Microbiome Analysis and the Firmicutes/Bacteroidetes ratio
The 16S-mapped reads obtained from all analyzed microbiome samples were 211854+/-56420and resulted in the identification of 1797+/-400 operational taxonomic units (OTUs) in the simple obese patients (OB-G),1761+/-509 OTUs in the obese with complications (OBc-G) samples and 1747+/-602 OTUs in normal weight Healthy Donors (nwHD). A visual analysis of the Firmicutes and the Bacteroidetes phyla abundances and their Firmicutes/Bacteroidetes (F/B) ratio indicated that the samples from obese patients had always a lower F/B-ratio value than the group of nwHDs, indicating that F/B ratio is not a biomarker for obesity, differently from it was proposed by some authors11,12. These data are confirmed both by the analysis of the consensus of 7 16S-polymorphic regions (F/B-ratio: Obese:1.2+/-0.8 vs nwHD:1.6+/-1.8, respectively) and by analyzing the V4-region alone (F/B-ratio: Obese:0.9+/-0.7 vs nwHD:1.2+/-1.3, respectively). It is to be stressed that the V4 region is the more common 16S polymorphic region reported in the literature on metagenomics.
Alpha And Beta Diversity Analysis
Different alpha diversity profiling indices have been used that estimate either the community richness (Chao1-index) or richness and evenness (Shannon or Simpson indexes). It is important to highlight that none of the analyzed alpha diversity indexes reached statistical significance in the comparison between the complete casistics of obese patient (55 fullOB) and the 25nwHD. In detail, the trend ofChao1 mean values were slightly higher in pathological samples (143.2+/-22.3) rather than in nwHDs (141.5+/-30.1) (Supplementary Fig. 1a). Simpson (Si) and Shannon (Sh) indexes behaved in the opposite way, where both slightly higher in nwHDs (Si: 0.946+/-0.014 - Sh: 5.057+/-0.320) rather than full OB (Si: 0.87+/-0.10- Sh: 2.83+/-0.45) (Supplementary Fig. 1a). The Bray-Curtis index used for beta diversity analysis clearly showed statistical significance (PERMANOVA p < 0.02) in the comparison between full OB and nwHDs, thus indicating a difference in microbiome composition between patients and normal weight controls (Supplementary Fig. 1b).
Comparative analysis: fullOB vs nwHD.
The specific associations between taxa present in fullOB and nwHD were analyzed using the Calypso13,14 package which considers the sparsity (i.e., a dataset with many values equal to zero) and the compositional origin of microbiome data. Specifically, we used the sparse Partial Least Squares-Discriminant Analysis (sPLS-DA)14 that associates the importance of a specific taxon to describe a group of samples (Fig. 1a-b). The results indicated that among the first 5 genus ordered for importance, 3 of them showed a positive correlation with fullOB (Streptococcus, Acidaminococcus and Sutterella with importance scores of 0.40, 0.39, and 0.36, respectively). The relative abundance analysis, using 4 different algorithms within MicrobiomeAnalyst, and the sparse correlation for compositional data (SparCC) analysis confirmed this result (Fig. 2a and Supplementary Table 1). A positive correlation with fullOB was found for Sutterella wadsworthensis and Streptococcus thermophilus, both characterized by 10 times higher abundances in obese patients compared to nwHD (Supplementary Table 1). Prevotella genus and Prevotellacopri, were positively correlated with fullOB rather than nwHD (Fig. 2a and Supplementary Table 1). The Microbial Dysbiosis index (MD-index) for this comparison is 1.5764 (Fig. 2a) indicating a high imbalance (dysbiosis) in the microbial flora of obese patients compared to the one of controls (nwHD).
Comparative analyses: OB-G vs nwHD and OBc-G vs nwHD.
The analysis of the more relevant genera using supervised Random-Forest indicated that Streptococcus was a genus tightly linked to OB-G, followed by Sutterella, Clostridium, and Lactobacillus (Fig. 2b and Supplementary Table 2). The species that showed a positive association with OB-Group of samples were Sutterellawadsworthensis and Blautiaproducta (Supplementary Table 2). Both the Sutterella genus and the Sutterella wadsworthensis species have confirmed a strongerassociation with OB-G rather than OBc-G,considering the number of different methods reaching statistical significance and also inthe analysis of relative abundances(Fig. 2b-c and Supplementary Table 2–3). The same was true for the Streptococcus genus and their descendent species that were better described in the OB-Gof patients (Supplementary Table 2). Random-forest's supervised analysis indicated that the genera Lactobacillus, Gemminger, and the species Coprococcus comes and Bacteroides massiliensis showed higher "mean decrease accuracy" and thus a positive association with the OBc-G patients (Supplementary Table 2). More in detail, in relative abundance analysis the Bacteroides massiliensis showed a statistically significant value only in the comparison between OBc-G and nwHD (Supplementary Table 3), but not in the comparison between OB-G and nwHD (Supplementary Table 2). On the contrary, Coprococcus comes showed a positive association with both OB-G and with OBc-G group of patients (Supplementary Tables 2–3).
It is to be noted that some Bacteroides genera behave in opposite ways, indeed Bacteroides fragilis, Bacteroides plebeius, and Bacteroides thetaiotaomicron showed higher abundances in OB-G patients compared to nwHD. While, on the contrary Bacteroides faecisandBacteroides massiliensis showed a positive association with OBc-G compared to nwHD (Supplementary Tables 2–3). The MD-index behaves in opposite ways in the OB-G and OBc-G correlation networks respect to nwHD. Indeed, MD-index computed for OB-G was 1.3993 showing high dysbiosis, while in OBc-G was − 0.2786 (Fig. 2b-c, respectively) which underline still a dysbiosis but much less evident as for OB-G samples.
Supervised Random Forest analysis and normal weight healthy donors (nwHD).
We applied the Random Forest machine-learning algorithm15 to identify taxa able to discriminate between patients with obesity and normal-weight donors with good classifications performances. We already described the correlations between specific taxa and the OB-G or OBc-G. Here we analyzed taxa closely that classify the normal-weight healthy donor group. Alistipes genus, different Alistipes species, and the Bifidobacterium longum were strongly associated with nwHD (Supplementary Tables 1, 2 and 4). In addition, the genus Akkermansia and its descendants Akkermansiamuciniphila were mainly associated with nwHD (Supplementary Tables 1 and 4). Among Blautia species a different behavior was observed for Blautiawexlerae and other Blautia species since it appears closely linked to nwHD rather than obese samples (Supplementary Tables 3 and 4).
Correlation Analysis Of Taxa Clusters With Physiological Parameters
In the analysis of taxa related to samples of OBc-G, we found interesting positive correlations between some clusters of species and the total cholesterol, LDLlevels (Figs. 3a-c and Supplementary Tables 5a) and CRP (Figs. 3a-b and Supplementary Tables 5a-b). Specifically, the increase of the species enclosed in Cluster 5 is significantly associated with the increase ofboth total cholesterol and LDL levels; seeminglythe increase of the species enclosed in Clusters 18 (Fig. 3a) and 14 (Fig. 3b), is significantly associated with the increase of CRP, which is a marker of inflammation. On the contrary, Glycemia (0', 60' and 120') showed a negative correlation with the clusters of species 20, 21 (Fig. 3a) and 11 (Fig. 3b).In the comparisons of the species analyzed entirely versus the same species analyzed after the feature reduction (Figs. 3a vs. 3b), we found common species (i) positively associated with CRP and (ii) negatively associated with glycemia(0',60',120'): specifically we found (i) Alistipesindistinctus, Clostridium innocuum, Desulfovibriopiger, Prevotellaruminicola and Prevotellain common between clusters 18 and 14, while (ii) Acidaminococcusfermentans, Clostridium cocleatum and Clostridium ramosumin common between clusters 20–21 and cluster 11 (see Supplementary Tables 5a-b).While the connection of Alistipesindistinctusand Clostridiuminnocuumwith obesity or with clinical parameters related to obesity is not known in the literature, Desulfovibriopiger, Prevotellaruminicola and Prevotellaspecies are already known to be associated with inflammation, insulin-resistance, hyperglycemia and type 2-diabetes16.It is important to highlight that Desulfovibriopiger (Clusters 6 for OB-G and Clusters 14, 18 forOBc-G) showed that the increase of abundance of this species is always associated with the increase in the value of some clinical parameters critical for obesity (such as TRG for OB-G and CRP for OBc-G, respectively).
Interestingly enough, inthe clusters of taxa found negatively correlated with glycemia (Figs. 3d-e, and Supplementary Tables 5d-e) we found that the family Oxalobacteraceae and two genera descending from it, namely Herbaspirillum and Oxalobacter, were already known to be associated with a decreased of insulin-resistance and glycemia17.
Finally, in the analysis performed on the nwHD group, BMI-SDS showed a negative association with five species, enclosed in Cluster 19 (Fig. 3f and Supplementary Tables 5f). This data strongly suggests that BMI-SDS could represent the most sensible clinical parameter to correctly classify children and adolescents as normal weight, overweight, and obese subjects and that the species belonging to this cluster could be considered protective against obesity.
Inferring Functional (Metabolic) Pathways Characterizing Ob-g And Obc-g
At least 20 different metabolic pathways, inferred with PICRUSt2 and present in the MetaCycdatabase18, showed statistical significances in the comparison between OB-G vs nwHD (Fig. 4a). Different pathways involved in amino acid biosynthesis showed a positive correlation with obesity in pediatric patients. Thus, pathways involving the phenylalanine (PWY-6628) and the tyrosine (PWY-6630) aromatic amino acids were among the entries with the highest statistical importance. In addition, tryptophan (PWY-6629) and methionine biosynthesis (HSERMETANA-PWY and HOMOSER-METSYN-PWY) showed higher abundances in OB-G, while an additional pathway (PWY-5345) for methionine biosynthesis showed an opposite behavior. More, different pathways involved in polyamine biosynthesis known to play a role in bacterial pathogenicity and biofilm formation showed higher abundances in the microbiome of OB-G. Among them, the POLYAMINSYN3-PWY is the one showing the higher importance of all pathways analyzed by Random-Forest analysis (mean decrease accuracy 0.006). Others like POLYAMSYN-PWY and the pro-inflammatory lipopolysaccharide19pathway (LPSSYN-PWY) showed higher abundances in OB-G compared to lean subjects. More, the second in order of importance in the Random-Forest analysis (mean decrease accuracy 0.004) was the super-pathway of (R,R)-butanediol biosynthesis (P125-PWY) which was overabundant only in nwHDs.
Obesity in adolescents is often associated with clinical complications such as insulin resistance, hyperglycemia, dyslipidemia, and hypertension, which together are termed "metabolic syndrome". Regarding this issue, we found 3 pathways more abundant in OBc-G with respect to nwHD: PWY-341 (glycolysis V), PWY-5532 (adenosine nucleotides degradation), and PWY-6478 (GDP-D-glycero-alpha-D-manno-heptose biosynthesis) (Fig. 4b). The heptose-sugars are components of bacterial cell surface common in the pro inflammatory lipopolysaccharide (LPS)20.