d-serine as a metabolic immune checkpoint in the tumour microenvironment Authors: Kai Tsugaru, Shohei Suzuki, Kentaro Miyamoto, Kazuhiro Murakami, Akihiko Chida, Chihiro Kato, Naoto Fukasawa, Arina Shigehara, Mai Hasegawa, Yusuke Yoshimatsu, Jumpei Sasabe, Masashi Mita, Nick Barker, Yosuke Harada, Yasutaka Sukawa, Toshiaki Teratani, Nobuhiro Nakamoto, Hirofumi Kawakubo, Kaoru Takabayashi, Motohiko Kato, Naohisa Yahagi, Taisuke Kondo, Eri Arai, Yuko Kitagawa, Kenro Hirata, Yasuo Hamamoto, Takanori Kanai, Tomohisa Sujino Published: 2026 ABSTRACT SummaryBackgroundd-amino acids (D-AAs), the enantiomers of proteinogenic l-amino acids, are detectable in mammals, yet their biological roles in cancer immunity remain largely unexplored. Whether specific D-AAs modulate tumour progression or influence responsiveness to immunotherapy in gastrointestinal cancer is unknown. We aimed to determine how D-AAs, particularly d-serine (D-ser), shape the tumour immune microenvironment and affect clinical outcomes in gastrointestinal cancers.MethodsMechanistic studies were conducted using murine MC38 tumours and orthotopic gastric cancer (GC) organoid allografts with or without D-AAs supplementation. Immune landscape alterations were assessed using single-cell RNA sequencing of tumour-infiltrating immune cells, flow cytometry, ex vivo macrophage–T cell co-culture assays, and microbiome manipulation experiments. D-AAs concentrations in plasma, urine, and stool were quantified in healthy controls (HCs; n = 87) and patients with GC across three cohorts (Cohort 1, n = 14; Cohort 2, n = 108; Cohort 3, n = 28). Associations between plasma D-ser levels, disease stage, immune cell infiltration, and clinical outcomes following anti–PD-1 antibody therapy were analysed.FindingsD-ser promoted tumour progression by suppressing CD8+ T cell immunity and enhancing SPP1-associated immunosuppressive macrophage signalling in murine model. Across all clinical cohorts, the plasma, urine, and stool levels of several D-AAs, most prominently D-ser, were significantly elevated in patients with GC compared to HCs. Plasma concentration of D-ser strongly correlated with disease stage (I–IV). Elevated plasma D-ser is associated with an immunosuppressive tumour microenvironment and poor response to anti–PD-1 monotherapy in patients with advanced gastric cancer.InterpretationThis study identifies D-ser as a previously unrecognised immunosuppressive metabolite that promotes tumour immune evasion by increased macrophages and reduced CD8+ T cell effector function, thereby shifting the tumour microenvironment toward an immunosuppressive phenotype. Clinically, D-ser is a potential metabolite to predict cancer progression and immunotherapy resistance.FundingThis work was supported by 10.13039/501100002241The Japan Science and Technology Agency (JST) 10.13039/501100020964Fusion Oriented Research for Disruptive Science and Technology (FOREST)[JPMJFR210P], Grants-in-Aid from the Japanese Society for the Promotion of Science (JSPS) (25K10430, 21K18272, 23H02899, 23K27590, 25K22627), KGRI challenge grant, Sakaguchi Memorial Foundation, 10.13039/100009619Japan Agency for Medical Research and Development (CREST 21gm1510002h0001), and 10.13039/100016567Miyarisan Pharmaceutical Grant. Research in contextEvidence before this studyd-amino acids (D-AAs), the enantiomers of proteinogenic l-amino acids (L-AAs), are present in human biological fluids and are derived from the diet, microbiota, and host metabolism. Among them, d-serine (D-ser) is well known for its role in neurotransmission and mucosal immunity. However, evidence linking D-AAs to cancer immunity remains limited. A small number of reports have detected D-AAs in the blood or urine of patients with cancer, but their functional contribution to tumour progression or immune evasion has not been systematically evaluated. No prior studies have addressed the mechanistic role of D-ser in antitumour immunity or its potential impact on immunotherapy outcomes.Added value of this studyThis study provides the clinical and mechanistic evidence that D-ser acts as an immunosuppressive metabolite in tumour. Using MC38 and B16F10 murine tumour models and a gastric cancer (GC) organoid transplant model, we demonstrated that 1% of exogenous D-ser promotes tumour growth by reducing cytotoxic CD8+ T cell infiltration and enhancing the accumulation of immunosuppressive macrophages. Importantly, these effects were independent of changes in the microbiota. We also showed that plasma concentration of D-ser levels elevated in patients with advanced GC and were associated with a poor response to anti–PD-1 antibody therapy.Implications of all the available evidenceOur findings establish D-ser as a metabolic checkpoint in cancer immunity. Clinically, the plasma concentration of D-ser may serve as a potential predictive biomarker for immunotherapy resistance in GC, enabling improved patient stratification before anti–PD-1 treatment. Therapeutically, interventions that limit D-ser availability could restore CD8+ T cell activity and shift macrophages toward anti-tumoural states, thereby improving immunotherapy efficacy. More broadly, this study integrates D-ser into the growing landscape of amino acid metabolism in cancer, alongside tryptophan, glutamine, and arginine pathways, and highlights its potential as a targetable axis for tumour immune evasion. Research in context Evidence before this study d-amino acids (D-AAs), the enantiomers of proteinogenic l-amino acids (L-AAs), are present in human biological fluids and are derived from the diet, microbiota, and host metabolism. Among them, d-serine (D-ser) is well known for its role in neurotransmission and mucosal immunity. However, evidence linking D-AAs to cancer immunity remains limited. A small number of reports have detected D-AAs in the blood or urine of patients with cancer, but their functional contribution to tumour progression or immune evasion has not been systematically evaluated. No prior studies have addressed the mechanistic role of D-ser in antitumour immunity or its potential impact on immunotherapy outcomes. Added value of this study This study provides the clinical and mechanistic evidence that D-ser acts as an immunosuppressive metabolite in tumour. Using MC38 and B16F10 murine tumour models and a gastric cancer (GC) organoid transplant model, we demonstrated that 1% of exogenous D-ser promotes tumour growth by reducing cytotoxic CD8+ T cell infiltration and enhancing the accumulation of immunosuppressive macrophages. Importantly, these effects were independent of changes in the microbiota. We also showed that plasma concentration of D-ser levels elevated in patients with advanced GC and were associated with a poor response to anti–PD-1 antibody therapy. Implications of all the available evidence Our findings establish D-ser as a metabolic checkpoint in cancer immunity. Clinically, the plasma concentration of D-ser may serve as a potential predictive biomarker for immunotherapy resistance in GC, enabling improved patient stratification before anti–PD-1 treatment. Therapeutically, interventions that limit D-ser availability could restore CD8+ T cell activity and shift macrophages toward anti-tumoural states, thereby improving immunotherapy efficacy. More broadly, this study integrates D-ser into the growing landscape of amino acid metabolism in cancer, alongside tryptophan, glutamine, and arginine pathways, and highlights its potential as a targetable axis for tumour immune evasion. Introduction The tumour microenvironment integrates malignant cells, immune populations, and host-derived metabolites to shape cancer progression and therapeutic response.1, 2, 3, 4, 5, 6, 7, 8, 9, 10 Cytotoxic CD8+ T cells are central to effective antitumour immunity, yet their activity is frequently curtailed by immunosuppressive programs within tumours, including the accumulation of tumour-associated macrophages (TAMs)11, 12, 13, 14, 15, 16 and the induction of T cell exhaustion.6,11, 12, 13, 14, 15, 16, 17, 18 While immune checkpoint blockade has transformed the management of several cancers, including intestinal tract cancer, a substantial fraction of patients fails to respond, underscoring the need to identify additional mechanisms that constrain antitumour immunity. Metabolic cues are increasingly recognised as critical regulators of immune function in the patients with cancer. Amino acid metabolism governs T cell differentiation, effector function, and persistence. Studies on glutamine, arginine, and tryptophan metabolism have revealed how tumours exploit metabolic pathways to evade immune surveillance.19,20 In contrast, the role of d-amino acids (D-AAs), which are enantiomers of proteinogenic l-amino acids (L-AAs), has received little attention in cancer immunity.21 D-AAs are present in mammals and are derived from dietary intake, microbial metabolism, and host enzymatic activity.19,20 Previously, D-AAs, especially d-serine (D-ser) and d-alanine (D-ala), were known as neuromodulators acting at N-methyl-d-aspartate (NMDA) receptors in the brain.22 Whether D-AAs contribute to tumour immune regulation or influence responsiveness to immunotherapy remains unknown. In this study, we investigated the functional and clinical relevance of D-AAs in gastrointestinal cancer progression. Using complementary murine tumour models, single-cell transcriptomics, and ex vivo functional assays, we demonstrated that D-ser acts as an immunosuppressive metabolite that reshapes the tumour immune microenvironment through macrophages and CD8+ T cells. Extending these findings to human disease, we showed that the concentration of circulating D-ser is markedly elevated in patients with gastric cancer (GC), correlates with disease stage, and predicts poor response to anti–PD-1 antibody therapy. Together, these data identify D-ser as a previously unrecognised metabolic checkpoint that constrains CD8+ T cell-mediated antitumour immunity and contributes to immunotherapy resistance in the patients with GC. Methods Patient recruitment Patients with GC were prospectively enrolled at Keio University Hospital between October 2017 and March 2022. Diagnosis and clinical staging (I–IV) were determined according to the Eighth Edition American Joint Committee on Cancer (AJCC) Cancer Staging Manual.23 Healthy controls (HCs) were recruited from individuals undergoing routine annual health screenings and who had no history of cancer or chronic gastrointestinal disease. Biological specimens, including plasma, urine, and feces, were collected after an overnight fast (≥8 h) and before the initiation of any treatment. The study included three patient cohorts Cohort 1 (Exploratory): This cohort included 87 healthy controls and 14 treatment-naïve patients with GC for the initial screening of D-AA profiles. Cohort 2 (Validation): This cohort included 108 additional treatment-naïve patients with GC stratified by clinical stage (I: n = 52; II/III: n = 24; IV: n = 32) to assess d-amino acid concentrations in relation to tumour stage. Patients who underwent curative surgery were stratified based on pathological stage (pStage) to accurately assess the extent of cancer progression. Patients who were ineligible for curative surgery were stratified based on clinical stage (cStage). Cohort 3 (Immunotherapy Response): This cohort included 28 patients with cStage IV GC who were undergoing third-line or later chemotherapy. Plasma was collected immediately before the initiation of anti–PD-1 antibody therapy to evaluate associations between concentration of D-AA and treatment outcomes. Demographic and clinical characteristics, including medical history, laboratory data, histological findings, treatment regimens, and radiological responses, were collected from medical records. Cohorts 1 and 2 were retrospective studies, and cohort 3 was a prospective observational study. All the participants provided written informed consent. The study protocol was approved by the Institutional Review Board of Keio University (approval number: 20170322) and was conducted in accordance with the Declaration of Helsinki. The study was registered in the University Hospital Medical Information Network (UMIN) clinical trial registration system (UMIN 000018068). Demographic data and sex of all the cohorts involved in this study The demographics and clinical characteristics for Cohorts 1–3 are described (Table 1, Table 2, Table 3). Sex was self-reported by study participants.Table 1Patient demographic data of cohort 1.Clinical characteristicsHealthy control (n = 87)Cohort 1 GC (n = 14)Age, median (range)59 (31–85)62 (41–79)Sex, male, n (%)55 (63)8 (57)Race. Asian, n (%)87 (100)14 (100)Plasma creatinine (mg/dL), median (range)0.82 (0.47–1.63)0.77 (0.54–1.14)Plasma eGFR (mL/min/1.73 m2), median (range)71 (33–103)78 (39–88)Disease stage, cStage IV or recurrence, n (%)14 (100)Table 2Patient demographic data of cohort 2.Clinical characteristicsHealthy control (n = 87) from cohort 1Stage I (n = 52)Stage II/III (n = 24)Stage IV (n = 32)Age, median (range)59 (31–85)71 (50–89)73 (40–87)74 (43–91)Sex, male, n (%)55 (63)36 (69)14 (58)23 (72)Race. Asian, n (%)87 (100)52 (100)24 (100)32 (100)Plasma creatinine (mg/dL), median (range)0.82 (0.47–1.63)0.80 (0.51–1.27)0.75 (0.43–2.56)0.81 (0.38–1.19)Plasma eGFR (mL/min/1.73 m2), median (range)71 (33–103)70 (35–96)63 (20–100)69 (36–116)T stage, 1/2/3/4/unknown or unclassifiablea, n (%)45/7/0/0/0 (87/13/0/0/0)0/4/9/11/0 (0/17/37/46/0)1/0/9/15/7 (3/0/28/47/22)N stage, 0/1/2/3/unknowna, n (%)51/1/0/0/0 (98/2/0/0/0)2/10/8/4/0 (8/42/33/17/0)6/5/10/10/1 (19/16/31/31/3)M stage, 0/1, n (%)52/0 (100/0)24/0 (100/0)0/32 (0/100)Number of metastases, 0/1/2/≥3/unknowna, n (%)51/1/0/0/0 (98/2/0/0/0)3/21/0/0/0 (12/88/0/0/0)0/10/18/3/1 (0/31/57/9/3)Histology (Lauren classification), intestinal/diffuse/mixed/unknown or unclassifiable, n (%)31/16/4/1 (60/30/8/2)11/11/1/1 (46/46/4/4)12/13/5/2 (37/41/16/6)ECOG performance status, 0/1/≥2, n (%)12/17/3 (38/53/9)HER2 status, positive/negative/unknown, n (%)8/20/4 (25/63/12)Previous gastrectomy, yes, n (%)6 (19)aPatients with recurrent disease were included as “unclassifiable”. Patients with unknown or unclassifiable data were excluded for univariate and multivariate analyses in Fig. 7.Table 3Patient demographic data of cohort 3.Clinical characteristicsHealthy control (n = 87) from cohort 1Stage IV > 3rd line (n = 28)Age, median (range)59 (31–85)69 (41–89)Sex, male, n (%)55 (63)9 (32)Race. Asian, n (%)87 (100)28 (100)Plasma creatinine (mg/dL), median (range)0.82 (0.47–1.63)0.84 (0.50–1.07)Plasma eGFR (mL/min/1.73 m2), median (range)71 (33–103)68 (22–94)Number of metastases, 0/1/2/≥3/unknown, n (%)0/12/8/8/0 (0/42/29/29/0)Histology (Lauren classification), intestinal/diffuse/mixed/unknown or unclassifiable, n (%)11/15/2/0 (39/54/7/0)ECOG performance status, 0/1/≥2, n (%)3/19/6 (11/68/21)HER2 status, positive/negative/unknown, n (%)4/24/0 (14/86/0)Previous gastrectomy, yes, n (%)16 (57) Table 1Patient demographic data of cohort 1.Clinical characteristicsHealthy control (n = 87)Cohort 1 GC (n = 14)Age, median (range)59 (31–85)62 (41–79)Sex, male, n (%)55 (63)8 (57)Race. Asian, n (%)87 (100)14 (100)Plasma creatinine (mg/dL), median (range)0.82 (0.47–1.63)0.77 (0.54–1.14)Plasma eGFR (mL/min/1.73 m2), median (range)71 (33–103)78 (39–88)Disease stage, cStage IV or recurrence, n (%)14 (100) Patient demographic data of cohort 1. Table 2Patient demographic data of cohort 2.Clinical characteristicsHealthy control (n = 87) from cohort 1Stage I (n = 52)Stage II/III (n = 24)Stage IV (n = 32)Age, median (range)59 (31–85)71 (50–89)73 (40–87)74 (43–91)Sex, male, n (%)55 (63)36 (69)14 (58)23 (72)Race. Asian, n (%)87 (100)52 (100)24 (100)32 (100)Plasma creatinine (mg/dL), median (range)0.82 (0.47–1.63)0.80 (0.51–1.27)0.75 (0.43–2.56)0.81 (0.38–1.19)Plasma eGFR (mL/min/1.73 m2), median (range)71 (33–103)70 (35–96)63 (20–100)69 (36–116)T stage, 1/2/3/4/unknown or unclassifiablea, n (%)45/7/0/0/0 (87/13/0/0/0)0/4/9/11/0 (0/17/37/46/0)1/0/9/15/7 (3/0/28/47/22)N stage, 0/1/2/3/unknowna, n (%)51/1/0/0/0 (98/2/0/0/0)2/10/8/4/0 (8/42/33/17/0)6/5/10/10/1 (19/16/31/31/3)M stage, 0/1, n (%)52/0 (100/0)24/0 (100/0)0/32 (0/100)Number of metastases, 0/1/2/≥3/unknowna, n (%)51/1/0/0/0 (98/2/0/0/0)3/21/0/0/0 (12/88/0/0/0)0/10/18/3/1 (0/31/57/9/3)Histology (Lauren classification), intestinal/diffuse/mixed/unknown or unclassifiable, n (%)31/16/4/1 (60/30/8/2)11/11/1/1 (46/46/4/4)12/13/5/2 (37/41/16/6)ECOG performance status, 0/1/≥2, n (%)12/17/3 (38/53/9)HER2 status, positive/negative/unknown, n (%)8/20/4 (25/63/12)Previous gastrectomy, yes, n (%)6 (19)aPatients with recurrent disease were included as “unclassifiable”. Patients with unknown or unclassifiable data were excluded for univariate and multivariate analyses in Fig. 7. Patient demographic data of cohort 2. Patients with recurrent disease were included as “unclassifiable”. Patients with unknown or unclassifiable data were excluded for univariate and multivariate analyses in Fig. 7. Table 3Patient demographic data of cohort 3.Clinical characteristicsHealthy control (n = 87) from cohort 1Stage IV > 3rd line (n = 28)Age, median (range)59 (31–85)69 (41–89)Sex, male, n (%)55 (63)9 (32)Race. Asian, n (%)87 (100)28 (100)Plasma creatinine (mg/dL), median (range)0.82 (0.47–1.63)0.84 (0.50–1.07)Plasma eGFR (mL/min/1.73 m2), median (range)71 (33–103)68 (22–94)Number of metastases, 0/1/2/≥3/unknown, n (%)0/12/8/8/0 (0/42/29/29/0)Histology (Lauren classification), intestinal/diffuse/mixed/unknown or unclassifiable, n (%)11/15/2/0 (39/54/7/0)ECOG performance status, 0/1/≥2, n (%)3/19/6 (11/68/21)HER2 status, positive/negative/unknown, n (%)4/24/0 (14/86/0)Previous gastrectomy, yes, n (%)16 (57) Patient demographic data of cohort 3. Statistical methods used in clinical studies In cohort 2, both univariate analysis and multivariable linear regression analysis were performed to identify factors independently associated with concentration of plasma D-ser, d-asparagine (D-asn), D-ala, and d-proline (D-pro). In multivariate linear regression analysis, each concentration of D-AA was logarithmically transformed because of their skewed distribution. Covariates were selected a priori based on their clinical relevance. Patients with missing data (n = 10) were excluded from the multivariate linear regression analysis. In cohort 3, OS was calculated from the initiation of palliative chemotherapy (nivolumab) to the day of death. Patients without a recorded date of death were assessed as censored based on their last recorded date. Progression-free survival (PFS) was calculated from the initiation of chemotherapy to the documented disease progression or death from any cause. Objective response was evaluated according to the Response Evaluation Criteria in Solid Tumours (RECIST) version 1.1 for patients with measurable lesions. Tumour response was categorised as complete response (CR), partial response (PR), stable disease (SD), non-CR/non-PD, or progressive disease (PD). Cases with target lesions but without imaging available for treatment efficacy assessment were classified as not evaluable (NE).24 The cut-off value was determined from Receiver Operating Characteristic (ROC) curve analysis by selecting the threshold with higher specificity among candidate thresholds with comparable Youden's index (= Specificity + Sensitivity–1). Survival analyses were performed using the Kaplan–Meier method and log-rank tests. Hazard ratios (HRs) were estimated using the Cox proportional hazards model. All P values were based on a two-sided hypothesis, and P values < 0.05 indicated statistical significance. Animal Animal experiments were conducted using female mice (8–11 weeks old) of the following strains: C57BL/6J-CD45.1 or CD45.2 (wild-type) purchased from Japan CLEA; Rag2−/− mice from JLA (Tokyo, Japan); and germ-free (GF) males purchased from Sankyo Labo Service (Tokyo, Japan) and maintained in vinyl isolators. Except for GF mice, all animals were housed under SPF conditions at the Animal Care Facility of Keio University School of Medicine. All experiments were approved by the Regional Animal Study Committee of Keio University (approval nos. D-2006-008 and A2022-006) and were conducted in accordance with institutional guidelines and Home Office regulations. Isolation of mononuclear cells from spleen and tumour Spleens were smashed into a 100-μm cell strainer and then added to erythrocytes lysed with 0.84% ammonium chloride solution. The dissolved solution was centrifuged at 600×g for 5 min. The tumours were dissected, skin was removed, and the tumour tissues were finely minced. The tumour tissues were digested and incubated in roswell park memorial institute (RPMI) 1640 medium (nacalai tesque) containing 2 mg/mL collagenase (Wako) and 0.1 mg/mL DNase (Sigma–Aldrich) and 10% foetal bovine plasma (FBS) (Thermo Fisher Scientific) for 30 min at 37 °C. The dissolved solution was centrifuged at 600×g for 5 min at 4 °C. The pellet was resuspended in 40% Percoll and overlaid on 75% Percoll. Percoll gradient separation was performed by centrifugation at 800×g for 30 min at 20 °C. The LP mononuclear cells, which were present at the interface, were collected. Countess II (Thermo Fisher Scientific) or LUNA-FX7 (Logos Biosystems) determined the number of live cells.25, 26, 27 Organoid culture Primary gastric tumour-derived organoids (GC organoids, ∗C6434dTKO cells) were generated from Apc −/− Kras G12D Trp53 −/− mice and cultured as described previously.28 To produce luciferase-expressing GC organoids, luciferase cDNA from the pGL3-Promoter vector (Promega) was subcloned into the pPB-CAG-IB PiggyBac transposon expression vector (kindly provided by Dr. Hitoshi Niwa). Organoids were co-transfected with this construct and a PiggyBac transposase plasmid using Lipofectamine LTX (Thermo Fisher Scientific) and selected in blasticidin-containing medium to establish stable luciferase expression. In vivo imaging Female C57BL/6 mice (Charles River Laboratories) were used for all in vivo imaging experiments. One week before transplantation, the mice received either Water or 1% D-ser in their drinking water. On the day of surgery, 5 × 105 dissociated cells were orthotopically transplanted into the gastric wall of the anaesthetised mice. At 21 and 42 days of post-transplantation, mice were injected intraperitoneally with D-luciferin (75 mg/kg; Wako), and bioluminescent signals were captured using the IVIS Lumina LT system and analysed with Living Image Software (PerkinElmer). Determination of D-AAs Plasma, urine, and fecal specimens were mixed with ultrapure water and methanol at a ratio of 1:1:18 (v/v/v), vortexed for 10 s, and centrifuged at 13,000×g for 5 min at 4 °C. PBS homogenates were centrifuged at 14,000×g for 10 min, after which the supernatant was mixed with 18 volumes of methanol, vortexed for 2 min, and centrifuged again at 12,100×g for 5 min at 4 °C. All supernatants were stored at −80 °C until analysis. D-AAs and L-AAs concentrations were determined by two-dimensional high-performance liquid chromatography (HPLC) using a NANOSPACE SI-2 system (Osaka Soda, Japan) equipped with an octadecylsilyl column (KSAARP, 1.0-mm inner diameter (ID) × 400 mm; Kyushu University/KAGAMI Inc, Japan) and a Pirkle-type enantioselective column (KSAACSP-001S, 1.5 mm ID × 250 mm; Kyushu University/KAGAMI Inc.), as previously described.19 MC38, B16F10 cell line culture and mouse injection The murine colorectal cancer cell line MC38 was generously provided by A. Yoshimura (Keio University, Japan). The murine melanoma cell line B16F10 was provided by T. Kondo (Keio University, Japan). Cultured MC38 cells or B16F10 cells were injected subcutaneously (5 × 105 cells/mouse). Tumour volume was calculated as 1/2× (major axis) × (minor axis)2 every other day until day 12–18. The recent STR profiling of cell lines and organoids The murine MC38 colorectal cancer cell line (RRID: CVCL_B288) and B16F10 melanoma cell line (RRID: CVCL_0159) were originated from ATCC. Authentication was performed by the commercial cell repository prior to distribution. Cell identities were verified using the Cellosaurus database and confirmed not to be listed in the ICLAC Register of Misidentified Cell Lines. Cells were used within 20 passages after receipt.29 The cell lines had previously been tested for mycoplasma contamination and were negative. Mouse gastric cancer organoids were generated as described previously. Although dedicated mycoplasma testing was not performed for organoid cultures, they were routinely examined by DAPI staining during serial passage, and no microscopic evidence of mycoplasma contamination was observed. Fluorescence-activated cell sorting (FACS) analysis After blocking with anti-mouse CD16/CD32 antibody (BD Pharmingen) for 20 min, cells were incubated with specific cell surface fluorescence-labelled monoclonal antibodies (mAbs) at 4 °C for 30 min. For intracellular cytokine staining, cells were stimulated for 3 h with 50 ng/mL PMA (Sigma–Aldrich) and 500 ng/mL ionomycin (Sigma–Aldrich) in the presence of 10 μg/mL brefeldin A (BD Bioscience). Subsequently, the cells were permeabilized using Permeabilization Buffer (Invitrogen) and stained intracellularly with antibodies against TNF-α, and IFN-γ. The following monoclonal antibodies were used for FACS analysis: anti-mouse CD45, CD45.1, CD45.2, CD4, CD8α, CD11b, CD11c, Ly6C, CX3CR1, TCR-β, PD-1, TNF-α, and IFN-γ. Dead cells were excluded using the Fixable Viability Dye eFluor 780 (Invitrogen). Events were acquired using a FACS Canto II flow cytometer (BD Biosciences) and analysed using FlowJo software (v 10.10.0) (BD Biosciences).30, 31, 32, 33 The total number of macrophages and CD8+ T cells was counted using live cells after isolating all cells and calculated using FlowJo. All cell numbers were normalised to weight (cells/mg). Please refer to Extended Data Table 1 for detailed information on the antibodies used. d-amino acids/l-amino acids administration D-/L-ser, D-/L-asn, D-/L-ala, and D-/L-pro (each at 1% w/v; FUJIFILM Wako Pure Chemical Corp., Tokyo, Nacalai Tesque Inc., Tokyo)20 were dissolved in the drinking water of the animals. Ex vivo proliferation assay Naïve CD8+ T cells were isolated from the spleen using a magnetic cell sorting (MACS) isolation kit labelled with 5 μM CellTrace™ Violet dye (Thermo Fisher Scientific) for 20 min at 37 °C in the dark. CD11b+ Macrophages used as antigen-presenting cells (APCs) were isolated and plated in 96-well flat-bottom plates at 105 cells/well. CD8+ T cells were added to the macrophages at a 1:1 T cell:APC ratio in complete RPMI medium. The co-cultures were incubated for 72 h. Controls included unstimulated T cells (no APCs) and T cells stimulated with plate-bound anti-CD3ε (Biolegend) and anti-CD28 antibodies (Biolegend). CD8+ T cell proliferation analysis CD8+ T cell proliferation was assessed using CFSE dilution. After gating on singlet, live CD8+ T cells, sequential CFSE peaks (F0–Fn) were identified. CD8+ T cells were co-cultured with 1 μg/mL of anti-CD3ε (BD Biosciences), 2 μg/mL of anti-CD28 (BD Biosciences), 100IU/mL of IL-2 (PeproTech).34 The percent divided was calculated as the proportion of cells that underwent at least one division: %Divided=(1–N0/∑i=0niNi)×100 The division index was defined as the average number of divisions per cell:DivisionIndex=∑i=0niNi/∑i=0nNi Antibiotics experiment A cocktail of vancomycin (0.5 g/L), ampicillin (1.0 g/L), neomycin (1.0 g/L; Sigma–Aldrich), and metronidazole (0.5 g/L; Sigma–Aldrich) was prepared in water and administered by oral gavage at a dose of 200 μL every day. Fecal microbiome transplantation Feces were obtained from C57BL/6J mice and stored at −80 °C until use. The feces were dissolved in 1.5 mL PBS, homogenised, and filtered through a 100-μm nylon mesh strainer (Corning, Tewksbury, MA, USA). The filtered contents were resuspended in PBS to a total volume of 2 mL, of which 150 μL per mouse was administered to GF mice in a vinyl isolator by oral gavage. CD8+ T cell depletion Mice received 200 μg every other day of either isotype antibody (Rat IgG2b; Bio X cell, USA) or anti-CD8α antibody (Bio X Cell, USA) to ablate CD8+ T cells. Macrophage depletion Mice were treated with clodronate liposomes using the Macrokiller kit (MKV300, Cosmobio, Tokyo, Japan). Control: Control animals received empty (vacant) liposomes provided within the same Macrokiller kit. Anti-mouse SPP1 antibody CD11b+ Macrophages were treated with an anti-mouse/human osteopontin (SPP1) monoclonal antibody (Bio X Cell, Lebanon, NH, USA). Control: An isotype control antibody (Mouse IgG2c isotype control, Bio X Cell, Lebanon, NH, USA) was used as a negative control. Each antibody was used in 10 μg/mL at final concentration. Immunofluorescence staining Formalin-fixed, paraffin-embedded (FFPE) GC tissues obtained from biopsy or surgical resection were sliced into 4-μm-thick sections and used for immunofluorescence staining. Sections from tumour were deparaffinized in xylene (5 min × 3), followed by rehydration through a graded ethanol series (100%, 90%, and 70%; 5 min each), and rinsed in distilled water. Antigen retrieval was performed using HistoVT One at 90 °C for 30 min, followed by cooling at room temperature for 30 min. Sections were then permeabilized with 0.5% PBST (PBS containing Triton X-100) for 10 min and subsequently blocked with blocking buffer. Sections were incubated overnight at 4 °C with primary antibodies diluted in blocking buffer. The following primary antibodies were used: mouse monoclonal anti-human CD68 antibody (Abcam, USA) and rabbit anti-human CD8α antibody (Abcam, USA). After washing with PBS, sections were incubated with fluorophore-conjugated secondary antibodies for 2 h at room temperature in the dark: Goat anti-Mouse IgG (H + L) Alexa Fluor 488 (Abcam, USA) and F (ab')2 Goat anti-Rabbit IgG (H + L) Alexa Fluor 555 (Abcam, USA). Nuclei were counterstained with DAPI and sections were mounted using antifade mounting medium. Fluorescent images were acquired using a confocal laser scanning microscope (Cell-Voyager CQ1; Yokogawa, Tokyo, Japan) with 20 × and 40 × objective lenses. Image processing (brightness, contrast, and colour balance) was applied uniformly across entire images. Cell counting was performed in five randomly selected fields per tissue section at 200 × magnification and obtained the data with mean of their counts in each patient.35 Bone marrow-derived macrophage polarisation Femurs, fibulae, and humeri were harvested from 8 to 11-week-old female C57BL/6 mice. Bone marrow cells were flushed from the bones with 5 mL of HBSS using a 21-gauge needle. The collected cells were treated with RBC lysis buffer (pluriSelect, USA). Cells were then cultured in 150-mm cell culture dishes in high-glucose DMEM (Nacalai Tesque) supplemented with 20 ng/mL M-CSF (PeproTech, USA) and 10% FBS for 7 days in a 37 °C CO2 incubator. On day 7, the supernatant was removed, cells were washed with PBS, and adherent cells were detached by treatment with TrypLE Express (Gibco) for 8 min. A total of 5 × 105 cells were seeded into each well of a 24-well plate. On day 8, IL-4 (R&D Systems) and IL-13 (R&D Systems) were added at a concentration of 20 ng/mL together with d-serine, l-serine, or control medium at a final concentration of 1%. After 24 h, the supernatant was removed, and adherent cells were lysed and harvested using Buffer RLT from the RNeasy Micro Kit (Qiagen). Real-time quantitative reverse-transcription PCR analysis RNA was isolated and purified using RNeasy Micro Kit (Qiagen) according to manufacturer's instructions. Reverse transcription was performed using an iScript cDNA Synthesis Kit (Bio-Rad). Real-time quantitative PCR amplification was performed using the KAPA SYBR Fast qPCR Kit (Kapa Biosystems) and a CFX Opus 96 Real-Time PCR System (Bio-Rad). Gene expression levels were normalised to Gapdh mRNA. The following primer sequences (5′–3′) were used: Gapdh, forward, CATCACTGCCACCCAGAAGACTG, reverse, ATGCCAGTGAGCTTCCCGTTCAG; Spp1, forward, AGCAAGAAACTCTTCCAAGCAA, reverse, GTGAGATTCGTCAGATTCATCCG36; Vegfa, forward, CCACGACAGAAGGAGAGCAGAAGTCC, reverse, CGTTACAGC AGCCTGCACAGCG.37 Serine racemase immunohistochemical staining Formalin-fixed, paraffin-embedded (FFPE) gastric cancer tissues obtained from biopsy or surgical resection were sectioned at 4 μm thickness and subjected to immunohistochemical (IHC) staining. IHC staining was performed on a BOND RX automated staining platform (Leica Biosystems, Wetzlar, Germany). Deparaffinization was carried out according to the manufacturer's standard protocol. Serine racemase (SRR) IHC staining was performed using an anti-serine racemase antibody (Santa Cruz Biotechnology, Dallas, TX) at a dilution of 1:100. Antigen retrieval was performed using BOND Epitope Retrieval Solution 1 (pH 6.0, Leica Biosystems) at 98 °C for 10 min. Chromogenic development was carried out for 10 min. Following staining, the sections were dehydrated through five changes of 100% ethanol, cleared in xylene, and cover slipped. Staining intensity was scored as follows: SRR-negative, 0; SRR weak-positive, 1; and SRR strong-positive, 2. The proportion of stained area was scored as follows: 0%, 0; 0 < x < 10%, 1; 10% ≤ x < 50%, 2; and x ≥ 50%, 3. Based on these two parameters, the H-score was defined as the sum of the products of the staining intensity score and the corresponding proportion score for each staining category. (Max = 8. For example, (strong positive, 2) x (x ≥ 50%, 3) + (strong positive, 2) x (10% ≤ x < 50%, 2) = 8). 16S rRNA gene sequencing analysis Fresh fecal samples were collected, stored at −80 °C, and used for microbial composition analysis.20,30,31,38, 39, 40, 41, 42, 43, 44, 45 The V3–V4 hypervariable regions of the 16S rRNA gene were amplified using Ex Taq Hot Start (Takara, Kusatsu, Japan) and purified using AMPure XP beads (Beckman Coulter, Brea, CA, USA). Equimolar amplicon pools were sequenced on an Illumina MiSeq using the MiSeq Reagent Kit v3 (600 cycles), according to the manufacturer's instructions. Primer sequences were trimmed from the raw reads using Cutadapt (v4.6; forward: CCTACGGGNGGCWGCAG; reverse: GACTACHVGGGTATCTAATCC).46 DADA2 (v1.30.0) was employed for quality filtering, paired-end merging, PCR chimera removal, and exact amplicon sequence variant (ASV) inference.47 Barrnap (version 0.9) was used to remove ASVs other than those of bacteria (https://github.com/tseemann/barrnap). Non-bacterial ASVs were filtered out using Barrnap (v0.9; https://github.com/tseemann/barrnap). Representative sequences and ASV tables were generated using the QIIME 2 plugin (v2023.7.0) after removing chimeric and low-abundance sequences.30,48 Taxonomy was assigned, and phylogenetic relationships were inferred against the SILVA 138 99% OTU reference using the feature-classifier plugin.49 Alpha diversity metrics were calculated on data rarefied to 17,522 sequences per sample, and beta diversity was assessed via weighted UniFrac distances and visualised using principal coordinate analysis (PCoA), with statistical significance determined using PERMANOVA. Taxonomic bar plots were created using the microshades package (v1.12),50 and differential abundance was determined by analysing the composition of microbiomes with bias correction 2 (AMCOM-BC2)51 using the qiime2 command. Single-cell RNA sequencing analysis We inoculated a 1% solution of D-ser or saline (Cont) into MC38-bearing mice. CD45+ and FVD− cells from tumour-infiltrating mononuclear cells were sorted using FACS Aria II (BD Biosciences).30,31,33,52, 53, 54 CD45+ and FVD− cells were pooled from the tumour-infiltrating mononuclear cells of three control or three D-ser mice, and 30,000 cells were loaded into a Chromium Controller (10× Genomics). RNA-seq libraries were prepared using Chromium Single Cell 5′ Reagent Kits (v2 Chemistry Dual Index) according to the manufacturer's instructions (10× Genomics). The libraries were sequenced using DNBSEQ-G400 (MGI Tech). Cell Ranger (v 7.1.0) processed sequencing reads from all samples (10× Genomics). Seurat (v 5.2.1)55 was used to aggregate and analyse the processed data following the official guidelines (https://satijalab.org/seurat/articles/pbmc3k_tutorial.html). Specifically, PCA was performed to identify clusters, and 17 gene clusters (0–16) were projected onto the Uniform Manifold Approximation and Projection (UMAP) space with normalised gene expression. Cell–cell communication analysis was performed using CellChat (v. 1.6.1). Volcano plots were generated using EnhancedVolcano (v. 1.22.0) (https://github.com/kevinblighe/EnhancedVolcano) (https://github.com/kevinblighe/EnhancedVolcano). Vlnplot2 from Seurat Extend (v 1.0.8) (https://github.com/huayc09/SeuratExtend) was used to plot and compare gene expression in both groups using the Wilcoxon test. Quantification and statistical analysis The results are generally presented as mean ± SEM. Statistical analyses were performed using GraphPad Prism v10.3.1 (GraphPad Software, Boston, Massachusetts USA). Data normality was assessed using the Shapiro–Wilk test. For normally distributed variables, comparisons between the two groups were made using a two-tailed unpaired Student's t-test. For analyses involving three or more groups, one-way ANOVA followed by Tukey's post-hoc test was applied. Clinical data were analysed using JMP® 18 (JMP Statistical Discovery LLC, Cary, NC, USA). Role of funders All funders did not participate in this study's design, data collection, data analysis, interpretation, or writing of the report. Key resources table Please refer to Supplementary Table S1. List of software and algorithms used in this study Please refer to Supplementary Table S2 provides information on the software and algorithms. List of reagents used in this study Please refer to Supplementary Table S3 provides information on the reagents used. Results d-serine administration selectively promotes tumour growth To determine whether specific amino acids influence tumour progression, MC38 cells were transplanted subcutaneously into mice that were administered 1% solutions of D- or l-serine (ser), D- or l-alanine (ala), D- or l-proline (pro), D- or l-asparagine (asn), or water as a control (Cont) (Fig. 1A). When tumour growth was compared within each amino acid pair, D-ser supplementation uniquely and significantly accelerated tumour growth relative to its corresponding l-enantiomer, whereas no significant differences were observed between D- and L-forms of ala, pro, or asn (Fig. 1B). Moreover, D-ser supplementation significantly increased tumour volume compared to that in the control, D-asn, and D-ala groups (Fig. 1B, Supplementary Fig. S1A). Collectively, these results demonstrate that D-ser selectively enhances tumour growth in vivo, distinguishing it from both its l-enantiomer and other tested D-AAs.Fig. 1d-serine administration enhances the tumour volume by altering immune cells. (A and B) Shema of the Experiment. Water (Control: Cont) or 1% of D- or l-amino acid (D-ser, D-asn, D-ala, D-pro, L-ser, L-asn, L-ala, L-pro) was administered to SPF C57BL/6J mice for 14 days, followed by MC38 tumour transplantation. (B) MC38 tumour volume on day 18 in each condition (n = 6–18, combined 4 independent experiments). (C) Single-cell RNA-seq analysis: tumour-infiltrated lymphocytes from control mice and 1% D-ser-treated mice 18 days after MC38 transplantation (n = 3 per group). CD45+ cells were sorted by FACS. (D) UMAP of CD45+ cell clusters following data collection and quality control. (E) Frequency of each cluster in the control (Cont) and d-serine-treated (D-ser) mice. (F) Frequency of each cluster in individual mice (n = 3). (G) UMAP of macrophage clusters from (D). (H) Frequency of each macrophage cluster in the control and D-ser-treated mice in (G). (I) Differentially expressed genes in the macrophage cluster comparing Cont and D-ser conditions. (J) Dot plots showing differentially expressed genes in each macrophage cluster comparing Cont and D-ser conditions. (K) Cx3cr1 and Ly6c expression in tumour-infiltrated macrophages (gated on FVD−CD45+TCRb−CD11b+CD11c− cells) (n = 4 per group). (K) Representative flow cytometry plots and percentage of Ly6C+CX3CR1+ cells and Ly6C−CX3CR1+ cells within FVD−CD45+TCRb−CD11b+CD11c− cells of tumour-infiltrated lymphocytes. (L) Representative histogram and percentage of MHC-II+ cells within FVD−CD45+TCRβ−CD11b+CD11c− tumour-infiltrating leukocytes. All data are presented as the mean ± SEM. p values were determined using the Kruskal–Wallis test post Dunn's multiple comparisons tests (B), the Student's t-test (F, H, K, L), and the Wilcoxon rank-sum test (I). All data are combined in two to four independent experiments. d-serine administration enhances the tumour volume by altering immune cells. (A and B) Shema of the Experiment. Water (Control: Cont) or 1% of D- or l-amino acid (D-ser, D-asn, D-ala, D-pro, L-ser, L-asn, L-ala, L-pro) was administered to SPF C57BL/6J mice for 14 days, followed by MC38 tumour transplantation. (B) MC38 tumour volume on day 18 in each condition (n = 6–18, combined 4 independent experiments). (C) Single-cell RNA-seq analysis: tumour-infiltrated lymphocytes from control mice and 1% D-ser-treated mice 18 days after MC38 transplantation (n = 3 per group). CD45+ cells were sorted by FACS. (D) UMAP of CD45+ cell clusters following data collection and quality control. (E) Frequency of each cluster in the control (Cont) and d-serine-treated (D-ser) mice. (F) Frequency of each cluster in individual mice (n = 3). (G) UMAP of macrophage clusters from (D). (H) Frequency of each macrophage cluster in the control and D-ser-treated mice in (G). (I) Differentially expressed genes in the macrophage cluster comparing Cont and D-ser conditions. (J) Dot plots showing differentially expressed genes in each macrophage cluster comparing Cont and D-ser conditions. (K) Cx3cr1 and Ly6c expression in tumour-infiltrated macrophages (gated on FVD−CD45+TCRb−CD11b+CD11c− cells) (n = 4 per group). (K) Representative flow cytometry plots and percentage of Ly6C+CX3CR1+ cells and Ly6C−CX3CR1+ cells within FVD−CD45+TCRb−CD11b+CD11c− cells of tumour-infiltrated lymphocytes. (L) Representative histogram and percentage of MHC-II+ cells within FVD−CD45+TCRβ−CD11b+CD11c− tumour-infiltrating leukocytes. All data are presented as the mean ± SEM. p values were determined using the Kruskal–Wallis test post Dunn's multiple comparisons tests (B), the Student's t-test (F, H, K, L), and the Wilcoxon rank-sum test (I). All data are combined in two to four independent experiments. As exogenous D-ser could theoretically be converted to L-ser in vivo, we quantified the circulating amino acid levels. D-ser administration did not upregulated L-ser concentration in plasma, whereas plasma D-ser levels were substantially elevated in D-ser–treated mice (Supplementary Fig. S1B), indicating that the observed phenotype was attributable to D-ser. To determine whether D-ser directly promoted tumour cell proliferation, MC38 cells were cultured in vitro with D-ser. Even at high concentrations (0.1%), D-ser did not alter MC38 proliferation rates (Supplementary Fig. S1C). Together, these findings demonstrate that D-ser administration promotes tumour progression in vivo through host-dependent mechanisms rather than direct effects on tumour cell growth. d-serine remodels the tumour immune microenvironment by increasing macrophages and reducing CD8+ T cells To elucidate the mechanisms underlying D-ser driven tumour progression, we first examined tumour-infiltrating immune cells. The total number of CD45+ immune cells isolated from MC38 tumours, normalised to tumour weight, was comparable between D-ser-treated and control mice (hereafter referred to as D-ser and Cont mice, respectively) (Supplementary Fig. S2A, left two panels). We therefore input an equal number of CD45+ cells (10,000 cells per mouse) for single-cell RNA sequencing (scRNA-seq). Following quality control, 12,752 cells from Cont mice and 14,201 cells from D-ser mice (n = 3 per group) were retained for downstream analysis (Fig. 1C; Supplementary Fig. S2A, right panel). Unsupervised clustering identified 27 distinct clusters (26 immune cell clusters and one endothelial cell cluster), which were annotated based on the expression of canonical marker genes (Fig. 1D; Supplementary Fig. S2B and C). These included three CD8+ T cell subsets, two CD4+ T cell subsets, γδ T cells, four NK cell clusters, plasmacytoid dendritic cells (pDCs), conventional dendritic cells (DCs), monocytes, ten macrophage clusters, B cells, mast cells, osteoclasts, and endothelial cells. Next, we compared the relative abundance of each tumour-associated immune cell cluster between the Cont and D-ser mice. D-ser administration significantly increased the proportion of several macrophage clusters, including angiogenic macrophages, Fn1+ macrophages, and three clusters of proliferative macrophages in tumours (Fig. 1E and F). In contrast, the proportion of antigen-presenting macrophages among total tumour-infiltrating immune cells was significantly reduced in D-ser-treated tumours. Additionally, cytotoxic CD8+ T cells were markedly decreased in D-ser mice, whereas other immune populations, including CD4+ T cells, NK cells, and B cells, remained largely unchanged. Together, these findings suggest that D-ser shapes the tumour immune landscape by promoting angiogenic, Fn1+, and proliferative macrophage subsets while reducing antigen-presenting macrophages and cytotoxic CD8+ T cell populations. d-serine drives pro-tumoural macrophage alteration Given the well-established role of tumour-associated macrophages (TAMs) in tumour progression,15,18,56 we next analysed TAM subsets from D-ser and Cont mice. The total number of TAM was higher in D-ser–treated mice than in control mice; however, this difference did not reach statistical significance after normalisation to tumour weight (Supplementary Fig. S2D, middle panel). This trend was consistent with the scRNA-seq data, which showed an increased representation of macrophage clusters in D-ser–treated tumours (Supplementary Fig. S2D, right panel). When cluster proportions were reanalysed within the macrophage compartment, TAMs from D-ser mice showed significant enrichment of angiogenic macrophages and one proliferative macrophage cluster compared with those from Cont mice. In contrast, antigen-presenting macrophages were significantly more abundant in control tumours than in D-ser mice (Fig. 1G and H). In addition, the proportion of Cxcl9+ macrophages showed a trend toward enrichment in the Cont mice compared to the D-ser mice. These results demonstrate that D-ser does not merely increase overall macrophage abundance but actively establishes the macrophage compartment toward pro-tumoural and proliferative phenotypes at the expense of antigen-presenting subsets. We then examined the differences in the expression of genes in macrophages in D-ser treated and Cont mice. Consistent with the cluster analysis, differential gene expression analysis in tumour macrophages from D-ser mice and Cont mice revealed that antigen-presentation and inflammatory genes (H2-Ab1, Ciita, Irf7, Tap1, Tap2, Ifitm1, Cxcl9, Ccl5)57 were enriched in macrophages from Cont mice, whereas immunosuppressive and pro-tumoural genes (Spp1, Arg1, Hmox1, Fn1, Vegfb, Tmem119)57 were preferentially expressed in macrophages from D-ser–treated mice (Fig. 1I, Supplementary Fig. S2E). Moreover, the expression levels of these inflammatory genes among most macrophage clusters were higher in the Cont mice than in the D-ser mice, whereas those of immunosuppressive genes among most macrophage clusters were lower in the Cont mice than in the D-ser mice (Fig. 1J), suggesting that D-ser drives immunosuppressive gene expression, promotes the proportion of angiogenic macrophages, and reduces pro-inflammatory clusters, such as antigen-presenting macrophage and Cxcl9+ macrophage clusters, compared with those in the Cont mice. Flow cytometric analysis in tumour macrophages confirmed that the proportion of Ly6C+CX3CR1+ macrophages among TAMs which correspondence with the proinflammatory macrophages predominated in Cont mice, whereas the proportion of Ly6C−CX3CR1+ macrophages among TAMs was enriched following D-ser treatment (Fig. 1K, Supplementary Fig. S2F). In addition, the ratio of MHC-II+ cells in TAMs of D-ser mice was less than that of Cont mice, supporting the reduction of antigen-presenting macrophages in D-ser mice (Fig. 1L). Together, these results suggest that D-ser promotes tumour progression by modulating macrophages toward immunosuppressive and pro-tumoural states. d-serine reduces cytotoxic CD8+ T cell abundance and effector function Given the reduction in the cytotoxic CD8 T population observed in D-ser-treated tumours, we next examined CD8+ T cell heterogeneity using scRNA-seq data (Fig. 2A, Supplementary Fig. S3A). Nine CD8+ T cell subsets were identified, including IFN-response, effector, effector-memory, early activated, exhausted progenitor (effector like exhausted CD8+ T cell, named Texeff), central-memory (CD8 Tem), proliferating CD8, and naïve populations (Fig. 2A). Both flow cytometry and scRNA-seq analyses consistently demonstrated a reduction in CD8+ T cells in D-ser-treated tumours. Specifically, the absolute number of total CD8+ T cells and tumour weight–normalised CD8+ T cell counts measured by FACS were significantly lower in D-ser mice than in Cont mice (Fig. 2B, left 2 panels). Consistent with these findings, scRNA-seq data also showed a reduced total number of CD8+ T cells in D-ser mice (Fig. 2B, middle panel).Fig. 2D-ser administration reduces tumour infiltrated CD8+ T cells. (A) UMAP of CD8 T cell clusters. UMAP was reannotated from CD8 clusters (Fig. 1D). (B) Total number of CD8+ cells in MC38 tumours from mice treated with or without 1% D-ser in drinking water (D-ser and Cont mice). Cell numbers were quantified by FACS and normalised to tumour weight (left 2 panels, n = 3 each group). Cell counts of total CD8+ T cells and each CD8+ T cell subset in scRNA-seq analysis, comparing the Cont and D-ser mice (middle and right panels, n = 3). (C) Differential gene expression of CD8+ T cells between the Cont and D-ser mice. (D) Expression levels of each gene in CD8+ T cells from control and D-ser mice. (E–G) Flow cytometric quantification of IFN-γ+CD8+ T cells, TNF-α+CD8+ T cells, and PD-1+CD8+ T cells. (E) Representative flow cytometry plots. (F) Percentage of IFN-γ, TNF-α, and PD-1 expressing cells in FVD−CD45+TCRβ+ cells of tumour-infiltrated lymphocytes. (G) Percentage of IFN-γ, TNF-α, and PD-1 expressing cells in FVD−CD45+TCRβ+CD8α+ cells of tumour-infiltrated lymphocytes. All data are presented as the mean ± SEM. p values were determined using Wilcoxon test (C, D), Student's t-test (B, F, G) (n = 9–11 per group; all data were combined from three independent experiments). D-ser administration reduces tumour infiltrated CD8+ T cells. (A) UMAP of CD8 T cell clusters. UMAP was reannotated from CD8 clusters (Fig. 1D). (B) Total number of CD8+ cells in MC38 tumours from mice treated with or without 1% D-ser in drinking water (D-ser and Cont mice). Cell numbers were quantified by FACS and normalised to tumour weight (left 2 panels, n = 3 each group). Cell counts of total CD8+ T cells and each CD8+ T cell subset in scRNA-seq analysis, comparing the Cont and D-ser mice (middle and right panels, n = 3). (C) Differential gene expression of CD8+ T cells between the Cont and D-ser mice. (D) Expression levels of each gene in CD8+ T cells from control and D-ser mice. (E–G) Flow cytometric quantification of IFN-γ+CD8+ T cells, TNF-α+CD8+ T cells, and PD-1+CD8+ T cells. (E) Representative flow cytometry plots. (F) Percentage of IFN-γ, TNF-α, and PD-1 expressing cells in FVD−CD45+TCRβ+ cells of tumour-infiltrated lymphocytes. (G) Percentage of IFN-γ, TNF-α, and PD-1 expressing cells in FVD−CD45+TCRβ+CD8α+ cells of tumour-infiltrated lymphocytes. All data are presented as the mean ± SEM. p values were determined using Wilcoxon test (C, D), Student's t-test (B, F, G) (n = 9–11 per group; all data were combined from three independent experiments). Proportional analysis indicated a shift toward naïve and central memory (Tcm) CD8+ T cell subsets in D-ser–treated mice, with a concomitant decrease in IFN-response and Texeff CD8+ T cell populations; however, these differences were not statistically significant (Supplementary Fig. S3B). Importantly, despite the equivalent number of immune cells being input for scRNA-seq analysis (Supplementary Fig. S2A), the absolute number of Texeff CD8+ T cells was reduced in D-ser mice compared to that in the Cont mice (Fig. 2B, right panel). Taken together, these results indicate that D-ser reduces the absolute number of the effector CD8+ T cells, thereby impairing antitumour immunity. To quantify the expression level of each gene, DEG analysis of CD8+ T cells in Cont mice and D-ser mice revealed that CD8+ T cells from Cont mice were enriched for cytotoxic and inflammatory gene programs, including Gzmb, Gzme, Gzmf, Cxcr6, Ccl5, and Entpd1. In contrast, CD8+ T cells from D-ser-treated tumours preferentially expressed genes associated with naïve or central memory states, including Ccr7, Tcf7, Slamf6, and Itgb1 (Fig. 2C and D). Flow cytometric analysis validated these transcriptomic findings, revealing significantly lower frequencies of IFN-γ+, TNF-α+, and PD-1+ CD8+ T cells in D-ser mice than in Cont mice (Fig. 2E–G, Supplementary Fig. S3C). These data demonstrate that D-ser suppresses the quantity and functional capacity of tumour-infiltrating CD8+ T cells. d-serine modulates the tumour immune microenvironment through altered cell–cell communication To investigate whether D-ser affects cell–cell communication in the tumour microenvironment, we performed a comprehensive analysis of cell–cell signalling pathways in D-ser and Cont mice (Fig. 3A, Supplemental Fig. 4A–D). Top 5% increased signals in D-ser mice were composed of Trem2+ macrophages-cytotoxic NK, osteoclasts, proliferating macrophages-DCs, and top 5% decreased signals in D-ser mice were composed of Cxcl9+ macrophages and antigen-presenting macrophages-CD8 T cells. Notably, other macrophage populations, including IFN-response, Trem2+, angiogenic, Fn1+, and proliferating macrophage clusters, were also associated with CD8+ T cell clusters. Other small clusters, such as endothelial cells, showed decreased communication with CD8+ T cells and macrophages in D-ser mice.Fig. 3D-Ser modulates interaction between macrophages and CD8 T cells. (A) CellChat analysis was performed to compare cell–cell communication networks among tumour-infiltrating lymphocytes (TILs) from control (Cont) and D-ser–treated (D-ser) tumours. (A) Circle plots display the top 5% of signalling pathways ranked by the magnitude of change in overall communication strength between conditions (Upper panel: interactions increased in D-ser, Lower panel: interactions decreased in D-ser). Edge thickness indicates interaction strength, and node size reflects the relative contribution of each cell type to the signalling network. (B–D) CD11b+ macrophages (105 cells) were isolated form tumour in Cont mice or D-ser treated mice and co-cultured with BV-labelled naïve CD8 T cells (105 cells) from CD45.1 mice for 72 h. (C) BV staining among CD8+CD45.1+TCRβ+FVD− cells in each condition (medium; naïve CD8 T cells only, anti-CD3+CD28; naïve CD8 T cells plus anti-CD3 antibody and anti-CD28 antibody). (D) Division index and percent divide index were calculated in each condition. (E,F) CellChat analysis was performed to compare normalised cell–cell communication signals from macrophages to CD8+ T cell populations in control (Cont) and d-serine–treated (D-ser) tumours. (E) Circle plots display the top 5% of signalling pathways ranked by the magnitude of change in overall communication strength between conditions (Upper panel: interactions increased in D-ser, Lower panel: interactions decreased in D-ser). Edge thickness indicates interaction strength. (F) Heatmap showing the top 10 signalling pathways with the largest changes in communication strength between macrophages and CD8+ cell population in Cont and D-ser mice. Red indicates increased signalling in D-ser, and blue indicates decreased signalling. (G–I) CD11b+ macrophages (105 cells) were isolated form tumour in D-ser mice and co-cultured with BV-labelled naïve CD8 T cells (105 cells) from CD45.1 mice for 72 h plus anti-SPP1 antibody or IgG2c antibody (control Ab). (H) BV staining among CD8+CD45.1+TCRβ+FVD− cells in each condition (medium; naïve CD8+ T cells only, D-ser Mac + control Ab; naïve CD8+ T cells plus anti-CD3 ab and control Ab, D-ser Mac + SPP1 Ab; naïve CD8+ T cells plus anti-CD3 Ab and anti-SPP1 Ab). (I) Division index and percent divide index were calculated in each condition. All data are presented as the mean ± SEM. p values were determined by two-way ANOVA with correction using the Geisser–Greenhouse method (D, I). All data are combined in two independent experiments. D-Ser modulates interaction between macrophages and CD8 T cells. (A) CellChat analysis was performed to compare cell–cell communication networks among tumour-infiltrating lymphocytes (TILs) from control (Cont) and D-ser–treated (D-ser) tumours. (A) Circle plots display the top 5% of signalling pathways ranked by the magnitude of change in overall communication strength between conditions (Upper panel: interactions increased in D-ser, Lower panel: interactions decreased in D-ser). Edge thickness indicates interaction strength, and node size reflects the relative contribution of each cell type to the signalling network. (B–D) CD11b+ macrophages (105 cells) were isolated form tumour in Cont mice or D-ser treated mice and co-cultured with BV-labelled naïve CD8 T cells (105 cells) from CD45.1 mice for 72 h. (C) BV staining among CD8+CD45.1+TCRβ+FVD− cells in each condition (medium; naïve CD8 T cells only, anti-CD3+CD28; naïve CD8 T cells plus anti-CD3 antibody and anti-CD28 antibody). (D) Division index and percent divide index were calculated in each condition. (E,F) CellChat analysis was performed to compare normalised cell–cell communication signals from macrophages to CD8+ T cell populations in control (Cont) and d-serine–treated (D-ser) tumours. (E) Circle plots display the top 5% of signalling pathways ranked by the magnitude of change in overall communication strength between conditions (Upper panel: interactions increased in D-ser, Lower panel: interactions decreased in D-ser). Edge thickness indicates interaction strength. (F) Heatmap showing the top 10 signalling pathways with the largest changes in communication strength between macrophages and CD8+ cell population in Cont and D-ser mice. Red indicates increased signalling in D-ser, and blue indicates decreased signalling. (G–I) CD11b+ macrophages (105 cells) were isolated form tumour in D-ser mice and co-cultured with BV-labelled naïve CD8 T cells (105 cells) from CD45.1 mice for 72 h plus anti-SPP1 antibody or IgG2c antibody (control Ab). (H) BV staining among CD8+CD45.1+TCRβ+FVD− cells in each condition (medium; naïve CD8+ T cells only, D-ser Mac + control Ab; naïve CD8+ T cells plus anti-CD3 ab and control Ab, D-ser Mac + SPP1 Ab; naïve CD8+ T cells plus anti-CD3 Ab and anti-SPP1 Ab). (I) Division index and percent divide index were calculated in each condition. All data are presented as the mean ± SEM. p values were determined by two-way ANOVA with correction using the Geisser–Greenhouse method (D, I). All data are combined in two independent experiments. To further delineate how D-ser shapes immune interactions within the tumour microenvironment, we performed a network analysis of cell–cell communication, comparing outgoing and incoming signalling between D-ser and control mice. Differential heatmap analysis (D-ser minus control) revealed distinct pathway alterations induced by D-ser (Fig. 3B; Supplementary Fig. S4A–E). D-ser administration markedly enhanced both outgoing and incoming FN1 and SPP1 signalling, particularly in macrophage populations, with strengthened interactions toward cytotoxic NK, osteoclast, and dendritic cell clusters (Fig. 3B; Supplementary Fig. S4A and B). In contrast, antigen presentation–related pathways, including MHC-I, MHC-II, and CD86, were modestly reduced in the signals sent from macrophage subsets to T cell clusters (Supplementary Fig. S4C–E). Specifically, recruitment and activation signals mediated by the CXCL axis, including Cxcl16-Cxcr6, exhibited markedly lower interaction strengths between multiple macrophage clusters and CD8+ T clusters in D-ser mice. These data suggest that D-ser suppresses the inflammatory crosstalk between macrophages and CD8+ T cells, potentially contributing to a reorganised immune landscape. To assess the immunosuppressive capacity of macrophages against CD8+ T cells, CD11b+ TAMs were isolated from tumours of control or D-ser mice and co-cultured with naïve CD8+ T cells. The Percent Divided and Division Index of CD8+ T cells were comparable between D-ser cultured or L-ser cultured medium without macrophages, suggesting that D-ser did not directly regulate the activation of CD8+ T cells (Supplement Fig. S5A–C). CD8+ T cells cocultured with macrophages from D-ser mice displayed significantly reduced Percent Divided and Division Index compared with those from Cont mice (Fig. 3B–D), demonstrating that D-ser–conditioned macrophages suppress the proliferative capacity of CD8+ T cell proliferative capacity. To clarify the macrophage subset and signalling pathways responsible for D-ser–mediated immunosuppression, we performed additional CellChat analyses focussing on macrophage–CD8+ T cell interactions. Consistent with our initial findings, D-ser treatment markedly altered intercellular communication between TAM subsets and CD8+ T cells (Fig. 3E). These data suggest that D-ser administration did not only increase or decrease communication from a specific macrophage subset, but also it shifted the predicted macrophage interactions from proliferating CD8+ T cells toward cytotoxic CD8+ T cells. Among the identified signalling pathways, SPP1- and FN1-associated signalling was significantly increased in macrophage populations from D-ser–treated tumours (Fig. 3F, Supplementary Fig. S5D and E). Notably, previous studies have demonstrated that SPP1-expressing macrophages are associated with T cell dysfunction and immunosuppressive tumour microenvironments.58, 59, 60, 61 To functionally test the suppressive effect of D-ser–conditioned macrophages on CD8+ T cells through SPP1 signalling, we co-cultured naïve CD8+ T cells with macrophages isolated from D-ser–treated mice in the presence of an anti-SPP1 neutralising antibody or IgG2c control antibody (Fig. 3G). Importantly, blockade of SPP1 signalling partially restored CD8+ T cell proliferation (Fig. 3H and I), indicating that SPP1 signalling contributes to the immunosuppressive interaction between macrophages and CD8+ T cells under D-ser conditions. Furthermore, to investigate whether D-ser directly induces an immunosuppressive macrophage phenotype, we cultured bone marrow derived macrophage (BM-Mac) under IL-4-polarising conditions plus D-ser or L-ser62,63 (Supplementary Fig. 5F). The expression level of SPP1 and Vegfa in BM-Mac with D-ser condition increased significantly than that in BM-Mac with L-ser and control conditions. Together, these results support D-ser promotes an SPP1 expression on macrophage, which subsequently suppresses CD8+ T cell proliferative responses through macrophage–T cell communication pathways. d-Serine promotes tumour growth through macrophage and CD8+ T cell dependent manner To determine whether macrophages and CD8+ T cells are key to D-ser mediated tumour progression, we perturbated macrophages and CD8 T cells in D-ser treated mice. First, to determine whether macrophages mediate the immunosuppressive effects of D-ser in vivo, we performed macrophage depletion using clodronate liposomes in MC38 tumour-bearing mice treated with or without D-ser (Fig. 4A–C). Notably, macrophage depletion completely abolished the tumour-promoting effect of D-ser, and tumour volumes were comparable between D-ser–treated and control mice following macrophage depletion. These findings demonstrate that macrophages are essential mediators of D-ser–induced tumour progression and support a central role for tumour-associated macrophages in the immunosuppressive effects of D-ser in vivo.Fig. 4D-ser promotes tumour growth through macrophages and CD8+ T cells dependent manner. (A–C) Tumour growth curve of the MC38 tumour model treated with liposomal control or clodronate liposome in B6J mice. liposomal control or clodronate liposome was intravenously injected on day 5, 8, and 11 in C57BL/6J mice treated with water or 1% of D-ser. (liposomal control with water; ContEmp, liposomal control with D-ser; D-serEmp, Clodronate liposome with water; ContΔMac, Clodronate liposome with D-ser; D-serΔMac) (D) Tumour growth curve of the MC38 tumour model treated with water or 1% of D-ser in Rag2−/− mice or C57BL/6J mice (Rag2−/− mice, n = 8 per group; B6J control mice, n = 7). (E and F) Control Ab (Rat IgG2a Ab) or 200 μg of anti-CD8a antibody was intraperitoneally injected every three days into MC38-inoculated C57BL/6J mice. Mice were treated with water or 1% D-ser. Tumour growth curve of MC38 tumour model treated with water or D-ser (control IgG2a Ab with water; Contiso (n = 8), control IgG2a with D-ser; D-seriso (n = 10), anti-mouse CD8α with water; ContΔCD8α (n = 7), anti-mouse CD8α with D-ser; D-serΔCD8α (n = 9) for each group). All experiments were combined with the results of two independent experiments. All data are presented as the mean ± SEM. p values were determined by two-way ANOVA with correction using the Geisser–Greenhouse method (C, D, F). D-ser promotes tumour growth through macrophages and CD8+ T cells dependent manner. (A–C) Tumour growth curve of the MC38 tumour model treated with liposomal control or clodronate liposome in B6J mice. liposomal control or clodronate liposome was intravenously injected on day 5, 8, and 11 in C57BL/6J mice treated with water or 1% of D-ser. (liposomal control with water; ContEmp, liposomal control with D-ser; D-serEmp, Clodronate liposome with water; ContΔMac, Clodronate liposome with D-ser; D-serΔMac) (D) Tumour growth curve of the MC38 tumour model treated with water or 1% of D-ser in Rag2−/− mice or C57BL/6J mice (Rag2−/− mice, n = 8 per group; B6J control mice, n = 7). (E and F) Control Ab (Rat IgG2a Ab) or 200 μg of anti-CD8a antibody was intraperitoneally injected every three days into MC38-inoculated C57BL/6J mice. Mice were treated with water or 1% D-ser. Tumour growth curve of MC38 tumour model treated with water or D-ser (control IgG2a Ab with water; Contiso (n = 8), control IgG2a with D-ser; D-seriso (n = 10), anti-mouse CD8α with water; ContΔCD8α (n = 7), anti-mouse CD8α with D-ser; D-serΔCD8α (n = 9) for each group). All experiments were combined with the results of two independent experiments. All data are presented as the mean ± SEM. p values were determined by two-way ANOVA with correction using the Geisser–Greenhouse method (C, D, F). Then, we evaluate the importance of CD8+ T cells in D-ser treated mice. MC38 tumours were implanted into Rag2−/− mice. In contrast to wild-type (B6J) mice, D-ser did not further accelerate tumour growth in Rag2−/− mice, indicating the requirement for adaptive immunity (Fig. 4D). Consistently, antibody-mediated depletion of CD8+ T cells abolished the tumour-promoting effect of D-ser, resulting in comparable tumour growth between D-ser and Cont mice (Fig. 4E and F). All together, these findings suggest that D-ser promotes tumour growth in a macrophage and CD8-dependent manner. We further evaluated the administration of D-ser increase the size in different types of tumours, so we used B16F10 melanoma cell and GC organoids derived from Apc −/− Kras G12D Trp53 −/− mice. D-ser administration enhanced B16F10 tumour growth as similar findings in MC38 cells, with increased number of macrophages and decreased number of CD8+ T cells (Fig. 5A–C). Next, we examined whether elevated D-ser levels enhance tumour progression in an orthotopic GC model. Luciferase-expressing GC organoids derived from Apc −/− Kras G12D Trp53 −/− mice were transplanted into the gastric walls of B6 mice (Fig. 5D). Engraftment was confirmed by IVIS imaging at three weeks, after which the mice were randomised to receive either water or 1% D-ser (WaterGC or D-serGC). After an additional three weeks, D-serGC mice showed significantly greater tumour progression, as reflected by higher bioluminescence intensity, increased stomach weight, and larger tumour burden, compared with WaterGC controls (Fig. 5E and F). Together, these results indicate that elevated D-ser accelerates tumour growth in both subcutaneous and orthotopic GC models and that this effect is mediated by the suppression of anti-tumour CD8+ T cell responses.Fig. 5D-Ser promotes B16F10 tumour and GC organoid tumour. (A) Tumour growth curve of the subcutaneously inoculated B16F10 tumour model treated with water or 1% of D-ser in B6J mice (Cont, n = 7; D-ser, n = 9). (B) Subcutaneously inoculated B16F10 tumour volume on day 15 in each group. (C) Total intratumoural cell counts of macrophage (gated on FVD−CD45+TCRb−CD11b+CD11c− cells) and CD8a+ T cell of the samples in each group (Cont, n = 7 per group; D-ser, n = 9). (D) Schema of mouse gastric cancer organoid model: After seven days of oral administration with water or 1% D-ser in each group, mouse gastric cancer 5 × 105 cells were injected into the stomach wall with anaesthesia. Twenty-one days after organoid transplantation, we selected mice in which the transplanted cells settled in the stomach wall by using IVIS imaging (Cont: n = 10, D-ser: n = 11). The transplanted mice were scanned again using IVIS imaging on day 42 since organoid transplantation, and all mice were sacrificed for tumour assessment on a few days later. (E) IVIS imaging at three and six weeks after organoid implantation. (F) Representative haematoxylin and eosin (H&E)-stained images of gastric cancer organoids implanted in the stomach wall of mice. (G) Stomach weight from D-ser-treated mice and control mice. All experiments were combined with the results of two independent experiments. All data are presented as the mean ± SEM. p values were determined by Student's t-test (A, B, C, G). D-Ser promotes B16F10 tumour and GC organoid tumour. (A) Tumour growth curve of the subcutaneously inoculated B16F10 tumour model treated with water or 1% of D-ser in B6J mice (Cont, n = 7; D-ser, n = 9). (B) Subcutaneously inoculated B16F10 tumour volume on day 15 in each group. (C) Total intratumoural cell counts of macrophage (gated on FVD−CD45+TCRb−CD11b+CD11c− cells) and CD8a+ T cell of the samples in each group (Cont, n = 7 per group; D-ser, n = 9). (D) Schema of mouse gastric cancer organoid model: After seven days of oral administration with water or 1% D-ser in each group, mouse gastric cancer 5 × 105 cells were injected into the stomach wall with anaesthesia. Twenty-one days after organoid transplantation, we selected mice in which the transplanted cells settled in the stomach wall by using IVIS imaging (Cont: n = 10, D-ser: n = 11). The transplanted mice were scanned again using IVIS imaging on day 42 since organoid transplantation, and all mice were sacrificed for tumour assessment on a few days later. (E) IVIS imaging at three and six weeks after organoid implantation. (F) Representative haematoxylin and eosin (H&E)-stained images of gastric cancer organoids implanted in the stomach wall of mice. (G) Stomach weight from D-ser-treated mice and control mice. All experiments were combined with the results of two independent experiments. All data are presented as the mean ± SEM. p values were determined by Student's t-test (A, B, C, G). d-Serine accelerates tumour growth independent of microbiota composition As the gut microbiota also influences tumour progression, we questioned whether the microbiota contributed to D-ser–induced tumour progression. 16S rRNA gene sequencing of fecal samples from MC38-bearing mice treated with or without D-ser revealed distinct microbial compositions (Fig. 6A, Supplementary Fig. S6A). D-ser–treated mice were enriched in Ruminococcaceae, Actinobacteria, Bifidobacterium, Faecalibacterium, and Akkermansia, whereas control mice showed higher abundances of Eubacterium and Parabacteroides (Fig. 6B). These data suggest that the microbiota composition in D-ser–treated mice differed from that in control mice. To determine whether the microbiota in D-ser-treated mice was sufficient to impact tumour growth, we performed fecal microbiota transplantation. Ex-germ free mice received fecal material from D-ser–treated or control donors and were subsequently inoculated with MC38 cells without further D-ser supplementation (Fig. 6C). Although donors treated with D-ser developed larger tumours (Fig. 6D), recipient mice displayed comparable tumour volumes regardless of the donor microbiota origin (Fig. 6E). These findings indicate that D-ser–induced microbial shifts alone do not drive tumour progression in mice.Fig. 6D-Ser accelerates tumour growth independent of D-ser induced microbiota composition. (A) 16S rRNA-seq of the microbiome from control mice and D-ser-treated mice eighteen days after subcutaneously injected MC38 cancer cells (n = 4 per group). (B) List of the microbiota enriched in each group. (C) Regimens for fecal microbiota transplantation (FMT). MC38 cancer cells were subcutaneously injected into B6J donor mice and treated with water or 1% D-ser-containing water (Day −25 to −7). Eighteen days after MC38 injection to the donor mice, fecal samples were collected from each mouse, and 16S rRNA-seq was performed to evaluate the microbiota (day −7) (A). Filtered extraction of fecal contents from each mouse gavage into germ-free recipient B6J mice (day −7). After one week of FMT, all recipient mice were transplanted MC38 cells (day 0). Eighteen days after fecal microbiota transplantation and MC38 transplantation (day 18). All mice were sacrificed. (D) Donor mice: Tumour growth curve of the MC38 tumour model treated with saline or D-ser (n = 11 per group). (E) FMT-Recipient mice: Tumour growth curve of FMT germ-free mice treated with water (n = 10 FMT from control mice, n = 9 FMT from D-ser administered mice). (F) Treatment regimen. Water or 1% D-ser with or without an antibiotic cocktail (A: ampicillin, V: vancomycin, M: metronidazole, N: neomycin) was administered to SPF C57BL/6J mice for 21 days, followed by MC38 tumour inoculation. (G) Tumour growth curve of the MC38 tumour model treated with water or 1% D-ser with or without an antibiotic cocktail. C57BL/6J mice (n = 8–12 per group) were subcutaneously inoculated with 5 × 105 MC38 tumour cells (MC38 mice) and treated with water or an antibiotic cocktail for 18 days. (H) Tumour volume on day 18. All experiments were combined with the results of two independent experiments. All data are presented as the mean ± SEM. p values were determined by PERMANOVA (A), Student's t-test (D, E) and two-way ANOVA with correction using the Geisser–Greenhouse method (H). D-Ser accelerates tumour growth independent of D-ser induced microbiota composition. (A) 16S rRNA-seq of the microbiome from control mice and D-ser-treated mice eighteen days after subcutaneously injected MC38 cancer cells (n = 4 per group). (B) List of the microbiota enriched in each group. (C) Regimens for fecal microbiota transplantation (FMT). MC38 cancer cells were subcutaneously injected into B6J donor mice and treated with water or 1% D-ser-containing water (Day −25 to −7). Eighteen days after MC38 injection to the donor mice, fecal samples were collected from each mouse, and 16S rRNA-seq was performed to evaluate the microbiota (day −7) (A). Filtered extraction of fecal contents from each mouse gavage into germ-free recipient B6J mice (day −7). After one week of FMT, all recipient mice were transplanted MC38 cells (day 0). Eighteen days after fecal microbiota transplantation and MC38 transplantation (day 18). All mice were sacrificed. (D) Donor mice: Tumour growth curve of the MC38 tumour model treated with saline or D-ser (n = 11 per group). (E) FMT-Recipient mice: Tumour growth curve of FMT germ-free mice treated with water (n = 10 FMT from control mice, n = 9 FMT from D-ser administered mice). (F) Treatment regimen. Water or 1% D-ser with or without an antibiotic cocktail (A: ampicillin, V: vancomycin, M: metronidazole, N: neomycin) was administered to SPF C57BL/6J mice for 21 days, followed by MC38 tumour inoculation. (G) Tumour growth curve of the MC38 tumour model treated with water or 1% D-ser with or without an antibiotic cocktail. C57BL/6J mice (n = 8–12 per group) were subcutaneously inoculated with 5 × 105 MC38 tumour cells (MC38 mice) and treated with water or an antibiotic cocktail for 18 days. (H) Tumour volume on day 18. All experiments were combined with the results of two independent experiments. All data are presented as the mean ± SEM. p values were determined by PERMANOVA (A), Student's t-test (D, E) and two-way ANOVA with correction using the Geisser–Greenhouse method (H). We next evaluated tumour growth under gut microbiome-depleted conditions. MC38-bearing mice were treated with a broad-spectrum antibiotic cocktail (ampicillin, vancomycin, metronidazole, and neomycin; AVMN) with or without D-ser supplementation (Fig. 6F). AVMN reduced tumour volume in control mice, consistent with previous studies. However, D-ser supplementation fully restored tumour growth in antibiotic-treated mice, resulting in significantly larger tumours than those in antibiotic-only controls (Fig. 6G and H). Together, these results demonstrate that D-ser promotes tumour progression independently of microbiome composition, supporting a direct effect of D-ser on the tumour immune microenvironment rather than an indirect microbiota-mediated mechanism. We next assessed the natural source of D-ser in mice and human. Plasma D-ser levels in mice before and after MC38 tumour implantation were unchanged, suggesting that MC38 tumour cells do not directly upregulate plasma D-ser level (Supplementary Fig. 6B). Because D-ser is generated from L-ser by serine racemase (SRR) in mammals, we next investigated whether tumour cells contribute to systemic D-ser production in human. In contrast to previous reports in colorectal cancer, SRR expression was broadly detected in human gastric tissues by IHC staining of archival specimens, whereas expression was comparable between cancerous or adjacent non-cancerous gastric lesions (Supplementary Fig. S6C and D). SRR expression of normal gastric mucosa or cancer cells showed no significant correlation with plasma concentration of D-ser in patients with GC. Together, these findings suggest that GC cells themselves are unlikely to be the primary source of plasma D-ser level. The expression of srr was not detected in murine immune cell populations in scRNA-seq data (Supplementary Fig. 6E), suggesting that immune cells are unlikely to represent a major source of D-ser production from L-ser. We therefore examined whether the gut microbiota contributes to systemic D-ser production. Following treatment with AVMN, plasma concentrations of D-ser and D-pro were markedly reduced in AVNM-treated mice, whereas D-ala levels were increased (Supplementary Fig. S6F and G). Consistently, fecal concentrations of D-ser, D-pro, and D-ala were also altered after AVNM treatment (Supplementary Fig. S6H). These findings suggest that the gut microbiota contributes to systemic D-ser levels, implicating the microbiome as a potential source of D-ser in the tumour-bearing state. Circulating d-Amino acids are elevated in patients with gastric cancer and correlate with disease stage To translate our mechanistic findings into a clinical context, we first profiled the spectrum of L- and D-AAs in the plasma of healthy controls (HCs) and patients with GC (Cohort 1; n = 14 patients with stage IV GC, n = 87 HCs; Fig. 7A, Supplementary Fig. S7A). All l-amino acids, except for allo-threonine and allo-isoleucine, were detected in both groups. A small subset of patients with GC, but not HCs, showed detectable levels of d-leucine, d-phenylalanine, and d-lysine. In contrast, D-asn, D-ser, D-ala, and D-pro were consistently detected in both GC and HC samples and were selected for further quantitative analysis of plasma, urine, and fecal samples.Fig. 7Circulatingd-amino acids are elevated in the patients with gastric cancer and correlate with disease stage. (A) Schematic representation of the study. Plasma, urea, and feces from HCs and patients with GC were obtained before treatment. All samples were analysed by two-dimensional HPLC-FLD. (B) Concentration of plasma D-ser, D-asn, D-ala, and D-pro (determined as biomarker candidates in cohort 1, shown in Supplementary Fig. S6A) in HCs and patients with GC stratified by disease stage (pathological stage (pStage) I, pStage II or III, and clinical stage (cStage) IV). (C) Concentration of urine D-ser, D-asn, D-ala, and D-pro in HCs and patients with GC stratified by disease stage (pStage I, pStage II or III, and cStage IV). (D) Concentration of fecal D-ser, D-asn, D-ala, and D-pro from HCs and patients with clinical stage IV GC. (E) ROC curve to divide HCs and patients with GC (stage I) by using concentration of serum d-amino acid. (F) Univariate analysis of the associations between concentration of plasma D-ser, D-asn, D-ala, and D-pro and clinicopathological characteristics in patients with GC. † Seven pts with recurrent GC were excluded. †† Four pts with unclassifiable histology were excluded. ††† Four pts with unknown status of HER2 were excluded. p values were determined by two-way ANOVA with correction using the Geisser-Greenhouse method (C, D, E and G). ∗∗p < 0.01, ∗∗∗p < 0.001, All data are presented as the mean ± SEM. Tumour staging was determined according to the Union for International Cancer Control (UICC) TNM Classification, 8th edition. Circulatingd-amino acids are elevated in the patients with gastric cancer and correlate with disease stage. (A) Schematic representation of the study. Plasma, urea, and feces from HCs and patients with GC were obtained before treatment. All samples were analysed by two-dimensional HPLC-FLD. (B) Concentration of plasma D-ser, D-asn, D-ala, and D-pro (determined as biomarker candidates in cohort 1, shown in Supplementary Fig. S6A) in HCs and patients with GC stratified by disease stage (pathological stage (pStage) I, pStage II or III, and clinical stage (cStage) IV). (C) Concentration of urine D-ser, D-asn, D-ala, and D-pro in HCs and patients with GC stratified by disease stage (pStage I, pStage II or III, and cStage IV). (D) Concentration of fecal D-ser, D-asn, D-ala, and D-pro from HCs and patients with clinical stage IV GC. (E) ROC curve to divide HCs and patients with GC (stage I) by using concentration of serum d-amino acid. (F) Univariate analysis of the associations between concentration of plasma D-ser, D-asn, D-ala, and D-pro and clinicopathological characteristics in patients with GC. † Seven pts with recurrent GC were excluded. †† Four pts with unclassifiable histology were excluded. ††† Four pts with unknown status of HER2 were excluded. p values were determined by two-way ANOVA with correction using the Geisser-Greenhouse method (C, D, E and G). ∗∗p < 0.01, ∗∗∗p < 0.001, All data are presented as the mean ± SEM. Tumour staging was determined according to the Union for International Cancer Control (UICC) TNM Classification, 8th edition. To assess the relationship between D-AAs and disease stage, we quantified the plasma levels of these D-AAs in an independent cohort of 108 treatment-naïve patients with GC (stage I: n = 52; stage II/III: n = 24; stage IV: n = 32) and 87 HCs (Cohort 2; Fig. 7A). Plasma concentrations of D-asn, D-ser, D-ala, and D-pro were significantly higher in patients with GC than in HCs, including those with stage I disease (Fig. 7B). The relative proportions of D-ser, D-ala, and D-pro were also elevated in patients with stage I GC compared to those in HCs (Supplementary Fig. S7B). Moreover, plasma D-ser, D-asn, and D-pro concentrations increased progressively with advancing disease stage, reaching their highest levels in the patients with stage IV GC (Fig. 7B). Because D-AAs are eliminated through renal filtration, we evaluated their relationship with the estimated glomerular filtration rate (eGFR). Although plasma D-AAs concentrations correlated with eGFR, stage-stratified regression analyses demonstrated that predicted values in the patients with stage IV GC remained higher than those in HCs (Supplementary Fig. S7C), indicating that the elevation in D-AAs was not solely driven by renal function differences. The urinary concentrations of D-ser, D-asn, and D-ala normalised to creatinine were also higher in patients with GC than in HCs, whereas D-pro levels did not differ significantly (Fig. 7C). As many D-AAs originate from food and gut microbiota, we analysed fecal samples and found higher concentrations of D-ser, D-ala, and D-pro, but not D-asn, in patients with GC, mirroring the plasma and urine findings (Fig. 7D). To evaluate the diagnostic utility, we generated ROC64 curves using plasma D-ser, D-ala, and D-pro levels to distinguish the patients with stage I GC from HCs. The combined panel achieved an AUC of 0.96, with 96.2% sensitivity and 89.7% specificity (Fig. 7E). Finally, we assessed whether the plasma D-AAs concentrations were influenced by clinicopathological variables that could potentially confound their association with GC. Univariate stratified analyses revealed that elevated plasma concentrations of D-AAs, particularly D-Ser, were associated not only with age and renal function (eGFR), but also with clinicopathological features reflecting disease burden, including lymph node metastasis, distant organ metastasis, and the number of metastatic organs (Fig. 7F, Supplementary Table 4). Notably, the association between plasma D-Ser levels and these clinicopathological factors remained significant in multivariable linear regression analyses (Supplementary Fig. S7D). Together, these data demonstrate that plasma concentrations of D-ser, D-asn, D-ala, and D-pro are elevated early in the patients with GC and progressively increase with disease burden, with D-ser showing the strongest association with advanced clinical stage. Elevated plasma D-ser is associated with an immunosuppressive tumour microenvironment and poor response to anti–PD-1 therapy in gastric cancer Based on our murine findings demonstrating that D-ser reduced PD-1+ CD8+ T cells within the tumour microenvironment, we hypothesised that circulating D-ser may influence the clinical efficacy of anti–PD-1 therapy in GC. To test this, we evaluated the association between circulating D-AAs and clinical outcomes in patients with stage IV GC treated with anti–PD-1 antibodies. Baseline plasma samples were collected from 28 patients prior to the initiation of anti–PD-1 therapy (Cohort 3) and analysed alongside samples from healthy controls (HCs) (Fig. 8A; Supplementary Fig. S8A). Consistent with the findings from Cohort 2, the absolute concentrations and relative abundances of D-ser, D-asn, D-ala, and D-pro were significantly higher in patients with GC than in HCs. Among the 28 patients, 15 achieved disease control (non-progressive disease, non-PD), whereas 13 experienced progressive disease (PD) following anti–PD-1 therapy. The proportion of patients with a poorer performance status (PS) was significantly higher in the PD group than in the non-PD group (Supplementary Fig. S8B), whereas all other baseline characteristics were comparable between the two groups. Plasma concentrations and relative abundance of D-ser and D-asn were significantly higher in the PD group, whereas D-ala, D-pro, and all l-amino acids showed no differences between the groups (Fig. 8B, Supplementary Fig. S8C). These associations remained significant after adjusting for renal function (Supplementary Fig. S8C, E). Concordantly, the urinary levels of D-ser/Cr and D-asn/Cr, but not other D-AAs, were elevated in patients with PD. Finally, we assessed whether baseline plasma D-AAs levels were associated with progression-free survival (PFS) and overall survival (OS) in patients treated with anti–PD-1 antibody and evaluated their potential predictive value for immunotherapy response. To optimise the predictive performance, we selected the best concentration of plasma D-ser based on the corrected Akaike information criterion (AICc) and Bayesian information criterion (BIC). The maximum AUC was 0.8513 when the cutoff value of the concentration of plasma D-ser was 2.18 nmoL/ml (Fig. 8C). Based on the ROC curve, we defined the cutoff value and stratified the patients into two groups: D-serhi group (plasma concentration of D-ser >2.18 nmoL/ml, n = 15) and D-serlo group (plasma concentration of D-ser <2.18, n = 13). The patients in D-serhi group exhibited significantly poorer OS but not PFS than those in D-serlo group (Fig. 8D).Fig. 8Elevated plasma D-ser is associated with an immunosuppressive tumour microenvironment and poor response to anti–PD-1 therapy in gastric cancer. (A) Plasma concentrations of D-Ser, D-Asn, D-Ala, and D-Pro in HC and patients with GC in cohort 3 (immediately prior to nivolumab administration) (B) Plasma concentrations of D-Ser, D-Asn, D-Ala, and D-Pro in cohort 3 (immediately prior to nivolumab administration), divided by the best clinical response to nivolumab monotherapy into the group of progressive disease (PD, n = 13) and the group of non-progressive disease (non PD, n = 15). Non-PD consists of one complete response (CR), two partial response (PR), 7 stable disease (SD), and 5 non-CR/non-PD. (C) ROC curve to define patients with PD and non-PD using plasma D-ser. Area under curve (AUC), Corrected Akaike Information Criterion (AICc), and Bayesian Information Criterion (BIC) were calculated. When the cut-off value for D-ser was set at 2.18 nmol/dL, a sensitivity of 76.9% and a specificity of 80.0% were obtained. (D) Kaplan–Meier curves for PFS and OS between two groups determined by the cut-off value of D-ser at 2.18 nmol/dL based on (C). (E and F) Representative image of immunohistochemical analysis of tumour from patients with GC. D-serhi group (n = 10) and D-serlo group (n = 9) was divided by (C and D) (DAPI: Cyan, CD8: Red, CD68: Green). (F) The number of CD8-positive and CD68-positive cells within the tumour, as well as the CD8/CD68 cell ratio. Cell counting was performed in five randomly selected fields with cancer cells per tissue section at 200 × magnification and obtained the data with mean of their counts in each patient. All data are presented as the mean ± SEM. p values were determined by two-sided unpaired Student's t-test (A, B, F), log-rank test (D) and gehan-breslow-wilcoxon test (D, left panel). Elevated plasma D-ser is associated with an immunosuppressive tumour microenvironment and poor response to anti–PD-1 therapy in gastric cancer. (A) Plasma concentrations of D-Ser, D-Asn, D-Ala, and D-Pro in HC and patients with GC in cohort 3 (immediately prior to nivolumab administration) (B) Plasma concentrations of D-Ser, D-Asn, D-Ala, and D-Pro in cohort 3 (immediately prior to nivolumab administration), divided by the best clinical response to nivolumab monotherapy into the group of progressive disease (PD, n = 13) and the group of non-progressive disease (non PD, n = 15). Non-PD consists of one complete response (CR), two partial response (PR), 7 stable disease (SD), and 5 non-CR/non-PD. (C) ROC curve to define patients with PD and non-PD using plasma D-ser. Area under curve (AUC), Corrected Akaike Information Criterion (AICc), and Bayesian Information Criterion (BIC) were calculated. When the cut-off value for D-ser was set at 2.18 nmol/dL, a sensitivity of 76.9% and a specificity of 80.0% were obtained. (D) Kaplan–Meier curves for PFS and OS between two groups determined by the cut-off value of D-ser at 2.18 nmol/dL based on (C). (E and F) Representative image of immunohistochemical analysis of tumour from patients with GC. D-serhi group (n = 10) and D-serlo group (n = 9) was divided by (C and D) (DAPI: Cyan, CD8: Red, CD68: Green). (F) The number of CD8-positive and CD68-positive cells within the tumour, as well as the CD8/CD68 cell ratio. Cell counting was performed in five randomly selected fields with cancer cells per tissue section at 200 × magnification and obtained the data with mean of their counts in each patient. All data are presented as the mean ± SEM. p values were determined by two-sided unpaired Student's t-test (A, B, F), log-rank test (D) and gehan-breslow-wilcoxon test (D, left panel). Although late-phase PFS became similarly poor in both groups, the patients in D-serhi group showed earlier disease progression after anti–PD-1 therapy. Because primary resistance to immune checkpoint blockade is typically reflected by early progression events, we additionally applied the generalised Wilcoxon test, which places greater weight on early events, and observed a significant difference between the D-serhi and D-serlo groups (Fig. 8D). We additionally performed multivariate Cox regression analysis incorporating performance status (PS) and other clinical covariates (Supplementary Fig. S8F). In this analysis, PS remained independently associated with prognosis, whereas plasma D-ser did not retain statistical significance. We speculate that this result may reflect the limited sample size and the close association between elevated plasma D-ser levels and overall disease severity in patients with advanced GC. Despite the lack of independent significance in multivariate analysis, we then address the question whether plasma concentration of D-ser associated with tumour immune microenvironment. Immunohistochemical analysis of CD8+ T cells and macrophages in pretreatment biopsy samples from patients with GC revealed that tumours from patients in D-serhi group exhibited increased macrophage infiltration together with reduced CD8+ T cell infiltration compared with tumours from patients in D-serlo group (Fig. 8E and F). Consistent with these findings, the intratumoural CD8+ T cells/macrophage ratio was significantly lower in the D-serhi group. These findings indicate that elevated circulating D-ser levels reflect a tumour microenvironment with increased macrophages and reduced CD8+ T cell infiltration, consistent with our functional studies in mice. supporting its value as a potential predictive biomarker for anti–PD-1 antibody resistance. Together, these results identify D-ser as a metabolic immune checkpoint that promotes tumour progression by macrophage-mediated suppression of cytotoxic CD8+ T cells and associated with poor response to immune checkpoint blockade in GC. Discussion The metabolic regulation of antitumour immunity has emerged as a central determinant of cancer progression and therapeutic response; however, the contribution of D–AAs to this process remains largely unexplored. In this study, we identified D-ser as a previously unrecognised metabolic regulator of antitumour immunity in cancer. By integrating functional analyses in murine tumour models with clinical observations across multiple human cohorts, we demonstrated that D-ser functions as an immunosuppressive metabolite that promotes tumour progression by disabling CD8+ T cell responses and shifting the macrophage population toward pro-tumoural states. Notably, these effects occur independently of microbiome alterations, supporting a microbiota-independent immunomodulatory role for D-ser within the tumour microenvironment. Our findings expand the concept of metabolic immune checkpoints beyond the established l–amino acid pathways. Prior studies have shown that glutamine metabolism regulates T cell proliferation and exhaustion,9,65,66 arginine availability governs effector T cell fitness,67, 68, 69 and tryptophan catabolism through IDO/TDO generates immunosuppressive kynurenines.70, 71, 72, 73 In contrast, D-ser represents a distinct and previously unappreciated immune regulation axis. The coordinated suppression of cytotoxic CD8+ T cells and enrichment of immunosuppressive TAMs observed here suggests that D-ser imposes a multilayered blockade on antitumour immunity.56,74, 75, 76, 77, 78, 79, 80, 81, 82, 83 This mode of immune suppression is conceptually reminiscent of known metabolic checkpoints, yet functionally distinct, positioning D-ser as a component of metabolic immune evasion. Cell–cell communication analysis revealed a complex regulatory circuit between cytotoxic CD8+ T cells and specialised macrophage subsets within the tumour microenvironment. These interactions appear to reflect a dynamic balance between inflammatory recruitment signals and immunosuppressive feedback. In D-ser mice, the “incoming” FN1 and SPP1 signalling pathways were markedly enhanced in CD8+ T cells, indicating that these cells are frequent recipients of signals from the myeloid compartment. Notably, Fn1+ and Spp1+ TAMs are well characterised as immunosuppressive subsets that inhibit CD8+ T cell–mediated anti-tumour responses.16,57,60,84, 85, 86, 87 Importantly, functional blockade of SPP1 partially restored CD8+ T cell proliferation in co-culture experiments, supporting SPP1 signalling as a functional mediator of D-ser–induced macrophage immunosuppressive activity. Conversely, analysis of “outgoing” signalling demonstrated how CD8+ T cells influence macrophage populations, primarily through costimulatory and antigen-presentation–related pathways. In D-ser mice, MHC-II and CD86 signalling interactions were modestly reduced compared to controls,88, 89, 90 suggesting the attenuation of a feedback loop in which CD8+ T cells promote macrophage maturation or antigen-presenting polarisation. Collectively, these findings suggest that D-ser impairs anti-tumour CD8+ T cell immunity through at least two complementary functions: suppression of inflammatory macrophage–T cell communication required for CD8+ T cell recruitment and maintenance, and enhancement of SPP1-associated immunosuppressive macrophage signalling. This disruption may favour the expansion and maintenance of immunosuppressive TAM subsets in the TME. Such a bidirectional circuit, where dysfunctional CD8+ T cells permit the accumulation of pro-tumoural macrophages, which in turn further suppress CD8+ T cell function, could amplify immune evasion and reinforce tumour progression. Clinically, elevated circulating D-ser levels were associated with poorer outcomes following anti–PD-1 therapy in patients with GC. Although plasma D-ser did not remain an independent predictor in multivariate analyses incorporating performance status and other clinical covariates, patients with high D-ser levels exhibited earlier disease progression and a TME characterised by increased macrophage infiltration and reduced CD8+ T cell accumulation. These findings are consistent with our experimental data demonstrating that D-ser promotes macrophage-mediated immune suppression and impairs CD8+ T cell responses. Taken together, our results suggest that plasma D-ser reflects an immunosuppressive tumour state. Our study was small cohort, so larger prospective studies will be required to determine its independent predictive value and clinical utility. This study has several limitations. First, the cellular and enzymatic sources of elevated D-ser levels in patients with GC remain undefined. The concentration of circulating L- and D-AAs may be influenced by multiple factors, including tumour cells, immune cells, the microbiota, and dietary intake.19,20,91, 92, 93, 94 Our findings suggest that elevated D-ser levels in the patients with GC are unlikely to originate primarily from tumour cells. We demonstrated the altered microbiota derived from D-ser administration did not directly enhance the tumour burden, however, our result did not support the idea that the both microbiota-derived production and dietary sources may contribute substantially to systemic D-ser levels, resulting to enhance tumour burden. Given the complex interplay among host metabolism, microbial activity, and nutrient availability, elucidating the relative contribution of each source will be essential for understanding the origin, regulation, and biological significance of D-ser in cancer. Second, our murine models were based on transplanted tumours and short-term D-ser supplementation. These conditions may not fully recapitulate the chronic metabolic and inflammatory environments present in patients, including long-standing atrophic gastritis or Helicobacter-associated diseases. Thus, the long-term effects of sustained D-ser elevation in more physiologically relevant models require further investigation. Third, although several L-AAs influence immune responses through defined transporters and receptors during cancer progression,9,65,66,95 we did not detect the expression of canonical D-ser–related receptors, such as NMDA receptors, on tumour-infiltrating immune cells in D-ser mice. This observation raises the possibility that D-ser may enter immune cells, especially macrophages via alternative amino acid transport systems or passive transport mechanisms, or that its effects are mediated indirectly through metabolic or paracrine pathways. Clarifying the molecular function of D-ser uptake and signalling will be an important direction for future studies. Finally, whether D-ser exerts similar immunosuppressive effects in other malignancies is unknown. Addressing these questions is critical for translating our findings into therapeutic strategies targeting the D-ser axis. In summary, we uncovered D-ser as a metabolic immune checkpoint that constrains macrophages and CD8+ T cell-mediated antitumour immunity and contributes to immunotherapy resistance in GC. These findings broaden the landscape of amino acid–mediated immune regulation and provide a rationale for incorporating D-ser-targeted strategies into future cancer immunotherapy paradigms. Contributors Conceptualisation, Tomohisa Sujino; methodology, Kai Tsugaru, Shohei Suzuki, Kentaro Miyamoto and Tomohisa Sujino; investigation, Kai Tsugaru, Shohei Suzuki, Kentaro Miyamoto, Kazuhiro Murakami, Akihiko Chida, Chihiro Kato, Naoto Fukasawa, Arina Shigehara, Mai Hasegawa, Yusuke Yoshimatsu, Jumpei Sasabe, Masashi Mita and Yosuke Harada; visualisation, Shohei Suzuki, Kentaro Miyamoto, Kai Tsugaru; resources, Shohei Suzuki, Kazuhiro Murakami, Nick Barker, Yasutaka Sukawa, Toshiaki Teratani, Nobuhiro Nakamoto, Hirofumi Kawakubo, Kaoru Takabayashi, Motohiko Kato, Naohisa Yahagi, Taisuke Kondo, Eri Arai, Yuko Kitagawa, Kenro Hirata, Yasuo Hamamoto, and Takanori Kanai; writing-original draft, Tomohisa Sujino, writing-review & editing; Tomohisa Sujino, Kai Tsugaru, Shohei Suzuki, Kentaro Miyamoto; supervision, Tomohisa Sujino, funding acquisition, Tomohisa Sujino, Kai Tsugaru, Takanori Kanai. All authors read and approved the final version of the manuscript. All authors agreed on their roles in this manuscript. Kai Tsugaru, Shohei Suzuki, Kentaro Miyamoto and Tomohisa Sujino have accessed and verified the underlying data in mouse experiment. Kai Tsugaru, Chihiro Kato, and Tomohisa Sujino have accessed and verified the underlying data in human cohort. Data sharing statement 16S rRNA gene sequencing data were deposited in DDBJ under accession number PRJDB20363. The scRNA-seq data were deposited in the DDBJ PRJDB20370. Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Tomohisa Sujino (tsujino1224@keio.jp). Declaration of interests Kentaro Miyaamoto is an employee of Miyarisan Pharmaceutical Co., Ltd. Masashi Mita is the founder of KAGAMI Inc. Masashi Mita, Kai Tsugaru, Takanori Kanai, Tomohisa Sujino submitted a patent regarding for the d-amino acids in cancer. The rest of other authors declare no potential conflicts of interest.