INS-1 Cells Undergoing Caspase-Dependent Apoptosis Enhance the Regenerative Capacity of Neighboring Cells

Rat insulinoma INS-1 cells overexpressing wild-type HNF1A (WT-HNF1A) or DN-HNF1A under control of a doxycycline-dependent transcriptional activator have been described previously (4,8). Cells were cultured in RPMI-1640 at 6 mmol/l glucose supplemented with 10% FBS (PAA, Cölbe, Germany), 2 mmol/l l-glutamine, 1 mmol/l pyruvate, 100 units/ml penicillin, 100 μg/ml streptomycin, 10 mmol/l HEPES (pH 7.4), and 50 μmol/l 2-mercaptoethanol (Sigma, Dublin, Ireland). For experiments investigating glucose dependence of PSP/reg induction, expression of DN-HNF1A was induced for 24 h and continued for a further 2 h in medium supplemented with glucose at 3, 6, 12, and 18 mmol/l. For caspase inhibition, cells were preincubated with 100 μmol/l of the broad-spectrum caspase inhibitor zVAD.fmk (Bachem, St. Helen's, U.K.) for 45 min prior to induction.

cDNA synthesis was performed using 1.5 μg total RNA as the template and Superscript II reverse transcriptase (Invitrogen, Paisley, U.K.) primed with 50 pmol random hexamers (New England Biolabs, Ipswich, MA). Real-time PCR was performed using the LightCycler 2.0 (Roche Diagnostics, Indianapolis, IN) and the QuantiTech SYBR Green PCR kit (Qiagen). Specific PCR primers were designed using Primer3 software (sequences available on request). For absolute quantification of hnf1a copy number, a gene-specific PCR amplicon of known concentration was prepared as a standard. Melting curve analysis and gel electrophoresis was utilized to verify specificity of all PCR products. The data were analyzed using LightCycler software, version 4.0, with all samples normalized to β-actin.

For Western blotting analysis, 25 μg protein lysates were obtained from INS-1 cells overexpressing WT-HNF1A and DN-HNF1A as described (9). The rabbit polyclonal anti-PSP/reg antibody (22) was diluted 1:20,000 in Tris-buffered saline containing 1% bovine serum albumin. The primary mouse monoclonal anti–β-actin antibody was utilized at a 1:10,000 dilution (Sigma). Horseradish peroxidase–conjugated secondary antibodies were obtained from Pierce and detected using SuperSignal West Pico Chemiluminescent Substrate (Pierce) and imaged using a FujiFilm LAS-3000 imaging system (Fuji, Sheffield, U.K.).

Paraffin-embedded pancreatic sections from rat insulin promoter (RIP)-DN-HNF1A and control wild-type C57BL/6JBomTac mice (13) were deparaffinized and incubated overnight at 4°C with the rabbit polyclonal anti-PSP/reg antibody diluted 1:20 (22). Specificity of the antibody was confirmed by Western blotting experiments using recombinant human PSP/reg protein as a control. Slides were incubated for 1 h at room temperature in rhodamine-labeled anti-rabbit secondary antibody (1:100; Jackson Immunoresearch, Suffolk, U.K.). Slides were incubated again overnight at 4°C with the second primary guinea pig anti-insulin antibody (1:20; DakoCytomation A0564) followed by 1 h at room temperature in fluorescein isothiocyanate (FITC)-labeled anti-goat secondary antibody (1:100; Jackson Immunoresearch) and mounted in Vectashield with DAPI (Vector Shield; Vector Labs, Burlingame, CA). Images were taken with a Zeiss LSM710 confocal microscope equipped with a 40 × 1.3NA oil-immersion objective (Carl Zeiss, Jena, Germany). FITC was excited at 488 nm with an argon laser, using a 488/543-nm multichroic beam splitter, and the emission was collected at a 490- to 550-nm spectral band. Rhodamine was excited at 543 nm with a helium neon laser, using the same beam splitter, and the emission was collected at a 570- to 650-nm spectral band. DAPI was excited using the 405 nm DPSS laser and a 405-nm dichroic beam splitter. DAPI fluorescence was detected in the 420- to 480-nm spectral band. Images were processed using Zeiss LSM 4.2 software (Carl Zeiss). Rhodamine-positive cells in the periphery and interior of the islet were quantified as a percentage of total islet cells using Image J software.

INS-1 cells were seeded at 1 × 10⁵ per ml on a 24-well plate and allowed to settle overnight. Following treatment cells were lysed in 200 μl lysis buffer (10 mmol/l HEPES, pH 7.4; 42 mmol/l KCl; 5 mmol/l MgCl2; 1 mmol/l phenylmethylsulfonyl fluoride; 0.1 mmol/l EDTA; 0.1 mmol/l EGTA; 1 mmol/l dithiothreitol [DTT]; 1 μg/ml pepstatin A; 1 μg/ml leupeptin; 5 μg/ml aprotinin; and 0.5% 3-[{3-cholamidopropyl}dimethylammonio]-1-propane sulfonate [CHAPS]). Fifty microliters of this lysate was then added to 150 μl reaction buffer (25 mmol/l HEPES, 1 mmol/l EDTA, 0.1% CHAPS, 10% sucrose, and 3 mmol/l DTT, pH 7.5) containing 10 μmol/l Ac-DEVD-AMC, which is efficiently cleaved by the apoptotic executioner caspases 3 and 7, among others (23). Cleavage of the substrate DEVD-AMC resulting in the accumulation of fluorescent AMC was measured on a Genios fluorescence plate reader (Tecan) using 355 nm excitation and 460 nm emission wavelengths as previously described (9). Protein content was determined using the Pierce Coomassie Plus Protein assay reagent (Perbio) and activity expressed as change in fluorescent units per microgram of protein per hour.

Cells were stained using DAPI to visualize nuclear morphology. Condensed and/or fragmented nuclei were considered apoptotic and observed using an Eclipse TE 300 inverted microscope (Nikon, Dusseldorf, Germany) and a 20 × dry objective, with DAPI excited at 340 ± 25 nm and the emission collected between 450 and 500 nm.

Cellular proliferation was assessed by means of bromodeoxyuridine (5-bromo-2-deoxyuridine [BrdU]) incorporation using the BrdU Cell Proliferation Assay (Calbiochem). BrdU was added to cells for 24 h, and cells were fixed following treatment. Following incubations with the primary and secondary antibodies, the peroxidase-labeled secondary antibody was detected by addition of substrate solution for 15 min and the reaction stopped. Absorbance was read using a Bio-Rad 550 microplate reader at dual wavelengths of 450/540 nm. The values for the background wells were subtracted from all values. The data were normalized to noninduced control.

Cell supernatants were centrifuged (500g; 15 min) to remove floating cells and medium containing microparticles released from INS-1 cells collected following an established protocol (24). The medium was filtered using 0.2-μm filters (VWR International, Dublin, Ireland) in order to remove microparticles. Microparticles were analyzed on a Coulter EPICS XL flow cytometer (Beckman Coulter, Villepinte, France) as previously described (25). Regions corresponding to microparticles were identified in forward light scatter and side-angle light scatter dot plots using a logarithmic gain. The gate for microparticles was defined as events with a 0.1- to 1-μm diameter, in comparison with calibrator beads (Megamix fluorescent beads of 0.5, 0.9, and 3 μm in diameter; Biocytex, Marseille, France), and plotted on an FL/FSC dot plot to determine the microparticles. Microparticles were defined as membrane vesicles <1 μmol/l in diameter and expressing externalized phosphatidylserine. Phosphatidylserine was labeled using FITC-conjugated annexin V (Roche Diagnostics, Meylan, France) in the presence or absence (negative control) of CaCl2 (5 mmol/l) as reported earlier (25).

The recombinant human PSP/reg protein was generated as previously described (22,26). Briefly, the coding region of PSP/reg was cloned into a transfer vector (pPIC9; Invitrogen) containing the signal sequence of the yeast α-mating factor to drive the protein into the secretory pathway.

Sixteen diabetic subjects with mutations in the HNF1A gene (P291fsinsC, n = 9 and p.Ser353fs, n = 7) were identified from an Irish MODY collection. Sequencing of the HNF1A gene was performed by IntegraGen (Bonn, Germany). Control subjects were first-degree relatives of HNF1A-positive subjects who were nondiabetic and negative for the known HNF1A mutation in their pedigree (n = 7). In addition, 10 GAD- and/or islet antibody–positive patients with type 1 diabetes were recruited in the Mater Misericordiae University Hospital. Anthropometric measurements including weight, height, and BMI were obtained. The plasma glucose concentration was measured using an YSI analyser (Roche). A1C was determined using high-performance liquid chromatography (Menarini HA81-10, Rome, Italy). Anti-GAD was analyzed using competitive fluid-phase radioimmunoassay by the neuroscience group at John Radcliffe Hospital, Oxford, U.K., and islet cell antibodies by University College London Medical School by similar means. Study groups were matched for age, sex, and BMI as shown in Table 1 and approved by the Mater Misericordiae University Hospital Ethics committee.

The enzyme-linked immunosorbent assay (ELISA) to quantify human PSP/reg was performed using anti-sera from rabbits and guinea pigs immunized with recombinant human PSP/reg protein as previously described (22,27). Serum was prepared by centrifugation, and the IgG were purified by affinity chromatography on protein A columns. Subsequently, a sandwich ELISA was designed on 96-well ELISA plates. Antibody of the first species (guinea pig) was coated to the bottom, blocked with BSA, and aliquots of serum were then incubated for 2 h. After washing, antibodies of the other species (rabbit) were incubated. Finally, a phosphatase-coupled anti-rabbit IgG was used. Patient serum PSP/reg levels were compared with standard amounts of protein of recombinant human PSP/reg.

Results were expressed as means ± SE. Statistical analysis was conducted using the SPSS version 15.0 software package for Windows (SPSS). Differences between treatments were analyzed by Student t test, as well as one-way ANOVA and a subsequent Tukey tests. Nonparametric data were analyzed by Mann-Whitney U test. Differences were considered to be significant at P < 0.05.

Inducible, dominant-negative suppression of HNF1A function in INS-1 cells reduces the expression of HNF1A target genes involved in glucose and lipid homeostasis (4) and induces a caspase-dependent apoptosis (8). To investigate the relationship between DN-HNF1A expression, activation of apoptosis, and induction of the regenerative PSP/reg gene in particular, INS-1 cells stably transfected with respective rtTA systems were induced to express either DN-HNF1A or WT-HNF1A. We observed a time-dependent induction of DN-HNF1A or WT-HNF1A mRNA in the respective INS-1 cells (Fig. 1,A). This correlated with a significant reduction in mRNA levels of pdk1, a known HNF1A target gene (7), in response to DN-HNF1A expression but not WT-HNF1A (Fig. 1,B). Examination of the expression of the Reg family of genes in INS-1 cells demonstrated that DN-HNF1A led to a potent induction of the PSP/reg gene at 24 h (∼15-fold) and 48 h (∼30-fold). Induction of WT-HNF1A for up to 48 h did not significantly regulate PSP/reg expression (Fig. 1 C).

We next investigated PSP/reg protein levels after induction of DN-HNF1A. Cleavage of PSP/reg converts the 16-kDa proteins into a 14-kDa insoluble fibrillar protein (28). Western blot analysis demonstrated a significant upregulation of the PSP/reg protein (16 kDa) following DN-HNF1A induction and a time-dependent increase in the 14 kDa form of PSP/reg (Fig. 1 D).

We next determined whether the induction of PSP/reg mRNA was glucose dependent. Interestingly, we found that the DN-HNF1A induction of PSP/reg mRNA expression was modulated by varying extracellular glucose concentrations for 2 h following DN-HNF1A induction. The increase in PSP/reg mRNA expression was potentiated in a concentration-dependent manner, with the highest induction evident at 18 mmol/l (Fig. 2 A).

We analyzed paraffin-embedded pancreatic sections from 5-month-old diabetic (Fig. 2,B) mice expressing DN-HNF1A in β-cells and compared them with wild-type C57BL/6JBomTac mice. Sections were costained for PSP/reg and insulin. PSP/reg was expressed at low levels in wild-type islets (36 ± 1 PSP/reg-positive cells per islet; n = 8 islets from n = 2 animals) and was found almost exclusively in peripheral islet cells. In contrast, DN-HNF1A islets showed significantly elevated PSP/reg-positive islet cells (82 ± 2 PSP/reg-positive cells per islet; n = 8 islets from n = 2 animals). Moreover, PSP/reg-positive cells were found to be widespread throughout the transgenic islets and were also detectable in insulin-negative islet cells. Strong expression was also observed in the surrounding acinar cells (Fig. 2,B). Interestingly, high magnification analysis suggested that many islet cells with elevated PSP/reg immunoreactivity were positioned in the vicinity of cells displaying apoptotic nuclear morphology (Fig. 2 C).

We next sought to determine whether increased PSP/Reg1a levels could also be detected in HNF1A-MODY diabetic subjects. We analyzed serum PSP/Reg1a levels in a group of 16 diabetic subjects with HNF1A-MODY using ELISA. Results were compared with serum levels in seven HNF1A-MODY–negative, nondiabetic family members. PSP/Reg1a levels were significantly elevated in HNF1A-MODY subjects (Fig. 2,D). Interestingly, we also detected elevated PSP/Reg1a serum levels in the serum of patients with type 1 diabetes (n = 10 patients) (Fig. 2 D).

Previously we demonstrated that DN-HNF1A–induced apoptosis of INS-1 cells required caspases and involved the activation of the mitochondrial apoptosis pathway (8). To assess whether the activation of executioner caspases influenced DN-HNF1A–induced PSP/reg mRNA induction, we first determined the time course of DEVDase activation (indicative of executioner caspase 3/7 activity) by monitoring the cleavage of a fluorigenic caspase substrate. In agreement with our earlier report (8), there was a significant caspase activity in the time frame of 24–48 h post induction. Expression of WT-HNF1A or doxycycline treatment of parental INS-1 cells did not activate caspases (Fig. 3,A). Furthermore, DN-HNF1A induction mediated nuclear morphological changes indicative of apoptosis with condensation and fragmented nuclei. Pretreatment with the pancaspase inhibitor zVAD.fmk inhibited the occurrence of this apoptotic morphology confirming the caspase dependence (Fig. 3 B).

Given the similar kinetics of PSP/reg mRNA induction and caspase activation, we explored the possibility that caspase activation regulated PSP/reg expression. Indeed, DN-HNF1A induction of PSP/reg expression was completely inhibited by pretreatment with zVAD.fmk (Fig. 3,C and D). Next we addressed whether DN-HNF1A induction of PSP/reg mRNA could be sufficiently explained by its ability to activate apoptosis. Indeed, induction of apoptosis in INS-1 cells with the topoisomerase inhibitor, etoposide, was sufficient to increase PSP/reg mRNA levels in a caspase-dependent manner (Fig. 3 E).

Executioner caspases such as caspase 3 inactivate transcription and translation processes (29), suggesting that apoptosing cells were unlikely to be the source of increased PSP/reg mRNA during DN-HNF1A–induced apoptosis. We investigated whether this increased expression was due to apoptosing cells sending a paracrine signal to neighboring cells. After 48 h of DN-HNF1A induction in the presence and absence of zVAD.fmk, the conditioned medium was added directly to naïve INS-1 cells resulting in a 16-fold increase in PSP/reg mRNA levels in the naïve INS1-cells (Fig. 4,A). This induction was absent in cultures incubated with the conditioned medium of zVAD.fmk pretreated DN-HNF1A–induced cells (Fig. 4,A) or heat-inactivated conditioned medium (Fig. 4 B), supporting the hypothesis that apoptotic cells secrete heat-sensitive factor(s) that stimulate(s) PSP/reg gene induction in neighboring cells.

Microparticles are small (100 nm) membrane vesicles originating from blebbing membranes of apoptotic cells and can be identified by annexin V staining. These microparticles may contain nuclear proteins as well as nucleic acids (30,31). We examined if microparticles were present in our conditioned medium and, if so, whether these could modulate PSP/reg expression. After 48 h of DN-HNF1A induction in the presence and absence of zVAD.fmk, we enriched for microparticles in the supernatant (25) and applied either filtered or unfiltered medium to naïve INS-1 cells for a further 24 h. Naïve INS-1 cells treated with the unfiltered conditioned medium from DN-HNF1A–induced cells had increased PSP/reg mRNA levels that were abolished by filtration of this conditioned medium or by using unfiltered medium from DN-HNF1A–induced cultures in the presence of zVAD.fmk (Fig. 4,C). Importantly, flow cytometry identified increased microparticle levels in the unfiltered, compared with the filtered, conditioned medium from the same experiment, indicating that microparticles were not generated by the isolation procedure (Fig. 4,D). Indeed, microparticles were at least 150-fold more abundant in unfiltered compared with filtered conditioned medium with a significantly reduced level in zVAD.fmk pretreated unfiltered medium (Fig. 4 D). These results suggest that caspases are required for microparticle formation, and their removal abolishes the conditioned medium–induced PSP/reg mRNA expression in naïve INS-1 cells.

Animal models of diabetes have demonstrated that recombinant PSP/reg (rPSP/reg) increases β-cell ductal cell proliferation (21,32). DN-HNF1A induction led to significant decreases in cell proliferation as detected by BrdU incorporation, which was reversed by rPSP/reg treatment (Fig. 5,A). Decreased cell proliferation has been attributed to DN-HNF1A–induced expression of the cell cycle inhibitor p27^Kip1 (8,9). Induction of DN-HNF1A led to increased p27^Kip1 mRNA expression but did not affect p21^WAF1 expression (data not shown), with rPSP/reg cotreatment reducing DN-HNF1A–induced p27^Kip1 expression (Fig. 5,C). Furthermore, the rPSP/reg treatment rescued the decrease in insulin gene expression induced by DN-HNF1A expression (Fig. 5,B). Taken together, our data suggest that rPSP/reg treatment reverses the DN-HNF1A–induced phenotype. Indeed, BrdU incorporation was also significantly elevated in naïve INS-1 cells treated with conditioned cultured medium from DN-HNF1A–induced cells compared with those induced in the presence of zVAD.fmk (Fig. 5 D).

The present study provides evidence that β-cell apoptosis may stimulate β-cell regeneration. We demonstrate 1) that dominant-negative suppression of HNF1A leads to apoptosis and a potent induction of PSP/reg; 2) this induction was potentiated by increased extracellular glucose concentrations; 3) the induction of PSP/reg was inhibited by the broad-spectrum caspase inhibitor, zVAD.fmk, which also inhibited DN-HNF1A–induced apoptosis; 4) conditioned medium from DN-HNF1A–induced cells was sufficient to induce PSP/reg expression in naïve INS-1 cells, suggesting that apoptosing INS-1 cells secrete factors that stimulate PSP/reg induction in neighboring cells; 5) removal of microparticles from this conditioned medium inhibited PSP/reg induction, suggesting that microparticles released from blebbing membranes mediated this effect; 6) treatment of INS-1 cells with rPSP/reg protein reversed the DN-HNF1A–induced decrease in cell proliferation and insulin mRNA; and 7) moreover, PSP/reg serum levels were significantly elevated in HNF1A-MODY patients and type 1 diabetes compared with HNF1A-MODY–negative, nondiabetic family members.

Evidence is growing that functional β-cell mass is decreased in animal models of MODY and in type 2 diabetes, with an increased apoptosis rate largely contributing to this decrease (12,13,33). However, a decrease in β-cell mass does not necessarily implicate that new β-cells are not generated during disease progression. Indeed, β-cell mass is a result of a homeostatic balance between the generation of new β-cells and the rate of their apoptosis (34). As HNF1A-MODY and type 2 diabetes evolve over several years or decades, the destruction or apoptosis rate of β-cells may eventually exceed the rate of β-cell generation during disease progression.

Although β-cell regeneration may occur during disease progression, it is not clear whether this happens through self-replication of mature β-cells, through neogenesis of progenitor cells residing in the pancreas or recruited from other tissues, or through transdifferentiation. Here, we supply to our knowledge for the first time evidence that the interaction of β-cell apoptosis and regeneration is more complex and in fact that both processes may be mechanistically linked. In our study, we observed that dominant-negative suppression of HNF1A or apoptosis induced by genotoxic stress (etoposide) led to caspase-dependent apoptosis of INS-1 cells, associated with the induction of the regenerative PSP/reg gene. Furthermore, conditioned medium from apoptosing cells was sufficient to stimulate PSP/reg mRNA expression and BrdU incorporation in naïve INS-1 cells. These findings allow for the conclusion that apoptosing β-cells secrete factor(s) that can stimulate regenerative gene induction and cellular proliferation in neighboring cells. Finally, we demonstrate that administration of rPSP/reg reversed the DN-HNF1A–induced decrease in cell proliferation, supporting the concept that β-cell replication may be positively regulated by β-cell apoptosis. The ability of apoptotic cells to provide signaling cues for their environment has been demonstrated in other settings, such as inhibition of inflammation. Interestingly, recent studies (35,36) have demonstrated that caspase activation is required for tissue regeneration during wound healing and planarian regeneration, suggesting that tissue regeneration in response to caspase-dependent apoptosis may be an evolutionary conserved process.

The Reg family of proteins constitute a multigene family in humans and rodents (19). PSP/reg increases β-cell mass and stimulates β-cell proliferation under physiological conditions and in 90% depancreatized and nicotinamide-injected rats (19,37). Studies in mice lacking the gene Reg1, the mouse homologue of PSP/reg, revealed that Reg1 is involved in cell-cycle progression and normal islet growth (37). Reg1 knockout mice have significant reductions in islet cell mass and rate of DNA synthesis in isolated islet cells (37,38). We identified that rPSP/reg reversed the DN-HNF1A–induced upregulation of p27^Kip1gene expression and the concomitant decrease in insulin expression. Previous studies have shown that PSP/reg induces extracellular signal–related kinase 1/2 signaling in MKN45 gastric cancer cells (39). However, the signal transduction pathways activated by PSP/reg in pancreatic islets warrant further investigation. Interestingly, we observed that elevated extracellular glucose concentrations rapidly potentiated PSP/reg gene expression. This may represent an important physiological feedback loop for the regulation of β-cell mass.

PSP/reg is physiologically secreted from pancreatic acinar cells (20). In the adult rat, PSP/reg has been shown to be present only in small amounts in β-cells (20,40). We detected increased expression of PSP/reg in insulin-positive islet cells in 5-month-old transgenic HNF1A-MODY mice, confirming previous findings that PSP/reg is expressed in islets during injury and regeneration processes (17,19). Moreover, PSP/reg was also found in insulin-negative islet cells (Fig. 2 B). In this context, a recent study has suggested α-cells to be the major source of β-cell regeneration/conversion in mice in vivo (41). We also detected increased PSP/reg levels in the serum of HNF1A-MODY patients. We can only speculate whether the source of PSP/reg in the serum of HNF1A-MODY patients was the endocrine or exocrine pancreas (i.e., islet or acinar cells). We can likewise not exclude that these levels came from other organs potentially affected in HNF1A-MODY, such as the liver, where PSP/reg is also expressed (42). However, we also detected elevated levels of PSP/reg in subjects with type 1 diabetes, arguing for a role for pancreatic injury and β-cell apoptosis as a source for elevated PSP/reg serum levels.

In our study, we detected that apoptotic INS-1 cells released annexin V–positive microparticles into the medium in a caspase-dependent manner. Microparticles are being increasingly described as efficient vehicles for the release of signaling molecules. Microparticles enable communication between different cell types (43) and also deliver proteins to cells that do not normally express them (44). Microparticles deriving from apoptotic cells have been shown to carry nuclear proteins as well as nucleic acids (31,45). Filtration experiments suggested that the absence of microparticles abolished the effect of the conditioned medium to induce PSP/reg mRNA in naïve INS-1 cells. This suggested that cytokines, peptide hormones, or other soluble messengers present in the conditioned medium did not mediate this effect. Interestingly, heat inactivation of the conditioned medium was sufficient to inhibit PSP/reg mRNA expression in naïve INS-1 cells. It is therefore tempting to speculate that microparticles may contain signaling molecules or nuclear proteins which, upon uptake by neighboring cells, enhanced PSP/reg gene expression in these cells.

In conclusion, our study demonstrates that the execution of caspase-dependent apoptosis in INS-1 cells stimulates the induction of regenerative genes in neighboring cells, suggesting a role for MPs and PSP/reg in this process. These findings provide new mechanistic insights into a fundamental aspect of β-cell mass regulation.

This study was supported by grants from Science Foundation Ireland (08/IN1/1949) and the Health Research Board (RP/2008/14) to J.H.M.P.; grants from the Health Research Board (RP/2007/316), the Mater Foundation cofunded by the Medical Research Charity Groups/Health Research Board, and a Mater College Grant to M.M.B; and the Swiss National Foundation (grant no. 32-66907.01) to C.B.W.

No potential conflicts of interest relevant to this article were reported.

C.B. researched data and wrote the manuscript. S.R.R., S.B., M.B., A.M.F., H.D., M.W.W., S.M.K., and C.M.B. researched data. C.G.C. reviewed/edited the manuscript and contributed to the discussion. C.B.W., R.G., and M.M.B. researched data and reviewed/edited the manuscript. J.H.M.P. wrote the manuscript and contributed to the discussion.

We thank M. Bain (Department of Cell Physiology and Metabolism, University Medical Center, Geneva, Switzerland) for excellent assistance and advice.