Using the online RNASeqPower Sample Size Calculator (https://rodrigo-arcoverde.shinyapps.io/rnaseq_power_calc/), a sample size of n = 11 was determined to achieve 89.03% statistical power (α = 0.05). This sample size ensures both statistical validity for 16S rRNA sequencing and compliance with animal ethics requirements. Therefore, a total of 22 specific pathogen-free (SPF) healthy male C57BL/6 mice (15-month-old, weighing 30.46 ± 3.05g) were obtained from Chang Zhou Cavens Laboratory Animal Ltd. The mice were housed under standardized conditions (12-hour light/dark cycle, 22 ± 1°C, 50-60% humidity) with ad libitum access to water and a standard chow diet. Mice were randomly divided into two groups: (1) Control group (CON, n = 11): administered vehicle (sterile water) via oral gavage daily; (2) Metformin-treated group (TEST, n = 11): administered metformin (Sangon biotech, Shanghai, China) dissolved in sterile water at 300 mg/kg body weight/day via oral gavage for 5 months. Body weight was monitored weekly. All experimental procedures were approved by the Animal Care Ethics Committee of Bengbu Medical University. The Animal Ethical Approval number was 2020-050.

At 20 months of age, fecal samples were collected and stored at -80°C for microbiota analysis. After specimen collection, the mice were fasted for 6 hours and euthanized by CO₂ asphyxiation. Euthanasia was induced using a small animal gas anesthesia system (Yuyan Scientific Instrument Co., Ltd, Shanghai, China). Animals were placed in a sealed chamber, and compressed CO₂ was introduced at a flow rate of 30% of the chamber volume per minute. Once deep anesthesia was confirmed by the absence of a pedal reflex (toe pinch) and respiratory arrest, mice were promptly removed from the chamber. Terminal blood collection was performed via cardiac puncture while the animals remained under deep anesthesia. Death was confirmed following blood collection by cervical dislocation or exsanguination. Blood was centrifuged (3,000 × g, 15min, 4°C) for isolate serum, and stored at -80°C for biochemical analysis. Spleens were aseptically excised, weighed, and divided into three portions for distinct processing protocols. For RNA real-time quantitative polymerase chain reaction (RT-qPCR) analysis, tissues were either processed immediately or flash-frozen in liquid nitrogen followed by storage at -80°C. For flow cytometric analysis, spleen tissues were mechanically dissociated by pressing through a 45-μm nylon mesh using a syringe plunger. The resulting cell suspensions underwent purification through Percoll gradient centrifugation (Solarbio, Beijing, China). For histological processing, spleens were post-fixed in 4% PFA for 12 hours at 4°C, then cryoprotected by immersion in 30% sucrose solution (prepared in 0.01 M PBS, pH 7.4) for 24 hours at 4°C. Tissues were embedded in optimal cutting temperature (OCT) compound (Tissue-Tek, Miles, Elkart, IN) and sectioned into 6-μm slices using a cryostat (Leica CM1900, Bannockburn, IL). Sections were mounted on poly-L-lysine-coated slides and stored at -80°C until further processing for staining.

Hepatic, renal, and metabolic parameters were quantified using commercial assay kits: alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin (TBIL), cholinesterase (CHE), creatinine (CRE), urea (UREA), albumin (ALB), globulin (GLB), creatine kinase (CK), and lactate dehydrogenase (LDH) (all from Ortho-clinical diagnostics, inc., NY, USA). Measurements were performed on a VITROS 5600 Integrated System (Ortho-clinical diagnostics, inc.) following manufacturer protocols.

Fecal DNA was extracted using the QIAamp DNA Stool Mini Kit (QIAGEN, Hilden, Germany). The 16S rRNA gene was amplified with primers 27F (5’-AGRGTTYGATYMTGGCTCAG-3’) and 1492R (5’-RGYTACCTTGTTACGACTT-3’) and sequenced on an Illumina NovaSeq 6000 platform (2 × 250 bp). Raw reads were processed in QIIME2 (v2021.11) using DADA2 for denoising and amplicon sequence variant (ASV) clustering. Taxonomic assignment was performed against the SILVA (v138) database.

Bioinformatic analysis of the gut microbiota was carried out using the Majorbio Cloud platform (https://cloud.majorbio.com).

Rarefaction curves were calculated with Mothur v1.30.1. Alpha diversities (Chao1 and Shannon) were analyzed with R-3.3.1 (stat) package. Hierarchical clustering and principal coordinate analysis (PCoA) were analyzed with R-3.3.1 (vegan) based on Bray-curtis dissimilarity. ANOSIM analysis was used to confirm statistically significant separation between groups. Beta diversity difference analysis performed with python-2.7 package and Wilcoxon rank-sum test. The linear discriminant analysis (LDA) effect size (LEfSe) (http://huttenhower.sph.harvard.edu/LEfSe) was performed to identify the significantly abundant taxa of bacteria between the two groups (LDA score > 2, p<0.05).

Putative functional profiles for Clusters of Orthologous Groups (COG) categories and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were inferred from 16S rRNA gene amplicon sequences using PICRUSt2 (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States, version 2.5.2). The PICRUSt2 workflow involved placing ASVs into a reference phylogenetic tree, followed by hidden-state prediction of gene family copy numbers for KEGG orthologs (KOs) and COG categories. The predicted copy numbers were then multiplied by ASV abundance counts to generate metagenome predictions. For KEGG pathways, differential abundance testing for pathways was performed via ANCOM-BC2 (QIIME2 plugin). Differentially abundant KEGG pathways (Pathway Level 3) were filtered using a Benjamini-Hochberg-adjusted p-value threshold of ≤ 0.05. Only pathways meeting this criterion were included. All significantly enriched pathways were systematically curated by cross-referencing them with established prokaryotic metabolic capabilities using MetaCyc’s bacterial pathway database (https://metacyc.org/) and literature evidence. Pathways with only eukaryotic associations were excluded from biological interpretation. COG category abundances were normalized as relative abundances per sample.

Correlation analyses were performed using R-3.3.1, python-2.7 package. In the correlation analysis between gut microbiota composition and clinical factors, dominant bacterial genera were operationally defined as the top 10 most abundant genera at the genus level based on mean relative abundance across all samples. This selection criterion ensured inclusion of taxa with the highest biological relevance to community structure. Correlations were assessed using Spearman’s rank correlation coefficient (ρ), a non-parametric method chosen for its robustness against non-normally distributed data and ability to capture monotonic relationships. Statistical significance (p-values) was derived from the Spearman test statistic, with the null hypothesis of no correlation rejected at p<0.05. To mitigate false discovery risks inherent in multiple hypothesis testing, all p-values underwent rigorous adjustment via the Benjamini-Hochberg false discovery rate (FDR) correction procedure. This approach controlled the expected proportion of false positives among significant results at α = 0.05. The final significance threshold for correlations was FDR-adjusted p<0.05, with correlation strength and direction visualized in heatmaps.

Single-cell suspensions were stained with fluorochrome-conjugated antibodies and corresponding isotype-matched controls (detailed in Table 1 ). Briefly, antibody cocktails containing isotype controls and specific antibodies, at manufacturer-recommended concentrations, were added to 200 µL aliquots of cell suspension. Following 30-minute incubation at room temperature with light protection, cells underwent two successive washes with 5 mL phosphate-buffered saline (PBS) using centrifugation (400 ×g, 5min). Washed cells were fixed in 2% paraformaldehyde (PFA) in PBS (pH 7.4) for subsequent analysis. Flow cytometry acquisition was performed using a flow cytometer (RaiseCare Biotechnology Co., Ltd., Qingdao, China) with standardized voltage settings. Data analysis was conducted using Raiseflower software (v2.1.3, RaiseCare Biotechnology). For the flow cytometry gating strategy, gates defining positive populations were established based on fluorescence thresholds set using isotype-matched control antibodies ( Supplementary Figure S1 ). The hierarchical gating approach first selected singlet events via FSC-A/FSC-H. Subpopulations were subsequently gated using lineage-defining markers: CD3⁺ for T cells, with CD4⁺ and CD8⁺ subsets ( Supplementary Figure S1A ); CD3^-B220⁺ for B cells and CD3^-NK1.1⁺ for NK cells ( Supplementary Figure S1B ); and CD45⁺Ly-6G⁺ for neutrophils and CD45⁺F4/80⁺ for macrophages ( Supplementary Figure S1C ). Macrophage subsets were further classified as CD68⁺CD86⁺ (M1) and CD68⁺CD163⁺ (M2) based on established polarization markers ( Supplementary Figure S1D ).

The immunofluorescence (IHF) assay was conducted following established protocols (21, 22). Briefly, frozen spleen sections fixed in 4% paraformaldehyde (PFA/PBS) were incubated with primary antibodies (see Table 1 for specifications) at 4 °C for 16–18 hours in a humidified chamber. Following three 5-minute washes with 0.01 M PBS (pH 7.4), sections were incubated with corresponding fluorescein-conjugated secondary antibodies (refer to Table 1 for dilutions) for 1 hour at room temperature under light-protected conditions. After subsequent PBS washes, slides were mounted using ProLong™ Gold Antifade Mountant containing DAPI nuclear counterstain (Thermo Fisher Scientific, Waltham, MA, USA) and sealed with nail polish. Fluorescence imaging was performed using an Axio Observer Z1 inverted microscope equipped with ApoTome.2 structured illumination (Carl Zeiss AG, Oberkochen, Germany). All images were acquired with consistent exposure settings using ZEN Blue 3.1 software (Zeiss).

Total RNA was isolated from murine spleen tissues using TRIzol™ Reagent (Thermo Fisher Scientific, Carlsbad, CA, USA) following the manufacturer’s protocol. RNA integrity was verified using an Agilent 2100 Bioanalyzer with RNA Integrity Numbers (RIN) > 7.0, and quantification was performed using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific) with A260/A280 ratios between 1.8 and 2.0. First-strand cDNA synthesis was conducted using the BeyoRT™ II First Strand cDNA Synthesis Kit with gDNA Eraser (Beyotime Biotechnology, Shanghai, China), starting with 1 μg total RNA input. Quantitative PCR amplification was performed in triplicate reactions using BeyoFast™ SYBR Green qPCR Mix (Beyotime Biotechnology) on an Applied Biosystems 7500 Real-Time PCR System (Thermo Fisher Scientific). Primer sequences are detailed in Table 2 . We specifically selected qPCR targets to profile polarized immune pathways relevant to microbiome-immune crosstalk. These targets represent key T helper subsets: Th1 (Ifng), Th2 (Il4 and Il10), Th17 (Il17a), and Treg (Foxp3); and macrophage polarization states: M1 (Il1b and Il6) and M2 (Arg1 and Tgfb1). This selection of canonical cytokine genes reflects functional axes known to be altered in gut-microbiota interactions (27, 28). Although markers such as Il21, Cxcl13, Il2, or NF-κB pathway components could provide supplementary insights, our focused panel aligns directly with the study’s core hypothesis centered on T cell and macrophage polarization. Gene expression quantification was calculated using the comparative 2−^ΔΔCt method with normalization to the Gapdh reference gene (29).

Data are presented as mean ± SD. Group comparisons were performed using unpaired Student’s t-test or Wilcoxon rank-sum test for parametric and non-parametric data, respectively. Multiple testing corrections were applied via the Benjamini-Hochberg method. Correlations between microbiota and immune parameters were assessed using Spearman’s rank correlation. Statistical significance was set at p<0.05. All analyses were performed in GraphPad Prism v9.3.1 or R v4.1.2.

As summarized in Table 3 , the baseline body weight of 15-month-old mice did not differ significantly between the control (30.43 ± 2.85g) and metformin-treated (30.50 ± 3.37g) groups (unpaired t-test, p=0.96, n = 11). After 5 months of oral metformin administration (300 mg/kg/day), body weights remained comparable between groups (34.15 ± 3.38g vs. 32.70 ± 4.10g; p=0.38, n = 11), indicating no treatment-related effects on growth or systemic metabolism.

Serum analyses revealed no treatment-related toxicity across hepatic, renal, or cardiac systems. Hepatic function markers, including alanine aminotransferase (ALT: 51.18 ± 30.13 U/L vs. 51.82 ± 21.10 U/L; p=0.95, n = 11) and aspartate aminotransferase (AST: 132.09 ± 95.85 U/L vs. 133.45 ± 75.99 U/L; p=0.97, n = 11), showed no significant differences. The AST/ALT ratio (3.82 ± 3.49 vs. 3.07 ± 2.06; p=0.52, n = 11), total bilirubin (TBIL: 3.44 ± 1.24 μmol/L vs. 3.24 ± 1.51 μmol/L; p=0.71, n = 11), and cholinesterase (CHE: 3111.82 ± 1187.19 U/L vs. 2403.76 ± 1098.53 U/L; p=0.15, n = 11) remained unaffected. Renal function parameters, creatinine (CRE: 23.45 ± 6.64 μmol/L vs. 28.18 ± 8.77 μmol/L; p=0.18, n = 11) and urea (UREA: 12.96 ± 1.88 mmol/L vs. 12.22 ± 1.20 mmol/L; p=0.29, n = 11), also exhibited no statistically significant changes. Protein metabolism markers, albumin (ALB: 26.62 ± 9.13 g/L vs. 23.87 ± 4.40 g/L; p=0.41, n = 11), globulin (GLB: 26.57 ± 4.04 g/L vs. 25.93 ± 4.05 g/L; p=0.70, n = 11), and the albumin/globulin ratio (A/G: 1.04 ± 0.46 vs. 0.94 ± 0.23; p=0.54, n = 11), were similarly unaltered. Cardiac and muscle injury markers, creatine kinase (CK: 1343.00 ± 999.57 U/L vs. 1437.36 ± 1126.96 U/L; p=0.83, n = 11) and lactate dehydrogenase (LDH: 839.55 ± 231.24 U/L vs. 913.82 ± 336.69 U/L; p=0.54, n = 11), showed no treatment-associated elevations. Collectively, these data demonstrate that prolonged metformin treatment at the tested dosage does not induce systemic toxicity or clinically significant perturbations in metabolic or organ function in aged mice.

To delineate the tripartite relationship between long-term metformin treatment, gut microbiota, and splenic immune microenvironment in aged mice, 16S rRNA gene sequencing was employed to investigate metformin-induced alterations in gut microbiota composition. Supplementary Tables S1 and S2 showed the post-processed 16S-count at ASV- and genus-level, respectively. The rarefaction curves in Figure 1A reached plateaus as sequencing depth increased, indicating adequate coverage for capturing species diversity. This asymptotic pattern demonstrated sufficient sequencing depth for downstream analyses. Hierarchical clustering analysis using Bray-Curtis dissimilarity revealed distinct microbial composition-based segregation between two groups ( Figure 1B ). To assess the taxonomic richness and evenness of the gut microbiota community, we performed alpha diversity analysis. The Chao1 and Shannon indices were used to evaluate sequencing depth adequacy and quantify species diversity, respectively. The Chao1 index showed no significant difference between the two groups ( Supplementary Figure S2A , p > 0.05). In contrast, the Shannon index revealed significantly higher diversity in the metformin-treated group ( Supplementary Figure S2B , p<0.05). This indicates that while sequencing depth was comparable between groups, the metformin-treated group exhibited greater species diversity. Subsequently, beta diversity analysis was used to measure differences in community composition between the two groups. The dendrogram topology showed tighter clustering of biological replicates, reflecting high intra-group similarity. Color-coding according to experimental conditions confirmed metformin-treated samples (TEST, green) formed a distinct clade separate from controls (CON, red). Principal coordinates analysis (PCoA) based on Bray-Curtis dissimilarity revealed significant β-diversity patterns between control and treatment groups ( Figures 1C, D ). The two-dimensional ordination plot illustrated distinct clustering patterns, where sample proximity reflected similarity in microbial community composition ( Figure 1C ). ANOSIM analysis confirmed statistically significant separation between groups ( Figure 1D , R=0.48, p=0.001), with inter-group distances substantially exceeding intra-group variations.

Metformin-induced gut microbiota modulation was characterized through amplicon sequence variant (ASV) distribution analysis and taxonomic profiling. Venn diagram analysis revealed group-specific ASV patterns, with 1,790 ASVs identified across cohorts, comprising 669 control-exclusive (CON) and 568 metformin-exclusive (TEST) variants, while 553 ASVs (30.9% of total) were shared ( Figure 2A ). QIIME2-processed data (DADA2 denoising, 99% identity clustering) demonstrated phylum-level restructuring, where CON microbiota was dominated by Firmicutes (73.55 ± 7.37%), Bacteroidetes (18.73 ± 5.18%), and Proteobacteria (4.18 ± 2.93%), collectively representing 96.45% of community composition. In Metformin treatment group (TEST), the above three microbial communities still dominate, with proportions of 51.82 ± 14.48%, 31.18 ± 13.44%, and 7.82 ± 5.78%, respectively. Metformin treatment also significantly increased Verrucomicrobia abundance from 0.26% to 7.31%, establishing a four-phylum dominance pattern ( Figures 2B, D ).

Genus-level analysis (threshold: relative abundance >2%, prevalence >80% samples) identified differential taxa through ANCOM-BC2. Metformin-treated mice exhibited increased proportions of Muribaculum (Fold Change, FC=3.08), Akkermansia (FC=37.58), Escherichia (FC=50.70), Helicobacter (FC=4.58), Duncaniella (FC=2.02), and Allobaculum (FC=2.95). Conversely, Lactobacillus (FC=0.25), unclassified Lachnospiraceae (FC=0.33), Desulfovibrio (FC=0.33), and Mucispirillum (FC=0.27) were significantly decreased (p<0.05, Figures 2C, D ). The above results were further validated by Wilcoxon rank-sum test (p<0.05, Figure 2E ). For microbiome visualization, the ANCOM-BC2 results with a heatmap showing relative abundances of differentially abundant genera were shown in Supplementary Figure S3 .

These findings suggest that metformin can modulate the intestinal microbiota composition in mice.

To identify differentially abundant bacterial taxa, linear discriminant analysis effect size (LefSe) was applied with a logarithmic LDA score threshold of 2.0 and a significance level of α = 0.05. Taxonomic cladograms from phylum to species level are shown in Figure 3A . At the genus level, the control group was dominated by Lactobacillus in class Firmicutes; Desulfovibrio in class Deltaproteobacteria; Eisenbergiella and unclassified_f:Lachnospiraceae in class Clostridia; and Ileibacterium in class Erysipelotrichia. The Metformin-treated group exhibited higher abundances of Parasutterella and Turicimonas in class Betaproteobacteria; Parabacteroides, Bacteroides, Duncaniella and Muribaculum in class Bacteroidia; Helicobacter in class Epsilonproteobacteria; Escherichia in class Gammaproteobacteria; and Akkermansia in class Verrucomicrobiae ( Figures 3B and Supplementary Table S3 , LDA score > 2, p<0.05, Kruskal-Wallis test with Benjamini-Hochberg correction).

To investigate the functional impact of metformin treatment on the gut microbiota of aged mice, Phylogenetic Investigation of Communities by Reconstruction of Unobserved States 2 (PICRUSt2) was employed to predict clusters of orthologous groups (COGs) and metabolic pathways using Kyoto Encyclopedia of Genes and Genomes (KEGG). Significant variations in COG functional categories were observed for Replication, recombination and repair; Inorganic ion transport and metabolism; Posttranslational modification, protein turnover, chaperones; Signal transduction mechanisms; Lipid transport and metabolism (adjusted p<0.05, Benjamini-Hochberg correction; Figures 4A, B ).

To ensure clarity in distinguishing biologically relevant prokaryotic pathways from eukaryotic hits that may arise from database limitations, we systematically curated all significantly enriched KEGG pathways by cross-referencing them with established prokaryotic metabolic capabilities using MetaCyc’s bacterial pathway database. Pathways with functions exclusive to eukaryotic organisms were explicitly flagged as ‘non-bacterial’ in Table 4 and excluded from biological interpretation. KEGG pathway analysis at Pathway Level 3 revealed significant alterations in microbial functionality, primarily associated with metabolism and biosynthesis, such as Secondary bile acid biosynthesis, Lipoic acid metabolism, beta-Lactam resistance, Mismatch repair, RNA transport, Lipopolysaccharide biosynthesis, and Phosphonate and phosphinate metabolism (adjusted p<0.05; Figure 5A ). Among the top 30 differentially abundant pathways, notable differences were detected in metabolic and biosynthetic processes such as Secondary bile acid biosynthesis, Lipopolysaccharide biosynthesis, Vitamin B6 metabolism, beta-Alanine metabolism, and Arginine and proline metabolism (adjusted p<0.05; Figure 5B , Supplementary Table S4 ). Additionally, pathways potentially linked to aging, including Mismatch repair and Oxidative phosphorylation (30, 31), were enriched in the metformin-treated group ( Figure 5B , Supplementary Table S4 ).

Having established that metformin induces significant alterations in gut microbiota structure and metabolic potential, we next evaluated whether these microbial shifts associate with modulations in the splenic immune microenvironment.

To evaluate the impact of prolonged metformin treatment on splenic immune cell dynamics, FCM ( Figures 6A , Supplementary Figure S1 ) was employed. Immune cell subsets were defined as follows: total T cells (CD3⁺), helper T cells (Th, CD3⁺CD4⁺), cytotoxic T cells (Tc, CD3⁺CD8⁺), B cells (CD3^-B220⁺), natural killer cells (NK, CD3^-NK1.1⁺), neutrophils (NEUT, CD45⁺Ly-6G⁺), macrophages (MAC, CD45⁺F4/80⁺), M1 macrophages (M1, CD68⁺CD86⁺), and M2 macrophages (M2, CD68⁺CD163⁺). Statistical analysis ( Figure 6B ) revealed that in the metformin-treated (TEST) group compared to control (CON), the percentage of Tc cells increased significantly from 5.67 ± 2.79% to 10.34 ± 4.06% (p<0.01, n = 11). Macrophage proportions also rose from 3.45 ± 0.87% to 5.52 ± 0.82% (p<0.05, n = 11). The Th/Tc (CD4/CD8) ratio decreased markedly (2.13 ± 0.51 vs. 1.31 ± 0.31; p<0.01, n = 11). Additionally, M1 macrophages decreased from 18.49 ± 6.23% to 11.33 ± 3.72% (p<0.01, n = 11), while M2 macrophages increased from 4.81 ± 2.77% to 9.32 ± 3.26% (p<0.01, n = 11). Consequently, the M1/M2 ratio declined from 4.63 ± 2.09 to 1.28 ± 0.44 (p<0.01, n = 11).

To further validate FCM results, immunohistochemical fluorescence (IHF) was performed. As shown in Figure 7A , CD3⁺ total T, CD3⁺CD4⁺ Th, CD3⁺CD8⁺ Tc, CD3^-CD19⁺ B, CD3^-NK1.1⁺ NK, CD45⁺Ly-6G⁺ NEUT, CD45⁺F4/80⁺ MAC, CD68⁺CCR7⁺ M1, and CD68⁺Arg1⁺ M2 were quantified in the CON and TEST groups. Statistical analysis ( Figure 7B ) indicated that in the CON group, cell densities (cells/mm²) were as follows: T cells (383.33 ± 88.65), Th cells (230.00 ± 51.72), Tc cells (108.00 ± 16.92), B cells (748.33 ± 113.70), NK cells (168.92 ± 35.85), NEUT (184.83 ± 63.85), MAC (203.83 ± 34.38), M1 (102.17 ± 17.24), and M2 (40.67 ± 6.95). In the TEST group, these values were 435.33 ± 66.95 (T cells), 230.00 ± 41.84 (Th cells), 174.17 ± 26.78 (Tc cells), 835.00 ± 169.64 (B cells), 174.50 ± 59.74 (NK cells), 256.67 ± 82.66 (NEUT), 384.67 ± 47.26 (MAC), 76.50 ± 9.22 (M1), and 115.33 ± 14.25 (M2). Comparisons between groups demonstrated significant increases in Tc cells, macrophages, and M2 macrophages in the TEST group (p<0.01, n = 6), while M1 macrophage numbers and the M1/M2 ratio were reduced (p<0.05 or 0.01, n = 6). No significant differences were observed in other cell populations (p > 0.05, n = 6).

To further investigate the impact of long-term metformin treatment on immune cell differentiation in the spleen of aged mice, reverse transcription quantitative polymerase chain reaction (RT-qPCR) was used to assess mRNA levels of Th1-, Th2-, Th17-, Treg-, M1-, and M2-associated cytokines or markers. As shown in Figure 8 , the mRNA levels of cytokines or markers of Th1 (Infg), Th17 (Il17a), and M1 (I11b and Il6) were significantly higher in CON group compared to the TEST group (all p<0.01, n = 11). In contrast, mRNA levels of Th2 (Il4 and Il10), and M2 (Arg1 and Tgfb1) were markedly reduced in the TEST group (all p<0.01, n = 11). Notably, no significant difference was observed in Foxp3 expression, a Treg-specific marker, between the two groups (p > 0.05, n = 11).

To investigate the relationship between splenic immune cell subsets and gut microbial composition in aged mice, we analyzed the correlation between the relative abundance of dominant gut bacterial genera and immune cell subsets in the spleen. As illustrated in Figure 9 , the top 10 bacterial genera at the genus level showed distinct correlation patterns with immune parameters.

As shown in Figure 9A and Supplementary Table S5 , Lactobacillus exhibited significant negative correlations with Tc cells (cytotoxic T cells), macrophages (MAC), and M2 macrophages, while showing positive correlations with the CD4/CD8 ratio and M1/M2 ratio. Allobaculum was exclusively positively correlated with macrophages. Duncaniella negatively correlated with total T cells and CD4/CD8 ratio, but positively with MAC. Unclassified Muribaculaceae negatively associated with total T cells and Th cells (helper T cells), yet positively with MAC. Unclassified Lachnospiraceae demonstrated negative correlations with Tc cells, MAC, and M2 macrophages, but positive correlations with CD4/CD8 ratio, M1 macrophages, and M1/M2 ratio. Akkermansia and Muribaculum showed reciprocal correlation patterns: Akkermansia negatively correlated with CD4/CD8 ratio, M1 macrophages, and M1/M2 ratio, but positively with Tc cells, MAC, and M2 macrophages; Muribaculum showed negative correlations with CD4/CD8 ratio and M1/M2 ratio, but positive correlations with MAC and M2 macrophages.

Furthermore, correlations between microbial abundance and cytokine/marker mRNA levels were analyzed. As shown in Figure 9B and Supplementary Table S6 , Lactobacillus and Unclassified Lachnospiraceae were negatively associated with anti-inflammatory markers (Il4, Il10, Arg1, and Tgfb1) and positively correlated with pro-inflammatory cytokines (Ifng, Il17a, Il6, and Il1b). Conversely, Akkermansia and Muribaculum showed opposite trends, positively correlating with anti-inflammatory markers (Il4, Il10, Arg1, and Tgfb1) and negatively with pro-inflammatory cytokines (Ifng, Il17a, Il6, and Il1b). Duncaniella and Unclassified Muribaculaceae exhibited mixed profiles: Duncaniella negatively associated with Il17a, Il6, and Il1b but positively with Arg1, while Unclassified Muribaculaceae only showed a negative correlation with Il1b.

These correlation results suggest that higher abundances of Lactobacillus and unclassified Lachnospiraceae are associated with a pro-inflammatory state, whereas Akkermansia and Muribaculum are linked to anti-inflammatory responses.