Bacillus subtilis pb2441 partially attenuates hepatic steatosis with transcriptomic changes broadly similar to dapagliflozin in a Western diet mouse model Authors: Yangrae Cho, Min-Kyu Kim, Kyoung-Sik Moon Published: 2026 ABSTRACT AbstractNonalcoholic fatty liver disease (NAFLD) is characterized by hepatic lipid accumulation accompanied by inflammation and fibrosis. While pharmacological interventions such as SGLT2 inhibitors have demonstrated clinical benefits, the potential of probiotic-based approaches remains under investigation. In this study, we evaluated the effects of Bacillus subtilis pb2441 in a Western diet (WD)-induced mouse model of NAFLD and compared its effects with dapagliflozin. WD feeding resulted in increased body weight, liver weight, hepatic lipid accumulation, and expression of inflammation-, fibrosis-, and lipid metabolism–related genes, including increased lipogenesis and reduced β-oxidation. Transcriptomic analyses showed that pb2441 reduced lipogenesis-related gene expression while restoring β-oxidation–associated genes relative to the WD group, albeit to a lesser extent than dapagliflozin. H&E and Oil Red O staining demonstrated WD-induced hepatic steatosis and lipid accumulation, both of which were partially attenuated by pb2441 and dapagliflozin. Hepatic triglyceride levels were also reduced in both treatment groups, with a statistically significant decrease observed in the dapagliflozin group. Integrative analyses demonstrated that hepatic lipid accumulation was positively associated with inflammation scores, fibrosis-related gene expression, and serum ALT levels. Together, these findings indicate that pb2441 partially attenuates WD–induced hepatic steatosis and associated transcriptional changes, with broadly similar effects to dapagliflozin.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-56352-5.Keywords: Nonalcoholic fatty liver disease (NAFLD), Hepatic steatosis, Probiotics, Bacillus subtilis pb2441, Dapagliflozin, RNA sequencing (RNA-seq)Subject terms: Diseases, Gastroenterology Introduction Nonalcoholic fatty liver disease (NAFLD) is now the most common chronic liver disease worldwide1. It is strongly linked to modern dietary patterns characterized by excess calories, fat, and sugar2. NAFLD spans a clinical spectrum from simple steatosis to nonalcoholic steatohepatitis (NASH), which can progress to fibrosis, cirrhosis, and hepatocellular carcinoma3,4. Importantly, NAFLD/NASH frequently coexists with obesity, insulin resistance, type 2 diabetes, and dyslipidemia, reflecting shared disruptions in energy balance, glucose regulation, and hepatic lipid metabolism5,6. These connections highlight the need for interventions that can improve systemic metabolic status while mitigating diet-driven hepatic steatotic changes. Preclinical models of NAFLD commonly employ high-fat diets (HFD); however, these models do not fully recapitulate the complexity of human dietary patterns, particularly the combined effects of excess fat, cholesterol, and simple sugars7. In contrast, Western diet (WD) models incorporate these components and are known to induce not only hepatic lipid accumulation but also inflammatory and fibrotic responses that more closely resemble human NAFLD progression8,9. For this reason, we employed a WD model to better capture diet-induced hepatic pathology beyond simple steatosis. Therapeutic options for metabolic disease have expanded rapidly in recent years. Incretin-based therapies, including glucagon-like peptide-1 (GLP-1) receptor agonists, have demonstrated robust glycemic benefits and clinically meaningful effects on body weight and cardiometabolic risk10,11. Sodium-glucose cotransporter 2 (SGLT2) inhibitors have also shown substantial glycemic improvements, along with favorable effects on body weight12–14, cardiometabolic outcomes15,16, and hepatic steatosis endpoints17,18. In this study, dapagliflozin was selected as a pharmacological comparator because it represents a clinically established metabolic therapy that improves glycemic control and body weight, while also demonstrating beneficial effects on liver-related outcomes19,20. Mechanistically, SGLT2 inhibition lowers glycemia by blocking renal glucose reabsorption and increasing urinary glucose excretion21,22, providing a relevant benchmark for evaluating metabolic and hepatic effects of pb2441. In parallel with chemical-based drugs, probiotic candidates are being explored as dietary interventions to support metabolic and liver health23. Probiotics are typically developed as functional foods or dietary supplements rather than conventional pharmaceuticals, which places a premium on safety for preventive use in broadly healthy or at-risk populations. Mechanistically, probiotics may influence NAFLD pathogenesis through multiple, non-mutually exclusive routes, including remodeling of the gut microbiota, strengthening of intestinal barrier integrity, and immunomodulatory effects that reduce metabolic inflammation23–25. Consistent with this concept, several probiotic strains have been reported to attenuate weight gain and reduce hepatic steatotic features in diet-induced obesity models, supporting their potential as adjunctive or preventive approaches for NAFLD26–28. Although SGLT2 inhibitors have been extensively studied for metabolic benefits, whole-liver transcriptomic profiling under diet-induced steatotic conditions—particularly in direct comparison with probiotic interventions—remains limited. Here, we investigate Bacillus subtilis pb2441, a probiotic strain that attenuates body weight gain and hepatic steatotic features in a WD model. We hypothesized that pb2441 reduces net hepatic lipid accumulation by suppressing lipogenesis while enhancing fatty-acid beta-oxidation in the liver. To benchmark pb2441 against a well-characterized metabolic therapy and to probe whether pb2441 exerts effects through mechanisms distinct from pharmacologic glucose lowering, we compared pb2441 with dapagliflozin. Using phenotypic measurements, liver histology, and whole-liver RNA-seq, we evaluated whether pb2441 induces transcriptional remodeling consistent with reduced lipid synthesis and increased lipid utilization, and whether these patterns converge with or diverge from those observed under dapagliflozin. Results Effects of Western diet, dapagliflozin, and pb2441 on body weight In this study, dapagliflozin was included as a pharmacological comparator because of its well-established metabolic benefits and emerging evidence in NAFLD. At study initiation, the mean body weight was 20.7 ± 0.63 g. After 8 weeks, mice fed the NCD increased to 26.2 ± 1.59 g, whereas WD-fed mice reached 37.0 ± 3.73 g. Continued WD feeding for an additional 40 days further increased body weight to 46.6 ± 2.7 g in the WD control group. Mice fed a WD containing dapagliflozin or B. subtilis pb2441 showed attenuated weight gain from week 2 onward compared with the WD control group (p < 0.01; Fig. 1a–c), resulting in lower final body weights at day 40 (pb2441: 44.3 ± 2.8 g; dapagliflozin: 44.0 ± 3.2 g). pb2441 and dapagliflozin attenuate Western diet (WD)-induced weight gain with distinct energy-intake patterns. (a) Body weight trajectories during 8 weeks of diet induction (NCD vs WD), followed by a 40-day intervention with WD alone, WD supplemented with Bacillus subtilis pb2441, or WD supplemented with dapagliflozin. (b) Enlarged view of the boxed region in (a), highlighting body-weight changes during the intervention period. (c) Relative body weight change during the intervention period (percent change from the start of intervention). (d–f) Energy intake (kcal/day) from total intake (d), solid food (e), and drinking (f) before and during the intervention; bars show mean per cage, and P values indicate within-group before–after comparisons. (g) Tail-vein whole-blood glucose at week 8 (pre-intervention). (h) Tail-vein whole-blood glucose after the 40-day intervention. (i) Liver weight at necropsy. (j–l) Serum enzyme levels of AST (j), ALT (k), and ALP (l). (m) Hepatic triglyceride (TG) levels. For time-course panels (a–c), asterisks indicate significant differences versus the WD control at the corresponding time point (*p < 0.05; **p < 0.01; ***p < 0.001; two-way repeated-measures ANOVA with Sidak’s multiple-comparisons test). For box plots (g–l), different letters indicate significant differences among groups (p < 0.05; one-way ANOVA with Tukey’s multiple-comparisons test); groups sharing a letter are not significantly different. Box plots show the median and interquartile range, with whiskers indicating the range; points represent individual mice. Energy intake Total energy intake increased from the pre-intervention (8-week) period to the 40-day intervention period in the NCD group (p < 0.01) and the dapagliflozin group (p < 0.05), showed a trend toward increase in the WD control group (p = 0.09), and showed a trend toward decrease in the pb2441 group (Fig. 1d). This increase was driven by both solid-food– and drinking-derived energy in the dapagliflozin group, whereas in the NCD and WD control groups it was primarily driven by solid-food intake (Fig. 1e-f). During the pre-intervention 8-week WD feeding period, energy intake did not differ among WD-fed mice later assigned to the WD control, pb2441, or dapagliflozin groups (Suppl. Figure 1a–c online). During the subsequent 40-day intervention, the dapagliflozin group increased drinking-derived energy intake, whereas the pb2441 group tended to reduce solid-food intake (Suppl. Figure 1d–f online). Blood glucose At week 8, tail-vein whole-blood glucose levels were higher in the WD group than in the NCD group (Fig. 1g). After the subsequent 40-day intervention, both pb2441 and dapagliflozin lowered glucose relative to the WD control (p < 0.01), with dapagliflozin also lower than pb2441 (p < 0.05) (Fig. 1h). Because fasting duration differed between weeks 8 and 14, glucose values were interpreted within each time point. Liver weight, hepatic enzyme levels, and hepatic triglyceride (TG) levels Liver weight was higher in the WD group than in the NCD group (Fig. 1i; p < 0.001) and was reduced by both pb2441 and dapagliflozin treatment (p < 0.001), with no appreciable difference between the two intervention groups. Serum AST and ALT levels were elevated in the WD group. pb2441 significantly reduced AST and ALT, whereas dapagliflozin showed a tendency toward lower AST and ALT that did not reach statistical significance (Fig. 1j-k). Serum ALP also differed among groups and was reduced in the pb2441- and dapagliflozin-treated groups relative to the WD group (Fig. 1l). Hepatic TG levels differed significantly among groups (Kruskal–Wallis test, p = 0.023). Compared with the WD group, hepatic TG levels were markedly reduced in the dapagliflozin-treated group (Mann–Whitney U test, p < 0.01) and showed a decreasing trend in the pb2441-treated group (p = 0.065). No significant difference was observed between the pb2441 and dapagliflozin groups (p = 0.72). The reduction in hepatic TG levels was consistent with histological findings indicating reduced hepatic steatosis, including decreased Oil Red O staining and lower NAS values. Histological assessment of hepatic steatosis Histological assessment of liver tissues was performed using H&E, Oil Red O, and Sirius Red staining (Fig. 2a-c). H&E staining revealed characteristic features of hepatic steatosis in the WD group, including hepatocellular vacuolation, which appeared less prominent in the pb2441- and dapagliflozin-treated groups. To quantitatively assess steatotic changes, NAS was evaluated. NAS values were elevated in the WD group and were lower in both pb2441 and dapagliflozin groups (Fig. 2d). Oil Red O staining further confirmed increased hepatic lipid accumulation in WD-fed mice, which was reduced following pb2441 and dapagliflozin treatment (Fig. 2e). Together, these histological and quantitative analyses support the presence of diet-induced hepatic steatosis and its partial attenuation by both pb2441 and dapagliflozin, with comparable overall effects. pb2441 and dapagliflozin attenuate Western diet–induced hepatic steatosis with limited effects on collagen deposition. (a) Representative H&E-stained liver sections showing steatotic morphology and hepatic architecture across experimental groups. (b) Representative Oil Red O–stained liver sections highlighting neutral lipid accumulation. (c) Representative Sirius Red–stained liver sections showing collagen deposition in liver tissues. (d) NAFLD activity score (NAS) quantified from H&E-stained sections. (e) Quantification of Oil Red O–positive lipid area in liver tissues. (f) Quantification of Sirius Red-positive collagen area in liver tissues. Limited effects on hepatic collagen deposition Sirius Red staining demonstrated mild collagen deposition in WD-fed mice with moderate inter-individual variability. The relatively limited collagen deposition suggested that the WD model used in this study primarily represented hepatic steatosis with early fibrotic remodeling rather than advanced fibrotic NASH. Quantification of Sirius Red-positive area showed a tendency toward reduced collagen deposition in the pb2441-treated group, although the difference did not reach statistical significance (Fig. 2c and f). Differential gene expression profiles Using padj < 0.05 (no |log2FC| cutoff), the WD vs NCD comparison identified 6734 DEGs. Relative to WD, dapagliflozin induced 1475 DEGs, whereas pb2441 induced 168 DEGs (Fig. 3; Suppl. Table 1 online). A direct comparison between pb2441 and dapagliflozin further highlighted broader transcriptomic remodeling under dapagliflozin (Fig. 3). Reversal/rescue analysis based on the WD vs NCD disease signature showed broadly similar directional reversal rates across interventions, while the number of significantly reversed genes was higher with dapagliflozin than with pb2441 (Suppl. Table 2 online). Volcano plots of differential expression in pairwise comparisons. Volcano plots show DESeq2 results for the indicated contrasts (log2 fold change versus − log10 p-value). Each point represents a gene; genes passing the significance threshold are highlighted. Dashed lines indicate log2 fold change = 0 and p = 0.05. Targeted metabolic gene set analyses Targeted analyses of metabolic gene sets implicated in hepatic lipid handling showed concordant directional changes in the dapagliflozin and pb2441 groups. Across adipogenesis/lipogenesis-associated genes, multiple transcripts related to de novo lipogenesis were reduced during the intervention period (Fig. 4a–c and Suppl. Figure 2), whereas β-oxidation–associated transcripts were increased in both treatment groups compared to the WD group, with a more pronounced induction observed in the dapagliflozin group (Fig. 4d–g and Suppl. Figure 3). Representative cytochrome P450 transcripts also showed intervention-associated remodeling under WD (Fig. 4h–I and suppl. Figure 4). Expression of representative hepatic lipid-handling genes. Box-and-point plots show variance-stabilizing transformation (VST) expression values for selected genes involved in hepatic lipid handling. (a–d) Lipogenesis/adipogenesis-associated genes: Acaca (acetyl-CoA carboxylase; malonyl-CoA production), Elovl5 (fatty-acid elongase), Fasn (fatty-acid synthase), and Cebpa (C/EBPα; transcriptional regulator of adipogenic/lipogenic programs). (e–g) Mitochondrial fatty-acid β-oxidation genes: Acaa2 (mitochondrial thiolase), Cpt2 (carnitine palmitoyltransferase 2; mitochondrial fatty-acid import), and Acadvl (very-long-chain acyl-CoA dehydrogenase). (h–i) Representative cytochrome P450 genes within lipid-oxidation programs: Cyp4a32 and Cyp4a10 (fatty-acid ω-oxidation/PPARα-linked genes). Points represent individual animals; boxes indicate the median and interquartile range (whiskers, 1.5 × IQR). p values for pairwise comparisons are shown above brackets (statistical approach as described in Methods). NCD, normal chow diet; WD, Western diet; pb2441, WD supplemented with B. subtilis pb2441; Dapa, WD supplemented with dapagliflozin. Metabolic gene–based unsupervised clustering and response heterogeneity Unsupervised analyses based on the metabolic gene set (adipogenesis/lipogenesis, β-oxidation, and cytochrome P450 genes), including hierarchical clustering/heatmaps and PCA, separated NCD, WD, and dapagliflozin samples, whereas pb2441 samples were more broadly distributed, suggesting greater within-group variability in this dataset (Fig. 5a-b). Centroid-based similarity scoring using the top 2,000 variable transcripts (Fig. 5c) and NCD-like scores derived from WD–NCD DEGs (Fig. 5d) indicated that both intervention groups remained closer to a WD-like than an NCD-like transcriptomic state overall, with intermediate NCD-like scores relative to the NCD and WD controls. Unsupervised clustering and centroid-based similarity of hepatic metabolic gene expression. (a) Heatmap and hierarchical clustering of a predefined metabolic gene set (adipogenesis/lipogenesis, β-oxidation, and cytochrome P450 genes). VST expression values were z-scored per gene across samples; rows were clustered by hierarchical clustering. Colored markers indicate gene-set membership. (b) Principal component analysis (PCA) of the same metabolic gene set. (c) Sample-level similarity to NCD and WD transcriptomic centroids computed using the top 2000 variable transcripts (Pearson correlation to each centroid). (d) NCD-like scores derived from centroid correlations using WD vs NCD differentially expressed genes; higher scores indicate greater similarity to the NCD centroid. Points represent individual samples; boxes indicate the median and interquartile range. Integration of hepatic transcriptomic signatures with histological and physiological phenotypes To assess the biological relevance of transcriptomic alterations, we integrated gene expression–derived scores with histological and physiological parameters (Fig. 6). Western diet markedly increased the expression of canonical hepatic inflammation markers (Tnf, Il1b, Il6, Ccl2, Cxcl10, Nlrp3, and Socs3) compared to the NCD group (Fig. 6a). Treatment with B. subtilis pb2441 or dapagliflozin showed a consistent trend toward reduced expression, although not all comparisons reached statistical significance. A similar pattern was observed for the expanded TNF-related gene set (Fig. 6b), supporting modulation of inflammatory signaling pathways. Fibrosis-associated genes (Col1a1, Col3a1, Tgfb1, Acta2, Timp1, Mmp9, Mmp20, and Pdgfrb) were also upregulated in the WD group, while pb2441 and dapagliflozin treatments partially attenuated this increase (Fig. 6c). Integrated analysis of hepatic transcriptomic signatures and physiological phenotypes. (a) Canonical hepatic inflammation marker score (Tnf, Il1b, Il6, Ccl2, Cxcl10, Nlrp3, and Socs3). (b) Expanded TNF-related gene score (Tnf, Ccl2, Cxcl10, Socs3, C1qtnf6, Tnfrsf10b, Tnfrsf22, Tnfaip1, and Tgfb1). (c) Fibrosis-associated gene score (Col1a1, Col3a1, Tgfb1, Acta2, Timp1, Mmp9, Mmp20, and Pdgfrb). (d) Correlation between canonical inflammation and fibrosis scores. (e) Correlation between expanded TNF-related and fibrosis scores. (f) Correlation between fibrosis score and NAFLD activity score (NAS). (g–i) Correlations between hepatic lipid accumulation (Oil Red O–derived lipid area) and inflammation score (g), fibrosis score (h), and serum ALT levels (i). (j–l) Correlations between canonical inflammation scores and physiological parameters, including body weight (j), liver weight (k), and serum ALT levels (l). Each point represents an individual animal. Gene expression scores were calculated from variance-stabilized RNA-seq counts. Correlation lines indicate linear regression fits. Correlation coefficients (r) and p-values are shown where applicable. To examine relationships between transcriptional programs, composite scores for inflammation and fibrosis were calculated. Canonical inflammation scores were positively correlated with fibrosis scores (Fig. 6d), and similar relationships were observed for the expanded TNF-related gene set (Fig. 6e), indicating coordinated regulation of inflammatory and fibrotic pathways. Notably, fibrosis scores were positively correlated with NAFLD activity score (Fig. 6f), linking transcriptomic changes to histological disease severity. Hepatic lipid accumulation, quantified by Oil Red O staining, showed positive associations with both inflammation scores (Fig. 6g) and fibrosis scores (Fig. 6h). Lipid accumulation was also positively correlated with serum ALT levels (Fig. 6i), suggesting a relationship between steatosis and liver injury. Furthermore, canonical inflammation scores were significantly associated with physiological parameters, including body weight (r ≈ 0.65), liver weight (r ≈ 0.63), and serum ALT levels (r ≈ 0.56) (Fig. 6j–l). Similar trends were observed for fibrosis scores. Collectively, these results demonstrate that hepatic lipid accumulation, inflammatory signaling, and fibrotic responses are tightly interconnected, and that pb2441 partially attenuates these pathological features in a Western diet–induced model. RNA-seq QC and evaluation of necropsy/cage confounding Library sizes and DESeq2 size factors were broadly comparable across groups, and neither PCA nor sample-to-sample similarity patterns showed an apparent association with necropsy sequence. Distance-based testing (PERMANOVA) indicated no significant contribution of necropsy order to the expression-distance structure; because cages were fully nested within groups, group and cage effects could not be fully disentangled in multivariable models (full QC analyses in Supplementary Information online). Discussion In this study, we evaluated the effects of B. subtilis pb2441 in a WD-induced model of hepatic steatosis and compared its effects with the SGLT2 inhibitor dapagliflozin. By integrating histological, physiological, and transcriptomic analyses, we demonstrate that pb2441 partially attenuates WD-induced hepatic steatosis and modulates associated inflammatory and fibrotic transcriptional programs. The use of a WD model represents an important aspect of this study. Unlike conventional high-fat diet models, WD incorporates excess fat, cholesterol, and simple sugars, which more closely recapitulate the complex metabolic and inflammatory features of human NAFLD progression7. In this context, WD feeding induced not only hepatic lipid accumulation but also coordinated upregulation of inflammation- and fibrosis-associated gene expression programs, supporting its relevance as a model of diet-induced liver pathology8,9. Histological analyses provided direct evidence of hepatic steatosis and its attenuation by pb2441. H&E staining revealed characteristic steatotic morphology in WD-fed mice, while Oil Red O staining and NAS scoring provided quantitative confirmation of increased lipid accumulation and disease severity. These histological features were partially reduced in pb2441-treated animals, indicating a protective effect at the tissue level. Transcriptomic analyses further revealed coordinated regulation of inflammatory and fibrotic pathways. Canonical inflammation markers and expanded TNF-related gene sets were upregulated in the WD group and showed a trend toward attenuation with pb2441 treatment. Similarly, fibrosis-associated genes were elevated in WD-fed animals and partially reduced following pb2441 administration. Although not all individual genes reached statistical significance, the consistent directional changes across gene sets suggest coordinated modulation of disease-relevant pathways. Although fibrosis-associated genes showed coordinated directional changes in transcriptomic analyses, Sirius Red staining revealed only limited collagen deposition across groups. These findings suggest that the present WD model primarily represents hepatic steatosis with early fibrotic remodeling rather than advanced fibrotic NASH. Therefore, fibrosis-related transcriptomic alterations should be interpreted cautiously as molecular signatures associated with early fibrotic responses rather than definitive evidence of overt fibrosis resolution. These findings suggest that pb2441 and dapagliflozin may primarily attenuate early fibrotic remodeling and progression rather than reverse established fibrosis. An important finding of this study is the relationship between hepatic lipid metabolism and downstream inflammatory and fibrotic responses. pb2441 treatment was associated with reduced expression of lipogenesis-related genes, while β-oxidation–related genes were increased relative to the WD group, although the magnitude of induction was less pronounced than that observed with dapagliflozin. Notably, pb2441 consistently increased β-oxidation–related gene expression relative to the WD group. These patterns suggest that pb2441 primarily limits hepatic lipid synthesis, with a supportive contribution from enhanced β-oxidation. Consistent with this interpretation, histological analyses demonstrated reduced hepatic lipid accumulation in pb2441-treated animals. Given the established role of lipid accumulation in driving hepatic inflammation and fibrosis, the attenuation of inflammatory and fibrotic gene expression is likely, at least in part, a secondary consequence of reduced lipid burden. Integration of transcriptomic and phenotypic data further supported this interpretation. Inflammation scores were positively correlated with fibrosis scores, indicating tight coupling between inflammatory and fibrotic transcriptional programs. In addition, inflammation and fibrosis scores were positively associated with NAS, linking gene expression changes to histological disease severity. Hepatic lipid accumulation quantified by Oil Red O staining was also positively correlated with inflammation scores, fibrosis scores, and serum ALT levels, suggesting that lipid accumulation is closely linked to transcriptional activation of inflammatory and fibrotic pathways as well as biochemical indicators of liver injury. The comparison with dapagliflozin provides additional insight into potential mechanistic differences. Both pb2441 and dapagliflozin reduced hepatic lipid accumulation and showed similar directional effects on inflammatory and fibrotic gene expression. Despite fewer DEGs, pb2441 induced coherent directional changes in key metabolic pathways. However, their transcriptomic signatures suggest distinct modes of action. Dapagliflozin promotes systemic energy loss through urinary glucose excretion, which can enhance metabolic flux and lipid utilization29,30. In contrast, pb2441 appears to act primarily by limiting hepatic lipid synthesis, with a secondary contribution from enhanced β-oxidation relative to the WD group. This is consistent with the observed upregulation of β-oxidation–related genes relative to the WD group. These findings suggest that the two interventions converge on similar phenotypic outcomes through partially distinct upstream mechanisms, highlighting the potential of pb2441 as a complementary approach to pharmacological metabolic therapies. In addition to metabolic effects, pb2441 may also influence inflammatory responses through host–microbe interactions. Components of B. subtilis, including flagellin, have been reported to interact with pattern recognition receptors such as Toll-like receptor 5 (TLR5), which can modulate innate immune signaling31. However, given that inflammatory gene expression changes were modest and closely associated with reductions in lipid accumulation, these effects are likely secondary to metabolic improvements, with potential contributions from direct host–microbe signaling. One important consideration is the reduction in food intake observed in pb2441-treated animals, which may contribute to some of the observed metabolic and transcriptional changes. However, the consistent relationships between transcriptomic scores, histological features, and physiological parameters suggest that pb2441-associated effects are not solely attributable to reduced caloric intake, but also involve coordinated modulation of disease-relevant pathways. This study has several limitations. First, inflammatory gene expression changes were modest and did not reach statistical significance for all markers, highlighting the need for further validation using targeted approaches. Second, mechanistic studies, including loss-of-function experiments or microbiome analyses, were not performed and will be required to establish causality. Third, colonization or persistence of pb2441 in the host was not directly assessed, which may influence interpretation of its biological activity. In addition, Sirius Red staining demonstrated only limited collagen deposition, indicating that the present WD model primarily reflected hepatic steatosis with early fibrotic remodeling rather than advanced fibrotic NASH. In conclusion, pb2441 partially attenuates WD–induced hepatic steatosis and is associated with coordinated changes in inflammatory and fibrotic transcriptional programs, with effects comparable to dapagliflozin. The integration of transcriptomic, histological, and physiological data highlights a close relationship between lipid accumulation, inflammation, and fibrosis in diet-induced liver pathology. These findings support the potential of pb2441 as a complementary strategy for metabolic liver disease and provide a framework for future mechanistic and translational studies. Material and methods IACUAC statement All procedures for mouse were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of the Animal and Plant Quarantine Agency of Korea (KBIO-IACUC-2025-033). Every effort was made to reduce animal use and minimize discomfort in accordance with the approved protocols. All experimental procedures adhered to applicable ethical standards and regulatory guidelines. This research was conducted in full compliance with the ARRIVE (Animal Research: Reporting of In Vivo Experiments) guidelines. Animals and experimental design Five-week-old male C57BL/6 mice were purchased from Central Lab Animal Inc. (Seoul, Republic of Korea) and housed under specific-pathogen-free conditions (23 °C, 55 ± 10% humidity, 12-h light/dark cycle). Mice were housed five per cage with ad libitum access to water and their assigned diets. After 1 week of acclimation, mice were randomly assigned to either a normal control diet (NCD; 2 cages, n = 10) or a Western diet (WD; 9 cages, n = 45). The WD was based on a high-fat formulation providing 60% kcal from fat (Research Diets, New Brunswick, NJ, USA; or Raonbio, Republic of Korea) and was supplemented with 10% (w/w) fructose and 10% (w/w) sucrose (Sigma-Aldrich, St. Louis, MO, USA). The NCD group was maintained on NCD throughout the study. After 8 weeks of WD feeding, WD-fed mice were allocated into three groups (n = 15/group; 3 cages/group) and maintained for an additional 40 days on WD only (WD control), WD supplemented with B. subtilis pb2441, or WD supplemented with dapagliflozin. pb2441 spores were incorporated into the WD at 3.0 × 108 CFU/g prior to pelleting; after gamma irradiation, viable spore counts were re-assessed and 2.0 × 108 CFU/g remained. Based on mean daily food intake (1.90 g/day) and mean body weight (42.8 g) in the pb2441 group, the estimated pb2441 intake was 3.8 × 108 CFU/mouse/day (8.88 × 109 CFU/kg/day). Dapagliflozin was incorporated into the WD at 22 mg/kg (22 μg/g), corresponding to an estimated intake of 42.9 μg/mouse/day (0.0429 mg/mouse/day; 1.03 mg/kg/day) based on mean daily food intake (1.95 g/day) and mean body weight (41.7 g). Body weight and food intake were measured twice weekly during the first 8 weeks and once weekly during the subsequent 40-day intervention period. Feed consumption was measured per cage by subtracting the remaining feed from the initial amount at each time point, and average intake per mouse was calculated within each cage; cages were treated as the experimental unit for statistical analyses. Two animals with extremely low body weight were excluded from statistical analyses. After 8 weeks of feeding, mice were fasted for 16 h prior to blood glucose measurement. Tail-vein glucose was measured using a handheld glucometer after 16 h and 12 h of fasting at weeks 8 and 14, respectively. At the end of week 14, mice were euthanized by inhalation of a volatile anesthetic, isoflurane, delivered via an induction chamber until respiration ceased, and blood was collected from the inferior vena cava. Dissected organs were weighed and processed promptly: tissues were fixed in formalin for histology or preserved in RNAlater for RNA extraction, and then stored at − 80 °C. Histology and whole-slide imaging Liver tissues were fixed in 10% neutral buffered formalin (10% NBF), processed through graded ethanol for dehydration, cleared in xylene, and embedded in paraffin. Paraffin-embedded tissue blocks were sectioned at 5 μm and stained with hematoxylin and eosin (H&E) and Sirius red using standard procedures. The stained slides were digitized using a whole-slide scanner to generate whole-slide images (typically > 1 GB per slide). Digital slides were converted and analyzed using Slide Converter and CaseViewer software (3DHISTECH, Budapest, Hungary). NAFLD activity score evaluation NAFLD activity score was evaluated from hematoxylin and eosin (H&E)-stained liver sections according to established criteria. Steatosis (0–3), lobular inflammation (0–3), and hepatocellular ballooning (0–2) were scored, and the sum of these components was used as the NAS (range 0–8). Histological evaluation was performed on representative sections from each animal. When multiple images per animal were available, scores were averaged to obtain a single value per animal. Oil red O lipid quantification Quantification of hepatic lipid accumulation from Oil Red O–stained sections was performed by measuring the lipid-positive area using image analysis software. The results are expressed as the percentage of Oil Red O–positive area relative to the total tissue area (% area). Analysis of hepatic lipid content Approximately 100 mg of liver tissue was homogenized in methanol (0.5 ml) using a plastic motar pistle. Subsequently, 1 ml of chloroform was added and vigorously mixed for lipid extraction. The samples were centrifuged at 10,000 g for 5 min at 4 °C, and a portion of the supernatant was dried with nitrogen gas. The lipid extract was dissolved in 2-propanol, and the triglyceride concentrations were measured using the corresponding product kits (Roche Diagnostics, Germany) and an automated analyzer (Hitachi, Japan). Hepatic TG levels were normalized to tissue weight and expressed as mg TG/g liver tissue. RNA extraction from tissue Total RNA was extracted from mouse liver tissue using the RNeasy® Mini Kit (QIAGEN, Hilden, Germany) according to the manufacturer’s instructions. RNA extraction was performed by JS Link (JS Link Inc., Seoul, Republic of Korea). Liver tissues were preserved in RNAlater and stored at 4 °C for 7 days prior to RNA extraction. Samples were placed in microcentrifuge tubes pre-cooled with liquid nitrogen and homogenized by bead beating at 50 Hz for 10 s, repeated three times. Following homogenization, 350 µL of Buffer RLT was added to each sample and mixed thoroughly by vortexing. The lysate was transferred to a 2 mL microcentrifuge tube and centrifuged at maximum speed for 3 min to pellet cellular debris. The supernatant was carefully transferred to a new microcentrifuge tube without disturbing the pellet. An equal volume of 70% ethanol was added to the cleared lysate and mixed immediately by pipetting. The sample, including any precipitate, was then transferred to an RNeasy Mini spin column placed in a 2 mL collection tube and centrifuged for 15 s at 8,000 × g. On-column DNase treatment was performed to remove genomic DNA contamination. After discarding the flow-through, the spin column was washed with 700 µL of Buffer RW1 and centrifuged for 15 s at 8,000 × g. This was followed by two washes with 500 µL of Buffer RPE, centrifuged for 15 s and 2 min at 8,000 × g, respectively. The spin column was then placed in a new 1.5 mL collection tube, and RNA was eluted by adding 50 µL of RNase-free water directly to the membrane and centrifuging for 1 min at 8,000 × g. RNA concentration and purity were assessed using 1 µL of total RNA on a NanoDrop 8000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). RNA integrity was evaluated using a Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, USA), and RNA quality was reported as the RNA Integrity Number (RIN). Only RNA samples with RIN ≥ 7.0 were used for library preparation. mRNA sequencing mRNA sequencing libraries were prepared according to the manufacturer’s instructions for the TruSeq Stranded mRNA Library Prep kit (Illumina, San Diego, California, USA). The entire process from library preparation to sequencing was performed by JS Link (JS Link Inc., Seoul, Republic of Korea). mRNA was purified and fragmented from total RNA using poly-T oligo-attached magnetic beads with two rounds of purification. Cleaved RNA fragments primed with random hexamers were reverse transcribed into first-strand cDNA using reverse transcriptase and random primers, with dUTP incorporated in place of dTTP to preserve strand specificity. These cDNA fragments then underwent end repair with the addition of a single ‘A’ base and subsequent adapter ligation. Unique dual index (UDI) adapters were used for multiplexing. The products were purified and enriched by PCR to create the final strand-specific cDNA libraries. Library quality was verified using a TapeStation (Agilent Technologies, Santa Clara, CA, USA). After quantitative PCR (qPCR) using KAPA SYBR FAST qPCR Master Mix (Kapa Biosystems, Wilmington, MA, USA), index-tagged libraries were pooled in equimolar amounts. Sequencing was performed on an Illumina NovaSeq X Plus system with 10B flow cells following provided protocols for paired-end 2 × 100 bp reads, targeting approximately 50 million paired-end reads per sample. Raw sequencing data have been deposited in the NCBI Sequence Read Archive (SRA) (accession number: PRJNA1396692, reviewer link: https://dataview.ncbi.nlm.nih.gov/object/PRJNA1396692?reviewer=c9k2rn9avolstk9d3541jql1eb). Read QC, pseudo-alignment, and quantification FASTQ files were generated using BCL Convert (v4.4.4; mismatch = 0) without adapter/quality trimming. Sequencing quality was summarized using MultiQC (v0.11.9), and all libraries met the criterion of ≥ 80% bases with Phred quality score Q30. Reads were pseudo-aligned and quantified using kallisto (v0.46.1)32 against the mouse reference transcriptome (GRCm39; Ensembl annotation), producing transcript-level abundance estimates. Transcript-level estimates were summarized to gene-level count matrices using tximport33. Differential expression analysis Raw gene-level count matrices were analyzed in R (v4.5.1) using DESeq234 within RStudio (v2025.09.1 + 401). Lowly expressed genes were prefiltered prior to differential testing (e.g., CPM ≥ 2 in at least one sample, or an equivalent count-based DESeq2 prefilter). DESeq2 fits a negative binomial generalized linear model with median-of-ratios library-size normalization (size factors) and gene-wise dispersion estimation with empirical shrinkage. A model was fitted with diet/treatment group as the main factor, and pairwise contrasts were evaluated using Wald tests (WD vs NCD, pb2441 vs WD, dapagliflozin vs WD, and pb2441 vs dapagliflozin). Unless otherwise stated, DEGs were defined as genes with Benjamini–Hochberg adjusted p value (padj) < 0.05, without applying an absolute log2 fold-change cutoff. Cook’s distance was computed to assess potential outlier counts; however, no samples were excluded based on outlier diagnostics. For each contrast, DEGs and summary statistics (baseMean, log2FoldChange, standard error, p value, and padj) were exported for reporting and downstream analyses. Variance-stabilizing transformation and unsupervised QC/visualization For unsupervised analyses and visualization, raw counts were transformed using DESeq2’s variance-stabilizing transformation (VST) (blind = TRUE unless otherwise specified)34. VST values were used for principal component analysis (PCA), hierarchical clustering, and sample-to-sample similarity assessment. Sample similarity was quantified using Pearson correlation of VST profiles and visualized as a correlation heatmap. For PCA, VST values were optionally standardized per gene (z-scored across samples) and projected onto the first two principal components. For heatmaps, VST values were z-scored per gene and clustered by genes (rows) and/or samples (columns) as specified in the figure legends. In addition to transcriptome-wide analyses, selected gene sets (adipogenesis/lipogenesis, β-oxidation, and cytochrome P450 genes) were analyzed separately, and pathway annotations were referenced to KEGG35. These gene sets were curated by the authors based on KEGG pathway membership and observed expression patterns across experimental groups. Functional enrichment and pathway-level analyses Over-representation analysis was performed by submitting DEG lists to DAVID using Mus musculus as the background36. Gene Ontology, pathway, and disease-related categories were queried. To obtain a more inclusive set of terms, the EASE score threshold (a modified Fisher’s exact test) was set to 1.0, and functional annotation charts were used to summarize enriched terms and associated genes. In parallel, gene set enrichment analysis (GSEA) was performed in R for each contrast using all tested genes ranked by DESeq2-derived statistics. Enrichment was evaluated against curated pathway gene sets, including KEGG pathways35 and immune/inflammatory pathways (e.g., cytokine–cytokine receptor interaction, chemokine signaling, NF-κB signaling, and TNF signaling). The gene universe was defined as all genes tested in DESeq2. Pathways were considered significantly enriched based on an FDR-adjusted q value threshold (as specified in the Results/Figure legends), and leading-edge genes were examined to interpret treatment-associated transcriptional changes relative to WD and NCD controls. Visualization of differential expression Volcano plots were generated for each contrast using DESeq2 results, displaying log2 fold change on the x-axis and − log10(p value) or − log10(padj) on the y-axis (as specified). Genes meeting the DEG threshold (padj < 0.05) were highlighted to visualize the magnitude and significance of transcriptional differences. Centroid-based transcriptomic similarity scoring and heterogeneity To quantify sample-level similarity to NCD versus WD transcriptomic states, centroid-based similarity scores were computed from VST-transformed data. In one approach, the feature set was restricted to genes differentially expressed between WD and NCD (padj < 0.05; no |log2FC| cutoff). VST values were standardized per gene (z-scored across samples), and group centroids were defined as the mean standardized expression vector for the NCD and WD groups. For each sample, similarity to NCD and WD centroids was quantified by Pearson correlation (cor_NCD and cor_WD), and Δcor was defined as cor_NCD − cor_WD. Δcor values were linearly rescaled to a 0–1 range to generate an NCD-like score, with a complementary WD-like score defined as 1 − NCD-like score. Within-group heterogeneity was summarized using dispersion metrics (SD and IQR), and differences in dispersion among groups were evaluated using the Brown–Forsythe test (median-centered Levene’s test). Integrated analysis of gene expression and physiological phenotypes To integrate transcriptomic and phenotypic data, gene expression–based scores were calculated for predefined gene sets related to inflammation and fibrosis. Variance-stabilized transformed (VST) RNA-seq counts obtained from DESeq2 were used for all analyses. For each gene set, expression values were standardized (z-score) across samples and averaged to generate a composite score per animal. These gene expression scores were then correlated with histological parameters (NAFLD activity score and Oil Red O–derived lipid area) and physiological measurements (body weight, liver weight, and serum biochemical markers including ALT and AST). Correlation analyses were performed using Pearson’s correlation coefficient, and linear regression was used for visualization. Assessment of cage and necropsy-sequence effects Potential confounding effects of necropsy sequence and cage were assessed using PCA-based and distance-based approaches. Linear models were fitted to PCA scores (PC1 and PC2) with group, cage_id, and necropsy sequence as covariates. In addition, PERMANOVA (adonis2; 999 permutations) was performed on Euclidean distances computed from VST profiles to test associations with group, cage_id, and necropsy sequence. Because each cage contained animals from only one experimental group (i.e., cage nested within group), cage and group effects could not be fully disentangled; therefore, tests involving cage were interpreted cautiously under this housing structure. Statistical tests Longitudinal body-weight trajectories were analyzed using a two-way repeated-measures ANOVA (group and time as factors) followed by Sidak’s multiple-comparisons tests versus the WD control at each time point. Pre- versus during-intervention energy intake within each group was evaluated using paired two-sided Student’s t-tests at the cage level (cages as matched units). End-point comparisons among groups (e.g., glucose, liver weight, and plasma enzymes) were performed using one-way ANOVA followed by Tukey’s multiple-comparisons tests. When assumptions of normality or homoscedasticity were not met, appropriate nonparametric alternatives were applied. Data are presented as mean ± SD unless otherwise stated; p < 0.05 was considered statistically significant. Supplementary Information Supplementary Information. Acknowledgements We thank the K-Bio Osong Foundation, Yeongmi Koo, and Jaewon Jang for technical assistance with H&E staining and mouse experiments, and Woo Sik Oh for RNA-seq sequencing. Author contributions Conceptualization: YC, MK; Experimentation: MK; Formal analysis and investigation: YC, MK; Writing-original draft: YC; Writing-review and editing: KSM, YC; Supervision: MK, YC. Funding This work was supported by the Health Functional Food Development Support Program (RS2023-00282826) and Technology Development (R&D) Support Program (RS-2024-00509945, S3374815) of the Ministry of SMEs and Startups (MSS, Korea), and the Korea Institute of Toxicology (Grant No. 2710008773). Data availability The RNAseq sequencing dataset is available in the NCBI SAR repository under accession number PRJNA1396692 (Reviewer link: https://dataview.ncbi.nlm.nih.gov/object/PRJNA1396692?reviewer=c9k2rn9avolstk9d3541jql1eb). Other data will be available from YC upon request. Declarations Competing interests YC is a major stock holder of Proxenrem. Bacillus subtilis pb2441 is under a Korean patent application, but we have no restrictions on sharing the data in this manuscript. All authors declare no additional competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Yangrae Cho, Email: yangchorae@gmail.com. Kyoung-Sik Moon, Email: ksmoon@kitox.re.kr. References Associated Data Supplementary Materials Supplementary Information. Data Availability Statement The RNAseq sequencing dataset is available in the NCBI SAR repository under accession number PRJNA1396692 (Reviewer link: https://dataview.ncbi.nlm.nih.gov/object/PRJNA1396692?reviewer=c9k2rn9avolstk9d3541jql1eb). Other data will be available from YC upon request.