Dietary Intervention Reverses Fatty Liver and Altered Gut Microbiota during Early-Life Undernutrition

Nonalcoholic fatty liver disease (NAFLD) remains a global epidemic, but it is often studied in the context of obesity and aging. Nutritional deficits, however, also trigger hepatic steatosis, influencing health trajectories in undernourished pediatric populations. Here, we report that exposure to specific gut microbes impacts fatty liver pathology in mice fed a protein/fat-deficient diet. We utilize a multiomics approach to (i) characterize NAFLD in the context of early undernutrition and (ii) examine the impact of diet and gut microbes in the pathology and reversal of hepatic steatosis. We provide compelling evidence that an early-life, critical development window facilitates undernutrition-induced fatty liver pathology. Moreover, we demonstrate that sustained dietary intervention largely reverses fatty liver features and microbiome shifts observed during early-life malnutrition.

O besity-associated nonalcoholic fatty liver disease (NAFLD), with a global prevalence over 25%, remains a leading cause of liver disease around the world (1). While NAFLD incidence increases with age, recent studies estimate that NAFLD also affects 3 to 12% of the pediatric population (1,2).
Up to 30% of NAFLD cases may progress to NASH (nonalcoholic steatohepatitis), a reversible condition characterized by hepatocellular ballooning, inflammation, and fibrosis (3,4). As NASH advances, irreversible damage, including liver cirrhosis or hepatocellular carcinogenesis, may occur (5,6). Linked to obesity and diabetes mellitus, NAFLD is considered both a driver and manifestation of metabolic syndrome, and it has been largely examined as a consequence of overnutrition (1,7). Severe undernutrition, however, has also been shown to promote fatty liver through impaired lipid metabolism (8)(9)(10).
Over 10% of the worldwide population experiences undernutrition (11). Malnutrition accounts for nearly half of all deaths in children under the age of five (12). A critical, environmental burden contributing to the persistence of early-life malnutrition is fecal-oral contamination. Poor sanitation, including the lack of clean drinking water and access to hygienic toilet facilities, promotes chronic exposure to fecal microbes. These microbial exposures alter the gut microbiota and impair nutrient absorption (12)(13)(14)(15). In pediatric populations, fecal-oral contamination and subsequent gut dysbiosis are linked to growth stunting and lasting malnutrition consequences, from neurocognitive impairment to poor metabolic function (14,(16)(17)(18).
Both undernutrition and gut microbes shape health trajectories, including systemic metabolic activity (12,13,16). Indeed, the gut microbiome has been implicated in fatty liver as changes in microbial composition and function, notably alterations of bacterium-mediated bile acid metabolism, have been reported in obese-associated NAFLD (5,19). However, the precise influence of fecal-oral contamination on the etiology, pathology, and persistence of undernutrition-triggered fatty liver remains largely unexplored.
We previously reported chronic fecal exposures, via bacterial gavage, trigger growth faltering, gut dysbiosis, and broad metabolic alterations in malnourished mice-C57BL/6J mice placed on a malnourished diet that received iterative E. coli/Bacteroidales gavage (MBG) model (13,20). The MBG bacterial gavage is comprised of specific fecal commensals (Escherichia coli/Bacteroidales). Subsequent MBG bacterial colonization not only reflects increased relative abundance of E. coli and Bacteroidales commensals reported in pediatric malnourished communities but also elicits gastrointestinal and immune features linked to fecal-oral contamination (13,(20)(21)(22). In an independent research study, colonization by E. coli and Bacteroidetes members, originally isolated from a malnourished pediatric microbiota, impaired weight gain in a murine model of malnutrition (21).
We have previously reported that E. coli/Bacteroidales exposures fail to robustly colonize the gut or trigger growth deficits in control (CON) mice. Furthermore, we demonstrated that enteric Salmonella enterica serotype Typhimurium infection increased hepatic lipidosis and inflammatory markers within the MBG liver, accompanied by a striking increase of hepatic S. Typhimurium burden. Enteric infection, however, failed to trigger immune and hepatic alterations in CON mice repeatedly exposed to E. coli/Bacteroidales (13). These reported findings indicate that the deleterious effects of E. coli/Bacteroidales fecal exposures require malnutrition. Whether fecal-oral contamination promotes fatty liver features in the absence of enteric insult remained undetermined.
Here, we show that exposure to specific fecal commensals (Escherichia coli/Bacteroidales) exacerbates triglyceride accumulation in the context of undernutrition. MBG livers display diffuse macrovesicular lipidosis accompanied by a striking shift in the liver metabolome, notably polyunsaturated fatty acid (PUFA) and retinol metabolism. Multiomic analyses linked phenylacetate and glycerophospholipid metabolism to hepatic steatosis and the MBG gut microbiome. Fatty liver histology was observed only in young mice, not in adult mice, exposed to MBG conditions. Importantly, we demon-strate that dietary intervention reversed fatty liver pathology and largely mitigated gut microbiota compositional and functional alterations, putatively driving undernutritioninduced fatty liver. Our work indicates a critical window of metabolic development that, when disrupted, may significantly impact liver function. Moreover, this study reveals the complexity of fatty liver pathology, characterizing a multihit model involving diverse metabolomic and microbial alterations. We anticipate that these findings will provide valued insights into the roles of diet and gut microbes in driving fatty liver pathology, highlighting potential therapeutic targets to address undernutritioninduced fatty liver.

RESULTS
Malnutrition and fecal-oral contamination promote fatty liver features. To explore the influence of diet and fecal gut commensals on liver function, we utilized the MBG murine model, an established malnutrition model previously characterized by our lab (13,20). Briefly, newly-weaned C57BL/6J mice were placed on a malnourished diet (MAL mice). This protein/fat-deficient, carbohydrate-rich diet reflects dietary shifts previously reported in undernourished communities (13,(23)(24)(25). To model chronic fecal exposure, a subset of MAL mice received iterative E. coli/Bacteroidales gavage (MBG mice). Control (CON) mice, placed on a standard chow diet of equivalent caloric value, provided a healthy control (Fig. 1A).
We further characterized malnourished growth deficits with X-ray micro-computed tomography (micro-CT) (five mice per group). Following micro-CT scanning, threedimensional images were reconstructed, and bone, lean, and adipose tissues were segmented using MicroView software (Fig. 1B). MAL and MBG mice exhibited a modest, albeit not significant, decrease in total volume (see Fig. S1A in the supplemental material). Fecal-oral contamination exacerbated growth alterations, notably loss of bone and lean body percentage. In contrast, both total volume and percentage of adipose tissues increased within MAL and MBG mice ( Fig. 1B and C and Fig. S1), indicative of impaired nutrient storage and metabolism, a process requiring healthy liver function (26).
As overnutrition-associated NAFLD and metabolic syndrome are highly connected (7), we also assessed clinical features of metabolic disruption. Insulin levels were comparable across groups under nonfasting and fasting conditions. While nonfasting mice exhibited comparable glucose concentrations, fasting glucose levels were elevated within MAL and MBG sera, possibly indicative of early insulin resistance and altered glucose metabolism (Fig. S2D). These results indicate that MAL/MBG mice display undernutrition-induced fatty liver features, with fecal-oral contamination promoting fat/glycogen-associated steatosis and impaired triglyceride metabolism.
To further characterize metabolic shifts, we conducted untargeted metabolomics for less polar and polar metabolites via reversed-phase ultrahigh-performance liquid chromatography-Fourier transform mass spectrometry (RP-UPLC-FTMS) and hydrophilic interaction chromatography-FTMS (HILIC-FTMS), respectively, with four mice per group. Over 1,000 differentially abundant hits were detected following FTMS (one-way analysis of variance [ANOVA] Fisher's least significant difference [LSD], adjusted P value [cor-rected for the false discovery rate] [Padj] Ͻ 0.05). Of these hits, ϳ350 differentially abundant metabolic features were annotated using the METLIN database. Diet predominantly shifted the liver metabolome as reported by unsupervised principalcomponent analysis (PCA) (Fig. 3A), and pathway analyses found no significantly enriched MAL versus MBG metabolomic pathways following false-discovery rate (FDR) correction (data not shown). Subsequent metabolite set enrichment analyses using Metaboanalyst 4.0 focused on dietary-driven metabolomic shifts (27).
Metabolite set enrichment analysis identified phosphatidylethanolamine (PE) biosynthesis, sphingolipid metabolism, and phospholipid biosynthesis as the top enriched metabolomic pathways identified in malnourished (MAL and MBG) livers (Fig. 3B), metabolic shifts observed in both undernutrition and hepatic steatosis models (13,28). Mid-coronal plane with lean and adipose tissue highlighted in blue and green, respectively, is overlaid on surface-rendered bone tissue (white) (five mice per group). (C) Percent lean, bone, and adipose tissues from murine body (shoulder blade to tail). Bars indicate means Ϯ standard errors of the means (SEM) (error bars). Statistical significance was determined by Kruskal-Wallis with post hoc Dunn's test and indicated as follows: *, P Ͻ 0.05; **, P Ͻ 0.01; ns, not significant.
Compared to malnourished counterparts, CON livers exhibit enriched pathways linked to bile and PUFA metabolism, specifically ␣-linolenic acid (␣LA) and linoleic acid (LA) metabolism (Fig. 3B), broadly reflecting shifts previously reported within the healthy small intestine (13).
To extend untargeted metabolomic data, we conducted fatty acid profiling of total fatty lipids from liver tissue via gas chromatography. Malnourished mice displayed a reduction in saturated fatty acid (SFA) percent content (Fig. 4A). In contrast, malnutrition elevated relative unsaturated fatty acid (UFA) percent content (Fig. S3A). This increase was largely driven by monounsaturated fatty acid (MUFA) content, as MAL and MBG livers display nearly half the PUFA mol% of CON livers ( Fig. 4A and B).
Dietary LA and ␣LA serve as precursors to downstream 6 and 3 PUFAs, respectively. MAL and MBG mice exhibit a persistent loss of LA and ␣LA metabolism supporting metabolic enrichment analyses ( Fig. 3B and Fig. 4A and B). Elevated 6/3 ratios-an inflammatory marker often associated with Western diets, have been associated with NALFD and NASH (29,30). In our model, healthy and malnourished mice exhibit comparable 6/3 ratios (Fig. S3B), further suggesting that moderate malnutrition may trigger hepatic steatosis uncoupled from significant inflammation.
Dietary intervention largely reverses fecal microbiota shifts and fatty liver features. We next sought to assess whether malnutrition and fecal-oral contamination intensifies fatty liver features specifically within early critical windows (childhood) or whether these perturbations trigger comparable fatty liver features in adult mice. Furthermore, we explored whether dietary intervention halts or reverses fatty liver pathology of early-life malnutrition.
To address these unknowns, we conducted diet reversal experiments with eight mice per group. As MAL and MBG mice exhibited similar liver metabolomic profiles, we chose to utilize the MBG model during subsequent intervention studies. CON and MBG mice served as healthy and undernutrition-induced fatty liver controls, respectively. Following the initial 4-week phase, a subset of control mice was switched to the MBG model-C-MBG (adult-onset malnutrition)-while a subset of MBG mice received the control diet-MBG-R (reversal arm)-in order to assess the impact of dietary intervention on MBG fatty liver, model set-up reported in Fig. 5A.
After early-life malnutrition (7-week time point), MBG mice displayed significant weight faltering (t ϭ 3.73, P ϭ 0.0008). During the subsequent reversal phase, C-MBG mice exhibited a modest, but not significant, weight loss. The final weights of MBG-R mice, however, were comparable with those of CON and C-MBG mice by week 11 (F 3,28 ϭ 6.786, P ϭ 0.0014; Fig. 5B). We observed no significant difference in tail size at week 11, suggesting that tail length differences previously reported in young malnourished mice (13), are largely resolved by adulthood and not influenced by adult-onset malnutrition (Fig. S4A).
As we introduce E. coli/Bacteroidales commensals in the initial malnutrition phase (MBG and MBG-R) and reversal phase (C-MBG), we also assessed fecal microbiota composition by 16S rRNA sequencing across time with fecal pellets taken upon weaning (arrival), after the initial malnutrition phase (week 7), and following reversal (week 11) and report the relative abundance of bacterial members by family classifi-  Fig. 5C). The relative abundance of specific bacterial gavage members was increased in C-MBG and/or MBG and/or MBG-R mice, but not CON animals (Fig. S4C).
Diet significantly influenced the relative abundance of select bacterial members with family and genus annotations reported in Fig. 5E and Fig. S4E. Bacteria from Coriobacteriaceae (H ϭ 19.034, Padj ϭ 0.002) and Streptococcaceae families (H ϭ 22.537, Padj ϭ 0.001) exhibit divergent shifts in response to malnourished diet. The relative abundance of Coriobacteriaceae species increased upon malnutrition, while dietary intervention somewhat mitigated Coriobacteriaceae bloom in MBG-R mice. Increased relative abundance of Coriobacteriaceae has been reported in rodent models following chronic stress (31,32), suggesting that this bacterial family may be a marker of systemic strain. In contrast, the relative abundance of Streptococcaceae bacteria was decreased in malnourished mice (C-MBG and MBG), while MBG-R mice displayed increased relative In addition to LA and ␣LA, major PUFAs include DGLA (dihomo-␥-linolenic acid), AA (arachidonic acid), and DHA (docosahexaenoic acid). Fatty acid profiling from the same experiment. Bars indicate means Ϯ SEM. Statistical significance was determined by ANOVA with post hoc Dunnett's test and indicated by asterisks as follows: *, P Ͻ 0.05; **, P Ͻ 0.01; ***, P Ͻ 0.001; ****, P Ͻ 0.0001. abundance approaching CON abundance. Like Streptococcaceae, Erysipelotrichaceae bacteria have been linked with higher fat intake (33,34). However, the relative abundance of Erysipelotrichaceae remained elevated in C-MBG mice and reduced in MBG-R mice (H ϭ 16.366, Padj ϭ 0.005), matching the original, early-life diet. This finding may indicate that early-life malnutrition sets a long-term trajectory for Erysipelotrichaceae abundance, which is resilient against sustained dietary shifts during murine adulthood. Relative abundance of Peptostreptococcaceae was also increased in CON mice (H ϭ 15.366, Padj ϭ 0.005), but showed a striking reduction in the C-MBG, MBG, and MBG-R gut microbiota, suggesting that bacteria within this family are highly sensitive to malnourished diet and may not recover, even after prolonged dietary intervention. As these alterations may reflect model-specific bacterial shifts, we also predicted microbiome metabolic signatures of health and malnutrition using predictive PICRUSt analyses (35). Like compositional alterations, metabolic pathways of the C-MBG and MBG-R microbiota largely shifted toward MBG and CON counterparts, respectively, highlighting a robust microbial response to diet. Top differentially abundant PICRUSt hits following FDR correction included amino acid biosynthesis and degradation pathways, broadly matching metabolomic patterns between CON and MAL intestinal content (13). In addition, malnourished mice (MBG and C-MBG) exhibited elevated pathways contributing to the tricarboxylic acid cycle (TCA) (or citric acid cycle), potentially reflecting increased carbohydrate load in the MAL diet ( Fig. S4F and Table S2).
Fecal-oral contamination and diet drive lasting features of early-life malnutrition, notably physical stunting and gut dysbiosis. Our findings suggest that sustained dietary intervention considerably mitigates these consequences in our early-life malnourished model.
Dietary reversal also mitigated fatty liver features. CON hepatocytes exhibited low-fat/glycogen content as observed by H&E staining, while 11-week-old MBG livers displayed increased hepatic macrovesicular lipidosis compared to 7-week-old MBG mice. Unexpectedly, undernutrition-induced fatty liver histology was not observed in either C-MBG or MBG-R livers ( Fig. 6A and B). In contrast, triglyceride content was significantly elevated in both C-MBG and MBG mice, while triglyceride levels in MBG-R livers were comparable with those in CON mice (Fig. 6C).
PCA of untargeted metabolomics for both less polar and polar metabolites confirmed a significant change in liver metabolomes, with the C-MBG and MBG-R metabolomic profile shifting toward MBG and CON, respectively (Fig. 7A). Over 2,000 differentially abundant hits were detected following RP-UPLC-FTMS and HILIC-FTMS analyses (one-way ANOVA Fisher's LSD, Padj Ͻ 0.05). Nearly 800 differentially abundant hits were putatively annotated by m/z values against the METLIN database.
We first explored metabolomic distinctions between chronic, early-onset malnutrition (MBG) and adult-onset malnutrition (C-MBG). While MBG and C-MBG livers exhibit largely similar metabolomic profiles (Fig. 7A), we identified metabolites elevated in persistent early malnutrition versus adult-onset malnutrition (45 metabolites: MBG enriched over C-MBG plus MBG enriched over CON) for metabolite set enrichment analyses. The top enriched pathway for the MBG profile was phenylacetate metabolism (Fig. 7B), a bacterial product from aromatic amino acid metabolism (AAAM) linked to hepatic steatosis (36). To explore whether the malnourished microbiome might contribute to enriched phenylacetic acid metabolism, we returned to predictive microbiome PICRUSt analyses. The MBG microbiome exhibits significantly elevated AAAM pathways compared to healthy controls (ARO-PWY, Padj ϭ 0.001; COMPLETE-ARO-PWY, Padj ϭ 0.002). While these predicted pathways did not reach statistical significance postreversal in CON, C-MBG, MBG, and MBG-R samples (ARO-PWY, Padj ϭ 0.105; COMPLETE-ARO-PWY, Padj ϭ 0.078), the relative frequency of AAAM pathways was higher in the MBG microbiome than in the C-MBG microbiome, as well as the microbiome of mice fed a healthy diet (CON and MBG-R), indicating a putative causal role for the MBG microbiota in undernourished-induced fatty liver (Fig. S5).
We then categorized MBG-R metabolomic features as "reversible" or "resilient" to dietary intervention. We considered reversible metabolites as those significantly different between MBG-R and MBG but not MBG-R and CON, while resilient metabolites were significantly altered between MBG-R and CON but not MBG-R and MBG. Of the differentially abundant metabolite hits, 505 were categorized as reversible, while only 106 metabolite features were considered resilient, supporting robust reversal of the fatty liver metabolome upon dietary intervention (Fig. 7C). Metabolites were then classified and grouped into metabolomic pathways using the small molecule pathway database (SMPDB). Adaptive immune pathways-B cell receptor (BCR) signaling pathway and T cell receptor signaling pathway-were observed in both reversible and resilient metabolomic profiles. The resilient profile featured many pathways associated with arachidonic acid (AA) metabolism within the liver. In contrast, the reversible profile included metabolites contributing to retinol (vitamin A 1 ) metabolism. Moreover, nearly 60% of reversible metabolites were involved in lipid and fatty acid metabolism, notably amino acid metabolism and various phospholipid biosynthesis pathways (Fig. 7C).
To confirm untargeted metabolomic profiling, we assessed vitamin A metabolites and profiled long-chain fatty acids following dietary reversal. While the healthy and malnourished diets have distinct macronutrient profiles (e.g., reduced fat, elevated carbohydrates), both diets contain identical micronutrient content, including vitamin availability (13). Following absorption within the small intestine, dietary retinol can be esterified into retinyl ester (storage form) or oxidized to retinal and retinoic acid (37). We quantified retinoid levels within murine liver tissue with liquid chromatography (LC)-MS. While retinol levels were comparable across groups, malnourished mice (MBG and C-MBG) displayed reduced retinal and retinoic acid levels. As expected from metabolomic pathway analyses, dietary intervention mitigated retinoid shifts in MBG-R mice (Fig. 7D), supporting a reversible metabolic pattern.
As altered retinol metabolism also influences fatty acid metabolism (38), we assessed long-chain fatty acid profiles following dietary reversal. Fatty acid profiles of 11-week-old CON and MBG livers exhibit patterns similar to those of their 7-week counterparts (Fig. 4A and B and Fig. S3A). As expected, SFA and PUFA percent content remained elevated within healthy livers and reduced in malnourished mice, while the MBG liver displayed increased MUFA mol% (Fig. 8A). The reversal (end) diet, rather than early-life diet, shaped fatty acid content as C-MBG and MBG livers exhibited fatty acid profiles similar to those of MBG and CON livers, respectively (Fig. 8A and B and Fig. S6A). We specifically assessed whether dietary intervention increased the relative abundance  Table S3 and Text S1 in the supplemental material. of 6 and 3 PUFAs within the liver. Decreased mol% of LA and LA-derived 6 PUFAs in malnourished mice (C-MBG and MBG mice) were reversed upon dietary intervention (MBG-R mice). While ␣LA relative abundance was elevated in CON livers, dietary intervention failed to shift ␣LA percent content in MBG-R mice. Dietary intervention, however, reversed shifts in downstream PUFA 3 members, including 20:5 3 and 22:6 3 (docosahexaenoic acid [DHA]). Despite reported PUFA alterations, 6/3 PUFA ratios remained comparable across all groups following dietary reversal (Fig. 8B and  Fig. S6B).
While PCA and metabolite set enrichment analysis revealed diet-induced alterations of phenylacetate, retinol, and fatty acid metabolism, the critical metabolites specifically linked to hepatic steatosis remained uncertain. Finally, we also sought to explore whether these metabolic shifts were associated with, or uncoupled from, microbiome features.
To address these unknowns, we conducted undirected, weighted gene coexpression network analysis (WGCNA) with untargeted metabolomic data. Using the WGCNA R package (39), highly correlated metabolites were clustered into 52 modules across samples without foreknowledge of metabolomic function. Module relationship to clinical traits, including hepatic histology and triglyceride content, the definitive diagnostic features of NAFLD (4,8), was determined by Spearman rank correlation (Padj Ͻ 0.05). Two modules significantly correlated with fatty liver traits: the "yellow" module (positive correlation) and "turquoise" module (negative correlation). These modules also correlated with the end diet (reversal diet), but not starting diet, indicating that metabolites within these modules were responsive to dietary intervention (Fig. 8C). Metabolites within these modules and the "red" module, which was not correlated with any group or clinical trait, were selected for further study and annotation.
Significantly correlated modules were predominantly comprised of glycerophospholipids, hereafter referred to as GP1 (for glycerophospholipid 1; yellow) (105 annotated, 183 nonannotated metabolites) and GP2 (for glycerophospholipid 2; turquoise) (243 annotated, 418 nonannotated). In contrast, the red module (28 annotated, 56 nonannotated metabolites) was largely comprised of cholanoic and taurocholic bile acid metabolites and was designated BA (Table S3). While GP1 and GP2 contain PE and phosphatidylcholine (PC) members, modules also exhibit distinct phospholipid patterns. The total number and relative abundance of glycerophosphoglycerols were more prevalent in GP1, while GP2 was enriched with glycerophosphoserines ( Fig. S6C and Table S3). In addition, GP2, but not GP1, contains SFAs. We also examined the relationship between these modules and predicted microbiome functionality. GP1 and GP2, but not BA, display divergent and significant correlations with key PICRUSt pathways (Fig. S6D). This multiomic perspective not only identifies glycerophospholipid metabolism and fatty acid metabolism as key pathways linked to hepatic steatosis but also supports a causal role for the gut microbiota in driving undernutrition-induced fatty liver.

DISCUSSION
While diet significantly influences NAFLD progression, NAFLD has largely been studied as a condition associated with overnutrition rather than undernutrition (7,8). The MAL/MBG diet reflects dietary aberrations often observed during food insecurity-a poor diet comprised of refined carbohydrates and reduced intake of unsaturated fats and lean proteins (23,40,41). This form of malnutrition exists across both developed and developing countries (40,42,43). In addition, MBG mice model a "secondary hit" contributing to persistent undernutrition-a chronic exposure to fecal commensals due to poor sanitation/hygiene access and fecal-oral contamination (13,44). MBG fecal contamination consists of E. coli and Bacteroidales, commensal microbes associated with both fatty liver and undernourished cohorts (21,22,36).
Although this model cannot capture the complex spectrum of malnutrition and the nonbiological forces driving health disparities in malnourished communities (e.g., socioeconomic status) (12), MBG pathology provides a unique opportunity to examine undernutrition-induced fatty liver. Here, we assessed the liver metabolome and gut microbiome, as altered metabolic and microbial pathways drive obese-associated NAFLD (36,45). Further study of the undernourished gut-liver axis is needed to address additional molecular features (e.g., liver proteomic profiling) and/or go beyond presented analyses (e.g., metatranscriptomic studies to examine hypotheses generated from predictive PICRUSt data). Despite study limitations, MBG findings further knowledge of fatty liver within the context of dietary deficiency and gut microbial dysbiosis.
Prior studies have examined how undernutrition triggers fatty liver. In a proteindeficient rodent model, hepatic steatosis accompanied impaired mitochondrial fatty acid oxidation and hepatic peroxisome loss. Fenofibrate treatment, a peroxisome proliferator-activated receptor ␣ (PPAR␣) stimulant, not only restored peroxisome deficits and improved mitochondrial function but also reduced hepatic steatosis, demonstrating a critical peroxisome-mitochondrion role in undernutrition-induced fatty liver (9). In addition, proinflammatory mediators, epigenetic modification, and reactive oxygen species (ROS) have also been implicated in the progression of pediatric malnutrition (2, 10). Fatty liver features have also been associated with intrauterine growth restriction and maternal undernutrition, further supporting a critical developmental window shaping liver health trajectories (10).
Beyond diet, the gut microbiome has also been implicated in the pathology of obese-associated fatty liver, particularly via modulation of bile acids (5,19). Synthesized within the liver, bile acids are secreted in the small intestine. Gut microbes modify these primary bile acids, forming secondary bile acids (46). We have previously reported shifts in bile acid metabolism in our malnourished model, notably reduction of both primary and secondary tauro-conjugated bile acids, indicative of impaired host function and gut dysbiosis. In contrast, obese NAFLD/NASH cohorts exhibit increased plasma taurocholate levels (5,47), perhaps suggestive of an overnourished NAFLD biomarker or systemic plasma profile. Surprisingly, undirected WGCNA following dietary intervention found no correlation between taurocholic acid module (BA) and fatty liver features, as well as between the BA module and key microbiome pathways. These findings suggest that alterations in taurocholic metabolism are potentially a consequence of, rather than contributor to, hepatic steatosis within the MBG and dietary reversal models.
Here, we report two hepatic metabolomic pathways linked to the MBG microbiome-phenylacetate and glycerophospholipid metabolism. While diet largely shaped the liver metabolome, hepatic steatosis was not observed following adult-onset malnutrition (C-MBG mice). To identify potential metabolomic distinctions between the MBG and C-MBG liver, we conducted metabolite set enrichment analysis which identified phenylacetic acid metabolism enriched in the MBG versus C-MBG metabolome. Phenylacetic acid was recently identified as a driver of hepatic steatosis in a cohort of obese, nondiabetic women (FLORINASH study). Researchers combined hepatic transcriptome, plasma/urine metabolomics, and fecal metagenomics to identify signatures and metabolic contributors of fatty liver. These multiomic analyses revealed disruption of AAAM, a bacterial pathway producing phenylacetic acid. Chronic phenylacetic acid exposure elevated hepatic triglyceride content, triggering NAFLD-like features in mice (36). PICRUSt predictions from our study also revealed elevated aromatic biosynthesis pathways in the malnourished microbiome prior to and following reversal treatment, supporting a microbiome-dependent role in undernutrition-induced NAFLD progression.
Both metabolite set enrichment analyses and WGCNA independently report aberrant lipid metabolism during undernutrition, notably altered glycerophospholipid and fatty acid metabolism. Altered glycerophospholipid profiles have been reported in murine and human cohorts of fatty liver disease and are implicated in hepatic steatosis pathology. For example, PE, a highly abundant mammalian glycerophospholipid, contributes to lipid signaling and serves as a precursor to PC. Both elevated PE metabolism and altered PC/PE glycerolipid ratios have been associated with fatty liver progression (28,(48)(49)(50). In murine livers, disruption of PE biosynthesis via disruption of the CDP-ethanolamine pathway triggered a 10-fold increase of triacylglycerol content in murine livers (28). While our methodology lacks the capacity to identify specific glycerophospholipid species driving hepatic steatosis, further study to explore glycerophospholipids as a mechanism driving hepatic steatosis and steatohepatitis is warranted.
Metabolite enrichment set analyses did highlight specific shifts in fatty acid profiles-PUFA metabolism. Key cell membrane components, PUFAs modulate inflammatory processes, lipid signaling, and triglyceride accumulation (29,51,52), with 3 PUFA supplementation recently examined as a promising NAFLD treatment (29,53). While MAL/MBG mice displayed a striking reduction of hepatic PUFA content, dietary intervention largely restored 6 and 3 PUFA profiles.
PUFAs are metabolized via fatty acid oxidation, a catabolic process influenced by vitamin A metabolites or retinoids. Storage and metabolism of retinoids largely occur within liver hepatocytes and hepatic stellate cells (37). Retinoids are key regulators of hepatic adiposity with retinaldehyde administration inhibiting diet-induced weight gain in mice (38,54). Beyond fatty acid regulation, retinol, retinal, and retinoic acid contribute to diverse biological functions, including vision, adaptive T cell immunity, and gene transcription (55,56). Prevalent in malnourished communities, vitamin A deficiencies drive vision impairments, growth deficits, and even mortality rates (57)(58)(59). Largely stored in the liver, hepatic steatosis is linked to vitamin A deficiencies (60). As both CON and MBG mice consume diets with equivalent vitamin A availability and exhibit comparable dietary retinol levels within the liver, retinal/retinoic acid deficits in malnourished mice likely reflect liver dysfunction (60). Like improved PUFA profiles, dietary intervention mitigated impaired vitamin A metabolism in MBG-R mice, most likely due to reduced hepatic steatosis. Clinical trials assessing vitamin A supplementation on anthropometric measurements in pediatric populations have reported promising, albeit inconsistent results (61,62). These conflicting findings may result as a consequence of undernutrition-induced fatty liver and subsequent impairment of retinol metabolism.
In summary, we demonstrate that diet and the gut microbes alter multiple pathways that contribute to fatty liver features in a mouse model of early-life malnutrition. Malnutrition triggered diffuse macrovesicular lipidosis accompanied by (i) microbiome alterations and (ii) metabolomic shifts in phenylacetate, glycerophospholipid, PUFA, and vitamin A metabolism within the MBG liver. Beyond characterizing malnutrition-induced hepatic steatosis, our work highlights microbial-dependent shifts in composition and function which may contribute to fatty liver pathology and persistence. Sustained dietary intervention largely mitigated these aberrant features, while improving growth markers and reducing fatty liver histology (Fig. 9). In contrast, mice failed to exhibit diffuse macrovesicular lipidosis following adultonset malnutrition and fecal-oral contamination, despite marked shifts in microbiome and metabolomic profiles. These alterations may precede fatty liver pathology in C-MBG mice maintained on a malnourished diet and/or may involve additional disruptions not captured in our study. Alternatively, our findings raise an intriguing possibility of a critical developmental window programming undernutritioninduced fatty liver within this model.
Childhood malnutrition and NAFLD remain global health concerns. The prevalence of malnutrition-induced fatty liver, specifically among pediatric populations, is expected to rise during the oncoming decades (2). Much research has examined NAFLD linked to one arm of the malnutrition spectrum-overnutrition and obesity. Our work provides a multifaceted assessment of undernutrition-induced fatty liver within an early-life model that addresses global health burdens, dietary deficiency, and gut microbiota dysbiosis (12,13,44). We anticipate that these findings will provide critical launching points to identify putative dietary, microbial, and/or metabolomic targets that address fatty liver pathology within undernourished communities.

MATERIALS AND METHODS
Mouse studies. Newly-weaned female C57BL/6J mice were purchased from Jackson Laboratory and housed at the University of British Columbia Modified Barrier Facility (12-h light-dark cycle, ad libitum chow and water access). Mice were randomized into experimental groups with comparable starting weights and housed in ventilated cages filled with wood chip bedding (three to five mice per group). All mouse studies were approved by the Animal Care Committee at the University of British Columbia and the Canadian Council on Animal Care guidelines.
Micro-CT. Micro-CT scans were completed on 7-week-old anesthetized (isoflurane) mice within the Centre for High-Throughput Phenogenomics at the University of British Columbia using the eXplore CT 120 (TriFoil Imaging, Chatsworth, CA, USA). Micro-CT scanning was conducted with in-house protocols (rotation mode, continuous; single scan time, 4 min; entrance dose, 175 mGy). Image data sets were reconstructed into three-dimensional volumes (isotropic voxel size, 100 m). On the basis of published methodology (63), we classified tissues into adipose, lean, and bone tissue with MicroView software (GE Healthcare Biosciences) with the following signal-intensity thresholds Ϫ200 to Ϫ275, Ϫ30 to Ϫ40, and 190 to 250 HU, respectively.
Histology measurements. Individual liver lobes were stored in 10% formalin for 12 to 24 h at room temperature. Following formalin storage, tissues were transferred into 70% ethanol. Paraffin-embedded tissues were sliced and stained with H&E using established practices by the Biomedical Research Centre (Ingrid Barta) or Wax-it Histology Services at the University of British Columbia. H&E tissues were imaged under a light microscope at 40ϫ, and the percentage of fat-or fat/glycogen-associated per image was determined by Fiji (Image J) on 8-bit images. The threshold of "open" space was set by CON histology, and the same threshold settings were applied to all samples.
Ex vivo cytokine quantification. Liver tissues were collected for cytokine analysis. Tissue samples were stored in 1 ml of PBS with cOmplete EDTA-free protease inhibitor prior to homogenization and Notably, enriched metabolism of phenylacetic acid, a bacterial product of AAAM metabolism, corresponds with increased relative frequency of AAAM predicted microbiome pathways in MBG mice, while altered glycerophospholipid metabolism correlates with both microbiome functional profiles and hepatic steatosis. Adultonset malnutrition elicits metabolomic shifts largely uncoupled from hepatic steatosis (not shown), highlighting the importance of an early-life development period in liver function. In contrast, sustained dietary intervention largely mitigates microbial and host metabolic shifts during malnutrition, reducing hepatic steatosis and improving growth. Collectively, these findings demonstrate a putative role for commensal gut microbes in NAFLD and highlight putative host/microbial targets to reduce fatty liver burden in undernourished communities. This figure was created using Biorender.

Reversing Undernourished NAFLD Involves Gut Microbiome
September/October 2020 Volume 5 Issue 5 e00499-20 msystems.asm.org 15 frozen at Ϫ70/Ϫ80°C. Tissue homogenates were centrifuged at top speed (Ն16,000 ϫ g) for 15 min at 4°C, and the resulting supernatants were stored at Ϫ80°C. Cytokine levels from liver supernatants were measured using the BD Biosciences cytometric bead array mouse inflammation kit. All cytokine concentrations were normalized to starting tissue weight. Triglyceride, glucose, and insulin. Triglyceride measurements were determined from liver supernatants using the abcam triglyceride assay kit (ab65336). Triglyceride levels were first normalized to starting tissue weight and then compared against CON samples. Following euthanasia, blood was collected from nonfasted and fasted (overnight) mice. Immediately upon collection, blood glucose was measured via glucometer, while insulin levels were measured from mouse sera by ALPCO Mouse Insulin ELISA kit (catalog no. 80-INSMS-E10). Enzyme-linked immunosorbent assays (ELISAs) were completed according to the manufacturer's recommendations.
(i) RP-UPLC-FTMS analysis. Individual mouse liver tissues in 5 l water/mg of liver tissue, plus two 4-mm metal balls were homogenized on a MM 400 mill (2ϫ, 30 Hz for 1 min). Following a 5-s spin down, methanol-chloroform (4:1) at 25 l/mg liver tissue was added to the Eppendorf tube. Following repeated homogenization method, samples were placed in an ice-water bath for sonication (5 min) and centrifuged for 20 min (15,000 rpm, 10°C). A 60-l aliquot of supernatant from each sample was dried using the same nitrogen evaporator, and subsequent residue was reconstituted in 80% methanol (40 l). Ten-microliter portions of these samples were used for reversed-phase RP-UPLC-FTMS. RP-UPLC-FTMS runs utilized the Waters BEH C 8 (2.1 ϫ 50 mm, 1.7 m) column with 0.01% formic acid in water (mobile phase A) and 0.01% formic acid in 1:1 acetonitrile-isopropanol (mobile phase B). The mobile phase elution gradient was 5 min for 5% to 50% in mobile phase B, 15 min for 50% to 100% in mobile phase B, and 2 min in 100% mobile phase B. The column temperature was maintained at 60°C with a flow rate of 400 l/min. Full RPLC-FTMS methodology is reported at the NIH Common Fund's Data Repository and Coordinating Center website (studies ST001367 and ST001368).
(ii) HILIC-FTMS analysis. Individual sample supernatants were mixed with 120 l of water, 180 l of methanol, and 195 l of chloroform. The mixture was vortex mixed at 3,000 rpm for 30 s before centrifugal clarification. Three hundred microliters of the upper, aqueous phase was precisely taken out and transferred to a "V"-shape LC injection microvial and dried down under a gentle nitrogen gas flow in the nitrogen evaporator. The residue was reconstituted in 50 l of 80% acetonitrile. Ten microliters was injected for HILIC-FTMS.
HILIC-FTMS was performed on a Waters HILIC column (2.1 ϫ 100 mm, 1.8 m) for chromatographic separation of very polar metabolites. The mobile phase was as follows: mobile phase A was 0.01% formic acid in water, and mobile phase B was 0.01% formic acid in acetonitrile. For binary gradient elution, 85% mobile phase B for 1 min; 85% to 25% mobile phase B in 8 min, followed by column equilibration at 85% mobile phase B for 6 min between injections. The flow rate was 0.3 ml/min, and the column temperature was 30°C. The MS instrument was run in the survey scan mode with FTMS detection at a mass resolution of 60,000 full width at half maximum (FWHM) at m/z 400. Two HILIC-FTMS data sets were acquired for each sample, one with positive-ion detection and the other with negative-ion detection. The mass scan range was m/z 80 to 800.
Vitamin A metabolomics. Vitamin A metabolites were assessed at The Metabolomics Innovation Centre. Mouse liver tissue was homogenized in 50% aqueous methanol (25 l/mg tissue) in Eppendorf tubes with two 4-mm metal balls/tube using the MM 400 mixer mill (shaking frequency, 30 Hz for 1 min ϫ 2), followed by sonication in a water bath for 2 min. Hexane (50 l/mg tissue) containing 20 g/ml butylated hydroxytoluene (BHT) (antioxidant) was added to the tube. The mixture was vortex mixed at 3,000 rpm for 30 s before 6-min centrifugation at 15,000 rpm, and at 10°C, the whole phase was split into an upper organic phase and a lower aqueous phase. The organic phase was removed with a gel-loading tip, and the aqueous phase was extracted with hexane again at 50 l/mg tissue. After centrifugation, the organic-phase extracts from two rounds of liquid-liquid extraction were combined and then dried in a nitrogen evaporator. The residue was dissolved in methanol (5 l per mg tissue), containing 0.5 g/ml of beta-tocopherol-D3 as internal standard. 10 l of sample was injected to a C 8 UPLC column (2.1 ϫ 50 mm, 1.7 m) to run UPLC-high-resolution MS on a Thermo Scientific LTQ-Orbitrap mass spectrometer, which was operated with positive-ion FTMS detection at 60,000 FWHM (m/z 400) in a mass scan range of m/z 100 to 1800. Serially diluted, mixed standard solutions of fat-soluble vitamin A (retinol, retinal, and retinoid acid) in a concentration range of 0.01 to 100 nmol/ml per compound were prepared in the same internal standard solution, and 10-l aliquots were injected to acquire the data to construct the linear calibration curves for the quantitation. The mobile phases were a 5 M silver-ion solution (mobile phase A) and acetonitrile-isopropanol (1:1) (mobile phase B) for binary-solvent gradient elution, with a gradient of 30% to 100% mobile phase B in 10 min at a flow rate of 250 l/min. Vitamin A concentrations were calculated from the linear regression calibration curves of their standard compounds.
(i) Fatty acid methyl ester preparation. Total lipids from liver homogenates were extracted via chloroform-methanol (2:1 [vol/vol]; 3 times) with 0.01% butylated hydroxytoluene. The chloroform phase was evaporated under nitrogen, and the fatty acids were transesterified by incubation in 2.5 ml of 5% methanolic HCl at 75°C for 90 min. Following transesterification, 2.5 ml of n-pentane and 1 ml of saturated NaCl solution were added to extract fatty acid methyl esters (FAMEs). The n-pentane phase was separated, evaporated under N 2 gas, and redissolved in 80 l of carbon disulfide. Two microliters was used for subsequent GC analysis.
(ii) Gas chromatography conditions. Gas chromatography analyses was performed on a GC system 7890A with a series injector 7683B and a flame ionization detector (Agilent Technologies, Barcelona, Spain), equipped with a DBWAX capillary column (length of 30 m, inner diameter of 0.25 mm, and film thickness of 0.20 m). The injection port was maintained at 220°C and the detector at 250°C. Injections were performed using the splitless mode. The flow rate of carrier gas (helium 99.99%) was maintained at 1.8 ml/min. The temperature program was 5 min at 145°C, then 2°C/min to 245°C, and finally, the temperature was held at 245°C for 10 min with a postrun of 250°C for 10 min.
(iii) Data analysis. Identification of FAMEs was made by comparison with authentic standards (Larodan Fine Chemicals, Malmö, Sweden). The fatty acid profile detected plus identified plus quantified represents more than 95% of the total chromatogram. Results were expressed as percent moles. Here, we normalized against pooled CON samples.
Microbiome and multiomic analyses. (i) 16S sequencing and analyses. Collected fecal pellets were stored in Ϫ70°C prior to DNA isolation with the QIAamp PowerFecal DNA kit (Qiagen catalog no. 12830). Library preparation for 16S rRNA sequencing was performed with barcoded primers (V4 region) as described in reference 65. Upon ensuring successful amplification via gel electrophoresis, PCR amplicons were cleaned and normalized with the Sequal-Prep kit (ThermoFisher catalog no. A1051001), pooled, and sequenced on an Illumina MiSeq (v2 kit, 2 ϫ 250 bp reads).
Demultiplexed reads were analyzed and annotated in QIIME2 (v 2018.2) using the DADA2 pipeline (sampling depth of 22051 bp) and Greengenes 97% operational taxonomic units (OTUs) (66)(67)(68). Additional filtering excluded contaminants (mitochondria, chloroplast). QIIME provided Padj for bacterial families. Downstream microbiome analyses and visualization were performed in R; further details are provided in Text S1 in the supplemental material.
(iii) WGCNA. WGCNA R package (39) identified metabolomic modules (modular eigengene) that correlated with both clinical features, traits, and PICRUSt output (Spearman rank correlation test, Padj Ͻ 0.05). WGCNA was completed on the less polar metabolomic data (positive run) following normalization (as described above in "Untargeted metabolomics and metaboanalyst analyses"). On the basis of a scale-free topology, we chose a soft threshold ␤ ϭ 13. Modules containing Ն5 metabolites were identified, and full clustering criteria and the R code are given in Text S1. Modules were auto-labeled by color. Metabolites within modules of interest were auto-annotated using MassTRIX: Mass TRanslator into Pathways v. 3 (70) (Table S3).
Statistical analysis. Statistical analyses provided were performed with GraphPad Prism software version 7.00/8.00. Results are expressed as the means with standard error of the means (SEM), unless otherwise stated.
Data availability. The data sets generated and/or analyzed during the current study are available from the corresponding author upon request. Microbiome and WGCNA pipelines are reported in Text S1. Raw metabolomic data have been deposited at the NIH Common Fund's National Metabolomics Data Repository (NMDR) website, the Metabolomics Workbench (https://www.metabolomicsworkbench.org), where it has been assigned study identifier (ID) PR000935. These data can be accessed directly via https://doi.org/10.21228/M8TT3R. NMDR is supported by NIH grant U2C-DK119886. Raw 16S sequencing has been deposited in the NCBI Sequence Read Archive (SRA) under accession number PRJNA629327.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. TEXT S1, PDF file, 0.4 MB.

ACKNOWLEDGMENTS
We thank Lisa Thorson, Finlay lab manager, for assistance and support throughout the project. We are grateful for the histological suggestions and analytical advice provided by Ian Welch. In addition, we thank members of the Finlay and Ayala labs for their input and critique during the research process, notably Haggai Bar-Yoseph, Manuel Portero, Rebeca Berdún, Paula Littlejohn, and Zack Gerbec, and for valued research and nutritional advice. We also thank Guobin Sun and Nancy Ford at the CHTP for their critical insights and assistance with micro-CT analyses. Finally, we thank Emma Allen-Vercoe and Michelle Daigneault at the University of Guelph for the initial bacterial strains utilized in the MBG model.
UBC research trainees were funded by UBC, the Canadian Tri-Agency Council, and Vanier Canada Graduate Scholarships (CGS). K. C. Bauer is the recipient of an NSERC-Vanier scholarship. The Ayala and Pamplona labs received funding through the Generalitat of Catalonia, Spanish Government, Instituto de Salud Carlos III, and ERDF (European Regional Development Fund -European Commission). B. B. Finlay serves as a Peter Wall Distinguished Professor at UBC, CIFAR-Humans and the Microbiome Director, and CIHR Foundation Grant recipient. The Finlay lab is grateful for support from operating grants received from CIHR and the Bill and Melinda Gates Foundation.
We declare that we have no competing financial interests.