Non-alcoholic fatty liver disease (NAFLD) is a severe liver disease that is increasing in prevalence with the worldwide epidemic of obesity and its related insulin-resistance state. A ‘two-hit’ mechanism has been proposed; however, the complete physiopathogenesis remains incompletely understood. Evidence for the role of the gut microbiota in energy storage and the subsequent development of obesity and some of its related diseases is now well established. More recently, a new role of gut microbiota has emerged in NAFLD. The gut microbiota is involved in gut permeability, low-grade inflammation and immune balance, it modulates dietary choline metabolism, regulates bile acid metabolism and produces endogenous ethanol. All of these factors are molecular mechanisms by which the microbiota can induce NAFLD or its progression toward overt non-alcoholic steatohepatitis.
The epidemic of obesity [1, 2] has led to the dramatic increase of its related metabolic diseases, namely insulin resistance, type 2 diabetes and non-alcoholic fatty liver disease (NAFLD) . The mechanisms involved in weight gain and the development of obesity are numerous and complex, and research continues to uncover new factors. In the past few years, a potential role of the gut microbiota has emerged in weight regulation [4-8]. The gut microbiota is now considered as a major metabolic internal organ, composed of >1014 microorganisms and containing a second genome (named the metagenome), which is up to 100–400 times that of humans . Culture-independent, large-scale tools  and associated projects such as the Human Microbiome Project  or the MetaHit consortium  have enabled major breakthroughs in the understanding of gut microbiota composition and functions in different pathological conditions. Data suggest an important impact of the gut microbiota on health  and in the pathogenesis of certain inflammatory and metabolic  diseases such as type 2 diabetes  and obesity. Recent literature also points to a potential role in the development of NAFLD.
Non-alcoholic fatty liver disease is a very prevalent and severe disease that can lead to cirrhosis, liver carcinoma  and death related to liver morbidity. In addition, data show that NAFLD correlates with increased cardiovascular risk assessment scores and most of the clinical surrogates of cardiovascular diseases. A few smaller studies have suggested that NAFLD induces not only increased risk of patent cardiovascular disease, independently of usual risk factors [16, 17] and of other components of metabolic syndrome, but also increased risk of related mortality [18, 19]. However, no treatment has yet proven effective to improve non-alcoholic steatohepatitis (NASH) [20, 21]. Finding novel mechanisms for the pathogenesis of NASH, in particular involving the microbiota, could emphasize new research areas to develop new therapeutic targets.
This review summarizes what is currently known of microbiota composition in obesity and the physiopathogenesis of NAFLD in that context. Recent literature suggests a role of the gut microbiota in NAFLD but mainly in animal models. We therefore aim to present the putative molecular mechanisms linking fatty liver and gut microflora.
NAFLD and Obesity
Non-alcoholic fatty liver disease is a frequent disease, occurring in 16–30% of the general population, depending on the assessment method [22, 23]. Its prevalence rises  in parallel with the worldwide epidemic of obesity and metabolic diseases , reaching 50–90% of obese individuals, among whom more than one-third present with overt NASH . The diagnosis of NAFLD is made on histological findings, first described by Kleiner and Brunt and their colleagues [27, 28]. It includes a wide spectrum of lesions, starting with steatosis, characterized by the accumulation of triglycerides within the hepatocytes , in the absence of other liver disease or significant alcohol consumption . Some patients will also develop hepatocyte injuries, such as ballooning and inflammatory infiltrates, both responsible for NASH, with or without concomitant collagen deposition (fibrosis) that potentially leads to end-stage cirrhosis . Recently, a new algorithm also based upon histological determination has been proposed and validated to better define early stages of NASH, most particularly in morbidly obese patients . Although liver biopsy is the gold standard to confirm the diagnosis, it is an invasive and costly procedure, potentially responsible for secondary effects [32, 33]. Furthermore, considering the ever-increasing prevalence of both diabetes  and obesity , and the subsequent number of patients at risk for liver alterations, it cannot be considered a practical, efficient and large-scale tool to identify those at risk of NASH or advanced fibrosis. In that context, a clinical scoring system including type 2 diabetes, hypertension, sleep apnea and ethnicity has been suggested to propose liver biopsy for those at high risk of NASH . Other imaging techniques include fast magnetic resonance imaging , computed tomography, ultrasound  and emerging proton magnetic resonance spectroscopy, which is a reliable non-invasive tool to quantify hepatic triglyceride content . The last of these correlates closely with histopathological grade  and allows larger sampling than liver biopsy (8–27 cm3). Recently, non-invasive tests, initially used in viral hepatitis B  and C , have been developed, that are currently validated to screen NAFLD in at-risk populations. For example, Fibromax®, which is an algorithm including an association of gender, age, weight, height and numerous serum biomarkers, has shown good reliability in the prediction of liver abnormalities including steatosis , NASH  and fibrosis, in both overweight patients and in those at different stages of obesity [45-47].
Regarding the pathogenesis of NAFLD, for a long time, a ‘two-hit’ mechanism was proposed to explain the progression in liver alteration stages, where the first hit consists in hepatocyte lipid accumulation mainly due to obesity and insulin resistance . Then a second hit, occurring in some patients only, plays a role in the shift from steatosis to NASH [49, 50]. Several factors have been incriminated in the pathophysiology of NAFLD such as oxidative stress, systemic inflammatory mediators and chronic intermittent hypoxia [51-53]. Furthermore, adipose tissue inflammatory tone  also seems to play an important role, because patients with more severe stages of NASH display increased macrophage accumulation in visceral adipose tissue . Moreover, their deep subcutaneous adipose tissue not only displays increased macrophage infiltration but also increased inflammatory gene expression compared with superficial adipose tissue . These associations between macrophage accumulation in adipose tissue and NASH stages were found independently of the diabetes status in morbid obesity.
More recently, an alternative theory favours the determinant effect of free fatty acid lipotoxicity in liver injury, which leads to NASH and occurs in parallel with triglyceride droplet accumulation (i.e. steatosis) . Hepatic steatosis is known to develop in the context of an imbalance of triglycerides afflux in the liver, which is dramatically increased during obesity and even more so when associated with insulin resistance. Indeed in that context, there is a concomitant increased supply (coming from diet, de novo lipogenesis  and adipose tissue) and decreased degradation (via impaired β-oxidation) . A special focus should be placed on fructose consumption, which is dramatically increased in the western diet . Indeed, its metabolism induces both de novo hepatic lipogenesis and reactive oxygen species (ROS) production and therefore participates in the two ‘hits’ of NAFLD physiopathology . Yet to date, the complete pathogenesis of NAFLD is still unknown and many more factors remain still to be uncovered, in particular the precise role of gut microbiota.
Microbiota Composition in Obesity
Studies in mice indicated a relation between gut microbiota and weight regulation. Indeed, germ-free mice displayed reduced adiposity despite increased food consumption when compared with conventional animals. Most importantly, colonization with caecal content from conventional mice induced a rapid fat mass gain, even though those mice ingested less food [4, 61, 62]. The physiopathogenesis by which microbiota plays a role in obesity development was finally confirmed after manipulation in mice and implies multiple mechanisms. The gut microbiota is able to process otherwise indigestible dietary polysaccharides [62-64] into short-chain fatty acids  that can subsequently be absorbed by the intestine. Moreover, the microbiota directly impacts gene regulation to favour increased storage into adipose tissue . Finally, experiments comparing the feces of obese and lean individuals demonstrated that the level of short-chain fatty acids was higher in the obese whereas residual calories from food were concomitantly reduced. These results suggested that the microbiota from obese animals displayed increased capacity to extract and subsequently store energy compared with that of lean animals [65, 66]. Since then, studies in mice using 16S rRNA sequencing methods have demonstrated that microbiota composition differs in obese and lean individuals, with increased Firmicutes and decreased Bacteroidetes levels in the obese although they ingested the same amount of food . The same result was also found in a small study in humans, where obese individuals also displayed increased Firmicutes and decreased Bacteroidetes compared with lean ones , data that were later confirmed in larger cohorts [69, 70]. However, this result remains under debate because conflicting data have also emerged [66, 71, 72]. Indeed, using two techniques, Duncan et al. observed no differences in the numbers of Bacteroidetes in obese or lean patients [71, 72]. Likewise, Jumpertz et al. found the same level of the three dominant phyla (namely Bacteroidetes, Fimicutes and Actinobacteria) in groups with different body mass index .
Those discrepancies have led to further studies, using large-scale metagenomic tools that enabled interesting discoveries:
First, recent data in humans have focused on the importance of bacterial diversity rather than the composition in different phyla . Interestingly, in obese populations with a high prevalence of obesity-related disease, the overall number of gut microbiota was reduced. This is in line with previous results in mouse models  and humans , where greater corpulence was associated with reduced diversity. Likewise, children from rural Africa displayed increased microbiota richness and biodiversity than same-age healthy children from Western Europe .
Second, although gut microbiota display large inter-individual variability, a ‘core’ microbiota was pointed out since 40% of the genes from each individual were shared with at least half of the individuals in the cohort . Most importantly, based on studies of the microbiome in different large groups of patients, three Enterotypes, which can be compared with blood types, have been identified, that are not driven by corpulence. These Enterotypes differ according to the abundance of the three dominant genera, each of them able to process certain types of nutrients . Indeed, Enterotype 1, enriched in Bacteroides spp., has been associated with long-term consumption of animal proteins and saturated fat whereas Enterotype 2 (dominated by Prevotella spp.) is associated with a carbohydrate-based diet . Gut microbiota composition seems to have a strong link with long-term food habits in adults . Moreover, comparison between the microbiota from children in Western Europe and in rural Africa revealed that the latter had low levels of Firmicutes and increased levels of Bacteroidetes in a link with major food consumption differences, namely high amounts of plant polysaccharides . Other aspects of the diet have also been related to the microbiota. Indeed, the overall increase in calorie content induced a concomitant rapid increase in numbers of Firmicutes and a decrease in Bacteroidetes in lean and obese humans . Likewise, when submitted to a high-fat [79, 80] or western [61, 81] diet, both mouse and humanized mouse models displayed the same changes. Finally, Ravussin et al. suggested that the increase in fat content rather than weight modifications drove the increase in numbers of Firmicutes . If those studies demonstrate the unquestionable effect of acute changes in the diet on the rapid changes in microbiota composition, it does not translate in to a switch from one Enterotype to the other .
NAFLD and Microbiota
Several lines of evidence suggest a strong interaction between gut flora and liver. Deriving from its anatomical position, the liver receives 70% of its blood supply from the intestine through the portal vein, so it represents the first line of defence against gut-derived antigens, and one of the most exposed organs to gut-derived toxic factors, such as bacteria and bacterial by-products . The link between gut microbiota and the development of NAFLD has been shown in mice and in humans. In mice, Bäckhed et al. observed that the transplantation of normal caecal microbiota to germ-free mice induced, 15 days later, a 60% increase in body fat along with a more than two-fold increase in hepatic triglyceride content . In the early 1980s, a parallel development of NASH and small intestinal bacterial overgrowth was reported in humans after intestinal bypass. Interestingly, hepatic steatosis regressed after antibiotic treatment (namely metronidazole) , suggesting a possible causative role for microbiota in NAFLD.
Gut microflora may stimulate hepatic fat deposition and promote NASH though several mechanisms (Fig. 1):
It promotes obesity by improving energy yield from food (see Microbiota and Obesity)
It regulates gut permeability, low-grade inflammation and immune balance
It modulates dietary choline metabolism
It regulates bile acid metabolism
It increases endogenous ethanol production by bacteria
Lipopolysaccharide and Toll-like-receptor 4 signalling
NAFLD and other insulin-resistant states are associated with activation of the innate immune system. The gut microbiota participate in the development and homeostasis of the overall immunity of the host . Indeed, the cross-talk between host and bacteria at the mucosal interface is responsible for innate and adaptive immune responses that protect the host and maintain intestinal homeostasis. This depends on specific pattern recognition receptors, including Toll-like receptors (TLRs) and NOD-like receptors that recognize highly conserved microbial molecules called ‘pathogen-associated molecular patterns’ (PAMPs), whereas the endogenous products are distinguished in damage-associated molecular patterns (DAMPs) [83, 86, 87]. The TLRs, acting as immune sensors of PAMPs and DAMPs, initiate an adaptive immune response and a signalling cascade leading to activation of pro-inflammatory genes, such as tumour necrosis factor-α (TNF-α), interleukin-6 (IL-6), IL-8, IL-12 . The most studied PAMP is the lipopolysaccharide (LPS), a component of the gram-negative bacteria cell membrane, the active component of endotoxin. The latter binds to lipopolysaccharide-binding protein (LBP), which subsequently binds to CD14. The LPS–LBP–CD14 complex activates TLR-4, present in Kupffer cells, triggering an essential inflammatory cascade, including stress-activated and mitogen-activated protein kinases, Jun N-terminal kinase, p38, interferon regulatory factor 3 and nuclear factor-κB pathway [88, 89]. Translocation of nuclear factor-κB to the nucleus induces the transcription of numerous pro-inflammatory genes such as TNF-α and IL-1β . Other bacterial endotoxins such as peptidoglycan also contribute to stimulate pro-inflammatory cascades though activation of the NOD1 receptor . Many studies, mostly carried out in animal models, enabled LPS–TLR-4 signalling activation to be related to insulin-resistance and NASH. Cani et al., first described how a 4-week high-fat diet led to a moderate increase in plasma LPS concentration in mice, defining this effect as ‘metabolic endotoxaemia’, because this increase was still 10–50 times lower than values that could be reached during septicaemia [80, 92]. Importantly, 4 weeks of continuous infusion of LPS in mice mimicked the high-fat diet phenotype, namely an increase in insulin resistance, liver triglyceride content and adipose tissue inflammation. Besides, they demonstrated that these effects were CD14-dependent, because the phenotype was significantly reversed in CD14 knockout mice . The role of the gut microbiota was further demonstrated by antibiotic treatment reducing the intensity of the disease in high-fat diet and ob/ob mice [93-95]. Since then, other studies in TLR-4 null mice (–/–) have confirmed that TLR-4 is essential for hepatic fat deposition and NASH development [96-98]. Hepatic stellate cells might also have an important role in generating the liver inflammatory cascade linked with endotoxinaemia. Indeed, they have been shown to be the target through which TLR-4 promote fibrogenesis via enhancement of transforming growth factor-β (TGF-β) signalling [99, 100].
Another important contributor to microbiota dysbiosis, participating in the activation of lipid peroxidation and ROS production—both involved in the second-hit mechanism in NAFLD/NASH progression—is the inflammasome. Inflammasomes are cytoplasmic multi-protein complexes composed of leucin-rich-repeat containing proteins and nucleotide-binding domain (NLRPs) [101, 102], which are sensors of PAMPs and DAMPs. They govern cleavage of pro-inflammatory cytokines such as IL-1β and pro-IL-18. Most DAMPs induce ROS production, which is known to activate the NLRP3 inflammasome . Recently, Henao-Mejia et al., showed that changes in gut microbiota associated with NLRP6 and NLRP3 inflammasome deficiency were linked with exacerbated hepatic steatosis and enhanced TNF-α expression. Furthermore, co-housing inflammasome-deficient mice with wild-type mice transferred the phenotype, providing direct evidence that inflammasome-mediated dysbiosis is implicated in NASH progression . Studies in humans have demonstrated increased endotoxin levels in adults and children with NASH [105-107]. Furthermore, several studies showed that endotoxin levels increased with the severity of the disease [108, 109]. Likewise, chronic endotoxinemia, which behaves as a real ‘hepatotoxin’, may be of particular importance in inducing hepatic inflammation, fibrosis and insulin resistance.
Alteration of intestinal permeability (leaky gut) and small intestinal bacterial overgrowth
The gut epithelium plays a central role in demarcating microbes in the gut from the host immune system. Gut epithelial cells are linked to one another with tight junctions, which play a pivotal role in maintaining intestinal barrier integrity . NAFLD and steatohepatitis have been associated with small intestinal bacterial overgrowth and increased intestinal permeability [4, 111]. Miele et al., provided the first evidence of an increased intestinal permeability (leaky gut) and tight junction alterations in 35 patients with biopsy-proven NAFLD compared with healthy subjects. Furthermore, both gut permeability and the prevalence of small intestinal bacterial overgrowth correlated with the severity of steatosis, although not with the presence of NASH . Wigg et al. found small intestinal bacterial overgrowth in 50% of patients with NASH, compared with 22% in controls ; More recently, Gäbele and co-authors have presented novel experimental evidence about the association between impaired intestinal barrier function and hepatic fibrogenesis and inflammation . The authors induced NASH by feeding C57BL/6 mice with a high-fat diet and then studied the effects of exposing the mice to 1% dextran sulphate sodium. Such exposure led to intestinal epithelial injury. Importantly, they observed that the combined administration of a high-fat diet and dextran sulphate sodium not only worsened steatohepatitis, but also induced a pro-fibrogenic response in the liver. All together these data support the theory that disturbances in the homeostasis between bacteria and host at the intestinal epithelial cell level lead not only to altered intestinal barrier but may also promote bacterial translocation from the gut into the portal circulation, further inducing liver damage [87, 115]. However, the nature and type of translocated microbial components remains to be studied. Lastly, growing evidence suggests that at the early onset of a high-fat diet, not only bacterial products but also complete living bacteria can be translocated from the intestinal lumen towards tissues, such as the adipose tissue [116, 117], enhancing the role of bacterial translocation and gut permeability in tissue injury. The accumulation of translocated bacteria in the metabolically active tissues can also participate in exacerbating hepatic inflammation and fibrosis
Altered choline metabolism
Choline is an important phospholipid component of the cell membrane, and a key partner of fat metabolism in the liver, and of very-low-lipoprotein assembly. Lastly, it promotes lipid transport from the liver . A choline-deficient diet promotes liver steatosis, which is reversible by choline infusion . Enzymes produced by the gut microbiota catalyse the conversion of dietary choline into toxic methylamines (dimethylamine and trimethylamine). The uptake by the liver of those amines and their transformation to trimethylamine-N-oxide can induce liver inflammation [120, 121]. Interestingly, Spencer et al. showed that variations in levels of Gammaproteobacteria and Erysipelotrichi in human faecal microbiota were directly associated with changes in liver fat during choline depletion . Hence, microbiota dysbiosis can promote NASH both by decreasing choline levels and increasing toxic methylamines [123, 124].
Modification of bile acid metabolism
Representing the main components of bile, bile acids are secreted into the duodenum and work to emulsify liposoluble dietary nutrients to facilitate their digestion and absorption. They also have a strong antimicrobial activity. Bile acids damage bacterial cell membranes by interacting with membrane phospholipids, which results in bactericidal activity . Recent studies have demonstrated that dietary fats (high in saturated fat), by promoting changes in host bile acid composition, can markedly alter conditions for gut microbial assemblage, resulting in dysbiosis [126, 127]. Reciprocally, the gut microbiota is able to modulate bile acid metabolism, through farsenoid X receptor stimulation . Dekaney et al. showed that farnesoid X receptor (FXR) null and germ-free mice failed to induce the expression of bile acid transport genes, following ileo-caecal resection, compared with wild type mice . Indeed, bile acids are ligands for a G-protein-coupled receptor (TGR5/Gpbar-1) and activate nuclear receptors such as FXR (NR1H4). FXR and its downstream targets play a key role in the control of hepatic de novo lipogenesis, very-low-density lipoprotein-triglyceride export and plasma triglyceride turnover , whereas TGR5 binds secondary bile acids and promotes glucose homeostasis, by stimulating secretion of glucagon-like peptide 1 . Administration of specific TGR5 agonists lowers serum and liver triglyceride levels, thereby reducing liver steatosis . By modifying bile acid metabolism and FXR/TGR5 signalling, gut flora could therefore contribute indirectly to the development of NAFLD .
Hepatotoxic microbial by-products: role of ethanol
Intestinal microflora produces a number of potentially hepatotoxic compounds such as ethanol, phenols, ammonia, which are delivered to the liver by the portal circulation. Those compounds activate Kupffer cells (resident hepatic macrophages) and stimulate their production of nitric oxide and cytokines . Acetaldehyde and acetate are two major metabolites of ethanol. While acetate is a substrate for fatty acid synthesis, acetaldehyde may lead to the production of ROS. This could be involved in liver injury by contributing to the disruption of intestinal barrier function and to the two-hit mechanisms of NASH. Remarkably, Zhu et al. studied gut microbiota of patients with and without NASH in obese individuals and in healthy children. They observed an increase of alcohol-producing bacteria in the gut microbiota of children with NASH, associated with an elevated blood alcohol concentration, without dietary alcohol consumption . Furthermore, they reported that Escherichia (a well-known ethanol producer ) stood out as the only abundant genus that differed between NASH and obese patients without NASH. Interestingly, changes in Enterobacteria seem to be a frequent feature in obesity and weight changes. Importantly, this endogenously produced alcohol has a well-established role in the generation of ROS and consequently liver inflammation . Moreover, this could participate in the increase of gut permeability [88, 136, 137].
Gut microbiota modulation as a new therapeutic strategy
Probiotics are defined as live commensal microorganisms, that when consumed in adequate quantities, confer a health benefit to the host (FAO/WHO 2001) . Probiotics have been suggested as a treatment to prevent chronic liver damage, because they prevent bacterial translocation and epithelial invasion, inhibit mucosal adherence by bacteria, produce antimicrobial peptides, decrease inflammation and stimulate host immunity [139, 140]. The main probiotics on the market are lactobacilli, streptococci and bifidobacteria . Although the literature on the subject is large, it is difficult to draw conclusions about the true effect of probiotics on NAFLD. Indeed the animal models and bacterial strains tested across the studies are different. Likewise, it must be noted that gut microbiota will always outnumber the probiotics that can be administered. Moreover, probiotic processes will always be confounded by the variability of the human microbiome, diet and genetics [142, 143]. Several animal models have provided evidence that probiotics may reduce NASH progression [144-148]. The most studied probiotic is VSL#3 mixture, which was first given for 4 weeks in ob/ob mice fed with a high-fat diet . Treatment with VSL#3, improved liver histology, reduced hepatic total fatty acid content, and decreased serum alanine aminotransferase levels, a surrogate marker of NAFLD. Ma et al. provided further evidence that VSL#3 ameliorates insulin resistance and hepatic steatosis by preventing the expected natural killer T-cell depletion induced by the high-fat diet . Several strains of lactobacillus have demonstrated a protective effect on NAFLD [149-151], whereas a recent meta-analysis pointed out the association of certain species (L. fermentum and L. ingluviei) with weight gain . Probiotics have been shown to have several anti-inflammatory effects, which could contribute to clinical benefit in NAFLD, reviewed elsewhere . Regarding human studies, preliminary data showed that both VSL#3 and a symbiotic (association of pro/prebiotic) given to NAFLD patients for 2–3 months improved liver enzymes, TNF-α and oxidative stress markers . More recently, two randomized double-blind placebo-controlled studies, showed a significant decrease in liver aminotransferases with probiotics in children  and in adults . These promising results are strongly indicative of a great potential benefit to use probiotics in the treatment of NAFLD. Nonetheless, stated in a Cochrane meta-analysis, larger randomized studies are still needed .
Prebiotics are indigestible carbohydrates that stimulate the growth and activity of beneficial bacteria, particularly lactobacilli and bifidobacteria . Lactulose is a commonly used prebiotic, which has proven its efficiency in reducing symptoms in liver patients by increasing bifidobacteria . Other prebiotics include inulin-type fructans that can be differentiated by their degree of polymerization such as long-chain and short-chain, i.e. oligofructose and fructo-oligosaccharides . Several beneficial effects of prebiotics on gut microbiota, serum lipids and liver physiology have been shown . Besides, in several animal models of NAFLD, lactulose and oligo-fructose have been shown to reduce hepatic steatosis development [160, 161]. The potential mechanisms include reduced de novo fatty acid synthesis, reduced body weight through stimulation of gut peptides (glucagon-like peptide 1 and peptide YY), reduced inflammation and pro-inflammatory cytokines (IL6, TNF-α), improved glycemic control, modulation of gut microbiota and alterations in volatile organic compounds [83, 159]. Yet, studies assessing the effects of prebiotics on NAFLD in humans are lacking. Nevertheless, a pilot study already gives encouraging results, since Daubioul et al. showed a reduction in liver enzymes and fasting insulinaemia in seven patients with NASH who were treated with oligo-fructose for 8 weeks .
Gut–liver axis plays a central role in the pathogenesis of NAFLD, mainly through the crosstalk of the intestinal microbiota with the host immune system modulating inflammation, insulin resistance and intestinal permeability. Large-scale randomized trials using antibiotics, probiotics and prebiotics with imaging and histological endpoints are needed.