The present study utilized ACE2 knockout (ACE2 KO, ACE2 ^-/y) mice, as previously described (Niu et al., 2008). Twelve-week-old male ACE2 KO mice and their wild-type littermates (WT) were fed a normal chow diet (13.5% calories from fat; Vital River Laboratory Animal Technology Co., Ltd, Beijing, China). A polymerase chain reaction (PCR) approach was used to confirm the ACE2 KO mice with genomic DNA extracted from tail biopsies. Six-week-old male C57BL/6J mice, purchased from Vital River Laboratory Animal Technology (Beijing, China), were used to induce obesity by feeding them a high-fat diet (HFD; 60 kcal% fat) (Research Diets, New Brunswick, NJ, USA) for 8 weeks. The mice were then randomly divided into control and ACE2-overexpression groups. Adenovirus coding rat ACE2 (Ad-ACE2) and the control green fluorescent protein (Ad-GFP) were respectively injected into the obese mice via the tail vein [5 × 10⁸ particle-forming units (pfu) in 100 μL of saline]. The animals were used on the 6th day post-virus injection. All mice were housed in standard animal laboratories with constant temperature and humidity, in an artificial 12-h light/dark cycle, and given free access to food and water. All animal care and experimental protocols were approved by the Ethics Committee of Animal Research at Beijing Tongren Hospital, Capital Medical University, Beijing, China.

The experimental mice were housed in individual cages to minimize fecal contamination. Approximately 2 g of mouse stool samples was collected, placed in sterile DNA/RNA-free tubes, frozen with liquid nitrogen, and preserved in an −80°C freezer until further analyses. Total genome DNA from the samples was extracted using the Cetyltrimethylammonium bromide (CTAB) method. DNA concentration and purity were monitored on 1% agarose gels, and 10 ng of template DNA was chosen for subsequent amplification. The 16S rRNA genes of distinct V4 hypervariable region were amplified using specific the primer 515F (5′-GTGCCAGCMGCCGCGGTAA-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′) with a barcode. Phusion^® High-Fidelity PCR Master Mix (New England Biolabs, Ipswich, MA, UK) and efficient high-fidelity enzymes were used to ensure amplification efficiency and accuracy. Thermal conditions consisted of an initial denaturation at 98°C for 1 min, followed by 30 cycles of denaturation at 98°C for 10 s, annealing at 50°C for 30 s, and elongation at 72°C for 30 s, with a final extension of 5 min at 72°C. The PCR products were detected by electrophoresis and purified with the Qiagen Gel Extraction Kit (Qiagen, Germany).

The real-time PCR was performed on the Light Cycler 480 real‐time PCR system (Roche, Basel, Switzerland) using the SYBR Green PCR Master Mix kit (Applied Biosystems, Foster City, CA). The primer sequences are as follows: β-actin, Forward primer 5′-AGTGTGACGTTGACATCCGTA-3′, Reverse primer 5′-GCCAGAGCAGTAATCTCCTTCT-3′; ACE2, Forward primer 5′-TCCAGACTCCGATCATCAAGC-3′, Reverse primer 5′-TGCTCATGGTGTTCAGAATTGT-3′.

Sequencing work was performed by the NovelBio Bio-Pharm technology company using an Illumina MiSeq platform. Raw sequence data were obtained and paired-end reads were merged using FASTA36.3.6. Raw tags were processed by the Quantitative Insights into Microbial Ecology pipeline (QIIME, v1.9.0, http://qiime.org) to obtain high-quality clean tags. In summary, sequences with ambiguous bases, a length of less than 150 bp, or a Phred quality score <20 were deleted. The chimera sequences were identified and removed using the UCHIME algorithm. Finally, all effective tags were clustered using the Uparse algorithm, and sequences with ≥97% similarity were assigned to the same operational taxonomic units (OTUs). A representative sequence of each OUT was screened to annotate taxonomic assignments using the Silva Database (SSU128, https://www.arb-silva.de) based on the Mothur algorithm. Normalized OTU abundance information was used for further alpha and beta diversity with the QIIME sequence analysis package. The relative abundance and proportion of species at different classification levels (phylum, class, order, family, genus, and species) of each sample were presented and analyzed using a microbial community barplot based on species annotation. Furthermore, biomarkers and metabolic functional traits of the microbiome were analyzed using LDA Effect Size (LEfSe) and Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUS).

The extracted, purified DNA of each sample was used to construct the sequencing library using the NEBNext^® Ultra™ DNA Library Prep Kit for Illumina (NEB, USA) following the manufacturer’s recommendations. Briefly, genomic DNA was randomly fragmented by restriction enzyme to a size of 350 bp. Then, the DNA fragments were end-polished, A-tailed, and ligated with the full-length adaptor for Illumina sequencing, followed by PCR amplification. The prepared libraries were evaluated for size distribution using Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, California, USA) and quantified using real-time PCR. Whole-genome sequencing was performed on an Illumina HiSeq platform. The sequencing raw data were preprocessed using Readfq (V8, https://github.com/cjfields/readfq) and Bowtie2.2.4 software (http://bowtiebio.sourceforge.net/bowtie2/index.shtml) to remove the reads that were unqualified and of host origin. The SOAPdenovo software (V2.04, http://soap.genomics.org.cn/soapdenovo.html) was employed for the single sample and mixed assembly of the obtained clean data. Fragments shorter than 500 bp in all Scaftigs generated from assembly were filtered out. The remaining Scaftigs were used for open reading frame prediction and redundancy filtering for the gene catalog using MetaGeneMark (version: 2.10, http://exon.gatech.edu/GeneMark/meta_gmhmmp.cgi) and CD-HIT software (version: 4.5.8, http://www.bioinformatics.org/cd-hit/). The clean data of each sample were mapped to the gene catalog using Bowtie2.2.4 for further statistical analysis. Taxonomical annotation was performed using the DIAMOND software (Buchfink et al., 2015) to blast the obtained gene catalog against reference sequences from the Non-Redundant Protein Sequence Database (NR database) (version: 2014-10-19) of NCBI. The taxonomic profiling of the organisms in each sample was quantified by MataPhlAn (http://huttenhower.sph.harvard.edu/metaphlan/). DIAMOND software was implemented to blast unigenes against the KEGG and eggnog databases for further functional profiling.

Approximately 0.1 g of fecal sample was pretreated with 50 μL of 15% phosphoric acid, 100 mg of glass bead, 100 μL of internal standard solution (125 μg/mL of hexanoic acid), and 400 μL of ether. Then, the mixture was ground at 60 Hz for 60 s in a high-throughput tissue grinder, then centrifuged at 12,000 rpm for 10 min at 4°C. The supernatant was collected for SCFA measurement using a Thermo TRACE 1310-ISQ LT Gas Chromatography-Mass Spectrometry (GC-MS) (Thermo, USA).

A total of 0.1 g of fecal samples was individually ground with liquid nitrogen, and the homogenate was resuspended in 400 μL of a methanol and acetonitrile mixture (1:1) by vortexing for 5 min. The samples were incubated at −20°C for 1 h and then centrifuged at 12,000 rpm, 4°C for 15 min. The supernatant was collected and vacuum-dried for further detection. Finally, the supernatant was dissolved in 100 μL of acetonitrile and injected into the liquid chromatography–mass spectrometer (LC-MS) system using the Acquity UPLC Nexera X2 system (SHIMAZDU, JAP, Kyoto Prefecture, Japan) coupled to Triple TOF 5600+ (sciex, USA). Appropriately, 3 μL of the sample was injected onto a ZORBAX Eclipse Plus C18 (3.5 μm * 2.1 mm * 100 mm, Agilent) using a 17-min linear gradient at a flow rate of 0.5 mL/min. Mobile phase A was water containing 0.1% of formic acid, and mobile phase B was acetonitrile containing 0.1% of formic acid. The gradient was set as 2% phase B for 1 min, linearly increased to 90% in 12 min, then a decrease to 2% over the next 1 min, and held at 2% for 4 min. The column was equilibrated for 5 min before every single sample introduction, with the temperature maintained at 40°C. The mass spectrometry was conducted by electrospray ionization in both positive and negative modes, with the following parameters: ion source temperature of 120°C, desorption temperature of 500°C, desolvent nitrogen flow of 600 L/h, cone gas flow of 50 L/h, sampling cone voltage of 27 eV, extraction cone voltage of 4 eV, and a quadrupole scan range of 50–1,500 m/z.

For multivariate analysis, orthogonal partial least-squares discrimination analysis (OPLS-DA) was constructed to determine the distributions and find the metabolic difference using the MetaboAnalyst (http://www.metaboanalyst.ca/MetaboAnalyst/). The variable importance in projection (VIP) scores from OPLS-DA were used for metabolite ranking. The parameters R²Y were used to evaluate the fitting condition of the OPLS-DA models, and Q² was used to assess the predictive ability. The volcano plot was used to filter metabolites of interest; the peaks exhibited statistically significant metabolites and the intensity data of these regions were used for further hierarchical cluster and metabolic pathway analysis. For metabolite identification, the assigned modified metabolite ions were confirmed by matching in the Human Metabolome Database (HMDB) (http://www.hmdb.ca/spectra/ms/search) database and KEGG database (http://www.genome,jp/kegg).

All quantitative data were presented as the mean ± SEM. Comparison of ACE2 KO vs. WT mice and ACE2-overexpressed vs. GFP mice was tested by the Mann–Whitney U-test using GraphPad Prism 8 (GraphPad Software, San Diego, California, USA). The differences were deemed significant at p <0.05 and a false discovery rate (FDR) q-value <5%. For 16S rRNA and metagenomic sequencing, the bacterial taxa analyses, Venn diagrams, 3D-PCoA, and cluster analyses were carried out using R software (Version 2.15.3). Significant species between groups were assessed using a Kruskal–Wallis (KW) sum-rank test by LEfse. PICRUSt2 was adopted for functional prediction, with statistical significance determined using Welch’s t-test with corrections for multiple comparisons via FDR on the STAMP platform. For untargeted metabolomics analysis, the raw data were processed using Mass Spectrometry-Data Independent Analysis software (MS-DIAL) (Tsugawa et al., 2015) and MetaboAnalyst 5.0 (https://www.metaboanalyst.ca/) to obtain clean data, which were counted using the statistical software R (R version R-3.4.3), Python (2.7.6 version), and CentOS (release 6.6). The metabolites with VIP scores>1.0, FC>1.2 or FC<0.833, and p-value<0.05 were considered statistically significant contributors. Volcano plots were performed on the basis of log₂ (fold change) and −log₁₀ (p-value) of metabolites by ggplot2 in R. The data of differential metabolites were normalized using z-scores and plotted by Pheatmap package in R for clustering heatmaps. The metabolic pathway enrichment of differential metabolites was performed using the KEGG database, with p-values less than 0.05 considered statistically significant.

To investigate the effects of ACE2 on gut microbial community structure, we used the ACE2 knockout mouse model and analyzed the fecal microbiome by sequencing the V3–V4 regions of 16S rRNA. After filtering unqualified tags, we obtained a total of 327,194 reads, with an average of 23,371 ± 5,977 per sample. As sequencing depth increases, the rarefaction index increases sharply with sequences below 5,000, and the rarefaction curve approached a saturation plateau with sequences exceeding 10,000, suggesting comprehensive species coverage in the samples ( Figure 1A ). The rank abundance curves showed similar slopes and widths between the WT and ACE2 KO group, indicating no significant difference in microbial richness and evenness ( Figure 1B ). According to the abundance of OTUs in each group, the Venn diagram illustrates that a total of 10,272 OTUs were identified as major microbiota, shared by two groups, accounting for approximately 54% of each compartment ( Figure 1C ). For the evaluation of richness and diversity of microbial communities, the alpha diversity values were analyzed. Compared with the WT mice, the ACE2 KO mice had slightly lower Chao1 and ACE index, but no significant difference in Shannon indexes ( Figures 1D–F ). The microbial community structure (beta diversity) was performed using the 3D-PCoA based on UniFrac distances. As shown in Figure 1G , there is an apparent separation between WT and ACE2 KO groups according to gut bacterial composition.

To clarify the influence of ACE2 in microbial diversity, the obtained bacterial OTUs were further classified according to the species database. The dominant phyla in both groups were Firmicutes, Bacteroidetes, and Proteobacteria. The relative abundance of Deferribacteres was significantly increased in the ACE2 KO mice, considered pathogenic bacteria harmful to the intestinal mucus layer (Ravussin et al., 2012).

The Firmicutes-to-Bacteroidetes ratio of bacteria abundance, which is linked to obesity, showed no differences between the two groups ( Figure 1H ). At the family level, the proportion of Alcaligenaceae increased markedly in ACE2 KO mice compared to WT mice. The bacteria of Clostridiales vadinBB60 group, Erysipelotrichaceae, and Deferribacteraceae were also enriched in the ACE2 KO mice, but these changes were statistically insignificant ( Figure 1I ). At a genus level, ACE2 deletion substantially decreased the relative abundances of Marvinbryantia and Alistipes, whereas it increased the relative proportion of Anaerotruncus, Catenibacterium, and Parasutterella. Moreover, the OTUs corresponding to ambiguous taxa of the Bacteroidales S24-7 group decreased, whereas those of the Mucispirillum and [Eubacterium] xylanophilum group were enriched in ACE2 KO mice as compared to the controls ( Figure 1J ). These data collectively demonstrated that the absence of ACE2 results in the modification of natural taxonomic profiling of the gut microbiota.

We next examined how ACE2 overexpression affects gut microbiota in obese mice; Ad-ACE2 injected mice resulted in ACE2 upregulation significantly in the liver, but not in the ileum ( Supplementary Figure 1 ). On the other hand, 16S rRNA analysis was also conducted in the ACE2-overexpressed obese mice. The species accumulation boxplot ( Figure 2A ) showed that the number of observed species increased continuously, achieving a saturation plateau when the sample size reaches 10, suggesting sufficient sampling for data analysis. The rank abundance curve, used to describe species richness and evenness, suggested that the ACE2-overexpressed group had wider distribution ranges on the horizontal axis and shallower slopes than those in the control group, indicating more abundant and uniform species in the former ( Figure 2B ). In the Venn diagram analysis, there were 1,031 common OTUs shared by the two groups ( Figure 2C ). The ACE2-overexpressed group had more unique OTUs than the control mice (461 vs. 306). For the microbial community richness, Chao1 and ACE indexes in ACE2-overexpressed mice tended to increase, albeit not significantly, while the Shannon diversity index exhibited a similar level between the GFP and ACE2-overexpressed group ( Figures 2D–F ). The 3D-PCoA revealed that the gut microbial communities in the two groups could be clearly separated, with Anosim analysis showing consistent results ( Figures 2G, H ). Collectively, these results suggested the overexpression of ACE2-enriched fecal bacterial diversity.

For the specific compositional changes, the top 10 species with the highest relative abundance at different classifications were analyzed using Perl software. At the phylum level, ACE2-overexpressed mice had a lower abundance of Firmicutes and a higher abundance of Bacteroidetes, resulting in a decreased Firmicutes/Bacteroidetes ratio, which was closely related to weight loss ( Figure 2I ). At the family level, we found a decreased relative abundance of Lactobacillaceae and increased richness in Muribaculaceae, Rikenellaceae, Oscillospiraceae, Marinifilaceae, Bifidobacteriaceae, and Ruminococcaceae in the ACE2-overexpressed group compared to the control group ( Figure 2J ). Similarly, there were evident differences between the two groups at the genus level. Compared to the control group, the ACE2-overexpressed group showed a significant increase in the proportion of Alistipes, Colidextribacter, Bifidobacterium, butyricimonas, and Rikenellaceae-RC9-gut group, while the abundance of Lactobacillus and Limosilactobacillus decreased ( Figure 2K ).

Furthermore, the LEfSe method based on linear discriminant analysis (LDA) was employed to identify the most differentially abundant intestinal flora between ACE2 KO and WT mice. However, the cladogram showed no predominant bacteria between the two groups ( Figure 3A ). To further screen for differential microbiota, we used the deep sequencing shotgun metagenomics. As outlined in Figure 3B , Adlercreutzia_equolifaciens (t:GCF_000478885) and Anaerotruncus_sp_G3_2012 (t:GCF_000403395), belonging to the Actinobacteria phylum and Firmicutes phylum, respectively, showed significant abundance increases in ACE2 KO mice. Moreover, sequencing reads were assigned to the Kyoto Encyclopedia of Genes and Genomes (KEGG) database for functional modules and pathways analysis. A total of 6,040 KEGG modules and 316 KEGG pathways were identified. Twelve pathways showed significant differences between the two groups. Among them, six KEGG pathways related to amino acid and pyruvate metabolism, geraniol degradation, novobiocin biosynthesis, and tropane, piperidine, and pyridine alkaloid biosynthesis were upregulated in ACE2 KO mice. In contrast, pathways related to galactose, fructose, and mannose metabolism, thyroid hormone and sphingolipid signaling, ribosome biogenesis in eukaryotes, and type I polyketide structures were downregulated compared to the control group ( Figure 3C ). For Cluster of Orthologous Groups of proteins (COG) functional classification, the results showed that only carbohydrate transport and metabolism were statistically significant and enriched in WT mice microbiomes ( Supplementary Figure 2A ). Then, the gene expression profile related to glucose metabolism was analyzed on the basis of the CAZy database, detecting 290 carbohydrate-active enzymes (CAZymes) families. Among them, glycosyl transferases (GTs), glycoside hydrolases (GHs), and carbohydrate esterases (CEs) were the most abundant classes in both groups, but there were no significant changes in the distribution of these CAZymes ( Supplementary Figure 2B ). Meanwhile, 11 CAZymes families exhibited significant differences between the two groups ( Figure 3D ). For the three enzymes with higher abundance, GH1 and polysaccharide lyase family 0 (PL0) were decreased in the ACE2 KO mice, while CBM56 was increased compared to the control group. These data indicated that the lack of ACE2 may result in the deregulation of glucose metabolism, particularly affecting the hydrolysis of glycosidic bonds and the decomposition of polysaccharides.

On the flip side, the LEfSe comparison identified 18 discriminative species biomarkers (LDA score>4, p<0.05) between ACE2-overexpressed and control groups. Compared to the control group, the ACE2-overexpressed group had an enriched abundance of nine taxa: the phylum Bacteroidota and its corresponding class Bacteroidia, the order Bacteroidales, the families Rikenellaceae and Marinifilaceae, and the genera Alistipes, Odoribacter, and Colidextribacter, as well as the order Peptostreptococcales-tissierellales. Conversely, nine taxa were depleted in the ACE2-overexpressed group: the phylum Firmicutes and its corresponding order Lactobacillales, the family Lactobacillaceae, the genera Lactobacillus and Limosilactobacillus, the species Lactobacillus_johnsonii and Lactobacillus_reuteri, and the genera Dubosiella and Lachnospiraceae_NK4A136_group ( Figures 3E, F ). For the functional prediction using PICRUSt2 based on the KEGG ontology (KO) database, as shown in Figure 3G , the ACE2-overexpressed group exhibited more specific genetic information than the control group (112 vs. 73 genes). Additionally, 56 genes showed a significant difference between the ACE2-overexpressed and GFP groups ( Supplementary Figure 3 ), with eight genes involved in carbohydrate and lipid metabolism. Specifically, the pathway related to 3-oxoacyl-[acyl-carrier protein] reductase was over-represented in the ACE2-overexpressed group, while pathways for sucrose-6-phosphatase, 2,3-bisphosphoglycerate-dependent phosphoglycerate mutase, L-lactate dehydrogenase, UDP-glucose 4-epimerase, cardiolipin synthase A/B, ribose-phosphate pyrophosphokinase, and enolase were under-represented ( Figure 3H ).

As the major metabolic end products, SCFAs have been reported to play a positive role in improving insulin sensitivity and gut barrier integrity, and reducing fat mass (Cani, 2019). To evaluate how ACE2 deletion changes the fecal SCFA profile, quantitative analysis of SCFAs was conducted using GC-MS technology. As illustrated in Figure 4A , the total SCFA concentrations were similar between ACE2 KO mice and WT mice. Similarly, there were no differences in the contents of acetic acid, propionic acid, butyric acid, valeric acid, and caproic acid between the two groups. However, a significant increase in branched short-chain fatty acids (BSCFAs), including isovaleric acid and isobutyric acid, was observed in ACE2 KO mice. These observations confirm that the ACE2 absence has no effect on the contents of major SCFAs but causes a sharp rise in the levels of BSCFAs, which have been closely associated with the gut environment and glucolipid metabolism homeostasis.

The PLS-DA models were employed to characterize the metabolic differences between classes and the score scatter plot showed a clear separation trend of WT and ACE2 KO mice ( Figure 4B ). The cumulative R² at 0.939 and Q² at 0.401 for the model indicated the well goodness of fit and predictive ability of the PLS-DA. As shown in the volcano map in Figure 4C , a total of 88 differential metabolites that met the criteria with fold changes (FC)>1.5 or <0.667 and p<0.05 were identified after the HMDB database screening. The metabolites included carboxylic acids and derivatives, prenol lipids and fatty acyls, nucleosides, and other markers, among which, 43 were upregulated, while 45 were downregulated in ACE2 KO mice ( Additional File 1 ).

Anti-inflammatory factors, such as propyl propane thiosulfonate and secoisolariciresinol diglucoside, and antidiabetic effectors like myricetin, kaempferol, and kobusone were decreased in ACE2 KO mice compared to WT mice, while harmful factors, such as 2’-deoxyguanosine, 4-acetamidobutyric acid, alprenolol, and deoxyadenosine, were significantly increased. The differential metabolites were further analyzed using hierarchical clustering. The heatmap ( Figure 4D ) showed that the metabolic profiles of ACE2 KO mice and normal mice were significantly distinguished by clustering.

Moreover, pathway analysis showed that altered metabolites were significantly enriched in seven KEGG pathways (p < 0.05, impact > 0.01), namely, purine metabolism; nicotinate and nicotinamide metabolism; glycine, serine, and threonine metabolism; pyrimidine metabolism; steroid hormone biosynthesis; arachidonic acid metabolism; and tryptophan metabolism ( Figure 4E ). The correlations between untargeted fecal metabolites, SCFA, and gut microbiota were analyzed using Spearman’s correlation coefficient ( Figure 4F ). The results showed no statistically significant correlation in acetic acid and caproic acid. Other metabolites and gut microbiota in ACE2 KO and control groups constitute a dynamic relationship of mutual influence. Furthermore, the metabolites related to glucose and lipid metabolism altered by ACE2 might impact host health. For instance, the BSCFAs (isobutyric acid and isovaleric acid) were positively correlated with the Erysipelotrichaceae family (Catenibacterium phylum) and the Alcaligenaceae family (Parasutterella phylum). As a natural flavonoid compound, kaempferol was positively correlated with the Alistipes genus, and negatively correlated with the Alcaligenaceae and Anaerotruncus families.

To explore the impact of ACE2 on fecal metabolites, stool samples from control and Ad-ACE2-treated mice were analyzed using untargeted LC-MS. The chemical classification of identified metabolites showed that the top three categories with the higher proportions were lipids and lipid-like molecules, organic acids and derivatives, and organoheterocyclic compounds ( Figure 5A ). PLS-DA revealed that the fecal metabolic phenotype of Ad-ACE2-treated mice was obviously different from that of control mice ( Figure 5B ). A total of 75 metabolites were significantly altered in the stool of ACE2-overexpressed mice compared to control mice ( Additional File 2 ), with 53 metabolites increased and 22 metabolites decreased ( Figures 5C, D ). Furthermore, these dysregulated metabolites were enriched in various metabolomic pathways, with the top 20 KEGG pathways illustrated in Figure 5E . Most differential metabolites were enriched in the metabolic pathways and amino acid biosynthesis. Specifically, these altered metabolites were mainly related to lipid and lipid-like molecules, organic acids and derivatives, organoheterocyclic compounds, and purine nucleosides. The potential correlation between differential bacterial species and fecal metabolites related to glucolipid metabolism, inflammation, and intestinal incretin was examined ( Figure 5F ). Firmicutes were negatively correlated with guggulsterone and positively correlated with 9-Oxo-ODE. Bacteroidota and Lactobacillaceae were positively correlated with guggulsterone, 2’-deoxyguanosine, and N-oleoyl glycine, respectively. Significant negative correlations were found between Rikenellacea and Oscillospiraceae and 9-Oxo-ODE, as well as between Marinifilaceae and Bifidobacteriaceae and N-Oleoyl Glycine. Alistipes showed a positive correlation with guggulsterone and taurocholic acid, and a negative correlation with 9-Oxo-ODE and 2’-deoxyguanosine. Limosilactobacillus showed a positive correlation with N-oleoyl glycine and 9-oxo-ODE, and a negative correlation with guggulsterone. Colidextribacter and Bifidobacterium were negatively correlated with 9-Oxo-ODE and N-oleoyl glycine, respectively. Interestingly, serotonin and all-trans-13,14-dihydroretinol were not correlated with any species.