We used two murine HCC cell lines in this study: RIL-175 (a p53/Hras mutant line syngeneic to C57Bl/6 mouse strain background),26 kindly provided by Dr Tim Greten (NIH), and HCA-1, a line derived from a spontaneous HCC in a C3H mouse established in our laboratory.27 28

For orthotopic tumor generation, RIL-175 cells were implanted in male mice of C57Bl/6 strain background (wild-type, Rag1 ^–/– , Cxcr3 ^–/–, Jackson Labs) and HCA-1 cells in male C3H mice.29 30 For spontaneous HCC generation, we used Mst1 ^–/– Mst2 ^f/– mice (both genders).

Regorafenib was provided by Bayer and administered by oral gavage daily at a dose of 5, 10 and 20 mg/kg in 34% 1,2-propandiol (Sigma-Aldrich), 34% PEG400 (Sigma-Aldrich), 12% pluronic F68 (Thermo Fischer Scientific, Waltham, Massachusetts, USA), and 20% water, as per manufacturer’s recommendation. Mouse anti-PD1 antibody (clone RMP-014) was purchased from BioXcell (Lebanon, New Hampshire, USA) and Mouse IgG isotype control was purchased from Thermo Fischer Scientific. The STAT3 inhibitor LLL12 was purchased from BioVision (Milpitas, California, USA). Antimouse PD1 antibody or IgG (control) were given intraperitoneally at a dose of 10 mg/kg thrice weekly, as described.11 Daily gavage of vehicle and intraperitoneal injection of 10 mg/kg of isotype-matched IgG were given as control treatments for anti-PD1 antibody. For the survival studies, moribund status was used as the end point and moribund was defined as symptoms of prolonged distress, >15% of weight loss compared with the starting date, body condition score >2 and tumor size of >24 mm in diameter.

Formalin-fixed paraffin-embedded HCC tissue and blood samples were obtained with informed consent from a published cohort of deidentified patients who underwent liver resection at the department of Surgery at Fundeni Clinical Institute, Bucharest, Romania.31 Furthermore, we collected blood samples with informed consent from patients treated by regorafenib as standard of care (HCC) or in a phase I clinical trial of regorafenib monotherapy in patients with acute myeloid leukemia (AML). Peripheral blood was collected before regorafenib therapy and after 2 weeks of treatment and processed in a Clinical Laboratory Improvement Amendments-certified facility of the Steele Laboratories at MGH.

Prior to immunostaining, cells were washed with the buffer and fixed and permeabilized with FoxP3/Transcription Factor Staining Buffer Set (eBioscience/Thermo Fischer Scientific) to stain the intracellular markers. Harvested cells were incubated in Dulbecco's Modified Eagle Medium with cell activation cocktail with brefeldin A (Biolegend) for 4 hour at 37°C. The cells were stained with the antibodies of cell surface and intracellular marker in the buffer with brefeldin A.

Antimouse CD16/32 antibody (clone 93, Biolegend, San Diego, California, USA) was added for FcR blockade, and incubated for 5 min at room temperature. After another washing step, antibodies for cell phenotyping were added, and cells were incubated for 40 min at room temperature. The monoclonal antibodies used for flow cytometry analysis were for: CD45 (30-F11), CD3e (145-2 C11), CD4 (RM4-5), CD8a (5H10-1), Foxp3 (FJK-16s), CD39 (Duha59), CXCR3 (CXCR3-173), CD11b (M1/70), Ly6C (HK1.4), Ly6G, CD206 (C068C2), MHC-II (M5/114.15.2), interferon gamma (IFN-γ) (XMG1.2), PD1 (29F.1A12) (eBioscience).

For analyses of endothelial and perivascular cells, tumor tissues containing the total areas of CD31 (clone DIA-310, Dianova, Germany) positive endothelial cells and α-smooth muscle actin (α-SMA) (Sigma-Aldrich, Saint Louis, Missouri, USA) positive pericytes were identified by scanning tumor sections under ×10 magnification and were counted in five random fields under ×200 magnification. CD8+ lymphocytes were detected using anti-CD8b antibody (eBioscience). Five random fields were selected from the tumor center, which did not include the edge of the tumor. Hypoxia was identified and quantified using immunostaining for anticarbonic anhydrase (CA)-IX (rabbit monoclonal; dilution, 1:100, Abcam). These data were analyzed using ImageJ (US NIH) and Photoshop (Adobe Systems, San Jose, California, USA) software. Analysis was performed using a laser-scanning confocal microscope (Olympus, FV-1000). Anticleaved caspase-3 antibody (Abcam) was used to assess the cell apoptosis in the tumor tissues, and antiphospho-STAT3 antibody (Abcam) for quantitative assessment of p-STAT3. For the latter, immunostained whole slide tumor sections were scanned on a Hamamatsu NanoZoomer S360 slide scanner. For quantification purpose, whole images were analyzed for at least three representative areas of vital tumor which was defined as region-of-interest.

The RNA-Scope assay was performed according to manufacturer’s instructions.32 Briefly, formaldehyde-fixed and paraffin-embedded human and mouse HCC tissue blocks were cut in 5 µm sections, pretreated for 30 min, following digestion and hybridization at 40°C in a HybEZ oven with both dual labeling probes antimurine CXCR3 probe (Mm-CXCR3; 402511-C2) and CXCL10 (Mm-CXCL10; 408921) or human CXCL10 (Hs-CXCL10; 311851), all from Advanced Cell Diagnostics. Counterstaining was performed using hematoxylin for 10 s.

From the orthotopic HCC and liver damage mouse models described above, blood samples were collected at 1, 4, 7 and 24 hours after once-daily oral administration of 5, 10 and 20 mg/kg regorafenib, respectively, for 5 days. Anti-PD1 antibody was given on days 1 and 4 with dose of 10 mg/kg by intraperitoneal injection. Blood samples were drawn from four to five animals per sampling time point. Regorafenib and its active metabolites M-2, M-4 and M-5 were analyzed in plasma samples using a validated bioanalytical liquid chromatography-tandem mass spectrometry.25 33 Pharmacokinetic (PK) parameters were derived from the area under the plasma concentration versus time curve (AUC) using a non-compartmental analysis.

All statistical analyzes were performed using Stata software, V.14.1 (StataCorp, College Station, Texas, USA). Error bars indicate the SEM. Differences were considered significant when p values were <0.05. Quantitative variables were compared using Student’s t-test. Analysis of experiments with more than two groups were performed using one-way analysis of variance with Scheffe’s correction for multiple comparisons. Wilcoxon signed-rank test was used for analysis of marker changes over time. Based on tumor growth delay data, we estimated that an n>10 mice/group were required to ensure an alpha level of 5% (two-sided) and beta level of 20% (two-sided) for the survival analysis. Median overall survival was estimated using the Kaplan-Meier method. Statistical analyses in the survival experiments were performed by Cox proportional hazard model and HR and 95% CIs were calculated as well.

See additional materials and methods in online supplemental material.

We first evaluated the dose-dependent effects of regorafenib in combination with anti-PD1 antibodies on HCC vessels in a time-matched manner (after 1 week of treatment). CD31+ surface area—a measure of microvascular density (MVD)—was increased after treatment with 10 mg/kg regorafenib and anti-PD1 antibody compared with all the other groups in RIL-175 model (online supplemental figure S1A). Moreover, pericyte coverage was significantly increased, and tissue hypoxia showed a tendency for decrease, after treatment with anti-PD1 antibody with 10 mg/kg but not with 20 mg/kg of regorafenib (figure 1A,B and online supplemental figure S1B,C).

To further understand the effects of the intermediate dose (10 mg/kg) of regorafenib, we examined its effects alone or in combination with anti-PD1 treatment at a later time-point in the same model (day 12). We found that the significant increase in MVD after 10 mg/kg regorafenib alone and with anti-PD1 treatment compared with control was maintained at this time-point (online supplemental figure S1B). Moreover, the hypoxic tissue area, measured by CA-IX staining, was significantly decreased in the intermediate dose regorafenib and combination treatment groups compared with control (figure 1C,D and online supplemental figure S1D,E). Furthermore, the mature (pericyte-covered) MVD was significantly increased only in the combination group, consistent with both structural and functional vascular normalization in HCC after regorafenib at the 10 mg/kg dose with PD1 blockade (figure 1E,F).

In addition, we evaluated PK parameters in HCC-bearing mice after regorafenib administration alone at doses of 5, 10 and 20 mg/kg or in combination with anti-PD1 therapy at the 10 mg/kg regorafenib dose for 5 days to achieve a steady state. We found differential plasma versus tumor PK for regorafenib and anti-PD1 therapy. In plasma, the PK parameters area under the curve within the observation period of 24 hours (AUC_0–24) and maximum/peak concentration (C_max) showed a dose-proportional increased for regorafenib (online supplemental table S1). However, while tumor AUC_0–24 and C_max also showed a dose-proportional increase for regorafenib, addition of anti-PD1 therapy to regorafenib 10 mg/kg further increased tumor exposure of the drug (online supplemental table S1 and online supplemental figure S1F,G). Consistent with these data, we found an increase in the biologically active metabolite of regorafenib M-4, which occurs prominently in mice, from the lower to higher dose group. While plasma PK of M-4 were not changed by addition of anti-PD1 antibody to regorafenib 10 mg/kg, tumor PK parameters also showed trends for increased M-4 (online supplemental table S1). The concentrations of the active metabolites M-2 and M-5 were below the detection limit of the assay in most samples. Collectively, these results show that an intermediate dose of regorafenib with anti-PD1 blockade is required to promote vascular normalization, which reduces hypoxia and increases regorafenib delivery and activity in the HCC tissue.

Having demonstrated that regorafenib given at 10 mg/kg dose with anti-PD1 antibody normalized tumor vessels’ structure and function, we next investigated the effects of this combination therapy on antitumor immunity in HCC models. First, we evaluated the intratumoral infiltration and distribution of CD8 T cells in RIL-175 tumors. The total number of CD8 T-cells infiltrating inside the tumor, evaluated by IF, was significantly increased after 1 week of treatment in the tumors from mice receiving anti-PD1 therapy, alone or with regorafenib at 5, 10 or 20 mg/kg. However, the CD8 T cell infiltration inside the tumor was significantly higher after 10 mg/kg regorafenib and anti-PD1 antibody compared with all other groups (figure 2A,B).

In separate experiments, we tested whether this effect of regorafenib 10 mg/kg/anti-PD1 therapy was maintained at a later time-point (day 12 of treatment). We found that the total number of CD8 T cells measured by flow cytometry in tissue was significantly increased in the combination therapy group as compared with all the other treatment groups (figure 2C), despite the persistence of immunosuppressive myeloid cells (online supplemental figure S2A,B). Furthermore, when we evaluated the intratumoral distribution of the CD8 T cells by IF, we found that their infiltration was significantly increased in the combination group versus other treatment groups (figure 2D,E and online supplemental figure S2C). In addition, combination therapy promoted the activation of CD8 T cells, as measured by IFN-γ expression. Using western blot analysis, we found that overall tissue IFN-γ expression was increased after combination therapy (online supplemental figure S2D); flow cytometry validated that CD8+IFN-γ+activated CTLs were significantly increased only in the combination therapy group (figure 2F).

To investigate the mechanism mediating the increased intratumoral CTL infiltration after combined regorafenib 10 mg/kg and anti-PD1 treatment in HCC, we examined the transcriptional changes in whole tumor lysate using RNA sequencing analysis after 1 week of treatment (online supplemental dataset S1 and figure S2E). Gene set enrichment analysis (GSEA) using curated gene sets (Gene Ontology) showed significant changes in multiple pathways, including those related to immune response and factors related to CTL trafficking in RIL-175 tumor tissues (figure 3A). We found that an unbiased GSEA using a previously published ‘poor prognostic angiogenic gene’ signature20 showed that these genes were enriched in the control group compared with regorafenib (10 mg/kg) plus anti-PD1 therapy (online supplemental figure S2F). These data further support the vascular normalization effect which was seen in the time-matched study presented in figure 1.

On the other hand, we found significant upregulation in the gene expression of both the Cxcr3 receptor and its ligand Cxcl10 after combination treatment (compared with control) among the top hits (online supplemental table S2). In addition, we measured protein concentration in the whole tumor tissue (used in the PK experiments) by ELISA and found that CXCL10 showed a non-significant trend for increase after treatment with intermediate dose regorafenib and anti-PD1 antibody (n=4) (online supplemental figure S3A). Of note, we have previously reported RNA sequencing analysis after dual antibody blockade of VEGFR2 and PD-1 antibody using the same orthotopic HCC model, but found no difference in CXCL10 or CXCR3 expression.21 These data indicate that upregulation of the CXCL10/CXCR3 axis may be due to the multitargeted (VEGFR-independent) activity of regorafenib.

In the same samples used for the PK study, we also found that regorafenib treatment inhibited phosphorylated (p)STAT3 levels in HCC tissues (online supplemental figure S3B,C). Thus, we evaluated the pSTAT1/pSTAT3 ratio after combination of regorafenib (10 mg/kg)/anti-PD1 treatment. We found an increase in the ratio of STAT1/STAT3 phosphorylation levels after combination treatment (figure 3B,C). These results indicate that regorafenib treatment can shift the ratio between activated STAT1 and STAT3, potentially leading to increased CXCL10 production. We also used the TCGA database to confirm the correlation between the CXCL10 expression and the other immune-related genes. We found that the CXCR3, STAT1, CD8A and IFNG genes were positively correlated in human HCC tissue (online supplemental figure S3D).

To determine if CXCL10 was produced by murine HCC cells in response to regorafenib treatment, we next cultured RIL-175 cells with various concentration of regorafenib in vitro and measured the CXCL10 levels by ELISA. Regorafenib treatment significantly increased the level of CXCL10 secreted by RIL-175 cells and to a lesser extent in HCA1 cells, with a peak concentration at 1 µM (figure 3D and online supplemental figure S3E). Moreover, we also found that a selective and direct STAT3 inhibitor (LLL12) significantly increased CXCL10 expression in RIL-175 cells (online supplemental figure S3F). Furthermore, we confirmed that the STAT1 activation and STAT3 inhibition were also seen in the RIL-175 cells culture with 1 µM of regorafenib in vitro (figure 3E,F).

To validate whether regorafenib treatment can increase expression of CXCL10 in vivo, and to identify a cell population in HCC expressing CXCL10 and CXCR3, we performed an RNA-Scope in situ hybridization (ISH) dual labeling for both markers. In line with the in vitro findings, we found an increased level of CXCL10 mRNA in the combination group of regorafenib and anti-PD1 antibody in the murine HCC tissue (figure 3G–I). In addition, dual labeling of both CXCL10 and CXCR3 confirmed that TILs in proximity of CXCL10-expressing HCC cells showed expression of CXCR3 mRNA. CXCL10 expression was almost exclusively detected in cancer cells, based on histological analysis performed by a trained liver pathologist (AQ). Moreover, flow cytometric analyses confirmed the increased levels of CXCR3+CD8+ T cells after combined regorafenib/anti-PD1 therapy in the tumor tissue but not in the spleen (online supplemental figure S3G,H).

Finally, we measured the dose-dependent changes in intratumoral CXCL10 protein expression after regorafenib treatment in mice. We found that regorafenib treatment alone at 10 mg/kg and 20 mg/kg increased protein levels of CXCL10 evaluated in whole tumor tissue lysate by western blot analysis as early as 6 hours after treatment (online supplemental figure S3I,J). Moreover, when we evaluated CXCL10 expression after prolonged combination treatment with regorafenib and anti-PD1 antibodies (for 1 week in the PK studies), we found increased protein expression level of CXCL10 for the intermediate (10 mg/kg) and the high dose (20 mg/kg) of regorafenib but not for the low dose (5 mg/kg) by western blot analysis in whole tumor tissue lysate (figure 3J,K).

Collectively, these results indicate that CXCL10 expression by HCC cells is increased after regorafenib treatment due to inhibition of STAT3 activity leading to increased intratumoral infiltration CD8+CXCR3+ T cells.

We finally tested the efficacy of combining anti-PD1 antibody with regorafenib dose versus each agent alone or control in three different mouse models (orthotopic RIL-175 and HCA-1 grafts, Mst1/Mst2-KO model) with underlying liver damage (figure 4A). In the RIL-175 HCC model, we found that combination treatment showed tumor growth delay but only when regorafenib was used at the dose of 10 mg/kg with anti-PD1 therapy (figure 4B). Moreover, regorafenib (10 mg/kg) plus anti-PD1 group showed a significant and substantial survival benefit (median survival of 29 days, more than double compared control group in this model) with an HR (HR=0.17; p<0.001) (figure 4C). The low dose (5 mg/kg) of regorafenib plus anti-PD1 therapy showed no significant growth delay and only a non-significant trend for survival benefit (HR=0.47; p=0.089) (online supplemental figure S4A). When we compared the efficacy of the combination treatment of anti-PD1 therapy with regorafenib at 10 mg/kg versus anti-PD1 therapy with a high dose (20 mg/kg) of regorafenib, we found that only regorafenib at 10 mg/kg significantly improved survival compared with control group (HR=0.24; p=0.004) (online supplemental figure S4B).

We also tested the efficacy of regorafenib combined with anti-PD1 treatment in highly metastatic HCC model with underlying liver damage (HCA-1 grafted in C3H mice). Similar to the RIL-175 model, we found a trend for increased survival for the combination when regorafenib was used at 10 mg/kg (figure 4D); of note, no activity was seen when we combined regorafenib at 20 mg/kg with anti-PD1 therapy (online supplemental figure S4C). Moreover, we tested anti-PD1 therapy with regorafenib at 10 mg/kg against spontaneous HCCs developed in Mst1/Mst2-deficient mice and found that it significantly delayed tumor growth (figure 4E). We did not detect any overt toxicity (eg, body weight loss) with any of the regimens tested, consistent with prior experience with this drug in mouse models (online supplemental figure S4D).34

Finally, when we repeated the survival experiment in an orthotopic RIL-175 model in Rag1 ^–/–/C57Bl/6 mice, which lack functional T cells, we found no significant benefit for regorafenib 10 mg/kg with or without anti-PD1 antibody (figure 4F).

These results show the safety and efficacy of combining regorafenib at 10 mg/kg with PD1 blockade in mice, and the requirement of functional T cells for the antitumor effects of this combination therapy.

To demonstrate that intratumoral CD8 T-cell infiltration after combination therapy of regorafenib with anti-PD1 antibody was mediated by CXCL10/CXCR3 axis, we next evaluated the effects of combined regorafenib (10 mg/kg)/anti-PD1 antibody therapy on RIL-175 HCC growing in Cxcr3-deficient (Cxcr3 ^–/–) versus Cxcr3-proficient (Cxcr3 ^+/+) C57Bl/6 mice. In time-matched experiments, we found that the number of CTL infiltrating HCC tissue was significantly decreased in Cxcr3 ^–/– mice compared with Cxcr3 ^+/+ mice after 1 week of combination treatment (figure 5A,B). Of note, despite this decrease, combination therapy induced a small increase in intratumoral CD8+ T cells even in Cxcr3 ^–/– mice, likely mediated by vascular normalization. Moreover, when we evaluated the efficacy of regorafenib at 10 mg/kg with anti-PD1 antibody treatment in Cxcr3 ^–/– mice in survival experiments, we found no additional survival benefit for this combination over anti-PD1 therapy alone (figure 5C); these data are in contrast to the benefits seen with regorafenib at 10 mg/kg with anti-PD1 antibody treatment in Cxcr3 ^+/+ tumor-bearing mice (figure 4C).

These results demonstrate that CXCR3 is a mediator of intratumoral CD8 T-cell infiltration—along with vascular normalization—and is critical for the added survival benefit after combination of regorafenib at 10 mg/kg with anti-PD1 therapy in HCC models.

We next performed RNA-ISH analysis of human tumor tissues and ELISA in blood samples from a published cohort of resectable patients with HCC.31 Furthermore, we collected blood samples from patients treated by regorafenib as standard of care (HCC); in addition, we examined the effect of regorafenib on blood CXCL10 levels in a clinical trial in a different disease—AML (NCT03042689). In line with preclinical data, we found that CXCL10 expression was detectable primarily in human HCC cells; >60% (30/48) untreated (surgically resected) tumors showed positivity for CXCL10 expression in HCC cells (figure 6A–D). Seven cases showed elevated levels of CXCL10 expression, with 13%–17% of tumor cells showing positivity for CXCL10 mRNA (figure 6A–D). Moreover, CXCL10 was detectable in the blood concentration of all patients with HCC (median: 617.9 pg/mL, n=20) and AML (median: 547.4 pg/mL, n=8). Finally, we found an increased concentration of plasma CXCL10 in both patients with advanced HCC (n=1) and AML (n=7) after 2 weeks of treatment (figure 6E,F). These data indicate that CXCL10 is objectively measurable and its concentration can increase in the blood of patients with HCC—supporting its future examination as a potential biomarker of the immunomodulatory activity of regorafenib treatment.