CCL8-dependent recruitment of natural killer cells enhances the antitumor activity of neoadjuvant chemotherapy in gastric cancer.

Introduction
Gastric cancer (GC) is one of the most common malignancies of the digestive system, ranking fifth in global cancer incidence and mortality [1]. Based on robust evidence from large-scale randomized controlled trials, neoadjuvant chemotherapy (NACT) has been established as the standard of care for patients with locally advanced GC in both Western and Asian populations. In East Asia, the SOX regimen, comprising oxaliplatin and S-1 (tegafur), is the most widely recommended protocol, holding a Grade I recommendation in treatment guidelines across China, Japan, and South Korea [2–6]. NACT improves the likelihood of radical surgical resection, eliminates occult micrometastases, enables in vivo evaluation of treatment responsiveness, and has been shown to improve survival outcomes in GC patients [7–10]. Nonetheless, a subset of patients derives limited benefit from NACT, highlighting the urgent need to explore combination strategies and identify predictive immunologic mechanisms of therapeutic response.
Traditionally, chemotherapeutic agents have been understood to kill tumor cells by inducing non-immunogenic forms of programmed cell death. However, emerging evidence suggests that certain cytotoxic drugs can also induce immunogenic cell death (ICD), enhance tumor immunogenicity, and modulate the tumor immune microenvironment (TIME) through direct interactions with immune cell subsets [11, 12]. These immunomodulatory properties may underlie the synergistic effects observed when chemotherapy is combined with immune checkpoint inhibitors. For instance, NACT has been reported to increase CD8⁺ T cell infiltration and reduce the abundance of macrophages and regulatory T cells (Tregs) in GC patients, thereby enhancing antitumor immunity [13–16]. Moreover, individual variation in the composition of immune cells and the cytokine milieu within the TIME is increasingly recognized as a key determinant of response and prognosis [17–20]. As the dynamic crosstalk between tumor cells and immune components plays a critical role in shaping chemotherapy response [18, 21, 22], elucidating the immunological consequences of NACT is crucial for identifying biomarkers and designing rational combination regimens. Yet, the specific impact of NACT on NK cells within the TIME and the molecular mechanisms involved remain poorly understood.
In this study, we investigated the immunomodulatory effects of NACT in GC, with a particular focus on NK cell dynamics. By analyzing paired tumor biopsies before and after NACT, as well as using murine tumor models and in vitro assays, we identified significant alterations in NK cell infiltration and uncovered a mechanistic link involving tumor-derived CCL8. Specifically, we demonstrate that NACT activates the mitogen-activated protein kinase (MAPK) pathway in tumor cells, promoting the secretion of CCL8, which subsequently drives NK cell recruitment. This mechanism was further validated using in vivo neutralization and adoptive NK cell transfer experiments. Our findings provide mechanistic insight into how NACT reshapes the immune landscape of GC and establish a rationale for combining chemotherapy with NK cell-targeted immunotherapies.

Results

NACT enhances the infiltration of NK cells in GC patients
A total of 33 patients were included in the study, and the clinicopathological characteristics are summarized in Table S3. As shown in the table, 78.8% of patients had clinical stage II or III, and 15.2% of patients had HER2-positive disease. The predominant NACT regimen administered to patients was SOX (66.7%), and approximately half of the patients responded to treatment, with 42.4% displaying tumor regression grade 0 or 1.
To evaluate the impact of NACT on the TIME of GC patients, immunohistochemistry and immunofluorescence were utilized to analyze 55 specimens, consisting of 22 matched pre-NACT biopsy and post-NACT surgical samples. Compared with biopsy specimens, the abundance of CD4+ T cells, CD8+ T cells, NK cells, B cells, macrophages, and dendritic cells (DCs) increased after NACT (Fig. 1A), and the SOX regimen showed a similar trend (Figure S1A). Analysis of immune marker expression in 22 paired samples revealed that the changing trends observed in the majority of patients were consistent with those in the entire cohort. However, there was heterogeneity among these patients, which could potentially contribute to treatment divergence (Figure S1B).
We next assessed the relationship between the number of immune cells and the efficacy of NACT by a Chi-square test. Increased CD8+ T cells and NK cells following NACT were correlated with a more favorable clinical response (Fig. 1B). However, the expression levels of other markers did not exhibit a significant correlation with clinical response (Figure S2). These results indicate that NACT promoted the infiltration of antitumor immune cells, and the increase in CD8+ T cells and NK cells following NACT was associated with a more favorable clinical response.

NACT promotes NK cell infiltration and TIME remodeling in tumor-bearing mice
The SOX regimen is a grade I recommended NACT regimen in China and is also the most widely used regimen in the patient cohort of our study. We further constructed a subcutaneous tumor model in C57BL/6 mice to investigate the effect of the SOX regimen on the TIME of GC (Fig. 2A). As expected, the SOX regimen effectively suppressed tumor growth, evidenced by reduced tumor weight and volume compared to the control group (Fig. 2B, C, D). We observed increased proportions of CD8+ T cells, B cells, DCs, and NK cells, while the proportion of inhibitory subset Treg cells and neutrophils decreased significantly. Conversely, no significant changes in CD4+ T cells, macrophages, myeloid-derived suppressor cells (MDSCs), or DC subsets (cDC1 and cDC2) were shown (Fig. 2E).
Besides, we explored the effect of the SOX regimen on peripheral immunity in tumor-bearing mice, revealing an increased proportion of CD4+ T cells, CD8+ T cells, and cDC1 and a decreased proportion of macrophages, neutrophils, and polymorphonuclear MDSCs (pmn-MDSCs; Figure S3A, B). Overall, these findings indicate that the SOX regimen attenuates tumor growth and reshapes the TIME of GC in mouse subcutaneous tumors by promoting the infiltrating proportion of antitumor immune cells.

NK cells mediate the antitumor efficacy of the SOX regimen
Transcriptome sequencing of mouse tumor tissues was then performed to clarify the precise mechanism of the SOX regimen in modulating the TIME. KEGG enrichment analysis revealed that the upregulated genes were primarily clustered in the cytokine interaction signaling pathways, indicating an augmented immune response following NACT (Fig. 3A). After deconvolution analysis of these differential genes, we discovered that the immune microenvironment was altered and NK cells were increased post-treatment (Figure S4A, B), consistent with the results of flow cytometry. Moreover, chemokines associated with NK cell recruitment (CCL2, CCL7, CCL8, and CCL11) were also upregulated (Fig. 3B). These findings suggested that NK cells might play a crucial role in the antitumor effect of the SOX regimen.
The aforementioned findings suggest that in GC patients and GC-bearing mice, NACT promotes the recruitment of NK cells, and increased CD8+ T cells and NK cells following NACT were associated with a more favorable clinical response. Previous research has largely focused on T cells following NACT, whereas the alterations in NK cells and the underlying mechanisms have been less frequently explored. Next, we applied C57BL/6 mice, Rag1−/− mice (T and B cell-deficient), and NCG mice (T, B, and NK cell-deficient) to validate the role of NK cells in the antitumor process of NACT. As shown in Fig. 3C–E, compared with the group of Rag1−/− mice or C57BL/6 mice treated with the SOX regimen, the tumors in NCG mice exhibited significantly greater size. These observations imply that NK cells are closely related to the antitumor effects elicited by the SOX regimen.

Tumor cells mediate NK cell recruitment during NACT
Studies have shown that malignant cells can release danger-associated molecular patterns (DAMPs), pro-inflammatory cytokines, and inflammatory mediators to activate and recruit immune cells, ultimately improving the antitumor efficacy [23–25]. To explore the mechanisms underlying the recruitment of NK cells by NACT and the role of tumor cells in this process, we established a cell-derived xenograft (CDX) model in NCG mice using the MGC803-18.2 cell line, thereby eliminating the influence of immune cells (Fig. 4A). Mice were randomly allocated to receive either SOX, CLDN18.2-CAR NK, or their combination therapy on day 26. As a result, the combination therapy highlighted superior efficacy in inhibiting tumor growth and reducing tumor volume, especially when CLDN18.2-CAR NK treatment was administered before SOX on the second day (Fig. 4B, C, D). Crucially, in the MGC803-18.2 GC mouse model, the combination of the SOX regimen with CAR NK further promoted the infiltration of NK cells (Fig. 4E, F), indicating that tumor cells can potentiate NK cell recruitment by NACT.
Due to the inability to replicate the in vivo conversion of S-1 to 5-FU by liver enzymes in vitro, we replaced S-1 with 5-FU in subsequent in vitro experiments. After determining the IC50 values of oxaliplatin and 5-FU (Figure S4C), we evaluated the recruitment effect of tumor cell culture supernatants on NK92 cells using the Transwell assay. Consequently, the recruitment index of NK92 cells increased in the combined treatment group, demonstrating that the SOX regimen effectively promotes the recruitment of NK cells by tumor cells in vitro (Fig. 4G, H). Additionally, major histocompatibility complex (MHC) class I chain-related protein A (MICA) and MHC class I chain-related protein B (MICB), serving as ligands for the natural killer group 2 member D (NKG2D) receptor, contribute to triggering NK cell-mediated cytotoxicity. The upregulated expression of MICA and MICB in GC cells (Figure S4D) and the heightened NK cell cytotoxicity (Figure S4E) indicate that the combination therapy boosts the sensitivity of GC cells to NK cell-mediated killing. Therefore, NACT not only facilitates NK cell recruitment by tumor cells but also enhances the susceptibility of gastric cancer cells to NK cell cytotoxicity.

NACT induces CCL8 expression to promote NK cell recruitment
We then set out to elucidate how tumor cells facilitate the recruitment of NK cells during NACT. According to the results of RNA-seq, we measured the levels of chemokines released from GC cells after combined treatment by RT-qPCR and found increased expression of CCL2, CCL7, CCL8, and CCL11(Fig. 5A). We also noticed a significant increase in the expression of NK cell recruitment-related chemokines in the treated tumor cell culture supernatant, including CCL3, CCL5, CCL7, CCL8, CCL13, and CCL16, revealed by the cytokine chip (Fig. 5B). Based on the overlapping results from cytokine chip analysis and RNA-seq, we further examined the correlation between CCL7, CCL8, and NK cells in gastric cancer. Among them, CCL8, a chemokine that can recruit various types of immune cell populations [26, 27], showed a significant correlation with NKG2D in GC (Figure S5A).
To further elucidate the role of CCL8 in promoting NK cell recruitment, we employed specific antibodies to neutralize CCL8 in the tumor cell culture supernatants. The recruitment index of NK92 cells by combined treatment decreased upon CCL8 neutralization (Fig. 5C). In vivo, the proportion of NK cells in tumor tissue decreased upon CCL8 blockade (Fig. 5D), accompanied by weakened tumor control in the SOX and CCL8 block group (Fig. 5E, F). These results highlight the role of NACT in boosting CCL8 secretion by GC cells to facilitate the recruitment of NK cells.

MAPK activation drives CCL8 expression in response to NACT
To decode the mechanism by which the SOX regimen promotes the secretion of CCL8 in GC cells, we conducted GSEA using the above transcriptome sequencing data from mice. The analysis revealed activation of the MAPK signaling pathway following treatment with the SOX regimen (Figure S5B), and the increased phosphorylation levels of the MAPK signaling pathway in GC cells were further determined by western blot (Fig. 6A). In addition, decreased CCL8 expression was observed at both protein (Fig. 6B) and mRNA levels (Fig. 6C) after treatment with specific inhibitors targeting the MAPK p38 signaling pathway, respectively. These results collectively demonstrate that NACT promotes the secretion of CCL8 from GC cells by activating the MAPK signaling pathway, thereby promoting the recruitment of NK cells (Fig. 6D).

Combination therapy with NACT and NK cell infusion achieves clinical benefit
An adult patient, who developed poorly differentiated gastric adenocarcinoma with moderate to large amounts of ascites, received treatment with 5-FU and oxaliplatin combined with NK cell transfer. The chemotherapy regimen involved administering 5-FU at a dose of 180 mg/m2 intravenously on days 1–3 and oxaliplatin at a dose of 85 mg/m2 intravenously on day 1, with 7.4 × 109 NK cells administered intravenously one week later (Fig. 7A). The triplet therapy was given every 3 weeks. Flow cytometry analysis showed that the combination therapy resulted in an increased number and proportion of NK cells, as well as elevated levels of IL-2R in the patient's peripheral blood, while the levels of the gastric cancer-related tumor marker CA199 were decreased (Fig. 7B). The elevation of IL-2R indicates a shift in the immune response toward the activation of NK cells, since the binding of IL-2R with IL-2 is a critical step for the activation of NK cells [28]. Moreover, the combination therapy resulted in a significant reduction in ascites volume and lesion size of GC tissue (Fig. 7C). Together, these observations suggest the potential of NACT combined with NK infusion in GC, which deserves further investigation in clinical trials.

Materials and methods

Cell culture and reagents
MGC803 (ATCC: CL-0158) and HGC27 (ATCC: CL-0107) were provided by the Cell Bank of the Chinese Academy of Sciences. MGC803 cells are identified as correct by short tandem repeat. MFC (QuiCell-M019) and NK92 (QuiCell-N262) cells were purchased from Shanghai QuiCell Biotechnology Corporation, originally sourced from Kunming Cell Bank, Chinese Academy of Sciences. The maintenance medium for MGC803, HGC27, and MFC cells comprised RPMI1640 (Basal Media) supplemented with 10% fetal bovine serum (FBS, Gibco). The complete medium for NK92 cells consisted of Alpha Minimum Essential medium (ScienCell, HX-M001) with a concentration of 0.2 mM inositol (ScienCell, HX-S004), 0.1 mM 2-mercaptoethanol (ScienCell, HX-S005), 0.02 mM folic acid (ScienCell, HX-S006), 20 ng/mL recombinant interleukin-2 (IL-2; ScienCell, HX-S007), 12.5% horse serum (ScienCell, HX-S002), 1% penicillin–streptomycin (ScienCell, HX-S003), and 12.5% fetal bovine serum (FBS; ScienCell, HX-S001). All cells were cultured at 37 °C and 5% CO2. The human CCL8 (AF-281) antibody was from R&D Systems. S-1 (H20090045) was obtained from Taiho Pharmaceutical, and oxaliplatin (M2290) was from Ambole. 5-fluorouracil (5-FU; H31020593) was from Shanghai Xudong Haipu Pharmaceutical. Adezmapimod (HY-10256), SCH772984 (HY-50846), and SP600125 (HY-12041) were acquired from MedChemExpress. The antibodies used are summarized in Table S1.

Immunohistochemistry (IHC) and immunofluorescence staining
A retrospective analysis was conducted on primary GC patients who underwent NACT in the Oncology Department of Changhai Hospital between January 2015 and June 2021. The study protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki and was approved by the Research Ethics Committee of Changhai Hospital (Committee on Ethics of Medicine, Navy Medical University, PLA, No.82072707). Written informed consent was obtained from each patient.
A total of 55 specimens were analyzed in the study, including 33 patient-derived samples, 22 of which were paired samples. Continuous 4 μm sections were made from wax blocks of GC tissues and stored in a 4 °C refrigerator. After deparaffinization, slices were subjected to high-temperature and high-pressure antigen retrieval for 2 min and 20 s in EDTA (pH = 8.0) for CD4, CD8, and CD68, or citrate buffer (pH = 6.0) for CD15, CD20, and CD56. Following cooling to room temperature, endogenous peroxidase was blocked with 3% H₂O₂ for 20 min. Sections were then incubated with goat serum for 30 min at 37 ℃ before incubation overnight at 4 ℃ with primary antibodies at the specified dilutions listed in Table S1. For immunohistochemistry, a horseradish peroxidase (HRP) secondary antibody was incubated for 30 min at 37 ℃, followed by counterstaining with 3,3'-diaminobenzidine (DAB) for 20 s, and the nuclei were stained with hematoxylin for 10 min. Meanwhile, immunofluorescence staining involved incubation using a secondary antibody conjugated to a fluorescent dye for 30 min at 37 ℃, with subsequent nuclear staining using the DNA-specific dye DAPI. Aperio ImageScope software was used to count immune cells at 40 times the field of view, and the results were independently determined by technicians blinded to the patients' clinical data.

Generation of engineered cells expressing Claudin18.2
Claudin18.2 (NM_001002026.3) was fused with a His-tag and then cloned into pLVX-EF1a-IRES-Puro (Addgene, 85,132). The lentivirus was generated by co-transfection with the packaging plasmids psPAX2 and pMD2.G into lenti-X-293 cells cultured in DMEM medium with 10% FBS. MGC803 cells were seeded into 24-well plates at 2 × 105 cells/well and then infected at a multiplicity of infection (MOI) of 10. The cells were harvested 7 days post-infection and subsequently sorted by flow cytometry.

Manufacturing of Claudin18.2-specific CAR NK (CLDN18.2-CAR NK)
NK92 cells were transduced with a lentiviral vector encoding the humanized CLDN18.2-specific hu8E5 single-chain fragment variables (scFv). Briefly, the hu8E5 scFv was combined in frame with CD8 hinge and transmembrane domain, CD137 co-stimulatory domain, and CD3ζ activation domain. A leader peptide derived from GM-CSFRα was included to facilitate chimeric antigen receptor (CAR) cell surface expression.

Mice and tumor models
The 4- to 6-week-old C57BL/6, Rag1−/− mice, and female NOD/ShiLtJGptPrkdcem26Cd52 Il2rgem26Cd22/Gpt (NCG) mice were ordered from Jiangsu Gempharmatech Corporation. Mice were fed in a pathogen-free vivarium under standard conditions. All experiments were approved by the Committee for the Ethical Use of Laboratory Animals at the Naval Medical University.
The 5 × 106 MFC cells were resuspended with phosphate buffered saline (PBS) and matrigel (Corning, 356,234) mixed at a ratio of 1:1 and then injected subcutaneously at the right flank of mice. Mice were stratified by tumor size into treatment groups to ensure a similar distribution of tumor size at baseline. S-1 was administered at 10 mg/kg every 2 days, and oxaliplatin was used at 10 mg/kg every 4 days for one week. For in vivo anti-CCL8 treatment, the anti-CCL8 antibody (BioLegend, A16070K) was intraperitoneally injected (50 μg/mL, 100 μL) once daily. In the context of CAR NK combination therapy, 5 × 106 MGC803 cells overexpressing claudin18.2 (MGC803-18.2) were injected at the right flank of NCG mice, and 1 × 106 CLDN18.2-CAR NK cells were injected intravenously once a week. Tumor size was measured every 2 days. Tumor volume was calculated as follows: (length × width2)/2. Mice were anesthetized with 5% isoflurane to induce rapid unconsciousness, followed by cervical dislocation for euthanasia.

Flow cytometry
Flow cytometry analysis was performed using the Attune NxT system (Invitrogen). To prepare a single-cell suspension, the tumor tissues from GC-bearing mice were minced and enzymatically digested using collagenase type IV (Worthington, LS004188) and DNase I (Worthington, LS002138) in RMPI1640 for 30 min at 37 ℃, passed through a 70 μm cell strainer, and centrifuged. Blood samples were lysed with ACK lysis buffer (Beyotime, C3702) for 5 min at 4 °C. Cells were stained with the fluorochrome-conjugated antibodies at the indicated dilutions listed in Table S1 for 30 min at 4 °C in the dark and washed twice with sterile PBS. All samples were further passed through a 100 μm cell strainer. Dead cells were excluded by staining with FVS-780 (eBioscience, 65–0865-14). Data were analyzed using FlowJo (version 10).

RNA sequencing
For RNA sequencing (RNA-seq), mouse GC tissues were collected in liquid nitrogen before being transferred to a − 80 °C freezer. RNA extraction and sequencing were performed by Guangzhou Gene Denovo Biotechnology Company. Gene set enrichment analysis (GSEA) was conducted to assess enriched gene sets in the RNA-seq data. The CIBERSORTx algorithm was used to predict the degree of infiltration of 22 immunocyte types within the GC tissues.

CCK8
Gastric cancer cells were cultured overnight and then exposed to fresh medium containing varying concentrations of oxaliplatin or 5-FU, ranging from 0 to 100 μg/mL (0, 0.5, 1, 2, 5, 10, 20, 50, 80, 100 μg/mL) for 48 h. Steps were performed following the instructions supplied by Beyotime (C0038). The half-maximal inhibitory concentration (IC50) values were determined by constructing dose-response curves.

Recruitment of NK92 cells
GC cells were treated with oxaliplatin and 5-FU according to the corresponding IC50 for 2 days. Then, the tumor cell culture supernatants were collected after filtering with a 0.22 μm sterile filter and stored in a − 80 °C freezer. 2 × 105 NK92 cells labeled with 5 μM CFSE (BD Horizon, AB_2869649) were added to the upper chamber of a 5-μm transwell system (Corning, 3421), with tumor cell culture supernatants in the lower chamber. After 2 h of coculture, the chambers were removed, and cells in the lower chambers were visualized and counted under a microscope. The migration index was calculated using the formula, where the number of cells migrating toward the supernatants is divided by the total number of cells. For the antibody neutralization experiment, the tumor cell culture supernatants in the lower chamber were pre-treated with CCL8 neutralizing antibody (1.5 μg/mL, 3 μg/mL) for 1 h before coculture.

Real-time quantitative polymerase chain reaction (RT-qPCR)
GC cells were exposed to oxaliplatin and 5-FU at their respective IC50 concentrations for 48 h. RNA extraction (Fastagen, 220,010) and cDNA Synthesis (Vazyme, R333-01) were performed following the kits according to the manufacturer’s instructions. cDNA was amplified by qPCR with the Hieff® qPCR SYBR Green PCR Kit (YEASEN, 11201ES08). The primers used are listed in Table S2. The target gene was calculated using Formula 2−△△Ct.

Cytotoxicity assay
GC cells were divided into 3 groups: the control sample (untreated control cells), the maximum enzyme activity control sample (untreated cells for subsequent lysis), and the drug-treated sample wells. After treating tumor cells with oxaliplatin and 5-FU for 48 h, NK92 cells were introduced and cocultured [Effector (E): Target (T) ratio 1:1, 5:1, 10:1] for 4 h. The release of lactate dehydrogenase (LDH) into the supernatants was quantified. The percentage of cytotoxicity was calculated according to the manufacturer’s protocol (Beyotime, C0016).

Cytokine chips
A Human Chemokine Antibody Array G1 (RayBiotech, AAH-CHE-G1-4) was used to assess the levels of relative chemokines in cell culture supernatants. After treatment with oxaliplatin and 5-FU in low serum medium for 48 h, the culture supernatants of GC cells were harvested and stored in a − 80 °C refrigerator for subsequent detection. The concrete experiments and analysis were performed by Guangzhou RayBiotech Company.

Western blot
After treatments with oxaliplatin and 5-FU for 48 h, GC cells were collected, and total protein was extracted. For mitogen-activated protein kinase (MAPK) inhibition, cells were initially treated with their respective inhibitors for 24 h separately (SP600125: 5 μM; Adezmapimod: 5 μM; SCH772984: 500 nM). Cells were lysed using RIPA lysis buffer (Beyotime, P10013) containing protease and phosphatase inhibitors (Beyotime, P1005) at 4 °C for 30 min. Protein concentrations in the lysates were quantified with the BCA assay kit (Beyotime, P0012S). Gels were prepared following the instructions provided in the gel rapid preparation kit (Epizyme, PG112). Equal amounts of protein were separated by SDS–PAGE, transferred to nitrocellulose membranes, and probed with the specified antibodies at 4 °C overnight. The primary antibody dilution ratio was 1:1000, except for CCL8, which was diluted at 1:500 (Table S1). The secondary antibodies were incubated at room temperature for 2 h. Images were captured using a chemiluminescence imaging system.

Statistical analysis
GraphPad Prism 7.0 was used for statistical analysis and charting. FlowJo V10 software was used to analyze flow cytometry results. The t-test was used for comparison between two groups, the pairwise t-test was used for the comparison of paired samples, and the Chi-square test was used for correlation analysis. One-way analysis of variance (ANOVA) was used for comparison among three or more groups.

Discussion
Although NACT has been widely adopted as a standard treatment for locally advanced GC, its immunomodulatory effects on the TIME remain incompletely understood. Our study demonstrates that NACT can effectively reshape the TIME by enhancing antitumor immunity, particularly via the recruitment and activation of natural killer (NK) cells. We observed increased infiltration of multiple immune effector subsets, especially CD8⁺ T cells and NK cells, after NACT in both GC patients and murine models, with their elevated presence correlating with improved clinical response. The use of matched pre- and post-NACT samples enabled us to capture individualized immune alterations, highlighting TIME heterogeneity and its potential influence on treatment outcomes.
CD8⁺ T cells have long been recognized as key antitumor effectors, and their post-NACT enrichment has been associated with improved prognosis in GC [13, 14]. Our findings confirm these results but also emphasize a complementary role of NK cells. NK cells are cytotoxic lymphocytes capable of tumor cell lysis independent of prior antigen sensitization [29–31], yet their role in the context of chemotherapy has been underexplored. Here, we show that NK cell abundance significantly increases following NACT and that this increase is positively associated with better therapeutic outcomes. In vivo depletion experiments further confirmed that NK cells contribute functionally to NACT-mediated tumor suppression. These results extend previous findings in other malignancies, such as pancreatic and lung cancer, where chemotherapy has been shown to modulate NK cell infiltration and cytotoxicity [32, 33].
Mechanistically, we identified CCL8 as a key chemokine mediating NK cell recruitment following NACT. Transcriptomic and cytokine profiling revealed upregulation of several NK-recruiting chemokines post-treatment, among which CCL8 showed the strongest correlation with NK cell-associated markers. Functional blockade of CCL8 in vitro and in vivo significantly reduced NK cell infiltration and attenuated the therapeutic effects of NACT, establishing a causal link. CCL8, a CC chemokine structurally related to CCL2, has been shown to attract various immune subsets through receptors such as CCR1, CCR2, and CCR5 [34–37]. Our study is the first to implicate CCL8 as a mediator of chemotherapy-induced NK cell trafficking in GC. We further demonstrated that NACT activates the MAPK signaling pathway in tumor cells, and pharmacologic inhibition of p38 or JNK reduced CCL8 expression, suggesting that MAPK signaling is required for this effect. These results provide new mechanistic insight into how tumor-intrinsic responses to chemotherapy can regulate immune recruitment, adding to prior evidence linking MAPK activity to chemokine expression and immunogenic cell death [38–40].
Translationally, our findings suggest that combining NACT with NK cell-based immunotherapy may yield enhanced clinical benefit. In the MGC803-18.2 GC mouse model, SOX chemotherapy combined with CLDN18.2-targeted CAR NK cells produced synergistic antitumor effects and further increased NK cell infiltration, indicating that tumor cells exposed to chemotherapy become more receptive to NK cell-mediated killing. This was supported by increased expression of NK ligands such as MICA and MICB, as well as improved NK cytotoxicity in vitro. Notably, an advanced GC patient receiving 5-FU and oxaliplatin in combination with adoptive NK cell transfer exhibited increased peripheral NK cell counts, elevated IL-2R levels, and significant tumor regression. These findings are consistent with early-phase studies reporting favorable outcomes for NK-based immunotherapy in hepatocellular carcinoma and metastatic GC [41–43]. As NK cell therapies continue to evolve, including CAR-engineered NK cells and cytokine-activated NK products, chemotherapy may serve not only as a cytotoxic agent but also as an immune-priming strategy that enhances NK cell trafficking and function within the tumor.

Conclusion
In conclusion, our study reveals that NACT enhances antitumor immunity in GC by promoting NK cell infiltration through tumor-derived CCL8. We identify MAPK pathway activation as the upstream driver of CCL8 secretion in tumor cells, establishing a mechanistic link between chemotherapy and NK cell recruitment. Functional validation in both animal models and a clinical case supports the therapeutic potential of combining NACT with NK cell-based immunotherapy. These findings provide a rationale for future clinical studies exploring this combination strategy in GC.

Supplementary Information
Below is the link to the electronic supplementary material.