T Cell-Intrinsic Vitamin A Metabolism and Its Signaling Are Targets for Memory T Cell-Based Cancer Immunotherapy

Memory T cells play an essential role in infectious and tumor immunity. Vitamin A metabolites such as retinoic acid are immune modulators, but the role of vitamin A metabolism in memory T-cell differentiation is unclear. In this study, we identified retinol dehydrogenase 10 (Rdh10), which metabolizes vitamin A to retinal (RAL), as a key molecule for regulating T cell differentiation. T cell-specific Rdh10 deficiency enhanced memory T-cell formation through blocking RAL production in infection model. Epigenetic profiling revealed that retinoic acid receptor (RAR) signaling activated by vitamin A metabolites induced comprehensive epigenetic repression of memory T cell-associated genes, including TCF7, thereby promoting effector T-cell differentiation. Importantly, memory T cells generated by Rdh deficiency and blocking RAR signaling elicited potent anti-tumor responses in adoptive T-cell transfer setting. Thus, T cell differentiation is regulated by vitamin A metabolism and its signaling, which should be novel targets for memory T cell-based cancer immunotherapy.


INTRODUCTION
The efficient generation of memory T cells is critical to develop effective cancer immunotherapy and robust protective immunity against infectious diseases. Accumulating evidence demonstrates that nutrient metabolism plays a critical role in T cell differentiation; glycolysis and fatty acid oxidation accelerate effector and memory T cell differentiation, respectively (1)(2)(3)(4). Thus, it may be possible to guide T cells in a favorable direction by manipulating nutrient metabolism (4).
Vitamin A is an essential nutrient for reproduction, development, cell differentiation, and vision (5)(6)(7)(8)(9). Vitamin A (retinol, ROL) is metabolized into retinoic acid (RA) via retinaldehyde (also known as retinal, RAL) by two oxidation steps that are strictly regulated by retinol dehydrogenases (RDHs) and retinaldehyde dehydrogenases (RALDHs) (9). Vitamin A and its metabolites are multifunctional and can positively or negatively regulate the acquired immune response. RA, the most active form of these metabolites, enhances the extra-thymic induction of regulatory T (Treg) cells (10,11) and is required for the proper function of effector CD8 + T cells (12) as well as the development of both Th1 and Th17 cells (13). Thus, vitamin A metabolites can influence T cell function and differentiation. However, it remains unsolved whether T cell itself can metabolize vitamin A, and how the metabolites regulate the T cell differentiation.
In this study, we describe that T cell itself can metabolize vitamin A into RAL by RDH10, a rate-limiting enzyme for RA biosynthesis, and that lack of the vitamin A metabolism by Tcell-specific Rdh10 knockout enhances the induction of memory, especially, central memory T cells (T CM ) in Listeria infection model, indicating that vitamin A metabolites such as RAL and RA regulate T cell differentiation. Furthermore, we describe that RA comprehensively induces repressive chromatin states and deletes the memory T cell profile through retinoic acid receptor (RAR), thereby promoting effector T-cell differentiation. Conversely, blockage of RAR signaling suppresses terminal T cell differentiation and efficiently increases T CM with a potent anti-tumor immunity. Thus, vitamin A metabolism and RAR signaling are novel targets to enhance anti-cancer immunity.

LM-OVA and Infection
LM-OVA (14) was kindly provided by Dr. H. Shen (University of Pennsylvania, Philadelphia, PA, USA). LM-OVA was prepared as described previously (14,15) and injected into the tail vein. Unless otherwise stated, mice were infected with 2 × 10 4 colonyforming units (CFU) of LM-OVA on the day after adoptive transfer.

Adoptive Transfer and Isolation of Lymphocytes
For the adoptive transfer of naïve T cells, CD8 + T cells (OT-I cells) were isolated from the spleen of naive Rdh10 f/f Cd4Cre Rag1 -/-OT-I CD45.1 + CD45.2 + (Rdh10CKO) or Rdh10 f/f Rag1 -/-OT-I CD45.1 + (control) mice using the Pan T Cell Isolation Kit II (Miltenyi Biotec, Bergisch Gladbach, Germany). For the adoptive transfer of memory T cells, CD45.1 + OT-I cells were isolated from pooled cells of the spleen and lymph nodes from several mice >30 days post-infection by magnetic selection using biotin-conjugated anti-CD45.1 mAb and streptavidin beads. The isolated cells were stained with anti-CD45.2-PE mAb and Rdh10CKO (CD45.1 + CD45.2 + ) and control (CD45.1 + ) memory OT-I cells were separately sorted by FACSAria. The isolated Rdh10CKO and control OT-I cells were mixed at a ratio of 1:1 (0.5-1 × 10 4 each) and then transferred into sex-matched C57BL/6J (CD45.2 + ) mice after the confirmation of the ratio by flow cytometry.

Adoptive T-Cell Therapy Using Rdh10CKO Memory T Cells
For assessment of anti-tumor activity of Rdh10CKO and control memory OT-I cells that were generated in LM-OVA infection model as described above, six-week-old B6 mice were irradiated (3 Gy) and then subcutaneously injected on the left side of abdomen with 2.0 × 10 6 OVA-expressing EL-4 (EG-7) cells, immediately followed by i.v. injection with 1.5 × 10 5 memory OT-I cells. In this experiment, we set a humane endpoint; mice were sacrificed when tumor volume reached 15 mm in length. For measurement of tumor volume, blinding was not performed but there were two measurers to obtain accurate data.

Reagents, Antibodies, and Flow Cytometry
All recombinant interleukins (ILs) were purchased from PeproTec (London, UK), except for human IL-2 (Shionogi, Osaka, Japan). All retinoids (all-trans form), LE540, and UVI3003 were purchased from Sigma-Aldrich (St. Louis, MO, USA), Wako (Osaka, Japan), and Tocris Bioscience (Ellisville, MO, USA), respectively. All antibodies used in this study are listed in Supplementary Table 4. Cells were incubated with FcR blocking Reagent (Miltenyi Biotec) and then stained with the appropriate combinations of mAbs and fluorochromeconjugated streptavidin. For intracellular cytokine staining, cells were incubated with or without SIINFEKL peptide (1 mM) or PMA + ionomycin in the presence of brefeldin A for 4 h and stained using the Cytofix/Cytoperm Kit (BD Bioscience, Franklin Lakes, NJ, USA). Foxp3-positive cells and apoptotic cells were detected using the BD Foxp3 Staining Kit (BD Bioscience) and Apoptosis Detection Kit (BD Bioscience), respectively. For cell surface-stained samples, 7-aminoactinomycin D (7AAD; eBioscience) was added just before the flow cytometric analysis to exclude dead cells. Flow cytometry was performed using FACSAria. The data were analyzed using FlowJo software (TreeStar, San Carlos, CA, USA).

Homing Assay
T cells were isolated from splenocytes of Rdh10 f/f Cd4Cre and Rdh10 f/f mice and labeled with either CFSE or CellTrace Violet (CTV). Both labeled cells were mixed at a ratio of 1:1 (4 × 10 6 per each) and transferred into sex-matched 5-7-week-old B6 mice after the actual ratio was determined by flow cytometry. Sixteen hours later, the ratio of transferred cells from the spleen and lymph nodes was determined by flow cytometry.

Culture of Human Cells and Assessment of Their Functions
CD4 + CD45RO + T cells were freshly isolated from PBMCs of healthy donors using the BD IMag Cell Separation System, stimulated with plate-bound anti-CD3 and soluble anti-CD28 mAbs, and cultured in the presence of IL-2 (20 IU/ml) and RA (1 mM), LE540 (10 mM), or dimethyl sulfoxide (DMSO). Seven days later, the cells were harvested, washed twice, and cultured in an IL-2-free medium overnight. On the next day, the cell phenotypes and functions were evaluated, as described below. To induce apoptosis, the cells were incubated with anti-Fas Ab (100 ng/ml) for 4 h. To assess their proliferation and reconstitution capacity in vitro, the cells were labeled with CellTrace Violet (CTV; Thermo Fisher Scientific) and then cultured in the presence of IL-7 (5 ng/ml) or anti-CD3/CD28 mAbs. For reconstitution in NOG mice, the cells (1 × 10 6 ) were co-transferred with CD3 + cell-depleted autologous PBMCs (2 × 10 6 ) isolated using CD3 Microbeads and the LD Column (Miltenyi Biotec) into 5-to 7-week-old female mice. Four weeks later, splenocytes were prepared from the mice and the reconstituted T cells were evaluated by flow cytometry. For assessment of T cell differentiation in RDH10-overexpressed and -silenced T cells, CD4 + CD45RO + T cells infected with lentivirus vector were purified by sorting and stimulated with anti-CD3/CD28 mAbs in serum-free medium supplemented with the indicated concentration of all-trans ROL and IL-2. Jurkat cells were cultured in the presence of Retinoid (1 mM) or DMSO for 4 days. In treated Jurkat cells, the expression of CD62L was measured at the protein and mRNA levels by flow cytometry and real-time PCR, respectively. All experiments were performed in duplicate or triplicate.

Anti-Tumor Activity and Persistency of Tumor Antigen-Specific Human T Cells in NOG Mice
B10-T cells were stimulated with plate-bound anti-CD3 (2 mg/ ml) and soluble anti-CD28 (2 mg/ml) mAbs, and cultured in the presence of IL-2 (40 IU/ml) and RA (1 mM), LE540 (10 mM), or dimethyl sulfoxide (DMSO). Three days later, the cells were harvested, washed twice, and further cultured for 4 days in 40 IU/ml of IL-2-supplemented medium. For assessment of antitumor activity, NOG mice were subcutaneously injected with HLA-A*24:02-and Luciferase gene-expressing K562 leukemia cell line (K562-A24-Luc, 3 × 10 5 cells). After confirmation of engraftment by In Vivo Imaging System IVIS (Xenogen, Alameda, CA) on day 3, mice were irradiated (2 Gy) and then transferred with LE540-or DMSO-treated B10-T cells (1 × 10 7 cells) and modified WT1 235 peptide-pulsed CD3-depleted autologous PBMCs (5 × 10 6 cells), followed by ip injection of recombinant human IL-2 (500 IU) on day 4. Mice were allocated into groups of equal average base line tumor burden prior to treatments. Tumor volumes were evaluated with IVIS imaging system. In some experiments, blood was collected from tail vein at the same time as that of measurement of tumor volumes and examined for frequency of the transferred GFP + CD8 + T cells by using flow cytometry. Cell number of transferred GFP + CD8 + T cells in 1 × 10 6 living cells were estimated with both frequency of transferred T cells (hCD45 + hCD3 + cells) and frequency of GFP + hCD8 + cells in the transferred T cells.

Preparation and Transfection of the Viral Vector
Human RDH10 (NM_172037.4) and mouse Rdh10 (NM_133832.3) cDNA were isolated from PBMCs and splenocytes, respectively. The RARa open reading frame (NP_000955) was sub-cloned from the original plasmid (pFN21AE1591, Promega KK, Tokyo, Japan) into the pFC14K plasmid (Promega). Lentivirus vectors encoding full-length RARa, RARa D81-153 (RARa-dDBD), or RARa403 (RARa-dAF2) (16) with a C-terminal Halo-tag were generated using the In-Fusion Cloning Kit (TaKaRa Bio Inc, Shiga, Japan). Lentiviruses were generated using CSII-EF-MCS-IRES2-Venus (for overexpression), CS-RfA-EG (for shRNA expression), pCAG-HIVgp, and pCMV-VSVG-RSV-Rev by a standard method. Target sequences for shRNAs are listed in Supplementary Table 5. Retroviruses were generated from the plat-E packaging cell line by the transduction of MSCV-IRES-GFP encoding Rdh10 or from plat-gp packaging cell line by the transduction of MSCV-IRES-GFP encoding B10-TCR and pCMV-VSVG-RSV-Rev. For transfection with the lentivirus, CD4 + CD45RO + T cells were stimulated for 3 days as described above and then spin-infected in the presence of polybrene on a RetroNectin-coated plate. Three days later, the cells were maintained in IL-2-free medium overnight, analyzed for their surface phenotype by flow cytometry, and then transfected cells (Venus + cells) were sorted for various experiments, as described above. Lentivirus-transfected Jurkat cells were also generated by a spin-infection method, purified by sorting Venus + or GFP + cells, and then used for experiments. For a generation of B10-T cells, T cells were activated with anti-CD3/ CD28 mAbs in the presence of IL-2 for 2-3 days and then spininfected with B10-TCR-encoding retrovirus as described above. Six hours later, retrovirus infection was repeated. B10-T cells were further cultured for 5-4 days and then treated with RAR agonist/antagonist as described above.

High-Performance Liquid Chromatography (HPLC)
Naïve and activated T cells (0.5-1.0 × 10 7 cells) were plated in 3 ml of complete medium supplemented with IL-2 (20 IU/ml) and incubated with 1 ml of 3 H-labeled ROL (PerkinElmer, Boston, MA, USA) for 4 h. Cells were harvested and washed three times with PBS and intracellular retinoids were extracted as described previously (17). The extracted retinoids and non-labeled standard all-trans ROL, RAL, and RA were mixed and then fractionated using the HPLC system (Shimadzu, Kyoto, Japan). Separation was performed using the Synergi 4 mm Hydro-RP 80A Column (Phenomenex, Torrance, CA, USA) and an isocratic elution by the solvent composed of 75% acetonitrile and 25% 50 mM ammonium acetate (pH 7) at a flow rate of 1.5 ml/min. Retinoid standards were detected by measuring ultraviolet absorption at 350 nm.
To evaluate the relationship between CD62L expression and plasma ROL, heparinized blood collected from 6-to 7-week-old female mice was used to measure CD62L expression and the concentration of plasma ROL, as described previously (18). To investigate the effect of a high vitamin A (HVA) diet on CD62L expression, all 6-to 7-week-old female mice were fed an AIN93G-based control (vitamin A, 4 IU/g) diet for 1 week prior to feeding AIN93G-based HVA (250 IU/g). These diets were purchased from Clea Japan, Inc. Extracted retinoids were dissolved in acetonitrile and then analyzed by HPLC as described above. To determine the concentration of ROL, a standard curve was obtained using serial dilutions of all-trans ROL.

Chromatin Immunoprecipitation (ChIP) Assay
ChIP assays were performed using a SimpleChIP Plus Enzymatic Chromatin IP Kit (Cell Signaling Technology, Danvers, MA, USA) according to the manufacturer's instructions.
Quantitative Real-Time PCR mRNAs were extracted using TRIzol reagent (Thermo Scientific) and reverse-transcribed into cDNA using SuperScript VILO (Thermo Scientific). Quantitative real-time PCR was performed using Power SYBR Green Master Mix (Thermo Scientific) following a standard protocol. Primers are listed in Supplementary Table 6. Data were analyzed by a comparative quantification method.

ChIP-Seq Analysis
A ChIP-seq and bio-informatics analysis were performed by the TaKaRa Bio Dragon Genomics Center (Yokkaichi, Mie, Japan) using the Illumina HiSeq2500 system. Data quality was checked using FastQC software and all data were found to be of good quality. Reads were aligned to the hg19 human genome using bowtie2 version 2.2.5. Duplicate reads were removed before peak calling using MACS version 1.4.2. Peaks were assigned to genomic regions using BEDTools. According to RefSeq, genes located within 10 kb upstream or downstream of peaks were identified and extracted.

Statistical Analysis
Data were analyzed using Prism 6, 8, or 9. Normally and nonnormally distributed data were analyzed by parametric (unpaired t-test) and nonparametric (Mann-Whitney test and Wilcoxon test) tests, respectively. For multiple comparisons, ANOVA with post-hoc Dunnett's test was used. In the case that the objective cell number was less than 10, the samples were excluded from statistical analysis due to unreliable data ( Figure 3F left and Supplementary Figure 5D).

Vitamin A Metabolism Promotes Human T Cell Differentiation
Wilms' tumor gene 1 (WT1) -specific CD4 + T cell clone differentiated during culture into four subsets, CD62L + CD127 + (P1), CD62L -CD127 + (P2), CD62L + CD127 -(P3), and CD62L -CD127 -(P4) cells, which were considered as central memory (T CM ), effector memory, effector, and terminal effector T (T EFF ) cells, respectively, according to cytokine production, proliferation and self-renewal capacity, and pluripotency (Supplementary Figures 1A-D). Differential gene expression among P1, P2, and P4 was examined using a microarray approach, followed by quantitative PCR. RDH10 expression increased as a function of the differentiation stage (Supplementary Table 1 and Supplementary Figure 1E). This was also observed in freshly isolated CD4 + and CD8 + T cell subsets (Supplementary Figure 1F). Furthermore, RDH10 mRNA expression levels in CD4 + and CD8 + T cells increased sharply and then decreased following T cell receptor (TCR) stimulation (Supplementary Figure 1G). RDH10 catalyzes the oxidation of ROL to RAL, which is the rate-limiting step in RA biosynthesis (19,20). To investigate whether human T cells themselves can metabolize vitamin A via RDH10, RDH10knocked down and -overexpressed CD4 + T cells were generated and examined for RAL and RA production from tritium ( 3 H) -labeled ROL. Knockdown ( Figures 1A, B) and overexpression ( Figures 1C, D) of RDH10 decreased and increased the production of the RAL metabolite from ROL, respectively. Importantly, RA, the most active form of vitamin A, was not detected in the activated T cells ( Figures 1B, D).
Next, to examine the effect of vitamin A metabolism on T cell differentiation, RDH10-knocked down and -overexpressed CD4 + T cells were activated with TCR stimulation and cultured in serum-free medium supplemented with IL-2 and various concentrations of ROL. Seven days later, the cell expansion and phenotypes (CD62L expression and cytokine production capacity) were analyzed. When RDH10 was knocked down, both T cell expansion and IFN-g-positive T cells decreased, while both IL-2-single positive T cells and CD62L-positive memory phenotype T cells increased, suggesting the blockage of T cell differentiation to terminal stages ( Figure 1E). In contrast, the overexpression of RDH10 resulted in increases in both T cell expansion and IFN-g-positive T cells, but a decrease in CD62Lpositive memory phenotype T cells, suggesting the promoted T cell differentiation ( Figure 1F). The blockage and promotion of T cell differentiation were dependent on the concentration of ROL in serum-free medium without contamination of vitamin A. These results indicate that human T cells themselves metabolize vitamin A, the metabolites of which regulate T cell differentiation, via RDH10.

Rdh10 and Vitamin A Regulate CD62L Expression In Vivo
To examine the roles of Rdh10 in T cell differentiation, T cellspecific Rdh10 conditional knockout (Rdh10CKO) mice were  Figure 2A). A deficiency of Rdh10 was confirmed by PCR and the loss of RAL production from ROL in T cells (Supplementary Figures 2B, C). Rdh10-deficient thymocytes exhibited CD4, CD8, CD24, and TCR-b expression levels comparable to those of control mice (Supplementary Figure 3A). In addition, the frequencies of Foxp3 + regulatory and a4b7 integrin-expressing T cells, which are induced by RA (10,11,17), were not different between Rdh10-deficient and control mice (Supplementary Figures 3B, C). CD4 + and CD8 + T cells in lymphoid organs expressed CD62L at higher levels in Rdh10-deficient mice than in control mice (Figures 2A-C). Enhanced CD62L expression was observed at as early as the double-positive stage when the expression of Cre recombinase was initiated during T cell development in the thymus ( Figure 2D). It is well known that CD62L expression is enhanced in non-activated T cells. However, no difference in the expression of CD44, an activation marker of T cells, between Rdh10-deficient and control mice (Supplementary Figure 3D) suggested that the enhanced CD62L expression was not due to the change in T cell activation, but was caused by altered vitamin A metabolism. This was consistent with the observation that an Rdh10 deficiency also enhanced CD62L expression in OT-II cells (21) that are unresponsive to environmental and self-derived antigens in a Rag1-/-background ( Figure 2E). Next, the functional effect of enhanced CD62L expression on T cell homing in vivo was examined. The mixtures of Rdh10deficient and control T cells in the ratio of 1:1 were intravenously injected into B6 mice, and the ratio was investigated in secondary lymphoid organs 16 h later. Consistent with the enhanced expression of CD62L, which was a receptor for T cell migration into lymph nodes, the ratio of Rdh10-deficient T cells to control T cells significantly increased in the lymph nodes ( Figure 2F). These results demonstrated that the ability of Rdh10-deficient T cells to migrate into lymph nodes was potentiated compared to that of control T cells. Thus, these results indicate that Rdh10 regulates the homing capacity of T cells to lymph nodes by the regulation of CD62L expression.
ROL down-regulated CD62L expression in both dose-and Rdh10-dependent manners even in OT-I cells (mouse CD8 + T cells) (Supplementary Figure 4A). Furthermore, RAL and RA, which are ROL metabolites downstream of Rdh10, decreased CD62L expression in OT-I cells independently of Rdh10 (Supplementary Figure 4B). These results show that ROL metabolism regulates CD62L expression, suggesting that ROL concentration in blood closely links to CD62L expression level on T cells. Indeed, an inverse correlation was observed between plasma ROL concentration and CD62L expression in T cells in wild-type mice ( Figure 2G). These results indicated that foodderived vitamin A can regulate CD62L expression level in T cells.
To confirm them, female B6 mice were fed with control diet (4 IU/g) for one week and then with either a high-vitamin A (HVA) (vitamin A, 250 IU/g) or control diet for the next 1 week. As expected, HVA diet reduced CD62L expression in both CD8 + and CD4 + T cells, with a concomitant increase in ROL concentration in plasma ( Figures 2H, I). Taken together, these results demonstrate that vitamin A (ROL) and its metabolites are physiologically involved in the regulation of CD62L expression in T cells.

Rdh10 Deficiency Promotes Memory T Cell Differentiation
The immune response of Rdh10-deficient CD8 + T cells was investigated in an ovalbumin-expressing Listeria monocytogenes (LM-OVA) infection model (14,22). B6 recipient mice (CD45.  Figures 3A-C). Similar kinetics were also observed in the number of transferred OT-I cells ( Figure 3D). These results indicated that an Rdh10 deficiency in OT-I cells induced differentiation toward memory T cells. Next, Rdh10CKO OT-I cells in the memory phase were characterized in more detail. Rdh10CKO OT-I cells expressed CD62L at higher levels and contained T CM (CD62L + CD127 + ) cells at a higher frequency in both the spleen (Figures 3E, F) and lymph nodes (Supplementary Figures 5D, E), compared to control OT-I cells. The T CM marker CD127, but not the T EFF marker KLRG1, was also expressed at higher levels in Rdh10CKO OT-I cells than in control OT-I cells (Supplementary Figures 5F, G). To confirm that Rdh10CKO OT-I cells were potential memory T cells, the recall responses of these memory T cells were examined after LM-OVA re-challenge because a strong recall response is a hallmark of T CM . A large number of Rdh10CKO OT-I cells were observed in the spleen and liver on day 5 after re-challenge ( Figure 3G). Importantly, these Rdh10CKO OT-I cells were resistant to the attenuation of T CM generation induced by repetitive antigen exposure (24) with a higher frequency and a greater absolute T CM count, compared to control OT-I cells (Figures 3H, I). Thus, these results indicate that Rdh10 deficiency not only enhances memory T cell generation, but also improves the potential for the memory T cells to rapidly expand after re-challenge and to re-generate the T CM pool. The properties of Rdh10-deficient memory T cells suggest Rdh10 as a potential target to improve an effect of cancer immunotherapy because T CM has been shown to have a great anti-tumor activity (25). To confirm this concept, Rdh10deficient memory OT-I cells were investigated for anti-tumor activity against OVA-expressing EL-4 tumor cell line (EG-7) in an adoptive T cell therapy model. As expected, Rdh10-deficient memory OT-I cells suppressed the tumor growth greater than control memory OT-I cells, resulting in prolonged survival ( Figures 3J, K).
In Rdh10 (+/-)-lacZ knock-in reporter mice (Supplementary Figure 2A), Rdh10 expression was transiently enhanced in OT-I cells after LM-OVA infection (Supplementary Figure 6A), which was consistent with the transient production of RAL in activated T cells (Supplementary Figures 6B-D). These findings raised the possibility that RAL produced by Rdh10 in the effector phase could promote the differentiation of T cells into effector T cells. Therefore, the recall response was compared between Rdh10 hi and Rdh10 lo T cells isolated from Rdh10-lacZ knockin reporter mice in the effector phase (Supplementary Figure 7A). Rdh10 lo T cells showed a greater recall response than Rdh10 hi T cells (Supplementary Figures 7B, C), indicating that low Rdh10 expression in the effector phase strengthened the recall response as a result of the increase in T CM cells. This was also confirmed by the lower proliferative capacity and lower T CM generation in the recall response in Rdh10-overexpressing T cells (Supplementary Figures 7D, E). Thus, these results demonstrate that Rdh10 regulates T cell differentiation in the effector phase via RAL.

RA Receptor Signaling Regulates Human T Cell Differentiation
Not only RA but also RAL binds to retinoic acid receptor (RAR) and the ligand-RAR complex regulates the expression of various genes (26). To examine the effect of RAR signaling downstream of RDH10 on human T cell differentiation, CD45RO + CD4 + T cells were stimulated with anti-CD3 and -CD28 mAbs in the presence of RA or LE540 (an RAR antagonist) (27), and the expression levels of T cell differentiation markers (CD62L, CD127, CCR7, and CD27) (28-30) were examined after 8 days (Figures 4A, B, and Supplementary Figure 8A). RAR signaling promoted by RA decreased T CM phenotype T cells, such as CD62L + CD127 + and CD62L + CD27 + T cells, whereas the blockage of RAR signaling by LE540 decreased T EFF phenotype T cells, such as CD62L -CD127 -, CD62L -CCR7 -, and CD62L -CD27 -T cells. These results show that the up-and down-regulation of RAR signaling respectively suppress and promote the expression of CD62L indicating that RAR signaling is a key pathway for the regulation of CD62L expression.
Next, it was examined whether these RA-and LE540-treated T cells (named as T RA and T LE cells, respectively) exhibited the functional characteristics of T EFF and T CM , respectively. DMSO-treated T cells (T DM ) were used as a control. It is generally known that T EFF is lower in resistance to apoptosis, proliferative capacity, T CM production after cell division, and in vivo expansion capacity, compared to T CM . Consistent with these findings, T RA cells showed high frequencies of apoptosis regardless of the presence or absence of anti-Fas mAb ( Figure 4C), and showed decreases in both the proliferative capacity ( Figure 4D and Supplementary Figure 8B) and the frequency of T CM (CD62L + CD127 + ) under two different stimulatory conditions ( Figure 4E and Supplementary Figure 8C). Furthermore, the number of the transferred T RA cells in NOG mice was lower compared to that of T DM cells 4 weeks after T cell transfer ( Figure 4F). On the other hand, the down-regulation of RAR signaling induced by LE540 had the opposite effect on T cells. Thus, these results demonstrate that RAR signaling promotes T cell differentiation into T EFF . It should be noted that the effect of RA on T cell differentiation required TCR signaling (Supplementary Figure 8D), and that UVI3003 (31), an antagonist of retinoid X receptor (RXR) whose heterodimer with RAR acts as a transcription factor (32, 33) could not induce T CM (Supplementary Figure 8E). In addition, RAL down-regulated CD62L expression and decreased T CM generation although these functions of RAL were weaker than those of RA (Supplementary Figure 8F). Importantly, even in human CD8 + T cells, RAR signaling promoted T EFF differentiation as evidenced by low expression of CD62L and weak proliferative capacity in response to IL-7-or TCRstimulation in RA-treated CD8 + T cells (Supplementary Figure 9).

Blocking of RAR Signaling Confers a Strong Anti-Tumor Activity on T Cells
Anti-tumor activity of T LE cells was examined by using an adoptive T cell therapy model. T cells were transduced with B10-TCR (34,35) which recognized a WT1-derived CTL epitope, WT1 235 in an HLA-A*24:02-restricted manner, and the T cells were treated with DMSO or LE540. The LE540treated B10-TCR-transduced T cells (B10-T LE cells) expressed more highly memory or undifferentiated T cell markers such as CD62L, CCR7, and CD45RA, compared to B10-T DM cells (Supplementary Figure 10). B10-T DM cells or B10-T LE cells were intravenously transferred into K562-A24-luc tumorbearing NOG mice. In addition, autologous WT1 235 -pulsed, CD3 + T cell-depleted PBMCs were co-transferred as antigenpresenting cells into the NOG mouse ( Figure 5A). B10-T LE cells began to suppress tumor growth at day 11 (7 days after T cell transfer) and let the tumor decrease on day 28 (24 days after T cell transfer) whereas statistically significant anti-tumor effect of B10-T DM cells was not observed (Figures 5B-D). Importantly, B10-TCR + (GFP + ) CD8 + T cells in B10-T LE cells reached a peak on day 18 and were maintained whereas those in B10-T DM cells once reached a peak on day 18 but then remarkably decreased ( Figure 5E). Persistence of tumor-specific CD8 + T cells has been shown to be a major factor for eliciting an anti-tumor response in adoptive T-cell therapy (36,37). Therefore, the strong antitumor responses of B10-T LE cells likely resulted from the persistence of GFP + CD8 + T cells. Thus, these results show that blocking RAR signaling by LE540 confers a potent antitumor activity through the increase in memory CD8 + T cells that play a central role in anti-tumor immunity.

Epigenetic Regulation of CD62L Expression by RAR Signaling
CD62L expression in memory T cells is critical for the elimination of tumor cells (25,38) and viral infection (39), and the results described above clearly demonstrate that vitamin A metabolism and RAR signaling regulate CD62L expression. Therefore, the mechanism underlying the regulation of CD62L expression was investigated. Enforced expression of RARa decreased CD62L expression at both the protein and mRNA  Figures 6A,  B), indicating that shedding of CD62L from the cell surface (40) was not the cause of the down-regulation of surface CD62L expression. KLF2, a key transcription factor that induces CD62L expression (40)(41)(42), increased regardless of the decrease in CD62L expression by RA treatment (Figure 6C), suggesting the existence of a novel regulatory mechanism of CD62L expression. RARa directly binds to DNA, recruits corepressors, and suppresses the transcription of target genes. Both the DNA-binding domain (DBD) and activation function 2 (AF2) domain are critical for regulating target gene transcription (16,43). Neither DBD-nor AF2 domaindeficient RARa proteins down-regulated CD62L expression, regardless of the presence of RA ( Figure 6D). Furthermore, two co-repressor proteins, i.e., nuclear receptor co-repressor (NCoR) (44) and silencing mediator for retinoid/thyroid hormone receptor (SMRT) (45), were needed for RAR signaling-dependent CD62L repression, as evidenced by significant restoration of CD62L expression after the knockdown of both co-repressors ( Figure 6E). Since these corepressors recruit histone deacetylases (HDACs) and confer a deacetylated status on histones, the role of histone deacetylation at the CD62L locus in the downregulation of CD62L expression was investigated. RA treatment decreased histone acetylation (H3K9/14ac) at the CD62L promoter region (Figures 6F, G and Supplementary Figure 11). On the other hand, LE540-treatment increased H3K9/14ac (Supplementary Figure 11). Furthermore, Trichostatin A (TSA), an HDAC inhibitor, could abrogate the RA-induced CD62L downregulation ( Figure 6H). Thus, these results demonstrate that RAR signaling induced by RAL or RA represses CD62L expression via the deacetylation of H3K9/14 at the CD62L promoter region.

RA Signaling Induces Comprehensive Epigenetic Repression of Memory T Cell-Associated Genes
In addition to the down-regulation of CD62L, RA signaling also induced various effector T cell properties, as described in Figure 4, suggesting that the epigenetic regulation of multiple genes was associated with T cell differentiation. To examine broad-scale epigenetic regulatory effects, a ChIP-seq analysis was performed for T DM , T RA , and T LE cells with antibodies specific for H3K9/14ac and H3K27me3, which are markers for "opened" and "closed" chromatin structures, respectively (46,47). The number of H3K9/14ac-enriched regions was lower in T RA cells (1400 peak calls) than in T DM (2298 peak calls) and T LE cells (2590 peak calls) ( Figure 7A). The number of H3K27me3enriched regions was comparable among T DM , T RA , and T LE cells, but was slightly lower in T LE cells than in the other two types of cells. Analysis using the Cis-regulatory Element Annotation System (CEAS) (48) demonstrated that the proportion of the regions occupied by H3K9/14ac in both the regulatory regions (promoter and downstream) and gene bodies (UTRs, coding exons, and introns) was lower in T RA cells than the other two types of cells ( Figure 7B, upper panel). In contrast, the proportion of the regions occupied by H3K27me3 in these regions, especially in promoter regions was lower in T LE cells than the other two types of cells ( Figure 7B, lower panel). These results revealed that T RA and T LE cells had the "closed" and "opened" chromatin states, respectively, for gene expression. Next, to investigate whether these chromatin states define T cell differentiation, the genes located within 10 kilobases (kb) upstream and downstream of the regions occupied by H3K9/ 14ac or H3K27me3 were focused. The majority of occupied genes, especially of H3K9/14ac-occupied genes, were different and unique among T DM , T RA , and T LE cells ( Figure 7C and Supplementary Figure 12A). A Metascape analysis (49) of the uniquely occupied genes showed that the cluster of T LE cells was distinct from that of T RA cells, but similar to that of T DM cells ( Figure 7D). T LE cells had, in the uniquely H3K9/14ac-occupied genes, more highly enriched GO terms, including "cellular protein catabolic process", "protein ubiquitination involved in ubiquitin-dependent protein catabolic process", and "mitochondrion organization", which were likely to be associated with memory T cells based on the requirement for autophagy (50) and mitochondrial oxidative phosphorylation (51) for memory T cell differentiation, compared to T DM cells ( Figure 7D). Furthermore, an enrichment analysis using Reactome Gene Sets that T LE cells had highly enriched memory-associated genes related to "fatty acid, triacylglycerol, and ketone body metabolism", "mitochondrial protein import", and "TCA cycle and respiratory electron transport" ( Figure 7E). Surprisingly, the majority of GO terms for the H3K27me3occupied genes enriched in T RA and T DM cells were not related to T cell biology, but rather related to the development and differentiation of organs and tissues, such as the heart, eyes, and neurons (Supplementary Figure 12B). To clarify epigenetic changes among T DM , T RA (more differentiated), and T LE cells (less differentiated), the genes with two histone modifications were extracted by a comparison of differentiated T cells with less differentiated T cells (T DM vs T LE cells, T RA vs T DM cells, and T RA vs T LE cells), and categorized into "Opened" and "Closed" genes ( Supplementary Tables 2 and 3). This approach also revealed that the most differentiated T RA cells had the pronounced closed chromatin state, compared to T DM and T LE cells ( Figure 7F). Notably, in the comparison of T LE with T RA cells, T RA cells had the closed chromatin state in the regions of memory T cellassociated genes, including TCF7, BCL2, KIT, DUSP4, CABLES1, and SMAD4 (52-55) ( Figure 7G and Supplementary Table 3). It is well known that TCF7 is a key transcription factor for the maintenance of T cell stemness in CD8 + T cells (56,57) and the generation of memory T cells. Consistent with this, H3K27me3 at the TCF7 locus more increased in RA-treated CD8 + T cells compared to LE540-or DMSO-treated CD8 + T cells  Figure 7H). Thus, RA signaling comprehensively induces repressive chromatin states and deletes the memory T cell profile.

DISCUSSION
Although it was well known that RAR signaling was required for eliciting effector function in both CD4 + and CD8 + T cells (12,13,58,59), vitamin A metabolism in T cells and its role for T cell differentiation were unclear. In this study, we revealed that T cells intrinsically metabolized vitamin A to RAL by RDH10 and that this metabolism promoted T cell differentiation into terminal effector T cells. Interestingly, although RDH10 is a rate-limiting enzyme for RA production, T cells could not produce RA, which is a more potent RAR signal activator than RAL. This unique vitamin A metabolism suggests the existence of interesting and important kinetics of T cell differentiation. If T cells can produce RA upon activation, they will rapidly differentiate into terminal T EFF cells and disappear by apoptosis, resulting in the discontinuation of the immune response. On the other hand, RAL not completely but appropriately induces T cell differentiation into T EFF cells to keep T CM cells with the avoidance of the complete T cell terminal differentiation, resulting in the continuation of the immune response. Accordingly, T cells can protect themselves from excessive differentiation by a lack of RA production and maintain a portion of undifferentiated T CM cells. These findings demonstrated a surprising dynamism of T cell differentiation. Importantly, RA is abundant in tumors (12,60). When T cells infiltrate into tumors, the T cells can take in the exogenous RA that abundantly exists in tumors and rapidly differentiate into T EFF cells. Inflammatory stimuli can also induce RA production from dendritic cells (DCs) (61), and thus RA is abundant in inflammatory sites. Therefore, T cells infiltrated into the inflammatory sites can rapidly differentiate into T EFF cells by taking in the exogenous RA and eradicate infectious agents. Moreover, RAL from T cells might be taken up and used for RA synthesis by DCs at inflammatory sites because RA production from DCs is likely to be enhanced by the interaction with T cells (17). Taken together, while T cells themselves maintain memory function with self-renewal activity and avoid complete differentiation into T EFF cells by a lack of RA production, they can rapidly differentiate into T EFF cells in RA-rich regions such as tumor and inflammatory sites, where prompt effector function is needed.
We demonstrated that RA signaling induced epigenetic repression of memory-associated genes via the decrease and increase in H3K9/14ac and H3K27me3 deposition, respectively. This finding is consistent with a recent report indicating that the accumulation of H3K27me3 deposition on "pro-memory genes", such as Tcf7, guides terminal effector T cell differentiation (62). In addition, the epigenetic repression by RA signaling also resulted in the disappearance of unique signatures such as the catabolic process and mitochondrial oxidative phosphorylation in memory T cells. At the effector phase, numerous effector T cells against a pathogen can be generated by clonal expansion that is supported by anabolic processes, but most of these cells subsequently die and only~5% of them differentiate into memory T cells whose differentiation is supported by catabolic processes, such as autophagy and mitochondrial fatty acid oxidation. Importantly, recent studies have shown that T cells can become memory T cells by shifting their metabolic process from the anabolic to catabolic state (1,63), implying that most (95%) effector T cells physiologically fail to change their metabolism to the catabolic state after pathogen clearance. Accordingly, there is surely a "gate pass" for the metabolic change. Our present results indicate that the "gate pass" should be a permissive epigenetic signature, which is turned into a repressive signature by RA signaling.
The mechanisms by which RA signaling induces comprehensive epigenetic repression of memory-associated genes remain unclear. As both the DBD and AF2 domain of RARa are required for the repression of CD62L via the deacetylation of H3K9/14 at the CD62L promoter induced by RA, it is likely that RARs bind to genome DNA and recruit functional proteins such as co-repressor proteins, resulting in the induction of comprehensive repression. However, we could not detect any functionally or evolutionarily conserved RAresponsive elements (RAREs) within ±20 kb of the CD62L gene (data not shown). In addition, we did not find co-localization of closed regions with RAREs (data not shown). Therefore, it is not clear whether RARs control epigenetic modifications by direct binding to RAREs. It should be noted that RA signaling alone was insufficient for the differentiation into effector T cells, and that TCR signaling was simultaneously required for it, as shown in Supplementary Figure 8D. Interestingly, upon TCR stimulation, the expression of Ezh2, which encodes a methyltransferase, is enhanced and the deposition of H3K27me3 increases in CD8 + T cells (62). Furthermore, in activated T cells with TCR stimulation, HDACs co-localize with histone acetyltransferases (HATs) and RNA Pol II at transcribed regions to remove acetylation marks and reset the chromatin state (64). RAR signaling also causes long-term and stable epigenetic changes during the differentiation of stem cells into other specialized cell types, thus silencing or activating large sets of genes either directly or indirectly (65,66). Thus, RAR signaling may exhibit cross-talk with TCR signaling in the context of epigenetic modification and achieve the directional differentiation into terminal effector T cells.
Allie et al. reported that blockage of RAR signaling enhance the development of central memory-like CD8 + T cells in mouse model (59). However, it was no evidence that blockage of RAR signaling in human T cells enhance anti-tumor immunity through inducing memory T cell. We here succeeded in demonstrating for the first time that RAR signaling in human T cells is targets for improving the clinical efficacy of cancer immunotherapies such as adoptive TCR-or chimeric antigen receptor-transduced T-cell therapy, whose clinical effect is largely dependent upon the amount of memory and undifferentiated T cells (36,67). Furthermore, we demonstrated that RDH10-dependent vitamin A metabolism in T cells is also the target. RDH10-targeted immunotherapy should be safe because the conditional RDH10 knockout adult mouse did not present any detectable organ damages (unpublished data). In fact, some chemical agents that could inhibit the RDH10 enzyme activity increased memory T cells (unpublished data). Therefore, the present study contributes to develop memory T cell-based cancer immunotherapy by in vitro and in vivo intervention of T cell differentiation.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.ncbi.nlm. nih.gov/, GSE137142, https://www.ncbi.nlm.nih.gov/, GSE137714.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by the Ethics Committee of Osaka University Graduate School of Medicine the Osaka University Research Ethics Committee. The patients/participants provided their written informed consent to participate in this study. The animal study