Cross-tissue multiomics reveals that Akkermansia muciniphila counteracts metabolic syndrome by reprograming gut microbiota, oleoylethanolamide and the gut-hypothalamus axis
1Department of Integrative Biology and Physiology, University of California, Los Angeles, Los Angeles, CA 90095, USA
3Molecular Toxicology Interdepartmental Ph.D. Program, University of California, Los Angeles, Los Angeles, CA 90095, USA
4UCLA Goodman-Luskin Microbiome Center, Vatche and Tamar Manoukian Division of Digestive Diseases, Department of Medicine, David Geffen School of Medicine, Los Angeles, CA 90095, USA
5Institute for Quantitative and Computational Biosciences, University of California, Los Angeles, Los Angeles, CA 90095, USA
6Department of Molecular and Medical Pharmacology, University of California, Los Angeles, Los Angeles, CA 90095, USA
*Correspondence: Dr. Xia Yang, Ph.D., Department of Integrative Biology and Physiology and Department of Molecular and Medical, Pharmacology, UCLA Email: xyang123@ucla.eduAbstract
High fructose diet is a major risk factor for metabolic syndrome (MetS). The gut bacterium Akkermansia muciniphila (A. muciniphila) has been shown to improve fructose-induced MetS, but the underlying mechanism remains unclear. Here, we investigated how A. muciniphila modulates fructose-induced MetS using multitissue, multiomics studies encompassing gut microbiota, plasma and gut metabolome, and hypothalamus single cell RNA-sequencing. A. muciniphila colonization enriched beneficial gut bacteria, increased metabolites including bile acids, endocannabinoids, and vitamins, and activated genes related to oxytocin and vasopressin signaling in hypothalamic neurons. Multiomics network analysis prioritized the metabolite oleoylethanolamide (OEA), an endocannabinoid analogue, as a potential regulator of gut-hypothalamic interaction conferred by A. muciniphila, its associated beneficial bacteria, and bile acid remodeling. Oral administration of OEA to fructose-fed mice recapitulated A. muciniphila effects, including counteracting body weight gain, enhancing thermogenesis, and ameliorating glucose intolerance. Concomitantly, OEA supplementation stimulated expression of its receptors and tight junction genes in the intestine, as well as neuronal activation marker c-Fos and oxytocin and vasopressin signaling genes in the hypothalamus. These findings underscore the regulatory role of A. muciniphila in gut microbiota homeostasis and metabolomic reprogramming, and pinpoint OEA as a key mediator of its action on the gut-hypothalamus axis in alleviating fructose-induced MetS.
Article notes
Competing Interest Statement
The authors have declared no competing interest.
Summary of Updates:
Introduction
Metabolic syndrome (MetS), characterized by three or more of conditions including hyperlipidemia, hyperglycemia, hypertension, insulin resistance, and abdominal fat accumulation, is a risk factor for type 2 diabetes, cancer, metabolically-associated fatty liver disease, and cardiovascular disease in humans and animal models 1–5. Unhealthy diets including high fat western diets and high fructose diets are main contributors to MetS. The mechanisms underlying MetS induced by different risk diets, especially the less studied fructose, is critical to counteract the MetS epidemic and develop precision medicine.
Recent studies of high fructose-induced MetS have revealed multiple mechanisms. First, high fructose reprograms the metabolic, epigenomic, and transcriptomic profiles of the host, especially in tissues such as the hypothalamus, liver, and adipose tissue 6–9. Fructose also interacts extensively with the gut to affect disease risk as fructose alters the gut microbiota and its metabolic outputs, and modulates gut permeability 10. Additionally fructose is first absorbed and metabolized in the small intestine; high fructose consumption overwhelms intestinal fructose metabolism and absorption, leading to fructose spillover to the liver and colon 11. Conversely, host genetics and gut microbiome composition influence fructose metabolism as well as susceptibility to fructose-induced metabolic diseases 8,11–13. For example, C57BL/6J (B6) mice harbor high Akkermansia muciniphila (A. muciniphila) abundance and strong resistance to obesity and glucose intolerance, whereas DBA mice, which have markedly low levels of this bacterium, have high susceptibility to those phenotypes 12. Treatment of susceptible DBA mice with A. muciniphila rendered mice resistant to fructose-induced obesity and glucose intolerance, supporting A. muciniphila as a potential therapeutic probiotic against fructose-induced MetS 12. Additionally, we found A. muciniphila abundance was uniquely correlated with the expression of energy homeostasis-related genes such as Oxt, which encodes oxytocin, in the hypothalamus 12, suggesting communication between A. muciniphila and the hypothalamus in the context of fructose consumption. However, how A. muciniphila confers gut-brain crosstalk to mitigate MetS remains unresolved.
We carried out a cross-tissue multiomics study that utilized single cell RNA-sequencing (scRNA-seq) of the hypothalamus, 16S rRNA gene sequencing of the fecal microbiota, and global metabolomics of fecal and plasma samples, to dissect the molecular cascades involved in the protective effect of A. muciniphila (Figure 1A). Our integrative multiomics analysis reveals metabolite-mediated communication from gut A. muciniphila to the host hypothalamus to regulate energy and metabolic homeostasis that is unique to high fructose-induced MetS.
Materials and methods
Experimental Model and Subject Details
Five-week-old DBA/2J (DBA) male mice (18-23g) were obtained from the Jackson Laboratory (Bar Harber, ME, USA) and housed in a pathogen-free barrier facility at the University of California, Los Angeles. Mice were fed chow diet (Lab Rodent Diet 5001, LabDiet, St. Louis, MO). DBA mice were chosen due to their high susceptibility to fructose-induced obesity and their lack of A. muciniphila based on previous studies 8,12.
Animals and study design for fructose and A. muciniphila treatment
After one-week acclimation, 6-week old mice (baseline, week 0) were subjected to antibiotics (Ab) with a solution of vancomycin (50 mg/kg, Sigma, Saint Louis, MO), neomycin (100 mg/kg, Sigma, Saint Louis, MO), and metronidazole (100 mg/kg, Sigma, Saint Louis, MO) twice daily (9 a.m. and 6 p.m.) for 7 days according to methods previously described before weekly gavage with A. muciniphila (ATCC BAA-835) 14. Ampicillin (1 mg/mL) was also provided ad libitum in drinking water. Antibiotic-treated mice were maintained in sterile cages with sterile food and water and handled aseptically for the remainder of the experiments. Animals showing signs of illness after antibiotics treatment were excluded from the study. Antibiotic-treated mice were randomly divided into four groups (n=8-14/group): A. muciniphila gavage with regular water consumption (AkkW), A. muciniphila gavage along with ad libitum consumption of 8% fructose (NOW Real Food, Bloomingdale, IL) dissolved in water (weight/volume) starting at one week after the initiation of A. muciniphila gavage for 8 weeks (AkkF), PBS gavage (as control for A. muciniphila) with regular water (PBSW), PBS gavage with 8% fructose consumption starting at one-week after the initiation of PBS gavage for 8 weeks (PBSF). For A. muciniphila treatment, 200 μL A. muciniphila suspension (5 x 109 cfu/mL in pre-reduced PBS) was gavaged throughout the experiment as previously described15.
Metabolic phenotypes, plasma, feces, and hypothalamic samples were collected from DBA mice using the protocols previously described 12. In the Ahn et al. paper 12, we have reported body weight gain and glucose level during intraperitoneal glucose tolerance test (IPGTT) in A. muciniphila-treated DBA mice. On top of these, additional phenotypic traits such as body composition (fat mass, lean mass), adiposity, and tissue weights (adipose, liver, heart, kidney, spleen) were showcased in this study. Body mass composition was determined by NMR in a Bruker minispec series mq10 machine (Bruker BioSpin, Freemont, CA). We monitored daily food and water intake as well as a total calorie intake. Fecal samples were collected at weeks 0 (beginning of Ab treatment), 1 (beginning of A. muciniphila treatment), 2 (beginning of fructose treatment), and 6 at the beginning of the 12-h dark cycle (6 p.m.). Mice were fasted overnight before sacrifice, Blood samples were collected from the retro-orbital sinus at the end of fructose feeding (week 10 from baseline) after a 12-h overnight fast and transferred to blood collection tubes with K2 EDTA (BD, Becton Dickinson, USA). Plasma was obtained by centrifugation for 10 minutes at 1,500 x g. After blood collection, mice were sacrificed, and hypothalamus samples were dissected as described in the previous study 16. All samples were snap frozen and stored at -80°C until further analysis. Plasma samples were used for metabolomics; fecal samples were used for 16S rRNA sequencing and metabolomics; hypothalamic samples were used for scRNA-seq.
Animal studies were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. All experimental protocols were approved by the Institutional Animal Care and Use Committee at the University of California, Los Angeles.
16S rRNA sequencing of fecal samples
Total bacterial DNA was isolated from fecal samples using ZymoBIOMICS DNA Miniprep extraction kit (Zymo Research, USA) according to manufacturer’s instructions. Bacterial 16S rRNA was amplified for amplicons spanning the variable region 4 (V4) using barcoded primers. The amplicons from different samples were pooled, purified, and sequenced using the Illumina MiSeq4000 instrument with average sequencing depth of 10,531 reads per sample. Sample processing and sequencing were performed at the Microbiome Core Lab at UCLA.
scRNA-seq of hypothalamus samples
Freshly dissected hypothalami were dissociated with papain (Worthington, Lakewood, NJ, USA) for single cell isolation 20, and cells were suspended in 0.04% BSA-PBS at a final concentration of 100 cells/μl as previously described 21. Single cells were encapsulated in droplets together with barcoded primer beads (ChemGenes, Wilmington, MA, USA) to generate single-cell transcriptomes attached to microparticles (STAMPs), followed by generating cDNA library with the Drop-seq protocol by Macosko et al 2015 22. Paired-end sequencing data was generated using Illumina NovaSeq 6000 at a sequencing depth of ∼40,000 read pairs per cell, with custom length of 26 bp for read 1, 76 bp for read 2, and 8 bp for read 1 index.
Real time qRT-PCR
To test downstream targets of OEA treatment in the intestine, total RNA was extracted from ileum and hypothalamus using Direct-zol Miniprep Kit (Zymo Research, Irvine, CA) and cDNA was synthesized using cDNA Reverse Transcription Kit (Thermo Fisher Scientific, Madison, WI). Real time qRT-PCR was performed in QuantStudio 3 Real-Time PCR System (Thermo Fisher Scientific, Madison, WI) for the relative quantification of Ppara, Gpr119, Nape-pld, Faah, or Cd36 gene expression from ileum which were previously implicated in the activities of OEA. We also quantified Avp, Oxt, c-Fos, Fosb, Jun, and Ganq gene expression from hypothalamus. PCR primers were designed using PrimerBank 25 and the Primer Blast tool available from NCBI web site. Melt curve was checked to confirm the specificity of the PCR product. Relative quantification was performed using the 2 ^ (−△△ CT) method. Beta-actin was used as an endogenous control gene to evaluate the target gene expression levels. All data are presented as the means ± SEM of n = 4-7/ group.
Western blot
Ileum was homogenized in lipa buffer and protein was extracted. Protein concentration was measured using BCA protein assay (Bio-Rad, Irvine, CA, USA). Proteins (20μg per lane) were separated on 4-12% or 12% Mini-PROTEAN® TGX gels (BioRad, Hercules, CA, USA) and transferred onto a PVDF membrane (Pall, Ann Arbor, MI, USA). Membranes were blocked in 5% milk in TBS-T (TBS, 0.1% Tween 20), and incubated at 4 °C overnight with primary antibodies (actin or occludin) diluted in TBS-T with 5% milk. After washing, membrane was incubated in HRP (horseradish peroxidase)-conjugated secondary antibodies for 1 hour. Proteins were visualized using ECL reagents (BioRad, Hercules, CA, USA). Blot images were captured using ChemiDoc Image System (BioRad, Hercules, CA, USA).
Mucin staining
Tissue sections of 6 μm thickness were prepared from frozen ileum embedded in OCT. Alcian blue staining was performed to stain mucin according to manufacturer’s protocol (Abcam, Cambridge, UK).
Bioinformatics analysis 16S rRNA data analysis
Microbial sequence data was analyzed using QIIME2 pipeline (version 2020.8) 26. Raw FASTQ files demultiplexed by sample barcodes were used as input files for QIIME2. Denoising was conducted with DADA2 followed by amplicon sequence variant (ASV) analysis 27. The representative sequences from ASVs were targeted for taxonomy classification via naïve Bayesian classifier with Silva ver. 138 database 28. Alpha and Beta diversity were calculated with Shannon and weighted Unifrac values as input, respectively. Linear discriminant analysis effect size (LefSe) was used to compare the abundances between A. muciniphila-treated and control groups at each time point to obtain the linear discriminant analysis (LDA) score 29.
Representative ASVs from all samples were subject to further functional profiling using PICRUSTs2 30. A series of standard workflows were conducted to place sequence into reference phylogeny to infer 16S rRNA copy number. Using phylogeny and 16S rRNA copy number, gene family profiles were predicted and pathway abundances were measured. All statistical differential abundance (ASV and pathways) were tested using Kruskal-Wallis with pairwise Wilcoxon rank sum exact test as a post hoc test. P-values were adjusted with Benjamini-Hochberg.
Analysis of global metabolomic profiling of plasma and fecal samples
Raw data was extracted and proceeded with peak identification and quality control using Metabolon’s hardware and software31. Metabolites were identified by matching to the library based on authenticated standards that contains the retention time/index (RI), mass to charge ratio (m/z), and chromatographic data on all molecules present in the library. Metabolites were quantified using area under the curve of the peaks. All metabolomics samples were processed and analyzed (differentially abundant metabolites and pathway enrichments) as described previously 32. Briefly, differential metabolites analysis between AkkF and PBSF groups was performed using Welch’s Two Sample t-Test on log transformed data with q-val < 0.05 as an indication of significance with high confidence. In cases where significant metabolites at q-val < 0.05 were not enough to run downstream pathway analysis, p-val < 0.05 was used to select differential metabolites. Pathway enrichment analysis was performed by mapping metabolites to the Metabolon proprietary database. Pathway enrichment score was calculated by The statistical significance was calculated using a Fisher’s exact test and the top representative metabolic pathways were selected with an enrichment score greater than 1.5 and a hypergeometric p-val < 0.05.
scRNA-seq data analysis
DropSeqPipe version 1.13 (https://github.com/Hoohm/dropSeqPipe) was used to preprocess the scRNA-seq sequencing data and generate digital gene expression matrix files. Gene expression matrix from each sample was loaded into the Seurat R package version 4.0.3. Quality control was conducted based on number of genes (200-4000), unique molecular identifiers (UMIs; 500-5000) and percent of mitochondrial genes (<10%) to filter out cells that did not meet these thresholds. Potential doublets were removed using DoubletDecon version 1.1.6 with default parameters 33.
After quality control, all samples were integrated together using the FindIntegrationAnchors function in Seurat. Next, the integrated Seurat object was used for dimension reduction using principal component analysis (PCA), k-nearest neighbors (KNN) for graph construction, and Louvain clustering to cluster cells. Cell clusters were visualized using t-distributed stochastic neighbor embedding (tSNE) and uniform manifold approximation and projection (UMAP).
Cell type identity was determined for each cluster using the reference mapping function in Seurat where hypothalamic scRNA-seq data from Romanov et al. (GSE74672) was used as a reference 34. The data from Romanov et al. was used as a reference because they used tissue specimen that comprises all regions in hypothalamus; paraventricular nucleus, anterior nucleus, suprachiasmatic nucleus, Dorsomedial nucleus, Ventromedial nucleus, and arcuate nucleus. Initially, the mapping resulted in seven different cell types, namely, neurons, ependymal cells, oligodendrocytes, microglia, vascular and smooth muscle (VSM) cells, and astrocytes. Of these, ependymal cells, oligodendrocytes and microglia were subject to further refinement based on cell subtype marker genes from Chen et al 35. Ependymal cells were divided into ependymal and tanycytes based on the expression of tanycyte specific genes (Ccdc153 and Rax). Oligodendrocyte cells were separated into myelinating oligodendrocyte (MO; top marker Mobp), newly formed oligodendrocytes (NFO; top marker Fyn), and Oligodendrocyte precursor cells (OPC; top marker Pdgfra). Microglia was further separated into different states based on the expression of Mrc1, Dab2, and Trem119 to represent under-differentiated microglia 36 (Figure S1A-C). Lastly, neuronal cells were subclustered to elucidate neuronal subtypes that express different neuropeptides. Markers for each of these sub-clusters were found using the FindAllMarkers function. These neuronal subclusters were then named after their corresponding uniquely expressed markers or neuropeptides.
To obtain differentially expressed genes (DEGs) resulting from A. muciniphila treatment for each cell type, we performed non-parametric Wilcoxon rank sum test and Poisson regression using the Seurat package for general cell types and neuronal subtypes, respectively.
Poisson test was used to identify DEGs of neuronal subtypes because the assumption of Poisson model fits better with the data within each neuronal subtype where homogenous cells expressing mRNA of a gene at fixed rate and low UMI counts (average of 1,874 UMIs/cell) 37. The transcriptomes of each cell cluster or subcluster were compared between groups with or without A. muciniphila treatment, namely AkkF versus PBSF groups, and AkkW versus PBSW groups. DEG p-values were adjusted for multiple testing using Bonferroni correction. For pathway analysis which requires a larger number of DEGs, DEGs at p-val <0.05 were used. Pathway analysis was carried out using EnrichR version 3.0, with KEGG_2019_Mouse, BioCarta_2016, Reactome_2016, and GO_Biological_Process_2018 as reference pathway databases 38 and significant pathways were determined using adjusted p-val < 0.05 cutoff.
Integrative analysis across multi-omics data
Within the microbiome data, pair-wise correlation between gut bacteria was calculated using FastSpar to avoid compositional bias from relative abundances in microbiome data 39,40. Pair-wise correlation between A. muciniphila-responsive metabolites in plasma or feces and A. muciniphila gut microbiota, as well as between DEGs from scRNA-seq and metabolic phenotypes including body weight and weight gain were assessed using Biweight midcorrelation (bicor) 41. A network was generated with bacterial taxon, metabolites, DEGs, and phenotypes that had pair-wise correlations at p-val <0.05. Hub nodes were found using the weighted key driver analysis (wKDA) method in Mergeomics with default parameter 42–44. Subnetworks of the hub nodes were extracted using degree of 3. Network was visualized using Cytoscape (version 3.9.1) 45.
Results
A. muciniphila reprograms hypothalamic signaling involved in endocannabinoid–oxytocin–vasopressin axes
A. muciniphila has been reported to contribute to gut-brain crosstalk by modulating neurotransmitter pathways such as serotonin 48,49. Agreeing on the involvement of the gut-brain axis, our previous study showed that A. muciniphila abundance was correlated with the expression of several metabolic genes in the hypothalamus, including Oxt, Th, and Bmp7 12. To understand how A. muciniphila remodels gene expression in the hypothalamus, a key control center of whole-body metabolism, we performed hypothalamic transcriptome profiling at the single-cell level. For scRNA-seq, after pre-processing and quality control, we analyzed 4,198, 3,229, 3,046, and 3,124 hypothalamic cells from the AkkF, AkkW, PBSF, and PBSW groups, respectively. Each group consisted of three mice, for a total of 12 independent scRNA-seq samples. Among the 11 major hypothalamic cell types identified (Figure 2A,B, Figure S1,2A) 34–36,50, cell-type proportions were unchanged by A. muciniphila treatment or fructose diet for most cell types except for neurons (fructose diet effect, p = 0.009) and tanycytes (Akk x diet interaction, p = 0.012) as assessed by 2-way ANOVA (Figure 2C, Figure S2B).
To determine genes that were affected by A. muciniphila in each hypothalamic cell type, we analyzed differentially expressed genes (DEGs) between Akk and PBS groups on either the fructose or the control diet within each cell type. Neurons exhibited the highest DEG numbers indicating stronger transcriptional responses to A. muciniphila, particularly under fructose feeding with 60 DEGs at adjusted p-val < 0.05 and 646 DEGs at p < 0.05, followed by endothelial, mature oligodendrocytes (MO), astrocytes, and microglia (Figure 2D,E). A. muciniphila-induced DEGs in the neurons showed significant overlap between the fructose and control diet conditions (Figure 2E), but many DEGs (e.g., Zfp91, Jund, Usp32, Jun, Mt3, Gnaq, and Apoe) were found only in the fructose condition (Figure S2C, Table S1). Across cell types, DEGs such as those involved in stress responses (Fos, Fosb, Egri, Atf3) showed consistent direction of change under both diets, although DEGs with opposite direction of change between diets were also identified, such as Lamp1, Zfp1, Mt1 and Actb.
Pathway analysis of the A. muciniphila-induced DEGs revealed protein processing in ER, leptin signaling, oxidative phosphorylation, thermogenesis, VEGF signaling, and Wnt signaling as shared across several cell types under fructose diet (Figure 2F, Table S2). In hypothalamic neurons under both fructose and control dietary conditions, mitochondrial oxidative phosphorylation and thermogenesis pathways were upregulated, suggesting A. muciniphila enhanced neuronal energy expenditure and metabolic signaling. Notably, although genes involved in retrograde endocannabinoid signaling were also significantly enriched in neuronal DEGs under both diets, the pathway genes showed an overall increase under fructose diet and an overall decrease under the control diet (Figure 2F, G). Retrograde endocannabinoid signaling is known to regulate presynaptic neurotransmitter release and synaptic plasticity, and to modulate neuropeptide signaling in hypothalamic circuits 51. Consistent with this, neuropeptide signaling pathways including oxytocin (OXT) and vasopressin (AVP) signaling, both of which regulate energy balance and metabolism 52, were upregulated in neurons under fructose diet only (Figure 2F, H). Among DEGs included in OXT signaling pathway, Fos (neuronal activation marker) expression was upregulated in both fructose and water groups, indicating that A. muciniphila activated neurons (Figure 2H, Figure S2D) 53.
To further characterize neuronal changes induced by A. muciniphila, we subclustered neurons into eleven subtypes based on the neuropeptide markers (Figure 2I, Figure S3A). Fructose significantly increased the proportions of Gal (p = 0.007 by Two-way ANOVA) and Npy/Agrp (p = 0.032) subtypes, whereas A. muciniphila induced a non-significant decrease (p=0.09) in Foxb1 neurons (Figure S3B). Significant expression changes in Oxt, Avp, Agrp, Gal and Meg3, regulators of hypothalamic neuroendocrine activity, energy balance, and glucose homeostasis 54–56 were found in the neuronal subtypes (Figure 2J). Among DEGs in the OXT signaling pathway, Fos, Gnaq, Jun, and Actb were mostly upregulated in the neuronal subtypes by A. muciniphila, and Fos was upregulated mainly in Pdyn/Oxt and Npy/Agrp subtypes; the patterns are stronger in fructose group (Figure 2K, Table S1).
Together, these results indicate that A. muciniphila induces broad transcriptomic changes particularly in neurons under the fructose diet condition to upregulate retrograde endocannabinoid signaling, oxytocin and vasopressin neuropeptide pathways, and thermogenic and oxidative phosphorylation gene programs.
A. muciniphila treatment inhibits pathogenic microbes and promotes beneficial gut bacteria
To understand whether A. muciniphila drives the phenotypic and hypothalamic gene expression changes directly or via secondary shifts in the microbiota, we tracked microbial dynamics throughout the treatment process of A. muciniphila. A. muciniphila was present in most of DBA fecal samples at low levels (average 1.3%) at baseline (week 0) before antibiotic treatment (Figure S4A). After a week of antibiotic treatment, the alpha diversity (Shannon index) of all samples decreased significantly as expected (Figure 3A and Figure S4B). One week treatment with A. muciniphila further decreased the overall diversity at week 2. By contrast, the PBS control group showed restoration of microbial diversity at week 2 after antibiotic treatment. At week 6 (5 weeks after A. muciniphila treatment), A. muciniphila increased and became one of the dominant species in the A. muciniphila-treated group compared to week 2 (Figure S4C,D). Principal coordinate analysis (PCoA) plot of beta diversity (weighted UniFrac) showed samples from the A. muciniphila treatment group clustered apart from the PBS control samples (Figure 3B). These results support successful colonization of A. muciniphila by 5 weeks of treatment.
At week 6, the Firmicutes/Bacteroidetes (F/B) ratio, which is positively correlated with obesity 57, showed a significant decrease in response to A. muciniphila treatment in both control and fructose diet groups (Figure 3C). Additionally, Akkermansiaceae and Lactobacillaceae are dominant families in the A. muciniphila treatment group (Figure 3D). At the genus level, A. muciniphila significantly promoted the growth of Lactobacillus, a well-known beneficial bacterium, while reducing Staphylococcus and Aerococcus (Figrue 3E-H). Additional analysis of all amplicon sequence variants (ASV) using the linear discriminant analysis effect size (LefSe) analysis and cladogram confirmed that genus Akkermansia is enriched in the A. muciniphila treated group, whereas in the control PBS group Romboutsia, Peptostreptococcaceae, Aerococcus, Agathobacter, and Staphylococcus were enriched (Figure S5A,B). Among these, Peptostreptococcus, Aerococcus, and Staphylococcus are known to be clinically pathogenic 58,59. Overall, A. muciniphila treatment inhibited bacterial taxa known to be pathogenic and promoted select bacterial taxa that are regarded as beneficial.
To determine the functional implications of these gut microbial changes with respect to A. muciniphila treatment, PICRUSt2 was used to predict the functional profiles of the bacterial community altered by A. muciniphila 30. Our results indicated upregulation of pantothenate (Vitamin B5), menaquinol (Vitamin K2), and phylloquinol (Vitamin K1) biosynthesis pathways by A. muciniphila treatment in both water and fructose diets (Figure 3I-K). Supplementation of all these nutrients has been reported to reduce obesity and insulin resistance 60,61. To the best of our knowledge, an increase in the biosynthesis of these vitamins due to A. muciniphila treatment has not been reported elsewhere. These observed changes in gut bacterial community and functional pathways may contribute to downstream metabolite alterations that further mediate A. muciniphila-host interactions.
A. muciniphila induces metabolomic changes including endocannabinoids related metabolites
To determine whether A. muciniphila modulates metabolites that may mediate subsequent communication with host tissues, we carried out global metabolomic analysis from the fructose-consuming groups with or without A. muciniphila treatment. A total of 813 and 786 metabolites were identified in plasma and feces, respectively. PCA revealed a distinct separation between control (PBSF) and A. muciniphila (AkkF) groups for both plasma and fecal metabolite profiles (Figure 4A). Differential metabolite analysis using Welch’s Two Sample t-test identified 11 plasma metabolites and 344 fecal metabolites that were differentially abundant between AkkF and PBSF at q-val < 0.05 (Table S3 and S4). Since the small number of differential plasma metabolites identified using the stringent cutoff limited our ability to perform pathway analysis, we used a less stringent threshold (p-val < 0.05), identifying 134 plasma and 328 fecal differential metabolites for downstream analysis (Figure 4B-D).
A. muciniphila induced stronger alterations in fecal metabolites compared with plasma metabolites (Figure S6). Metabolites in the lipid family were mostly increased by A. muciniphila in fecal samples but were decreased in the plasma (Figure 4E,F and Figure S6A,B), suggesting reduced lipid absorption into the circulation and increased excretion into feces. Indeed, A. muciniphila has been reported to reduce jejunal lipid absorption and lower serum triglyceride while increasing fecal triglycerides in HFD-fed mice 65. Agreeing with our functional prediction based on the microbial community changes in the previous section (Figure 3I), pantothenate (Vitamin B5) was the most significantly upregulated metabolite in feces, but not in plasma (Figure 4C and 4I).
Thirty-seven metabolites overlapped between plasma and fecal samples. These include nisinate and oleoylethanolamide (OEA) which are increased by A. muciniphila in both fecal and plasma samples (Figure 4C,E-G). Nisinate is a very long chain polyunsaturated omega-3 fatty acid involved in DHA biosynthesis 62. OEA is an endocannabinoid analogue primarily synthesized in the small intestine and a member of the N-acylethanolamine (NAE) family that acts as a neurotransmitter regulating the gut-brain axis via vagus nerve or by directly reaching the area postrema in the brainstem 63. OEA is known to regulate feeding, lipid metabolism, and energy homeostasis by modulating hypothalamic neuropeptide signaling 64–66. Other metabolites in the NAE family were also increased in feces by A. muciniphila treatment, including N-palmitoylethanolamine (PEA) and N-stearoylethanolamine (SEA) (Figure S7A,B). Among these NAEs, only OEA was increased in both plasma and feces (Figure 4C,E-G). Furthermore, the fecal bile acid profile was significantly altered, with increased levels of both primary bile acids such as chenodeoxycholic acid (CDCA) and secondary bile acids, notably ursodeoxycholic acid (UDCA) (Figure 4F,H and Figure S7C,D). CDCA is known to promote the activity of N-acyl phosphatidylethanolamine phospholipase D (NAPE-PLD), an enzyme that plays a crucial role in OEA biosynthesis 67.
Pathway analysis using MetaboAnalyst68 also supported the importance of OEA and the endocannabinoid pathway. A. muciniphila treatment significantly affected five metabolic pathways in plasma (Figure 4J), including phosphatidylethanolamine and phosphatidylcholine, both of which are OEA substrates69. In fecal samples, A. muciniphila affected nine metabolic pathways (Figure 4K) including the endocannabinoid pathway, branched fatty acids, primary bile acids, and pyrimidine metabolism. Therefore, across fecal microbiota, fecal and plasma metabolomics, and hypothalamic scRNA-seq data, the endocannabinoid pathway, known to participate in gut-brain axis signaling 70 was identified as a consistent feature affected by A. muciniphila under fructose diet; within the endocannabinoid pathway, OEA appears to be an important metabolite.
Intestinal gene expression supports changes in OEA biosynthesis and signaling genes
As the intestine is a main site for OEA synthesis and signaling, we investigated genes involved in these pathways upon A. muciniphila treatment. In support of the OEA increase in our metabolomic results, real-time qPCR analysis of ileal tissues showed that A. muciniphila significantly increased the expression of Napepld (encoding enzyme NAPE-PLD responsible for OEA synthesis) without changes in Faah (OEA degradation) and Cd36 (OEA substrate uptake) in fructose-fed mice (Figure 4L). OEA is a specific ligand for peroxisome proliferator-activated receptor-alpha (PPARα) and G protein-coupled receptor 119 (GPR119), both of which regulate food intake and glucose homeostasis via gut-brain signaling 51. Indeed, Pparα and Gpr119 expression levels were significantly upregulated by A. muciniphila treatment (Figure 4L). These results support the notion that A. muciniphila treatment induces OEA biosynthesis and promotes downstream intestinal signaling genes Pparα and Gpr119 to mediate communications between gut microbiota and the host.
Multiomics integration reveals a molecular network involved in A. muciniphila treatment
To further elucidate the molecular cascades involved in A. muciniphila-to-host crosstalk based on our multiomics datasets, we performed a correlation network analysis across the microbiome, metabolome, scRNA-seq, and phenotype data (Figure 5A). Within the microbiome families, we found that Akkermansiaceae was positively correlated with beneficial bacteria Lactobacillaceae, Enterobacteriaceae, Erysipelotrichaceae, and Clostridiaceae, but negatively correlated with pathogenic bacteria Staphylococcaceae, Peptostreptococcaceae, Aerococcaceae, and Streptococcaceae. Thus, the beneficial effect of A. muciniphila may arise not only from itself but also from the establishment of a healthier microbial community.
Between microbiota and metabolome, we observed significant correlations between different bacterial families and bile acids, which supports the important role of microbiome in regulating the bile acid pool 71. Primary bile acids that are produced in the liver can be further modified into secondary bile acids in the gut, and both steps are regulated by bacteria 72. In our network, four primary bile acids (chenodeoxycholate, taurochenodeoxycholate, chenodeoxycholic acid sulfate, cholate sulfate) and three secondary bile acids (hyocholate, taurochenodeoxycholate, 6-beta-hydroxylithicholate) were directly correlated with Akkermansiaceae, suggesting a role of Akkermansiaceae in bile acid regulation. Hagi et al. reported that the ratio of primary and secondary bile acids is important for the growth of A. muciniphila in the culture medium 73. This suggests there is a potential environmental reconstruction feedback loop, where A. muciniphila creates more suitable growth conditions for itself by recruiting other bile acid metabolizing bacteria such as Lactobacillus, Clostridium and Enterococcus 73,74. In our network, the metabolite OEA, whose biosynthesis is modulated by bile acids 75,76, was also positively correlated with Akkermansiaceae. Lastly, Oxt expression in the hypothalamus was positively but indirectly correlated with Akkermansiaceae via inhibition of pathogenic Peptostreptococcaceae and Aerococcaceae.
Taken together, these integrative multiomic results suggest a molecular cascade linking intestinal A. muciniphila to altered hypothalamic signaling (Figure 5B). A. muciniphila orchestrates with other beneficial bacteria (e.g., Lactobacillus, Clostridium and Enterococcus) to affect the bile acid pool, increase OEA substrate (phosphatidylethanolamine and phosphatidylcholine) metabolism, and induce OEA biosynthesis in the intestine through the upregulation of NAPE-PLD. We note that A. muciniphila activated PPARα and GPR119, known receptors of OEA in the intestine, which can transmit signals to the hypothalamus via the vagus nerve 77, 78,79. We also observed that A. muciniphila activated hypothalamic oxytocin and vasopressin signaling as well as energy metabolism pathways (thermogenesis and oxidative phosphorylation), all of which can regulate energy and metabolic homeostasis in fructose-induced MetS. Based on these results, we hypothesize that OEA may mediate the effects of A. muciniphila.
Administration of OEA partially recapitulates the metabolic benefits of A. mucinipihla
To validate whether OEA is one of the metabolites mediating the beneficial effects of A. muciniphila, we orally gavaged OEA (or vehicle) into DBA mice with or without fructose feeding and examined phenotypes and gene expressions in small intestine and hypothalamus (Figure 6A). We observed a significant decrease of food intake and increase of liquid intake in fructose fed mice; however, OEA did not modify these effects. Overall calorie intakes were the same across all groups (Figure 6B,C). Decreased food intake by OEA has been observed in previous studies in control diet-fed rats and high-fat diet mice 80,81, but in our fructose diet study OEA did not influence food intake, demonstrating that the effect of OEA on food intake is dependent on dietary condition. OEA treatment significantly mitigated fructose-induced body weight gain over 8 weeks as well as glucose intolerance (p = 0.036 and p = 0.021 by Two-way repeated measures ANOVA test, respectively; Figure 6D,E). OEA also non-significantly dampened the fructose-induced increases in total and individual fat mass including gonadal and inguinal white adipose tissues (gWAT and iWAT) (Figure 6F). To test whether thermogenesis was involved in the effects of OEA on body weight, we measured body temperature at week 7. OEA significantly increased rectal temperature compared to vehicle under fructose consumption, supporting that OEA induces thermogenesis and energy expenditure (Figure 6G).
To determine whether OEA administration recapitulates the effects of A. muciniphila on intestinal and hypothalamic gene expression, we performed real-time qPCR analysis. In ileum, OEA treatment induced a significant increase in Gpr119 expression and a non-significant increasing trend in Pparα expression compared with vehicle under fructose feeding (Figure 6H). We also tested the tight junction-related genes, Ocln encoding occludin and Tjp1 encoding tight junction protein 1 (ZO-1). Expression of both genes were significantly decreased by fructose diet and OEA reversed the fructose effect (Figure 6I). In hypothalamus, we also tested genes involved in oxytocin and vasopressin signaling which were found to be DEGs in our hypothalamus scRNA-seq analysis for A. muciniphila (Oxt, Jun, c-Fos, Fosb, Jun, Gnaq, Avp). Among these, OEA significantly increased hypothalamic expression of c-Fos, a key immediate early gene and a marker for neural activity 53, Avp, and genes involved in oxytocin signaling including Oxt and Fosb (Figure 6J,K).
Taken together, the metabolic phenotypes and gene expression patterns altered by OEA partially recapitulate the effects of A. muciniphila, with consistencies in decreasing body weight and glucose intolerance, increasing intestinal OEA receptors and tight junction markers, and activating hypothalamic genes involved in neuronal activation and vasopressin and oxytocin signaling.
Comparison of convergent and divergent effects of A. muciniphila between fructose and high-fat diet mouse models
To elucidate whether the effects and mechanisms of A. muciniphila observed in our fructose-induced MetS model converge or diverge from those in the widely used HFD-induced MetS model, we compared our results with previously published data on the HFD mouse model at phenotypic, gut microbiome, metabolomics, and transcriptome levels (Table 1).
At the phenotypic level, A. muciniphila improved metabolic phenotypes in both models by decreasing weight gain, increasing glucose sensitivity, and improving gut barrier function in both fructose and HFD MetS models 47,82–87, but the effect on fat mass was less consistent.
At the microbiome level, in our fructose MetS model A. muciniphila mainly increased Lactobacillus and decreased Lachnospiraceae, but these taxa exhibited opposite trends or no change in the HFD model. A. muciniphila also uniquely decreased Romboutsia, Agathobactor, Fermicutes, and Staphylococcus in the fructose model.
At the metabolomics level, A. muciniphila increased beta-muricholate and decreased taurochenodeoxycholate in both HFD and fructose diets, but several other major primary bile acids (chenodeoxycholate and alpha-muricholate) and secondary bile acids (e.g., ursodeoxycholate) were increased by A. muciniphila only in fructose MetS. All other tau-conjugated bile acids such as tauro-beta-muricholate and tauroursodeoxycholate were decreased by A. muciniphila in fructose diet, but were either increased or unchanged under HFD 84. Apart from bile acids, A. muciniphila increased endocannabinoids in feces under both diets, but the class of endocannabinoids varied according to the diet: acylglycerols (2-PG, 2-AG, 2-OG; activate cannabinoid receptors (CB1/CB2) and exhibit more variable, and often pro-lipogenic, effects on metabolic regulation88) in HFD 89 versus endocannabinoid-like N-acylethanolamines (OEA, PEA, SEA; ligand of PPAR-a and associated with regulation of lipid metabolism and energy homeostasis) in high fructose. Trigonelline, a methylated form of vitamin B3 previously reported to inhibit hepatic lipid accumulation90, as well as long-chain omega-3 fatty acids involved in DHA biosynthesis, docosapentaenoate (DPA) and nisinate, were significantly increased by A. muciniphila in fructose model; however, its effects on these metabolites have not been reported in HFD model despite evidence that DPA exerts beneficial effects 91. SCFA production by A. muciniphila, especially acetate and propionate, is well documented in vitro and is hypothesized to mediate part of its metabolic benefits. In a high-fat high-fructose diet (HFHFD) model, A. muciniphila increased SCFAs in the stool and liver 92,93,94. However, in both HFD 86 and our fructose models, SCFA was not affected by A. muciniphila.
Overall, our systematic comparison of the A. muciniphila effects between our fructose MetS model and the HFD model reveals convergence on improving body weight, glucose sensitivity, and intestinal tight junctions, but highlights remarkable differences in gut bacteria and metabolites between the two models. For hypothalamic or intestinal genes, direct comparison between the two dietary models was not feasible due to the limited availability of reports of A. muciniphila effect on the brain and intestine in the HFD model.
Discussion
In the present study, using a multitissue multiomics systems biology approach coupled with experimental validation, we unveiled the molecular cascades involved in the beneficial effect of A. muciniphila on fructose-induced MetS. At the microbiota level, A. muciniphila treatment promoted the proliferation of beneficial bacteria such as Lactobacillus, Bacteroides,
Clostridium, and Enterococcus while inhibiting the growth of pathogenic taxa including Staphylococcus and Aerococcus. Global metabolomics revealed major increases in bile acids (both primary and secondary) and the endocannabinoid analogue OEA, accompanied by reductions in tau-conjugated bile acids, consistent with the activity of Lactobacillus, Bacteroides, Clostridium, and Enterococcus species known to deconjugate bile acids 67,72,76,95–98. scRNA-seq further revealed hypothalamic alterations in pathways linked to energy metabolism and endocannabinoid signaling across neuronal and glial populations. Experimental validation confirmed that OEA partially mimics A. muciniphila effects in improving fructose-induced obesity and glucose intolerance, reinforcing gut barrier function and activating intestinal receptor signaling, and stimulating oxytocin and vasopressin signaling in the hypothalamus, thereby supporting OEA as a functional regulator of A. muciniphila’s beneficial effects in the gut-brain axis. Together, these data support a gut–brain cascade from A. muciniphila to bile acids to OEA to hypothalamic reprogramming in counteracting fructose-induced MetS.
Previous work has highlighted diverse mechanisms underlying the metabolic benefits of A. muciniphila. For instance, under high-fat feeding, Chelakkot et al. showed that A. muciniphila strengthens gut barrier integrity through TLR2 signaling and extracellular vesicles 99, thereby reducing lipopolysaccharide leakage and pro-inflammatory cytokines such as IFNγ and TNF-α 100,101. It also converts dietary fiber into SCFAs that regulate glucose and lipid metabolism and has been linked to serotonin elevation in the colon and serum 49,93.
In our fructose model, we observed parallel improvements in gut barrier function and dampened hypothalamic inflammation by A. muciniphila (Figure 1I,J, Figure S8). Reduced caloric absorption through enhanced gut barrier function and transit 102 may partially explain the reduced overall body weight. Additionally, activation of hypothalamic energy-expenditure pathways (thermogenesis, oxidative phosphorylation) suggests that elevated energy expenditure drives weight reduction 89,103. Our scRNA-seq analysis indeed supports that A. muciniphila reprograms hypothalamic signaling through both neuronal and non-neuronal cells. Oxytocin (OXT) and vasopressin (AVP) pathways were among the most strongly altered, with cell-type and diet specificity. Under fructose feeding, OXT signaling and AVP-regulated water reabsorption were enriched in neurons as well as other cell types. These neuropeptides are known to increase energy expenditure, lipolysis, insulin sensitivity, and thermogenesis, supporting the observed improvements in metabolic outcomes 52,104,105. Gene-level changes reinforced these pathway alterations: immediate-early genes (Fos, Jun) and neuromodulators (Gal, Meg3) were also induced, consistent with OXT/AVP-driven neuronal activation and glucose homeostasis regulation. Among the neuronal subtypes, Fos upregulation was the strongest in the Pdyn/Oxt and Npy/Agrp subtypes particularly in the fructose group, suggesting A. muciniphila may activate neurons related to energy homeostasis. Other DEGs such as Pkhd1, Ndfip1, and Gas5 are associated with ciliary signaling, proteostasis, and stress adaptation, respectively. Together, these results demonstrate that A. muciniphila exerts central metabolic benefits through diet-specific modulation of hypothalamic neuropeptides and diverse signaling pathways central to energy homeostasis.
In searching for the regulators of A. muciniphila-to-hypothalamus crosstalk, a central discovery from our plasma and fecal metabolomics analysis is the coordinated regulation of bile acid metabolism and OEA synthesis by A. muciniphila in fructose-induced MetS. We observed significant increases in both primary (CDCA, UDCA) and secondary bile acids. In mice, CDCA and UDCA are usually at low levels due to the rapid conversion into muricholic acids, but both were strongly elevated together with muricholic acids under A. muciniphila treatment, suggesting enhanced hepatic synthesis and/or incomplete conversion 106.
Mechanistically, CDCA has the highest affinity for NAPE-PLD, the key OEA-synthesizing enzyme in gut epithelial cells 67. Consistent with this, A. muciniphila upregulated expression of NAPE-PLD, an enzyme involved in OEA biosynthesis, and PPARα and GPR119, the receptors through which OEA acts to activate gut–brain signaling. While OEA shares structural similarity with endocannabinoids, it does not activate CB1/CB2 107, instead triggering PPARα- and GPR119-mediated signaling to promote oxytocin release and hypothalamic signaling reprogramming 108,109. Additional metabolite changes by A. muciniphila under fructose diet, including increased vitamin B5 and vitamin K, further point to potential complementary mechanisms which activate brown or white adipose tissue, respectively, and modulate glucose/fat metabolism 110,111.
Connecting the multiomics data together, our network analysis provides a systems-level perspective of A. muciniphila’s actions. The bacterium fosters a favorable microbial environment, enriching Lactobacillaceae, Erysipelotrichaceae, and Clostridiaceae while inhibiting pathogenic taxa 112–114. Most bile acids were positively correlated with A. muciniphila treatment, recapitulating their mutual reinforcement 106. Importantly, OEA emerged as the most coherent metabolite linking gut microbiota changes to hypothalamic gene regulation via PPARα and GPR119 75,76,115–117. This places OEA at the core of the A. muciniphila–driven gut–brain axis.
While decreasing weight gain, increasing glucose sensitivity, and gut barrier improvement are a common effect of A. muciniphila in both fructose and HFD models, the microbiota and metabolite responses significantly differ (Table 1). In the fructose model, A. muciniphila increased Lactobacillus and decreased Staphylococcus, Aerococcus, and Romboutsia, whereas in HFD feeding, Lactobacillus was primarily reduced 84,118. Metabolite responses also diverged: although some bile acids (e.g., taurochenodeoxycholate, beta-muricholate) shifted similarly, others (e.g., tauro-beta-muricholate, tauroursodeoxycholic acid sulfate) moved in opposite directions. Most notably, A. muciniphila regulated NAEs endocannabinoid analogues (OEA, PEA, SEA) in the fructose model, whereas acylglycerols dominated in the HFD study 14. This distinction is significant because acylglycerols are primarily ligands for the CB1 receptor, which has been associated with psychiatric adverse effects. In contrast, endocannabinoid analogues such as OEA do not act through the CB1 receptor and have emerged as potentially safer anti-obesity therapeutics 51,107. These results highlight MetS subtype- specific molecular mechanisms and identify OEA as a key mediator of metabolic reprogramming by A. muciniphila under fructose-induced MetS. Taken together, these results point to the versatile functions of A. muciniphila to modulate different metabolites with therapeutic effects in different disease settings and offer precision medicine targets for different MetS subtypes.
We acknowledge the following limitations. Causal inference across microbes, metabolites, and host genes requires deeper experimental validation. Although our OEA experiments confirmed activation of gut and brain signaling pathways, whether OEA acts mainly through the vagus nerve or also directly via brain entry remains to be clarified. Caution for translational generalization to humans is also needed. In a human MetS cohort without specific dietary intervention, A. muciniphila did not significantly increase overall circulating endocannabinoids but selectively increased 1-PG and 2-PG 119, which differs from the OEA increase we found under fructose-induced MetS and the previously reported increases in acylglycerols (2-AG, 2-OG, 2-PG) in HFD induced MetS in mice 14. Broader testing across diets, hosts, and disease conditions will be critical to identify universal versus condition-specific mediators.
In summary, our multiomics study delineates how A. muciniphila confers metabolic benefits in fructose-induced MetS in mice. By shaping the gut microbiota, remodeling bile acid pools, and enhancing OEA synthesis, A. muciniphila triggers gut–brain interactions that reprogram hypothalamic energy metabolism and neuropeptide signaling. These findings not only provide mechanistic understanding but also point to the endocannabinoid OEA and bile acid pathways as diet-specific therapeutic targets for fructose-associated metabolic disorders.
Supporting information
Acknowledgments
This work was supported by NIH DK104363 (XY) and QCB Collaboratory Fellowship (S.M.H.). We thank Wini Suryavanshi for her valuable assistance with the animal experiments.
Declaration of interest statement
The authors declare no competing interests.
Ethics approval statement
The ethics application (Approval No: ARC-2012-059) for mouse study was approved by the UCLA Institutional Animal Care and Use Committee.
Lead Contact and Materials Availability
Further information and requests for resources and code should be directed to and will be fulfilled by the Lead Contact, Xia Yang (xyang123@ucla.edu).
Data availability statement
scRNA-seq data have been deposited at NCBI Gene Expression Omnibus (GEO) database (Accession: GSE212491). Raw 16S rRNA sequencing data for all samples have been deposited in the open-source repository NCBI Sequence Read Archive (SRA) database (Accession: PRJNA876737). All mass spectrometry data analyzed in this study have been deposited to Mendeley Data and are publicly available (URL: https://doi.org/10.17632/f55rsncm9y.1). All the reagents and primers used in this study are provided in Table S5.