Mitochondria-Associated Endoplasmic Reticulum Membrane (MAM) Integrity Is Required for Insulin Signaling and Is Implicated in Hepatic Insulin Resistance

Mitochondria and endoplasmic reticulum (ER) networks are interconnected, sharing structural and functional interactions essential for the maintenance of cellular homeostasis. The contacts between ER and mitochondria, known as mitochondria-associated ER membranes (MAMs), play a pivotal role in calcium (Ca²⁺) signaling, lipid transport, energy metabolism, and cell survival (1). The physical interactions between both organelles depend on complementary membrane proteins, which tether the two organelles together at specific sites. For example, the voltage-dependent anion channel (VDAC) of the outer mitochondrial membrane interacts with the inositol 1,4,5-triphosphate receptor (IP3R) on the ER through the molecular chaperone glucose-regulated protein 75 (Grp75), allowing Ca²⁺ transfer from the ER to mitochondria (2). Recently, mitofusin 2 (Mfn2) was discovered as a direct ER-mitochondria tether, regulating also the interactions and Ca²⁺ transfer between both organelles (3). Now it appears that MAM architecture involves a large number of proteins with various functions (4).

Accumulating evidence suggests that both ER and mitochondria are key actors of energy homeostasis of liver and that hepatic insulin resistance is associated with mitochondrial dysfunction (5), ER stress (6), and altered lipid and Ca²⁺ homeostasis (7). Interestingly, the specific loss of Mfn2 in hepatocytes was recently associated with hepatic insulin resistance (8), whereas overexpression of Mfn2 improved diet-induced hepatic insulin resistance (9). In addition, different proteins of insulin-signaling pathways were recently identified at the MAM interface, such as protein kinase B (PKB) (10), mammalian target of rapamycin complex 2 (11), and phosphatase and tensin homolog (12). However, whether perturbation of organelle interactions participates to hepatic insulin resistance is still unknown.

Until now, the visualization of ER-mitochondria contacts in situ in pathological contexts was limited by the lack of adapted analytical methods. Conventional imaging approaches to assess ER-mitochondria interactions rely on transmission electronic microscopy (13,14), electron tomography (15,16), or immunocolocalization of ER- and mitochondria-specific fluorophores/fluorescent proteins (17). However, these strategies are not particularly sensitive and/or quantitative and not easily amenable to large screening. Alternative methods have emerged using genetic approaches, which may, however, limit analysis of interactions at endogenous expression levels of proteins (18).

We therefore sought to develop and validate a quantitative method to detect and quantify ER-mitochondria interactions by in situ proximity ligation assay (PLA) and to study MAM implication in insulin signaling and their potential contribution to hepatic insulin resistance. Our data demonstrate that MAM are an important signaling hub contributing to insulin action and that their dysfunction in liver is associated with hepatic insulin resistance.

HuH7 cells were cultured in DMEM (1 g/L glucose) supplemented with 10% FBS and antibiotics, at 37°C and 5% CO₂. For lipotoxicity studies, HuH7 cells were treated with BSA or palmitate (200 µmol/L, 18 h) in serum-free culture medium. For mRNA analysis, HuH7 cells were cultured for 6 h in serum-free DMEM (1 g/L glucose) supplemented with forskolin (10 µmol/L) in the presence or absence of insulin (10⁻⁷ mol/L).

Primary mouse hepatocytes were isolated by a collagenase perfusion method as previously described (19) and cultured in DMEM (3 g/L glucose) supplemented with 10% FBS and antibiotics at 37°C and 5% CO₂. Primary human hepatocytes were isolated as previously described (20). All experiments were performed after 24 h of incubation. For the measurement of glucose production, cells were incubated for 24 h in Krebs buffer supplemented in lactate (20 mmol/L), pyruvate (2 mmol/L), and forskolin (10 µmol/L) in the presence or absence of insulin (10⁻⁷ mol/L). For mRNA analysis, hepatocytes were cultured for 6 h in serum-free DMEM (3 g/L glucose) supplemented in forskolin (10 µmol/L) in the presence or absence of insulin (10⁻⁷ mol/L).

Human pcDNA4/TO-Grp75 vector was purchased from CliniSciences. Human pcDNA3-Mfn2 vector was generated from IMAGE clone (pCMV-SPORT6-MFN2, Life Science) by digestion with EcoRI and NotI and subcloning into the pcDNA3 vector (Invitrogen). The constructs used for inducing tethering between mitochondria and ER (AKAP1-FKBP-YFP and CFP-FRB-ER) were a generous gift of Gyorgy Hajnoczky and were used as previously described (16,18). Recombinant adenoviral genome carrying cyclophilin D (Ad-CypD) was generated from a vector encoding rat cyclophilin D (CypD) (21) by homologous recombination in the VmAdcDNA3 plasmid and amplified in HEK293 cells (22).

HuH7 cells were transfected in six-well plates for 48 h with 2 μg expression plasmids for specific target or with the empty vector as control using ExGen 500 transfection reagent (Roche Diagnostics). For silencing experiments, HuH7 cells were transfected with specific siRNA (Qiagen) for VDAC1 (25 nmol/L), Grp75 (50 nmol/L), IP3R1 (50 nmol/L), and Mfn2 (25 nmol/L) using High Perfect transfection reagent (Qiagen) following the technical recommendations.

HuH7 cells or primary hepatocytes were fixed with 4% paraformaldehyde (10 min at room temperature) and permeabilized using 0.1% triton ×100 (15 min at room temperature). Then saturation was performed using PBS-BSA 1% for 1 h, followed by incubation with primary antibodies at 4°C overnight: IP3R1 (Santa Cruz, 1/500, rabbit), Grp75 (Santa Cruz, 1/500, mouse), CypD (Abcam, 1/500, mouse), VDAC1 (Abcam, 1/100, mouse), OXPHOS (MitoSciences, 1/500, mouse), ANT (MitoSciences, 1/500, mouse), and SERCA2 (Cell Signaling, 1/500, rabbit) (Supplementary Fig. 1A). After several washes, the secondary antibodies coupled to a fluorochrome (Alexa Fluor) were added for 1 h. Slides were washed three times in PBS and mounted in VECTASHIELD mounting medium (H-1200).

Duolink II in situ PLA (Olink Bioscience) enables the detection, visualization, and quantification of protein interactions (<40 nm) as an individual fluorescent dot by microscopy. Cell fixation and permeabilization were performed similarly to immunofluorescence. The proximity ligations were performed according to the manufacturer’s protocol. Preparations were mounted in Duolink II mounting medium containing DAPI 18 (Eurogentec). Fluorescence was analyzed with a Zeiss inversed fluorescent microscope equipped with an ApoTome using the AxioVision program. Quantification of signals was done with the BlobFinder software (Centre for Image Analysis, Uppsala University) and expressed as percentage of blobs per nucleus compared with the controls. Experiments were performed at least three times, with a minimum of five fields taken per condition.

Animal studies were performed in accordance with the French guidelines for the care and use of animals and were approved by a regional ethic committee. Male C57BL/6J and ob/ob mice (12 weeks old) were purchased from Harlan and adapted to the environment for 1 week before study. Four-week-old C57BL/6J males were either fed with a standard diet (SD) or with a high-fat and high-sucrose diet (HFHSD; prepared by the Unit of Experimental Food Preparation, Jouy-en-Josas, France) for 16 weeks as previously described (23). One group of HFHSD mice were treated with rosiglitazone (20 mg/kg in the food) during the last 4 weeks of feeding. CypD knockout (KO) male mice were an initial gift from S.J. Korsmeyer’s laboratory (Dana Farber Cancer Institute, Boston, MA) (24), and mice were backcrossed for eight generations on a C57BL/6 genetic background. Wild-type (WT) and CypD-KO mice (24 weeks old) were daily force fed with either methylcellulose or metformin (300 mg/kg/day) for 4 weeks. Glucose, insulin, and pyruvate tolerance tests were performed on 6 h–fasted mice. Glucose (2 mg/g body weight), insulin (0.75 mU/g body weight), or pyruvate (2 mg/g body weight) were injected intraperitoneally, and blood glucose levels were monitored using a glucometer at the indicated time points.

Preparation of total lysates, electrophoresis by SDS-PAGE (10, 12, or 4–15% ready gels, Bio-Rad), immunodetection, and quantification were performed as previously described (25).

Isolation of MAM fractions were performed by differential ultracentrifugation as previously described (26).

Data are expressed as the mean ± SEM. Statistical analysis was performed with the StatView software. Comparisons among more than two groups were analyzed by one-way ANOVA. When a significant F value was obtained, multiple comparisons were made using a Fisher protected least significant difference test. For all other analyses, data were compared by Student t test. Statistical significance was defined as a value of P < 0.05.

We used in situ PLA to detect and quantify interactions between VDAC1 and IP3R1, two organelle-surface proteins involved in the Ca²⁺ channeling complex at the MAM interface. We probed one protein of the outer mitochondrial membrane and one protein of the MAM with mouse anti-VDAC1 and rabbit anti-IP3R1 primary antibody (Supplementary Fig. 1B, panel a) and then with anti-mouse IgG and anti-rabbit IgG that are conjugated to oligonucleotide extensions. If the oligonucleotides are at the requisite distance (<40 nm), they hybridize with the subsequently added connector oligos to form a circular DNA template (Supplementary Fig. 1B, panel b), which is ligated and subsequently amplified to create a ssDNA product (Supplementary Fig. 1B, panel c). Since the distance between the ER and mitochondria at contact sites ranges from 10 to 25 nm, we hypothesized that the ER-mitochondria junction would enable proximity ligation and subsequent detection by hybridization of Texas red-labeled oligonucleotides probes (Supplementary Fig. 1B, panel d). Each fluorescent dot would represent the formation of one VDAC1/IP3R1 interaction, thus allowing quantification of in situ ER-mitochondria interactions in individual cells. Similar in situ PLA experiments were also performed with the Grp75/IP3R1 pair of antibodies. As prerequisites, we validated the primary antibodies by both gene silencing strategy and immunocolocalization experiments in HuH7 cells (Supplementary Fig. 2A and B) and confirmed the colocalization of VDAC1, Grp75, and IP3R1 by conventional immunofluorescence in HuH7 cells (Supplementary Fig. 3), indicating that these proteins are good candidates to study ER-mitochondria interactions. Then we investigated interactions between endogenous VDAC1/IP3R1, VDAC1/Grp75, or Grp75/IP3R1 proteins in fixed HuH7 cells and observed fluorescent signals for all pairs of proteins (Supplementary Fig. 4), indicating that the VDAC1/Grp75/IP3R1 complex can be visualized by in situ PLA. Appropriate controls confirmed the specificity of the assay (Supplementary Fig. 4).

Next we challenged organelle coupling by silencing or overexpressing MAM proteins and quantified these changes by in situ PLA in HuH7 cells. Silencing of VDAC1 (−45%; P < 0.05) (Supplementary Fig. 2A) induced a significant reduction of VDAC1/IP3R1 interactions (−99.5%; P < 0.001) (Fig. 1A). Interestingly, this silencing of VDAC1 also reduced the Grp75/IP3R1 signal counts (−41%; P < 0.001) (Fig. 1A), indicating a reduction of ER-mitochondria interactions. Silencing of Grp75 corroborates these results since we observed that downregulation of Grp75 (−64%; P < 0.05) (Supplementary Fig. 2A) reduced VDAC1/IP3R1 (−47%; P < 0.05) and Grp75/IP3R1 (−57%; P < 0.005) interactions, compared with control cells (Fig. 1B). Furthermore, inhibiting either IP3R with xestospongin C (1 µmol/L, 18 h) or the IP3 production with a PLC inhibitor (U73122, 1 µmol/L, 18 h) also reduced VDAC1/IP3R1 interactions (−22 and −41%, respectively; P < 0.05) (Supplementary Fig. 5A), indicating that pharmacological approach is also an efficient strategy to alter MAM integrity. At the opposite, the induction of Grp75 expression significantly induced the interactions of both VDAC1/IP3R1 and Grp75/IP3R1 proteins (+67 and +70%, respectively; P < 0.05) (Fig. 1C).

Then we modulated organelle coupling independently of the VDAC1/Grp75/IP3R1 complex by modulating either the expression of Mfn2, a direct tether of both organelles whose invalidation reduced ER-mitochondria interactions (3), or by overexpressing an artificial tether (16,18). Reduction of Mfn2 expression (−58%; P < 0.001) (Supplementary Fig. 2A) using a specific siRNA significantly reduced the in situ PLA signal for both VDAC1/IP3R1 and Grp75/IP3R1 (−62 and −66%, respectively; P < 0.05) (Fig. 1D). In contrast, we observed an increase of both VDAC1/IP3R1 and Grp75/IP3R1 interactions (+48 and +64%, respectively; P < 0.05) (Fig. 1E) following the transient overexpression of Mfn2 (+593%; P < 0.003) (Supplementary Fig. 2A). Alternatively, we transfected HuH7 cells with CFP-FRB-ER and OMM-FKBP-YFP constructs (16,18) and then added rapamycin (100 nmol/L) for 2 min to increase ER-mitochondria contacts. Rapamycin treatment significantly increased the VDAC1/IP3R1 signal counts (+56%; P < 0.04) (Fig. 1F), confirming that in situ PLA is a reliable method to investigate MAM integrity and organelle interactions.

We then explored whether genetically induced perturbations of ER-mitochondria interactions could affect insulin action in HuH7 cells. Insulin action was estimated at the level of the signaling cascade by measuring phosphorylation of insulin receptor substrate 2 (IRS2), PKB, and glycogen synthetase kinase 3β (GSK3β), as well as at the level of gene expression by measuring neoglucogenic gene expression, such as the glucose-6-phosphatase (G6P) and the phosphoenolpyruvate carboxykinase (PEPCK). Silencing of Grp75 (Fig. 2A) or Mfn2 (Fig. 2B) reduced insulin-stimulated phosphorylation of IRS2, PKB, and GSK3β and significantly prevented the inhibitory action of insulin of G6P and PEPCK mRNA levels (Fig. 2C). Pharmacological inhibition of IP3R, using xestospongin C, also reduced insulin-stimulated PKB phosphorylation (−41%; P < 0.05) (Supplementary Fig. 3B). On the other hand, overexpression of Grp75 or Mfn2 increased both insulin signaling (Fig. 2D and E, respectively) and action (Fig. 2F). Furthermore, palmitate treatment of HuH7 cells, which altered insulin signaling (Fig. 3A) and action on gene expression (Fig. 3B), was associated with a marked reduction of VDAC1/IP3R1 interaction (Fig. 3C). Importantly, palmitate inhibitory action on insulin signaling could be prevented by Mfn2 or Grp75 overexpression, which increased organelle interactions (Fig. 3D and E, respectively). The opposite modulation of MAM by silencing experiments did not amplify palmitate-induced alterations of insulin signaling (data not shown). Altogether, these results indicate that there is a strong relationship between MAM integrity and insulin action in the human hepatic cell line.

Using conventional Percoll gradient fractionation, we isolated MAM fractions from liver of fasted mice injected or not with insulin. Fraction purity was verified by electronic microscopy (Supplementary Fig. 6). Insulin-stimulated PKB Ser 473 phosphorylation in both liver homogenates and MAM fractions (Supplementary Fig. 7A) suggests that part of phosphorylated PKB localized at the MAM interface in response to insulin. Interestingly, the effect of insulin on PKB phosphorylation is higher in MAM fractions compared with homogenate (Supplementary Fig. 7A, #P < 0.0001), suggesting that MAM could contribute to insulin signaling in hepatocytes. We further demonstrated by in situ PLA that both PKB and p(S473)PKB were in close proximity with IP3R1 (Supplementary Fig. 7B) and that insulin treatment increased both PKB/IP3R1 and p(S473)PKB/IP3R1 interactions without changing VDAC1/IP3R1 interactions (Supplementary Fig. 7B). These data support a recruitment of p(S473)PKB at the MAM interface in response to insulin.

Recently, we identified CypD as a new mitochondrial protein interacting with the VDAC1/Grp75/IP3R1 Ca²⁺-channeling complex at the MAM interface in cardiomyocytes (27). We confirmed that CypD/IP3R1 interactions could be visualized by in situ PLA in primary hepatocytes of WT mice. The absence of CypD/IP3R1 signals in CypD-KO hepatocytes confirmed the specificity of the assay (Supplementary Fig. 8A). We then found a significant reduction of MAM amount in liver of CypD-KO mice compared with WT mice (Fig. 4A), confirming the implication of CypD in MAM integrity also in liver. Interestingly, CypD-KO mice also showed altered responses to insulin (Supplementary Fig. 8B) and pyruvate (Fig. 4B) tolerance tests, indicating a state of insulin resistance and an increase of hepatic neoglucogenesis in this model. Furthermore, CypD-KO mice treated during 4 weeks with metformin displayed improved insulin sensitivity (Supplementary Fig. 8B) and hepatic neoglucogenesis (Fig. 4B) and, concomitantly, an increase in MAM amount in liver (Fig. 4A), indicating association between MAM integrity and hepatic insulin action in vivo in mice. To define this association, we investigated insulin action and VDAC1/IP3R1 interactions in primary hepatocytes of these mice. Alterations of insulin signaling (Fig. 4C) and action, on both gene expression (Fig. 4D) and hepatic glucose production (Fig. 4E), were clearly associated with a reduction of the VDAC1/IP3R1 signal counts (Fig. 4F) in CypD-KO hepatocytes compared with WT hepatocytes. Interestingly, primary hepatocytes prepared from metformin-treated CypD-KO mice had improved insulin action (Fig. 4C–E) and increased MAM interactions (Fig. 4F). We further extended these observations to human primary hepatocytes, in which addition of NIM811 (2 µmol/L, 48 h), a specific inhibitor of CypD, significantly reduced CypD/IP3R1 and VDAC1/IP3R1 interactions (−57 and −34%, respectively; P < 0.05) (Fig. 5A) and decreased insulin-stimulated PKB phosphorylation (−37%; P < 0.05) (Fig. 5B).

Next we measured organelle coupling in liver of genetically or diet-induced insulin-resistant mice. As expected, both ob/ob and 16-week HFHSD mice are glucose intolerant (Supplementary Fig. 9A and C) and insulin resistant (Supplementary Fig. 9B and D) compared with control mice. Furthermore, ob/ob mice showed increased neoglucogenesis during pyruvate tolerance test (Fig. 6A). In addition, the treatment of HFHSD mice with the antidiabetic agent rosiglitazone during the last 4 weeks of feeding improved responses to glucose and insulin tests (Supplementary Fig. 9A and B). Interestingly, the metabolic alterations of both ob/ob and HFHSD mice were associated with a reduction of ER-mitochondria interactions in liver (Figs. 6B and 7A, respectively), whereas rosiglitazone treatment was associated with an increase of MAM (+16%; P < 0.05) (Fig. 7A) in HFHSD mice. Furthermore, when investigating the expression of MAM-associated proteins in the MAM fractions in liver of HFHSD-fed mice, we found marked alteration of key MAM protein (Fig. 7B) when compared with SD mice, with noticeable reduction of VDAC1, CypD, and PACS2 amounts, increased levels of Mfn2, and no change of IP3R1. In addition, basal phosphorylation of PKB was increased in MAM of HFHSD mice compared with SD mice (Fig. 7B). Interestingly, rosiglitazone treatment was able to restore CypD and PACS2 expression, as well as the phosphorylation levels of PKB (Fig. 7B).

We further analyzed insulin action and MAM integrity by in situ PLA in primary hepatocytes of both models of insulin resistance. In agreement with in vivo data, the alterations of insulin signaling and/or action are associated with a significant reduction of VDAC1/IP3R1, Grp75/IP3R1, and/or CypD/IP3R1 interactions in both ob/ob (Fig. 6C–E) and HFHSD (Fig. 7C and D) hepatocytes compared with their respective controls, confirming disruption of MAM integrity in hepatocytes of diabetic mice. However, primary hepatocytes prepared from rosiglitazone-treated mice did not show beneficial effect on insulin-stimulated PKB and GSK3β phosphorylation (Fig. 7C), as well as on protein interactions at the MAM interface (Fig. 7D). These results suggest that the effect of this antidiabetic agent, well measurable in crude liver samples (Fig. 7A), was lost after plating cells in culture.

To further confirm the strong relationship between MAM integrity and insulin action, we sought to restore organelle coupling in hepatocytes of diabetic mice and to determine whether it resulted in enhanced insulin action. As overexpression of MAM proteins increased organelle coupling in HuH7 cells, we reasoned that it would likely also be effective in primary hepatocytes of diabetic mice. Therefore, we chose to overexpress CypD in ob/ob and in HFHSD hepatocytes. Firstly, we controlled that the alterations of organelle interaction and insulin action in ob/ob hepatocytes were conserved after adenovirus infection of the cells (Fig. 8A–C). Then overexpression of CypD increased the VDAC1/IP3R1 interactions compared with ob/ob hepatocytes infected with a green fluorescent protein (GFP)-expressing adenovirus (Fig. 8A), indicating that the strategy was successful in reconstituting structural ER-mitochondria interactions ex vivo. Importantly, CypD overexpression in ob/ob hepatocytes also improved insulin signaling (Fig. 8B) and the metabolic action of the hormone (Fig. 8C). Similar types of results (Fig. 8D–F) were obtained when overexpressing CypD (Supplementary Fig. 10) in HFHSD primary hepatocytes. Altogether, these data demonstrated that improving MAM integrity in primary hepatocytes of diabetic mice is able to restore hepatic insulin action.

In this study, we validated an approach based on in situ PLA to detect and quantify ER-mitochondria interactions and we revealed for the first time that the MAM interface is an important novel player in the regulation of hepatic insulin action and that disruption of MAM could be involved in hepatic insulin resistance.

It has been suggested that ER-mitochondria interactions can be regulated by drug exposure (28) or by disease states (29), but until now, MAM integrity was difficult to access in vivo with conventional imaging approaches. We sought to use in situ PLA to quantify ER-mitochondria interactions in fixed cells, and to our knowledge, this is the first application of in situ PLA to detect organelle interactions. This technique raises several interesting advantages: 1) in situ PLA confers dual-binder specificity for the detection of organelle interactions in situ and reveals proximity of proteins in normal cells without being subject to artifacts of overexpression or ectopic expression (30); 2) because the signal produced by in situ PLA can be amplified, the method is exquisitely sensitive, ensuring that transient and weak interactions can be visualized as a single spot and quantified; and 3) the system is rather simple to use and suitable for multiple condition testing. In this study, we have carefully validated the method to ensure that in situ PLA is suitable to visualize ER-mitochondria coupling. In addition, we found very reproducible results when targeting several pairs of proteins (VDAC/IP3R, Grp75/IP3R, CypD/IP3R) at the MAM interface, strongly supporting the robustness of the technique to monitor organelle coupling. However, the fact that the in situ PLA staining involved almost all the cellular volume of hepatocytes contrasts with previous data reporting that MAM may represent ∼20% of total mitochondrial network (17). This could rely on the fact that images were analyzed by conventional fluorescent microscopy, allowing simultaneous visualization of protein interactions at different focal planes. An important result is that organelle interactions can be modulated by disruption or reinforcement of the interorganellar protein linkage, with different efficiencies depending on the targeted protein, suggesting that both the nature and the amount of tethering proteins directly control ER-mitochondria contacts in agreement with previous studies (3,31–33).

Using this method to evaluate ER-mitochondria interactions, we found that MAM integrity is required for efficient insulin action in hepatocytes. Indeed, we demonstrated in vitro that hepatic insulin signaling and action could be diminished or strengthened by disruption or enforcement of organelle interactions, respectively. This finding was confirmed whatever the targeted proteins used at the MAM interface (IP3R, Grp75, Mfn2, or CypD). Furthermore, we found that palmitate-induced alterations of insulin action in HuH7 cells are associated with a decrease in MAM integrity, while the induction of organelle interactions is able to prevent palmitate-induced insulin resistance. These in vitro data were confirmed and extended in vivo in mice and in mouse and human primary hepatocytes. Indeed, this close association between MAM integrity and hepatic insulin action was observed in three different mice models, including CypD-KO mice, ob/ob mice, and HFHSD-fed mice. Importantly, pharmacological improvement of insulin sensitivity using metformin or rosiglitazone in CypD-KO and HFHSD insulin-resistant mice improved ER-mitochondria interactions. Reciprocally, increasing MAM amount in primary hepatocytes of diabetic mice by overexpressing CypD enhanced insulin signaling and action. Altogether, these data strongly support a close functional relationship between MAM integrity and hepatic insulin response.

Considering this strong association, it is still, however, difficult to conclude that MAM dysregulation is the cause of hepatic insulin resistance in diabetic state or that altered insulin action contributes to MAM defects. Indeed, we found parallel alterations in MAM amount and insulin action in vivo in mice liver and concomitant restoration upon antidiabetic treatments. The fact that the beneficial effects of metformin, but not of rosiglitazone, were conserved in primary hepatocytes prepared from the treated animals suggested the possibility of different regulatory mechanisms of MAM dynamics, depending on the pharmacological treatment. This might be related to the well-known different mechanisms of action of these drugs, with metformin known to act primarily in liver (and potentially directly on mitochondria), while rosiglitazone affects hepatic metabolism indirectly through activation of peroxisome proliferator–activated receptor γ in adipose tissue and modulation of adiponectin production. In line with our data, recent studies demonstrated that loss or mutations of MAM proteins could be associated with impaired glucose homeostasis in mice (8,34) and that liver-specific overexpression of a MAM protein, such as Mfn2, is sufficient to improve diet-induced insulin resistance (9). However, whether this improvement involves a reinforcement of MAM integrity was not evaluated in these works, and thus additional studies are clearly required to demonstrate that an improvement of MAM integrity in vivo could correct hepatic insulin resistance.

In line with recent studies (10,11), we also found that PKB, an important signaling protein mediating the metabolic effects of insulin, is recruited at the MAM interface and phosphorylated on S473 in response to insulin. The fact that interactions between both PKB/IP3R1 and p(S473)-PKB/IP3R1 were found in hepatocytes using in situ PLA strongly supports a true localization of PKB at the MAM interface. Importantly, we further demonstrated that acute insulin treatment increased these interactions without modifying VDAC1/IP3R1 interactions, indicating that insulin is likely able to induce a recruitment of PKB at the MAM interface. As the interactions of both phosphorylated and total PKB protein with IP3R1 were increased in response to insulin treatment, it is presently not clear whether phosphorylation is required for the recruitment of PKB at the MAM interface or whether PKB is directly phosphorylated in MAM. The presence of mammalian target of rapamycin complex 2 at the MAM interface (11) might suggest that PKB could be directly phosphorylated in MAM. These results reinforce the concept that there are different pools of PKB in hepatocytes, and thus MAM might be a novel, not yet considered, insulin-signaling hub in hepatocytes. Therefore, loss of MAM integrity could be directly involved in hepatic insulin resistance. This hypothesis is also supported by the observation that diet-induced diabetic mice were characterized by a reduction of VDAC1, CypD, and PACS2 expression and an increase of PKB phosphorylation at MAM interfaces. These defects were partly restored by rosiglitazone treatment, suggesting, therefore, that the relationship between MAM integrity and hepatic insulin sensitivity could involve the regulation of MAM proteins. In agreement, it was reported that PACS2 (31) and CypD (27) are required for MAM integrity and that high PKB activity is associated with reduced Ca²⁺ flux from the ER through the IP3R (35,36). On the other hand, we found that the expression of Mfn2 is increased at the MAM interface of diabetic mice, whereas loss (8) or overexpression (34) of Mfn2 was, respectively, associated with hepatic insulin resistance or increased insulin sensitivity, suggesting potential adaptive mechanisms.

Interestingly, PKB has been reported to phosphorylate the IP3R channel, reducing its IP3-dependent Ca²⁺ release capacity (35,36). Here we found that inhibition of IP3R activity (by xestospongin C or by U73122) altered MAM integrity. This may indicate a potential link between IP3-dependent ER-calcium release and MAM in liver cells. The fact that diabetic mice were characterized by increased phosphorylated PKB in MAM fraction, and also that insulin phosphorylates PKB in MAM, could suggest that regulation of calcium homeostasis is eventually another process to take into account in the relationship between MAM and insulin resistance. Therefore, further studies are required to prove this hypothesis and to evaluate whether a disruption of Ca²⁺ transfer from ER to mitochondria in conditions of altered MAM integrity could contribute to hepatic insulin resistance.

ER stress (6) is a classical hallmark of hepatic insulin resistance, and some studies suggested an interplay between ER stress and mitochondrial dysfunction in the reduction of insulin responsiveness (37). Our data indicate that MAM dynamics and regulation could contribute to this interplay. This is in line with recent data showing that mitochondria-ER coupling was enhanced during early phase of ER stress (28,38) but also that alteration of these interactions results in a disruption of Ca²⁺ transfer (3) and subsequent ER stress (31). We also observed that modulation of MAM in vitro in HuH7 cells could impact ER stress (Supplementary Fig. 11) and that disruption of MAM in CypD-KO mice was associated with hepatic ER stress (data not shown). We did not investigate in the current study the consequence of MAM modulation on mitochondria function in liver cells. Previous results indicated that mitochondria function could either be altered (28,38) or enhanced (11) by MAM disruption, indicating complex relationships. Similarly, the implication of mitochondrial dysfunction in liver insulin resistance is still a matter of debate (5,39). Therefore, future studies should be designed to define the causal links between MAM integrity, ER stress, and mitochondria dysfunction in relation with insulin resistance. Nevertheless, our study adds new important information regarding the role of the interactions between mitochondria and ER in liver, suggesting that defective ER-mitochondria contacts could contribute to hepatic insulin resistance.

In summary, we validated that in situ PLA offers a reliable method to visualize ER-mitochondria interactions in a native environment, and we demonstrated that MAM should be regarded as a novel hub for insulin-stimulated PKB phosphorylation, its integrity being strongly associated with correct insulin signaling and action in hepatocytes. Furthermore, defective ER-mitochondria coupling is closely associated with impaired hepatic insulin sensitivity both in vivo and in vitro, and increasing organelle coupling improves insulin signaling in primary hepatocytes. Our data clearly indicate, therefore, that the MAM interface and interorganelle contacts are novel actors in the mechanisms of action of insulin in liver and that these interorganelle domains should be taken into consideration in the dysregulations associated with insulin resistance.

Acknowledgments. The authors thank Michel Rivoire and Maud Michelet of the INSERM U1052 (Lyon, France) for the preparation of human primary hepatocytes, Elisabeth Errazuriz for her technical help at the CeCILE Imaging Center (Lyon, France), Gyorgy Hajnoczky of the Thomas Jefferson University (Philadelphia, PA) for the generous gift of ER-mitochondria linkers, and Roman Eliseev of the University of Rochester Medical Center (Rochester, NY) for the generous gift of the expression vector encoding rat CypD.

Funding. This work was supported by INSERM and the National Research Agency (ANR-11-BSV1-033-02 to M.O. and ANR-09-JCJC-0116 to J.R.). E.T. and P.T. were supported by a research fellowship from the French Ministry of Higher Education and Research.

Duality of Interest. This work was supported by Servier Laboratories. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. E.T. and J.R. designed the experiments, researched data, contributed to discussion, and wrote the manuscript. P.T., G.V., N.B., A.B., M.-A.C., J.J.-C., and B.B. researched data. F.Z., M.O., and H.V. contributed to discussion and reviewed/edited the manuscript. J.R. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Some of the data in the article were presented as a poster at the 49th Annual Meeting of the European Association for the Study of Diabetes, Barcelona, Spain, 23–27 September 2013 (abstract published in Diabetologia 2013;56:S283) and as an oral presentation at the 74th Scientific Sessions of the American Diabetes Association, San Francisco, CA, 13–17 June 2014.