Tumor endothelial cell‐induced CD8+ T‐cell exhaustion via GPNMB in hepatocellular carcinoma

Abstract Tumor endothelial cells (TECs) promote tumor angiogenesis and regulate cytotoxic T cells in the tumor microenvironment. However, the roles of TECs for tumor‐infiltrating T‐cell in hepatocellular carcinoma (HCC) is still unknown. Here, we aimed to investigate how TECs influenced tumor growth and immune responses of HCC focusing on CD8+ T‐cell infiltration and exhaustion. First, TECs were isolated from subcutaneous HCC tumors with murine HCC cell lines (BNL‐T) with magnetic selection of CD31+ cells, and normal endothelial cells (NECs) were isolated from normal liver. Second, immunocompetent mice were injected with BNL‐T alone, BNL‐T + NECs, or BNL‐T + TECs for tumor formation, and the functions and exhaustion of tumor‐infiltrating CD8+ T cells were evaluated. The mice injected with BNL‐T + TEC showed rapid tumorigenesis and a decrease in the number of infiltrating CD8+ T cells. In addition, the percentage of CD8+ T‐cell exhaustion was significantly higher in tumors from the administration of BNL‐T + TEC. Third, the next‐generation sequencing on TECs was performed to identify mRNAs that might be a novel treatment target. The molecule of glycoprotein nonmetastatic melanoma protein B (GPNMB) was identified and the functions of GPNMB was analyzed by silencing of GPNMB expression using small interfering RNAs. The silencing of GPNMB expression in TECs induced the suppression of tumor growth and T‐cell exhaustion. In conclusion, TECs induced tumor‐infiltrating T‐cell exhaustion via GPNMB expression and GPNMB might be a novel therapeutic target in HCC.


| INTRODUC TI ON
Hepatocellular carcinoma accounts for nearly all primary liver cancers.
It is the fourth most common cause of cancer-related death worldwide.
Treatment options for HCC include surgical resection, radiofrequency ablation, trans-arterial chemoembolization, and systemic chemotherapy. Recently, the combination of an anti-programmed cell death ligand-1 (PD-L1) inhibitor and a VEGF inhibitor has shown clinically meaningful benefits for patients with unresectable HCC. 1 Currently, the field of immunotherapy with VEGF inhibitors is rapidly growing, but the mechanism underlying the anti-tumor effects provided by immunotherapy combined with anti-angiogenic therapy remains unclear. [2][3][4] Tumor blood vessels in the TME play important roles in solid tumor progression, and act as gatekeepers for tumor cells to metastasize to distant organs. 5 The characteristics and functions of TECs in tumor blood vessels are different from those of NECs in normal blood vessels. Indeed, compared with NECs, TECs have different morphology, altered blood flow, enhanced permeability, and structural abnormalities, due to differences in cell-to-cell connection and pericyte attachment. 3,6,7 Recent studies have reported that TECs could facilitate the escape of cancer cells from immune surveillance and induce cancer progression. In HCC, when TECs were functionally and structurally normalized, CD8 + T cells became activated and T-cell infiltration increased. 8 The mechanism by which TECs facilitate immune escape remains unclear, but several studies have indicated that metabolic changes in the TME were related to CD8 + T-cell expression of immune checkpoint proteins and CD8 + T-cell exhaustion. [9][10][11] Cytotoxic T cells comprise an essential component of protective host immunity against malignant tumors. However, prolonged exposure to cognate antigens attenuates the effector capacity of T cells and limits their therapeutic potential. This process, known as T-cell exhaustion, is involved in T-cell dysfunction, tolerance, and senescence. 12 In the present study, we investigated the influence of TECs on HCC tumor growth and the host immune response, in vitro and in vivo. We compared TECs to NECs for their effects on CD8 + T-cell functions, with a focus on T-cell exhaustion. We performed nextgeneration sequencing on TECs and NECs to investigate potential treatment targets, and we identified glycoprotein nonmetastatic melanoma protein B (GPNMB) as a candidate target. Finally, we investigated the effects of targeting GPNMB in TECs.

| Reagents
We purchased BSA from Sigma-Aldrich and ProLong Glass Antifade Mountant with NucBlue™ Stain from Thermo Fisher Scientific.

| Cell line and culture conditions
Experiments were performed with the BNL 1 ME A.7R.1 murine HCC cell line (American Type Tissue Culture Collection). Cells were maintained in high-glucose DMEM (Nacalai Tesque), supplemented with 10% FBS (Invitrogen, Life Technologies), and cultivated in a humidified incubator at 37°C with 5% CO 2 .
A new, highly tumorigenic HCC cell line (BNL-T) was established as previously described. 13 We injected a suspension of 5 × 10 6 BNL cells in a volume of 100 μl into mice. At 3 or 4 weeks after the injection, subcutaneous tumors were harvested, minced, then incubated with collagenase type I. Dissociated tumor cells were placed in DMEM in a 10-cm dish and incubated until they reached confluency.
Then, tumor cells were again injected subcutaneously into mice. We repeated this procedure for three cycles. The original HCC cell line was designated BNL, and the HCC cells obtained after three cycles were designated BNL-T.

| Animal experiments
Animal experiments were performed with 8-week-old male BALB/ cAJcl-nu/nu immunodeficient mice and BALB/cAJcl immunocompetent mice (CLEA). To produce tumors in vivo, we injected mice with 5 × 10 6 BNL cells subcutaneously. To obtain TECs and NECs, we harvested subcutaneous BNL tumors or healthy livers of BALB/cA-Jcl-nu/nu mice, respectively, and isolated endothelial cells as previously reported. 14 CD31 + cells were plated and grown in Endothelial Tumor volume (V) was determined three times per week with the formula V = d2 × D/2, where d was the minor tumor diameter and D was the major tumor diameter. This study was approved by the Animal Experiments Committee, Osaka University (approval number, 02-065-001).

| Flow cytometry
Tumor endothelial cells and NECs were detached with trypsin-EDTA. Then, all cells were incubated with anti-CD31 and anti-CD105 antibodies. Subcutaneous tumors were minced and incubated using a Tumor dissociation kit (Miltenyi Biotec). After washing, single-cell suspensions were filtered and incubated with anti-CD4, anti-PD-1, anti-CD45, anti-TCRβ, anti-CD8, anti-T-cell immunoglobulin and mucin domain-containing protein 3 (anti-Tim-3), and Live/dead antibodies. All antibodies were purchased from BioLegend. These cells were analyzed using a FACS Calibur flow cytometer (BD Biosciences) and FACS Diva ™ software (BD Biosciences). For the cell evaluation method, viable cells were gated from tumor cells, CD45-positive cells were gated, lymphocytes were gated by size, doublets were removed, TCRβ-positive CD3 cells were gated, and the percentages of CD4 and CD8 in CD3 cells were evaluated, followed by gating with CD8, and the percentages of PD-1 positive and Tim-3 positive cells in CD8 cells was evaluated.

| In vitro functional assays
Tube formation was analyzed as previously described. 16 Briefly, diluted Matrigel (Corning) was placed in each well of a 24-well plate.
Tumor-cell viability was assessed using the CCK-8 assay (Dojindo), as previously described. 17 Cell migration of NECs and TECs was evaluated by using the Radius™ 24-Well Cell Migration Assay (Collagen I Coated; Cell Biolabs). Briefly, each cell line (5 × 10 5 cells/well) was incubated in a 24-well plate. Then, cell migration was observed into the cleared area using the ImageJ/Fiji ® plugin "Wound Healing Tool." Wound closure was expressed as the ratio of the percentages of wound-closing areas, measured in high-power fields.

| Evaluation of tumor-infiltrating CD8 + T-cell function and metabolism
Tumors were minced and dissociated using the Tumor dissociation kit. Tumor-infiltrating CD8 + T cells were purified with antibodycoated magnetic beads (CD8 [TIL] MicroBeads mouse, Miltenyi Biotec). ELISpot assays were performed as previously described. 19 Briefly, tumor-infiltrating CD8 + T cells were sorted and plated Each condition was tested in duplicate wells. Spots were counted using computer-assisted image analysis (ELISpot analyzer; Cellular Technology). The total production of ROS was assessed using a ROS assay kit (Dojindo). Briefly, tumor cells were sorted to isolate tumor-infiltrating CD8 + T cells. ROS production was determined using a fluorescence microscope (BZ-X700; Keyence). ized mRNA expression data, and the values were expressed as logarithms, with 2 as the base (i.e., log 2 ). We excluded mRNAs that were undetectable. Candidate target mRNAs were defined as those that showed more than a five-fold difference between NEC and TEC expression levels.

| GSEA
We analyzed mRNA expression levels to determine associations between TEC or NEC gene expression profiles and previously de-

| Quantitative reverse transcriptionpolymerase chain reaction
qRT-PCR was performed as previously described. 21 Briefly, cDNA was synthesized using a Reverse Transcription System (Promega).
qRT-PCR was performed on an Applied Biosystems ® ViiA™7 Real-Time PCR System (Thermo Fisher Scientific). Amplification products were quantified with the THUNDERBIRD ® SYBR qPCR Mix (TOYOBO). Each gene expression level was normalized to the expression of GAPDH, an endogenous control.

| Small interfering RNA in TECs
We downregulated GPNMB expression in TECs by transfecting with 5 nM ssiRNA (Life Technologies) and Lipofectamine RNAiMAX reagent (Invitrogen), according to the manufacturer's instructions.
As a control, we transfected TECs with Scrambled siRNA (Life Technologies). 18 We analyzed transfected TECs with functional, cell wound-healing, and tube formation assays. We also tested the effects of GPNMB-siRNA-transfected TECs on in vivo tumor formation. Briefly, we injected 4.5 × 10 5 BNL-T cells + 5 × 10 4 transfected TECs subcutaneously into each mouse (n = 10 per group).

| Statistical analysis
Continuous variables were compared using Mann-Whitney U-tests or Student's t tests. Statistical analyses were performed with JMP software version 15.0 (SAS Institute Inc.) and R version 3.6.2 (http:// www.R-proje ct.org/).

| NEC and TEC isolation and functional assessments
To compare the characteristics of NEC and TEC, each one was isolated from tumor or normal livers. Both TECs and NECs showed high expression levels of CD31 and CD105 that were specific markers for endothelial cells ( Figure 1A). In western blotting, TECs, NECs, and BNL cells showed similar high expression levels of VEGFR1 and ZO-1, which are well known as endothelial receptor and adhesion markers. Only BNL cells showed strong expression of AFP that was a specific marker for HCC ( Figure 1B). These results suggested that we had successfully separated NEC and TEC from the tumors.
Compared with NECs, TECs showed greater proliferative ability ( Figure 1C) and migration activity in wound-healing assays (36.4% vs. 60.8%, p = 0 0.0031; Figure 1D). In tube-forming assays, NECs and TECs formed round tubes characteristic of cultured endothelial cells, but BNL cells did not exhibit tube formation ( Figure 1E). Therefore, cultured TECs and NECs maintained endothelial cell properties after cell isolation.

| A highly tumorigenic cell line (BNL-T) established with in vivo selection
The HCC parental cell line (BNL) showed a high rate of tumorigenesis (80%) in immunodeficient mice. However, in immunocompetent mice, the tumorigenesis rate was relatively low (60%) 2 weeks after injection. 14 To create a highly tumorigenic HCC cell line, we applied

| NEC and TEC tumorigenesis in a subcutaneous tumor model
We subcutaneously injected BNL-T cells, alone, or combined with NECs or TECs (5 × 10 4 cells). The total number of injected cells in each group was adjusted to 5 × 10 5 cells. The BNL-T + TEC group exhibited rapid tumorigenesis compared with the BNL-T and BNL-T + NEC groups (Figure 2A). When tumor tissues were stained with anti-CD31 antibody to detect tumor vessels, the BNL-T + TEC group showed significantly larger numbers of vessels (32.4/field) compared with the BNL-T and BNL-T + NEC groups (25.2 and 26.8/field, respectively, p < 0.001; Figure 2B). The proportion of Ki-67-positive cells was significantly larger in the BNL-T + TEC group (36.3%) than in the BNL-T and BNL-T + NEC groups (15.8% and 18.3%, respectively; p < 0.001; Figure 2C). Notably, double immunofluorescence CD31 and CD8 staining demonstrated significantly fewer tumorinfiltrating CD8 + T cells in the BNL-T + TEC group (8.2 cells/field), compared with the BNL-T and BNL-T + NEC groups (11.6 and 11.8 cells/field, respectively; p = 0.004; Figure 2D).
We investigated IFNγ production with ELISpot assays, in which CD8 + T cells sorted from BNL-T tumors were used as target cells. The results showed that CD8 + T cells sorted from BNL-T + TEC tumors produced significantly less IFNγ (22 spots, p = 0.002), compared with CD8 + T cells sorted from BNL-T (74 spots) and BNL-T + NEC (66 spots) tumors ( Figure 3C). These results indicated that the tumor-infiltrating CD8 + T cells from BNL-T + TEC tumors were exhausted and produced low levels of IFNγ against the BNL-T cells. Then, we measured total ROS production to evaluate metabolic deficiencies in tumor CD8 + T cells. We found that ROS accumulated in CD8 + T cells from BNL-T + TEC tumors. The total ROS production was lower in CD8 + T cells from BNL-T and BNL-T + NEC tumors than in CD8 + T cells from BNL-T + TEC tumors ( Figure 3D).

| Transcriptome analysis of NECs and TECs with next-generation sequencing
Based on the preceding results, we suspected that TECs played important roles in tumor growth and T-cell exhaustion in HCC. We also hypothesized that specific molecules expressed by TECs affected the TME and led to CD8 + T-cell exhaustion. To identify candidate molecules, we performed a transcriptome analysis of NECs and TECs with next-generation sequencing. We identified 42 mRNAs that showed greater than five-fold differences between NECs and TECs ( Figure 4A). To investigate functional differences between NECs and TECs, we performed a GSEA with the NEC and TEC RNA sequence data. The GSEA results showed that TECs were more with GPNMB expression in NECs ( Figure 4C). Western blotting and immunocytochemical staining confirmed that GPNMB expression was upregulated in the cytoplasm of TECs ( Figure 4D,E). Double immunofluorescence staining of CD31 and GPNMB demonstrated that GPNMB was expressed in tumor blood vessels, but not in normal blood vessels of normal liver ( Figure 4F).
Proliferation was inhibited in TECs transfected with siGPNMB-1 or siGPNMB-2, to ~20-30% of the levels observed in TECs transfected with siScramble ( Figure 5C). The wound-healing assay showed that TEC migration was significantly interrupted by downregulating GPNMB expression; wound closure was 33.4% in cells transfected with siGPNMB-1, 34.2% in cells transfected with siGPNMB-2, and 56.5% in cells transfected with siScramble (p < 0.001; Figure 5D, Figure S2A). Tube formation assays demonstrated that the length of tubes in TECs was found to be significantly shortened by suppressing the GPNMB expression; the average tube length was 4750 μm in cells transfected with siGPNMB-1, 4567 μm in cells transfected with siGPNMB-2, and 7500 μm in cells transfected with siScramble ( Figure S2B). In western blotting, the expression of adhesion factors, ICAM-1 and VCAM-1, was increased by suppressing GPNMB expression in TECs ( Figure 5B).  Figure 5E). We counted the numbers of vessels that expressed CD31 in harvested subcutaneous tumors.

| Functional changes in tumor-infiltrating CD8 + T cells after suppressing GPNMB expression
Flow cytometry demonstrated that the ratios of CD8 + T cells to

| DISCUSS ION
This study demonstrated that TECs induced tumor-infiltrating CD8 + T-cell exhaustion in HCC (Figure 3). Our next-generation sequencing results revealed that GPNMB could serve as a novel treatment target (Figures 4 and 5). Indeed, we demonstrated that downregulating GPNMB suppressed tumor growth and T-cell exhaustion ( Figure 6). The treatment modalities for HCC have included surgical resection, ablation, trans-arterial chemoembolization, and systemic chemotherapy. The recent development of antimolecular targeted therapy has brought the several chemotherapeutic regimens for HCC. 44 However, long-term use of these therapies for HCC induced resistance to chemotherapy. Alternatively, immunotherapy may be an attractive therapeutic option for HCC, because an inflammatory TME was associated with improved survival. 45 Previous studies have shown that intertumoral accumulation of infiltrating cytotoxic T cells was a prognostic factor in HCC. 46 Recently, studies on TECs have generated a widely held hypothesis that tumor vasculature is highly abnormal and dysfunctional; this impairs the ability of immune cells to penetrate solid tumors. The normalization of tumor vasculature might potentiate immunotherapy efficacy by enhancing tissue perfusion and improving immune effector cell infiltration. Furthermore, tumor vessel normalization might increase the infiltration and activation of CD8 + T cells. 47 Our results showed that TECs facilitated tumor growth and progression by promoting CD8 + T-cell exhaustion and inhibiting CD8 + T-cell infiltration. We showed that TEC GPNMB expression led to immune suppression by inducing T-cell exhaustion, but knocking down GPNMB expression led to the recovery of CD8 + T-cell infiltration and IFNγ production. Hypervascular HCC tumors exhibited crosstalk between vascular cells and immune cells; therefore, a treatment modality that combines immune therapy and anti-angiogenic agents could be a reasonable, promising treatment strategy for HCC.
In conclusion, we showed that TECs induced tumor-infiltrating T-cell exhaustion via GPNMB expression. These findings suggested that GPNMB in the tumor vasculature might serve as a novel therapeutic target in HCC treatments.

ACK N OWLED G M ENTS
We acknowledge the NGS core facility of the Genome Information Research Center at the Research Institute for Microbial Diseases of Osaka University for support in RNA sequencing and data analysis.

D I SCLOS U R E S
The authors of this manuscript declare no conflict of interest associated with this study.