Arid5a Promotes Immune Evasion by Augmenting Tryptophan Metabolism and Chemokine Expression

Immunotherapy is a paradigm shift in the treatment of advanced cancers (1). Particularly, immune checkpoint blockade (ICB), which can induce long-lasting tumor regression and appears to be more effective than any other type of treatment for patients with tumors that have already metastasized. However, the majority of tumors show resistance to ICB (2). Although a possible explanation for this is the plasticity and heterogeneity of tumors and their microenvironment, there are still many issues to be understood about the mechanism by which advanced cancers avoid immunosurveillance.

Tumors are generally classified as hot or cold based on the abundance of tumor-infiltrating lymphocytes (TIL; ref. 3). Hot tumors are T-cell inflamed and demonstrate an initial immune response, which is weakened over time by the upregulation of immune checkpoints or by an increase in the number of suppressive immune cells. In contrast, cold tumors lack a sufficient number of preexisting TILs, and are thus considered nonimmunogenic (4, 5). Immune evasion is a crucial cancer hallmark, often caused by increased activation of an immune checkpoint pathway, such as the programmed death 1 (PD-1)–PD ligand 1 (PD-L1) axis, or accumulation of immunosuppressive metabolites (6). Hot tumors fully or partially respond to ICB. Cold tumors, on the other hand, are refractory to most immunotherapies. A noninfiltrated “cold” tumor microenvironment (TME) is associated with various epithelial cancers, such as colorectal carcinoma and pancreatic ductal adenocarcinoma (PDAC; ref. 7). In particular, a subtype of these cancers that shows the tumor-intrinsic mesenchymal phenotype has been reported to facilitate immune evasion via crosstalk with stromal immune cells in the TME (2).

The TME comprises tumor cells, stromal cells, and immune cells. Stromal and immune cells affect tumor growth by interacting directly with tumor cells or by modulating innate and adaptive immunity in the TME (8). Small molecules, including kynurenine (Kyn), adenosine, and chemokines secreted by the tumor and stromal cells, recruit immunosuppressive cells, such as myeloid-derived suppressor cells (MDSC) and regulatory T cells (Treg) to the TME (8, 9). Immune responses are further modulated by the inhibition of T-cell activation or by the emergence of T-cell exhaustion within the TME (7, 10).

Indoleamine 2,3-dioxygenase 1 (Ido1) is a rate-limiting metabolic enzyme that promotes an immunosuppressive microenvironment, and high levels of Ido1 expression are associated with advanced cancers (9, 11). Ido1 catabolizes tryptophan (Trp) to produce Kyn, which inhibits effector T-cell activation and promotes Treg differentiation (9, 11). Pharmacologic inhibition of Ido1 results in T cell–dependent antitumor responses in murine models, but not complete tumor regression (11, 12). Chemokine signals, such as those triggered by the CCR2–CCL2 axis, recruit neutrophils, monocytes, and MDSCs into the TME and contribute to the progression of various cancers (13, 14). However, therapies designed to interfere with either CCR2 or CCL2 showed disappointing results in clinical trials (14). Thus, understanding the expression pattern and regulation of immune metabolites and chemokines has the potential to lead to the development of novel strategies for cancer treatment.

AT-rich interactive domain 5a (Arid5a) binds to the 3′-UTR of several immune-related mRNAs, including Il6 Stat3, Tbet, and Ox40 (15, 16). The importance of Arid5a in inflammatory and autoimmune responses is highlighted by the fact that Arid5a-deficient mice are resistant to experimental autoimmune encephalomyelitis, bleomycin-induced lung injury, and septic shock (15, 16). However, the role of Arid5a in cancer was not studied.

In this study, we found that Arid5a was an essential molecule in regulating immune evasion in models of PDAC and colorectal carcinoma. Our findings provide new insights toward understanding the roles of Arid5a in the progression of mesenchymal tumor subtypes.

C57BL/6 mice and nu/nu BALB/c mice were purchased from CLEA Japan. All experiments were performed in accordance with the guidelines approved by the Animal Care and Use Committees of Immunology Frontier Research Center and the Research Institute for Microbial Diseases, Osaka University (Osaka, Japan).

HPAF-II, Capan-2, BxPC-3, Panc-1, Mia-PaCa-2, SW1990, DLD1, Caco-2, LS174T, HCT116, SW480, and SW620 cells were from the ATCC. Mia-PaCa-2, Panc-1, Caco-2, LS174T, HCT116, SW480, and SW620 cells were cultured in DMEM (Nacalai, catalog no. 08456–36) containing 10% FCS (HyClone, catalog no. SH30071.03); BxPC-3, SW1990, and DLD1 cells were cultured in RPMI1640 medium (Nacalai, catalog no. 30264–56) containing 10% FCS; Capan-2 cells were cultured in McCoy 5A medium (Sigma, catalog no. M9309) containing 10% FCS; and HPAF-II cells were cultured in MEM (Gibco, catalog no. 11095080) containing 10% FCS. KPC cells were prepared from a Pdx1-cre; LSL-KrasG12D; LSL-Trp53R172H mouse (17), as described previously (18), and cultured in DMEM containing 10% FCS. MC38 cells were purchased from Kerafast and cultured in DMEM containing 10% FCS and supplemented with 2 mmol/L glutamine (Gibco, catalog no. 25030081), 0.1 mmol/L nonessential amino acids (Gibco, catalog no. 11140050), 1 mmol/L sodium pyruvate (Gibco, catalog no. 11360070), and 10 mmol/L HEPES (Nacalai, catalog no. 17557–94). 293FT cells were from Invitrogen (catalog no. R70007) and were cultured according to the manufacturer's instructions. Plat-E cells were a gift from Dr. Kitamura (Tokyo University, Tokyo, Japan), they were cultured in DMEM containing 10% FCS. All cell lines used in the experiments were obtained from 2012 to 2020. Cell viabilities were measured using Cell Counting Kit-8 (Dojindo, catalog no. CK04), according to the manufacturer's instructions. All cells used were determined to be free of Mycoplasma using 4′,6-diamidino-2-phenylindole (DAPI, Sigma-Aldrich, catalog no. D9542) before starting the experiments, and no antibiotics were added to the cell cultures throughout the experiments to avoid the latent infection of Mycoplasma. All cell lines used in this study were cultured for less than 1 month for the experiments described.

To generate Arid5a-knockout (KO) cells, KPC and MC38 cells were transfected with one of three plasmids from Santa Cruz—Arid5a/Mrf-1 CRISPR/Cas9 (catalog no. sc-431969), Arid5a/Mrf-1 HDR (catalog no. sc-431969-HDR), or control CRISPR/Cas9 (catalog no. sc-418922; Supplementary Table S1)—using UltraCruz Transfection Reagent (Santa Cruz Biotechnology, catalog no. sc-395739) according to the manufacturer's instructions. Two days after transfection, cells were cloned by limiting dilution in 96-well plates and single-cell clones were selected and analyzed by genomic PCR to verify Arid5a coding sequence integrity using primers listed in Supplementary Table S1 and per the protocol in “Genomic PCR.” Arid5a depletion was confirmed by quantitative PCR and immunoblotting

The DNeasy Blood and Tissue Kit (Qiagen, catalog no. 69506) was used to isolate total DNA from wild-type (WT) and KO cells (KPC and MC38). KOD FX (Toyobo) was used according to manufacturer's protocol for genomic PCR. PCR was performed at 98°C for 10 seconds, 60°C for 30 seconds followed by 30 cycles and 68°C for 3 minutes. Primers used in the experiments are provided in Supplementary Table S1.

Mouse xenograft experiments were performed according to our previously published protocol (18). Briefly, KPC (2 × 10⁵) or MC38 cells (1 × 10⁵) in 100 mL of 50% growth factor reduced BD Matrigel Matrix (BD Biosciences, catalog no. 354230) were injected into both flanks of either nu/nu BALB/c mice (8- to 10-week-old females) or C57BL/6 mice (8- to 10-week-old females) and tumor growth was measured twice a week. Tumor volume was calculated using the formula, volume = (a × b²)/2, in which a and b are the largest and the smallest tumor diameters, respectively.

Tumors were dissected from the skin of mice at the designated time points and chopped into smaller fragments. The tumor fragments were then transferred to gentleMACS C tubes (Miltneyi Biotec, catalog no. 130–093–237) containing 2.5 mL of tumor dissociation enzyme solution (Mouse Tumor Dissociation Kit, Miltneyi Biotec, catalog no. 130–096–730). Tubes were processed on gentleMACS octo-Dissociator (Miltenyi Biotec, catalog no. 130–095–937) using the 37°C_mTDK_2 protocol designated for hard tumor tissue. Before further procedures, cell suspensions were filtered using 40-μm cell strainers (Falcon, Corning brand, catalog no. 352340) and were centrifuged at 500 × g for 10 minutes at 4°C. The pellets were then resuspended in 8 mL of 40% Percoll (GE Healthcare, catalog no. 17–5445–01) and centrifuged at 500 × g for 10 minutes at 4°C. Supernatants were aspirated, then the cell pellets were quenched with 10 mL RPMI containing 10% FBS and used for subsequent analyses.

For the enrichment of CD45⁺ leukocytes, after washing the cells from tumor tissues in FACS buffer (PBS +2% FBS), CD45⁺ (TIL) MicroBeads, mouse (Miltenyi Biotec, catalog no. 130–110–618) were added according to the manufacturer's instructions. The samples were incubated at 4°C for 15 minutes, then the cells were washed and resuspended in 2 mL of FACS buffer. The autoMACS Pro Separator (Miltenyi Biotec, catalog no. 130–092–545) was used to isolate CD45⁺ leukocytes according to the manufacturer's instructions.

Antibodies were conjugated to all metals other than platinum using the MaxPar Conjugation Kit (catalog no. 2011698) according to the manufacturer's instructions. Platinum-labeled antibodies were conjugated with cisplatin as described previously (19). Metal isotopes were obtained from Fluidigm with the exception of Indium(III)chloride-113 and Indium(III)chloride-115 (Trace Sciences) and Cisplatin-195 and -196 (BuyIsotope) as shown in Supplementary Table S2. Items from Trace sciences and BuyIsotype were custom orders. The conjugated antibodies were stored in PBS-based antibody stabilizer (Candor Biosciences, catalog no. 130050). All antibodies were titrated for optimal staining concentrations with control murine tissues. Full details of the antibodies used and the metals to which they were conjugated are provided in Supplementary Table S2.

For each tumor sample, 1.5 × 10⁶ or fewer cells were labeled with metal-conjugated CD45-specific antibodies (Supplementary Table S2). Individual tumor samples were incubated with CD45–113In, CD45–115In, CD45–194Pt, CD45–195Pt, CD45–196Pt, or CD45–198Pt antibodies together with Fc-block reagent, and fluorochrome-labeled antibodies for 30 minutes in room temperature, and then washed twice in CyFACS buffer (PBS with 0.1% BSA and 2 mmol/L EDTA).

Barcoded cells were pooled together and washed in CyFACS buffer then stained with a metal-conjugated surface stain antibody cocktail for 45 minutes at room temperature. Cells were then washed twice in CyFACS buffer, stained for viability with the cisplatin analogue dichloro-(ethylenediamine) palladium(II) (20) in PBS for 5 minutes at room temperature, and then fixed and permeabilized using the Foxp3 Transcription Factor Staining Buffer Set according to the manufacturer's protocol (eBiosciences, catalog no. 00–5523–00). Cells were subsequently stained with a metal-conjugated intracellular antibody cocktail for 45 minutes at 4°C, and then washed twice in CyFACS buffer and once in PBS. Cells were then fixed overnight in 1.6% paraformaldehyde solution containing DNA Cell-ID Intercalator-103Rh (Fluidigm, catalog no. 201103A). All antibodies used for mass cytometry are listed in Supplementary Table S2.

Prior to data acquisition, cells were washed once in CyFACS buffer and twice in MilliQ H₂O. Cells were then diluted to 1 × 10⁶ cells/mL in MilliQ H₂O containing 15% EQ Four Element Calibration Beads (Fluidigm, catalog no. 201078) and filtered. Cells were acquired at a rate of 200 to 300 cells/second using a Helios mass cytometer (Fluidigm). Flow Cytometry Standard files were normalized to EQ bead signal.

For analysis of mass cytometry results, gating and manual debarcoding was performed using CytoBank software. Live CD45⁺ live cells from all samples in each analysis were clustered together by Clustering Large Analysis (CLARA) using the Statistical Single-Cell Analysis by Fixed Force- and Landmark-Directed (SCAFFoLD) R package (21). Statistical analysis using the Significance Across Microarrays algorithm was performed within statistical SCAFFoLD. SCAFFoLD maps were then generated with manually gated populations as landmarks. Secondary analysis of gated CD3⁺TCRb⁺B220⁻CD11b⁻NK1.1⁻CD4⁺/CD8a⁺ cells was performed in the same manner, with the exception that X-shift clusters generated within Vortex software (22) were used as the landmarks to help visualize major subpopulations within the SCAFFoLD map. In all experiments, clustering and map generation were performed on all samples to enable the direct comparison between tumor models. Heatmaps and dendrograms were made by heatmap.2 within the gplots package of R software.

Freshly harvested MACS-sorted CD45⁺ lymphocytes, generated as described in “MACS enrichment of CD45⁺ lymphocytes,” were plated onto 96-well U-bottom plates in FACS buffer. The cell suspensions were blocked with CD16/CD32 antibody (Tonbo Biosciences, catalog no. 70–0161-M001). Then, cells were incubated in a cocktail of Live/Dead near-IR stain (Invitrogen, catalog no. L10119) and surface antibodies prepared in FACS buffer for 30 minutes. For intracellular staining, cells were permeabilized with the FoxP3 transcription factor staining buffer kit (eBioscience, catalog no. 00–5523–00) according to the manufacturer's instructions. Cells were then incubated in a cocktail of intracellular antibodies prepared in 1X Perm/Wash buffer (eBioscience, catalog no. 00–8333–56) for 45 minutes, and then washed twice in Perm/Wash buffer prior to data collection. Samples were collected on an LSR Fortessa (BD Biosciences), and compensation and data analysis were carried out with FlowJo vX.0.7 (TreeStar). For t-distributed stochastic neighbor embedding analysis, precompensated and gated data were exported from FlowJo. All surface and intracellular antibodies used in this study are listed in Supplementary Table S1.

MACS-sorted CD45⁺ lymphocytes from mouse tumors were plated onto 96-well U-bottom plates. A cocktail of Phorbol 12-myristate 13-acetate (50 ng/mL; Sigma, catalog no. P1585), Ionomycin (0.5 μg/mL; Sigma, catalog no. I3909) and Golgi Stop (1:1,000 dilution; BD Biosciences, catalog no. 554724) was prepared in RPMI containing 10% FBS. The cells were incubated in the cocktail for 2 hours at 37°C, and then processed for FACS staining.

Total RNA was extracted from WT or KO KPC and MC38 cells using RNeasy Mini kit (Qiagen, catalog no. 74134) according to the manufacturer's instructions. Preparation of cDNA libraries was performed using a TruSeq stranded mRNA sample prep kit (Illumina, catalog no. 20020594) according to the manufacturer's instructions. Sequencing was performed on an Illumina HiSeq 2500 platform in the 75-base single-end mode. Illumina Casava1.8.2 software was used for basecalling. Sequenced reads were mapped to the mouse reference genome sequences (mm10) using TopHat version 2.0.13 in combination with Bowtie2 version 2.2.3 and SAMtools version 0.1.19. The fragments per kilobase of exons per million mapped fragments (FPKM) was calculated using Cuffnorm version 2.2.1. Analysis of the main functions regarding gene expression changes was performed with Ingenuity Pathway Analysis software (IPA; Qiagen, https://www.qiagenbioinformatics.com/products/ingenuity-pathway-analysis/). The data have been deposited in the NCBI Gene Expression Omnibus (GEO) database under GEO accession number GSE159967.

Total RNA was extracted from cells (WT or KO or shRNA-treated or Ido1-overexpressing or Ccl2-overexpressing KPC, and WT or KO or shRNA-treated MC38) using RNeasy Mini kit according to the manufacturer's instructions. RNAs (15 ng each) were then reverse-transcribed using the SuperScript VILO Master Mix (Invitrogen, catalog no. 11756050), and Quantitative reverse transcription PCR (qRT-PCR) reactions were performed in triplicate using TaqMan Universal PCR Master Mix and TaqMan gene expression assays (catalog no. 4331182, Applied Biosystems), and measured using 7300 Real Time PCR System (Applied Biosystems). Delta C_t values, which were the C_t values of Gapdh subtracted from the C_t values of each gene, were used for normalization, and relative mRNA levels were calculated by 2^ΔΔC_t method. Primers for the genes analyzed, Ido1, Ccl2, and Arid5a, and the normalizing gene Gapdh are listed in Supplementary Table S1.

For stable short hairpin RNA (shRNA)–mediated gene silencing, pLKO.1-puro-based recombinant lentiviruses, either generated by ourselves or purchased from Addgene, were used. In brief, shRNAs were constructed in pLKO.1-puro from the Sigma Mission shRNA library, as listed in the Supplementary Table S1. A control scramble shRNA in pLKO.1-puro (Addgene, catalog no.1864) were transfected into 293FT cells, together with the envelope plasmid pMD2.G (Addgene, catalog no. 12259) and the packaging plasmid psPAX2 (Addgene, catalog no. 12260), using Lipofectamine LTX (Invitrogen, catalog no. 15338100) according to the manufacturer's instructions. Forty-eight hours after the transfection, culture supernatants were harvested, filtered through 0.45-μm filters (Advantec), and the resultant lentivirus preparations were then applied onto target cells in the presence of Polybrene (8 μg/mL; Nacalai catalog no. 12996–81). After 24 hours, 4 μg/mL of puromycin (Nacalai catalog no. 14861–71) was added to the cell culture for one week to select infected cells.

Cells were lysed on ice with RIPA buffer (1% Nonidet P-40, 150 mmol/L NaCl, 20 mmol/L Tris–HCl, pH 7.4, 5 mmol/L EDTA, 1% Na-deoxycholate, 0.1% SDS, 1 mmol/L Na₃VO₄, 10 μmol/L Na₂MoO₄, 5 μg/mL aprotinin, 2 μg/mL leupeptin, 3 μg/mL pepstatin A and 1 mmol/L phenylmethylsulfonyl fluoride). Protein concentrations were determined using a DC protein assay kit (Bio-Rad Laboratories, catalog no. 5000111JA) with BSA as a standard. Protein samples (20 μg each) were separated by SDS-PAGE and transferred to a polyvinylidene difluoride membrane (Millipore). Membranes were then incubated with the antibodies, indicated in the Supplementary Table S1, followed by horseradish peroxidase–conjugated anti-mouse IgG (GE Healthcare, catalog no. NA931V). Specific binding was detected using Chemi-Lumi One Ultra, according to the manufacturer's instructions (Nacalai, catalog no. 11644). Data are shown as representative results of at least two independent experiments.

WT and KO KPC cells were stimulated with or without IFNγ (2.5 ng/mL; R&D Systems, catalog no. 485-MI) and the supernatants were collected at 0, 24, 48, and 72 hours after stimulation. The concentration of Kyn in cell supernatants was measured by the Kyn ELISA kit (ImmuSmol, #BA E-2200), according to the manufacturer's instructions.

WT and KO KPC or MC38 cell supernatants were collected at day 2 and the concentration of Ccl2 was measured using the Mouse CCL2/JE/MCP-1 immunoassy Kit (R&D Systems, catalog no. MJEOOB), according to the manufacturer's instructions.

Serum and tumor samples were collected at 10 and 21 days after the implantation of tumors, respectively. The collected samples were immediately frozen in liquid nitrogen and stored in a freezer until analysis. Tumors were homogenized in 4-fold (v/w) of saline, and serum samples were added to acetonitrile containing 1 μg/mL internal standard [Tryptophan-d5 (catalog no. DLM-1092–0.5, Kynurenine-d6 (catalog no. CLM-9884-PK), and Kynurenic acid-d5 (catalog no. DLM-7374–0.01), purchased from Cambridge Isotope Laboratories Inc.], and centrifuged at 13,000 rpm for 5 minutes at 4°C. Supernatants were injected into the LC/MS-MS system for analysis. The NexeraX2 liquid chromatography system (Shimadzu) with QTRAP 4500 mass-spectrometer (AB SCIEX) equipped with an electrospray ionization (ESI) source was used for the quantification of Trp, Kyn, and kynurenic acid. The analytes were separated on an Intrada amino acid column (2.1 mm × 50 mm; 3.0 μm particle size; Imtakt), with the column temperature set at 40°C. The mobile phase consisting of an aqueous phase (A: 100 mmol/L ammonium formate in water) and an organic phase (B: 0.1% formic acid in acetonitrile) was used for a gradient elution at a flow rate of 0.5 mL/minute. The HPLC elution program was as follows: 80% B (0.5 minute) → 0% B (linear decrease in 1.5 minutes) → 80% B (linear increase in 0.5 minute) → 80% B (0.5 minute). Multiple reaction monitoring was performed using individually optimized conditions (Supplementary Table S3).

WT or KO KPC cells were stimulated with IFNγ (5 ng/mL) for 6 hours, which is shown as the zero-time point. Cells were then treated with actinomycin D (2 μg/mL; Sigma, #A9415), and total RNA was harvested at the indicated times using RNeasy Mini kit. Ido1 transcripts were quantified by qRT-PCR analysis, and the amount of the Ido1 transcripts at 0 minute were normalized to 100%. The percentage of mRNA remaining at serial time points compared with the 0 minute time point were plotted and the half-lives of cytokine mRNAs were determined by nonlinear regression curve fitting (one-phase decay) using GraphPad software.

To generate the luciferase plasmid constructs, DNA fragments of the 3′-UTRs of the Ido1 and Ccl2 mRNAs were prepared through PCR amplification using cDNAs of KPC cells, as a template. The PCR cycling conditions were 98 °C for 2 minutes, followed by 30 cycles of 98°C for 10 seconds, 60°C for 30 seconds, and 68°C for 3 minutes, and single cycle of 68°C for 10 seconds. The PCR products were subcloned into the pmirNanoGlo luciferase plasmid (Promega, catalog no. CS194105) using the In-Fusion HD Kit (Takara, catalog no. Z9649N) according to the manufacturer's instructions. The primers used for amplifying the cDNAs for cloning are listed in the Supplementary Table S1. Arid5a-deficient KPC cells cultured in 24-well plates were transfected with either the pmirNanoGlo luciferase plasmid encoding the 3′-UTR region, or the pmirNanoGlo luciferase control plasmid (50 ng). All transfections were performed in combination with Arid5a pcDNA3-Flag (600 ng) or the empty pcDNA3-Flag vector (600 ng; Supplementary Table S1). The pmirNanoGlo luciferase control plasmid was used as a negative control, and the empty vector was used to equalize the total amount of DNA. Transfections were performed using the Lipofectamine 3000 kit (Thermo Fisher, catalog no. L3000015). After 8 hours, cells were lysed using passive lysis buffer. Luciferase activity was measured using the Nano-Glo Dual-Luciferase Reporter Assay System (Promega, catalog no. N1630). Firefly luciferase activity was normalized by Renilla luciferase activity, and expressed as the fold stimulation relative to the activity in vector-transfected cells.

The RNA binding assay was performed using RiboTrap Kit (MBL, cat. #RN1011) according to the manufacturer's instructions. Briefly, the 3′-UTRs of Ido1 and Ccl2 mRNA were amplified by PCR and subcloned into the pBluescript SK(-) plasmid, from Dr. Sabe (Hokkaido University, Sapporo, Japan), using the In-Fusion HD kit. To modify these RNAs, using Riboprobe Systems (Promega, catalog no. P1440), 5-bromo-UTP (BrU) included in the RiboTrap Kit, was randomly incorporated into the 3′-UTR of Ido1 or Ccl2 upon in vitro transcription. The quality and concentration of BrU-labeled RNAs were assessed using the Bioanalyzer 2100 (Agilent) with the RNA 6000 Nano Kit (Agilent, catalog no. 5067–1511). Anti-BrU was conjugated to Invitrogen Dynabeads (Invitrogen, catalog no. 10004D) overnight at 4°C, after which the beads were washed with the wash buffer provided in the RiboTrap Kit. Anti-BrU–conjugated beads were then incubated for 3 hours at 4°C with the BrU-labeled RNAs. Then, recombinant Arid5a protein was transferred to the mixture followed by 2 hours of incubation at 4°C. The samples were then washed, eluted, and subjected to SDS-PAGE.

Normalized RNA-sequencing (RNA-seq) expression data for pancreatic cancer (Study ID: QCMG) were downloaded from cBioPortal (23, 24). Then, 96 samples tagged as ICGC were extracted for the subtype analysis mentioned in the article (25).

For colorectal cancer, data were obtained via Synapse (ID:syn2623706) mentioned in the article (26) and 512 samples tagged as colorectal cancer in The Cancer Genome Atlas (TCGA) were analyzed. Dunn multiple comparisons test was used for the comparison of mRNA expression levels among the cancer subtypes.

For Spearman correlation analysis between ARID5A and other genes, datasets of TCGA PanCancer Atlas (168 and 592 samples for pancreas and colon cancer, respectively) were used and their analyses were conducted on the cBioPortal site (https://www.cbioportal.org/).

Immunofluorescence microscopy analysis was performed according to the method described previously (18). Briefly, WT KPC cells were incubated with or without TGFβ (5 ng/mL; R&D Systems, catalog no. 240-B) for 48 hours before fixation with 4% paraformaldehyde in PBS (with no detergent). Cells were then subjected to immunostaining using an anti–E-cadherin antibody (Supplementary Table S1), followed by an Alexa Fluor 488–conjugated secondary antibody (Molecular Probes, catalog no. A11006). F-actin was visualized using Texas Red-X-conjugated phalloidin (Molecular Probes, catalog no. T7471), and nuclei were stained with DAPI before mounting with 50% glycerol in PBS. Fluorescence images were obtained using a confocal laser-scanning microscope (Model A1R with NIS-Elements, Nikon) using a CFI Plan Apo VC 60X H oil-immersion objective with an NA of 1.4, and analyzed using the attached software. Data were collected from two independent experiments, each analyzing at least 10 cells. Images were handled using Photoshop version 7 (Adobe).

Statistical significance was calculated between two groups by the Student unpaired t test. One-way ANOVA followed by the Tukey Honestly Significant Difference posttest was used to calculate statistical significance between multiple groups. Statistical significance of the survival data was calculated by Kaplan–Meier with log-rank analysis. Analyses were performed using GraphPad Prism 8. Error bars represent the SEM, and a P value of less than 0.05 was considered to indicate a statistically significant difference between groups.

The unpaired t test, one-way ANOVA (Dunn multiple comparison test), and two-way ANOVA were performed using GraphPad software. Figure legends specify the test used, criteria for statistical significance, and the number of experimental, biological, and technical replicates.

We first analyzed the expression of Arid5a in commercially available human pancreatic cancer cell lines retaining epithelial and quasi-mesenchymal (QM) characteristics (27, 28). The HPAF-II and Capan-2 cell lines show epithelial phenotypes, whereas Panc-1 and Mia-PaCa-2 cells have the QM phenotype. BxPC3 and SW1990 cells retain both epithelial and mesenchymal characteristics, resulting in partial epithelial-to-mesenchymal transition (EMT; ref. 29). We found that mesenchymal-like and partial EMT cell lines showed much higher expression levels of ARID5A than epithelial-like cell lines (Fig. 1A). The tumor-intrinsic mesenchymal phenotype has been reported to facilitate immune evasion via crosstalk with stromal immune cells in the TME (1, 2). Among PDACs, tumor subtypes with QM characteristics are associated with highly aggressive disease (25). Utilizing RNA-seq data from human PDAC patients (25), we found that QM and immunogenic PDACs expressed significantly higher levels of the ARID5A transcript than the other PDAC subtypes (Fig. 1B). Consistent with this, analysis of PDAC RNA-seq data from TCGA demonstrated that ARID5A expression significantly correlated with the expression of cytokines associated with EMT induction, such as TGFB1 and IL6, and of representative EMT-associated transcription factors (EMT-TF), such as ZEB1, ZEB2, SNAI1, SNAI2, TWIST1, and TWIST2, but not STAT3, which has been shown to be involved in IL6-induced Arid5a expression in macrophages and mouse embryonic fibroblasts (MEF; refs. 15, 16; Supplementary Fig. S1A). Moreover, using KPC cells we found that TGFβ stimulation induced EMT-like morphologic changes in KPC cells (Supplementary Fig. S1B), and, under these conditions, promoted the expression of Arid5a (Supplementary Fig. S1C). These results indicate that EMT-TFs might be involved in the upregulation of ARID5A expression in the mesenchymal subtypes of PDAC.

To study the possible roles of Arid5a in tumor progression, we generated Arid5a KO KPC cells using the CRISPR/Cas9 system (Fig. 1C; Supplementary Fig. S1D–S1F). RNA-seq analysis of these cells demonstrated that EMT-TFs and EMT markers, such as Zeb1, Zeb2, Snai1, Snai2, Twist2, Acta2, and Itgb1, were significantly downregulated in KO KPC cells compared with WT KPC cells, whereas expression of the representative epithelial marker E-cadherin (Cdh1 gene) was substantially increased (Supplementary Fig. S1G). Using IPA, we found that signaling pathways associated with EMT and metastasis, such as the regulation of EMT by growth factors/development, IL8, oncostatin M, and stemness signals, were decreased upon the deletion of Arid5a (Supplementary Fig. S1H). In addition, the IL6, Stat3, and JAK/Stat signaling pathways were downregulated in KO KPC cells (Supplementary Fig. S1H). Activation of the IL6/JAK/STAT3 signal upregulates EMT-TFs and enhances metastasis via the induction of EMT (29). Therefore, our data indicate the involvement of Arid5a in inducing the mesenchymal properties of the partial EMT cell line KPC.

We next subcutaneously injected WT or KO KPC tumor cells into immunocompetent mice and monitored tumor growth. KO KPC xenografts showed significantly less tumor growth than WT xenografts, indicating a possible role for Arid5a in tumor progression (Fig. 1D). We also analyzed whether Arid5a acted in a cell-intrinsic or an immune cell–dependent manner to promote tumor growth, by injecting WT or KO KPC cells subcutaneously into immunodeficient mice. KO KPC tumors showed comparable growth to WT tumors (Fig. 1E), indicating that Arid5a promotes PDAC progression in an immune cell–dependent manner.

To corroborate the role of Arid5a in immune evasion, we performed mass cytometry (Cytometry by Time-Of-Flight: CyTOF) analysis of immune cells infiltrating the TME of WT and KO KPC tumors. Characterization and quantification of the expression of 28 surface markers and 8 intracellular markers (6 of which were TFs) among a wide range of immune cell populations demonstrated that the genetic ablation of Arid5a in KPC cells promoted an antitumor immune response (Fig. 2A and B). Among the infiltrated immune cells, immunosuppressive granulocytic MDSCs (gMDSC), also known as polymorphonuclear MDSCs (30), were much more abundant in the TME of WT KPC tumors than in the TME of KO KPC tumors (Fig. 2A; Supplementary Fig. S2A and S2B). Furthermore, we found a larger number of Ki67⁺ actively proliferating dendritic cells (DC), natural killer (NK) cells, natural killer T (NKT) cells, and T cells in the KO KPC-derived tumors than in the WT KPC-derived tumors (Fig. 2A and B). Global analysis demonstrated that numerous clusters of T cells were substantially expanded upon the deletion of Arid5a, including effector memory (EM; CD44⁺CD62L⁻) CD4⁺ T cells, central memory (CM; CD44⁺CD62L⁺) CD8⁺ T cells, and EM (CD44⁺CD62L⁻) CD8⁺ T cells (Fig. 2A and B). By performing in-depth analysis of the CD4⁺ and CD8⁺ T-cell compartments, we found that the exhausted PD1^hiTbet⁺ EM Th1 cell population was significantly decreased in the TME of KO KPC tumors and this PD1^hi cell population was replaced by PD1^int and PD1^lo EM Th1 cells in the TME of KO KPC tumors (Fig. 2C and D). Consistent with a supportive role for Arid5a in immunosuppression, we also found that activated KLRG1⁺ Tregs were significantly reduced in the TME of KO KPC tumors compared with WT KPC tumors (Fig. 2C and D). Furthermore, flow cytometry analysis demonstrated that tumor-infiltrating Tregs were significantly decreased on day 14 and 21 in the TME of KO KPC tumors (Supplementary Fig. S2C and S2D). Like CD4⁺ T cells, the number of PD1^hiLAG3^hiCD39⁺CD8⁺ T EM cells with an exhaustion-like phenotype was significantly decreased in the TME of KO KPC tumors compared with WT KPC tumors (Fig. 2E and F). These findings suggested that Arid5a facilitates the immune evasion of PDAC by promoting intratumoral infiltration of suppressive immune cells (gMDSCs and Tregs) and exhausted T cells.

We further characterized the cytotoxic activity of CD4⁺ and CD8⁺ T cells by analyzing the production of granzyme B (GzmB) and cytokines (Supplementary Fig. S3A). Flow cytometry analysis demonstrated that a significantly greater proportion of TILs were CD8⁺ T cells upon the deletion of Arid5a in KPC cells and that CD4⁺ T cells were reciprocally decreased (Supplementary Fig. S3B). Effector CD8⁺ T cells infiltrating KO KPC tumors expressed significantly higher levels of GzmB compared with TILs of WT tumors (Fig. 3A), and effector CD4⁺ T cells infiltrating KO KPC tumors showed a similar trend (Supplementary Fig. S3C and S3D). Concurrently, infiltrating effector CD4⁺ and CD8⁺ T cells in KO KPC tumors significantly induced the production of IFNγ, a potent antitumor cytokine (Fig. 3B and C). Taken together, our results indicate that Arid5a intrinsically suppresses the cytotoxic activity of CD4⁺ and CD8⁺ T cells in WT KPC tumors.

We next sought to identify the molecular mechanisms underlying the immune evasion mediated by Arid5a. Comparing the gene expression profiles of KO KPC cells and WT cells by RNA-seq demonstrated substantially lower expression levels of genes associated with inflammation and immune evasion, as manifested by the downregulation of Stat3, IL6ra, Ido1, Ido2, Adora2a, and chemokine-encoding genes, in KO KPC cells (Supplementary Fig. S4A). Using IPA, we identified that the deletion of Arid5a decreased canonical signaling pathways associated with the hepatic fibrosis signal, proinflammatory response (NF-κB, IL1, IL6, acute phase response), cell proliferation, invasion (Stat3, P13K/AKT, ERK/MAPK, PDGF), and Trp degradation III. In contrast, tumor suppressor PTEN signaling pathway genes were increased in KO KPC cells (Supplementary Fig. S4B).

Increased levels of Ido1 in both mouse and human tumor models contributes to tumor immune evasion (9, 11). Analysis of PDAC RNA-seq data from TCGA showed a positive correlation of ARID5A and IDO1 and the significant association of IDO1 with a poor prognosis (Supplementary Fig. S4C and S4D). These results highlight Ido1 as a possible downstream target of Arid5a in the immune evasion of pancreatic tumors.

Ido1 expression in the TME promotes tumorigenesis via the induction of numerous tolerogenic immune phenotypes, such as the suppression of effector T-cell activation and the enhanced infiltration of Tregs and MDSCs (8, 9, 30). The enzyme Ido1, which is upregulated by IFNγ, promotes Kyn production by catalyzing the amino acid Trp (31). Upon IFNγ stimulation, KPC cells expressed both Ido1 mRNA and protein; this induction was significantly decreased upon the deletion of Arid5a (Fig. 4A and B). IFNα and IFNβ did not induce the expression of Ido1 in KPC cells (Supplementary Fig. S4E). Consistent with these data, KO KPC cells stimulated with IFNγ released significantly smaller amounts of Kyn into the supernatant than WT cells (Fig. 4C). Using LC-MS/MS we found that Kyn levels were significantly decreased in both the sera and tumors of mice 10 days after inoculation with KO KPC cells (Fig. 4D). On day 21, Kyn production was significantly decreased in the serum and showed a decreasing tendency in KO KPC tumors (Fig. 4E). The production of kynurenic acid, a downstream metabolite of Kyn, was significantly suppressed in the tumors of mice with KO KPC xenografts on day 10 and 21, but not in the sera (Supplementary Fig. S4F). Similarly, Trp levels were significantly increased in KO KPC tumors on day 21 (Supplementary Fig. S4G). Collectively, these results suggest that Arid5a regulates Ido1-mediated Trp metabolism in KPC cells.

We have shown previously that Arid5a promotes the inflammatory response underlying various diseases through stabilization of the mRNAs of several genes, such as IL6, Stat3, Tbet, and IL17-induced targets, and contributes to the inflammatory response and a variety of diseases (15, 16). Here, we found that the half-life of Ido1 mRNA was significantly decreased in IFNγ induced KO KPC cells treated with actinomycin D compared with similarly treated WT cells (Fig. 4F). We then generated a luciferase vector encoding the 3′-UTR region of Ido1 to determine whether Arid5a stabilizes Ido1 mRNA. We analyzed the luciferase activity of Ido1 by transiently expressing Flag-tagged full-length Arid5a (Arid5a-FL), an Arid5a mutant lacking the Arid domain (Arid5a-ΔM), or an empty vector in KO KPC cells. We found that Arid5a stabilized Ido1 mRNA (Fig. 4G; Supplementary Fig. S4H). To determine the direct binding of Arid5a to Ido1 mRNA by RNA pulldown, we immunoprecipitated recombinant Arid5a proteins using the bromo-UTP–incorporated 3′-UTR of Ido1, or an empty vector, and found that Arid5a binds directly to Ido1 mRNA (Fig. 4H; Supplementary Fig. S4I).

IFNγ-mediated JAK/Stat1 signaling activates Ido1 transcription (31, 32). We found that the expression kinetics of tyrosine-phosphorylated Stat1 (pYStat1) and Stat1 were comparable between WT and KO KPC cells (Supplementary Fig. S4J). Furthermore, the silencing of Stat1 expression substantially reduced Ido1 expression, but did not affect Arid5a expression (Supplementary Fig. S4K and S4L). Taken together, these results indicate that pYStat1 might activate the initial transcription of Ido1, followed by the posttranscriptional stabilization of Ido1 mRNA by Arid5a in KPC cells.

Consistent with our in vitro data, we found that the reconstitution Arid5a-FL in KO KPC cells (Arid5a^hi KO) recovered tumor growth to a comparable level to that of WT cells in C57BL/6 mice, whereas KPC cells reconstituted with Arid5a-ΔM (Arid5a^ΔM KO) demonstrated tumor growth comparable to that of KO KPC cells. On the other hand, all four cell lines showed similar growth in immunodeficient nu/nu BALB/c mice (Fig. 4I and J; Supplementary Fig. S4M). These results indicated that the Arid domain was important for the immune evasion mediated by Arid5a. To investigate the in vivo biological role of the Arid5a-mediated Ido1–Kyn axis, we generated two stable cell lines using lentiviral transduction; Ido1 knockdown KPC cells (shIdo1) and Ido1-expressing KO KPC cells (Ido1^hi KO cells; Supplementary Fig. S4N and S4O). We found that mice inoculated with shIdo1 KPC cells showed suppressed tumor growth compared with mice inoculated with WT cells (Fig. 4K). Consistent with a role for Ido1 as a downstream regulator of Arid5a, Ido1^hi KO KPC cells partially recapitulated the growth of KO KPC cells (Fig. 4M). However, Ido1^hi KO KPC did not fully recover the tumor growth of KO KPC cells, indicating that Arid5a regulates KPC growth through other downstream targets beside Ido1 (Fig. 4M). In contrast, nu/nu BALB/c mice showed comparable tumor growth when inoculated with WT, KO KPC cells, or Ido1^hi KO KPC (Fig. 4L and N). These results demonstrate that Arid5a-mediated tumor growth in KPC cells is partially regulated through Ido1-mediated Trp metabolism.

In response to the specific chemokines that are expressed by tumor cells and stromal cells, various immune cell subsets migrate into the TME and regulate tumor immune responses in a spatiotemporal manner (14). We found that chemokines, including Ccl2, Ccl5, Ccl7, Ccl8, and Cxcl3, were expressed at lower levels in KO KPC cells than in WT cells (Fig. 5A). To test whether chemokine signaling was involved in Arid5a-mediated immune evasion, we analyzed the PDAC data from TCGA. We found that expression of CCL2, CCL5, CCL7, and CCL8 positively correlated with ARID5A expression, whereas expression of CXCL3 did not (Supplementary Fig. S5A). KPC cells expressed the Ccl2 transcript at a high basal level compared with other chemokines (Supplementary Fig. S5B). Consistent with these data, protein levels of Ccl2 in KO KPC cell supernatants were significantly lower than in WT cell supernatants (Fig. 5B).

An increased level of CCL2 is a poor prognostic indicator in patients with PDAC (33). Ccl2 plays a pivotal role in MDSC recruitment and tumor progression in colorectal carcinoma, glioma, renal cancer, lung cancer, liver cancer, and melanoma models (13, 14, 34), but this is not seen in pancreatic cancer. Other studies show that MDSCs mediate Treg development via TGFβ and IL10 production in the TME (35). Thus, we analyzed the role of Arid5a in regulating Ccl2. We found that Arid5a stabilizes Ccl2 mRNA (Fig. 5C), and binds directly to the 3′-UTR of Ccl2 mRNA (Fig. 5D; Supplementary Fig. S4I).

To investigate the biological role of Arid5a-mediated Ccl2 stabilization in KPC tumor growth in vivo, we generated two stable cell lines using lentiviral transduction; Ccl2 knockdown KPC cells (shCcl2) and Ccl2-expressing KO KPC cells (Ccl2^hi KO; Supplementary Fig. S5C and S5D). We found that shCcl2 KPC cells showed reduced tumor growth in immunocompetent mice compared with WT KPC cells (Fig. 5E). The rate of tumor growth of Ccl2^hi KO KPC cells was similar to that of WT cells in immunocompetent mice (Fig. 5G). In contrast, shCcl2 KPC and Ccl2^hi KO KPC tumor growth was comparable to WT tumor growth in immunodeficient mice (Fig. 5F and H). These results suggest that the posttranscriptional stabilization of Ccl2 mRNA is a common mechanism controlling Arid5a-mediated tumor progression.

We next sought to determine whether our findings in PDAC reflected a general mechanism common to various tumors. We performed immunoblotting analyses to investigate the expression of ARID5A in human colorectal carcinoma cell lines, including DLD1, Caco-2, LS174T, HCT116, SW480, and SW620. Consistent with PDAC cells lines, colorectal carcinoma cell lines with a partial-EMT phenotype, namely, HCT116, SW480, and SW620 cells, expressed high levels of the ARID5A protein (Supplementary Fig. S6A). Analysis of a previously published colorectal cancer RNA-seq dataset (26), showed that cells with consensus molecular subtype 4 (CMS4), a mesenchymal subtype of colorectal carcinoma, expressed significantly higher levels of ARID5A than the other subtypes (Supplementary Fig. S6B). Like PDAC, analysis of colorectal carcinoma data from TCGA demonstrated that expression of ARID5A was weakly but statistically significantly correlated with the expression of TGFB1, IL6, and representative EMT-TFs (Supplementary Fig. S6C).

The MC38 mouse model of colorectal cancer is highly aggressive (36, 37). MC38 tumors show extensive infiltration of immunosuppressive cells, such as MDSCs, and exclusion of T cells (38, 39). To investigate the role of Arid5a in MC38 tumor progression, we generated KO MC38 cells using CRISPR/Cas9 (Fig. 6A; Supplementary Fig. S6C and S6D). Tumors formed by Arid5a-deficient MC38 cells showed remarkably reduced growth in syngeneic mice compared with WT MC38 cells, but not in immunodeficient mice (Fig. 6B and C). These results indicated that as in our models of PDAC, Arid5a did not play a crucial intrinsic role in MC38 cell tumor growth in vivo, but that it might be involved in immune evasion by MC38 cells. As in PDAC, we found the frequency of immunosuppressive MDSC subsets, mainly gMDSCs, was reduced in KO MC38 tumors compared with WT MC38 tumors (Fig. 6D; Supplementary Fig. S7A and S7B), whereas the frequencies of DCs, NK, NKT, CM CD8⁺, EM CD8⁺, and EM CD4⁺ T cells were increased, indicating activation of antitumor immune responses in mice upon the deletion of Arid5a in inoculated tumors (Supplementary Fig. S7A and S7B). Similarly, KLRG1 and CD44-expressing intratumoral CD103⁺-activated Treg subsets were significantly reduced, as they were in the microenvironment of KO KPC tumors (Supplementary Fig. S7C and S7D). The subsets of Tbet⁺ EM cells were significantly increased in KO MC38 tumors showing an upregulation of CD4⁺ T cells with effector-like phenotypes. On the other hand, naïve (CD62⁺CD44⁻) CD4⁺ T-cell subsets were expanded in KO MC38 tumors, likely owing to differentiation of naïve CD4⁺ T cells into multiple effector subsets that can mediate various types of antitumor immunity (ref. 40; Supplementary Fig. S7C and S7D). EM cells circulate through tumor tissues and are known to have cytotoxic properties (41). Infiltration of the Ly6C⁺PD1^hi EM subset of CD8⁺ T cells was upregulated in KO MC38 tumors (Supplementary Fig. S7E and S7F). However, the expansion of naïve CD8⁺ T cells was downregulated upon Arid5a deletion, indicating that naïve cells might have differentiated into EM subsets. These results suggest a general role for Arid5a in promoting tumor immune evasion through disrupting the balance between immunosuppressive and antitumor immune cells in the TME.

To understand the common underlying mechanism regulated by Arid5a, we monitored the expression of Ido1 and chemokines in WT and KO MC38 cells. In contrast to a previous report (42), IFNγ-induced Ido1 protein expression was not detected in MC38 tumors (Supplementary Fig. S8A). Analysis of the colorectal carcinoma RNA-seq dataset from TCGA demonstrated that ARID5A expression correlated with the levels of chemokine transcripts, such as CCL2, CCL5, CCL7, and CCL8 (Supplementary Fig. S8B), as confirmed by comparing the RNA-seq data of WT and KO MC38 cells (Supplementary Fig. S8C). Similar to KPC cells, Ccl2 expression was markedly decreased in KO MC38 cells compared with the other chemokines (Fig. 6E). In addition, WT MC38 cells expressed higher levels of Ccl2 than the other chemokines (Supplementary Fig. S8D), indicating Ccl2 as a possible downstream target of Arid5a. Consistent with these data, protein levels of Ccl2 in WT MC38 cell supernatants were significantly higher than those in KO MC38 cell supernatants (Fig. 6F). Next, we generated four stable cell lines using shRNA lentiviral transduction; that is, MC38 KO cells expressing Arid5a-FL (Arid5a^hi KO), Arid5a-ΔM (Arid5a^ΔM KO), or Ccl2 (Ccl2^hi KO), and Ccl2 knockdown MC38 cells (shCcl2). We found that KO MC38, Arid5a^ΔM KO, and shCcl2 cells showed suppressed tumor growth in immunocompetent mice compared with WT cells (Fig. 6G; Supplementary Fig. S8E). Notably, the rate of tumor growth was a similar level to that of WT cells in MC38 Arid5a^hi KO and Ccl2^hi KO cells (Fig. 6G). In contrast, in immunodeficient mice, tumors derived from all of our generated MC38 cell lines showed comparable growth to WT tumors (Fig. 6H). Thus, Arid5a promotes the immune evasion of MC38 tumors through the regulation of Ccl2 transcripts. These results suggest that the posttranscriptional stabilization of Ccl2 mRNA is a common mechanism controlling Arid5a-mediated tumor progression.

In this study, we demonstrated that Arid5a expression is associated with the mesenchymal phenotypes of PDAC and colorectal carcinoma, and that Arid5a is involved in immune evasion by promoting tumor infiltration by gMDSCs and Tregs, and by suppressing the recruitment and activation of antitumor effector T cells. Mechanistically, Arid5a augments the expression of Ido1 and Ccl2 through posttranscriptional stabilization of Ido1 and Ccl2 mRNAs by binding to their 3′-UTRs. Together, our results indicate that Arid5a acts as a dual regulator in malignant tumors, such as the mesenchymal subtypes of PDAC and colorectal carcinoma, to generate an immunosuppressive TME. First, Arid5a induces the suppressive effects of Ido1 on T cells via a reduction in intratumoral Trp concentration and promotes Treg differentiation/activation. Second, Arid5a induces the expression of chemokines in the TME that recruit immunosuppressive cells, such as Treg and gMDSCs, to the TME. Thus, our findings provide insights into the molecular basis of immune evasion by PDAC and colorectal cancer via Arid5a, and indicate that Arid5a is a promising target for tumor immunotherapy.

Previous studies indicate a link between the mesenchymal properties of tumor cells and immunosuppression, but the precise molecular mechanisms that confer mesenchymal tumor subtypes with resistance to immunosurveillance remain unknown (2). Our study demonstrated that Arid5a expression correlated with the mesenchymal subtypes of PDAC and colorectal cancer, such as the QM and CMS4 subtypes, respectively, and that ARID5A expression significantly correlated with the levels of TGFB1, IL6, and representative EMT-TFs, but not STAT3, in PDAC and colorectal carcinoma. Furthermore, during TGFβ-induced EMT, Arid5a expression was augmented in KPC cells, whereas it has been previously reported that Arid5a expression in macrophages and MEFs is enhanced by Stat3 through IL6 stimulation or by NF-κB via lipopolysaccharide stimulation (15, 16). Our results strongly indicate that EMT-TFs are involved in the upregulation of ARID5A in the mesenchymal subtypes of PDAC and colorectal carcinoma. The molecular mechanisms as to how ARID5A expression is regulated in mesenchymal tumors needs to be determined in the future.

It is well established that tumor infiltration by gMDSCs and Tregs leads to tumor growth and metastasis (8, 9, 30). Similarly, prognostic analyses of tumor-infiltrating immune cells show that infiltration of CD8⁺ T, Th1, Tfh, B, and NK cells, and DCs into a tumor is associated with a favorable prognosis in most cancers (1, 43). In this regard, we found that infiltration of immunosuppressive gMDSCs and Tregs was suppressed, whereas that of antitumor immune cells, such as DCs, NK, NKT, and T cells, was increased in Arid5a-deleted KPC tumors. Furthermore, the presence of cytotoxic and effector memory T cells correlates with increased survival of cancer patients (43). PD1^hiCD8⁺ T cells share features with tissue-resident memory T cells, and show an aberrantly activated status with an apoptosis-prone potential (44). In our current study, cells demonstrating exhaustion-like characteristics (Ly6C⁻PD1^hi EM CD8⁺ T cells) were prominent in WT KPC tumors. The deletion of Arid5a in KPC cells suppressed the infiltration of this exhausted population. Tissue-resident (CD103⁺) memory (Trm) CD8⁺ T cells have direct tumor-killing ability and enhanced cytotoxic responses, thereby strengthening antitumor immunity (45, 46). Schenkel and colleagues showed that Trm cells express IFNγ, which triggers the production of local inflammatory chemokines and induces the recruitment of circulating memory CD8⁺ T cells (47). In our study, Trm cells were significantly increased in KO KPC tumors, highlighting a role for Arid5a in tumor-intrinsic immune evasion. Furthermore, the Tbet⁺ EM CD4⁺ T-cell population with antitumor activity was larger in KO KPC tumors than in WT tumors. MC38 showed similar results to KPC tumors, highlighting the role of Arid5a in the immune evasion of mesenchymal tumor subtypes.

Expression of IFNγ and GzmB correlates with the effector functions of CD8⁺ and CD4⁺ T cells (40, 48). Previous single-cell RNA-seq analyses show the presence of cytotoxic effector CD4⁺ T cells in non–small cell lung and hepatocellular carcinoma (49, 50). In addition, cytotoxic CD4⁺ subsets are clonally expanded in bladder tumors and can kill autologous tumors in an MHC class II–dependent manner (51). In our current study, KO KPC tumors showed increased secretion of IFNγ and GzmB by both CD4⁺ and CD8⁺ T cells, suggesting more robust T-cell antitumor activity upon the deletion of Arid5a. Together, our data suggest that Arid5a is involved in creating an immunosuppressive TME by recruiting gMDSC and Treg cells, and by attenuating the infiltration of tissue-resident EM CD8⁺ and Tbet⁺ EM CD4⁺ T cells.

We found that Arid5a promotes immune evasion by PDAC through regulating Trp metabolism and chemokine production in the TME. The expression of Ido1 in tumor tissues produces Kyn through Trp metabolism, which ultimately activates the Aryl hydrocarbon receptor (AhR) and promotes an immunosuppressive microenvironment. The interaction between Kyn and AhR drives the generation of Tregs, and AhR activation leads to the extensive mobilization of gMDSCs (52, 53). Therefore, understanding the regulatory mechanisms governing Ido1 expression and its downstream Kyn production is expected to provide pharmacologically novel strategies to effectively target this metabolic pathway, which has the potential to reverse tumor immunosuppressive mechanisms. Previous studies show the initiation of Ido1 transcription through IFNγ-mediated JAK/Stat1 signaling (31, 32). We further demonstrated that Arid5a posttranscriptionally stabilizes Ido1 mRNA by binding to its 3′-UTR.

Previous work shows that absence of the CCR2–CCL2 axis increases tumor infiltration by CD103⁺ DCs and activates CD8⁺ T cells, and decreases infiltration by gMDSCs (54). Similarly, PDACs express a high level of CCL2, which creates an immunosuppressive TME by recruiting monocytes, and blockade of CCR2 restores antitumor immunity in preclinical models (55). In this study, we showed that Arid5a posttranscriptionally stabilizes Ccl2 mRNA, resulting in high expression levels of Ccl2 in the TME.

The 3′-UTRs of Ido1 and Ccl2 have conserved adenylate-uridylate–rich elements (56, 57). However, the posttranscriptional alterations and functions of these genes remain to be clarified. Taken together, our findings support a dual role for Arid5a in orchestrating the immune suppression of cancer; that is, a metabolic role in which Arid5a promotes the Ido–Kyn–AhR signaling axis and an immunomodulatory role in which Arid5a enhances Ccl2/CcR2 signaling. We propose that Arid5a links creation of an immunosuppressive microenvironment with maintenance of the mesenchymal phenotype in cancer cells, and that this orchestrates the immunopathology of the TME.

J.B. Wing reports personal fees and nonfinancial support from Fluidigm Corporation outside the submitted work. T. Itokazu is an employee of Osaka University and Neuro-Medical Science Lab. This lab is one of the "Research Alliance Laboratories" and is established within the university to undertake joint research with Mitsubishi Tanabe Pharma Corporation (MTPC). No disclosures were reported by the other authors.

G. Parajuli: Conceptualization, resources, data curation, software, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. M. Tekguc: Data curation, investigation. J.B. Wing: Resources, data curation, formal analysis, methodology, performed the CyTOF analysis. A. Hashimoto: Data curation, investigation, writing–review and editing. D. Okuzaki: Data curation, performed RNA-seq analysis. T. Hirata: Investigation, performed LC-MS/MS analysis. A. Sasaki: Investigation, performed LC-MS/MS analysis. T. Itokazu: Investigation, performed LC-MS/MS analysis. H. Handa: Data curation. H. Sugino: Data curation. Y. Nishikawa: Resources. H. Metwally: Resources, methodology. Y. Kodama: Resources. S. Tanaka: Data curation. H. Sabe: Data curation. T. Yamashita: Investigation, performed LC-MS/MS analysis. S. Sakaguchi: Formal analysis. T. Kishimoto: Conceptualization, resources, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing. S. Hashimoto: Conceptualization, resources, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing.

The authors thank Y. Gemechu, K.K. Nyati, S. Kang, C.-Y. Tsai, S. Sakakibara, and H. Kikutani for critical discussion; T. Akagi for the pCX4 vector; H. Inoue and M. Okawa for their assistance; and H.A. Popiel for critical reading of the manuscript.

This work was supported by the Kishimoto Foundation. S. Hashimoto was supported by Grants-in-Aid from the Ministry of Education, Science, Sports and Culture of Japan (17K07151). H. Metwally was supported by KAKENHI Research Activity Start-up grant (19K23864).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.