The visceral adipose tissue bacterial microbiota provides a signature of obesity based on inferred metagenomic functions

Metabolic inflammation mediated obesity requires bacterial molecules to trigger immune and adipose cells leading to inflammation and adipose depot development. In addition to the well-established gut microbiota dysbiosis, a leaky gut has been identified in patients with obesity and animal models, characterized by the presence of a tissue microbiota in the adipose fat pads. To determine its potential role, we sequenced the bacterial 16 S rRNA genes in the visceral adipose depot of patients with obesity. Taking great care (surgical, biochemical, and bioinformatic) to avoid environmental contaminants. We performed statistical discriminant analyses to identify specific signatures and constructed network of interactions between variables. The data showed that a specific 16SrRNA gene signature was composed of numerous bacterial families discriminating between lean versus patients with obesity and people with severe obesity. The main discriminant families were Burkholderiaceae, Yearsiniaceae, and Xanthomonadaceae, all of which were gram-negative. Interestingly, the Morganellaceae were totally absent from people without obesity while preponderant in all in patients with obesity. To generate hypotheses regarding their potential role, we inferred metabolic pathways from the 16SrRNA gene signatures. We identified several pathways associated with adenosyl-cobalamine previously described to be linked with adipose tissue development. We further identified chorismate biosynthesis, which is involved in aromatic amino-acid metabolism and could play a role in fat pad development. This innovative approach generates novel hypotheses regarding the gut to adipose tissue axis. This innovative approach generates novel hypotheses regarding the gut to adipose tissue axis in obesity and notably the potential role of tissue microbiota.


INTRODUCTION
Over the last decade, numerous scientific reports describe a specific microbiota signature in the gut that characterizes patients with obesity notably, through a reduced taxonomic and bacterial gene diversity content [1].Some mechanisms related to the molecular interplay between gut microbiota and adiposity have been proposed.An initial enhanced feed efficiency was suggested to be mediated by a change of the Bacteroidetes to Firmicutes ratio [2].Furthermore, an increased gram-negative bacterium has been reported [3] characterized by a moderate low grade but persistent elevation of plasma LPS concentration.The latter would stimulate what is known as metabolic inflammation [4], linking it [5] to the development of obesity.Moreover, LPS has been shown to trigger preadipocyte proliferation leading to an increased fat mass [6].A rate limiting step is the transport of LPS from the gut to the tissue to trigger inflammation.Recently, we and others described that obesity is characterized by increased intestinal permeability [3,7] leading to the translocation of LPS-containing alive bacteria towards tissues such as the adipose depots [8].Thanks to dramatic technical improvements and numerous validations throughout the world of the specific sequencing of the bacterial 16SrRNA gene way above contaminant background in tissues [9,10] we have been able to establish the existence of a tissue microbiota [11].This new tissue microbiome ecology is suggested to be causal of metabolic inflammation leading to obesity [12].Furthermore, type 2 diabetes, a comorbidity of obesity, induces a specific tissue microbiota dysbiosis [13] which could further reinforce insulin resistance and hyperglycemia.Hence, there is a tissue microbiota signature associated with obesity and its comorbidities.Whether this signature is specific of obesity is unknown.Moreover, the potential functions and biochemical pathways associated with a tissue bacterial ecology specific to obesity remains to be established.To this aim, we analyzed patients from the ROLINASH cohort, sequenced and analyzed the 16SrRNA gene in visceral adipose tissue to establish signatures of the potential tissue microbiota according to obesity.Visceral obesity is considered the adipose depot more relevant for the development of insulin resistance and chronic low-grade inflammation.It is noteworthy that extreme care was taken to avoid any contaminant and to report only adipose tissue specific 16SrRNA gene sequences.

MATERIAL AND METHODS Human cohort
The study was performed on a subset of the ROLINASH cohort (registration 4065/2014).A monocentric observational study was conducted in the Second Department of Surgery, Emergency Mureş County Hospital of Romania.Informed consent was obtained from patients.Exclusion criteria were serious diseases notably linked to inflammation (Hepatitis B and Hepatitis C infection, chronic diseases, inflammatory systemic diseases, acute or chronic infections in the previous month, use of antibiotic, antifungal, antiviral drugs, proton-pump inhibitors, anti-obesity drugs, laxatives, worth to mention, fiber supplements or probiotics or participation in a weight loss program or weight change of 3 kg during the previous 6 weeks, pregnancy or breastfeeding, or major psychiatric antecedents; neurological diseases, history of trauma or injured brain, language disorders, and excessive alcohol intake (≥40 g/day in women or 80 g OH/ day in men) or intravenous drug abuse, and previous bariatric surgery.The cohort consists of 62 Caucasian patients classified according to BMI: with BMI < 30 as people without obesity, between 30-40 as patients with obesity, and ≥40 (Table 1) as people with severe obesity.Informed written consent was obtained from each patient and the study protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki and was approved by the Ethics Committee of the Hospital of Girona, Spain, and the Hospital of Roma, Italia.
Adipose tissue biopsies.During laparoscopic surgical procedures, omental adipose tissue biopsies were conducted without using any energy devices to collect the samples.Hemostasis was done after the samples were extracted from the abdomen.On all patients, antiplatelet drugs and oral anticoagulation therapy were paused for 1 week before the biopsies.Were performed on all patients.After following the biopsies, all patients were monitored for any signs of pain or clinically suspected bleeding by nursing staff over a 3-day period.If no deleterious secondary events were evident, all patients would be discharged after the mandatory days follow up observation, a stable blood count and a normal ultrasound examination.All patients were followed-up in 2 weeks to review the results of the histology.The samples were stored in a sterile container and kept at −80 °C until assayed.
Clinical assessments.Trained nurses performed anthropometric measurements of each subject in the morning after fasting for at least 8 h BMI was defined as body weight (in kilograms) divided by the square of body height (in meters).Waist circumference was measured in the horizontal plane midway between lowest rib and the iliac crest to the nearest 0.1 cm at the end of a normal expiration repeatedly in men and women by 3 trained nurses on 3 consecutive days.Fasting plasma glucose was measured after fasting for at least 12 h.Hypertension was defined in accordance to the Guidelines of the European Heart Association or if the subject was taking medication for hypertension.Diabetes was diagnosed when fasting plasma glucose was ≥126 mg/dL (7 mmol/L), 2-h postprandial plasma glucose ≥200 mg/dL (11.1 mmol/L), and HbA 1c ≥ 6.5% or if the subject was taking medication for diabetes.

16SrRNA gene sequencing and bioinformatic analysis
Total DNA was extracted from visceral adipose using the NucleoSpin® Blood kit (Macherey-Nagel, Germany) after a mechanical lysis step of 2

Adipose tissue 16SrRNA bacterial DNA analyses
To ensure the accuracy of the analysis, it is important to differentiate between bacterial DNA sequences specifically present in the adipose depots and those that may be potential contaminants.It should be noted that starting in 2015, results from independent researchers sequencing the 16 SrRNA gene from tissues such as the blood, the adipose depots and the liver, in the field of metabolic diseases, were published [9,10,[15][16][17][18], notably in human tissues.In such reports, as in the present study, the risk of bacterial contaminants from reagents was deeply investigated, as shown in recent publications [19,20].Furthermore, Negative controls were systematically run in each experiment, including extraction procedure negative controls (water at DNA extraction step) and PCR negative controls (water at first PCR step).The 16 S rRNA genes amplified by PCR from the negative controls have been sequenced and showed a very low abundance (1000-fold lower) of the contaminant when compared to the tissue 16SrRNA gene concentration (to be provided upon request).The bacterial gene richness of the samples was found to be several folds higher than in the control, further supporting the conclusion that the identified bacterial DNA sequences are not contaminants.Notably, the contaminants identified by Hornung et al. were not found in our sequences, demonstrating that, at least when compared with other groups, those contaminants are excluded [16].In preliminary sets of data, we identified different bacteria from culture-omics, i.e., the cultured bacteria extracted from the adipose depots and secondarily grown on plates and clones sequenced.These analyses performed on different human tissues are available upon request.

In silico analytical and statistical procedures
Normalization procedure.The clinical parameters were normalized as follows: xiÀmeanðxÞ sdðxÞ , where mean(x) is the mean of x values, and sd(x) is the standard deviation (SD).For the graphical representation of the clinical variables, we performed a principal component analysis (PCA) using the R software and the FactoMineR package.This function automatically standardizes the data.We used the missMDA (1.18) R Packages to handle the missing values.
In contrast to other omics analyses such as RNASeq 16S rRNA MiSeq sequencing data (targeted microbiota) are very sparse with zero counts in most if not all samples.In addition, many rare taxa are observed and due to sequencing artifacts, contaminant removal and sequencing errors.To overcome these issues, we performed a two steps analysis.The primary step involves pre-processing: (1) Pre-filtering the raw count data to remove OTUs with low counts across all samples.We considered an incidence of an OTU below 5 observations as negligible.(2) we performed Centered logratio (CLR) transformation which is a normalization method implemented in aldex.clr()function in ALDEx2 (version 1.4.0)ALDEX2 estimates technical variation within each sample per taxa by using the Dirichlet distribution.It furthermore applies the centered-log-ratio transformation to remove compositional constraints [21].The result of the normalization was used for the beta diversity analysis, and the discriminant analyses.
Taxonomic composition.Stacked bar plots were performed for the relative abundance of the top 10 bacteria at the phylum and family taxonomic levels.The analyses were generated using the "ggplot2" (version 3.3.4)and "reshape2" (version1.4.4) packages.Statistical analyses were performed using "ggpubr" (version > = 0.1.3)package.Bray-Curtis distances between patients were calculated using "Phyloseq" (version 1.16.2) package.Hierarchical cluster analysis was determined using "complete" as the agglomeration method.For the visualization and annotation of phylogenetic trees the "ggtree" (version 3.13) package was used.
Alpha diversity and beta diversity analyses.Alpha microbial diversity graphs generated using the R package "Phyloseq" and "ggplot2" for the following alpha diversity indexes: Observed Species, Chao1, AEC, Shannon, Simpson, InvSimpson, and Fisher.Statistical significance between more than two groups was calculated using the Kruskal-Wallis test for nonparametric analyses with the default number of Monte Carlo permutations.Beta microbial diversity was calculated, and the Manhattan method was used to estimate the distances between patients from the BMI groups.The scatter plot was generated and the principal coordinate analysis (PCoA) calculated.To determine the statistical significance of differences between all groups PERMANOVA was used, with P < 0.05 were considered significant.
To quantify the similarities between groups, we performed an ANOSIM (Analysis of Similarities).The corresponding Rs range between −1 and 1.
Positive numbers indicate similarity within groups, while values close to zero indicate absence of difference between groups.
Discriminant analyses.Initially, a Venn diagram analysis (VennDiagram package, version 1.6.20)was conducted to identify OTUs that were present in less than 5 patients in the overall cohort.These OTUs were eliminated from the overall OTU table since they add a lot of non-specific background.The filtered OTU table was then used to perform a sparse Partial Least Squares Discriminant Analysis (sPLS-DA using counts at the family level after normalization using the ALDEx2 R package.From these analyses, the Area Under Curve (AUC) of the ROC (Receiver Operating Characteristics) curve was calculated using R package "mixOmics".In addition, a Random Forest analysis was also performed using the R package "randomForest" (version 4.6-14) to compare outcomes with other discriminant analyses, thereby refining the discriminant OTUs.
Functional metagenomic prediction.To predict the functional potential of a bacterial community, we inferred the abundance of corresponding microbial genomes and associated metabolic pathways from the 16S rRNA gene profiles using the PICRUSt2 (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States v2) algorithm.The total OUT table and FASTA file were used for this prediction, following PICRUSt2 tutorial (https://github.com/picrust/picrust2/wiki)and the generated pathway and enzyme code abundance tables were then imported into the different discriminant algorithms used for the OTUs: PLS-DA, Network, and Random Forest.We considered differences to be significant when p < 0.05 and calculated the fold change between groups as an indicator of differences.Furthermore, a volcano plot analysis was performed based on the DESeq2 R package to identify discriminant pathways and enzymes between groups.To visualize the metabolites and enzymes associated with the discriminant pathways according to the different BMI groups, the identified pathway codes were imported into the MetaCyc website.

Cohort clinical parameters
To identify the tissue microbiota signature associated with obesity, patients from the ROLINASH cohort were recruited and divided into people without obesity, (BMI < 30) from the patients with obesity (30 ≥ BMI < 40), and people with severe obesity (BMI ≥ 40) patients since the physiology of the latter's is most likely different from the other in patients with obesity.To analyze the distribution of the patients based upon their anthropomorphic and clinical features, a principal component analysis was first performed (Fig. 1A).Although some outliers were observed the first two components explained about 35% of the total variance.Body weight and related variables were the main drivers of the distribution on the first component (Fig. 1B, C).However, It is interesting to observe that liver biochemical parameters such as circulating serum liver enzymes were the next most contributors to the variance (Fig. 1C).Adipose tissue 16SrRNA bacterial DNA analyses It is well known that changes in microbiota ecology are related to alpha and beta diversities, which compare bacterial ecologies based upon their richness and similarities, respectively.For alpha diversity, different indexes were calculated that exemplify different OTUs based on their relative abundances.Different calculation modes were used for each index, which provided a wide and objective analysis of bacterial richness and diversity in the adipose tissue (Fig. 2A).Differences in richness were observed when using the Chao1 and the Fisher indexes, which favor the importance of low abundance taxa (Fig. 2B).Furthermore, the beta diversity, which analyzes similarities between the sequences of each OTU to establish a distance matrix, showed that the people without obesity group was mostly separated from the others two on both dimensions, strongly suggesting taxonomical differences (Fig. 2C-E).Differences in the beta diversity were observed when comparing the people without obesity, and people with severe obesity.The performance of Anosim Analysis R > 0 showed that the intra-group distance is smaller than the inter-group distance, validating the group analyses (Fig. 2F).
To characterize potential adipose tissue microbiota signatures specifically associated with obesity status, we first characterized the normalized relative abundance of each phylum per group (Fig. 3A).No major changes were observed between groups.Proteobacteria was the dominant phylum (~80%) followed by Firmicutes and Actinobacteria (Table 2).Individual data showed large variations in relative abundance between patients, suggesting different states of microbiota dysbiosis (Fig. 3B).At the family taxonomic level (Fig. 3C), besides the large heterogeneity of the bacterial ecologies per individual, the data showed that some differences appeared according to obesity status (Fig. 3D), characterized by Burkholderiaceae, Xanthomonadaceae, and Yersiniaceae (Fig. 2E-G and Table 3).Therefore, we calculated the relative taxon abundances for each individual patient and for each taxon.We identified that five families were significantly different between groups such as the Alcaligenaceae, Xanthomonadaceae, Burkholderiaceae, Yersiniaceae, and the Morganellaceae.Interestingly, the patients with no obesity were characterized by a total lack of this taxon (Fig. 3H-L).
Altogether, this first taxonomical analysis suggested that some differences exist between subjects with and without obesity but without providing a clear specific discriminant signature.To further refine our analysis, we performed discriminant multivariate analyses.A sparse Partial Least Square Discriminant Analysis (sPLS-DA) (Fig. 4A) showed that the different groups were much more discriminated between each other's.The circle plot analysis also discriminated 5 families Alcaligenaceae, Burkholderiaceae, Morganellaceae, Xanthomonadaceae and Yersiniaceae, (Fig. 4B).To confirm our initial results, we analyzed the database using the Random Forest analysis strategy, which ranked each family according to their significance within each group and showed that some families were specific to the groups that were identified when performing sPLS-DA analyses (Fig. 4C).The heatmap analysis also showed that the above families were clustered together with features related to body weight (Fig. 4D, and insert).Therefore, multiple families could classify each group although with varying levels of significance.
Then, we calculated ROC curves for the specificity and sensitivity of the identified variables as biomarkers (Fig. 5A).As expected from the above data, the prediction score for subjects with and without obesity was extremely high, with a score of 0.99.However, the prediction score of the patients with obesity and people with severe obesity when compared to each other and to the people without obesity were lower at 0.72 and 0.69, respectively, indicating that the discrimination between the two groups of patients with obesity was weak.In addition, a network of signatures was calculated for the prediction of the people without obesity, and showed again the importance of the Burkhoderiaceae family (Fig. 5B) while the Random Forest analysis also classified the Morganellaceae as the most discriminant taxon, although at a very low abundance (Fig. 5C).By using different statistical approaches altogether, we could hierarchize the discriminant taxa from the three groups (Table 4).Although we did not aim at identifying whether these signatures were from live bacteria, we suggest that they may have interacted with the host at some point during the development of the tissue and the obesity.To explore their potential role, we run prediction analysis software to identify potential metabolic pathways inferred from the taxonomy and the public databases.

Function and pathway predictive metagenomics analyses from adipose tissue bacteria 16SrRNA gene
The bacterial 16SrRNA gene is a biomarker allowing the identification of bacteria at different taxonomic levels from OTU analyses Table 5.By using currently enriched online databases, we predicted the full genomic content and corresponding relative abundances from the OTU abundance table.We then quantified the predicted enrichment within gene functions and pathways associated with the identified OTUs.PICRUSt2 analyses were performed for these predictions.The corresponding results should be considered as working hypotheses.
Firstly, a PCA plot was created to visualize the pathways per BMI group (Fig. 6A).No clear discrimination was observed between the three groups.Therefore, we analyzed the PCoA based on Manhattan Distances between individuals which was more discriminant and identify pathways statistically different between people without obesity and patients with obesity or people with severe obesity (Fig. 6B-D).These results allowed us to perform Partial Least Square Discriminant Analysis (PLS-DA) (Fig. 6E).Since the PLS-based model relies on prediction distances, which can be seen as a determined optimal cut-off, we quantified sensitivity and specificity indexes from a ROC analysis.As expected from the above data, the index of prediction of the people without obesity was extremely high with a score of 0.96 while the prediction score of the patients with obesity and people with severe obesity were lower 0.71 and 0.67, respectively (Fig. 6F).
Finally, a Network analysis displayed relevant positive and negative associations with the BMI of the people without obesity only.We calculated correlation values between Pathways and BMI, as variables, in a pair-wise manner as represented by the scalar product value between every pair of vectors in dimension length representing the variables (Pathways and BMI) on the correlation "circle" (Fig. 6G).Pathways with a similarity value above of 0.5 and below −0.5 were plotted.The following vitamin-associated pathways i.e., PWY-6269, PWY-5509 and COBALSYN-PWY (B12-vitamin) are positively associated with the people without obesity while the PWY-6165 (chorismate synthesis) and GOLPDLCAT-PWY, involved in the propanediol biosynthesis, were negatively associated.We compared such analyses with the Random Forest analysis and showed again that the PWY-6165 pathway was also strongly discriminating between the people without obesity with the patients with obesity (Fig. 6H).Furthermore, we performed a volcano plot analysis which identified pathways with decreased or increased abundances in people without obesity, when compared with the patients with obesity (Fig. 6I) or people with severe obesity (Fig. 6J).We similarly identified the PWY-6165 as highly associated with the patients with obesity and people with severe obesity.Interestingly, this pathway is responsible for the synthesis of chorismite, which is a key substrate of aromatic amino-acid biosynthesis.In an attempt to suggest some mechanisms to the identified pathway we treated mice for 1 month with the bacterial precursor of chorismate i.e., the shikimate.However, no change on body weight was observed suggesting that the pathway was not relevant or that the experimental design was not appropriate (not shown).
We then represented all discriminant pathways on the metabolic map of Ipath3 database.We found that the people F analysis of similarities between groups (ANOSIM) where the dissimilarity is shown on the Y axis for each group and between the 3 groups.The dissimilarity between groups and the similarity within the people without obesity are significant.without obesity was associated with galactose metabolism while several pathways implicated the metabolism of aromatic aminoacids, notably phenylalanine, tyrosine, and tryptophan biosynthesis, in the patients with obesity (Fig. 6K).Interestingly, we also observed an increased abundance in pathways involved in valine, leucine and isoleucine biosynthesis i.e., branched chain amino acids (Table 6) in both groups of patients with obesity as well.Additionally, we correlated such pathways with the clinical parameters and showed positive correlations between BMI and PWY 6269/5509/COBALSYNTH (B12 vitamin) and 6143/5651 while the body weight was correlated positively with 7446/GOLD-PDLCAT (Fig. 6L, M).Some negative correlations were observed as well between pathways and clinical parameters (Fig. 6L, M).Interestingly, two major clusters of pathways were correlated with the height suggesting an important interaction between growth and gut adipose tissue microbiota.
In summary, our analyses show clear evidence of discriminant taxonomic and metabolic signatures associated with obesity status.

DISCUSSION
This study is the first to infer molecular pathways from adipose tissue 16SrRNA signatures that are specific to people without obesity.While alpha diversity of 16SrRNA gene sequences was not found to be a clear discriminant feature of the adipose tissue signatures using discriminant analyses, specific taxonomic signatures were identified, including the total absence of the Moganellaceae family from to people without obesity groups.The Burkholderiaceae, Yeasinieaceae, and Xanthomonadaceae were highly represented and part of the discriminant signatures.Furthermore, molecular pathways were inferred from the 16SrRNA gene sequences, revealing differences in the biosynthesis of B12 vitamins and chorismate, a precursor of aromatic amino-acid biosynthesis.
Over the last decade numerous pieces of evidence from our and other groups demonstrated the existence of adipose tissue bacterial 16SrRNA gene sequences far above what observed for contaminants [8][9][10][11][12].We previously analyzed the accumulation of 16SrRNA in the adipose tissue by qPCR following the natural    colonization of the germ-free mice within a conventional environment.It took 48 h for bacteria to colonize the visceral fat depots (not shown).Eventually, the oral administration of E. coli expressing ampicillin and a GFP genes to high-fat diet-fed mice leads to the accumulation, in different adipose depots, of ampicillin resistant and GFP-expressing E. coli that can be isolated on ampicillin containing plates [22].In a therapeutic area different from that of metabolic disease such as cancer, numerous evidences demonstrated the existence of intra-tissue intracellular live bacteria [23][24][25][26][27][28].The authors have identified live bacteria, notably from breast tumors, which were isolated and showed their causal role in the controlling chemotherapy efficacy.In addition, they used non-cancerous tissues as controls and identified live bacteria as well which were different from those found in the tumors.The existence of adipose tissue bacterial DNA in humans [11,18,29] and rodents [9,10] has been described by us and others.Importantly, since other independent groups [12,13,30] have reproduced our discoveries, this last argument also contributes to validating the existence of bacterial DNA in adipose depots, and other tissues as well far way above contaminants.This was true in humans and in rodent models.These arguments and the important controls performed in our study demonstrate that the bacterial DNA identified in our study is not related to any technical or surgical contaminant risk.Therefore, this specific bacteria DNA signature led us to suggest that bacterial antigens, and functions, and/or live bacteria could interact with the host cells notably with those from the stroma vascular fraction of adipose depots imprinting the adipose development.We challenged this hypothesis and observed 16 SrRNA signatures in the visceral adipose depot associated with the BMI.Previous studies showed in different cohorts that the adipose fat pads were characterized with specific signatures of bacterial DNA associated with inflammation and type 2 diabetes [12,30].They convincingly report the presence of bacteria through histological observation such as Fluorescent in Situ Hybridization.Such bacteria were associated with inflammation and type 2 diabetes biomarkers [12,30].While they report a signature mostly composed of Proteobacteria and Firmicutes we further show that associated with elevated BMI, Actinobacteria are also present.Since our bacterial DNA signature was not associated with type 2 diabetes (not shown), we could suggest that Actinobacteria could be a specific signature of in patients with obesity.Such bacterial DNA signature was also observed by others [13].They show that type 2 diabetes signatures were most evident in mesenteric adipose tissue, in which individuals with diabetes displayed reduced bacterial diversity concomitant with fewer Gram-positive bacteria, such as Faecalibacterium, as opposed to enhanced levels of typically opportunistic Gram-negative Proteobacteriaceae.Hence, we here confirmed that the visceral adipose fat pad is associated with bacterial DNA signatures of metabolic diseases.According to the metabolic factor considered some taxa are discriminant.We here observed that numerous gram-negative bacteria the Burkholderiaceae, the Yersiniaceae, the Morganellaceae and the Xanthomonadaceae were strong discriminants of the patients with obesity and people without obesity.Specially, Morganellaceae were not observed in any patient from the people without obesity while present in more than 80% of the patients with obesity.Previous report shows that genera belonging to the Morganellaceae family, when indicate high levels of Morganellaceae often cause infections in the immunocompromised hosts, such as diabetic patients [31].However, it is possible that the accumulation of LPS determinants from these families in fat pads could contribute to adiposity development, as observed in rodents [6].We have also observed a slight increase in the alpha diversity (within-group) of bacterial DNA sequences in patients with obesity, along with increased intestinal permeability that characterizes both patients [32] and animal models [3,[33][34][35][36] with metabolic diseases.This increased intestinal permeability is strongly suggestive of an increased bacterial translocation through the gut epithelium, which our group initially in metabolic diseases [8].It is worth noting that while the fecal microbiota is mostly composed of Firmicutes and Bacteroidetes, the adipose microbiota is composed of Proteobacteria, Firmicutes, and Actinobacteria this observation implies that there is an intestinal filter which enriches and selects for Proteobacteria in the adipose depots.This conclusion is supported by our initial report, which showed reduced IL17-producing cells in the ileum lamina propria, resulting in lower defensin production, such as RegIII, and a weaker nonspecific barrier against commensals in mice [37], confirmed by others [38] and in humans [39].
Another important step forward to the identification of the potential mechanisms associating changes in adipose bacteria  signature and obesity is related to the bacterial metabolic pathways inferred from the metagenomics analyses.Thanks to novel algorithms PICRUSt2, we can predict bacterial genes and molecular pathways characterizing a set of 16 SrRNA genes from taxonomic signatures and public metagenomic databases.From numerous approaches such as PLS-DA, Random Forest, Volcano and Network analyses we identified a handful of inferred molecular pathways associated with obesity.Notably, the PWY-6165 which leads to the production of Shikimate then Chorismate using substrate from the glycolysis and the aspartic amino acid.
Showed an increased gene richness in patients with obesity.Interestingly, recent observations predicted a reduced chorismate content in the gut in mice treated with a functional seaweed "codium fragile" as an anti-obesity agent [40].Additionally, maternal obesity alters the human milk metabolome in a way that the shikimic acid predicted higher infant adiposity over the first 6 months of life [41], further suggesting that this pathway is involved in obesity development at least during infancy.It is noteworthy that the shikimate/chorismate pathway is important in the production of aromatic amino acids such as tryptophan [42], which has been reported to be a key intestinal substrate of bacteria through the indole pathway while depleted in patients through the kynurenine pathway, notably in patients with cardiometabolic diseases [43].Along the same line of investigation, we identified numerous microbial pathways encoding for the biosynthesis of the B12 vitamin which are positively associated with non-obesity.It is noteworthy that gynoid obesity is as well negatively correlated with vitamin B12 biosynthesis [44].This finding has been as well documented in humans with hepatic steatosis [45].Interestingly, the lipid lowering treatment silymarin induced bacterial B12 production in male rats but not in male germ-free mice [46].In humans the delta changes of serum B12 associate negatively with the fluctuations of serum triglycerides, thereby suggesting the key role of microbiota on the metabolism of B12 vitamin notably in patients with dyslipidemia and obesity.Our data could suggest that part of this metabolism could be due to adipose tissue bacteria which when reduced in frequency would no longer supply the adipose depot with B12 vitamin.Such observations have been many times relayed in literature and comfort our hypothesis [47].Eventually, the gut microbiota is involved in the biosynthesis of B12 vitamin.A low vitamin B 12 in plasma of women with obesity coincides with the decreased presence of B vitamin-producing bacteria and increased presence of inflammatory-associated bacteria [48,49].It is noteworthy that the patient with obesity and severe obesity may have different levels of insulin resistance thereby predictive of diabetes.It would be of importance in other studies to evaluate the signature of tissue microbiota along with features of insulin resistance to potently predict type 2 diabetes.We did identify microbial pathways related to LPS biosynthesis as part of the signatures in patients with obesity as reported in [4].Altogether, we here add to the knowledge a specific signature of bacterial 16 SrRNA gene in the visceral adipose depots of patients with obesity.Inferring the corresponding metabolic pathway suggest potential mechanisms to be evaluated in rodent models.
Fig. 6 Multivariate analyses of the Pathways inferred from OTUs.A PCA of pathways; B PCoA of Manhattan distances; C, D box plots showing statistical differences between groups for each component according to patients classified per BMI group.E Sparse Partial Least Square Discriminant Analysis (sPLS-DA) of the OTUs.Ellipses are shown for the group distribution.Each axis shows the % of variance for the two major components used in the discrimination.F ROC curve analyses of pathways as variables predicting people without obesity, patients with obesity, and people with severe obesity (BMI groups).The score of prediction of each group versus the others are calculated and the corresponding curves are represented.G Network between the most predicting pathways of the group of people without obesity.H Best discriminant predictors of all groups as analyzed by Random Forest analysis; the most discriminant pathway is circled.I, J Volcano plot graphical representation of the log10 of the p-value of the most upregulated and downregulated pathways as calculated by log2 fold change for the people without obesity vs patients with obesity (I) and patients with obesity vs people with severe obesity (J), patients respectively.The consensus discriminant pathway is circled.K Graphic representation of the discriminant microbial metabolic pathways (green: patients with obesity and people with severe obesity; blue: people without obesity).L Heatmap of correlations (−0.44 to 0.44) between of metabolic pathways and clinical parameters and linear regression analyses between pathways M.

Fig. 1
Fig. 1 Principal component analysis of the clinical features of the cohort of patients.A Principal component analysis (PCA) biplot of anthropomorphic and biochemical variables.The biplot shows the PCA scores of the clinical variables as vectors (in black arrows).Individuals of each BMI group are represented in Gray circle for people without obesity, blue circle for patients with obesity and Orange circle for people with severe obesity.The variance scores of the first two dimensions are shown on the axes.B Importance of each clinical feature within the 5 first components (Dim.).The size of the dots and the color intensity are proportional to the importance of the clinical variables within each component.The corresponding scale is shown.C Barplot of the most contributing variables in principal component 1 (Dim1).The contribution of variables in accounting for the variability in Dim.1 the principal component analysis are expressed in percentage.Variables that are correlated with.

Fig. 2
Fig. 2 Alpha and beta diversity indexes.A Box plots of alpha diversity indexes per group.B representation of the Boxplots of diversity indexes Chaos and Fisher, which are the only indexes with statistical significance as analyzed by Kruskal-Wallis tests on non-parametric values.Differences between groups are shown when P < 0.05.C Principal Coordinate (1-2) Analysis of Manhattan Distances.The variances of each coordinate are shown on axes.Individuals of each BMI group are represented in Gray circle for patients without obesity, blue circle for patients with obesity and Orange circle for people with severe obesity.D, E barplot analyses corresponding to both components and for each group.F analysis of similarities between groups (ANOSIM) where the dissimilarity is shown on the Y axis for each group and between the 3 groups.The dissimilarity between groups and the similarity within the people without obesity are significant.

Fig. 3
Fig. 3 Individual patient relative abundances and classification.Stacked barplots of the mean relative abundances (%) at the phylum (A), and family (C) taxonomy levels.Right: Stacked barplot of the top relative abundances (B phyla; D families).Left: dendrogram representation (hierarchal classification) of Bray-Curtis similarity-based distances between the 62 individuals (B phyla; D families).Only OTUs with relative abundance >0.1% were considered.The patient groups are indicated with colored circles: 16 people without obesity (Gray), 20 patients with obesity (Blue), 16 people with severe obesity (Orange).E-G Boxplots (relative abundance) of some individual taxa (E Burkholderiaceae, F Yersiniaceae, G Xanthomonadaceae). Statistical analyses (Kruskal-Wallis test) were performed on non-parametric values.Differences between groups are shown when P < 0.05.H-L barplot analyses of relative abundances of some families represented per patient and per group.

Fig. 4
Fig. 4 Multivariate analyses.A sparse Partial Least Square Discriminant Analysis (sPLS-DA) of the OTU signatures.Ellipses are shown for the group distribution.Each axis shows the % of variance for the two major components used in the discrimination.B circle plot showing the projection of the main discriminant variables at the family taxonomic level.Alcaligenaceae, Morganellaceae, Xanthomonadaceae, Yersiniaceae, were negatively associated with component 1, while Burkholderiaceae were positively associated.C Random Forest classification of taxa at the family taxonomic levels.The variables were attributed to groups according to their index of correlation either positive (right) or negative (left).D heatmap of correlation indexes (from −0.65 to +0.65) between variables and clinical parameters.The taxa associated with body weight indexes are boxed and showed in the upright inset.

Fig. 5
Fig. 5 Summary of specific group signatures.A ROC curve analyses of variables predicting people without obesity, patients with obesity, and people with severe obesity.The score of prediction of each group versus the others are calculated and the corresponding curves are represented.B network of correlations (−0.58 to 0.58) between the most predicting taxa of the people without obesity.C best discriminant taxon predictors of all groups.

Table 1 .
Anthropomorphic and clinical characteristics of patients, as means +/− standard deviation.

Table 2 .
Relative abundances.Relative abundance of the top 10 bacteria at the Phyla level of taxonomy *Statistically different when Wilcoxon P < 0.05.Phyla (%) in groups of patients and statistical significance between groups.
O: patients with obesity, E: people with severe obesity, N: people without obesity.*Statistically different when Wilcoxon P < 0.05.**Statistically different when Wilcoxon P > 0.01.

Table 5 .
Discriminant molecular pathways as a signature of people without obesity and patients with obesity.

Table 4 .
Summary of the discriminant taxa identified by multiple multivariate analyses.