Adult male K18-hACE2 transgenic mice (B6.Cg-Tg(K18-ACE2)2Prlmn/J, JAX stock no. 034860) were purchased from Jackson Laboratories and used for all experiments. Adult mice (8–10 weeks old) were used for standard infection and treatment experiments, while aged mice (77 weeks old) were used for aging-associated comparisons. The animal procedures were conducted in a biosafety level 3 (BSL3) facility in compliance with the Public Health Service Policy on Humane Care and Use of Laboratory Animals. The experimental protocol was approved by the Institutional Animal Care and Use Committee (IACUC) of Department of Laboratory Animal Resources of Yonsei University College of Medicine, accredited by the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) International (protocol no. 001071). For infection, the mice were anesthetized with a mixture of ketamine and xylazine and intranasally inoculated with SARS-CoV-2. The infection dose was determined based on the experimental objective: 1 × 10⁵ PFU was used for longitudinal microbiome profiling (Figures 1, 2), and 1 × 10² PFU was used for survival and therapeutic efficacy studies (Figure 3).

African green monkey kidney epithelial Vero E6 cells were cultured in Dulbecco’s modified Eagle’s medium (Hyclone) supplemented with 10% fetal bovine serum (Hyclone) and 1% penicillin–streptomycin (Hyclone). The cell line was maintained at 37°C and in a 5% CO₂ atmosphere. The cells were authenticated and routinely tested negative for mycoplasma contamination. The original Wuhan strain of SARS-CoV-2 (accession number: NCCP43326/Korea) and the SARS-CoV-2 Omicron variant (BA.1, accession number: NCCP43408/Korea) were obtained from the Korea Centers for Disease Control and Prevention. The virus was produced by infecting Vero E6 cells cultured in DMEM containing 2% FBS for 2 to 3 days (MOI of 0.5). Supernatant containing the virus was collected and clarified by centrifugation (4,000 rpm for 15 min) before storage at –80°C. All experiments used viral stocks with ≤3 passages. Titer of viral stock was determined by plaque assay. All work relating to SARS-CoV-2 virus was conducted in the BSL-3 facility of Avison Biomedical Research Center in accordance with institutional biosafety committee regulations.

Infectious SARS-CoV-2 plaque-forming units were quantified by plaque titration on Vero E6 cells. At 1 day before infection, the cells were seeded in 12-well plates at a density of 1 × 10⁵ cells/mL. After washing once with DMEM without FBS, the cells were inoculated with viruses serially diluted in DMEM containing 2% FBS at 1:10 dilution for 1 h at 37°C. The inoculum was removed, and the cells were washed with DMEM without FBS and subsequently overlaid with a 1:1 mix of 2% sea-plaque agarose (Lonza) and 2× DMEM supplemented with 2% FBS and 1% penicillin–streptomycin. After 48–72 h of incubation at 37°C, the cells were fixed with 4% paraformaldehyde (Biosesang) at 4°C overnight and visualized by crystal violet staining. The number of plaques was counted, and the virus titer was calculated. For infected mice, mouse tissues were chopped in phosphate-buffered saline (PBS) and stored at −80°C until further processing. Viruses were extracted from the chopped tissues by three freeze–thawing cycles. The virus supernatant was collected and clarified by centrifugation and 0.2-µm filter. The virus titers in the supernatant were quantified using the plaque assay method described previously.

Akkermansia muciniphila (ATCC BA-835) cultures were prepared according to a modified protocol adapted from a previous study (26). The bacteria were cultivated in Gifu Anaerobic Medium (GAM; MB Cell, Seoul, South Korea), supplemented with 0.2% mucin (porcine stomach mucin type III, Sigma Aldrich, USA). A frozen glycerol stock stored at -80°C was initially streaked onto a GAM–mucin agar plate and incubated at 37°C for 24 h. A single colony from this plate was subsequently selected for expansion in a 2-L culture. Following large-scale cultivation, bacterial cells were harvested, transferred to sealed 50-mL tubes, and centrifuged at 3,000 rpm for 30 min at 4°C. The pellets were then resuspended in fresh medium and stored as 50% glycerol stocks with a concentration of 6.06 × 10¹⁰ CFU/mL. Colony-forming units (CFU) were quantified by serial dilution of one stock, plating on GAM–mucin agar, and counting colonies after 72 h of incubation at 37 °C. All procedures, except centrifugation, were conducted under strict anaerobic conditions in an anaerobic chamber (Coy Lab Products) with a gas mixture of 85% N₂, 5% H₂, and 10% CO₂.

To deplete the gut microbiota, mice were treated with a broad-spectrum antibiotic cocktail administered via oral gavage for 7 days, followed by an additional 7 days of antibiotic treatment provided in their drinking water. The antibiotic mixture for gavage included ampicillin (1 mg/mL), neomycin (1 mg/mL), metronidazole (1 mg/mL), and vancomycin (0.5 mg/mL) (Sigma Aldrich). After completing the 7-day gavage treatment, the same antibiotic cocktail was supplied in the drinking water for another 7 days. Following the 14-day antibiotic regimen, drinking water was replaced with sterile water for 24 h to ensure the clearance of residual antibiotics. For Akkermansia muciniphila treatment, the mice were orally gavaged with 1 × 10⁸ CFU of A. muciniphila suspended in 200 μL of sterile PBS every other day for a total of 10 treatments. The control mice received an equivalent volume of sterile PBS without bacteria. To confirm successful colonization, fecal samples were collected and analyzed for the presence of A. muciniphila by qPCR targeting the A. muciniphila-specific 16S rRNA genes as described previously (27). Once colonization was verified, the mice were used for SARS-CoV-2 infection experiments or other study protocols.

For the isolation of primary immune cells, lungs and spleens were harvested at 7 days post-infection. Lung tissues were minced into small pieces and digested in RPMI 1640 medium containing 1 mg/mL collagenase type V (Sigma Aldrich) and 20 U/mL DNase I (Sigma Aldrich) for 30 min at 37 °C with gentle stirring. After digestion, the cell suspension was passed through a 40-µm cell strainer to remove debris and obtain single-cell suspensions. Red blood cells (RBCs) were lysed using RBC lysis buffer (BioLegend). For spleen cell isolation, the spleens were mechanically dissociated and filtered through a 40-µm strainer to obtain single-cell suspensions. RBC lysis was performed with RBC lysis buffer (BioLegend).

Single-cell suspensions were washed with ice-cold PBS before being resuspended in 40 μL of PBS for surface marker staining. The following antibodies were used for surface staining: Live Dead (Fixable aqua), CD45 (BUV395), MHCII (BUV496), CD8 (BUV496), NK1.1 (BUV563), CD44 (BUV615), F4/80 (BUV661), CD62L (BUV737), TCRβ (BV421), CD11c (BV480), LY6G (BV570), TCRβ (BV605), CD4 (BV650), TCRβ (BV650), Ly6C (BV786), CD69 (PE), CD103 (PE-Dazzle594), CD19-biotin (PE-Cy5), TCRγδ (APC), CD8 (APC-Cy7), and CD11b (SparkBlue550). For intracellular staining of transcription factors and cytokines, the cells were first surface-stained, followed by fixation and permeabilization using Fixation/Permeabilization Buffer (eBioscience). The intracellular markers used for staining included IFN-γ (BUV737), GRANZYME B (Pacific Blue), TNFα (BB700), IL-2 (PE), PERFORIN (PE-Dazzle594), and FOXP3 (AF488). Tetramer staining was performed by, first, staining the cells with the SARS-CoV-2 peptide-MHC class I tetramer (NIH Tetramer Core Facility), followed by subsequent staining with surface and intracellular markers. Detailed information regarding the antibodies used, including clone numbers and manufacturers, is listed in Supplementary Table S1. The gating strategies for identifying immune cell populations are presented in Supplementary Figure S4. Data acquisition was carried out using a SONY ID7000 flow cytometer, and analysis was performed using FlowJo software (Tree Star). For dimensionality reduction and visualization of immune cell populations, the UMAP (Uniform Manifold Approximation and Projection) algorithm was employed.

Lung homogenates and serum were prepared at 7 days post-infection, and cytokine levels were measured using LEGENDplex™ Mouse Th Cytokine Panel (cat. no. 741043), Mouse Cytokine Panel 2 (cat. no. 740134), and Mouse Proinflammatory Chemokine Panel (cat. no. 740007) (BioLegend, San Diego, CA, USA) according to the manufacturer’s instructions. Data were analyzed using a Luminex 200 system and LEGENDplex software.

Lung tissues were fixed in 10% neutral-buffered formalin (Sigma, St. Louis, MO, USA) for 24 h and embedded in paraffin, sectioned to a thickness of 4 μm, and processed for hematoxylin and eosin (H&E) staining or immunohistochemistry (IHC). For H&E staining, slides were deparaffinized, immersed in 0.1% Mayer’s hematoxylin for 10 min, and counterstained with 0.5% eosin. The slides were dehydrated through a graded ethanol series (50%, 70%, 95%, and 100%) and mounted using a mounting solution (Thermo Fisher Scientific, Waltham, MA, USA). The stained slides were evaluated by veterinary pathologists. The stained slides were independently analyzed by three veterinary pathologists.

For IHC, deparaffinized slides were rehydrated through xylene, 100%, 95%, and 70% ethanol, followed by distilled water. Antigen retrieval was performed using a pH 6.0 citrate buffer (Dako S1699, Agilent Technologies, Santa Clara, CA, USA) under high temperature in a pressure cooker, followed by cooling on ice for 1 h. To block endogenous peroxidase activity, the slides were incubated with 3% H₂O₂ in PBS for 30 min, followed by treatment with M.O. reagent (Vector Laboratories, Burlingame, CA, USA) for 1 hour. The slides were then incubated overnight at 4°C with primary antibodies, including SARS-CoV-2 N protein (NB100-56576, Novus; 40143-MM08, Sino Biological), CD4 (25229s, Cell Signaling), CD8 (98941, Cell Signaling), F4/80 (ab6640, Abcam), LY-6G/LY-6C (ab2557, Abcam), and CD45R(B220) (ab64100, Abcam). Afterward, the slides were treated with a protein-blocking solution (Dako) for 1 h, followed by incubation with an HRP-conjugated secondary antibody (Dako) for 15 min. The IHC signal was developed using DAB substrate (Dako), and nuclear counterstaining was performed with Mayer’s hematoxylin. For immunofluorescence staining, Alexa 488-conjugated anti-mouse, anti-rat, and anti-goat IgG antibodies, along with Cy3-conjugated anti-rabbit IgG antibodies, were applied. Images were acquired using a Zeiss LSM980 confocal microscope. Areas of bronchus-associated lymphoid tissue (BALT), characterized by B220-positive follicles, were identified as PTPRC (B220)-positive regions (28).

Mouse weight change and body temperature were monitored daily for 21 days post-infection. Weight change was calculated as a percentage of the baseline value recorded on the day of infection, and body temperature was measured using an implantable programmable temperature transponder (IP55-300, BMDS).

Fecal DNA extraction from mice was performed on all surviving animals at each time point for microbiome analysis. The total genomic DNA from fecal samples was extracted using the phenol-chloroform extraction method. Briefly, fecal samples were suspended in 300 μL of TE buffer and subjected to mechanical disruption using a bead beater. The disrupted samples were then treated with an equal volume of phenol/chloroform/alcohol (25:24:1) solution and 10% SDS solution centrifuged at 12,000 × g for 10 min at 4°C. The aqueous phase containing DNA was transferred to a new tube and precipitated by adding 1 volume of 100% ethanol. The precipitated DNA was pelleted by centrifugation at 12,000 × g for 10 min at 4°C, washed with 70% ethanol, and resuspended in nuclease-free water.

The 16S rRNA gene was amplified using Illumina-adapted universal primers, 341F and 805R, targeting the V3–V4 region. Polymerase chain reaction (PCR) was performed with ExTaq polymerase (Takara Bio, Kusatsu, Japan), starting with an initial denaturation at 98°C for 3 min, followed by 25 cycles of denaturation at 95°C for 30 s, annealing at 55°C for 30 s, and extension at 72°C for 30 s, with a final extension at 72°C for 5 min. The PCR products were purified and quantified using AMPure XT beads (Beckman Coulter Genomics, Danvers, MA, USA) and Qubit dsDNA High-Sensitivity Assay Kit (Invitrogen; Carlsbad, CA, USA), respectively. The normalized amplicons were prepared for sequencing with Nextera XT DNA Library Preparation Kit (Illumina, San Diego, CA, USA) in combination with Nextera XT Index Kit. Sequencing was carried out on the MiSeq platform using the MiSeq Reagent Kit v3 (600 cycles) (Illumina). Sequencing reads were demultiplexed, and paired-end reads were processed by QIIME2 pipeline (version 2024.5) (29). DADA2 workflow was applied for the quality trimming, denoising, and filtering against chimeric PCR artifacts (30). The resulting exact amplicon sequence variants (ASVs) were then assigned taxonomy using the Greengenes 2 reference database (2022.10) through q2-greengenes plugin within QIIME2 (31). The 16S rRNA gene amplicon datasets obtained by this research were deposited in National Center for Biotechnology Information (NCBI) under accession numbers.

Alpha diversity indices such as observed features and Shannon were calculated using QIIME2 platform. For the diversity comparison among groups, Kruskal–Wallis test was used. Unweighted UniFrac metrics was used to assess beta diversity, and principal coordinate plot was generated by MicrobiomeAnalyst (32). To statistically evaluate the differences in microbial community composition (beta-diversity) between groups, permutational multivariate analysis of variance (PERMANOVA) was performed with 999 permutations using unweighted UniFrac distance. The R package “edgeR” was used to extract differentially abundant genus for volcano plots, and the volcano plots were produced using VolcaNoseR online platform (33, 34). For differential abundance analysis, raw counts were first filtered to exclude low-abundance taxa: samples with fewer than 8,000 total reads and genera with a relative abundance of less than 0.1% were removed. The remaining data were normalized using the trimmed mean of M-values (TMM) method. Differential abundance was determined using a generalized linear model (GLM), and P-values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) approach (FDR < 0.05). To identify differentially abundant microbes considering multivariable associations, we used the R package “MaAsLin2” (Microbiome Multivariable Associations with Linear Models) with two fixed effects (virus types and dpi) and one random effect (mice ID) (35). Longitudinal analysis was performed using the q2-longitudinal plugin and volatility analysis was used to determine how the bacterial abundance changed over the virus infection period (36). The R package “ALDEx2” was used to analyze significant differences in gut microbiota composition and predicted metabolic pathways between groups for merged (aged and young) and paired samples, employing the Wilcoxon rank-sum test (37). Functional pathway prediction was performed using PICRUSt2 based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) database using the 16S rRNA gene sequencing data (38). The ALDEx2 results were visualized as heatmap, PCA plot, and bar chart of effect sizes using the R package, including “ALDEx2” and “pheatmap”.

To investigate changes in Akkermansia abundance following SARS-CoV-2 infection in human patients, we utilized public data from the Sequence Read Archive (SRA) and National Genomics Data Center (NGDC). We analyzed over 700 samples from five independent Bioprojects (PRJNA624223, PRJEB43555, PRJNA650244, PRJNA758913, and CRA003945) and extracted the relative abundance percentage values of “Bacteria” and “Akkermansia muciniphila” using the SRA Taxonomy Analysis Tool (STAT) (Supplementary Table S2). To minimize technical variability and batch effects across different SRA datasets, we applied stringent metadata filtering, including only samples with clearly defined clinical history and infection status. COVID-19 disease severity was classified according to each study’s definitions: asymptomatic and mild cases were designated as severity I, moderate cases showing pneumonia symptoms as severity II, and severe and critical cases as severity III. Considering that Akkermansia, unlike other common gut microbiota, tends to be characterized by carrier and non-carrier status, we first identified Akkermansia presence in each sample and then conducted further analyses specifically on Akkermansia carriers.

All data were expressed as means ± standard error of the mean. Statistical comparisons between groups were performed using unpaired t-test. Correlations between microbiota abundance and weight or temperature change were analyzed using Spearman’s correlation. Statistical significance was set at p <0.05. All statistical analyses were conducted using GraphPad Prism (GraphPad Software).

To investigate the effects of SARS-CoV-2 variants (WA and Omi) on gut microbiota dynamics, we infected B6 K18-hACE2 mice with each variant (10⁵ PFU) and collected fecal samples at multiple time points. Due to the high lethality of WA infection, samples were analyzed up to 5 days post-infection (dpi), while Omi-infected mice, which survived longer, provided samples up to 7 dpi (Figure 1A). Microbial diversity analyses revealed clear differences in the impact of the two variants on gut microbial composition. In WA-infected mice, species richness, measured by observed features, declined significantly at 2 dpi and remained low at 5 dpi. By contrast, Omi-infected mice exhibited only a transient decrease at 2 dpi, followed by recovery at 5 dpi (Figures 1B, C). Similarly, the Shannon diversity index, which captures both richness and evenness, showed a sustained reduction in WA-infected mice but remained stable in Omi-infected mice over the same period. These findings indicate that WA infection induces more profound and persistent loss of microbial diversity, whereas Omi infection allows partial recovery, reflecting differential host–microbe interactions. Beta diversity analysis further highlighted the divergence between the two variants. Principal coordinate analysis (PCoA) of beta diversity demonstrated a significant shift in microbial community structure in WA-infected mice at 5 dpi, indicating persistent dysbiosis (Figure 1D). In contrast, the microbiota composition of Omi-infected mice remained relatively stable, with clustering patterns similar to pre-infection states. This suggests that WA infection causes more severe and lasting disruption of gut microbial homeostasis compared to Omi infection. Taxonomic analysis revealed variant-specific changes in microbial composition. Akkermansia muciniphila (A. muciniphila), a mucin-degrading bacterium implicated in intestinal barrier modulation and immune regulation, was enriched in both WA- and Omi-infected mice. However, its relative abundance was significantly higher in WA-infected mice, where it dominated the microbial community (Figure 1E). This disproportionate increase in A. muciniphila in WA-infected mice suggests a potential role in the exacerbation of gut dysbiosis and host immune dysregulation. Maaslin2 analysis further identified additional taxa enriched in WA-infected mice, including Nanosyncoccus sp003979185, Eubacterium sp000436835, and Lactobacillus johnsonii. Conversely, Omi-infected mice were enriched with taxa such as CAG_485 sp002362485, Kineothrix sp000403275, and Acetatifactor muris (Figure 1F). Volatility analysis of the most significantly altered taxa highlighted the dynamic nature of gut microbiota changes during infection. In WA-infected mice, microbial disruption persisted, with sustained decreases in beneficial taxa and enrichment of dysbiosis-associated species. In contrast, Omi-infected mice exhibited transient disruptions followed by recovery, suggesting the resilience of the gut microbiota to Omi infection (Figure 1G). Collectively, these findings highlight the greater impact of WA infection on gut microbiota composition compared to Omi infection, with WA inducing more profound and persistent dysbiosis. The significant enrichment of A. muciniphila in WA-infected mice underscores its potential role as a key modulator of the gut–lung axis, contributing to the pathophysiology of SARS-CoV-2 infection.

We sought to determine whether WA infection induces specific gut microbiota and functional changes. We analyzed microbial composition and predicted metabolic pathways in both young and aged K18-hACE2 mice since aging is key determinant for COVID-19 severity (39). Consistent with prior findings, aged mice exhibited higher baseline microbial diversity compared to young mice (40, 41), which remained unchanged following WA infection (Supplementary Figures S1A–C). Despite this, a significant increase in Akkermansia abundance was observed in both age groups at 5 days post-infection (dpi), reaffirming its enrichment during WA infection (Figure 2A; Supplementary Figure S1D). To identify microbial taxa specific to WA infection, we performed ALDEx2 pairwise analysis by combining datasets from young and aged mice. This analysis revealed the significant enrichment of Akkermansia and Emergencia after WA infection (p-adjust < 0.05), along with increases in Bacteroides faecalis and Faecalimonas phoceensis. In contrast, the Christensenellales order, previously abundant in pre-infection samples, showed a marked reduction at 5 dpi (Figure 2A). We next analyzed potential functional shifts in the gut microbiome. We used PICRUSt2 to predict microbial metabolic pathways from the same dataset. Principal component analysis (PCA) based on the inferred metagenomic profiles indicated a separation in predicted functional potential before and after WA infection (Figure 2B). While most pathways showed minimal changes (effect size < 1), glycosaminoglycan degradation—a hallmark function of mucin-degrading bacteria like Akkermansia—was significantly increased at 5 dpi (Figure 2C). Furthermore, the predicted abundance of several lipid-related pathways, including carotenoid biosynthesis, steroid biosynthesis, steroid hormone biosynthesis, and sphingolipid metabolism, were inferred to be significantly enriched following WA infection (Figure 2C). These findings align with the lipid-encapsulated nature of SARS-CoV-2 virions and reported lipid metabolism alterations in infected patients (42, 43). These results together suggest that WA infection induces specific changes in gut microbial composition and predict functional potential, characterized by the dominance of Akkermansia and elevated lipid-related pathways. These alterations likely reflect complex host–microbe interactions during SARS-CoV-2 infection and may contribute to the pathophysiology of severe COVID-19.

We first investigated gut microbiota changes associated with SARS-CoV-2 pathogenesis by assessing correlations between microbial taxa and physiological parameters—namely, weight loss and body temperature reduction—following WA strain infection in mice. Among the microbial taxa identified, members of the Akkermansiaceae (Akkermansia muciniphila) and the Eubacterium fissicatena groups showed significant positive correlations with both disease-related parameters, suggesting a potential link between their expansion and infection-induced physiological decline (Supplementary Figures S2A–C). Conversely, several taxa within the Lachnospiraceae and Ruminococcaceae families were negatively associated with weight and temperature loss, suggesting a potentially protective role in the context of infection. To assess whether these patterns were conserved in humans, we analyzed gut microbiota profiles from five independent COVID-19 cohorts (7, 11, 44–46). A. muciniphila was consistently detected at higher frequency and relative abundance in COVID-19 patients compared to healthy controls across all datasets (Supplementary Figures S2D, E). Although data were integrated from multiple studies, the consistent enrichment of Akkermansia across independent cohorts suggests a robust biological signal that outweighs study-specific batch effects. When stratified by disease severity, individuals with mild disease (severity I) exhibited significantly higher A. muciniphila levels than controls, whereas those with moderate to severe disease (severity II and III) showed greater inter-individual variability and a trend toward reduced abundance (Supplementary Figure S2F). These findings suggest that A. muciniphila expansion may represent a host-compensatory or protective response to SARS-CoV-2 infection, prompting us to examine its functional role in disease modulation.

To directly test the potential immunomodulatory role of A. muciniphila in SARS-CoV2 infection, we utilized a pseudo-germ-free mouse model generated by broad-spectrum antibiotic (ABX) pretreatment followed by mono-colonization with A. muciniphila. Notably, infection severity—measured by weight and survival rate—was comparable between ABX-treated infected mice and untreated infected controls, suggesting that microbiota depletion alone had minimal impact on disease outcomes (data not shown). This established a clean baseline to evaluate the specific contributions of A. muciniphila. Following mono-colonization (1 × 10⁹ CFU), the mice were challenged with SARS-CoV-2 (WA strain, 1 × 10² PFU) (Figure 3A). Stool analysis confirmed successful colonization, with A. muciniphila levels in mono-colonized mice reaching approximately fivefold higher than those in SPF animals, suggesting that mono-colonization led to robust and physiologically relevant engraftment of A. muciniphila (Supplementary Figure S3A). Remarkably, A. muciniphila pre-treatment significantly mitigated weight and temperature loss compared to vehicle-treated controls (Figure 3B). Moreover, lung tissue analysis revealed a reduction in SARS-CoV-2 nucleocapsid-positive cells in colonized mice (Figure 3C), although viral titers remained unchanged (Supplementary Figure S3B), indicating that A. muciniphila limits local viral spread without affecting total replication levels. Notably, despite these protective effects, histopathological assessment of the lung showed no significant differences in lung inflammation, edema, or capillary dilatation between the VEH and A. muciniphila groups (Figure 3D), suggesting that the observed benefits were not associated with overt tissue pathology. Immunological profiling of the lung microenvironment demonstrated that A. muciniphila treatment selectively enhanced local cytokine responses. Th2-associated cytokines (IL-4, IL-5, and IL-13) and Th17-associated cytokines (IL-1β, IL-17A, IL-17F, IL-22, and MIP-3α) were significantly upregulated in lung homogenates, whereas systemic cytokine levels in the serum remained unaltered (Figures 3E, F; Supplementary Figure S3C). Flow cytometry and histological analysis further showed increased frequencies of lung-resident CD4⁺ T cells, CD8⁺ T cells, and B cells in A. muciniphila-colonized mice, alongside elevated macrophages and neutrophils (Figures 3G, H; Supplementary Figures S3D, E). These changes were accompanied by the formation of organized lymphoid aggregates resembling inducible bronchus-associated lymphoid tissue (iBALT). Collectively, these data indicate that A. muciniphila fosters localized antiviral immunity and lymphoid tissue organization, potentially contributing to the containment of SARS-CoV-2 pathogenesis without triggering systemic inflammation.

To investigate the immune changes in the lung mediated by A. muciniphila pre-treatment, we analyzed various immune cell populations in lung tissue and compared them to vehicle-treated controls (Supplementary Figure S4). UMAP analysis revealed diverse immune cell populations, including NK cells, CD8⁺ T cells, γδ T cells, monocytes, neutrophils, B cells, and both regulatory and conventional T cells (Figure 4A). Consistent with earlier histological observations (Figure 3), A. muciniphila pre-treatment significantly increased T cell abundance while reducing B cell proportions across lung samples (Figure 4A). Notably, this was accompanied by a significant overall increase in the total number of lung immune cells compared to vehicle-treated mice (Figure 4B). A further subset analysis revealed a pronounced shift toward T cell dominance. While the proportion of B cells decreased, CD4⁺ T cells increased significantly following A. muciniphila treatment (Figures 4A, C). In contrast, no significant changes were observed in NK cells, γδ T cells, or myeloid cell populations (Supplementary Figures S5A, B). Focusing on CD4⁺ T cell subsets, we identified a significant increase in effector CD4⁺ T cells (CD44⁺ CD62L^-) within the lungs of A. muciniphila-treated mice (Figures 4D, E), indicating localized activation of effector T cells. Notably, these changes were lung specific, as no comparable activation was detected in the spleen (data not shown), underscoring that A. muciniphila primarily modulates local, rather than systemic, T-cell responses. To evaluate A. muciniphila’s impact on SARS-CoV-2-specific responses, we analyzed antigen-specific CD8⁺ T cells using a SARS-CoV-2 tetramer. A. muciniphila pre-treatment significantly increased the proportion of activated tissue-resident memory T (T_RM) cells expressing CD69⁺ and/or CD103⁺ markers among SARS-CoV-2-specific CD8⁺ T cells (Figures 4F, G). Although the total numbers of T_RM cells showed only marginal increases, their enrichment highlights A. muciniphila’s role in promoting SARS-CoV-2-specific T_RM cell establishment (Supplementary Figure S5D). Importantly, the abundance of SARS-CoV-2-specific CD69⁺ CD103⁺ T_RM cells negatively correlated with temperature loss during SARS-CoV2 infection (Supplementary Figure S6), suggesting their functional relevance in mitigating infection-induced pathology. These findings collectively indicate that A. muciniphila enhances localized T cell responses, promotes T_RM cell formation, and supports durable immune memory in the lungs, contributing to improved host resilience against viral infections.

To assess the functional impact of A. muciniphila on lung immune responses, we analyzed cytokine secretion profiles in lung-resident immune cells. While no significant changes were observed in non-T-cell populations (Supplementary Figures S7A, B), A. muciniphila-treated mice exhibited a marked enhancement in cytokine production by lung-resident T cells compared to vehicle-treated controls (Figure 5). Notably, the polyfunctionality index, which reflects the ability of individual T cells to secrete multiple cytokines simultaneously, was significantly elevated in A. muciniphila-treated mice (Figures 5A–C; Supplementary Figure S8A). This enhanced polyfunctionality highlights the increased functional capacity of CD4⁺ T cells, particularly in their antiviral responses. Specifically, secretion of TNF-α and IFN-γ, two key cytokines for antiviral immunity, was significantly upregulated in CD4⁺ T cells following A. muciniphila treatment (Figure 5D). A similar enhancement was observed in CD8⁺ T cells, with A. muciniphila-treated mice displaying significantly higher cytokine secretion and polyfunctional capacity compared to controls (Figures 5E–G; Supplementary Figure S8B). The cytokine profile of CD8⁺ T cells revealed a notable increase in TNF-α and IFN-γ production (Figure 5G), further emphasizing the improved antiviral functionality of these cells. Collectively, our results demonstrate that A. muciniphila promotes localized, lung-restricted immune modulation, leading to iBALT formation, T-cell activation, TRM enrichment, and enhanced polyfunctionality. These findings align with clinical observations of gut–lung crosstalk in COVID-19 (6–8) and further highlight A. muciniphila as a potential modulator of respiratory immune defense, supporting its therapeutic potential in viral infections.