a higher capacity to infiltrate the tumor parenchyma to exert their effector function.

Aligning with these results, transcriptomic analysis revealed that all anti-PD1 and combination-treated tumors recapitulated the HCC Immune Class (25), compared with none in the placebo arm (Fig. 3D). Combination further contributed to the proinflammatory and antitumor immunity effect of anti-PD1 with the highest enrichments in pathways of cytokine and chemokine signaling, T-cell activation and Th1-related responses, B-cell signaling, classically activated M1 macrophages, dendritic cells, antigen presentation, and features of response to nivolumab in melanoma (ref. 37; P < 0.05 vs. placebo; Fig. 3D; Supplementary Fig. S4D). Flow cytometry did not reveal significant changes in the immune infiltrate composition for B cells or dendritic cells, but showed a cabozantinib-related reduction in macrophage percentages (Supplementary Fig. S4G).

In terms of immune suppression features, IHC analysis revealed decreased infiltration of Tregs (FOXP3⁺) in combination and anti-PD1 arms versus placebo (P < 0.05), with a clear trend for cabozantinib (Fig. 3B). In addition, combination and cabozantinib-treated tumors excluded Tregs from intratumoral areas, with mean 1.7% and 3.2% of the total FOXP3⁺ positivity in the IHC sections being intratumoral (P < 0.05 vs. 15% in placebo; Fig. 3C). In parallel, gene expression data revealed that combination treatment also caused a significantly reduced enrichment in TGFβ and β-catenin signaling pathways, associated with immune exclusion features in HCC and resistance to immunotherapies (refs. 25, 38–40; P < 0.05 vs. placebo; Fig. 3D). Furthermore, cabozantinib and combination arms were also linked to the strongest PD-L1 expression in tumor cells and increased CTLA4 staining (Supplementary Fig. S4H), which could potentially be associated with the activation of feedback loops to negatively regulate T-cell activation. Overall, tumor molecular data suggest that combination enhanced antitumor immunity by stimulating both innate and adaptive components.

We then investigated treatment-related effects on systemic immunity in our Hepa1-6 model through flow cytometry and chemokine profiling in blood. In contrast to what was observed in tumor tissue, the proportions of circulating overall T lymphocytes (CD45⁺CD3⁺) as well as CD4⁺ and CD8⁺ subpopulations were significantly increased by cabozantinib and combination (mean 38% and 34% of CD45⁺ cells) as compared with placebo and anti-PD1 (20% and 21%, P < 0.05; Fig. 3E). In addition, combination induced a significant increase in the proportions of proliferating CD8⁺ T cells in blood (P < 0.05 vs. placebo; Fig. 3E), and the greatest enrichment in the subset of memory/effector T cells (CD45⁺CD3⁺CD44⁺), particularly memory/effector CD8⁺ T cells (P < 0.01 vs. placebo), which were seen increased to a lesser extent by cabozantinib monotherapy (P < 0.05 vs. placebo; Fig. 3E). We also assessed the neutrophil-to-lymphocyte ratio (NLR), associated with poor prognosis in patients with HCC (41), as the rate between CD3⁺ T cell and CD11b⁺Ly6G⁺ cell percentages. Cabozantinib and combination induced a significant reduction of the NLR as compared with placebo (mean 0.47 and 0.38 vs. 1.4, P < 0.05; Fig. 3F). Of note, no significant changes were observed for circulating Tregs, B cells, macrophages, or dendritic cells (Supplementary Fig. S5A). Finally, cytokine and chemokine profiling of peripheral blood revealed significantly increased levels of two chemoattractants in the combination arm: CTACK/CCL27 (chemotactic agent for T lymphocytes) and IL16 (leukocyte chemoattractant and modulator of T-cell activation; P < 0.05 vs. placebo; Supplementary Fig. S5B). Overall, cabozantinib and especially combination impacted systemic immunity by promoting an enrichment in adaptive immune cell subsets.

We then elucidated whether these active neutrophil profiles were linked to any specific molecular or clinical features in human HCC, using two cohorts of patients (cohort 1: n = 167, and cohort 2: n = 57 including paired tissue and blood samples), and the MCPCounter (42) neutrophil signature which correlated strongly with the neutrophil infiltration and activation profile in combination-treated mice (Supplementary Fig. S6). Unsupervised clustering of human HCC based on enrichment of the neutrophil signature revealed three distinct clusters of samples—Neutrophil Enriched, Neutrophil Depleted, and Tumor-only Neutrophil Depleted (TND; Supplementary Fig. S7A and S7B).

The Neutrophil Enriched cluster, accounting for 25% (42/167) of patients, was characterized by high neutrophil enrichment both in the tumor and in the adjacent tissue. These samples exhibited significant enrichment in the S3 HCC molecular class (27) and liver metabolic pathways—capturing more differentiated tumors and suppression of TGFβ signaling (43), but a reduced activation of proliferation and DNA repair pathways (Fig. 4A). In terms of immunity, Neutrophil Enriched HCCs were neither associated with adaptive immunity pathways, the HCC Inflamed (44) or Immune Classes (25) nor their active or exhausted subclasses, and they lacked HCC immune exclusion features derived from β-catenin pathway activation (39).

The Neutrophil Depleted cluster captured 31% of cases (52/167) which presented low neutrophil enrichment in both the tumor and adjacent nontumor tissue. Finally, the TND cluster encompassed 44% of cases (73/167) displaying low neutrophil enrichment in the tumor but high enrichment in the adjacent tissue. Both the Neutrophil Depleted and TND groups presented pro-proliferative traits, with a higher prevalence of stem cell features and an enrichment in oncogenic TGFβ signaling (late TGFβ; Fig. 4A; ref. 43). In addition, Neutrophil Depleted HCCs were more polyploid (FDR < 0.05 vs. TND) and displayed a marked enrichment in poor prognosis signatures in the adjacent tissue (FDR < 0.05 vs. TND and Neutrophil Enriched). Furthermore, we observed a reduced proportion of hepatitis B and C virus infections in Neutrophil Enriched cases (Supplementary Fig. S8A) and a higher proportion of cirrhotic cases in the Neutrophil Depleted subgroup (FDR < 0.05). We also observed an overall lack of correlation between tumor and adjacent neutrophil enrichment values (Supplementary Fig. S7C and S7D).

To gain insight into the neutrophil phenotypes and potential molecular mechanisms driving neutrophil recruitment in the tumors, we analyzed the DEGs in tumor tissue among clusters (FDR < 0.05, fold change > 1.5). We identified 74 genes commonly upregulated in the Neutrophil Enriched group when compared with TND, some of which were linked to innate immune cells like the macrophage-related CD14 and the neutrophil chemoattractant CXCL2 (Fig. 4B; Supplementary Fig. S9B). In parallel, we confirmed that Neutrophil Enriched HCCs were significantly associated with transcriptomic features of antitumor activity like the N1 phenotype, neutrophil migration, activation, and cytotoxicity (Fig. 4C; Supplementary Fig. S9C). We also detected an enrichment in neutrophil chemokines mostly driven by CXCL2 and CXCL12 expression, which were more strongly correlated with MCPCounter tumor enrichment (P < 0.05; Fig. 4D; Supplementary Fig. S9D). In addition, following up on the absence of β-catenin–driven immune exclusion among Neutrophil Enriched tumors, we observed that β-catenin activation (assessed through the presence of the Chiang CTNNB1 subclass) was linked to lack of active neutrophil features and a reduced expression of neutrophil chemoattractants (P < 0.05 for CXCL3 and CXCL12 in the CTNNB1 subclass vs. rest; Supplementary Fig. S9E).

Importantly, and consistent with the favorable molecular features of Neutrophil Enriched tumors, patients of this cluster presented significantly longer survival as compared with the rest of the cohort [Fig. 4E; HR = 0.49 (95% CI, 0.28–0.85)], log-rank P = 0.011, which remained significant in a multivariate analysis (Cox proportional hazards model, P = 0.025; Table 1). On the other hand, the Neutrophil Depleted subgroup presented reduced survival in the univariate analysis [Fig. 4E; Table 1; log-rank P = 0.02, HR = 1.62 (95% CI, 1.07–2.45)]. Furthermore, we observed that tumors with the highest MCPCounter signature enrichment were associated with significantly better outcomes as compared with the rest [n = 164, HR = 0.45 (95% CI, 0.25–0.81), multivariate Cox P = 0.007] and these results were recapitulated in TCGA [n = 319, HR = 0.37 (95% CI, 0.21–0.66), multivariate Cox P < 0.001; Supplementary Fig. S10; Supplementary Table S9]. In addition, we confirmed all the above molecular findings in an independent cohort (cohort 2: n = 57; Supplementary Figs. S8B and S9A).

Overall, we identified that a subset of human HCC recapitulates the specific neutrophil-related molecular effect of cabozantinib and anti-PD1 combination from our preclinical model, and these tumors of the Neutrophil Enriched cluster were linked to favorable molecular and clinical features.

The recent approval of atezolizumab + bevacizumab has been a pivotal milestone in the treatment of advanced HCC and suggests that combination therapies with ICIs are the future for HCC treatment. However, responses are typically observed in a subset of patients, and there is currently a surge of research into broadening the spectrum of responders by overcoming potential tumor-intrinsic resistance to immune checkpoint blockade. The fact that recent reports suggest that immune therapies might be more effective in viral than nonviral-related HCCs (8, 45) also further highlights the need to understand the benefits of combining distinct agents with ICIs, because not all TKIs lead to equal immunomodulatory effects. In this study, we determined using preclinical models that the experimental combination of cabozantinib with anti-PD1 was associated with a greater proportion of antitumor responses in a shorter time, and enhanced molecular features of antitumor immunity including higher neutrophil recruitment and activation. To our knowledge, this is the first study reporting a dominant role of neutrophils, particularly of the N1 phenotype, in the immune effects associated with a TKI combined with anti-PD1 therapy in HCC. We also identified distinct subgroups of human HCCs based on neutrophil features with potential clinical relevance (Fig. 5).

The observed increase in neutrophil infiltration after cabozantinib administration in our murine models had only been previously reported in a prostate cancer model, where neutrophil recruitment proved to be essential for the antitumor effect of the drug (23). These results were also supported by the transcriptomic data from a human colorectal cancer PDX model (24), contributing to the notion that this effect of cabozantinib may not be limited to a specific cancer type. Our transcriptomic analysis revealed that cabozantinib could be driving neutrophil infiltration through the upregulation of ligands for the neutrophil receptor CXCR2 (CXCL3 in our subcutaneous tumors, CXCL1 in the colorectal cancer PDX model), and that the upregulation of CXCL12 could promote their retention in the tumor, as previously reported in prostate cancer (23). In light of the phenotypic plasticity attributed to neutrophils in cancer biology (46), we observed that cabozantinib in combination with anti-PD1 could further enhance a molecular phenotype of neutrophil activation, possibly due to a contributing proinflammatory effect of the immunotherapy. In fact, the combination had the greatest impact upregulating genes linked to N1 phenotype, neutrophil maturity and degranulation, as well as immunoglobulin-related genes, which could also contribute to an antitumor neutrophil phenotype through antibody-dependent cellular cytotoxicity (47). Moreover, combination treatment was linked to reduced β-catenin signaling, which is associated with immune exclusion in HCC (38, 39), as well as lower enrichment of TGFβ signaling, which has been demonstrated to polarize neutrophils toward a protumor N2 phenotype (35, 46). Overall, this demonstrates the potential of cabozantinib to trigger a neutrophil-mediated antitumor innate immune response, which is enhanced when combined with anti-PD1 therapy.

In addition, cabozantinib significantly reduced infiltrating CD8⁺PD1⁺ T cells, a clinically interesting effect given that the accumulation of a dysfunctional CD8⁺PD1⁺ T-cell subpopulation could be responsible for the reduced efficacy of ICIs in nonviral HCC, particularly NASH-HCC (8). Further studies are required to confirm whether cabozantinib and anti-PD1 combinations could improve NASH-HCC patient response by depleting this CD8⁺PD1⁺ T-cell subset while increasing neutrophil infiltration. Of note, the evaluation of this combination in preclinical NASH-HCC models (e.g., high-fat diet) and a subgroup analysis of the COSMIC-312 trial (21) could shed light on this question.

In parallel to these effects, the combination of cabozantinib and anti-PD1 had the most favorable impact on the tumor immune microenvironment out of the treatment arms investigated, such as a significant reduction of Tregs, enriched Th1 and M1 macrophage phenotypes, and enriched innate/adaptive immune pathways. The increased proportions of memory/effector T cells found in blood suggest that adaptive immunity was also activated with this treatment, a feature associated with a significant lowering of the NLR (P < 0.001 for combination). HCC progression is closely linked with inflammation, and the importance of this is underscored by the wealth of evidence linking inflammation-based scores such as the NLR with HCC prognosis (48). Finally, consistent with the inhibitory activity of cabozantinib against tyrosine kinases including VEGFR2 (19), our data indicate that at least part of the antitumor mechanism of action of cabozantinib is derived from its antiangiogenic properties (reduced endothelial cells and VEGF signaling, and absence of VETCs, which have been proposed as a predictor of aggressive HCC; ref. 48).

In patients with HCC, using MCPCounter (42) as a transcriptomic tool to evaluate active neutrophil infiltration in tumor and nontumor adjacent tissue, we confirmed that neutrophil enrichment in the adjacent tissue or in blood is not always linked to intratumor enrichment. This suggests that neutrophil recruitment mechanisms may be independent in liver adjacent and tumor tissues, and the circulating NLR associated with poor prognosis (41) may not reflect intratumoral neutrophil levels nor their phenotype. This finding has significance given we found that a high enrichment of active neutrophils in the tumor was linked to better outcomes in two independent cohorts, along with less proliferative and more differentiated molecular features, and no enrichment in inflammation profiles (HCC Immune Class; ref. 25). In addition, intratumor neutrophil enrichment was linked to the expression of the chemokines CXCL2 (a CXCR2 ligand like CXCL1 and CXCL3, upregulated by cabozantinib in mice) and CXCL12 (upregulated in our model and previously described as a cabozantinib effect in prostate cancer; ref. 23), suggesting that relevant mechanisms of neutrophil chemoattraction in human HCC were recapitulated in the preclinical models. Moreover, β-catenin pathway activation (39) may negatively influence the presence of this neutrophil phenotype in HCC as we observed it was linked to reduced enrichment in neutrophil signatures and expression of neutrophil chemokines; this may represent an additional immune exclusion effect of β-catenin activation that is worth further exploration.

Taken together, these observations suggest that cabozantinib, particularly in combination with anti-PD1 treatment, contributes to inducing neutrophil infiltration, decreasing an immune suppressive environment, and enhancing antitumor activity compared with the monotherapies in HCC. Those patients with a reduced enrichment of active neutrophil phenotypes in the tumor (∼30% of cases) could potentially gain the greatest benefits from this combination (Fig. 5), although the molecular assessment of tumor samples from patients with HCC exposed to cabozantinib in combination with PD1/PDL1 inhibitors will be necessary to test these hypotheses.

Overall, the combination of cabozantinib and anti-PD1 has the potential to bring together the activation of adaptive and innate immune responses against the tumor, and this provides a mechanistic rationale for combining cabozantinib and anti-PD1 therapy to render enhanced antitumor immune responses among patients with HCC.

J.M. Llovet reports grants and personal fees from Bayer HealthCare Pharmaceuticals, Eisai Inc, and Ipsen; grants from Boehringer Ingelheim; and personal fees from Eli Lilly, Merck, Bristol Myers Squibb, Glycotest, Nucleix, Genentech, Roche, AstraZeneca, Omega Therapeutics, Iylon, Mina Alpha, and Boston Scientific outside the submitted work. No disclosures were reported by the other authors.

R. Esteban-Fabró: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. C.E. Willoughby: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. M. Piqué-Gili: Validation, investigation, visualization, writing–review and editing. C. Montironi: Formal analysis, investigation, methodology, writing–review and editing. J. Abril-Fornaguera: Formal analysis, investigation, writing–review and editing. J. Peix: Investigation, methodology, writing–review and editing. L. Torrens: Formal analysis, investigation, methodology, writing–review and editing. A. Mesropian: Investigation, writing–review and editing. U. Balaseviciute: Investigation, writing–review and editing. F. Miró-Mur: Formal analysis, methodology, writing–review and editing. V. Mazzaferro: Resources, writing–review and editing. R. Pinyol: Conceptualization, formal analysis, supervision, investigation, methodology, project administration, writing–review and editing. J.M. Llovet: Conceptualization, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing.

This study was sponsored by Ipsen Pharmaceuticals. R. Esteban-Fabró is supported by a doctoral training grant (BES-2017-081286) from MCIN/AEI/10.13039/501100011033 and the European Social Fund (ESF) and a mobility grant from Fundació Universitària Agustí Pedro i Pons. C.E. Willoughby is supported by a Sara Borrell fellowship (CD19/00109) from the Instituto de Salud Carlos III (ISCIII) and ESF. M. Piqué-Gili is supported by a doctoral training grant (PRE2020-094716) from MCIN/AEI/10.13039/501100011033 and ESF, and a mobility grant from Fundació Universitària Agustí Pedro i Pons. C. Montironi is supported by a Rio Hortega fellowship (CM19/00039) from ISCIII and ESF. J. Abril-Fornaguera is supported by a doctoral training grant from the University of Barcelona (PREDOCS-UB 2020) and a mobility grant from Fundació Universitària Agustí Pedro i Pons. J. Peix is supported by a PERIS ICT-Suport grant from the Departament de Salut de la Generalitat de Catalunya (SLT017/20/000206). A. Mesropian is supported by a FI-SDUR pre-doctoral support grant (BDNS 550325) from the Agency for Management of University and Research Grants (AGAUR) and the Generalitat de Catalunya. U. Balaseviciute is supported by an EILF-EASL Juan Rodés PhD Studentship (EASL_JR_12_20) from the European Association for the Study of the Liver (EASL) and the EASL International Liver Foundation (EILF). J.M. Llovet is supported by grants from the European Commission (EC) Horizon 2020 Program (HEPCAR, proposal number 667273-2), the NIH (RO1DK56621 and RO1DK128289), the Samuel Waxman Cancer Research Foundation, the Spanish National Health Institute (MICINN, SAF-2016-76390 and PID2019-105378RB-I00), through an Accelerator award in partnership between Cancer Research UK, Fondazione AIRC and Fundación Científica de la Asociación Española Contra el Cáncer (HUNTER, ref. C9380/A26813), and by the Generalitat de Catalunya (AGAUR, SGR-1358). We are indebted to the Cytometry and Cell Sorting Core Facility of the Institut d'Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS), particularly Dr. Isabel Crespo, for excellent flow cytometry technical assistance.

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.