Targeting AQP5-mediated arginine deprivation in gastric cancer stem cells restores NK cell anti-tumor immunity

Summary

Natural killer (NK) cells exhibit impaired anti-tumor activity upon entering the tumor microenvironment (TME); however, the precise mechanism(s) remains elusive. In this study, we demonstrate that AQP5⁺ gastric cancer stem cells contribute to the dysfunction of NK cells by reprogramming the urea cycle (UC). Mechanistically, AQP5 competitively binds ATP-dependent RNA helicase A (DHX9) over karyopherin subunit beta 1 (KPNB1), inhibiting DHX9 nuclear translocation and transcriptionally down-regulating argininosuccinate synthase 1 (ASS1). Low-arginine condition in the TME reshaped by AQP5⁺ tumor cells weakens NK cell function by limiting NO synthesis. Notably, preclinical murine models confirm that oral arginine supplements improve the NK cell-directed killing against organoids generated by AQP5^High GC (gastric cancer) tissues. Besides, AQP5⁺ tumor cells also redirect the UC to the TCA cycle, which stores the saved nitrogen in glutamine by promoting glutamate-ammonia ligase (GLUL) stability. This study uncovers the evidence of AQP5⁺ cancer stem cells impairing NK cell cytotoxicity by changing self-metabolism patterns.

Graphical abstract

Highlights

• •AQP5⁺ GCSCs establish arginine-deficient TME, leading to impaired NK cell function
• •AQP5 blocks DHX9 nuclear translocation, suppressing ASS1 to deplete arginine
• •AQP5-TRIM21 stabilizes GLUL, accruing glutamine to fuel tumor bioenergetics
• •ULK1 inhibition combined with L-arginine is an effective therapy for AQP5⁺ gastric cancer

Zhao et al. reveal that AQP5⁺ gastric cancer stem cells impair NK cell via creating low-arginine microenvironment. AQP5 blocks DHX9 nuclear translocation to suppress ASS1 while promoting GLUL stabilization, diverting nitrogen flux to cancer bioenergetics. Combining ULK1 inhibition with L-arginine synergistically suppresses AQP5⁺ tumor progression, offering a therapeutic strategy.

Introduction

Cancer stem cells (CSCs) with long-term tumorigenic capacity play a pivotal role in tumor progression, recurrence, metastasis, and drug resistance, making them important therapeutic targets.¹ Like normal stem cells, CSCs also require a stem cell niche, which exhibits characteristics typical of the tumor microenvironment (TME), to promote cancer progression.² The TME is composed of diverse cell populations and the extracellular matrix. CSCs crosstalk with immune cells to create a favorable immunosuppressive TME by secreting various factors, such as interleukin-33 and macrophage migration inhibitory factor.^3,4,5 CSCs also upregulate membrane proteins such as SIRPγ (signal regulatory protein γ) to activate Hippo/YAP (Yes-associated protein) signaling, sustaining CD47 expression to transmit immune escape signals.⁶ Aquaporin-5 (AQP5) has been recently identified as a specific surface marker for malignant stem cells in the stomach.^7,8 We previously reported that an AQP5^hi population of gastric cancer cells displayed CSC properties (i.e., are GC-CSCs).⁹ Moreover, studies have suggested that increased AQP5 expression by tumor cells is closely related to the immune status of the TME.¹⁰ However, it remains unclear whether AQP5⁺ GC-CSCs affect immune cells in the TME to escape immune attack.

Natural killer (NK) cells are cytotoxic lymphocytes of the innate immune system and play crucial roles in immuno-surveillance of cancers, recently emerging as a promising avenue for cancer immunotherapy.¹¹ However, various studies have reported that NK cells rapidly lose their anti-tumor functionality once recruited to the TME, which is a pivotal barrier to extending their therapeutic potential to solid tumors.¹² A pan-cancer single-cell panorama analysis indicates that human NK cells enriched in tumors show impaired anti-tumor activity, which correlates with unfavorable prognosis and resistance to immunotherapy.¹³ Poznanski et al. uncover that human NK cell dysfunction is due to suppression of glucose metabolism caused directly by the TME via lipid peroxidation-associated oxidative stress.¹⁴ Nutrient deprivation in the TME caused by poor vascular exchange or competition from rapidly proliferating cancer cells drives immune cells toward immunosuppressive or tolerogenic phenotypes.¹⁵ Recently, He et al. reveal that pancreatic ductal adenocarcinoma cells can suppress NK cell cytotoxicity by actively consuming vitamin B6 in the TME.¹⁶ Strategies that suppress or alter cancer metabolism to improve TME nutrient availability or that modulate immune metabolism now offer promising opportunities for cancer therapies.

In this study, we comprehensively investigated the effects of AQP5⁺ GC-CSCs on the immune microenvironment and revealed that anti-tumor activities of NK cells were significantly impaired. Then, we identified AQP5 as a molecular determinant that favors CSCs’ escape from NK cells’ attack by competitively binding to DHX9 over KPNB1 and promoting glutamate-ammonia ligase (GLUL) stability. Collectively, this study demonstrated that AQP5⁺ gastric CSCs reduce arginine production in the urea cycle (UC) to impair NK cell function while redirecting the conserved nitrogen to fuel their own tricarboxylic acid (TCA) cycle. This finding reveals a promising strategy to enhance NK cell-mediated anti-tumor effects in GC therapy.

Results

AQP5⁺ gastric CSCs impair NK cell activity to escape NK cell-mediated anti-tumor killing

To explore whether AQP5⁺ GC-CSCs affect immune cells in the TME, we confirmed the credibility of AQP5 as a potential biomarker for gastric CSCs. Consistent with others’ and our previous reports, AQP5⁺ GC cells exhibit higher expression of classic CSC surface markers, and almost all AQP5⁺ GC cells express Lgr5 and Cd44 (Figures S1A–S1C) and have higher ability of spheroid colony (Figures S1D–E) and tumorigenesis (Figure S1F), which reveals that AQP5⁺ GC cells are a group of GC-CSCs.

Next, to elucidate the contribution of the immune microenvironment to AQP5⁺ GC-CSCs-driven tumor progression, we compared tumor growth and metastatic differences of AQP5⁺ vs. AQP5⁻ MFC (mouse forestomach carcinoma cells) cells between immunocompetent C57BL/6J mice and immunodeficient NSG mice (NOD-scid IL2Rγ^null, Figure S2). The tumorigenicity and metastatic ability of AQP5⁺ MFC cells were significantly higher than AQP5^– MFC cells within the two mouse models (Figures S2A–S2C, S2E, and S2F). Notably, the variance multiplier for tumor volume (AQP5⁺ vs. AQP5⁻) in C57BL/6J mice was significantly higher than that in NSG mice (Figure S2D). A similar trend was observed in pulmonary metastatic nodules (Figure S2G), suggesting that the immune microenvironment likely provides a more favorable niche for AQP5⁺ gastric cancer progression.

Then, we performed single-cell RNA sequencing (scRNA-seq) on subcutaneous tumors constructed by AQP5⁺ and AQP5^- MFC cells in C57BL/6J mice to further analyze how AQP5⁺ GC-CSCs reshape the favorable immune microenvironment, and AQP5 expression in epithelial cells of subcutaneous tumor was validated (Figures S3A and S3B). We found that the proportions of NK cells, B cells, and macrophage clusters differed markedly between AQP5⁺ and AQP5^– subcutaneous tumors (Figure 1A). Accordingly, we performed flow cytometric analysis to quantify the proportions of these three cell populations in subcutaneous tumors. The results demonstrated that only NK cells exhibited a statistically significant reduction in AQP5⁺ tumors (Figures 1B, S3C, and S3D), while excluding potential confounding effects of the TME on NK cell apoptosis (Figure S3E). Subsequently, we performed cell sorting of T cells, NK cells, B cells, and macrophages from subcutaneous tumors to evaluate whether AQP5 expression affected their proliferative capacity and cytotoxic functions. We observed no significant differences in proliferative activity or cytotoxic function among tumor-infiltrating B cells, T lymphocytes, and macrophages when comparing between AQP5⁺ and AQP5⁻ subcutaneous tumors (Figures S3F–S3J). Notably, NK cells infiltrating AQP5⁺ subcutaneous tumors displayed statistically significant impairment in both proliferation and cytotoxic function compared to AQP5⁻ tumors (Figures 1C and 1D). To evaluate whether AQP5 expression affected NK cell maturation, we stratified tumor-infiltrating NK cells into four distinct maturation subsets based on scRNA-seq data. Quantitative analysis showed a pan-subset reduction in AQP5⁺ tumors, suggesting AQP5 affected NK cell abundance rather than maturation status (Figures S3K–S3L). Gene set variation analysis revealed that, compared with NK cells in AQP5^– tumors, biological processes and pathways in NK cells infiltrated in the AQP5⁺ tumors, including killing effect-related pathways such as tumor necrosis factor (TNF)-⍺ signaling and interferon (IFN)-⍺ response, were significantly inhibited (Figure S3M). scRNA-seq analysis further demonstrated that AQP5⁺ subcutaneous tumors might induce NK cell dysfunction, characterized by significant downregulation of cytotoxic effector molecules (granzyme A-Gzma, granzyme C-Gzmc, and natural killer cell granule protein 7-Nkg7), tumor-homing capacity (Xcl1 and Sell), and impaired inflammatory signaling (Tnf and Il10ra) compared to AQP5⁻ tumors (Figure 1E). Subsequently, we performed in vitro validation of AQP5⁺ GC cells’ impact on NK cell proliferation and cytotoxicity. Primary NK cells isolated from the peripheral blood of healthy donors exhibited markedly attenuated proliferation and cytotoxic activity and decreased effector molecule expression (IFN-γ, TNF-α, and GZMA) when co-cultured with AQP5⁺ GC cells compared to AQP5⁻ GC cells (Figures 1F–1H). These results revealed that AQP5⁺ TME might induce more profound NK cell dysfunction.

Figure 1.

AQP5⁺ GC cells create a favorable immune microenvironment

(A) Seven clusters with distinct transcriptional profiles in subcutaneous tumors established with 5 × 10⁶ AQP5⁺ or AQP5⁻ MFC cells in C57BL/6J mice visualized using tSNE (t-distributed stochastic neighbor embedding) plots.

(B) Flow cytometry analysis of the proportion of NK (CD3⁻NK1.1⁺) cells in AQP5⁺ or AQP5⁻ tumors (subcutaneous tumors in C57/B6J mice, AQP5⁺ group n = 6, AQP5⁻ group n = 4).

(C and D) NK cells isolated from AQP5⁺ or AQP5⁻ tumors (subcutaneous tumors in C57/B6J mice, AQP5⁺ group n = 6, AQP5⁻ group n = 4); proliferation of NK cells was assessed by carboxyfluorescein succinimidyl ester (CFSE) assay (C); cytotoxic activity of NK cells was measured by LDH (lactate dehydrogenase) release assay (D).

(E) Expression of NK cell-related genes in subcutaneous tumors from (A).

(F–H) NK cells separated from peripheral blood of healthy volunteers and co-cultured with AQP5⁺ or AQP5⁻ AGS (adenocarcinoma gastric stomach) cells in transwell chamber for 48 h; then, proliferation detected by CCK8 (F), relative expression of IFN-γ, TNF-α, and granzyme A (Gzma) (G), and cytotoxicity (H) in NK cells were performed (n = 3).

(I) Flow cytometry analysis of the proportion of NK (CD3⁻CD56⁺) cells in gastric cancer tissues with high or low AQP5 expression (n = 7/group).

(J) Representative images of immunofluorescence staining of AQP5, EPCAM, CD3, and CD56 in gastric cancer tissue microarrays (n = 95). Scale bar: 500 and 200 μm.

(K) Proportion of cells with high and low NK cell infiltration grouped by the median in each group.

(L–N) Isolation and culture of NK cells derived from gastric cancer tissues of patients with high or low AQP5 expression. Relative IFN-γ and TNF-α expression (L) and cytotoxicity (M) in NK cells isolated from GC tissues with high or low AQP5 expression (n = 3). NK cells (green) infiltrating PDOs (red) were evaluated by microscopy after 10⁵ NK cells were co-cultured with PDOs in matrix gel for 48 h; the NK cells per organoid were calculated (n = 5) (N). Scale bar: 100 μm.

(O) PBS or 10⁵ NK cells separated from the spleen of C57BL/6J mice and injected intravenously into NSG mice to reestablish anti-tumor immunity the day before injecting MFC cells. After the 5 × 10⁶ AQP5⁺ or AQP5⁻ MFC cell subcutaneous injection, NK cells or PBS were intravenously injected on days 7, 14, and 21. Images and volume of subcutaneous tumors (n = 6/group).

(P) NSG mice were treated with PBS or 10⁵ NK cells as in (O); representative images and statistical analysis of metastatic nodules in the lungs established with 5 × 10⁶ AQP5⁺ or AQP5⁻ MFC cells (n = 3/group). Scale bar: 2 mm.

p values in (B)–(D), (G)–(I), (L)–(N), and (P) were calculated using a two-tailed Student’s t test; those in (F) and (O) were calculated using two-way ANOVA. Data in (C)–(H) and (L)–(P) are represented as mean ± SD; violin plot shows all points in (B) and (I).

To elucidate AQP5’s contribution to NK cell dysfunction in gastric cancer, we categorized primary tumor specimens into AQP5-high and AQP5-low-expression groups based on quantitative immunohistochemical (IHC) staining intensity (Figure S4). Consistent with the observations in subcutaneous tumors, flow cytometry analysis also revealed that only NK cell frequency was significantly decreased in AQP5-high gastric cancer tissues (Figures 1I and S3N–S3O). To further evaluate the impact of AQP5 expression on infiltrating NK cell numbers in gastric cancer tissues, we performed multi-color immunofluorescence analysis on 95 gastric cancer tissue microarrays (Table S2; Figure 1J). The median proportions of AQP5/EPCAM-positive or CD3-negative/CD56-positive (NK cell marker) cells were defined as cutoff values. Gastric cancer tissues with AQP5/EPCAM-positive rates above the median were classified as the AQP5-high group, and those with CD3⁻CD56⁺ rates above the median were classified as the NK cell-high group, while those below the median were designated as the low-expression groups. Moreover, AQP5 positive cell rates were significantly higher in the AQP5-high group than in the AQP5-low group (Figure S5A). Across the cohort, NK cell-high (CD3⁻CD56⁺) tissues were observed in 51% of patients, whereas the remaining 49% displayed NK cell-low infiltration (Figure S5B). However, only 35% of the tissues in the AQP5-high group exhibited high NK cell infiltration, a proportion significantly lower than the 64% observed in the AQP5-low group (Figure 1K). These results suggest that high AQP5 expression in gastric cancer tissues reduces NK cell infiltration. Additionally, we found that NK cells isolated from AQP5-high gastric cancer tissues exhibited significantly reduced cytotoxicity and lower levels of effector molecules (IFN-γ and TNF-α), as well as a markedly impaired ability to attack gastric cancer organoids (Figures 1L–1N).

To further validate the effects of AQP5⁺ GC cells on NK cell-mediated anti-tumor activity, we established a xenograft model and a tail-vein injection-induced lung metastasis model in NSG mice. Adoptive transfer of 10⁵ NK cells significantly suppressed the growth and metastasis of AQP5^– tumors but had minimal impact on AQP5⁺ tumors (Figures 1O–1P). These findings further supported the notion that NK cell function was impaired in the context of AQP5⁺ tumors. Collectively, our results suggested that AQP5⁺ GC cells might foster an immunosuppressive TME that compromised the cytotoxic efficacy of NK cells.

AQP5⁺ GC cells create a low-arginine microenvironment to weaken the anti-tumor capacity of NK cells

Cell culture media are composed of diverse components, including proteins, nucleic acids, and metabolites. To pinpoint the specific factor responsible for NK cell dysfunction, we treated NK cells with conditioned media from AQP5⁺ GC cells that had been enzymatically digested with DNase I, RNase, or Proteinase K. Intriguingly, depletion of DNA, RNA, or proteins from the supernatant had no significant effect on NK cell proliferation, suggesting that the suppressive factor may belong to another class of molecules (Figures S6A–S6C).

Metabolites play an important role in regulating cell activity and biological behavior.¹⁷ To investigate the metabolic reprogramming associated with AQP5 expression in gastric cancer, we performed untargeted metabolomic profiling of AQP5⁺ and AQP5^– GC cells, revealing distinct metabolic signatures between the two groups (Table S3). Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis highlighted profound alterations in AQP5⁺ GC cells across several key metabolic pathways, including arginine biosynthesis, glutamate metabolism, and central carbon metabolism in AQP5⁺ GC cells (Figure 2A). To validate these metabolic findings, we quantitatively analyzed amino acid levels using targeted liquid chromatography-tandem mass spectrometry (LC-MS/MS). Consistent with the untargeted metabolomics results, AQP5⁺ GC cells exhibited significantly reduced levels of UC metabolites (arginine and ornithine), while demonstrating elevated concentrations of glutamate and glutamine (Figures 2B and S7; Table S4). Hence, the UC might be impaired in AQP5⁺ GC cells (Figure 2C). Considering that the TME reshaped by AQP5⁺ GC cells significantly impaired NK cell-mediated anti-tumor ability, these amino acids were measured in AQP5⁺ and AQP5^– GC cell media by LC-MS/MS. Interestingly, only arginine levels were decreased in AQP5⁺ GC cell media, while ornithine, glutamate, and glutamine showed no significant changes compared to AQP5^– GC cell media (Figure 2D; Table S5). Importantly, arginine levels also decreased significantly with increasing AQP5 expression levels—classified into four grades by IHC—in primary GC specimens (Figures 2E and S8; Table S6). Cumulatively, these results indicated that AQP5 expression in GC cells might create a low-arginine microenvironment (TME).

Figure 2.

Explore how AQP5⁺ GC cells blunt the anti-tumor activity of NK cells by decreasing the arginine production

(A) KEGG enrichment analysis of metabolic pathways for differential metabolites in AQP5⁺ and AQP5⁻ AGS cells.

(B) L-glutamate, L-glutamine, L-arginine, and L-ornithine levels in AQP5⁺ and AQP5⁻ AGS cells detected by targeted metabolic mass spectrometry (n = 5).

(C) Schematic diagram of amino acid metabolism; blue: decreased in AQP5⁺ AGS cells; red: increased in AQP5⁺ AGS cells.

(D) L-arginine, L-ornithine, L-glutamate, and L-glutamine levels in the medium of AQP5⁺ and AQP5⁻ gastric cancer cells detected by targeted metabolic mass spectrometry (n = 3).

(E) Targeted metabolic mass spectrometry detection of L-arginine in GC tissues with different AQP5 expressions (n = 22).

(F–H) Images (F), tumor volume (G), and tumor incidence (H) in subcutaneous tumors established with 5 × 10⁶ AQP5⁺ or AQP5⁻ MFC cells in C57BL/6J mice treated with daily oral gavage of different concentrations of L-arginine or PBS (n = 6/group).

(I and J) Lung metastatic models established with AQP5⁺ and AQP5⁻ MFC cells (GFP-labeled) in C57BL/6J mice by tail vein injection; mice received daily oral gavage of different concentrations of L-arginine or PBS. Images (I) and statistical analysis (J) of metastatic nodules in the lungs (n = 3/group). Scale bar: 2 mm.

(K) Proportion of NK cells in subcutaneous tumors corresponding to (F) detected by flow cytometry (AQP5⁺, n = 6); AQP5⁺+L-arginine (1 g/kg daily oral gavage), n = 6; AQP5⁺ + L-arginine (2 g/kg daily oral gavage), n = 4; AQP5⁻, n = 4).

(L and M) Subcutaneous tumors established with 5 × 10⁶ AQP5⁺ MFC cells in C57BL/6J mice treated with anti-NK1.1 antibody (200 μg per mouse, twice weekly) or isotype control, followed by daily oral administration of L-arginine (2 g/kg) or PBS. (L) NK cell depletion efficiency was validated by flow cytometry analysis of peripheral blood (n = 12/group). (M) Representative images and tumor volume measurements of subcutaneous tumors are shown (n = 6/group).

(N and O) NK cells co-cultured with AQP5⁺ AGS cells, AQP5⁻ AGS cells, or AQP5⁺ AGS cells supplemented with 40 nmol/mL L-arginine for 48 h and co-cultured with AGS cells. Images detected by live cell workstation (N); NK cells were labeled green by CFSE, and AGS cells were labeled red; statistical analysis for the mortality rates of tumor cells (O) (n = 3). Scale bar: 20 μm.

(P–S) NK cells separated from the peripheral blood of healthy volunteers co-cultured with AQP5⁺ AGS cells, AQP5⁻ AGS cells, and AQP5⁺ AGS cells supplemented with 40 nmol/mL L-arginine in transwell chamber for 48 h. The L-arginine levels detected by ELISA (P), proliferative capacity detected by CFSE (Q), cytotoxicity (R), and relative expression of IFN-γ, TNF-α, and Gzma (S) in NK cells (n = 3).

Statistical analyses, p values in (B), (D), (E), (J), (K), (L), and (P)–(S) were calculated using a two-tailed Student’s t test; in (G), (O), and (M) by two-way ANOVA test; and in (H) by log rank test. Data in (B), (D), (G), (J), (M), and (O)–(S) are represented as mean ± SD; violin plot shows all points in (K) and (L).

Given the essential role of arginine in NK cell function,¹⁸ we next examined whether L-arginine supplementation could rescue the NK cell dysfunction induced by AQP5⁺ GC cells in vivo. Following daily oral gavage of different L-arginine doses (1–2 g/kg), we found that L-arginine (2 g/kg) significantly suppressed the occurrence, progression, and metastasis of AQP5⁺ MFC tumors (Figures 2F–2J). Notably, this treatment regimen markedly enhanced the tumor infiltration of NK cells within AQP5⁺ tumors (Figure 2K). And, L-arginine (40 nM) did not inhibit or promote the proliferation or clonal expansion of AQP5⁺ GC cells in vitro (Figures S9A and S9B). Importantly, NK cell elimination using an anti-NK1.1 monoclonal antibody (200 μg per mouse, twice weekly) completely reversed the tumor-inhibitory effects of arginine replenishment (2 g/kg), establishing the essential role of NK cells in mediating arginine therapeutic efficacy (Figures 2L and 2M). Thus, low-L-arginine-condition-induced impairment of NK cell anti-tumor immunity might be a pivotal factor driving the increased progression of AQP5⁺ tumors compared to AQP5^- tumors within the TME.

To further validate these findings, we confirmed whether in vitro L-arginine supplementation (40 nM) could restore the impaired NK cell function caused by AQP5⁺ GC cells. NK cells co-cultured with AQP5⁺ GC cells in transwell assays exhibited significantly reduced tumor-killing capacity compared to those co-cultured with AQP5⁻ GC cells, while L-arginine supplementation effectively rescued their anti-tumor activity (Figures 2N and 2O). Moreover, we found that NK cells co-cultured with AQP5⁺ GC cells in transwell chambers exhibited significantly lower intracellular arginine levels than those co-cultured with AQP5⁻ GC cells. Exogenous L-arginine supplementation effectively restored the depleted arginine levels caused by AQP5⁺ GC cells (Figures 2P and S10A; results shown for both AGS cells and HGC-27 cells). Notably, supplementation with exogenous L-arginine substantially ameliorated the AQP5⁺ GC cell-mediated impairments of NK cell proliferation, cytotoxicity, and effector molecule expression (Figures 2Q–2S and S10B–S10D; results shown for both AGS cells and HGC-27 cells). Furthermore, we observed no detectable changes in NK cell apoptosis following either co-culture with AQP5⁺ GC cells or exogenous L-arginine supplementation (Figure S10E). Collectively, these data provided compelling evidence that L-arginine deficiency in the TME induced by AQP5⁺ GC cells served as a pivotal mechanism underlying impaired NK cell anti-tumor activity.

AQP5 prevents DHX9 nuclear translocation to reduce arginine production

To investigate how AQP5⁺ GC cells established a low-L-arginine TME, we performed proteomic profiling of key UC enzymes (Figure 3A; Table S7). Strikingly, argininosuccinate synthase 1 (ASS1) expression was reduced in both GC cells with AQP5 overexpression (Figure 3B) and AQP5⁺ GC cells (Figures 3C and S11A), whereas ARG (Arginase) levels in AQP5⁺ GC cells were unaffected (Figures 3C and S11B). Moreover, ASS1 overexpression in AQP5⁺ GC cells not only significantly reversed the L-arginine reduction in the culture medium (Figure S11C) but also restored the proliferative and cytotoxic activity of NK cells co-cultured in transwell assays (Figures 3D and S11D; results shown for both AGS cells and HGC-27 cells). Notably, in vivo ASS1 overexpression not only remarkably inhibited the subcutaneous tumor progression of AQP5⁺ GC cells but also enhanced NK cell recruitment and restored L-arginine levels within the TME (Figures 3E–3G and S11E–S11G). Consistently, ASS1 depletion in AQP5⁻ GC cells promoted tumor progression in vivo, accompanied by decreased intratumoral L-arginine levels and diminished NK cell recruitment (Figures S12A–S12F). Collectively, these results further supported the critical role of ASS1-dependent arginine synthesis in AQP5⁺ GC cells in regulating NK cell anti-tumor activity. Validation of clinical specimens further demonstrated that ASS1 expression was significantly downregulated in AQP5^High GC tissues compared to AQP5^Low counterparts (Figure 3H; Table S2; same cohort as Figure 1J). Moreover, ASS1 expression showed a strong negative correlation with the quantity of AQP5⁺EPCAM⁺ tumor cells across GC specimens (Figure S12G). Cumulatively, these data suggested that AQP5 reduced arginine production by inhibiting ASS1 expression.

Figure 3.

Analysis of the mechanism of AQP5 down-regulating ASS1 expression in GC

(A) Schematic diagram of the urea cycle.

(B) Proteomics analysis of the expression of the key enzyme in the UC (ASS1, CPS1, ASL, and OTC) in AQP5-OE and Ctrl (control group) AGS cells (n = 1).

(C) Expression of ASS1 and ARG in AQP5⁺ and AQP5⁻ AGS cells detected by western blotting.

(D) NK cells treated with supernatant collected from AQP5⁺ AGS cells, AQP5⁻ AGS cells, and ASS1-OE-AQP5⁺ AGS cells. Proliferative capacity (left) and cytotoxicity (right) in different groups of NK cells were evaluated (n = 3).

(E–G) Images (E), tumor volume (F), and proportion of infiltrating NK cells (G) in subcutaneous tumors established with 5 × 10⁶ Ctrl AQP5⁺ or ASS1-OE AQP5⁺ MFC cells in C57BL/6 mice (n = 6/group).

(H) Immunohistochemical analysis of ASS1 expression in gastric cancer tissues stratified by AQP5 expression levels (the same patient cohort as in Figure 1K). Scale bar: 500 and 200 μm.

(I) qPCR detection of ASS1 expression in AQP5 AGS-knockdown (KD) cells or AQP5-KD AGS cells transfected with siRNA targeting DDX21, DDX5, DDX3X, DDX17, DHX9, DDX46, or FUS (n = 3).

(J and K) Expression of ASS1 (J) and L-arginine levels in the medium (K) of AQP5-KD AGS cells or AQP5-KD AGS cells transfected with DHX9-KD siRNA or corresponding control cells (n = 3).

(L) Confocal microscopy analysis of the location of AQP5 (red) and DHX9 (green) in AGS cells transfected with Ctrl or AQP5-OE lentivirus (n = 1). Scale bar: 20 μm.

(M) Western blot analysis of DHX9 in the nucleus and cytoplasm of AGS cells transfected with Ctrl or AQP5-OE lentivirus (n = 1).

(N and O) AQP5-overexpressing (N) and AQP5-knockdown (O) AGS cells co-transfected with DHX9-HA and KPNB1-Myc. Cell lysates were subjected to immunoprecipitation (IP) with anti-HA agarose and immunoblotted (n = 1).

(P and Q) Gradient overexpression of AQP5 (P) or KPNB1 (Q) in 293T cells co-transfected with KPNB1-Myc or AQP5-FLAG and DHX9-HA. The cell lysates were subjected to IP with anti-HA agarose and immunoblotted (n = 1).

(R and S) Western blotting (R) and confocal microscopy (S) detection of DHX9 (green) expression in the nucleus and cytoplasm of Ctrl or AQP5-OE AGS cells transfected with Ctrl or KPNB1-OE plasmid (n = 1). Scale bar: 50 μm.

(T) Diagram summarizing the mechanism of AQP5 downregulation of L-arginine. AQP5 competitively binds DHX9 with KPNB1 to inhibit the nuclear translocation of the transcriptional regulatory protein, DHX9, and downregulate the expression of the key enzyme ASS1 and L-arginine production.

p values in (D), (G), and (K) were calculated using a two-tailed Student’s t test and in (F) with a two-way ANOVA. Data in (D), (F), (I), and (K) are represented as mean ± SD; violin plot shows all points in (G).

Considering that AQP5 transcriptionally inhibits ASS1 in GC cells, we screened transcription factors or regulatory factors mediating this effect using reverse chromatin immunoprecipitation analyses (Table S8). Seven transcriptional regulation-related proteins were selected for further validation (Figure S13A); DExHBox helicase 9 (DHX9) was identified as a key regulator of ASS1 expression, since its knockdown in GC cells significantly reversed the effects of AQP5 knockdown—namely, ASS1 upregulation (Figures 3I and 3J), elevated L-arginine production (Figure 3K), and enhanced ASS1 promoter activity (Figure S13B). Moreover, DHX9 was significantly enriched at the ASS1 promoter (Figure S13C). Notably, AQP5 did not affect DHX9 expression levels (Figure S13D). DHX9 translocates to the nucleus as a transcriptional regulator to modulate transcriptional activity.¹⁹ We speculated that AQP5 might impact the nuclear translocation of DHX9. Indeed, AQP5 overexpression induced DHX9 cytoplasmic retention (Figures 3L and 3M), while AQP5 knockdown promoted nuclear translocation (Figure S13E). Thus, AQP5 decreased ASS1 expression by inhibiting DHX9 nuclear translocation.

Most cytoplasmic proteins are transferred to the nucleus via nuclear transporter proteins.²⁰ Hence, we screened proteins that interact with DHX9 and found KPNB1 (karyopherin subunit beta 1) as the only nuclear transporter protein that binds DHX9 (Table S9). Meanwhile, AQP5 significantly inhibited the binding of DHX9 to KPNB1 (Figure S14A), which was further validated in AGS and 293T cell lines by AQP5 overexpression or knockdown (Figures 3N–3O, S14B, and S14C). In addition, the interaction between KPNB1 and DHX9 gradually decreased with increasing AQP5 expression (Figure 3P); similarly, AQP5 binding to DHX9 also gradually decreased with increasing KPNB1 expression (Figure 3Q). Importantly, DHX9 could bind to AQP5 or KPNB1 (Figures S14D and S14E). These results suggested that AQP5 competes with KPNB1 for binding to DHX9. Notably, while KPNB1 overexpression promoted DHX9 nuclear translocation in control GC cells, co-overexpression of KPNB1 in GC cells that overexpressed AQP5 obviously reversed the inhibitory effect of AQP5 on DHX9 nuclear translocation (Figures 3R and 3S). These findings demonstrated that AQP5 competes with KPNB1 for binding to DHX9, thereby inhibiting DHX9 nuclear translocation and subsequently suppressing ASS1 transcriptional expression and arginine biosynthesis (Figure 3T).

Decreased L-arginine in the TME limits NO synthesis in NK cells weakening their anti-tumor ability

L-arginine is a substrate for the production of polyamines, creatine, and proline, essential for cell survival and proliferation. We therefore assessed the metabolic alterations in NK cells following L-arginine treatment using untargeted metabolomics. However, L-arginine supplementation did not significantly increase polyamine, ornithine, creatine, or proline in NK cells (Table S10).

The synthesis of NO by NK cells is mediated by constitutively expressing the endothelial NOS (nitric oxide synthase), which supports their cytotoxic functions.²¹ We found that the NO level in NK cells co-cultured with AQP5⁺ GC cells in transwell assays was significantly decreased, while L-arginine supplementation restored NO levels (Figures 4A and S15A; results shown for both AGS and HGC-27 cells). Three types of NOS isoforms (inducible nitric oxide synthase-iNOS, endothelial nitric oxide synthase-eNOS, and neuronal nitric oxide synthase-nNOS) participate in NO synthesis. Pharmacological inhibition of iNOS significantly weakened NO synthesis, proliferation, and cytotoxicity in NK cells, regardless of AQP5 expression or exogenous L-arginine supplementation. Strikingly, iNOS blockade completely abolished the enhancement of L-arginine on NK cells co-cultured with AQP5⁺ GC cells (Figures 4B–4D and S15B–S15D; results shown for both AGS cells and HGC-27 cells). In contrast, inhibition of neither eNOS nor nNOS altered NO levels in NK cells under any co-culture condition (Figures S15E and S15F). Furthermore, NK cells isolated from human gastric cancer tissue also showed comparable levels of iNOS expression (Figure S15G). Together, these findings demonstrate that iNOS is the key enzyme responsible for arginine catabolism and NO generation in NK cells.

Figure 4.

Explore the mechanism of AQP5⁺ GC cells influencing anti-tumor activity of NK

(A) NO OD(optical density) value in NK cells co-cultured with AQP5⁺ or AQP5⁻ AGS cells in transwell chamber treated with PBS or 40 nmol/mL L-arginine (n = 3).

(B–D) NO OD value (B); proliferative capacity (C); and cytotoxicity (D) in NK cells co-cultured with AQP5⁺ AGS cells, AQP5⁻ AGS cells, or AQP5⁺ AGS cells with 40 nmol/mL L-arginine in transwell chamber and treated with DMSO or SMT (S-methylisothiourea, iNOS inhibitor) (n = 3).

(E–K) Images (E), tumor volume (F), tumor incidence (G), L-arginine levels (H), NK cell proportion (I), proportion of TNF-α⁺ NK cells (J), and proportion of IFN-γ⁺ NK cells (K) in subcutaneous tumors established with AQP5-overexpressed MFC cells in Inos^fl/fl Ctrl mice and Inos^fl/flNcr1^cre mice or with AQP5-overexpressed MFC cells in Inos^fl/fl Ctrl mice and Inos^fl/flNcr1^cre mice receiving daily oral gavage of L-arginine (2 g/kg; n = 6/group).

(L and M) Representative images (L) and statistical analysis (M) of metastatic nodules in the lungs established with AQP5-overexpressed MFC cells in Inos^fl/fl Ctrl mice and Inos^fl/flNcr1^cre mice or with AQP5-overexpressed MFC cells (GFP-labeled) in Inos^fl/fl Ctrl mice and Inos^fl/flNcr1^cre mice receiving daily oral gavage of L-arginine (2 g/kg; n = 3/group). Scale bar: 2 mm.

p values in (A)–(D), (H)–(K), and (M) were calculated using a two-tailed Student’s t test; in (F) with two-way ANOVA; and in (G) with log rank test. Data in (A)–(D), (F), (H)–(K), and (M) are represented as mean ± SD.

Then, we further analyzed whether the low-arginine TME created by AQP5⁺ GC cells impaired NK cell anti-tumor immunity by limiting NO synthesis in transgenic mice with NK cell-specific iNOS deletion (Inos^fl/flNcr1^cre/-). Compared with Inos^fl/flCtrl mice, subcutaneous tumors established by MFC cells exhibited accelerated growth in Inos^fl/flNcr1^cre/- mice, accompanied by reduced NK cell infiltration (Figure S16). This suggests that NK cell-specific iNOS deletion impairs their anti-tumor function. Notably, conditional iNOS deletion in NK cells did not affect the incidence or growth of tumors established by MFC cells stably overexpressing AQP5. However, its deletion completely abrogated the L-arginine supplementation-induced suppression of tumor growth observed in the Inos^fl/flCtrl mice (Figures 4E–4G). Second, while exogenous L-arginine supplementation significantly increased intratumoral arginine levels in subcutaneous tumors of both mouse models, it only significantly enhanced the proportion of tumor-infiltrating NK cells and cytotoxic NK cells (TNF-α⁺ or IFN-γ⁺) in Inos^fl/flCtrl mice; this enhancement was not observed in Inos^fl/flNcr1^cre/- mice (Figures 4H–4K). Consistent with these findings, L-arginine supplementation failed to suppress lung metastasis of AQP5-overexpressing MFC cells in Inos^fl/flNcr1^cre/- mice, while it significantly reduced the metastatic burden in Inos^fl/flCtrl mice (Figures 4L and 4M). These data demonstrate that the L-arginine deficiency in the TME caused by AQP5⁺ GC cells critically impairs NK cells-mediated tumor immunity by restricting NO synthesis.

Decreased L-arginine anabolism promotes the TCA cycle in AQP5⁺ GC cells

Arginine affects cellular bioenergetics, including ATP levels, the ADP/ATP ratio, and oxygen consumption rates (OCRs).²² Intriguingly, we observed significantly lower L-arginine uptake in AQP5⁺GC cells compared to AQP5^– GC cells (Figure 5A). To determine whether L-arginine deficiency affects bioenergetics in AQP5⁺ GC cells, we performed bioenergetic analyses. Surprisingly, AQP5⁺ GC cells exhibited increased ATP levels and OCRs (Figures 5B–5D). Moreover, ASS1 overexpression in AQP5⁺ GC cells reversed this bioenergetic phenotype (Figure S17), suggesting that AQP5⁺ GC cells activate alternative metabolic pathways to compensate for L-arginine deficiency.

Figure 5.

Evaluate the effect of AQP5 on the bioenergy of GC cells

(A) L-Arginine uptake ratio in AQP5⁺ AGS cells and AQP5⁻ AGS cells calculated by evaluating the L-arginine concentration in media by ELISA (n = 3).

(B) Intracellular ATP levels in AQP5⁺ AGS cells and AQP5⁻ AGS cells at 72 h (n = 3).

(C and D) Oxygen consumption rates (OCRs) of AQP5⁺ and AQP5⁻ AGS cells cultured under the same conditions as described for (B) (n = 3).

(E) L-glutamic acid level detected by targeted metabolic mass spectrometry in tumor tissues expressing AQP5 (n = 22).

(F) Amino acid levels in AQP5⁺ AGS cells and AQP5⁻ AGS cells detected by targeted metabolic mass spectrometry (n = 5).

(G) Schematic diagram of the TCA cycle and UC.

(H) Schematic diagram of intracellular fluxes in AGS cells quantified using U-13C6 glutamate tracer. Red circles, 13C atoms.

(I) Bar charts represent the relative abundance of metabolites between AQP5⁺ AGS cells and AQP5⁻ AGS cells (n = 5 independent biological replicates).

p values in (A), (B), (D)–(F), and (I) were calculated using a two-tailed Student’s t test. Data in (A)–(D), (F), and (I) are represented as mean ± SD; violin plot shows all points in (E).

The UC in AQP5⁺ GC cells was significantly suppressed (Figures 3A and 3B; Tables S3 and S4). Given that cancer cells modulate the expression of UC enzymes to minimize nitrogen waste and optimize nitrogen utilization for biomass,^23,24 along with the well-established intimate metabolic link between the UC and the TCA cycle,²⁵ we speculated that AQP5⁺ GC cells redirect nitrogen metabolites into the TCA cycle to meet their heightened bioenergetic demands. In mitochondria, glutamate serves as a primary anaplerotic substrate feeding the TCA cycle.²⁶ Notably, elevated glutamate levels were observed in both AQP5⁺ GC cells (vs. AQP5^– controls; Figure 2B) and AQP5^High tumor tissues (Figure 5E). Targeted metabolomic profiling revealed significant accumulation of TCA cycle intermediates—including malate, succinate, isocitrate, and citrate—in AQP5⁺GC cells (Figures 5F and 5G; Table S11). These findings indicate that increased glutamic acid enters the TCA cycle rather than the UC in these cells.

To define the impact of AQP5 expression on glutamate metabolism in GC cells, we performed [U-¹³C₆] glutamate tracing in AQP5⁺ and AQP5^– GC cells, followed by mass spectrometric analysis (Figure 5H; Table S12). Our results revealed significant increases in M⁺4 succinate, M⁺4 fumarate, and M⁺4 glutamine, along with decreased M+2 ornithine enrichment in AQP5+ GC cells (Figure 5I). These findings indicate that AQP5⁺ GC cells channel glutamate toward glutamine and TCA cycle intermediates, while reducing UC flux. This metabolic rewiring enables AQP5⁺ GC cells to minimize nitrogen loss while sustaining bioenergetics through enhanced TCA cycle.

AQP5⁺ GC cells redirect nitrogen from the UC to glutamine synthesis by sustaining GLUL stability, thereby facilitating the TCA cycle

Glutamine metabolism plays a crucial role in cellular bioenergetics, as its carbon skeleton can be converted to glutamate via transaminase activity and subsequently to α-ketoglutarate, a key TCA cycle intermediate.²⁷ Our analyses revealed significantly elevated glutamine levels in both AQP5⁺ GC cells (compared to AQP5^– controls; Figure 2B) and AQP5^High GC tissues (Figure 6A). The glutamine metabolism is tightly regulated by two key enzymes: (1) glutamine synthetase (GLUL/GS), which catalyzes glutamine synthesis from glutamate and NH4⁺, and (2) glutaminase (GLS), which mediates glutamine catabolism to generate glutamate (Figure 6B). [U-¹³C₆] glutamate tracing has demonstrated enhanced glutamine synthesis flux in AQP5⁺ GC cells (Figures 5H and 5I). These findings indicate that AQP5⁺ GC cells undergo metabolic reprogramming to accumulate glutamine as an energy reservoir for the TCA cycle by suppressing the UC to minimize nitrogen loss, thereby optimizing their bioenergetic capacity.

Figure 6.

Explore how AQP5 influences glutamine level in GC cells

(A) L-glutamine levels detected by targeted metabolic mass spectrometry in tumor tissues expressing AQP5 (n = 22).

(B) Schematic diagram of glutamate metabolism.

(C and D) L-glutamine level was detected by ELISA (C) and intracellular ATP levels (D) in AQP5⁺ AGS cells, AQP5⁻ AGS cells, and AQP5⁺ AGS cells transfected with GLUL-KD or GLS-KD siRNA for 24 or 48 h (n = 3).

(E) Western blot detection of GLUL and GLS expression in AQP5⁺ AGS cells and AQP5⁻ AGS cells (n = 1).

(F and G) L-glutamine level (F) and L-arginine level (G) were detected by ELISA in AQP5⁺ AGS cells, AQP5⁻ AGS cells, and AQP5⁺ AGS cells transfected with GLUL-KD siRNA (n = 3).

(H and I) AGS cells transfected with Ctrl or AQP5-OE lentivirus were treated with cycloheximide (CHX) for the indicated times (n = 1). (H) Western blots detect GLUL expression and (I) the quantification of relative GLUL levels (n = 3).

(J–L) AQP5 overexpressed or knocked down in AGS cells, which were transfected with GLUL-HA. The cell lysates were subjected to immunoprecipitation (IP) with anti-HA agarose and immunoblotted.

(M) AGS cells transfected with Ctrl or AQP5-OE lentivirus were co-transfected with UBB/UBC siRNA, K63-Ub-HA, or K63R-HA. IP assay was performed with anti-HA agarose, followed by immunoblotting (n = 1).

(N) AGS cells transfected with Ctrl or AQP5-OE lentivirus were co-transfected with GLUL-HA and TRIM21-KD siRNA; cells were analyzed by IP assays with anti-HA agarose and immunoblotted (n = 1).

(O) L-Glutamine level was detected by ELISA in AGS cells co-transfected with AQP5-OE lentivirus and TRIM21-KD siRNA or corresponding controls (n = 3).

(P and Q) AQP5-knockdown (P) and AQP5-overexpressing (Q) AGS cells were co-transfected with GLUL-HA and TRIM21-Myc (n = 1). Cell lysates were subjected to IP with anti-HA agarose and immunoblotted.

(R) Gradient overexpression of AQP5 in 293T cells co-transfected with TRIM21-Myc and GLUL-HA. The cell lysates were subjected to an IP assay with anti-HA agarose and immunoblotted (n = 1).

p values in (A), (C), (D), (F), (G), and (O) were calculated using a two-tailed Student’s t test. Data in (C), (D), (F), (G), (I), and (O) are represented as mean ± SD; violin plot shows all points in (A).

Next, we defined the impact of GLUL and GLS on glutamine metabolism in AQP5⁺ GC cells by silencing GLUL and GLS with small interfering RNA (siRNA) (Figures S18A and S18B). Silencing of GLS led to time-dependent accumulation of glutamine accompanied by progressive ATP decrease (Figures 6C and 6D), indicating that glutamine catabolism was critical for maintaining cellular bioenergetics through TCA cycle input. In contrast, GLUL silencing exhibited distinct temporal effects; while 24 h silencing only partially reduced glutamine levels (remaining higher than AQP5⁻ controls) without affecting ATP production, extended silencing for 48–72 h decreased glutamine to levels observed in AQP5^- GC cells and significantly reduced ATP content (Figures 6C and 6D). Notably, baseline expression analysis revealed significantly higher GLUL but comparable GLS levels in AQP5⁺ versus AQP5⁻ GC cells (Figure 6E). Importantly, silencing GLUL for 48 h not only reduced glutamine levels but also reversed L-arginine decrease observed in AQP5⁺ GC cells (Figures 6F and 6G) and markedly inhibited proliferation and migration of AQP5⁺ GC cells (Figures S18C and S18D). These results established that AQP5⁺ GC cells maintain elevated glutamine synthesis through GLUL upregulation to sustain TCA cycle, achieving this by diverting nitrogen flux from the UC to glutamine accumulation.

To elucidate how AQP5 regulates GLUL expression, we performed comprehensive molecular analyses. Intriguingly, while AQP5 overexpression in GC cells did not alter GLUL transcript levels, it markedly enhanced GLUL protein stability (Figures S19A, 6H, and 6I). Given the crucial role of the ubiquitin-proteasome system in regulating protein degradation and homeostasis through a highly coordinated process involving ubiquitin tagging and proteasomal degradation,²⁸ we examined GLUL ubiquitination patterns and found that AQP5 overexpression increased GLUL ubiquitination (Figure 6J). Ubiquitin contains seven lysine residues (K6, K11, K27, K29, K33, K48, and K63) and an N-terminal methionine that can form eight distinct types of ubiquitin chains, among which K48-, K63-, and K27-linked chains are most abundant.²⁹ AQP5 overexpression specifically promoted K63-linked ubiquitination of GLUL, without affecting K27- or K48-linked ubiquitination (Figures 6K, 6L, S19B, and S19C). This finding aligns with our previous work establishing that K63-linked ubiquitination, unlike canonical K48-linked ubiquitination, promotes protein stabilization rather than proteasomal degradation.^30,31 To further validate this mechanism, we employed a genetic complementation approach: after depleting endogenous ubiquitin, we reconstituted cells with either wild-type hemagglutinin (HA)-tagged ubiquitin (K63-Ub) or a K63-mutant ubiquitin (K63R). Notably, the K63 mutation completely abolished AQP5-mediated GLUL stabilization (Figure 6M). These results demonstrate that AQP5 promotes GLUL protein stability through specific K63-linked ubiquitination, revealing a previously unrecognized post-translational regulatory mechanism in GC metabolism.

To elucidate the mechanism underlying AQP5-mediated GLUL ubiquitination, we performed mass spectrometry-based interactome analysis. Among all identified GLUL-interacting proteins, TRIM21 emerged as the only E3 ubiquitin ligase closely associated with ubiquitination (Table S13). As a well-characterized E3 ligase, TRIM21 typically regulates protein stability through ubiquitin-mediated degradation³²; however, our co-immunoprecipitation experiments identified a ternary complex comprising AQP5, TRIM21, and GLUL (Figures S19D–S19F). Surprisingly, TRIM21 significantly promoted rather than decreased GLUL protein stability in 293T cells (Figure S19G). Therefore, we hypothesized that AQP5 induced the ubiquitination of GLUL via TRIM21. Moreover, TRIM21 knockdown not only decreased K63-linked ubiquitination of GLUL and glutamine level but also abolished AQP5-induced K63-linked ubiquitination of GLUL and the increase in glutamine levels (Figures 6N and 6O). We thus postulated that AQP5 might enhance GLUL protein stability by promoting the GLUL-TRIM21 interaction.

Consistent with our hypothesis, overexpression of AQP5 significantly promoted the interaction of GLUL and TRIM21, while AQP5 knockdown attenuated their binding (Figures 6P and 6Q). To further verify that AQP5 regulates the GLUL-TRIM21 interaction, we performed graded overexpression of AQP5 in 293 T cells. Our results demonstrated that graded AQP5 overexpression progressively enhanced TRIM21-GLUL interaction, K63-linked ubiquitination of GLUL, and consequent GLUL protein stabilization (Figure 6R). In addition, AQP5⁺ cells formed a DHX9-GLUL-TRIM21 complex, which was not observed in AQP5⁻ cells (Figure S19H). While DHX9 interacts with TRIM21, this interaction did not affect DHX9 protein stability (Figure S19I). These data suggested that AQP5 promotes K63-linked ubiquitination of GLUL by recruiting the E3 ligase TRIM21 to GLUL, thereby sustaining GLUL protein stability.

Combining L-arginine administration with ULK1 inhibition induces repression of AQP5^High tumors

Our previous studies demonstrated that AQP5 activates autophagy by regulating ULK1.⁹ Consistent with our previous findings, AQP5⁺ GC cells exhibited enhanced LC3Ⅰ-to-LC3Ⅱ conversion compared to AQP5⁻ cells, indicating elevated autophagic activity (Figure S20A). Moreover, inhibition of autophagy significantly decreased the malignant biological functions of AQP5⁺ cells (Figures S20B–S20D). ULK1 inhibitors have demonstrated preliminary clinical efficacy in various tumors, including colorectal and pancreatic cancers, but their therapeutic benefits remain limited to a subset of patients (e.g., clinical study identifier: S22-00776). In this study, we found that AQP5 impaired NK cell activity through decreasing arginine production. We hypothesized that AQP5 exerts its pro-tumorigenic effects via both ULK1-mediated autophagic activation in tumor cells and impairment of NK cell cytotoxicity. Consequently, combining ULK1 inhibitors with exogenous L-arginine supplementation could represent a synergistic therapeutic strategy for AQP5^High GC. To validate this strategy, we utilized patient-derived organoids (PDOs) and patient-derived xenograft (PDOX) models to evaluate the therapeutic efficacy of ULK1 inhibition, either alone or in combination with L-arginine supplementation (Figure 7A). In a co-culture system where NK-92 cells were embedded with PDOs in Matrigel, we observed that L-arginine treatment significantly enhanced NK-92 cell into AQP5^High PDOs and that the combination of L-arginine with the ULK1 inhibitor substantially disrupted their structural integrity. However, neither ULK1 inhibitor monotherapy nor L-arginine treatment alone could fully dismantle the organoid architecture (Figure 7B). Notably, this combinatorial effect was markedly attenuated in AQP5-low PDOs compared to AQP5-high PDOs (Figure 7B). These results demonstrated the improved therapeutic efficacy of this combination strategy for AQP5⁺ GC in the in vitro PDO model.

Figure 7.

Validation of anticancer effects of L-arginine and ULK1 inhibitors on PDO and PDOX models

(A) Schematic diagram of PDO and PDOX model construction.

(B) Snapshots of NK-92 cells co-cultured with PDOs generated with gastric tissues with AQP5 high or low expression were performed by microscopy in 24 h (left), and the NK cells per organoid were calculated (right) (n = 3). PDOs pre-labeled with red dye were inlaid in Matrigel, and NK cells were pre-labeled with CFSE. Scale bar: 100 μm.

(C and D) PDOs generated with gastric tissues with AQP5 high or low expression were injected subcutaneously into mice (3–4 million cells per mice). When the tumor grew to 0.15–0.2 cm³, the mice were randomly grouped, and the drug treatment started. Tumors were examined over 27 days (n = 6/group); the tumor volume (left: images, right: the statistical analysis) (C) and tumor inhibition rate (D) were evaluated in the indicated groups at the indicated time points.

(E) Proportion of NK cells in different groups of subcutaneous tumors detected by flow cytometry.

Statistical analyses, p values in (B), (C), (D), and (E) were calculated using two-tailed Student’s t test. Data in (B–D) are represented as mean ± SD; violin plot shows all points in (E).

To further validate our combination strategy, we established PDOX models using gastric cancer tissues with either AQP5^High or AQP5^Low expression in NV-NSG-hIL15 mice reconstituted with human peripheral blood mononuclear cells (PBMCs) to establish a functional human immune system. Initial characterization confirmed both AQP5 expression levels in subcutaneous tumors and the successful repopulation of NK cells in peripheral blood (Figures S21A and S21B). The combination therapy of ULK1 inhibitor and L-arginine demonstrated significant tumor regression specifically in AQP5-high PDOX models, while showing minimal effect in AQP5-low models (Figures 7C and 7D). Interestingly, L-arginine monotherapy exhibited greater efficacy against AQP5-high tumors compared to AQP5-low tumors (Figures 7C and 7D). Consistent with the results of the in vitro PDO model, L-arginine treatment also promoted NK cell infiltration into tumors, especially in AQP5-high tumors (Figure 7E). These results collectively indicated that the synergistic combination of ULK1 inhibition and L-arginine supplementation represents a promising therapeutic approach for gastric cancer patients with high AQP5 expression.

Discussion

An 11-year follow-up study revealed that high NK cell cytotoxic activity was associated with better therapeutic efficacy and reduced cancer recurrence.³³ While clinical trials have shown promising results for NK cell therapy in various cancers, including hematological tumors, neuroblastoma, ovarian cancer, breast cancer, and GC, with good therapeutic outcomes,³⁴ the rapid functional impairment of NK cells upon recruitment into tumors remains a significant limitation for their therapeutic application as immunotherapeutics.³⁵ Thus, efforts have been made to develop strategies to maintain functional NK cells in the TME.³⁶ In this study, we demonstrated that arginine deficiency in TME induced by AQP5⁺ GC-CSCs was an important factor impairing NK cell anti-tumor activity.

Extensive research has established the therapeutic potential of arginine supplementation in cancer immunotherapy. Multiple studies have demonstrated that arginine supplements can improve the recovery of defective NK cells after surgery to prevent metastases and increase CD8⁺ T cell effectiveness to reduce tumor burden.^18,37,38,39 The clinical significance of targeting arginine metabolism is underscored by several ongoing investigations. Notably, a phase 2 clinical trial demonstrated that L-arginine supplementation significantly improved radiotherapy outcomes in patients with brain metastases, leading to prolonged progression-free survival.⁴⁰ Furthermore, comprehensive analyses of immune checkpoint inhibitor trials have identified baseline plasma L-arginine levels as a robust predictor of therapeutic response.⁴¹ A study has revealed that serum arginine concentrations in patients with gastric cancers were significantly decreased, which might be a potential therapeutic target.⁴² Our study revealed that arginine supplementation could restore NK cell anti-tumor immunity, and its combination with a ULK1 inhibitor achieved an enhanced therapeutic effect on AQP5^High tumors in PDO and PDOX models. Several reports also indicate that tumor cells tend to plunder arginine in the TME to support oncogenic metabolism.^43,44 However, our results showed that oral administration of arginine alone did not promote tumor growth, potentially attributed to lower arginine uptake in AQP5⁺ GC cells. Importantly, early study has reported that in vivo arginine supplements (30 g/day for 3 days) increased the number of circulating CD56⁺ cells and the NK cell cytotoxicity in healthy volunteers,⁴⁵ which further supported that the combination of arginine supplementation would be an effective treatment strategy for GC patients with AQP5^High expression.

Recent studies have characterized AQP5 as a specific surface marker for GC-CSCs that is not expressed in other CSCs, including lung cancer, liver cancer, colorectal cancer.^7,8,9 AQP5 expression not only contributes to the tumorigenesis of gastric mucosa and malignant progression of GC but is also associated with various types of tumor-infiltrating immune cells with unclear mechanism.^7,8,9,10 Indeed, the reciprocal crosstalk between CSCs and infiltrating immune cells in the TME contributes to tumor progression⁴⁶; for example, CSCs can repress cytotoxic T lymphocytes to evade immune attack through receptor/ligand interactions or metabolites.⁴⁷ However, few reports have explored the impact of CSCs on the anti-tumor activity of NK cells. In this study, high AQP5 expression in GC cells reduced the quantity and cytotoxicity of tumor-infiltrating NK cells, confirming the reshaping effect of AQP5⁺ GC-CSC subset on NK cell anti-tumor immunity. Further studies exploring other cell subpopulations that induce defective NK cells and the potential mechanisms in various types of tumors are urgently needed, which will provide effective strategies for NK cell-mediated immunotherapy for malignant diseases.

AQP5⁺ GC cells were found to reprogram the UC to decrease arginine production, resulting in arginine deficiency in the TME and consequently impaired NK activity. The TME is frequently scarce in arginine due to its poor supply and high consumption.^48,49 L-arginine is a substrate for multiple metabolites (including polyamines and NO) and has strong immunomodulatory properties.^21,37 Immune cells that cannot express ASS1 are far less tolerant to arginine deficiency.⁵⁰ Low-arginine condition decreases T cell survival and confers Treg (regulatory T cell)-like properties upon activated CD4⁺ T cells, creating an immunosuppressive microenvironment.^49,51 More specifically, tumor-associated macrophages consume arginine in the TME to synthesize proline and secrete ornithine, creating an inhospitable milieu for anti-tumor CD8⁺ T cells.⁵² In this study, we demonstrated that AQP5 expression in GC tumor tissues did not affect the proportion of tumor-infiltrating T cells or macrophages, nor their proliferation or cytotoxicity. However, AQP5⁺ GC cells significantly reduced the anti-tumor activity of NK cells. The modulatory mechanism of arginine on NK cells remains unclear.²¹ Herein, we analyzed the expression levels of iNOS in the immune cells of GC tissues and found that NK cells expressed comparable levels of iNOS, implying NK cells may require more arginine to perform their immunomodulatory functions. By inhibiting iNOS in vitro and in Inos^fl/fl Ncr1^cre transgenic mice, this study confirms that arginine triggers the anti-tumor activity of NK cells depending on iNOS metabolism.

The predominant purpose of the UC in the liver is to convert excess nitrogen into disposable urea, preventing cell death triggered by toxic ammonia.²⁵ Multiple types of tumors reduce the UC by altering UC enzyme expression, and undischarged nitrogen is incorporated into glutamine by transamination. This glutamine serves as a ready source of carbon and nitrogen to support biosynthesis, energetics, and cellular homeostasis.^23,24,53 Although AQP5⁺ GC cells dysregulated the UC by suppressing ASS1 expression, AQP5 expression could accumulate glutamine by stabilizing the GLUL protein level to store unwasted nitrogen. Unlike ammonia, which is generated from glutamine via mitochondrial glutaminase 1 (GLS1) catalysis and induces effector CD8⁺ T cells death,⁵⁴ AQP5⁺ GC cells did not change the expression of GLS. Importantly, interfering with GLUL expression in AQP5⁺ GC cells for 24 h showed no impact on bioenergetics. However, cellular bioenergetics were significantly decreased after 48 h of interference, supporting the notion that accumulated glutamine served as a bioenergetic storage. Under hypoxia conditions, glutamine-carbon was overused in tumor cells.⁵⁵ The high influx of glutamine-derived carbon into the TCA cycle supported tumor cell proliferation.⁵⁶ Metabolomics further revealed that AQP5 expression strongly altered glutamate and central carbon metabolism while exerting minimal effects on nitrogen metabolism. Glutamate is the major source of α-ketoglutarate that feeds the TCA cycle. Hence, [¹³C6]glutamate tracing further confirmed that glutamate—acting as a carbon donor in AQP5⁺ GC cells—enters the TCA cycle to support bioenergetics.

Herein, these findings suggest that the specialized metabolic programming of AQP5⁺ CSCs not only blocks the generation of an effective anti-tumor immune response but can also reserve energy for themselves. Notably, we demonstrated that arginine deficiency in the TME critically impairs NK cell function, providing a rationale for developing improved NK cell-based immunotherapy strategies.

Limitations of the study

While this study elucidates the role of AQP5⁺ gastric CSCs in impairing NK cell function through arginine deprivation and metabolic reprogramming, several limitations should be noted. First, the mechanistic insights were primarily derived from preclinical models (e.g., murine xenografts and patient-derived organoids), which might not fully recapitulate the complexity of human tumor-immune interactions. Second, the therapeutic efficacy of L-arginine supplementation combined with ULK1 inhibition, though promising, requires validation in larger cohorts and clinical trials to assess translational potential. Finally, we observed a negative correlation between arginine levels and AQP5 expression in clinical samples, but the metabolic adaptations of AQP5⁺ cells were mainly analyzed in vitro. Future work should investigate these pathways under dynamic TME conditions, such as hypoxia or nutrient fluctuations, to better model in vivo conditions. Addressing these limitations could further refine therapeutic strategies targeting AQP5⁺ GC stem cells.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Qingli Bie (xiaobie890101@163.com).

Materials availability

No material resources were generated.

Data and code availability

The single-cell sequence data have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2021) in the National Genomics Data Center (Nucleic Acids Res 2022), China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (GSA: CRA015435), which are publicly accessible at https://ngdc.cncb.ac.cn/gsa.

The proteomics and the mass spectrometry raw data have been deposited in iProX via PRIDE and Zenodo (IPX0012430001, IPX0012430002, and IPX0012430003; https://doi.org/10.5281/zenodo.16777120). The metabolomics data have been deposited at Zenodo (https://doi.org/10.5281/zenodo.15646147, https://doi.org/10.5281/zenodo.15646453, https://doi.org/10.5281/zenodo.15646497, https://doi.org/10.5281/zenodo.15646966, https://doi.org/10.5281/zenodo.15647006, https://doi.org/10.5281/zenodo.15742594, and https://doi.org/10.5281/zenodo.15647052). All data are publicly available as of the date of publication. Accession numbers are listed in the key resources table.

Acknowledgments

This study was supported by grants from the National Natural Science Foundation of China (no. 82173371, 82273447, 82273069, 82372679, 82472959, and 82403283), a project funded by China Postdoctoral Science Foundation (2022M711320 and 2022M711322), Tai Shan Young Scholar Foundation of Shandong Province (no. tsqn201909192 and tsqn202312383), Shandong Provincial Natural Science Foundation (no. ZR2021QH021, ZR202112020099, ZR2024MH139, and ZR2024QH089), and Shandong Postdoctoral innovation project (no. SDCX-ZG-202201002).

Author contributions

B.Z., Q.B., and H.X. designed the study, directed the project, and supervised data analysis. R.Z., B.H., and J.L. performed and analyzed most of the experiments. L.H. provided primary GC specimen and corresponding serum samples and followed up the survival status and prognosis. Y.W. and T.L. collected peripheral blood of healthy volunteers. M.Z. and T.Z. provided iNOS mice and performed corresponding experiments. Y.Z. and X.Z. performed data organization and statistical analysis. Q.B., H.X., B.Z., R.Z., and B.H. wrote the manuscript. All authors edited the manuscript.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit multiclonal antibody anti-AQP5	Abcam	Cat# ab315855
Rabbit monoclonalantibody anti-EpCAM	Abcam	Cat# ab223582; RRID：AB_2762366
Rabbit monoclonalantibody anti-CD3	Abcam	Cat# ab16669; RRID：AB_443425
Rabbit monoclonalantibody anti-CD56	Abcam	Cat# ab220360
Rabbit monoclonalantibody anti-CD3	Biolegend	Cat# 100205; RRID：AB_312662
Rabbit monoclonalantibody anti-CD56	R&D Systems	Cat# FAB7820A-025UG; RRID：AB_3652722
Mouse monoclonal antibody anti-CD3	Biolegend	Cat# 980008;RRID：AB_2810818
Mouse monoclonal antibody anti-CD56	R&D Systems	Cat# FAB2408A-025; RRID：AB_562656
Rabbit monoclonalantibody anti-ASS1	Cell Signaling Technology	Cat# 70720; RRID：AB_2799790
Mouse monoclonal antibody anti-GAPDH	Abcam	Cat# ab8245; RRID：AB_2107448
Rabbit monoclonalantibody anti-DHX9	Cell Signaling Technology	Cat# 71286
Rabbit polyclonal antibody anti-ARG	BOSTER	Cat# BM3973; RRID：AB_331563
Rabbit monoclonal antibody anti-H3	Cell Signaling Technology	Cat# 9715
Rabbit polyclonal antibody anti-β-tubulin	Bioss	Cat# bs-4511R, RRID:AB_11114300
Mouse monoclonal antibody anti-KPNB1	Abcam	Cat# ab2811; RRID：AB_2133989
Mouse monoclonalantibody anti-DHX9	Thermo Fisher	Cat# MA5-35534; RRID：AB_2849434
Rabbit monoclonal antibody anti-TNF-α	R&D Systems	Cat# MAB4101-SP; RRID：AB_2240643
Rabbit monoclonal antibody anti-IFN-γ	R&D Systems	Cat# MAB485-SP; RRID：AB_2123047
Mouse monoclonal antibody anti-GLS	BOSTER	Cat# M01272-2; RRID：AB_3081831
Rabbit monoclonal antibody anti-GLUL	Cell Signaling Technology	Cat# 80636; RRID：AB_2799956
Rabbit monoclonal antibody anti-Ubiquitin	Cell Signaling Technology	Cat# 58395; RRID：AB_3075532
Rabbit monoclonal antibody anti-K63 Ubiquitin	Abcam	Cat# ab179434; RRID：AB_2895239
Rabbit monoclonal antibody anti-K27 Ubiquitin	Abcam	Cat# ab181537; RRID：AB_2713902
Rabbit monoclonal antibody anti-K48 Ubiquitin	Abcam	Cat# ab140601; RRID：AB_2783797
Rabbit monoclonal antibody anti-TRIM21	BOSTER	Cat# A02079-2; RRID：AB_3081516
Mouse monoclonal antibody anti-LGR5	R&D Systems	Cat# FAB8078P; RRID：AB_3652859
Mouse monoclonal antibody anti-CD44	R&D Systems	Cat# FAB4948A; RRID：AB_1096713
Rabbit monoclonal antibody anti-CD3	Cell Signaling Technology	Cat# 86603; RRID：AB_2800082
Mouse monoclonal antibody anti-NK1.1	Thermo Fisher	Cat# MSM1-2010-P0
Mouse monoclonal antibody anti-HA	Thermo Fisher	Cat# 26183; RRID：AB_2533052
Mouse monoclonal antibody anti-FLAG	Thermo Fisher	Cat# MA1-91878; RRID：AB_1957945
Mouse monoclonal antibody anti-MYC	Thermo Fisher	Cat# MA1-980; RRID：AB_558470
Rabbit monoclonal antibody anti-LC3	Cell Signaling Technology	Cat# 12741; RRID：AB_2617131
Rabbit monoclonal antibody anti-P62	Cell Signaling Technology	Cat# 5114; RRID：AB_10624872
Mouse monoclonal antibody anti-CD45	R&D Systems	Cat# FAB11444P-100
Rabbit monoclonal antibody anti-B220	R&D Systems	Cat# FAB1217P; RRID：AB_357006
Rabbit monoclonal antibody anti-F4/80	R&D Systems	Cat# FAB5580P-100UG; RRID：AB_2044655
Mouse monoclonal antibody anti-B220	R&D Systems	Cat# FAB1430A-025; RRID：AB_2174121
Mouse monoclonal antibody anti-CD68	R&D Systems	Cat# IC20401G-100UG; RRID：AB_3654926
Chemicals, peptides, and recombinant proteins
Anti-Flag Affinity Gel	Bimake	Cat# B23101
Anti-Myc tag Rabbit mAb conjugated Agarose Beads	Engibody Biotechnology	Cat# AT0541
href = "http://engibody.com/products/HA-tag-Mouse-mAb-conjugated-Agarose-Beads-AT0079.html" Anti-HA tag Rabbit mAb conjugated Agarose Beads	Engibody Biotechnology	Cat# AT1794
L-Arginine	MedChemExpress	Cat# E5036
SMT	Beyotime	Cat# S0008
7-NI	Abcam	Cat# ab120233
L-NIO	Macklin	Cat# L881893
MRT68921	CSNpharm	Cat# CSN11346
CFSE	Beyotime	Cat# Y229292
CD3 MicroBeads, human	Miltenyi Biotec	Cat# 130-097-043
CD56 MicroBeads, human	Miltenyi Biotec	Cat# 130-097-042
CD19 MicroBeads, mouse	Miltenyi Biotec	Cat# 130-121-301
CD11b MicroBeads	Miltenyi Biotec	Cat# 130-093-634
CD3 MicroBead Kit, mouse	Miltenyi Biotec	Cat# 130-094-973
Critical commercial assays
CytoTox 96 Non-Radioactive Cytotoxicity Assay Kit	Promega	Cat# G1780
DAF-FM DA	Thermo Fisher	Cat# D23844
EnVision-HRP kit	Dako	Cat# K4001
PrimeScript RT Reagent Kit	TAKARA	Cat# RR037A
Dual-Luciferase Reporter Assay System	Promega	Cat# E1910
ChIP assay kit	Beyotime	Cat# P2078
L-arginine ELISA kit	MyBiosource	Cat# MBS728648-96
ATP Determination Kit	Thermo Fisher	Cat# A22066
Glutamine Assay Kit-WST	DOJINDO	Cat# G268
NK Cell Isolation Kit, human	Miltenyi Biotec	Cat# 130-092-657
NK Cell Isolation Kit, mouse	Miltenyi Biotec	Cat# 130-115-818
Reverse-chip Kit	BersinBio	Cat# Bes5005
Experimental models: cell lines
Human: 293T	ATCC	Cat# CRL-3216
Human:AGS	ATCC	Cat# CRL-1739
Human:HGC-27	National Collection of Authenticated Cell Cultures	Cat# SCSP-5263
Human:NK-92	ATCC	Cat# CRL-2407
Mouse:MGC	Procell	Cat# CL-0156
Experimental models: organisms/strains
Mouse: NSG	Beijing Viewsolid Biotech	N/A
Mouse: C57BL/6J	Beijing Viewsolid Biotech	N/A
Mouse: NV-NSG-hIL15	Shanghai Model Organisms Center	N/A
Primers used in this paper, see Table S1
Recombinant DNA
pCMV FLAG-AQP5	This paper	N/A
pCMV HA-ASS1	This paper	N/A
pcDNA3.1-HA-DHX9	This paper	N/A
pcDNA3.1-MYC-KPNB1	This paper	N/A
pcDNA3.1-HA-GLUL	This paper	N/A
pcDNA3.1-MYC-KPNB1	This paper	N/A
software and algorithms
CRISPR design tool	IDT design tool	https://design.synthego.com/
TCGA data source	TCGA Hub	https://xenabrowser.net/datapages/
Deposited data
Differential metabolites in AQP5+ and AQP5- gastric cancer cells identified by untargeted metabolomics analyses	This paper	https://doi.org/10.5281/zenodo.15646147
Differential amino acid metabolites in AQP5+ and AQP5- gastric cancer cells identified by targeted metabolomics analyses	This paper	https://doi.org/10.5281/zenodo.15646453
Differential amino acid metabolites in AQP5+ and AQP5- gastric cancer cells medium identified by targeted metabolomics analyses	This paper	https://doi.org/10.5281/zenodo.15646497
Differential amino acid metabolites in gastric cancer tissue identified by targeted metabolomics analyses	This paper	https://doi.org/10.5281/zenodo.15646966
Differentially expressed protein between exogenous overexpression of AQP5 and control group	This paper	IPX0012430002
Potential proteins interacted with ASS1 promoter identified using Reverse-ChIP analyses	This paper	IPX0012430003
Differential metabolites in NK cells treated with PBS or L-Arginine identified by untargeted metabolomics analyses	This paper	https://doi.org/10.5281/zenodo.15647006
Differential center carbon metabolites in AQP5+ and AQP5- gastric cancer cells identified by targeted metabolomics analyses	This paper	https://doi.org/10.5281/zenodo.15742594
Potential proteins interacted with GLUL	This paper	https://doi.org/10.5281/zenodo.16777120
Intracellular fluxes in AQP5+ and AQP5- gastric cancer cells quantified using U-13C6 glutamate tracer.	This paper	https://doi.org/10.5281/zenodo.15647052
Differential proteins that potentially interacted with DHX9 protein in exogenous overexpression of AQP5 and control group	This paper	IPX0012430003
Single-cell sequencing of subcutaneous transplanted tumors in AQP5+ and AQP5- cells	This paper	GSA: CRA015435

Experimental model and subject details

Human subjects

Gastric cancer specimens were sourced from the Affiliated Hospital of Jining Medical College. Patients had not received preoperative radiotherapy or chemotherapy. The study received ethical approval (No. 202403B50) from the hospital’s Research Ethics Committee, and informed consent was secured from all participants before inclusion.

Mouse experiments

Approval for the animal experiment protocols was granted by the Affiliated Hospital of Jining Medical University (No. 202403B50). Mice were housed in groups of six in each cage. The mice were housed under specific pathogen-free conditions, where they were maintained under controlled environmental conditions (temperature: 22 ± 2 C; humidity: 45% ± 5%) with a 12h light-dark cycle. They were provided with a standard commercial diet and sterile water ad libitum throughout the study, in accordance with standard guidelines for animal care and husbandry.

In vivo tumor xenograft models

Subcutaneous cell injections were administered to C57BL/6 (6–10 weeks, Female, average body weight of 22.4 ± 0.8 g) or NSG mice (NOD-scid IL2Rγ^null, 6–10 weeks, Female, average body weight of 20.8 ± 0.8 g) obtained from Beijing Viewsolid Biotech Co. Ltd. Mice were anesthetized and euthanized upon losing 20% of body weight or when tumors reached 1.4 cm in diameter. Xenografts were excised, weighed, and preserved in formalin, each treatment arm consisted of six mice (n = 6).

In Figure 1A, the single-cell RNA sequencing was performed on an AQP5+ tumor tissue with a volume of 628 mm³ and an AQP5− tumor tissue with a volume of 144 mm³. The AQP5+ tumor tissue yielded 6,969 cells, while the AQP5− tumor tissue yielded 5,934 cells. All protocols were approved by the Animal Ethics Committee of the Affiliated Hospital of Jining Medical University.

Tail vein injection metastasis model

In the tail vein injection model, GFP-labeled MFC gastric cancer cells were injected (2 million) into the tail veins of C57B6/J (6–10 weeks, Female, average body weight of 22.4 ± 0.8 g) or NSG mice (NOD-scid IL2Rγ^null, 6–10 weeks, Female, average body weight of 20.8 ± 0.8 g) obtained from Beijing Viewsolid Biotech Co. Ltd. After 28 days, mice were sacrificed, and lung metastasis was assessed via GFP signals using an OV100 microscope (Olympus), each treatment arm consisted of five mice (n = 3).

Generation of transgenic inos^fl/flNCR1^cre mice

Single-guide RNA (sgRNA) targeting inos exons was designed for precise Cas9-directed cleavage. Mutated DNA templates were synthesized through chemical synthesis or PCR, incorporating repair sequences compatible with CRISPR/Cas9. The sgRNA and repair template were inserted into a CRISPR/Cas9 vector for genome editing. This vector was injected into fertilized eggs, which were implanted into a surrogate mouse. Some offspring carried the desired sequence. Genotyping by PCR and sequencing confirmed the presence of the target mutation. Inos flox mice were crossed with NK cell-specific Cre (NCR1^Cre) mice to produce flox homozygous, Cre-positive mice for conditional inos knockout in NK cells.

For subcutaneous xenograft studies, inos^fl/flNCR1^cre and wild-type (WT) mice (female, 6–10 weeks old; mean body weight 22.4 ± 0.8 g) were injected with tumor cells. Mice were anesthetized and humanely euthanized upon either a 20% body weight loss or when tumor diameter exceeded 1.4 cm. Xenografts were surgically excised, weighed, and fixed in formalin for further analysis. Each experimental group contained six mice (n = 6).

For metastatic colonization assays, GFP-labeled MFC gastric cancer cells (2 × 10⁶ cells per mouse) were intravenously injected into the tail veins of inos^fl/flNCR1^cre and WT mice (same age and weight range). After 28 days, mice were euthanized, and lung metastases were quantified by GFP fluorescence imaging using an Olympus OV100 microscope. Each treatment group comprised five mice (n = 3).

PDO and PDOX model

According to the manufacturer’s protocol (bioGenous), harvested tissues were minced in basal medium after removal of fatty tissues and blood vessels. Tissue fragments were digested in organoid dissociation solution at 37°C for 30 min with periodic mechanical agitation (pipetting). The digested suspension was sequentially filtered through a 100 μm strainer, treated with RBC lysis solution (1 min), and centrifuged to collect cellular pellet. The pellet was resuspended in a 1:1 mixture of basal medium and Matrigel (ABWBIO), plated in 24-well flat-bottom plates, and allowed to solidified (20 min, 37°C/5% CO₂). Finally, 400 μL of PDO medium was added to cover the culture wells and replaced every 48 h.

For organoid-NK cell interaction analysis: After 7 days of culture, mature tumor organoids were stained with SPY555 (red fluorescence) for visualization. Peripheral blood NK cells were labeled with CFSE (green fluorescence) and co-cultured with organoids in Matrigel-containing medium. The cell-Matrigel mixture was plated in confocal dishes, solidified (20 min, 37°C/5% CO₂), and overlaid with 800 μL PDO medium containing experimental treatments:1μM MRT62981 (ULK1 inhibitor) alone, 40 nM L-Arg alone or combination therapy. Organoid-NK interactions were visualized and analyzed using laser scanning confocal microscopy after 24/48 h incubation.

The patient-derived xenograft (PDOX) models: NV-NSG-hIL15 mice (6–10 weeks, Female, average body weight of 20.8 ± 0.8 g) received subcutaneous injections of 3-4×10⁶ organoid cells in 50% Matrigel (right flank). Tumor growth was monitored using caliper measurements. When tumors reached 150–200 mm³ (calculated as [length × width²]/2), mice were randomized into treatment groups: L-Arg: 2 g/kg in sterile PBS, daily oral gavage; MRT62981: 20 mg/kg (formulated in 10% DMSO, 40% PEG300, 5% Tween-80, 45% saline), i.p. every other day; both agents at above dose combinations, each treatment arm consisted of six mice (n = 6).

Treatments continued for 27 days with regular tumor monitoring. All procedures were performed in accordance with institutional animal care guidelines (max tumor diameter <15 mm, volume <1500 mm³).

Cell culture and transfection

The MFC mouse gastric cancer cell line (iCell Bioscience) is grown in RPMI 1640 medium with 10% fetal bovine serum (FBS), while the AGS human gastric cancer cell line (iCell) is maintained in DMEM/F-12 medium containing FBS. The NK-92 cell line (iCell Bioscience) is cultured in a specialized medium. All cell lines were maintained at 37°C in a humidified 5% CO₂ atmosphere. Cell line authentication was performed via short tandem repeat (STR) profiling (Genetica DNA Laboratory Services). Routine mycoplasma contamination testing confirmed the absence of contamination. Transfection was conducted with Lipofectamine 3000 reagent according to the manufacturer’s instructions. The sequences for shRNAs and siRNAs are listed in Table S1.

Method details

Cell sorting

We designed single-guide RNAs targeting the 3′ end of the AQP5 gene. A donor plasmid containing a mCherry coding sequence flanked by homology arms matching the AQP5 genomic region was constructed. The sgRNA (carrying puromycin resistance), Cas9, and donor plasmid (carrying neomycin resistance) were co-transfected into target cells. After 48–72 h, cells were screened for fluorescence to identify successful knock-in. The transfected cells were subjected to antibiotic selection to identify positive cells. Subsequently, the cells were sorted based on mCherry fluorescence intensity: mCherryhigh cells were designated as AQP5+, while mCherrylow/non cells were designated as AQP5-. The sorted populations were collected and used for downstream experiments.

Reverse-ChIP

The reverse-chip kit was utilized for this study. Cells (3 × 10⁸) were crosslinked at room temperature for 10 min using 3% formaldehyde, and the reaction was halted by the addition of 0.125 M glycine. Following this, cells were scraped and harvested, and chromatin DNA was sonicated to produce fragments approximately 500 bp in size. Biotin-labeled probes for ASS1, provided by BersinBio, are listed in Table S8. A mixture of pre-denatured probes (85°C for 3 min was prepared at a final concentration of 1 nM per probe and added to the chromatin supernatant for hybridization, which proceeded under the following conditions: 85°C for 10 min, 37°C for 30 min, 70°C for 5 min, 37°C for 30 min, 55°C for 2.5 min, and 37°C for 60 min. The supernatant was subsequently incubated with streptavidin magnetic beads at 37°C for 2 h. After five washes, the beads were resuspended in elution buffer, and proteins were eluted through heating and shaking. Finally, the protein samples were analyzed by mass spectrometry.

Tumor dissociation and single-cell preparation

Subcutaneous tumors were aseptically harvested from tumor-bearing mice and placed in cold PBS supplemented with 2% FBS. Tumors were mechanically dissociated using sterile scissors and further digested enzymatically in RPMI-1640 medium containing 1 mg/mL collagenase IV (Sigma-Aldrich), 0.1 mg/mL DNase I (Roche), and 2% FBS at 37°C for 30 min with gentle agitation. The digested tissue was filtered through a 70-μm cell strainer to obtain a single-cell suspension, followed by red blood cell lysis using ACK lysis buffer (Gibco).

T cell isolation and culture

T cells were enriched from the single-cell suspension using magnetic-activated cell sorting (MACS). Briefly, cells were incubated with anti-CD3 microbeads (Miltenyi Biotec) for 15 min at 4°C, followed by positive selection using an LS column (Miltenyi Biotec) according to the manufacturer’s protocol.

Isolated T cells were resuspended in complete T cell medium consisting of RPMI-1640 supplemented with 10% FBS, 1% penicillin-streptomycin (Gibco), 50 μM β-mercaptoethanol (Sigma-Aldrich), 10 mM HEPES (Gibco), 1 mM sodium pyruvate (Gibco), and 20 U/mL recombinant murine IL-2 (PeproTech). Cells were cultured at a density of 1 × 10⁶ cells/mL in 24-well plates pre-coated with 1 μg/mL anti-CD3 (BioLegend) and 2 μg/mL anti-CD28 (BioLegend) antibodies to stimulate T cell activation. Cultures were maintained at 37°C in a humidified incubator with 5% CO₂, and medium was replenished every 2–3 days. All primary cells were routinely tested for mycoplasma contamination by PCR with negative results.

Macrophage isolation and culture

Tumor-associated macrophages (TAMs) were enriched using a two-step protocol. First, myeloid cells were isolated by incubating the single-cell suspension with anti-CD11b microbeads (Miltenyi Biotec) for 15 min at 4°C, followed by positive selection using an LS column (Miltenyi Biotec) according to the manufacturer’s protocol. To further purify macrophages, CD11b+ cells were stained with anti-F4/80 (BioLegend) and sorted by fluorescence-activated cell sorting (FACS) on a BD FACSAria III.

Isolated TAMs were resuspended in complete macrophage medium consisting of RPMI-1640 supplemented with 10% FBS, 1% penicillin-streptomycin (Gibco), 50 μM β-mercaptoethanol (Sigma-Aldrich), 10 mM HEPES (Gibco), and 20 ng/mL recombinant murine M-CSF (PeproTech). Cells were plated at a density of 5 × 10⁵ cells/mL in 24-well plates and cultured at 37°C in a humidified incubator with 5% CO₂. Non-adherent cells were removed after 2 h by gentle washing with warm PBS, and adherent macrophages were cultured for an additional 24–48 h before functional assays. All primary cells were routinely tested for mycoplasma contamination by PCR with negative results.

B cell isolation and culture

B cells were then positively selected by incubating the cells with anti-CD19 microbeads (Miltenyi Biotec) for 15 min at 4°C, followed by magnetic-activated cell sorting (MACS) using an LS column (Miltenyi Biotec).

Isolated B cells were resuspended in complete B cell medium consisting of RPMI-1640 supplemented with 10% FBS, 1% penicillin-streptomycin (Gibco), 50 μM β-mercaptoethanol (Sigma-Aldrich), 10 mM HEPES (Gibco), and 10 ng/mL recombinant murine IL-4 (PeproTech). For activation, B cells were plated at a density of 1 × 10⁶ cells/mL in 24-well plates pre-coated with 10 μg/mL anti-IgM F(ab')₂ and cultured at 37°C in a humidified incubator with 5% CO₂. Cultures were supplemented with 1 μg/mL CpG ODN 1826 (InvivoGen) to enhance B cell activation and proliferation. All primary cells were routinely tested for mycoplasma contamination by PCR with negative results.

NK cell isolation and culture

NK cells were isolated from the enriched immune cell population using a mouse NK cell isolation kit (Miltenyi Biotec) according to the manufacturer’s instructions. Briefly, non-NK cells were labeled with a biotinylated antibody cocktail and subsequently depleted using magnetic bead-conjugated anti-biotin antibodies. Isolated NK cells were cultured in complete RPMI-1640 medium supplemented with 10% FBS, 1% penicillin-streptomycin, 1 mM sodium pyruvate, and 100 IU/mL recombinant mouse IL-2. All primary cells were routinely tested for mycoplasma contamination by PCR with negative results.

NK cell sorting and culture from human tissues

Tissue was dissociated into single-cell suspensions using a tissue dissociation kit (Bio-leader Incorporation Co., Lto) according to the manufacturer’s instructions. Mononuclear cells were isolated from the suspension via gradient centrifugation with Percoll (Gibco). Human natural killer (NK) cells were then purified using immunomagnetic beads specific for CD3 and CD56 (Miltenyi Biotec, Germany). These NK cells were cultured in NK cell medium with 500 U/mL of recombinant interleukin-2 (rIL-2) (Peprotech), achieving a concentration of 1 × 10⁶ cells/mL. All primary cells were routinely tested for mycoplasma contamination by PCR with negative results.

NK cell sorting and culture from human peripheral blood

PBMCs were isolated using density gradient centrifugation. Briefly, blood was diluted 1:1 with PBS and carefully layered over Lymphoprep or Ficoll-Paque PLUS in a sterile centrifuge tube. The sample was centrifuged at 400 × g for 30 min at room temperature with the brake turned off. The PBMC layer at the interface was collected and washed twice with PBS supplemented with 2% FBS. NK cells were isolated from PBMCs using a negative selection kit (Miltenyi Biotec) according to the manufacturer’s instructions. Non-NK cells were labeled with a biotinylated antibody cocktail and subsequently depleted using magnetic bead-conjugated anti-biotin antibodies. typically achieving >95% purity.Isolated NK cells were cultured in complete RPMI-1640 medium supplemented with 10% FBS, 1% penicillin-streptomycin, 1 mM sodium pyruvate, and 100 IU/mL recombinant human IL-2. Cells were seeded at a density of 1 × 10⁶ cells/mL in a humidified incubator at 37°C with 5% CO₂. All primary cells were routinely tested for mycoplasma contamination by PCR with negative results.

Arginine concentration and uptake assay

AGS cells (5 × 105) were grown in 6-well plates with amino acid-free DMEM (Guangzhou Bost Biotechnology Co., Ltd., Guangzhou, China). After the addition of arginine, the cells were incubated for 24 h. Arginine concentration in the culture medium was assessed using ELISA. The arginine uptake ratio was calculated as follows: 1- (concentration of arginine in the final medium sample - concentration of arginine secreted by the cells)/concentration of arginine added initially.

Single-cell transcriptome sequencing

scRNA sample collection, library preparation, and sequencing

Single-cell RNA sequencing was performed by Genedenovo Biotech Co., Ltd. (Guangzhou, China) utilizing the Illumina platform. Tumors from AQP5+ and AQP5- MFC cells were transplanted into C57B6/J mice and excised after 28 days for analysis. The tumors were cut into 1 mm³ pieces and dissociated enzymatically using the MACS Tumor Dissociation Kit (Miltenyi, #130-095-929). The resulting cell suspension was filtered through a 70 μm cell filter and centrifuged. Following the removal of the supernatant and lysis of red blood cells, transcriptome amplification was carried out using the SMART-seq2 protocol. Library preparation was completed using a Chromium Controller and the 10× Chromium Next GEM Single Cell 3′ v3.1 protocol. The cell suspension was mixed with the master mix and loaded onto the Chromium Next GEM Chip G, along with the 3' v3.1 gel beads and dispensing oil. In each droplet, RNA transcripts were uniquely barcoded and reverse-transcribed. The resulting cDNA was pooled, end-repaired, an “A” base was added, and adapters were ligated. The final products were purified and amplified by PCR to create the cDNA library, which was sequenced on the Illumina HiSeq platform, following the read length specifications in the user guide.

scRNA-seq data processing

Raw sequencing data from single-cell RNA sequencing (scRNA-seq) was processed with Cell Ranger 3.0.2 (10x Genomics). The scRNA-seq data was screened to generate a gene expression matrix using DNBelab C series scRNA analysis software (MGI). The reference genome (Ensemble assembly: Sscrofal1.1) was downloaded for this analysis. Cells were preserved based on the following criteria: detection of over 200 and under 5,000 genes, with mitochondrial gene expression (MT%) limited to below 30%. Following the generation of the unique molecular identifier (UMI) count profile, quality control and subsequent analyses were conducted using Seurat 3.0. Gene expression for each cell was normalized with the “LogNormalize” method, which scales total expression by a factor of 10,000 and applies a logarithmic transformation. For data alignment, 1,000 highly variable genes were selected from each matrix, and the ‘FindIntegrationAnchors' and ‘IntegrateData' functions in Seurat were implemented. Clustering was achieved with the “FindClusters” function to identify sub-cell type clusters. t-SNE visualizations depicted the clusters from datasets originating from two donors. The “FindAllMarkers” function in Seurat was applied to identify differentially expressed marker genes in each cluster, utilizing the Wilcoxon rank-sum test for statistical analysis.

Flow cytometry

Tissues or adherent cells were converted into single-cell suspensions and stained with the antibodies listed in key resources table. Analysis was performed using a CytoFLEX Flow Cytometer, and the resulting data were evaluated with FlowJo software.

Tissue microarrays construction

Tissue microarrays were constructed from gastric cancer and adjacent normal tissues (Shanghai Biochip Co., Ltd., Shanghai, China). The arrays were stained for AQP5, EPCAM, CD3, CD56, and ASS1. Two pathologists independently evaluated the staining intensity and the level of protein expression across the sections.

qPCR

Total RNA was extracted and purified with RNA Easy Isolation Reagent (Vazyme) following the manufacturer’s protocol. One microgram of total RNA was then reverse transcribed using the PrimeScript RT Kit (Vazyme). Real-time quantitative RT-PCR (qPCR) was conducted with the SYBR qPCR Mastermix Kit (Vazyme). Primer sequences are listed in Table S1, with β-Tubulin used as the normalization reference gene.

Targeted metabolomics

UHPLC-MS/MS high-throughput targeted metabolomics of cell supernatants

Metabolite extraction from cell supernatants

The samples were thawed in an ice-water bath and vortexed for 30 s. A 50 μL aliquot from each sample was transferred to an Eppendorf tube. Following this, 200 μL of pre-cooled extraction solution (acetonitrile-methanol, 1:1, with isotopically labeled internal standards at −40°C) was added, and the mixture was vortexed for an additional 30 s before sonication for 15 min in the ice-water bath. After incubation at −40°C for 1 h, the samples were centrifuged at 12,000 rpm (RCF = 13,800 × g, R = 8.6 cm) for 15 min at 4°C. A 100 μL aliquot of the resulting clear supernatant was then transferred to an auto-sampler vial for UHPLC-MS/MS analysis.

UHPLC-MS/MS high-throughput targeted metabolomics of cell sediment samples

Metabolites extraction from cell sediment

An extraction solution (1,000 μL) composed of pre-cooled acetonitrile-methanol-water (2:2:1) with an isotopically labeled internal standard was added to the sample. The mixture was vortexed for 30 s and then chilled in dry ice. A freeze-thaw cycle was performed three times using liquid nitrogen, followed by an additional 30-s vortex and 15 min of sonication in an ice-water bath. The samples were then incubated at −40°C for 1 h. Centrifugation was carried out at 12,000 rpm (RCF = 13,800 × g, R = 8.6 cm) and 4°C for 15 min. Finally, a 100 μL aliquot of the clear supernatant was transferred to an auto-sampler vial for UHPLC-MS/MS analysis.

UHPLC-MRM-MS analysis

UHPLC separation was achieved using an Agilent 1290 Infinity II system equipped with a Waters ACQUITY UPLC BEH Amide column (100 × 2.1 mm, 1.7 μm). Mobile phase A comprised 1% formic acid in water, and mobile phase B was 1% formic acid in acetonitrile. The column temperature was maintained at 35°C, with the auto-sampler set to 4°C and an injection volume of 1 μL. An Agilent 6460 triple quadrupole mass spectrometer, outfitted with an AJS electrospray ionization interface, facilitated assay development. Key ion source parameters included capillary voltages of +4000/-3500 V, nozzle voltages of +500/-500 V, nitrogen gas temperature of 300°C, gas flow rate of 5 L/min, sheath gas temperature of 250°C, sheath gas flow rate of 11 L/min, and nebulizer pressure of 45 psi. MRM parameters were optimized via flow injection analysis using standard solutions of the individual analytes introduced into the mass spectrometer’s API source. The most sensitive transitions were determined during MRM scans, allowing for the optimization of collision energy for each Q1/Q3 pair. Q1/Q3 pairs exhibiting the highest sensitivity and selectivity were designated as “quantifiers” for quantitative analysis, while additional transitions served as “qualifiers” to confirm the identity of the target analytes. Data acquisition and processing were performed using Agilent MassHunter Work Station Software (B.08.00).

Untargeted metabolomics

Metabolite extraction for untargeted analysis

Approximately 107 cultured cells were harvested, followed by the addition of 800 μL of cold methanol/acetonitrile (1:1, v/v) to precipitate proteins and extract metabolites. The resulting mixture was transferred to a centrifuge tube and centrifuged at 14,000 × g for 5 min at 4°C. The supernatant was carefully collected, and the remaining solution was evaporated using a vacuum centrifuge at 4°C. For liquid chromatography-mass spectrometry (LC-MS) analysis, the dried samples were reconstituted in 100 μL of acetonitrile/water (1:1, v/v) and placed into LC vials.

LC-MS analysis

Polar metabolites were analyzed in untargeted metabolomics using a Sciex TripleTOF 6600 quadrupole time-of-flight mass spectrometer, coupled with hydrophilic interaction chromatography via electrospray ionization (ESI). Liquid chromatography was performed on an ACQUITY UPLC BEH Amide column (2.1 mm × 100 mm, 1.7 μm; Waters, Ireland) employing a gradient of solvent A (25 mM ammonium acetate and 25 mM ammonium hydroxide in water) and solvent B (acetonitrile). The mass spectrometer operated in both negative and positive ionization modes, utilizing information-dependent acquisition for product ion scans in high-sensitivity mode. Parameters included fixed collision energy at 35V ± 15 eV, declustering potentials of 60 V for positive and −60 V for negative modes, exclusion of isotopes within 4 Da, and monitoring of 10 candidate ions per cycle.

Cell proliferation assay

For cell proliferation assays, 2,000 cells were plated in each well of 96-well plates. Daily assessments of cell viability were conducted over four days using the Cell Counting Kit-8 (GLPBIO), following the manufacturer’s guidelines. Absorbance readings at OD450 were recorded to create cell growth curves.

Additionally, cell proliferation was analyzed using carboxyfluorescein succinimidyl ester (CFSE) with the CellTrace CFSE Cell Proliferation Kit (Thermo Fisher). Sorted lymphocytes were resuspended in PBS supplemented with 5% FBS at 1 × 106 cells/mL and incubated with 1.25 μM CFSE for 5 min at room temperature. After thorough washing, the stained cells were cultured in medium for 6 to 7 days. Cell division was quantified by measuring the reduction in CFSE fluorescence using flow cytometry.

Immunochemistry

Human tissues embedded in paraffin were deparaffinized using xylene, and antigen retrieval was achieved by boiling the samples for 20 min. The tissues were then incubated with primary antibodies overnight at 4°C. Afterward, secondary antibodies were applied for 30 min at room temperature, and staining was finalized with an EnVision-HRP kit.

Differential protein display analysis

Cells were lysed with WB&IP buffer (Beyotime) containing protease inhibitors (Boster Biological Technology). A total of 20 μg of protein from each sample was combined with 5× loading buffer and heated for 5 min. Proteomic analysis was performed using iTRAQ (Shanghai OE Biotech Co., Ltd). Differential expression was defined by a p-value of <0.05.

Co-immunoprecipitation and western blotting

Cells were lysed with WB&IP buffer (Beyotime) containing protease inhibitors (Boster Biological Technology). The resulting lysates were subjected to overnight immunoprecipitation at 4°C with designated primary antibodies, followed by either protein A/G precipitation for 2 h or incubation with magnetic beads linked to tagged antibodies (Biomake). The beads were washed three times with lysis buffer and eluted in SDS sample buffer. The eluted immune complexes were separated via SDS-PAGE and analyzed by Western blotting.

Equal protein amounts were loaded onto 7.5%, 10%, or 12.5% SDS-polyacrylamide gels, which were then transferred to nitrocellulose membranes (Millipore). The membranes were incubated overnight with specific primary antibodies. Following three washes with 1× TBST buffer, secondary antibodies were applied for 1 h at room temperature. Signal detection was performed using Luminata Crescendo western horseradish peroxidase substrate (Vazyme).

Immunofluorescence

A total of 1 × 105 cells were cultured in a 24-well plate with pre-inserted slides. After 48 h, the culture medium was removed. The cells were fixed in 4% paraformaldehyde and permeabilized with 0.5% Triton X-100. Blocking was achieved with 5% BSA. Primary antibodies were added and incubated overnight at 4°C with gentle agitation. Following this, a fluorescent secondary antibody, diluted in the blocking solution, was applied in the dark for 1 h at room temperature. Nuclei were stained with DAPI for 5 min, and the slides were mounted using fluorescent mounting medium. An upright fluorescence microscope was used to observe and assess the positive cell staining index.

ChIP assay

A ChIP assay was performed on AGS cells using the ChIP kit (Abcam, #ab117137) following the manufacturer’s protocol. The promoter region of the ASS1 gene, which contains the DHX9 regulatory element, was immunoprecipitated with an anti-DHX9 antibody, using rabbit IgG as a control. The resulting purified DNA fragments underwent real-time PCR with primers specific to the ASS1 promoter. The ChIP signal from the anti-DHX9 antibody was normalized to the IgG control signal to determine the fold change. Details of the primers used are provided in Table S1.

Luciferase reporter assays

The pGL3-ASS1-promoter recombinant luciferase reporter vector and pRL-TK plasmid were co-transfected into AGS cells to perform a dual luciferase reporter assay, with pRL-TK serving as an internal control. After 48 h, the cells were harvested, and relative light units were quantified using a Promega luciferase assay kit. Luciferase activity was calculated as the ratio of firefly to Renilla luciferase fluorescence intensity.

Protein profiling

Cells were lysed with WB&IP buffer (Beyotime) containing protease inhibitors (Boster Biological Technology). A total of 20 μg of protein from each sample was combined with 5× loading buffer and heated for 5 min. Proteins were separated via 12.5% SDS-PAGE at a constant current of 14 mA for 90 min. Visualization of protein bands was achieved using Coomassie Blue R-250 staining. Trypsin digestion was performed on the proteins. The resulting peptides were desalted using C18 cartridges (Empore SPE Cartridges, standard density, bed I.D. 7 mm, 3 mL volume; Sigma), concentrated via vacuum centrifugation, and reconstituted in 40 μL of 0.1% (v/v) formic acid. LC-MS/MS analysis was performed with a Q Exactive mass spectrometer (Thermo Scientific). The raw MS data for each sample were compiled and analyzed for identification and quantification using MaxQuant software (FDR <1%, ≥2 unique peptides).

Cell migration assay

Assays were performed using 24-well Transwell plates with 8-μm polyethylene terephthalate membrane filters (Falcon, Becton-Dickinson). In the upper chamber, 105 cells were seeded in serum-free DMEM, while the lower chamber contained DMEM with 15% FBS. Cells were permitted to migrate for 18 h. Migrated cells on the lower filter surface were fixed in 4% paraformaldehyde, stained with 1% crystal violet, and counted in three randomly chosen fields via microscopy.

NK cell cytotoxicity assay

The CytoTox 96 Non-Radioactive Cytotoxicity Assay Kit was utilized to assess the cytotoxic activity of NK cells. In this assay, target cells (1.5×10⁴ cells/well) were co-cultured with NK cells (1.5×10⁵ cells/well) for 4 h at 37°C in a 5% CO2 environment, with three replicates per condition. Following incubation, the culture supernatant was harvested, and cytotoxicity was quantified using the formula outlined in the kit’s instructions. Cytotoxicity = [D(490) experimental group - D(490) effector cell spontaneous group - D(490) target cell spontaneous group]/[D(490) target cell maximal group - D(490) target cell spontaneous group) x 100%.

ATP measurement

ATP levels were quantified using the ATP Determination Kit, following the manufacturer’s guidelines.

Arginine/glutamine enzyme-linked immunosorbent assay

Arginine and glutamine concentrations in mouse or human tissues and cellular lysates were determined using the L-arginine/Glutamine kit following the manufacturer’s guidelines.

NO activity assay

Cells were collected and resuspended in diluted DAF-FM DA at a concentration of 107cells/mL, followed by a 20-min incubation at 37°C in a cell culture incubator. This process was carried out in a centrifuge tube, with the tube inverted every 3 to 5 min to maximize interaction between the probe and the cells. Subsequently, the cells were washed three times with PBS (pH 7.4) to eliminate any unincorporated DAF-FM DA. Fluorescence intensity was then assessed using a microplate reader.

Oxygen consumption rate (OCR) assay

Cells were plated in an XF96 cell culture microplate at a density of 104 cells per well, with three replicates. On the same day, 180 μL of hydration solution from the XF96 Extracellular Flux Assay Kits was added to the lower compartment, and the cells were hydrated overnight at 37°C in a CO2-free incubator. The Seahorse XF Basic Medium was prepared, with its pH adjusted to 7.4. The following day, cells were washed twice with Seahorse XF Basic Medium, and 175 μL of the medium was added to each well, followed by a 1-h incubation at 37°C in a CO2-free environment. Drugs, including Oligomycin, DMSO, FCCP, DMSO, Antimycin A, and Rotenone, were diluted to the desired concentrations and introduced into the upper layer of the XF96 Extracellular Flux Assay Kits, with 25 μL added to each well. After 30 min, the XF96 kits were removed, and the lower layer was substituted with a cell plate that had been incubated in a 37°C CO2-free incubator for 1 h before analysis.

Quantification and statistical analysis

Statistical analysis was carried out using SPSS 19.0 software and GraphPad Prism 5. Two-tailed and unpaired Student’s t tests were used for two group comparisons. Spearman’s correlation test was performed to analyze the correlation of two genes. Survival curves were estimated using the Kaplan-Meier method and compared using the log rank test. Data are shown as the mean ± SD. p < 0.05 was considered statistically significant.

Published: September 16, 2025

Supplemental information

Table S2. Clinical and molecular characteristics of gastric cancer patients stratified by AQP5 and NK cell status

(A–J) Patient demographics (gender and age), tumor morphology (diameter, grade, and TNM stage), and survival data (alive/dead and survival months).

(K–L) Group classification based on AQP5 expression (high/low) and NK cell activity (high/low), with corresponding ASS1 staining scores.

Table S3. Differential metabolite profiles in AQP5⁺ versus AQP5⁻ gastric cancer cells identified by untargeted metabolomics

(A) Metabolite name.

(B–D) Metabolite abundances (peak intensities) in AQP5⁻ cells (triplicate samples: AQP5⁻1, AQP5⁻2, and AQP5⁻3).

(E–G) Metabolite abundances in AQP5⁺ cells (triplicate samples: AQP5⁺1, AQP5⁺2, and AQP5⁺3).

Table S4. Differential amino acid profiles in AQP5⁺ versus AQP5⁻ gastric cancer cells quantified by targeted metabolomics

(A) Metabolite name.

(B–Z) Metabolite abundances (peak intensities) in AQP5⁻ cells (triplicate samples: AQP5⁻1 to AQP5⁻5) and AQP5⁺ cells (triplicate samples: AQP5⁺1 to AQP5⁺5).

Table S5. Differential amino acid secretion profiles in culture medium of AQP5⁺ versus AQP5⁻ gastric cancer cells

(A) Metabolite name.

(B–Z) Extracellular concentrations of 22 amino acids in conditioned medium from AQP5⁻ cells (triplicate samples: AQP5⁻1 to AQP5⁻3) and AQP5⁺ cells (triplicate samples: AQP5⁺1 to AQP5⁺3).

Table S6. Tumor tissue amino acid profiles stratified by AQP5 expression levels in gastric cancer patients

(A) Metabolite name.

(B–Z) Absolute concentrations of 22 amino acids in tumor tissues from patients with low (AQP5⁻), intermediate (AQP5^±), and high (AQP5⁺, AQP5⁺⁺) AQP5 expression (5–6 replicates per group).

Table S7. Differentially expressed protein between exogenous overexpression of AQP5 and control group

(A and B) Accession and gene names: Each protein is identified by its UniProt accession number and gene name.

(C) Descriptions: detailed protein names and functions are provided.

(D–K) Technical metrics validate protein detection.

(L–N) Indicate the expression difference between AQP5-overexpressing and control groups.

(O–AF) Functional annotations.

Table S8. Potential proteins interacted with ASS1 promoter identified using Reverse-ChIP analyses

(A) prot_hit_num: unique identifier for each protein hit.

(B) prot_acc: protein accession number.

(C) prot_desc: description of the protein, including its name, organism, gene name, and other details.

(D) prot_score: score indicating the likelihood or strength of interaction.

(E) prot_mass: molecular mass of the protein in daltons.

(F) prot_matches: number of matches for the protein.

(G) prot_matches_sig: number of significant matches.

(H) prot_sequences: number of sequences associated with the protein.

(I) prot_sequences_sig: number of significant sequences.

(J) prot_cover: coverage percentage of the protein.

(K) prot_pi: isoelectric point (pI) of the protein.

(L) emPAI: emPAI value, which estimates protein abundance.

Table S9. Differential proteins that potentially interacted with DHX9 protein in exogenous overexpression of AQP5 and control group

(A) Unique protein identifier used to distinguish different proteins.

(B) Common or descriptive name of the protein.

(C) Name of the gene encoding the protein.

(D) Biological function or pathway involvement of the protein.

(E) Protein group assignment, which may include homologous proteins or isoforms sharing peptides.

(F) Original description line from the FASTA file.

(G) All proteins associated with the protein group.

(H) Total number of peptides identified.

(I) Number of “razor” unique peptides assigned to this protein group.

(J) Number of peptides uniquely mapping to this protein.

(K) Sequence coverage percentage.

(L) Molecular weight of the protein.

(M) Protein abundance/intensity in the experimental group.

(N) Protein abundance/intensity in the control group.

(O) Indicates whether the protein was only identified or also quantified.

Table S10. Differential metabolites in NK cells identified by untargeted metabolomics analyses

(A) Unique identifier of the metabolite.

(B) Mass-to-charge ratio, used in mass spectrometry analysis.

(C) Retention time (seconds), the retention time in chromatographic analysis.

(D) Chemical name of the metabolite.

(E) Adduct form.

(F) Human Metabolome Database accession number.

(G) KEGG database accession number.

(H) Superclass classification of the metabolite.

(I) Class of the metabolite.

(J) Subclass of the metabolite.

(K–M) Metabolite abundance values in the three PBS-treated groups.

(N–P) Metabolite abundance values in the three arginine-treated groups.

(Q–S) Metabolite abundance values in the three quality control samples.

Table S11. Differential center carbon metabolites in AQP5⁺ and AQP5⁻ gastric cancer cells identified by targeted metabolomics analyses

(A) Metabolite name.

(B–Z) Metabolite abundances (peak intensities) in AQP5⁻ cells (triplicate samples: AQP5⁻1 to AQP5⁻5) and AQP5⁺ cells (triplicate samples: AQP5⁺1 to AQP5⁺5).

Table S12. Intracellular fluxes in AQP5⁺ and AQP5⁻gastric cancer cells quantified using U-13C6 glutamate tracer

(A) Chemical name of the metabolite.

(B) Molecular formula of the metabolite.

(C) Isotopic labeling state.

(D–H) Relative abundance of isotopically labeled metabolites in AQP5⁺ gastric cancer cells (5 biological replicates).

(I–M) Relative abundance of isotopically labeled metabolites in AQP5⁻ gastric cancer cells (5 biological replicates).

Table S13. Potential proteins interacted with GLUL

(A) Unique protein identifier.

(B) Full name of the protein.

(C) Name of the gene encoding the protein.

(D) Functional annotation of the protein.

(E) Protein group to which this protein belongs, listing multiple related protein identifiers.

(F) FASTA header information, including source organism, gene name, and other metadata.

(G) Number of proteins in the group.

(H) Total number of identified peptides.

(I) Number of “razor” peptides uniquely assigned to this protein group.

(J) Number of peptides uniquely assigned to this protein.

(K) Protein sequence coverage (%).

(L) Molecular weight of the protein (kDa).

(M–O) Label-free quantification (LFQ) intensity value in the AGS1-3 sample.