Non-alcoholic fatty liver disease (NAFLD) is the commonest form of liver disease in all regions of the world with modern industrialised economies, including Korea and many other Asian countries.1-6 Patients usually present without symptoms or clinical features are non-specific. Instead, liver abnormalities are found incidentally by hepatic imaging, particularly ultrasonography, and/or there are raised liver enzymes (alanine aminotransferase [ALT] and gamma-glutamyltranspeptidase).7-9 The diagnosis of NAFLD requires exclusion of other disorders, particularly viral hepatitis, significant alcohol intake, and exposure to potentially hepatotoxic medications. By agreements such as the Asia-Pacific Guidelines on NAFLD,6 the term NAFLD is now retained for cases of fatty liver associated with metabolic complications of over-nutrition, usually with central obesity and overweight.

We and others have stressed that NAFLD is closely allied to pre-diabetes and metabolic syndrome.3,10,11 As recently reviewed,3 the evidence for this includes the strong risk factors for NAFLD posed by obesity, insulin resistance, glucose intolerance and one or more components of metabolic syndrome, and the corresponding strong risk for onset of type 2 diabetes and cardiovascular disease/events conferred by a fatty liver.12-16 Community based studies from Korea, Japan and other areas in North Asia have been highly informative for understanding that NAFLD is not so much a "Western disease" as the inevitable result of changes in prosperity and lifestyle that have increased the prevalence of overweight/obesity, insulin resistance, type 2 diabetes and cardiovascular risk factors (clustered as metabolic syndrome).3,4 Thus the community prevalence of NAFLD in this region increased from less than 10% in the 1980s, through 10% to 20% in the 1990s, to current rates of 15% to 30% or higher.4,17

The known ethnic differences in metabolic complications of over-nutrition, such as insulin resistance, diabetes, metabolic syndrome and hypoadiponectinemia, are also consistent with the proposition that, like them, NAFLD is a genetic disorder.3,18,19 Thus, an encompassing concept for NAFLD pathogenesis is that it represents the outcome of genetically determined interactions between a changing environment and a susceptible host. In this case, the environmental factors include too much energy intake, particularly in the form of cheap, highly processed simple carbohydrates and saturated fats, and reduced levels of physical fitness resulting from sedentary lifestyles.20,21 Of particular interest to the present review, one prevalent genetic polymorphism predisposing to steatosis in overweight persons of European or Hispanic ancestry, PNPLA3, does not operate by increasing the risks of diabetes or metabolic syndrome.18,22-25 Instead, it correlates with serum ALT levels,26 reflecting liver injury or inflammation, and with more severely fibrotic liver disease in both NAFLD/non-alcoholic steatohepatitis (NASH) and alcoholic cirrhosis.27,28 This point emphasises that not all cases of NAFLD have the same implications for liver disease.

NAFLD embraces a pathological spectrum of liver disease, from cases of steatosis with virtually no evidence of hepatocellular injury or liver inflammation, often referred to as simple steatosis or "not NASH," through steatohepatitis (NASH), to cases with cirrhosis.29-31 The latter are often complicated by portal hypertension and hepatic decompensation, and occasionally present with hepatocellular carcinoma (HCC).32 At this late stage, steatosis and liver inflammation may both have resolved; they are cases of "cryptogenic cirrhosis." As discussed next, natural history and clinical outcome studies based on community and liver clinic cohorts indicate a nearly 2-fold increase in standardized mortality rates among persons with NAFLD.33-37 Further, while cardiovascular disease and common cancers remain the two most common causes of death, liver-related mortality ranks the third most common, as compared to 13th in the general community.36 A key question emerges: what aspects of liver pathology, and what disease mechanisms, account for progression of NAFLD to cirrhosis and its fatal complications?

1. Fibrotic severity

The observation that histologic characteristics are useful in predicting the outcome of patients with NAFLD is best exemplified for patients at either end of the pathological spectrum. At one end, individuals with only hepatic steatosis (simple steatosis) infrequently show signs of any histologic progression, and are not at significant long-term risk of liver-related death.33,34,38 By contrast, those with advanced hepatic fibrosis (bridging fibrosis [F3] and/or cirrhosis [F4]) are likely, in time, to experience liver-related complications (ascites, variceal bleeding, and/or HCC).35-37 While cardiovascular disease and cancer head the list of causes of death, 7- to 10-year liver-related mortality (12% to 25%) ranks third overall.2,36,37 In fact, the outcome of patients with advanced NAFLD (Child-Pugh B and C) is similar to that of individuals with hepatitis C virus-related cirrhosis.35,37

In reaching these general conclusions, certain assumptions are implied. First, the necessity for histologic appraisal is problematic because liver biopsies are performed less often outside research studies and clinical trials due to patient and clinician perceptions that the result will not influence management, and the concerns about biopsy-related complications. While non-invasive assessment of hepatic necroinflammatory activity and hepatic fibrosis (serum biomarkers, transient elastography) is increasingly advocated,39-44 it is most reliable at either end of the clinical spectrum of severity (mild, severe), when histology is most predictable. It remains suboptimal in the substantial number of patients in patients with mild-moderate hepatic fibrosis (F1, F2), among whom liver disease may progress.34

Second, in patients with only hepatic steatosis there can be changes in host characteristics over time, such as increasing body weight or worsening insulin resistance and/or development of diabetes, and baseline steatosis and necroinflammatory severity have not been correlated with such progression of metabolic disease.12,34 These considerations not withstanding, most gastroenterologists and hepatologist would generally reassure patients with isolated hepatic steatosis about their liver prognosis, but recommend primary care follow-up of cardiovascular risk factors and lifestyle interventions to address these. Conversely patients with advanced hepatic fibrosis should enter a more rigorous liver follow-up protocol.

2. Presence of NASH (versus "not NASH")

Current uncertainty about how "progressive' this condition really is at least partly stems from the use of differing operational definitions for NASH.45,46 Thus, NASH has been variously defined to include cases with hepatic steatosis and lobular inflammation (regardless of hepatic fibrosis),29 hepatic steatosis with lobular inflammation and ballooning of hepatocytes with or without fibrosis,33,47,48 or as separate scoring systems for "activity" (the NAFLD activity score, which assigns numerical scores to steatosis, lobular inflammation and ballooning and fibrosis (the latter usually F0-F4).30 The Brunt system29 was developed by correlating histologic changes with serum aminotransferases (AT) as a measure of hepatic necroinflammatory activity, and not with clinical outcome, whereas the scoring system proposal by Kleiner et al.30 was never intended for diagnosis but was to be used as a tool for assessing serial liver biopsies in clinical trials. The premise has been that small changes could be identified more clearly and reliably by assigning numerical values than by descriptive remarks.45

A head-to-head comparison of these different histologic classification systems has recently been reported,47 and an editorial based on additional data from Korea reached similar conclusions.46 Both authors recommended the following. First, for routine clinical use (i.e., for diagnosis), an indication that there is or is not steatohepatitis is probably sufficient, with an intermediate category of "borderline" steatohepatitis where there is some uncertainity. Second, among the various components of steatohepatitis, ballooning degeneration of hepatocytes is broadly favoured for defining NASH.45-47 In one study, ballooning degeneration was found to correlate with liver-related mortality, but only by univariate analysis.47

In summary, the combination of hepatic fat and lobular inflammation is now regarded as insufficient for a diagnosis of NASH. However, other features such as the presence of "more than mild" portal inflammation,29,33 or the presence of panacinar steatosis (as compared to isolated zone 3 steatosis),48 have also been associated with advanced hepatic fibrosis. The latter is the best histologic predictor of liver-related mortality irrespective of the degree of steatohepatitis.49 As expected from the earlier discussion, classification systems incorporating hepatic fibrosis in the definition of NASH correlate well with liver-related mortality,33,47 while systems that do not are not predictive of future outcome.29,30 It needs to be stated, however, that the latter systems do include staging for hepatic fibrosis, but do not require its presence for the definition of NASH.

3. Extent of necroinflammatory activity

Having established that fibrotic NASH is all that matters, is there any value in assessing the degree of necroinflammatory activity? It would be if it could be determined that the grade of inflammation is a predictor of future hepatic fibrosis (in the case of liver outcomes) or metabolic syndrome-related disorders (in the case of overall mortality). Some evidence supports this view,50 although negative studies have also been reported.12 A systematic review showed clearly that age and inflammation on the initial biopsy (hazard ratio, 2.5) were the main independent risk factors for fibrosis progression.51 These findings are not surprising because clinicians are familiar with the need to 'damp down' hepatic inflammation in chronic viral hepatitis B or C and autoimmune hepatitis in order to achieve a favourable clinical outcome by preventing or reversing progression of hepatic fibrosis.

4. Liver histology and cardiovascular outcomes

After establishing that NASH is the hepatic component of the metabolic (insulin resistance) syndrome,2,3,52,53 it was to be anticipated that morbidity and mortality from cardiovascular disease would be highlighted in long-term studies of NAFLD. Surrogate markers of atherosclerosis (e.g., carotid intima-media thickness) are present even in adolescents with NAFLD, and clinical endpoints such as deaths from myocardial infarction/need for coronary revascularisation have been documented in several natural history studies of NAFLD.36,37 The concept that fatty liver may also drive the inflammatory cascade of atherosclerosis is now gaining acceptance. There is some evidence that individuals with NASH have a worse atherogenic profile,54 and are more likely to have overt cardiovascular disease than patients with hepatic steatosis alone.12 In summary, based on present somewhat limited evidence, it can be concluded that ongoing hepatic necroinflammatory activity in patients with NAFLD increases the risk of future cardiovascular disease, and confers a higher risk of unfavourable liver-related outcomes by promoting development of hepatic fibrosis.

Inflammation is a critical response to tissue damage or infection in which secreted mediators such as cytokines, chemokines and eicosanoids coordinate cellular defences and tissue repair. Since this is generally a whole body response, it is possible that inflammation affecting or infiltrating the liver in NASH may originate outside the liver. One site of interest is the adipose, particularly visceral adipose which is expanded in NAFLD.3,55,56 Visceral adipose is inherently pro-inflammatory,57-59 but inflammation also occurs in stressed, de-differentiated subcutaneous adipose tissue in obesity. Important consequences include the release of macrophage chemokines and cytokines, notably macrophage chemotactic protein 1 (MCP-1) and tumor necrosis factor-α (TNF-α). A recent time course study showed that in mice fed a high fat (HF), cholesterol-enriched diet, macrophage and cytokine transcripts were up-regulated in adipose at 6-16 weeks, before their appearance in liver from 16 to 26 weeks.60 Key inflammatory signals, including interleukin (IL)-1β, IL-1 receptor antagonist, TNF-α and CD11b+ and CD11c+ macrophages, were particularly associated with liver inflammation. Lanthier et al.61 have likewise shown that macrophage inflammation of adipose is responsible for the early stages of both hepatic and peripheral insulin resistance in HF-fed mice, but deletion of adipose macrophages cannot reverse the later phase once liver inflammation is established.

In other research, a consistent increase in serum MCP-1 has been noted to be part of the systemic and adipose inflammatory state in metabolic syndrome.62-67 Adipocyte-derived MCP-1 (also known as CCL-2) stimulates recruitment of chemokine (C-C motif) receptor 2 (CCR-2)-expressing macrophages into adipose. MCP-1 and its cognate receptor, CCR2, are potentially important molecules in NASH since, like NF-κB and c-Jun N-terminal kinase (JNK), they unite the inflammatory response with insulin resistance,68,69 as reviewed by Maher et al.70 and depicted in Fig. 1. MCP-1 also stimulates lipogenesis in the liver.71 In this way, adipose inflammation can exacerbate steatosis and connect to innate inflammatory responses within the liver.

Inflammation and de-differentiation of adipose also alters release of the key insulin-sensitizing and anti-inflammatory adipokine, adiponectin. Adiponectin blocks elaboration and release of TNF-α.72,73 Serum adiponectin levels fall in metabolic syndrome and type 2 diabetes, while low serum adiponectin levels in NAFLD are inversely related to steatosis severity, and in some studies to the presence of NASH.72-75 Key signalling pathways that explain some of the connections between hepatic inflammation and insulin resistance include the IκB kinases (IKK)/nuclear factor-kappaB (NF-κB) and JNK, as discussed later and reviewed.70

In addition to macrophages recruited to inflamed adipose, circulating lymphocytes and macrophages also contribute to systemic inflammation in metabolic syndrome. For instance, raised serum cholesterol levels are associated with increased secretory function of circulating lymphocytes.76 Conversely, treatment with simvastatin and/or ezetimibe reduced plasma levels of highly-sensitivity C-reactive protein and intercellular adhesion molecule 1. Statin or combination treatment also significantly reduced lymphocyte release of TNF-α, interferon-gamma (IFN-γ) and IL-2, an anti-inflammatory effect that was most marked for patients with insulin resistance.76

Another tissue compartment that could contribute to liver inflammation in NASH is the gastrointestinal tract, more specifically, the gut microbiota. There is evidence of altered gut flora in obesity,77 and of increased mucosal permeability in NASH.77-80 Further, in some animal models sterilisation of gut contents or their modification by probiotic administration to suppress endotoxin production altered liver inflammation or liver injury,81 albeit the models do not conform to what we now categorize as NASH. The topic of intestinal-liver interactions in obesity and fatty liver disease has been reviewed elsewhere,70,80,82,83 and will be mentioned later in respect to activation of innate immunity in the liver.

Notwithstanding the potential relevance of adipose inflammation,84 circulating chemokines, cytokines and inflammatory cells, and the gut microbiota to NASH pathogenesis, the perspective we will take in this review is that one may not need to look much further than at the liver itself to understand the origins of inflammation in NASH.

The liver is comprised of several cell types, each of which could potentially activate or be influenced by hepatic inflammation. Hepatocytes comprise 60% to 80% of all liver cells and conduct the metabolic, biosynthetic, detoxification and biliary secretory functions of the liver. In fatty liver, hepatocytes stain positive for triacylglycerides (TG), and in NASH the defining pathological element is hepatocellular injury, evident as ballooning, Mallory bodies and apoptosis. Among other liver cell types, Kupffer cells (KCs), the liver's resident macrophage population, natural killer (NK) cells, NK T cells, T cells, sinusoidal endothelial cells (SECs) and hepatic stellate cells (HSCs) can each play pro-inflammatory roles.85,86

Several possible mechanisms activate pro-inflammatory pathways in livers with NASH, leading to release of chemokines, cytokines and other pro-inflammatory molecules, as summarised in Table 1. Chemokine release is particularly responsible for recruitment of infiltrating monocyte-derived macrophages, and neutrophils, which together with lymphocytes comprise the mixed cell type inflammatory infiltrate in NASH. Oxidative stress and necrosis can provoke a neutrophil inflammatory response.87 In general, pro-inflammatory signalling in NASH is mediated by activation of innate immune mechanisms. These may be primed by gut-derived endotoxin, but there is increasing evidence that this is in response to lipotoxicity and/or molecules released by stressed hepatocytes (discussed below).

1. Lipotoxicity

The appearance of simple steatosis in the majority of cases indicates that fatty livers are not necessarily pro-inflammatory. However, it now seems likely that the steatotic hepatocytes in NASH contain excess lipid molecules other than TG, and there is mounting evidence that such non-TG lipid molecules are implicated in the pathogenesis of NASH by the process of lipotoxicity.3,88-92 Conversely, formation of TG may actually be a cytoprotective mechanism in liver.89,90 Candidate lipotoxic molecules in NASH have been reviewed;90,92,93 they are summarized in Table 2.

Lipidomic analyses of human fatty livers have identified free cholesterol (FC) but not free fatty acids (FFA), diacylglycerides (DAG) or ceramide among the potential lipotoxic molecules that accumulate selectively in NASH but not in "not NASH' NAFLD livers.84,91,93-95 Lysophosphatidylcholine has also been implicated in a small study.95 Another consistent feature is depletion of very long chain polyunsaturated fatty acids (PUFA); the potential relevance could be impaired production of hepatoprotective eicosanoids. Consistent with this proposal, the plasma lipidomic signature of NASH indicates over-production of proinflammatory (15-hydroxyeicosatetraenoic acid) rather than anti-inflammatory products of lipooxygenase.96

Some potential lipotoxic lipid species implicated in NASH have been explored experimentally, particularly saturated FFA and FC, but also (mostly in dietary studies) PUFA,97,98 sucrose,99 and fructose.100 Such studies demonstrate the unequivocal potential of such lipid molecules to kill cells of hepatocyte lineage, by directly or indirectly activating JNK and the mitochondrial/lysosomal cell death pathway,101 and also to stimulate pro-inflammatory signalling via NF-κB and JNK/activator protein 1 (AP-1), as discussed later. In general, saturated long chain fatty acids (such as palmitic and stearic acids) are more toxic than mono-unsaturated FFA.102-105 There are also data that the effects of palmitic acid may be exerted via formation of lysophosphatidylcholine,95,106 or via reactive oxygen species (ROS),107 or endoplasmic reticulum (ER) stress.105 To date, however, most such studies have been in immortalised cell lines (typically human HCC cells) whose biology differs from well-differentiated hepatocytes and the intact liver (Table 2).

Alternatively, investigators have used animal models whose pathology may resemble NASH but the pathogenesis does not involve obesity, insulin resistance and hypoadiponectinemia.108,109 For example, Mar? et al.110 have elegantly demonstrated how FC accumulates in livers of animals fed a high (2%) cholesterol, choline-deficient diet or high cholesterol/cholate-supplemented diet, sensitizing hepatocytes prepared from such livers to apoptosis via the mitochondrial cell death pathway. In this work, cholesterol-loaded hepatocytes were also exquisitely sensitive to TNF-α-mediated cytolysis, despite unchanged NF-κB expression, which usually confers hepatoprotection to hepatocytes, unless they are depleted of reduced glutathione (GSH).111 Cholesterol loading appears to deplete mitochondrial GSH, rendering cells susceptible to apoptosis via cytokine death receptor signalling (Fas or TNF-R). Such a phenomenon has also been demonstrated for Fas-mediated apoptosis of FC-loaded macrophages.112,113

The most compelling evidence that hepatocytes may be the source of liver inflammation in NASH comes from studies in obese rodents with insulin resistance that leads to hyperinsulinemia and diabetes. We have used mice with a spontaneous mutation of the murine homology of the Alstr?m gene (Alms1[termed foz/foz]),108,114,115 while others have used wildtype (WT) C57B6 mice or rats.100,116-118Foz/foz mice exhibit hyperphagia with early onset obesity and insulin resistance, the phenotype of Alstr?m syndrome. Feeding them a high carbohydrate, HF diet with 0.2% cholesterol accelerates onset of diabetes with marked hypoadiponectinemia.114 The resultant liver pathology shows NASH with fibrosis,108,114,115 whereas chow-fed foz/foz NOD.B10 mice and WT NOD.B10 mice fed the same diet develop only steatosis. Feeding WT C57B6 mice a similar HF diet, and particularly diets with higher cholesterol content (1% or 2%, often supplemented with cholic acid) also leads to unequivocal NASH; the onset is generally later, varying between 6 and 15 months in different reports.119-121 A similar approach, typically with cholesterol-enriched HF diet, can also produce NASH in some lines of rats116,117 and in a line of opossums (ABCB4) that are genetically predisposed to hypercholesterolemia.122 Finally, a HF diet rich in trans fats combined with high-fructose corn syrup equivalent and inactivity (the American Lifestyle-Induced Obesity Syndrome) also caused obesity-related steatosis with moderate necroinflammatory change, albeit in this and most other animal models (the foz/foz mouse is an exception), hepatocellular ballooning, a cardinal feature of human NASH is inconspicuous and there was no fibrosis.100

In HF-fed foz/foz mice, onset of NASH is associated with more than 200-fold increase in liver cholesterol esters (CE), and ~8-fold increase in FC.115 Removal of cholesterol from the HF diet reduces hepatic CE and FC content and ameliorates the severity of liver injury and steatohepatitis.115 Likewise, pharmacological treatments that lowered hepatic cholesterol dampened necroinflammatory severity in this NASH model.123 Conversely, increasing dietary cholesterol (to 2% in foz/foz mice,115 or 1% in other studies with C57B6 mice108-121) worsens inflammation and liver injury in experimental NASH. It is plausible that FC or other cholesterol fractions (7-ketocholesterol and other oxysterols are candidates) could activate KCs and recruited macrophages directly, analogous to processes implicated in atheroma,112,113,124 and demonstrated in low density lipoprotein receptor knockout and apoE knock-in mice.125,126 However, immunofluorescence studies in foz/foz mice (unpublished data) and human livers94 show that hepatocytes are the cell type most conspicuously laden with FC in NASH. The subcellular compartments involved are the plasma membrane, ER and mitochondria.94,115 A noteworthy feature of our studies has been the location of macrophages and neutrophils around heavily lipid-laden and swollen hepatocytes, some of which are ballooned (Fig. 2). Cellular processes could lead hepatocytes to incite inflammatory recruitment in NASH are discussed next.

2. Cytokines and oxidative stress

An earlier concept of NASH pathogenesis envisaged a "two hit" process, in which the abnormal metabolic milieu causing steatosis comprised the "first hit," and the vulnerability of a fatty liver to a separate injurious process ("second hit") resulted in cell death and inflammation.127 Fifteen years ago, the injurious processes of interest were oxidative stress and cytokines, particularly those stimulated by endotoxin (lipopolysaccharides), such as TNF-α.128,129 While both oxidative stress and cytokines are clearly evident in livers with NASH,130-133 the weight of evidence is that TNF-α is a consequence rather than cause of liver inflammation in NASH.2,72,134 Further, serum TNF-α levels increase in obese people, most likely originating from macrophages in the inflamed adipose;84 importantly, values in NAFLD patients do not discriminate NASH from "not NASH."74 It is also salient that some experimental forms of steatohepatitis, including a forced over-nutrition model,135 can occur in the absence of TNF-α or its NF-κB signalling type 1 receptor.136,137

Oxidative stress is a key pro-inflammatory pathway in acute liver injury, such as ischemia-reperfusion injury87 and in some types of steatohepatitis,134 including alcohol-related liver disease, methionine deficiency,138 and methionine and choline deficient (MCD).137,139-142 Older studies employing immunohistochemistry demonstrated evidence of oxidized proteins, lipids and DNA in NASH livers,143-145 but this could be a consequence of inflammation rather than its cause. A potential distraction has been identification of multiple sources of pro-oxidants in NASH, such as mitochondria (from uncoupling of oxidative phosphorylation to release reactive oxygen species), from ER (induction of cytochromes P450 [CYP] 2E1 and 4A),146,147 peroxisomes89 and inflammatory cells (NADPH oxidase).134,148,149 Hepatoprotection from anti-oxidants and anti-oxidant pathways (such as heme oxygenase) has been demonstrated in MCD steatohepatitis,150,151 and vitamin E may have some efficacy against necro-inflammatory change in NASH,152 but there is less evidence for operation of oxidative stress in murine models that link metabolic syndrome to NASH. We agree with the interim conclusion reached by several experts,89,91 that oxidative stress and/or cytokines are not likely to be the initiators of liver inflammation in NASH, although roles in insulin resistance, perpetuation of necroinflammatory change, fibrogenesis and progression towards cirrhosis and hepatocarcinogenesis remain likely.

3. ER stress

Accumulation of unfolded proteins within the ER is often observed in cells like hepatocytes that have high rates of protein synthesis. The cellular responses, collectively known as the unfolded protein response (UPR), involve provision of chaperones, such as 78 kDa glucose-regulated protein (GRP78), for protein refolding and transport out of the ER, and suppression of further protein synthesis.153-155 Failure to mount an adequate UPR triggers a set of intracellular molecular "switches" that comprise the ER stress response. The three key pathways are depicted in Fig. 3. Through these pathways, ER stress activates NF-κB, JNK and C/EBP, with downstream effects on inflammatory recruitment, phosphorylation of insulin receptor signalling intermediates (to worsen insulin resistance), lipogenesis, and oxidative stress. These processes can ultimately lead to dismantling of the cell by apoptosis, particularly involving C/EBP-homologous protein, which transcriptionally suppresses anti-apoptotic Bcl-2 and induces pro-apoptotic Bim (Fig. 3).

Relationships between hepatic ER stress, lipogenesis, insulin resistance and hepatic steatosis in obesity and metabolic syndrome have been the subject of intense scrutiny,155,156 and ER stress has been proposed as a mechanism in diverse experimental forms of liver injury (alcohol-related, drug-induced).154-157 In obese humans, UPR (typically GRP78 expression) and ER stress markers have been noted in the adipose, liver and pancreatic beta cells.158 To date, however, the evidence for operation of hepatic ER stress in human NAFLD/NASH is limited and inconsistent; some pathways seem to be activated, others are not,159 and there have not been informative correlations between pathways and disease phenotype. Likewise, the evidence for operation of ER stress in animal models is conflicting.155,160-163 In particular, there is little evidence that ER stress is a pro-inflammatory pathway in models that exhibit both the metabolic determinants of NAFLD and steatohepatitis pathology, such as HF-fed foz/foz mice (van Rooyen, unpublished data).

Impaired activity of sarcoplasmic-ER calcium ATPase-2b (SERCA), the ER calcium sequestering pathway, appears to be a key mediator of cellular responses to ER stress.164 Such inhibition could deplete ER calcium stores, causing cytoplasmic ionic calcium concentrations to rise, increasing its movement into mitochondria with implications for mitochondrial injury, but this has not yet been demonstrated. Enrichment of the ER membrane with cholesterol also inhibits SERCA activity in parallel with increased membrane order parameter.155 This has potential implications for NASH because ER is one site of increased cholesterol deposits (van Rooyen, unpublished data). Ultimately, the mechanistic relevance of ER stress as a disease pathway must come from in vivo studies of chemical chaperones that block its operation.165 One such chaperone is tauroursodeoxycholic acid, an agent that appears to have little if any therapeutic efficacy against NASH.166,167

4. Mitochondria, autophagy and the regulation of inflammation

Ultrastructural studies have consistently shown intra-mitochondrial crystals in NASH, the identity of which has not been resolved,168,169 and the association with decreased hepatic adenosine triphosphate (ATP) levels is also consistent with mitochondrial uncoupling or injury.170,171 Mitochondria are a major source of ROS. Physiologically, about 2% of oxidative phosphorylation is uncoupled, but during hibernation, obesity and in several experimental models of NAFLD expression of uncoupling proteins (UCP), particularly UCP2, increases.172,173 Damage to mitochondrial DNA and proteins, saturated FFAs and excessive ionic calcium could further uncouple oxidative phosphorylation, thereby generating oxidative stress. As mentioned earlier, FC impairs GSH uptake into mitochondria with similar deleterious effects.110,174 In addition, permeabilization of the inner mitochondrial membrane by opening of the mitochondrial permeability transition pore is a key pathway to initiation of cell death by apoptosis or necrosis.175

A critical cellular response to mitochondrial injury or starvation (energy depletion) is autophagy (termed mitophagy when confined to mitochondria).176-179 During mitophagy, damaged mitochondria are eliminated in a controlled process of lysosomal membrane and macromolecular turnover. This counters cellular degeneration and prevents unnecessary cell loss or, in the face of insurmountable damage, prepares residual cellular remnants (apoptotic bodies) for macrophage-mediated clearance in the more organised cell death pathway of apoptosis.176 By augmenting apoptosis, autophagy tends to dampen inflammation, whereas necrotic cell death can promote it.87,178-180 Mitochondria play a central role in inflammatory pathways, such as NF-κB and interferon-responsive factors (IRF), as depicted in Fig. 4, as well in the induction of inflammasomes (discussed below). There is also an interaction between impairment of autophagy and induction of ER stress.181 The recent interest in whether abrogation of autophagy contributes to inflammatory recruitment in NASH has been reviewed.179

5. The inflammasome

The inflammasome is a larger multimeric structure that regulates caspase 1 activation.182,183 The NLRP3 (nucleotide-binding domain, leucine-rich repeat containing) inflammasome (also known as cryopyrin or NALP-3) is expressed by myeloid cells and is up-regulated by pathogen-associated molecular patterns (PAMPs). It requires a caspase recruitment domain, and can recruit pro-caspase 1 in the presence of the adapter protein ASC (apoptosis-associated speck-like CRD-domain containing protein). Once all the components of the NALP3 inflammasome are assembled in the cytosol, caspase 1 is released and can promote the cleavage and therefore maturation of pro-inflammatory cytokines (pro-IL-1β, pro-IL-18, and IL33) to promote and sustain inflammation. NLRP3 inflammasome can be activated by several endogenous and exogenous agonists, as reviewed elsewhere.182,183 Salient to NASH, palmitic acid (but not oleic acid) induces activation of the NLRP3-ASC inflammasome to activate caspase 1 and cause IL-1β and IL-18 production.183 This pathway involves mitochondrial production of ROS (Fig. 4). Other agonists that could be relevant include uric acid crystals, which can precipitate in the extracellular space of dying cells, and extracellular DNA, possibly including mitochondrial DNA.

The inflammasome is activated in experimental alcohol-induced liver injury,83 and in mice fed the MCD diet, but not in HF diet-induced simple steatosis.184 Csak et al.184 exposed hepatocyte cultures to palmitic acid, and showed that this sensitised liver cells to release IL-1β following the further addition of lipopolysaccharide. In addition, palmitic acid provoked hepatocytes to release undefined "danger signals," which then activated the inflammasome in liver lymphocytes and macrophages to augment release of IL-1β and TNF-α. Other work has confirmed that, under certain circumstances, hepatocytes can themselves secrete chemokines and cytokines.185 Thus, activation of the inflammasome is one of several models by which hepatocytes could play a central role in inflammatory recruitment in NASH, but as indicated next, there are other potential pathways.

6. Ballooned hepatocytes and inflammatory recruitment; is the p53/senescence pathway involved?

Early studies identified ballooning as one of few histological features associated with risk of cirrhosis development in NAFLD.33 While not always confirmed by subsequent studies, in which presence of fibrosis and histology as "definite NASH" tend to over-ride ballooning in multivariate analyses,12,47 a link between ballooning and portal fibrosis has been emphasized by Richardson et al.186 These authors also found a strong link between ballooning and lobular inflammation in NASH, which is consistent with the proposal that ballooning attracts inflammatory cells, as indicated by their co-localisation in experimental studies (Fig. 2), and their implication in secretion of Hedgehog ligands.187-189 This family of fibrogenic transcription factors also plays a pro-inflammatory as well as pro-fibrotic role. Thus ballooned hepatocytes have been shown to be a focus for both HSC activation and hepatic precursor cell recruitment, both of which are under cytokine regulatory control.189

Ballooned hepatocytes often contain Mallory's hyaline (also known as Mallory-Denk bodies), which are derived from ubiquitin-modified intermediate (cytokeratin [CK]) filaments; ubiquitin staining can be used to identify ballooned cells more clearly.190-193 This destruction of intermediate filaments might indicate that cytoskeletal disruption leads to ballooning, but ultrastructural studies are limited. There is also evidence that foamy, lipid micro-droplets confer the glazed appearance of ballooned hepatocytes rather than hydropic change.190,192,193 Apoptosis is increased in livers with steatohepatitis,101,104,132,193 while circulating peptides liberated by caspase 3 cleavage of CK18, an hepatocyte-specific CK, serves as a biomarker for NASH versus "not NASH."194,195 The original term for apoptosis was "shrinkage necrosis"; therefore, the presence of ballooning seems more likely to reflect imminent cell necrosis rather than apoptosis. If so, the disintegration products could be pro-inflammatory, and it is well recognized that necrosis, an unregulated form of cell death, activates macrophages, neutrophils (e.g., by high mobility gel box 1 [HMGB1]) and other pro-inflammatory pathways,87,180,196 including the inflammasome discussed earlier.

An alternative possibility is that ballooned hepatocytes are a reflection of cellular senescence in the liver. In epithelial cells, stressors such as oxidative stress and DNA damage can lead to replicative senescence.196 In humans, this is particularly associated with shortened telomere length such that cell division is no longer possible.197 Most interest in senescence as a disease mechanism has been for neurodegenerative disorders and cancer;196,197 it does not appear to have been much studied in NASH. However, cirrhosis is associated with loss of telomere length,198 and p53, the guardian of senescence,196,199 is up-regulated in several types of fatty liver disease.200 Senescence arrests cell division by inducing cell cycle inhibitors (p21, p16^INK4A, Rb) and has a characteristic molecular expression profile closely linked to regulation of an inflammatory response in neighbouring tissues.197 The pro-inflammatory molecules involved include cytokines (IL-1, IL-6, IL-8), chemokines (IL-8, MCP-1, GRO α/β/γ) and chemokine receptors (CXCR2), most of which are consistently found to be up-regulated in experimental NASH (Table 1).72,86,132,137,201 Further research is required to establish whether hepatocyte senescence is inherent to inflammatory recruitment in the transition of steatosis to NASH.

1. Activation of NF-κB

NF-κB is a transcription factor comprised of five peptides that form homodimeric or heterodimeric complexes; p65 and p50 are highly expressed in liver. NF-κB p65:p50 heterodimers regulate the transcription of several hundred pro-inflammatory molecules (p50:p50 tends to be inhibitory), including cytokines, chemokines, adhesion molecules, nitric oxide and cyclooxygenase 2.141 In resting (G₀) hepatocytes, NF-κB is sequestered in the cytosol bound to inhibitory (IκB) proteins. Their phosphorylation, mediated by IKK, and subsequent ubiquitination targets the NF-κB -IκB complex to the 26S proteasome for degradation. This liberates NF-κB in a form that can be transported into the nucleus. Detection of p65 in nuclear extracts, or binding to cognate oligonucleotides in gel shift assays serve as indicators of NF-κB activation, together with increased levels of transcripts for "NF-κB-responsive genes." IKK is activated directly by oxidative stress and other cellular stressors (such as ER stress), or via liganding of NF-κB-signaling receptors.

NF-κB activation is uniformly found in human NASH202 and in all animal models in which it has been studied. Using MCD fed mice, we employed TNF-α and TNF-R1 knockout animals, and in vivo transfection of WT mice with non-degradable mutant-IκB to show that NF-κB activation is essential for hepatic inflammatory recruitment in steatohepatitis;137 further, such NF-κB activation occurs independently of TNF-α. Other work using the MCD dietary model has produced conflicting findings; curcumin, which blocks oxidative stress-mediated NF-κB activation provided protection,203 but TNF-α anti-serum reduced liver injury in rats administered the MCD diet,204 while Tomita et al.205 found that TNF-R knockout mice had protection against liver fibrosis in their MCD experiments.

Fractionation of livers from HF-fed foz/foz mice (Larter, unpublished data) and MCD-fed animals137 shows that NF-κB activation is most prominent in non-parenchymal cells (KCs, SECs, HSCs), but it is also evident in hepatocytes.137,206 The emerging concepts of metabolic stress mentioned earlier provide some indication that pro-inflammatory pathways in NASH could eminate from stressed hepatocytes via activation of NF-κB. Alternatively, TNF-α, IL-1β and other cytokines released from NF-κB-activated KCs could activate NF-κB in neighbouring hepatocytes.

Myeloid differentiation primary response gene 88 (Myd88) null mice are refractory to dietary steatohepatitis caused by a choline deficient and defined amino acid (CDAA) diet.207 Using bone marrow chimeric (WT/Myd88^-/-) mice, Miura and colleagues showed that the KC compartment was essential for inflammatory recruitment in this model.207 Further, the upstream stimulus to Myd88/NF-κB activation appeared to be Toll-like receptor 9 (TLR9),207 located in endosomes/lysosomes and most responsive to unmethylated CpG-containing DNA. The implication of TLRs and their role in the innate immune response and activation of NF-κB in NASH is discussed later.

2. JNK

Like NF-κB, the JNKs (1 and 2) can be activated directly by oxidative stress and by lipotoxic molecules (FFA, FC), or as the result of ligand binding to growth factor and TNF superfamily death-signalling receptors (Fas, TNF-R1, TNF-related apoptosis-inducing ligand death receptors) or TLRs.208 JNK activates the mitochondrial apoptosis pathway and forms the c-jun:c-fos heterodimer, AP-1; AP-1 is pro-inflammatory, typically inducing similar genes as NF-κB.

JNK appears always to be activated in lipotoxicity and in both experimental and human forms of NASH.159,209-213 In seminal work, Schattenberg et al.209 showed that activation of JNK1 (but not JNK2) was essential for inflammatory recruitment in MCD-induced steatohepatitis; others have confirmed this.210-212 Saturated fatty acids activate JNK in primary hepatocytes and tumour cells of hepatocyte lineage;95,101,106,214 and this was a critical pathway to cell death by the mitochondrial apoptosis pathway.101,214

In the foz/foz diabetes/metabolic syndrome model, we have noted that both JNK1 and JNK2 are activated with NASH, but not in genotype or dietary controls with simple steatosis.212 Further, dietary or pharmacological measures that lowered hepatic FC virtually abrogated JNK activation in association with mitigation of liver injury (ALT elevation), hepatocyte apoptosis and macrophage accumulation.123 These observations are consistent with the proposal that JNK activation is a key injury and inflammatory pathway in metabolic syndrome-related NASH.

1. Why could innate immunity be relevant to inflammation in NASH?

As mentioned, necrotic cell death elicits an inflammatory response. This concept was refined in 1994 when Matzinger215 proposed the "danger hypothesis" as a way in which the innate immune system can respond to key molecules released by damaged cells, thereby eliminating them. The mechanism by which stressed or dead cells trigger inflammation and adaptive immune responses involves damage-associated molecular patterns (DAMPS),180,216-218 also termed alarmins. Intracellular pro-inflammatory DAMPS include high-mobility group gel box 1 (HMGB1),218 heat shock proteins, fibrinogen and fibrinonectin, and mitochondrial products such as formyl peptides and mitochondrial DNA.217 Although they differ from PAMPs, some DAMPS can be recognised by similar receptors, particularly TLRs (e.g., TLR4 responds to both HMGB1 and lipopolysaccharide).217,219

2. TLRs and NASH

Eight TLRs are expressed in mammalian liver (TLRs 1, 2, 4, 6-10), with varying levels of expression on KCs, hepatocytes, SECs and HSCs.85 Most are expressed on the cell surface, but TLRs 1, 3 and 9 are intracellular (endosomal/lysosomal) proteins. TLRs recognise molecular patterns present on a broad range of pathogens and altered or specialised host molecules. Upon ligand binding and with recruitment of certain co-factors (e.g., myeloid differentiation factor 2 [MD2]), they signal via overlapping protein casettes to trigger inflammatory and antiviral responses, as well as maturation of dendritic cells to activate adaptive immunity.216 Individual TLRs interact with different combinations of adapter proteins (e.g., MD2) and activate transcription factors such as NF-κB, AP-1 (via JNK) and interferon-responsive factors (IRF). As shown in Fig. 5, MyD88 is shared by almost all TLRs and recruits members of the IL-1 receptor-associated kinase family. In fact, the intracellular domain of plasma membrane expressed TLRs exhibits IL-1 receptor motifs, and their intracellular signalling shares several intracellular adapter molecules with IL-1 (Fig. 5).

When released from necrotic cells, HMGB1 stimulates KCs and monocytes to produce pro-inflammatory mediators by acting as an endogenous ligand for TLR4, although it might do that by forming highly inflammatory complexes with other molecules (ssDNA, endotoxin, IL-1β, nucleosomes).218 TLR4 is involved in acute liver injury, such as hepatic ischemia-reperfusion injury,220-222 in alcoholic liver injury (when priming by gutderived endotoxin is pivotal).82 and is also up-regulated in MCD steatohepatitis219,223 and fructose-induced hepatic steatosis (not NASH).224,225 Saturated FFA can also bind to TLR4.70,90,226-228 TLR4 and MD2, its co-receptor for endotoxin, are expressed on KCs, hepatocytes and HSCs. Deletion of either TLR4 or MD2 dampens (but does not abolish) necroinflammatory activity of MCD steatohepatitis, with the most impressive effects being on NADPH oxidase expression and activation of inflammatory cells.229

Other research has identified activation of TLRs2 and 9 in various experimental models of NAFLD or NASH.225 As mentioned earlier, TLR9 is located within the cell and is most responsive to unmethylated CpG-containing DNA, but it also binds HMGB1. TLR9-deficient mice are protected from steatohepatitis in the CDAA model.207 TLR2 (but not TLR4) expression by hepatocytes can be induced by lipopolysaccharide, TNF-α and IL-1β via NF-κB activation, while signalling cross-talk between TLR4 and TLR9 amplifies the inflammatory response of macrophages,230 indicating other potential loops for perpetuation of inflammation in NASH. TLR5 is not expressed in the liver, but it has recently been reported that TLR5 knockouts have altered gut flora and develop obesity and metabolic syndrome, including insulin resistance and steatosis.231 Any relevance to NASH has yet to be established, although a fascinating finding was that transfer of the altered gut flora from TLR5^-/- mice to healthy animals resulted in a similar disease phenotype, including (non-inflamed) fatty liver.231

3. Kupffer cells

KCs are specialised tissue macrophages in the liver. They not only contribute to insulin resistance in fatty liver disease,61 but unite the inflammatory responses in many liver diseases.232 KCs are particularly sensitive to gut-derived endotoxin, acting through CD14, TLRs 2 and 4 and adapter proteins such as MD2 to activate NF-κB via MyD88.70,229 Other intracellular signalling molecules lead to induction of IFN-γ, which is important for lymphocyte recruitment.86,208 In chimeric mice with KCs derived from MyD88^-/- bone marrow donors,207 there was amelioration of the inflammation and fibrosis induced in the CDAA model of steatohepatitis compared with WT mice, demonstrating a key role for KC activation in this model. To the authors' minds, the data are more convincing than those in a recent study (using an irradiated, 2% cholesterol HF diet that is hepatotoxic) in which the authors proposed that TLR4 on hepatocytes, not KCs, responds to HMGB1 by NF-κB activation.206 Other earlier work showed that engulfment of cellular fragments denoted as apoptotic bodies from UV-treated murine hepatocytes (which would also causes oxidative stress) activated KCs to generate FasL and TNF-α.233 Ablation of KCs (e.g., by gadolinium or clodronate liposomes) reduces severity of liver injury and inflammation in alcohol-related liver injury in rodents, as does measures to change the gut flora in favour of non-endotoxin producing organisms. In a HF-fed mouse model, KC ablation ameliorated severity of steatosis by releasing hepatocytes from IL-1β and NF-κB-dependent suppression of peroxisome proliferator-activated receptor (PPAR)-α activity, thereby allowing PPAR-γ to exert its effects on fatty acid oxidation.234

4. Lymphocytes

Several types of lymphocytes are present in the normal liver, including NK cells, NK T cells, and T cells.86 Hepatic NK cells can be regulated by KC-derived cytokines (IL-1, IL-18), and in turn generate IFN-γ which participates directly in cell killing and in modulation of T cell responses. Lymphocytes accumulate in NASH livers, but which subpopulations predominate and their pathogenic roles in injury and inflammation have not yet been fully characterized.

5. Neutrophils

The presence of neutrophils (polymorphonuclear neutrophils, [PMNs]) among the liver inflammatory infiltrate of alcoholic steatohepatitis has long been recognized. Neutrophils are also present in NASH,29,31 where their possible pathogenic significance remains obscure. In the foz/foz metabolic syndrome model of NASH, the reduction of hepatic cholesterol stores which ameliorates liver injury, apoptosis and macrophage recruitment does not appear to alter accumulation of myeloperoxidase positive cells (PMNs) (van Rooyen, unpublished data). On the other hand, reduction of hepatic triglyceride stores and lipogenesis either by a dietary reversion strategy or with Wy-14,643 (a potent PPAR-α) agonist, has more impressive effects on neutrophils than on macrophages.212 Dietary reversion (from HF to chow) suppressed UCP2 expression and increased hepatic ATP levels, which would favour operation of apoptosis (and this was observed) rather than ROS-mediated necrosis (Larter and Farrell, unpublished data). Combined use of M30 and full-length CK8/18 in patients with NASH indicates that both apoptosis and necrosis occur in humans with the inflammatory form of NAFLD.195 It is therefore possible that neutrophil accumulation is associated with necrosis in NASH, and it may be regulated by different pathways than those important for macrophage and lymphocyte recruitment and activation.87 These important and rather neglected issues require further study.

The two hits concept of NASH pathogenesis served to dissect injury and pro-inflammatory pathways from the metabolic causes of steatosis. Insights gained since then indicate that the lipid molecules that accumulate, together with TG, in some NAFLD livers can themselves participate directly in pathogenesis of the necroinflammatory element of NASH. The fact that steatosis (which biochemically is TG accumulation) does not inevitably predispose to NASH is better understood by recent studies showing that TG formation is protective against injury and inflammation, not predisposing to such inflammation.235,236 On the other hand, FC, certain FFA, DAG and some phospholipids can directly injury liver cells and mediate subcellular stresses (mitochondrial, ER, oxidative) that incite hepatocellular injury, cell death and inflammatory recruitment in NASH. Thus, the essential difference between the two extremes of liver pathology (NASH versus not-NASH) is not attributable to the amount of fat (TG) in the liver, but rather the type of lipid molecules that accumulate.

Research in NASH pathogenesis has reached the exciting stage where investigation of potential lipotoxic molecules is being refined. Arguably the most critical future direction, however, is to perform more definitive lipidomic studies in human liver so as to clearly identify which lipid species are unambiguously implicated, and the genetic and environmental reasons for their accumulation. Such measurements must also establish correlations between candidate lipotoxic mediators, pro-inflammatory (and pro-fibrotic) pathways and liver pathology. In lieu of such human data, researchers (and journal editors) might better focus their attention on models where development of NAFLD across the pathological spectrum that includes NASH is clearly related at least to over-nutrition and insulin resistance, and ideally to obesity, type 2 diabetes and hypoadiponectinemia, the metabolic determinants of human NASH.2,3 Other models have taught us what can occur in steatohepatitis pathogenesis, but hepatologists are most interested in what does occur in NASH. Therefore, nutritional depletion models like the MCD dietary model developed in the author's laboratory in 1996, choline deficiency, the CDDAA and similar deprivations, 2% cholesterol (equivalent to 20 kg of cholesterol a day for humans!) plus cholate diets, or genetic knockout and knock-in models of disordered adipokine (leptin, adiponectin) or lipid and cholesterol handling should no longer, in our view, receive the high level of current attention simply because of the "cute" reductionist science used, when clinically more relevant (and equally convenient) alternatives have been characterized metabolically and pathologically.108,109,135

Establishing whether the pro-inflammatory pathways in NASH eminate from hepatocytes subjected mitochondrial injury, impaired autophagy, the inflammasome, or from processes like ER stress, oxidative stress/necrosis, senescence and p53 expression is pertinent to the design of novel, mechanism-based therapies for NASH. Current approaches, vitamin E, PPAR-γ ("glitazone") agonists, ursodeoxycholic acid, lipase inhibitors, generally effect some reduction in steatosis severity, but effects on inflammation are less consistent (ezetimibe may be an exception), and there are few reliable reports of fibrosis reduction. Despite interest in the gut-liver axis in obesity, type 2 diabetes and NAFLD, there are few data on clinical improvement with use of probiotics, other measures to alter the intestinal microbiota or use of incretin mimetics. Only bariatric surgery, or other effective means of weight loss coupled to increased physical activity that improve insulin sensitivity seem to combat both steatosis and inflammation in NASH. We therefore need to learn whether this is because of a primary effect on improving insulin sensitivity and reducing hyperinsulinemia, with possible secondary changes to turnover and storage of hepatotoxic lipid species,237 such as we recently clarified for disordered hepatic cholesterol homeostasis.115 If so, the findings could direct a radically different therapeutic approach, perhaps even finding a cure for NASH without what seems presently to be essential- a change in lifestyle and a decrease in body weight.