Distinct gut microbiota distribution and genera in HBV-ACLF
To uncover the microbiota distribution and genera in HBV-ACLF, HBV-Other and healthy groups, fecal samples were performed 16S rRNA sequencing and Shannon indexes calculated. The diversities of microbiome were significantly different between HBV-Other, HBV-ACLF and healthy group (Fig.1A). The overall gut microbiota distribution in each group was visualized using a t-distributed stochastic neighbor embedding (t-SNE) visualization and further demonstrated distinct microbiota distribution between groups, especially between the healthy and liver disease groups (Fig.1B).
To identify the predominant gut microbiota in HBV-ACLF, LEfSe analysis was performed. The results showed that there were a number of different genera of gut microbiota between the healthy and the liver disease groups, and a trend could be observed that the HBV-ACLF had more Enterococcus relative richness than the healthy group (Fig.1C). Clinically, cocci to bacilli ratio is a common parameter used to inform the status of gut microbiota and the choice of antibiotics, therefore are often tested for patients with ACLF or abdominal and intestinal infections[27, 28]. We found that the ratio of cocci to bacilli richness was significantly different among the three groups where HBV-ACLF group exhibited the highest ratio (Fig.1D), suggesting the balance of gut microbiota in these patients were severely disrupted.
Establishing a microbiota classification model for the healthy, HBV-Other and the HBV-ACLF group
A classification model for the healthy, HBV-Other and the HBV-ACLF group was established by Random Forest classifier. The classification model included 18 most important taxa of the 3 groups (Fig.2A), with an AUC value of 0.89. In addition, the decomposition visualization (Fig.2B) demonstrated that the 18 selected taxa could be well distinguished among the 3 groups, suggesting the model was validly established.
Correlation between clinical/demographic variable and gut microbiota
To investigate the correlation between each clinical/demographic variable and gut microbiota among the 3 groups, adonis analysis was performed. The analysis showed that with the exception of sex, AST, HBsAg and HBeAb, all the other clinical/demographic variables were significantly associated with gut microbiota differences among the 3 groups (P<0.05, Table 1), which were consistent with previous reports[27-30].
Gut microbiota taxa difference between the progression and regression groups
To investigated whether gut microbiota differs within the HBV-ACLF group, we sub-assigned the group into progression group (disease progression at discharge; n = 47) and regression group (improved outcomes at discharge; n = 165) according to the Model for End-Stage Liver Disease (MELD) score at discharge. Fifty-two genera with different community richness between the HBV-ACLF progression and regression groups were identified with the most abundant genera (p < 0.005) listed in Table 2 and Supplementary Table S1 (p < 0.05). Enterococcus and Faecalibacterium showed the highest richness within the 52 genera, highlighting the importance of these two genera in ALCF which may contribute to disease progression. The relative abundance of Enterococcus was significantly elevated in the progression group, and that of Faecalibacterium was significantly elevated in the regression group (Fig.3).
Gut microbiota genera associated with blood biochemical indicators
To investigate whether there is a link between the gut microbiota and clinical parameters, we evaluate the association between different genera and blood biochemical indicators in all groups. The blood biochemical indicators were divided into three categories according to their clinical relevance as follows: Liver inflammation - alanine aminotransferase (ALT) and aspartate aminotransferase (AST); Liver disease severity - total bilirubin (TBIL), international normalized ratio (coagulation function) (INR) and end-stage liver disease model (MELD); Degree of infection - white blood cell count (WBC), neutrophil percentage (NEUT%) and procalcitonin (PCT).
The gut microbiota genera associated with each blood biochemical indicator were identified using a Random Forest regressor via microbe's Mean Decrease Gini. We trained several models using microbiota richness to predict their clinical relevance. By comparing the feature importance of the trained Regressor, we detected that Filifactor, Rikenellaceae, Clostridium, Bilophila and Comamonas were associated with ALT and AST(Fig.4A); Enterococcus, Enterococcaceae and Abiotrophin were associated with TBIL, INR and MELD (Fig.4B); and Enterococcus and Streptococcus were associated with WBC, NEUT% and PCT (Fig.4C). These results have the potential to inform the use of intestinal microbial intervention to alleviate or prevent the progression of liver disease.
Metagenomic sequencing between the progression and regression group in HBV-ACLF patients
The results of genus Enterococcus and Faecalibacterium by 16S rRNA sequencing were validated by the metagenomic sequencing (Fig.5A) where the richness of Enterococcus was higher in the progression group than in the regression group, and the richness of Faecalibacterium was higher in the regression group than in the progression group.
The time series samples of HBV-ACLF patients
The dynamic change of gut bacteria in patients with liver failure is an important indicator to predict the optimal time to introduce therapeutic interventions and to adjust follow-up treatments. We performed the time series samples analysis on day 1, 7 and 14 upon patient admission by metagenomic sequencing. The results showed that the richness of Lactobacillus casei paracasei was significantly higher in the progression group compared with the regression group (P<0.05); while the richness of Alistipes senegalensis, Faecalibacterium prausnitzii and Parabacteroides merdae were significantly higher in the regression group (P<0.05, Fig.5B). The results of Faecalibacterium prausnitzii were consistent with the 16sRNA sequencing results that genus Faecalibacterium was higher in the regression group. Further analysis revealed that the regression group had a small increase in the richness of Enterococcus faecium, while the progression group had a marked increase in the richness of Enterococcus faecium during the period of 14 days. Importantly, the richness of Enterococcus was significantly higher in the progression group than the regression group in day 1 (Fig.5A).
Bayes network analysis to identify the key species of gut microbiota differences
Finally, the key species of gut microbiota which were different between the progression group and the regression group were identified using Bayes network analysis. As shown in Table 3, 7 species (Streptococcus vestibularis, Peptostreptococcus unclassified, Scardovia unclassified, Prevotella salivae, Prevotella histicola, Actinomyces odontolyticus, Streptococcus parasanguinis) were enriched in the regression group while 3 species (Ruminococcus obeum, Dorea longicatena, Clostridium citroniae) were enriched in the progression group. These results were further validated by qPCR (Fig.6). Consistently, the progression group of HBV-ACLF exhibited significantly abundant Enterococcus faecium and Lactobacillus casei paracasei, while the regression group presented significantly abundant Faecalibacterium prausnitzii, Clostridium citroniae and Dorea longicatena.