Patients diagnosed with NASH based on liver histology were recruited from the Hanyang University College of Medicine (Seoul, Republic of Korea). The diagnostic criteria for NASH were satisfied if patients met the following three conditions: an alcohol consumption of less than 20 g per day, biopsy-proven steatohepatitis, and the absence of other chronic liver diseases. The exclusion criteria comprised any consumption of probiotics or prebiotics within 3 months of the study, pregnancy and/or lactation, or any history of major gastrointestinal surgery. All participants provided written informed consent, and the study was approved by the institutional review board of Hanyang University College of Medicine (IRB number: 2014–03–008-005). Age-and sex-matched healthy subjects recruited in our previous study (Lim M. Y. et al., 2021) were used as the control group in this study.

Fresh stool samples were collected from the participants using OMR-200 fecal sampling kits (OMNIgene GUT; DNA Genotek, Kanata, ON, Canada) and stored at-80°C prior to DNA extraction. Bacterial genomic DNA was extracted from fecal samples according to the instructions of the QIAamp DNA Stool Mini Kit (Qiagen, Hilden, Germany). The V3/V4 hypervariable region of 16S rRNA genes was amplified and sequenced using the Illumina MiSeq 2 × 300 System (Illumina, San Diego, CA, United States) according to the manufacturer’s instructions. Raw sequencing reads were analyzed using the QIIME 2 pipeline (Bolyen et al., 2019). Briefly, raw sequence data were demultiplexed and filtered for quality using the DADA2 plugin (Callahan et al., 2016). Only features belonging to the bacterial domain and in the range of 380–450 base pairs were assessed. De novo chimera filtration was performed using the “vsearch uchime-denovo” program (Rognes et al., 2016). Following data filtration, taxonomy was assigned using a pre-trained naive Bayes classifier against Silva-138 reference sequences. Species assignment was conducted with the “vsearch usearch_global” tool and was based on a 99% identity threshold. The dataset was rarefied to the smallest sample before beta diversity analysis. All raw sequencing data presented in this study were deposited in the Sequence Read Archive database under accession number PRJNA901628 ¹.

Faecalibacterium prausnitzii bacteria were isolated from human feces according to the method described by Martín et al. (2017), with some modifications. Human feces were collected from healthy Koreans aged 7–60 years as approved by the Institutional Review Board of Dongguk University Ilsan Hospital in the Republic of Korea (2018–06–001-012). As detailed in a previous study (Lee et al., 2022), we performed polymerase chain reaction (PCR) tests using species-specific primers for F. prausnitzii (forward primer: 5’-ACTCAACAAGGAAGTGA-3′; reverse primer: 5’-AATTCCGCCTACCTCTG-3′) to identify the isolates, producing a single band of the available product size (192 bp). Following PCR confirmation, we conducted 16S rRNA gene sequencing using 27F primer (5’-AGAGTTTGATCCTGGCTCAG-3′) and 1492R primer (5’-GGTTACCTTGTTACGACTT-3′). For our bacterial studies, F. prausnitzii strains were cultured in soy peptone-based medium containing (per liter): 20 g soy peptone; 10 g yeast extract; 2.5 g K₂HPO₄; 0.5 g l-cysteine hydrochloride, and some supplements. The bacteria were cultured at 37°C in an anaerobic chamber containing 90% N₂, 5% CO₂, and 5% H₂. Cells were harvested via centrifugation at 12,000 × g for 5 min at 4°C. Thereafter, the pellets were resuspended in pre-reduced and sterile phosphate-buffered saline (anaerobic PBS), aliquoted, and stored at-80°C in 20% glycerol.

The minimum inhibitory concentrations (MIC) were determined for the isolates of seven antibiotic classes (piperacillin-tazobactam, ceftizoxime, chloramphenicol, clindamycin, meropenem, moxifloxacin, metronidazole, and ciprofloxacin) which are effective against anaerobic bacteria using the Wilkins–Chalgren medium according to Clinical and Laboratory Standards Institute (2017). All MICs were interpreted using the CLSI breakpoints for anaerobes. The hemolytic activity of the isolates was determined using tryptic soy agar containing 5% (v/v) defibrinated sheep blood, with the plates being incubated at 37°C for 24 h under the anaerobic conditions listed in Section 2.3. After incubation, hemolytic activity was evaluated and classified based on red blood cell lysis in the medium surrounding the colonies. Strains with no zones around the colonies (γ-hemolysis) were considered safe.

Six-week-old female C57BL/6 mice were purchased from Daehan Biolink Co., Ltd. (Chungbuk, Korea). The use and care of animals were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) at Dongguk University (approval number: IACUC-2019-041-1) and conformed with the guidelines of the International Association for the Study of Pain policies on the use of laboratory animals. After 1 week of acclimation, the animals were randomly assigned and housed in standard plastic cages (three mice per cage). A total of 96 mice were randomized into the following eight groups (each group n = 12): a normal control (CON), NASH, NASH with silymarin, A2-165, EB-FPDK3, EB-FPDK9, EB-FPDK11, or EB-FPYYK1 group. All groups were maintained for 16 weeks under different regimens. The normal group was fed a low-fat diet (10 kcal% fat; Research Diets, Inc., NJ, United States) and had free access to plain tap water. NASH model groups were fed a high-fat diet (60 kcal% fat; Research Diets, Inc.) and had free access to water enriched with 30% fructose (high-fructose, HF). Mice in the remaining groups started receiving oral administrations at 8 weeks, consisting of silymarin (NASH with silymarin group) or 1 × 10⁸ CFU of F. prausnitzii strains (A2-165, EB-FPDK3, EB-FPDK9, EB-FPDK11, and EB-FPYYK1 groups). The weight and calorific intake of all mice were measured weekly. After the experiment, the mice were anesthetized to collect blood samples (see Sections 2.6 and 2.7) and euthanized before removal of the spleen, liver, and large intestine. Fresh spleens and livers were weighed. The livers and large intestine were partially sectioned and fixed for histological analysis (Section 2.8), with the remaining tissues stored at-80°C for RNA analysis (Section 2.9).

An OGTT was performed during the last week of the study. After being subjected to 14 h of fasting, the mice were administered oral glucose (2 g/kg), and blood was obtained from the tail vein 0, 30, 60, and 120 min after glucose treatment. Glucose levels (mg/dL) were measured using Accu-Chek test strips on an Accu-Chek Active blood glucose meter (Roche Diagnostics, Rotkreuz, Switzerland). The glucose area under the curve was calculated by plotting the glucose concentration as a function of time (min).

Whole blood samples were centrifuged (2000 × g for 15 min at 4°C) to separate the serum. The serum triglyceride (TG), total cholesterol (TC), aspartate aminotransferase (AST), and ALT levels were measured using assay kits (Asan Pharmaceutical, Seoul, Korea). Lipids were extracted from the liver tissue according to the Folch protocol (Folch et al., 1957).

The liver and large intestine were fixed in neutral-buffered 10% formalin solution, embedded in paraffin wax, and sectioned at a thickness of 4 μm using a microtome. Hematoxylin and eosin (H&E) staining, Sirius Red staining, and α-SMA (ab5694; Abcam, Cambridge, MA, United States) immunohistochemistry (IHC) were performed on the liver sections. To assess the degree of steatosis, lobular inflammation, and hepatocyte ballooning, a NAFLD Activity Score (NAS) was assigned to each group. The NAS system was proposed by the National Institute of Diabetes and Digestive and Kidney Diseases–NASH Clinical Research Network, and the score range is described in Table 1. The thickness of the mucosa and muscularis externa in the large intestine tissue sections was measured with a Nikon Eclipse Ni microscope (Nikon Corporation, Tokyo, Japan). Large intestine sections were stained with anti-zonular-1 antibody (67–7,300; Invitrogen, Waltham, MA, United States) and anti-occludin antibody (71–1,500; Invitrogen) for the analysis of tight junctions. The area of liver fibrosis was quantified using ImageJ software (NIH, Bethesda, MD, United States).

Total RNA was extracted from homogenized liver and large intestine tissues using TRIzol Reagent (Life Technologies, Carlsbad, CA, United States) and purified using RNA PureLink RNA Mini Kits (Thermo Fisher Scientific, Waltham, MA, United States) according to the manufacturer’s instructions. Complementary DNA (cDNA) was synthesized with a reaction micture of volume 20 μl, containing 2 μg of pure RNA, oligo dT primer (M-MLV cDNA Synthesis Kit, Enzynomics), and an M-MLV reverse transcriptase (M-MLV cDNA Synthesis Kit, Enzynomics, Daejeon, Korea) according to the manufacturer’s instructions. For quantitative real-time polymerase chain reaction (qRT–PCR), the cDNA (2 μl) was mixed with primer pairs (250 nM each) and 10 μl of qPCR 2× SYBR Green Premix (Enzynomics, Daejeon, Korea) in reaction mixture of volume 20 μl. After initial denaturation at 95°C for 10 min, cDNA was amplified for 40 cycles of denaturation (95°C, 15 s) and annealing (60°C, 1 min) using QuantStudio3 (Applied Biosystems). The results were normalized to glyceraldehyde-3-phosphatase dehydrogenase (GAPDH). All primer sequences are listed in Table 2.

Using the QIAamp® DNA Stool Mini Kit, DNA was extracted from cecal samples of control, NASH-induced, EB-FPDK9, and EB-FPDK11-treated mice (n = 7 per group). 16S rRNA hypervariable regions in V1–V2 were amplified by using primers containing unique 10-base barcodes and sequenced by using the Ion Torrent PGM system according to manufacturer’s instructions. Quality-filtered raw sequence reads were clustered into amplicon sequence varients (ASVs) using the SILVA rRNA gene database with a 99% sequence identity threshold. The software Quantitative Insights into Microbial Ecology 2 (QIIME2) was used to select representative reads and calculate alpha-and beta-diversity. The linear discriminant analysis (LDA) effect size (LEfSe) was used to identify taxa with varying abundances between groups, with a logarithmic LDA score threshold of 2.0 and an alpha value of 0.05 for the factorial Kruskal-Wallis test.

To extract short chain fatty acids (SCFAs), the cecum (50 mg) was mixed with 800 μl of distilled water and 10 μl of 5 M HCl and after adding 400 μl of ether, the mixture was shaken at 4°C for 5 min. After spin down, 200 μl of ether layer was derivatized by adding 20 μl of N, O-bis (trimethylsilyl) trifluoroacetamide (BSTFA) at 70°C for 20 min and then incubated at 37°C for 2 h. The derivatized SCFAs were analyzed using a GC/MS system (Shimadzu Corp., Kyoto, Japan) equipped with a DB-5MS column (30 m × 0.25 mm, 0.25 μm film thickness, Agilent Technologies, Santa Clara, CA, United States) at a split ratio of 1: 50. The injector temperature was set at 200°C, and helium was used as the carrier gas at a flow rate of 0.89 ml/min. The oven temperature program was set as holding at 40°C for 2 min, increasing from 40 to 70°C at a rate of 10°C/min, increasing from 70 to 85°C at a rate of 4°C/min, increasing from 85 to 110°C at a rate of 6°C/min, increasing from 110 to 290°C at a rate of 90°C/min, and holding 290°C for 5 min. The effluent was detected using a GCMS-TQ 8030 MS (Shimadzu Corp.) system with selected ion monitoring (SIM). The ion source and interface temperatures were 200 and 250°C, respectively, and detector voltage was 0.1 kV. SCFAs were detected by SIM mode with m/z 117, 131, and 145 of acetic acid, propionic acid, and butyric acid, respectively. Authentic standard SCFAs were used to quantitative analysis of SCFAs.

MAM sequences were identified from four F. prausnitzii genomes (EB-FPDK3, EB-FPDK9, EB-FPDK11, and EB-FPYYK1) by BLAST search with low e-values against the MAM sequence from A2-165 as the query (Quévrain et al., 2016; Auger et al., 2022; Hong and Nam, n.d.). Translated MAM sequences were aligned by using Clustal Omega (Sievers et al., 2011) to investigate MAM-derived peptides which were previously identified as pep1–5 (Quévrain et al., 2016). Further, homology models of MAM proteins derived from four F. prausnitzii strains were provisionally built using the Modeler program (Webb and Sali, 2016).

All data are expressed as the arithmetic mean ± standard error of the mean (SEM). We used GraphPad Prism 7 (GraphPad, San Diego, CA, United States) for all statistical analyzes, comprised of Mann–Whitney U tests (two-tailed) or Kruskal–Wallis one-way ANOVAs, with Dunn’s test to correct for multiple comparisons (unless indicated otherwise). We considered a p value <0.05 to indicate statistical significance.

To explore the dysbiosis of gut microbiota in NASH, we performed 16S rRNA gene amplicon sequencing on fecal samples from patients with NASH and compared the results with those obtained for healthy individuals. We used our previously published data (Lim M. Y. et al., 2021) on the gut microbiota composition of healthy Korean subjects (n = 99), who were age-and sex-matched with patients with NASH (n = 45). Gut microbiota compositional discrimination was observed via unweighted UniFrac principal component analysis between the NASH and healthy control groups (Figure 1A). Bacterial diversity (according to the Shannon index, Faith’s PD index, observed features, and Chao1 index) in NASH patients was significantly lower compared to the healthy controls (Figure 1B). We also observed dysbiosis in the gut microbiota profile, particularly deficient F. prausnitzii and abundant Fusobacterium mortiferum (Figure 1C). As a next step, we found that two ASVs belonging to F. prausnitzii, 5fdd92ad3225b67f02453b5c4590b968 and 692ed0000e9f6e5d47f92a7c59d88434, showed significant differences between the groups (Supplementary Figure S1A). Using these ASVs, strains EB-FPDK3 and EB-FPDK9 were matched to ASV 692ed0000e9f6e5d47f92a7c59d88434, whereas strains EB-FPDK11 and EB-FPYYK1 were matched to ASV 5fdd92ad3225b67f02453b5c685 with only minor variations (Supplementary Figure S1B). Therefore, we hypothesized that restoration of these strains in the gut could alleviate NASH symptoms. To test this hypothesis, we isolated four F. prausnitzii strains (EB-FPDK3, EB-FPDK9, EB-FPDK11, and EB-FPYYK1) from fecal samples collected from four healthy individuals (Supplementary Figure S2). Before administering these strains to mice, we confirmed their basic safety via laboratory tests for hemolytic activity and antibiotic susceptibility. The hemolytic activities of five strains (the four isolates plus the type strain, A2-165) were evaluated on blood agar plates. None of the tested strains showed α-hemolytic or β-hemolytic activities when grown on blood agar plates. All five strains showed γ-hemolytic activity, that is, negative or no hemolytic activity. Susceptibility to antimicrobial agents was assessed according to the CLSI guidelines. All strains, including the type strain A2-165, were susceptible to clindamycin (MICs <0.125) and metronidazole (MICs ranging from ≤0.125 to 4 μg/ml). However, all strains were resistant to meropenem (MICs >64 μg/ml) and fluoroquinolones (moxifloxacin and ciprofloxacin, MICs ranging from 8 to >32 μg/ml; Supplementary Table S1).

A high-fructose and high-fat (HFHF) diet is widely used to induce NASH in animal models to achieve a state that most closely resembles the human NAFLD (Rafiq et al., 2009; Im et al., 2021). To examine the effect of F. prausnitzii on NASH, five F. prausnitzii strains (comprising the reference strain A2-165 and four F. prausnitzii isolates) were orally administered to HFHF-fed mice for 9 weeks. Silymarin was used as a positive control; it contains a mixture of flavonolignans extracted from milk thistle and has been used as a natural drug against liver disease for centuries (Kim et al., 2012; Ni and Wang, 2016).

HFHF feeding of mice in the NASH group led to a significant increase in body weight and calorie intake compared to mice following a normal diet in the CON group (Figures 2A,B). There were no significant differences in body weight or calorie intake in any of the treatment groups (referring to the groups in which either silymarin or a bacterial strain was administered) compared to the NASH group. To confirm the effect of F. prausnitzii on glucose homeostasis, an OGTT was performed after 16 weeks using blood from mice that had been subjected to a fasting state for 14 h. The blood glucose level reached its highest level 30 min after glucose administration and gradually decreased (Figure 2C). The area under the receiver operating characteristic curve was significantly higher in the NASH group than in the CON group (Figure 2D). Silymarin, A2-165, EB-FPDK9, and EB-FPDK11 supplementation significantly reduced the blood glucose levels compared to those in the NASH group (p < 0.05). This indicates that A2-165, EB-FPDK9, and EB-FPDK11 bacterial strains were effective at improving glucose tolerance.

NASH mouse models are characterized by the accumulation of hepatic lipids (Saito et al., 2015). To investigate lipid accumulation in the livers of our study animals, we measured the concentrations of TG and TC. Their levels in both the liver and serum were significantly increased in the NASH group compared to the respective levels in the CON group (Figures 3A–D). All treatments significantly decreased TG and TC levels in the liver compared to those in the NASH group (Figures 3A,B). Serum TG levels were significantly decreased in the groups administered with silymarin and all F. prausnitzii strains except EB-FPYYK1; the largest decreases were recorded in the EB-FPDK3 and EB-FPDK11 groups. Serum TC levels were significantly decreased by silymarin, EB-FPDK9, and EB-FPDK11 treatment (Figures 3C,D). Serum AST and ALT levels were significantly higher in the NASH group than in the CON group (Figures 3E,F). Compared to those in the NASH group, the serum AST levels were significantly lower in all experimental groups, while the serum ALT levels were significantly decreased in all experimental groups except EB-FPYYK1.

Liver images were captured to investigate the extent of liver damage. The livers of mice in the NASH group displayed an enlarged volume of lipid accumulation characterized by a yellow color and hard texture compared to the CON group, in which hepatic lipids exhibited a soft texture with a smooth, red-brown surface (Figures 4A,C). For histopathological analysis, microscope images were obtained of H&E-stained liver tissues (Figure 4B). In the NASH group, we observed microvascular steatosis by lipid deposition and abnormal hepatocyte morphology, such as lobular inflammation and ballooning. When each group was scored according to the NAS criteria listed in Table 1, no abnormal morphology was observed in the CON group, whereas the NAS scores of the EB-FPDK3, EB-FPDK9, and EB-FPDK11 groups were significantly lower than that of the NASH group (Figure 4D). These results show that the oral administration of F. prausnitzii strains reduces the amount of lipid accumulation in and subsequent damage to the liver, as induced by NASH.

The progression to fibrosis and cirrhosis is a key challenge in human NAFLD. To assess the degree of fibrosis that NASH can induce in the liver, we performed Sirius Red assays (Figures 5A–C). By measuring the Sirius Red-positive areas of livers with ImageJ software, hepatic fibrosis progression was mainly observed in the periportal region of all mouse groups. The livers of NASH group mice showed a significantly larger Sirius red-positive area than those of the EB-FPDK3, EB-FPDK9, EB-FPDK11, EB-FPYYK1, and silymarin groups (Figures 5A,B). Our IHC assay used α-SMA as a marker for evaluating stellate cell activation and fibrosis progression (Akpolat et al., 2005). It identified α-SMA-positive areas predominantly in lipid-accumulated foci. These regions were significantly larger in the NASH group than in the CON group, while they were significantly smaller in all treatment groups compared to the NASH group (Figures 5A,C). We also found that the expression levels of collagen type 1 a 1 and metallopeptidase inhibitor 1 were significantly decreased by F. prausnitzii infection (Figures 5D,E). These results indicate a reduction in fibrosis and collagen deposition in the liver after the oral administration of F. prausnitzii.

The consumption of an HFHF diet can shift the intestinal luminal composition and increase the risk of gut leakiness (Moreira et al., 2012). Using microscope photographs of H&E-stained large intestine tissues, we measured the thickness of the mucosa and muscularis externa where the damaged colonic barrier appeared thinner (Figures 6A–C). In the NASH group, the mucosa and muscularis externa were significantly thinner compared to those in the CON group. The silymarin group showed significantly increased mucosal and muscularis externa thicknesses compared to the NASH group. Among the F. prausnitzii strains, EB-FPDK9, EB-FPDK11, and EB-FPYYK1 showed significant efficacy in improving both the mucosa and muscularis externa thickness.

We investigated the efficacy of barrier function in the large intestine via IHC assay, using the expression of zonula occludens-1 (ZO-1) and occludin (OCLN) proteins as indicators (Figures 6D–F). The tight junction protein ZO-1 is essential for barrier function and plays a critical role in the effective mucosal repair (Kuo et al., 2021). In the NASH group, the ZO-1-positive area was significantly smaller compared with that in the CON group, suggesting barrier damage (Figure 6E). Among the F. prausnitzii treatments, only that of the EB-FPDK11 strain resulted in a significant increase in ZO-1-positive area compared to the NASH group. OCLN is another tight junction protein that is crucial for maintaining the epithelial barrier (Chelakkot et al., 2018). The OCLN-positive area was significantly smaller in the NASH group than in the CON group, but the area was recovered in mice receiving EB-FPDK9 and EB-FPDK11 treatments (Figure 6F).

To understand the mechanisms underlying the observed effects of F. prausnitzii treatment in a NASH mouse model, we evaluated the key signaling pathways involved in the modulation of lipid metabolism, as quantified by real-time reverse-transcriptase PCR. The levels of mRNA expressing the transport proteins CD36 and fatty acid transport protein 5 (FATP5) were approximately 1.6- and 1.99-fold higher in the NASH group than in the CON group, respectively (Figures 7A,B). Additionally, the levels of mRNA expressing peroxisome proliferator-activated receptor gamma (PPAR-γ), sterol regulatory element-binding protein-1c (SREBP-1c), fatty acid synthase (FAS), and lipoprotein lipase (LPL) were also significantly higher in the NASH group. Our PCR results revealed that the expression level of genes involved in lipid metabolism was significantly reduced by oral administration of F. prausnitzii (Figures 7C–F). This indicates that F. prausnitzii bacteria can regulate the genetic mechanisms underlying lipid metabolism as related to the alleviation of hepatic steatosis in NASH mice.

We also investigated the expression levels of mRNA that encode key inflammatory cytokines, namely, tumor necrosis factor-α, monocyte chemoattractant protein-1, and interleukin-6. The levels of these cytokines in the NASH group were significantly higher than those in the CON group (Figures 8A–C). However, oral administration of F. prausnitzii strains attenuated the severity of the hepatic inflammatory state by preventing an increase in inflammatory gene expression levels induced by NASH (Figures 8A–C). In particular, F. prausnitzii EB-FPDK9, EB-FPDK11, and EB-FPYYK1 showed a significant reduction in gene expression for all three cytokines compared with the NASH group. The mRNA expression of Toll-like receptor 4 was also significantly higher in the NASH group than in the CON group. However, the oral administration of F. prausnitzii significantly reduced the mRNA expression of Toll-like receptor 4 as induced by NASH (Figure 8D). This receptor is responsible for pathogen detection and the initiation of cytokine production. Our results, therefore, show that F. prausnitzii alleviated liver inflammation by regulating the expression of related genes in NASH-induced mice.