Metformin differentially activates ER stress signaling pathways without inducing apoptosis

Metformin is a commonly used drug in the treatment of type 2 diabetes. Metformin suppresses endogenous glucose output and increases peripheral insulin sensitivity (Stumvoll et al., 1995; Fery et al., 1997). Diabetes is often associated with heart disease and worsens the prognosis of patients with heart conditions. Until recently it was unclear whether metformin was a suitable therapeutic in heart patients because unstable or acute congestive heart failure was reported to be associated with an increased risk of lactic acidosis with metformin (Khurana and Malik, 2010). A recent study, however, suggested that metformin is not contraindicated in conditions of heart failure. Results from a recent meta-study show that it might in fact be the most suitable drug to reduce mortality in diabetic patients with heart conditions (Eurich et al., 2007). For example, one study showed that diabetic patients with heart failure have a lower risk of readmission for heart failure if they are treated with metformin than do patients treated without or with other insulin sensitizers (Masoudi et al., 2005). The mechanisms of this metformin effect are unknown, and the question arises: does this cardioprotective effect directly manifest in the cardiomyocyte?

The ER is the major hub for protein sorting and folding in the cell. Protein overload of the ER can lead to activation of the unfolded protein response (UPR). UPR pathways are crucial in the pathogenesis of many human disorders, including neurophysiological diseases, metabolic disorders such as insulin resistance and type 2 diabetes, and cardiovascular diseases such as cardiac hypertrophy, heart failure, atherosclerosis and ischemic heart disease (Araki et al., 2003; Kim et al., 2008; Minamino and Kitakaze, 2010; Thoms et al., 2009).

Three signaling branches have been described whereby unfolded proteins can be sensed and UPR activated (Ron and Walter, 2007; Rutkowski and Hegde, 2010). Protein kinase RNA (PKR)-like ER kinase (PERK) initiates one of these signaling pathways. PERK is an ER-resident transmembrane protein that couples ER stress signals to translation inhibition. PERK phosphorylates the eukaryotic initiation factor 2 α (eIF2α) and itself at its cytoplasmic kinase domain, which leads to reduction of global protein biosynthesis, one of the feedback loops to reduce protein stress in the ER. eIF2α phosphorylation promotes the expression of activating transcription factors (ATFs) such as ATF4 and ATF5. Inositol-requiring enzyme 1 (IRE1) is the ER-bound sensor of another UPR signaling pathway. IRE1 is a conserved transmembrane protein with an ER-luminal sensor for misfolded proteins. Activation of IRE1 initiates the non-conventional splicing of the transcription factor X-box-binding protein 1 (XBP1), which is responsible for the activation of a larger number of ER-stress responsive genes, including chaperones of the ER. The third ER sensor in UPR pathways is the membrane-bound transcription factor ATF6. Upon activation, ATF6 is cleaved and releases its cytoplasmic domain, which enters the nucleus and activates ER stress response element (ERSE)-dependent gene products, including the ER-luminal chaperone binding immunoglobulin protein [Bip; also referred to as 78-kDa glucose-regulated protein (GRP78)]. Unfolded proteins in the ER are not the only trigger for UPR signaling. Energy deprivation, hypoxia, disturbed ER Ca²⁺ levels, pathogens, drugs and secondary metabolites can also induce UPR signaling. Accordingly, the UPR response can be termed endoplasmic reticulum stress signaling (ERSS).

ER stress that is prolonged or severe can lead to apoptosis. Apoptotic cell death of cardiomyocytes is involved in several cardiac disorders. All three ERSS branches have been associated with proapoptotic signals, but, depending on the circumstances and the tissues involved, ERSS can also lead to anti-apoptotic signaling. A downstream component of the pathways leading to apoptosis is the mitochondrial apoptosis machinery, which is able to integrate a number of stressors, including ER stress (Minamino and Kitakaze, 2010). Mitochondrial function and membrane integrity are regulated by the Bcl-2 protein superfamily, which consists of three groups, depending on the number of Bcl-2 homology (BH) domains: the anti-apoptotic members Bcl-2 and Bcl-x_L – containing four BH domains – interact and inhibit proapoptotic Bax and Bak, which contain three BH domains. Alternatively, Bcl-2 members heterodimerize with BH3-only proteins, including Bim and Puma, thereby releasing Bax and Bak, which in turn promote apoptosis. It is believed that CHOP [C/EBP homologous protein; also referred to as GADD153 (growth arrest and DNA damage 153)] mediates ER-stress-dependent apoptosis by downregulating Bcl-2 and elevating oxygen stress (McCullough et al., 2001; Minamino and Kitakaze, 2010).

The effect of metformin on ER stress in cardiomyocytes has not yet been investigated. Therefore, we wanted to study whether metformin is involved in ER stress induction in the heart, and, more specifically, which ERSS pathways are regulated by metformin. Using a rat cardiomyocyte system, we found that metformin differentially induces ER stress pathways but is not associated with apoptotic signaling – in spite of the strong and persistent induction of CHOP. These results might be important in understanding the effect of metformin on the heart, but they also shed light on the possible mode of action of the drug in diabetes.

Freshly isolated and cultivated neonatal rat cardiomyocytes provide a unique cell system to study drug effects on heart cells. To analyze the effect of metformin on cardiomyocytes, we treated freshly isolated neonatal rat cardiomyocytes with 2.5 mM metformin for 48 hours. Metformin significantly increased the mRNA (P<0.0001) and protein levels of CHOP, a prominent marker of several ER-initiated apoptotic signaling pathways (Fig. 1A). Next we tested the concentration and time dependency of CHOP induction (Fig. 1B,C). Starting at a concentration of 2.0 mM metformin, CHOP mRNA levels were significantly increased (P<0.0001; Fig. 1B). CHOP mRNA induction over control levels was significant (P=0.0002) after 24 hours incubation with 2.5 mM metformin (Fig. 1C). After 32 hours incubation, a plateau was reached and longer time of incubation did not result in a further enhancement of CHOP expression (Fig. 1C). CHOP mRNA induction was paralleled by induction of the CHOP protein (Fig. 1D).

To identify possible upstream effectors of CHOP, we analyzed the expression of PERK, which induces CHOP via ATF4. We determined PERK activation by western blotting through its phosphorylation status. We found induction of PERK phosphorylation with a corresponding increase in CHOP protein levels (Fig. 2A). Similar levels of PERK phosphorylation and CHOP induction were found in response to tunicamycin, a mixture of related nucleosides that block N-glycosylation and lead to the induction of ER stress (Fig. 2A). Furthermore, in a timecourse experiment, PERK phosphorylation was found to increase concomitantly with CHOP expression (Fig. 2B, Fig. 1D). The activation of the PERK-ATF4 pathway was further shown by quantitative PCR (qPCR) analysis of the PERK-dependent transcription factor ATF4, which was found to be upregulated by metformin in a time- and concentration-dependent manner (Fig. 2C–E). Similar to CHOP induction, saturation of ATF4 induction was reached after 32 hours in the timecourse experiment (Fig. 2E).

To assess whether CHOP induction was indeed dependent on ATF4, we reduced ATF4 levels by knockdown with siRNA. Treatment with ATF4-specific siRNA led to an 87% reduction of ATF4 expression compared with treatment with control siRNA (Fig. 2F). With 13% ATF4 present, metformin could induce CHOP only by a factor of 1.6 compared with a factor of 4.7 in the experiment with control siRNA (Fig. 2G). We noticed that the reduction in CHOP inducibility correlated with the efficiency of the ATF4 knock down (Fig. 2H). These results show that CHOP induction is dependent on ATF4. In conclusion, we found that metformin activates the PERK-ATF4-CHOP axis in ERSS.

IRE1 catalyzes splicing of XBP1 mRNA to XBP1s mRNA, which is 26 bp shorter than XBP1 and encodes the basic-region leucine zipper (bZIP) transcription factor XBP1s. XPB1 splicing is therefore a readout of the activation of the IRE1 arm in UPR signaling. In rodent full-length XBP1, a PstI restriction site is destroyed during splicing. Treatment with of up to 2.5 mM metformin for 48 hours did not lead to an increase in XBP1 splicing in cardiomyocytes (Fig. 3A, lanes 4–8). By contrast, tunicamycin applied for 6 or 3 hours initiated XBP1 splicing (Fig. 3A, lanes 1 and 2). We quantified the response by expressing XBP1 splicing as the percentage of the spliced form of XBP1s (Fig. 3B). We conclude that the highly conserved IRE1-XBP1 pathway is not induced by metformin.

mRNA expression of ATF6, the stress sensor of the third ERSS pathway, was not increased by metformin (Fig. 3C,D). Expression levels of ATF6 were slightly reduced at lower concentrations or earlier time points only to reach baseline levels at the highest metformin concentration or the latest time points in the timecourse experiment (Fig. 3C,D). Upon activation, ATF6 protein is transported from the ER to the Golgi and cleaved by S1P and/or S2P proteases (Ye et al., 2000). The N-terminal fragment of ATF6 enters the nucleus and acts as a transcription factor for ER chaperones. Thus, ATF6 mRNA levels are not a sufficient indicator of ATF6 activation. To determine whether metformin activates ATF6, we transfected cells with FLAG-ATF6. This construct allows detection of the N-terminal ATF6 domain by western blotting using anti-FLAG antibodies. The control experiment with 2 hours 5 mM DTT treatment indicated strong activation of ATF6 (Fig. 3E). By contrast, metformin could not activate ATF6 after 24 or 48 hours (Fig. 3E).

We then investigated the influence of metformin on the expression of other ERSS-associated factors. The luminal ER chaperone Bip is a mediator of ER stress that is intimately connected to all three branches of ERSS (Bertolotti et al., 2000; Shen et al., 2002). Bip is induced in conditions of heart failure and its induction has been associated with protection against ischemic injuries (Nakagawa et al., 2000; Nickson et al., 2007). Bip tended to be upregulated by metformin, but increases were not statistically significant (Fig. 4A) and, in the timecourse and dose-response experiments, the highest increase corresponded to the latest time points and the highest concentrations of metformin, respectively (Fig. 4B,C). Expression levels were found slightly depressed at 1 mM in the concentration curve (Fig. 4B) or after 8–24 hours metformin treatment (Fig. 4C). ATF3 belongs to the ATF family of bZIP transcription factors and has been implicated in a number a cellular stress responses (Thompson et al., 2009). Metformin, however, does not induce ATF3 expression (Fig. 5A,B).

In summary, our experiments show that metformin differentially induces ERSS in cardiomyocytes, with a marked induction of CHOP and the PERK-ATF4 arm and no induction of the IRE-XBP1 or the ATF6 arm of the UPR. Bip is possibly induced, but ATF3 mRNA was not upregulated. Because CHOP activation is associated with apoptotic signaling, we next tested the effect of metformin on the induction of pro- and anti-apoptotic factors.

The anti-apoptotic gene Bcl-2 is repressed by CHOP, whereas the expression of the pro-apoptotic factor Bim and possibly also the pro-apoptotic and ER-stress-sensitive factor Puma are induced by CHOP (McCullough et al., 2001; Nickson et al., 2007; Puthalakath et al., 2007). We therefore analyzed the effect of metformin on the expression of these factors. Bcl-2 expression was significantly reduced, whereas metformin did not effect the expression of Bim and Puma (Fig. 5C). By contrast, 3 or 6 hours tunicamycin treatment led to upregulation of Bim and Puma (Fig. 5D). This observation stresses the remarkable difference between metformin-and tunicamycin-induced ERSS: although both drugs affect the anti-apoptotic Bcl-2 gene, only the canonical ER stressor tunicamycin affects the pro-apoptotic BH3-only proteins Bim and Puma.

Apoptotic ER stress has been suggested to be transmitted via activation of caspase 12 (Nakagawa et al., 2000), ultimately leading to the activation of caspase 3, the central executioner of apoptotic cell death. Treatment of cardiomyocytes with metformin did not lead to caspase 12 activation (Fig. 6A,B), whereas tunicamycin was a strong activator of caspase 12 (Fig. 6A,B). The apoptosis-inducer staurosporine is used to underline that caspase 12 activation is specific for ERSS-related apoptosis, because the staurosporine effect bypasses caspase 12 (Fig. 6B). Consequently, caspase 3 is also not activated by metformin, whereas staurosporine is a potent activator of caspase 3 (Fig. 6C).

Finally, we showed that the profound CHOP activation in response to metformin does not lead to apoptotic cell death. Terminal deoxynucleotidyl-transferase dUTP nick end labeling (TUNEL) of mock-, tunicamycin- or metformin-treated cardiomyocytes only showed apoptosis in the case of tunicamycin (Fig. 6D). In accordance with this observation, tunicamycin-treated cardiomyocytes assumed the granular appearance of apoptotic cells, whereas metformin-exposed cells maintained their smooth phenotype (Fig. 6E).

Cardiovascular disease is a common comorbidity in diabetic patients and is responsible for 80% of mortality in the diabetic population (Voulgari et al., 2010). In addition to an impairment of coronary vessels, pathophysiological changes in the metabolism of cardiomyocytes themselves are responsible for increased susceptibility of diabetic patients to heart failure leading to a diabetic form of cardiomyopathy and worsening the course of ischemic heart insults (Barsotti et al., 2009; Voulgari et al., 2010). Metformin – the most common non-insulin anti-diabetic drug –has been shown to influence cardiac metabolism and its safety for cardiac disease patients is still under debate (Barsotti et al., 2009; Khurana and Malik, 2010). A large-scale clinical trail by the UK Prospective Diabetes Study (UKPDS) Group suggested that metformin reduces the risk of cardiovascular infarction that is associated with diabetes (UKPDS, 1998; Masoudi et al., 2005) and, in a meta-analysis, metformin has been reported to be the only anti-diabetic drug to be beneficial in patients with diabetes and heart conditions (Eurich et al., 2007).

Experiments conducted in vitro and in vivo have shown a beneficial effect of metformin on ischemic cardiomyocytes (Bhamra et al., 2008; Calvert et al., 2008; Solskov et al., 2008; Paiva et al., 2010). Independently, other researchers have shown that ER stress induces apoptosis in ischemic cardiomyocytes (Okada et al., 2004; Azfer et al., 2006; Szegezdi et al., 2006; Thuerauf et al., 2006). In the light of these results it became necessary to investigate the direct effect of metformin on ER stress pathways in order to gain more insight into the complex action of metformin on cardiomyocytes and to understand the molecular basis of its impact on the heart. We therefore investigated the influence of metformin on the ER stress pathways in cardiomyocytes using isolated neonatal rat cardiomyocytes.

We found that metformin treatment leads to a strong and persistent induction of CHOP. Consistently with CHOP activation, we found a pronounced activation of the PERK pathway (Fig. 7). Interestingly, however, the induction of the PERK-CHOP axis was not associated with increased levels of apoptosis. This is an unexpected finding because inducers of CHOP are known to lead to the induction of apoptosis. For cardiomyocytes, it was shown that CHOP induction by ischemia or mechanical stress results in apoptosis (Szegezdi et al., 2006; Cheng et al., 2009). We therefore investigated possible downstream targets in CHOP-mediated apoptosis in order to investigate their activation status after metformin treatment.

Members of the Bcl-2 family are a possible CHOP target in apoptotic signaling. Pro-apoptotic Bcl-2 family members can trigger apoptosis by mitochondrial membrane permeabilization. For this to occur, the relative proportion of the anti-apoptotic and pro-apoptotic members is essential (Wong and Puthalakath, 2008; Weston and Puthalakath, 2010). Bcl-2 and the pro-apoptotic proteins Bim and Puma are transcriptionally regulated in response to cell stress (Wong and Puthalakath, 2008). The BH3-only family proteins Puma, Noxa, Bid and Bim have been described to mediate apoptosis after ER stress activation (Li et al., 2006; Puthalakath et al., 2007; Upton et al., 2008). A CHOP-dependent upregulation of the BH3-only protein Bim by ER stress, induced by thapsigargin, has been shown for fibroblast cell lines and thymocytes (Puthalakath et al., 2007). Puma has been shown to be responsive to chemical ER stressors and hypoxia and/or reoxygenation in cardiomyocytes (Toth et al., 2006; Nickson et al., 2007). We found that metformin leads to downregulation of Bcl-2 in cultured neonatal rat cardiomyocytes. This might be a response to CHOP induction, because CHOP has been shown to decrease Bcl-2 expression in HeLa cells and fibroblasts (McCullough et al., 2001). By contrast, metformin failed to upregulate Bim, providing a possible explanation for the absence of apoptosis. Similarly, we observed that the pro-apoptotic BH3-only protein Puma was upregulated by tunicamycin-induced ER stress but not by metformin (Fig. 7).

The results of a recent publication support our finding that CHOP induction does not necessarily result in the onset of apoptosis. Neuronal cultures from CHOP^−/− mice were found to be less sensitive to thapsigargin, and overexpression of CHOP protects neuronal cells against hypoxic injury (Halterman et al., 2010). Together with our results, these data challenge the common notion of CHOP as a pro-apoptotic factor.

Caspase 3 is the central executioner protein of apoptosis. The role of caspase 12 in ER-dependent apoptosis is still under debate, because functional caspase 12 has not been found in humans (Fischer et al., 2002). In neonatal rat cardiomyocytes, however, caspase 12 is activated when CHOP is induced by tunicamycin or thapsigargin (Szegezdi et al., 2006; Thuerauf et al., 2006). If CHOP is induced by hypoxia, some authors find caspase 12 activation (Terai et al., 2005; Szegezdi et al., 2006), whereas other authors do not find an involvement of caspase 12 (Thuerauf et al., 2006). In addition, knock down of caspase 12 has been shown to reduce hypoxia-induced apoptosis (Terai et al., 2005). It can be concluded that caspase 12 is involved in ER-stress-induced apoptosis in rodent cardiomyocytes. It is possible that caspase 12 is representative of another pathway or amplifies the Bcl-2 pathway. Our experiments show that metformin does not activate caspase 3 or caspase 12 and, accordingly, does not induce apoptosis in rat cardiomyocytes (Fig. 7). It has recently been suggested that metformin might exert its cardioprotective effect by enhancing the integrity of the mitochondrial membrane (Bhamra et al., 2008).

Beyond those proteins that are involved in specific ERSS pathways, the ER luminal chaperone Bip is one of the most general markers for a general induction of ER stress. Bip is induced in conditions of heart failure and its induction has been associated with protection against ischemic injuries and cardioprotection in general (Terai et al., 2005; Shintani-Ishida et al., 2006; Thuerauf et al., 2006; Groenendyk et al., 2010). Part of the Bip pool is found on the cell surface and reported to protect cells thereby (Hardy and Raiter, 2010). We found an approximate twofold induction of Bip upon metformin treatment, supporting the hypothesis that metformin induces ER stress in a manner that promotes the survival of cardiomyocytes.

In our experiments, Bip activation was, however, weak in comparison to classical activators of ER stress. With respect to the downstream effects of PERK-ATF4 activation, this might indicate that we have to look for another protective ER stress factor to explain the cardioprotective role of metformin. With respect to the upstream factors of the PERK-ATF4 axis, this might explain the specific activation of only one of the three ER stress pathways, if the ER stress sensors respond to Bip recruitment with different threshold levels. The late induction of PERK by metformin is consistent with the idea that PERK is not directly activated by metformin. We speculate that the effect of metformin effect on the energy status of the cell causes the PERK-specific induction of UPR signaling.

Our data show that metformin triggers the UPR without inducing apoptosis. We speculate that metformin-dependent UPR induction capitalizes on the beneficial sites of ERSS, such as chaperone induction, but keeps the stress response below a threshold level that would induce apoptosis. CHOP induction alone is not sufficient for induction of apoptosis in cardiomyocytes and might require the induction of secondary pathways. It will now be interesting to understand how metformin leads to the induction of ERSS and whether activation of the well-known metformin target, AMPK, is involved in this process. The model presented in this study can help to understand which parameters trigger UPR-mediated protection and under which circumstances UPR contributes to apoptosis and tissue damage.

For preparation of cardiomyocyte cultures, neonatal Wistar rats were sacrificed on day 1. Hearts were excised quickly, rinsed in PBS and cut into small pieces. Subsequently, tissue pulp was dissolved in a solution of 0.1% collagenase B (Cell Systems) with 0.2% trypsin (Biochrom) in PBS (Invitrogen) at 37°C. Cells were collected by centrifugation at 300 g for 5 minutes and resuspended in DMEM:F12 medium (Invitrogen) containing 10% FCS (Biochrom) with 1% penicillin/streptomycin (Invitrogen). Cells were plated for 90 minutes in 87-mm-diameter culture dishes at 37°C and 5% CO₂ to enrich the cardiomyocyte fraction and to remove fibroblasts. Under these conditions, fibroblasts adhere preferentially to the bottom of the culture dish. The non-adherent cells in the supernatant were collected by centrifugation as described above and plated on gelatine-coated (Sigma) 36-mm culture dishes (Nunc) at a density of approximately 100,000 cells per cm². At 1 day after isolation, the medium was renewed. At day 2 the cells formed a confluent spontaneously beating monolayer. After 48 hours, the medium was changed to the experimental medium containing 33% DMEM:F12 medium and 67% DMEM (no glucose, no sodium pyruvate; Invitrogen), and 1% FCS, 100 μM 5-bromo-2-deoxyuridine (Sigma) and 1% penicillin/streptomycin. All procedures involving experimental animals were performed in compliance with local animal welfare laws, guidelines and policies.

Where indicated, metformin (Sigma; stock solution: 300 mM in water), tunicamycin (Sigma; stock solution: 10 mg/ml in DMSO, final concentration 100 ng/μl) and staurosporine (Sigma; stock solution 1 mM in DMSO) were added to the medium as specified in the Results section.

Knock down of ATF4 was carried out at day 2 by transfection with siRNA. A pool of four different siRNAs (ON-TARGETplus SMARTpool Rat ATF4; Thermo Scientific Dharmacon) directed against rat ATF4 mRNA was used at 50 nM final concentration. A total of 6 μl DharmaFECT 4 transfection reagent (Thermo Scientific Dharmacon) were used per 36-mm culture dish. In each experiment, control cultures were performed using non-targeting siRNA (ON-TARGETplus non-targeting siRNA#1; Thermo Scientific Dharmacon). Cells were incubated with the transfection medium (DMEM:F12, 10% FCS, 100 μM 5-bromo-2-deoxyuridine, without penicillin/streptomycin). After 48 hours, transfection medium was changed to the experimental medium with or without metformin for another 24 hours.

In order to assay ATF6 processing, cells were transfected at day 2 with the p3×FLAG-ATF6 plasmid (Addgene plasmid 11975) described previously (Chen et al., 2002). Effectene Transfection Reagent (Qiagen) was used for transfection. 0.4 μg plasmid DNA dissolved in 100 μl EC buffer containing 3.2 μl Enhancer and 4 μl Effectene were used per 36-mm culture dish. Cells were incubated with transfection medium (DMEM:F12, 10% FCS, 100 μM 5-bromo-2-deoxyuridine, without penicillin/streptomycin). After 24 hours the transfection medium was changed to the experimental medium for a further 48 hours and metformin or dithiothreitol (Invitrogen) were added at the indicated time points. Transfection efficiency was determined by double immunofluorescence staining of the transfected cells to visualize the FLAG-taq and cardiomyocyte-specific desmin. Transfection efficiency was approximately 5% (p3×FLAG-ATF6 plasmid). More than 80% of the FLAG-taq-positive cells were also stained for desmin, demonstrating that cardiac fibroblasts (ca. 20% of the cultured cells) were not preferentially transfected.

RNA was extracted from samples using Trizol reagent (PEQLAB) according to the instructions of the manufacturer. The RNA-containing phase was purified using RNeasy kit columns (Qiagen) following the manufacturer’s protocol. RNA samples were stored at −80°C.

1 μg total RNA was transcribed into cDNA using random hexamer primers and Superscript II reverse transcriptase (Invitrogen) according to the manufacturer’s protocol in an i-Cycler thermocycler (Bio-Rad Laboratories). mRNA expression was determined using SYBR Green (Bio-Rad Laboratories). All measurements were performed as duplicates. For each gene, primer pairs spanning at least one intron (MWG Biotech and Metabion) were used in a final concentration of 0.2 μM (Table 1). Cycling conditions were as follows: in an initial step, Taq polymerase was activated at a temperature of 95°C for 15 minutes. Subsequently, amplification cycle conditions were 94°C for 30 seconds, 58°C for 30 seconds and 72°C for 30 seconds. Amplification cycles were repeated 40 times. After each run a melting curve analysis was performed to ensure amplification of the correct products. The expression of adiponectin receptor 1 (adipor1) mRNA was used to normalize the expression of the respective mRNA. Results were presented as percentage of change compared with the control group, the value of which was set arbitrarily at 100%. Statistical analyses were performed by comparing the normalized Ct-values by a paired t-test or one-way ANOVA using Prism software (Graph Pad). Error bars indicate the standard error of the mean (s.e.m.). CHOP inducibility (I; Fig. 2H) was calculated as follows: I=(2^ΔΔCt [CHOP_{ATF4 siRNA with metformin}]/2^ΔΔCt [CHOP_{ATF4 siRNA}])/(2^ΔΔCt [CHOP_{contr. siRNA with metformin}]/2^ΔΔCt [CHOP_{contr. siRNA}]) and plotted against ATF4 knockdown efficiency (2^ΔΔCt [ATF_{contr. siRNA}]/2^ΔΔCt [ATF4_{ATF4. siRNA}]).

For western blotting analysis, proteins were extracted from cultured cardiomyocytes using Trizol reagent. Two volumes of isopropanol were added to the phenol phase after removal of the aqueous phase containing the RNA and the interphase containing the DNA. After incubation at room temperature for 10 minutes, proteins were sedimented by centrifugation (10,000 g, 10 minutes, 4°C). The supernatant was discarded and the protein pellet was washed three times with ethanol containing 0.3 mM guanidine hydrochloride (20 minute incubation at room temperature, 20 minute centrifugation at 10,000 g, 4°C). After a further wash step with ethanol the protein pellet was dried in a vacuum centrifuge. The protein pellet was dissolved in 1% SDS containing Complete protease inhibitors and Phosstop phosphatase inhibitors (Roche) by incubation at 75°C for 30 minutes under agitation. Protein concentration was determined using the BCA Protein Assay Kit (Thermo).

A total of 50 μg cellular protein extract per lane was subjected to SDS-PAGE (10–14% polyacrylamide; Roth). Proteins were transferred to nitrocellulose membrane (Schleicher & Schuell) and the membrane was blocked in TBS buffer containing 5% BSA for 2 hours or, in the case of caspase 12, over night. Proteins were identified by incubation at 4°C over night, or, in the case of caspase 12, at room temperature for 2 hours, in TBS buffer containing 1% Tween and 5% BSA together with the specific antibody. Antibody dilutions were: monoclonal mouse anti-CHOP, clone L63F7 (Cell Signaling), 1:1000; monoclonal rabbit anti-PERK, clone C33E10 (Cell Signaling), 1:1000; monoclonal rabbit anti-phospho(Thr980)-PERK, clone 16F8 (Cell Signaling), 1:1000; polyclonal rabbit anti-caspase-12 (Biovision), 1:1000; monoclonal rabbit anti-GAPDH, clone 14C10 (Cell Signaling), 1:4000; and anti-FLAG tag M2 (Stratagene), 1:1000. HRP-conjugated secondary antibodies were used in TBS buffer containing 1% Tween, 5% BSA for 2 hours at room temperature: goat anti-mouse Ig (Pierce), 1:2000, and goat anti-rabbit Ig (BD Pharmingen), 1:2000. Chemiluminescence was detected using Super Signal West Femto reagent (Thermo) on an X-ray film (Fuji).

Isolated neonatal rat cardiomyocytes were plated on a glass slide coated with laminin (Sigma). Staining procedure was performed using TUNEL Enzyme and TUNEL Label Mix (Roche) according to the manufacturer’s instructions.

We are grateful to Ron Prywes for providing p3×FLAG-ATF6 through Addgene. We thank Noa Lipstein, Elke Oetjen and Ralph Krätzner for comments on the manuscript.