Alteration in gut microbiota associated with hepatitis B and non-hepatitis virus related hepatocellular carcinoma

Volunteers information was collected, including age, gender, weight, height, drinking history, body mass index (BMI) (kg/m²) and blood biochemical indexes (Table 1, Additional file 1). Only two HCC patients were assessed as the Child–Pugh class B, other HCC patients were Child–Pugh class A (Additional file 1). The average MELD score of HCC patients was five (Additional file 1). In total, 2047 operational taxonomic units (OTUs) were obtained from fecal microbiota of three groups for healthy controls, HBV related HCC (B-HCC) patients and non-HBV non-HCV (NBNC) related HCC (NBNC-HCC) patients, averaging 1749, 1285 and 1696, respectively. As shown in Table 1, the coverage values were nearly 1.00 for the sequences in three groups, which indicated that the sequencing depth was enough for the investigation of fecal microbiota of HCC patients or healthy controls.

The overlapping OTU of three groups were shown in a Venn diagram (Fig. 1a). These data demonstrated that 246, 46 and 141 OTUs were existed independently in healthy controls, NBNC-HCC and B-HCC patients, respectively. Based on the OTUs analysis, the bacterial communities of B-HCC patients tended to be more heterogeneous, whereas those of the healthy controls and NBNC-HCC patients exhibited similar patterns (Fig. 1b). According to the rank-abundance curves, species richness of B-HCC patients was much higher than other two groups, and all the OTUs were evenly distributed (Fig. 1c).

Generally, the bacterial alpha diversity indices (such as abundance-based coverage estimators (ACE) and CHAO1 index) in B-HCC patients was the highest (Fig. 1d, e). However, the Shannon and Simpson indices were not significantly different among three groups (P > 0.050, Table 1). The Principal Co-ordinate Analysis (PCoA) of beta diversity calculated on the unweighted UniFrac distances was used for clustering 90 samples into three distinct enterotypes (Fig. 1f).

Distinct differences in bacterial composition were observed among three groups. The microbiome contained 30 phyla, 125 orders and 479 genera in all fecal samples. Bacteroidetes, Firmicutes and Proteobacteria were the most abundant taxonomic groups (Fig. 1g, Additional file 2). The relative abundance of Proteobacteria (15.2%) in NBNC-HCC patients was individually higher than that in other two groups, while the Proteobacteria (8.5%) in B-HCC patients was less. On the contrary, the relative abundance of Firmicutes (36.8%) in NBNC-HCC patients was the lowest among three groups. Based on the order level (Fig. 1h, Additional file 3), Enterobacteriales, Clostridiales, Bacteroidales and Selenomonadales were the most abundant taxonomic groups. For genus level (Fig. 1i, Additional file 4), Escherichia-Shigella, Buchnera, Bacteroides, Prevotella, Megamonas and Faecalibacterium were predominant bacteria. The relative abundance of Escherichia-Shigella was much higher in NBNC-HCC patient (8.0%), however, the abundance of Buchnera and Megamonas were much smaller in NBNC-HCC and B-HCC patients. In B-HCC patients, the abundance of Prevotella was much greater than other two groups. Meanwhile, healthy controls had more Buchnera species.

Several similar findings were existed in the top 35 genera heatmap (Fig. 2a, Additional file 5), for example, Proteus, Lachnospiraceae UCG 010, Veillonella, Subdoligranulum, Prevotella 2, Barnesiella and Ruminococcaceae spp., were enriched both in NBNC-HCC and B-HCC patients. However, the differential abundance of bacteria have been found between NBNC-HCC and B-HCC patients displayed the reduced levels of Faecalibacterium, Pseudobutyrivibrio, Lachnoclostridium, Ruminoclostridium, Prevotella 9, Alloprevotella and Phascolarctobacterium (Fig. 2a, b), which may result in the decrease of potential anti-inflammatory short chain fatty acids (SCFAs), especially the butyrate [19, 20]. SCFAs which are intestinal microbial metabolites through dietary fiber play the anti-inflammatory effects on the immune systems [21‐23]. Butyrate, the energy for enterocytes, which influences the intestinal barrier through the mucous production and tight junction [24]. In addition, Faecalibacterium inhibits interleukin (IL)-12 secretion and stimulates IL-10 [25] (Fig. 2b). In contrary, potential pro-inflammatory strains including Escherichia-Shigella, Enterococcus, Proteus, Veillonella increased in the gut of NBNC-HCC patients. For instance, Enterococcus can produce polysaccharide A and lipopolysaccharide (LPS), which suppressed IL-17 production, resulting in experimental colitis and promoted LPS translocated into the cell [26, 27].

Interestingly, a group of potential anti-inflammatory bacteria (such as Prevotella, Alloprevotella, Faecalibacterium, Ruminiclostridium) were increased in feces of B-HCC patients (Fig. 2a, c). It is well known that these bacteria are essential for healthy status. For instance, Prevotella is well known to produce propionate in healthy gut [28], and it may be play a protective effect in adult NAFLD patients [29]. Meanwhile, pro-inflammatory bacteria (such as Escherichia-Shigella, Enterococcus) were declined in fecal of B-HCC patients.

The Monte-Carlo tests of canonical correspondence analysis (CCA) revealed the top 25 genera were extremely influenced by alcohol (P = 0.017) and BMI (P = 0.007) (Fig. 2d, Additional file 6). For instance, Escherichia-Shigella was positively associated with alcohol factor. However, the Ruminococcus 2 was also positively associated with BMI factor. Alpha-fetoprotein (AFP) is one of the most useful markers for diagnosing and monitoring HCC [30]. In our study, AFP also had a strong influence on genus assemblages in HCC patients (P = 0.024, Fig. 2e). Due to the gut flora at different stages of liver disease various, so we also predicated the relationship between clinical data and the top 35 genera (Additional files 1 and 7). We calculated correlation of Spearman in all samples. The P value was corrected using the Holm method of R (Version 3.4.4, psych package). The clinical data were mainly focused on the common hepatic function index, which were alanine aminotransferase (ALT), aspartate aminotransferase (AST), glutamyl transpeptidase (GGT), total bilirubin (TBil), albumin and AFP. Several genera (e.g., Enterococcus, Proteus, Tyzzerella 4, Parasutterella, Bifidobacterium) were negatively correlated with GGT, ALT and AST, while Dialister were negatively correlated with albumin. In addition, the index TBil showed positive correction with Parabacteroides.

At the genus level, Megamonas, Lachnospira, Eubacterium ventriosum and Lachnospiraceae UCG 001 were significantly decreased in NBNC-HCC patient samples as compared with healthy control samples (P < 0.050, Fig. 3a). In contrast, several genera such as Prevotella, Phascolarctobacterium, Anaerotruncus were particularly enriched in B-HCC patients than that in healthy controls (P < 0.050, Fig. 3b). The proportions of members of Buchnera, Lachnospira, Phascolarctobacterium, Eubacterium ventriosum were increased obviously in B-HCC patient samples compared with NBNC-HCC patients (Fig. 3c). Collectively, these differences revealed the dysbiosis involve in the development of HBV or non-HBV non-HCV related HCC.

Aberrant ecological networks of microbial communities occurred differently in B-HCC and NBNC-HCC patients.

To explore the relationships among various genera (top 35 and the significantly different genera data, Additional files 5, 8), the ecological networks of three groups were visualized. A striking feature was that taxonomically related genera tend to cluster in healthy controls (Fig. 4a). It was obvious that the within network highly associated connections of healthy controls occupied significant position and the interactions between these nodes were major balance. In NBNC-HCC patients (Fig. 4b), perhaps due to the differences in diet and excessively drinking levels (72.73%, Table 1), NBNC-HCC patients displayed a simpler concurrent network with fewer integrated symbiosis compared to healthy controls. All the interactions of bacteria in NBNC-HCC patients were positively interactions. Nevertheless, patients with B-HCC displayed multifaceted network with lots of genera, and totally grouped into a solo module with association to many other modules (Fig. 4c). Most of the bacteria related to inflammation gathered together. In this small symbiotic network, most of the interactions showed a stronger positive relation, such as Clostridium, Bryobacter, Lachnospiraceae, Buchnera, Burkholderia, Pseudobutyrivibrio. However, fewer interactions were negative, such as Alistipes, Bradyrhizobium and Sutterella, which implicated in competitive relationships for different genera. These observations implicated that the intestinal ecosystem becomes permissive for the development and maintenance of the related taxa in HCC patients.

The potential multiple biological pathways of NBNC-HCC were different from other two groups.

To further understand the biological functions of genera among HCC patients and healthy controls, we performed Kyoto Encyclopedia of Genes and Genomes database (KEGG) analysis associated with gut microbiota [31]. We identified 15,039 biological pathways in all data. The predicated functions showed unique 109 for healthy controls, 12 for NBNC-HCC patients, and 18 for B-HCC patients (Fig. 5a). The mapped results indicated that multiple biological pathways were divided into seven branches (Fig. 5b). For instance, membrane transport, replication and repair, carbohydrate metabolism and amino acid metabolism were the predominant pathways.

Healthy controls and B-HCC patients displayed similar pathways regarding to the top 35 multiple biological pathways (Fig. 5c, Additional files 8, and 9). However, B-HCC patients showed higher abundance of pathways related to chaperones and folding catalysts, general function prediction, DNA replication proteins and chromosome, which further supported that the HBV can destroy the normal function of DNA [5]. In addition, NBNC-HCC patients showed lower abundance of pathways related to amino acid metabolism (such as purine, cysteine and methionine, in red color) and glucose metabolism (such as starch and sucrose, glycolysis/gluconeogenesis, fructose and mannose, in green color). Meanwhile, KEGG analysis showed that microbial functional genes involved in oxidative phosphorylation, amino sugar and nucleotide sugar metabolism were also declined in B-HCC patients. It was also reported that marked depletion of amino acids and nucleotides metabolism in alcohol related cirrhosis patients [32]. In agreement with the study, we noted that some types of transport such as secretion system, transcription factors, other in coupled transporter and ABC transporters are enrichment in the multiple biological pathways of NBNC-HCC patients.

The relationships of multiple biological pathways were predicated to be associated with seven significantly different genera of top 35 genera (Fig. 5d). For instance, Phascolarctobacterium and Alloprevotella involved in the similar potential pathways, such as pyrimidine metabolism, cysteine and methionine metabolism and peptidases. Both genera had a negative relationship to arginine and proline metabolism and pyruvate metabolism. In addition, Ruminococcaceae UCG 002 involved in the potential pathways related to transporter, ribosome, ribosome biogenesis, chromosome, amino acid metabolism (Fig. 5d). Meanwhile, Lachnospira was potentially associated with nucleotide sugar metabolism, amino sugar, fructose and mannose metabolism.

We also identified significant changes in the multiple biological pathways of three groups (Fig. 5e). Five significantly discriminative metabolic pathways (D Arginine and D ornithine metabolism, bisphenol degradation, porphyrin and chlorophyll metabolism, linoleic acid metabolism, fructose and mannose metabolism) between healthy controls and NBNC-HCC patients, and three significantly discriminative pathways (energy metabolism, porphyrin and chlorophyll metabolism, D Arginine and D ornithine metabolism) between healthy controls and B-HCC patients. Due to the taxonomic microbiome, composition was different between two HCC groups, 11 significantly discriminative pathways (biosynthesis of 12, 14 and 16 membered macrolides, mRNA surveillance pathway, indole alkaloid biosynthesis, p53 signaling pathway, small cell lung cancer, toxoplasmosis, betalain biosynthesis, influenza A, viral myocarditis, colorectal cancer, cytochrome P450) between NBNC-HCC and B-HCC patients. The metabolic pathways such as tetracycline biosynthesis and tyrosine metabolism showed divergence enrichment degree among three groups (Additional files 10, 11, 12, 13, 14, 15).

A total of 57 HCC patients and 33 healthy controls who attended annual physical examination were recruited from September 2016 to May 2017 at Nanjing Medical University Affiliated Cancer Hospital. All participants were provided a written informed consent upon enrolment. This study was approved by the Ethics Committee of Nanjing Medical University.

HCC diagnosis is depending on three factors which includes chronic liver disease background, the positive iconography examination results, or the positive pathological examination. All HCC patients were free from other viral infections, such as Human Immunodeficiency Virus (HIV) [50]. These patients were also free from any other types of liver disease. The HCC patients underwent viral serologic testing (HBsAg and HCVAb). The HCC patients were separated into two groups. Based on the history of HBV or positive HBsAg for more than 6 months, the HBV infection was diagnosed. The patients with HBV were defined as HBV related HCC (B-HCC). While in the other group includes the patients without HBsAg or HCVAb, so-called non-HBV non-HCV related HCC (NBNC-HCC) [51]. The healthy controls were excluded diabetes, metabolic syndrome, hypertension, inflammatory bowel diseases, liver disease and cancers. All of them did not get any antiviral therapy or immunotherapy in the past 6 months.

The physiological characteristics of volunteers including age, weight and drinking condition were investigated (Table 1, Additional file 1). The standard of drinking history was considered as any alcoholic beverage (unit: gram) [52]. Men intake of alcohol less than 9.9 grams daily (or 4.9 g/days for women) was considered as low, while the consumption between 10 and 39.9 g/days (or women 5 and 19.9 g/days) was moderate, more than 40 g/days (or women 20 g/days) was high. Excepting only one HCC patients, 56 HCC patients had the blood test before enrolled, such as AFP, ALT, AST, GGT, TBil, albumin, serum creatinine (SCr) (Additional file 1). Further, the level of cirrhosis was evaluated according to the Child–Pugh and MELD [53, 54].

Fecal samples of each participant were obtained at hospital. The time span from sampling to Nanjing Medical University was intended within 24 h. Frozen samples were then stored at − 80 °C until analysis.

The genomic DNA of feces was extracted using kit (#DP328, Tiangen Biotech Co., Ltd., Beijing, China). The DNA concentration was detected using Qubit 2.0 Fluorometer (Thermo Fisher Scientific, USA). PCR was performed to produce V4 regions of the 16S rRNA gene using the conserved primers 515F (5′-GTGCCAGCMGCCGCGGTAA-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′), and no template DNA reaction was used as a negative control. PCR products were monitored using the 2% agarose gel. The strips between 400 and 450 bp were purified with GeneJET Gel Extraction Kit (Thermo Fisher Scientific, USA). PCR fragments were sequenced by Novogene Bioinformatics Technology Co., Ltd. (Tianjin, China).

To elucidate genera interactions in each group, we constructed three groups of topological overlap networks. The topological overlap of OTU was clustered into modules using WGCNA package of R (Version 3.4.4). The network analysis was visualized using Cytoscape 3.5.1. The threshold was set by Pearson r > 0 P < 0.1, and the topology overlap > 0.01 [56]. The genera including the top 35 and the significantly different genera were used to do network analysis.

To examine the distribution of genera associated with personal features (Body Mass Index, alcohol and AFP), canonical correspondence analysis (CCA) were visualized using the software CANOCO 4.5 [57]. Monte-Carlo permutation tests were performed to analyze the personal features played significant influence on the distribution of genera at P < 0.050. To avoid interference of rare species, top 25 genera were included.