Abstract

   Lenvatinib is a multiple receptor tyrosine kinase inhibitor targeting
   mainly vascular endothelial growth factor (VEGF) and fibroblast growth
   factor (FGF) receptors. We investigated the immunomodulatory activities
   of lenvatinib in the tumor microenvironment and its mechanisms of
   enhanced antitumor activity when combined with a programmed cell
   death-1 (PD-1) blockade. Antitumor activity was examined in
   immunodeficient and immunocompetent mouse tumor models. Single-cell
   analysis, flow cytometric analysis, and immunohistochemistry were used
   to analyze immune cell populations and their activation. Gene
   co-expression network analysis and pathway analysis using RNA
   sequencing data were used to identify lenvatinib-driven combined
   activity with anti-PD-1 antibody (anti-PD-1). Lenvatinib showed potent
   antitumor activity in the immunocompetent tumor microenvironment
   compared with the immunodeficient tumor microenvironment. Antitumor
   activity of lenvatinib plus anti-PD-1 was greater than that of either
   single treatment. Flow cytometric analysis revealed that lenvatinib
   reduced tumor-associated macrophages (TAMs) and increased the
   percentage of activated CD8^+ T cells secreting interferon (IFN)-γ^+
   and granzyme B (GzmB). Combination treatment further increased the
   percentage of T cells, especially CD8^+ T cells, among CD45^+ cells and
   increased IFN-γ^+ and GzmB^+ CD8^+ T cells. Transcriptome analyses of
   tumors resected from treated mice showed that genes specifically
   regulated by the combination were significantly enriched for type-I IFN
   signaling. Pretreatment with lenvatinib followed by anti-PD-1 treatment
   induced significant antitumor activity compared with anti-PD-1
   treatment alone. Our findings show that lenvatinib modulates cancer
   immunity in the tumor microenvironment by reducing TAMs and, when
   combined with PD-1 blockade, shows enhanced antitumor activity via the
   IFN signaling pathway. These findings provide a scientific rationale
   for combination therapy of lenvatinib with PD-1 blockade to improve
   cancer immunotherapy.

Introduction

   Immune escape of cancer cells is a major mechanism of cancer
   malignancy, mainly due to exhaustion of CD8^+ T cells that recognize
   tumor antigens [[54]1]. The PD-1 signaling pathway is a key regulator
   of CD8^+ T cell exhaustion [[55]2]. PD-1 blockade was first approved
   for use in malignant melanoma patients and then in several other tumor
   types, such as renal cell carcinoma and non-small cell lung cancer
   [[56]3–[57]6], and further clinical trials are ongoing. However,
   overall response rates to single treatment with PD-1 blockade are
   approximately 20%–30% in patients with solid cancers [[58]3, [59]4,
   [60]7]. Therefore, although PD-1 blockade therapy can be effective in
   many types of cancer, further improvements are required, and
   combination treatments with other therapies are being investigated in
   clinical trials.

   Lenvatinib mesilate (lenvatinib) selectively inhibits VEGFR1–3 and
   other proangiogenic and oncogenic pathway-related receptor tyrosine
   kinases (RTKs), including FGFR1–4, platelet-derived growth factor
   receptor α (PDGFRα), KIT, and RET [[61]8–[62]10]. Lenvatinib is
   currently used as monotherapy in patients with radio-iodine-refractory
   differentiated thyroid cancer (RAI-DTC) [[63]11] and in combination
   with everolimus to treat advanced renal cell carcinoma after VEGF
   therapy in the United States and the European Union [[64]12]. Recently,
   lenvatinib was approved for first-line treatment of unresectable
   hepatocellular carcinoma (HCC) [[65]13] in the United States, the
   European Union, Japan, China and other countries. Exploratory biomarker
   analysis of lenvatinib in a Phase 3 study of patients with RAI-DTC
   revealed that lenvatinib increased VEGF and FGF23 levels in serum and
   that the increased FGF23 levels were associated with progression-free
   survival, thus demonstrating the target engagement of lenvatinib to
   inhibit VEGFR and FGFR signaling [[66]14].

   In the present study, we investigated the antitumor and
   immunomodulatory activities of lenvatinib in syngeneic mouse tumor
   models. We also explored the mechanism of action of combination
   treatment with lenvatinib plus PD-1 blockade by using flow cytometry,
   immunohistochemistry, and transcriptome analyses of tumor tissues
   treated with lenvatinib, PD-1 blockade, or lenvatinib plus PD-1
   blockade. Our data suggest that lenvatinib exhibits immunomodulatory
   activity by decreasing TAM numbers and that combined treatment with
   PD-1 blockade induces potent antitumor activities via the activation of
   IFN signaling pathways in cancer immune microenvironments.

Materials and methods

In vivo antitumor activity study

   CAnN.Cg-Foxn1nu/CrlCrlj (Balb/c^nu/nu) and BALB/cAnNCrlCrlj
   (Balb/c^wt/wt) mice were obtained from Charles River Laboratories
   Japan, and C57BL/6JJcl mice came from CLEA Japan. BNL 1ME A.7R.1, a
   murine hepatocellular carcinoma cell line, and CT26, a murine colon
   carcinoma cell line, were purchased from ATCC. B16-F10, a murine
   melanoma cell line, was kindly provided by Dr. Isaiah J. Fidler (MD
   Anderson Cancer Center, Houston, Texas, USA). Lenvatinib was
   synthesized at Eisai (Ibaraki, Japan) and diluted with distilled water
   containing 3 mM HCl. Cells (2.0–5.0 × 10^6) in 100 μL of PBS were
   injected subcutaneously into the mice. When the mean tumor volume
   reached 30–170 mm^3, the treatments were initiated. All animals
   received lenvatinib (3 or 10 mg/kg orally) once daily for the duration
   of time indicated by the black arrow in the figures. Anti-mouse PD-1
   antibody (clone; RMP1-14), anti-mouse CD8 antibody (clone; YTS169.4),
   anti-mouse IFN-γ antibody (clone; R4-6A2), anti-mouse CSF1R antibody
   (clone; AFS98), and isotype control antibodies for each antibody were
   purchased from Bio X Cell (West Lebanon, NH, USA). All antibodies were
   diluted with PBS and mice were dosed as follows: anti-PD-1 antibody,
   200 μg/mouse, once every three days or twice weekly; anti-CD8 antibody,
   200 μg/mouse, two days before drug treatment and twice weekly
   thereafter; anti-IFN-γ antibody, 300 μg/mouse, one day before treatment
   and twice weekly thereafter; and anti-CSF1R antibody, 300 μg/mouse, one
   day before drug treatment and once every two days thereafter.

   The tumor volume was measured twice weekly. Tumor size was calculated
   as follows: Tumor volume (mm^3) = Long axis (mm) × Short axis (mm) ×
   Short axis (mm) / 2. The percentage of ΔT/C (% of control for Δgrowth)
   was calculated using the following formula: (ΔT/ΔC) × 100%, where ΔT
   and ΔC are the changes in tumor volume (TV) (Δgrowth) for the
   drug-treated and non-treated control groups, respectively. Animal care
   and experimental procedures were performed in an animal facility
   accredited by the Health Science Center for Accreditation of Laboratory
   Animal Care and Use of the Japan Health Sciences Foundation. Our ethics
   committee is “Institutional Animal Care and Use Committee”. All
   protocols were approved by the Institutional Animal Care and Use
   Committee and carried out in accordance with the Animal Experimentation
   Regulations of Eisai Co., Ltd.

Flow cytometric analysis

   Tumor tissues were collected from each mouse on Day 8 or 9 after
   treatment, dissociated into single cells by using a Tumor Dissociation
   Kit and a gentleMACS Dissociator (Miltenyi Biotec, Bergisch Gladbach,
   Germany). In cell mixtures, leukocytes positive for CD45 were isolated
   with mouse TIL (CD45) microbeads (Miltenyi Biotec) by using an OctoMACS
   Separator (Miltenyi Biotec). After washing and filtration, cells were
   blocked with Mouse BD Fc Block (BD Biosciences, San Jose, CA, USA), and
   stained with the antibodies listed in [67]S1 Table and with DAPI
   (4′,6-diamidino-2-phenylindole, Dojindo, Kumamoto, Japan). The gating
   strategy is shown in [68]S1 Fig. To detect IFN-γ-producing activated
   CD8^+ T cells, we incubated CD45^+ cells with RPMI1640 containing with
   10% FBS, PMA/Ionomycin (Sigma, St. Louis, Missouri, USA), and BD
   GolgiPlug (BD Biosciences) for 4 h at 37 °C, and then stained them with
   antibodies against CD45, CD3 (Thermo Fisher Scientific, Waltham, MA,
   USA), CD4, and CD8 (BD Biosciences). The cells were then fixed and
   permeabilized with BD Cytofix/Cytoperm (BD Biosciences) and then
   stained with antibody against IFN-γ or GzmB (BD Biosciences). The cells
   were sorted with a BD FACS AriaII SORP or LSRFortessa X-20 (BD
   Biosciences) and the data were analyzed by using FlowJo v10 (BD
   Biosciences) or Cytobank (Cytobank, Inc., Santa Clara, CA, USA).

Immunohistochemistry

   Tumor tissues were embedded in Tissue-Tek O.C.T. compound (Sakura
   Finetek, Nagano, Japan) and frozen. Section slices were placed on glass
   slides and fixed with cold methanol for 10 min, dried, and then blocked
   with Block Ace buffer (DS Pharma Biomedical, Osaka, Japan) for 20 min
   at room temperature. Rabbit anti-mouse CD11b antibody, rat anti-F4/80
   antibody (Abcam, Cambridge, UK), and anti-mouse CD31 antibody diluted
   with Block Ace buffer were added to the slides, which were then
   incubated for 60 min at room temperature. The slides were then washed
   three times with PBS. Anti-rat IgG-Alexa488, and anti-rabbit
   IgG-Alexa594, diluted with Block Ace buffer, were then added to the
   slides, which were again incubated for 60 min at room temperature
   before being washed three times with PBS. Image scanning was conducted
   on a Leica TCS SP8 (Leica Microsystems, Wetzlar, Germany) or NanoZoomer
   (Hamamatsu Photonics, Shizuoka, Japan); the images were analyzed on a
   LAS AF (Leica Microsystems) or HALO (Indica Labs, Corrales, NM, USA).

Single-cell analysis

   Tumor tissues were collected from mice bearing BNL 1ME A.7R.1 tumors
   after 1 week of treatment with lenvatinib. After dissociation with a
   tumor dissociation kit (Miltenyi Biotec), CD45^+ cells were isolated by
   using a magnetic activated cell sorting (MACS) system with CD45
   microbeads (Miltenyi Biotec). Single-cell RNA sequencing libraries of
   extracted CD45^+ cells were prepared by using a ddSEQ Single-Cell
   Isolator (Bio-Rad Laboratories, Hercules, CA, USA) and a SureCell WTA
   3’ Library Prep Kit (Illumina, San Diego, CA, USA). Data preprocessing,
   such as unique molecule identifier (UMI) counting, was performed on the
   BaseSpace according to the manufacturer’s instructions. Cells that
   passed quality control with total UMIs of between 400 and 10,000 were
   selected for further analysis. To standardize gene expression across
   cells, UMIs were converted to UMIs per million (UPM) by dividing the
   UMI of each gene by the total UMIs of the cell and then multiplying by
   a scale factor of 1 million. Finally, the UPM was further transformed
   to the logarithmic scale [log[2](UPM+1)]. To visualize variations in
   immune cells and estimate the effect of lenvatinib treatment on immune
   cells, t-distributed stochastic neighbor embedding (t-SNE) [[69]15] was
   performed using the gene expression of 750 immune-related genes
   [[70]16].

RNA sequencing (RNA-Seq)

   The CT26 tumor tissues were extracted on day 9 after each treatment,
   and using RNA from the tumor tissues, sequencing libraries were
   generated by using an Illumina TruSeq mRNA Library Kit according to the
   manufacturer’s instructions. Paired-end sequencing of 2 × 150 bp was
   conducted using an Illumina NextSeq 500 according to the manufacturer’s
   instructions. FASTQ files for each sample converted from the BCL file
   were entered into FASTQC
   ([71]http://www.bioinformatics.babraham.ac.uk/projects/fastqc/) for
   quality assessments. Sequencing data were processed by using
   Trimmomatic (v0.36) [[72]17] to remove all Illumina adaptor sequences
   and bases with low quality scores. Filtered reads were mapped to the
   mouse genome (Mm10 and Ensembl 84) by using STAR (v2.5.0c) [[73]18].
   Gene expression levels in transcripts per million (TPM) for all samples
   were estimated using RSEM (v1.2.31) [[74]19].

   Data analysis was conducted by using Strand NGS (Agilent Technology,
   Santa Clara, CA, USA) and Ingenuity Pathway Analysis (IPA, QIAGEN,
   Hilden, Germany).

Weighted gene co-expression network analysis

   Weighted gene co-expression network analysis (WGCNA) [[75]20] was
   applied to identify gene modules in the RNA-Seq transcriptome data of
   the CT26 tumor tissues by using the R Bioconductor package. WGCNA
   clusters genes into color-name assigned modules using a topological
   overlap that measures network connectedness from an adjacency matrix of
   genes. A soft-thresholding power was determined to approximate
   scale-free topology of the network. The gene modules were identified by
   using a dynamic branch-cutting algorithm [[76]21] where the
   dissimilarity correlation threshold and the minimum number of genes in
   each cluster were set as 0.05 and 30, where the modules reflecting
   potentiating effects emerged.

Gene enrichment analysis

   Pathway enrichment analysis for a given gene list was performed by
   simple overlap statistic (hypergeometric) analysis [[77]22]. The
   biologically meaningful gene ontology terms and pathways were collected
   from multiple databases including the Gene Ontology consortium (GO;
   [78]http://geneontology.org), Pathway Interaction Database (PID;
   [79]http://pid.nci.nih.gov), Reactome ([80]http://www.reactome.org),
   Wikipathways ([81]www.wikipathways.org/), and NetPath
   ([82]www.netpath.org/).

Quantitative real-time PCR (qRT-PCR)

   The CT26 tumor tissues were extracted on Day 8 after antibodies
   treatment, and using RNA from the tumor tissue samples, cDNA was made
   using high-capacity cDNA Reverse Transcription Kit (Thermo Fisher
   Scientific). The cDNA diluted with TaqMan Fast Advanced Master Mix
   (Thermo Fisher Scientific) was applied in custom designed TaqMan array
   microfluidic cards (Thermo Fisher Scientific). Gene expression was
   measured by QuantStudio7 (Thermo Fisher Scientific). The data analysis
   was conducted by using Expression Suite Software v1.1.

Statistical analysis

   Differences between vehicle- and mono- or combination therapy-treated
   groups were analyzed by using unpaired t tests. A value of P < 0.05
   (two sided) was considered statistically significant. Missing tumor
   volume data were imputed using the last observation carry forward
   (LOCF). Statistical analyses were performed by using Prism (v7.02,
   GraphPad Software, San Diego, California, USA).

Results

Immunomodulating and antitumor activity of lenvatinib under immunocompetent
conditions

   To investigate the immunomodulatory activity of lenvatinib, in addition
   to its known antiangiogenetic activity [[83]8], we compared the
   antitumor activity of lenvatinib in immunocompetent mice (Balb/c^wt/wt
   mice) with that in immunodeficient mice (Balb/c^nu/nu mice) by using
   the CT26 mouse colon carcinoma model (CT26 model) and BNL 1ME A.7R.1
   mouse HCC cells (BNL model). Lenvatinib (10 mg/kg) inhibited tumor
   growth in both mouse models compared with vehicle treatment, but the
   tumor growth of the CT26 isograft was delayed significantly in
   Balb/c^wt/wt mice compared with Balb/c^nu/nu mice ([84]Fig 1A and 1B).
   Lenvatinib at 3 and 10 mg/kg also inhibited tumor growth of the BNL
   model in Balb/c^nu/nu mice, but it caused shrinkage of BNL tumors in
   Balb/c^wt/wt mice only ([85]S2 Fig). These findings indicate that
   lenvatinib has more potent antitumor activity in the immunocompetent
   tumor microenvironment.

Fig 1. Antitumor activity of lenvatinib in immunocompetent and
immunodeficient mice in the CT26 model.

   [86]Fig 1
   [87]Open in a new tab

   A. Immunodeficient mice (Balb/c^nu/nu) and immunocompetent mice
   (Balb/c^wt/wt) inoculated with the CT26 cells were randomized into
   groups of 6 mice with an average tumor volume size (Day 1 mean TV:
   Balb/c^nu/nu mice, 76.7 mm^3; Balb/c^wt/wt mice, 80.0 mm^3), and were
   then treated with vehicle (blue circles) or 10 mg/kg lenvatinib (red
   squares) once daily (black arrows). Error bars indicate the SEM. B. The
   values of ΔT/C (%) were plotted for Balb/c^nu/nu mice (red-filled
   squares) and Balb/c^wt/wt mice (red-open squares). ****, P<0.0001; ***,
   P<0.001; Dunnett’s test vs vehicle, ***P<0.001, Dunnett’s test vs.
   Balb/c^nu/nu mouse (n = 6). C. Mice, inoculated with CT26 cells, were
   randomized on Day 9 after CT26 inoculation with an average tumor volume
   size (Day 1 mean TV, 138 mm^3) into groups of 6 or 7 mice, and were
   then treated with vehicle (blue bar) or 10 mg/kg lenvatinib (red bar).
   TILs were isolated on Day 8 after the start of treatment, and the
   percentage of TAMs gated by using Ly6C^− F4/80^+ (black square) in the
   CD11b^+ Ly6G^− population was analyzed and plotted. ****, P<0.0001,
   unpaired t-test vs. vehicle (n = 6 or 7). D. Immunohistochemical
   analysis of the TAM population in CT26 tumor tissues. CD11b is stained
   red, F4/80 is green, and DAPI is blue.

   To investigate effects of lenvatinib on tumor-infiltrating lymphocytes
   (TILs), we performed a single-cell gene expression analysis of TILs
   (CD45^+ cells) in BNL tumor tissues. We collected and sequenced RNA
   from 301 and 220 cells of non-treated and lenvatinib-treated tumors,
   respectively. tSNE analysis showed that the total TILs (521 cells) from
   the lenvatinib-treated and vehicle groups could be divided into three
   immune cell populations. Compared with non-treatment, lenvatinib
   increased the number of immune cells in the C1 category but decreased
   the number of cells in the C3 category ([88]S3A and S3B Fig). The gene
   markers of immune cell populations indicated that T cell, NK cell, and
   cytotoxic cell markers were expressed by the C1-categorized cells.
   Neutrophil markers were expressed by the C2-categorized cells.
   Macrophage markers such as Cx3cr1, Mrc1 and Csf1r were expressed by
   most of the C3-categorized cells ([89]S3C Fig). These results suggest
   that lenvatinib decreased the TAM population, but increased the T, NK,
   and cytotoxic cell populations.

   Consistent with the results of the single-cell analysis, flow
   cytometric analysis indicated that the TAM population (gated as CD45^+
   CD11b^+ Ly6G^− Ly6C^− F4/80^+) was significantly decreased by
   lenvatinib treatment compared with vehicle treatment in both the CT26
   model ([90]Fig 1C) and the BNL model ([91]S4A Fig). In addition,
   immunohistochemical analysis showed that lenvatinib treatment reduced
   the number of CD11b^+ F4/80^+ double-positive cells in the tumor
   (indicated in yellow in [92]Fig 1D and [93]S4B Fig). These results
   indicate that lenvatinib decreases the TAM population in both the CT26
   and BNL models.

   In the CT26 model, the effect of TAM depletion on T cell activation was
   examined by using an anti-CSF1R antibody. In the presence of the
   anti-CSF1R antibody, Prf1 and GzmB expression increased, whereas the
   expression of TAM-related genes, such as Csf1r, Cx3cr1 and Itgam,
   decreased ([94]S5 Fig). These data suggest that reduced TAM
   infiltration by lenvatinib might cause activation of CD8^+ T cells.

Attenuation of the antitumor activity of lenvatinib upon loss of CD8^+ T cell
activation in the CT26 model

   To evaluate whether the antitumor activity of lenvatinib was dependent
   on CD8^+ T cell activation, we compared the antitumor activity of
   lenvatinib with and without CD8^+ T cells in the CT26 model by using an
   anti-CD8 antibody in Balb/c^wt/wt mice. We used flow cytometry to
   confirm that CD8^+ T cell percentage were decreased as a result of
   antibody depletion ([95]Fig 2A). Compared with vehicle treatment,
   lenvatinib significantly inhibited tumor growth despite CD8^+ T cell
   depletion; the antitumor activity of lenvatinib was reduced under these
   CD8^+ T cell-depleted conditions. The ΔT/C values of lenvatinib in
   control IgG-treated mice were significantly smaller than those in CD8^+
   T cell-depleted Balb/c^wt/wt mice ([96]Fig 2B and 2C).

Fig 2. Antitumor activity of lenvatinib under CD8^+ T cell-depleted
conditions in the CT26 model.

   [97]Fig 2
   [98]Open in a new tab

   Mice, inoculated with CT26 cells, were treated with anti-CD8 antibody
   or control IgG on Day 6 and twice weekly thereafter. A. For flow
   cytometric analysis, the percentage of CD8^+ T cells in the blood was
   determined. B. Two days after the first injection of anti-CD8 antibody
   or control IgG, mice were randomized into groups of 8 or 9 with an
   average tumor volume size (Day 1 mean TV: anti-CD8 antibody, 65.9 mm^3;
   control IgG, 49.3 mm^3), and then treated with vehicle (blue circles)
   or 10 mg/kg lenvatinib (red squares), indicated by the black arrow.
   Error bars represent the SEM. C. The percentage of ΔT/C (tumor size of
   each mouse treated with lenvatinib divided by the average tumor size of
   vehicle-treated mice) is shown for the control IgG (red-filled squares)
   and anti-CD8 antibody (red-open squares) treatment groups. **, P<0.01,
   Dunnett’s test vs. vehicle or control IgG (n = 8 or 9).

Antitumor activity of lenvatinib plus anti-PD-1 in the CT26 model

   We investigated the combined antitumor activity of lenvatinib plus
   anti-PD-1 in the CT26 model. Treatment with lenvatinib or anti-PD-1
   alone significantly inhibited the in vivo tumor growth of CT26
   isografts compared with vehicle treatment ([99]Fig 3A and 3B). Combined
   lenvatinib plus anti-PD-1 treatment significantly suppressed tumor
   growth compared with either treatment alone. Similarly, lenvatinib or
   anti-PD-1 alone significantly delayed tumor growth compared with
   vehicle treatment in the B16-F10 mouse melanoma syngeneic model, and
   lenvatinib plus anti-PD-1 combination therapy was more potent than
   either treatment alone ([100]S6 Fig).

Fig 3. Antitumor effects of lenvatinib treatment alone or in combination with
anti-PD-1 in the CT26 model.

   [101]Fig 3
   [102]Open in a new tab

   A. Mice were inoculated with CT26 cells and randomized into groups of 8
   with an average tumor volume size (Day 1 mean TV, 32.6 mm^3) and were
   then treated with vehicle (blue circles), 10 mg/kg lenvatinib (red
   squares) once daily, anti-PD-1 at 200 μg/mouse (green triangles) once
   every 3 days, or a combination of lenvatinib plus anti-PD-1 (purple
   circles). Lenvatinib treatment is indicated by the black arrow;
   anti-PD-1 treatment is indicated by black triangles. B. Tumor volumes
   for individual mice are shown for each treatment group. Error bars
   represent the SEM. **** P<0.0001, Dunnett’s test vs. vehicle on Day 19;
   ## P<0.01, #### P<0.0001 vs. combination.

Effect of lenvatinib plus anti-PD-1 on the tumor microenvironment in the CT26
model

   Next, we investigated the effect of lenvatinib plus anti-PD-1 on the
   tumor microenvironment, including tumor blood vessels and immune cell
   populations by using immunohistochemical analysis and flow cytometric
   analysis, respectively, on Day 8 after each treatment. First, we
   evaluated effects on microvessel density (areas rich in CD31-positive
   blood vessels) in CT26 isografts. Lenvatinib treatment alone and the
   combination treatment decreased the number of tumor blood vessels, but
   anti-PD-1 treatment alone did not ([103]S7 Fig). The antiangiogenesis
   activity of lenvatinib was comparable to that of the combination
   treatment in the CT26 model, suggesting an alternative mechanism for
   the antitumor activity of lenvatinib plus anti-PD-1. Using flow
   cytometry, we compared changes in the percentages of immune cell
   populations among TILs in response to each treatment. Immune cell
   populations were determined as gated in [104]S1 Fig. The TAM population
   was significantly decreased by combination treatment and lenvatinib
   treatment alone at equivalent levels. Among the TAMs, lenvatinib plus
   anti-PD-1 decreased the M2 population (I-A/I-E^low), and the ratio of
   M1 (I-A/I-E^high)/M2 tended to be increased by lenvatinib plus
   anti-PD-1 combination treatment. In other myeloid cells, lenvatinib
   increased the percentage of plasmacytoid dendritic cells (pDCs), and
   combination treatment increased the pDC and monocyte populations
   ([105]Fig 4A). Combination treatment also increased the percentage of T
   cells, especially CD8^+ T cells. PD-1^+ Tim3^+ CD8^+ T cells were also
   increased in the combination treatment group compared with vehicle
   treatment. IFN-γ^+ CD8^+ T cells were increased in the lenvatinib and
   combination treatment groups compared with vehicle treatment. GzmB^+
   CD8^+ T cells were increased in each single-treatment group compared
   with vehicle treatment, and combination treatment further increased the
   number of GzmB^+ CD8^+ T cells compared with lenvatinib treatment
   alone. These results suggest that CD8^+ T cells and their cytotoxic
   activity are upregulated by lenvatinib plus anti-PD-1 combination
   treatment in the CT26 model ([106]Fig 4B).

Fig 4. Immune cell population analysis of lymphocytes in the CT26 tumor
tissues.

   [107]Fig 4
   [108]Open in a new tab

   A, Percentages of myeloid and TAM cell populations. Upper panel: TAMs,
   pDCs, granulocytes, and monocytes. Lower panel: M1, M2 in CD45^+ cells,
   and the M1/M2 ratio. B. Upper panel: T cells, CD4^+ T cells, and CD8^+
   T cells. Lower panel: PD-1^+ Tim3^+ CD8^+ T cells, IFN-γ^+ CD8^+ T
   cells, and GzmB^+CD8^+ T cells. The gating strategy is shown in [109]S1
   Fig. *P <0.05, **P <0.01, ***P<0.001, unpaired t-test vs. vehicle;
   #P<0.05, unpaired t-test vs. combination (n = 6).

Gene co-expression network analysis to identify immune-related gene sets
modulated in response to the combined treatment of lenvatinib plus anti-PD-1

   RNA-Seq analysis followed by WGCNA and the trait correlation analysis
   was performed. The RNA-Seq analysis was performed by using CT26 tumor
   tissues from mice treated with vehicle, lenvatinib, anti-PD-1, or the
   combination for 1 week. The gene expression data (TPM > 1.0) of 13,847
   unique genes for 20 samples comprising four sample groups (vehicle,
   lenvatinib [LEN], anti-PD-1 [PD], and the combination [Combo]) were
   divided into 170 modules labeled with distinct colors in WGCNA. Defined
   sample traits, each of which reflects a pattern of expression, were as
   follows: changes occurred in the presence of at least one of the drugs
   (Treatment), only in single drug treatment (Len only or PD only), in
   the combination and one of the two single treatments (with Len or with
   PD), only in the combination (Combo-only), in the combination with
   additive effects (Additive), and in the combination with synergistic
   effects (Len-dominant synergy or PD-dominant synergy) ([110]Fig 5A).
   Gene subsets that correlated with one of traits (6) through (9) were
   considered possible targets of the ‘potentiating effect’ of the two
   drugs. Eight modules containing potential targets of potentiating
   effects were determined by WGCNA and subsequent trait analysis. These
   modules were selected based on the criteria that the most correlated
   trait was Combo-only, Additive, Len-dominant synergy, or PD-dominant
   synergy; the Pearson’s correlation was |r|>0.6; and the P value was
   <0.01. The significance (FDR-adjusted P value) of the top
   over-represented pathway in each module is shown in the bar graph, and
   the significantly enriched pathway is indicated (FDR-adjusted P value
   <0.05).

Fig 5. Weighted gene co-expression network analysis (WGCNA) to identify gene
subsets modulated by combination treatment.

   [111]Fig 5
   [112]Open in a new tab

   RNA-Seq analysis of tumor tissues followed by WGCNA was performed by
   using CT26 tumor tissues. A. Defined sample traits, each of which
   reflects a pattern of expression as indicated in the text. Gene subsets
   that correlate with one of traits (6) through (9) are considered to be
   possible target gene sets of the ‘potentiating effect’ of the two
   drugs. B. Modules from the WGCNA selected for trait correlation
   analysis. The modules highly correlated with one of the modules in A
   (r>0.65 and P<0.01) are shown with the number of genes included in each
   module in parentheses. The correlation strength of each module with
   each trait is shown in the heatmap where the most highly correlated
   trait is indicated by an asterisk. The significance (FDR-adjusted P
   value) of the top over-represented pathway in each module is shown in
   the bar graph, and the significantly enriched pathway is indicated
   (FDR-adjusted P value <0.05). The genes in the antiquewhite2 module
   represented in the type-I IFN signaling pathway are shown in the arrow
   box to the right. C. Eigengene expression pattern for the antiquewhite2
   module.

   To predict the target ‘biological units’ of the potentiating modules, a
   pathway enrichment analysis was conducted on the module genes ([113]Fig
   5B). Interestingly, type-I IFN signaling pathway-related genes were
   remarkably enriched (FDR-adjusted P value was as low as 1.50e–9) in the
   antiquewhite2 module, where genes were upregulated upon lenvatinib
   treatment and even more so upon combination treatment ([114]Fig 5B and
   5C, [115]S2 Table). This module includes genes for the key positive and
   negative regulators in the type-I interferon pathway, and these genes
   were also enriched in the virus response pathway and the IFN-γ-mediated
   signaling pathway, where component genes are overlapped in some places
   ([116]S2 Table).

Role of activation of IFN signaling in the combined antitumor activity of
lenvatinib plus anti-PD-1 in the CT26 model

   Based on our WGCNA, we hypothesized that activation of type-I IFN
   signaling by lenvatinib might induce activation of CD8^+ T cells.
   Therefore, we examined the antitumor activity of anti-PD-1 with prior
   lenvatinib treatment. Tumor growth inhibition by anti-PD-1 with prior
   lenvatinib treatment for 1 week was greater than that achieved by
   anti-PD-1 alone, nontreatment, or nontreatment with prior lenvatinib
   treatment ([117]S8 Fig). To examine if the response of tumor to
   lenvatinib was an important component in the response of the tumors to
   IFN signal activation [[118]23], we divided the mice into two groups
   according to a mean of tumor sizes after 1 week of lenvatinib
   treatment: those with relatively lenvatinib-sensitive smaller tumors
   (LEN-S) and those with relatively larger tumors that were moderately
   sensitive to lenvatinib (LEN-L). Mice in the LEN-S group that were
   subsequently treated with anti-PD-1 showed greater tumor reduction than
   the other treatment groups ([119]S9 Fig). To investigate signaling
   pathways selectively regulated in relation to tumor size after 1 week
   of lenvatinib treatment, we examined upstream regulators using IPA
   based on RNA-Seq data from LEN-S and LEN-L tumors harvested on Day 8.
   The analysis showed that VEGFA was one of the top-ranked molecules
   among the inhibitory upstream regulators in both groups. In contrast,
   type-I and type-II IFN-related molecules were top-ranked activate
   upstream regulators only in LEN-S. These data suggest that IFN pathways
   play important roles in lenvatinib treatment-sensitive tumors ([120]S3
   Table). Heatmap data showing T cell related genes [[121]24] or type-I
   IFN-related gene expression in LEN-S and LEN-L also indicated that T
   cell activation and type-I IFN-related gene expression were higher in
   LEN-S compared with LEN-L ([122]S10 Fig).

   Finally, we examined the roles of IFN-γ in the antitumor activity of
   lenvatinib plus anti-PD-1 in the CT26 model by using an anti-IFN-γ
   antibody. The antitumor activity of vehicle and anti-PD-1 alone was
   unaffected by IFN-γ inhibition, but that of lenvatinib alone and in
   combination with anti-PD-1 was significantly inhibited by IFN-γ
   inhibition. These data demonstrate that the antitumor activity of
   lenvatinib plus anti-PD-1 combination treatment is dependent on
   lenvatinib’s dominant activation of the IFN-γ signaling pathway
   ([123]Fig 6).

Fig 6. Antitumor activity of lenvatinib under IFN-γ-inhibited conditions in
the CT26 model.

   [124]Fig 6
   [125]Open in a new tab

   A. Mice, inoculated with CT26 cells, were treated with anti-IFN-γ
   antibody or control IgG on Day 8 after CT26 inoculation and twice
   weekly thereafter. The mice were randomized into groups of 7 with an
   average tumor volume size (Day 1 mean TV 71.1 mm^3). The day after the
   first injection of anti-IFN-γ antibody or control IgG, mice were
   treated with vehicle (control IgG, dark blue; anti-IFN-γ antibody,
   light blue), 10 mg/kg lenvatinib (control IgG, dark red; anti-IFN-γ
   antibody, light red), anti-PD-1 (control IgG, dark green; anti-IFN-γ
   antibody, light green) or combination (control IgG, dark purple;
   anti-IFN-γ antibody, light purple). Lenvatinib treatment is indicated
   by the black arrow, and anti-PD-1 treatment is indicated as black
   triangles. Error bars represent the SEM. **P<0.01, ****P<0.0001,
   Dunnett’s test vs. control IgG (n = 7).

Discussion

   Recently, we reported that the immunomodulatory activity of lenvatinib
   leads to potent antitumor activity in the Hepa1-6 mouse HCC model
   [[126]25]. In the current study, instead of the highly T cell-inflamed
   Hepa1-6 tumor model, we used the CT26 tumor model, a frequently used
   syngeneic mouse model demonstrating moderate T cell-inflamed tumor, and
   the BNL model, a syngeneic non-T cell inflamed HCC model. Consistent
   with our results using the Hepa1-6 model, single-cell analysis of TILs
   in BNL tumors showed that lenvatinib reduced immune cell populations,
   including those with macrophage markers. Flow cytometric analyses
   showed decreased TAM populations in both the CT26 and BNL model
   ([127]Fig 1 and [128]S3 Fig). These findings suggest that lenvatinib
   has immunomodulatory activity against both T cell-inflamed and non T
   cell-inflamed tumor microenvironments. In contrast, anti-PD-1 did not
   affect the myeloid populations. These data suggest that the antitumor
   activity of the combination treatment a lenvatinib plus anti-PD-1 would
   be dependent on the activity of lenvatinib against the myeloid cell
   populations.

   The antitumor immune response generally varies depending on the
   duration of treatment, but recent findings suggest that analysis
   approximately 1 week is sufficient to reveal changes in immune cell
   populations [[129]26, [130]27]. In our current study, flow cytometric
   analysis on Day 8 also showed that lenvatinib plus anti-PD-1
   combination treatment increased T cell numbers, especially those of
   CD8^+ T cells and GzmB^+-activated CD8^+ T cells. These results are
   supported by our finding that TAM depletion by the use of an anti-CSF1R
   antibody induced activation of CD8^+ T cells, as indicated by
   upregulation of Prf1 and GzmB ([131]S5C Fig). Our results thus reveal
   that lenvatinib is not only a potent angiogenesis inhibitor but also an
   effective immunomodulator whose functions differ from those of immune
   checkpoint inhibitors.

   Voron et al. reported that VEGFA produced from cancer cells and stromal
   cells upregulated the expression level of PD-1 and other inhibitory
   checkpoints involved in CD8^+ T cell exhaustion, and that VEGFR
   inhibitors inhibit immune regulatory function by targeting VEGFR
   signaling [[132]28]. The gene expression patterns of angiogenesis
   factors such as VEGFA are major gene expression signatures in clinical
   samples from anti-PD-1-nonresponsive patients [[133]29].

   The VEGFR signaling pathway also plays important roles in the migration
   of monocytes from bone marrow into tumors [[134]30–[135]32]. In
   addition, the FGF2-mediated FGFR1 signaling pathway regulates the
   survival and migration of TAMs in the tumor microenvironment in
   esophageal cancer [[136]33]. Therefore, these reports suggested that
   lenvatinib may increase antitumor immunity through decrease of the TAM
   population by inhibiting the VEGFR and FGFR signaling pathways.

   In the current study, WGCNA using the RNA-Seq data revealed that the
   IFN signaling pathway was extracted as Len-dominant synergistic
   pathway. In addition, molecules involved in the type-I and type-II IFN
   signaling pathways were the main lenvatinib-activated upstream
   regulators in mice with small tumors ([137]S3 Table). The type-II IFN,
   IFN-γ, is a key cytokine in the stimulation of a host antitumor immune
   response, including antigen presentation and T cell trafficking
   [[138]34]. Type-I IFN signal activation promotes the death of cancer
   cells and restricts tumor growth by altering the immune
   microenvironment [[139]23, [140]35, [141]36]. Moreover, the efficacy of
   PD-1 blockade is dependent on the activation status of the IFN pathway
   [[142]23]. Therefore, the activation of type-I and -II IFN signaling in
   tumor after lenvatinib treatment could be preferable for anti-PD-1
   antibody therapy.

   In our previous report, we showed that the dual inhibitory activity of
   lenvatinib against both VEGF and FGF induced broad-spectrum antitumor
   activity due to the antiangiogenesis effects [[143]8]. These
   antiangiogenetic effects convert the immunosuppressive status of the
   tumor microenvironment to a pro-tumor milieu and lead to priming of
   increased IFN-γ production by cytotoxic T cells [[144]37]. Together,
   these reports suggest that lenvatinib’s strong antitumor activity
   associated with angiogenesis inhibition is related to the activate IFN
   signaling in the tumor microenvironment.

   In contrast, rather than IFN signaling-related molecules, CTLA-4
   emerged as an upstream regulator in tumors that were moderately
   sensitive to lenvatinib (LEN-L). In a Phase I study, combined treatment
   with anti-VEGFA antibody and anti-CTLA-4 antibody achieved favorable
   outcomes in patients with metastatic melanoma [[145]38]. For tumors
   moderately sensitive to lenvatinib, the addition of anti-CTLA-4
   antibody may yield an effective combination therapy.

   Our results suggest that two inhibitory mechanisms––antiangiogeneis and
   anti-TAM activity––are responsible for the antitumor and
   immunomodulatory activities of lenvatinib. Currently, Phase 1b/2 trials
   of lenvatinib plus pembrolizumab in patients with selected solid
   cancers ([146]NCT02501096/[147]NCT03006926), a Phase 3 trial of
   lenvatinib in combination with pembrolizumab in patients with advanced
   renal cell cancer ([148]NCT02811861) and Phase 1b of lenvatinib plus
   nivolumab in patients with HCC ([149]NCT03418922) are underway. Further
   analyses are needed to precisely delineate the contributions of
   lenvatinib-induced VEGFR and FGFR inhibition to the suppression of TAM
   infiltration and IFN signaling activation as a mechanism of action in
   this combination therapy.

Conclusion

   Our findings in this study revealed that lenvatinib, a potent VEGFR and
   FGFR signaling inhibitor, shows more potent antitumor activity when
   combined with PD-1 blockade by decreasing TAM numbers, thereby
   affecting antitumor immune responses. In addition, this combination
   treatment enhanced the activation of the IFN signaling pathway. These
   findings demonstrate the mechanism of action of lenvatinib alone and in
   combination with PD-1 blockade. Our results provide a scientific
   rationale for evaluating lenvatinib plus PD-1 blockade as combination
   therapy in clinical trials to improve cancer immunotherapy.

Supporting information

   S1 Fig. Gating strategy for the flow cytometric analysis in myeloid and
   T cell panels.

   Immune cell populations analyzed in the Myeloid panel (TAMs, monocytes,
   granulocytes, pDCs, M1, and M2) and T cell panel (T cells, CD4^+ T
   cells, CD8^+ T cells, PD-1^+ Tim3^+ CD8^+ T cells, IFN-γ^+ CD8^+ T
   cells, GzmB^+ CD8^+ T cells) shown in Figs [150]1, [151]2 and [152]4
   were gated as indicated in the sequence of blue arrows.

   (TIF)
   [153]Click here for additional data file.^ (2.1MB, tif)
   S2 Fig. Antitumor activity of lenvatinib in immunocompetent and
   immunodeficient mice in the BNL model.

   Immunodeficient mice (Balb/c^nu/nu) and immunocompetent mice
   (Balb/c^wt/wt) inoculated with BNL cells were randomized into groups of
   5 with an average tumor volume size of 155 mm^3 (Balb/c^wt/wt mouse) or
   170 mm^3 (Balb/c^nu/nu mouse) (Day 1) and then treated with vehicle
   (blue circle), 3 mg/kg lenvatinib (orange square), or 10 mg/kg
   lenvatinib (red square) once daily, indicated by the black arrow. Error
   bars represent the SEM. ****P<0.0001, Dunnett’s test vs. vehicle (n =
   5).

   (TIF)
   [154]Click here for additional data file.^ (315.9KB, tif)
   S3 Fig. Single cell and RNA-Seq analyses of tumor-infiltrating
   lymphocytes in the BNL model.

   A. Single cells (521) are depicted in the tSNE two-dimensional space,
   where each plot represents each of the cells. Vehicle-treated cells are
   shown in cyan on the left; lenvatinib-treated cells are shown in red on
   the right. Cells were separated into three distinct clusters (C1: blue,
   C2: green, and C3: pink). B. Comparison of the changes in the
   proportions of cells in the three clusters between vehicle- and
   lenvatinib-treated cells. C. Expression levels of eight cell-specific
   marker genes.

   (TIF)
   [155]Click here for additional data file.^ (1.4MB, tif)
   S4 Fig. Immune cell population analysis in the BNL model.

   A and B. Representative profiles from flow cytometric analysis showing
   TAMs gated as a CD45^+CD11b^+Ly6G^-Ly6C^-F4/80^+ population in tumors
   derived from vehicle- or 10 mg/kg lenvatinib-treated mice are shown.
   The percentage of the TAM population among the CD45^+ cells is plotted.
   **P<0.01, unpaired t-test (n = 10). C. Representative images from
   immunohistochemical analyses showing CD11b (red) and F4/80 (green).
   TAMs are stained yellow.

   (TIF)
   [156]Click here for additional data file.^ (2.5MB, tif)
   S5 Fig. Immunomodulatory activity of anti-CSF1R antibody in the CT26
   model.

   A. Tumor tissues were harvested on Day 8, and TAM depletion by the
   anti-CSF1R antibody was confirmed by flow cytometry. B. Gene expression
   profiles of Csf1r, Cx3cr1, Itgam, Tgfb1, Il15, Il6, Prf1, GzmB, and
   Ifng in tumor tissues from control IgG-treated mice (blue) and
   anti-CSF1R antibody-treated mice (orange) in quantitative PCR analysis.
   *P<0.05, ***P<0.001, unpaired t-test (n = 6).

   (TIF)
   [157]Click here for additional data file.^ (443KB, tif)
   S6 Fig. Antitumor activity of lenvatinib alone or in combination with
   anti-PD-1 in the B16-F10 model.

   A. Mice were inoculated with B16-F10 cells and randomized into groups
   of 10 with an average tumor volume size of 71 mm^3 (Day 1) and then
   treated with vehicle (blue circle), 10 mg/kg lenvatinib (red square)
   once daily, anti-PD-1 at 500 μg/mouse (green triangle) once every 3
   days, or the combination (purple triangle). Lenvatinib treatment is
   indicated by the black arrow, and anti-PD-1 treatment is indicated as
   black triangles. B. Changes in tumor size for individual mice are shown
   for each group. *P<0.05, ****P<0.0001, Dunnett’s test vs. vehicle;
   ##P<0.01, ####P<0.0001 vs. combination.

   (TIF)
   [158]Click here for additional data file.^ (487.4KB, tif)
   S7 Fig. Immune cell population analysis in the CT26 model.

   A. Representative images from immunohistochemical analysis showing CD31
   (red) and DAPI (blue). B. The number of CD31-stained cells and the area
   of the tumor tissue slice were determined for each mouse by using HALO;
   the CD31^+ cell number was then divided by the tumor tissue area (mm^2)
   and plotted. *P<0.05, ***P<0.001, Dunnet’s test vs. vehicle.

   (TIF)
   [159]Click here for additional data file.^ (5.1MB, tif)
   S8 Fig. Antitumor activity of anti-PD-1 with prior lenvatinib
   treatment.

   A. Mice were inoculated with CT26 cells and randomized into 3 groups
   with an average tumor volume of 78.4 mm^3 (Day 1) as follows:
   nontreatment, 8 mice; anti-PD-1 treatment, 8 mice; and lenvatinib
   treatment, 32 mice. Mice were then treated with nontreatment (blue
   circle) or anti-PD-1 at 200 μg/mouse (green triangle) once every 3
   days. The lenvatinib-treated mice were further randomized into 2 groups
   of 16 mice on Day 8 at an average tumor volume of 212 mm^3 and then
   treated with anti-PD-1 at 200 μg/mouse (pink circle) once every 3 days
   or nontreatment (orange square). Lenvatinib treatment is indicated by
   the pink arrow and anti-PD-1 treatment is indicated by the green arrow.
   Nontreatment is indicated by the blue-framed arrow. B. Changes in tumor
   size for individual mice are shown for each treatment group. Error bars
   represent the SEM. *P<0.05, ****P<0.0001, Dunnett’s test vs. vehicle;
   ###P<0.001, ####P<0.0001, vs. combination.

   (TIF)
   [160]Click here for additional data file.^ (699.7KB, tif)
   S9 Fig. Antitumor activity of anti-PD-1 based on tumor volumes with
   prior treatment of lenvatinib for 1 week.

   Using the same data as those used in the legend to [161]S8 Fig,
   lenvatinib-treated mice were further divided into two groups based on
   the tumor volume on Day 8: the lenvatinib-sensitive relatively small
   tumor group (LEN-S; mean TV, 127 mm^3; n = 16) and the lenvatinib
   moderate sensitive relatively large tumor group (LEN-L; mean TV, 296
   mm^3; n = 16). Tumor growth was then re-analyzed according to anti-PD-1
   treatment (n = 8) or nontreatment (n = 8) for both the LEN-S and LEN-L
   group. Tumor growth in mice treated with nontreatment (blue circle),
   anti-PD-1 (green), LEN-S followed by anti-PD-1 (red) or nontreatment
   (dark pink), and LEN-L followed by anti-PD-1 (orange) or nontreatment
   (pink). B. Changes in tumor volumes for individual mice are shown for
   each treatment group. Error bars represent the SEM. ****P<0.0001,
   ****P<0.0001, Dunnett’s test vs. LEN-L followed by anti-PD-1; ##P<0.01,
   ###P<0.001, ####P<0.0001, vs. LEN-S followed by anti-PD-1.

   (TIF)
   [162]Click here for additional data file.^ (754.7KB, tif)
   S10 Fig. Gene expression profiles of tumor tissues of the LEN-S and
   LEN-L groups in the CT26 model.

   Using RNA-Seq data from the LEN-S and LEN-L tumors, gene expression
   levels of T cell-related genes and type-I IFN-related genes compared
   with vehicle were plotted as heatmaps. The gene expression level
   indicator is shown at the bottom right.

   (TIF)
   [163]Click here for additional data file.^ (1.6MB, tif)
   S1 Table. List of antibodies for flow cytometry analysis.

   (TIF)
   [164]Click here for additional data file.^ (291.1KB, tif)
   S2 Table. The properties of the antiquewhite2 module in [165]S3 Fig.

   The module genes and their correlation with the module eigengene and
   enriched biological pathways.

   (TIF)
   [166]Click here for additional data file.^ (544.6KB, tif)
   S3 Table. Upstream regulators of gene expression changes in LEN-S and
   LEN-L tumors compared with nontreated tumors in [167]S9 Fig.

   The top ten activated or inhibited upstream regulators in LEN-S and
   LEN-L compared with vehicle are listed.

   (TIF)
   [168]Click here for additional data file.^ (782.8KB, tif)

Acknowledgments