Stromal Reprogramming through Dual PDGFRα/β Blockade Boosts the Efficacy of Anti–PD-1 Immunotherapy in Fibrotic Tumors

Gastric cancer is the fifth most common cancer in terms of incidence and the third leading cause of cancer-related death worldwide (1). To improve the insufficient clinical outcome, molecular-targeted therapies that interfere with specific molecules to block gastric cancer progression have been developed (2). Moreover, novel immunotherapies, such as anti–PD-1 immune checkpoint inhibitors, have been developed in an attempt to update the cancer treatment paradigm; however, the response rate in patients with advanced gastric cancer remains limited (3). Gastric cancers are divided into two subtypes according to Lauren's classification, intestinal-type gastric cancers (IGC), and diffuse-type gastric cancers (DGC; ref. 4). DGCs are characterized by rapid progression and abundant stroma and are likely to show resistance to conventional drugs and yield a worse prognosis. Therefore, novel therapeutic strategies that strengthen the response rate of the currently available chemotherapies and immunotherapies are needed.

Abundant stroma and severe fibrosis in cancer tissues is one of the major characteristics of intractable cancers, such as pancreatic cancer and DGCs. In fact, accumulating evidence indicates that a large amount of stroma is associated with a poor prognosis for patients with gastric cancer (5–7). Cancer-associated fibroblasts (CAF), a major component of the fibrotic tumor stroma, have been reported to be involved in gastric cancer progression and resistance to conventional therapies through interactions with gastric cancer cells (8–11). Although the tumor stroma consists of various types of immune cells in addition to CAFs, such as T cells, B cells, neutrophils, and macrophages, the mutual interaction between CAFs and immune cells remains unclear. Considering the mechanism by which tumor fibrosis occurs, TGFβ, and platelet-derived growth factors (PDGF) are considered fibrogenic growth factors. In particular, PDGF–PDGF receptors (PDGFR) signaling enhances the growth of fibroblasts in several pathological conditions, such as cardiac and kidney fibrosis (12, 13). However, the existence of four PDGF ligands and two PDGFR makes understanding the process of fibrosis more complicated (14). Therefore, the fundamental mechanism of fibrotic tumor stroma formation and the impact of CAFs on the immune microenvironment in advanced gastric cancer have not yet been elucidated.

This study investigated the underlying mechanism of PDGF–PDGFR–mediated fibrosis and the effect of stromal remodeling by PDGF–PDGFR signaling inhibition on the immune microenvironment. Here, we show that fibrotic tumors possess an immunosuppressive microenvironment with polymorphonuclear myeloid-derived suppressor cell (PMN-MDSC) accumulation and that stromal modification may occur through dual PDGFRα/β blockade, improving the efficacy of immune checkpoint inhibitors against fibrotic tumors.

Patients with gastric cancer who underwent radical gastrectomy at Kumamoto University Hospital and Saiseikai Kumamoto Hospital were enrolled after obtaining written informed consent. Patients with a tumor depth below the submucosal layer were eligible for this study. Lauren's classification was used to define histologic types. Tumor staging was performed according to the American Joint Committee on Cancer Staging Manual 8th edition. Clinical information was collected from patient medical records. For surgical samples, the largest histologic part of the gastric cancer tissue showing heterogeneity was selected. Primary CAFs and normal fibroblasts (NF) were isolated from patients who underwent gastrectomy without preoperative treatment. The study was approved by the Medical Ethics Committee of Kumamoto University (approval no.: 1277) and Saiseikai Kumamoto Hospital (approval no.: 362).

Eight- to 10-week-old male C57BL/6N mice (Clea Japan) were housed in a room under stable temperature and humidity on a 12-hour light and dark cycle. Both water and food were supplied ad libitum. All animal procedures and studies were conducted in accordance with the Institutional Animal Care and Use Committee at Kumamoto University (approval no.: A2021–156).

Human gastric fibroblast cell lines were derived from the surgically excised gastric tissue of 100 patients with gastric cancer. CAFs were established from surgically excised gastric wall tumors, and NFs were established from nontumor gastric wall tissue. The detailed protocol to establish CAFs and NFs has been reported previously (15). The murine cancer cell lines (B16-OVA, LL/2 (LLC1), GL261, 4T1, CT26, Myc-Cap, MB49, and PancO2) were kindly provided by Dr. T. Moroishi, Kumamoto University (Kumamoto, Japan), and the murine gastric cancer cell line, GAN-KP cells, was established using previously established protocols (16). Murine fibroblasts were established from samples of C57BL/6N mouse skin. Using scalpels, the skin tissue was minced into small pieces. After mincing, the tissue pieces were cultured in collagen-coated dishes (IWAKI). These cells were cultured in RPMI1640 containing 10% FBS and incubated at 37°C with 5% CO₂.

The Cancer Genome Atlas (TCGA) database for 378 patient with stomach adenocarcinoma samples was downloaded from OncoLnc (https://www.oncolnc.org/) to analyze the relationship between survival and the mRNA expression of PDGF A, B, C, and D.

Two expert gastrointestinal pathologists (Y.K. and K.O.) who were blinded to the conditions selected representative Azan-stained slides from each case and assessed the ratio of the collagen-stained area to the stromal area. The percentage of the tumor stromal area was semi-quantitatively scored using a 3-point scale: score 1 (stroma ratio <10%), score 2 (stroma ratio 10%–30%), and score 3 (stroma ratio >30%). During subsequent scoring, any confusing cases were discussed by the two observers, and the scores were determined. Scores of 2 to 4 points were defined as stroma-low, and scores of 5 and 6 points were defined as stroma-high. Slides containing mucinous-type gastric cancer were excluded from the selection because mucin is stained blue during Azan staining.

All tissue slices (4-μm thick) were cut from formalin-fixed, paraffin-embedded tissues and mounted on glass slides. The sections were prepared for heat-induced antigen retrieval by autoclaving in 1× Antigen Retrieval Solution at a pH of 9. Subsequently, each slide was washed with PBS and blocked with 5% low-fat dry milk in PBS for 30 minutes. The sections were stained with the primary antibody overnight at 4°C. For immunofluorescence (IF) staining, Alexa Fluor 488-, 555-, or 649-conjugated secondary antibodies were incubated with the sections for 1 hour at room temperature. For IHC, the sections were incubated with the HRP-conjugated anti-goat secondary antibody (Nichirei) and anti-rabbit secondary antibody (Dako). PDGF-IHC staining was scored in a blinded manner by two investigators. Staining of PDGFC and PDGFD was estimated as the intensity of staining of cancer cells. The PDGFRα- and PDGFRβ-positive areas in randomly selected fields of human sections were measured in the hybrid cell count mode with a BZ-X700 all-in-one fluorescence microscope (KEYENCE). The αSMA-, PDGFRβ-, and COL1A1-positive areas in randomly selected areas of mouse sections were measured with the hybrid cell count mode of the BZ-X700 all-in-one fluorescence microscope (KEYENCE). The antibodies used in this study are listed in Supplementary Table S1.

All tissue slices (4-μm thick) were cut from formalin-fixed, paraffin-embedded tissues and mounted on glass slides. The sections were incubated with a 0.1% Sirius Red solution dissolved in aqueous saturated picric acid for 1 hour. Each slide was washed in acidified water (0.5% hydrogen chloride), dehydrated and mounted with DPX mounting. Collagen components were stained red.

Dead cells were excluded by staining with BDHorizon Fixable Viability Stain 510 (BD Biosciences) in (D)-PBS at room temperature for 10 minutes prior to cell-surface staining. Cells were incubated with TruStain FcX (anti-mouse CD16/32) antibody in staining buffer (PBS containing 2% FBS and 10% mouse serum) to block Fc receptors. Cell surface antigens were labeled with 1 μg/mL antibodies (Supplementary Table S1) at 4°C for 30 minutes and washed with ice-cold staining buffer. Cells were fixed with BD stabilizing fixative (BD Biosciences) and permeabilized with Invitrogen saponin-based permeabilization and wash buffer (Thermo Fischer Scientific). Intracellular antigens were labeled at 4°C for 30 minutes at the same concentration as the antibody for cell surface antigen. Flow cytometry was performed with a FACSVerse instrument (BD Biosciences). The flow cytometry data were analyzed using FlowJo v10.8.0 software (BD Biosciences).

Recombinant human PDGF-CC (R&D Systems), PDGF-DD (R&D Systems), human CD140b (PDGFRβ) antibody (BioLegend), sorafenib (LC Laboratories), regorafenib (Kasei), recombinant human TGFβ1 (PeproTech), recombinant mouse CXCL1 (PeproTech), recombinant mouse CXCL3 (R&D), recombinant mouse VEGF (PeproTech), anti-CXCL1 antibody (R&D Systems), and anti-CXCL3 antibody (R&D Systems) were purchased. The anti-PDGFRα antibody olaratumab (LY3012207/IMC-3G3) was kindly provided by Eli Lilly (17).  Cells were treated with recombinant human PDGF-CC and PDGF-DD (100 ng/mL), human CD140b (PDGFRβ) antibody (10 μg/mL), recombinant human TGFβ1 (10 ng/mL), recombinant mouse CXCL1, CXCL3, or VEGF (50 ng/mL), anti-CXCL1 antibody (3 μg/mL), anti-CXCL3 antibody (3 μg/mL), sorafenib, regorafenib (10 μmol/L), or PDGFRα (100 nmol/L) in each in vitro experiment.

CAFs were seeded in 96-well plates at a density of 1.0 × 10³ cells/well and incubated at 37°C with 5% CO₂. At 24 hours after seeding, the CAFs were exposed to recombinant PDGF CC or PDGF DD with or without sorafenib, regorafenib or PDGFRα treatment. Cell proliferation was then measured with an IncuCyte S3 Live-Cell Analysis system (Essen BioScience). For the analysis with the IncuCyte system, CAFs were seeded into 96-well plates and stained with NucLight Rapid Red (Essen BioScience). The plate was inserted into the IncuCyte instrument for real-time imaging, with four fields imaged per well every 24 hours over 10 days. Data were analyzed using IncuCyte software, which quantified the percentage of red confluence values. All groups were tested in triplicate, and the data are presented as the means ± SEM.

Total RNA was extracted from cultured cells using the miRNeasy Mini Kit (Qiagen) according to the manufacturer's instructions.

Complementary DNA (cDNA) was reverse transcribed from total RNA using SuperScript III, RNaseOUT, Recombinant Ribonuclease Inhibitor, Random Primers and Oligo(dT)12–18 Primer (Thermo Fisher Scientific). mRNA expression was quantified using SYBER Green. All qRT-PCR experiments were conducted with a LightCycler 480 System II (Roche Diagnostics). All qRT–PCR data are shown as the means ± SEM. Primer sequences are listed in Supplementary Table S2.

RNA sequencing was performed by the Liaison Laboratory Research Promotion Center (LILA; Kumamoto University) using the method described below. A 2100 Bioanalyzer (Agilent) was used to detect the concentration and purity of the total RNA. All samples with an RNA integrity number (RIN) >8.0 were used for sequencing. A NextSeq 500 (Illumina) instrument was used for the analysis, and the data were converted to Fastq files. Quality control of the data was performed with FastQC. Then, the filtered reads were mapped to the UCSC hg19 reference genome using HISAT2 (v2.1.0). The fragments per kilobase of exon per million mapped reads (FPKM) values were calculated using Cufflinks (v2.2.1) software. Significant genes were extracted using Cuffdiff (P < 0.05).

Murine MDSCs (G-MDSC and M-MDSC) were isolated from the spleens of C57BL/6N mice bearing GAN-KP cells using a Myeloid-Derived Suppressor Cell Isolation Kit (Miltenyi Biotec). Each MDSC (3.0 × 10⁵cells/well) was seeded in the upper chamber of a CytoSelect 24-well Cell Migration Assay plate (Cell Biolabs). For conditioned media (CM) experiments, CM from GAN-KP cells, murine fibroblasts, or both cocultured with or without regorafenib (10 μmol/L) for 48 hours was collected. The CM from cells pretreated with or without the IgG control (3 μg/mL), anti-CXCL1 antibody (3 μg/mL), or anti-CXCL3 antibody (3 μg/mL) were added to the lower well of the plate. After 6 hours, migratory cells were dissociated with Cell Detachment Buffer and quantified using CyQuant GR fluorescent dye solution.

We utilized the Visium spatial gene expression platform (10× Genomics) for the spatial transcriptomics experiments. Tumor samples from a vehicle control- or a regorafenib-treated mouse were embedded in Tissue-Tek optimal cutting temperature (OCT) compound (Sakura Finetek) immediately after surgical resection and frozen with isopentane in a liquid nitrogen bath then stored at –80°C. OCT-embedded blocks were cryosectioned at a thickness of 10 mm using a cryostat, and tissue sections were placed within the frames on a Visium Spatial Transcriptomic Library Preparation Glass Slides (10× Genomics). The sections were then fixed with ice-cold methanol and stained with hematoxylin and eosin Y (Millipore Sigma). Visium Spatial Gene Expression Library construction was performed according to the manufacturer's instructions. Libraries were sequenced with a DNBSEQ-G400 (MGI Tech) at a depth of approximately 350 million reads per sample. Sequencing was performed using the following read protocol: read 1, 28 cycles; i7 index read, 10 cycles; i5 index read, 10 cycles; and read 2, 100 cycles. Space Ranger software v1.3.1 was used to process raw FASTQ files and histology images. Spatial transcriptomic data were analyzed using R version 4.1.1 and Seurat version 4.0.4 software (18). All data were normalized using the SCTransform (19) function and merged, followed by dimensional reduction and clustering using Seurat functions (RunPCA, FindNeighbors, FindClusters, and RunUMAP) with the default parameters. FindAllMarkers function was implemented to identify differentially expressed genes (DEG) in each cluster using a Wilcoxon rank sum test, and genes with a Bonferroni adjusted P value less than 0.01 and an average log₂-fold change greater than 0.25 were filtered. Clusters with high expression of the Col1a2 gene were identified as stromal clusters (clusters 3, 5, and 7) and DEGs in each cluster were subjected to a Gene Ontology (GO) enrichment analysis using the gprofiler2 package (20). Pathways with an adjusted P value based on a g:SCS method less than 0.01 were considered statistically significant. The enrichment of DEGs in each GO term was calculated as (the number of DEGs found in the corresponding GO term)/(the number of genes annotated to the GO term).

Eight- to 10-week-old male C57BL/6N mice (Clea Japan) were housed at Kumamoto University (approval no.: A2021–156). For the establishment of serial transplantation mouse models, GAN-KP cells (5.0 × 10⁵ cells) were implanted subcutaneously into the mouse flank. The mice were sacrificed 3 weeks after tumor cell inoculation. The tumor mass containing the various noncancerous components, such as fibroblasts, was minced and dissociated in 1× TDR (BD Biosciences) for 30 minutes at 37°C under continuous rotation. The samples were suspended in buffer (PBS containing 2% FBS and 2 mmol/L EDTA), and the cell suspension was applied to a 70 μm strainer. The red blood cells contained in each sample were lysed by the addition of 1 mL of VersaLyse (Beckman Coulter) for 10 minutes, which was then quenched in buffer. The pellet was plated on collagen-coated dishes (IWAKI) and cultured with RPMI1640 containing 10% FBS and Zell Shield (Minerva Biolabs). The cells were cultured for 1 week and implanted subcutaneously into the mouse flank. The mice were sacrificed 3 weeks after tumor cell inoculation. This process was repeated as the serial transplantation method.

Both control and regorafenib-treated tumors were minced and dissociated in TDR, as mentioned in the Serial transplantation method section. At that time, N = 3 mice or samples were pooled to control for biological variability. These dissociated tumors were stained with 7AAD for 10 minutes at 4°C, washed with PBS containing 1% BSA, and the live cell fraction (7AAD-negative fraction) was sorted using BD FACSAria III flow cytometer. These sorted single cells, which had a viability of >80%, were collected in tubes containing resuspension buffer (PBS with 0.04% BSA) and cooled to 4°C. The cells were then washed by centrifugation at 300 × g for 20 minutes, and the cell density was adjusted to 1,000 cells/μL with suspension buffer to yield a cell suspension with a viability >90%. As described below, single cells were barcoded, and the library was prepared according to the manufacturer's protocol (Single Cell 3′ Reagent Kit v3.1, 10× Genomics). Each cell suspension sample was duplicated and loaded into a chromium next GEM chip G for a target cell output of 10,000 cells. Reverse transcription was performed at 53°C for 45 minutes. Following GEM-RT, samples were cleaned using Dynabeads, and cDNAs were amplified for a total of 11 cycles using a Bio-Rad C1000 Touch thermocycler. Amplified cDNA cleanup was performed using SPRIselect beads (Beckman Coulter) and analyzed on an Agilent Bioanalyzer High Sensitivity DNA chip (Agilent Technologies) to calculate the total cDNA yield. Twenty-five percent of the cleaned and amplified cDNA samples were fragmented, followed by adapter ligation and index PCR for a total of 14 cycles. The constructed libraries were purified again with SPRIselect beads to give a library with an average length of 420 to 440 bp. These libraries were sequenced using the Illumina Novaseq 6000 platform, with a 2 × 150 bp lead length, 750 Gb/1 pool (2 library × duplicate), and paired ends.

Sequencing data were aligned to the mm10 mouse reference genome and gene expression was quantified with the Cell Ranger Single-Cell Software Suite (v6.1.2, 10× Genomics). The resulting output file, which is a gene-cell barcode matrix, was imported into R (v4.0.3) and analyzed using Seurat (v4.1.0). Ambient mRNA contamination was estimated and removed using SoupX (v1.5.2), followed by in silico doublet prediction and removal using DoubletFinder (v2.0.3). Low-quality cells were them removed using much more permissive thresholds to (i) avoid excluding cells with lower mRNA levels, such as neutrophils and fibroblasts; and (ii) to retain nonhematopoietic cells, which are more sensitive to dissociation and have higher mitochondrial gene counts. Cells with (i) number of UMI counts per cell > 150,000; (ii) number of genes detected per cell >10,000; and (iii) percentage of mitochondrial genes >20% were removed from the dataset before the downstream analysis. Afterward, data from all four libraries were normalized and then integrated using Harmony (v0.1.0). Clusters were annotated on the basis of an examination of cell lineage-specific module scores calculated using Seurat's “AddModuleScore” function. Differential expression between different neutrophil clusters and treatment groups was assessed with Seurat's “FindMarkers” function using the MAST method.

GAN-KP cells (5.0 × 10⁵ cells) were implanted subcutaneously into the flanks of C57BL/6N mice. The mice were sacrificed 3 weeks after tumor cell inoculation. The tumor mass was minced and dissociated in 1× TDR (BD Biosciences) for 30 minutes at 37°C with continuous rotation. The samples were suspended in buffer (PBS containing 2% FBS and 2 mmol/L EDTA), and the cell suspension was filtered through a 70 μmol/L strainer. The red blood cells contained in each sample were lysed by adding 1 mL of VersaLyse (Beckman Coulter) and incubating the sample for 10 minutes, which was then quenched in buffer. The samples were sorted on the FACSAria II flow cytometer (BD Biosciences) to obtain CD90.2+ and EpCAM+ cell populations as described in the “Flow cytometry” section of the Materials and Methods. Cells were lysed in 10 μL of cell lysis buffer containing 40 U of RNasin Plus (Promega), Roche Cell Lysis Buffer (Roche), 10% NP40 (Thermo Fisher Scientific), and RNase-free water (TaKaRa). As described previously (21, 22), one hundred sorted cells were simultaneously subjected to first-strand cDNA synthesis and the random displacement amplification (RamDA) reaction using the PrimeScript RT Reagent Kit (TAKARA Bio Inc.), T4 gene 32 protein (New England Biolabs), and “not-so-random” primers. The cDNA was then used as the template for qPCR.

Reverse-phase protein arrays (RPPA) were generated as described previously (23). Briefly, cells were lysed in RIPA buffer (Thermo Fisher Scientific) containing a protease inhibitor cocktail (Thermo Fisher Scientific) and a phosphatase inhibitor cocktail (Thermo Fisher Scientific). After the protein concentration of the lysates was determined using Direct Detect (Merck), SDS sample buffer (Thermo Fisher Scientific) and DTT were added to the lysates for denaturation at 100°C for 10 minutes. Four-step twofold serial dilutions of the denatured lysates were added in quadruplicate onto ONCYTE SuperNOVA nitrocellulose film slides (Grace Bio-Labs) using a robotic spotter (Genex Arrayer, Kaken Geneqs Inc.). The array slides were blocked with SuperG blocking buffer (Grace Bio-Labs) at room temperature for 1 hour and probed with 48 primary antibodies at 4°C overnight. Following tyramide signal amplification (Dako), IRDye 680RD-streptavidin conjugate (LI-COR) was applied to the slides. Fluorescence images were captured by an InnoScan 710AL microarray scanner (Innopsys) and quantified using Mapix software (Innopsys). Slides incubated without primary antibodies served as blank controls. Relative protein levels of the analytes in each sample were then determined using the SuperCurve 1.5.17 R package (24). The protein concentrations on each set of slides were normalized for protein loading (25). The primary antibodies used in this study are listed in Supplementary Table S3.

Serial GAN-KP cells were established by repeating the serial transplantation more than five times. S5 tumor mice were subcutaneously transplanted with 5.0 × 10⁵ serial GAN-KP cells into the mouse flank. S0 tumor mice were transplanted with GAN-KP cells. The tumor weight and volume were evaluated in these mice. Tumor volume was calculated using the formula: tumor volume (mm³) = length (mm) × width (mm)/2. Seven days after transplantation, mice were randomized to each group for further treatment. For the anti–PD-1 treatment study, the mice were administered 200 μg/mouse anti–PD-1 (Bio X Cell) via intraperitoneal injection. Control mice received 200 μg/mouse IgG (Millipore) equivalent to the volumes of anti–PD-1 used to treat the mice. For the regorafenib treatment study, the mice were orally administered 10 mg/kg regorafenib (Kasei). Control mice received vehicle (ethanol/Cremophor EL) equivalent to the volumes of regorafenib used to treat the mice.

All experiments were performed in triplicate, and the data shown are representative of consistently observed results. Data are presented as the mean ± SD. The Mann–Whitney U test was used to compare continuous variables between two groups. In the survival data analysis, statistical comparisons were performed using Wilcoxon tests for quantitative variables and χ² tests for qualitative variables. Relapse-free survival (RFS) was calculated as the time from surgery to recurrence or death from any cause. If patients were alive or free of recurrent disease at the last follow-up, their data were censored. Overall survival (OS) was calculated as the time from surgery to death from any cause. The data of patients who were alive at the last follow-up were censored. Survival curves for RFS and OS were described using the Kaplan–Meier method, and log-rank tests were utilized for survival analysis. Prognostic values for DGC cases and other clinical factors were assessed using univariate and multivariate analyses based on the Cox proportional hazard regression model. HRs and 95% confidence intervals (CI) were estimated for each factor. All statistical analyses were performed with JMP version 13.1 software (SAS Institute). All P values were two-sided, and P < 0.05 was considered statistically significant.

All data are available within the article. RNA-sequencing data that support the findings of this study have been deposited in DDBJ under the following accession numbers: DRA013546, DRA013547, and DRA014657. Spatial transcriptomic data have been deposited in GEO under the accession number GSE214363.

We performed Azan staining of surgically resected specimens from 473 patients with gastric cancer to evaluate the amount of stroma in gastric cancer tissues and excluded 135 patients with mucinous and early-stage gastric cancer according to the study scheme (Supplementary Fig. S1A). Then, the clinicopathologic examination of the 338 patients with gastric cancer comparing the stroma-high and stroma-low groups revealed that a large amount of stroma was significantly correlated with T3 to T4 stage disease (P < 0.001), lymph node metastasis (P = 0.006), venous invasion (P < 0.001), pStage III–IV (P < 0.001), and the recurrence rate (P < 0.001; Supplementary Table S4). We next conducted survival analysis to evaluate the prognostic impact of the amount of stroma on patients with gastric cancer. Patients with stroma-high tumors had a worse prognosis than those with stroma-low tumors (5-year OS: 58.3% vs. 72.8%, P = 0.004; Supplementary Fig. S1B).

In addition, gastric cancers were histologically divided into intestinal-type gastric cancer (IGC) and diffuse-type gastric cancer (DGC) according to Lauren's classification. The clinicopathologic examination of each histologic type revealed that a large amount of stroma was significantly correlated with an advanced tumor stage, regardless of the histologic type (Supplementary Tables S5 and S6). Moreover, we compared patient prognoses between the stroma-high and stroma-low groups of patients with IGC or DGC. Notably, a significant correlation was not observed between the amount of stroma and the prognosis of patients with IGC (5-year OS, stroma low vs. high groups: 70.3% vs. 68.0%, P = 0.357; Fig. 1A); however, the stroma-high group had a significantly poorer prognosis than the stroma-low group of patients with DGC (5-year OS: 75.5% vs. 45.5%, P = 0.002; Fig. 1B).

On the basis of these findings from clinical analysis, we expected that the difference in cancer–stromal interactions between IGC and DGC would impact prognosis. PDGF–PDGFR signaling is known to be involved in CAF growth and cancer progression, and the prognostic impacts of PDGF ligands have been proven in several types of cancer, including gastric cancer (26, 27). We first examined the significance of PDGF ligands (mRNA level) and found that PDGFC and D expression, but not PDGFA and PDGFB expression, was significantly associated with a poor prognosis in patients with gastric cancer in TCGA cohort (n = 378; Fig. 1C). Moreover, IHC staining revealed that PDGFC and PDGFD protein levels were significantly associated with a poor prognosis for patients with gastric cancer in the Kumamoto cohort (n = 338; Fig. 1D and E). Furthermore, patients with DGC presenting PDGFC- or D-high tumor cells experienced a significantly shorter OS than patients with IGC presenting PDGFC- or D-high tumor cells, whereas no difference in prognosis was observed between patients with DGC and patients with IGC presenting PDGFC- or D-low tumor cells (Supplementary Figs. S2A and S2B). Given this finding, we expected diverse PDGFR expression across the different histologic types of gastric cancer tissues and thus examined PDGFRα and β expression between IGC and DGC tissues. Higher PDGFRα expression was detected in the IGC stroma than in the DGC stroma (Fig. 1F). Notably, PDGFRβ expression was much higher in DGC stroma than in IGC stroma (Fig. 1G). These findings suggest that the PDGF–PDGFRβ interaction in DGC stroma exerts a negative effect on the prognosis of patients with DGC presenting stroma-high tumors.

We next examined PDGFRα and β expression in various CAF lines isolated from gastric cancer tissues. Supporting the finding in gastric cancer tissues, in isolated CAFs, PDGFRβ was consistently expressed, whereas PDGFRα expression was occasionally lost (Fig. 2A and B). The tendency toward PDGFRβ dominant expression was consistent across different CAF lines (n = 9; Fig. 2C). We next examined whether CAF growth mediated by PDGF-CC mainly stimulates PDGFRα and whether CAF growth mediated by PDGF-DD mainly stimulates PDGFRβ. As expected, PDGF-CC and PDGF-DD significantly promoted the growth of CAFs expressing both receptors (Fig. 2A and D), and PDGF-DD significantly promoted the growth of CAFs expressing only PDGFRβ (Fig. 2B and E). Although the mechanism underlying the equilibrium between PDGFRα and β expression has remained unclear, here, for the first time, we found that TGFβ1 stimulation significantly decreased PDGFRα expression and increased PDGFRβ expression in CAFs (Fig. 2F and G). We performed RPPA and RNA sequencing analyses with unstimulated CAFs, CAFs stimulated with PDGF-DD, and CAFs pretreated with TGFβ1 and then stimulated with PDGF-DD to investigate the significance of PDGFRβ induction by TGFβ1 in PDGF-DD/PDGFRβ signaling in CAFs. The RPPA analysis revealed that the levels of p-PDGFRβ, p-p44/42MAPK (T202/Y204), p-MAPKAPK2 (T334), p-S6RP (S240/244), p-SHC, and p-Src (Y416) were significantly increased by the TGFβ1 pretreatment (Fig. 2H). On the basis of these results, the TGFβ1 pretreatment increased tyrosine phosphorylation in downstream molecules in the PDGF-DD/PDGFRβ signaling pathway through PDGFRβ induction upon TGFβ1 stimulation.

In addition, comprehensive gene expression analysis by RNA sequencing showed that the number of DEGs was markedly increased in CAFs with TGFβ1 pretreatment plus PDGF-DD stimulation compared to CAFs with PDGF-DD stimulation (Fig. 2I). Moreover, the pathway analysis of DEGs revealed that gene sets related to the cell cycle, signaling by Rho GTPases, immune response (cytokine signaling) and DNA repair were markedly enriched in CAFs pretreated with TGFβ1 and then stimulated with PDGF-DD compared to control CAFs (Fig. 2J). According to our previous study, the comprehensive genomic meta-analysis identified the correlation between stromal super-module expression and TGFβ pathway activation in gastric cancers (5). Taken together with the clinical findings that PDGFRβ expression is increased in the DGC stroma, these results suggest that TGFβ pathway activation induces PDGFRβ expression and subsequent PDGF-DD/PDGFRβ signaling enhances CAF growth in the DGC stroma.

We first examined the effect of a PDGFRα mAb treatment on CAF growth to confirm the significance of PDGF–PDGFR signaling in CAFs. Although the growth of CAFs stimulated by PDGF-CC was completely blocked by the PDGFRα-specific antibody treatment, only a partial effect of PDGF-DD stimulation on CAF growth was observed due to the remaining activity of PDGFRβ signaling (Fig. 3A). We next examined the effect of the PDGFRβ mAb treatment on CAF growth. As expected, the PDGFRβ mAb suppressed the growth of CAFs stimulated by PDGFD, but not that stimulated by PDGFC (Fig. 3B). In addition, treatment with both antibodies (anti-PDGFRβ and anti-PDGFRα) significantly inhibited CAF growth stimulated by PDGF ligands (Fig. 3C). Because the effect of PDGFRβ mAb was not substantial, we further examined the effect of two types of multikinase inhibitors (sorafenib or regorafenib) that have been used in clinical practice as dual PDGFRα/β inhibitors on CAF growth. These multikinase inhibitor treatments strongly suppressed the growth of CAFs stimulated by PDGF-CC or DD at lower levels than CAFs without PDGF ligand stimulation (Fig. 3D and E). Although we were not able to exclude the effect of multikinase inhibitors on unrelated kinases, these findings suggest that multikinase inhibitors exert a sufficient suppressive effect on CAFs stimulated with PDGF ligands.

To examine the change in immune response-related gene expression by PDGF–PDGFR blockade, we performed RNA sequencing using unstimulated CAFs, CAFs stimulated by PDGF-CC/-DD and CAFs pretreated with sorafenib or regorafenib followed by PDGF-CC/-DD stimulation. Consequently, we identified that among specific genes involved in leukocyte migration, the expression of CXCL1, 3, 5, 8, VEGFA and IL-33 was increased by PDGF-CC/-DD stimulation and remarkably blocked by sorafenib or regorafenib treatment (Fig. 3F). We further confirmed the induction and repression of the expression of these factors by performing qRT-PCR (Fig. 3G). CXCL1, 3, 5, and 8 are known as CXCR2 ligands and are deeply involved in CXCR2+ PMN-MDSC migration into cancer tissues (28, 29). In addition, VEGF family members play a key role in the development of an immunosuppressive tumor microenvironment by inhibiting progenitor cell differentiation to CD4⁺ and CD8⁺ lymphocytes (30) while enhancing the effects of immune suppressive cells such as regulatory T cells (Tregs) and MDSCs (31). IL33 is a key regulator of the anti-inflammatory functions of type 2 immunity involved in tissue fibrosis (32). Given these findings, we expected that multikinase inhibitor treatment could reverse the immune suppression augmented by CAFs.

On the basis of the results from the analysis of human CAFs, we employed a Trp53 gene deletion and K-Ras^G12V-expressing mouse gastric cancer cell line (GAN-KP cell) established from K19-Wnt1/C2mE transgenic (gastric neoplasia, GAN) mice (33–35), and the experimental results showed that both Pdgfc and Pdgfd were expressed at high levels in GAN-KP cells among nine mouse cancer cell lines (Fig. 4A). We first conducted a coculture assay using GAN-KP cells and mouse primary fibroblasts. The expression of Pdgfc, Pdgfd, and Tgfb1, which are intimately involved in fibrosis, was upregulated in KP cells cocultured with fibroblasts (Fig. 4B), whereas the expression of Cxcl1, Cxcl3, and Vegfa was markedly induced in fibroblasts cocultured with GAN-KP cells (Fig. 4C). We isolated splenic CD11b⁺Gr-1⁺ MDSCs from C57/BL6 mice and performed a chemotaxis assay using conditioned medium (CM) from GAN-KP cells, fibroblasts, and GAN-KP + fibroblasts to examine the role of factors secreted from each cell type on MDSC chemotaxis. Consequently, the number of invaded MDSCs through the membrane was drastically increased in the cocultured CM group (Fig. 4D). We further distinguished the two main subsets of MDSCs based on Ly6G/Ly6C expression: CD11b⁺Ly6G⁻Ly6C^high monocytic MDSCs (M-MDSC) and CD11b⁺Ly6G⁺Ly6C^low PMN-MDSCs. Consistent with recent knowledge regarding PMN-MDSCs (36), the chemotaxis of PMN-MDSCs was enhanced by CXCL1 and CXCL3 but not VEGF, whereas M-MDSCs were not affected by any of them (Fig. 4E). In addition, the number of invaded PMN-MDSCs was significantly reduced by treatment with the anti-CXCL1 or anti-CXCL3 antibody (Supplementary Fig. S3). Notably, regorafenib treatment efficiently blocked the expression of Cxcl1, Cxcl3, and Vegfa in fibroblasts cocultured with GAN-KP cells (Fig. 4F), and the chemotaxis of PMN-MDSCs that was enhanced by cocultured CM was significantly suppressed by regorafenib treatment (Fig. 4G). These findings suggest that CXCL1 and CXCL3 expression in fibroblasts activated by coculturing with GAN-KP cells induce the chemotaxis of PMN-MDSCs.

We established a fibrotic tumor model through serial transplantation of GAN-KP cells to further investigate the role of fibroblasts in the formation of an immunosuppressive microenvironment (Supplementary Fig. S4). The numbers of fibroblasts (αSMA-, PDGFRβ-, or COL1A1-positive) and collagen levels in tumors were sequentially increased by serial transplantation (Fig. 5A; Supplementary Fig. S5). We first evaluated the ratio between GAN-KP cells and fibroblasts using flow cytometry immediately before serial transplantation after 1 week of in vitro culture. Consequently, we found that almost all of living cells (99.8%) were GAN-KP cells (Supplementary Fig. S6). In addition, GAN-KP cells and fibroblasts were sorted from parental (S0) and serially transplanted (S6) tumors (Fig. 5B), and we found that Pdgfc/d and Tgfb1 expression increased in GAN-KP cells from S6 tumors compared with those from S0 tumors (Fig. 5C). In association with the increased expression of Tgfb1 in GAN-KP cells from S6 tumors, Pdgfra expression was slightly decreased and Pdgfrb expression was increased in fibroblasts from S6 tumors compared with the levels detected in S0 tumors (Fig. 5D). On the basis of these results, we expect the characteristic change in GAN-KP cells might drive the development of fibrotic features in this mouse model.

Moreover, the expression of Cxcl1, Cxcl3, and Vegfa was significantly higher in fibroblasts than in GAN-KP cells in S6 tumors (Fig. 5E). Together with the increased fibroblasts in serially transplanted tumors, we next evaluated the density of MDSC subsets and infiltrating lymphocytes. The number of PMN-MDSCs was significantly increased in serially transplanted tumors (more than three times of transplantation; S>3) compared with those in S0 tumors, whereas the density of M-MDSCs was not affected by serial transplantation (Fig. 5F). We next evaluated the frequency of tumor infiltrating lymphocytes (TIL) and natural killer cells. The numbers of CD3⁺, CD4⁺CD3⁺, CD8⁺CD3⁺ T cells, and NK1.1⁺CD3⁺ NKT cells were significantly decreased in S>3 tumors compared with S0 tumors. On the other hand, the number of CD25⁺Foxp3⁺ Tregs was significantly increased in S>3 tumors compared with S0 tumors (Fig. 5G). In addition, PD-1 expression, as a marker of T-cell exhaustion, was upregulated in CD3⁺, CD4⁺CD3⁺ T cells but not CD8⁺CD3⁺ T cells (Fig. 5H). However, the frequency of IFNγ⁺ cells among CD8⁺CD3⁺ T cells, indicating CD8⁺ cytotoxic T lymphocytes (CTL), was significantly decreased in S>3 tumors compared with S0 tumors (Fig. 5I). These results suggest that CXCL1 and CXCL3 from increased fibroblasts promote PMN-MDSC accumulation and subsequently attenuate antitumor immunity by effector T cells in addition to the decrease in TILs in serially transplanted fibrotic tumors.

Because serially transplanted GAN-KP tumors induced an immunosuppressive microenvironment, we compared the effect of anti–PD-1 immunotherapy between S0 and S5 GAN-KP tumors. Although anti–PD-1 treatment exhibited significant tumor-suppressive activity in S0 tumors, serially transplanted S5 tumors showed resistance to anti–PD-1 immunotherapy (Supplementary Fig. S7; Fig. 6A and B). On the basis of the effect of regorafenib as a dual PDGFRα/β inhibitor on CAFs in vitro, we next examined whether regorafenib ameliorated the fibrotic tumor microenvironment in S5 GAN-KP tumors. We performed spatial transcriptomics, which enable the characterization of the spatial topography of gene expression, and single-cell RNA sequencing (scRNA-seq) of S5 GAN-KP tumors in mice treated with vehicle or regorafenib to assess this alteration (Fig. 6C). Regorafenib treatment showed tumor-suppressive activity in serially transplanted S5 tumors (Fig. 6D). Spatial transcriptomics revealed that S5 GAN-KP tumors were classified into 10 clusters (Fig. 6E; Supplementary Fig. S8). The tumor stroma was identified on the basis of a high expression level of Col1a2 and allocated into three clusters: clusters 3, 5, and 7 (Fig. 6F). Regarding the key biomarkers expressed in these three clusters, we found that Tnc, Tagln, Mmp10, and Mmp13 were expressed at high levels in cluster 3 (Fig. 6G), and Cxcl12, Pi16, Gsn, and Col14a1 were specifically expressed in cluster 5 (Fig. 6H). Although general fibroblast markers were highly expressed in cluster 7 (Fig. 6I), no specific biomarkers were identified in cluster 7. The pathway analysis revealed that specific pathways related to myeloid cell chemotaxis, particularly granulocyte chemotaxis, were upregulated in cluster 5 compared with the other two subpopulations (clusters 3 and 7; Fig. 6J). Notably, we found that cluster 5 was drastically reduced along with the decrease in tumor stroma following regorafenib treatment (Fig. 6K). PDGFRα expression was quite low in all clusters, whereas PDGFRβ expression was higher in clusters 3 and 5 than that in cluster 7. In addition, PDGFRβ expression was significantly decreased by regorafenib treatment (Fig. 6L).

Moreover, we generated droplet-based 5′ scRNA-seq libraries from three individual tumors per group (vehicle or regorafenib). We obtained scRNA-seq profiles from 20,510 cells that passed quality control. Next, we performed in silico sorting and clustering analysis of cells contained within the tumor to uncover the proportion of each population at the single-cell level. Using cell type–specific canonical markers defined in the literature (37–39), the cells were categorized into 12 cell types (Fig. 6M; Supplementary Fig. S9). Unfortunately, we were not able to detect a sufficient amount of fibroblasts to confirm the diversity due to the low viability. Next, we focused on neutrophils containing PMN-MDSCs, which were particularly increased in serially transplanted fibrotic tumors. We reclustered 6,971 neutrophils and identified five distinct clusters. Furthermore, Arginase-1 (Arg1), which is an enzyme contributing to the establishment of immunosuppression and is abundant in PMN-PDSCs (40, 41), was expressed at high levels in cluster 7 (Fig. 6N and O), suggesting that PMN-MDSCs are enriched in this cluster compared with the other four clusters. Consistent with the finding from the spatial transcriptomics analysis that the stromal subpopulation recruiting granulocytes was efficiently decreased, regorafenib treatment specifically reduced cluster 7 (Fig. 6P) and eliminated Arg1-expressing neutrophils overall (Fig. 6Q). These results suggest that regorafenib treatment reverses the immunosuppressive microenvironment caused by the fibrotic stroma.

As dual PDGFRα/β blockade by regorafenib modified the stroma and immune microenvironment, we further investigated the efficacy of combination treatment with regorafenib and anti–PD-1 immunotherapy in serially transplanted fibrotic tumors. The mice were subsequently treated with IgG control plus vehicle, anti–PD-1 antibody plus vehicle, IgG control plus regorafenib or anti–PD-1 antibody plus regorafenib according to the protocol shown in Fig. 7A. As expected, regorafenib restored the antitumor efficacy of the anti–PD-1 antibody in serially transplanted fibrotic tumors showing resistance to anti–PD-1 immunotherapy (Fig. 7B–D). The number of tumor fibroblasts was significantly decreased by regorafenib treatment (Fig. 7E; Supplementary Fig. S10). Considering that PMN-MDSCs were significantly increased in serially transplanted fibrotic tumors, we next examined each treatment to determine whether any treatments had an impact on the density of MDSC subsets and infiltrating lymphocytes. Although PMN-MDSCs were decreased and M-MDSCs were increased in the treatment groups, the difference was greater in the regorafenib treatment groups, especially the combination therapy group (Fig. 7F). Combination therapy substantially induced tumor infiltration by lymphocytes (Fig. 7G). Notably, the proportion of IFNγ⁺ CTLs in tumors was significantly increased by combination therapy (Fig. 7H). These findings suggest that combination therapy with regorafenib and an anti–PD-1 antibody restored antitumor immunity through stromal reprogramming in fibrotic tumors.

CAFs are a major component of the tumor stroma and enhance tumor progression and drug resistance through various mechanisms (10, 42). The roles of CAFs in the immune microenvironment have also been investigated in several types of cancer. For instance, fibroblast activation protein (FAP)–expressing CAFs, the main source of CXCL12, limit T-cell recruitment into tumor tissue and are involved in the local immunosuppressive environment (43). IL6 from pancreatic CAFs promotes the differentiation of MDSCs in a STAT3-dependent manner and drives immune escape (44). However, it is not clear how CAFs should be targeted therapeutically to improve the immune microenvironment. Here, we provide the rationale for stromal amelioration with immune checkpoint inhibitors and present molecular evidence from a fibrotic tumor mouse model showing that CAFs stimulated with PDGF ligands affect the immunosuppressive microenvironment through the expression of specific chemokines.

PDGFRs are predominantly expressed in stromal cells, not cancer cells. Although PDGF–PDGFR is known as a fibrogenic growth factor, as is TGFβ, the significance of these signaling pathways in cancer progression is still unclear due to the complexity of PDGFRα and β. We identified the interaction between TGFβ and PDGF signaling in this study. The results showed that TGFβ1 stimulation increased PDGFRβ expression and decreased PDGFRα expression in isolated CAFs. Although TGFβ and PDGF are considered to be independent fibrogenic growth factors, there is the possibility that these signaling pathways interact with each other to compensate for missing aspects of gene expression and signaling mechanisms to generate tumor fibrosis. The pathway analysis of RNA sequencing data revealed that gene sets related to the cell cycle, signaling by Rho GTPases, cytokine signaling and DNA repair were markedly enriched in CAFs pretreated with TGFβ1 and then stimulated with PDGF-DD. Activation of receptor tyrosine kinases signaling pathways modulates DNA repair pathways (45). Rac1 and Cdc42 Rho GTPases are downstream of PDGFD/PDGFRB signaling (46). Together with these previous reports, the current results from the pathway analysis suggest that a TGFβ1 pretreatment enhances PDGFD/PDGFRβ signaling by increasing PDGFRβ expression. According to previous evidence, PDGFR-β⁺ fibroblasts are deeply involved in cancer aggressiveness and confer a worse prognosis in various human solid tumors (47–49). Taken together, these results indicate that PDGF–PDGFR signaling, particularly PDGFRβ signaling, might be a potent target to improve the stromal environment in cancer tissues with TGFβ signaling-driven fibrosis.

Several challenges have been addressed in an attempt to deplete or normalize cancer stroma. Immunotherapies targeting FAP peptides have been tested in mouse models to explore whether FAP-expressing stroma can be eliminated, and the results showed a suppressive effect on tumor growth (50, 51). Further investigation is needed to determine whether these strategies can be combined with chemotherapy, radiotherapy, or other immunotherapies. To target the cancer-promoting stroma by drugs, vitamin D receptor (VDR; ref. 52) and all-trans retinoic acid (ATRA; ref. 53) were utilized to suppress ECM remodeling through the reprogramming of pancreatic stellate cells. Although these drugs successfully improved the desmoplastic stroma by restoring homeostatic features, specific targets have not been determined. On the basis of the clinical significance of PDGF–PDGFR signaling in the tumor stroma, we focused on PDGFRs as potent targets to modify the stromal environment. Notably, we found that multikinase inhibitors strongly blocked CAF growth, CAF-derived chemokine production involved in MDSC recruitment and CAF-mediated chemotaxis of PMN-MDSCs. In addition to the suppressive effects in vitro, spatial transcriptomics and scRNA-seq analyses of fibrotic tumor samples revealed that regorafenib treatment restored fibrosis and PMN-MDSC accumulation.

Although immune checkpoint inhibitors are a promising treatment option, the number of patients who benefit from them remains limited. Thus, combination immunotherapy is an attractive strategy for nonresponders. Combining a CSF1R inhibitor with a CXCR2 antagonist blocked granulocyte infiltration mediated by CAFs and enhanced the effect of immunotherapy with a PD-1 antibody (54). With respect to fibrotic tumors, the inhibition of focal adhesion kinase (FAK) reduced fibrosis and increased their responsiveness to immunotherapy (55). The reprogramming of the TME by a FAK inhibitor enhanced the responsiveness to PD-1 checkpoint antagonists through CD8⁺ CTL infiltration into pancreatic ductal adenocarcinoma tumors during therapy (56). In addition, TGFβ signaling drives immune evasion during colorectal cancer progression, and blockade of TGFβ signaling rendered tumors susceptible to anti–PD-1/PD-L1 therapy (57). The findings reported in this study provide evidence of the potential for multikinase inhibitors to be used as another option in combination with anti–PD-1 antibodies to treat severely fibrotic tumors exhibiting resistance to immune checkpoint inhibitors. However, if clinical applications are considered, validation in preclinical models that form metastases will be required. Notably, the safety and efficacy of regorafenib plus an anti–PD-1 antibody for patients with gastric cancer and colorectal cancer were assessed in a phase Ib clinical trial (58). At present, further investigations of this combination are warranted in larger cohorts of patients with gastric cancer and colorectal cancer. Our preclinical evidence of combination therapy in a mouse model of a fibrotic tumor may provide the rational for patient selection in a clinical setting.

In conclusion, PDGF stimulation markedly increased the growth of CAFs and the expression of CXCLs, which were involved in PMN-MDSC recruitment. Tumors with severe fibrosis showed a decrease in tumor-infiltrating lymphocytes, PMN-MDSC accumulation, and resistance to anti–PD-1 antibodies. Moreover, dual PDGFRα/β blockade restored the immunosuppressive microenvironment through stromal modification and exerted a synergistic effect with anti–PD-1 antibody treatment on fibrotic tumors (Fig. 7I). The findings from the present study provide molecular evidence that the stromal amelioration induced by PDGFRα/β inhibition is a promising strategy for use in combination therapy with immune checkpoint inhibitors.

K. Ohnishi reports grants from Grant-in-Aid for Scientific Research (C) from Japan Society for the Promotion of Science (JSPS) during the conduct of the study. M. Masuda reports grants from Japan Society for the Promotion of Science during the conduct of the study. No disclosures were reported by the other authors.

T. Akiyama: Data curation, formal analysis, validation, investigation, writing–original draft. T. Yasuda: Data curation, formal analysis, validation, investigation, writing–original draft. T. Uchihara: Data curation, formal analysis, validation, investigation, writing–original draft. N. Yasuda-Yoshihara: Data curation, formal analysis, methodology. B.J.Y. Tan: Software, formal analysis, methodology. A. Yonemura: Formal analysis, validation, methodology. T. Semba: Data curation, software, formal analysis. J. Yamasaki: Investigation, methodology. Y. Komohara: Supervision, validation. K. Ohnishi: Supervision, validation. F. Wei: Resources. L. Fu: Resources. J. Zhang: Resources. F. Kitamura: Resources. K. Yamashita: Methodology. K. Eto: Resources. S. Iwagami: Resources. H. Tsukamoto: Supervision, methodology. T. Umemoto: Data curation, investigation, methodology. M. Masuda: Data curation, investigation, methodology. O. Nagano: Data curation, validation. Y. Satou: Data curation, supervision, validation. H. Saya: Data curation, supervision. P. Tan: Data curation, supervision, validation. H. Baba: Resources, data curation, supervision, validation, writing–review and editing. T. Ishimoto: Conceptualization, data curation, supervision, funding acquisition, validation, writing–original draft, project administration, writing–review and editing.

The authors thank S. Usuki (Liaison Laboratory Research Promotion Center, IMEG, Kumamoto University) for assisting with the RNA sequencing and T. Moroishi (Department of Cell Signaling and Metabolic Medicine, Faculty of Life Sciences, Kumamoto University) for kindly providing the murine cancer cell lines. This work was supported by the FOREST program of the Japan Science and Technology Agency (JST, grant no. JPMJFR200H to T. Ishimoto), the Japan Society for the Promotion of Science (JSPS, KAKENHI grant nos. 20H03531, 21K19535, and 21KK0153 to T. Ishimoto; 20K08985 to T. Yasuda; 21K16384 to T. Umemoto; and 20K07690 to M. Masuda), the Eli Lilly and Company, the Naito Foundation, the Shinnihon Foundation of Advanced Medical Treatment Research, and the Inter-University Research Network for Trans-Omics Medicine program at the Institute of Molecular Embryology and Genetics at Kumamoto University.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.