Prevotella Copri Increases fat Accumulation in Pigs fed by Formula Diets

Background: Excessive fat accumulation of pigs is undesirable. It severely affects economic return of modern pig industry. Studies in humans and mice have examined the role of the gut microbiome in host energy metabolism. Commercial Duroc pigs are often fed formula diets with high energy and protein. Whether and how the gut microbiome under this type of diets regulates swine fat accumulation is largely unknown. Results: In the present study, we systematically investigated the correlation of gut microbiome with pig lean meat percentage (LMP) in a total of 698 commercial Duroc pigs. We demonstrate that the gut microbiome of fat pigs was dominated by P. copri which occupied 23.53% and 5.76% of relative abundance in average in the discovery and validation cohort, respectively. High abundance of P. copri in the gut resulted in a higher abundance of serum metabolites associated with chronic inammation, e.g., branched chain amino acids, aromatic amino acids, the metabolites of arachidonic acid metabolism and lipopolysaccharides. Host intestinal barrier permeability and chronic inammation response were increased. A gavage experiment using germ-free mice conrmed that the P. copri isolated from experimental pigs was a causal species increasing host fat accumulation. Host colon, adipose tissue, and muscle transcriptomes indicated that P. copri colonization signicantly upregulated the expression of the genes related to immune and inammatory responses, lipogenesis, and fat accumulation, but attenuated the genes associated with lipolysis, lipid transport, and muscle growth. Conclusions: Taken together, we identied and conrmed that P. copri in the gut microbial communities of pigs fed by commercial formula diets results in the signicantly increased fat deposition of pigs, and proposed a possible mechanism of P. copri affecting fat accumulation. The results provided fundamental knowledges for reducing pig fat accumulation through regulating the gut microbial composition in pig industry. Differential metabolites LEfse: Linear discriminant analysis effect size; LPS: Lipopolysaccharide; KEGG: Kyoto encyclopedia of genes and genomes; OTU: Operational taxonomic unit; PULs: Polysaccharide utilization loci; SparCC: Sparse Correlations for Compositional data; TNF-α: Tumour necrosis factor alpha.

Despite a decade of research establishing a strong association between the gut microbiota and obesity, con icting ndings provided by several studies have challenged this view [8,9]. There are few bacterial strains that have been isolated and con rmed as having causal roles in obesity [10][11][12]. Moreover, the underlying mechanism has not yet been clearly established [13], although several studies have indicated that obesity is associated with a low-grade systemic and chronic in ammatory condition [14,15]. In the modern pig industry, to obtain rapid body weight gain, commercial formula diets with high concentrations of proteins and energy have often been provided to production pig herds. Whether and how the gut microbiome regulates swine fat accumulation (e.g., LMP) is also largely unknown.
In the current study, to investigate the relationship of gut microbial species with fat accumulation of pigs, we performed the association study in 698 commercial Duroc pigs fed by commercial formula feeds (corn-soybean formula feeds containing 2,960-3,023 kcal/kg of digestible energy and 15-17% of protein). We identi ed P. copri as a main bacterial species increasing host fat accumulation (decreasing LMP) in pigs. To further con rm the causality of P. copri in host fat accumulation, colonization experiment of P. copri was also carried out in germ-free mice. High abundance of gut P. copri increases serum levels of lipopolysaccharide (LPS), branch chained amino acids (BCAAs), aromatic amino acids (AAAs), and the metabolites of arachidonic acid metabolism, thereby increasing host intestinal barrier permeability and chronic in ammation response. The expression levels of genes related to lipid metabolism, transport, and localization in adipose and muscle tissues were signi cantly changed.
However, butyrate-producing bacteria can relieve low-grade chronic in ammation and result in increased LMP.

Results
Identifying a signi cant association of P. copri with fat accumulation of pigs Excessive fat accumulation signi cantly decreases pig lean meat percentage (LMP). Therefore, in this study, we used the LMP as an index to assess the role of the gut microbiome in porcine fatness. We recorded the LMP of 698 commercial Duroc pigs raised in two farms, with the samples comprising 550 pigs (309 males and 241 females) from two farms (280 from Shahu and 270 from Jiangying) as the discovery cohort and 148 pigs (100 males and 48 females) from the Jiangying farm as the validated cohort (Methods). The phenotypic values generally tted the normal distribution (Additional le 1: Fig.   S1). All 698 pigs had fecal samples collected at the age of 160 days, and we performed hypervariable region sequencing of the 16S rRNA gene (V4 region for the test cohort and V3-V4 regions for the validated cohort). The descriptions of the sequencing procedures are summarized in Additional le 2: Table S1. We rst analyzed the association of enterotypes and co-abundance groups (CAGs) of OTUs with the LMP in the discovery cohort. All samples were clustered into two enterotype-like groups that were dominated by either Prevotella or Treponema, and the pigs with the Prevotella enterotype had signi cantly lower LMP (Additional le 1: Fig. S2). At the CAG level, all 1,159 OTUs that passed quality control were used to construct a co-abundance network. These OTUs were clustered into 12 coabundance groups (CAGs) based on SparCC correlation coe cients ( Fig. 1b and Additional le 1: Fig.  S3). The CAG3 containing the OTUs mostly annotated to Prevotella, especially P. copri, were negatively correlated with the LMP, while the CAG8 that contained the OTUs annotated to F. prausnitzii and R.
avefaciens showed strongly positive correlations with the LMP, suggesting the central roles of these CAGs in the functional guilds of gut microbiota related to the LMP. (Fig. 1b).
We then performed a bacteria-wide association study with a two-part model to identify the bacterial taxa signi cantly associated with the LMP in the discovery cohort. A total of 166 LMP-associated OTUs were identi ed at FDR < 0.01, including 82 OTUs positively associated with the LMP and 84 OTUs showing negative associations with the LMP. Those positively associated OTUs mostly belonged to the order Clostridiales, for example F. prausnitzii, Lachnospiraceae, and Ruminococcaceae, while the negatively associated OTUs were mainly aligned to Prevotella (40/84). In particular, 18 P. copri OTUs showed the strongest negative associations with the LMP (Fig. 1a and Additional le 2: Table S2).
The results of the enterotype analysis were well repeated in the validation cohort. The P. copri OTUs were clustered into the CAG8 that was negatively associated with the LMP, and these P. copri OTUs were the hub nodes in the co-abundance network (Additional le 1: Fig. S4). A total of 11 LMP-associated OTUs were identi ed in the validation cohort. Two out of the ve OTUs positively associated with the LMP belonged to Christensenellaceae, while the OTUs showing the most signi cant negative association with the LMP were annotated to Prevotella and P. copri (Additional le 2: Table S3). These results further suggest the signi cant association of P. copri with fat accumulation of pigs.
We further performed shotgun metagenomic sequencing of 16 fecal samples that were contained in the validation cohort, including eight samples with the highest LMP values and another eight samples with the lowest LMP (Additional le 1: Fig. S5a). The metagenomic sequencing data are summarized in Additional le 2: Table S4. Consistent with the previous ndings in humans [12,16], the fat pigs had a signi cantly lower number of genes and α-diversity (Shannon index) in the gut microbiome than their lean counterparts (Additional le 1: Fig. S5b,c). We identi ed 40 species responsible for the LMP by a linear discriminant analysis of effect size (LEfSe). The members from Prevotella predominated the bacterial species enriched in fat pigs (13/20). In particular, P. copri showed the strongest negative association with the LMP and was the hub species among the bacterial species decreasing the LMP (Additional le 1: Fig.   S5d). A total of 20 species were enriched in lean pigs, most of which were butyrate-producing bacteria from Treponema and Clostridiales, e.g. L. reuteri and B. longum (Fig. 1c). To extend the metagenomic sequencing data, we integrated the metagenomic sequencing data of 20 fecal samples from the discovery cohort that were chosen based on the phenotypic values of feed e ciency in our previous study [17]. Similar to the rst batch 16 samples' results, the integrated 36 samples' result showed that four species of Prevotella including P. copri were signi cantly associated with the decreased LMP, while the species from Treponema and Clostridiales had higher abundance in lean pigs (Additional le 1: Fig.  S6a).
Diet effect on the abundance of gut P. copri identi ed in this study Earlier researches have shown that different habitual diets can in uence the genomic function and strain representation of intestinal P. copri [18,19], we further carried out a comparison on the abundance of P. copri in the guts of pigs raised under different feeding patterns. All OTUs belonging to P. copri dominated the gut microbiota of the discovery and validated pigs with averages of 23.53% (1.01-62.23%) and 5.76% (0.04%-38.29%) relative abundance. We further used the metagenomic sequencing data from wild boars (n = 6; free-living, high ber diets), Tibetan pigs (n = 13; semi-grazing, high ber diets), and Duroc pigs described above (n = 20 and 16; formula diets with high energy and protein). A signi cantly higher abundance of P. copri was observed in both Duroc populations (7544. 81  CAZymes having signi cantly higher abundance in the gut microbiome of obese individuals are mainly involved in the metabolism of rhamnose and glucan, and the biosynthesis of lipopolysaccharide (e.g., GH28, PL11, GH22, PL10 and GT4) (Fig. 2a). Correlation analysis between the LMP-associated OTUs and CAZymes indicated the contribution of the LMP-associated bacteria to the changes in CAZymes (Fig. 2b).
The LMP-associated KEGG pathways are shown in Fig. 2c and Additional le 1: Fig. S6b. We identi ed 17 KEGG pathways having signi cantly higher abundance in the gut microbiomes of fat pigs, including lipopolysaccharide biosynthesis and arachidonic acid metabolism involved in mediating in ammatory reactions [11,20]; FoxO signaling pathway, insulin resistance, BCAA (valine, leucine, and isoleucine) biosynthesis, and metabolism of aromatic amino acids (tyrosine and phenylalanine, AAA) related to obesity and insulin resistance [12,15,21,22] along with two-component system, bacterial chemotaxis, agellar assembly, and carbohydrate digestion and absorption associated with increased capacity for energy harvest from bacteria [6,23]. The pathway bacterial invasion of epithelial cells that should impair gut barrier integrity was also highly enriched in fat pigs compared with lean pigs (Fig. 2c). All these pathways were positively correlated with the fatness-associated bacterial species, especially with P. copri (Fig. 2d), suggesting that the bacterial species from fat pigs could produce more factors related to in ammatory reactions, obesity, and insulin resistance, impaired host gut barrier integrity, and the capacity for energy harvesting.
We further isolated and cultured P. copri in vitro from the fecal samples of the experimental pigs with low LMP values. Whole-genome sequencing was performed using a Nanopore platform (Methods). The fulllength of P. copri genome comprised 3.44 Mb containing 3,039 coding genes (CDS) (Additional le 1: Fig.  S7). We rst constructed a phylogenetic tree based on the genome sequences of P. copri isolates, including 60 isolates from westernized human populations, 51 isolates from non-westernized human populations, and one pig isolate from this study. The P. copri isolated from pigs in this study was clearly located in the clade from the westernized Chinese population (Fig. 2e). A total of 24 polysaccharide utilization loci (PULs) were then identi ed in this P. copri isolate. More than 10 PULs had higher prevalence in westernized populations (Additional le 1: Fig. S8). The genes involving in arachidonic acid metabolism, BCAA biosynthesis, AAA biosynthesis and metabolism, the TCA cycle, and protein export were found in the genome of this P. copri isolate, and the abundances of these genes in the tested samples were determined by combining the metagenomic sequencing data. Consistent with the LMPassociated functional capacities identi ed only by the metagenomic sequencing data, the gut microbiome of fat pigs had signi cantly higher abundances of the P. copri genes involved in arachidonic acid metabolism, BCAA biosynthesis, AAA biosynthesis and metabolism, and insulin resistance, but had lower abundances of the genes participating in the TCA cycle and protein export compared with lean pigs (Fig. 2f). Considering the high abundance of gut P. copri in fat pigs, P. copri was a main contributor to the shifts in metagenomic functional capacity related to the LMP.
Conversely, 15 KEGG pathways had signi cantly higher abundance in lean pigs (Fig. 2c), including butanoate metabolism, citrate cycle (the TCA cycle), metabolism of cofactors and vitamins, lysine degradation, cysteine and methionine metabolism, and arginine and proline metabolism. All these pathways were positively associated with multiple high LMP-associated bacterial species (Fig. 2d).
The changes of serum metabolome in fat pigs and the correlation with shifts in gut microbiome We rst measured and compared the concentration of serum LPS using an enzyme-linked immunosorbent assay (ELISA) between lean pigs (n = 8) and fat individuals (n = 8). Consistent with the higher abundance of the functional capacity of LPS biosynthesis in the gut microbiome, fat pigs had signi cantly higher serum LPS concentrations compared to their lean counterparts (P < 0.005; Fig. 3a). We then determined non-targeted metabolome pro les of 38 serum samples randomly collected from the validated cohort. We identi ed 80 metabolite features showing signi cant association with the LMP by Spearman rank correlation analysis (FDR < 0.05) (Additional le 2: Table S5). These LMP-associated metabolites were clustered into 23 KEGG pathways covering most of the LMP-associated functional pathways of the gut microbiome (Additional le 1: Fig. S9). We next focused on some interesting LMPassociated metabolites based on the LMP-associated functional capacities of the gut microbiome. Serum concentrations of BCAA, AAA and their related metabolites were considerably higher in fat pigs than in lean individuals ( Fig. 3b and Additional le 1: Table S5). Furthermore, compared to their lean counterparts, fat pigs had signi cantly higher serum concentrations of the metabolites related to in ammatory reaction and metabolic syndrome such as 3-methyl-2-oxovaleric acid, an intermediate of BCAA metabolism that can induce the accumulation of BCAAs [24], L-rhamnose (an important component of lipopolysaccharides), and the metabolites of arachidonic acid metabolism (5-HETE, 9-HETE, leukotrienes, and prostaglandins) ( Fig. 3b and Additional le 2: Table S5). These were notably consistent with the proposed function capacity of the gut microbiome for BCAA biosynthesis and the metabolism of AAAs and arachidonic acid in fat pigs (Fig. 2b).
Compared to fat pigs, lean pigs had higher concentrations of the metabolites previously reported to reduce pig fat accumulation and increase lean muscle mass in serum, e.g., creatine [25] and the metabolites of betaine [26] (L-histidine trimethyl betaine and proline betaine) and anti-in ammatory factors (Lipoxin A4) [27] (Fig. 3b and Additional le 2: Table S5). Catecholamines, including dopamine, Nacetyl dopamine, and L-dopa, which can reduce lipid accumulation in adipose tissue by increasing lipolysis, thereby decreasing lipogenesis and promoting free fatty acid (FFA) transportation [28], also exhibited higher abundances in the serum of lean pigs (Additional le 2: Table S5 and Fig. 3b). Serum concentrations of vitamins (vitamin K1, K2, D3, biotin and pantothenic acid) were also signi cantly higher in lean pigs than in fat individuals (Additional le 2: Table S5).
We further evaluated the contribution of the LMP-associated bacteria to the shifts in serum metabolites at the OTU level in all 38 samples with metabolome pro les (Additional le 1: Fig. S10) and at the species level in 16 samples from the metagenomic sequencing (Fig. 3c). P. copri was positively correlated with nearly all fatness-associated metabolites mentioned above, but negatively associated with the leanassociated metabolites (Fig. 3d). The other LMP-associated bacteria also contributed to the LMPassociated metabolites to different extents. For example, both P. copri and two Bacteroides spp. were signi cantly associated with serum BCAA concentration, but serum BCAA was largely driven by P. copri (Additional le 1: Fig. S11). These results indicated a signi cant contribution of the LMP-associated bacteria to the shifts in host serum metabolites related to fat accumulation.
Taken together, chronic in ammation-associated metabolites, e.g., BCAA, AAA, and metabolites of arachidonic acid, had higher abundances in fat pigs, while the metabolites associated with antiin ammation, lipid metabolism, and energy expenditure were enriched in the serum of lean pigs. The P. copri and other LMP-associated bacteria responded to the shifts in serum metabolites in fat pigs.

Increased host intestinal barrier permeability and chronic in ammatory reaction in fat pigs
Given the increased concentrations of serum LPS and the metabolites related to in ammatory reactions in fat pigs, in order to examine the host intestinal barrier permeability and chronic in ammation response, we determined the serum levels of biomarkers zonulin and FABP2 [29] and the pro-in ammatory cytokines tumor necrosis factor α (TNF-α), interleukin-1β (IL-1β), interleukin-6 (IL-6), and interferon-γ (IFNγ). As expected, compared to high lean meat pigs (n = 8), fat pigs (n = 8) had higher serum concentrations of zonulin (P < 0.005) and FABP2 (P = 0.083) (Fig. 4a, b), suggesting an increased intestinal barrier permeability in fat pigs. Moreover, fat pigs also had higher serum concentrations of TNF-α, IL-1β, IL-6 and IFN-γ (Fig. 4c, d, e, f), indicating host chronic in ammation response. Taken together, the results of metabolome analysis suggest that a high abundance of gut P. copri induces host intestinal barrier permeability and promotes chronic in ammatory response through production of LPS, BCAA, AAA, and the metabolites of arachidonic acid, and this increases fat accumulation in pigs.
Gavage in germ-free mice con rmed the causal role of P. copri We next evaluated the possible causal relationship between P. copri isolated from experimental pigs and host fat accumulation using gavage with live P. copri in germ-free mice. A qPCR analysis of fecal DNA at the end of a one-month gavage experiment con rmed the successful colonization of P. copri in the guts of each treated mouse (Fig. 5a). Phenotype measurements found signi cantly increased body fat percentage (P < 0.01) and epididymis fat percentage (P < 0.05) in mice raised on a normal chow diet and treated by gavage with P. copri (Fig. 5b). P. copri gavage inducing fat accumulation became more severe in mice fed a high-fat diet (HFD) (P < 0.005) (Fig. 5b). We further investigated the diet effect on P. copri colonization and host fat accumulation with 18 germ-free mice (C57BL6) that were divided into three groups fed standard chow, a high-fat diet, and a high-ber diet (Methods). A signi cantly higher abundance of colonized P. copri was identi ed in gavage-treated mice fed high-fat diets (P < 0.01), but there was no signi cant difference between gavage-treated mice fed standard chow and those fed highber diets (Additional le 1: Fig. S12a). Compared to the mice fed standard chow, the mice fed a high-fat diet had signi cantly higher percentages of both body fat and epididymis fat (P < 0.005). However, the mice fed high ber diet showed signi cantly less fat accumulation (P < 0.05) (Fig. 5c).
Gavage with the bacteria signi cantly increased serum concentration of LPS in mice (P < 0.005), which was enhanced by the HFD (Fig. 5d, Additional le 1: Fig. S12b). P. copri colonization also caused increased serum concentrations of the intestinal barrier permeability biomarkers FABP2 (P < 0.01) and zonulin (P < 0.05). Feeding a high-fat diet reinforced intestinal barrier permeability (Fig. 5e, Additional le 1: Fig. S12c). Notably, serum concentrations of pro-in ammatory cytokines (IL-6, IL-1β, TNF-α, and IFN-γ) were signi cantly higher in P. copri colonized mice than in PBS gavage control mice. This was enhanced by feeding a high-fat diet to P. copri colonized mice (P < 0.01) (Fig. 5f, Additional le 1: Fig. S12d). There were no signi cant differences in the concentrations of LPS, biomarkers zonulin and FABP2, or proin ammatory cytokines between gavage mice fed standard chow and those fed a high-ber diet, except for IFN-γ which showed lower abundance in high-ber diet mice (P < 0.05; Additional le 1: Fig. S12). We identi ed a total of 222 serum metabolites showing differential abundances between controls and P. copri colonized mice, and 215 differential metabolite features between controls and P. copri colonized mice fed with HFD (Additional le 2: Table S6). The differential metabolites between controls and P. copri colonized mice were enriched to the pathways highly similar to those identi ed between lean and fat pigs (Additional le 1: Fig. S13). For example, P. copri colonization (in both normal chow and HFD) increased the richness of the pathways related to BCAA biosynthesis, AAA metabolism, and arachidonic acid metabolism. The pathways of biotin metabolism, butanoate metabolism, and pantothenate were in low abundance in P. copri colonized mice (Fig. 5g). Furthermore, as observed in pigs, the metabolites involved in in ammatory reactions and metabolic syndrome such as BCAA, AAA, leukotrienes, prostaglandins, HETE, and L-rhamnose had higher abundances in mice treated by gavage with P. copri, while the metabolites associated with lean muscle mass, energy expenditure and reduced lipid accumulation (e.g., betaine, vitamins, and dopamine) were lower in the serum of colonized mice ( Fig. 5h and Additional le 2: Table S6). Taken together, gavage with P. copri in germ-free mice generated the increased intestinal barrier permeability and chronic in ammatory reaction.
Transcriptome analysis of colon, adipose, and muscle tissues elucidated the mechanism of gut microbiome affecting host fat accumulation To elucidate the molecular mechanism of P. copri in uencing host body fat percentage, RNA-sequencing analysis was performed on the tissues of colon, white adipose, and muscle harvested from control and colonized mice on a normal chow diet. We identi ed a total of 166 differentially expressed genes (DEGs) in colon tissue between control and P. copri colonized mice (Additional le 2: Table S7). These DEGs were enriched into 54 functional terms of biological process (BP) (Fig. 6a). Most of these 54 BP terms were related to immune and in ammatory response, metabolic process, cell communication, and intracellular signal transduction. Here, we identi ed some interesting genes having higher mRNA levels in colon tissue of P. copri colonized mice, including genes involved in immune and in ammatory responses (e.g., Ccl2, Ccl24, Ccl3, Ccl4, Ccl7, Il1b, Il6ra, Ilf2, Tlr2, Tlr3, Tlr5, and six genes from the immunoglobulin superfamily), and genes involved in fat accumulation and obesity (such as Fabp2 and Ins2) ( Fig. 6b and Additional le 2: Table S7). This was consistent with the chronic in ammatory reaction and the increased intestinal permeability in both fat pigs and P. copri colonized mice. In white adipose tissue, 296 DEGs were identi ed between control and P. copri colonized mice (Additional le 2: Table S7). The 296 DEGs were signi cantly enriched in 12 BP terms related to lipid transport and localization and metabolic process, and ve KEGG pathways including fatty acid metabolism and the PPAR signaling pathway (Fig. 6c). We further identi ed some interesting DEGs involved in lipogenesis (Fabp9, Scd1, Scd2, and Scd3) and in ammatory response (Il13ra2) that showed higher expression levels in P. copri colonized mice than in control mice. However, several genes related to lipolysis and lipid transport (Abca1, Apoc1, Apoe, Pparg, Cpt2, and Adrb3) had lower expression levels in P. copri colonized mice (Fig. 6d). We also identi ed 335 DEGs in muscle between control and P. copri colonized mice. These DEGs were enriched in the BP terms associated with biosynthetic process, metabolic process, and catabolic process of protein and other macromolecules (Fig. 6e). Interestingly, several genes related to lipogenesis and lipid deposition (e.g., Adipoq, Adipor2, Apobr, Dgat2, Fabp3, Scd2, Pck1, and Ppargc1a) and in ammatory response (Il11ra2, Il6ra, and Ilf2) had higher expression levels in P. copri colonized mice, whereas Igf2r and Igfbp7 associated with skeletal muscle growth were attenuated in response to P. copri colonization (Fig. 6f). We next evaluated the correlations between DEGs (Fig. 6b, d, and f) and differential serum metabolites. The DEGs in colon and muscle tissue were associated with most of the differential serum metabolites (Fig. 6g). For example, the expression level of Igf2r in muscle was negatively associated with chronic in ammatory-associated metabolites but was positively correlated with anti-in ammatory metabolites. However, the DEGs in colon tissue, which were mostly associated with immune response and showed negative correlations with anti-in ammation metabolites, were positively correlated with chronic in ammatory-associated metabolites (Fig. 6g). Taken together, P. copri colonization increased host fat accumulation through upregulating the expression levels of the genes related to immune and in ammatory response and the genes associated with lipogenesis and fat accumulation while downregulating the expression levels of the genes associated with lipolysis, lipid transport, and muscle growth.

Discussion
Accumulated evidences have been indicating that gut microbiota may contribute to host fat accumulation. In this study, we have identi ed P. copri from the gut microbiome of pigs fed with formula diets as a hub bacterial species increasing pig fat accumulation. P. copri is a complex comprising several distinct clades [19]. It has been both positively and negatively associated with host health depending on habitual diets. For example, P. copri colonization in mice fed a ber-rich diet improved glucose homeostasis via intestinal gluconeogenesis [30] , [31]. Prevotella abundance or the Prevotella-to-Bacteroides ratio can predict body weight and fat loss success in overweight participants consuming a whole-grain or high-ber diet [32,33]. However, a clinical trial report showed that a higher relative abundance of Prevotellaceae and Veillonellaceae along with increased gut permeability elevated circulating succinate levels associated with obesity and impaired glucose metabolism [34]. P. copri is associated with human insulin resistance and aggravating glucose intolerance [15]. Different habitual diets lead to distinct genetic and functional traits of human intestinal P. copri strains [18], and human intestinal P. copri isolates show distinct polysaccharide utilization pro les [35]. In the modern pig industry, to exploit the maximum of pig growth potential, commercial formula diets that are processed and contain high amounts of digestible energy and protein are provided to pig herds. These diets have selected and shaped gut P. copri of commercial pigs. Indeed, we found low abundances of P. copri in wild boars and Tibetan pigs fed high-ber diets. P. copri can increase serum BCAA, LPS, and arachidonic acid metabolites levels and induce chronic in ammation response. The same formula feed was provided to the pigs in the same farm, but we observed signi cant variation in gut P. copri abundance (0.04-62.23%). This could have been caused by maternal effects [36] or/and the diets [37] before performance measurement (from birth to 30 kg of body weight). Moreover, host genetics may be another reason causing this signi cant variation of P. copri abundance in the gut.
Serum concentrations of LPS, BCAA, AAA, and the metabolites related to arachidonic acid metabolism were signi cantly higher in fat pigs than in lean pigs. Previous reports in mice have also indicated the role of gut bacterial LPS in obesity [38,39]. Several studies in humans indicated that P. copri largely drives the increase of the microbial potential for BCAA biosynthesis [15], and these studies have suggested a causative role for serum level of BCAAs or their breakdown products in type 2 diabetes [40], obesity [41], and insulin resistance [15]. Arachidonic acid is the substrate for the synthesis of a range of biologically active compounds, including prostaglandins and leukotrienes [20]. These compounds can act as mediators and regulators of in ammatory cytokine production and immune function [20]. As for AAAs, increased circulating concentration of AAAs has been reported to be associated with obesity and insulin resistance in humans [40,42,43]. The correlation between the LMP-associated bacterial taxa and serum metabolites suggested that the gut microbiota, especially P. copri, drives elevated levels of serum BCAA, in ammatory response via these metabolites, thereby resulting in host fat accumulation.
In contrast, the bacterial species that have been reported to have anti-in ammatory effects in humans were signi cantly enriched in lean pigs, including F. prausnitzii [44] whose abundance in the gut was negatively correlated with P. copri in both experimental pig cohorts (Additional le 1: Fig. S14). The function term of metabolism of cofactors and vitamins was enriched in the gut microbiome of lean pigs. Interestingly, the vitamin K, D3, pantothenic acid, and biotin, which have been reported to be associated with decreased common obesity [45], reduced in ammation [46], and increased energy expenditure and adiponectin expression [47], were enriched in the serum of lean pigs.
Overall, combining the LMP-associated bacterial taxa, metagenome functional capacity and serum metabolome, we propose a model of gut microbiome in uence on the porcine LMP: 1) High abundance of P. copri in the gut largely drives the increase of serum LPS, BCAA, and arachidonic acid metabolites, which should dampen host immune response [48], upregulate host intestinal barrier permeability, and lead to chronic in ammatory response; 2) the genes related to immune and in ammatory response, lipogenesis, and fat accumulation are upregulated, while the genes associated with lipolysis, lipid transport and muscle growth are downregulated; and 3) butyrate-producing bacteria and those metabolites that are involved in anti-in ammatory action can increase lean meat percentage.

Conclusions
In conclusion, we identi ed and con rmed that P. copri from the gut microbiome of pigs fed by commercial formula diets signi cantly increased the fat deposition of pigs. We also proposed a possible mechanism of P. copri affecting fat accumulation. The results provided fundamental knowledges for reducing pig fat accumulation and increasing the LMP through regulating the gut microbial composition in the pig industry.

Experimental animals and sampling
Two experimental pig cohorts were used in this study. The discovery cohort comprised 550 Duroc pigs from Shahu (280 pigs) and Jiangying (270 pigs) farms in southern China. Another 148 Duroc pigs from the Jiangying farm were used as the validation cohort. All experimental pigs were raised under similar feeding and management conditions. The commercial formula feeds provided to experimental pigs of each farm contained 60% corn, 15% soybeans, 10% wheat bran, and 8% rice polishings. The main nutrient components of the diets are listed in Additional le 2: Table S8. Diet and water were offered ad libitum. Backfat thickness and transection area of the longissimus dorsi muscle were measured in the middle of the last 3rd and 4th ribs using a B-model ultrasound instrument (Pie-Medical, Netherlands) when the body weight of experimental pigs achieved 120 ± 10 kg, around the age of 160 ± 10 days. The GPS software was used to adjust the backfat thickness and transection area of the longissimus dorsi. Lean meat percentage (LMP) was calculated by the model: adjusted PPL = [80.95-(16.44*adj.bf)+(4.693*adj. LMA)]*0.54 [49], where PPL represents lean meat percentage, and adj.bf and adj. LMA represent adjusted backfat thickness and transection area of longissimus dorsi, respectively. The fecal samples were collected from all experimental pigs at the age of 160 days, conserved in sterilized tubes, and immediately immersed in liquid nitrogen for transportation and then stored at −80°C until use. In the validation cohort, we chose 16 fecal samples with extreme phenotype values for metagenomic sequencing, including eight samples with high LMP values (57.83 ± 0.54, mean ± SD) and eight samples with low LMP values (54.57 ± 0.59). To investigate the abundance of P. copri isolated in this study in the gut of pigs fed diets with different ber contents, metagenomic sequencing data of six fecal samples from natural free-living wild boars (high-ber diets) and 13 fecal samples from semi-grazing Tibetan pigs (supplemented with potato and highland barley, high-ber diets) were also used in this study. All experimental pigs were healthy and had not received any antibiotics, probiotics, or prebiotics within at least two months before sample collection.

DNA extraction and 16S rRNA gene sequencing
Fecal microbial DNA was extracted using the QIAamp DNA Stool Mini Kit (QIAGEN, Germany) following the manufacturer's guidelines. DNA concentration was measured with a Nanodrop-1000 (Thermo Scienti c, USA), and the quality was assessed by agarose gel electrophoresis. The barcoded fusion forward primer 515F (5'-GTGCCAGCMGCCGCGGTAA-3') and the reverse primer 806R (5'-GGACTACHVGGGTWTCTAAT-3') were used to amplify the V4 hypervariable region of the 16S rRNA gene in the discovery cohort. The primers 338F (5'-ACTCCTACGGGAGGCAGCA -3') and 806R (5'-GGACTACHVGGGTWTCTAAT -3') were used to amplify the V3-V4 hypervariable region of the 16S rRNA gene in the validation cohort. The PCR ampli cation conditions were as follows: initial 95°C denaturation step for 10 min, 35 cycles of 95°C for 25 s, 55°C for 20 s, and 72°C for 5 min followed by a nal extension for 10 min at 72°C. All amplicons were sequenced using the paired-end method on a MiSeq platform (Illumina, USA) following the standard protocols.
The raw 16S rRNA gene sequencing data were ltered for the primer sequences, the barcodes, and the low-quality reads according to Illumina's quality control procedure. High-quality paired-end clean reads were assembled using FLASH (v1.2.11) [50]. The USEARCH (v7.0.1090) quality lter pipeline was used to lter the putative chimeras and to choose Operational Taxonomic Units (OTUs) at 97% sequence identity [51]. Only those OTUs that had relative abundance > 0.05% and were present in more than 1% of the experimental pigs were included for further analysis. Taxonomies were assigned for the aligned sequences using Quantitative Insights Into Microbial Ecology (QIIME, v1.80) with a Ribosomal Database Project (RDP) classi er [52].

Construction of enterotype-like clustering
Enterotype-like clustering was performed according to the method described previously [53]. In brief, Jensen-Shannon divergence (JSD) distances were calculated based on the relative abundances of bacterial taxa at the genus level using the Partitioning Around Medoids (PAM) method. The optimal number of clusters and the groups' robustness were evaluated with the Calinski-Harabasz (CH) index and silhouette value. Sparse Correlations for Compositional data (SparCC) was applied to determine coabundance (positive) and co-exclusion (negative) relationships between genera based on their relative abundances [54]. Signi cant correlations between bacterial genera were identi ed using the partial correlation and information theory (PCIT) algorithm [55]. The absolute correlations were transformed into links between two genera in the genus network, and the networks were visualized in Cytoscape (v3.4.0).
The comparison of the LMP values between enterotypes was performed by Wilcoxon rank sum tests in the R package (v3.5.1).

Association analysis between OTUs and pig LMP
The residuals of phenotypic values of the LMP corrected for the effects of sex and sampling batch (three and two sampling batches for discovery and validation cohort, respectively) were used for further association analysis between the LMP values and the relative abundances of OTUs. Because the relative abundances of OTUs exhibited a non-normal distribution pattern, the association analysis was performed using a two-part model as reported previously [56]. In brief, the two-part model accounts for both binary and quantitative characteristics of gut microbial abundance. The binary model (adj_p = β 1 b + e, adj_p represents LMP values adjusted for the effects of sex and batch, β 1 is the estimated binary effect, b is a binary feature, and e refers to the residuals) describes a binomial analysis that tests for association of detecting a microbe with the LMP. The binary feature of a microbe under investigation was coded as 0 for undetected or 1 for detected in each sample. The quantitative model (adj_p = β 2 q + e, where β 2 is the estimated quantitative effect, and q is a quantitative feature) evaluates the association between the abundances of the detected microbes and the LMP values. A meta-analysis was performed to assess the effects of both binary and quantitative models by using an unweighted Z method (Z = ∑ k i=1 z i / ~N(0,1); z i = -1 (P i )). The nal association P value was set as the minimum of P values of binary, quantitative, and meta-analyses. In total, 1000 permutation tests were performed to correct for false positives, and a false discovery rate (FDR) < 0.01 was set as the signi cance threshold.

Co-abundance group (CAG) analysis of OTUs
The OTUs having relative abundance > 0.1% were used to construct CAGs. We rst calculated the correlation coe cients among OTUs using the Sparse Correlations for Compositional data (SparCC) algorithm in both test and validation cohorts [54]. Then, CAGs were de ned by a heat plot using the SparCC correlation coe cient matrix and Ward's linkage hierarchical clustering through the Made4 (v3.40) package [57]. PERMANOVA was performed to assess the accuracy of clustering with 1000 permutations at P < 0.01 [58]. The network plot highlighting the SparCC correlations among CAGs was constructed in Cytoscape (v3.6.0) [59]. Spearman's correlation analysis was performed to test the correlations between CAGs and the LMP values in both test and validation cohorts.

Metagenomic sequencing analysis
A pair-end (PE) library with insertion size of 350 bp was constructed for each of 16 samples according to the manufacturer's instructions (Illumina, USA). Sequencing was performed on a Novaseq 6000 platform (Illumina, USA). High-quality reads were obtained by ltering out adaptors, low quality reads, and host genomic DNA contamination from the raw data.
We assembled the high-quality reads into contigs using the SOAPdenovo assembler (v.2.21) [60]. The USEARCH (v.7.0.1090) program was used to exclude the redundant contigs [51]. The contigs more than 300 bp in length were used to predict open reading frames (ORFs) by applying MetaGeneMark (v2.10) [61]. A non-redundant gene set containing 2,799,188 genes was constructed by excluding the redundant genes from all predicted ORFs using Cd-hit software (v4.6.1) [62]. A gene abundance pro le was generated by mapping the high-quality reads from each sample to the non-redundant gene set using the screen function in MOCAT (v2.0) [63]. To assess gene richness in the high and low LMP pigs, we calculated the total gene number in each sample using the pair-oriented counting method [16]. The αdiversity (Shannon index) was calculated using the gene abundance pro les using the vegan R package (v3.5.1). Comparisons of gene counts and the α-diversity between high and low LMP pigs were performed using the Wilcoxon rank sum test. Taxonomic assignments of the predicted genes were performed using the BLAST + Lowest Common Ancestor (BLAST + LCA) algorithm based on the sequence similarity to the reference genomes in the non-redundant (NR) database [64]. Functional annotations were performed by aligning the putative amino acid sequences that were translated from the predicted genes against CAZy and KEGG databases using BLASTP [65]. Linear discriminate analysis effect size (LEfSe) was used to identify the bacterial species and function capacities of gut microbiome having signi cantly different abundances between high and low LMP pigs. Correlations between the LMP-associated bacterial species and the LMP-associated function capacities of gut microbiome were evaluated in the 16 samples with metagenomic sequencing data using Permutational analysis of variance (PERMANOVA) based on 9,999 permutations using the vegan package in R (v3.5.1) [12]. The signi cance threshold was set at FDR < 0.05. The correlation coe cient was calculated as Spearman's rank correlation. The heatmap was plotted using the gplots package in R (v3.5.1) [66].
The metagenomic sequencing data of another 20 fecal samples from the discovery cohort were obtained in our previous study via the same method [17] and were also used in this study. The association of bacterial species with the LMP in the integrated 36 metagenomic sequencing data was analyzed by a two-part model as described above. The comparison of the abundance of P. copri among pigs fed diets with different ber contents was performed with the metagenomic sequencing data of fecal samples from six wild boars, 13 Tibetan pigs, and 36 Duroc pigs as described above. Clean reads of each sample were aligned to the gene catalog using BWA MEM (v0.7.17-r1188) [67], and then the number of successfully assigned reads was computed using FeatureCounts (v2.0.1) [68]. The abundances of genes were normalized to fragments per kilobase of gene sequence per million reads mapped (FPKM). The abundance of P.copri was calculated by summing the abundances of all the members belonging to this species. The comparison of gut P. copri abundances among wild boars, Tibetan, and Duroc pigs was performed by a Wilcoxon test and visualized using the ggpubr package in R (v3.6.2).
Isolation and culture of the bacterial strain of P. copri from pig fecal samples The fecal samples from 22 experimental Duroc pigs with both extreme phenotypic values of fat accumulation (low LMP) and high abundance of P. copri were collected and used for the P. copri isolation experiment. One-gram fecal samples were suspended in phosphate buffered saline (PBS) buffer and serially diluted to 10 -8 . Eighty-microliter diluted samples were plated anaerobically on Bacteroides mineral salt agar to isolate P. copri [69] . The plates were incubated at 37 o C for 2-7 days in an anaerobic workstation (ELECTROTEK AW500SG UK) lled with 80% N 2 -10% CO 2 -10% H 2 gases [70]. A single colony from plates was selected according to the main characteristics of the strain that we were looking for based on the previous description [71], i.e., white, circular, convex and gram-negative rods, and puri ed by streaking the single bacterial colony on modi ed PYG agar supplemented with 5% (v/v) sterile de brinated sheep blood with a sterile probe [71]. The plates were maintained under the culture conditions mentioned above for two days. The 16S rRNA gene of the single strain was ampli ed using two universal primers 27F (5'-AGAGTTTGATCCTGGCTCAG-3') and 1492R (5'-GGTTACCTTGTTACGACTT-3') and sequenced by the Sanger method. The 16S rRNA gene sequences were then aligned to the NCBI nucleotide sequence database to determine P. copri strains. In addition, we blasted the 16S rRNA gene sequences of the isolated strains with the V3-V4 sequence of the OTU1905 (P. copri) that was most signi cantly associated with the LMP in this study. The isolated strain with > 99% sequence identity was used for gavage in germ-free mice. The P. copri strain isolated above was cultivated in modi ed PYG medium for 36 h under anaerobic conditions, harvested in log phase, centrifuged at 1000 rpm for 10 min, and then washed twice with PBS. The precipitate was re-suspended with 5% sterile non-fat milk prepared by PBS and stored at −80°C until use.
Whole-genome sequencing of P. copri The isolated P. copri strain was recovered and grown on PYG liquid medium at 37℃ with 80%-N 2 -10%CO 2 -10%H 2 for 72 h. Ten milliliters of cultured PYG uid was centrifuged at 5000 rpm for 10 min. P. copri cells were washed twice using sterilized PBS solution and collected for DNA extraction. Genomic DNA of P. copri was extracted using QIAamp DNA Mini Kit according to the manufacturer's instructions. The quantity and quality of extracted DNA were evaluated by agarose gel electrophoresis, NanoDrop-2000 (ThermoFisher, USA) and Qubit (ThermoFisher, USA).
A Nanopore sequencing library was prepared according to Oxford Nanopore's "1D gDNA selection for long reads" protocol (Oxford Nanopore Technologies, UK). In brief, 2 μg of genomic DNA of P. copri was sheared using a g-Tube (Covaris, USA) with 150 μl of nuclease-free water at 5000 rpm for 2 min. Long DNA fragments were enriched using the Blue Pippin selection system (Sage Science, USA). Subsequent puri cation of the DNA fragments was performed using AMPure beads. Nanopore 1D adapters were ligated to the end-repaired and adenylated DNA fragments using NEB Blunt/TA Master Mix (NEB, UK). The libraries were sequenced on a GridION X5 (Oxford Nanopore Technologies, UK). To improve the sequence quality, a library for second generation sequencing was prepared according to the standard protocol and sequenced on an Illumina Hiseq-2500 platform using a paired-end strategy.
Protein-coding genes of P. copri genome were predicted using Prodigal (v2.6.3) [80]. The predicted protein-coding genes were further annotated with InterProScan using Blast2GO against Pfam (release 31.0), TIGRFAMs (release 15.0) and SMART (v8.0) databases [81,82]. Functional annotation of proteincoding genes was also performed by Blast2GO with the KEGG database. To compare the abundances of those interesting genes identi ed on the P. copri genome and participating in arachidonic acid metabolism, BCAA biosynthesis, AAA biosynthesis and metabolism, insulin resistance, and other glycan degradation between high and low LMP pigs, the sequences from the metagenomic sequencing data were mapped to the obtained P. copri genome. The relative abundances of these genes were determined and compared using Wilcoxon tests. FDR < 0.05 was set as the signi cance threshold.
Construction of phylogenetic tree and analysis polysaccharide utilization loci of P. copri isolates To construct the phylogenetic tree of P. copri isolates from humans and pigs, we downloaded 111 P. copri genomes from westernized and non-westernized human gut microbiome [19]. Gff le of each genome was generated using prokka (v1.11) [83] and used to produce the alignment of core genes by Roary (v3.11) [84]. The phylogenetic tree was constructed based on the alignments of core genes using neighbor-joining approach in Megan 7 and visualized by iTOL [85]. Polysaccharide utilization loci of P. copri isolates were predicted by using deCAN-PUL (http://bcb.unl.edu/dbCAN_PUL) with identity > 75% and E-value < 1e-50.

Mouse intervention study
Twenty-one germ-free mice having similar body weight and size (Kunming; 12 males and nine females, each six weeks of age) used in this study were housed in cages under sterile conditions. Male and female mice were kept separately. Feed and water were available ad libitum. After two weeks of acclimatization to the new environment and the standard chow diet, mice were randomly divided into three groups (four males and three females per group). One group received a chow diet with P. copri administered by gavage. A high-fat diet group (60% fat, Research Diet, D12492) was administered with P. copri by gavage, and a chow diet group without gavage was used as a control. For the two colonization groups, mice were given 100 μl of P. copri suspension (1 × 10 7 CFUs/μl) three times a week for four weeks. To further investigate the effect of diet on P. copri's role in host fat accumulation, we used another 18 germ-free mice with the similar body weights and sizes (C57BL6; nine males and nine females) to perform gavage experiments using P. copri. These germ-free mice were managed and administered the bacteria using the same gavage methods and procedures described above. The 18 mice were randomly divided into three groups (three females and three males for each group) comprising a standard chow diet group, a high-fat diet (60% fat, Research Diet, D12492) group, and a high-ber diet (35% ber) group. The feeding experiment lasted four weeks, and mice were administered P. copri by gavage three times per week as described above.
Fecal samples were collected at the end of the gavage experiment, dipped into liquid nitrogen immediately, and stored at −80°C until use. All mice in each group had lean mass measured and body fat percentage calculated by a whole-body composition analyzer (Niumag, China) following the manufacturer's instructions. After the body weight measurements, all mice were sacri ced by cervical dislocation. Epididymal fat was isolated and weighted for all mice. The epididymal fat percentage (EMP) was calculated. Tissue samples of colon, epididymal white adipose, and muscle were sampled from each experimental mouse for further RNA-seq analysis. Venous blood was taken from the inner canthus of each mouse for serum metabolomic analysis. The concentrations of lipopolysaccharide, intestinal barrier permeability plasma biomarkers, and pro-in ammatory cytokines were also determined in serum samples of phenotyped mice by ELISA using the method described above.
Quantifying the abundances of P. copri in treated mice Mouse fecal bacterial DNA was extracted using the QIAamp fast DNA stool mini kit (Qiagen, Germany) as described above. The quantitative PCR was performed using a 7500-Fast Real-Time PCR System (ABI, USA) and SYBR® Premix Ex Taq™ II (TaKaRa, Japan). The two-step Real-Time PCR conditions were described as follows: an initial denaturation for 10 s at 95°C, 40 cycles of denaturation at 95°C for 5 s, and annealing at 60°C for 25 s. The RQ value of P. copri was determined by normalization to the 16S rRNA gene using the 2 -ΔΔCt method [15]. Primer sequences are listed in Additional le 2: Table S9 [86].
Determination of metabolome pro ling of serum samples Metabolome pro les of serum samples were determined for 38 pigs randomly selected from the validation cohort, and for seven mice from each of control, P. copri gavage, and P. copri gavage + HFD feeding groups (a total of 21 samples). Blood samples were collected from the anterior vein. After being placed into serum separator tubes, all samples were centrifuged at 2500 × g for 15 min at room temperature to isolate the serum. Serum samples were immediately stored at −80°C until use. A 100-μL aliquot of serum sample was used for the extraction of metabolites using 3 ml of pre-cooled methanol (chromatographically pure) (Merck Corp., Germany). After vortexing for 1 min and incubation at −20°C in a refrigerator for 3 h, the mixture was centrifuged at 15,000 rpm for 15 min at 4°C to precipitate the protein. Then, 200 μl of the supernatant was processed in a Speedvac overnight. The concentrated product was resuspended by the addition of 150 μl of water/methanol (85:15, v/v) and then placed into a sampling vial pending ultraperformance liquid chromatography-quadrupole time-of-ight mass spectrometry (UPLC-QTOFMS) (Waters Corp., USA). The quality control was performed via a pooled QC sample by mixing equal volumes (15 μl) of each serum sample.
Chromatographic separations were performed on a UHPLC BEH C18 column (2.1 mm × 100 mm, 1.7 μm) (Waters Corp., USA) maintained at 40°C. The injection volume was 0.4 μl for each sample, and the samples for blank-QC-tests were run alternately. The column was eluted with a linear gradient of 1-20% B at 0-3 min, 20-50% B at 3-5 min, 50-70% B at 5-10 min, 70-85% B at 10-15 min, and 85-100% B at 15-17 min followed by a re-equilibration step of 5 min. For electrospray positive ion mode (ES+) analysis, the mobile phase was water with 0.1% formic acid (A) and acetonitrile with 0.1% formic acid (B). For Negative ion mode (ES−) analysis, eluents A with water and B with acetonitrile were used. The ow rate was set at 0.3 mL/min. All the samples were kept at 8°C during the analysis.
The mass spectrometric data in both positive and negative modes were collected using an electrospray ionization source. The source parameters were set as follows: capillary voltage: 3 kV; drying gas ow: 11 L/min, and gas temperature: 350°C. Centroid data were collected from 50 to 1200 m/z with a scan time of 0.3 s and interscan delay of 0.02 s over a 20-min analysis time. MassLynx software (Waters, USA) was used for system controlling and data acquisition. Data normalization was performed by QC samples using MetNormalizer in R (v 3.5.1) that generated a data matrix containing retention time, m/z value, and normalized abundance [87]. To obtain metabolite names and the molecular formulas, we aligned the molecular mass data (m/z) of ions to the metabolites in the HMDB database (http://www.hmdb.ca) with a mass error of 10 ppm or less [88].
Associations between serum metabolites and porcine LMP phenotypic values were tested by Spearman rank correlation in the 38 experimental pigs. The analysis was performed using both 16S rRNA gene sequencing and metabolome analyses at FDR < 0.05. The correlations between the LMP-associated serum metabolites and the LMP-associated OTUs were assessed by Spearman correlation coe cients (FDR < 0.05). To further evaluate the correlation between the LMP-associated bacterial species and the LMP-associated serum metabolites in the 16 tested samples from the metagenomic sequencing, the metabolites differing in normalized abundance between high (n = 8) and low LMP pigs (n = 8) were identi ed by LEfSe. The online MetaboAnalyst program (http://www.metaboanalyst.ca) was used to assign the differential metabolites to KEGG pathways [89]. PERMANOVA and Spearman's correlation analysis were performed to assess the correlations between the LMP-associated bacterial species and the LMP-associated serum metabolites as described above. The serum metabolites with different abundances between controls and gavage-treated mice were identi ed by LEfSe.
Quantifying serum concentrations of lipopolysaccharide, intestinal barrier permeability biomarkers, and pro-in ammatory cytokines We quanti ed the concentrations of serum lipopolysaccharide (LPS), fatty acid-binding protein 2 (FABP2), zonulin, IL-1, IL-6, IFN-γ, and TNF-α using the enzyme linked immunosorbent assay (ELISA) method with commercial ELISA kits (Keshun, China) following the manufacturer's instructions. Brie y, except for the blank control wells, 50 μl of standard samples or appropriately diluted serum samples were added into the 96-well microtiter plates coated with the primary antibodies, and then 100 μl of HRP-conjugated secondary antibodies was added to the microtiter plates and incubated for 60 min at 37°C. Microtiter plates were washed four times with washing buffer, and 50 μl of substrates A and B were added to each well of microtiter plates, mixed gently, and incubated for 15 min at 37°C under light shading conditions. Finally, 50 μl of enzymatic reaction termination solution was added to each well to stop the reaction. The O.D. value for each sample at 450 nm was measured and recorded using a microtiter plate reader (Tecan In nite 200 pro, Switzerland). A standard curve was plotted according to the O.D. values and the concentrations of standard samples. The serum concentrations of LPS, intestinal barrier permeability plasma biomarkers, and pro-in ammatory cytokines in each test sample were determined using the standard curve. Each standard and tested serum sample was measured in triplicate. The Wilcoxon rank sum test was used to compare the serum concentrations of LPS, FABP2, zonulin, and pro-in ammatory cytokines between high and low LMP pigs at an FDR < 0.05. The multiple group comparisons of these data among experimental mice were performed by the Kruskal-Wallis test 34 . All these analyses were carried out using the R software (v3.5.1).

RNA extraction, sequencing, and data analysis
The mice used for con rming the causality of P. copri were further used for RNA sequencing analysis. Six mice from the group administrated with P. copri and fed standard chow diet, and the other six mice from the control group were randomly chosen. Total RNA was extracted from colon, epididymal white adipose, and muscle tissues using Trizol (ThermoFisher, USA) according to the manufacturer's manuals. The RNA concentration and integrity were assessed using a Nanodrop-1000 spectrophotometer (ThermoFisher, USA) and a bioanalyzer-2100 (Agilent, USA). The cDNA libraries were prepared using the Illumina Truseq Stranded mRNA preparation kit (Illumina, USA) according to the manufacturer's guidelines. The libraries were sequenced on an Illumina HiSeq 2500 platform (Illumina, USA). Raw data were trimmed for adapter sequences, and low-quality reads were ltered out to generate clean data. After that, the HISAT, StringTie and Ballgown pipelines were used to explore differentially expressed genes (DEGs) between controls and colonized mice as described previously [90]. Brie y, Hisat2 (v2.1.0) was employed to build a reference genome index, and then high-quality read sequences were aligned to the mouse reference genome assembly (GRCm38) to generate SAM les. Samtools (v1.8.0) was used for SAM le transformation and read sorting to generate sorted bam les [77]. Transcript assembly and quanti cation were performed using StringTie (v1.3.4). The outputs of StringTie, including gene annotation and gene abundance les, were processed by Ballgown (v3.5) to identify DEGs based on FPKM values with FDR < 0.05. DEGs were aligned to the GO database to perform gene ontology analysis [91].

Statistical analysis
Shapiro-Wilk and Levene's tests were performed to evaluate the distribution and equality of variances of the LMP values in the tested pig populations. The signi cance levels of differential LMP values between two groups of pigs selected for metagenomic sequencing (8 vs. 8), and between two groups of pigs used for determining serum metabolome pro les (17 vs. 21) were determined by t-tests. All analyses were performed using R (v3.5.1). The bacterial species, function capacities of gut microbiome, and serum metabolite features showing differential abundance between high and low LMP pigs were identi ed using LEfse using the online version of Galaxy at a signi cance threshold criteria of LDA score > 2.5 and alpha value < 0.01 [92]. The associations between the relative abundances of bacterial species and serum metabolite features were analyzed with the PERMANOVA method. The correlations between the LMP-associated bacterial species and the LMP-associated functional capacities of gut microbiome, and between the LMP-associated bacterial species and the LMP-associated serum metabolites, were evaluated using Spearman's correlation at FDR < 0.05.
For the colonization experiments using mice, we rst tested the distributions of the phenotypic data in each group by the Shapiro-Wilk and Levene's tests. The multiple group comparisons of the phenotypic values of body fat percentage and epididymal fat percentage among controls, P. copri gavage mice, and HFD + P. copri gavage mice or among P. copri gavage, P. copri gavage + HFD and P. copri gavage + high ber diet groups were performed by TukeyHSD tests at FDR < 0.05. Differential serum metabolites between groups were identi ed by LEfSe at LDA score > 3.5 and alpha value < 0.01.