Multi-omics analysis reveals that ginsenoside Rb1 improves prognostic outcomes in sepsis by modulating mitochondrial metabolism MTHFD2 targets

Abstract

Sepsis has emerged as a major threat to human mortality. Increasing evidence explores the impact of mitochondrial metabolism on the prognosis of sepsis patients and its therapeutic potential. To enhance risk stratification and identify potential targets, we conducted a retrieval and analysis of differential expression of mitochondrial metabolism related genes (MMRG) between sepsis and normal samples from public databases. Immune infiltration analysis was preformed to gain comprehensive knowledge of features of the established risk model. Additionally, single-cell sequencing results suggested MTHFD2 may be a critical target in sepsis for regulating immune infiltration characteristics, potentially altering platelet metabolism pathways significantly, thereby influencing sepsis occurrence and progression. Utilizing molecular docking, we further screened Ginsenoside Rb1 (Grb1) as a key pharmacological target interacting with MTHFD2. Further animal experiments preliminarily indicated that Grb1 administration was associated with reduced MTHFD2 expression and improved organ function and survival in CLP-induced septic rats. These findings provide new insights and potential therapeutic targets for clinical treatment of sepsis in the future.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-37362-9.

Introduction

Sepsis is defined as a life-threatening organ dysfunction caused by a dysregulated response to infection¹. With over 19 million cases annually, sepsis has become a leading threat to human mortality². Identifying patients at high risk of deterioration is crucial for determining whether aggressive interventions or conservative treatments should be pursued³. Therefore, refining risk stratification for sepsis patients holds significant clinical implications for prognosis improvement⁴.

Mitochondria, as the most critical biological energy supporters, are closely linked to disease progression⁵. Apart from regulating cellular metabolism, mitochondria influence cellular signal transduction, gene expression regulation, and programmed cell death, exerting multifaceted effects on disease⁵. The association between mitochondrial dysfunction and the severity and outcomes of sepsis has been elucidated. Bioenergetic failure during the progression of sepsis is a key pathophysiological mechanism underlying multi-organ dysfunction and clinical outcomes⁶. Impaired mitochondrial metabolism is a critical mechanism contributing to organ dysfunction in sepsis^7,8. Severe damage to mitochondrial metabolism in the early stages of acute illness leads to biogenesis reduction, increased ROS generation, and decreased ATP synthesis^9,10. Given the pivotal role of mitochondria and their metabolism in sepsis, there is increasing evidence exploring their impact on patient outcomes and highlighting mitochondrial components as potential targets for therapeutic strategies^11–13. Thus, further analysis of mitochondrial metabolism in sepsis is warranted.

Although mitochondrial dysfunction has long been implicated in the pathogenesis and progression of sepsis, the specific mitochondrial metabolic determinants that drive immunometabolic dysregulation remain insufficiently defined. A comprehensive multi-omics approach focused on mitochondrial metabolism may provide new mechanistic insights and facilitate the identification of previously unrecognized therapeutic targets. In this study, we initially conducted comparative analyses of differential expression of MMRG between sepsis and normal samples using public databases. Subsequently, based on these differentially expressed MMRG, we performed prognostic analysis, constructed a risk model, and conducted molecular subtype analysis. Bioinformatics analysis revealed MTHFD2 to have the highest risk coefficient and single-cell sequencing results further suggested MTHFD2 as a crucial target in sepsis that may relate to immune infiltration regulation and potentially affecting platelet activation. We subsequently identified Ginsenoside Rb1 (Grb1) as a key pharmacological target interacting with MTHFD2. Further animal experiments preliminarily indicated that Grb1 administration was associated with reduced MTHFD2 expression and improved organ function and survival in CLP-induced septic rats. These findings provide new insights and potential therapeutic targets for clinical treatment of sepsis in the future.

Materials and methods

Ethics statement

All participants involved in the clinical study provided written informed consent prior to enrollment. The study protocol was reviewed and approved by the Ethics Committee of the Army Medical University, China (approval number: 2021-179). All human experiments were conducted in accordance with the Declaration of Helsinki and relevant guidelines and regulations, and informed consent was obtained from all participants and/or their legal guardians. Animal experimentation protocols were approved by the Animal Ethics Committee of the Army Medical University, China, and conducted in accordance with the guidelines established by the National Institutes of Health for the Care and Use of Laboratory Animals. All animal experimental procedures are reported in accordance with the ARRIVE guidelines (https://arriveguidelines.org).

Acquisition and standardization of microarray data

The clinical sample sequencing data for sepsis used in this study were downloaded from the publicly accessible GEO database. From the GEO database, we obtained a dataset containing 28-day survival prognosis information for sepsis (GSE65682 dataset: 42 normal samples and 760 sepsis samples) for subsequent research and analysis. Based on the platform annotation file for the GSE65682 dataset (GPL13667), we matched and annotated the gene symbol for each sample in the Perl programming environment. Excluding sepsis samples without 28-day survival prognosis information (survival time and survival status), we extracted 42 normal samples and 479 sepsis samples from the GSE65682 dataset for further analysis. In the R programming environment, we performed the “normalize Between Arrays function” script to standardize the transcriptome data of each sample. Based on the standardized raw data, we identified differentially expressed gene signatures (DEGs) between the healthy control (HC) and the sepsis patient (SP) using the “limma” script, with differential thresholds set at fold change (|FC|) ≥ 2 and p.adjust < 0.05.

Identification of differentially expressed mitochondrial metabolism-related genes (DE-MMRG) and prognostic value analysis

By integrating data from four mitochondrial-related databases (MitoCarta, MitoMiner, IMPI2, and UniProt), we identified and extracted a series of mitochondrial-related gene signatures (Supplementary Table 1). In the Perl programming environment, we extracted metabolism-related genes from the KEGG signaling pathway reference file “c2.cp.kegg.v7.2.symbols” (Supplementary Table 2). Using the “Venn” script, we identified overlapping genes from the differential expression analysis, mitochondrial-related genes, and metabolism-related genes as DE-MMRG for further analysis. The “ggplot2” script was employed to assess the correlation among DE-MMRG. We performed GO and KEGG enrichment analysis on DE-MMRG using the “clusterProfiler” R script to predict potential molecular mechanisms. Utilizing the 28-day prognosis data of 479 sepsis samples, we conducted univariate Cox analysis with the “survival” script to calculate the hazard ratio (HR) and P-value for each DE-MMRG variable, considering p < 0.05 as indicative of prognostic value. The “glmnet” script was used to construct a LASSO function model, screening characteristic MMRG prognostic variables. Finally, multivariate Cox analysis was performed to determine the independent prognostic variables for sepsis.

Identification of MMRG molecular subtype characteristics and immune infiltration status in sepsis

Based on the expression characteristics of independent MMRG prognostic variables, we developed an unsupervised consensus clustering algorithm to assess molecular subtype characteristics of MMRG in sepsis. Utilizing the “ConsensusClusterPlus” R package, we employed the “K-means” algorithm to evaluate the consensus matrix heatmap, cumulative distribution function (CDF), and proportion of ambiguous clustering (PAC) for each k value, with the range of k set between 2 and 9. The analysis was performed with 50 resampling iterations (reps = 50), 80% of samples resampled in each iteration (pItem = 0.8), all features included (pFeature = 1), Euclidean distance, and a fixed random seed (123,456). Based on the optimal consensus clustering model parameters, we randomly divided the sepsis samples into distinct MMRG molecular subtypes. To visualize the distribution and assess the independence of MMRG molecular subtypes, we constructed a principal component analysis (PCA) plot using the “ggplot2” R script. The 28-day clinical survival outcomes of sepsis within MMRG molecular subtypes were evaluated using the “survival” R script. For the assessment of differentially regulated KEGG signaling pathways among MMRG molecular subtypes, we used the “GSVA” R script based on the KEGG signaling pathway reference gene list “c2.cp.kegg.v7.2.symbols”^14–16. The immune score of each sepsis sample was evaluated using the “ESTIMATE” algorithm based on gene transcriptome characteristics. Additionally, we performed quantitative analysis of the relative proportions of 23 immune cell types in sepsis samples using the ssGSEA assessment algorithm, executed through the “GSVA” R script, and based on known marker genes for 23 immune cells (immune cell.gmt).

Construction of MMRG scoring model and development of nomogram diagnostic model in sepsis

Univariate Cox regression analysis was performed to identify MMRG significantly associated with prognosis outcome. The selected prognostic genes were then further refined using Least Absolute Shrinkage and Selection Operator (LASSO) regression to construct a robust set of candidate genes for subsequent risk modeling. Based on the risk coefficients and expression characteristics of independent MMRG prognostic variables calculated from multivariate Cox analysis, we computed the MMRG scores for each sepsis sample. The MMRG scoring formula is as follows: MMRG score = MAOA × 1.0647 + MTHFD2 × 1.8093 + NME4 × 1.1827. Using the optimal cut-off value for 28-day clinical survival outcomes in sepsis samples, we stratified the sepsis samples into low and high MMRG score subgroups. The expression levels of MMRG independent prognostic variables (MAOA, NME4, and MTHFD2) in the MMRG score subgroups were analyzed using the “pheatmap” script. PCA analysis was performed with the “ggplot2” R script to evaluate the distribution of MMRG score subgroups. To validate the reliability and stability of the MMRG scoring system in predicting sepsis survival outcomes, we split the sepsis samples into training and validation sets with a 6:4 ratio using the “caret” R script. The “survival” R script was used to analyze the 28-day survival outcomes in three independent MMRG scoring cohorts, assessing the stability and accuracy of the MMRG scoring system. A Sankey diagram was constructed with the “ggalluvial” R script to explore the potential associations between MMRG molecular subtypes, MMRG scores, and sepsis survival outcomes. Univariate and multivariate Cox analyses were performed using the “survival” R script to calculate the HR and P-values for age, gender, and MMRG score, assessing their independent prognostic value for predicting sepsis outcomes. Based on clinical baseline data (age and gender) and MMRG scores, we developed a nomogram diagnostic model to predict 7-day, 14-day, and 28-day survival probabilities in sepsis using the “rms” script. The “rms” and “pec” R scripts were utilized to calculate the concordance index for clinical baseline characteristics and MMRG scores, evaluating the accuracy of each variable in predicting 28-day survival outcomes.

Single-cell RNA sequencing (scRNA-seq) analysis

In this study, we obtained the scRNA-seq dataset GSE167363 from the GEO database, which includes single-cell transcriptomic data derived from peripheral blood mononuclear cells (PBMCs) of 5 sepsis patients and 2 healthy controls¹⁷. The dataset underwent rigorous quality control, and an expression matrix was generated based on gene counts and unique molecular identifier (UMI) counts. Cells with gene counts ranging from 200 to 5000 and UMI counts less than 30,000 were included, resulting in a total of 38,562 cells meeting these criteria. Next, we identified the top 2000 most variable genes and applied the FindClusters function to partition cells into 23 clusters. Subsequent subcluster analysis was conducted at a resolution of 1.2 to refine cell types. UMAP was employed for two-dimensional visualization of cells, followed by differential expression analysis to identify significantly DEGs across clusters. The FindMarkers function from the Seurat package, utilizing a Wilcoxon rank-sum test, identified DEGs expressed in over 10% of cells within a cluster and with a log-fold change greater than 0.25. To elucidate the functional roles of DEGs, Gene Ontology (GO) analysis was performed using the clusterProfiler R package v4.0.2. Additionally, we constructed a transcription factor regulatory network using the pyscenic v0.11.0 package, integrating transcription factor information from AnimalTFDB with the scRNA expression matrix. Initially, GRNBoost2 was utilized to predict regulatory networks based on co-expression relationships between regulators and targets. Subsequently, CisTarget refined indirect targets and identified transcription factor binding motifs. Finally, AUCell quantified regulator activity in each cell, and high regulon specificity scores (RSS) were visualized using the pheatmap R package for major transcription factor regulons. scMetabolism was employed for single-cell metabolism pathway analysis to complement insights from the transcriptional regulatory network.

Molecular docking for identification of potential drug targets

Molecular docking was performed using CB-Dock2 to analyze the affinity between MTHFD2 and major components of ginseng. CB-Dock2 is an automated blind docking server for protein–ligand interactions, capable of identifying binding sites, calculating pocket centers and sizes, and conducting docking and scoring based on Vina. In this study, the 3D structure of the MTHFD2 protein (PDB ID: 6JIB) was downloaded from the RCSB PDB database (https://www.rcsb.org/), while the structures of small molecule compounds were obtained from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/). The protein and ligand structures were uploaded to the CB-Dock2 website (https://cadd.labshare.cn/cb-dock2/php/index.php) and subjected to blind docking using default parameters. Conformations were evaluated based on Vina scores, with the lowest score indicating the most stable conformation. Visualization was performed using PyMOL software (Version 2.5.0) for further analysis and interpretation of docking results.

Study design and participant recruitment

Between December 2021 and April 2022, we recruited a cohort comprising 30 septic patients from the Intensive Care Unit (ICU) at Daping Hospital, Army Medical University, based on the Sepsis-3 criteria for sepsis and septic shock. Additionally, 15 age-matched healthy volunteers were enrolled from the State Key Laboratory of Trauma, Burns, and Combined Injury. Blood samples were collected from all septic patients within 24 h of admission, while blood samples from healthy volunteers were obtained at the time of enrollment.

Blood sample collection and metabolite extraction

A 10 mL peripheral venous blood sample was collected from each participant in both groups. The samples were then centrifuged at 3000 rpm for 10 min, and the resulting supernatant was aliquoted into 1.5 mL centrifuge tubes and stored at -80 °C. For analysis, 100 μL of each sample was transferred to an EP tube. To this, 400 μL of extraction solution (methanol containing an isotopically labeled internal standard mixture) was added, followed by 30 s of vortexing, 10 min of sonication in an ice-water bath, and 1 h of incubation at  −  40 °C to precipitate proteins. Subsequently, the samples were centrifuged at 12,000 rpm (RCF = 13,800 × g, R = 8.6 cm) for 15 min at 4 °C. The resulting supernatant was then transferred to a fresh glass vial for analysis. A quality control (QC) sample was prepared by combining equal aliquots of the supernatants from all individual samples.

LC–MS/MS analysis

LC–MS/MS analyses were performed using a UHPLC system (Vanquish, Thermo Fisher Scientific) coupled with an Orbitrap Exploris 120 mass spectrometer (Orbitrap MS, Thermo) equipped with a UPLC HSS T3 column (2.1 mm × 100 mm, 1.8 μm). The mobile phase comprised 5 mmol/L ammonium acetate and 5 mmol/L acetic acid in water (A) and acetonitrile (B). The auto-sampler temperature was maintained at 4 °C, and the injection volume was 2 μL. The Orbitrap Exploris 120 mass spectrometer was selected for its ability to acquire MS/MS spectra in information-dependent acquisition (IDA) mode, controlled by Xcalibur software (Thermo). In IDA mode, the acquisition software continuously evaluates the full scan MS spectrum. The electrospray ionization (ESI) source conditions were set as follows: sheath gas flow rate at 50 Arb, auxiliary gas flow rate at 15 Arb, capillary temperature at 320 °C, full MS resolution at 60,000, MS/MS resolution at 15,000, collision energy at 10/30/60 in normalized collision energy (NCE) mode, and spray voltage at 3.8 kV (positive) or -3.4 kV (negative), respectively.

Data preprocessing and annotation

The raw data were converted to the mzXML format using ProteoWizard and subsequently processed using a custom R program developed in-house. This program utilized XCMS for peak detection, extraction, alignment, and integration. Metabolite annotation was performed using an internal MS2 database named BiotreeDB, with an annotation cutoff threshold set at 0.3.

Sepsis model establishment and treatment

Adult Sprague–Dawley rats weighing 200–220 g were housed in the animal facility with ad libitum access to food and water. Rats were randomly assigned to three groups: Sham, CLP, and Ginsenoside Rb1 (Grb1)-treated CLP (CLP+Grb1) group (HY-N0039, MedChemExpress). To induce sepsis, CLP were performed according to previously published methods¹⁸. Grb1 (10 mg/kg) was administered via the tail vein 30 min before sepsis induction, while the CLP group received an equivalent volume of ddH2O by the same route. Survival time was recorded up to 72 h post CLP induction. At 12 h post-CLP induction, lung tissues were harvested for wet/dry weight ratio measurement, and blood samples were collected for qPCR to detect inflammatory factors, ALT, and Creatine levels.

Anesthesia and euthanasia procedures

For in vivo experiments, mice were anesthetized with pentobarbital sodium (40 mg/kg, intraperitoneally) to minimize suffering. At the endpoint, animals were euthanized using CO₂ inhalation followed by cervical dislocation, in accordance with the institutional and international guidelines for the care and use of laboratory animals.

Determination of blood inflammatory cytokine levels

At 12 h post-CLP induction, 2 mL of abdominal aortic blood was collected from rats, allowed to clot at room temperature for 30 min, and then centrifuged at 1000 × g for 20 min. The resulting upper serum was collected for analysis. Each well of an ELISA strip was prepared with either a standard substance or sample. ELISA kits specific for interleukin-1β (IL-1β) (E-EL-R0012c), interleukin-6 (IL-6) (E-EL-R0015c), and tumor necrosis factor-α (TNF-α) (E-EL-R2856c) were procured from Elabscience (China). The concentrations of IL-6, IL-1β, and TNF-α in rat serum were quantified using the ELISA method.

Statistical analysis

Data extraction and preprocessing in the bioinformatics section of this study were performed using R software (version 4.1.0) and Perl language environment. The Pearson correlation algorithm was employed to assess the degree of correlation between two variables. Statistical differences between two groups were evaluated using Student’s t-test and Wilcoxon rank-sum test methods. Experimental data analysis was conducted using GraphPad Prism software (version 8.0.1) and SPSS 17.0 (SPSS Inc., Chicago, IL, USA) for statistical and visual analysis. PCR data were replicated in three independent experiments. To account for multiple comparisons, all P values obtained from statistical tests were adjusted using the Benjamini–Hochberg (BH) false discovery rate (FDR) method, and adjusted P values (FDR < 0.05) were considered statistically significant. Data are presented as mean ± SD, and significance symbols denote: *p < 0.05, **p < 0.01, ***p < 0.001.

Result

Differential expression analysis of mitochondrial metabolism-related gene signature in sepsis

Using the GSE65682 dataset, we analyzed the potential roles of mitochondrial metabolism-related genes in sepsis, comprising 42 normal samples and 479 sepsis blood samples with 28-day survival prognosis information. Differential expression criteria were set at |FC|≥ 2 and p.adjust < 0.05. We evaluated the differential gene signatures between the HC group and SP group (Fig. 1A). Visualization with a heatmap (Fig. 1B) illustrated the expression characteristics of differential gene signatures between the HC and SP groups. Utilizing a Venn diagram to analyze overlapping genes among mitochondrial genes, metabolism-related genes, and differentially expressed genes, we identified 12 DE-MMRG signatures in sepsis blood (Fig. 1C). Correlation analysis revealed significant associations among these 12 DE-MMRG; for example, ACSL1 was significantly positively correlated with ACSL4, and PGS1 was significantly positively correlated with ACSL1 and ACSL4 (Fig. 1D). GO enrichment analysis based on the 12 DE-MMRG indicated their potential involvement in biological functions such as purine ribonucleotide metabolic process, ribonucleotide metabolic process, mitochondrial matrix, and mitochondrial outer membrane (Fig. 1E). KEGG pathway analysis suggested that these MMRG may mediate pathways including Purine metabolism, Biosynthesis of cofactors, and Fatty acid biosynthesis in sepsis (Fig. 1F).

Fig. 1.

Identification of differential expression MMRG signature and enrichment analysis of potential molecular mechanisms. (A) Volcano plot depicting differential gene expression between the HC and SP groups. Differential criteria: |fold change|≥ 2 and p.adjust < 0.05. Red dots indicate upregulation in the SP group, blue dots indicate downregulation in the SP group, and gray dots represent no significant difference. (B) Heatmap visualization of differential gene expression. (C) Venn diagram analysis of overlapping genes among metabolism-related genes, mitochondrial genes, and differentially expressed genes. (D) Analysis of potential interactions among DE-MMRG genes. (E, F) GO and KEGG enrichment analysis of DE-MMRG.

Identification of MMRG-based molecular subtype features and analysis of immune infiltration characteristics in sepsis

Based on the 28-day survival prognosis information of sepsis samples and the expression characteristics of 12 DE-MMRG, we evaluated the potential associations of each DE-MMRG variable with clinical prognosis in sepsis. Using the LASSO-univariate Cox analysis algorithm, we identified 5 MMRG feature variables significantly associated with 28-day survival prognosis in sepsis (Supplementary Fig. 1). Among these prognostic MMRG feature variables, one beneficial variable correlated positively with clinical prognosis in sepsis (FHIT, HR < 1), while four risk variables were associated with adverse prognosis in sepsis (FECH, MAOA, MTHFD2, and NME4, HR > 1). Furthermore, employing multivariate Cox analysis, we identified 3 independent prognostic variables associated with clinical prognosis in sepsis, including MAOA, MTHFD2, and NME4. Based on the expression characteristics of these three independent prognostic variables, we performed unsupervised consensus clustering analysis to randomly classify sepsis samples into two MMRG molecular subtypes, with MMRG subtype A comprising 337 sepsis samples and MMRG subtype B comprising 142 sepsis samples (Fig. 2A–C). Principal component analysis (PCA) plots revealed significant separation in distribution patterns between MMRG subtypes A and B, confirming the accuracy and independence of these two MMRG molecular subtype classifications (Fig. 2D). Clinical prognosis curves for 28-day survival indicated that sepsis samples in MMRG subtype A had longer survival rates compared to those in subtype B, suggesting better clinical survival benefits in MMRG subtype A (Fig. 2E). Additionally, higher immune scores were observed in MMRG subtype A, indicating a potentially greater immune activation status associated with clinical benefits in sepsis (Fig. 2F). Using the GSVA algorithm, we evaluated the potential regulatory molecular mechanisms between MMRG molecular subtypes. KEGG pathway analysis revealed significant upregulation of the spliceosome signaling pathway in MMRG subtype A, whereas metabolic pathways including nitrogen metabolism, histidine metabolism, arginine and proline metabolism, and phenylalanine metabolism were significantly upregulated in the poorer prognosis MMRG subtype B (Fig. 2G). Immunological cell infiltration analysis based on ssGSEA revealed significant differences in immune microenvironment between MMRG molecular subtypes. MMRG subtype A exhibited higher infiltration levels of activated CD8⁺ T cells, eosinophils, immature B cells, and myeloid-derived suppressor cells (MDSCs), while subtype B showed higher infiltration levels of activated dendritic cells, macrophages, and type 17 T helper cells, indicating a more immunosuppressive environment (Fig. 2H). In conclusion, our study identifies independent prognostic variables based on MMRG that accurately stratify sepsis samples into two distinct MMRG molecular subtypes with different survival prognoses, and provides initial insights into the potential relationship between immune infiltration status and clinical prognosis in sepsis.

Fig. 2.

Identification of MMRG Molecular Subtypes and Assessment of Immune Infiltration Characteristics in Sepsis. (A-C) Unsupervised consensus clustering analysis based on independent prognostic variables of MMRG. Optimal classification of MMRG molecular subtypes was determined using consensus matrix heatmap (k = 2 ~ 9), CDF curves, and delta area curves. (D) Principal component analysis (PCA) of MMRG molecular subtypes. (E) Analysis of 28-day clinical survival prognosis curves of MMRG molecular subtypes. (F) Differential analysis of immune scores across MMRG molecular subtypes. (G) Differential analysis of KEGG pathway enrichment based on GSVA algorithm. (H) Characterization of immune cell infiltration landscapes based on ssGSEA algorithm.

MMRG prognostic signature development of sepsis

Based on the expression profiles and hazard ratios of three independent MMRG prognostic variables (MAOA, MTHFD2, and NME4), we developed a novel MMRG scoring system to stratify risk and assess 28-day clinical survival outcomes in sepsis samples. Using the optimal cutoff for 28-day clinical survival outcomes, sepsis samples were categorized into two MMRG score subgroups (Fig. 3A, B). Principal component analysis (PCA) revealed distinct distribution patterns between MMRG score subgroups (Fig. 3C). The 28-day clinical survival curve results indicated that sepsis samples in the low MMRG score subgroup had significantly better survival rates compared to those in the high MMRG score subgroup (Fig. 3D, p < 0.001). Expression profiling showed significantly elevated levels of MAOA, MTHFD2, and NME4 in the high MMRG score subgroup with poorer prognosis (Fig. 3E). Integrating clinical-pathological features (age and gender) of sepsis samples, we evaluated the independent prognostic value of MMRG scores and clinical-pathological characteristics using univariate and multivariate Cox regression analyses. Univariate Cox analysis revealed that age (HR = 1.021 (1.007–1.035), p = 0.004) and MMRG score (HR = 1.586 (1.359–1.850), p < 0.001) were associated with adverse prognosis in sepsis (Fig. 3F). Further multivariate Cox analysis identified MMRG score (HR = 1.580 (1.352–1.847), p < 0.001) as an independent prognostic factor for sepsis, distinguishing it from other pathological features in predicting clinical outcomes (Fig. 3G). Using clinical variables and MMRG scores, we constructed a nomogram diagnostic model to assess the probabilities of 7-day, 14-day, and 28-day survival in sepsis (Fig. 3H). C-index curve results demonstrated that the accuracy of predicting clinical survival outcomes in sepsis using MMRG scores was significantly superior to other clinical-pathological features (Fig. 3I). Based on these findings, we conclude that the MMRG scoring system developed based on three independent prognostic variables accurately stratifies risk in sepsis samples. Additionally, as an independent prognostic factor, the MMRG scoring system distinguishes itself from other pathological features in indicating clinical survival outcomes in sepsis.

Fig. 3.

Risk stratification and independent prognostic analysis based on MMRG Score in Sepsis. (A, B) Subgroup division based on MMRG score. (C) PCA plot analysis of MMRG score subgroups. (D) 28-day clinical survival prognosis curve analysis based on MMRG score subgroups. (E) Expression profile analysis of MMRG prognostic variables. (F, G) Univariate and multivariate Cox regression analyses evaluating the independent prognostic value of age, gender, and MMRG score. (H) Construction of a nomogram diagnostic model based on clinical-pathological features and MMRG score. (I) Concordance index analysis of MMRG score and clinical-pathological features.

Validation of independence of MMRG score predicting clinical prognosis in sepsis

To validate the independence of MMRG score in predicting clinical survival prognosis in sepsis, we divided 479 sepsis samples in a 6:4 ratio into two independent cohorts. The training set comprised 288 samples, while the validation set comprised 191 samples. Within the MMRG molecular subgroups, the MMRG score was significantly higher in the clinically poorer prognosis MMRG subtype B compared to subtype A, further elucidating the association of high MMRG scores with adverse sepsis prognosis (Fig. 4A). Using Sankey diagrams, we demonstrated the potential associations between MMRG molecular subtypes, MMRG score subgroups, and clinical survival outcomes in sepsis. Results indicated that the more favorable prognosis MMRG subtype A exhibited a higher proportion of low MMRG scores, correlating with better clinical outcomes (Fig. 4B). Kaplan–Meier survival curve analysis at 28 days showed significantly lower clinical survival prognosis in high MMRG score subgroups compared to low MMRG score subgroups in both the training and validation sets (Fig. 4C, D). Additionally, PCA plot results based on MMRG prognostic signature in both cohorts demonstrated significant separation between MMRG score subgroups (Fig. 4E, F). Based on these findings, we conclude that risk stratification based on MMRG score accurately assesses clinical prognosis in sepsis, with higher MMRG scores correlating with adverse outcomes.

Fig. 4.

Development and Independence Validation of MMRG Score Model in Independent Cohorts. (A) Differential analysis of MMRG scores among MMRG molecular subtypes. (B) Sankey diagram illustrating the associations between MMRG molecular subtypes, MMRG scores, and sepsis survival prognosis. (C, D) Kaplan–Meier survival curve analysis of MMRG score subgroups in the training and validation sets. (E, F) PCA plot analysis of MMRG score subgroups in both cohorts.

MMRG score subgroup immune microenvironment infiltration characteristics and correlation analysis

Subsequently, we further assessed the potential relationship between MMRG scores and immune infiltration characteristics. GSVA results indicated significant downregulation of several immune-related signaling pathways in the high MMRG score subgroup, such as the T cell receptor signaling pathway, B cell receptor signaling pathway, and chemokine signaling pathway, while metabolism-related pathways like Glutathione metabolism and Porphyrin and chlorophyll metabolism were significantly upregulated in the high MMRG score subgroup (Fig. 5A). Quantitative results of immune cell infiltration showed significant correlation of decreased immune cell proportions in the high MMRG score subgroup, including activated B cells, activated CD8 T cells, neutrophils, and natural killer T cells (Fig. 5B). ESTIMATE assessment results indicated a significant decrease in immune scores in the poor prognosis high MMRG score subgroup (Fig. 5C). These findings indicate that high MMRG scores are statistically associated with an immunosuppressed profile and poorer clinical prognosis, but they do not establish that MMRG scores directly regulate immune cell infiltration. Through correlation analysis, we elucidated the relationship between MMRG scores and immune infiltration characteristics in sepsis. Correlation analysis revealed significant associations between MMRG scores and most immune cells; for instance, MMRG scores were positively correlated with activated dendritic cells, CD56bright natural killer cells, and type 17 T helper cells, while negatively correlated with activated B cells, regulatory T cells, and plasmacytoid dendritic cells (Fig. 5D). Additionally, we identified significant correlations between three independent prognostic variables of MMRG and immune cell infiltration characteristics in sepsis (Fig. 5E). Our analysis describes the immune infiltration patterns across MMRG score subgroups and their statistical association with clinical outcomes; however, these observations remain correlative and do not imply a causal relationship.

Fig. 5.

Assessment of Immune Infiltration Characteristics and Correlation Analysis in MMRG Score Subgroups. (A) KEGG signaling pathway analysis based on GSVA in MMRG score subgroups. (B) Relative proportions assessment of 23 immune infiltrating cells in MMRG score subgroups. (C) ESTIMATE immune score estimation. (D, E) Correlation analysis between MMRG scores, independent prognostic variables of MMRG, and immune cells in sepsis.

Single-cell sequencing analysis reveals the expression landscape of MMRG in peripheral blood mononuclear cells (PBMCs) of sepsis

Building upon our previous research, we initially elucidated the potential role of MMRG in immune infiltration characteristics of sepsis. To further clarify the role of MMRG in sepsis immune infiltration, we utilized the scRNA-seq dataset GSE167363 (2 HC samples, 5 SP samples) to evaluate the expression characteristics of MMRG in sepsis PBMCs. Following quality control and data cleaning of the scRNA-seq sequencing, we obtained 38,562 high-quality single-cell clusters and performed standardized processing and unsupervised dimensionality reduction analysis. Using the SynEcoSys database, we analyzed marker genes in each cell subtype and identified seven distinct cell types, including plasmacytoid dendritic cells (pDCs), erythrocytes, neutrophils, platelets, monocytes, T cells, and B cells. The UMAP model depicted the expression distribution of three MMRG prognostic genes across each cell subtype, highlighting that MTHFD2, most correlated with sepsis prognosis risk, exhibited higher expression abundance across the seven cell subtypes, suggesting its potential role as a critical regulator of sepsis immune function (Fig. 6A-C). Further analysis of MTHFD2 expression across the seven cell subtypes revealed significantly higher expression in platelets compared to other cell subtypes, implying that MTHFD2 may be involved in sepsis immune infiltration changes through regulation of platelet function (Fig. 6D-F). To clarify the potential regulatory mechanisms of MTHFD2 in platelets, we categorized platelets into two subgroups based on MTHFD2 expression levels and identified differentially expressed genes between the two subgroups (Fig. 6G). KEGG enrichment analysis indicated that these differentially expressed genes between platelet subgroups are involved in regulating pathways such as ribosome, platelet activation, oxidative phosphorylation, and chemokine signaling pathways (Fig. 6H). Additionally, we further screened potential transcription factors that may regulate high expression of MTHFD2, revealing candidates such as SRF, H2AFY, HMGN3, and UQCRB (Fig. 6I). Our single-cell findings highlight a notable association between MTHFD2 expression and platelet-related transcriptional programs. While this raises the hypothesis that MTHFD2 may influence platelet functional states in sepsis, this possibility remains speculative and requires direct functional validation.

Fig. 6.

Single-cell Sequencing Analysis of MMRG Prognostic Signature. (A–C) Identification of cell subtypes and expression localization analysis of MMRG in sepsis. (D) Calculation of MMRG expression abundance across different cell subtypes. (E, F) Analysis of MTHFD2 expression levels in platelets. (G) Differential analysis of low and high MTHFD2 expression subgroups in platelets. (H) KEGG signaling pathway enrichment analysis of differentially expressed genes in platelet subgroups. (I) Transcription factor regulation analysis of low and high MTHFD2 expression subgroups in platelets.

Analysis of potential metabolic pathways in cell subtypes

Based on scRNA-seq data from sepsis, we utilized the “scMetabolism” algorithm to assess metabolic pathway activity across seven cell subtypes. Notably, platelets exhibited a heightened metabolic activation compared to the other six cell subtypes, suggesting that MTHFD2 may potentially regulate metabolic activity through platelet activation (Fig. 7A). Metabolic pathway analysis highlighted distinct regulation of metabolic pathways across the seven cell subtypes. In the platelet subgroup, we observed significant upregulation of pathways including phenylalanine metabolism, purine metabolism, pyruvate metabolism, and glutathione metabolism (Fig. 7B). Subsequently, metabolomics analysis further validated metabolic pathway changes between clinical samples from HC and SP. Differential pathway analysis based on metabolites between the two groups indicated that phenylalanine metabolism and cysteine and methionine metabolism may be critical pathways regulating sepsis (Fig. 7C). Based on these findings, we conclude that high expression of MTHFD2, through its potential role in platelet activation, significantly alters metabolic pathways in platelets, thereby influencing the occurrence and progression of sepsis.

Fig. 7.

Single-cell Sequencing Reveals Metabolic Pathways in Cell Subtypes. (A) Assessment of metabolic activity in cell subtypes. (B) Analysis of differentially regulated metabolic pathways across 7 cell subtypes in PBMCs. (C) Pathway enrichment analysis based on metabolomics profiling of clinical samples from HC and SP groups.

Grb1 inhibition of MTHFD2 expression improves rat survival outcome and suppresses inflammatory response

Utilizing the TCMSP and TCMIP databases, we identified a series of drug targets interacting with MTHFD2. Employing CB-Dock2 molecular docking, we found that MTHFD2 interacts most significantly with Grb1, suggesting Grb1 as a critical pharmacological target for MTHFD2 (Fig. 8A). mRNA analysis indicated a significant increase in MTHFD2 expression in the CLP rat model, which was markedly suppressed upon Grb1 administration (Fig. 8B, p < 0.05). Additionally, survival analysis at 72 h showed a notable extension in survival time in the CLP+Grb1 group compared to the CLP group, indicating that Grb1 improves survival outcomes in CLP rats (Fig. 8C, D). Further evaluation of the potential mechanisms underlying Grb1’s improvement in survival outcomes in CLP rats revealed significant reductions in the levels of IL-1β, IL-6, and TNF-α in the CLP+Grb1 group (Fig. 8E-G). Compared to the CLP group, Grb1 administration also significantly reduced the wet/dry ratio in the left lung (Fig. 8H) and decreased levels of creatinine and ALT in the CLP group (Fig. 8I-J). Based on these findings, we identified Grb1 as a potential drug target for regulating MTHFD2 and preliminarily elucidated that Grb1 may suppress the inflammatory response in CLP rats by downregulating MTHFD2 expression, thereby improving organ function and survival outcomes in sepsis. These findings provide new insights and potential therapeutic targets for future clinical management of sepsis.

Fig. 8.

Grb1 significantly reduces inflammatory levels and improves survival outcome in CLP rats. (A) Identification of MTHFD2 drug targets using molecular docking techniques. (B) mRNA expression analysis of MTHFD2 in control, CLP, and CLP+Grb1 groups (n = 6). (C, D) Survival analysis of control, CLP, and CLP+Grb1 groups (n = 6). (E–G) Quantitative analysis of IL-1β, IL-6, and TNF-α levels in control, CLP, and CLP+Grb1 groups (n = 6). (H) Wet/dry ratio analysis of the left lung (n = 6). (I, J) Quantitative analysis of creatinine and ALT levels in CLP and CLP+Grb1 groups (n = 6). Data are presented as mean ± SD; statistical significance: *p < 0.05, **p < 0.01, ***p < 0.001.

Discussion

Through bioinformatics analysis and subsequent animal model validation, we successfully established an MMRG-related prognostic model and explored its potential clinic value in this study. We identified MTHFD2 as a potential intervention target and highlighted Grb1 as a key pharmacological target interacting with MTHFD2. Subsequent rat model experiments further confirmed that Grb1 may improve organ function and survival outcomes in sepsis by downregulating MTHFD2 expression to suppress inflammatory responses.

A mainstream view on the fundamental cause of mortality in severe sepsis is immune suppression¹⁹. Immunosuppressed patients are highly susceptible to opportunistic nosocomial pathogens in the ICU and may succumb rapidly to infections^20,21. Our immune infiltration results indicated that sepsis with high MMRG scores generally exhibit significantly lower levels of immune response-related components compared to another group, including activated and immature B cells, CD8⁺ T cells, NK cells, macrophages, and eosinophils. However, we observed a significant increase in activated dendritic cells (DCs) in the subgroup with poorer prognosis and high MMRG scores. The involvement of DCs in immune dysregulation following sepsis onset is well recognized^22,23. During sepsis progression, both the function and quantity of DCs undergo significant changes²⁴. While sepsis generally experiences decreased DC counts, there are no reports of differences in DC content among sepsis risk stratifications^25–27. Dysregulation of DCs is characterized by abnormal surface molecule expression, reduced cytokine secretion, and diminished T cell stimulatory capacity^28,29. Our analysis did not fully elucidate the functional status of DCs, leaving open the possibility of high DC levels with reduced functionality, necessitating further research.

Furthermore, our results showed that immunosuppressive cells such as regulatory T cells (Tregs) and myeloid-derived suppressor cells (MDSCs) were significantly lower in the high MMRG score group compared to another group. The widely accepted conclusion is that an increased proportion of MDSCs in peripheral blood of sepsis correlates positively with secondary infection risk and hospitalization time^30–32. The reasons for our differing results remain unclear. One possibility is the metabolic association of MDSCs with arginine³³. Given the critical role of arginine in mitochondrial metabolism, our MMRG classification inevitably introduces bias³⁴. Moreover, MDSCs are mainly divided into granulocytic/polymorphonuclear MDSCs (PMN-MDSCs) and monocytic MDSCs (M-MDSCs). Apart from difficulties in phenotypic distinction resembling neutrophils or monocytes, different types of infections leading to sepsis also appear to affect MDSC expression^35,36. Tregs exhibit heterogeneity in sepsis³⁷. In the early stages of sepsis, non-survivors had lower absolute numbers of Tregs compared to survivors³⁷. Increasing the proportion and absolute number of Tregs is crucial for restoring immune-inflammatory homeostasis, reducing tissue damage, and organ injury^38–40. As the disease progresses, survivors exhibit a decline in Treg percentages, potentially related to immune response function against infections, making Tregs a potential intervention target for sepsis³⁷.

Our multivariate Cox analysis identified MTHFD2 as having the highest risk coefficient. MTHFD2 is a key enzyme involved in mitochondrial folate one-carbon metabolism (OCM)⁴¹. Its typical role includes mediating folate-dependent mitochondrial OCM and contributing to the synthesis of formyl methionyl transfer RNA required for protein synthesis initiation^42,43. OCM contributes to nucleotide synthesis, NADPH production, and oxidative stress defense⁴⁴. Additionally, OCM regulates microbial virulence modulation⁴⁵. Dysfunction in OCM precursor synthesis limits microbial pathogenicity⁴⁵. Furthermore, OCM plays a significant role in T cell function during immune responses⁴⁵. Due to its role in maintaining OCM pathway function and responding to growth factor stimulation, MTHFD2 emerges as a promising therapeutic target for OCM⁴⁶. While MTHFD2 has been reported in cancer, autoimmune disorders, and inflammatory diseases, its relevance to sepsis has never been systematically explored. No study to date has integrated large-scale transcriptomics, single-cell profiling, metabolism-related signatures to investigate the potential contribution of MTHFD2 to sepsis. Our findings suggest its critical role in sepsis prognosis and potential antioxidant function⁴⁷, warranting further exploration.

Grb1is a major pharmacological extract from ginseng⁴⁸. Our results identified Grb1 as the optimal responder to MTHFD2, suggesting its potential therapeutic value in sepsis treatment. Previous studies have focused on Grb1 as an antioxidant and metabolic regulator, protecting the nervous and cardiovascular systems, improving blood glucose levels, and insulin resistance^49–51. Grb1 exerts these effects by regulating multiple mitochondrial-related biological processes, positioning mitochondria as key targets for Grb1⁵². Additionally, reports indicate Grb1 inhibits the inflammatory cascade of sepsis⁵³. Its mechanism may involve downregulating TLR4 mRNA expression and inhibiting TNF-α production⁵⁴. Moreover, Grb1 functions as an antioxidant and metabolic regulator to improve arterial blood pressure and protect vital organs^50,54. Therefore, Grb1 holds promise in sepsis treatment and merits further exploration.

MTHFD2 showed transcriptional enrichment in platelet-related pathways, suggesting a possible hypothesis regarding platelet activation, though this remains entirely speculative without functional validation. Platelet activation plays a crucial role in sepsis development. Infections lead to thrombin generation, resulting in platelet activation⁵⁵. During sepsis development, endothelial cell damage due to platelet activation promotes neutrophil extracellular traps and microthrombus formation⁵⁶. Enhanced pyogenic coagulation and inflammatory responses contribute to disseminated intravascular coagulation (DIC), exacerbating organ damage⁵⁷. Additionally, platelets stimulate and regulate immune cells such as neutrophils, monocytes, and macrophages during sepsis onset⁵⁸. Evidence suggests a significant correlation between the degree of platelet reduction and sepsis outcomes⁵⁹. Therefore, antiplatelet drugs hold promise as sepsis treatment^60,61. However, limited data accumulation currently necessitates further clinical research to validate their effects.

Our study provides several layers of novelty that extend beyond the existing knowledge of mitochondrial dysfunction in sepsis. First, by integrating bulk transcriptomics, prognostic modeling, immune infiltration profiling, metabolomics, and single-cell RNA sequencing, we conducted the first systematic multi-omics characterization of MTHFD2 in sepsis. This approach enabled us to uncover robust associations between MTHFD2 expression, immune-metabolic remodeling, and platelet-related transcriptional programs—relationships that have not been previously reported in any sepsis-related study. Second, although MTHFD2 has been studied in other metabolic or inflammatory diseases, its functional relevance in sepsis has remained unexplored. Our in vivo findings provide preliminary evidence that MTHFD2 may participate in sepsis progression, offering a new rationale for considering MTHFD2 as a potential therapeutic target. Finally, by proposing a hypothesis that links mitochondrial one-carbon metabolism, immune dysregulation, and platelet activation, this work generates a conceptual framework that may guide future mechanistic investigations and the development of targeted interventions. Together, these findings add a novel dimension to the understanding of sepsis pathophysiology and offer clinically relevant directions for future translational research.

Several limitations of the present study should be acknowledged. First, although our analyses suggest associations between MTHFD2 and immune infiltration, metabolic alterations, and platelet-related transcriptional programs, these findings are primarily derived from observational data and therefore remain correlative in nature, without establishing direct mechanistic causality. The available experimental data provide only limited functional indications and are insufficient to delineate the precise regulatory pathways or downstream effectors of MTHFD2 in sepsis. In particular, the absence of targeted cellular mechanistic investigations, such as MTHFD2 knockdown, overexpression, or rescue experiments in relevant immune cell populations or platelets, restricts causal interpretation, and animal models cannot fully substitute for cell-specific functional validation. In addition, the prognostic model was constructed based on a single public cohort, which may introduce cohort-dependent bias and limits its generalizability, underscoring the need for validation in independent and preferably multi-center datasets. Moreover, the proposed involvement of MTHFD2 in platelet activation is mainly inferred from single-cell transcriptional profiles and should be interpreted with caution in the absence of direct functional platelet assays. Finally, the molecular docking analyses are exploratory and hypothesis-generating in nature, requiring experimental confirmation through appropriate biochemical or biophysical approaches; likewise, the lack of MTHFD2-specific rescue experiments in vivo precludes definitive conclusions regarding whether MTHFD2 represents the principal mediator of Ginsenoside Rb1–associated effects in sepsis.

Conclusion

This study integrates multi-omics analyses with preliminary animal validation to explore the potential role of MMRG in sepsis, highlighting MTHFD2 as a candidate of interest. Our findings generate hypotheses regarding the association of MTHFD2 with immune infiltration patterns, metabolic alterations, and platelet-related transcriptional programs, and suggest that Grb1 may modulate these pathways. However, all mechanistic interpretations remain exploratory and are based primarily on correlative analyses. Further in-depth mechanistic studies will be essential to determine whether these associations reflect causal biological processes. These exploratory findings provide a foundation for future research aimed at elucidating the immune-metabolic mechanisms of sepsis and identifying potential therapeutic targets.

Author contributions

Qi Shu, Huan Luo and Like Zhong contributed the raw data collection. Jiaping Mo and Xiaoyan Wei conceived the original ideas and composed this manuscript. Zhenjie Zhu, Xiaojuan Chen, Fugen He and Junfeng Zhu contributed the table and figures of this manuscript. Han She and Qi Shu designed and completed the experimental component. Qi Shu, Huan Luo and Like Zhong contributed equally to this article and all authors contributed to the article and approved the submitted version.

Funding

This study was supported by Natural Science Foundation of Chongqing, China (CSTB2022NSCQ-MSX1509); Science and Technology Project of Traditional Chinese Medicine in Zhejiang Province (2023007693); Zhejiang Province Medical and Health Science and Technology Project (2024KY1407, 2022KY279).

Data availability

The datasets used during the current study are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

The study protocol was approved and reviewed by the Ethics Committee of the Army Medical University, China (approval number: 2021–179). Animal experimentation protocols were approved by the Laboratory Animal Welfare and Ethics Committee of Army Medical University (Approval No. AMUWEC20224867).

Consent for publication

Not applicable.