Largely driven by prior genetic associations between specific loci and IR or IR-related diseases and by observations of differential gene expression between healthy and IR states, a number of studies has investigated the association between liver DNA methylation changes at specific loci and hepatic IR ( Table 1 ). In this section, we describe key examples of candidate genes studies which explore the multi-layered nature of epigenetic regulation and the resulting coordinated changes in hepatic gene expression and insulin sensitivity.

Today, it is possible to quantify DNA methylation at base-pair resolution across the genome using whole-genome bisulphite sequencing (WGBS), although DNA methylation arrays are more frequently used when comparing cohorts of cases and controls due to their lower cost. These technologies have facilitated the undertaking of epigenome-wide association studies (EWAS) of DNA methylation across multiple IR-related states. A summary of the hepatic IR-related EWAS performed to date in human liver and blood is provided in Table 2 and key examples are discussed below, with a focus on two main pathologies that are characterised by hepatic IR: T2D and NAFLD.

Few studies have directly associated genome-wide DNA methylation levels with insulin sensitivity as the primary outcome, although one such study reported that methylation at PDGFA, which encodes platelet-derived growth factor subunit A (66) ( Figure 2 ), associated with IR measured as HOMA2-IR scores. PDGFA methylation was subsequently implicated in a novel mechanism of hyperinsulinemia-induced hepatic IR, as in vitro studies confirmed that insulin exposure reduced PDGFA methylation and increased its expression, while the direct exposure of cells to recombinant PDGF-AA interfered with insulin signalling by inhibiting insulin-induced AKT activation (66).

In contrast, several EWAS have assessed the levels of hepatic methylation in patients with T2D or NAFLD, and studying the methylome in these states has the potential to uncover novel mechanisms involved in hepatic IR. Firstly, the livers of T2D patients have been reported to show significant DNA hypomethylation alongside significantly lower levels of the dietary methyl-donor folate (vitamin B9) in the blood (64). Sites of hypomethylation occurred near genes previously implicated in T2D genetic risk (64) or involved in hepatic glycolysis and de novo lipogenesis (65), suggesting increased gene expression and pathway activity. This is exemplified by GRB10 (64, 65), encoding the growth factor receptor-bound protein 10, which acts as an inhibitor of pathways regulating growth and metabolism, and has been previously implicated in T2D genetic risk and insulin sensitivity (81). Reflecting the broader genome-wide trends observed in T2D, GRB10 presented lower hepatic methylation in individuals with T2D (64). Interestingly, more recent work in rodents has revealed an important axis between the activation of Grb10 and the promotion of hepatic steatosis (82). In this study, liver-specific GRB10 ablation suppressed to a large extent the lipogenic gene programme and steatosis that are induced by acute endoplasmic reticulum stress (82). It is therefore likely that the T2D-associated DNA methylation changes affecting GRB10 lead to its activation and hence promotion of hepatic steatosis and IR. More generally, significant changes in hepatic DNA methylation can also act as molecular fingerprints of specific metabolic states and have been used to discriminate between diabetic and non-diabetic individuals (67).

Hepatic IR is also deeply interconnected with NAFLD, since studies have shown that IR promotes the progression from simple steatosis to non-alcoholic steatohepatitis (NASH) (83), where hepatocyte injury and portal and lobular inflammation are also present, but also that hepatic steatosis and liver injury can interfere with insulin signalling (84). The EWAS carried out to date for NAFLD underscore the disruption of hepatic insulin signalling by DNA methylation as a major process underlying this disease, with many genes of this pathway showing differential methylation and associated changes in gene expression in NAFLD (some of which are highlighted in Figure 2 ). Furthermore, differential methylation in NAFLD was reported to promote inflammation, fibrogenesis, mitochondrial dysfunction and impaired lipid metabolism (75, 76, 78). Another important finding from these studies is the observation that some of the changes detected in the liver methylome reflect the progressive nature of NAFLD (75, 76, 78). In these studies, more advanced stages of NAFLD associated with global hypomethylation and concomitant over-activation of a pro-fibrogenic gene programme, as exemplified at FGFR2 (75). Fibroblast growth factor signalling is essential in normal liver function and its dysregulation is observed in chronic liver disease, including overexpression of fibroblast growth factor receptors (FGFRs) in hepatocellular carcinoma (85). In it perhaps not entirely unexpected that the methylation of FGFR2 was found markedly altered in advanced NAFLD, with 23 CpG sites hypomethylated, promoting its overexpression and the establishment of an inflammatory and pro-fibrotic niche (75). This trend of hypomethylation in advanced NAFLD, similar to observations in T2D (64, 65), may be of interest for the design of novel interventions to improve insulin sensitivity (addressed in more detail in the section “Epigenetic modifications can be long-term: Dietary methyl donors”). Interestingly, de Mello et al. reported only one site of differential methylation in cases with simple steatosis, compared with 1,292 sites in those with NASH (76), suggesting that prolonged metabolic disruption is associated with acquired changes in the DNA methylome.

One potential limitation of these studies is the association of differentially methylated sites with the closest gene, which is not always the target gene of cis-regulatory elements due to the 3D conformation of DNA in the nucleus (17). Nevertheless, TF binding motif analysis of NAFLD differentially methylated sites revealed strong enrichments for binding motifs of bona fide hepatic regulators of glucose and lipid metabolism such as PGC1α, SREBF2, FOXA1, and FOXA2 (48), further supporting the notion that appropriate DNA methylation is necessary for overall hepatic metabolic homeostasis.

Liver biopsies can provide important insight into liver-specific disease mechanisms in a research setting. However, their invasive nature limits the number of samples available and therefore the statistical power with which to detect associations. This is reflected by the fact that none of the EWAS discussed above included more than 50 cases (see Table 2 ). An alternative is to use the blood methylome as a proxy to detect changes in the liver, since cell-free DNA of liver origin can be detected in the blood, as can liver-specific DNA methylation patterns (77, 86). Blood samples can be easily obtained for large numbers of individuals, providing greater statistical power to research studies. In addition, biomarkers identified in blood are more likely to be useful in a clinical setting, where blood tests will be more appropriate for the early detection of IR. Two recent EWAS performed on whole blood used cohort sizes of 1,450 and 3,400 individuals to detect significant associations between CpG methylation and hepatic steatosis, as well as the levels of liver enzymes (79, 80). One example finding is the hypomethylation of cg08309687 (at LINC00649) which was associated with both NAFLD and T2D, suggesting it could be a robust biomarker of both hepatic fat accumulation and T2D (79). Other EWAS of blood samples from T2D and IR cases are listed in Table 2 and reviewed in detail in (87). These have identified CpG sites at which differential methylation associated with measures of insulin sensitivity and future risk of T2D, including at SREBF1 (69), which encodes the sterol regulatory element binding transcription factor 1 (SREBF1, also known as SREBP-1), a recognised regulator of insulin action in the liver (88) ( Figure 2 ).

Since DNA methylation is typically a more stable modification, it is of great interest whether these acquired changes can be reversed. This important question has been addressed in studies of the liver methylome following bariatric surgery, in which NASH-associated methylation was found to be reversible, but only at specific loci (48) (discussed in more detail in the section “Altering the epigenome: Weight loss and exercise”).

Firstly, HDACs play a widely known role in influencing insulin sensitivity and hepatic lipid metabolism through the deacetylation of both histone and non-histone proteins. Changes in histone acetylation have been reported to alter the expression of glucose-mediated gene expression (101). There are 18 HDACs in humans, divided into classes I (Rpd3-like), II (Hda1-like), III (Sir2-like) and the class IV protein (HDAC11), based on sequence similarity to yeast proteins and co-factor requirement. Despite belonging to the HDAC class of proteins, many HDACs are exclusively or partially involved in the acetylation of non-histone proteins. Classes I and II contribute majorly to histone deacetylation with class I (HDAC1, 2 3 and 8) localising to the nucleus and class IIa (HDAC4, 5, 7 and 9) localising to both the nucleus and the cytoplasm (102). The seven class III HDACs, known as sirtuins based on their similarity to the yeast protein Sir2, are further subdivided and include SIRT1, 2, 6 and 7 which can be found in the nucleus.

While HDACs catalyse the removal of acetylation from histones, HATs catalyse its addition. These include CBP and p300, which have been implicated in hepatic insulin sensitivity. Under physiological conditions, insulin inhibits gluconeogenesis by selectively disrupting the interaction between CBP/p300 and the cyclic AMP-responsive element-binding protein (CREB), a TF involved in the expression of metabolic and gluconeogenic genes (129). In functional studies, p300 overexpression in mice was reported to cause hepatic steatosis, insulin resistance, and inflammation (27). Inhibiting CBP/p300 in mouse liver decreased hepatic lipid content and the expression of lipogenic genes in vivo and also reduced the expression of gluconeogenic genes in primary mouse hepatocytes (130). While this study reported no difference in in vivo insulin sensitivity, mice had reduced fasting glucose and lower glucose levels after glucose loading (130). Similarly, CREB activity is positively associated with hepatic IR and NAFLD in mice (131) and its knockdown in mouse liver decreased hepatic and circulating lipid levels and improved insulin sensitivity (132). The p300/CBP‐associating factor (PCAF) is another HAT that acetylates histone H3 and other non-histone proteins including key regulators of gluconeogenesis (PGC-1α) (133) and lipogenesis (ACLY), effectively regulating insulin sensitivity and glucose production in the liver (134), while p300 is also known to acetylate and activate FOXO1 (109) ( Figure 2 ).

In comparison to the more widely studied histone acetylation, there have been fewer studies linking HMTs, HDMs and DNMTs to hepatic IR. Firstly, the major family of HMTs is the SET domain family, consisting of over 55 proteins in humans, of which around half have no known substrates and may have no enzymatic activity (135, 136). As an example of an association between a HMT and hepatic IR, haploinsufficiency of the HMT MII2 resulted in hyperglycaemia and hyperinsulinemia at fasting and peripheral IR, as well as hepatic fat accumulation and abnormal plasma triglycerides in mice (137). Another HMT, G9a, which catalyses the addition of the repressive histone marks, H3K9me1 and H3K9me2, was observed to have lower expression levels in diabetic mouse models (138). However, G9a has been reported to impact insulin signalling through a methyltransferase-independent mechanisms, by regulating the architectural TF HMGA1, a key regulator of the insulin receptor (138). There are other noteworthy interactions between HDMs and TFs; for example, the HDM Phf2 is required for transcriptional activation by the glucose-responsive TF ChREBP, by erasing the repressive mark H3K9me2 (139) ( Figure 2 ). ChREBP is a master regulator of lipid and glycolytic metabolism and plays a major role in enhancing lipogenesis in insulin resistant states (140). Liver-specific knockout of the HDM JMJD1C, which interacts with USF1 to promote the transcription of lipogenic genes, resulted in reduced expression of lipogenic genes and reduced lipogenesis (28).

In relation to DNA methylation, the mammalian DNA methyltransferases include DNMT1, DNMT3a and DNMT3b (141). DNMT3a has a well-known role in peripheral IR in adipose tissue (142). In the liver, DNMT1 regulated the expression of miR-9-3 in mice, which has been found to regulate hepatic glucose production and insulin sensitivity (143). Knockout of TET1, which catalyses the first step of DNA demethylation (the conversion of 5-methylcytosine into 5-hydroxymethylcytosine), inhibited NAFLD progression in mice and promoted fatty acid oxidation by activating PPARα through increased hydroxymethylation at the PPARA promoter (144).

As discussed above, long-term epigenetic modifications can result from external exposures, including those during development and early life. However, epigenetic modifications can also be transient or short-term, such as those obtained in response to the diet. There is also growing evidence to suggests that some long-term marks can eventually be reversed. Various lifestyle and environmental factors have been shown to influence the epigenome throughout life, either conferring deleterious or beneficial effects to insulin response. Two examples of epigenomic modifiers, the circadian rhythm and exercise, are discussed below.

The emerging field of epitranscriptomics has revealed that RNA modifications can also modulate gene expression levels (228). Whilst there are over 100 types of RNA chemical modifications, N⁶-methyladenosine position (m⁶A) is the most abundant internal mRNA modification in eukaryotic cells and it plays an important role in gene expression, including through altering protein binding, mRNA structure, stability, splicing and translation efficiency (229). mRNA m⁶A methylation has recently been implicated in liver disease. For example, the levels of m⁶A methylation and the m⁶A methyltransferase METTL3 were observed to be higher in the livers of patients with T2D and positively correlated with HOMA-IR (230). Several recent studies have also reported increased hepatic m⁶A levels in mice under HFD, while knocking-out METTL3 improved insulin sensitivity and reduced lipid accumulation, including by increasing specific mRNAs half-lives and reducing the total levels of mRNA encoding the fatty acid synthase, FASN (229–231). The topic of liver disease epitranscriptomics had been extensively reviewed by Zhao et al. (232).

It is clear that the contribution of aberrant RNA modifications to IR is currently less characterised than other epigenomic features. Still, these studies highlight the importance of this epigenetic mechanism as an additional layer of gene regulation that contributes to hepatic insulin homeostasis. Some interesting insights have already been gained regarding the relationship of m⁶A levels and disruption of the circadian rhythm. For example, mice with knockout of the circadian factor Bmal1 presented increased levels of hepatic m⁶A, leading to increased Ppara mRNA lifetime and expression and hepatic steatosis (233). In future studies, it will be interesting to investigate the interplay of RNA modifications and their associated epigenetic machinery with other epigenomic and genetic factors that influence insulin response.

Targeting the epigenome offers the opportunity, in theory, to reverse deleterious epigenetic marks and alter gene expression levels. Multiple drugs and compounds which target the epigenome have been proposed for the treatment of cardiometabolic disease. These include HDAC inhibitors and SIRT1 activators to treat IR and T2D (234, 235). Both in vitro and in vivo studies have aimed to assess whole-body and tissue-specific effects of these epigenetic drugs. For example, HDAC inhibitors and SIRT1 activators have been reported to improve insulin secretion by pancreatic β‐cells (87, 236).

HDAC inhibitors are currently in use to treat different diseases, including cancer, thalassemia and sickle cell anaemia (102, 237), although non-selective, ‘pan’-HDAC inhibitors such as vorinostat and romidepsin have been associated with severe side effects due to the diverse biological functions of HDACs (238). Similarly, the knockdown of multiple class II HDACs caused severe adverse effects in mice and, even though it lowered the expression of gluconeogenic genes, it did not have a significant impact on gluconeogenesis (112). More selective HDAC inhibitors will likely be more appropriate for the treatment of metabolic disorders, although this is challenged by the risk of off-target effects due to the high degree of HDAC similarity. It has been suggested that HDAC8 may be an appropriate target, since it has less structural similarity with other HDAC enzymes (239). The HDAC inhibitor tannic acid was recently reported to attenuate the accumulation of lipids in the liver and prevent NAFLD in mice fed a HFD, by decreasing H3K9 and H3K36 acetylation and transcription of lipogenesis-related genes, including fatty acid synthase (FASN), SREBP1C and ATP citrate lyase (ACLY) (240).

Also of interest is the effect that drugs commonly used to treat cardiometabolic disease have on the epigenome. For example, metformin, a widely used T2D treatment, has been reported to increase SIRT1 activity and decrease class II HDAC and HMT activity (241). In primary human hepatocytes, metformin treatment induced changes in chromatin modifications at an enhancer element within the ATM gene, a locus that has been associated with response to metformin (242), and it reversed H3K36me2 signatures in mouse models of prediabetes and diet-induced obesity (94). The protein encoded by ATM has a role in AMPK activation, a pathway which stimulates fatty acid oxidation and decreases glucose production and lipogenesis in liver (243). In mouse studies, metformin reversed histone marks associated with diet-induced-obesity (94) and activated AMPK, resulting in the phosphorylation of class II HDACs and nuclear exclusion of HDAC4, HDAC5 and HDAC7 in the liver (109).

In conclusion, targeting the epigenome offers the possibility to reverse deleterious epigenetic marks and reprogram unfavourable metabolic pathways. There are several therapies in use which alter the epigenome, including SIRT1 activators such as metformin and resveratrol. However, as this review has discussed, there is still much to learn about the effects that potential interventions such as HDAC inhibitors or dietary supplements may have on the hepatic epigenome and wider cellular functions. Exercise and weight loss are attractive treatment options which may effectively alter epigenomic programming and the collective evidence suggests that these interventions will be most effective if maintained for extended periods of time. Future research into treatments for hepatic IR have two main avenues (1): preventing the deposition of persistent epigenetic marks, such as those seen in NAFLD progression, and (2) developing treatments to reverse deleterious epigenetic marks and metabolic profiles.

Aside from potential treatments, the epigenome may also be a source of disease biomarkers. These may enable the diagnosis of disease progression without requiring invasive biopsies or may contribute towards estimates of future disease risk. For the former, novel biomarkers which can discern mild from severe disease states may prove particularly useful in a clinical setting. While IR can be effectively diagnosed using a blood test, complex IR-related states such as NASH, which is a major risk factor for cirrhosis and HCC, are difficult to distinguish from simple steatosis without a liver biopsy (244). Bloodborne biomarkers offer a potential non-invasive diagnostic. For example, circulating levels of cytokeratin 18, which is seen following hepatocyte death, has been reported as a strong candidate for distinguishing NASH patients (245). In terms of the epigenome, circulating liver-specific miRNAs may also provide rapid and non-invasive biomarkers of liver damage (246) and IR/NAFLD (247). Methylation of the PPARG gene in plasma DNA was reported to reflect the molecular pathology associated with fibrotic liver disease (248). Several EWAS have reported associations between methylation in blood samples and IR-related disease (see previous EWAS section). Promising studies have also identified sites of methylation which associate with the future risk of T2D onset (72).

A study by Dijk et al. analysed neonatal blood spot samples and identified 63 genomic regions which associated with insulin sensitivity at five years of age (149). Such regions could offer biomarkers which may inform appropriate lifestyle interventions to reduce risk of IR-related disease. More recently, Sadeh et al. developed a method to identify liver-specific histone modification signatures in plasma cell-free nucleosomes using ChIP-seq (249). Despite the relatively small sample size and patient heterogeneity, the authors were able to detect liver-disease associated changes in histone modification enrichment at specific loci (249). This methodology is expected to be taken up in the future to facilitate the high-throughput interrogation of disease signatures in patient blood samples.

A popular area of recent discussion is in predicting disease onset using genetic risk scores (250). This approach currently has limited use in a clinical setting, although future research is likely to benefit from incorporating genetic risk scores with epigenetic marks, in order to capture acquired as well as inherited risk. It should also be noted that genetic variation can cause epigenetic variation, which should be considered in efforts to integrate genetic and epigenetic risk. Another area of consideration is the use of biomarkers in identifying subgroups of patients. For example, genetic data can be used to distinguish T2D cases characterised by severe IR, which is closely linked with the risk of complications and choice of treatments (251, 252). There is some evidence to suggest that blood methylation may predict future risk of complications in diabetic cases (253). At present, investigating epigenetic markers of IR offers immediate promise in a research setting, as they continue to uncover novel mechanisms contributing to IR. Incorporating epigenetic biomarkers in mainstream healthcare must first provide enough benefit to outweigh the costs of additional infrastructure and training.

The studies discussed in this review support the notion that hepatic IR and its associated traits and comorbidities are not only characterised by broad changes of the liver epigenome, but also that the epigenome should be regarded as a modifier of metabolic programs and whose disruption may impact insulin sensitivity. The current studies also highlight potentially modifiable pathways to improve hepatic insulin sensitivity, but a tremendous amount of research work will be required to investigate the additional aspects of hepatic IR and the safety of so-called epigenetic drugs.

While the observational nature of many human studies still precludes the characterisation of mechanisms of disease progression at large, mounting evidence suggests that the epigenome should be regarded as a rich pool for the discovery of disease biomarkers. The observation that individuals with IR showed higher global methylation than their normoglycemic monozygotic twins emphasises the role of environmental factors in modulating the epigenome; information which can provide biomarkers independent from inherited genetic risk factors (74). This notion has truly taken off in regard to the profiling of liver methylomes in IR, as demonstrated by the large array of methylome-wide studies we discussed. The high degrees of correlation between DNA methylation patterns in blood and liver (77) make blood methylation biomarkers an appealing strategy for the early identification of individuals at risk of developing hepatic IR and related cardiometabolic conditions (254). It must be noted, however, that early DNA methylation array versions largely excluded informative methylation in the non-coding genome, instead focusing on gene promoters and bodies. Given the large body of evidence demonstrating that the non-coding genome is highly functional and hosts a myriad of different types of DNA regulatory elements (255), previous studies were inevitably limited in scope. For this reason, there has been an increased uptake by the community of true genome-wide profiling of DNA methylation landscapes with the application of techniques such as Methylated DNA Immunoprecipitation followed by sequencing (MeDIP-seq) (256), whole-genome bisulphite sequencing (WGBS) and, more recently, with the repurposing of long-read sequencing technologies to detect methylated DNA (257, 258). Whilst these techniques offer the opportunity to obtain a very detailed view of the DNA methylation profiles of different tissues and metabolic states [even of individual cells with their adaptation to single-cell use, as already demonstrated in mouse liver (259)], their application on a population level is not expected to be rolled out soon given the high cost per sample and need for training and infrastructure. Still, we do expect to obtain increased resolution in the picture of IR methylomes in the near future with the use of more recently developed DNA methylation array designs, which incorporate a higher number of distal regulatory elements. For example, the Infinium MethylationEPIC Chip [Illumina (260)], includes ENCODE open chromatin sites and enhancers. These studies will enable the increased integration of genetic associations from GWAS, which most often detect noncoding risk variants in distal regulatory elements, with EWAS signals.

Due to higher associated costs and technical challenges, the profiling of histone modifications has not really matched the pace of methylome-wide studies. Nevertheless, exciting research has recently demonstrated the potential to once more harness the correlation between liver and blood epigenomic profiles, this time by performing ChIP-seq on circulating cell-free DNA (249). Future studies are expected to leverage on this new technology to identify additional biomarkers of hepatic IR, assisting in the prevention of NAFLD, T2D and other cardiometabolic diseases, which together have an enormous health, societal and economic impact. It is also expected that future studies will provide much needed granularity to our view of liver IR epigenomes by, for instance, mapping accessible (and thus active) chromatin sites with single-cell resolution in liver biopsies from patients presenting varying degrees of disease severity. Importantly, recent technological advances now make it possible to perform multiomic analyses with single-cell resolution (261). Such powerful studies may revolutionise our understanding of hepatic IR and potentially uncover disease processes that are particular to one cell type, or even one sub-population of resident liver cells.

Understanding the interplay between the liver epigenome extrinsic factors such as in utero exposures or diet will inevitably still rely on appropriate animal models. These models are expected to remain central for studies of long-term effects on the epigenome. Still, improvements in stem cell differentiation protocols to produce mature human hepatocytes are expected to enable the further development of novel human in vitro models of IR. Together with rodent models, sophisticated in vitro models such as organ-on-a-chip will also enable the better understanding of how cues from other tissues important for glucose homeostasis, such as pancreatic islets (262), influence the liver epigenome to promote IR.

Finally, future studies of the epigenetic basis of hepatic IR should aim to better integrate epigenomic profiles with genetic susceptibility factors. This integration of genomic and epigenomic information will be further enriched by the analysis of more ethnically diverse panels, as the vast majority of current studies are centred around the analysis of Caucasian genetic risk factors. Such results are unlikely to arise from individual research labs and are instead expected to be produced by large international consortiums. Thus, a difficulty will not only be data gathering, but also standardization of data processing and homogenization. Overall, while there will be challenges ahead, the field of epigenomics is expected to offer important contributions to the identification of targetable pathways to improve hepatic insulin sensitivity.