Non-alcoholic fatty liver disease (NAFLD) is not a single disease entity, rather it describes a spectrum of liver conditions that range from fatty liver (steatosis) to more severe steatosis coupled with marked inflammation and fibrosis [non-alcoholic steatohepatitis (NASH)] to severe liver disease such as cirrhosis and possibly hepatocellular carcinoma. Obesity, notably abdominal obesity, is a common risk factor for NAFLD. The pathogenesis from steatosis to NASH is poorly understood, and the ‘two hit’ model, as suggested nearly two decades ago, provides a feasible starting point for characterization of underlying mechanisms. This review will examine the oxidative stress factors (‘triggers’) which have been implicated as a ‘second hit’ in the development of primary NASH. It would be reasonable to assume that multiple, rather than single, pro-oxidative intracellular and extracellular triggers act in conjunction promoting oxidative stress that drives the development of NASH. It is likely that the common denominator of these pro-oxidative triggers is mitochondrial dysfunction. Understanding the contribution of each of these ‘triggers’ is an essential step in starting to understand and elucidate the mechanisms responsible for progression from steatosis to NASH, thus enabling the development of therapeutic targeting to prevent NASH development and progression.
One of the most common liver diseases in developed countries that affects humans who do not consume significant amounts of alcohol, that is less than 30 and 20 g of alcohol daily for men and women, respectively [1, 2] is non-alcoholic fatty liver disease (NAFLD). NAFLD represents a spectrum of conditions ranging from steatosis (fatty liver) to marked inflammation and fibrosis (known as non-alcoholic steatohepatitis (NASH)) to cirrhosis and hepatocellular carcinoma (HCC) [1-4] (Fig. 1). Interestingly, NASH may progress to HCC without first progressing to cirrhosis [5, 6]. It is estimated that 17–33% of the general population in the UK have hepatic steatosis while the prevalence is reported to be higher (~70%) in individuals with type 2 diabetes . Although excessive fat mass, notably in the upper body (abdominal) depot, is a well-documented risk factor for NAFLD , it has also been reported in non-obese (body mass index <30 kg/m2) individuals . The prevalence of NAFLD is suggested to differ with gender, race and ethnicity . Genetic and epigenetic factors also are involved in the pathogenesis of the disease [8-12] and although of interest, are outside the scope of this review.
The biological mechanisms causing progression through the spectrum of NAFLD stages are not well defined. The initial hypothesis proposed by Day and James  was the ‘two hit’ model with the ‘first hit’ being steatosis, and the ‘second hit’ needed to initiate NASH requiring ‘other’ factor(s) that promote lipid peroxidation; suggesting that this ‘two hit’ model may also apply to ethanol-associated liver disease. More recently, additional factors have also been suggested to promote progression of NASH. Jou et al.  suggested that steatosis progresses to NASH when adaptive mechanisms that protect hepatocytes from fatty acid-mediated lipotoxicity are overwhelmed, increasing the amount of hepatocyte death and ultimately failure of the normal regenerative capacity of the liver. Cell death initiates an inflammatory cascade and disruption of hepatic architecture along with liver progenitor cell activation, which may increase the risk of HCC . A ‘multiple parallel hits’ model has been suggested to promote progression of steatosis to NASH because of failure of the antilipotoxic protection systems of the liver and multiple hits from the gut and/or adipose tissue .
A major factor proposed to be important in the development of primary NASH [as opposed to secondary NASH induced by drugs e.g. amiodarone ] is oxidative stress. Oxidative stress refers to an imbalanced cellular state in which the production of reactive oxygen/nitrogen species (ROS/RNS) are increased to an extent that overrides the normal operating free radical clearing mechanisms; such as glutathione peroxidase (GPx), superoxide dismutase (SOD) and catalase. Oxidative stress can occur by an increased production of pro-oxidant species, which saturate the anti-oxidant machinery, and/or by a direct insult to the anti-oxidant machinery. For example, if the availability of substrates such as glutathione (GSH) is insufficient for GPx activity or variations in genotypes occur (e.g. SOD2 polymorphisms), oxidative stress manifests as a result of improper clearance of pro-oxidant species. Pro-oxidant species have been reported to impair nucleic acid, protein and cell membrane functions . Importantly, these species can initiate lipid peroxidation by targeting polyunsaturated fatty acids (FAs), resulting in the formation of highly reactive aldehyde products, such as 4-hydroxy-2-nonenal (4-HNE) and malondialdehyde (MDA). These reactive lipid derivatives have the potential to amplify intracellular damage by mediating the diffusion of ROS/RNS into the extracellular space thus causing tissue damage.
Several intracellular [e.g. mitochondrial dysfunction and endoplasmic reticulum (ER) stress] and extracellular (e.g. iron accumulation and inflammation by gut flora) factors (‘triggers’) have been implicated in hepatic oxidative stress, and elucidating the contribution of these factors is fundamental to understanding the development of NASH. Recently, Rolo et al.  reviewed the role of oxidative stress in the pathogenesis of NASH; here we will discuss, using the ‘two hit’ model as a framework, the most likely factors that have been suggested to initiate oxidative stress in (primary) NASH. These factors include: mitochondrial dysfunction, iron accumulation, ER stress and inflammation mediated by gut flora.
There is a myriad of evidence for mitochondrial abnormalities in NASH patients  and in animal models of NASH . Histological analyses have identified ultrastructural changes notably; mega-mitochondria, loss of cristae and paracrystalline inclusion bodies present within the mitochondrial matrix [21, 22]. These mitochondrial abnormalities including mutations in mitochondrial DNA (mtDNA) and nucleic genes encoding mitochondrial proteins have been shown to elevate levels of oxidative stress . There are also specific syndromes such as Alpers' disease, where mutations in polymerase gamma exist, and patients develop liver steatosis with late stage liver failure .
It has been hypothesized, that to counter the steatotic state, fatty acid β-oxidation is increased; consistent with this it has been found that individuals with higher amounts of liver fat (>5%) had greater hepatic fatty acid oxidation than subjects with low amounts (<5%) of liver fat [25, 26]. The increase in fatty acid oxidation in the steatotic state could potentially induce an increased electron flux through the electron transport chain (ETC), which may lead to a notable decrease in oxidative capacity. Mutations in complex II  and decreased activity of all ETC complexes  have been reported to occur in NASH. Thus, it could be hypothesized that a situation of ‘electron leakage’ because of the reduced activity of ETC complexes [high (NADH/NAD+) and (FADH2/FAD) ratios] may result in electrons directly reacting with oxygen and forming ROS instead of electrons shuttling through cytochrome c oxidase and combining with oxygen and protons to form water. When recapitulated experimentally; mitochondria depleted of cytochrome c oxidase and exposed to high [NADH] produce three-fold more hydrogen peroxide (H2O2) than non-depleted mitochondria . Abnormal mitochondrial ROS production has been shown in rodent models of NASH with impaired complex I-linked respiration . Other indirect sources of ROS/RNS, such as α-ketoglutarate dehydrogenase, pyruvate dehydrogenase, the growth factor adapter p66-Shc and monoamine oxidase, which provide/shuttle electrons into the ETC, may also contribute to oxidative stress (Fig. 2).
The mitochondrial cytochrome P450 2E1 (CYP2E1) enzyme is directly capable of generating ROS, although its relative contribution to oxidative stress has not been conclusively determined. Nevertheless, increased activity of CYP2E1 has been observed in NASH patients . Moreover, animal studies using the methionine choline deficient (MCD) diet to induce NASH reported increased CYP2E1 activity in a rat model . A recent paper employing the use of CYP2E1 knock-out mice reported that CYP2E1 is necessary for NASH development . By using the NAFLD histological scoring system (NAS) and blind evaluation, it was found that only wild-type (WT) mice on a high fat diet developed NASH; with NASH defined as a NAS score ≥5 . The authors noted that intrahepatic levels of tumour necrosis factor alpha (TNFα) although higher in high fat fed CYP2E1 knock-out and WT mice when compared with animals on a chow (low fat) diet; CYP2E1 knock-out had lower levels compared with WT mice . However, interpretation of the role of mitochondrial specific CYP2E1 must be approached with caution, as an ER CYP2E1 isoform also exists. Additionally, CYP2E1 polymorphisms occur and in particular the presence of the c2 allele has been shown to increase the likelihood of developing NASH by 75% in obese, non-diabetic women compared with controls; although a small study (48 participants) it does go some way to explaining why not all people go on to develop NASH .
Decrease in anti-oxidant mechanisms
Detoxifying enzymes within the mitochondria, including GPx and manganese SOD (MnSOD), can be defective in NASH. Additionally, the level of catalase that is present in hepatic mitochondria remains unclear  but it is thought to be low, thus the ability of the mitochondria to reduce ROS/RNS is limited. A plausible explanation of reduced GPx activity is that it requires reduced GSH within the mitochondrial matrix for optimum catalytic activity and there is evidence of GSH depletion in NASH; which is thought to be because of impaired transport of cytosolic GSH into the mitochondria . The C47T polymorphism of the SOD2 gene (which encodes MnSOD), substantially increases H2O2 production and has been associated with determining susceptibility to NASH [36-38] and developing advanced fibrosis  i.e. greater than or equal to stage one on the Brunt criteria  (Fig. 2).
It must be acknowledged that the perturbation of mitochondrial redox balance can further exacerbate initial mitochondrial dysfunction through various mechanisms including: decreased ETC oxidative capacity where ROS/RNS interact with polyunsaturated FAs to produce highly reactive aldehydes such as MDA and 4-HNE. MDA inhibits cytochrome c oxidase while 4-HNE induces uncoupling protein 2 resulting in ‘electron leakage’. MDA and 4-HNE can also damage mitochondrial membranes which are integral to the function of the ETC. Additionally, ROS/RNS can directly lead to mtDNA mutations .
Dysregulation of iron homeostasis in NASH, as demonstrated by increased serum ferritin and transferrin saturation, was first shown in 1994 . Subsequently, investigations into the prevalence of the hemochromatosis (HFE) gene have found protein variants including 16189 , H63D and C282Y, with the latter two being associated with increased hepatic iron accumulation and are suggested to be proportional to severity of NASH . However, role of HFE variants in NASH is not clear. Most recently, a large study (786 patients) found that C282Y but not H63D variants were associated with higher hepatocellular iron content but reported that those subjects with the mutation had less liver ballooning or NASH . In support of this, others have reported only a trend towards higher levels of serum ferritin in C282Y heterozygotes; this was not associated with any differences in steatosis, inflammation or fibrosis in those with or without C282Y . Conversely, in a study of 51 NASH patients, 31% were either homozygous or heterozygous for the C282Y mutation . The presence of the mutation was associated with a higher content of hepatic iron and a significant association with liver damage . In different ethnic groups, the role of HFE variants in NASH severity is debated [48, 49]. Although iron plays a role in catalysing the production of ROS via the Fenton reaction  (Fig. 3), the precise mechanism by which iron accumulation leads to progression of NASH remains to be elucidated.
The saturation of plasma transferrin presents an opportunity for non-transferrin bound iron to be internalized by cells, including hepatocytes. High concentrations of unbound iron are able to saturate intracellular chelators, thus generating pro-oxidant products. According to the Haber–Weiss and Fenton reactions, ferrous iron (Fe2+) can catalyse the production of the more potent hydroxyl radical from H2O2, predominately generated from the peroxisomal β-oxidation of very long chain and branched chain FAs. Work in liver slices from rats demonstrated that cytosolic iron can potentially increase catalytic activity of 15-lipo-oxygenase resulting in disruption of the peroxisomal membrane, initiating further H2O2 production before peroxisomal anti-oxidant mechanisms can initiate detoxification . This process occurs physiologically for turnover of peroxisomes; however, it may be stimulated to a pathological level in NASH because of iron overload. Other pro-oxidant mechanisms caused by iron accumulation could exist, for example, in a rodent model chronic iron overload induced by dietary supplementation of carbonyl iron has been shown to upregulate inducible nitric oxide synthase (iNOS) expression through the extracellular signal-related kinase (ERK) signalling cascade .
Decrease in anti-oxidant mechanisms
Iron overload, via dietary supplementation also increases oxidative stress by decreasing GSH, thereby limiting GPx activity and reducing the cells anti-oxidant capacity . Alternatively, defective iron metabolism has been shown to affect anti-oxidant machinery of the cell. For example, under normal conditions, the transcription repressor BTB and CNC homology 1 (Bach1) inhibits expression of haem oxygenase-1 (HO-1), an enzyme responsible for the breakup of haem-bound iron  is reported to have anti-oxidant/anti-inflammatory properties . In times of pro-oxidant stress when cellular haem is high, Bach1 is removed from the HO-1 promoter leading to its increased expression. Therefore, iron deficiency could reduce the anti-oxidant capacity of the cells through Bach1-mediated HO-1 inhibition. Consistent with this Bach1 knock-out MCD mice have been shown to be resistant to development of NASH because of concomitant induction of HO-1 activity . It has also been shown that iron can directly act as a competitive inhibitor of enzymes such as enolase . Therefore, it is plausible that other enzymes that contain metal binding sites, such as members of the SOD family, could be inhibited in a similar manner by iron thus reducing anti-oxidant capacity.
Disulphide bonds within nascent polypeptides are formed within the ER, utilizing the sequential activities of protein disulphide isomerase and oxidoreductin-1 (ERO1)  which donates electrons to molecular oxygen giving rise to ROS. In a lipid-rich environment, the ER initiates a stress response involving a myriad of signalling cascades, collectively termed the ‘unfolded protein response’ (UPR) to ensure that misfolded proteins do not accumulate in the cell. It has been hypothesized that in NASH, ER UPR can generate substantial oxidative stress , but UPR independent mechanisms also exist simultaneously (Fig. 4).
Unfolded protein response upregulates the activity of the pro-apoptotic protein CCAAT/enhancer-binding protein homologous protein (Chop), which has been implicated in enhancing ROS generation . Concomitantly, ATP-requiring chaperone proteins attempt to correct protein misfolding, putting further strain on the potentially dysfunctional mitochondrial ETC. There is evidence that the UPR induces calcium leakage from the ER, which can act on both the inner and outer mitochondrial membranes opening the permeability transition pore for cytochrome c, effectively blocking ETC, stimulating ROS/RNS production and initiating apoptosis . While these are mainly ER-mediated pro-oxidant mechanisms acting via the mitochondria, increased activity of the ER isoform of CYP2E1 could also contribute to oxidative stress through an ER dependent and mitochondrial independent mechanism.
Decrease in anti-oxidant mechanisms
It has been observed in yeast that during UPR, levels of GSH decrease although the exact effect of UPR on GSH and its subsequent action on protein folding is debated . Less controversial is the UPR mediated inhibition of NFE2-related factor 2 (Nrf2), which activates transcription of other anti-oxidant enzymes by binding to the anti-oxidant response element. Recently, it was shown that deletion of Nrf2 rapidly triggers progression to NASH in a MCD diet mouse model .
Inflammation meditated by gut flora
An appreciation of the role of gut flora in mediating the progression from steatosis to NASH is beginning to emerge with evidence from animal and human studies . However, most of the mechanistic understanding is based on an increase in pro-oxidant species and the potential for decreasing functionality of anti-oxidant machinery remains to be explored.
An increase in Gram-negative bacteria within the gut can cause exposure of Kupffer cells (liver resident macrophages) to higher amounts of lipopolysaccharide (LPS). Alternatively, increased sensitivity of Kupffer cells to LPS by upregulation of Toll-like receptor 4 (TLR4)  or cluster of differentiation antigen14 (CD14)  receptors may be a more common mechanism operating in NASH. In support of this, it has been shown that obese mice with a high degree of steatosis were more responsive to a low dose of LPS and rapidly developed NASH compared with lean mice . A postulated mechanism for this is the endocytosis of LPS by Kupffer cells resulting in the production of pro-inflammatory cytokines such as TNFα . It has been reported that the expression of TNFα receptor is increased in patients with NASH  and this may be involved in increased ROS production .
Decrease in anti-oxidant mechanisms
To our knowledge, studies have not yet been undertaken to elucidate a mechanistic role for gut flora in decreasing anti-oxidant capacity and leading to the pathogenesis of NASH in humans or animals.
Therapeutic potentials for the ‘second hit’
There is a universal consensus among guidelines that the first-line option for NAFLD treatment (& prevention) is lifestyle change, including weight reduction, dietary modification and physical exercise [2, 69-72]. Indeed, modest weight loss (<10% body weight) reversed hepatic steatosis in individuals with type 2 diabetes . However, evidence for therapeutic strategies to decrease the progression to NASH is not clear cut . In a recent review, Lomonaco and colleagues  highlighted potential pharmacological agents that may be utilized for the treatment of NAFLD but commented that at this point in time, there are no approved pharmacological treatments for NASH. As the ‘second hit’ for NASH appears to be promoted by oxidative stress, it would seem logical that anti-oxidant therapeutics could potentially be efficacious in NAFLD/NASH patients. Consistent with this, studies have investigated the use of the anti-oxidant vitamin E (α-tocopherol), for the treatment of NASH and results suggest that vitamin E therapy (given at a daily dose of 800 IU) may be effective for the treatment [75-77]. Although guidelines recommend vitamin E as a first-line pharmacotherapy, it is not appropriate for all NASH patients . The anti-oxidant, curcumin has been shown in animal studies, to have potential to improve NASH  but has yet to be tested in humans.
Hepatic metabolism of methionine produces cysteine, which is a precursor of GSH, the major intracellular anti-oxidant in the body. Methionine is the immediate precursor of S-adenosylmethionine (SAM), a common cosubstrate involved in methyl group transfers . Human kinetic studies have suggested the significantly lower rates of transmethylation of methionine groups observed in individuals with NASH is because of a decreased synthesis of SAM , thus increasing SAM may be an effective strategy for treatment of NASH. Betaine, which is derived from the oxidation of dietary sources of choline, has been shown to raise SAM in animal models , although work in humans has yet to clearly demonstrate betaine to be effective in the treatment of NASH [81, 82]. In contrast, SAM supplementation in a rodent MCD diet model showed protection against NASH .
One other potential strategy is to target the factors that potentiate mitochondrial dysfunction; for example by promoting mitochondrial oxidative capacity . The use of MnSOD mimics have shown improved mitochondrial oxidative capacity in mice models [85, 86]. Oxidative stress caused by iron accumulation could be countered by phlebotomy and chelation therapy with deferoxamine or apotransferrin . Results from a phase II clinical trial suggest that iron reduction, as a result of phlebotomy therapy may improve liver histology, with the authors suggesting that future studies should focus on patients with a serum ferritin of at least 300 μg/L .
Targeting UPR mediators and CYP2E1 may also prove beneficial although inhibitors of CYP2E1 such as chlomethiazole can be hepatotoxic therefore limiting their clinical use. Gut flora profiles vary between individuals, but a reduction in downstream effectors such as TNFα can be achieved by pentoxifylline . Randomized controlled trials in subjects with biopsy-proven NASH have suggested long-term (12 months) administration of pentoxifylline results in improved histological features of NASH [90, 91]. Emerging data suggest that probiotic supplementation in individuals with NAFLD [92, 93] and NASH  improves plasma aminotransferase levels. The effect of a prebiotic on NAFLD/NASH requires exploration. Given that the discussed triggers are not exclusive, drug interactions must also be considered while making a pharmacological attempt to hinder progression to NASH.
This review has described some of the potential cellular mechanisms (Table 1) that contribute to an increase in oxidative stress in patients with hepatic steatosis, promoting an environment that drives the development from steatosis towards NASH. It must be appreciated, however, that NAFLD is a disease spectrum, inextricably linked to epistatic [8-12] and environment interfaces that remain a challenge to untangle and study independently. Moreover, it is generally difficult to study disease progression in humans in vivo and conclude whether the observed factor is a bona fide cause, or an effect of NASH.
Table 1. Overview of the postulated pro-oxidant mechanisms and the factors decreasing anti-oxidant mechanisms contributing to oxidative stress in non-alcoholic steatohepatitis (NASH)
Potential ‘triggers’ of oxidative stress
ER, endoplasmic reticulum; ETC, electron transport chain; GSH, glutathione; GPx, glutathione peroxidase; CYP2E1, cytochrome P 450 2E1; SOD2, superoxide dismutase 2; MnSOD, manganese superoxide dismutase; iNOS, inducible nitric oxide synthase; HO-1, haem oxygenase-1; Bach1, BTB and CNC homology 1; UPR, unfolded protein response; ERO1, endoplasmic reticulum oxidoreductin-1; Chop, CCAAT/enhancer-binding protein homologous protein; Nrf2, NFE2-related factor 2; TNFα, tumour necrosis factor alpha.
Decreased oxidative capacity of ETC leads to ‘electron leakage’
Antigens derived from gut flora activate the NADPH oxidase system
Discrepancies in findings among animal models (for example, owing to differences in gut microflora across different labs) and clinical studies (caused by different patient inclusion criteria and ethnicity-related predispositions) further complicate the picture. Even with the use of tightly controlled animal models of NASH, the extrapolation of the results to humans is restricted because of fundamental physiological differences – rodent livers; for instance, produce apoB48 and apoB100 while their human counterparts only produce apoB100. Additionally, the NASH diet models, while experimentally necessary and useful, are not completely feasible because it is improbable that human subjects consume such diets, and thus the recapitulation of a NASH phenotype using these dietary models can be questioned. While the Kleiner score  has made an attempt to standardize NAFLD diagnosis based on histology, there is still considerable subjectivity among histopathologists when using these diagnostic criteria . Liver biopsies represent a temporal and spatial ‘snapshot’ of the liver and should not be used to generalize the entire metabolic state of the liver. Utilization of histology may also underestimate the role of possible triggers of the ‘second hit’ as emphasized by the findings in NASH patients with some , but not all reporting iron accumulation . Therefore, it is likely that the ‘second hit’ in NASH occurs as a result of multiple intertwined pro-oxidative triggers that act in conjunction (Fig. 5) with the common denominator of oxidative stress most likely being mitochondrial dysfunction as evidenced by the key metabolic role of the ETC. The aforementioned potential ‘triggers’ interact with mitochondria in a variety of ways, for example, the mitochondria act as a source of superoxide anion which is a substrate for the Haber–Weiss reaction and it has been suggested that the ER and mitochondria can form physical junctions ; furthermore, gut derived inflammatory cytokines may exacerbate mitochondrial-mediated oxidative stress.
A greater understanding of the multiple triggers orchestrating the oxidative stress in NASH, and by extension NAFLD progression, requires a panoramic view; that is, studies, asking pivotal questions about NASH initiation and progression to be undertaken in cellular, animal and human models. For in-depth, longitudinal studies to be undertaken in humans, robust biomarkers and non-invasive diagnostic techniques for tracking the disease progression are required. Understanding the roles of the ‘triggers’ highlighted in this review, along with other contributors will aid in the development of novel diagnostic and therapeutic approaches.
The authors thank Dr Karl Morten and Dr Umbreen Ahmed for their expert and very helpful comments.
Financial support: Leanne Hodson is a British Heart Foundation Intermediate Fellow in Basic Science.
Conflicts of interest: The authors do not have any disclosures to report.