Demographic data
A total of 189 patients were randomized to receive P9 or placebo. After a 4-week intervention period, eleven and eight patients from the placebo and P9 group were excluded because of decline participation or missing diary (Fig. 1a; Figure S1). Thus, 170 patients (85 supplemented P9 and 85 receiving placebo) completed the study. Patients were aged 22.35 ± 4.07 years and 22.11 ± 4.90 years in P9 and placebo groups, respectively. The proportion of male subjects in the P9 and placebo groups was 47.06% and 50.59%, and that of female subjects was 52.94% and 49.41%, respectively. The ratio of Han ethnic to other ethnic groups was 82 to 3 and 81 to 4 in P9 and placebo groups, respectively (Table S2). There was no significant difference between the P9 and placebo groups in terms of the baseline age, gender ratio, ethnicity, history of drug allergy, preexisting disease, and previous drug treatment (P > 0.25 in all cases; Table S2).
P9 administration apparently improved diarrhea symptoms
The diarrhea symptom severity score was based on the Gastrointestinal Symptom Rating Scale. Before the intervention, no significant difference existed in the diarrhea symptom severity score between the P9 and placebo groups (2.013 ± 0.977 and 2.074 ± 0.874 in P9 and placebo groups, respectively, P = 0.588; Fig. 1b; Table S3). After the 28-day intervention, the diarrhea symptom severity score in P9 group was significantly lower than the placebo group, reduced by 20.6% (1.138 ± 0.694 and 1.433 ± 0.803 in P9 and placebo groups, respectively, P = 0.048; Fig. 1b), indicating that P9 intervention can relieve diarrhea symptoms.
At baseline (day 0), no significant differences were observed in all secondary endpoint parameters, including the scores of bowel frequency, stool consistency, fecal urgency, and DASS-21 questionnaire were noted between P9 and the placebo groups (P > 0.05; Fig. 1b; Table S3). However, compared with the placebo group, after the 28-day intervention, patients in the P9 group had statistically significant improvement in stool consistency score (4.717 ± 0.725 and 4.933 ± 0.626 in P9 and placebo groups, respectively; P < 0.05; Fig. 1b), though not the scores of bowel frequency, fecal urgency, and the depression, anxiety and stress levels (P > 0.05; Figure S1b).
Although no significant difference was observed in the number of adverse events between P9 and the placebo groups (P = 0.218; Table S3), obviously fewer adverse events were reported by the patients in P9 group than those in the placebo group at day 28 (six vs 14 events) and day 42 (three vs five events). No serious adverse reactions were reported in any of the groups at any time point.
P9 supplementation modulated patients’ fecal bacterial microbiota diversity and composition
The fecal microbiota of 169 patients was analyzed at three time points (days 0, 28, and 42) by metagenomic sequencing. One patient from P9 group did not provide stool sample for fecal metagenome analysis (Fig. 1a).
No significant differences were observed in the alpha diversity (Shannon and Simpson indices, P > 0.05) between the P9 and placebo groups at any time point (Fig. 2a). However, beta diversity analyzed by PCoA and adonis test revealed significant differences in the fecal microbiota structure between P9 and placebo groups at days 28 (R2 = 0.015, P = 0.012) and 42 (R2 = 0.014, P = 0.018; Fig. 2b).
We also found that P9 administration was associated with post-intervention changes in the species-level fecal microbiota composition. A total of 29 major SGBs (mainly belonging to Lactobacillaceae, Oscillospiraceae, Ruminococcaceae, Lachnospira ) were identified; none of them showed significant differences between the two groups at baseline (day 0), and they only became differentially abundant at days 28 and 42 (19 and 10 SGBs, respectively; Fig. 2c; Table S4). After 28-day intervention, the fecal microbiota of P9 group had significantly more Lactiplantibacillus plantarum and Ruminococcus_A faecicola compared with the placebo group, while an opposite trend was observed in the species, Mediterraneibacter torques, Eubacterium_I ramulus, and Enterocloster sp000431375 (P < 0.05 in all cases; Fig. 2c). At day 42, significantly more Butyricicoccus_A sp002395695 and Streptococcus thermophilus were detected in P9 group compared with the placebo group, and Phascolarctobacterium faecium and Faecalibacterium sp. showed an opposite trend (P < 0.05 in all cases; Fig. 2c). Interestingly, eight SGBs showed consistent differences at days 28 and 42, including significantly more Acutalibacteraceae sp000431775 and Paraprevotella xylaniphila, but significantly fewer Coprococcus sp and Butyricimonas virosa in P9 group compared with the placebo group (P < 0.05 in all cases; Fig. 2c). Taken together, these results suggested that P9 supplementation could significantly change the diversity and composition of gut microbiota in patients with chronic diarrhea.
We further analyzed the correlation between the species-level gut microbiota and clinical indicators of diarrhea to investigate whether the improvement of clinical indicators was related to changes in specific bacteria after P9 administration (Fig. 2d). Our results showed that the stool consistency score correlated positively with Eubacterium_I ramulus (r = 0.20, P = 0.019); the stress score correlated positively with Enterocloster sp000431375 (r = 0.21, P = 0.008); and the bowel frequency showed a significant positive correlation with Mediterraneibacter torques (r = 0.24, P = 0.002). Moreover, the anxiety score showed a significant negative correlation with Ruminococcus_A faecicola (r = -0.20, P = 0.019), suggesting that the symptom alleviation was associated with changes in some specific functional gut bacteria.
P9 supplementation modulated patients’ fecal phageome diversity and structure
Since phages play a part in modifying the gut microbiota, we then investigated intervention-associated changes in patients’ fecal phageome diversity and composition. A total of 94,384 nonredundant vOTUs were annotated by comparing our dataset against the Metagenomic Gut Virus catalogue; and 41,059 vOTUs were assigned into 13 bacteriophage families, including 7,587 prophages and 33,472 non-prophages (Fig. 3a). Taxonomic annotations of these sequences found a high prevalence of Siphoviridae (33.15%), Myoviridae (9.31%), Microviridae (5.07%), Podoviridae (1.91%), and crAss-phage (1.03%; Fig. 3a), most of which belonged to the order Caudovirales (87.67%; Fig. 3a).
Beta diversity analysis showed that no significant differences were found in the overall phageome between P9 and the placebo groups at days 0, 28, and 42 (Fig. 3b). Consistently, there were no significant differences in the alpha diversity of phageome between the two groups at days 0 and 42 (P > 0.05; Fig. 3c), except that, at day 28, the Simpson index of patients’ fecal phageome in P9 group was numerically lower than that of the placebo group (P = 0.056; Fig. 3c). Interestingly, we observed a significant positive correlation between the Shannon index of the gut bacteria microbiota and phageome (R = 0.928, P < 0.001; Fig. 3d), which was consistent with the results of Procrustes analysis (R = 0.818, P = 0.001; Fig. 3e), suggesting that there was a strong cooperativeness between the gut phageome and their bacterial hosts.
Phages are obligate intracellular parasites residing in bacterial host, so specific genomic associations between phages and bacteria reflect the phage infection history. To further explore the interplay between gut bacteriophages and bacteria, we then investigated specific distribution of phage sequence in bacterial host genomes (Fig. 3f). A total of 21,103 vOTUs annotated to 12 known bacteriophage families were analyzed; 94.2% of these vOTUs (corresponding to 19,874 vOTUs) were predicted to connect with specific bacteria hosts, and 12,870 of them were connected to known host bacterial genera. Siphoviridae, the most highly widespread gut phage family, were mainly associated with Firmicutes and Bacteroidota hosts (including the genera Ruminococcus, Bacteroides, Faecalibacterium, Eubacterium, and Lachnospira). Myoviridae and Microviridae are two widespread and abundant human gut phage families often infected Firmicutes and Bacteroidota hosts (including the genera Faecalibacterium, Lachnospira, and Bacteroides). In addition, crAss-phage mainly infected Bacteroides hosts (Bacteroides and Prevotella), and Herelleviridae mainly infected Firmicutes hosts (Flavonifractor). Surprisingly, most of these infected bacterial hosts, including Faecalibacterium, Eubacterium 28, and Lachnospira, changed significantly after P9 intervention.
Finally, we explored the association between patients’ gut bacteriophages and clinical indicators after P9 intervention. A correlation analysis of clinical indicators and 12 known bacteriophage families was performed (Fig. 3g). Our results showed that both Microviridae (r = 0.20, P = 0.012) and crAss-phage (r = 0.23, P = 0.003) were positively correlated with the fecal urgency score; and Herelleviridae showed a positive correlation with bowel frequency (r = 0.20, P = 0.019). These results suggested changes in specific phage sequences after P9 supplementation were associated with diarrhea improvement.
P9 supplementation modulated patients’ gut bioactive metabolites and CAZymes
A genome-centric metabolic reconstruction was established to identify intervention-associated changes in GMMs encoded in 629 SGBs using the MetaCyc and Kyoto Encyclopedia of Genes and Genomes databases. A total of 72 GMMs were identified, belonging to 11 metabolic modules, including SCFAs, amino acids, tryptophan and its derivatives, unsaturated fatty acids, monosaccharides, disaccharides, polysaccharides, neurotransmitters, vitamins, bile acids and other metabolic modules (Fig. 4a). These identified modules were encoded by 34 bacterial orders. The modules of acetate synthesis, quinolinic acid degradation, and S-adenosylmethionine synthesis were common to most orders (Fig. 4a). However, some orders exhibited a higher metabolic diversity than other orders. The top three metabolically diverse orders were Bacteroidales, followed by Lachnospirales and Oscillospirales (belonging to Firmicutes), encoding a wide array of metabolic modules related to SCFA, amino acid, vitamin, and bile acid metabolism. Interestingly, several bacteria belonged to these three orders (including Lactiplantibacillus plantarum, Paraprevotella xylaniphila, and Acutalibacteraceae sp000431775) showed significant differential abundance after P9 intervention. Modulation of the composition of these metabolic diverse taxa may drastically change the potential metabolic function of the gut microbiota.
We then analyzed changes of potential gut bioactive metabolites after P9 intervention using the MelonnPan pipeline, and we found no significant difference in the alpha diversity of gut bioactive metabolite profile between the P9 and placebo groups at days 0 and 42, but the Shannon index in the P9 group was almost significantly higher than that of the placebo group at day 28 (P = 0.059; Fig. 4b). The results of alpha diversity analysis were in line with that of the beta diversity analysis by PCoA and adonis test. Significant differences were observed in the gut bioactive metabolites profile between the P9 and placebo groups at day 28 (R2 = 0.028, P = 0.013), but not at day 0 (R2 = 0.004, P = 0.538) and 42 (R2 = 0.008, P = 0.069; Fig. 4c). Moreover, a total of 18 differentially gut active metabolites between the P9 and placebo groups were probiotic intervention-specific; these predicted metabolites only became significantly different between groups after 28-day P9 intervention but not at baseline (Fig. 4d, Table S5). Several of these probiotic-intervention responsive metabolites, such as cholate, chenodeoxycholate, C2 carnitine, C16 carnitine, cholesterol, X2 hydroxyphenethylamine, creatine, and bilirubin, were significantly enriched in P9 group compared with the placebo group (P < 0.05; Fig. 4d). To further explore the enzyme repertoire for complex polysaccharide metabolism encoded by the patients’ fecal microbiota, CAZyme genes were annotated using dbCAN2. A total of 26,170 CAZyme-encoding genes were found across 629 SGBs (Table S6), and most genes encoded the family glycoside hydrolases (GHs, 14,598 genes), followed by glycosyltransferases (GTs, 6,108 genes), carbohydrate esterases (CEs, 2,726 genes), carbohydrate-binding modules (CBMs, 1,690 genes), polysaccharide lyases (PL, 810 genes), and auxiliary activities (AAs, 238 genes). In addition, comparative analysis of CAZyme-encoding subfamilies between P9 and placebo groups found that 15 CAZyme subfamilies were enriched in P9 group, including: glycoside hydrolases (GH108, GH13_13, GH13_2, GH13_27, GH158, GH64, GH5_18), carbohydrate-binding modules (CBM4, CBM56), glycosyltransferases (GT21, GT60, GT74), polysaccharide lyases (PL1_3, PL10_2), and CE16 (Fig. 4e). These results together indicated that P9 administration could enrich the CAZyme-encoding genes in the gut microbiota, possibly contributing to a broadened carbohydrate utilization capacity.
P9 supplementation modulated patients’ fecal metabolome
Next, we analyzed specific changes in the fecal metabolome of the patients after P9 intervention. On the fecal metabolome PCA plot, symbols representing the quality control samples formed a close cluster (Fig. 5a), indicating a good stability of the instrumental conditions and the reliability of analysis of other samples. The PLS-DA analysis of the fecal metabolomes of samples from the two groups showed that there was a moderate degree of group-based separation at all three time points, although no significant difference was detected by adonis test (P > 0.05; Fig. 5b). We further analyzed the metabolite-level differences in the fecal metabolomes between two groups based on the VIP value generated by the PLS-DA model (VIP > 2.0) and the P value calculated by Wilcoxon test (P < 0.05). A total of 21 significant differential metabolites were identified at days 28 and 42 (Fig. 5c; Table S7). These metabolites were not significantly different between the two groups at baseline. Specifically, compared with the placebo group, P9 group had significantly more caffeic acid, taurine, 3-hydroxypentanoic acid, hexacosanoic acid, and cerotic acid after intervention (P < 0.05; Figure. 5d). We further explored the associations between differentially abundant metabolites and clinical indicators of diarrhea by performing Pearson correlation analysis. Interestingly, the results showed that caffeic acid was significantly and negatively correlated with the depression score (r = 0.21, P < 0.01); and taurine was significantly and negatively correlated with the diarrhea severity score (r = 0.21, P < 0.01; Figure. 5e).
Afterwards, we analyzed the SCFA composition in patients' feces by GC-MS to identify differential abundant SCFAs associated with P9 intervention. After 28 days of intervention, more acetic acid and butyric acid were detected in P9 group compared with the placebo group (P < 0.05; Fig. 6a), but the fecal contents of isovaleric acid, valeric acid, propionic acid, and isobutyric acid exhibited no significant differences between groups (P > 0.05; Fig. 6a). These results confirmed that P9 intervention could increase patients’ functional gut metabolites like acetic acid and butyric acid.