Glucose-induced toxicity in insulin-producing pituitary cells that coexpress GLUT2 and glucokinase. Implications for metabolic engineering.

We have shown that intermediate lobe (IL) pituitary cells can be engineered to produce sufficient amounts of insulin (ins) to cure diabetes in nonobese diabetic mice but, unlike transplanted islets, ILins cells evade immune attack. To confer glucose-sensing capabilities into these cells, they were further modified with recombinant adenoviruses to express high levels of GLUT2 and the beta-cell isoform of glucokinase (GK). Although expression of GLUT2 alone had negligible effects on glucose usage and lactate production, expression of GK alone resulted in approximately 2-fold increase in glycolytic flux within the physiological (3-20 mm) glucose range. GLUT2/GK coexpression further increased glycolytic flux at 20 mm glucose but disproportionately increased flux at 3 mm glucose. Despite enhanced glycolytic fluxes, GLUT2/GK-coexpressing cells showed glucose dose-dependent accumulation of hexose phosphates, depletion of intracellular ATP, and severe apoptotic cell death. These studies demonstrate that glucose-sensing properties can be introduced into non-islet cells by the single expression of GK and that glucose responsiveness can be augmented by the coexpression of GLUT2. However, in the metabolic engineering of surrogate beta cells, it is critical that the levels of the components be closely optimized to ensure their physiological function and to avoid the deleterious consequences of glucose-induced toxicity.

Type 1 diabetes in humans and in nonobese diabetic (NOD) 1 mice is an autoimmune disorder that results in the complete destruction of the insulin-producing pancreatic ␤ cells. Despite recent advances in insulin injection therapy, current regimens do not adequately mimic the normal physiological patterns of insulin release by ␤ cells, and patients with diabetes are thus at continued risk for developing severe long-term complications. These limitations have provided the impetus to develop alternative forms of therapy that would circumvent the need for insulin injections.
One appealing therapeutic strategy is to engineer insulin expression into non-␤ cells (1)(2)(3). Using transgenic mouse techniques, we targeted mouse preproinsulin 2 (ins) gene expression to the intermediate lobe (IL) of the pituitary in NOD mice via the proopiomelanocortin (POMC) promoter (1). Our studies showed that the IL cells secreted large amounts of insulin, sufficient to cure diabetes when implanted into spontaneously diabetic NOD mice. Moreover, in contrast to transplanted islets, the insulin-producing IL (hereafter, ILins) cells evaded immune recognition and damage. These studies were the first to show that a non-islet cell type could be engineered to secrete sufficient amounts of insulin to cure diabetes, yet escape the autoimmune process that kills ␤ cells in type 1 diabetes (1). However, a limitation of this insulin gene delivery system was that insulin secretion was not glucose-regulated.
The regulation of insulin release in ␤ cells is mediated by the metabolism of glucose (4). According to current models, glucose metabolism generates signals such as ATP or an increase in the ratio of ATP to ADP ratio that leads to closure of the ATPsensitive K ϩ (K ATP ) channels. The resulting plasma membrane depolarization activates voltage-gated L-type Ca 2ϩ channels, inducing the influx of Ca 2 that triggers insulin granule exocytosis (5). The proteins that control the metabolic flux of glucose in ␤ cells function as "glucose sensors." Rodent ␤ cells express two specialized proteins that have been considered as candidates for the glucose sensor: the facilitated glucose transporter isoform, GLUT2, and the low affinity glucose phosphorylating enzyme, glucokinase (GK). A large body of evidence suggests that GK is the flux-controlling enzyme for glycolysis in ␤ cells and, as such, serves as the "gatekeeper" for metabolic signaling (4). GK is unique among mammalian hexokinases in having a K m for glucose in the physiological (8 -10 mM) glucose range (6).  The high capacity glucose transporter, GLUT2, is also unique in having a high K m (ϳ17 mM) for glucose (7) but is believed to play a more "permissive" role in glucose sensing by allowing the rapid equilibration of glucose across the plasma membrane (8). However, several studies have suggested that the expression of GLUT2 is required for conferring glucose-sensing capabilities into non-␤ cell lines (9,10). In addition, it has been observed that GLUT2 and GK are coexpressed not only in ␤ cells and hepatocytes but also in glucose-responsive neurons in the hypothalamus and the gut (11), further suggesting that GLUT2 may be an important component of the glucose-sensing apparatus in non-islet cells.
It has been suggested that the simple "iterative engineering" of glucose-sensing components into cells or cell lines may simulate the performance of normal islet ␤ cells (12). The goal of this study was to determine whether the expression of GK or GLUT2, alone or in combination, would confer glucose-sensing capabilities into ILins cells. We found that the single expression of GK conferred glucose-sensing capabilities in the physiological glucose concentration range and that glycolytic flux was increased by the coexpression of GLUT2. However, unexpectedly, these manipulations also resulted in a state of glucose-induced toxicity with severe apoptotic cell death. These findings have important implications for future strategies to metabolically engineer surrogate ␤ cells.

EXPERIMENTAL PROCEDURES
Construction of Recombinant Adenoviruses-Recombinant adenoviruses were generated using the strategy of Cre-lox recombination (13). A 1.7-kb XbaI/SmaI cDNA fragment encoding the rat GLUT2 protein (14) or a 1.45-kb KpnI/XbaI cDNA fragment encoding the islet isoform of GK (GK.B1 (15)) were inserted into the polylinker of the shuttle vector pAdMDM (13). To construct the GK-GFP fusion protein, a PstI/ BamHI fragment containing part of the GK.B1 cDNA was first subcloned into the plasmid pEGFP-N3 (CLONTECH) resulting in pGK.B1PstI/pGFP. The remaining KpnI/PstI fragment of the GK.B1 cDNA was then ligated into pGK.B1PstI/GFP resulting in pGK.B1/ GFP. Fusion in-frame and orientation was confirmed by sequence analysis. An XhoI-NotI fragment containing the GK.B1-GFP fusion protein was then ligated into pAdMDM.
Cell Culture and Adenovirus Infection-Pituitary tissues were excised from POMC-ins transgenic NOD mice (1), and single cell suspensions were prepared as described previously (16). Cells were cultured in RPMI 1640 medium with 10% fetal bovine serum (Life Technologies, Inc.), 100 units/ml penicillin, and 100 g/ml streptomycin. Cells were incubated at 37°C, 5% CO 2 for 7-10 days prior to adenoviral infection. The multiplicity of infection (m.o.i.) was calculated based on estimates that the cell number doubled from the time of culture to the time of infection. An m.o.i. of 1 was used for all experiments unless otherwise indicated. Adenoviruses were added in a minimum volume of RPMI medium to cover the cells. After incubation at 37°C for 2 h, the viruses were aspirated and the cells were cultured for 48 h.
TUNEL Labeling and Double Immunofluorescence Microscopy-Cells (2.5 ϫ 10 5 /well) were grown in duplicate on SonicSeal chamber slides (NUNC) for 7-10 days. After adenovirus infection and 48-h culture in glucose-free RPMI containing various glucose concentrations (3, 10, or 20 mM glucose), cells were washed with phosphate-buffered saline and DNA strand breaks were detected with a commercially available kit (In Situ Cell Death Detection Kit, TMR red, Roche Molecular Biochemicals) following the manufacturer's instructions. Fluorescence was monitored with an Olympus IX 70 inverted tissue culture microscope, and images were captured on an AxioCam charge-coupled device camera equipped with AxioVision 2.05 software (Zeiss). The GFP channel was visualized using an Endow GFP bandpass emission filter cube (Chroma); the TUNEL-TMR was visualized using a Cy3 with a narrow band excitation filter set (Chroma). Multichannel images were composed in AxioVision, by exposing each channel individually and then superimposing the two channels to create merged images. Automatic exposure times for the GFP channel were adjusted to visualize the extremely dim GFP fluorescence of the GK-GFP/GLUT2-infected cells at 20 mM. These exposure times were kept constant throughout the image acquisition depicted in Fig. 5. To quantitate apoptosis, we counted labeled (GFP and TUNEL) cells in at least three separate fields; the number of double-labeled cells was expressed as a percentage of the total number of GFP-expressing cells.
Northern Blot Analysis-Total RNA was extracted from mouse tissues and the MIN6 cell line (17) with TRIzol (Life Technologies, Inc.). Northern blot analysis was carried out as described previously (1). Blots were sequentially hybridized with random-primed [ 32 P]dCTP radiolabeled probes encoding GLUT1, GLUT2 (14), GK, insulin (1), and ␤-actin, with stripping of the blots between hybridizations. The GLUT1 probe was generated by reverse transcription-polymerase chain reaction with mouse brain RNA as a template with the forward primer 5Ј-ACCTGCAGGAGATGAAAGAAGAGG-3Ј and the reverse primer 5Ј-TGGAAGAGCTCCTCAGGTGTCTTG-3Ј. The GK probe, containing sequence common to islet and liver GK mRNA (positions 587-1254), was generated from mouse liver RNA with the forward primer 5Ј-TCTG-GAGCAGAAGGGAACAACATC-3Ј, corresponding to exon 5, and the reverse primer 5Ј-GCTTTCGCGCATGCGATTTATGAC-3Ј, corresponding to exon 9. The membranes were exposed to x-ray film at Ϫ80°C for 2 h to 5 days.
Western Blot Analysis-Immunoblotting was performed as previously described (15). A sheep anti-rat GK antiserum that was raised against the common regions of the liver and islet GK protein (15) diluted at 1:10,000, or a rabbit anti-rat carboxyl-terminal GLUT2 antiserum (East Acres Biologicals), diluted at 1:500, was used to detect GK or GLUT2, respectively. The bound antibodies were visualized with horseradish peroxidase-conjugated secondary antibodies diluted at 1:10,000 and a chemiluminescent substrate (ECL, Amersham Pharmacia Biotech).
Metabolic Assays-Metabolic assays and insulin secretion studies were performed 48 h after adenovirus infection. Immediately prior to the assays, cells were washed and preincubated for 1 h in glucose-free Hanks' modified balanced buffer (137 mM NaCl, 4.8 mM KCl, 2.5 mM CaCl 2 , 1.2 mM MgCl 2 , 0.31 mM Na 2 HPO 4 , 0.4 mM KH 2 PO 4 , 5 mM NaHCO 3 , 20 mM Hepes, pH 7.4) with 0.5% (w/v) bovine serum albumin (Fraction V, Sigma Chemical Co.). At the end of the experiments, cells were solubilized in 0.1 N NaOH, and the protein content was determined with a Bio-Rad protein assay kit.
Glucose Phosphorylating Activity-Cells were cultured at 1 ϫ 10 6 cells/well in 6-well plates. After trypsinization and washing in glucosefree phosphate-buffered saline, cells were homogenized on ice in buffer (6,15). The supernatant fractions were assayed for glucose-phosphorylating activity in buffer containing 0.03-0.5 mM and 6 -100 mM glucose by a fluorometric method (6,15). The values obtained were extrapolated to a temperature of 37°C assuming a Q 10 of 2. The K m and V max were calculated from a Hanes plot.
Glucose Usage-Cells, cultured at 2 ϫ 10 5 cells/well in 24-well plates, were incubated for 30 min with Hanks' buffer containing 0.3, 3, or 20 mM glucose and 2 Ci of D-[5-3 H]glucose or D-[2-3 H]glucose (PerkinElmer Life Sciences). The production of 3 H 2 O was measured by a radiometric method based on water diffusion as described previously (18), and glucose usage was calculated as described (19).
Insulin Secretion Studies-ILins cells, cultured in 96-well plates at 5 ϫ 10 4 cells/well, were incubated in buffer containing 3 or 20 mM glucose for 15 min. Supernatants were centrifuged and assayed for rat insulin by radioimmunoassay (Linco Research). Cellular insulin content was measured after acid ethanol extraction. For perifusion studies, 2 ϫ 10 5 ILins cells were cultured on a 10.5-mm polyethylene terephthalate membrane (Becton Dickinson) for 10 -14 days. The membranes were then transferred into perifusion micro-chambers. Cells were perifused at 0.5 ml/min in 0.5% bovine serum albumin-modified Hanks' buffer containing 3 mM glucose and were stimulated for 10 min with 50 mM KCl, and 10 or 100 M acetylcholine (Sigma). Effluents were collected in 1-ml aliquots and were analyzed for insulin by radioimmunoassay.
Assays of Intracellular Metabolites-Cells (1 ϫ 10 6 cells/well in 6-well plates) were incubated for 1 h in buffer containing 3 or 20 mM glucose.
After removal of the incubation buffer, samples were deproteinized with perchloric acid and assayed for glucose 6-phosphate (G6P), fructose 1,6-bisphosphate (F1,6BP), and ATP by enzymatic methods (20) using a Hewlett-Packard model 8450 spectrophotometer system set to measure A 335-345 to A 390 -400 . Lactate released into the incubation buffer was similarly assayed.
Statistical Analysis-Values are expressed as mean Ϯ S.E. Statistical analysis was performed with unpaired t test or analysis of variance. Differences were taken to be statistically significant at p Ͻ 0.05.

Characterization of the Endogenous Metabolic Machinery in IL Pituitary
Cells-Northern blot analysis (Fig. 1A) revealed that IL cells did not express detectable GLUT2 mRNA transcripts, but they expressed the low K m erythroid/brain glucose transporter, GLUT1, at levels similar to brain. Further stripping and re-probing this blot with a 660-bp DNA fragment containing part of the GK sequence that is common to both liver and islets showed that IL cells expressed a 2.8-kb transcript of similar size and abundance to that of anterior lobe (AL) pituitary and MIN6 cells, compared with the smaller 2.4-kb transcript in liver. However, subsequent assays showed that IL cells did not express functional GK activity, suggesting that tissue-specific alternative mRNA splicing events may disrupt the GK reading frame in IL cells as has been described in AL cells (15,21). Hybridization with an insulin probe showed large amounts of insulin transcripts after only a 2-h exposure in ILins cells similar to MIN6 cells, attesting to the high levels of insulin production in these transgenically engineered cells (1). Finally, in contrast to AL cells and the AL corticotroph cell line AtT-20, IL cells did not express mRNA for either Kir6.2 or SUR1 (22), which together comprise the ␤ cell K ATP channel complex (data not shown).
Increases in the cytosolic Ca 2ϩ concentration constitute a main trigger for insulin exocytosis in ␤ cells (23). To assess whether insulin secretion in ILins cells would be stimulated by agents that increase intracellular Ca 2ϩ and to characterize the magnitude and dynamics of this process, perifusion studies were performed in which cells were exposed sequentially to high concentrations of KCl (50 mM) and the neurotransmitter, acetylcholine (ACh), 10 M and 100 M (24). Depolarization of ILins cells with KCl caused a brisk and potent secretory response, with insulin secretion ϳ10-fold over the basal levels (Fig. 1B). Further stimulation with ACh elicited a ϳ4-fold increase in insulin secretion over the basal levels. Thus, ILins cells could be stimulated to secrete insulin in a Ca 2ϩ -dependent manner.
ILins Cells Are Efficiently Infected with Recombinant Adenoviruses-To test the feasibility of using adenoviral vectors as a vehicle for introducing glucose-sensing genes into ILins cells, primary cultures were treated with recombinant adenovirus containing the ␤-galactosidase reporter gene LacZ (13). Subsequent X-gal staining revealed that an m.o.i. of 10 viral particles per cell resulted in an infection of more than 90% of cells, and that an m.o.i. of 1 resulted in an infection of ϳ30 -50% of cells (data not shown). To determine whether adenovirus was capable of specifically infecting ILins cells, primary cultures were transduced with recombinant adenovirus containing a human growth hormone (hGH) reporter gene. Double immunolabeling of ILins cells with antibodies to hGH and insulin showed colocalization of hGH and insulin expression in the majority of ILins cells (Fig. 1C); as expected, because GH-positive cells are never normally part of the IL pituitary, there was no specific GH staining in uninfected ILins cells.
Expression of Functional GLUT2 and GK Proteins-Because IL cells did not express GLUT2 or functional GK proteins, recombinant adenoviruses were constructed that contained the cDNA encoding GLUT2 (AdGLUT2) or the functional islet isoform of glucokinase (AdGK) under the control of the CMV promoter. Western blot analysis showed that the transduction of primary pituitary cell cultures with either AdGK or Ad-GLUT2 resulted in extremely large increases in immunoreactive GK or GLUT2 proteins, respectively ( Fig. 2A).
Cells treated with AdGLUT2 showed a concentrationdependent increase in 3-O-methyl-D-glucose uptake, in contrast to cells treated with the control LacZ-containing adenovirus that only express GLUT1 (Fig. 2B). LacZ-infected IL cells showed low K m glucose phosphorylating activity (K m ϭ 41 Ϯ 3.5 FIG. 1. A, Northern blot analysis of liver, brain, lung, MIN6, anterior lobe pituitary (AL), and ILins cells. Ten micrograms of total cellular RNA was loaded for each sample, and the blot was sequentially hybridized to 32 P-labeled GLUT1, GLUT2, glucokinase, insulin, and ␤-actin probes. B, representative perifusion of ILins cells demonstrating Ca 2ϩ -dependent insulin secretion. ILins cells were perifused at 0.5 ml/min with buffer containing 3 mM glucose and were sequentially stimulated with 10-min pulses (indicated by the bar) of buffer containing 50 mM KCl, 10 M acetylcholine (ACh), and 100 M ACh. Successive 1-min fractions were collected. C, recombinant adenoviruses infect ILins cells with high efficiency. Primary cultures of ILins cells were grown on chamber culture slides for 10 days and treated with adenovirus containing a hGH reporter gene. Indirect double-immunofluorescence staining with insulin (Ins) and human growth hormone (hGH) antibodies demonstrated colocalization of GH (green, right) and insulin (Cy3, left) staining in virtually all ILins cells.
M, V max ϭ 55 Ϯ 5 nmol/min/mg of protein, n ϭ 7 separate experiments), similar to non-infected parental IL cells. In contrast, IL cells infected with the GK-containing adenovirus exhibited high levels of functional GK activity (K m ϭ 8.5 Ϯ 0.5 mM, V max ϭ 2.08 Ϯ 0.13 mol/min/mg of protein, n ϭ 5 separate experiments), which comprised ϳ98% of the total glucose phosphorylating activity, comparing the phosphorylation rates at 100 and 0.5 mM glucose.
Glucose-sensing Capabilities in IL Cells Expressing GLUT2 and GK-To investigate the metabolic impact of expressing GLUT2 and GK individually and in combination, we measured the production of 3

Impairment in Insulin Secretion and Loss of Cell Viability by Apoptosis in IL Cells
Coexpressing GLUT2 and GK-To assess whether the increased glucose usage in metabolically engineered ILins cells was coupled to glucose-stimulated insulin secretion (GSIS), we examined insulin release in response to glucose challenge under static incubation conditions. As previously noted, IL cells lack K ATP channels but secrete abundant amounts of insulin in response to agents that increase intracellular Ca 2ϩ (Fig. 1B). In ␤ cells, GSIS can occur in the absence of functional K ATP channels (by the so-called "K ATPchannel-independent pathway") when intracellular Ca 2ϩ is raised (25). However, ILins cells failed to respond to glucose even when these agents were provided. Instead, as seen in Fig.  4, the insulin secretion in cells that coexpressed GK and GLUT2 was severely reduced (45%) compared with control infected cells (p Ͻ 0.05). These findings raised the possibility that the coexpression of GLUT2 and GK might be having adverse effects on cell viability.
To visualize the fate of the virally infected cells, we constructed recombinant adenoviruses expressing green fluorescence protein (GFP) alone (AdGFP) or GFP fused in-frame to the carboxyl terminus of GK.B1 (AdGK-GFP). The activity of AdGK-GFP was identical to that of native AdGK virus by glucose phosphorylation and glucose usage assays (data not shown). To assess whether there was toxicity due to nonspecific effects of adenoviral coinfection or due to the specific combination of virus used, we infected cells with various combinations of viruses at equivalent m.o.i. As can be seen in Fig. 5A, cells that were coinfected with GFP/LacZ-, GK-GFP/LacZ-, or GFP/ GLUT2-expressing adenoviruses showed equivalently bright fluorescence intensity with minimal evidence of cell damage. In marked contrast, cells that were coinfected with GLUT2/GK-GFP-containing adenoviruses exhibited severely reduced GFP fluorescence intensity, plasma membrane blebbing, and cell fragmentation, suggestive of apoptosis. These studies demonstrate that the cell toxicity observed was due to the specific coexpression of GLUT2 and GK.
In all the studies described so far, IL cells were cultured in RPMI 1640 medium (containing 11 mM glucose) for 48 h to allow for expression of the adenoviruses. To determine whether the loss of cell viability was glucose dose-dependent and to further characterize the mode of cell death, cells were infected and cultured for 48 h in glucose-free RPMI 1640 medium to which 3, 10, or 20 mM was added. Apoptosis was assessed by a TUNEL assay to identify DNA fragmentation. As shown in Fig.  5B, cells coexpressing control GFP/LacZ viruses were microscopically normal at all glucose concentrations, whereas cells coexpressing the GK-GFP/LacZ viruses showed some mild loss of cell viability, but this occurred only at high (20 mM) glucose concentrations. No significant TUNEL labeling was seen in GFP-positive cells with either of these virus combinations, even at high glucose concentrations. In contrast, the cells that were coinfected with GK-GFP/GLUT2-containing adenoviruses showed a glucose dose-dependent loss of cell viability by apoptosis. At 3 mM glucose, GK-GFP/GLUT2-expressing cells appeared relatively healthy with only occasional TUNEL-positive cells, but, at 10 mM glucose, there was severe loss of cell viability with decreased GFP fluorescence intensity and a dramatic increase in TUNEL-positive cells. These cells also appeared shrunken with plasma membrane blebbing and densely fragmented nuclei. The changes became further pronounced when cells were cultured at 20 mM glucose. In GK-GFP/ GLUT2-expressing cells, the percentage of GFP/TUNEL-TMR double-labeled cells increased from 3.5% (3 mM glucose) to 33% (10 mM glucose) to 41% (20 mM glucose), compared with the GFP/LacZ and GK-GFP/LacZ cells that showed fewer than 1% double-labeled cells within this glucose range.
Accumulation of Hexose Phosphates and Compromise of Cellular Energy Metabolism-To examine the biochemical basis for the lack of GSIS and the loss of cell viability, cells were virally infected and cultured at low (3 mM) glucose for 48 h. The cells were then exposed for 1 h to fresh buffer containing either 3 or 20 mM glucose, and the intracellular metabolites were assayed. These studies showed (Fig. 6, top panel) that, when challenged with high (20 mM) glucose, the GK-expressing cells exhibited ϳ3-fold increase in intracellular F1,6BP compared with AdGLUT-2 or control infected cells that were unchanged (p Ͻ 0.001). In contrast, in cells coexpressing GLUT2 and GK, F1,6BP levels were ϳ5-fold increased at low (3 mM) glucose (p Ͻ 0.001) and were ϳ7-fold increased at 20 mM glucose compared with control cells (p Ͻ 0.001). The levels of G6P in the virally transduced cells were ϳ5-fold lower than the levels of F1,6BP but otherwise showed a similar pattern of accumulation (data not shown). Intracellular ATP levels changed in the opposite direction to the levels of F1,6BP (Fig. 6, bottom  panel). When GK-expressing cells were challenged with 20 mM glucose, ATP levels decreased by 35% compared with control cells (p Ͻ 0.001) that remained unchanged. In GLUT2/GKcoexpressing cells, ATP levels decreased by 54 and 80% at 3 and 20 mM glucose, respectively, compared with control cells (p Ͻ 0.001). Of note, lactate release was in general agreement with the glucose usage data (Fig. 3), with the control GFP-and GLUT2-expressing cells showing similar lactate production with no increases from 3 to 20 mM glucose, the GK-GFPexpressing cells showing a 1.6-fold increase from 3 to 20 mM glucose in (12 Ϯ 1.8 to 19 Ϯ 2.1 nmol/min/mg of protein, p Ͻ .05), and the GK-GFP/GLUT2-coexpressing cells showing increased lactate production both at 3 and at 20 mM glucose (17 Ϯ 1.3 and 20 Ϯ 0.9 nmol/min/mg of protein, p Ͻ .05, compared with control GFP-infected cells). DISCUSSION We demonstrate that ILins cells can be engineered to express extremely high levels of functional GK and GLUT2 proteins. The expression of GLUT2 alone enhanced glucose usage only at very low glucose concentrations (i.e. 0.3 mM glucose), similar to a previous report (26). In contrast, the single expression of GK in ILins cells increased glucose usage and lactate production in the physiological glucose range with a 2-fold increase from 3 to 20 mM glucose, relative to control cells (p Ͻ 0.001). These results are consistent with the known regulation of GK enzyme activity that occurs over the same glucose range. The coexpression of GLUT2 and GK further increased glucose usage at 20 mM glucose, but unexpectedly, this was accompanied by a leftward shift in the glucose dose-response curve with disproportionately greater glycolytic flux at 3 mM glucose. Despite the stimulation of glycolysis in these engineered cells, insulin secretion was reduced with a pronounced accumulation of hexose phosphates, depletion of ATP, and apoptotic cell death.
Why did glycolysis and energy production go awry in these metabolically engineered cells? One factor may relate to the inherent organization of the glycolytic pathway. The term "turbo design" has been coined to describe the organization of many catabolic pathways that begin with one or more ATP-consuming steps, after which further metabolism and ATP-generating reactions yield a net production of ATP across the pathway (27). Turbo design is exemplified in glycolysis, in which two molecules of ATP are initially invested at the hexokinase (HK) and phosphofructokinase (PFK) steps as a prelude to the net synthesis of four molecules of ATP further down the pathway. This design feature makes tight regulation of the enzymes involved in the initial ATP-consuming steps absolutely essential (27). In mammalian cells the activity of HK is inhibited by its product, glucose 6-phosphate (G6P). However, GK is not subject to such feedback inhibition. Therefore, when GK is expressed at high levels, there is the risk of uncontrolled accumulation of G6P, especially at high glucose levels. In addition, because G6P is in equilibrium with fructose 6-phosphate, the fructose 6-phosphate would be expected to rise and increase flux through PFK.
In these studies, the levels of F1,6BP were ϳ5-fold higher than those of G6P but otherwise showed similar patterns of accumulation. A number of factors may contribute to the preferential accumulation of F1,6BP. First, PFK would be expected to be activated allosterically by the consumption of its inhibitor ATP (and presumably, the corresponding rise in the PFK activator AMP) during glucose phosphorylation. Second, because the product of the PFK reaction, F1,6BP, is itself a potent allosteric activator of PFK, the rise in F1,6BP levels may be autocatalytic. Third, because F1,6BP and glyceraldehyde 3-phosphate (GAP) and dihydroxyacetone phosphate form an equilibrium pool through the rapid aldolase and triose phosphate isomerase reactions, the F1,6BP concentration varies as the square of the GAP concentration (28), so there may be a disproportionate rise in F1,6BP as glycolytic flux increases. Finally, as more F1,6BP and GAP accumulate, the GAP concentration may approach or surpass the K m of GAPDH, further diminishing the responsiveness of GAPDH to the increased metabolic flux. It should be noted that the maximum accumulated F1,6BP is equivalent to only about 7% of the glucose utilization (or the lactate production), suggesting that there may not be a major block at the reaction catalyzed by GAPDH. However, accumulated F1,6BP is equivalent to nearly twice the normal level of intracellular ATP, and therefore, the sequestration of phosphate in F1,6BP and G6P likely accounts for the drop in cellular ATP levels. Admittedly, our measurements of these metabolites were at the end of 1 h of incubation, and it is possible that these changes did not occur at a constant rate but that the rise in F1,6BP and loss of ATP occurred more acutely toward the end of the incubation period such that phosphate depletion did inhibit the flow through GAPDH (which uses P i as a substrate) and the glycolytic ATP production. Regardless of the exact timing, it is likely that the sequestration of phosphate in G6P and F1,6BP accounts for the ATP depletion and loss of cell viability.
It has been suggested that the reason why GK is not hazardous to cell types where it is normally expressed, such as ␤ cells and hepatocytes, is that in the former, GK activity is the lowest of the activities of all glycolytic enzymes and, as such, constitutes the "rate-limiting" step of glycolysis (29); whereas in hepatocytes, GK is subject to a number of regulatory influences, including the GK regulatory protein. Furthermore, hepatocytes have additional reactions (e.g. glucose cycling through glucose-6-phosphatase and glycogen synthesis) that may prevent the excessive accumulation of glycolytic intermediates (30).
A striking feature of the cells that coexpressed GLUT2 and GK was the leftward shift in their glucose-response threshold with a disproportionate increase in glycolytic flux at 3 mM glucose. These findings were unexpected in view of the high K m of GLUT2 and GK and could not be predicted from studying the metabolic impact of each gene individually. A simple explanation is that when GK is expressed alone, endogenous GLUT1 limits the flux of glucose through the cells; but when GLUT2 and GK are coexpressed, neither transport nor phosphorylation is rate-limiting. However, it seems unlikely that glucose transport is rate-limiting, because even in LacZ-infected cells uptake at 3 mM glucose was ϳ2-fold higher than the glucose usage in cells coexpressing GLUT2 and GK (10.2 Ϯ 0.9 versus 5.6 Ϯ 0.4 nmol/min/mg of protein, respectively). These data suggest that in this setting there may be a direct interaction between GLUT2 and GK (31) or, more likely, in combination they may affect the activities of other components involved in the upper part of the glycolytic pathway (32).
Similar metabolic perturbations, i.e. activation of the first steps of glycolysis with the accumulation of hexose phosphates and loss of ATP, have been reported in several other systems. These include Saccharomyces cerevisiae yeast mutants that are unable to synthesize the HK inhibitor trehalose 6-phosphate (27), trypanosomes in which glycolysis is not compartmentalized (33), healthy subjects who received high doses of parenteral fructose, which is phosphorylated by fructokinase that bypasses the regulated HK step (34), and the INS-1 ␤-cell line that was engineered to overexpress large amounts of GK (35,36). In the latter setting, control of glycolysis at the GK step was also lost, with a marked increase in glucose usage at low (2.5 mM) glucose concentrations. These effects were attributed to high level GK expression alone (35), but our findings raise the possibility that the endogenous GLUT2 may have also adversely contributed to this phenotype.
Another remarkable feature of GLUT2-and GK-coexpressing cells was that, despite exhibiting an acute 80% drop in intracellular ATP levels at high glucose levels, these cells showed pronounced amounts of apoptotic cell death. These findings were unexpected, because apoptosis is an energyrequiring process and progression to necrotic or apoptotic cell death is thought to depend in part on the cellular ATP content, with rapid ATP depletion usually resulting in necrosis (37). These studies suggest that, although the ATP levels in GLUT2/ GK-coexpressing cells are severely reduced, these ATP levels are still sufficient to complete the apoptotic program. Recent studies have suggested that glucose may induce apoptosis in pancreatic ␤ cells (38,39). Although caution must be used against extrapolating our data to islets, these findings suggest that GLUT2 and GK may be involved in a glucose signaling pathway that, when imbalanced, results in metabolic toxicity and apoptotic cell death.
It has been suggested that the simple introduction of glucosesensing components into cells or cell lines may simulate the performance of normal islet ␤ cells. These findings raise important caveats to this notion and demonstrate the deleterious consequences that can result when the expression levels of the metabolic components are not closely optimized. We must emphasize that the metabolic toxicity described in this report was not due to the high level expression achieved with this particular viral gene delivery system. We have recently created a series of transgenic mouse lines that coexpress GLUT2 and GK in IL cells at levels that are markedly lower than those achieved with adenoviruses. Interestingly, the size of the IL tissues in these transgenic mice corresponded inversely to the levels of GK expression, with the most severe reduction in the size of the IL tissues in the lines that expressed the highest levels of GK. 2 Although the mechanism by which glucose stimulates insulin secretion in ␤ cells is complex and clearly requires more than just the expression of GLUT2 and GK, these transgenic tissues should provide an excellent starting material for determining the requirements for optimal glucose-sensing in surrogate ␤ cells.