Human umbilical vein endothelial cells (HUVECs) were purchased from iCell Bioscience Inc. (Shanghai, China) and cultured in endothelial cell medium (ScienCell Research Laboratories, Carlsbad, CA, USA) supplemented with 5% fetal bovine serum (FBS), endothelial cell growth supplement (ECGS), and antibiotic (penicillin/streptomycin) solution. HUVECs were passaged no more than 6 times prior to use in experiments. Human brain vascular pericytes (HBVPs) were purchased from Fenghbio Co., Ltd. (Changsha, China) and cultured in Dulbecco’s modified Eagle’s medium (DMEM; Gibco, Thermo Fisher Scientific Inc., Waltham, MA, USA) supplemented with 10% FBS (Gibco, Thermo Fisher Scientific Inc.). PLC/PRF/5 cells were cultured in minimum Eagle’s medium (MEM; Gibco, Thermo Fisher Scientific Inc.) containing 10% FBS and non-essential amino acids (NEAA, Gibco, Thermo Fisher Scientific Inc.). Hep1-6 cells were purchased from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China) and were cultured in DMEM containing 10% FBS.

The effect of lenvatinib on tumor vasculature independent of the host immunity was evaluated in female 4-week-old NOD-Prkdc-(em26cd52) Il2rg(em26Cd22)/Nju (NCG) mice and C57BL/6 mice purchased from the GemPharmatech Co., Ltd. (Nanjing, China). All animal experiments were approved by the Institutional Animal Care and Use Committee of Sun Yat-sen University Cancer Center (SYSUCC; Approval No. L102012021030D). The other materials and methods of animal experiments were shown in supplementary data.

The immunohistochemistry (IHC) staining procedure used was as described elsewhere1. The following anti-mouse primary antibodies were used: CD31 (Cell Signaling Technology, Danvers, MA, USA), α-SMA (α-Smooth Muscle Actin; Cell Signaling Technology), NG2 (Abcam Plc, Cambridge, UK), VEGFR2 (Cell Signaling Technology). Images of five random fields per sample were acquired on a light microscope (Olympus Corporation, Tokyo, Japan). The positive staining area per field was calculated by Image-pro Plus software (Media Cybernetics, Inc.).

Tissue muti-plex IHC staining was performed using a Pano-Panel IHC Kit (Panovue, Beijing, China) according to the manufacturer’s instructions. Immunofluorescence (IF) staining images were acquired using a Vectra Polaris Imaging System (Akoya Biosciences, Marlborough, MA, USA) and quantitatively analyzed using the HALO software (Indica Labs Inc., Corrales, NM, USA). The following anti-mouse primary antibodies were used: CD31 (Cell Signaling Technology), α-SMA (α-Smooth Muscle Actin, Cell Signaling Technology), NG2 (Abcam Plc), CD8 (Abcam Plc), VCAM1 (Abcam Plc), Ki67 (Abcam Plc), Granzyme B (GranB; Abcam Plc). For immunocytochemistry (ICC) staining, HUVECs were fixed in 4% paraformaldehyde for 20 minutes at room temperature, washed three times with PBS, and then incubated with the corresponding primary antibody at 4°C, overnight. Afterwards, after washing three times with PBS f, the cells were incubated with the appropriate fluorescent dye-conjugated secondary antibody for 1 hour at room temperature, and then nuclear stained with 4’,6-diamidino-2-phenylindole (DAPI) for 10 min. Images were obtained by confocal laser scanning microscopy using a Zeiss LSM880 Confocal Microscope (Carl Zeiss Microscopy GmbH, Jena, Germany). The co-localization analysis of the immunofluorescent markers was performed with the Image J software.

HUVECs were starved in serum-reduced (1%) RPMI 1640 medium overnight and treated with lenvatinib (Selleck Chemicals LLC) at different concentrations for 2 h, followed by 60 ng/ml VEGFA165 (PeproTech Inc., Cranbury, NJ, USA). Then, HUVECs were harvested and washed twice with PBS. Cell-surface staining was performed with APC-conjugated anti-human CD304 (neuropilin-1, NRP1) monoclonal antibody (mAb, 354506; BioLegend Inc., San Diego, CA, USA), phycoerythrin (PE)-conjugated anti-human VEGFR-2 mAb (359904; BioLegend Inc.) at 4°C, for 20 minutes. For intracellular staining, HUVECs were permeabilized with BD Cytofix/Cytoperm kit (BD Biosciences) and stained according to the manufacturer’s instructions. For xenograft-derived immune cell stained, tumor tissues were minced and dissociated using a Tumor Dissociation Kit (Miltenyi Biotec Inc., Auburn, CA, USA) according to the manufacturer’s instructions. The tissue homogenates were then filtered through a 70-µm filter to isolate single cells. Immune cells were identified by CD45, CD3, CD4 and CD8. To examine the expression of Granzyme B by T cells, cells were stimulated with phorbol 12-myristate 13-acetate (PMA) and ionomycin (MilliporeSigma, Burlington, MA, USA) in the presence of the protein transport inhibitor brefeldin A (BioLegend Inc.) for 5 h, and then stained with fluorochrome-conjugated primary anti-Granzyme B antibody. All the antibodies were purchased from Biolegend Inc.

HUVECs for Western blot analysis were washed twice with PBS and then lysed on ice with RIPA buffer (MilliporeSigma) with freshly added protease inhibitor cocktail (Roche AG, Basel, Switzerland). The clarified lysates were separated on 10% sodium dodecyl sulfate/polyacrylamide gel electrophoresis (SDS/PAGE) gels and transferred to polyvinylidene fluoride (PVDF) membranes (MilliporeSigma). The membranes were blocked with 5% nonfat milk and protein bands were reacted with the appropriate primary antibodies by incubating at 4°C overnight. After washing the membranes, the protein bands were reacted with horseradish peroxidase (HRP)-conjugated secondary antibody. Subsequently, the immunoreacted protein bands were visualized using an enhanced chemiluminescence detection system (Bio-Rad Laboratories Inc., Hercules, CA, USA).

Cell lysates for co-immunoprecipitation (co-IP) experiments were prepared by lysing cells in IP Lysis Buffer (Beyotime, Shanghai, China) containing a protease inhibitor cocktail (Beyotime). Then, the clarified lysates were incubated overnight at 4°C with the appropriate primary antibodies according to the specific antibody instruction. Subsequently, the sample-antibody mixtures were rotated together with Protein A/G magnetic beads (MedChemExpress LLC., Monmouth Junction, NJ, USA) for 2 h at 4°C and then washed five times with 0.5% Triton X-100 and collected after magnetic separation. The samples were ultimately eluted 1×SDS-PAGE (Beyotime) sample buffer and further assessed by Western blot analysis.

HUVECs viability was determined using the cell counting kit-8 (CCK-8) cell viability assay (Dojindo Laboratories, Kumamoto, Japan). HUVECs (3×103 cells per well) were seeded in flat-bottomed 96-well plates. After incubation for 24 h, the cells were treated with lenvatinib at various concentrations and 60 ng/ml VEGFA165 (PeproTech Inc.) in serum-reduced (1%) RPMI 1640 medium for 72 h. The final concentration of dimethyl sulfoxide (DMSO, as vehicle) was ≤1%. Results were expressed as the mean ± S.E.M. of at least three independent proliferation assays. The dose-response curve and half maximal inhibitory concentration (IC50) values of lenvatinib were calculated busing the GraphPad Prism software (GraphPad Software Inc., San Diego, CA, USA).

HUVECs were seeded on the upper chamber of 0.1% gelatin-coated transwell inserts (0.4 µm). After reaching confluence, HUVECS were treated with lenvatinib at different concentrations for 2 h. Afterwards, albumin-Evans Blue complex (2 mg of Evans Blue/100 mL of EndoGROTM supplemented with 0.5% FBS-0.1% BSA and 60 ng/ml VEGF) was added into the upper chamber and incubated for 90 min. The ratio of Evans Blue in the lower/upper chambers was calculated by measuring the optical density (at 620 nm) of the culture medium harvested from both chambers and used to determine in vitro permeability.

HUVECs were transduced with recombinant lentiviruses carrying a red fluorescent protein (RFP) expression plasmid and HBVPs were transduced with recombinant lentiviruses carrying a green fluorescent protein (GFP) expression plasmid. HUVECs-RFP and HBVPs-GFP were co-cultured at a 4:1 ratio in Corning Matrigel Matrix (Corning Inc., Corning, NY, USA), supplemented with extracellular matrix (ECM) medium (ScienCell Research Laboratories) containing lenvatinib (at different concentrations), 20 ng/ml PDGF-BB (PeproTech Inc.) and 60 ng/ml VEGFA165 (PeproTech Inc.). Images were acquired 24 and 48 h after seeding to assess the sprouting capacities of endothelial cells (ECs) by fluorescence microscopy using a 4x objective.

HBVPs were seeded in 96-well plates at a density of 1 × 103 cells/well (100 μL/well). After allowing cells to attech for 24 h, HBVPs were treated with lenvatinib at different concentrations, 20 ng/ml PDGF-BB (PeproTech Inc.) and 60 ng/ml VEGFA165 (PeproTech Inc.). At the testing time points (24, 48 and 72 h), 10 μL of CCK-8 solution (Dojindo Laboratories) was added to each experimental well and then incubated for 2 h at 37°C. The absorbance was measured at 450 nm with an automatic microplate reader (Molecular Devices LLC, San Jose, CA, USA).

HBVPs were starved in serum-reduced (1%) RPMI 1640 medium overnight and treated with lenvatinib at different concentrations, 20 ng/ml PDGF-BB (PeproTech Inc.) and 60 ng/ml VEGFA165 (PeproTech Inc.) for 12 h. After detaching by digestion with trypsin and 0.25% EDTA, HBVPs were resuspended and counted. Subsequently, for the migration assay 6.0 × 104 cells in 200 μL of serum-free medium were placed in the upper compartment of a Transwell chamber (Corning; 24-well insert, pore size: 8 μm). The lower chamber was filled with DMEM containing with 10% FBS as a chemoattractant and then incubated for 12 h. After 12 h, the cells on the upper surface of the membrane were carefully wiped off with a cotton swab. The cells on the lower surface were fixed, stained with 0.1% crystal violet, and washed with PBS. The number of stained cells in five randomly chosen visual fields of each insert were counted under a light microscope. The experiments were performed in triplicate.

Total RNA was isolated with TRIzol and purified using the RNeasy Plus Micro Kit (Qiagen, Hilden, Germany) according to the manufacturer’s protocol. The RNA libraries were then sequenced using the Illumina HiSeq4000 platform (Illumina Inc., San Diego, CA, USA) at the Novogene Genomic Sequencing Center (Beijing, China). The RNA sequencing (RNA-seq) sequence reads were aligned to the mouse reference genome using STAR (v2.5.1b). Differential expression analysis was performed using the DESeq2 R package (1.10.1). The resulting P-values were adjusted using the Benjamini and Hochberg’s method for controlling the false discovery rate. An adjusted p-value (padj) < 0.05 and an absolute fold change of 2 were set as the threshold for significantly differentially expressed genes (DEGs). The alignment results were converted to RNA-seq gene expression measurements as FPKM (fragments per kilobase of exon per million fragments mapped) using HTSeq v0.6.0.

Statistical analyses were performed using SPSS (version 19.0) software (IBM Corporation, Armonk, NY, USA) or GraphPad Prism 6 (GraphPad Software Inc.). Based on the distribution level, data are shown as the mean ± S.D. Independent-sample or paired t test was performed to analyze the difference between the two groups with normally distributed continuous variables. For the comparison among treatment groups in the in vivo study, one-way analysis of variance (ANOVA) was performed. In all cases, a two-tailed P-value <0.05 was considered statistically significant, *P <0.05, **P <0.01, ***P <0.001, ****P <0.0001.

First, we chose NCG mouse with severe combined immunodeficiency as in vivo model to test the efficacy of lenvatinib therapy in modifying the tumor vasculature of HCC to avoid the potential effect of CD4+ T lymphocytes (18) by establishing subcutaneous xenograft tumors derived from the PLC/PRF/5 cell line, which is not sensitive to lenvatinib (19). The administration of lenvatinib was started at 7 days after transplantation of PLC/PRF/5 cells. Previous reports indicated that 3 to 30 mg/kg/day lenvatinib was necessary to achieve anticancer effects in different mouse models (19–21), thus we selected 3, 10 and 30 mg/kg/day as our dose gradients. During the treatment period, the results revealed that compared with the control group treated with CMC, the tumor growth in mice from groups treated with the three different doses was dramatically suppressed with no significant difference among the three groups ( Figures 1A-C ), and a slight decrease in body weight was found only in the Len-30 group ( Figure S1A ). We also assessed the tumor vasculature perfusion and structure after 1 week of treatment, based on the tumor vascular “normalization window” theory used in most studies (10, 22, 23) by real-time contrast-enhanced ultrasonography (CEUS) and IHC staining. According to the perfusion images ( Figure 1D ) and videos (Vid. S1), blood perfusion inside the xenograft tumor was dramatically improved only by the dose of 10 mg/kg/day lenvatinib, whereas inside the tumor of the other three groups there were still large non-enhanced areas that lacked blood perfusion. The results suggested that 10 mg/kg/day of lenvatinib had superior effect in inducing the change in the tumor vascular normalization.

As shown by the IHC staining of the tumor vasculature microstructure, after the treatment with lenvatinib, the area of microvessels inside the tumor region dramatically decreased in all the three treated groups ( Figure 1E ). Additionally, the IHC staining results with alpha-smooth muscle actin (α-SMA), which is a classical marker of pericytes, showed that the positive area was significantly higher in the Len-10 group compared with the other three groups. Further evaluation by the multiplex IHC assay with CD31, NG2 and α-SMA directly revealed a typical feature of the mature and stable vasculature with better coverage by pericytes ( Figure 1F ). In addition, the assessment of the degree of hypoxia inside the tumor environment, after injecting a hypoxia detective probe, revealed that the treatments with 10 mg/kg/day lenvatinib significantly alleviated the hypoxia inside the tumor tissue compared with the control group. In addition, we also observed its tendency to alleviate hypoxia in the Len-3 group, but the statistical analysis showed no significant difference between this group and the control group. Noteworthy, there was no significant relief of hypoxia in the group treated with the higher dose of 30 mg/kg/day of lenvatinib ( Figures 1E, G ).

It is well established that VEGFR2 is a highly active receptor that modulates various signaling pathways to promote neovascularization and generate leaky immature blood vessels (24). Therefore, we evaluated the expression of VEGFR2 by IHC, which revealed, as shown in Figures S1B-D , markedly reduced expression of VEGFR2 in endothelial cells (ECs) from mice in the Len-10 group. Remarkably, the expression of VEGFR2 in ECs from mice in the Len-30 group was upregulated instead and the relative expression level was even higher than that in the control group. In summary, we confirmed that an escalating dose of lenvatinib exerted a similar inhibitory effect on neovascularization, but a high dose of lenvatinib showed a tendency to aggravate the dysfunction of tumor vasculature.

We next investigated whether the same effect occurs in the TME of mice with competent immune system in a C57BL/6 mouse model. As shown in Figure 2A , we selected the time points at day 4, 8 and 14 to further investigate the effect of variables, such as dose and time, on the dynamic changes in vascular normalization. The analysis of tumor growth inhibition showed that there was no significant difference in tumor volume among the three treatment groups, but their tumor growth was markedly suppressed compared with the control group ( Figure 2B ). Regarding the changes in tumor vasculature, the results showed that the 10 mg/kg/day dose of lenvatinib had superior effect to normalize the tumor vessels at all the time points as indicated by the expression of α-SMA and NG2 (a more specific marker of pericytes) ( Figures 2C , S2A-C ). Additionally, their expression levels relative to the expression level of CD31 were dramatically increased in the Len-10 group compared with the other three groups ( Figures S2D, E ). We also found that both the total and relative expression of α-SMA and NG2 reached the highest level on day 8 ( Figures S2D, E ). Moreover, the multiplex IHC analysis of α-SMA, NG2 and CD31 expression, shown in Figure 2D , revealed a strong co-localization of α-SMA and NG2 surrounding CD31-stained tumor vessels in the Len-10 group. We also measured blood perfusion inside the tumor by CEUS on day 8 and representative images are shown in Figure 2E . The results showed that after the administration of contrast agents only in the Len-10 group the light intensity inside the tumor increased rapidly, following the increase in its intensity outside the tumor. However, the increase of the light intensity inside the tumor in the other three groups was blocked at a significantly lower level than the intensity outside the tumor ( Figure 2F ). In addition, we compared the maximum of the respective intensity value and perfusion area among the four groups, and both parameters showed the same tendency that the Len-10 group had the optimal condition of blood perfusion ( Figures 2G, H ). Overall, we concluded that the dose of 10 mg/kg/day exerted the optimal normalizing effect both in immunodeficient and immunocompetent mouse models.

To investigate the effect of concentration-escalation of lenvatinib in vitro, we performed a cell viability assay to evaluate the antiproliferative activity of lenvatinib over a range of escalating concentrations against HUVECs stimulated with 60 ng/ml VEGFA₁₆₅ for 48 h, which simulated the angiogenesis induced by the high level of VEGF produced within the TME. According to the dose-response curve ( Figure S1E ), the half maximal inhibitory concentration (IC₅₀) of lenvatinib against HUVECs was 10.72 μM. Therefore, we chose sequential escalating doses of 2.5, 5, 10 and 20 μM of lenvatinib, which were proportional to the IC₅₀. We then examined the integrity of vascular junctions using the Evans blue/Albumin permeability assay. The results showed that the treatment with 5 μM lenvatinib successfully blocked the VEGF-induced increased permeation of Evans blue. However, increasing the concentration of lenvatinib to 10 μM and 20 μM failed to further decrease the permeation of Evans blue and aggravated it instead ( Figure 3A ). In order to confirm the impairment of tight junctions resulting in the hyperpermeability of confluent HUVEC monolayer, we used Western blot analysis to measure the expression and phosphorylation of vascular endothelial (VE)-cadherin, which is an essential component of tight junctions between ECs. The results revealed that the phosphorylation of VE-cadherin was upregulated after the stimulation with 60 ng/ml VEGF₁₆₅, indicating the internalization of VE-cadherin and the impairment of tight junctions (25), but its phosphorylation was significantly reduced by 5 μM lenvatinib, while expression of total VE-cadherin remained unchanged. Noteworthy, although the phosphorylation of VE-cadherin was reduced when the concentration of lenvatinib was increased to 10 and 20 μM, the expression of total VE-cadherin was also downregulated ( Figures 3B , S3A-B ).

The analysis of the locational distribution of the expressed VE-cadherin by IF staining revealed, as shown by the images in Figure 3C , that its expression was localized to the plasma membrane in areas of cell-cell contact, which was disrupted by treatment with VEGFA. After the pre-treatment with lenvatinib at 2.5 and 5 μM, the effect of VEGFA on VE-cadherin was blocked. However, the higher doses of lenvatinib (10 and 20 μM) had the opposite effect, resulting in the disruption of staining continuity of VE-cadherin in areas of cell-cell contact ( Figure S3C ).

Previous research has well established that perivascular cells (PVCs) that surround and support the endothelium, including pericytes and vascular smooth muscle cells, are critical components for the integrity of vessels (26). Thus, we next investigated the effect of different concentrations of lenvatinib on pericytes. We used a cell viability assay to evaluate the effect of lenvatinib (at different concentrations) on the proliferation of HBVPs treated with 60 ng/ml VEGFA₁₆₅ and 20 ng/ml PDGF-BB to simulate the high levels of VEGFA and PDGF-BB secreted into the TME. The results suggested that lenvatinib at 2.5 and 5 μM had no inhibitory effect on the proliferation of HBVPs induced by VEGFA₁₆₅ and PDGF-BB. However, lenvatinib tended to suppress the proliferation of HBVPs at the higher concentrations of 10 and 20 μM ( Figure 3D ). The difference in proliferation among the cells treated with higher and lower concentrations of lenvatinib appeared to be most obvious at 72 h after the stimulation VEGFA₁₆₅ and PDGF-BB ( Figure 3E ). We also evaluated the effects of lenvatinib at different concentrations on the migration of HBVPs stimulated with high levels of VEGFA₁₆₅ and PDGF-BB. After the combined treatment for 12 h, the results revealed that the stimulation with VEGFA₁₆₅ and PDGF-BB slightly promoted the migration ability of HBVPs, which was further increased by the addition of 2.5 and 5 μM lenvatinib, with 5 μM lenvatinib resulting in the optimal migration. However, the number of migratory cells dramatically decreased when the dose of lenvatinib was increased to 10 and 20 μM ( Figures 2F, G ).

In addition, we used a HUVEC/HBVP three-dimensional (3D) co-culture model to analyze their morphology by fluorescent microscopy after culturing for 24 and 48 h. At 24 h of culturing, this co-culture model produced vessel-like and elongated endothelial structures with associated pericytes. In the control co-culture without VEGFA₁₆₅ and PDGF-BB stimulation, no distinct vessel-like networks formed. In the group stimulated with VEGF and PDGF-BB, the vessel-like structures irregularly extended out from the main stalk of the vessels and the pericytes were still loosely associated with ECs both at 24 and 48 h, which resembled the abnormalities of tumor vasculature. In the groups treated with 2.5 and 5 μM lenvatinib, the pericytes were recruited and closely connected with the ECs at 24 h and the appeared to be more closely connected at 48 h. At the same time, the number of vessel-like tubular structures was less and their diameter was smaller, especially at the time point of 48 h. These features resembled the normalized vasculature. In the groups treated with 10 and 20 μM, the vessel-like tubular structures were significantly pruned, along with the loosely associated pericyte ( Figures 2H, I ). Taken together, the above results indicated that lenvatinib at the concentrations of 5 μM had an optimal effect on restoring the integrity of vessels by reinforcing the tight junction between ECs and increasing the coverage by pericytes in vitro.

Recently, some studies suggested that a cross-family compensatory mechanism might have a promising effect in regulating angiogenesis, especially through treatment with multi-target small molecule inhibitors (27). Pfister et al. (28) reported that VEGFA can also bind to PDGFRβ and induce its phosphorylation. These findings inspired us to investigate whether the diverse effects of different concentrations of lenvatinib were due to cross-family interaction. We assessed whether the treatment with lenvatinib could regulate the binding of VEGF to PDGFRβ by co-immunoprecipitation (co-IP) analysis. Encouragingly, the treatment with lenvatinib actually promoted the binding of VEGFA to PDGFRβ, and this effect was most evident in the group treated with 5 μM lenvatinib. In addition, the tyrosine-phosphorylation of PDGFRβ was also enhanced ( Figures 4A , S3D ), which triggered specific regulatory mechanisms. At the same time, we also observed that the tyrosine-phosphorylation of VEGFR2 was inhibited by lenvatinib in a concentration-dependent manner without affecting the interaction with VEGFA₁₆₅ ( Figures 4B , S3E ).

NRP-1 is a multifunctional transmembrane receptor protein that interacts with a variety of activated tyrosine kinase receptors (29). Combined with the above hypothesis of cross-family compensation, these results encouraged us to investigate whether the compensatory binding of VEGF could further influence the interaction of NRP-1 with different tyrosine kinase receptors. We first examined the internalization of NRP-1 and VEGFR2 in HUVECs stimulated with 60 ng/ml VEGF and treated with different concentrations of lenvatinib by flow cytometry, which reflected the activation of their downstream signaling pathways. The results revealed that upon single stimulation of HUVECs with VEGFA, the expression of both the NRP-1 and VEGFR2 on the surface of HUVECs was significantly downregulated. Upon further pre-treatment of HUVECs with 2.5 and 5 μM lenvatinib, the surface expression of NRP1 was still downregulated but the downregulation of surface VEGFR2 was blocked. However, in HUVECs pre-treated with 10 μM and 20 μM lenvatinib, both the downregulation of surface NRP-1 and VEGFR2 were both blocked ( Figures 4C, D ). The measurement of total cell expression of NRP-1 and VEGFR2 after cell permeabilization revealed an upward tendency following the stimulation with VEGF, but there was only a slight difference between the groups treated with lenvatinib ( Figures 4E, F ). Additionally, the subsequent locational distribution analysis of NRP-1 and VEGFR2 by IF staining revealed that NRP-1 and VEGFR2 were located on the membrane of resting HUVECs. After stimulating HUVECs with VEGFA, the fluorescence intensity of immunostained NRP-1 and VEGFR2 on the membrane was immediately diminished. However, pre-treatment with 2.5 and 5 μM lenvatinib led to the recovery of the strong fluorescence signal of immunostained VEGFR2 on the cell membrane, suggesting that the internalization of VEGFR2 was inhibited, while the signal of NRP-1 on the membrane remained diminished. Increasing the concentration of lenvatinib to 10 and 20 μM inhibited the internalization of both NRP-1 and VEGFR2, which persistently remained localized on the membrane after stimulation with VEGFA ( Figures 4G , S3F ).

Having found the compensatory binding of VEGF to PDGFRβ, we sought to examine the locational correlation between NRP-1 and PDGFRβ. The results in Figure 4H reveal that, after stimulation with VEGFA, both the fluorescence intensity of immunostained NRP-1 and PDGFRβ markedly decreased and there was no significant co-localization between them. Also, although the treatment with lenvatinib recovered their fluorescence intensity, only HUVECs treated with lenvatinib at 5 μM showed strong co-localization of the expression of NRP-1 and PDGFRβ according to the scatter plot results and the analysis of overlap coefficient ( Figures 4H, I ). The co-IP assay further validated their interaction at the protein level, which showed that the stimulation with VEGFA markedly promoted the interaction between VEGFR2 and NRP-1, but their interaction was significantly blocked by the treatment with lenvatinib. Instead, the treatment with 5 μM lenvatinib induced the formation of the NRP-1-PDGFRβ complex, which was consistent with the results of the IF staining ( Figures 4J , S3G ). In summary, the treatment with 5 μM lenvatinib concurrently blocked the VEGF-induced interaction between VEGFR2 and NRP-1 and promoted the formation of the NRP1-PDGFRβ complex.

To investigate the downstream signaling pathway of the NRP1-PDGFRβ complex induced by lenvatinib, RNA-seq analysis was performed on HUVECs treated with different concentrations of lenvatinib and 60 ng/ml of VEGF. We then identified DEGs between HUVECs treated with 5 μM lenvatinib and 0 μM lenvatinib. Gene ontology (GEO) functional classification enrichment analysis of the DEGs identified from the RNA-seq data assigned the DEGs that were significantly enriched in GEO terms to a series of functional groups associated with the regulation of vascular development and morphogenesis ( Figure 5A ), which was consistent with our previous in vitro experiments. In addition, the Kyoto Encyclopedia of Genes and Genes pathway enrichment analysis showed that the top enriched pathway was the Rap1 signaling pathway ( Figure 5B ).

We also examined the activation of the Crkl-C3G-Rap1 axis, which is a typical signaling pathway that regulates the activity of Rap1 (30, 31). The western blot analysis results in Figure 5C reveal that the activation of Crkl, C3G and Rap1 showed the same tendency for all six treatment groups, and the activation of all three proteins was highest in the group treated with 5 μM lenvatinib ( Figures 5C , S4A-C ). In addition, we treated the HBVPs with lenvatinib at different concentrations followed by stimulation with 60 ng/ml VEGFA₁₆₅ and 20 ng/ml PDGF-BB. The IP assay showed that only the treatment of 5 μM lenvatinib was able to significantly block the interaction between VEGFR2 and PDGFRβ, but it promoted the interaction between NRP-1 and PDGFRβ. Additionally, the tyrosine-phosphorylation of PDGFRβ was clearly increased by lenvatinib ( Figures 5D , S4D ). Based on these results, we propose a schematic model to illustrate the mechanism of the various effects of the treatment with lenvatinib at different doses on vascular normalization ( Figure 5E ).

We next evaluated the effect of the optimal dose of lenvatinib on the infiltration of different immune cells inside the tumor. Flow cytometry analysis revealed that the number of infiltrated leukocytes inside the subcutaneous tumors significantly increased in the Len-10 group compared with the other three groups at all checked time points (day 4, 8 and 14) ( Figures S3A-D ). Regarding functional lymphocytes, both the proportion of CD45+CD3+ cells and CD45+CD3+CD8+ cells was significantly increased in the Len-10 group compared to the control, Len-3 and Len-30 groups at all the checked time points, and the difference was more evident at day 8 and 14 ( Figures 6A-D ). Meanwhile, the proportion of intratumoral CD45+CD4+ showed no difference among these four groups at day 4. Additionally, at day 8 of treatment, the Len-10 group had the largest proportion of infiltrated CD45+CD4+ cells as CD3+ and CD8+ T cells. However, the number of infiltrated CD45+CD4+ cells significantly increased in the Len-30 group, but their number in the other three groups decreased ( Figures S5E-H ).

It has previously been established that the infiltration of immune cells is closely related to the expression density of vascular cell adhesion molecule 1 (VCAM1) in vascular ECs (32). We evaluated the expression levels of VCAM1, CD31 and CD8. At days 4 and 8 of the treatments, the area ratio of VCAM1 versus CD31 significantly increased in the Len-10 group compared to the other three groups ( Figure 6E ). Noteworthy, the area ratios of the control, Len-3 and Len-10 groups collectively decreased while the area ratio of the Len-30 group markedly increased and surpassed the ratio of the Len-10 group at day 14 ( Figures 6F, G , S6A ). However, the number of infiltrated CD8+ T cells was always the highest in the Len-10 groups at all checked time points ( Figure 6H ), which suggested that the upregulated expression of VCAM1 early in the treatment benefited immune cell infiltration. These results showed that, in contrast to the higher dose of 30 mg/kg/day, the 10 mg/kg/day dose of lenvatinib had an optimal effect on inducing vascular normalization and increasing the immune infiltration in vivo.

We also evaluated the PD-1 expression in the tumor samples from the control, Len-10 and Len-30 groups at different time points by multiplex IHC staining of PD-1 and CD8. The results revealed that both the number of PD-1+ cells and PD-1+CD8+ cells was higher in the Len-10 group than in the Len-30 group at all checked time points ( Figures S7B-E ). These findings led us to hypothesize that the combination treatment with 10 mg/kg/day lenvatinib and anti-PD-1 antibody could produce enhanced antitumor effect in the murine model. To test this hypothesis, we treated the tumor-bearing mice with two doses (10 or 30 mg/kg) of lenvatinib or CMC treatment once a day, combined with the administration of anti-PD-1 antibody (100 μg, i.p.) or IgG control once every three days ( Figure 7A ) to investigate their synergetic effect.

As hypothesized, the combined treatment with lenvatinib and anti-PD-1 antibody suppressed tumor growth compared with either the monotherapy group or control group. In addition, tumor growth was more prominently inhibited and even regressed by the 10 mg/kg/day lenvatinib plus anti-PD-1 antibody treatment compared to the higher dose (30 mg/kg/day) combination treatment ( Figures 7B , S7B ). The photo flux statistics and representative fluorescence images of live imaging also showed the same tendency ( Figures 7C, D ). No statistical difference in body weight was found among all groups ( Figure S7A ), suggesting that the combination treatment had no noticeable toxicity.

Moreover, the examination of the number and functions of infiltrated lymphocytes in the harvested tumor samples at the end of the therapy revealed that the Len-10+αPD-1 group had a better immune-supportive environment, accompanied by more CD45+CD3+ T cells, CD45+CD3+CD8+ T cells and CD45+CD3+CD8+GranB+ activated T cells ( Figures 7E-G ). Representative flow cytometry graphs showed clearly different clusters of CD3+CD8+ inside the tumors from the six groups ( Figure 7H ). In addition, multiplex IHC staining of CD8, Ki67 and GranB showed that both the number of CD8+ cells and CD8+Ki67+GranB+ cells per field were highest in the Len-10+αPD-1 group ( Figures 7I-K ). At the same time, we assessed the stability of tumor vessels by IHC. The results showed that there were more pericytes surrounding the vessels in the tumor tissue from the tumor-bearing mice treated with 10 mg/kg/day lenvatinib as well as 10 mg/kg/day Lenvatinib plus anti-PD-1 antibody, while the treatment of anti-PD-1 antibody demonstrated no effect on the vascular normalization in our study ( Figures S8A-D )

Additionally, the analysis of the proportion of Ki67+GranB+ cells in total CD8+ cells revealed that the combination treatment with anti-PD-1 antibody and lenvatinib upregulated the proportion of Ki67+GranB+ cells in CD8+ T cells, with the best effect being obtained with the 10 mg/kg/day dose of lenvatinib. However, the monotherapy with anti-PD-1 antibody failed to increase the proportion of Ki67+GranB+ cells compared with the control group ( Figure 7L ). A summary of the treatment with the dose-escalating lenvatinib to enhance vascular normalization and increase the efficacy of ICB immunotherapy is shown in Figure 8 .