The biochemical standard for normal triglyceride content in the human liver is 5.5% of triglyceride of wet liver tissue weight [15,16]. Histologically, the liver is considered steatotic when ≥5% of hepatocytes in a tissue section stained with hematoxylin and eosin contain macrovesicular steatosis [17,18,19,20]. Steatosis is graded by the pathologist from 0 to 3 based on its severity: grade 0 (normal) = <5%, grade 1 (mild) = 5%–33%, grade 2 (moderate) = 34%–66%, and grade 3 (severe) = ≥67% of hepatocytes characterized by macroscopic steatosis [17]. As discussed below, these percentages seem quite different from those obtained by proton magnetic resonance spectroscopy (¹H-MRS) (Table 1).

Steatosis can most accurately be measured using ¹H-MRS [28]. This technique enables sampling of a large volume fraction of the liver compared to a biopsy [29,30] and provides an accurate and reproducible measurement of liver fat content [30]. However, ¹H-MRS is expensive, as it requires use of magnetic resonance imaging (MRI) scanner and special expertise to perform proton magnetic resonance spectroscopy (¹H-MRS) at the time of MRI scanning. ¹H-MRS has been used in one population-based study, the Dallas Heart Study (DHS), to define normal liver fat content [23]. In this study, ¹H-MRS was performed on 2349 subjects, of which 345 were considered healthy based on the following criteria: no history of liver disease or risk factors for hepatic steatosis (alcohol consumption ≤30 g/day in men, ≤20 g/day in women, body mass index (BMI) <25 kg/m², normal fasting serum glucose, non-diabetic and normal serum alanine aminotransferase (ALT) (≤30 IU/L in men, ≤19 IU/L in women)). The upper limit of normal liver fat content was defined based on the upper 95th percentile in the healthy subjects and was 5.56% [23].

The ¹H-MRS studies determine the hepatic triglyceride content rather than the percentage of hepatocytes with macroscopic lipid droplets. The relationship between ¹H-MRS and histological liver fat content has been analyzed in two small studies, which included 13 [31], 12 [32] and 50 [33] subjects. In the first two studies, the ¹H-MRS-determined normal liver fat in the DHS, i.e., the 5.56% value corresponded to 15.7% [31] and 13.9% [32] of hepatocytes with macroscopic steatosis. On the third study, histological grade 1 (5%–33% macroscopic liver fat) corresponded to 11% (7%–14%), grade 2 (33%–66%) to 18% (14%–23%) and grade 3 (>66%) to 25% (10%–28%) ¹H-MRS liver fat [33]. ¹H-MRS-measured liver fat corresponds well to triglyceride content measured in a liver biopsy (r = 0.90, p < 0.001) [34]. These data show that the technique used to define normal liver fat influences the normal value.

Hepatic steatosis can be diagnosed with MRI using an out-of-phase and in-phase imaging technique developed by Dixon WT et al. [35]. This method involves acquisition of MR images at echo times in which fat proton and water proton signals are either out-of-phase (water and fat signals cancel) or in-phase (water and fat signals add up) [35,36,37]. Once the out-of-phase and in-phase images are acquired by using constant calibration and other scanner settings, a quantitative fat signal fraction can be calculated from the hepatic signal [38]. Modified versions of the early Dixon method have been introduced. These include the hepatic fat fraction by Fishbein MH et al. which uses fast gradient echo techniques [25,39] and correlates well with histological liver fat content (r = 0.77, p < 0.001). The newer MRI-determined proton density fat fraction (PDFF) technique provides a quantitative, standardized and objective MRI measurement of hepatic fat based upon inherent tissue properties [40,41]. The MRI-PDFF method is reproducible and correlates closely with ¹H-MRS (r = 0.99) [33,42] and liver histology (8.9%–9.4% at grade 1, 15.8%–16.3% at grade 2, and 22.1%–25.0% at grade 3, p < 0.0001) [33,43,44]. With this technique, the 5% macroscopic liver fat determined by histology corresponds to a PDFF value of 6% to 6.4% [45,46].

Ultrasound (US) is an inexpensive and widely available tool to visualize the liver and its fat content. Hepatic steatosis appears as a diffuse increase in parenchymal brightness and echogenicity on US images, and is often compared to hypoechogenity of the kidney cortex. Most studies score steatosis semiquantitatively as “mild”, “moderate” and “severe” based upon the visual assessment of hepatic echogenicity [27,47,48,49]. Lack of standardization precludes accurate comparison of data acquired by different machines and investigators. US lacks sensitivity in obese subjects [50] and in subjects with low liver fat content [51]. The sensitivity of diagnosing fatty liver increases from 55% to 80% when liver fat increases from 10%–20% to over 30% [51]. A recent study [52] suggested that the optimum sensitivity for US was achieved at a ¹H-MRS-measured liver fat content greater than 12.5%. A meta-analysis of 44 studies comprising 4720 subjects concluded that US has a sensitivity of 85% and a specificity of 94% for detecting 20%–30% macroscopic steatosis [53]. The sensitivity and specificity were 65% and 81% for detecting 0%–5% steatosis and 93% and 88%, respectively, for detecting >10% steatosis.

Xia MF et al. created an equation for accurate quantification of liver fat content using US in Chinese subjects [54]. A tissue-mimicking phantom was used as a standard and the US hepatic/renal ratio was measured to calculate liver fat content in 127 subjects, in whom liver fat was also measured using ¹H-MRS. The adjusted R² for the model was 80%. The optimal cut-off for the US-measured liver fat content to diagnose hepatic steatosis was 9.15%, which yielded a sensitivity and specificity of 95% and 100%, respectively. The utility of this technique in other ethnic groups which are more obese than the Chinese in the face of a similar amount of liver fat [55,56] remains to be tested.

Hepatic steatosis can also be assessed by using computed tomography (CT) by comparing attenuation of the liver parenchyma to that of the spleen [57]. Tissue fat deposition lowers attenuation, hence fatty areas are less dense and appear darker than the non-fatty tissues [22]. The attenuation value in the healthy liver is 50 to 57 Houndsfield Units (HU) and 8 to 10 HU higher than that of spleen [22]. It decreases by 1.6 HU for every 1 mg of triglycerides per gram of liver tissue [58]. In subjects with steatosis, the mean attenuation value of the liver is lower than that of the spleen, and the liver appears darker than the spleen. Attenuation values less than 40 HU in the liver or 10 HU less in the liver than in the spleen are indicative of marked hepatic steatosis (>30%). Smaller fractions of fatty infiltration cannot be accurately and reliably assessed [59,60].

In subjects with NAFLD and the metabolic syndrome (MetS), i.e., in “Obese/Metabolic NAFLD”, liver fat is closely correlated with direct measures of insulin resistance such as the inability of insulin to suppress hepatic glucose production [61], and indirect measures such as fasting serum insulin and the product of fasting insulin and glucose (Homeostasis model assessment for insulin resistance [HOMA-IR]) [62]. Indeed, liver fat correlates better with fasting insulin than with liver enzymes such as serum ALT and aspartate aminotransferase (AST) [63,64]. This close association between fasting insulin and liver fat is physiologically feasible as the main action of insulin after an overnight fast is to restrain hepatic glucose production. The inability of insulin to suppress hepatic glucose production increases fasting glucose, which stimulates insulin secretion leading to hyperglycemia and hyperinsulinemia.

Lipolysis is the main source of fatty acids used for synthesis of intrahepatocellular triglycerides [65,66]. Liver fat is closely correlated with the ability of insulin to suppress lipolysis [67,68]. The ability of insulin to suppress very low density lipoprotein (VLDL) production is also impaired in NAFLD, which contributes to hypertriglyceridemia and a low high density lipoprotein (HDL) cholesterol concentration. Damaged hepatocytes release increased amounts of C-reactive protein (CRP) and coagulation factors, which could contribute to increased risk of cardiovascular disease and atherothrombotic vascular disease (Figure 1).

Any obese person with NAFLD and features of the MetS can be considered to have “Obese/Metabolic NAFLD” irrespective of genetic risk factors. The most recent proposal defines the MetS in 10 different ways [69]. The presence of any three out of five features (hypertriglyceridemia, low HDL cholesterol, hyperglycemia, hypertension, increased waist circumference) is required for diagnosis of the MetS [69]. For clinical practice, this definition still remains the best tool to diagnose insulin resistance, although the extent to which the 10 different definitions increase the risk of endpoints such as type 2 diabetes and cardiovascular disease is unclear. Measurement of fasting insulin and glucose concentrations and their calculation of their product HOMA-IR might seem more attractive direct tools to measure insulin sensitivity in subjects with NAFLD. The problem with this approach is that insulin assays are not internationally standardized and give highly variable results [70].

Approximately 30% of Europids and several other ethnic groups carry the PNPLA3 I148M variant [12]. The association between the PNPLA3 gene variant and NAFLD [12] has been replicated in over 50 studies, including eight genome wide association studies [71,72,73]. In a meta-analysis carriers of the I148M variant had 73% more liver fat, a 3.2-fold higher risk of necro-inflammation and a 3.2-fold greater risk of developing fibrosis than the non-carriers [71]. In a meta-analysis comprising 12 Asian studies, the risk of NAFLD was 1.9-fold increased in carriers compared to non-carriers [72]. Recent meta-analyses have also shown that this gene variant increases the risk of cirrhosis by 1.9-fold [74] and hepatocellular carcinoma (HCC) by 1.8-fold [75].

In vitro, the PNPLA3 I148M gene variant abolishes intrahepatocellular lipolysis [76,77] and by acting as a lysophosphatidic acid acyl transferase stimulates triglyceride synthesis from long unsaturated fatty acids containing coenzyme A (CoA) more than from saturated fatty acid CoAs [78]. The contribution of each these mechanisms to function of the PNPLA3 gene variant in the human liver is uncertain. It is clear, however, that the human liver lipidome markedly differs between “Obese/Metabolic NAFLD” and “PNPLA3 NAFLD” [14]. The increase in liver fat in the carriers of the PNPLA3 I148M gene variant is due to polyunsaturated triglycerides, whereas in “Obese/Metabolic NAFLD” the concentration of saturated triglycerides and insulin resistance-inducing ceramides is increased [14].

Table 2 summarizes the 14 studies that include data on insulin sensitivity in carriers and non-carriers of the I148M variant [12,79,80,81,82,83,84,85,86,87,88,89,90,91]. Carriers of the PNPLA3 I148M variant had more liver fat in their liver than non-carriers. Insulin sensitivity as evaluated by HOMA-IR [62], the hyperinsulinemic clamp technique, fasting or post-glucose insulin and glucose concentrations did not, however, differ between carriers and non-carriers of the gene variant. These studies included obese and non-obese, diabetic and non-diabetic as well as pediatric cohorts. Serum triglycerides were either similar or lower in variant allele carriers as compared to non-carriers, consistent with lack of insulin resistance (Table 2).

Approximately 7% of all subjects carry the TM6SF2 E167K variant. This gene variant increases the risk of NAFLD, independent of genetic variation in PNPLA3 at rs738409, obesity and alcohol intake [92]. A recent meta-analysis reported that carriers of the TM6SF2 E167K gene variant have a 2.1-fold higher risk of NAFLD than non-carriers [93]. They also had lower circulating total and low density lipoprotein (LDL) cholesterol, and triglyceride concentrations than non-carriers [93].

Four in vitro studies have examined the mechanism by which the TM6SF2 E167K gene variant could increase liver fat. Recombinant adeno-associated viral vectors expressing short hairpin RNAs were used to reduce Tm6sf2 transcripts in the mouse liver, which increased hepatic triglyceride content three-fold [92]. TM6SF2 knock-out mice developed hepatic steatosis and had a three-fold reduced plasma VLDL triglyceride levels due to decreased lipidation [94]. In another study, TM6SF2 small interfering RNA inhibition also decreased export of triglyceride-rich lipoproteins and lipid droplet content in human hepatoma cell lines (Huh7 and HepG2) [95]. Overexpression of TM6SF2 in Huh7 cells reduced cellular triglyceride content [96]. Transient overexpression of human TM6SF2 in mice using a liver-targeting adenovirus containing the human TM6SF2 coding region increased, while knockdown of endogenous TM6SF2 decreased circulating total cholesterol [96]. In the latter study, no change in hepatic fat content was observed. This was hypothetized to be due to the transient exposure, compared to the lifetime exposure of humans carrying the gene variant [96].

Table 3 summarizes seven studies that have reported data on liver fat content and insulin sensitivity in carriers and non-carriers of TM6SF2 E167K gene variant [13,81,92,97,98,99,100]. In all but one of these studies, carriers had a significantly higher liver fat content as determined by ¹H-MRS, MRI, histology or US [13,92,97,98,99,100] than non-carriers. Insulin sensitivity, as determined by HOMA-IR or from oral glucose tolerance test measures did not differ between carriers and non-carriers. Triglyceride concentrations were either lower [81,98,100] or similar [13,97,99] but also in one study higher [92] in TM6SF2 E167K variant allele carriers compared to non-carriers.

Although NAFL defined as macroscopic steatosis affecting >5% of hepatocytes predicts fibrosis [7,8,9], it is unknown how various degrees of steatosis predict liver outcomes. Such information would help the clinician to decide which patients to refer to the hepatologist. The same applies to the non-invasive markers of NAFL proposed to be used by the recent European NAFLD guideline if imaging tools are not available [101]. This guideline also recommended testing for the I148M gene variant in “selected cases and in clinical trials”. The latter might be helpful in identifying patients with NAFLD who are at risk for advanced liver disease but who lack features of the MetS and are therefore not at risk for cardiovascular disease or type 2 diabetes. A cost–benefit analysis of this suggestion is warranted.