The gut microbiota of nonalcoholic fatty liver disease: current methods and their interpretation

There are only a limited number of studies that have examined microbiota composition in patients with simple steatosis or NASH, and these have very dissimilar results (Table 1). First of all, patients with NASH were recently shown to have a decreased abundance of bacteria belonging to the phylum Bacteroidetes compared to subjects with simple steatosis and healthy individuals as shown by qPCR [9]. In contrast, studies using sequencing techniques showed an increase of Bacteroides, one of the most important genera within the Bacteroidetes phylum, and a decrease of Firmicutes in NASH patients as compared to healthy subjects [8]. The lower representation of Firmicutes in NASH patients was especially due to a reduced abundance of the Lachnospiraceae and Ruminococcaceae families. However, another study demonstrated an increase of Lachnospiraceae and Lactobacillaceae in NAFLD patients, albeit without distinguishing between simple steatosis and NASH [10]. Both studies observed a decrease in members of the Ruminococcaceae family in NASH [8, 10]. The apparent lack of consistent changes of gut microbiota composition in NASH is further exemplified by two recent studies, showing either an overrepresentation of the genus Escherichia from the Enterobacteriaceae family in subjects with NASH [8] or no difference in Escherichia coli abundance in NASH patients compared to subjects with simple steatosis [9]. Despite the lack of consistent NAFLD-related gut microbiota changes, the possible overgrowth of these ethanol-producing bacteria may underlie the increased circulating ethanol levels in NASH [8]. The endogenous production of ethanol might, in turn, contribute to the formation of free fatty acids and oxidative stress (Fig. 1), further underscoring the potential role of ethanol-producing bacteria in the pathogenesis of NAFLD.

Large scale trials, designed to identify alterations of microbiota composition in patients with simple steatosis versus NASH, are required to shed more light on the nature of the gut microbiota shifts characteristic of specific stages of NAFLD. These trials should also pay attention to the fact that most NAFLD patients are obese, since obesity itself is linked to gut microbiota composition changes, as reviewed elsewhere [12, 13]. Future studies should ideally include obese non-NAFLD patients or non-obese NAFLD patients [14] to exclude the impact of obesity, or control for obesity in statistical analyses. It should be further noticed that many studies so far excluded all taxa with an abundance below 1 %. However, even low-abundant bacteria such as Akkermansia muciniphila have the potential to profoundly affect host metabolism [15]. In addition, without any fundamental direct evidence provided by fecal transplantation or antibiotic studies, one cannot exclude that the described alterations in the intestinal microbiota are a consequence rather than a cause of liver disease.

Surprisingly little evidence exists to date for an effect of gut microbiota on liver fibrosis. However, very recently, an elegant study was published showing that in a bile duct ligation-induced liver fibrosis model, transplantation of the gut microbiota of mice fed a high fat diet (HFD) aggravated fibrosis relative to transplanting gut microbiota of mice fed chow [16]. This was mainly attributable to an increased abundance of Gram-negative Proteobacteria; a marked decrease of Bifidobacteriaceae was also observed. Specific bacteria of both the Proteobacteria and Firmicutes phyla produce enzymes catalyzing choline conversion into methylamines [17]. The latter may promote liver inflammation via the portal vein, and decreased choline levels have already been associated with fibrosis progression [18]. However, future studies should ascertain to what extent the described microbiota composition changes in fibrosis affect choline metabolism.

Cirrhosis patients often have a higher proportion of potentially pathogenic bacteria and a reduction of autochthonous (resident) bacteria compared to healthy individuals (Table 2) [19‐21]. Common differences include a decrease in families with the potential to convert primary into secondary bile salts (Lachnospiraceae and Ruminococcaceae) in cirrhosis, and overgrowth of the Gram-negative Enterobacteriaceae family, similar to what is observed in NASH [22]. This suggests that bile salts and endotoxin [23] may play a role in the pathogenesis of cirrhosis. Additional gut microbiota changes in patients with cirrhosis include a decrease in Clostridiales XIV and an increase in Enterococcaceae and Staphylococcaeae [19] as well as overgrowth of Veillonellaceae [21].

Another cohort of cirrhotic patients revealed a decrease of the genera Bacteroides, Eubacterium, and Alistipes, whereas Clostridium and Prevotella were increased compared to healthy controls [24]. However, the most abundantly enriched species in these cirrhotic patients belonged to the Streptococcus and Veillonella genera. Remarkably, these genera comprise oral species that might invade the gut and contribute to small-intestinal bacterial overgrowth, which frequently occurs in NASH and cirrhosis [7, 25]. Furthermore, the lower microbial richness and reduced abundance of butyrate-producing species with anti-inflammatory properties (F. prausnitzii, Coprococcus comes, Lachnospiraceae spp., Ruminoccaceae spp.) suggest that patients with cirrhosis have a less “healthy” microbiota [24, 26, 27].

Cirrhotic patients with NASH further demonstrated a decrease in Veillellaceae and an increase in Bacteroidaceae and Porphyromonadaceae families compared to cirrhotic patients without NASH [19]. Although the abundance of Enterobacteriaceae was increased in NASH patients compared to healthy individuals, Enterobacteriaceae in NASH patients within a cirrhotic cohort were not affected. The latter might be due to the high proportion of Enterobacteriaceae in cirrhotic patients [19, 21, 22]. Interestingly, Bajaj and colleagues showed that the microbiome in cirrhotic patients with stable disease remains unaltered over time, suggesting that the composition of the microbiome can be used as a potential disease marker [19]. All in all, it appears that different stages of NAFLD are associated with different microbiota compositions, although these need to be better defined.

It is essential that sampling and storage methods do not modify microbiota composition by themselves. An important consideration in this respect concerns the type of samples collected, which are commonly stool or endoscopic biopsies. Stool samples are non-invasive and easy to obtain. However, the colonic mucosal microbiota has been shown to deviate considerably from stool microbiota, also in cirrhotic patients [49‐51]. Moreover, whereas microbiota composition along the colon was considered homogenous only a decade ago [50, 51], recent high-throughput microbial profiling techniques revealed that the mucosal microbiota varies along the length of the gut [52, 53].

Immediate freezing of microbiota-containing samples is regarded the gold standard for long-term storage. However, this is a logistic challenge in large cohort studies which usually rely on fecal swabs [54]. Fecal swabs are commonly stored for short-term at room temperature in specific media in which obligate anaerobes can survive [55]. Recently, the impact of different sampling (fecal aliquots and fecal swabs) and storage techniques (−80, −20, +4 °C for 1 week and RT for 24 h) on fecal microbiota composition was examined in healthy and diseased individuals [56]. Although no significant effect of storage temperature on microbiota composition was observed during transport (24 h), fecal swabs stored in Cary-Blair medium showed an enrichment of Ruminococcus and Enterobacteriaceae in comparison to fecal aliquots stored at −80 °C. Therefore, it is recommended to use a single uniform method within one study to minimize possible bias. Additionally, it is crucial to only compare results that have been obtained with the same sample methodology and from the same type of sample.

As for the design of studies, large-scale well-defined prospective cohorts of patients that have been carefully phenotyped are essential. In these studies, factors known to influence gut microbiota composition should be taken into account. In particular, diet and use of antibiotics, probiotics, and prebiotics should be well-documented. Furthermore, sex-specific differences in the human colonic microbiota have recently been shown [52] and should be considered when setting up microbiota studies.

The human gut mainly harbors strictly anaerobic microbial species that are difficult to culture. Therefore, several culture-independent techniques have been introduced in recent decades to analyze composition and complexity of the intestinal microbiota. These techniques include (quantitative) polymerase chain reaction [(q)PCR], PCR followed by denaturing gradient gel electrophoresis, fluorescent in situ hybridization (FISH), and DNA microarrays that hybridize ribosomal RNA (rRNA) sequences with probes.

Although these methods are useful for rapid microbial profiling, they usually do not provide detailed taxonomic data. For that reason, the most important techniques nowadays are high-throughput next-generation sequencing followed by bioinformatics analysis. These sequence methods are commonly based on analysis of the 16S rRNA gene which is present in all bacteria and archaea, consisting of nine unique hypervariable regions (V1–V9) and bordered by highly conserved regions [57]. Despite the fact that other genes have been suggested, sequencing of 16S rRNA remains the gold standard to analyze the microbiota in view of the completeness of reference databases, low costs, and advanced bioinformatics software available. Different next-generation sequencers can be used to analyze the 16S rRNA gene. However, the majority utilizes either Illumina sequencing (Illumina Inc., San Diego, CA) or 454 pyrosequencing (Roche, Brandfort, CT). These widely popular approaches have distinct coverage ratios, sequence lengths, and construction mechanisms. 454 pyrosequencing generates longer sequences which increases taxonomic accuracy, whereas Illumina provides higher coverage at lower cost per sample [58]. It should be noted that the selection of primers, the number of sequences, costs, and aim of the experiment are strongly interdependent. Therefore, it is essential to balance these factors to achieve the optimum amount of information.

Although 16S sequencing is the most commonly used technology in microbiome studies, it is important to understand its limitations. Differences in DNA extraction methods, PCR errors, and discrepancy in 16S gene copy numbers all affect the proportions of bacteria detected [59‐61]. The latter leads to bias in the detection of unknown or unclassified bacteria (dead or alive) and taxonomic characterization. Moreover, both 454 pyrosequencing and Illumina display sequencing error rates; pyrosequencing is associated with relatively more insertions and deletions, whereas Illumina has more mismatches [62, 63]. Therefore, it is crucial to consider standardized control sequences to estimate the exact error rate for each experiment. Additionally, despite the fact that sequencing of 16S rRNA provides insight into overall microbiota composition, it does not provide data on their functions and interactions. Finally, it should be noted that it was recently found that in certain conditions, next-generation sequencing data were found to be less representative of the ‘real’ microbiota than culture-based methods [64].

As opposed to sequencing of marker genes such as the 16S rRNA gene, sequencing of the entire genomic content of microorganisms (the ‘microbiome’) provides more specific information on their potential functional roles. These metagenomic or metatranscriptomic analyses, referred to as shotgun sequencing, are especially appropriate for linking microbial communities with functional potential and activity in the human gut. In addition, metabolomics—the quantitative assessment of metabolic responses of organisms to genetic or pathophysiological changes—is a powerful approach and also key to unraveling specific host-microbe interactions. Metagenomic approaches relate the microbiome to phenotype changes in disease, whereas metabolomic approaches relate the metabolic profile with disease phenotypes [65]. The latter analyses are principally based on proton nuclear magnetic resonance (1H-NMR) spectroscopy and mass spectrometry methods. Ideally, these analytical techniques should be complemented for the most detailed characterization of microbial metabolites obtained from feces, blood, urine, or intestinal tissues. In contrast to just sequencing the microbiota, shotgun-sequencing and metabolomics require direct freezing of fecal samples.

16S rRNA gene sequences can be analyzed with several tools, especially with QIIME [66] and Mothur software [67], which are reviewed in detail elsewhere [65, 68, 69]. These approaches produce phylogenetic trees and assessments of bacterial diversity within samples (α-diversity) and between samples (β-diversity). To this end, sequences are clustered into taxonomic groups, referred to as operational taxonomic units (OTUs), by comparing them across samples (de novo) or to reference data such as Greengenes [70]. Importantly, a reference-based approach provides a more straightforward interpretation and enables the comparison of data with different sequenced regions of the 16S rRNA gene. However, a crucial consideration and limitation of this technique is the common existence of unclassified bacteria in these databases and the complex taxonomic clustering including discrimination of related bacterial types. To reduce the effect of sequencing errors, it is recommended to disable the formation of new clusters with sequences that were not detected in any reference database. Fortunately, the increasing developments in the field of microbiota contribute to a high rate of classifying and discovering novel bacteria species.