Ketogenic diet impairs NK cell cytotoxic function in colorectal cancer liver metastasis by inducing ferroptosis via suppression of the p62-Keap1-Nrf2 pathway.

1
Introduction
Colorectal cancer (CRC) ranks as the third most commonly diagnosed malignancy and the second leading cause of cancer-related deaths worldwide [1]. In China, CRC has recently surpassed other cancers to become the second most prevalent neoplasm [2]. Among newly diagnosed colorectal cancer patients, 20 % present with metastatic disease at initial diagnosis, while an additional 25 % of those initially presenting with localized disease will subsequently develop metastases [3,4]. As the primary site of hematogenous spread in CRC, hepatic metastasis development correlates with a precipitous drop in five-year survival from 60 % to 13 %, constituting a major clinical challenge in disease management [5,6].
Emerging evidence underscores the critical influence of dietary factors across the entire cancer continuum, encompassing tumorigenesis, disease progression, and therapeutic response [7]. Among various nutritional interventions, the KD - characterized by high fat and very low carbohydrate intake - has gained attention as a potential adjunctive therapy in oncology [8]. Some animal studies indicate that KD may attenuate tumor growth kinetics and improve survival outcomes, with particularly promising results reported in CRC [9,10]. The therapeutic potential of KD appears partially mediated by β-hydroxybutyrate (BHB), a key ketone body that functions as an epigenetic modulator to enhance CD8+ T-cell memory formation [11]. However, the application of KD remains controversial due to conflicting scientific findings. Victoire et al. demonstrated that BHB paradoxically increased the metastatic potential of pancreatic ductal adenocarcinoma, particularly to the liver [12]. Similarly, Ferrer et al. reported that while KD promoted ferroptosis in tumor cells, it concurrently induced corticosterone deficiency that exacerbated cancer-associated cachexia [13]. Notably, the effects of KD and its metabolic byproducts on CRLM remain poorly characterized, representing a critical gap in our current understanding.
The immune cells residing within the TME exert crucial roles in the process of tumorigenesis. NK cells, a pivotal subset of tumor-antagonizing immune cells, orchestrate the tumoricidal response by secreting perforin, granzymes, and pro-inflammatory cytokines and chemokines, including IFN-γ, TNF-α, and CCL5 [14]. Current evidence suggests that diminished NK cell infiltration levels and/or compromised cytolytic activity are significantly correlated with reduced overall survival rates and increased relapse in colorectal cancer patients [[15], [16], [17]]. In the present study, we demonstrate that KD and its metabolite impair NK cell-mediated antitumor immunity, thereby facilitating metastatic progression of colorectal cancer in hepatic microenvironments. Thus, the robust viability and potent efficacy of NK cells are pivotal to the inhibition of tumorigenesis. Recently, novel therapies that enhance antitumor immunity have been developed. Among these therapies, several have focused on NK cell-based immunotherapies aimed at combating CRC, including monoclonal antibodies, immune checkpoint blockade approaches, and adoptive NK cell therapies [18]. Elucidating the mechanisms by which KD and its metabolites in our model suppress NK cell quantity and functionality will enable the development of potentiated NK cell-based immunotherapies to combat CRC and CRLM.
In this study, we investigated the role of the KD in promoting liver metastasis of colorectal cancer and its immunomodulatory effects. We identified a KD-derived metabolite that promotes CRLM by inducing ferroptosis in NK cells. Our mechanistic study demonstrated that this metabolite impairs NK cell survival and cytotoxic function through the p62-Keap1-Nrf2 axis. Through targeted intervention, we successfully restored NK cell survival and tumor-killing capacity. Our findings reveal the tumor-promoting effects of KD, highlighting potential risks in clinical applications. Critically, we demonstrate that KD drives an immunosuppressive reprogramming of NK cells within the TME, establishing a proof-of-concept that targeted epigenetic or metabolic interventions could rescue their tumor surveillance function.

2
Materials and methods
2.1
Cell culture and treatment
All cell lines were obtained from the Kunming Cell Bank of the Chinese Academy of Sciences. The human colorectal cancer (CRC) cell line HT-29 and the NK cell line NK-92MI were maintained in RPMI 1640 medium (HyClone, Logan, UT, USA), whereas the murine CRC cell line MC38 was cultured in Dulbecco's modified Eagle's medium (DMEM; HyClone, Logan, UT, USA). All media were supplemented with 10 % fetal bovine serum (FBS; Gibco, Newcastle, Australia) and 1 % penicillin-streptomycin (Biological Industries, Shanghai, China), and cells were incubated at 37 °C in a humidified 5 % CO2 atmosphere.
To evaluate the effects of PE on murine and human NK cells in a tumor microenvironment, NK cells were isolated from mouse spleens using the EasySep™ Mouse CD49b Positive Selection Kit (STEMCELL Technologies, Vancouver, Canada). Subsequently, 2 × 105 cells were cultured with 5 μg/mL PE (GLPBIO, California, USA) in conditioned medium from MC38 cells. Similarly, NK-92MI cells were treated with PE in the presence of HT-29 cell-conditioned medium. After 24 h of incubation, cells were harvested for further analysis.

2.2
Human subjects
The use of human tissues was conducted in accordance with the Helsinki Declaration and approved by the Ethics Committee of The First Affiliated Hospital, College of Medicine, Zhejiang University (Approval No. IIT20250911A).

2.3
Animal experiments
Five-week-old female C57BL/6J mice were obtained from Hangzhou Ziyuan Experimental Animal Technology Co. and maintained under specific pathogen-free (SPF) conditions. To evaluate the impact of a ketogenic diet (Xietong Pharmaceutical Bio-engineering Co., Ltd., Nanjing, Jiangsu) on colorectal cancer liver metastasis, mice were randomly assigned to either a standard diet or ketogenic diet for two weeks prior to tumor inoculation. Subsequently, animals received an intrasplenic injection of 3 × 105 MC38 cells in 100 μL phosphate-buffered saline (PBS). Dietary regimens were maintained until sacrifice at either day 14 or 21 post-injection.
To assess the effects of PE (derived from the ketogenic diet) on metastatic progression, mice received intraperitoneal injections of either control solution or PE (5 μg/mouse, three times weekly) prior to MC38 cell inoculation. Serum and liver tissues (including tumor lesions) were collected for subsequent analysis. All experimental procedures were approved by the Institutional Animal Care and Use Committee of Zhejiang University School of Medicine (Protocol No. 2024-1738) and conducted in compliance with relevant ethical guidelines.
To enhance the therapeutic potential of NK-92MI cells in a model of CRC liver metastasis, four-week-old female nude mice were intrasplenically injected with 4 × 106 HT-29 cells suspended in 100 μL of PBS. Four weeks post-inoculation, NK-92MI cells were pretreated with 10 μM tert-butylhydroquinone (TBHQ; MCE, New Jersey, USA) for 24 h and subsequently administered to the mice via tail vein injection three times per week for three weeks.

2.4
Flow cytometry and image stream
For the isolation of tumor-infiltrating lymphocytes (TILs), tumor tissues were enzymatically digested in a dissociation medium consisting of DMEM supplemented with 2 % fetal bovine serum (FBS), 0.1 % collagenase IV (Gibco, Newcastle, Australia), 0.01 % hyaluronidase (Solarbio, Beijing, China), and 0.02 % DNase I (Roche Diagnostics, Rotkreuz, Switzerland) for 30 min at 37 °C with constant agitation. Lymphocytes were subsequently purified by density gradient centrifugation using Percoll at 800×g for 25 min. For surface marker analysis, isolated cells were stained with fluorochrome-conjugated antibodies (detailed in Table S1) for 30 min at room temperature in the dark. Intracellular cytokine staining was performed following 4-h stimulation with Leukocyte Activation Cocktail containing BD GolgiPlug (BD Pharmingen, New Jersey, USA) to enable cytokine accumulation (detailed in Table S1).
Cellular fluorescence was quantified using one of the following analytical systems: a BD FACSCanto II flow cytometer (BD Biosciences, New Jersey, USA), a Cytek Aurora spectral cytometer (Cytek Biosciences, Fremont, CA, USA), or a Millipore ImageStream®X Mk II imaging cytometer (MilliporeSigma, Massachusetts, USA). Acquired data were subsequently analyzed with FlowJo™ software (v10.8.1; FlowJo LLC, Oregon, USA) or the IDEAS® application suite (v6.2; Luminex Corporation, Texas, USA).

2.5
Measurement of liver injury
A 10 μl aliquot of serum was dispensed onto Fuji dry-chem slides (Fujifilm, Kyoto, Japan), which had been pre-loaded into the dry biochemical analyzer (Fujifilm, Kyoto, Japan). The concentrations of ALT and AST were then directly measured and recorded individually, in accordance with the manufacturer's instructions.

2.6
Western blotting analysis
Tissues or cells were lysed in RIPA lysis buffer (Beyotime, Shanghai, China), supplemented with a protease and phosphatase inhibitor cocktail (Beyotime, Shanghai, China), for 30 min at 4 °C. Subsequently, the lysate was centrifuged at 12,000×g for 10 min at 4 °C. The resulting supernatant was collected, and its protein concentration was determined using the Enhanced BCA Protein Assay Kit (Beyotime, Shanghai, China). The proteins were then separated by 10 % sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and incubated with specific antibodies, which are detailed in Table S2.

2.7
EasyLight luciferase assay (ELLA)
Supernatants from NK-92MI cells, treated with or without TBHQ, were collected and analyzed for cytokine secretion (IL-2, IFN-γ, and TNF-α) using ELLA kits (ProteinSimple, California, USA) according to the manufacturer's protocol.

2.8
RNA extraction and quantitative real-time PCR
Total RNA was isolated using the Trizol reagent (Takara, Kyoto, Japan). cDNA synthesis was carried out with HiScript II Q RT SuperMix (Vazyme, Nanjing, China). Subsequently, qPCR was performed on the cDNA using SYBR Green PCR Master Mix (Vazyme, Nanjing, China). The sequences of the primers utilized for PCR analysis are provided in Table S3. The relative gene expression levels were determined by the 2−ΔΔCT method, with GAPDH serving as an endogenous control.

2.9
Untargeted metabolomics
Metabolite extraction and detection were performed at Metware Co., Ltd. (Wuhan, China). A 20 mg tissue sample was homogenized using a grinder operating at 30 Hz, followed by extraction with a 400 μL solution consisting of a 7:3 (v/v) methanol mixture containing an internal standard. Subsequently, 200 μL aliquots of the supernatant were transferred for LC-MS analysis. Each sample underwent analysis using two distinct LC/MS methods: one aliquot was analyzed in positive ion mode, while the other aliquot, analyzed in negative ion mode, employed the same elution gradient as the positive mode. Data acquisition was conducted in information-dependent acquisition (IDA) mode using Analyst TF 1.7.1 Software (Sciex, Concord, ON, Canada). Metabolic identification information was then derived by searching through a combination of the laboratory's proprietary database, integrated public databases, AI-driven databases, and metDNA.

2.10
Single-cell RNA sequencing
Tumor tissues were dissociated at 37 °C with a shaking speed of 50 rpm for approximately 30 min. The resulting cell suspensions were filtered through a 40 μm nylon cell strainer. Subsequently, their viability was assessed using a Countess® II Automated Cell Counter (Thermo Fisher, Massachusetts, USA). The prepared sample was then sent to Majorbio Bio-pharm Technology Co., Ltd. (Shanghai, China) for further testing.
Beads, each bearing a unique molecular identifier (UMI) and cell barcode, were loaded to near saturation to ensure that each cell was paired with a bead within a Gel Beads-in-Emulsion (GEM) system. These 10 × beads then underwent second-strand cDNA synthesis, adaptor ligation, and universal amplification. The sequencing libraries were quantified using both a High Sensitivity DNA Chip (Agilent, California, USA) on a Bioanalyzer 2100 and the Qubit High Sensitivity DNA Assay (Thermo Fisher, Massachusetts, USA). Sequencing was performed on a Novaseq Xplus platform using PE150 mode. Both the sequencing and bioinformatic analysis were conducted on the platform provided by Majorbio Co., Ltd. (Shanghai, China).
The reads were processed through the Cell Ranger (v7.1.0) pipeline, adhering to default and recommended parameters. The FASTQ files generated from the Illumina sequencing output were aligned to the mouse genome, specifically version GRCm38, utilizing the STAR algorithm [19].

2.11
Detection of mitochondrial function and ATP production
Mitochondria were visualized using Mito-Tracker Green and Mito-Tracker Red CMXRos (Beyotime, Shanghai, China). The dyeing solution was prepared in accordance with the manufacturer's instructions. Cells were incubated with the dyeing solution for 30 min and subsequently washed twice with 1 × PBS. Analysis was conducted using either the STELLARIS 5 Laser Scanning Confocal Microscope (Leica, Wetzlar, Germany) or the BD FACS Calibur Canto II (BD Pharmingen, NJ, USA). ATP production was quantified using an ATP assay kit (Beyotime, Shanghai, China), following the manufacturer's instructions, and measured with the GloMax® 20/20 Luminometer (Promega, Madison, Wisconsin, USA).

2.12
Transmission scanning electron microscope (TEM)
Fresh tissues or cells were fixed in 2.5 % glutaraldehyde at 4 °C for a minimum of 4 h, followed by fixation with 1 % osmic acid for 1 h at room temperature. The samples were then washed three times with ddH2O. Subsequently, they were stained with 2 % uranyl acetate for 30 min. The samples underwent dehydration with a series of alcohol concentrations (50 %, 70 %, 90 %, 100 %, and again 100 %) and twice with 100 % acetone, with each step lasting 15–20 min. The samples were then treated with a mixture of acetone and embedding medium in a 1:1 ratio for 2–4 h, followed by infiltration with a 1:3 ratio overnight. Next, the samples were embedded in pure embedding medium within an embedding plate and kept at 37 °C overnight. Polymerization was carried out at 60 °C for 48 h. Ultrathin sections of 60–80 nm were obtained using an ultrathin microtome. Finally, the slices were observed under a Talos 120 kV transmission electron microscope (Thermo Fisher, Massachusetts, USA).

2.13
Flow cytometric measurement of the mitochondrial permeability transition pore (MPTP)
The opening of the MPTP in PE-treated NK-92MI cells was analyzed using a Mitochondrial Permeability Transition Pore Assay Kit (Biosharp, Hefei, China) coupled with flow cytometry detection. Cell suspensions (1 × 106 cells/mL) were incubated with calcein-AM, a non-fluorescent probe that distributes into cytoplasmic compartments, including the mitochondria. Following loading, cytoplasmic esterases hydrolyzed the AM moiety, yielding the membrane-impermeant fluorescent compound calcein. To isolate mitochondrial-specific signals, CoCl2 was introduced to selectively quench cytoplasmic calcein fluorescence, while mitochondrial-localized fluorescence remained intact for quantification. Disruption of mitochondrial integrity resulted in a marked decrease in calcein fluorescence intensity.

2.14
Extracellular flux assays
Mitochondrial oxygen consumption rates (OCR) in NK-92MI cells were quantified using the Seahorse XFe24 Analyzer (Agilent, California, USA). Following 24-h exposure to PE, cells were rinsed twice with pre-warmed assay medium (Seahorse XF RPMI containing 1 mM sodium pyruvate, 2 mM l-glutamine, and 10 mM glucose) and seeded at 10,000 cells/well into Cell-Tak-coated 24-well Seahorse plates (Corning, New York, USA). Real-time OCR measurements were performed using the Mito Stress Test kit under standardized protocol conditions.

2.15
Cell cytotoxicity
To prepare for subsequent co-culture, 1 × 105 MC38 cells, prestained with CFSE staining solution from Beyotime (Shanghai, China), were carefully plated in a 24-well plate. Following this, NK cells were isolated from mouse spleens utilizing the EasySep Mouse CD49b Positive Selection Kit (STEMCELL Technologies, Vancouver, Canada). These NK cells were treated with or without PE for 24 h and then introduced into the 24-well plate, where they were mixed with the prestained MC38 cells at effector-to-target (E: T) ratios of 2.5:1, 5:1, and 10:1, respectively. After a 12-h period, the cells were harvested and subsequently stained with 7-AAD Viability Staining Solution (Thermo Fisher, Massachusetts, USA). The stained cells were then analyzed using the BD FACS Calibur Canto II flow cytometer (BD Pharmingen, NJ, USA).
A total of 1 × 104 HT-29 cells were plated in an E-Plate Cardio 24 (ACEA Biosciences, San Diego, California, USA) and allowed to adhere to the plate's surface. Subsequently, NK-92MI cells were either treated with or without PE for 24 h and then added to the E-Plate Cardio 24 containing the HT-29 cells. The proliferation of the adherent HT-29 cells was then monitored using the xCELLigence RTCA Cardio system (ACEA Biosciences, San Diego, California, USA).

2.16
Migration assays
NK-92MI cells were treated with PE in the presence of HT-29 cell-conditioned medium. After 24 h of incubation, cells were harvested for further analysis. For the migration assay, a 24-well Transwell plate with 8 μm pore inserts (Corning, NY, USA) was used. The upper chamber was seeded with 100 μL of HT-29 cell suspension (1.5 × 105 cells/mL) in serum-free medium, while the lower chamber was filled with 600 μL of NK-92MI cells (with or without PE treatment) in medium containing 20 % serum. The plate was incubated at 37 °C in a humidified 5 % CO2 atmosphere for 72 h. Following incubation, migrated cells on the lower membrane surface were fixed with 4 % paraformaldehyde (15 min), stained with crystal violet (20 min), and quantified using an Olympus IX83 fluorescence microscope (Olympus, Tokyo, Japan).

2.17
Lipid peroxidation and FerroOrange measurement
Cells were incubated with 5 μM C11-BODIPY 581/591 (Invitrogen, Carlsbad, California, USA) for 30 min in a humidified incubator at 37 °C with 5 % CO2 to assess lipid peroxidation. Subsequently, cells were harvested, and lipid peroxidation levels were quantified using a flow cytometer.
Cells were incubated with 1 μM FerroOrange (DOJINDO, Kyushu Island, Japan) for 30 min at 37 °C in a humidified incubator with 5 % CO2 to evaluate the presence of Fe2+. Following this incubation, the cells were collected and examined for lipid peroxidation levels using an Olympus IX83 fluorescence microscope (Olympus, Tokyo, Japan).

2.18
Kaplan-Meier plotter (KM plotter)
The Kaplan-Meier plotter (KM plotter) database (https://kmplot.com/) was utilized to assess the mRNA expression levels of characteristic immune cells infiltrating tumor tissues in CRC patients across different stages, in order to evaluate their prognostic significance.

2.19
Statistical analysis
Statistical analysis was conducted using GraphPad Software Prism 9.0 (San Diego, CA, USA). Significance between two groups was assessed using the student's t-test. For comparisons involving three or more groups, one-way analysis of variance (ANOVA) was employed, followed by Tukey's multiple comparison test. The data presented are the mean ± standard error of the mean (S.E.M.) and are representative of at least three independent experiments. Statistical significance was defined as ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001 and nsp> 0.05.

3
Results
3.1
Ketogenic diet promotes the metastatic growth of colorectal cancer in the liver
To investigate the role of the KD in liver CRC metastasis, we performed an experiment using a mouse model. We established liver CRC metastasis by intrasplenic injection of MC38 in mice fed either a ND or a KD (Fig. 1A). Following this, we evaluated and compared the liver metastatic burden between the KD and ND groups. Strikingly, our results showed that on Day 21 post-MC38 injection, mice in the KD group had significantly more metastatic lesions and heavier liver tumors than those in the ND group (Fig. 1B and C). Histological analyses, including hematoxylin and eosin (H&E) staining and Ki67 immunostaining, further revealed a notably increased invasion and proliferation capacity of tumor cells in the KD group (Fig. 1D). Additionally, elevated serum levels of AST and ALT in the KD group suggested more severe liver function impairment compared to the ND group (Fig. 1E). Epithelial-mesenchymal transition (EMT) is intricately linked to tumor initiation, invasion, resistance to therapy, and particularly metastasis [20]. Given its significance, we conducted a thorough examination of several characteristic protein markers of EMT in the livers of both groups. These markers included Slug, Snail and Vimentin, which are known to be indicative of EMT processes in tumors. The expressions of these EMT markers were significantly higher in the KD group compared to the ND group, indicating the establishment of metastatic niches in the KD group (Fig. 1F). Additionally, MMP12 and Collagen I, representative markers of extracellular matrix (ECM) remodeling, were also upregulated in the KD group, further supporting their contribution to metastatic progression (Fig. 1F).
Given the observed promotion of metastatic growth in the KD group, we hypothesized that the KD induced an immunosuppressive liver microenvironment. To validate this immunosuppressive effect, we analyzed the RNA expression levels of immune-suppressive and immune-activating genes in the liver tissue between the two groups. Gene expression analysis revealed that the KD group, relative to ND controls, showed significantly elevated levels of immunosuppressive factors (Tgfb, Il4, Il6, Il23) coupled with reduced expression of immune-activating cytokines (Il15, Il18) (Fig. 1G). To determine whether metabolites derived from a ketogenic diet altered the liver microenvironment in our model, we performed untargeted liquid chromatography-tandem mass spectrometry (LC-MS/MS) to compare hepatic metabolites between the ND and KD groups. The analytical robustness of our model was confirmed by Partial Least Squares Discriminant Analysis (PLS-DA) (Fig. 1H and I), while hierarchical clustering analysis revealed distinct metabolomic profiles between the two groups (Fig. 1J). Taken together, our findings reveal that a ketogenic diet induces significant alterations in the liver's immune-metabolic microenvironment, creating conditions that promote the metastatic progression of colorectal cancer to the liver.

3.2
Ketogenic diet enhances liver CRC metastasis via metabolite phosphatidylethanolamines
It has been reported that the ketone bodies, acetoacetate (AcAc) and BHB, which are effector metabolites produced by a ketogenic diet, have the potential to reduce the proliferation of colon cancer cells and even inhibit tumor growth [9]. In our study, volcano plot analysis revealed 261 significantly upregulated and 205 downregulated metabolites in the KD group compared to the ND group (Fig. 2A). To identify key differential metabolites, we performed KEGG (Kyoto Encyclopedia of Genes and Genomes) enrichment analysis on the altered metabolites. Glycerophospholipid metabolism emerged as the most significantly enriched pathway (Fig. 2B). Further analysis via heatmap visualization highlighted upregulated metabolite profiles within this pathway, particularly showing a pronounced increase in multiple phosphatidylethanolamines (PEs) in the KD group (Fig. 2C). Alterations in the PC and/or PE levels across different tissues have been linked to metabolic disorders such as atherosclerosis, insulin resistance, and obesity [21]. The schematic overview of the Kennedy pathway revealed key enzymes and metabolic intermediates involved in phosphatidylcholine (PC) and PE biosynthesis (Fig. 2E). Rich ethanolamine derived from the KD may play a crucial role in the accumulation of PE. Interestingly, Pemt expression of the liver was downregulated in the KD group, resulting in diminished PC synthesis via methylation of PE (Fig. S1A). Notably, the elevated phosphatidylethanolamines in our study included multiple species harboring polyunsaturated fatty acyl (PUFA) chains (Fig. 2D).
The correlation analysis revealed that higher PE abundance was associated with increased tumor weight, implying that increased PE levels may contribute to tumor growth (Fig. 2F). To evaluate the in vivo effects of PE, we administered PE via intraperitoneal (i.p.) injection to mice with CRC liver metastasis (Fig. 2G). As expected, PE treatment markedly exacerbated the progression of metastatic liver tumors, as demonstrated by a significant increase in both the number and weight of tumor nodules (Fig. 2H and I). Histopathological analysis further revealed enhanced tumor aggressiveness and proliferation in the PE-treated group compared to controls (Ctrl), evidenced by hematoxylin and eosin (H&E) staining and elevated Ki67 expression (Fig. 2J). Additionally, the PE group exhibited significantly higher serum levels of aspartate aminotransferase (AST) and alanine aminotransferase (ALT) than the Ctrl group, indicating more pronounced liver dysfunction (Fig. 2K). These findings demonstrate that PEs, ketogenic diet-derived metabolites, exacerbate colorectal cancer liver metastasis in this model.

3.3
Ketogenic diet administration impairs NK cell immunity against liver CRC metastatic growth
Given the KD-induced immunosuppressive microenvironment in the liver, we hypothesized that KD might reshape the hepatic immune landscape, particularly within the TME of liver CRC metastases. To test this hypothesis, we performed comprehensive immune profiling of tumor-infiltrating leukocytes in a murine model of liver CRC metastasis using flow cytometry. Our analysis revealed a significant reduction in tumor-infiltrating NK cell populations following KD administration (Fig. 3A). Quantitative assessment confirmed markedly decreased NK cell infiltration in the TME of KD-fed mice compared to ND controls (Fig. S1B). This finding was further corroborated by immunofluorescence staining (IF), which showed fewer NK cells in tumors from the KD group (Fig. 3H). Notably, despite these NK cell-specific changes, we observed almost no significant alterations in other immune cell populations, including CD4+ and CD8+ T cells, NKT cells, γδT cells, macrophages, dendritic cells (DCs), or MDSCs (Fig. 3A–G and S1E-G). These results suggest that KD selectively modulates NK cell recruitment to the TME while maintaining other immune cell compartments in liver CRC metastases. NK cells play a critical role in antitumor immunity by mediating direct cytotoxicity and orchestrating immune responses. Our findings demonstrated that a KD significantly modulated the functional properties of NK cells. Notably, tumor-infiltrating NK cells from KD-treated mice exhibited reduced IFN-γ and TNF-α production alongside elevated PD-1 expression (Fig. 3I–J, S1C-D, H). Together, these data suggest that a ketogenic diet may suppress NK cell accumulation and cytotoxic function, thereby facilitating metastatic progression of colorectal cancer in the liver.
Although the role of NK cells in controlling primary tumors remains controversial, multiple studies across various cancers (including colorectal cancer) have demonstrated that higher levels of NK cells in circulation or tumor tissue are inversely correlated with clinical metastasis occurrence [22,23]. To confirm this observation, IF staining was conducted to characterize immune cell composition in the TME of CRC patients stratified by liver metastasis status. Notably, NK cell infiltration was markedly decreased in metastatic cases, whereas macrophages and T cells showed comparable levels between groups (Fig. 3K). In addition, we performed Kaplan-Meier survival analysis to evaluate the prognostic significance of CD56 expression in the TME of CRC patients among different clinical stages. Our analysis revealed that high CD56 expression was significantly associated with prolonged survival in stage IV colorectal cancer patients, though this association was not observed in earlier disease stages (Fig. 3L). Given that patients with stage IV disease present with distant metastases, we conducted Kaplan-Meier survival analysis to assess the prognostic value of CD56 expression in metastatic versus non-metastatic CRC patients. The results showed that high CD56 expression was significantly associated with prolonged survival in metastatic CRC patients, though this association was not observed in non-metastatic CRC patients (Fig. 3M). Kaplan-Meier survival analysis was also conducted to assess whether key gene expression in other immune cells predicts prognosis in metastatic versus non-metastatic CRC patients. Our analysis revealed that only NK cells, but not other immune cell populations (B cells, macrophages, dendritic cells, or CD8+ T cells), maintained prognostic significance in metastatic CRC disease (Fig. S2A–D). Collectively, our results indicate that the ketogenic diet compromises NK cell function, a key immune defense mechanism against hepatic metastatic progression in colorectal cancer.

3.4
PE-mediated depletion of NK cells contributes to enhanced liver metastasis in colorectal cancer
To investigate the mechanisms by which PE influences NK cells in liver CRC metastasis, we performed scRNA-seq on liver tumor tissues from both Ctrl and PE-treated groups (Fig. 4A). This unbiased approach comprehensively characterized the transcriptional landscape, capturing diverse cell populations, including tumor cells, fibroblasts, monocytes, macrophages, DCs, NK cells, T cells, and other immune subsets (Fig. 4B). Notably, the proportion of NK cells among all detected cells was significantly lower in the PE-treated group than in the Ctrl group (Fig. 4B). To characterize NK cell heterogeneity, we performed hierarchical clustering on single-cell RNA sequencing data from all NK cells, which revealed three distinct subsets at a resolution of 0.2 (Fig. 4C). By integrating differential gene expression analysis with RNA Velocity, we defined these subsets as: Ccl5+ NK cells – exhibiting cytotoxic effector properties; Ube2c + NK cells – displaying proliferative potential; and Il1b + NK cells – showing exhaustion markers (Fig. 4C–S3A-B). Notably, all three NK cell subsets were significantly reduced in the PE-treated group compared to controls (Fig. 4C).
Volcano plots derived from differential gene expression analysis (Ctrl vs. PE-treated groups) revealed significant downregulation of Nkg7, Ifng, Ccl5, and Pfn1 in the PE-treated group (Fig. 4D). Notably, the PE-treated group exhibited not only a marked reduction in NK cell numbers but also a pronounced impairment in their functional capacity compared to controls. To corroborate these findings at the protein level, we performed flow cytometry to quantify NK cell abundance and assess functional markers. The results confirmed our transcriptomic data, showing consistent reductions in both NK cell frequency and activity (Fig. S3C–E). Collectively, these data demonstrate that PE treatment compromises NK cell-mediated immunity, thereby promoting liver metastatic growth in CRC.

3.5
PE attenuates mitochondrial function in NK cells within the TME
To elucidate the underlying mechanisms, we performed Gene Ontology (GO) enrichment analysis, which identified significantly enriched pathways in the PE-treated group, including mitochondrial function, cellular metabolism, and cell death-related processes (Fig. 4E). Furthermore, gene set enrichment analysis (GSEA) revealed that PE treatment downregulated genes involved in mitochondrial protein pathways, particularly those encoding inner mitochondrial membrane protein complexes and mitochondrial protein-containing complexes (Fig. 4F). These results suggest a potential impairment of mitochondrial function in NK cells following PE treatment, providing critical mechanistic insights for further investigation.
We conducted transmission electron microscopy (TEM) analysis of liver tumor tissues from both control and PE-treated groups to examine intracellular structures, with particular focus on immune cell morphology. TEM analysis of PE-treated samples revealed several key ultrastructural alterations: significantly diminished cell-cell contacts between immune cells and other cell types (blue dashed boxes), reduced mitochondrial numbers (red arrows), and pyknotic mitochondrial morphology (red solid boxes), indicative of mitochondrial dysfunction and impaired cytotoxicity (Fig. 4G). Subsequently, we isolated NK cells from liver tumor tissues of both groups and performed MitoTracker Green/Red staining. Flow cytometric analysis of liver tumor-derived NK cells showed that PE treatment significantly impaired mitochondrial biogenesis (reduced Green fluorescence) and depolarized mitochondrial membranes (decreased Red fluorescence) compared to untreated controls (Fig. 4H and I), indicating substantial mitochondrial impairment.
To determine whether the mitochondrial damage induced by the metabolite PE persists in vitro, we isolated NK cells from mice and cultured them with tumor-conditioned medium supplemented with PE. MitoTracker Green staining of both mouse NK cells and human NK-92MI cells revealed a marked decrease in mitochondrial mass following PE stimulation, as visualized by confocal microscopy and quantified by fluorescence intensity (Fig. 4J–S3F-H). Furthermore, we observed a significant, dose-dependent reduction in ATP production in PE-treated mouse NK cells (Fig. S3I). A similar effect was confirmed in the human NK-92MI cell line (Fig. S3J). Given that GSEA of single-cell RNA sequencing data suggested potential defects in mitochondrial membrane protein complexes upon PE treatment, we assessed mitochondrial membrane permeability. Our results demonstrated a dose-dependent increase in mitochondrial membrane permeability in PE-exposed NK cells (Fig. 4K). To further evaluate mitochondrial function, we performed real-time metabolic profiling using a mitochondrial stress assay in PE-treated NK-92MI cells. Both maximal respiration and spare respiratory capacity exhibited dose-dependent impairment (Fig. 4L), indicating compromised mitochondrial oxidative phosphorylation function. Taken together, these results demonstrate that PE disrupts both structural integrity and physiological function of mitochondria in NK cells residing in the TME.

3.6
PE suppresses the p62-Keap1-Nrf2 pathway to induce ferroptosis in NK cells
The scRNA-seq analysis demonstrated distinct expression patterns of mitochondrial function-related genes in NK cells from the PE group versus controls, as visualized by heatmap (Fig. 5A). Notably, the Sqstm1 gene—encoding the protein Sqstm1/p62—was differentially expressed and has been implicated in oxidative stress and cell death pathways. Previous studies indicate that the p62-Keap1-Nrf2 axis inhibits ferroptosis by stabilizing Nrf2 through competitive binding to Keap1, thereby promoting Nrf2 nuclear translocation and transcriptional upregulation of antioxidant genes [24]. We examined the expression of Nrf2, p62, and Keap1 in NK-92MI cells cultured in HT-29-conditioned medium following PE treatment. Western blot analysis revealed a dose-dependent decrease in p62 and Nrf2 protein levels, while Keap1 expression showed a corresponding increase with higher PE concentrations (Fig. 5B), supporting the inhibition of the p62-Keap1-Nrf2 axis.
To evaluate the potential role of p62-Keap1-Nrf2 pathway suppression in ferroptosis induction, we employed flow cytometry with Annexin V/PI double staining to characterize cell death patterns in NK-92MI cells following PE treatment (Fig. 5C). The results demonstrated a dose-dependent increase in cell death rates with escalating PE concentrations (Fig. 5D). We also examined cell death and proliferation in mouse colon cancer cells (MC38) and human colon cancer cells (HT-29) following PE stimulation. The results demonstrated that the metabolite PE did not inhibit tumor cell growth, thereby excluding any direct effect of PE on tumor cells (Fig. S4A–D). To investigate the specific cell death mechanism induced by the metabolite PE in NK cells, we treated PE-stimulated NK-92MI cells with various cell death inhibitors, including the ferroptosis inhibitor Ferrostatin-1 (Fer-1), the necroptosis inhibitor Necrostatin-1 (Nec-1), and the apoptosis inhibitor z-VAD-FMK (zVAD). Flow cytometry analysis revealed that only Fer-1 significantly attenuated PE-induced NK cell death. These findings suggest that PE induces ferroptosis in NK cells (Fig. 5E).
Subsequently, we examined these ferroptosis-related characteristic markers to further validate the ferroptosis phenotype. Western blot analysis demonstrated a dose-dependent reduction in GPX4 protein expression in NK-92MI cells following PE treatment (Fig. 5F). PE treatment induced a dose-dependent increase in lipid peroxidation in NK-92MI cells, as measured by C11 BODIPY staining (Fig. 5G). Staining with the ferrous iron (Fe2+) indicator FerroOrange revealed significantly higher levels of free Fe2+ in PE-treated NK-92MI cells, as evidenced by intense fluorescence signal accumulation (Fig. 5H). The GSH/GSSG ratio in NK-92MI cells decreased progressively with increasing PE concentrations (Fig. 5I). TEM of PE-stimulated NK-92MI cells revealed decreased mitochondrial numbers, shrinkage, and cristae disappearance, displaying characteristic features of ferroptosis (Fig. 5J). Furthermore, immunofluorescence staining of liver tumor tissues from PE-treated mice revealed not only a reduction in NK cell infiltration but also a significant decrease in intracellular GPX4 expression within these NK cells (Fig. 5K). To further validate these findings, we isolated tumor-infiltrating NK cells from PE-treated mice and performed C11 BODIPY staining, which demonstrated a marked increase in lipid peroxidation levels (Fig. S4E). These in vivo results corroborate our earlier conclusions regarding PE-induced ferroptotic stress in NK cells.

3.7
PE compromises the cytotoxic capacity of NK cells against tumor cells
To investigate whether the elevated ferroptosis in NK cells leads to a decline in their tumor-killing capacity, we further examined the cytokine secretion and tumor-killing function of NK cells after PE intervention. We isolated mouse NK cells and cultured them in the presence of tumor supernatant with or without PE stimulation, then assessed their cytokine secretion. Flow cytometry results showed that PE stimulation significantly reduced the levels of activation-associated cytokines Granzyme B, IFN-γ, and TNF-α in NK cells (Fig. 5L). Next, we conducted a Transwell assay by co-culturing PE-treated or untreated NK-92MI cells with HT-29 cells (Fig. S4F) and observed the migration of HT-29 cells to the lower chamber containing NK cells. Crystal violet staining showed that HT-29 cell migration was significantly enhanced in the presence of PE-treated NK cells compared to untreated NK cells (Fig. 5M).
Additionally, we co-cultured mouse NK cells (with or without PE stimulation) with mouse colon cancer cells MC38 to evaluate the death of MC38 cells, reflecting the tumor-killing ability of NK cells. The results demonstrated that the tumor-killing effect increased with higher NK-to-MC38 cell ratios, but PE-treated NK cells exhibited significantly reduced cytotoxicity against MC38 cells (Fig. 5N). We also co-cultured PE-treated or untreated NK-92MI cells with human colon cancer cells HT-29 and monitored HT-29 cell proliferation using a Real-Time Cellular Analysis (RTCA) system. The results revealed that PE-treated NK cells had a significantly weaker inhibitory effect on HT-29 cell growth, allowing better proliferation of HT-29 cells (Fig. 5N). These findings collectively indicate that, in the presence of tumor supernatant, the metabolite PE induces ferroptosis in NK cells and significantly compromises their tumoricidal capacity.

3.8
Activation of the Nrf2 pathway protects against PE-induced ferroptosis and restores the anti-tumor activity of NK cells
Nrf2, as the central regulator of the p62-Keap1-Nrf2 signaling axis, enhances cellular antioxidant capacity and modulates ferroptosis. Notably, Nrf2 activation further upregulates p62 expression through the p62-Keap1-Nrf2 positive feedback loop, thereby sustaining antioxidant defense and suppressing ferroptosis. We treated NK-92MI cells with the Nrf2-specific agonist TBHQ and found that TBHQ significantly suppressed PE-induced cell death (Fig. 6A). Western blot analysis revealed that TBHQ treatment markedly upregulated the expression of both Nrf2 and p62 proteins in NK-92MI cells, whereas PE treatment substantially reduced their levels. Notably, the combination of TBHQ and PE effectively reversed the downregulation of Nrf2 and p2 and enhanced GPX4 expression (Fig. 6B). To further investigate the subcellular localization of Nrf2, we performed imaging flow cytometry. Representative images demonstrated that TBHQ-treated cells exhibited significantly increased Nrf2 expression, predominantly localized in the nucleus. In contrast, PE treatment markedly reduced Nrf2 levels, while co-treatment with TBHQ effectively counteracted PE-induced Nrf2 downregulation (Fig. 6C). Flow cytometry-based quantification further confirmed these observations (Fig. 6D). To validate the protective effect of TBHQ against PE-induced ferroptosis, we systematically assessed key ferroptosis markers, including lipid peroxidation levels, GSH/GSSG ratio, and Fe2+ content. All these results consistently demonstrated that TBHQ significantly alleviated PE-triggered ferroptosis in NK cells (Fig. 6E–G).
To investigate the role of this mechanism in colorectal cancer liver metastasis progression, we performed comparative immunofluorescence analysis of tumor tissues from non-metastatic and metastatic CRC patient groups. The results revealed a significant reduction in the infiltration of p62+ and/or Nrf2+ NK cells within the TME of the metastatic group compared to the non-metastatic group (Fig. 6H). This result not only suggests the potential association between NK cell dysfunction and colorectal cancer metastasis, but also provides possible molecular markers for clinical prognosis evaluation.
To investigate whether upregulating Nrf2 expression in intratumoral NK cells could enhance their anti-tumor capacity, we treated NK cells with TBHQ in the presence of tumor cell-conditioned medium. Subsequent analysis of cytokine secretion revealed that TBHQ-treated NK cells exhibited significantly increased production of IL-2, IFN-γ, and TNF-α compared to untreated controls (Fig. S5A–C). To evaluate the functional impact of this intervention in vivo, we established a colorectal cancer liver metastasis model in nude mice. After four weeks, mice received intravenous injections of either TBHQ-treated or untreated NK cells three times per week for three weeks (Fig. 6I). Assessment of hepatic tumor burden demonstrated that TBHQ-activated NK cells markedly suppressed liver tumor growth, whereas untreated NK cells exhibited inconsistent anti-tumor efficacy (Fig. 6J and K). These findings suggest that pharmacological activation of the Nrf2 pathway augments the therapeutic potential of NK cells against tumors, providing a novel strategy to optimize NK cell-based immunotherapy.

4
Discussion
Whereas localized CRC is potentially curable by surgery, mCRC invariable progresses due to minimal residual disease comprising metastasis-initiating cells with therapy-refractory properties. Current therapeutic strategies for mCRC demonstrate limited subtype-specific effectiveness and considerable adverse effects, driving the development of innovative targeted approaches such as adoptive cell transfer, microbial ecosystem remodeling, and evidence-based dietary modulation [3]. While dietary factors significantly influence tumor biology, evidence-based dietary interventions remain underdeveloped in clinical oncology practice due to insufficient evidence [7]. In this study, we developed an experimental mouse model to assess the impact of ketogenic dietary intervention on CRC hepatic metastasis, offering translational insights for clinical oncology.
Our study revealed a paradoxical tumor-promoting effect of KD in our murine model of CRC liver metastasis. To delineate how dietary interventions reprogram the hepatic metastatic niche, we employed untargeted metabolomics to identify the metabolic signatures distinguishing ND and KD groups under the CRC liver metastasis model. Our metabolomic analysis revealed a significant enrichment of PE in the KD group, and subsequent functional studies demonstrated that exogenous PE administration enhances CRC liver metastasis. Our findings contrast with the well-documented prominence of β-hydroxybutyrate (BHB) in ketogenic diet studies. While BHB level was significantly elevated in the KD group, this metabolite ranked only within the top 20, indicating its relatively minor metabolic importance compared to PE. This may partially explain why the ketogenic diet promoted rather than suppressed colorectal cancer liver metastasis in our study.
NK cells (Lin− NK1.1+ CD49a− CD49b+ Eomesᴴi T-betᴹed) and ILC1s (typically defined as Lin− NK1.1+ CD49a+ CD49b− Eomesᴸᵒ/- T-betᴴi) both significantly inhibit MC38 cell seeding and growth in the liver [25,26]. Nevertheless, these two populations exhibit distinct functional roles in related disease models. For example, Ducimetière et al. showed that ILC1s are essential for suppressing early metastatic seeding in colorectal cancer, whereas NK cells mediate long-term antimetastatic immunity in MC38-injected mice [25]. In our study, NK cells constituted the vast majority (75 %) of lymphocytes in liver metastases (LM), with ILC1s representing only 10 % (Fig. S1I). In line with this, a recent study by Eleonora Russo et al. reported that NK cell infiltration substantially increased during liver metastasis progression, while ILC1s were progressively excluded from the metastatic sites [17]. Collectively, these findings indicate that the NK1.1+ cells in our model are predominantly conventional NK cells. Furthermore, this specific NK cell population is significantly depleted following KD treatment. Therefore, it is crucial to interrogate how emerging NK cell phenotypes arise in the metastatic niche and what role they play in antitumor immunity in the future.
PE serves not merely as a structural membrane component but actively participates in protein biogenesis and functional regulation, including oxidative phosphorylation, mitochondrial biogenesis, autophagy and ferroptosis [27,28]. The accumulation of PE has also been documented in hepatic steatosis and steatohepatitis [[28], [29], [30]]. In particular, a reduced PC:PE ratio was detected in liver biopsies of patients with nonalcoholic steatohepatitis (NASH) [29,31]. Thus, maintaining the balance of PC:PE ratio in hepatocyte is crucial for cellular integrity, though each phospholipid contributes distinctly to hepatic lipid metabolism. Notably, the observed increase in phosphatidylethanolamines in our study comprised several species containing polyunsaturated fatty acyl (PUFA) chains. Strikingly, a study published in Cell (2025) demonstrated that metastatic tumor cells exhibit markedly higher levels of PUFA-enriched lipid species compared to primary tumor cells, supporting our own findings [32]. Nevertheless, the involvement of PE in cancer progression, especially in the context of CRC liver metastasis, has not been fully elucidated. This represents one of the major highlights of our research.
Additionally, PUFA-containing phospholipids (PUFA-PLs) in cell membranes are key substrates for lipid peroxidation, enabling ferroptosis, a regulated cell death process [33]. Our results demonstrated that NK cell infiltration in the TME was diminished by both the KD and the metabolite PE. Integrated analysis of scRNA-seq and supporting experiments revealed mitochondrial dysfunction and subsequent ferroptosis in tumor-infiltrating NK cells from the PE-treated group. Based on these observations, we revealed that KD and its metabolite PE impaired anti-tumor immunity by inducing ferroptosis of NK cells in the TME. Accumulating evidence indicates that ferroptosis serves as a critical mechanism in tumor suppression, thereby creating new avenues for cancer therapy [34]. However, as research progresses, increasing evidence suggests that simply inducing ferroptosis often fails to suppress tumors and may even promote tumor growth [[35], [36], [37]]. Since ferroptosis-inducing agents lack cellular specificity, they indiscriminately affect all cell types in the TME, including both malignant and immune cells. Cancer cells could even employ genetic/epigenetic changes to resist ferroptosis, including SLC7A11 upregulation or Nrf2 activation [38,39]. Thus, investigating ferroptosis in immune cell subsets within the tumor microenvironment is also of critical importance [40].
Recently, Pooranee K. Morgan and colleagues characterized the lipid profiles of human and mouse immune cells, uncovering that the distinct ferroptosis sensitivity observed across immune cell types stems from their differential PUFA-PLs composition [41]. Specifically, Utilizing established hydrogen atom transfer kinetics of fatty acids, they developed a quantitative Cellular Phospholipid Peroxidation Index (CPI) to assess ferroptosis susceptibility in immune cells [42,43]. Notably, NK cells exhibited the highest CPI among murine immune cell populations, indicating their pronounced vulnerability to ferroptosis. These may explain why the ketogenic diet and its metabolic product PE preferentially induced ferroptosis in NK cells in our study. However, research on NK cell ferroptosis remains limited. Current studies by Lizhong Yao et al. demonstrated that cancer-associated fibroblasts (CAFs) suppress NK cell anti-tumor activity through ferroptosis induction, while Jian-Xin Cui et al. reported that l-kynurenine promotes NK cell depletion via ferroptosis in the gastric cancer microenvironment [44,45]. Our study provides novel insights by revealing the unique susceptibility of NK cells to KD-induced ferroptosis, representing a significant advancement in this emerging field.
As previously stated, lipid peroxidation serves as a primary trigger for ferroptosis. Nrf2 plays a pivotal role in counteracting lipid peroxidation and ferroptosis, making it a compelling research target [24,46]. In general, Nrf2 levels exhibit an inverse correlation with ferroptosis susceptibility, wherein elevated Nrf2 expression confers ferroptosis resistance, while Nrf2 depletion sensitizes cells to ferroptosis inducers. Our scRNA-seq analysis revealed that Nrf2 expression in NK cells did not differ between the Ctrl and PE-treated groups, while p62 levels were significantly decreased in PE-treated NK cells compared to controls. Consistent with our findings, emerging evidence indicated that the p62-Keap1-Nrf2 pathway suppresses ferroptosis by stabilizing Nrf2 through Keap1 inactivation, leading to enhanced nuclear accumulation of Nrf2 and subsequent transcriptional activation of antioxidant genes [[47], [48], [49]]. Furthermore, emerging evidence indicates that lipid peroxidation-mediated oxidative stress in the tumor microenvironment impairs NK cell function. Notably, pharmacological Nrf2 activation has been shown to rescue both metabolic activity and cytotoxic capacity of NK cells, resulting in potentiated anti-tumor immunity [50,51]. Given Nrf2's multifaceted role in promoting cell survival, targeted modulation of its activity remains an exceptionally viable therapeutic avenue for associated disease states. Our findings provide compelling evidence supporting the anti-tumorigenic role of Nrf2 activation in this experimental model.
However, our study has several limitations that should be acknowledged. Firstly, the concept of ketogenic diet is overly broad, as both the source and ratio of dietary components may significantly influence metabolic outcomes, thereby affecting its therapeutic efficacy against colorectal cancer liver metastasis. Therefore, precise formulation of dietary composition represents the cornerstone of standardized nutritional therapy. A second limitation is our exclusive focus on NK cell-mediated mechanisms in the TME, as other immunomodulatory pathways affecting T cells, macrophages, or dendritic cells may also influence the observed effects. Finally, the mechanism by which the metabolite PE downregulates p62 expression remains to be elucidated, which will constitute a primary focus of our subsequent investigations.
Collectively, our findings demonstrate that the KD and its metabolite PE exacerbate CRC liver metastasis through impairment of NK cell quantity and cytotoxic function. Mechanistically, PE mediates NK cell depletion via suppression of the p62-Keap1-Nrf2 antioxidant pathway. Importantly, pharmacological activation of Nrf2 in NK cells restored their tumoricidal capacity, suggesting a targetable metabolic vulnerability (Fig. 7). This work yields two key translational insights: (1) it provides critical safety considerations for clinical KD implementation in CRC patients, and (2) establishes a therapeutic paradigm whereby targeted metabolic or epigenetic modulation of NK cells can reinstate their tumor surveillance capabilities.

Funding
This work was supported by the 10.13039/501100012166National Key Research and Development Program of China (2021YFA1301100, 2021YFA1301101), the Fundamental Research Funds for the Central Universities (2022ZFJH003), Research Project of Jinan Microecological Biomedicine Shandong Laboratory (JNL-2022012B).

CRediT authorship contribution statement
Rui Cai: Project administration, Writing – original draft, Writing – review & editing. Yuting Meng: Project administration, Supervision. Minghui Ru: Resources, Software. Xueyao Wang: Conceptualization, Methodology. Wenting Li: Data curation, Investigation. Xinlan Liu: Formal analysis. Shulin Zhuang: Software. Yong Huang: Visualization. Hongyan Diao: Funding acquisition, Supervision.

Declaration of competing interest
The authors declare no competing financial interests. This work was supported by the National Key Research and Development Program of China (2021YFA1301100, 2021YFA1301101), the Fundamental Research Funds for the Central Universities (2022ZFJH003), Research Project of Jinan Microecological Biomedicine Shandong Laboratory (JNL-2022012B).