Gut microbiome and metabolite profiles are associated with subsequent histologic MASH in a longitudinal mouse model

ABSTRACT

Detecting metabolic dysfunction-associated steatohepatitis (MASH) before fibrosis develops remains a major challenge. We used a longitudinal streptozotocin/high-fat diet (STAM/HFD) mouse model to identify microbial and metabolic features that emerge before histological MASH onset. Integrative 16S rRNA and untargeted LC-MS/MS analyses revealed microbial and metabolite shifts weeks before disease development. Lachnospiraceae and Oscillospiraceae were enriched in non-MASH mice and were negatively correlated with fatty acids elevated under MASH conditions. Bile acid and fatty acid metabolites enriched under MASH and non-MASH states varied with collection time, showing that diurnal variation strongly shapes biomarker profiles. Analysis of a human MAFLD cohort confirmed higher abundances of Lachnospiraceae bacteria in healthy participants. Our results identify early microbial and metabolic features that are associated with subsequent histologic MASH onset and reveal how temporal factors influence biomarker detection. These time-sensitive features, together with cross-species concordance in a human MAFLD cohort, lay the groundwork for developing and validating microbiome-based diagnostics and circadian-informed interventions that target the gut-liver axis to prevent metabolic liver disease.

IMPORTANCE

Metabolic dysfunction-associated steatohepatitis (MASH) often develops silently, limiting opportunities for early intervention before fibrosis occurs. In this study, we identify gut microbial taxa and stool metabolites that are associated with future histologic MASH development in a longitudinal mouse model. We show that members of the Lachnospiraceae and Oscillospiraceae families are enriched in mice that remain disease-free and that members of the Lachnospiraceae family exhibit similar patterns in a human MAFLD cohort. Importantly, we demonstrate that time of sample collection has a strong effect on metabolomic profiles, with disease-associated lipid and bile acid signals detectable only at specific sampling times. These findings highlight circadian timing as a critical variable in microbiome-based biomarker discovery and support the development of time-aware diagnostic strategies for early MASH risk assessment.

INTRODUCTION

Metabolic dysfunction-associated steatotic liver disease (MAFLD) affects up to 39% of individuals worldwide and is rapidly increasing in prevalence (1, 2). Current estimates indicate that by 2040, MAFLD will impact over 55% of the global population (3). MAFLD is characterized by excessive fat accumulation in hepatocytes (i.e., steatosis). Over time, the accumulated injury of steatosis can lead to inflammation and fibrosis of the liver, i.e., metabolic dysfunction-associated steatohepatitis (MASH). If left unchecked, MASH can lead to cirrhosis and hepatocellular carcinoma (HCC) (4, 5), and it is the leading indication for liver transplants (6, 7). However, despite its prevalence, there are presently few specific therapies for MASH, particularly for fibrosis, or methods to distinguish those with MAFLD who are at high risk for MASH (8). Current treatments seek to reduce disease severity by focusing on addressing comorbidities, such as obesity and type 2 diabetes (T2D) (9, 10). Recently developed pharmacologic agents, including the thyroid hormone receptor-β agonist resmetirom and the dual GIP/GLP-1 receptor agonist tirzepatide, show promise for patients with fibrotic MAFLD by improving liver fibrosis in a subset of patients (11). These advances underscore the importance of accurately identifying individuals with established fibrosis or at high risk of progression so that they can be appropriately prioritized for these emerging, costly therapies.

Recent efforts have focused on identifying non-invasive biomarkers that capture early pathological changes (12, 13). Among these, alterations in the gut microbiome and metabolite profiles (across blood, urine, stool, and tissue) have emerged as promising candidates, reflecting the central role of the gut-liver axis in MASH pathogenesis. Large-scale, multi-omic analyses have identified conserved microbial, metabolomic, and, more recently, viral signatures associated with MAFLD/MASH across diverse populations (12, 14–18). For example, an integrated analysis of seven global cohorts (n = 1,892) revealed nine cross-ethnicity microbial markers that distinguished MAFLD with an AUC of 0.95 (17), while another study reported enrichment of 66 gut bacterial species, including oral-typical taxa, and transkingdom dysbiosis involving bacterial and viral communities within a health worker population (18). Yet despite these advances, microbiome and metabolite profiles still vary widely across individuals. Factors such as comorbid conditions (e.g., obesity and diabetes), diet, circadian rhythms, feeding status, geography, and host genetics, can all modulate the microbial and metabolite landscape, leading to inconsistent results across human studies (12, 17, 19). Moreover, nearly all these studies have examined cross-sectional cohorts in which disease was already established. Thus, researchers still need to identify early predictors of MASH in longitudinal cohorts.

Mouse models provide a controlled environment to investigate early predictors before disease onset. In our recent work, we employed the streptozotocin/high-fat diet (STAM/HFD) mouse model of MASH to test whether time-restricted feeding (TRF), a regimen that consolidates food access to the active period of these mice, could mitigate disease progression in a non-obese background. Although TRF did not prevent MASH development in this context (20), it did reduce hepatic triglyceride accumulation, consistent with our previous findings (2, 21, 22). Leveraging the cohort’s incomplete penetrance, with more than half of HFD-fed mice developing histologic MASH and roughly half developing fibrosis despite similar exposures, we analyze longitudinal samples collected before and after disease onset to identify early microbial and metabolite signatures that are associated with subsequent histologic progression. Specifically, we ask whether fecal microbial and metabolic profiles differ under TRF compared with ad libitum feeding, and whether early MASH-associated signals can be detected weeks before overt liver pathology is present. Furthermore, because TRF is a circadian intervention, we collected samples at both day and night timepoints to capture diurnal variation in microbial and metabolic activity (23). Through this longitudinal and temporal design, we aim to define early disease trajectories and uncover candidate signals and therapeutic targets for MASH.

RESULTS

Diet and TRF have the largest influence on gut microbiome composition in STAM/HFD mice

To characterize early, pre-disease signifiers of MASH, we leveraged our previously described streptozotocin–high-fat diet (STAM/HFD) mouse model data set (20). In brief, four-week-old mice were assigned to one of three conditions: STAM mice with time-restricted feeding (TRF) access to HFD (FT; Zeitgeber time [ZT] 13–21), STAM mice with ad libitum access to HFD (FA), and STAM mice with ad libitum access to a normal chow diet (NA). Stool samples were collected at 4 and 7 weeks post-intervention at two ZT timepoints (ZT1 and ZT13) for 16S rRNA amplicon sequencing and untargeted LC-MS/MS metabolomics (Fig. S1A). As shown in our earlier study, TRF did not protect against MASH development in this STAM/HFD model (Fig. S1B), although it modestly reduced hepatic triglyceride accumulation (Fig. S1C) (20). Despite these mice having similar genetic background, diet, and feeding regimen, approximately 40% of mice in the HFD conditions did not develop MASH. By leveraging our longitudinal sampling, this experiment investigated whether differences in the gut microbiome and metabolome at an earlier, pre-disease time point were associated with which mice later develop MASH.

In our microbiome analysis, we observed that diet had the most significant impact on microbial composition, as evidenced by its strong influence on the first principal component (PC1) in our robust principal component analysis (RPCA) plot showing microbiome dissimilarities across mice (PERMANOVA, NA vs FA P = 0.001; NA vs FT P = 0.001) (Fig. 1A) (24). Following diet, TRF emerged as the second most influential factor, driving dissimilarity along PC2 (PERMANOVA, FA vs FT P = 0.001). This contrasts with our histologic liver findings, where we observed no TRF-driven differences. To minimize the possibility that diet masked other subtler drivers of changes in microbial composition, we refined our analysis by separating samples by disease stage (week 4, prior to advanced disease, or week 7, when advanced disease was apparent) and excluding the NA condition (Fig. 1B). These results showed that TRF drove microbial compositional separation at both disease stages (PERMANOVA, FA vs FT week 4 P = 0.001; week 7 P = 0.002), and that there may be diurnal collection time differences at week 7 but not week 4 (PERMANOVA, ZT1 vs ZT13, week 4 P = 0.152, week 7 P = 0.009). As expected, diet is the main driver of microbiome compositional change. After diet, feeding condition was the next strongest driver, with TRF exerting a stronger effect on gut microbiome composition than collection time.

Fig 1.

Diet and TRF have the largest influence on gut microbiome composition in STAM/HFD mice. (A) RPCA plot of stool 16S results (PERMANOVA α < 0.05). (B) RPCA plot removing the NA condition and separating week 4 (left) and week 7 (right) samples.

Enrichment of Lachnospiraceae is associated with non-MASH status before histologic onset

Given the clear TRF-driven differences and minimal impact of collection time on microbial diversity, we next performed differential abundance analyses comparing non-MASH and MASH samples at weeks 4 and 7 for both FA and FT conditions, aggregating across ZT timepoints. This decision was guided by the lack of a detectable ZT effect at week 4 (Fig. 1B), allowing us to maximize power for identifying early taxa associated with later deleterious histology. This analysis revealed only a small number of taxa with significant differential abundance between non-MASH and MASH samples, many belonging to the Lachnospiraceae family (BIRDMAn, credible if the 94% highest density interval [HDI] excludes 0; Fig. S1A and B; Table S1) (25). Specifically, seven differentially abundant microbes were shared between weeks 4 and 7 in the FA condition and four in the FT condition (Fig. 2A). Among the shared taxa in FA mice, five exhibited consistent abundance trends across disease stages, including members of the Lachnospiraceae and Oscillospiraceae families (Fig. 2B, top, red and blue ribbons). All these taxa, except f__Lachnospiraceae (65bc2), were enriched under non-MASH conditions at both timepoints (credible if 94% HDI excludes 0; Fig. 2C; Fig. S2C, top), suggesting that these taxa are associated with reduced MASH risk and that their enrichment precedes histologic disease onset by weeks. Because these signals are derived from 16S rRNA profiling and several key taxa are classified only to the family level, these associations should be interpreted as family-level patterns that may encompass multiple species or strains with distinct functional effects.

Fig 2.

Lachnospiraceae is enriched in non-MASH mice before histologic onset. (A) Overlap of differentially abundant microbes between non-MASH and MASH at week 4 and week 7 under (left) FA and (right) FT conditions. (B) Microbes that are differentially abundant in non-MASH vs MASH (BIRDMAn, credible if 94% highest density interval [HDI] excludes 0) before (week 4) and after (week 7) disease onset under FA (top) and FT (bottom) conditions. Ribbons indicate microbes that were differentially abundant at both disease stages. Red and blue ribbons represent microbes consistently higher under MASH or non-MASH, respectively. Abundance plots of selected microbes that are consistently differentially abundant at both week 4 and week 7 under either (C) FA or (D) FT conditions. We used the histologic NAS scores to define MASH and non-MASH mice.

Meanwhile, among the four shared microbes in the FT condition, only two displayed consistent abundance trends across disease stages, f__Oscillospiraceae_88309 (a6c1d) and g__Acetatifactor; s__Acetatifactor sp011959105 (be65d) (credible if 94% HDI excludes 0; Fig. 2B, bottom, blue ribbons; Fig. 2D). Both taxa were consistently enriched in non-MASH mice at week 4 and week 7 but were distinct from those identified in the FA condition, suggesting that TRF induces a distinct set of microbial shifts that may be associated with reduced MASH risk. The remaining taxa exhibited inconsistent patterns, either reversing direction between weeks 4 and 7 or differing between FA and FT conditions (Fig. 2D; Fig. S2C). The identification of taxa that are consistently enriched in non-MASH mice at week 4, well before overt histologic evidence of disease, and at week 7 highlights their potential as candidate early markers of subsequent MASH classification.

Collection time has a greater impact than TRF on stool metabolomic composition in STAM/HFD mice

In our metabolomics analysis, we found that, like the microbiome, diet exerted the strongest influence on the stool metabolite pool, as reflected by its dominant effect on PC1 in the RPCA plot (PERMANOVA, NA vs FA P = 0.001; NA vs FT P = 0.001) (Fig. 3A). However, unlike the microbiome, TRF did not significantly alter metabolite composition among the remaining principal components (PERMANOVA, FA vs FT P = 0.162). Unlike mice with diet-induced obesity (DIO), which spread their caloric intake throughout day and night and exhibit disrupted activity and metabolic rhythms (26), mice under the STAM/HFD conditions retain diurnal feeding, motor activity, and sleep patterns (Fig. S3A) (20). To better resolve additional sources of variation outside of diet, we stratified samples by disease stage (week 4 or week 7) and excluded the NA condition (Fig. 3B). This analysis showed that TRF failed to drive significant metabolomic separation at either stage (PERMANOVA, FA vs FT: week 4 P = 0.062, week 7 P = 0.134). In contrast, collection time strongly influenced metabolite diversity (PERMANOVA, ZT1 vs. ZT13: week 4 P = 0.001; week 7 P = 0.001). These observations are likely due to the similar temporal feeding behavior between FA and FT mice in this model, because FA mice do not consume food uniformly throughout the day.

Fig 3.

Collection time has a greater impact than TRF on stool metabolomic composition in STAM/HFD mice. (A) RPCA plot of stool metabolomics profiles (PERMANOVA α < 0.05). (B) RPCA plot removing the NA condition and separating week 4 (left) and week 7 (right) samples.

Microbial fatty acids from the light phase may indicate MASH development weeks before disease onset

Given the differences in the RPCA plot between ZT1 and ZT13, but not between condition, we conducted a differential abundance analysis to identify differences between non-MASH and MASH mice, combining conditions and separating them by ZT time. This analysis revealed 29 differentially abundant annotated metabolites between non-MASH and MASH mice at both week 4 and week 7 for ZT1 (light phase) and 15 at ZT13 (dark phase) (BIRDMAn, credible if 94% HDI excludes 0; Fig. 4A; Table S2). Approximately half of these metabolites are potentially of microbial origin, based on matches in the microbe-derived metabolite search tool microbeMASST, in addition to matching to known or predicted bile acids and fatty acids (Fig. 4B) (27–29). The annotation of these metabolites was performed using MS/MS spectral libraries, which include libraries available in the GNPS ecosystem. To reflect the intrinsic limitation of MS/MS spectral matching, annotations are are reported at the class level and represent level 3 annotations according to the Metabolomics Standards Initiative (30–32). Among the 29 metabolites (mtb) shared between week 4 and week 7 at ZT1, five had spectral matches to the following fatty acid derivatives: FA 18:2 (mtb 1264 and 3552), FA 20:3;O (mtb 3597), FA 20:4;O (mtb 1992), and FA 12:2;O3 (mtb 3748; Fig. 4B). These fatty acids consistently showed enrichment under MASH conditions at both week 4 and week 7 (credible if 94% HDI excludes 0; Fig. 4B and C, top; Fig. S3B), supporting their association with later MASH development. Furthermore, these fatty acids were enriched only at ZT1, highlighting the need to account for collection time when searching for metabolomic markers.

Fig 4.

Microbial fatty acids from the light phase may indicate MASH development weeks before disease onset. (A) Overlap of differentially abundant metabolites between non-MASH and MASH mice at week 4 and week 7 under (left) ZT1 and (right) ZT13. (B) Metabolites that are differentially abundant in non-MASH vs MASH (BIRDMAn, credible if 94% HDI excludes 0) before (week 4) and after (week 7) disease onset at ZT1 (top) and ZT13 (bottom). Abundance plots of selected metabolites that are consistently differentially abundant at both week 4 and week 7 under either (C) ZT1 or (D) ZT13. * = these metabolites were annotated based on matches to a library of suspects (29). All metabolites were annotated based on matches to spectral libraries, representing level 3 annotation according to the Metabolomics Standards Initiative (30, 31). Lipid nomenclature follows the LIPID MAPS simplified nomenclature (32).

Additionally, we annotated several bile acids among the 29 annotated metabolites that showed differential abundance at ZT1 for both week 4 and week 7. Specifically, bile acids with spectral matches to candidate penta- (mtb 2245) and tetra- (mtb 2568, mtb 4446, and mtb 2691) hydroxylated bile acids were differentially abundant under ZT1 and consistently enriched under MASH conditions (credible if 94% HDI excludes 0; Fig. 4B and C, bottom; Fig. S3B) (28). However, these bile acids had a smaller effect size than fatty acids, potentially suggesting that lipids may dominate over bile acids for early MASH detection at ZT1.

Most metabolites detected at ZT1 showed consistent enrichment under MASH at both week 4 and week 7, but these patterns reversed among the shared differentially abundant metabolites at ZT13. For example, mtb 2245, a candidate pentahydroxylated bile acid, increased at week 4 in mice that later developed MASH but shifted toward higher abundance in non-MASH mice at week (Fig. 4B and D, left). Several other bile acids followed similar trajectories (Fig. S3C). In contrast, mtb 4200, a tauro-dihydroxylated bile acid derivative, remained elevated in non-MASH mice at both time points (Fig. 4B and D, right), making it a potential early indicator of MASH in samples collected during the dark phase. The greater variability in metabolite patterns at ZT13 compared with ZT1 underscores the need to examine metabolomic changes during the light phase to capture more stable early disease signatures.

Lachnospiraceae are associated with non-MASH status in mice and non-MAFLD status in humans

To explore relationships between metabolites and microbes differentially abundant between non-MASH and MASH mice at both week 4 and week 7, we performed correlation analyses integrating these data sets. Importantly, these cross-omic correlations were computed on per-sample matched microbiome and metabolome profiles collected from the same stool specimens at each ZT timepoint (ZT1 or ZT13). Although most correlations varied between the two stages, two taxa classified only to the family level, f__Lachnospiraceae (e6136) and f__Oscillospiraceae_88309 (a6c1d), consistently correlated with five fatty acid-related metabolites before and after MASH onset (Fig. 5A; Fig. S4A). Both taxa were enriched under non-MASH conditions and were negatively correlated with derivatives of FA 18:2 (mtb 1254, mtb 3552), FA 20:4;O (mtb 1992), FA 12:2;O3 (mtb 3748), and FA 20:3;O (mtb 3597), fatty acid-related metabolites that were elevated under MASH. These consistent inverse relationships suggest that f__Lachnospiraceae (e6136) and f__Oscillospiraceae_88309 (a6c1d) are associated with lower levels of these disease-associated fatty acid-related metabolites in this model.

Fig 5.

Lachnospiraceae are associated with non-MASH status in mice and non-MAFLD status in humans. (A) Heatmap showing the correlation of the 10 microbes and 41 metabolites with significant differential abundance between non-MASH and MASH at week 7 in the STAM/HFD mice. White asterisks highlight the microbe and metabolite associations of interest. (B, C) BIRDMAn results showing (B) microbial taxa and (C) metabolites from a human MAFLD cohort that differed between MAFLD with no AF and non-MAFLD individuals (credible if the 94% highest density interval [HDI] excludes 0). Pink bars indicate taxa belonging to the Lachnospiraceae family, and orange bars indicate metabolites classified as fatty acids or their derivatives. (D) Heatmap showing correlations between Lachnospiraceae taxa identified in panel B and annotated metabolites with the strongest associations. * = these metabolites were annotated based on matches to a library of suspects (29). All metabolites were annotated based on matches to spectral libraries, representing level 3 annotation according to the metabolomics standards initiative (30, 31). Lipid nomenclature follows the LIPID MAPS simplified nomenclature (32).

To extend our findings beyond mice, we analyzed microbial data from a human twin and family cohort comprising 98 probands spanning the full spectrum of MAFLD (16). This cohort includes related individuals, enabling partial control for shared genetic background and environmental exposures, and incorporates detailed clinical metadata that allows adjustment for key covariates (e.g., age, sex, and BMI). Becasue the data set does not include a specific MASH designation, we compared non-MAFLD individuals to those with MAFLD and no advanced fibrosis (AF), representing the closest available disease stage to our mouse model. Differential abundance analysis identified enrichment of Lachnospiraceae taxa in non-MAFLD individuals (credible if 94% HDI excludes 0; Fig. 5B; Table S3), consistent with their enrichment in STAM/HFD mice. Several lipid-related metabolites, including fatty acid derivatives, were also differentially abundant by MAFLD status (Fig. 5C; Table S4). Correlation analysis between these microbes and metabolites revealed associations between Lachnospiraceae taxa and fatty acids and fatty acid-derived metabolites (Fig. 5D; Table S5), although these relationships did not directly recapitulate the specific microbe-metabolite pairings observed in mice. Our analysis comparing non-MAFLD individuals to those with MAFLD-cirrhosis similarly revealed comparable microbe and metabolite association patterns (Fig. S4B and C). Together, these findings indicate that Lachnospiraceae are associated with non-disease states compared with MASH/MAFLD across both data sets, while the specific microbe-metabolite relationships differ between systems.

DISCUSSION

We identified microbial taxa that were associated with subsequent histologic MASH classification across different feeding conditions. These microbes appeared weeks before histologic disease onset and were differentially enriched under MASH or non-MASH conditions. f__Lachnospiraceae (e6136) and f__Oscillospiraceae_88309 (a6c1d) were consistently enriched in non-MASH mice prior to disease development (Fig. 2B). Members of the Lachnospiraceae family also showed higher abundance in healthy, non-MAFLD participants within a twin and family cohort that covered the full MAFLD spectrum (Fig. 5B) (16). These results are consistent with earlier studies reporting reduced levels of Lachnospiraceae and Oscillospiraceae in MAFLD patients, including those with advanced fibrosis and cirrhosis (33–35). A decline in Oscillospiraceae also often marks the microbial signature of MAFLD onset (36). We classified f__Oscillospiraceae_88309 (a6c1d) only to the family level, but it may be related to members of the Oscillospiraceae family, such as Faecalibacterium prausnitzii. F. prausnitzii, improves metabolic health, produces anti-inflammatory metabolites, and reduces hepatic fat and fibrosis in mice (37, 38). Thus, the abundances of Lachnospiraceae and Oscillospiraceae are associated with subsequent histologic MASH classification in this model. While similar patterns are observed in human data sets, these comparisons are descriptive, and prospective diagnostic performance will require model training and validation in longitudinal human cohorts.

In our murine model, f__Lachnospiraceae (e6136) and f__Oscillospiraceae_88309 (a6c1d) showed an inverse relationship with fatty acids enriched under MASH (Fig. 5A). While Lachnospiraceae bacteria also correlated with fatty acid-related metabolites in the human cohort, these relationships did not directly recapitulate the specific microbe-metabolite pairings observed in mice. This is in contrast to other studies that have identified eicosanoids, encompassing two of our detected fatty acids, FA 20:3;O (mtb 3597) and FA 20:4;O (mtb 1992), as key inflammatory mediators, including derivatives, such as 15-oxo-eicosatetraenoic acid and 20-hydroxyarachidonic acid (39). Moreover, eicosanoids together with polyunsaturated fatty acids, including linoleic acid (FA 18:2, mtb 1264, mtb 3552), are associated with fibrosis progression in human MASH cohorts (40, 41). These findings suggest that the associations between microbial taxa and fatty acids we observed in mice may not translate directly to humans, possibly because of differences in diet and metabolic context. Nevertheless, these taxa may be associated with beneficial host phenotypes through alternative mechanisms that remain to be determined. Indeed, a recent study showed that a microbially derived bile acid, 3-succinylated cholic acid (3-sucCA), improved MASH outcomes by altering the abundance of beneficial gut microbes (42).

Several bile acid markers were consistently enriched under MASH at week 4 and week 7 during the light phase (ZT1), but these patterns disappeared during the dark phase (ZT13; Fig. 4B). These diurnal shifts underscore that collection time is a critical determinant of metabolomic state, as disease-associated bile acid signatures were apparent at ZT1 but not ZT13, highlighting the need to account for sampling time in biomarker discovery and interpretation, particularly in preclinical models (23). Patients with MAFLD have elevated circulating bile acid levels compared with healthy individuals (38, 43), consistent with our ZT1 findings where all bile acids increased under MASH conditions. However, primary and secondary bile acids can exert divergent effects in MAFLD pathogenesis (44). For example, taurocholic acid is associated with greater steatohepatitis risk (45), whereas hyodeoxycholic acid (HDCA) and taurohyodeoxycholic acid (THDCA) decline in rats with MAFLD (46). This pattern parallels our results, where a tauro-dihydroxylated bile acid derivative (mtb 4200) was enriched under non-MASH conditions at both week 4 and week 7 during the dark phase. In rats, THDCA alleviates hepatic steatosis by activating the farnesoid X receptor (FXR) (46). This raises the possibility that impaired insulin signaling in the STAM model might disrupt FXR activity, potentially limiting the ability of THDCA-like bile acids to improve steatosis. FXR regulates bile acid, lipid, glucose, and cholesterol metabolism, and recent evidence strengthens its causal link to MAFLD progression (44). Clinical trials targeting this pathway, such as those testing the FXR agonist obeticholic acid, further underscore FXR’s central role in disease modulation (47, 48).

Although TRF did not prevent MASH development in our original study (20), it induced distinct shifts in the gut microbiome before and after disease onset (Fig. 1A and B). Consistent with other TRF studies, these changes likely reflect microbial adaptation to restricted feeding windows (2, 49, 50). In this STAM/HFD model, diurnal differences appeared only after disease onset in TRF-treated mice. In diet-induced obesity models, TRF typically restores rhythmic feeding and metabolic patterns disrupted by ad libitum high-fat diet feeding, thereby reinstating light-dark differences (2, 21, 49). In contrast, our insulin-deficient STAM mice lacked diurnal variation at week 4 despite TRF, and ad libitum HFD mice displayed pronounced diurnal differences at week 7. This reversal of the expected pattern may result from altered feeding behavior in the insulin-deficient background. TRF also failed to modify the metabolome, which remained similar to that of ad libitum-fed HFD mice (Fig. 3A and B). This similarity likely reflects comparable feeding patterns between groups, with collection time capturing only feeding-driven metabolomic variation. Consistent with this, metabolic cage measurements confirmed preserved diurnal alignment of feeding behavior, locomotor activity, and sleep across groups (Fig. S3A), although fecal sampling at only ZT1 and ZT13 limits resolution of intrinsic circadian oscillations. Together, these findings suggest that TRF-associated microbial and metabolic changes may depend, in part, on intact insulin signaling, although this relationship requires further investigation.

Our study has a few noteworthy limitations. Despite using advanced analytical approaches, our ability to annotate microbial and metabolic features remained restricted. We identified most microbial markers only at the family level, and our metabolite coverage was incomplete because extraction and LC-MS/MS methods do not capture all compound classes. Compound annotations relied on spectral library matching, which often provides subclass or molecular family-level identification rather than compound-level assignments. These assignments represent level 3 annotations according to the Metabolomics Standards Initiative (30, 31). Some annotated metabolites may also represent structural isomers, and follow-up investigations are necessary to unequivocally determine their structures. The STAM/HFD model does not develop obesity or insulin resistance, whereas most MAFLD/MASH occurs in the context of obesity and insulin resistance. This distinction may influence metabolic and feeding dynamics relevant to TRF and limit generalizability to obesity-associated MASH. Nonetheless, this model was selected for its reliable progression through the full histologic spectrum of MAFLD, making it well-suited for identifying microbiome signatures associated with disease progression. 16S rRNA gene sequencing provides limited taxonomic resolution and does not capture microbial function, constraining mechanistic interpretation. Additionally, the human cohort analysis is cross-sectional and not explicitly matched across all clinical and environmental variables and should therefore be interpreted in the context of potential residual confounding. Future studies using higher-resolution approaches, including shotgun metagenomics and functional profiling, will be required to define microbial mechanisms and validate the diagnostic classifiers identified here.

Together, our findings identify early microbial and metabolic features that precede histologic MASH onset, highlight Lachnospiraceae and Oscillospiraceae taxa associated with healthier phenotypes, and provide complementary human data showing enrichment of Lachnospiraceae in non-disease states. These results integrate microbial composition, feeding behavior, and insulin signaling in relation to MASH progression, providing a foundation for developing microbiome- or diet-based strategies for early detection and intervention. Future studies should mechanistically test whether and how these taxa exert potentially protective effects that may reduce MASH development in high-risk settings.

MATERIALS AND METHODS

Animal model

The streptozotocin (STZ) +high fat diet (HFD) murine model of MASH (STAM/HFD) was used for this study (51). In brief, C57BL/6 male mice were injected with 0.2 mg of STZ (Cayman chemical item no. 13104) per day for five consecutive days starting at 10–12 days after birth. Male mice born within a week of each other were considered as a single cohort for the study. Mice were housed three to a cage, with each mouse coming from a different dam to prevent pre-wean cage effects that could bias the data. Only male mice were used for the experiment as female STAM/HFD mice do not develop MASH. At 4–5 weeks of age (post-wean), hyperglycemia (>200 mg/dL) was confirmed by colorimetric urinalysis strips (Diastix, SKU: 562806). After hyperglycemia was confirmed, mice were maintained on a high-fat diet (60% kcal fat, Research Diets, D12492i) or normal chow diet (Teklad Diet LM-48) depending on condition assignment for the duration of the study. Individual animal weights and cage-averaged food intake were measured at the same time once per week following weaning for the duration of the study using a standard digital scale. Mice were sacrificed at the time of MASH onset (7 weeks post-intervention). Mice were euthanized at ZT1 and ZT13, with half of the mice in each condition euthanized at each timepoint. Prior to euthanasia, a submandibular bleed was completed to collect blood for serum insulin and glucose quantification. Liver and ileum tissues were harvested at the time of sacrifice. A section of the right medial lobe was harvested for liver histology, and a section of the left medial lobe was harvested for transcriptomics analysis. Liver and ileum tissue samples for RNA-sequencing were flash frozen in liquid nitrogen and stored at −80℃.

Feeding patterns

Mice were maintained on a 12-h light (ZT0-12)/12-h dark (ZT12-24) light cycle. Mice either had ad libitum access to food or TRF access to food from ZT13-21. From ZT13-21, mice in the TRF group were physically moved from a cage without food to a cage with access to food. Mice with ad libitum access to food were also moved between cages at the same times as the TRF group to control for handling. For this study, we used 12 mice in each of the three conditions (36 mice total): (i) STAM + HFD + ad libitum access to food (FA), (ii) STAM + normal chow diet (NCD) + ad libitum access to food (NA), and (iii) STAM + HFD + TRF access to food from ZT13-21 (FT).

Fecal collection

Prior to starting a high-fat diet, we collected a single time point stool sample from all mice to rule out founder effects in our microbiome results. Stool was collected from mice at two time points: before MASH pathology (4 weeks post-intervention) and at 7 weeks post intervention, at ZT1 (light phase) and ZT13 (dark phase). For the experiment, we used a subset of six mice in each of the three conditions. Every mouse came from a different home cage because each cage was considered a single independent unit (n = 1) for microbiome characterization. Mice were placed in clean cups at ZT1 and ZT13, and at least three pellets of fresh stool were collected at each time point before returning mice to their home cage. Stool was stored at −80°C and later processed for 16S rRNA gene amplicon (16S) sequencing and untargeted metabolomics.

16S sample processing

DNA extraction was performed on stool samples using the Earth Microbiome Project 16S Illumina Amplicon Protocol (52). For ileum samples, a total nucleic acid extraction was performed using MagMAX Microbiome Ultra (catalog #A42357) on powderized tissue (mortar and pestle on dry ice), as previously described (53). Raw sequencing data were uploaded to Qiita under the study ID 13785 (54). Raw sequences were demultiplexed, trimmed to 150 bp, and processed using Deblur 2021.09 with default parameters on Qiita with the exception of our taxonomic annotation which we assigned using Greengenes2 (55). 16S feature tables pre-processed in Qiita as described were analyzed using QIIME 2 version 2022.2 (56). After rarefying the samples to the same sequencing depth, diversity and differential abundance analysis were performed. Ileum samples were rarefied to 4,800, resulting inthe loss of one sample per condition. Stool samples were rarefied to 10,000 with 100% sample retention.

LC-MS/MS sample processing

Flash-frozen (liquid nitrogen) tissue samples were pulverized by mortar and pestle on dry ice and stored at −80°C until extraction. At the time of extraction, samples were thawed on ice, weighed, and a clean/sterile stainless-steel bead was added to each sample. Extraction was performed with 50% MeOH/water (10–20 µL of solvent was added for each 1 mg of sample), samples were homogenized at 25 Hz for 5 min, then centrifuged at max speed for 15 min. Supernatants were concentrated using a centrifugal low-pressure system then stored at −80°C once dry. Samples were reconstituted using reconstitution solvent (80% MeOH/20% water + 1 µM sulfadimethoxine), centrifuged, and supernatant was removed for analysis.

Next, 5 µL was injected into a Vanquish UHPLC system coupled to a Q-Exactive orbitrap mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). For the chromatographic separation, a C18 porous core column (Kinetex C18, 150 × 2 mm, 1.8-µm particle size, 100-Å pore size, Phenomenex, Torrance, CA, USA) was used. For gradient elution, a high-pressure binary gradient system was used. The mobile phase consisted of solvent A (H₂O + 0.1% formic acid [FA]) and solvent B (acetonitrile [ACN ]+ 0.1% FA), unless otherwise specified. The flow rate was set to 0.5 mL/min, unless otherwise specified. After injection, the samples were eluted with the following linear gradient: 0–1 min, 5% B, 1–4 min, 5%–60% B, 4–10 min, 60%–99% B, followed by a 3-min washout phase at 99% B and a 3-min re-equilibration phase at 5% B. Data-dependent acquisition (DDA) of MS/MS spectra was performed in positive mode. Electrospray ionization (ESI) parameters were set to 53 L/min sheath gas flow, 14 L/min auxiliary gas flow, 0 L/min sweep gas flow, and 400°C auxiliary gas temperature; the spray voltage was set to 3.5 kV, and the inlet capillary to 320°C, and a 50 V S-lens level was applied. The MS scan range was set to m/z 150–1,500 with a resolution at m/z 200 (Rm/z 200) of 17,500 with one microscan. The maximum ion injection time was set to 100 ms with an automated gain control (AGC) target of 1.0 × 10⁶. Up to five MS/MS spectra per MS1 survey scan were recorded in DDA mode with Rm/z 200 of 17,500 with one microscan. The maximum ion injection time for MS/MS scans was set to 100 ms with an AGC target of 5 × 10⁵ ions. The MS/MS precursor isolation window was set to 1 m/z. The normalized collision energy was set to a stepwise increase from 20 to 30 to 40% with z = 1 as default charge state. MS/MS scans were triggered at the apex of chromatographic peaks within 2 to 15 s from their first occurrence. Dynamic precursor exclusion was set to 5 s. Ions with unassigned charge states were excluded from MS/MS acquisition, as well as isotope peaks.

An in-house modified version of MZmine2.37, corr.17.7, available at https://mzmine.github.io/iin_fbmn, was utilized to perform all feature finding steps. MS/MS spectra were converted to .mzML files using msConvert (ProteoWizard) (57). MS1 feature extraction and MS/MS pairing were performed with MZmine2. Intensity threshold of 5 x 10⁴ for MS1 spectra and of 5 × 10² for MS/MS spectra was used. MS1 chromatogram building was performed within 10 ppm mass windows and a minimum peak intensity of 5E4. Extracted ion chromatograms (XICs) were deconvoluted using the local minimum search algorithm with a chromatographic threshold of 80%, a search minimum in RT range of 0.2 min, and a median m/z center calculation, with m/z range for MS2 pairing of 0.01 and RT range for MS2 scan pairing of 0.2 min. MS1 feature lists were joined using an m/z tolerance of 10 ppm and retention time tolerance of 0.3 min; alignment was performed by assigning weights of 75 to m/z and 25 oto retention time. Feature areas and feature correlation pairs were exported as .csv files. The corresponding consensus MS/MS spectra were exported as an .mgf file. For spectral networking and spectral library matching, all files were analyzed through the feature-based molecular networking workflow (FBMN) on GNPS using default parameters and public spectral libraries, in addition to two expanded bile acid libraries and suspect libraries (28–31, 58, 59). The metabolite were renamed based on the literature of LIPIDMAPS and the candidate bile acids library to account for different possibilities of isomers (28, 32). The results are available at the following link: https://gnps.ucsd.edu/ProteoSAFe/status.jsp?task=f41b3349635c4db3951083ad68727fc2. Data set and analysis files, including MZmine2 parameters, are available on MassIVE (MSV000088063).

Statistical analysis

To calculate 16S and metabolomics dissimilarities among our three conditions, we calculated the robust Aitchison β-diversity calculated using DEICODE in QIIME 2 v 2021.4 (--p-min-feature-count 10 and --p-min-sample-count 500) for each omic data set and visualized the results using robust principal component analysis (RPCA) plots (24, 25). We used PERMANOVA to test whether there were any significant diversity differences in pairwise comparisons among conditions.

To determine what microbes are differentially abundant between non-MASH and MASH mice at 4 and 7 weeks of disease progression, we used BIRDMAn, a Bayesian framework for performing differential abundance analysis (25). Because there were minimal differences driven by time of collection but clear separation of FA and FT in the 16S RPCA plots, we ran BIRDMAn to compare non-MASH and MASH in FA and FT separately but combined the ZT1 and ZT13 samples. The model we used was a negative binomial model from the SingleFeatureModel class with a beta prior of 5, individual dispersion lognormal prior of 0.5, and 500 iterations. while combining the ZT1 and ZT13 samples. The model we used was a negative binomial model from the SingleFeatureModel class with a beta prior of 5, individual dispersion lognormal prior of 0.5, and 500 iterations.

To determine if the differential abundance of a metabolite was credible, we filtered for microbes whose 94% highest density interval (HDI) did not cross 0 for its mean ratio value. We plotted the 16S results using the ggalluvial R package to show microbes that were differentially abundant before and during MASH (60). We performed a similar analysis on our metabolomics to find non-MASH vs. MASH differences present at week 4 and week 7. However, because there were no differences between FA and FT, but clear differences based on time of collection in the metabolomics RPCA plots, we ran BIRDMAn to compare non-MASH and MASH on ZT1 and ZT13 separately but combined the FA and FT samples. Except for setting the individual dispersion lognormal prior value of 1, the BIRDMAn parameters for the metabolomics analysis were the same as those used for the 16S analysis.

To find the correlation between the microbial and metabolomic signatures of interest, we used the correlation function in joint-RPCA from Gemelli in QIIME 2 v 2021.8 (24, 56). Correlations were calculated using paired stool samples with both omics measured from the same specimen and analyzed within each disease stage. We ran BIRDMAn on the Caussy et al. 16S and fecal metabolomics human data set (Qiita ID: 11635) to identify differences between individuals with non-MAFLD and MAFLD using the same method as for the 16S in this study (16). This data set was processed using the same method used for the 16S in this study. A search for the presence of the metabolites of interest in public data sets, particularly microbial data sets was performed using the Mass Spectrometry Search Tool (MASST) through FASST search with a precursor and fragment ion tolerance of 0.02 Da, a cosine similarity threshold of 0.7, and a minimum of four matched signals (27, 61, 62). These searches were conducted with a script available at https://github.com/robinschmid/microbe_masst. The human metabolomics data were processed using the same method as the metabolomics data in this study, and the corresponding FBMN job is available at https://gnps.ucsd.edu/ProteoSAFe/status.jsp?task=778cfcb06dec4af3bf8bc8c50bb388dc.

DATA AVAILABILITY

All data and code are available for review at https://github.com/ZarrinparLab/MASH-TRF and are also available upon request. The 16S rRNA gene amplicon next-generation sequencing data analyzed in this study are available in the European Nucleotide Archive under accession number PRJEB116074, and the LC-MS/MS data are available in massIVE under MSV000088063.

ETHICS APPROVAL

Animal experiments were conducted in accordance with the guidelines of the IACUC of the University of California, San Diego (protocol #S16285).

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/msystems.00720-26.

Supplemental Figures. msystems.00720-26-s0001.pdf.

Fig. S1 to S4.

Supplemental Tables. msystems.00720-26-s0002.xlsx.

Tables S1-S5.

OPEN PEER REVIEW. reviewer-comments.pdf.

An accounting of the reviewer comments and feedback.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.