Strain-level Screening of Human Gut Microbes Identies Blautia Producta as a Novel Anti-hyperlipidemic Probiotic Via the Production of 12-methylmyristic Acid

Background: Compelling evidence has linked the commensal gut microbiota to human metabolic syndromes and provided new therapeutic potentials against diseases, such as hyperlipidemia. However, the precise regulatory effect of each bacterial species on human lipid homeostasis remains largely unknown. Results: We set up a cell-based high-throughput screening platform and screened 2250 human gut bacterial strains from 186 species for the lipid-decreasing activity in HepG2 cells, in which 388 strains steadily inhibited lipid accumulation. Different strains in the same species usually displayed distinct lipid-modulatory actions, revealing an obvious strain-specicity. Blautia producta, Lactobacillus gasseri, and Bidobacterium pseudolongum contained a much higher portion of hypolipidemic strains. Among all the tested strains, the mucosal bacterium Blautia producta exhibited the most potency to suppress lipid accumulation, and gavage of live Bl. producta effectively ameliorated hyperlipidemia in mice. 12-Methylmyristic acid (12-MMA) was identied as an important active metabolite of Bl. producta by pan-genomics and comparative metabolomics, which exerted potent anti-hyperlipidemic effect in vivo and activated G protein-coupled receptor 120 (GPR120), thus stimulating white adipose tissue browning. Conclusions: Together, these data reveal a previously unreported large-scale lipid-modulatory prole of gut microbes at the strain level, and raise the possibility of developing therapeutics based on Bl. producta and microbial metabolite 12-MMA to treat hyperlipidemia. oil red O (b scale bars, 200 μm (d), intracellular TG quantication (e), Fenobirate (10 μM) positive control. in vivo anti-hyperlipidemic effect of live Bl. producta was assessed


Introduction
Mounting progress in recognition of the inherent relationship between gut microbiota and host health not only reveals novel insights into the etiology of critical illnesses but also opens up new avenues for therapeutic interventions [1][2][3][4][5]. It is widely accepted that gut microbes play crucial roles in regulating host energy homeostasis and metabolism, especially lipid metabolism [6,7]. A comprehensive understanding of the lipid-modulatory landscape of various bacterial species will facilitate the development of novel therapeutics against lipid metabolic diseases. Multiple species, such as Lactobacillus spp, Bi dobacterium spp, and Akkermansia muciniphila, are well-documented to possess bene cial effects on hyperlipidemia and non-alcoholic fatty liver disease (NAFLD) [8,9]. For gut microbiome researches, the cultured microbial repository can play indispensable role to facilitate the interpretation of microbial functions and host-microbiome interactions. Several large-scale cultivation efforts have been made to unravel the 'dark matters' of human gut microbiota [10][11][12][13][14]. However, our understanding of the human gut microbiome is still very insu cient. Which species or strains in our gut microbiota have therapeutic effects on lipid disorders remains poorly understood.
Free fatty acids (FFAs) are the best-characterized microbial metabolites to mediate various bene cial effects of gut microbiota on host metabolism and immunity [3,15]. Elevating short-chain fatty acids (SCFAs) production by gut microbes or direct supplementation of SCFAs in foods can signi cantly fecal samples donated by healthy adults in Hainan province, China. Each strain was purely cultivated, identi ed by mass spectrometry, named with a unique code, and then cryopreserved at -80 °C refrigerator.
For screening assay, each bacterial clone was pin transferred from the cryopreserved tube onto a solid YCFA medium (Solarbio) plate and was grown for 24-48 h at 37 °C under anaerobic conditions. A single colony was inoculated into liquid YCFA medium and cultured at 37 °C for 48 h under anaerobic conditions to obtain the rst generation (F1) bacterial solution. 10% (v/v) of F1 bacterial solution was inoculated into fresh liquid YCFA medium and cultured at the same conditions for 48 h to generate F2 solution. Following the similar step as F2 production, the working bacterial solution F3 was further obtained from 10% (v/v) F2. After centrifugation at 1,600 × g for 15 min, the supernatant part was collected and then ltered using a Millipore lter (0.22 μm). These sterile spent culture broth ltrates were used as screening samples in the study.

Cellular lipid accumulation assay
HepG2 cells were purchased from ATCC and were cultured in Dulbecco's modi ed Eagle's medium (DMEM, Thermo Fisher Scienti c) supplemented with 10% fetal bovine serum (Thermo Fisher Scienti c) at 37 °C with 5% CO 2. For lipid accumulation assay, HepG2 cells were seeded in 96-well plates containing 100 μl DMEM. When the con uence reached 85%, the medium was replaced with 70 μl fresh DMEM supplemented with 100 μM oleic acid (OA, Sigma) and 30 μl spent culture broth ltrate or YCFA medium. After incubation for 22-24 h, lipid accumulation was evaluated by oil red O staining or TG quanti cation kit (Solarbio) as described previously [57]. Each experiment (n = 8 for oil red O staining, n = 4 for TG determination) was repeated in triplicate. Liquid YCFA medium and feno brate (10 μM, Sigma) were used as the negative and positive control, respectively. Speci cally, (1) for the large-scale bacteria screening, HepG2 cells were rst stained with oil red O solution at room temperature for 30 min, and then DMSO was added to dissolve stains attached to cells, followed by measuring OD value at 358nm in a microplate reader. Lipid-lowering rate of each examined bacterial strain was evaluated and ranked according to the calculation of [(YCFA OD 358 -Sample OD 358 ) / YCFA OD 358 ]*100%. (2) To con rm the lipid-lowering e cacy of Blautia producta, BODIPY staining and intracellular TG content were measured to corroborate the screening result according to the manufacturers' instructions.

Animal experiment
All the animal experiments were performed in accordance with the National Institutes of Health regulations for the care and use of animals in research. The protocol was approved by the medical ethics committee of Peking Union Medical College (Nos. YZS201904021; YZS201910013; YZS202105022). All efforts were made to minimize animal suffering.
(1) To assess the anti-hyperlipidemic effect of live Bl. producta, 24 male C57BL/6J mice (8-week-old, 20-24 g, Vital River Laboratory Animal Technology) were divided into three groups with eight animals in each group. One group was used as blank control and continued to feed on normal chow (Chow group) while other groups were fed HFD (60% kcal fat as indicated, Beijing HuaFuKang Bioscience). Animals on HFD were gavaged with Bl. producta (HFD+Bl. producta group, 10 9 CFU per animal per day) or an equal volume of YCFA medium (HFD group). Bodyweight was assessed weekly. After 4 weeks of treatment, mice were fasted overnight, anesthetized in chambers saturated with iso urane, then sacri ced by cardiac puncture. Stools for metagenomic analysis were collected on the day before animals were euthanized. Blood was drawn in 1.5 ml centrifuge tubes, and sera were separated for estimation of serum levels of total triglyceride (TG), total cholesterol (TC), low-density lipoprotein cholesterol (LDL-c) and highdensity lipoprotein cholesterol (HDL-c) by respective kits (Nanjing jiancheng Bioengineering Institute). Liver samples were weighed and snap-frozen in liquid nitrogen for sequential biochemical analysis.
(2) To assess the dynamic impact of live Bl. producta on gut microbiota in mice, 16 male C57BL/6J mice (8-week-old, 20-24 g) were randomly divided into two groups, fed on HFD, and gavaged with Blautia producta (HFD+Bl. producta group, 10 9 CFU per animal per day) or an equal volume of YCFA medium (HFD group) for 8 weeks. The fecal materials were obtained from each mouse on Days 1, 3, 7, 14 and 28 after treatment. At the end of the experiment, mice were fasted overnight, anesthetized with iso urane, and sacri ced by cardiac puncture. Blood was drawn in 1.5 ml centrifuge tubes, and sera were separated for parameters evaluation. Liver samples were weighed and snap-frozen in liquid nitrogen for sequential biochemical analysis or xed in 4% paraformaldehyde (Solarbio) for histological analysis.
(3) To assess the anti-hyperlipidemic effect of 12-methylmyristic acid (12-MMA), 16 male C57BL/6J mice (8-week-old, 20-24 g) were randomly divided into two groups and fed HFD. The test group (HFD+12-MMA) was gavaged with 12-MMA (40 mg/kg, Sigma), while the HFD group was given an equal volume of solvent (1% tween 80 + 0.5% carboxymethyl cellulose sodium (CMC-Na, Sigma)) for four weeks. At the end of the experiment, mice were fasted overnight, anesthetized in chambers saturated with iso urane and then sacri ced by cardiac puncture. Blood was drawn in 1.5 ml centrifuge tubes, and sera were separated for parameters evaluation. Liver and fat samples were weighed and snap-frozen in liquid nitrogen for sequential biochemical analysis or xed in 4% paraformaldehyde (Solarbio) for histological analysis.

Histology and immunohistochemistry analysis
The liver and fat tissues of each mouse were xed in 4% paraformaldehyde, embedded in para n, and cut into slides with a thickness of 4 μm. Liver and fat tissue sections were stained with hematoxylin and eosin (H&E) for histological analysis. For oil red O staining, liver tissues from the same liver lobe were cut into small pieces, then the frozen sections were rinsed in distilled water and stained with 0.2% oil red O (Sigma) and 60% 2-propanol (Sigma) for 10 min at 37 o C. For immunohistochemistry analysis, fat slides were rinsed in 0.01 mol/L sodium citrate (pH 6.0) and heated for 20 min in a microwave to retrieve antigen. The sections were blocked in blocking buffer containing 5% goat serum, 2% BSA, 0.1% Triton X-100 and 0.1% sodium azide in PBS, then incubated overnight with anti-UCP1 (Cell Signaling) by a dilution of 1:100 at 4 o C. After being washed twice in PBS, slices were incubated with secondary antibodies (Cell Signaling) for 1 h at room temperature. Slides were counterstained with H&E. All digital pictures were acquired using an EVOS X1 microscopy (Thermo Fisher Scienti c).

DNA extraction
The microbial genomic DNA of fecal samples were extracted by DNeasy PowerSoil kit (QIAGEN) and subjected to 1% agarose gel electrophoresis for evaluation. Concentration and purity of microbial DNA were determined with NanoDrop 2000 UV-vis spectrophotometer (Thermo Fisher Scienti c) and Qubit 3.0 uorometer (Thermo Fisher Scienti c).

Shotgun sequencing
The gut microbial composition was determined by shotgun sequencing of the fecal samples collected from each mouse. Libraries were prepared using KAPA HyperPlus Library Preparation kit (KAPA Biosystems) and quanti ed by KAPA Library Quanti cation Kits (KAPA Biosystems) following the manufacturer's instructions. Shotgun sequencing was performed on Illumina NovaSeq 6000 System (Illumina) at a depth of 1Gb. Cluster generation, template hybridization, isothermal ampli cation, linearization, blocking, denaturing and hybridization of the sequencing primers were performed according to the work ow indicated by Illumina.
As previously described [58], low-quality reads were removed from the raw data by using MOCAT2 Sequencing adapters were removed by using Cutadapt software (version v1.14,-m 30). Then SolexaQA package was used to remove the reads with a threshold of less than 20 or the length of less than 30bp. The reads which could be aligned with the mouse genome (Mus musculus, GRCm38) were cleaned by using SOAP aligner software (v2.21, -M 4 -l 30 -v 10). The clean reads were assembled by SOAP de novo software (an iterative De-Bruijn Graph De Novo Assembler) using the parameters of -d 1,-M 3,-R ,-u,-F to get the scaftigs of at least 500bp. Genes were predicted using MetaGeneMark. A non-redundant gene catalogue was constructed with CD-HIT using the parameters of c 0.95 -aS 0.9. The clean reads were mapped onto the gene catalogue with the length of at least 100bp using BWA software to calculate the gene abundance.
Sequencing data analysis α-Diversity was calculated by a vegan (2.5-6) package and presented by Shannon and Simpson indices. The principal coordinate analysis (PCoA) and nonmetric multidimensional scaling (NMDS) were calculated based on the Bray-Curtis distance using vegan (2.5-6) package. Microbial community composition was analyzed using Metaphlan2 software. Brie y, the query reads were mapped against the reference genomes in RefSeq database (version 82) with a 97% identity threshold. The linear discriminant analysis (LDA) effect size (LEfSe) method (https://github.com/biobakery/lefse) was used to identify species that show statistically signi cant differential abundances among groups. Heat maps were generated by using the R package "heatmap".

Pan-genomics
Twelve reference genomes of Bl. producta were used to perform pan-genomic analysis, in which seven genomes were downloaded from NCBI RefSe or GenBank database and ve genomes were constructed by de novo sequencing of pure Bl. producta cultures. Genes annotation was obtained using prokka software (https://github.com/tseemann/prokka). The pan-genomic analysis was performed by ppanggolin software (https://github.com/labgem/PPanGGOLiN) to acquire gene matrix, jaccard distance-based UPGMA clustering tree, and KEGG functional annotation.

Metabolomics analysis
The untargeted metabolomics pro ling of the pure microbial culture was performed on the metabolomic platform of Shanghai Biotree biomedical technology Co. Ltd.
(1) Sample preparation: The frozen samples were kept on the dry ice-ethanol bath. About 50 mg material was accurately weighed in an Eppendorf Safelock microcentrifuge tube, to which 25 mg of pre-chilled zirconium oxide beads, 10 μl of internal standard, and 50 μl of 50% pre-chilled methanol were added for automated homogenization (BB24, Next Advance, Inc.). After centrifugation at 14,000 g and 4 o C for 20 min (Microfuge 20R, Beckman Coulter, Inc., Indianapolis, IN, USA), the supernatant was carefully transferred to an autosampler vial (Agilent Technologies, Foster City, CA, USA). Each aliquot of 175 μl of pre-chilled methanol/chloroform (v/v=3/1) was added to the residue for the second extraction. After centrifugation at 14,000 g and 4 o C for 20 min, the supernatant was carefully transferred to an autosampler vial. All samples in autosampler vials were evaporated brie y to remove chloroform using a CentriVap vacuum concentrator (Labconco) and further lyophilized with a FreeZone freeze dryer equipped with a stopping tray dryer (Labconco). Oral glucose tolerance test (OGTT) As previously described [57], before the OGTT test, mice were fasted for 6 h and then gavaged of 2 g/kg glucose. The blood glucose concentration in the tail vein was monitored at 0, 30, 60, 90 and 120 min after glucose administration (p.o.) using a glucose meter (Roche).
Quantitative real-time quantitative PCR (qRT-PCR) assay The inguinal fat tissue was used to evaluate the mRNA levels of key genes involved in fat browning. Total RNA extraction, rst-chain cDNA synthesis, and quantitative PCR assays were performed as previously reported [59]. The primers used for each gene were listed in Additional le 2: Table S7.
Fluorescence-based GPR120 activation assay To evaluate the activating effect of 12-MMA on GPR120, we set up a uorescence-based GPR120 activation assay using STC Data are presented as means ± SEM. SPSS 17.0 and Prism 7 (Graphpad) were used for statistical analysis. The signi cance of group differences for normally distributed data was assessed by one-way ANOVA followed by Bonferroni post hoc tests. PERMANOVA (Permutational multivariate analysis of variance) was performed to evaluate the signi cance of group differences for PCoA and NMDS analysis. P* < 0.05 was considered statistically signi cant. All nal gures were assembled using Adobe Illustrator.

Results
Gut bacteria exhibit diverse and strain-speci c effects on intracellular lipid accumulation To get a comprehensive insight into the roles of gut microbes on lipid metabolism, we set up a cell-based high-throughput screening platform to assess the lipid-modulatory activity of the conditioned culture medium of the individual gut bacterium at the strain level (Fig. 1a). A total of 2250 strains were screened for their modulation on OA-elicited intracellular lipid accumulation. These gut bacterial strains contained 5 phyla, 12 classes, 48 genera, and 186 species (Additional le 2: Table S1 and S2), covering most of the dominant genera in humans except Faecalibacterium and Akkermansia [29]. After three cycles of oilred O staining-based screening, 388 strains showed a steady lipid-lowering effect in which 294 strains were from Firmicutes, 76 from Actinobacteria, 13 from Proteobacteria, and 5 from Bacteroidetes ( Fig. 1b; Additional le 2:  (Fig. 1c). When focused on individual species with ≥ 25 tested strains, Lactobacillus gasseri (43.88%), Bi dobacterium pseudolongum (28%) and Bi dobacterium catenulatum (25%) showed higher positive rates, while Enterococcus faecails (8.82%) and Escherichia coli (8%) contained less lipid-lowering strains ( Fig. 1d; Additional le 2: Table S4). These data provided several valuable reservoirs for seeking bene cial strains to maintain lipid homeostasis.
Interestingly, the positive rate rose sharply with screened strain numbers increasing. Although many species with <5 tested strains did not exhibit lipid-decreasing effect in these screenings, all species with ≥14 tested strains provided at least one positive strain regardless of their taxonomy ( Fig. 1e; Additional le 2: Table S2). Moreover, many species, such as B. adolescentis, L. gasseri, B. licheniformis contained lipid-increasing, lipid-lowering, and neutral strains at the same time (Table S3), suggesting an apparent strain-speci c modulation on host lipid accumulation. Among all the tested bacterial species, Blautia producta is of the most notice because all ve tested strains exhibited prominent lipid-decreasing activity with an e cacy comparable to feno brate, an effective lipid-lowering drug widely used in clinic ( Additional le 2: Table S2 and S3).

Bl. producta exerts potent anti-hyperlipidemic action
Blautia producta is a popular mucosal bacterium belonging to the Lachnospiraceae group [30]. Prior studies have demonstrated Bl. producta as a bene cial bacterial to resist pathogen invasion [31,32] and inhibit in ammation [33], but its anti-hyperlipidemic effect has not been investigated. We, therefore, systemically assessed the lipid-lowering effect and identi ed the active metabolite of Bl. producta following the ow chart in Fig. 2a. The cell-based experiments showed that the conditioned culture medium of Bl. producta dose-dependently decreased the lipid accumulation in HepG2 cells, and its 30% (v/v) conditioned culture medium displayed a lipid-lowering e ciency comparable to feno brate (Fig. 2b-e). To further con rm its hypolipidemic effect, we treated HFD-fed mice with live Bl. producta via continuous gavage of 10 9 CFU per animal per day for 4 weeks. Live Bl. producta signi cantly suppressed HFD-induced body weight gain (Fig. 2f-g) and fat deposition (Fig. 2h), and alleviated hyperlipidemia (Fig. 2i), thus leading to the improvement of liver steatosis (Fig. 2j-k). When prolonged the treatment period to 8 weeks, gavage of live Bl. producta displayed a similar pharmacological e ciency on the aforementioned indicators, including body weight, hyperlipidemia and liver steatosis (Additional le 1: Fig.S1). These results collectively demonstrated the prominent lipid-lowering effects of Bl. producta.
We further inspected the impact of live Bl. producta on host gut microbiota. Oral administration of Bl. producta did not change the diversity of the gut microbes but shifted its compositional structure (Fig. 3ad). At the genus level, the most striking change was that the dramatic decrease of Akkermansia resulted from HFD feeding was restored by Bl. producta (Fig. 3e). More speci cally, the relative abundance of Akkermansia muciniphila was signi cantly increased by Bl. producta (Fig. 3f). At the same time, bene cial mucosal bacteria such as Bi dobacterium animalis, Lactobacillus pentosus and Lactobacillus sakei, were elevated while opportunistic pathogens Desulfovibrio piger, Desulfovibrio sp. G11 were declined in response to Bl. producta administration (Fig.3f). Lefse analysis also revealed that Erysipelotrichales and Verrucomicrobiales were enriched after gavage of Bl. producta (Fig. 3g,h).
However, gavage of live Bl. producta did not increase its relative abundance in feces (Fig. 3f). To con rm this observation, we performed an extra dynamic detection by repeating the animal experiment of Bl. producta and collecting feces on Day 1, 3, 7, 14 and 28. The results displayed that oral administration of Bl. producta dynamically shifted the overall structure of gut microbiota (Additional le 1: Fig.S2a). Bl. producta showed the greatest impaction on gut microbiome at Day 1, then the in uence turned weaker but it still reached signi cance on the 28 th day after treatment (Additional le 1: Fig.S2a). The in vivo abundance of Bl. producta was also regulated in a dynamic manner. Gavage of exogenous Bl. producta signi cantly increased the fecal content of Bl. producta on Days 1 and 7, while this stimulation gradually disappeared afterward and the abundance of fecal Bl. producta on Day 28 in Bl. producta-treated mice was mildly less than that in control mice (Additional le 1: Fig.S2b), which was similar to our prior data.
To certify the uniform of these two experiments, we quanti ed the abundance of several key genera and species in the fecal samples collected on the 28 th day after treatment and compared them with previous results. Gavage of Bl. producta showed the same modulatory trend on Akkermansia spp., B. animalis, L. pentosus, L. sakei, D. piger and Desulfovibrio sp. G11 in both experiments (Fig.2e,f; Additional le 1: Fig.S2c,d). These ndings indicated that Bl. producta is a bene cial microorganism with poor colonizing capacity. Hence, we guessed Bl. producta might exert lipid-lowering action via secretion of active compounds rather than stimulating intestinal Bl. producta growth.

12-Methylmyristic acid (12-MMA) is an important active metabolite of Bl. producta
To identify the active metabolites that mediate the bene cial effect of Bl. producta, pan-genomic analysis and UPLC-MS/MS-based untargeted metabolomics were conducted. The pan-genomic analysis included 5 new genomes of Bl. producta by de novo sequencing, namely I2DA, ID8A, ID9B, I24C and I31A, and 7 reference genomes from public databases. The rank of lipid-lowering e ciency of 5 Bl. producta strains was I2DA>ID8A>ID9B>I24C>I31A. The 5 lipid-lowering strains clustered in two neighbor branches (Additional le 1: Fig.S3a) and genes involved in lipid metabolism were uniformly enriched in these strains (Additional le 1: Fig.S3b), implying that active lipid metabolism might be a unique feature the effective Bl. producta strains.
We further performed untargeted metabolomics by using UPLC-MS/MS method. To nd the preferentially produced metabolites by Bl. producta, 500 typical strains were selected from 2250 bacteria tested in our study, which covered all the genera screened in this study. The 500 strains were cultured under the same condition for lipid-lowering activity assay, and their monocultures were pooled with equal weight as an entirety mixture. The signal strength of each recognized metabolite in Bl. producta was compared with that in the entirety mixture, to assess the priority of Bl. producta in producing each metabolite. A total of 614 metabolites were identi ed and 13 metabolites were highly (>2.8 fold) generated by Bl. producta over the entirety (Fig. 4a; Additional le 2: Table S5). 12-Methylmyristic acid (12-MMA) was the top priority product with a yield of 26.4 times over the whole. Therefore, 12-MMA was the most preferentially produced metabolite by Bl. product, although it is unnecessary to be the most abundant metabolite in this strain. Generally, an effective approach to identify the active metabolite of a particular species is to nd different strains with distinct pharmacological e ciencies and correlate the abundance of their metabolite with the differential effects. Because all tested Bl. producta strains were highly effective, it is impossible to identify key active metabolites following this strategy. We, therefore, modi ed the method by selecting another gut bacterial species harboring both lipid-decreasing and increasing strains as a reference. In our screening, Bacilli licheniformis virtually met the aforementioned requirements and the great majority of strains exhibited metabolite pro les highly resembling Bl. producta. As revealed by PLS-DA and Hierarchical Clustering, the effective (lipid-decreasing) and ineffective (lipid-increasing) strains were markedly separated (Fig. 4b-c). Volcano plot showed that several metabolites differently distributed among bacteria, in which 12-MMA and anandamide were enriched in effective strains, whilst pyrrolidonecarboxylic acid was higher in ineffective ones (Fig. 54). Furthermore, random forest analysis clearly exhibited that 12-MMA was the most critical metabolite to distinguish the effective bacteria from the ineffective ones ( Fig. 4e; Additional le 2: Table S6).

12-MMA effectively alleviates hyperlipidemia
In light of 12-MMA as a critical metabolite candidate of Bl. producta, we treated HFD-fed mice with 12-MMA for four weeks to verify its anti-hyperlipidemic effect. Oral administration of 12-MMA signi cantly inhibited HFD-induced body weight gain and fat deposit ( Fig. 5a-b). High serum lipids levels were remarkably reduced by 12-MMA (Fig. 5c). Consequently, liver steatosis was also alleviated after 12-MMA treatment, as revealed by lipid quanti cation and oil-red O staining (Fig. 5d-e). These results demonstrated that 12-MMA is an active metabolite of Bl. producta to alleviate hyperlipidemia.
Besides the bene cial effect on hyperlipidemia, supplementation of 12-MMA also improved oral glucose tolerance (Fig. 5f), indicating the bene cial roles in both lipid and glucose homeostasis. The 12-MMAtreated mice displayed multiple multilocular adipocytes in inguinal white fat (iWAT), representing a typical characteristic of WAT browning (Fig. 5g). Consistently, the expression of uncoupling protein 1 (Ucp1) and the transcription levels of crucial thermogenic genes in the iWAT were signi cantly upregulated (Fig. 5h-i), indicating a stimulatory effect of 12-MMA on WAT browning. Moreover, oral administration of 12-MMA signi cantly increased the serum level of glucagon-like peptide-1 (GLP-1), a popular downstream peptide of GPR120 (Fig. 5j). Recent reports showed that GPR120 activation contributes to both glucose consumption and WAT browning [27]. We, therefore, evaluated the in uence of 12-MMA on GPR120 activity by a uorescence-based assay [45] in STC-1 intestine endocrine cells.
Both 12-MMA and its counterpart myristic acid (MA) drastically strengthened GPR120 activity in a dosedependent manner (Fig. 5k), the secretion of GLP-1 was enhanced as well (Fig. 5l). These ndings provided a potential mechanism by which 12-MMA exerts therapeutic e cacy against hyperlipidemia.

Discussion
Over the past decade, it has been generally accepted that the gut microbiota plays vital role in regulating host lipid metabolism [34]. However, due to the lack of species/strain-level investigations, the exact effect of individual species on lipid homeostasis remains poorly understood. Here we, for the rst time, displayed a considerable functional pro le regarding lipid-modulatory capacities of human gut microbiota and identi ed a key active microbial metabolite, 12-MMA, which triggers GPR120 activation and thereby ameliorates hyperlipidemia and obesity.
Currently, intensive studies have illustrated the bene cial effects of the gut microbiome, providing an essential rationale to develop novel therapeutic approaches based on the vast bacteria reservoir. Several groups have performed screening investigations on gut microbes to discover more functional connections between the gut microbiome and hosts, such as carbohydrate degradation [34] and antibiotic resistance [35]. In the present study, we set up a high-throughput screening platform, aiming to seek the functional human gut bacteria with lipid-decreasing effects. A total of 2250 strains were screened, covering 186 species and representing an essential part of human gut microbiota. These . In our study, more than 60.99% of positive Lactobacillus strains were from L. gasseri, a well-known species with great potential to help with weight loss in humans and rodents [40,41]. Another notable species is Lysinibacillus sphaericus, 16.83% (17/101) of its tested strains stably inhibited lipid accumulation though the biological activity of this species is rarely reported. Hence, L. gasseri and L. sphaericus could be valuable sources to deeply mining for hypolipidemic strains. Moreover, we conclude from the screening analysis that strains belonging to the same species usually function diversely, obviously presenting strain-speci c regulatory roles on host lipid metabolism. Therefore, close attention should be paid to this when future researches are arranged to explore active functional bacteria or probiotics.
One important nding of this work is the identi cation of Bl. producta, which is a mucosal bacterium belonging to the Lachnospiraceae group [42]. In our screening, Bl. producta is of the most attraction because all tested strains robustly decreased lipid accumulation with comparable e cacy to the positive drug feno brate. Analogously, gavage of live Bl. producta isolated from human feces was effective to suppress HFD-induced body weight gain, decrease serum levels of lipids, and ameliorate liver steatosis in mice, exhibiting potent e cacy against hyperlipidemia. Prior studies have revealed that Bl. producta plays an essential role in the colonization resistance to the invasion of Vancomycin Resistant Enterococci (VRE) [43,44] and exhibits anti-in ammatory effect in HT-29 intestinal epithelial cells [45]. Besides, recent studies revealed that Blautia spp. are closely associated with the improvement in glucose and lipid homeostasis by drugs, such as metformin [46], resveratrol [47] and erythrocyte n-6 polyunsaturated fatty acids [48], yet the speci c species are still not determined. Our work not only demonstrates the antihyperlipidemic effect of Bl. producta in the mouse model, paving a path for the development of Bl. producta as a novel probiotics, but also releases an inspiring clue that Bl. producta might be a candidate to mediate the lipid-modulatory effects of some drugs.
Another interesting nding of this study is that we identi ed 12-MMA as a key active metabolite of Bl. producta and veri ed its marked e cacy against hyperlipidemia in mice. Apart from SCFAs, increasing evidence has documented the contributing roles of LCFAs in maintaining the balance of energy metabolism [49,50], 12-MMA found in our work again highlights the importance of LCFAs, together with another LCFA, myristoleic acid (MA), which reduces obesity through brown fat activation [51]. Previous studies have shown that some ai-FAs can be major constituents in bacteria and 12-MMA, also known as 12methyltetradecanoic acid, accounting for 25-30% of the total fatty acids in Bacillus megaterium [52,53]. Thus, 12-MMA might be indispensable for the growth of some species, and it could be constantly produced by the bacterial both in vitro or in vivo, although this needs further certi cation. This provides an explanation for the observation that 12-MMA was identi ed from the pure culture, but it was as effective as the whole bacterial strain in the animal experiment. Overall, the identi cation of 12-MMA as an active metabolite to improve hyperlipidemia widens our knowledge of ai-FAs. However, we currently cannot exclude the potential that other metabolites of Bl. producta also possess potent antihyperlipidemic effect, which is worth further investigation in the future.
Besides hyperlipidemia amelioration, 12-MMA also dose-dependently stimulated GPR120 and improved oral glucose tolerance. It has been well-acknowledged that GPR120 activation can effectively provoke brown fat activation and WAT browning [27,54]. Correspondingly, oral administration of 12-MMA remarkably up-regulated key thermogenic factors and elicited many multilocular adipocytes in iWAT, a typical phenotype of WAT browning. In light of the similar characteristics of anteiso-and unsaturated-FAs counterparts (12-MMA/MA), we speculate that MA may activate GPR120 as well, consequently resulting in BAT activation and improved obesity [51]. Of note, G-protein coupled receptors (GPCRs) are, to date quite successful therapeutic targets for various diseases [55,56], so 12-MMA might be a potential candidate against hyperlipidemia by targeting GPR120, yet massive further investigation is awaited to verify this conjecture.

Conclusions
In the present study, we have rst displayed a large-scale lipid-modulatory pro le of gut microbes at the species/strain level and demonstrated Bl. producta as a novel probiotic to exert prominent antihyperlipidemic effect via 12-MMA-GPR120-WAT browning axis to improve hyperlipidemia and obesity. This work opens up new avenues to systematically understand the modulatory roles of gut microbiota and provides possible therapeutic exploitation based on the gut microbiome and derived metabolites.  study than that shown on the x-axis. The percentage in red indicates the rate of species providing one or more positive strains.

Figure 2
Blautia producta ameliorates HFD-induce hyperlipidemia. a Graphical outline for the lipid-lowering effect assessment and identi cation of the potential active metabolite of Bl. producta. Bl. producta strain with the best e cacy in the cell-based screening was re-assessed for the lipid-lowering effect in HepG2 cells and signi cantly altered species by Bl. producta. g Cladogram of gut microbiota community. n = 8 for each group. *P < 0.05, **P < 0.01, ***P < 0.001.
producta. a The top 13 metabolites that are markedly enriched in Bl. producta. b PLS-DA plot. c Clustering analysis based on Euclidean distance. d Volcano plot and the abundance of key metabolites in effective/ineffective stains. e The top 15 metabolites that contribute to the discrimination of effective/ineffective stains. *P < 0.05. activation assay and GLP-1 secretion by STC-1 intestine endocrine cells. 12-MMA was suspended in solvent (1% tween 80 + 0.5% carboxymethyl cellulose sodium (CMC-Na)) and the HFD group was treated with an equal volume of solvent. n = 6 for each cell group.*P < 0.05, **P < 0.01.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Supplementalmaterials.pdf