This study comprises four distinct cohorts designed to analyze the metagenomic profiles of gut microbiota in obese individuals before and after bariatric surgery across different geographical locations. The study population undergoing weight loss interventions excluded individuals with basal metabolic diseases, cancer, and other related conditions. The samples were derived from consistent tissue types and uniform sequencing methods were employed. After screening, a total of four cohorts were included in the study ( Table 1 ), as detailed below:

Cohort 1 - PRJEB12123 (Liu et al., 2017). This cohort includes 256 fecal samples collected from Beijing, China. The samples include 105 healthy individuals, 88 obese individuals, and 63 individuals at various stages post-Sleeve Gastrectomy (SG). Specifically, the post-SG samples are categorized into SG 0M (23 samples), SG 1M (17 samples), and SG 3M (23 samples). This cohort uses paired-end sequencing with a read length of 100bp.

Cohort 2 - PRJNA597839 (Nie et al., 2020): Comprising 76 fecal samples from the USA, this cohort focuses on the longitudinal analysis of patients undergoing laparoscopic Sleeve Gastrectomy (LSG). It includes 30 healthy individuals and 46 obese individuals pre- and post-surgery. The samples collected include LSG pre (32 samples) and LSG 6M (14 samples). Sequencing is performed with paired-end reads of 150bp.

Cohort 3 - PRJEB12947 (Palleja et al., 2016): This cohort consists of 33 fecal samples collected from Beijing, China, specifically focusing on patients undergoing Roux-en-Y gastric bypass (RYGB). This cohort focuses on longitudinal samples from patients undergoing RYGB, without healthy controls: RYGB pre (13 samples), RYGB 1M (1 sample), RYGB 3M (12 samples), and RYGB 12M (8 samples). Paired-end sequencing with 2x100bp reads is utilized for this cohort.

Cohort 4 - PRJNA668357 (Fouladi et al., 2021): The final cohort involves 135 fecal samples from the USA, targeting the analysis of gut microbiota in patients undergoing RYGB surgery. This cohort includes 52 obese individuals and post-surgery samples: RYGB pre (38 samples), RYGB 1M (0 samples), RYGB 3M (27 samples), and RYGB 12M (18 samples). The sequencing for this cohort is performed with paired-end reads of 2x150bp.

Sequence read archive (SRA) files of all samples were downloaded using prefetch software. Raw sequence data were subjected to quality control using fastp (https://github.com/OpenGene/fastp). Human-derived sequences were removed by aligning the reads to the human genome using Bowtie2 (v2.3.4.3). Next, these non-human sequences were aligned to a reference database with Kraken2 (Wood et al., 2019). As a reference database, we used the Kraken-build utility to download bacterial, archaeal, viral and fungal libraries including National Centre for Biotechnology Information (NCBI) taxonomic information as well as complete genome sequences from RefSeq (Jing et al., 2021; Nearing et al., 2022; Sola-Leyva et al., 2021). In addition, we also incorporated the human and protozoa genome into the Kraken2 database to further reduce false-positive microbes introduced by human-derived or other eukaryotic sequences. Based on the effective clean data volume from the sequencing of each sample, we used the RPTM (Reads Per Ten Million) normalization method to standardize the data at the species level, followed by subsequent diversity and differential species analyses. Besides, the filtered reads were then assembled denovo using megahit (https://github.com/voutcn/megahit). Metagenome-Assembled Genomes (MAGs) were constructed using metaWRAP (https://github.com/bxlab/metaWRAP). The bin refinement module in metaWRAP was employed to optimize MAG quality, followed by dereplication using dRep (https://github.com/MrOlm/drep,v3.2.2) to remove redundant MAGs. Taxonomic classification of the MAGs was performed using the classify-wf module in GTDB-Tk. Genomic k-mer distances among MAGs were calculated using mash (v2.3, https://mash.readthedocs.io/en/latest/). A phylogenetic tree was constructed with the Neighbor-Joining method (nj) using the ape package in R. The resulting tree was visualized using the Interactive Tree Of Life (iTOL) web tool. Genes within the MAGs were predicted by prodigal (v2.6.3) and annotated by aligning them to the CAZy and KEGG databases using the blastp module of DIAMOND (https://github.com/bbuchfink/diamond). Functional genes were identified and their abundances were quantified for each MAG. CoPTR (v1.1.6) was used to compute peak-to-trough ratios (PTRs) from MAGs to accurately reflect microbial growth rates. Gapseq (v1.3.1) was used to predict metabolic processes (Zimmermann et al., 2021). Prediction of secondary metabolites in each MAG was performed using antiSMASH (v7.0.1).

The effects of different cohorts, sample locations, and experimental groups on microbial community composition were assessed using the PERMANOVA test in the vegan package of R (p-value < 0.05 was considered significant). Differential abundance testing to identify differentially abundant species between obese and healthy individuals (OB-Healthy) and between obese individuals pre- and post-treatment (OB-Treat) employed three methods: edgeR v3.36.0 (FDR < 0.05, FC = 5), ANCOM v2.1 (detection threshold 0.9), and LEfSe v1.0 (logarithmic LDA score > 2). Core differential species were identified based on consensus results from these methods across multiple cohorts, ensuring robust identification of key species affected by obesity and treatment. The Kruskal-Wallis test was used to compare PTR differences between groups. Differential KOs were subjected to KEGG enrichment analysis using R clusterProfiler. We calculated the within-sample α-diversity using Shannon and ACE index on species level to estimate the richness of samples using QIIME (Caporaso et al., 2010) v1.9.1. Besides, the post-surgery α-diversity index increased ratio refers to the proportional increase in α-diversity index after surgery compared to preoperative levels. Specifically, we calculated this ratio by: (postoperative α-diversity index - preoperative α-diversity index)/preoperative α-diversity index for each cohort. Variance partitioning analysis was performed by varpart function of vegan packages.

Three surgical methods—SG, LSG, and RYGB—all enhanced the α-diversity index of the gut microbiota, which was consistent with previous studies (Liu et al., 2017; Paganelli et al., 2019)( Figures 1A, B ; Supplementary Table S1 ). Additionally, even within the same group (Healthy or OB) across different cohorts, significant differences were observed in alpha diversity ( Supplementary Figure S1 ). To compare the effects of different surgical methods on gut microbiota diversity, for each cohort, we computed the ratio of the post-surgical α-diversity index increase relative to the OB group. The results showed that, in terms of species richness (ACE index), LSG > RYGB > SG. From the perspective of both species richness and evenness (Shannon index), RYGB > LSG > SG ( Supplementary Table S2 ).

Principal Coordinates Analysis (PCoA) and PERMANOVA can be used to assess the impact of different studies and confounding factors on microbiota outcomes (Wang et al., 2020). The analysis revealed that cohort locations and surgical interventions significantly influence microbial community structure, with the impacts of obesity and surgical intervention being the most pronounced (indicated by higher R² values). Figure 1C shows significant differences between the four cohorts (PRJEB12123, PRJEB12947, PRJNA597839, and PRJNA668357), with an PERMANOVA R² value of 0.03539 and a p-value of 0.001, indicating significant grouping by cohort. Figure 1D illustrates the distribution of microbial communities across various health and treatment statuses, including healthy individuals and those undergoing LSG, RYGB, and SG. The analysis yielded a PERMANOVA R² value of 0.06557 and a p-value of 0.001, indicating significant differences among these groups, with the impact of surgical treatment (a higher R²) being more pronounced than the influence of cohorts. To further validate the robustness of PERMANOVA results, we conducted variance partitioning analysis (VPA) ( Figure 1E ), which confirmed that the dominant effect of Treat is not influenced by the collinearity of project cohorts or country site, while the PERMANOVA R² of country site is inflated due to collinearity. Specifically, Treat exhibited an independent contribution of 6.75% (after excluding project cohorts and country site), whereas project cohorts contributed only 0.77% (after excluding country site and Treat), and country site showed no significant independent contribution (after excluding project cohorts and Treat), reinforcing that Treat is the primary variable with significant independent explanatory power. Overall, these results highlight the significant effects of cohort and surgical treatment on gut microbiota diversity.

Due to the significant impact of different cohorts on gut microbiota, and to uncover significantly enriched microbes for subsequent MAGs analysis, we compared the results by three differential analysis methods for each cohort to identify differential species which enhances the accuracy of differential species analysis. For LEfSe differential analysis method: using LDA > 2, 51 differential species were identified between the OB and Treat groups, and 32 differential species were identified between the OB and Healthy groups. For ANCOM differential analysis method: using a detection threshold > 0.9, 32 differential species were identified between the OB and Treat groups, and 16 differential species were identified between the OB and Healthy groups. For edgeR differential analysis method: using FDR < 0.05 and fold-change > 5, 542 differential species were identified between the OB and Treat groups, and 244 differential species were identified between the OB and Healthy groups. The three differential analysis methods yielded varying results, with the number of identified markers ranked from highest to lowest as follows: edgeR, LEfSe, and ANCOM. However, there was relatively little overlap between the results of the three methods ( Figure 2 ). Specifically, the three methods identified only 3 common markers in the “Healthy vs. OB” group and 7 common markers in the “Treat vs. OB” group, which is significantly fewer than the unique markers identified by each method. This highlights the necessity of integrating different cohorts and multiple differential analysis methods to comprehensively identify differential species.

Theoretically, potential beneficial microbes are more likely to be enriched in healthy or treatment groups compared to the OB group, a strategy also utilized in a published study (Wu et al., 2024). More cohort data and difference methodological support could help minimize the likelihood of false-positive markers. In this context, the more diverse the differential analysis methods and the more supported cohorts, the more reliable the identified biomarkers will be. Supplementary Table S3 lists species that are significantly enriched in the healthy group (16 species in total) and treatment group (11 species in total), focusing on those with the strongest support. For instance, Coprobacter fastidiosus and Lachnospira eligens were identified in two cohorts (PRJEB12123 and PRJNA597839) and supported by three differential analysis methods (edgeR, ANCOM, and LEfSe for the former; and ANCOM and LEfSe for the latter). Other species, such as Bacteroides stercorisrooris, Bacteroides timonensis, and Brotilimocola acetigignens, were identified in two cohorts and supported by edgeR analysis. In the treatment group, several species were identified across four cohorts, indicating strong support for their enrichment post-treatment. For example, Bifidobacterium dentium, Raoultella planticola, and Streptococcus mutans were consistently enriched across all four cohorts (PRJEB12123, PRJNA597839, PRJEB12947, and PRJNA668357), supported by edgeR analysis. Additionally, species such as Veillonella atypica and Veillonella parvula were identified in three cohorts and supported by three differential analysis methods (edgeR, ANCOM, and LEfSe). Similarly, we also analyzed the species enriched in OB group ( Supplementary Table S4 ), including Faecalimonas umbilicata, Fusobacterium varium, Acidaminococcus provencensis, Alkalihalobacillus bogoriensis, Collinsella tanakaei, Ellagibacter isourolithinifaciens, Enterococcus casseliflavus, Fusobacterium ulcerans, Limosilactobacillus mucosae, Peptacetobacter hiranonis, Slackia isoflavoniconvertens, and Turicibacter bilis. Overall, 39 species biomarkers were identified, including 27 microbes enriched in the NonOB_Enrich group and 12 microbes enriched in the OB_Enrich group. These 39 identified species biomarkers could serve as potential beneficial or non-beneficial microbes for future wet-lab functional validation.

To elucidate the genomic mechanisms underlying postoperative species abundance variations, we conducted a MAG construction analysis. We obtained 7,336 high-quality MAGs (completeness >90%, contamination <5%). After dRep-based deduplication, 4,201 unique MAGs were retained. We plotted the assembly completeness and contamination statistics for these MAGs. The quality statistics of the MAGs ( Figure 3A ) show that 69.2% of MAGs exceed 95% completeness, with the majority (57.87%) having contamination rates below 1%. The phylogenetic tree ( Supplementary Figure S2 ) illustrates the diversity and distribution of MAGs across different groups, projects, and phyla. From the overall structure of the phylogenetic tree, MAGs from different phyla are classified into distinct branches based on their evolutionary relationships. The genomes are distributed across different research cohorts and groups, demonstrating overall high genomic diversity. According to Figure 3B , Firmicutes_A has the highest representation across all categories, particularly in the PRJEB12123 project and the Healthy group. Bacteroidota and Actinobacteriota also show significant presence, with notable counts in both the Healthy and Treat groups across multiple projects. Other phyla, such as Desulfobacterota_I, Fusobacteriota, and Verrucomicrobiota, have lower representation but are still present across different categories, indicating a diverse genomic composition.

Based on the significantly enriched microbe lists in Supplementary Tables S3 , S4 , the number of MAGs identified through binning for each bacterium was summarized ( Supplementary Table S5 ; Figure 3C ). The analysis identified 185 gut microbiota MAGs (171 MAGs in the NonOB_Enrich group and 14 MAGs in the OB_Enrich group). These MAGs include five core species in the healthy group: Lachnospira eligens, Bacteroides uniformis, Coprobacter fastidiosus, Desulfovibrio fairfieldensis, and Klebsiella michiganensis. In the treatment group, eight core species were identified: Akkermansia muciniphila, Streptococcus salivarius, Eisenbergiella tayi, Bifidobacterium dentium, Enterocloster aldenensis, Enterocloster lavalensis, Streptococcus mutans, and Klebsiella variicola. Notably, Lachnospira eligens exhibited the highest number of MAGs in the healthy group, followed by Bacteroides uniformis and Coprobacter fastidiosus. In the treatment group, Akkermansia muciniphila had the highest number of MAGs. Additionally, five core species were identified, including Faecalimonas umbilicata, Collinsella tanakaei, Ellagibacter isourolithinifaciens, Peptacetobacter hiranonis, and Acidaminococcus provencensis. The phylogenetic tree results ( Figure 3D ) show that Bacteroides uniformis, Coprobacter fastidiosus, and Akkermansia muciniphila in the NonOB_Enrich group, cluster into one branch, while Collinsella tanakaei and Ellagibacter isourolithinifaciens in the OB_Enrich group, form a separate branch. These species may exhibit certain similarities in their biological functions. We further conducted a literature review to validate the aforementioned microbes as potential beneficial or non-beneficial ones ( Supplementary Table S6 ).

A key characteristic of obesity-related microorganisms is their ability to metabolize carbohydrates, converting specific sugars into propionic acid, lactic acid, or acetic acid. This metabolic activity plays a significant role in modulating the host’s response to a carbohydrate-rich diet, which is particularly relevant in the context of obesity development (Derrien and van Hylckama Vlieg, 2015; Le Barz et al., 2015; Schwiertz et al., 2010). NMDS based on CAZY enzymes count of MAGs ( Figure 3E ) showed a significant separation between microbes in the NonOB_Enrich and OB_Enrich group, indicating distinct carbohydrate degradation and utilization profiles. The bubble plot ( Figure 3F ) further shows the top 50 enzyme families by relative gene abundance across various species. The analysis reveals that key enzyme families, including Carbohydrate-Binding Modules (CBM), Carbohydrate Esterases (CE), Glycoside Hydrolases (GH), GlycosylTransferases (GT), and Polysaccharide Lyases (PL), have a higher relative proportion of metabolic enzyme genes detected in MAGs of the NonOB_Enrich group, such as Bacteroides uniformis, Lachnospira eligens, Akkermansia muciniphila, and Eisenbergiella tayi, compared to the MAGs in OB_Enrich group, implying that they have a stronger ability to metabolize and utilize a variety of complex carbohydrates.

We compared the shared and unique CAZy enzymes detected in microbes between the NonOB_Enrich and OB_Enrich group ( Figure 4A ; Supplementary Table S7 ). The results indicate that microbes in NonOB_Enrich group exhibit a significantly higher number of CAZy enzyme types compared to the OB_Enrich group microbes. The top three enzymes are associated with Bacteroides uniformis, Eisenbergiella tayi, and Coprobacter fastidiosus, while the bottom three are linked to Ellagibacter isourolithinifaciens, Peptacetobacter hiranonis, and Acidaminococcus provencensis. This suggests a greater diversity in the carbohydrate-active enzyme profiles of microbes in the NonOB_Enrich group. Additionally, Bacteroides uniformis possesses the highest number of unique enzymes, followed by a small number of enzyme families shared among some microbes. The relative proportion of genes from GH105, GH2, GH23, GH43, and GT0 gene families is significantly higher in microbes of the NonOB_Enrich group compared to the microbes in OB_Enrich group ( Figure 4B ).

Based on the 74 significantly different KOs ( Supplementary Table S8 ), these KOs were significantly enriched in four pathways: “amino acid biosynthesis,” “glycine, serine, and threonine metabolism,” “phenylalanine, tyrosine, and tryptophan biosynthesis,” and “galactose metabolism” ( Figure 4C ). The result in Figure 4D shows that species enriched in the NonOB_Enrich group produced more secondary metabolites compared to those in the OB_Enrich group. Furthermore, the predominant secondary metabolites produced by different species varied. For instance, Akkermansia muciniphila produced more terpenes, Bacteroides uniformis produced more RRE-containing compounds, and Lachnospira eligens produced more cyclic-lactone autoinducers. Additionally, the clustering results for the NonOB_Enrich and OB_Enrich groups displayed a clear separation trend, with the exception of Bifidobacterium dentium. In healthy human gut microbiota, secondary metabolite genes, such as those involved in NRPS, terpenes, and RiPPs, helped maintain microbial diversity and inhibited excessive pathogen growth. In contrast, in obese individuals, secondary metabolite gene diversity of MAGs was reduced, and the metabolite profile was altered, potentially contributing to chronic low-grade inflammation and metabolic disorders.

Microbes that activate the fatty acid biosynthesis pathway can induce body weight gain (Henneke et al., 2022). We compared the presence of fatty acid biosynthesis genes across species in the OB_Enrich and NonOB_Enrich (each species containing at least three MAGs included in the analysis). The Venn analysis results in Figure 5A show that more genes were detected in the NonOB_Enrich group, with seven unique genes, while both groups shared nine detected genes. This is consistent with a previous report, where the activity of genes involved in fatty acid metabolism is relatively low, particularly those related to enzymes in the tricarboxylic acid (TCA) cycle, with their function being significantly reduced (Liu et al., 2017). Figure 5B shows that the MAGs of different species clustered together, indicating a preference for the fatty acid biosynthesis genes carried by each species. The beneficial gut bacterium Lachnospira eligens carried the most diverse set of genes, including fabG, MCH, accB/bccP, fabK, fabD, accC, fabF, fabZ, accA, and accD. Bacteroides uniformis carried fabG, fabD, fabF, fabI, IpxC-fabZ, fabH, and ACSL/fadD. Akkermansia muciniphila carried fabG, fabD, fabF, fabI, IpxC-fabZ, and fabH. The species enriched in OB_Enrich, Collinsella tanakaei, carried the fewest genes, only including fabF and fabZ. The result in Figure 5C further shows a tendency for separation between the OB_Enrich MAGs and NonOB_Enrich MAGs, indicating a distinct difference in the detection profiles of fatty acid biosynthesis genes between the two groups. Overall, NonOB_Enrich MAGs exhibit higher gene diversity, which may contribute to SCFA production, promote fatty acid oxidation rather than storage, and reduce lipid accumulation, thereby regulating and improving host obesity (Den Besten et al., 2013; Gurung et al., 2020). This may explain why obesity is significantly associated with a reduction in SCFA concentration (Ecklu-Mensah et al., 2023).

We performed a clustering heatmap analysis ( Figure 5D ) to compare the average number of key reactions (z-score normalized) related to the fatty acid biosynthesis pathway, which were predicted from the genome fasta sequences of species in the obesity-associated microbiome group (OB_Enrich) and the non-obesity-associated group (NonOB_Enrich). The analysis revealed that beneficial gut microbes related to short-chain fatty acid production, such as Bacteroides uniformis, Akkermansia muciniphila, Streptococcus salivarius, and Lachnospira eligens, clustered into one subgroup. This indicates that these four species exhibit more similar fatty acid biosynthesis pathway reactions, suggesting potential synergistic effects. However, Coprobacter fastidiosus, Eisenbergiella tayi, and Desulfovibrio fairfieldensis, which were enriched in the NonOB_Enrich group, did not cluster with the other four species mentioned above, indicating distinct metabolic regulation mechanism. The enrichment of these three species in NonOB_Enrich may not directly regulate body weight through fatty acid biosynthesis, but rather exert their effects through other mechanisms, such as improving inflammatory responses.

Microbial growth rate, as reflected by PTR, was linked to disease status and largely independent of relative abundances capturing a unique biological variation that complements relative abundance data (Joseph et al., 2022a). Based on PTR analysis ( Supplementary Table S9 ), we discovered that microbes in the NonOB_Enrich group and OB_Enrich group mainly clustered into two distinct branches ( Supplementary Figure S3A ). The growth rates of the same species exhibited varying distribution patterns across different dimensions, such as geographic regions, study cohorts, and participant groups. Further analysis using differential boxplots revealed that in the PRJNA668357 cohort, Akkermansia muciniphila ( Supplementary Figure S3B ) and Bacteroides uniformis ( Supplementary Figure S3C ) had significantly higher PTRs in the RYGB group compared to the OB group. Considering the longitudinal data across different postoperative time points, Akkermansia muciniphila showed a significant increasing trend in PTR from the 1st to the 6th month after RYGB surgery ( Supplementary Figure S3D ), although this trend became non-significant by the 12th month (p-value = 0.075). For Bacteroides uniformis, there was a significant increase in PTR from the 1st to the 12th month post-surgery ( Supplementary Figure S3E ), with the increase at the 6th month being less pronounced (p value = 0.18). No statistically significant differences were observed in the PTRs of other microbes after surgery. This further indicates that Akkermansia muciniphila and Bacteroides uniformis, play a more crucial role in improving obesity, with different microbes playing crucial roles at various stages of recovery.