Patients aged 18–75 years with unexplained intestinal gas, bloating, diarrhea, abdominal pain, and functional dyspepsia that were suspected to be due to SIBO were prospectively recruited. Exclusion criteria included the following: a history of active inflammation, microscopic colitis, inflammatory bowel disease, food allergies, cancer, immune diseases, and severe systemic, cardiac, renal, hepatic, metabolic, and psychiatric diseases; intake of antibiotics or proton-pump inhibitors (PPI) in the last 4 weeks; intake of promotility agents, laxatives, or probiotics in the past week; current pregnancy; inability to endure endoscopy; or inability to obtain informed consent.

Demographic information, height, weight, and previous medical history of abdominal surgery were obtained at the time of enrollment. Clinical symptoms were evaluated using the gastrointestinal symptom rating scale (GSRS) (Svedlund et al., 1988) consisting of 15 items, including abdominal distension, heartburn, acid regurgitation, abdominal pain, sucking sensations in the epigastrium, nausea and vomiting, borborygmus, eructation, increased flatus, decreased passage of stools, increased passage of stools, loose stools, hard stools, urgent need for defecation, and feeling of incomplete evacuation. Each item was evaluated using a scale of 0 to 3, with 0 showing no symptoms, and 3 showing extremely severe symptoms. The baseline information of each patient was collected by the same nurse prior to the evaluation of SIBO to minimize bias.

All participants underwent lactulose hydrogen and methane breath testing (LBT) for diagnosis of SIBO and classified as either SIBO + or SIBO−. The participants were instructed not to eat fermentable foods and alcohol on the day prior to the test. They were also asked to fast for at least 12 h before the breath test, avoid smoking on the day of the test, and avoid exercise during the breath test. The participants cleaned their teeth on the morning of the breath test and gargled prior to testing to reduce fermentation of lactulose by oral bacteria. After the baseline expiratory samples were collected, the participants were instructed to take 10 g of lactulose orally (Duphalac, Abbott). Breath samples were then collected every 15 min until 90 min for a total of seven times. Hydrogen and methane concentrations were detected using a Quintron BreathTracker (Quintron Instrument Company, Milwaukee, WI, United States). A positive breath test standard was determined as follows (Rezaie et al., 2017; Pimentel et al., 2020): hydrogen rise above baseline ≥ 20 ppm by 90 min or methane ≥ 10 ppm at any point during the test within the 90 min.

Fresh fecal samples were collected before or at the time of gastroscopy in the hospital. The participants were instructed to remove the outer layer of their feces with a sampling spoon and collect the central portion into a sterile tube. Gastroscopy was performed to obtain one sample of the distal duodenum (5–10 cm below the major duodenal papilla) for biopsy as previously described (Li et al., 2018). The following day, colonoscopy was performed after bowel preparation with the oral compound polyethylene glycol electrolyte powder. One mucosal sample of the sigmoid colon (25–30 cm from the anus) was obtained during the insertion to prevent contamination by bacteria in the proximal colon. The enteroscope was then advanced past the ileocecal valve to obtain one sample of the distal ileum (5–10 cm from the ileocecal valve) for biopsy. All endoscopic procedures were performed by the same senior gastroenterologist. Each sample was placed in a sterile tube and immediately stored at −80°C for subsequent 16S rRNA sequencing analysis.

Fecal DNA was extracted using the E.Z.N.A. Soil DNA Kit (Omega Bio-tek, Norcross, GA, United States), and mucosal DNA was extracted using the FastDNA Spin Kit for Soil (MP Biomedicals) according to the manufacturers’ instructions. The concentrations and purities of the extracted DNA were measured using a NanoDrop2000 UV-vis spectrophotometer (Thermo Scientific, Wilmington, United States), and the qualities of the extracted DNA were evaluated using 1% agarose gel electrophoresis. The V3-V4 regions of the bacterial 16S rRNA gene were amplified using primers 338F (5′-ACTCCTACGGGAGGCAGCAG-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′). Sequencing was performed using a MiSeq PE300 platform (Illumina, San Diego, CA, United States) at Shanghai Majorbio Bio-pharm Technology Co., Ltd.

Raw sequencing data were qualified and stitched using the Trimmomatic and FLASH software, respectively. Briefly, low-quality reads (those with nucleic acid lengths less than 50 bp after quality control, with quality values less than 20, or containing N bases) were filtered. Based on the overlapping relationships between the paired-end (PE) reads, they were merged into sequences with minimum overlapping lengths of 10 bp. The maximum error ratio of the overlapping area of the spliced sequences was set at 0.2, and sequences that did not meet this requirement were removed. The samples were distinguished according to the barcodes and primers at both ends of the sequences, with the directions of the sequence being adjusted. The data obtained were then optimized. The resulting sequences were processed using the QIIME software (version 1.9.1). All sequences with 97% similarities were clustered using the URARese software (version 7.1), and chimeras were removed in the clustering process to obtain sequences representative of OTUs. Lastly, taxonomic analysis of the OTUs was performed using the RDF Classifier software (version 2.11) and compared with the Silva database (SSU132) at a comparison threshold of 70%.

The richness and diversity of the microbial community were compared using alpha-diversity analysis, which considers diversity indices (Shannon and Simpson), richness indices (Sobs, Chao, and Ace), and sequencing depth indices (coverage index), using mothur software (version v.1.30.1). Alpha-diversities between groups were analyzed using the Wilcoxon rank-sum test. Beta-diversities were used to compare the differences in species compositions between groups analyzed using PCoA and PCA analysis based on binary_jaccard, abund_jaccard, and bray_curtis distance subjected to ANOSIM analysis. The differences in species abundance of the microbial community among several groups were evaluated using the Kruskal–Wallis rank-sun test. The differences in species abundance of the microbial community between groups and the effects of each differentially abundant taxon were assessed using the Wilcoxon rank-sum test or the linear discriminant analysis (LDA) effect size (LEfSe) method (Zhuang et al., 2018; Liu et al., 2019), which emphasized statistical significance and biological correlation. The threshold was set at LDA level > 3. Random forest analysis, a machine learning procedure, was used to identify taxa to build prediction models to distinguish between the SIBO + and SIBO− subjects, area under a receiver operating characteristic (ROC) curve (AUC) was used to assess the performance of random forest prediction model. Specifically, random forest analysis was conducted to rank the importance of species, and AUC scores were calculated for the prediction model with the top N species in importance. In the coordinate axis of validation figure, X represented the number of genera in the importance ranking and Y represented the AUC value. In building the prediction model, the corresponding X value of the solid point with the highest Y value was considered the ultimate number of top species. The performance of the prediction model was evaluated for its discriminative power using ROC curve. All sequencing data were analyzed using the R software (version 3.3.1); randomForest package and plotROC package were the machine learning packages used.

In addition to the sequencing data, other data were statistically analyzed using the IBM SPSS Statistics software (version 25.0). For continuous data analysis, the Mann–Whitney U-test was used. For categorical data analysis, Fisher’s exact test was used. In all test results, the differences were considered statistically significant when the double-tailed P value was < 0.05.

The sample size of the subjects was determined based on the number of cases available during the study period. No formal sample size calculations were conducted prior to the commencement of this pilot study.

The study protocol was approved by the Institutional Ethical Review Committee of the Chinese PLA General Hospital (approval reference number: S2018-081-02). All participants provided written informed consent.

The sequencing data is available in National Center for Biotechnology Information under the Sequence Read Archive (SRA) with the BioProject No. PRJNA728515.

A total of 34 subjects were recruited, of which three were excluded from the analysis due to lack of information or factors related to sample handling, storage, or experimental operation. A total 31 subjects (19 men and 12 women) were eventually included. In some subjects, samples from all four sites were not collected. A total of 25 fecal, 24 duodenal, 26 terminal ileal, and 25 sigmoid colonic samples were collected.

According to the LBT, 17 subjects were classified into the SIBO + (10/17 male) group, and 14 subjects were classified into the SIBO− (9/14 male) group. There were no significant differences in sex distribution, age, BMI, or history of abdominal surgeries. The scores of abdominal distension and feeling of incomplete evacuation were higher in the SIBO + group than in the SIBO− group (P = 0.002 and P = 0.012, respectively). There were no differences in the scores of abdominal pain, heartburn, acid regurgitation, sucking sensations in the epigastrium, nausea and vomiting, borborygmus, eructation, increased flatus, decreased passage of stools, increased passage of stools, loose stools, hard stools, and urgent need for defecation, and total scores between the SIBO + and SIBO− groups. The demographic and clinical characteristics of all subjects are displayed in Table 1.

The ACE richness index showed that the microbial richness values of the duodenum, ileum, and sigmoid colon were significantly higher than those of the feces (P < 0.001) (Figure 1A), and the Shannon diversity index indicated that the microbial diversities of the three gut mucosal sites were significantly higher than those of feces (P < 0.001) (Figure 1B). Similar results were obtained using the Chao and Sobs indices (Supplementary Figures 1A,B). Principal coordinates analysis (PCoA) based on the Bray-Curtis distance showed that the bacterial composition of either the fecal or duodenal mucosal samples was significantly different from those of the other three groups (ANOSIM R = 0.305, P = 0.001) (Figure 2B). The differences in microbial compositions among sample types and gut segments were further demonstrated by the partial least squares discriminant analysis (PLS-DA) (Supplementary Figure 1C). There were also no differences in alpha and beta diversities between the mucosal microbiota of the ileum and sigmoid colon (Supplementary Figure 1D).

Species comprising more than 1% of the total microbiota were defined as predominant species. At the phylum level, the predominant phyla in the fecal microbiome were Firmicutes (54.06%), Bacteroidetes (32.59%), Proteobacteria (9.49%), and Actinobacteria (2.84%). The representative phyla were more diverse in the mucosal samples. The predominant phyla in the duodenal mucosal microbiome were Firmicutes (33.32%), Proteobacteria (22.79%), Bacteroidetes (17%), unclassified_k_norank_d_Bacteria (15.61%), Actinobacteria (4.91%), Fusobacteria (1.06%), and Epsilonbacteraeota (1.02%). The predominant phyla in the mucosal microbiome of the ileum and sigmoid colon were Firmicutes (40.45 and 38.54%, respectively), Bacteroidetes (23.55 and 23.54%, respectively), Proteobacteria (17.78 and 19.94%, respectively), unclassified_k_norank_d_Bacteria (9.18 and 8.9%, respectively), and Actinobacteria (5.07 and 5.23%, respectively) (Figure 2A and Supplementary Figures 2A–D). The mucosal microbiota in all three gut segments, especially in the duodenum, harbored more Proteobacteria, unclassified_k_norank_d_Bacteria, and Actinobacteria (P < 0.05), while those in the fecal samples harbored more Firmicutes and Bacteroidetes (P < 0.05) (Figure 2C). The mucosa of the duodenum showed lower relative proportions of Firmicutes and Bacteroidetes than the ileum (P < 0.05) (Figure 2D). It also showed higher relative proportions of Fusobacteria, Epsilonbacteraeota, Patescibacteria, and Spirochetes than the sigmoid colon (P < 0.01) (Figure 2E).

The six genera in the mucosa of the duodenum with the highest average proportions were unclassified_k_norank_d_bacteria (15.61%), Bacteroides (7.33%), Escherichia-Shigella (6.65%), Streptococcus (4.28%), Megamonas (3.86%), and Lactobacilli (3.41%); in the mucosa of the ileum were Bacteroides (13.87%), unclassified_k_norank_d_bacteria (9.18%), Escherichia-Shigella (6.42%), Faecalibacterium (5.58%), Blautia (3.14%), and Lactobacilli (3.12%); in the mucosa of sigmoid colon were Bacteroides (13.52%), unclassified_k_norank_d_bacteria (8.9%), Escherichia-Shigella (7.12%), Faecalibacterium (4.86%), Ralstonia (3.36%), and Blautia (3.17%); and in the feces were Bacteroides (26.06%), Escherichia-Shigella (8.21%), Faecalibacterium (5.9%), Agathobacter (5.64%), Phascolarctobacterium (4.27%), and Subdoligranulum (3.55%) (Figure 3A and Supplementary Figures 3A–D).

Bacteroides (P < 0.001), Agathobacter (P < 0.001), and Roseburia (P < 0.01) were more abundant in the fecal samples than in the mucosal samples (Figure 3B). On the other hand, the relative numbers of unclassified_k_norank_d_bacteria (P < 0.001), Lactobacillus (P < 0.001), Ralstonia (P < 0.001), and Streptococcus (P < 0.001) were higher in the mucosa than in the feces (Figure 3B). Compared to the ileum and sigmoid colon, Bacteroides (P < 0.05), Faecalibacterium (P < 0.01), Blautia (P < 0.001), Subdoligranulum (P < 0.01), [Ruminococcus]_torques_group (P < 0.01), Bifidobacterium (P < 0.01), and Agathobacter (P < 0.05) were relatively less abundant, whereas Streptococcus (P < 0.01), Neisseria (P < 0.001), Veillonella (P < 0.001), and Prevotella_7 (P < 0.001) were more abundant in the duodenal mucosa (Figures 3C,D). Although there was a high similarity between the bacterial species in the mucosa of the ileum and sigmoid colon, there were some genera with significantly different proportions. [Eubacterium]_ventriosum_group (P < 0.05) and norank_f__TRA3-20 (P < 0.05) were significantly more abundant in the mucosa of the ileum, while Anaerococcus (P < 0.05), Ruminiclostridium_1 (P < 0.05), Roseomonas (P < 0.05), Norank_f_veillonellacece (P < 0.01), and Herbaspirillum (P < 0.05) were more abundant in the mucosa of the sigmoid colon (Supplementary Figure 1E).

Compared with the SIBO− group, the microbiota richness values based on the Sobs index of the mucosa of the ileum and sigmoid colon were significantly decreased (P < 0.001 and P < 0.05, respectively) (Figures 4A,C) and the microbiota diversity based on the Shannon index of the ileum mucosa was significantly reduced (P < 0.001) (Figure 4B) in the SIBO + group. No significant differences in the microbial diversity were observed in neither duodenum mucosa nor feces between the SIBO + and SIBO− groups (Supplementary Figures 4A,B). Clustering in the bacterial community was significantly different between the SIBO + and SIBO− groups in the duodenal mucosa based on abund_jaccard distance (P = 0.039) (Figure 5A), in the ileac mucosa based on binary_jaccard (P = 0.002) (Figure 5B), and in the mucosa of the sigmoid colon based on binary_jaccard (P = 0.007) (Figure 5C). However, this finding was not observed in the fecal samples (Supplementary Figure 4C). Overall, the mucosal microbiota of the duodenum, ileum, and sigmoid colon were structurally different in the SIBO + subjects compared to SIBO− subjects.

Differential expression analysis showed that the relative abundances of multiple genera in the duodenum, ileum, and sigmoid colon mucosa were significantly different between the SIBO + and SIBO− groups. LEfSe (LDA score > 3) was used to detect the microbial species that most likely explains the differences between the SIBO− and SIBO + groups. In the duodenal mucosa, six genera were significantly enriched in the SIBO− group compared to the SIBO + group: Lactobacillus, Prevotella_1, Bifidobacterium, Dialister, norank_f__Ruminococcaceae, and Clostridium_sensu_stricto_1. Additionally, only the genus norank_f__norank_o__Absconditabacteriales_SR1 was enriched in the SIBO + group (Figure 5D). In the ileac mucosa, 13 genera were more abundant in the SIBO− group: Lactobacillus, Prevotella_1, norank_f__norank_o__Chloroplast, Agathobacter, Clostridium_sensu_stricto_1, [Ruminococcus] _gnavus_group, norank_f__Xanthobacteraceae, Dialister, norank_f__Ruminococcaceae, Butyrivibrio_2, Ruminococcus_2, Klebsiella, and Rhodococcus. However, five genera were more abundant in the SIBO + group: [Ruminococcus]_torques_group, Enterococcus, Sutterella, Holdemanella, and Butyricimonas (Figure 5E). In the sigmoid mucosa, five genera were more abundant in the SIBO− group: Lactobacillus, Prevotella_1, Ruminococcus_2, Dialister, and g__norank_f__Ruminococcaceae. However, three genera were more abundant in the SIBO + group: Sutterella, Holdemanella, and Catenibacterium (Figure 5F).

To further identify the shared genera involved in the pathophysiology of SIBO, we found five shared genera in the mucosal samples of the duodenum and ileum and seven shared genera in the mucosal samples of the ileum and sigmoid colon which were significantly different between the SIBO + and SIBO− subjects. Of these, four genera (Lactobacillus, Prevotella_1, Dialister, and norank_f__Ruminococcaceae) had similar changes among the duodenal, ileac, and sigmoid colonic samples in the SIBO + subjects compared to the SIBO− subjects.

To identify a characteristic microbiome with the potential to distinguish between SIBO− and SIBO + patients, a random forest analysis was done to screen for the most important biomarker. In the duodenum, the random forest analysis showed that the top 4 important features (Ruminococcus_1, [Eubacterium]_ventriosum_group, norank _f__NS9_marine_group, and unclassified_c__Parcubacteria) exhibited the highest predictive power. ROC analysis was then performed to evaluate the performance of the model, wherein an AUC value of 0.9 (95% CI: 0.76–1) was obtained (Figures 6A,B). In the ileum, the random forest analysis yielded the top 3 important features (Rhodococcus, norank_f__norank_o__NB1-j, and Candidatus_Solibacter) that were used to construct the prediction model, in which ROC analysis showed an AUC value of 0.93 (95% CI: 0.81–1) (Figures 6C,D). In the sigmoid colon, the random forest analysis showed the top 6 important features (Gallionella, Aggregatibacter, Gordonibacter, Selenomonas_1, Pediococcus, and Oribacterium) which were used to build a prediction model with an AUC value of 0.87 (95% CI: 0.71–1) based on ROC analysis (Figures 6E,F).