Inhibition of Foxo1 Protects Pancreatic Islet β-Cells Against Fatty Acid and Endoplasmic Reticulum Stress–Induced Apoptosis

Thapsigargin, tunicamycin, oleic acid, palmitic acid, formalin, 4′6-diamidino-2-phenylindole (DAPI), Hoechst dye, and propidium iodide were purchased from Sigma (Saint Louis, MO). SP610025 was from Tocris Bioscience (Ellisville, MO). Antibodies used were anti–α-tubulin (monoclonal; Sigma), anti-BiP, anti-Gadd153/CHOP (Santa Cruz Biotechnology, Santa Cruz, CA), anti–phospho-eukaryotic initiation factor-2α (eIF2α), anti–phospho-JNK1/2, anti–total JNK1/2, anti–phospho-c-Jun, anti–cleaved caspase 3, anti-Foxo1, anti–phospho-Ser473-Akt, anti–phospho-Thr308-Akt (Cell Signaling Technology, Beverley, MA). The anti-hemaglutinin antibody is a mouse monoclonal from Covance Research Products (Cumberland, VA). All primary antibodies were used at a 1:1,000 dilution in 5% BSA/Tris-buffered saline with Tween/0.1% NaN₃, with the exception of the CHOP antibody, which was used at a 1:200 dilution in 5% milk/Tris-buffered saline with Tween.

MIN6 cells were maintained in Dulbecco's modified Eagle's medium (DMEM) containing 25 mmol/l glucose, with 15% fetal bovine serum (FBS), 100 units/ml penicillin, 100 μg/ml streptomycin, 100 μg/ml l-glutamine, and 5 μl/l β-mercaptoethanol in humidified 5% CO₂, 95% air at 37°C (19). MIN6 cells used were between passages 21 and 31.

All procedures were performed in accordance with Washington University's animal studies committee. Animals were killed in a carbon dioxide chamber. Islets of 8- to 10-week-old wild-type male C57BL/6 mice were isolated by ductal collagenase distension/digestion of the pancreas (20), followed by filtering and washing through a 70-μmol/l nylon cell strainer (BD Biosciences, San Jose, CA). Isolated islets were then maintained in RPMI medium containing 11 mmol/l glucose, 10% FBS, 200 units/ml penicillin, and 200 μg/ml streptomycin in humidified 5% CO₂, 95% air at 37°C.

MIN6 cells were incubated in modified DMEM media with 0.5% (wt/vol) BSA alone or 0.2, 0.3, or 0.4 mmol/l palmitate or 0.4 mmol/l oleate complexed to 0.5% (wt/vol) BSA for 24 h. Preparation of the 0.4 mmol/l fatty acid media was carried out as previously described (8). Briefly, a 20 mmol/l solution of the fatty acid in 0.01 mol/l NaOH was incubated at 70°C for 30 min. Then, 330 μl of 30% BSA and 200, 300, or 400 μl of the free fatty acid/NaOH mixture was mixed together and filter-sterilized with 20 ml of either the DMEM or RPMI media. These conditions were chosen because we originally showed that 25 versus 5 mmol/l glucose inhibited Foxo1 activation (17) and was protective of apoptosis (74). In the current studies, we chose standard culture media (25 mmol/l) to begin the experiments with Foxo1 activity and apoptosis at a minimum to optimize the effects of FFA. The approximate molar ratio of fatty acids to BSA is 6:1 with 0.4 mmol/l palmitate. The addition of BSA or a fatty acid/BSA mixture has not been shown to affect the pH of the media. For transferase-mediated dUTP nick-end labeling (TUNEL) staining, cells were kept in culture in Lab-Tek II, CC²-administered chamber slides (Nunc, Rochester, NY).

MIN6 cells were grown on glass coverslips within the wells of a 6-well plate and incubated with either normal media, thapsigargin, BSA alone, or 0.2, 0.3, or 0.4 mmol/l palmitate complexed with BSA for 24 h. For the last hour of incubation, 10 μg/ml of propidium iodide and 20 μg/ml DAPI were added directly to the media. After this incubation, the MIN6 cells were washed 3× with PBS and fixed with 3.7% formalin for 15 min at 4°C. After fixation, the MIN6 cells were washed again 3 times with PBS and then mounted with antifading gel mounting medium (Biomeda, Foster City, CA) onto glass slides. Each condition reported represents >600 cells counted by randomized field selection. The percentage of cell-death is reported as the number of propidium iodide–stained nuclei over the total number of nuclei stained by DAPI, as quantitated by ImageJ version 1.3.8s (National Institutes of Health) (21). TUNEL staining was performed using a rhodamine ApopTAG kit (Amersham BioSciences, Piscataway, NJ) according to the manufacturer's instructions.

MIN6 cells were plated on 10-cm dishes and allowed to grow to 60–70% confluence. Media was then exchanged for 0.5% BSA or 0.4 mmol/l palmitate/0.5% BSA or 0.4 mmol/l oleate/0.5% BSA media for the times indicated. Cells were trypsinized and fractionated according to a previously published protocol evaluating Foxo1 localization (18).

Protein was extracted with a cell lysis buffer (diluted from 10× cell lysis buffer from Cell Signaling Technology and an additional protease cocktail tablet from Roche at one tablet/10 ml final buffer volume). Protein samples (30 μg) were separated by SDS-electrophoresis through either 4–15% gradient or 15% polyacrylamide gels (BioRad, Hercules, CA) and transferred to nitrocellulose membranes, followed by immunoblotting using all primary antibodies according to the manufacturer's instructions. Immunodetection was developed with ECL Advance (Amersham Biosciences, Buckinghamshire, U.K.) and imaged with a charge-coupled device camera (Alpha Innotech, San Leandro, CA).

Hemaglutinin-tagged Δ256-Foxo1 in the pCMV5 vector (DNFoxo) was a gift from D. Accili (Columbia University, New York). The IGF-binding protein-1 (IGFBP-1) promoter/luciferase gene construct (p925GL3) was a gift from M. Rechler (National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, Bethesda, MD). The CHOP promoter/luciferase construct (pGL2/3) was a gift from the lab of D. Ron (Skirball Institute, New York University, New York). The pRL-TK control vector contains the thymidine kinase promoter of the herpes simplex virus upstream of the Renilla luciferase (Promega, Madison, WI).

MIN6 cells were plated in 12-well plates 2 days before transfection. At ∼60–70% confluence, each well was transfected with 100 ng IGFBP-1/luciferase or CHOP/luciferase plasmid, 20 ng pRL-TK control vector, and either 100 ng of the pCMV5-Δ256-Foxo1 or an empty pCMV5 vector in 2 μl lipofectamine 2000 (Invitrogen, Carlsbad, CA) in 100 μl of OptiMem. For cell lysis, 200 μl of passive lysis buffer (Promega) was used. The firefly and Renilla luciferase activities were measured after the indicated hours of incubation with BSA or 0.4 mmol/l palmitate in a Monolight 3010 luminometer (BD Biosciences) using the dual-luciferase reporter assay system (Promega).

The hemaglutinin-tagged dominant-negative Foxo1 adenovirus (AdV-Δ256-Foxo1) and the hemaglutinin-tagged wild-type Foxo1 adenovirus were generous gifts from D. Accili (22). The green fluorescent protein (GFP) adenovirus was a gift from D. Kelly (Washington University, St. Louis, MO). Infection of the MIN6 cells was carried out at the indicated multiplicity of infection (MOI) for 1 h in serum-free media. The MIN6 cells were then washed in PBS and maintained in the DMEM/15% FBS media, and then experiments were carried out 24 h after infection.

The truncated and hemaglutinin-tagged dominant-negative Foxo1 allele (DNFoxo) (23) was a generous gift from D. Accili (Columbia University, New York, NY). It was inserted into a RIP-I/β-globin expression vector (24,25), sequenced, and microinjected into fertilized eggs of C57BL/6 × CBA mice according to standard protocol of the Mouse Genetics Core of the Washington University School of Medicine. Nine founders were obtained, of which five passed the transgene through germline transmission, as evidenced by genotyping and backcrossing to C57BL/6J mice (The Jackson Laboratory). Of the five that transmitted, one line maintained consistent and strong hemaglutinin staining within pancreatic islets, designated as RIP-DNFoxo. Transgenic and nontransgenic littermate male (designated as wild type) mice from C57BL/6J-backcrossed F4-F7 generations of this one line, known as RIP-DNFoxo, were used in all experiments in accordance with Washington University's animal studies committee. These mice are currently being phenotyped.

Primary islets from wild-type or RIP-DNFoxo mice were pooled (three per genotype) and divided to be incubated in modified RPMI media with 0.5% (wt/vol) BSA alone or with 0.4 mmol/l palmitate complexed to 0.5% (wt/vol) BSA for 24 h. Preparation of the 0.4 mmol/l free fatty acid media was carried out as previously described (8). Briefly, a 20 mmol/l solution of palmitic acid (Sigma) in 0.01 mol/l NaOH was incubated at 70°C for 30 min. Then, 330 μl of 30% BSA and 400 μl of the palmitic acid/NaOH mixture was mixed together and filter-sterilized with 20 ml of the RPMI media. The approximate molar ratio of fatty acids to BSA is 6:1 with 0.4 mmol/l palmitate. MIN6 cells were incubated in modified DMEM media with 0.5% (wt/vol) BSA alone or with 0.4 mmol/l palmitate plus 0.5% (wt/vol) BSA for 24 h.

After isolation and a 6-h recovery period after isolation in RPMI/10% FBS media, primary islets from wild-type or RIP-DNFoxo mice were pooled and divided into 12-well plates with 1 ml of RPMI/10% FBS media to be administered with either 10 μmol/l thapsigargin (10 μl of a 1 mmol/l stock; Sigma) or vehicle (10 μl of ethanol) alone for 48 h.

Western blot quantitation consisted of acquiring a per-lane ratio of the chemiluminescent signal intensities from phospho-Ser (256)-Foxo1 or Gadd153 to its respective α-tubulin using ImageJ (21) and then normalizing each signal to the first condition of the starved state in each respective blot. The Western blot figures indicate concentrations for which there is a significant difference (P < 0.05) from the starved state and a significant difference (P < 0.05) from the intensity of the glucose phosphorylation of Foxo1. For the luciferase assays, ratios of luciferase to Renilla were generated with standard deviations and error. Error was propagated for fold calculations between different conditions. Student's two-tailed t test for independent samples was used in significance calculations. The P value for significance is mentioned in the figure legends.

MIN6 cells infected with either a GFP or DNFoxo adenovirus were maintained in triplicate samples within 6-well plates in DMEM with either 0.5% BSA or 0.5% BSA plus 0.4 mmol/l palmitate for 24 h. Each well was then aspirated and rinsed once with PBS. Then, 1 ml of ice-cold Trizol (Invitrogen) was added to each well. The protocol was followed according to manufacturer's instructions until the aqueous phase was isolated. RNase-free 70% ethanol was added (twice the aqueous volume), and the mixture was run through a purification column within the RNeasy mini-kit (Qiagen, Valencia, CA). Purification of total RNA was completed using the kit and eluted with RNase-free water. RNA quality was assessed with gel electrophoresis and an Agilent bioanalyzer (Agilent Technologies, Santa Clara, CA).

We used a MouseRef-8 Expression BeadChip (Illumina, San Diego, CA), which allows for the processing of eight samples to be analyzed simultaneously for the quantitation of absolute expression of up to 16,435 genes and controls with a 30-fold redundancy of 50-mer oligos per gene. For control samples, four independent RNA preparations from the AdV-GFP BSA were pooled together, as were four independent RNA samples from the AdV-DNFoxo–infected samples administered with BSA and without palmitate. These two control groups occupied two arrays. For the experimental condition, administered with 0.4 mmol/l palmitate, three independent RNA samples for each of the GFP and DNFoxo virally infected samples occupied the other six gene expression arrays on the chip. For each gene, the Illumina software calculated the P value of the reliability of the spot. Only spots that had a significant P value were kept for further analysis. For each gene, the per-chip variance was integrated across the 30 repeats per gene on each array and the array variance. For each ratio (GFP Palm–to–BSA, DNFoxo Palm–to–BSA), 95% confidence intervals were calculated taking into account internal repeats (∼30) on multiple gene arrays. The cutoff fold change was arbitrarily chosen at 2.0: genes up- or downregulated more than twofold and with a 95% confidence interval not overlapping 0 (in Log₁₀ scale) were considered for further analysis. Annotation for the microarray platform was collected through the SOURCE repository (26). All transcripts significantly regulated by fatty acids were then clustered according to their Gene Ontology (27) functions using Genesis 1.7.2 (28).

The normalized unprocessed data are currently available online at http://drtc.im.wustl.edu/files/apermutt/microarray/FA/normalized_data.csv and will be made available in the MIAME (Minimum Information About a Microarray Experiment) standard on the NCBI (National Center for Biotechnology Information) Gene Expression Omnibus http://www.ncbi.nlm.nih.gov/geo/.

The microarray data were validated by real-time PCR on selected genes. cDNA was generated (Superscript III; Invitrogen) and amplified by real-time PCR (Power SYBR Green Master Mix; Applied Biosystems) on an ABI 7500 real-time PCR system (Applied Biosystems). Primers are shown in Supplemental Table S2 (available in an online appendix at http://dx.doi.org/10.2337/db07-0595). Results were normalized to the acidic ribosomal protein 36B4 and quantified using the comparative Ct method (29).

For each gene analyzed, 2,500 bp were collected from the NCBI Entrez databases (http://www.ncbi.nlm.nih.gov/) using EZ-Retrieve (http://siriusb.umdnj.edu:18080/EZRetrieve/). Perl global regular expression patterns were then used in BBEdit 8.6.1 (Bare Bones Software, Bedford, MA) to find all consensus sequences or their reverse complements in the sequences.

Glucose-responsive MIN6 insulinoma cells were assessed for apoptosis at 24 h after administration with a saturated fatty acid, palmitate (16:0), or a monounsaturated fatty acid oleate (18:1). Effects of palmitate versus oleate were compared because of the known different potencies of saturated versus monounsaturated fatty acids in inducing cell apoptosis (6,30,31) and in contributing to symptoms of diabetes and cardiovascular disease (32). Cells were incubated at a constant 0.5% BSA (wt/vol) and at concentrations of palmitate or oleate of 0.2, 0.3, and 0.4 mmol/l. Apoptosis, characterized by propidium iodide staining, normalized with DAPI staining, was visibly enhanced with increasing concentrations of fatty acids, as illustrated with palmitate in Fig. 1A. A dose response was observed in cell death rates reaching 11.6 and 3.4%, respectively, for administration with 0.4 mmol/l palmitate and oleate (Fig. 1B). That apoptosis contributes at least partly to palmitate-induced cell death in MIN6 insulinoma cells was evidenced by increased TUNEL staining (Supplemental Fig. S1).

Several recent studies in peripheral insulin target tissues have documented the relationship between fatty acid administration and the activation of the stress-activated kinases known as JNK1/2 in conjunction with decreased insulin signaling (10,11,15). We confirmed these observations using a 24-h administration of MIN6 cells with palmitate and oleate. As fatty acid concentration was increased, an apparent dose-response increase in JNK1/2 activation was observed, as evidenced by the two bands representing increased JNK1/2 phosphorylation and concomitant phosphorylation of its substrate target, c-Jun (Fig. 2A). This treatment did not affect the total amount of the JNK proteins. The effects on JNK activation appeared to be greater for palmitate than for oleate. A decrease in Akt phosphorylation, most evident with 0.4 mmol/l palmitate or oleate, was noted (Fig. 2B), consistent with previous findings (8).

Pdx1 is a transcription factor known to be involved in the regulation of insulin production and protection of pancreatic islet β-cells from apoptosis (33). A recent report observed that palmitate administration of rat islets resulted in decreased insulin gene expression that was associated with a nuclear to cytoplasmic translocation of Pdx1 without a change in total Pdx1 protein (34). In our studies a dose-dependent decrease in Pdx1 protein expression was observed with fatty acid administration (Fig. 2C). This was associated with dose-dependent increases in expression of the proapoptotic transcription factor CHOP. Cleaved and activated caspase 3 also increased, consistent with its role as a major participant in the apoptotic cell death pathway.

To define the sequence of molecular events that occur in β-cells on fatty acid exposure, MIN6 cells were incubated with either 0.4 mmol/l palmitate or oleate and analyzed over an 8-h period (Fig. 3A). JNK activation, as evidenced by phosphorylation of what appeared to be three isoforms, was noted as early as 1 h after administration with either fatty acid. The lowest phosphorylated JNK band, possibly JNK3, which has been shown to be expressed primarily in neurons (35), demonstrated transient expression. Activated JNK1/2 persisted over the 8-h incubation, as further evidenced by phosphorylation of the JNK substrate c-Jun.

The temporal relationship between fatty acid–induced ER stress and JNK activation was evaluated next. Whereas fatty acid administration resulted in insignificant changes in BiP (HSPA5, GPR78) protein, a major chaperone of the ER lumen (36), there was a progressive increase in phosphorylation of eIF2α, a protein that attenuates translation and initiates a program of apoptosis with continued stress exposure (37). This was most apparent at 4 h (Fig. 3B) and was followed by increased expression of CHOP and cleaved caspase 3. These data showed that JNK activation preceded induction of ER stress markers that in turn preceded induction of apoptotic markers.

To confirm the role of ER stress in fatty acid–induced apoptosis, cells were examined 24 h after palmitate administration in the presence and absence of the ER stress inhibitor tauroursodeoxycholic acid (TUDCA), a chemical chaperone (38). Addition of TUDCA reduced the levels of cleaved caspase 3, phospho-eIF2α, and CHOP protein expression by Western analysis (Fig. 3C), although incomplete inhibition suggests the possibility of other mechanisms contributing to fatty acid–induced cell death. TUDCA is not a selective inhibitor of ER stress and also inhibits Bax translocation and mitochondrial stress.

To examine the relationship between fatty acid administration, ER stress, and Foxo1 activity, MIN6 cells were administered with palmitate or oleate for 4 h, and the intracellular localization of Foxo1 was examined. As shown, fatty acids induced a shift of Foxo1 from the cytoplasm to the nucleus (Fig. 4A ). Consistent with the increased effect of palmitate on apoptosis and intracellular signaling, palmitate administration induced more Foxo1 translocation than oleate administration. Using an IGFBP-1 promoter-luciferase construct containing a Foxo1 binding site (39), palmitate administration resulted in a sustained increase in endogenous Foxo1 transcriptional activity from 4 to 24 h of incubation (Fig. 4B). Next, the relationship between ER stress and Foxo1 transcriptional activity in insulinoma cells was assessed. We previously demonstrated that the ER stress agent thapsigargin reduced Akt phosphorylation (9). As shown in Fig. 4C, the addition of thapsigargin (1 μmol/l) to MIN6 cells resulted in significant cytosolic to nuclear translocation of GFP-tagged Foxo1 within 1 h of administration that was sustained up to 18 h. Using an IGFBP-1 promoter-luciferase construct containing a Foxo1 binding site (39), we assessed Foxo1 activity of MIN6 cells after 12 h of administration with two ER stress–inducing agents, thapsigargin (1 μmol/l) and tunicamycin (2 μg/ml). Consistent with an increase in Foxo1 nuclear translocation, a significant increase in Foxo1 activity was observed (data not shown). The results of these studies indicated that the effect of fatty acid administration to induce ER stress could contribute to activation of Foxo1.

The data in Fig. 3 indicated that JNK activation is observed early after fatty acid administration. To test JNK's contribution to the action of fatty acid to induce Foxo1 and apoptotic markers, we used SP600125, a JNK-specific inhibitor (40). The addition of SP600125 resulted in a dose-dependent decrease of palmitate-induced phosphorylation of c-Jun along with expression of cleaved caspase 3 (Fig. 5A). Administration with 0.4 mmol/l palmitate resulted in an increase of nuclear Foxo1 at 4 h (Fig. 5B). A 1-h preincubation and continued incubation with 300 nmol/l of SP600125 resulted in maintenance of Foxo1 in the cytoplasm and decreased Foxo1 in the nucleus, despite the presence of 0.4 mmol/l palmitate. A repeat of this experiment with a 12-h incubation yielded similar findings. Because cell fractionation can result in cross-contamination of nuclear and cytoplasmic proteins, to confirm these findings, the experiment was repeated by direct visualization with GFP-tagged Foxo1, showing significant cytosolic to nuclear translocation after 4-h administration with palmitate and a reduction after JNK inhibition with SP600125 (Fig. 5C and D).

These results demonstrated that JNK activation contributes an important part of fatty acid–induced cytoplasmic to nuclear translocation of Foxo1. JNK activation enhances transcriptional activity of Foxo1 via direct binding and active phosphorylation of Foxo1 (14). However, JNK could also activate Foxo1 indirectly through inhibition of insulin signaling (41). The latter mechanism was previously suggested because decreased Akt activity was measured 24 h after oleate administration of insulinoma cells, and overexpression of a constitutively active Akt prevented oleate-induced apoptosis (8). We confirmed a decreased Akt phosphorylation 24 h after fatty acid administration (Fig. 2B). However, the time course of Akt phosphorylation after fatty acid administration shown in Fig. 5E was complex. An early decrease at 1 h after fatty acid administration was consistently followed by increased phosphorylation at 4 h when Foxo1 activity was increased. Thus, the JNK-dependent Foxo1 activation (Fig. 5B) appeared to reflect a direct action of JNK rather than JNK effects to decrease insulin signaling. We have observed a time course for glycogen synthase kinase 3β (Gsk3β) phosphorylation that is very similar to that of Akt phosphorylation (data not shown). This observation supports our interpretation that Akt is phosphorylated/activated after its early transient dephosphorylation and that this does not coincide with FoxO retention in the cytoplasm. The progressive decline in Akt phosphorylation after 4 h (Fig. 5E) suggested that decreased insulin signaling likely contributes additionally to JNK-induced Foxo1 activation at later time points.

To directly examine the contribution of Foxo1 to fatty acid–induced ER stress and cell death, we used an adenoviral construct of an hemaglutinin-tagged dominant-negative Foxo (AdV-DNFoxo) and evaluated its effects on ER stress–and fatty acid–induced apoptosis in MIN6 cells. This DNFoxo is a truncated allele of Foxo1 that retains the DNA-binding domain (amino acid residues 1–256) but lacks the transactivation domain (23). This construct was created to examine the contribution of the three Foxo proteins expressed in β-cells under normal conditions, and with loss of β-cell mass in insulin-resistant models. This allele has been demonstrated to function as a dominant-negative inhibitor of endogenous Foxo activity in the liver (42), in myoblasts (43), and in the β-cell (17,18). MIN6 cells transfected with 200 MOI of the DNFoxo adenovirus exhibited a nearly 60% reduction in propidium iodide staining and cell death when exposed to 100 nmol/l thapsigargin, compared with that in control GFP adenovirus (AdV-GFP)-transfected cells (Fig. 6A and B). Similar protection against cell death was also observed with palmitate administration of MIN6 cells (Fig. 6C and D). The effects of adenoviral administration after 24 h of palmitate were further examined by Western blot analysis of protein extracts (Fig. 6E). Increasing the MOI of the adenovirus resulted in augmented expression of the hemaglutinin-tagged DNFoxo protein, as indicated by Western blot using a hemaglutinin antibody. Increasing titers of the AdV-DNFoxo in cells exposed to 0.4 mmol/l palmitate for 24 h resulted in decreased expression of CHOP, along with a significant reduction in cleaved caspase 3 compared with cells similarly administered with palmitate but transfected with AdV-GFP. These results correlate with the observation of DNFoxo reducing the percentage of propidium iodide–positive cells (Fig. 6D). They support the interpretation that Foxo1 activation is important in mediating cell death after ER stress and fatty acid administration.

The results obtained with MIN6 cells expressing DNFoxo were confirmed with primary islets, as shown in Fig. 7. Islets were isolated from transgenic RIP-DNFoxo mice expressing the dominant-negative Foxo driven by the rat insulin promoter and were cultured in the presence of 10 μmol/l thapsigargin for 48 h to induce ER stress. This resulted in cleaved caspase 3 expression only in wild-type islets (Fig. 7A). The resistance to apoptosis in RIP-DNFoxo islets also applied to fatty acid administration. Isolated islets from wild-type and RIP-DNFoxo mice cultured in the presence of 0.4 mmol/l palmitate for 24 h exhibited similar JNK phosphorylation without changes in total JNK protein. (Fig. 7B). However, protein expression of the proapoptotic transcription factor CHOP was only observed in wild-type islets. RIP-DNFoxo islets also resisted caspase 3 activation by palmitate, which was reproduced with wild-type primary islets. Additionally, the abundance of phosphorylated Akt and the transcription factor Pdx1 were elevated in RIP-DNFoxo islets compared with similarly administered wild-type islets.

To define the contribution of Foxo activation to fatty acid regulation of gene expression relevant to apoptosis, we performed two parallel microarray experiments. The first experiment compared expression profiles in cells administered or not administered with fatty acid and infected with Adv-GFP. The second experiment further evaluated the gene expression profiles of MIN6 cells in fatty acid–administered and unadministered cells but infected with AdV-DNFoxo. This latter experiment permitted comparison of the effects of fatty acids in the presence and absence of Foxo signaling. Expression profiles of ∼16,000 genes were assessed. After 24-h administration with 0.4 mmol/l palmitate and AdV-GFP, 124 genes were significantly regulated (see research design and methods) more than twofold (91 genes upregulated and 33 downregulated). The Foxo-dependent fatty acid–regulated genes were clustered according to their ontology physiological processes (27,44). The degree of activation of the individual regulated genes and the effects of administration with AdV-DNFoxo are illustrated in Fig. 8. There was a pronounced participation of genes involved primarily in cellular metabolism, cellular physiology, and transport. These included a number of genes that could influence fatty acid–induced apoptosis. Immediately apparent was that the overwhelming majority of genes that were regulated by palmitate in the control condition were not regulated to the same extent when Foxo was inhibited, suggesting that the fatty acid–induced transcriptional regulation of these genes is predominantly Foxo dependent (Supplemental Fig. S2). The magnitude of expression of the genes up- and downregulated by palmitate in control conditions (AdV-GFP) and in the presence of AdV-DNFoxo are further illustrated in Supplemental Figs. S2A and B. These results were validated by an independent assessment of changes in mRNA by quantitative RT-PCR on selected transcripts (Supplemental Fig. S3).

One of the genes that contained multiple putative conserved Foxo binding sites whose transcription was activated by fatty acid administration was CHOP (DNA damage–inducible transcript 3 [Ddit3]) (Fig. 8). Previous studies had noted an association between fatty acid–induced ER stress and CHOP expression in insulinoma cells (7,30). We previously studied genes activated in MIN6 insulinoma cells on nutrient withdrawal by microarray analysis and noted that the most highly expressed gene was Ddit3 (CHOP, Gadd153). Observing a conserved canonical Foxo binding site in the CHOP promoter, we demonstrated with a CHOP-luciferase assay increased transcription of CHOP with nutrient withdrawal that was inhibited by coexpression with a plasmid expressing DNFoxo (17). This established that CHOP-induced expression with nutrient withdrawal was Foxo dependent. In the current studies, we sought to determine how fatty acid–and ER stress–induced CHOP expression and Foxo activation were related. MIN6 cells were transfected with a luciferase construct driven by 10 kb of the CHOP promoter (45), either an empty pCMV vector or pCMV-DNFoxo, and a thymidine kinase promoter driving Renilla expression as a control vector. CHOP promoter activity was induced in cells administered with 0.4 mmol/l palmitate for 24 h, as shown in Fig. 9. Cotransfection of the DNFoxo plasmid significantly reduced CHOP promoter activity in a dose-dependent manner. Nonspecific effects of pCMV-DNFoxo on CHOP-luciferase were corrected by normalization with inclusion of pTK-Renilla. These results clearly demonstrated that ER stress–induced CHOP expression is regulated at least in part by Foxo activation. In the absence of mutation of the Foxo binding sites in the CHOP promoter, however, these results are consistent with an indirect effect of Foxo activation on CHOP expression.

This study was designed to examine the sequence of molecular signaling events in fatty acid–induced apoptosis in insulinoma cells and islets and to test the hypothesis that Foxo1 activation plays a significant role in this process. Fatty acids have been shown to contribute to β-cell apoptosis by decreasing insulin/Akt signaling (7,8). Only one study examined Foxo1 involvement in this process, measuring changes in phosphorylation of Foxo1 as an indicator of protein activity but not definitively or directly implicating Foxo in the process (8). In the current study, Foxo involvement is documented in a definitive way, showing changes in its nuclear localization and its role in transcriptional effects of fatty acids. In addition, these studies determined the earliest events in fatty acid–induced apoptosis and documented the involvement of JNK (as opposed to Akt) in the early action of fatty acid to induce nuclear localization of Foxo1. Compelling evidence is presented to indicate that ER stress via activation of JNK and Foxo1 contributes to the apoptotic outcome after fatty acid administration. Finally, novel evidence shows that fatty acids alter expression of a large number of proapoptotic genes in islets and that many of these effects are mediated by Foxo1.

Administration of MIN6 cells with palmitate or oleate at various concentrations established a pattern of intracellular signaling that was common to both fatty acids. The data indicated that both fatty acids regulate the same signaling events, with palmitate being more potent possibly as a result of its less effective processing by esterification (6,30,31). In various mammalian cell lines including β-cells, lower fatty acid toxicity was associated with increased cellular capacity for triglyceride synthesis and with upregulation of stearoyl CoA desaturase 1 (SCD1) (31,46). SCD1 functions in fatty acid desaturation before incorporation into triglycerides, and its expression was shown to be downregulated by Foxo1 (47). This effect of Foxo1 could contribute to cell death indirectly by inhibiting fatty acid incorporation into the neutral triglyceride pool, consequently increasing available fatty acids and the potential for accumulation of toxic metabolites such as ceramides (48,49). Possibly, fatty acid activation of Foxo1 with subsequent inhibition of SCD1 would create a self-sustained cycle with deleterious cellular effects (31,46,48,49). The relatively greater effects of the saturated fatty acid on Foxo1 activation and apoptosis may have relevance to nutritional treatment of patients with type 2 diabetes.

A time course of administration with either palmitate or oleate in MIN6 cells suggested that JNK phosphorylation precedes markers of ER stress and apoptosis, including phosphorylation of eIF2α, expression of CHOP, and the eventual increase of cleaved caspase 3. In addition, JNK inhibition decreased nuclear Foxo1 along with the reduction in fatty acid–induced apoptosis. These findings position JNK as an important, early participant upstream of Foxo1 in fatty acid activation of the proapoptotic pathway. Previous studies on islets and insulinoma cells demonstrated JNK activation via oxidative stress, cytokines, and ER stress–inducing agents (9,18,50), but the underlying mechanisms remain unclear. A recent study demonstrated palmitate binding to the proinflammatory TLR2 (Toll-like receptor 2), leading to enhanced JNK activation and insulin resistance in murine myocytes (51). Another report documented dependence of JNK activation in macrophages on signaling via the membrane fatty acid translocase CD36 (75).

The important role of decreased insulin signaling and Akt phosphorylation in fatty acid–induced β-cell apoptosis has been reported previously (8). However, our data would indicate that early activation of Foxo1, which would set in motion the apoptotic program after fatty acids, is not dependent on Akt activity. The transient apparent increase in Akt activity at 4 h after fatty acid administration is unexplained, although it could be secondary to the acute stimulatory effect of fatty acids on insulin secretion (5). Alternatively, a recent report has described a complex interaction between the activities of Foxo and Akt (52). Administration of primary hepatocytes with either wild-type, constitutively nuclear, or dominant-negative Foxo adenovirus resulted in activation of protein kinase B pathways. In the current report, at 4 h after fatty acid administration, a time when Foxo1 was observed to be nuclear and transcriptionally active, we also observed increased phosphorylation of Akt (Fig. 3B). Possibly JNK-activated Foxo1 may have contributed to this transient increase in Akt phosphorylation. We repeatedly observed decreased Akt phosphorylation after 24 h of fatty acid administration, a finding consistent with that in another insulinoma cell line (8). Taken together, these results suggest that reduced Akt phosphorylation chronically contributes to maintenance of the early, direct activation of Foxo1 by JNK.

Whereas the current results show that fatty acid administration results in increased Foxo activity, previous studies (8) have shown that fatty acid administration results in reduced phosphorylation of another Akt substrate, Gsk3β, that could also contribute to apoptosis. In preliminary studies we have confirmed that palmitate and oleate administration of MIN6 cells results in reduced phosphorylation of Gsk3β, and we extended these studies to show that inhibitors of Gsk3β activity reduced fatty acid–induced apoptosis (76). These combined results thus indicate that reduced insulin signaling leads to activation of at least two proapoptotic proteins, Foxo and Gsk3β, that contribute to the deleterious effects of fatty acid on β-cells.

A total of 2,500 bp of the promoter regions upstream of the start codon for all of the fatty acid regulated genes were assessed in the mouse, rat, and human genomes for those that contained an evolutionarily conserved 7-bp sequence (T[A/G]TT[T/C][A/C]C) encoding Forkhead binding sites (53,54). By chance alone, the consensus binding site can be found once every 1,024 bp (4⁴ × 2³ × 0.5). The number of putative Forkhead binding sites for each gene is represented in Supplemental Table 1 by an asterisk appended to the symbol of the gene. Notable is that very few genes in this group had fewer than two potential Foxo binding sites. However, whether these genes are directly regulated by Foxo binding can only be determined by direct experimentation.

The fatty acid–induced genes that could be mediating apoptosis in β-cells were examined by a whole-genome expression study. We found that >100 genes exhibited a more than twofold up- or downregulation in expression after 24 h of administration with palmitate (Fig. 8). Validation of these expression results is suggested by the concomitant observation of fatty acid–reduced expression of Pdx1 and increased expression of CHOP proteins by Western analysis (Figs. 2C and 3B) and by RT-PCR of selected transcripts (Supplemental Fig. S3). A number of fatty acid–regulated genes that may contribute to reduce insulin signaling or secretion were noted. Trib3 (Trb3) is a known inhibitor of Akt activation and potentiator of apoptosis that has been incriminated in the insulin resistance of type 2 diabetes (55). Its 3.6-fold induction by fatty acids suggests the involvement of Trib3 in fatty acid impairment of insulin signaling. Trib3 was also identified as a target of ER stress–induced apoptosis in human embryonic kidney (293) cells (56). Furthermore, it was shown that Trib3 was downstream of CHOP activation and that upregulation of Trib3 by ER stress was required for apoptosis. The hypothesis that Trib3 is downstream of fatty acid–induced ER stress–and CHOP-mediated apoptosis in insulinoma cells can now be tested.

Other fatty acid–regulated genes included Eif4ebp1, a protein that binds Eif4E to inhibit translation. Phosphorylation of Eif4ebp1 leading to its dissociation from Eif4E has been implicated at least in part in insulin-mediated enhancement of translation and cell growth (57). Similarly, cyclin-dependent kinase inhibitor 1A (Cdkn1a or p21) is a protein that binds to cyclin complexes and inhibits cell cycle progression (58), and its activation is associated with ER stress and apoptosis of β-cells (59). Peroxisome proliferator–activated receptor-γ is a transcription factor that senses and regulates lipid metabolism. Its induction in β-cells has been associated with increased fatty acid uptake, triglyceride synthesis, and inhibited insulin secretion (58,60–62).

Remarkably, regulation of >95% of the fatty acid–sensitive genes was blunted or completely reversed with concomitant overexpression of dominant-negative Foxo1. For example, a sixfold induction by fatty acids of Cxcl12 (chemokine [C-X-C motif] ligand 12), a proinflammatory cytokine responsive to lipopolysaccharides, tumor necrosis factor, or interleukin 1 (63), was blunted by 85%. The threefold upregulation of Eif4ebp1 was reduced by 70%. Genes upregulated by palmitate administration that exhibited blunted response in the presence of dominant-negative Foxo1 also included Dgat2 (diacylglycerol O-acyltransferase 2), responsible for conversion of diglycerides to triglycerides (64), and Acaa2 (acetyl-CoA acyltransferase 2), which catalyzes the last step of mitochondrial β-oxidation (62,65).

Among the genes downregulated by fatty acid administration that may contribute to β-cell apoptosis, Pdx1 is especially interesting. There was a twofold reduction in Pdx1 gene expression after palmitate administration that correlated with the decrease in Pdx1 protein (Fig. 2C). Fatty acid reduction of Pdx1 expression was blunted by 69% with ADv-DNFoxo administration, documenting Foxo dependence of the effect. Previous work had shown that insulin receptor substrate-2 deficiency was associated with Pdx deficiency, and in this model β-cell destruction was corrected by either Foxo haploinsufficiency or overexpression of Pdx1 (61). Together, these results emphasize the link between fatty acid/ER stress reduction of insulin signaling, Foxo activation, and subsequent Pdx regulation and apoptosis.

We examined potential mechanisms for the Foxo1-dependent induction or repression of genes by fatty acids. β-Cells have a highly developed and active ER to maintain insulin secretion, which translates to a high susceptibility to stress (66). Foxo1 has been shown to contribute to apoptosis through transcription of several proapoptotic genes, such as Bim, FasL, and TRAIL (54). Our study identifies a novel mechanism by which Foxo1 could promote β-cell apoptosis through its activation of CHOP (66). Conditions that increase CHOP expression have consistently been associated with ER stress, decreased insulin signaling, and apoptosis in β-cells (7,9). Disruption of the CHOP gene delayed but did not prevent the onset of hyperglycemia and β-cell death in an in vivo model of pancreatic β-cell ER stress, the Akita mouse, heterozygous for a point mutation in the insulin 2 gene (Ins2C96Y) (67). Thus, these results suggest redundancy in ER stress–mediated pathways leading to apoptosis, potentially exclusive of, or in addition to, Foxo1 and CHOP.

This study documented the role of FoxO activation by fatty acids or ER stress in mediating β-cell death. The data do not rule out effects of fatty acid and oxidative stress in inducing mitochondrial dysfunction that triggers programmed cell death, as reviewed recently (68,69). There is limited information related to Foxo proteins and mitochondrial stress, but a recent study suggested differential roles of Foxo1 and Foxo3a. Silencing of Foxo1 did not alter susceptibility of endometrial cells to oxidative cell death, whereas silencing of Foxo3a prevented apoptosis induced by hydrogen peroxide (70).

Our findings suggest that further in vivo characterization of the role of Foxo1 within β-cells is merited especially in the context of the β-cell failure associated with diabetes. Prolonged administration of islets or insulinoma cells with fatty acids results in decreased glucose-stimulated insulin secretion (5,71). Foxo1 may mediate some of this effect as it contributes to the phenotype of insulin resistance by increasing fatty acid utilization and decreasing glucose utilization in skeletal muscle and by increasing gluconeogenesis in the liver (22,72,73). Foxo1 activation in β-cells could likewise downregulate glucose metabolism and thus insulin secretion and synthesis. Nutrient withdrawal and decreased insulin signaling have previously been shown to activate Foxo1 in β-cells (17), and now we extend the agonists of Foxo1 to include fatty acids and pharmacological ER stress inducers. We were able to significantly decrease β-cell apoptosis from ER stress or fatty acids by overexpressing a dominant-negative Foxo1. Inhibiting Foxo1 could be clinically relevant for promoting the survival of β-cells in obesity and diabetes, conditions associated with insulin resistance, increased serum fatty acid levels, and ER stress. These results thus identify Foxo1 as a potential new therapeutic target for preservation of β-cell mass and function.

This work was supported in part by an National Institutes of Health (NIH) grant (R37 DK16746, to M.A.P.), a Ruth L. Kirschstein Pre-Doctoral National Research Service Award (F32 DK06537, to S.C.M.), and a grant from the Philip Morris USA External Research Program. The Washington University Diabetes Research and Training Center (NIH P60 DK20579) is acknowledged for assistance from the Immunoassay, Morphology, and Transgenic Mouse Cores, and the Adipocyte Biology Core, Clinical Nutrition Research Unit (NIH DK56351).

The authors thank H. Harding and D. Ron (New York University) for reagents, D. Accili (Columbia) for insightful comments and reagents, K. Polonsky for helpful comments on the experiments, Cris M. Welling for lab assistance, and other members of the Permutt lab for helpful discussions.