The reagents and chemicals used in this study are listed in Supplementary Table S4.

The murine macrophage line RAW264.7 was pretreated with GAA (20 or 40 μM) for 1 h, followed by induction of inflammation with TNFα (50 ng/mL) for 12 h. GAA was prepared in dimethyl sulfoxide (DMSO), with a vehicle control containing an equivalent DMSO concentration (<0.1%). All cultures were maintained at 37 °C in a 5% CO₂ atmosphere.

An SRLI model was generated in male C57BL/6 mice (8 weeks old; n = 5 mice per group) as reported (Liu et al., 2025). GAA (20 or 40 mg/kg) or vehicle was administered daily by oral gavage for 3 days prior to lipopolysaccharide (LPS) challenge (10 mg/kg, i.p.). Mice were monitored for 10 h post-LPS injection before tissue collection under sodium pentobarbital anesthesia. The TNFα antibody group was administered 15 mg/kg anti-TNFα-antibody intraperitoneally along with the LPS injection. All animal studies were conducted per the approved protocol (Shanghai Tongren Hospital A2025-041-01).

ETCM (http://www.tcmip.cn/ETCM), ChEMBL (https://www.ebi.ac.uk/chembl/) and SwissTargetPrediction (http://swisstargetprediction.ch/) were used to predict GAA targets. Using the terms “sepsis-related liver injury” and “septic acute liver injury,” therapeutic targets associated with SRLI were obtained from GeneCards (https://www.genecards.org/) (relevance score ≥10) and OMIM (https://www.omim.org/). To identify the key genes, we first converted the intersecting targets of GAA and SRLI into official gene symbols. We then constructed a protein-protein interaction (PPI) network based on these targets using a minimum interaction score of 0.4 (medium confidence). Subsequently, we conducted functional enrichment analysis on the common targets to systematically explore the core signaling pathways involved in GAA’s treatment of SRLI.

Bioinformatics Analysis of Transcriptomic Data The gene expression dataset GSE217695 was obtained from the Gene Expression Omnibus (GEO) database. This dataset includes liver tissue transcriptomic profiles from wild-type (WT) mice and mice with LPS-induced SRLI. Raw gene count matrices were obtained from the GEO dataset and pre-processed using the TMM (Trimmed Mean of M-values) method implemented in the edgeR package to correct for differences in sequencing depth and library size. Subsequently, the voom function in the limma package was applied to transform count data into log2-counts per million (logCPM) and estimate the mean-variance relationship for precision weighting. Differential expression analysis was then performed using linear models in limma. Genes with an absolute log2 fold change (Log2FC) ≥ 1 and an adjusted P value (FDR) < 0.05 were identified as differentially expressed genes (DEGs). Volcano plots and heatmaps were generated using the ggplot2 and pheatmap packages in R, respectively, to visualize the expression patterns.

We obtained the crystal structures of TNFα, NF-κB1, CASPASE-3, and MAPK3 from the Protein Data Bank (PDB; https://www.rcsb.org/). Next, we downloaded the 3D structure of GAA from PubChem (https://pubchem.ncbi.nlm.nih.gov/). Finally, we performed molecular docking simulations between GAA and each target using AutoDock Vina 1.1.2 and visualized the docking results with Discovery Studio 2019 Client 19.1 software (BIOVIA, San Diego, CA, United States).

RAW264.7 cells were seeded in a 96-well plate at a density of 1 × 10^4 cells per well and incubated at 37 °C with 5% CO₂. After treatment with GAA at concentrations of 0, 2.5, 5, 10, 20, 40, or 80 μM for 24 h, 10 μL of CCK-8 reagent was added to each well, and absorbance was measured at 450 nm. All procedures were performed according to the manufacturer’s instructions for the Cell Counting Kit-8.

To evaluate macrophage polarization phenotypes, RAW264.7 cells were harvested after the indicated treatments and washed twice with cold PBS. To exclude dead cells from the analysis, cells were first stained with BD Horizon™ Fixable Viability Stain 780 for 15 min at room temperature in the dark. After washing, cells were incubated with an Fragment crystallizable receptor (FcR) blocking solution for 10 min at 4 °C to minimize non-specific binding. For surface marker analysis, cells were stained with PE-Cy7-conjugated anti-CD86 antibody for 30 min at 4 °C. Following surface staining, cells were fixed and permeabilized. Subsequently, cells were stained intracellularly with APC-conjugated anti-CD206 antibody for 30 min. Data acquisition was performed on a flow cytometer (BD biosciences), and data were analyzed using FlowJo v10.8.1 software. The gating strategy involved initial debris exclusion, singlet selection, and dead cell exclusion (FVS780-negative), followed by the quantification of CD86⁺ (M1) and CD206+ (M2) populations (Supplementary Figure S1).

GROMACS 2022 was used for all calculations. The protein–ligand complex was constructed using CHARMM36 for the protein and AutoFF-CGenFF for the ligand. It was solvated in a 1 nm TIP3P water box and neutralized with NaCl. Electrostatics were treated using the Particle Mesh Ewald (PME) method with a 1 nm cutoff, and van der Waals interactions were truncated at 1 nm. Hydrogen bonds were constrained using the SHAKE algorithm, and a 1 fs integration time step was employed. A three-stage energy minimization was performed sequentially on water molecules only, ions only, and then the entire all-atom system. This was followed by heating from 0 to 310 K under NVT conditions using the Berendsen thermostat, and a 1 ns NPT equilibration using the Parrinello-Rahman barostat. The production run lasted 100 ns under NPT conditions at 310 K and 1 bar, with coordinates saved every 10 ps. For analysis, the first 10 ns were discarded, and the remaining 90 ns were processed using GROMACS tools to calculate root-mean-square deviation (RMSD), root-mean-square fluctuation (RMSF), hydrogen bond counts, radius of gyration, and solvent-accessible surface area (SASA).

Surface plasmon resonance (SPR) assay SPR experiments were conducted at 25 °C using a Biacore 8K instrument (Cytiva, MA, United States) equipped with a Series S Sensor Chip CM5. Mouse TNFα protein was immobilized on the active flow cell (Fc2) via standard amine coupling to a target level of approximately 6,000 response units (RU), while the reference flow cell (Fc1) was left unmodified to serve as a control for non-specific binding. The running buffer consisted of 1.05× PBS-P+ containing 5.25% DMSO to match the solvent composition of the analyte samples. To eliminate bulk refractive index variations caused by the high solvent concentration, a solvent correction cycle was performed using eight standard solutions with DMSO concentrations ranging from 4.0% to 5.8%. Ganoderic acid A was serially diluted in the running buffer (ranging from 0.3125 µM to 5 µM) and injected over both flow cells. The dissociation was monitored for 120 s. Data were processed using Biacore Insight Evaluation Software with double referencing (subtraction of the reference surface signal and blank buffer injections) and fitted to a 1:1 Langmuir binding model to determine the kinetic parameters.

Follow the instructions provided with the ELISA kit to measure the expression levels of cytokines (IL-1β, IL-6, IL-10 and TNFα) in mouse serum and cell culture supernatant. Briefly, a sandwich ELISA employing two antibodies was used. The high-affinity enzyme plate was pre-coated with a specific anti-mouse antibody. Samples were added to the wells, and after incubation, the target proteins bound to both the solid-phase and detection antibodies. Following a wash to remove unbound material, streptavidin-HRP was added. After a subsequent wash, the chromogenic substrate TMB was added, and the plate was incubated in the dark to develop color. Finally, absorbance was measured using a microplate reader.

Under specific conditions, total RNA was extracted from RAW264.7 cells. Reverse transcription was performed using the PrimeScript™ RT reagent kit and qRT-PCR was conducted with the SYBR Green PCR master mix. β-Actin served as the internal reference gene. Primer sequences are listed in Supplementary Table S1.

RAW264.7 cells were treated with vehicle (Control) or TNFα (50 ng/mL) for 12 h, after which they were harvested for proteomic analysis. Total proteins were extracted using SDT lysis buffer and digested with trypsin using the Filter Aided Sample Preparation (FASP) method. The resulting peptide mixtures were analyzed on an Orbitrap Astral mass spectrometer (Thermo Scientific) coupled with a Vanquish Neo UHPLC system in Data-Independent Acquisition (DIA) mode. Raw data were processed using DIA-NN software (version 1.8) against the Mus musculus UniProt database. To correct for systematic variation and potential differences in sample loading, Global Normalization was applied to the precursor ion intensities during data processing. Differentially expressed proteins (DEPs) were identified based on a fold change >1.5 and an adjusted P value (FDR) < 0.05 to rigorously control the false discovery rate.

Mix the tissues and cells with radioimmunoprecipitation assay (RIPA) buffer, add protease and phosphatase inhibitors, sonicate, and incubate on ice. Centrifuge the mixture at 12,000 rpm for 10 min at 4 °C. Then, heat the supernatant at 95 °C for 5 min. Load the tissue or cell lysate onto a 15% sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) gel for electrophoresis. Transfer the separated proteins onto a polyvinylidene fluoride (PVDF) membrane. Block the membrane at room temperature with 5% non-fat milk for 1 h, then incubate it at 4 °C with the primary antibody for 10 h. Next, incubate the membrane at room temperature with the secondary antibody for 1 h. Apply the enhanced chemiluminescence (ECL) reagent to the membrane and capture the image using the Tanon 5200 imaging system.

Animal samples will be fixed with 4% paraformaldehyde, embedded in paraffin, and sectioned. The sections will then undergo dewaxing, cleaning, and antigen retrieval using citrate buffer at high temperature. Subsequently, 5% BSA will be applied for blocking. The sections will be incubated with anti-CD86 (1:200) antibodies at room temperature for 2 h, followed by incubation with a fluorescently labeled secondary antibody for 1 h. After washing with PBS, cell nuclei will be stained with 4′,6-diamidino-2-phenylindole (DAPI), and the samples will be mounted for imaging using a fluorescence microscope.

Data analysis and graph generation were performed using GraphPad Prism 10.1.2. Data are presented as mean+/-standard deviation (SD). The normality of data distribution was assessed using the Shapiro-Wilk test, and homogeneity of variance was verified using the Brown-Forsythe test. For datasets satisfying these assumptions, comparisons between two groups were performed using an unpaired Student’s t-test. For comparisons among multiple groups, one-way analysis of variance (ANOVA) was conducted, followed by Dunnett’s post hoc test for comparisons against a control group or Tukey’s post hoc test for pairwise comparisons between groups. A P value less than 0.05 was considered statistically significant.

To identify potential targets of GAA in SRLI, we employed a systematic strategy integrating five public databases: ETCM, SwissTargetPrediction, ChEMBL, GeneCards, and OMIM. We retrieved a total of 287 potential targets for GAA and 1432 targets associated with SRLI. A Venn diagram was constructed to illustrate the overlapping targets identified from these databases (Figure 1A), which revealed 42 overlapping targets between GAA and SRLI as candidate therapeutic targets (Figure 1B). Subsequently, we input these 42 candidate targets into the STRING database to construct a PPI network (Figure 1C) and used Cytoscape for further analysis. The core targets were identified based on their degree values (Supplementary Table S2). Eight key targets, including TNF, TP53, NF-κB1, ESR1, CASP3, PPARG, MAPK3, and SRC, were identified as central nodes (Figure 1D). Notably, TNF, NF-κB1, and TP53 exhibited the highest degree values. The prominence of these well-established inflammatory mediators among the hub targets strongly suggests that the therapeutic effect of GAA against SRLI is primarily associated with the modulation of inflammatory responses. Subsequently, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses of the 42 overlapping targets were performed to elucidate their functional context. In Gene Ontology analysis, terms such as “response to molecule of bacterial origin”, “response to lipopolysaccharide,” and “regulation of inflammatory response” were ranked among the top 20 biological processes (BP) (Figure 1E). Cellular component (CC) analysis highlighted structures like “focal adhesion” and “membrane raft” (Figure 1F). Molecular function (MF) terms were dominated by transcription factor binding and nuclear receptor activity (Figure 1G), indicating that the interaction of GAA with its targets primarily involves signaling transduction and transcriptional regulation. KEGG pathway mapping identified pathways related to inflammation and stress response, including “Lipid and atherosclerosis,” “Chemical carcinogenesis - receptor activation,” and notably, the “TNF signaling pathway” and “IL-17 signaling pathway” (Figure 1H). Collectively, these computational results highlight the TNF signaling pathway as a key potential node for GAA’s intervention in SRLI, forming the central hypothesis for our subsequent experimental validation.

To independently validate the hub genes identified through network pharmacology, we reanalyzed the publicly available murine hepatic transcriptome dataset GSE217695. Eight hours after LPS challenge, differential expression analysis revealed a clear distinction between control and septic livers (Figure 2A). From the PPI network, we identified 11 high-degree nodes (degree ≥18) as putative drivers of the LPS-induced signature (Figure 2B). Expression profiling of these candidates showed significant upregulation of Tnf, *Nf-κb1*, Casp3, and Egfr, along with downregulation of Mapk3, while Pparg, Esr1, Tp53, Mtor, Ctnnb1, and Src remained unchanged (Figure 2C). The significant upregulation of Tnf, *Nf-κb1*, Casp3, and Egfr provided strong independent validation for our network pharmacology predictions, and these four genes, together with the downregulated Mapk3, were therefore selected as the core set for further investigation (Figure 2D). Collectively, the dysregulation of these five genes underscores their central role in the pathogenesis of SRLI and, in conjunction with our network pharmacology results, strongly nominates them as high-priority candidate targets for GAA. KEGG enrichment analysis of the 8-h LPS-induced hepatic transcriptome identified “Cytokine-cytokine receptor interaction” as the top significant pathway. Additionally, inflammation-related pathways such as “TNF signaling,” “NF-kappa B signaling,” and “IL-17 signaling” were significantly enriched, along with “Salmonella infection”-related immune cascade (Figure 2E). A Sankey plot (Figure 2F) further illustrated the flow of DEGs into these pathways, positioning Tnf and Nfκb1 as central nodes propagating the inflammatory signal. These findings align with the GO terms related to immune and migration processes (Supplementary Figure S2) and reinforce that hyperactivation of the TNFα/NF-κB axis is a principal transcriptional hallmark of SRLI, thereby providing a strong rationale and a focused direction for evaluating GAA intervention.

Convergent evidence from network pharmacology and transcriptomics identified several central hubs, including TNFα, NF-κB1, MAPK3, and CASP3. The selection of TNFα for subsequent experimental validation was driven by a convergent multi-layered analysis: it emerged as a top-ranked hub node in both our network pharmacology and independent transcriptomic PPI networks, and its position as a master upstream regulator of the NF-κB pathway—the most significantly enriched pathway in our KEGG analysis—suggested that its direct targeting could mechanistically explain the suppression of the entire inflammatory cascade. Molecular docking revealed that all four proteins displayed strong affinities for GAA, with binding energy ≤ −7.3 kcal/mol (MAPK3 = −7.9 kcal/mol, CASP3 = −7.6 kcal/mol, TNFα = −7.3 kcal/mol, NF-κB1 = −7.3 kcal/mol) (Figures 3A–D). Inspection of the TNFα–GAA complex showed the triterpenoid docked into the cytokine’s active site, forming four direct hydrogen bonds (Ser95, Asn92, Thr79, Ser81) that provide enthalpic stability through key molecular interactions (Supplementary Table S3).

To confirm the physical interaction, we measured GAA binding to TNFα by surface plasmon resonance (SPR). Fitting the sensorgrams to a 1:1 binding model yielded an equilibrium dissociation constant (K_D) of 2.3 µM (Figure 4A), an affinity typical for small-molecule cytokine inhibitors. We then performed a 100-ns molecular dynamics (MD) simulation to assess the stability of the complex. The root-mean-square deviation (RMSD) plateaued within 20 ns and remained stable below 3 Å (Figure 4B), indicating a well-equilibrated system. Simultaneous decreases in solvent-accessible surface area (SASA) and radius of gyration (Rg) (Figures 4C,D) indicated that GAA binding induces a more compact conformation in TNFα. This ligand-induced structural tightening may potentially interfere with the conformational dynamics required for TNFα trimerization or receptor binding. At the residue level, the root-mean-square fluctuation (RMSF) remained low (>95% of residues <3 Å) (Figure 4E), while one to three persistent hydrogen bonds maintained the ligand’s position throughout the simulation (Figure 4F). In summary, the micromolar affinity from SPR and the multi-parameter stability profile from MD confirm that GAA directly binds to TNFα. Our prior bioinformatic analyses (Figures 1H, 2E,F) consistently positioned the NF-κB signaling pathway as the most significantly enriched cascade downstream of TNFα in SRLI, and PPI networks highlighted NF-κB1 as a core hub. We therefore hypothesize that the primary mechanism by which GAA-TNFα binding exerts its anti-inflammatory effect is through the suppression of this dominant NF-κB axis. This central hypothesis was tested and validated in our subsequent cellular experiments.

To identify the primary cellular effector of TNFα in septic liver injury for downstream validation, we analyzed a public scRNA-seq dataset. Tnf expression was significantly upregulated across cell populations (Figures 5A,B). A critical comparative analysis revealed that although hepatocytes showed the highest fold-change, this was attributable to their exceptionally low baseline expression. In contrast, macrophages not only maintained a substantially higher baseline level of Tnf but also exhibited a pronounced fold-increase upon insult, establishing them as the dominant and most relevant source of pathogenic TNFα signaling in the septic liver microenvironment (Figure 5C). This finding rationally guided our selection of the RAW264.7 macrophage model for functional validation. Cell viability assays confirmed no significant cytotoxicity of GAA at the tested concentrations (20 and 40 µM), ensuring that subsequent functional interpretations were not confounded by reduced cell viability (Figure 5E). We next established a model of TNFα challenge using recombinant cytokine (rTNFα) to test GAA’s functional antagonism. To unbiasedly identify the key pathways engaged by TNFα in this model, we first performed proteomic analysis of rTNFα-stimulated macrophages. KEGG enrichment analysis of the differentially expressed proteins identified the NF-κB signaling pathway as the most significantly enriched cascade (Supplementary Figure S3), providing a data-driven rationale for our subsequent mechanistic focus. Stimulation with rTNFα robustly triggered the secretion of the pro-inflammatory cytokines IL-1β and IL-6 (Figures 5F,G). Pre-treatment with GAA potently suppressed this release in a dose-dependent manner. In contrast to some anti-inflammatory agents, GAA did not significantly alter the secretion of IL-10 under these conditions (Figure 5H), indicating that its primary action is the suppression of pro-inflammatory signaling. Guided by the proteomic results, we investigated the NF-κB pathway. Western blot analysis confirmed that rTNFα stimulation for 30 min induced rapid phosphorylation of the p65 subunit and its inhibitor IκBα (Figures 5I–K), marking pathway activation. GAA pre-treatment effectively blunted this early signaling event, demonstrated by a dose-dependent reduction in the levels of both p-p65 and p-IκBα. This result provides direct evidence that GAA intercepts TNFα signaling at the level of the early NF-κB activation cascade. Furthermore, flow cytometric analysis revealed that GAA pre-treatment dose-dependently decreased M1/M2 ratio (Figures 5L,M), indicating a functional restoration of a balanced phenotypic state. Collectively, our data demonstrate that GAA functions as a potent antagonist of TNFα-induced pro-inflammatory signaling in macrophages. It effectively blocks the early activation of the NF-κB pathway, leading to reduced synthesis of key pro-inflammatory cytokines and a concomitant decrease in M1 polarization.

To validate the therapeutic efficacy and target specificity of GAA in vivo, we established an LPS-induced sepsis-associated liver injury model in mice. The experimental design included six groups: Control, LPS, LPS + Low GAA, LPS + High GAA, LPS + anti-TNFα (positive control), and LPS + anti-TNFα + High GAA (Figure 6A). Administration of GAA markedly improved LPS-induced liver injury, as evidenced by a significant reduction in serum alanine aminotransferase (ALT) and aspartate aminotransferase (AST) levels (Figure 6B) and a clear amelioration of histopathological lesions, including disrupted hepatic cord structure and focal necrosis (Figure 6C). The hepatoprotective effect of the High GAA dose was comparable to that of the anti-TNFα antibody, with no significant difference observed between the two groups. At the molecular level, Western blot analysis of liver lysates (Figure 6H) revealed that LPS challenge significantly enhanced p65 phosphorylation (p-p65) and promoted IκBα degradation, which were significantly reversed by GAA treatment, substantiating its inhibition of the NF-κB pathway in vivo. Crucially, the combination of High GAA with anti-TNFα antibody yielded no significant additive inhibitory effects on NF-κB pathway activation compared to anti-TNFα treatment alone (Figures 6H–J). Consistent with the suppression of systemic inflammation, the LPS-triggered surge in serum pro-inflammatory cytokines (IL-1β, IL-6, TNFα) was significantly suppressed by GAA (Figure 6D). Furthermore, GAA significantly reduced the LPS-induced upregulation of inflammatory genes (Cd86, Nos2, Tnf) (Figure 6E) and inhibited pro-inflammatory macrophage polarization, as evidenced by decreased CD86⁺ immunofluorescence in liver sections (Figures 6F,G). In summary, GAA demonstrates potent hepatoprotective and anti-inflammatory effects in a murine model of SRLI, mechanistically linked to the suppression of the NF-κB pathway. The powerful non-additivity observed upon co-treatment with an anti-TNFα antibody demonstrates that the therapeutic mechanisms of GAA and anti-TNFα antibody are not independent and likely share a critical, rate-limiting node within the TNFα signaling pathway.