Obesity increases the risk of metabolic diseases, including insulin resistance and type 2 diabetes, as well as cardiovascular disease. In addition to lipid accumulation in adipose tissue, obesity is associated with increased lipid storage in ectopic tissues, such as skeletal muscle and liver. Furthermore, lipid accumulation in the heart may result in cardiac dysfunction and heart failure. It has recently been demonstrated that intracellular lipid accumulation in ectopic tissues leads to pathological responses and impaired insulin signalling. Here, we will review the current understanding of how lipid storage and lipid droplet physiology affect the risk of developing metabolic diseases.
Obesity and the associated metabolic diseases, such as insulin resistance and diabetes, are among the most prevalent disorders adversely affecting world health. Increased lipid storage in ectopic tissues, including skeletal muscle and liver, is associated with insulin resistance . Recent evidence indicates that intracellular lipid accumulation in ectopic tissues triggers pathological responses with subsequent impairment of insulin signalling.
The purpose of this review was to summarize current understanding of the role of lipotoxicity in the development of insulin resistance and discuss whether the relationship between ectopic lipid storage and insulin resistance is indeed harmful.
Lipid droplet organization and structure
Neutral lipids such as triglycerides and cholesteryl esters are insoluble in water and thus are included in a highly hydrophobic oil phase (the core of the so-called lipid droplet) to facilitate intracellular storage . This hydrophobic phase is surrounded by a monolayer of amphipathic structures, in particular phospholipids and unesterified cholesterol (Fig. 1a).
Lipid droplets are present in most cells, but their size varies considerably from very large droplets in adipocytes to small droplets in skeletal muscle (Fig. 1b). They were originally considered to be simple storage depots but are now recognized as highly dynamic organelles that actively participate in numerous cellular processes . Several proteins are associated with the lipid droplet (Fig. 1a), and these are critical for lipid droplet formation and stability, trafficking events in the cell, and regulation of lipid droplet size through control of lipolysis and the promotion of fusion events . Recent genome-wide screens have revealed a diverse array of proteins involved in lipid droplet physiology [3, 4].
The best characterized and quantitatively most important lipid droplet proteins are the perilipins, of which five related proteins are currently known: perilipin-1–5. The perilipin subtypes associated with the lipid droplets change as the droplets mature; for example, nascent lipid droplets in adipocytes are coated with perilipin-3 and 4, but as they gain triglycerides and enlarge, these perilipins are replaced by perilipin-2 and ultimately perilipin-1 .
Perilipin-1 is mainly expressed in adipocytes and steriogenic cells . It exists in three isoforms (A, B and C) , formed by alternative splicing, of which the A isoform is by far the most abundant . Mice lacking perilipin-1 are lean with small adipose lipid droplets, are resistant to diet-induced obesity and have an enhanced basal lipolytic rate and a blunted response to hormone-stimulated lipolysis .
Perilipin-2 [also known as adipocyte differentiation-related protein (ADRP)] is expressed ubiquitously . It is always associated with droplets and is degraded in the absence of neutral lipids . Its expression is highly related to the amount of neutral lipids in the cell , and overexpression of perilipin-2/ADRP results in increased formation of droplets [12-14]. Thus, perilipin-2/ADRP seems to be important for the assembly of lipid droplets.
Perilipin-3 [also known as tail interacting protein 47 (TIP47)] was originally reported to be involved in the intracellular transport of mannose 6-phosphate receptors between the trans-Golgi and endosomes [15, 16], but is now also known to be present on lipid droplets . In contrast to perilipin-2/ADRP, perilipin-3/TIP47 can be observed in a diffuse pattern throughout the cytosol and becomes localized to lipid droplets in the presence of increased levels of fatty acids .
Perilipin-4 (also known as S3-12) has some sequence similarity with other perilipins and is mainly expressed in white adipose tissue . It has been suggested that nonlipid droplet pools of perilipin-4/S3-12 constitute a reservoir of coat proteins that allow rapid packaging of newly synthesized triglycerides and maximize energy storage during nutrient excess .
Perilipin-5 [also known as lipid storage droplet protein 5 (LSDP5)] is mainly expressed in oxidative tissues, such as heart, type I skeletal muscle, liver and brown adipose tissue . Overexpression of perilipin-5/LSDP5 in cultured cells results in a substantial increase in the accumulation of triglycerides in response to fatty acid treatment , which might be explained by a decrease in both basal and stimulated lipolysis . Thus, perilipin-5/LSDP5 seems to protect the triglyceride core of lipid droplets from degradation in a similar manner to perilipin-1. Perilipin-5/LSDP5 is transcriptionally regulated by peroxisome proliferator-activated receptor (PPAR)α in striated muscle and liver and by PPARγ in white adipose tissue [19-21].
Additional lipid droplet proteins
In addition to perilipins, several other proteins have been associated with lipid droplets. These include proteins involved in the transport and sorting of lipid droplets, such as motor proteins, Rab proteins and soluble N-ethylmaleimide-sensitive factor (NSF) attachment protein receptors (SNAREs). Rab18, a small GTPase, localizes to lipid droplets and mediates interactions with the endoplasmic reticulum (ER) [22, 23]. The SNARE protein soluble NSF attachment protein of 23 kDa (SNAP23) has also been identified on purified lipid droplets and we have shown that it plays a role in lipid droplet fusion .
A variety of enzymes required for lipid metabolism have also been identified on lipid droplets. In proteomic analyses, acyl-CoA synthase was identified on lipid droplets; this enzyme converts free fatty acids into fatty acyl-CoA esters and thereby has a key role in lipid biosynthesis . In addition, diacylglycerol O-acyltransferase 2 (DGAT2), which catalyses the final reaction in the synthesis of triglycerides in which diacylglycerol (DAG) is covalently bound to long-chain fatty acyl-CoAs, localizes to lipid droplets when fatty acids are added to cells to drive triglyceride synthesis and storage [26, 27].
It has also been shown that members of the cell death-inducing DFF45-like effector (CIDE) protein family, including CIDEA, CIDEB and CIDEC (also known as FSP27), are associated with lipid droplets. CIDE proteins appear to have a unique role in controlling the size of cytosolic lipid droplets in various cell types by promoting lipid droplet clustering and fusion [28, 29]. Animals that are deficient in CIDEA, CIDEB or CIDEC have lean phenotypes with higher energy expenditure and are resistant to diet-induced obesity and insulin resistance . Furthermore, the CIDE proteins have been shown to be important for the development of metabolic disorders in humans [31, 32].
Numerous proteins have also been identified that do not appear to be related to lipid metabolism and storage. Such proteins could simply be artefacts that appear during the isolation of lipid droplets. An alternative proposal by Welte  is that the lipid droplet surface is a reservoir for proteins that could be used in other organelles. The author suggested that the droplets deliver their protein cargo to the appropriate organelle following their transfer on microtubules. In addition, the droplets could function as a ‘rubbish dump’ for proteins before destruction by proteasomal degradation .
Turnover of lipid droplets
The formation of lipid droplets
Lipid droplets are formed in the ER where the enzymes to synthesize neutral lipids are located . However, the exact mechanism of formation is not understood. Models in which the formed neutral lipids are ‘oiled out’ between the two leaflets of the membrane have been proposed but not proven, and several additional models have been suggested [2, 35]. A major reason for the lack of understanding is the small size of the newly formed primordial droplets and the need to study the process in real time by time-lapse photography or similar technologies.
A number of proteins have been linked to the assembly of lipid droplets, but in all cases, their roles in this process are unclear. The ADP ribosylation factor (ARF)-dependent adaptor protein coat protein 1 (COP1), which is involved in membrane trafficking, has been linked to the assembly process [36, 37]. In addition, we demonstrated that phospholipase D1 (PLD1) and extracellular signal-regulated kinase 2 (ERK2) are essential for the assembly process . PLD1 catalyses the formation of phosphatidic acid , which has been implicated in the budding of vesicles . Thus, a mechanism reminiscent of a budding process may be involved in the process of forming lipid droplets. We also showed that ERK2 phosphorylates the motor protein dynein, and it is feasible that this protein provides the energy needed to allow the droplet to bud from the microsomal membrane . It is interesting that ERK2 is a downstream effector of insulin signalling , and inhibition of ERK2 abolishes insulin-stimulated lipid droplet formation .
Several other proteins that are associated with the ER but not lipid droplets are also involved in lipid droplet formation. One such example is fat storage-inducing transmembrane (FIT) protein 2, a six transmembrane domain-containing protein located in the ER membrane . It has been shown that the activity of FIT protein 2 in mediating lipid droplet formation is regulated by its conformation, but the mechanism by which the assembly process is influenced is unknown .
Lipid droplet growth and fusion
Lipid droplets are formed as small primordial droplets that increase in size, either by expansion or by fusion of existing lipid droplets. The fusion process is dependent on transport on microtubules, a process that requires motor proteins. We have demonstrated the need for the motor protein dynein and shown that ERK2 is important for targeting dynein to lipid droplets . SNARE proteins are known to be involved in fusion between transport vesicles and target membranes during intracellular transport of secretory proteins , and we have shown that the SNARE proteins SNAP23, syntaxin-5 and vesicle-associated membrane protein (VAMP)4 have a role in the lipid droplet fusion process .
It should be noted that triglyceride droplets would fuse spontaneously with each other if not provided with a surface that protects them against this fusion. Such protection is important because spontaneous and unregulated fusion could lead to large amounts of neutral lipids in the cytoplasm, which could have an impact on cell function. Because lipid droplets only contain neutral lipids and not hydrophilic substances, they are surrounded simply by a protective monolayer, in contrast to transport vesicles which are surrounded by a bilayer. However, only the outer monolayer is involved in fusion between transport vesicles and target membranes. We have thus proposed that the SNARE proteins use the same mechanism for lipid droplet fusion as for fusion between vesicles, that is they remove water between the fusing droplets and overcome the repelling force between the surrounding monolayer .
The availability of phospholipids, in particular phosphatidylcholine, is considered to be important for the expansion of lipid droplets. When lipid droplets expand to store more neutral lipids, their phospholipid content must also increase. It was recently shown that expanding lipid droplets recruit CTP:phosphocholine cytidylyltransferase (CCT), the rate-limiting enzyme for phosphatidylcholine synthesis and that this enzyme is activated when it binds to the surface of lipid droplets . When CCT availability is limited, lipid droplets may grow by fusion instead of expansion.
It has been proposed that DGAT2 promotes an increase in size by catalysing de novo triglyceride synthesis directly in the droplet . In addition, we presented a model in which small newly assembled lipid droplets (0.1–0.2 μm in diameter and thus barely visible using light microscopy) fuse with larger droplets and thereby deliver newly formed triglycerides to these droplets .
Lipid droplet catabolism
Lipid droplets contain enzymes such as adipose triglyceride lipase (ATGL) and hormone-sensitive lipase (HSL) that promote hydrolysis of triglycerides to generate fatty acids. In adipocytes, perilipin-1 interacts with comparative gene identification-58 (CGI-58) [46, 47], a coactivator of ATGL, to prevent triglyceride hydrolysis. When ATP is needed, β-adrenergic signalling activates lipolysis through increasing cyclic AMP (cAMP) levels and activating cAMP-dependent protein kinase A (PKA). Perilipin-1 is phosphorylated by PKA, resulting in the release of CGI-58, which then associates with and activates ATGL on lipid droplets [48, 49]. This enables the release of the first fatty acid from triglycerides to generate DAG, the substrate for HSL.
Perilipin-1 also promotes triglyceride degradation; β-adrenergic stimulation does not promote lipolysis in perilipin-1-knockout mice, and cells derived from these mice fail to show translocation of HSL to the lipid droplet [50, 51] (reviewed in ). Based on these and more direct results, it has been proposed that concomitant phosphorylation of perilipin-1 and HSL results in translocation of HSL to the lipid droplet where it catalyses the hydrolysis of triglycerides . By contrast, in oxidative tissues, although perilipin-5/LSDP5 interacts with ATGL leading to recruitment of ATGL to lipid droplets, this interaction protects the lipid droplets from lipolysis .
Lipotoxicity and insulin resistance
The development of insulin resistance is highly related to the accumulation of neutral lipids in locations other than adipose tissue  (Fig. 2). Skeletal muscle is extremely important for the development of systemic insulin resistance because it is the tissue in which most of the insulin-dependent disposal of glucose occurs. However, the mechanism by which increased lipid accumulation induces insulin sensitivity in this tissue is still debated.
Lipid accumulation in skeletal muscle not only occurs in patients with insulin resistance, but also in endurance-trained marathon runners, that is individuals with excellent insulin sensitivity [55, 56]. Thus, there may be other factors coupled to the accumulation of triglycerides that are important to promote insulin resistance. Extreme lipid-loading during pathological conditions, such as obesity, may lead to an imbalance in lipid homoeostasis to induce toxicity in the tissue, a phenomenon termed lipotoxicity. Bioactive lipid metabolites (e.g. ceramides, DAG and long-chain fatty acyl-CoAs) accumulate under these conditions and may cause cellular dysfunction and insulin resistance . In agreement, Bosma et al.  recently showed that in vivo muscle-specific overexpression of perilipin-2/ADRP, one of the most abundantly expressed lipid droplet proteins in skeletal muscle, resulted in increased lipid droplet accumulation but improved skeletal muscle insulin sensitivity. These data indicate that perilipin-2/ADRP promotes the distribution of excess fatty acids to triglyceride storage in lipid droplets, and thereby blunts lipotoxicity-associated insulin resistance .
Potential mediators of lipotoxicity
Ceramide accumulation is associated with impaired insulin signalling and insulin resistance . Ceramide blocks the activation of the serine/threonine-specific protein kinase Akt [also known as protein kinase B (PKB)] by inhibiting the translocation of Akt/PKB to the plasma membrane . It has been suggested that ceramide disables 3-phosphoinositide binding to the pleckstrin homo-logy domain of Akt/PKB by a protein kinase C (PKC)ξ-dependent mechanism .
In addition, elevated DAG levels and the resulting activation of PKCθ and PKCε are associated with impaired insulin signalling and insulin resistance [61, 62]. PKCθ inhibits insulin receptor substrate (IRS), an important ligand in the insulin response, by mediating phosphorylation on one or several serine residues and thus preventing the insulin-dependent tyrosine phosphorylation of IRS. PKCε inhibits insulin signalling by serine phosphorylation of IRS and by direct association with the insulin receptor to impair its kinase activity .
Diacylglycerol kinases (DGKs) play a key role in reducing levels of DAG, and Chibalin et al.  recently showed that DGKδ is involved in protecting against insulin resistance. These authors demonstrated that both expression of DGKδ protein and total DGK activity are reduced in skeletal muscle from type 2 diabetic patients and diabetic rodents; these reductions are normalized upon correction of hyperglycaemia. They concluded that reduced DGKδ activity causes aberrant insulin signalling, defective glucose uptake in skeletal muscle and adipose tissue, and systemic insulin resistance .
In contrast to the prevailing view that increased DAG levels promote insulin resistance, it has been shown in a recent study that total myocellular DAG levels are markedly higher in insulin-sensitive highly trained athletes than in normal-weight or obese sedentary volunteers. These findings suggest a more complex role for DAGs in insulin action, and instead the authors propose that particular molecular species of DAGs or ceramides are linked to insulin resistance .
Saturated fatty acids  and, in particular, branched-chained amino acids  have also been shown to induce insulin resistance. Saturated fatty acids such as palmitic acid signal though Toll-like receptor 4 (TLR4), which is involved in modulating innate immunity, and this signalling is important for insulin-induced insulin resistance [68-70]. TLR4 agonists promote biosynthesis of sphingolipids, including ceramides, and TLR4 deficiency has been shown to prevent ceramide formation induced by saturated fatty acids . Partially oxidized fatty acids resulting from increased fatty acid oxidation have also been linked to the development of insulin resistance , but the underlying mechanism remains to be resolved.
Lipotoxicity and insulin resistance in the heart
Myocardial triglycerides constitute a highly dynamic fatty acid storage pool that can be used as a reserve source of energy. In cardiomyocytes, triglycerides are synthesized by acyltransferases and phosphatases at the sarcoplasmic reticulum and mitochondrial membrane and then packaged into either cytosolic lipid droplets for temporary storage [72-75] or into lipoproteins for secretion . A complex interplay among lipases, lipase regulatory proteins and lipid droplet scaffold proteins leads to the controlled release of fatty acids from the cardiac triglyceride pool for subsequent mitochondrial β-oxidation and energy production.
It is now evident that regulation of myocardial triglyceride metabolism is critical for both cardiac energy metabolism and function as elevated levels of myocardial lipids, which are prevalent in patients who are obese and/or have type 2 diabetes, are linked to cardiac dysfunction [1, 77]. Pathological heart conditions, such as myocardial ischaemia and heart failure, are also associated with lipid accumulation [78, 79]. Recent research has shown that lipid overload in cultured cardiomyocytes inhibits insulin signalling in these cells . Furthermore, reducing myocardial lipid accumulation in patients with advanced heart failure reverses cardiac insulin resistance and normalizes cardiac metabolism . Proposed mechanisms for the link between increased intracellular lipid content and heart dysfunction include ceramide-induced apoptosis, increased formation of reactive oxygen species, mitochondrial dysfunction and/or ER stress [81-83].
Potential mechanism of lipid droplet-induced insulin resistance
Although lipid droplets appear to have a central role in the development of insulin resistance, the mechanisms involved have not been fully elucidated. We found that SNAP23 provides a link between lipid droplets and insulin resistance .
Under normal conditions, insulin promotes the uptake of glucose in skeletal muscle and adipose tissue by promoting the translocation of the glucose transporter 4 (GLUT4) from transport vesicles in the cell to the plasma membrane. GLUT4 is integrated into the plasma membrane following fusion between its transport vesicle and the plasma membrane, a complex process that involves the v-SNARE VAMP2 and the t-SNARE complex SNAP23 and syntaxin-4 . Thus, SNAP23 is important for both fusion between the GLUT4 transport vesicles and the plasma membrane and fusion between lipid droplets.
We showed that lipid accumulation in cells leads to distribution of SNAP23 away from the plasma membrane to the interior of the cell, resulting in insulin resistance . We translated these in vitro results into humans by studying skeletal muscle biopsies from patients with type 2 diabetes and lean and obese controls, and observed that triglyceride accumulation in human skeletal muscle is accompanied by both insulin resistance and redistribution of SNAP23 from the plasma membrane to the interior of the cell [85, 86]. Thus, a missorting of SNAP23 seems to be critical for the development of insulin resistance.
Increased expression of DGAT has been shown to prevent the development of insulin resistance by promoting the formation of lipid droplets [81, 87]. This observation demonstrates that it is not the droplet per se that removes SNAP23 from the GLUT4 transport vesicles but the availability of nonesterified fatty acids (NEFAs).
Interestingly, Schwenk et al.  have recently shown that overexpression of VAMP3, but not VAMP2, completely prevented lipid-induced inhibition of insulin-stimulated GLUT4 translocation in an in vitro model of cardiac insulin resistance.
Is the lipid droplet beneficial, harmful or neither?
Despite the clear relation between ectopic lipid accumulation and the development of insulin resistance, there is a growing body of evidence against a direct role of neutral lipids in the induction of insulin resistance but supporting an involvement of factors related to the structure of lipid droplets and their metabolism . It has further been shown that channelling more toxic lipids into triglyceride storage protects from insulin resistance and cell death [81, 87]. Thus, it is possible that the formation of lipid droplets could prevent the development of insulin resistance by ‘detoxifying’ fatty acids. However, this hypothesis is based on overexpression of enzymes involved in triglyceride biosynthesis and the ‘athletes’ paradox' and has not been directly proven.
Before it can be concluded that the lipid droplet is not harmful but rather a protector of insulin resistance or ‘an innocent bystander’, a number of issues should be addressed. First, lipid droplets are usually formed as a result of increased levels of fatty acids and are continuously turning over; the released fatty acids are either used or re-enter the lipid droplets as triglycerides. Thus, a large pool of lipid droplets is associated with increased availability of fatty acids and DAGs. Overexpression of DGAT or other enzymes involved in the formation of triglycerides has been shown to promote the storage of fatty acids and prevent cellular accumulation [81, 87]. However, these experimental findings have not been tested in humans, so the physiological relevance of an increased expression of DGAT remains unclear.
Second, the types of fatty acids should be determined. It is well known that saturated fatty acids are more toxic to the cell than unsaturated fatty acids  and that the former are linked to the development of insulin resistance . An increased pool of lipid droplets may serve as a substrate for biosynthesis of proinflammatory mediators such as ceramides.
Third, little is known about the fate of primordial droplets because their size makes them difficult to observe under light microscopy. The turnover of this pool of droplets needs to be investigated.
Hepatic steatosis induces metabolic dysfunction and oversecretion of triglyceride-rich lipoproteins
Formation of very low-density lipoproteins
The disposal of lipids from the liver is maintained by secretion of triglyceride-rich very low-density lipoproteins (VLDL). The formation of VLDL in the liver is a complex process. It starts with the synthesis of apolipoprotein (apo)B100 in the ER of the cell . ApoB co-translationally associates with phospholipids to form partially lipidated precursor VLDL particles . These are further lipidated through multiple interactions between apoB and the microsomal triglyceride transfer protein (MTP) resulting in the formation of triglyceride-poor VLDL2 particles; these are either secreted from the cell or act as precursors to larger, triglyceride-rich VLDL1 particles [94, 95]. The conversion of VLDL2 to VLDL1 requires a bulk addition of triglycerides and thus differs from the stepwise lipidation of apoB to form the partially lipidated precursor VLDL particles .
Very low-density lipoproteins formation is highly dependent on the availability of lipids: if apoB fails to be lipidated and is incorrectly folded it will be sorted to posttranslational degradation. Degradation of apoB is accomplished by both proteasomal and nonproteasomal pathways and each pathway is regulated by one or more metabolic factors [96-102].
Studies indicate that the fatty acids used for the biosynthesis of VLDL-triglycerides are derived from triglycerides stored in cytosolic lipid droplets [103-105]. It has been proposed that the assembly of a droplet starts in the hydrophobic portion of the microsomal membrane with the formation of a triglyceride lens, which is then released into the cytosol . A triglyceride lens may also bud into the lumen of the secretory pathway, thereby giving rise to a luminal droplet that becomes the core of VLDL. This hypothesis remains to be tested experimentally [107-109].
Specific lipid droplet proteins also seem to be of importance for VLDL formation. Recent studies have clarified that CIDEB binds to apoB in partially lipidated precursor VLDL particles, thereby facilitating the transport of triglycerides synthesized in the ER to VLDL precursors during VLDL lipidation . In the absence of CIDEB, bulk triglycerides cannot be efficiently transported to VLDL precursors, resulting in the secretion of triglyceride-poor VLDL particles and the accumulation of triglycerides in the cytosol .
Overproduction of VLDL1 in nonalcoholic fatty liver disease drives dyslipidaemia
Increased liver fat in humans is linked to overproduction of large VLDL1 particles , which is not surprising given that VLDL formation is dependent on lipid availability. The most common form of liver steatosis is nonalcoholic fatty liver disease (NAFLD) [112, 113]. This is related to insulin resistance and type 2 diabetes, and probably explains (at least partly) the dyslipidaemia that is observed in subjects with insulin resistance and type 2 diabetes [114-117]. The different components of this dyslipidaemia (hypertriglyceridaemia, low HDL, accumulation of small, dense LDL and postprandial hyperlipidaemia) are not isolated abnormalities but are closely linked to each other metabolically [118-120]. It is now recognized that the key pathological mechanism underlying the dyslipidaemia is overproduction of VLDL1 [118, 120].
Insulin signalling regulates VLDL production
The formation of cytoplasmic lipid droplets requires MTP  and droplets can be stabilized by apoC-III . Of interest, insulin resistance is linked to enhanced forkhead box (Fox) O1 activity  and sustained nuclear localization of FoxO1 increases expression of MTP and apoC-III . Under physiological conditions, insulin signalling regulates VLDL production during the postprandial period by targeting apoB, leading to intracellular degradation, and thus limiting formation of apoB-containing lipoproteins. In addition, the acute effect of insulin on VLDL kinetics has been elucidated in humans in vivo [125-129]. Although modelling approaches differ, they all show decreased secretion of VLDL-triglycerides and VLDL-apoB. Furthermore, insulin infusion has a greater effect on the secretion of VLDL-triglycerides than VLDL-apoB [126-128], and it has been shown to suppress mainly VLDL1-apoB production, while having little effect on VLDL2-apoB production [125, 129]. Thus, insulin not only reduces the number of overall VLDL particles, but also shifts the balance between VLDL2 and VLDL1 to reduce the relative proportion of VLDL1 particles  (Fig. 3).
Liver fat correlates with reduced effect of insulin
Kinetic studies have shown that liver fat is associated with a lack of insulin-induced VLDL1 suppression; insulin downregulates VLDL1 secretion in subjects with low levels of liver fat (as discussed above), but fails to suppress VLDL1 secretion in subjects with high liver fat levels, resulting in overproduction of VLDL1  (Fig. 3). The reason for this lack of effect is not known. Insulin suppresses the NEFA pool to a similar extent regardless of liver fat level , and thus the high degree of VLDL1 production in individuals with a high liver fat content must be achieved by the liver using a greater portion of systemic NEFA or recruiting other sources of triglycerides, such as from hepatic stores.
The lack of insulin-mediated VLDL1 suppression in subjects with increased liver fat content explains why conversion of VLDL2 to VLDL1 and secretion of triglyceride-rich VLDL1 particles are enhanced in individuals with type 2 diabetes [111, 130]. It is important to note that these subjects secrete more – not larger – VLDL1 particles than nondiabetic control subjects [111, 120, 130]. This indicates that the amount of lipid added to an individual VLDL2 particle to produce VLDL1 is equal in subjects with type 2 diabetes and nondiabetic individuals, but the rate of conversion is increased in those with type 2 diabetes [111, 131, 132]. These results indicate that cytoplasmic lipid droplets are of similar sizes or that only lipid droplets with a defined size can fuse with VLDL2 particles.
Is the observed association between high liver fat levels and reduced suppression of VLDL1 by insulin the result of fatty liver, or is fatty liver merely a consequence of hepatic insulin resistance? This is still controversial , but loss of insulin suppression of apoB secretion seems to occur before global hepatic insulin resistance [133, 134]. In agreement, Sorensen et al.  have shown that insulin suppression of hepatic glucose production is preserved in obese men in whom insulin-mediated suppression of VLDL1 secretion has been lost.
Increased de novo lipogenesis in subjects with insulin resistance
In fasting healthy human subjects, hepatic de novo lipogenesis (DNL) of fatty acids from carbohydrates in the liver contributes less than 5% to VLDL triglyceride content . By contrast, DNL has been shown to account for approximately 25% of liver and VLDL triglyceride content in hyperinsulinaemic subjects with NAFLD . In healthy individuals, DNL is elevated following meals , which can be accounted for by elevations in the circulating levels of lipogenesis precursors. However, in subjects with NAFLD, DNL is already elevated in the fasted state, and further postprandial elevation is not observed . It is likely that the increased DNL augments the loss of insulin-mediated suppression of VLDL1 secretion, contributing to the hypersecretion of VLDL1. However, to our knowledge, a correlation between hepatic DNL and VLDL1 secretion in humans has not been reported. If in addition in subjects with NAFLD hyperglycaemia prevails, upregulation of apo-CIII by glucose creates a milieu to further promote VLDL1 synthesis and secretion . Resistin has also been shown to enhance VLDL production by increasing the abundance of apoB mRNA and stimulating expression of MTP and DNL while inhibiting insulin-signalling pathways .
Molecular mechanisms of insulin-mediated suppression of VLDL1 formation
The molecular mechanisms involved in the direct suppression of VLDL1 formation by insulin are elusive and several mechanisms have been proposed. Sparks and co-workers have shown that activation of phosphatidylinositol 3-kinase (PI3-K) is necessary for the insulin-stimulated decrease in apoB secretion from rat hepatocytes [141-143]. PI3-Ks are a family of lipid kinases capable of phosphorylating the 3′ OH of the inositol ring of phosphoinositides to generate phosphatidylinositol (3,4,5)-triphosphate (PIP3). PIP3 binds with high affinity to the pleckstrin homology domain of Akt/PKB and mediates translocation of Akt/PKB to the plasma membrane .
The signal for increased export of hepatic triglycerides is likely to involve interruption of insulin-stimulated activation of PI3-K, which can occur through induction of the nontransmembrane protein tyrosine phosphatase (PTP)-1B [145, 146]. Liver-specific deletion of PTP-1B improves hepatic insulin signalling, enhances insulin suppression of hepatic gluconeogenesis and lowers circulating levels of triglycerides and cholesterol [147, 148]. Evidence that generation of PIP3 is necessary for insulin-mediated suppression of VLDL formation has been provided by recent studies of phosphatase and tensin homolog (PTEN) , a lipid phosphatase that catalyses removal of the 3′ phosphate of phosphoinositides. Data suggest that loss of PTEN, and the associated increase in cellular PIP3, result in suppressed VLDL secretion .
Insulin may also decrease VLDL secretion by inhibiting MTP expression via activation of the mitogen-activated protein kinase pathway [150, 151]. Furthermore, studies in ob/ob mice have shown that Foxa2 and its co-activator PPARγ co-activator β promote fatty acid oxidation and stimulation of MTP in liver, resulting in increased VLDL secretion [152, 153].
Liver steatosis, dyslipidaemia and hyperglycaemia may share a common origin
Williams and co-workers have recently proposed a remarkably simple molecular explanation that links fatty liver, dyslipidaemia and hyperglycaemia in type 2 diabetes [154, 155]. They postulated that these features share a common origin, termed pathway-selective insulin resistance and responsiveness (SEIRR). Analysis of a set of 18 insulin targets showed that insulin-stimulated liver from diabetic mice exhibits a specific defect in the ability of the NAD(P)H oxidase 4 (NOX4) to inactivate PTP gene family members; furthermore, hepatic SEIRR could be induced by impairment of NOX4 in cultured hepatocytes [154, 155]. This model needs to be tested in other systems, but it may help to explain how Akt/PKB can regulate lipid-lowering and glucose-lowering pathways that become insulin resistant but also lipogenic pathways that remain insulin responsive.
Lipds induce the innate immune system
Role of inflammation in the development of hepatic insulin resistance
Obesity and insulin-resistant states are associated with low-grade inflammation, which involves not only the adipose tissue but also the liver [156-160]. It has been proposed that this inflammation is critical in the pathogenesis of hepatic insulin resistance and development of hepatic steatosis [156-160]. In particular, activation of Kupffer cells, specialized macrophages in the liver, is an essential part of the pathogenesis of hepatic insulin resistance . The development of insulin resistance in mice susceptible to this is prevented by depleting Kupffer cells . Furthermore, co-culturing hepatocytes with Kupffer cells in vitro leads to significant increases in fatty acid esterification and triglyceride accumulation, an effect that is blocked by tumour necrosis factor (TNF)-α neutralizing antibodies . Likewise, systemic administration of TNF-α in hamsters stimulates assembly and secretion of VLDL1 particles .
These results indicate that Kupffer cell activation is a causal factor in the pathogenesis of hepatic insulin resistance, but what is the underlying mechanism? Emerging results have shown that saturated fatty acids and enteric lipopolysaccharides induce the innate immune system through TLR4-mediated signal transduction in the Kupffer cells and that this is crucial for development of insulin resistance and inflammation [69, 70]. Downstream effects of TLR4 activation are increased expression of kinases JNK, IKK and p38, which impair insulin signal transduction directly through inhibitory phosphorylation of IRS on serine residues [68, 91, 163, 164]. TLR4 activation also leads to increased transcription of pro-inflammatory genes, resulting in elevated levels of cytokines, chemokines, reactive oxygen species and eicosanoids; these changes promote further insulin-desensitization within the target cell itself and in other cells via paracrine and systemic effects [68, 91, 163, 164]. The importance of TLR4 has been illustrated in mice lacking this receptor. These mice are partially protected against high-fat diet-induced insulin resistance , possibly due to reduced inflammatory gene expression in liver and fat [68, 163, 164].
Fetuin-A: the missing link in lipid-induced inflammation and insulin resistance
It is well known that saturated fatty acids can stimulate inflammatory pathways in a variety of cell types, leading to decreased insulin sensitivity [69, 70]. Although these effects are mostly dependent on TLR4, exactly how saturated fatty acids mediate stimulation of inflammatory pathways has remained a subject of debate. It was postulated that saturated fatty acids can act as direct ligands for TLR4 . However, more recent evidence shows that direct interactions between saturated fatty acids and TLR4 are unlikely .
Pal et al.  recently resolved this debate by showing that the liver secretory protein fetuin-A (FetA) acts as an adaptor protein between free fatty acids and TLR4. The authors showed that serum concentrations of FetA are higher in obese humans and mice with diabetes than in healthy humans or wild-type mice. Knockdown of TLR4 or FetA in obese insulin-resistant mice dramatically improves glucose homoeostasis as a result of reduced activation of TLR4-mediated pro-inflammatory signalling cascades. The most convincing evidence for the importance of FetA was provided by the finding that in vivo infusion of saturated fatty acids leads to the expected state of insulin resistance in control mice but not in FetA knockout mice. Furthermore, Pal and co-workers also showed that the fatty acid palmitate directly binds FetA with high affinity and induces strong pro-inflammatory responses in human adipocytes and mouse macrophages, but fails to initiate pro-inflammatory effects in the absence of FetA or in the presence of a TLR4 inhibitor . This elegant study provided many answers, but did not clarify whether the concentration of circulating FetA is correlated to the degree of insulin resistance across individuals. If this is so, FetA may be a suitable risk factor for the development of decreased insulin sensitivity and possibly even type 2 diabetes. This was recently investigated by Sun et al.  and plasma Fet-A levels were indeed found to be independently associated with an increased risk of developing type 2 diabetes.
Obesity is a well-known risk factor for the development of type 2 diabetes and cardiovascular disease. It is noteworthy that obesity is not only associated with lipid accumulation in adipose tissue but also in nonadipose tissue, and the evidence for deleterious effects of lipid accumulation in nonadipose tissue is strong. In skeletal muscle and liver, lipid accumulation is associated with insulin resistance, an early hallmark of developing type 2 diabetes. Lipid accumulation in the heart has been associated with cardiac dysfunction and heart failure. Future research should focus on clarifying the molecular mechanisms involved, and developing strategies to prevent the deleterious effects of ectopic lipid accumulation.
Conflict of interest statement
No conflict of interest was declared.
At the very end of 2011, we received the totally unexpectedly and tragic news that Sven-Olof Olofsson had passed away at home at the age of 64. Sven-Olof had started writing this review and was working on it the day before he died. We have therefore, to the best of our abilities, followed the outline for the text that Sven-Olof proposed, and his thinking has greatly influenced the final manuscript. Sven-Olof was a very bright but modest and softly spoken gentleman with an exceptional personality and lovely sense of humour. We greatly miss him as an outstanding scientist, colleague and mentor, but most of all as a close friend. We also thank Dr. Rosie Perkins for scientific discussions and for editing the text.