Glucose-stimulated Single Pancreatic Islets Sustain Increased Cytosolic ATP Levels during Initial Ca2+ Influx and Subsequent Ca2+ Oscillations*

Background: The roles of ATP in glucose-stimulated insulin secretion have not been well described. Results: After glucose stimulation, ATP levels were elevated prior to an increase in Ca2+ levels. High nonoscillatory ATP levels were sustained during Ca2+ oscillations. Conclusion: High ATP levels may be necessary for initial Ca2+ influx and subsequent Ca2+ oscillations. Significance: This adds new insight into the mechanism of insulin secretion. In pancreatic islets, insulin secretion occurs via synchronous elevation of Ca2+ levels throughout the islets during high glucose conditions. This Ca2+ elevation has two phases: a quick increase, observed after the glucose stimulus, followed by prolonged oscillations. In these processes, the elevation of intracellular ATP levels generated from glucose is assumed to inhibit ATP-sensitive K+ channels, leading to the depolarization of membranes, which in turn induces Ca2+ elevation in the islets. However, little is known about the dynamics of intracellular ATP levels and their correlation with Ca2+ levels in the islets in response to changing glucose levels. In this study, a genetically encoded fluorescent biosensor for ATP and a fluorescent Ca2+ dye were employed to simultaneously monitor the dynamics of intracellular ATP and Ca2+ levels, respectively, inside single isolated islets. We observed rapid increases in cytosolic and mitochondrial ATP levels after stimulation with glucose, as well as with methyl pyruvate or leucine/glutamine. High ATP levels were sustained as long as high glucose levels persisted. Inhibition of ATP production suppressed the initial Ca2+ increase, suggesting that enhanced energy metabolism triggers the initial phase of Ca2+ influx. On the other hand, cytosolic ATP levels did not fluctuate significantly with the Ca2+ level in the subsequent oscillation phases. Importantly, Ca2+ oscillations stopped immediately before ATP levels decreased significantly. These results might explain how food or glucose intake evokes insulin secretion and how the resulting decrease in plasma glucose levels leads to cessation of secretion.

occur during extraction from the cells. Thus, ATP levels might be underestimated in the conventional ATP assay. To understand the actual dynamics of intracellular ATP during GSIS, a nondestructive method of measuring intracellular ATP is required. Recently, a genetically encoded fluorescent biosensor for the ATP/ADP ratio was employed in live pancreatic islets (14,15). However, the dynamics of [ATP] c itself and its correlation with [Ca 2ϩ ] c in the initial phase have not been well investigated. Moreover, it remains unclear how [ATP] c couples with [Ca 2ϩ ] c oscillations in the second phase.
We previously developed the genetically encoded fluorescent ATP biosensors ATeam (16) and GO-ATeam (17), which are based on the principle of FRET and insensitive to the related nucleotides, namely ADP, GTP, and dATP (16,17). Because GO-ATeam has little spectral overlap with the Ca 2ϩ -sensitive dye fura-2, it has allowed us to monitor ATP levels and Ca 2ϩ levels simultaneously in single living cells (17). In this paper, we visualized [ATP] c or mitochondrial ATP levels ([ATP] m ) together with [Ca 2ϩ ] c in each isolated islet by using GO-ATeam and fura-2 and examined how intracellular ATP dynamics regulate [Ca 2ϩ ] c in GSIS.
Isolation and Culture of Mouse Pancreatic Islets-Islets were isolated from the pancreases of C57BL/6J mice with the aid of collagenase (Roche Applied Science) according to the general protocol. In brief, the pancreas was injected with a solution of 0.5 mg/ml collagenase via the bile duct. After collagenase digestion for 40 min at 37°C and two or three wash steps, density gradient centrifugation was performed using Ficoll. Islets were collected manually and cultured in RPMI 1640 (Nacalai Tesque) containing 11 mM glucose at 37°C in a humidified atmosphere containing 5% CO 2 for 2-3 days. The culture medium was supplemented with 10% fetal calf serum (Nichirei Biosciences Inc., Tokyo, Japan), 100 units/ml penicillin, and 100 g/ml streptomycin.
Generation of Recombinant Adenovirus-Adenovirus for the expression of GO-ATeam biosensors was generated using the ViraPower TM adenoviral expression system (Invitrogen) according to the manufacturer's instructions. For construction of the adenoviral vector encoding GO-ATeam, the XhoI-PmeI fragment of pcDNA-GO-ATeam1 or pcDNA-mit-GO-ATeam1 (17) was subcloned into the pENTR-1A vector (Invitrogen), digested with SalI and EcoRV, and then recombined into the pAd/CMV/ V5-DEST destination vector.
Permeabilization of the Plasma Membrane-Plasma membranes of cells were permeabilized with 100 g/ml ␣-hemolysin from Staphylococcus aureus (Sigma) for 30 min. After permeabilization, cells were perfused with intracellular-like medium containing 140 mM KCl, 6 mM NaCl, 1 mM MgCl 2 , 0.465 mM CaCl 2 , 2 mM EGTA, and 12.5 mM HEPES (pH 7.0) and different concentrations of MgATP.  9). B, after permeabilization, cells were alternately perfused with intracellular-like medium including 7 or 8 mM MgATP (n ϭ 10). C, average time course of the fluorescence emission ratio (OFP/GFP) of MIN6 cells expressing GO-ATeam1 biosensors in the cytosol. OFP/GFP ratios were monitored when medium glucose was increased from 2.8 to 25 mM (n ϭ 22). Cells that did not exhibit clear increases in response to changing glucose levels were excluded from the data analysis. The error bars indicate S.D.
Imaging and Data Analysis-Islets were plated on a glassbottomed dish (Fine Plus International Ltd., Kyoto, Japan) and incubated at 37°C. For ATP imaging, ATeam1.03 (16) or a series of GO-ATeams (17) were transfected into islets using an adenoviral vector system. Before imaging, fura-2-AM was added to a final concentration of 2-5 M for 30  ; and for dual excitation ratio imaging of fura-2, 340/26 (fura-2S) or 387/11 (fura-2L) -DM400 -510/84. An ORCA-AG cooled CCD camera (Hamamatsu Photonics, Hamamatsu, Japan) was used to capture fluorescent images. The microscope system was controlled with NIS-Elements (Nikon). Imaging data were analyzed using MetaMorph (Molecular Devices, Sunnyvale, CA). Cer-tain parts of the islets were selected to quantify the average intensities of signals because of the nonuniform expression levels of ATP biosensors within an islet. Cross-correlation analysis was performed as reported previously (19).

Properties of GO-ATeam1 in Insulin-secreting Cells-The
GO-ATeam1 biosensor is composed of the ⑀ subunit of Bacillus subtilis F o F 1 -ATP synthase, a variant of Aequoria GFP (cp173-mEGFP), and a variant of Fungia orange fluorescent protein (OFP; mKO) (17). By inducing a conformational change in the ⑀ subunit, ATP binding to GO-ATeam1 increases FRET from GFP to OFP, thereby increasing the OFP/GFP emission ratio (FRET signal). To test whether GO-ATeam1 can sense ATP changes within insulin-secreting cells, we recorded FRET signals from GO-ATeam1-expressing MIN6 insulinoma cells, which were permeabilized with ␣-hemolysin. When the ATP concentration in the medium was altered, the FRET signal changed in an ATP concentration-dependent manner (Fig. 1A). Thus, GO-ATeam1 works properly in insulin-secreting cells. Moreover, we could monitor changes in the FRET signal when ATP levels were alternated between 7 and 8 mM (Fig. 1B), indicating that our system has the We next expressed GO-ATeam1 in the cytosol of isolated mouse pancreatic islets. When a GO-ATeam1-expressing islet was stimulated by increasing the glucose concentration in the medium from low (2.8 mM) to high (25 mM), rapid increases in the FRET signal were observed ( Fig. 2A). If factors besides ATP affect FRET signal of GO-ATeam1 independently of ATP binding, GO-ATeam3, which is an ATP-insensitive variant of GO-ATeam, will also show FRET changes. However, the FRET signal did not change in islets expressing GO-ATeam3 (Fig. 2B). Further, increases in FRET signal (YFP/CFP emission ratio) were also observed in islets when ATeam1.03 was used instead of GO-ATeam1 (Fig. 2C). These results strongly suggest that the increase in FRET signal of GO-ATeam1 and ATeam1.03 after stimulation with high glucose levels is actually due to the increase in [ATP] c . The time lag between glucose stimulation and the onset of FRET signal elevation was 22 Ϯ 6 s (mean Ϯ S.D., n ϭ 13). After the rapid increase, [ATP] c remained high. When islets were stimulated with various concentrations of glucose, [ATP] c increased rapidly in a concentration-dependent manner (Fig. 2D). We were able to observe 6 Ϯ 3, 13 Ϯ 2, and 15 Ϯ 3% (mean Ϯ S.D., n ϭ 10 for each) increases in FRET signals when the same islets were treated with 8.3, 16.7, and 25 mM glucose, respectively (Fig. 2E). Because 25 mM glucose stimulation, which is supraphysiological, is the most effective condition in which to observe obvious increases in FRET signals, we decided to use 25 mM glucose for high glucose stimulation. When the glucose concentration in the medium was gradually increased in a stepwise manner, rather than abruptly, [ATP] c also increased gradually (Fig. 2F). The mean amplitude of the increased FRET signal with gradually increasing glucose con-  centrations was not much different from that seen with an abrupt increase in glucose. FRET signals from single cells of isolated islets were also monitored ( Fig. 2G). At the single cell level, 25 mM glucose stimulation caused a 11 Ϯ 3% (mean Ϯ S.D., n ϭ 11) increase in FRET signal, whereas oligomycin A treatment caused a 40 Ϯ 3% decrease (mean Ϯ S.D., n ϭ 11), compared with basal levels.
Next, to examine whether mitochondrial energy metabolism is activated by glucose stimulation, FRET signals from a single islet expressing mitGO-ATeam1, a variant of GO-ATeam1 that is targeted to the mitochondrial matrix, were monitored. Like [ATP] c , [ATP] m rapidly increased when islets were treated with 25 mM glucose (Fig. 2H), whereas the FRET signal of ATPinsensitive mitGO-ATeam3 did not change (Fig. 2I), indicating that mitGO-ATeam1 indeed reflected the activation of mitochondrial energy metabolism.
Next, we monitored [ATP] c while decreasing the glucose concentration in the medium. Fig. 3A shows a representative time course of the FRET signal in an isolated pancreatic islet alternately subjected to 2.8 and 25 mM glucose. As expected, lowering the glucose concentration decreased [ATP] c . Interestingly, the decrease in [ATP] c induced by glucose reduction was much slower than the increase in [ATP] c caused by glucose addition. This means that [ATP] c is much more responsive to an increase than a decrease in extracellular glucose. The average rate of FRET signal decline was ϳ6-fold slower than that of FRET signal increase (Fig. 3B). Specifically, the time required for the FRET signal to reach 50% of maximum was ϳ90 s after the addition of glucose, whereas ϳ340 s were required for the signal to fall to this level after reduction in glucose (Fig. 3C). This phenomenon is not due to the kinetic property of GO-ATeam1, because the time constants of purified GO-ATeam1 for ATP association/dissociation have been determined to be less than 10 s (17). Moreover, treatment with either iodoacetate, which inhibits glyceraldehyde-3-phosphate dehydrogenase in the glycolytic pathway, or oligomycin A, which inhibits F o F 1 -ATP synthase, rapidly reduced [ATP] c in pancreatic islets (Fig. 4), indicating that the time constants of GO-ATeam1 expressed in islets were also fast enough.  (Fig. 5, A and B). The time lag between the onset of [ATP] c and that of [Ca 2ϩ ] c increase was 96 Ϯ 33 s (mean Ϯ S.D., n ϭ 19). This observation is consistent with two recent studies using Perceval, a genetically encoded biosensor for the ATP/ADP ratio, in which the cytosolic ATP/ADP ratio increased prior to [Ca 2ϩ ] c in the initial phase in glucose-stimulated mouse islets (14,15). Unlike the Perceval recording, however, we did not observe a transient drop in [ATP] c after glucose stimulation. It is noteworthy that the FRET signal of GO-ATeam is virtually unaffected by pH at physiological con-ditions (above pH 7), which is not the case for pH-sensitive Perceval. Moreover, although the reports using Perceval demonstrated that the ATP/ADP ratio increases prior to [Ca 2ϩ ] c in GSIS, they did not address whether the initial [Ca 2ϩ ] c increase can occur without the increase in ATP/ADP ratio or the actual [ATP] c . To address this question, we investigated the effects of pharmacological inhibitors of energy metabolism on [ATP] c and [Ca 2ϩ ] c responses in glucose-stimulated islets. Pretreatment with either iodoacetate for ϳ6 min or oligomycin A for 1.5-3 min abrogated the glucose-induced [ATP] c increase and, importantly, the subsequent [Ca 2ϩ ] c elevation (Fig. 5, C and E). On the other hand, these inhibitor treatments did not abrogate the tolbutamide-induced elevation of [Ca 2ϩ ] c (Fig. 5, D and F), indicating that the cells still sustained the ability to close K ATP channels and open voltage-dependent calcium channels immediately after treatment with iodoacetate or oligomycin A. These results indicate that the glucose-induced increase in energy metabolism depends entirely on both glycolysis and oxidative phosphorylation activity and is most likely crucial for subsequent Ca 2ϩ elevation.

Dynamics of [ATP] c or [ATP] m Together with [Ca 2ϩ ] c in Methyl
Pyruvate-stimulated Islets-We next investigated whether glucose-independent insulin secretion is also mediated by the activation of energy metabolism. Pyruvate, an end product of glycolysis, enters into mitochondria and is further metabolized through the TCA cycle after conversion to acetyl-CoA. However, stimulation with pyruvate does not lead to large effects on insulin secretion in isolated ␤-cells, presumably because of its low membrane permeability and the low expression levels of monocarboxylate transporter in isolated ␤-cells (20). On the other hand, the pyruvate analog methyl pyruvate (MP) is known to induce insulin secretion from isolated ␤-cells. Some reports have asserted that the insulinogenic effect of MP derives from its capacity to serve as a substrate for mitochon- drial energy metabolism (21,22), but others have suggested that MP directly affects K ATP channels in a metabolism-independent manner (13,23). These conflicting reports prompted us to test whether MP has a direct effect on ATP synthesis or not. Although MP decreased the pH of the medium from 7.4 to 7.0, the FRET signal of GO-ATeam is almost unaffected in this pH range (17). We stimulated islets with MP in the absence of glucose and monitored [ATP] m or [ATP] c (Fig. 6, A and B) (Fig. 6C). These data strongly suggest that MP induces the influx of Ca 2ϩ by increasing energy metabolism rather than directly affecting the K ATP channel.

Dynamics of [ATP] c or [ATP] m Together with [Ca 2ϩ
] c in Leucine-and Glutamine-stimulated Islets-Genetic data indicate that a mitochondrial enzyme, glutamate dehydrogenase, is important for insulin secretion. The constitutively active form of mutated glutamate dehydrogenase leads to hyperinsulinism syndrome (24), and deletion of glutamate dehydrogenase in ␤-cells partly impairs the insulin secretion response (25). Glutamate dehydrogenase catalyzes the reversible reaction: glutamate ϩ NAD(P) ϩ 7 ␣-ketoglutarate ϩ NH 4 ϩ ϩ NAD(P)H. It is known that leucine is an allosteric activator of glutamate dehydrogenase and an inducer of insulin secretion in pancreatic ␤-cells (26,27 (Fig. 7, A and B). Likewise, glutamine stimulation alone did not induce changes in [ATP] m and [ATP] c or in [Ca 2ϩ ] c (Fig. 7, C and D). Remarkably, when islets were pretreated with glutamine, stimulation with leucine led to an increase in both [ATP] m and [ATP] c , followed by rapid elevation of [Ca 2ϩ ] c (Fig.  7, E and F). Pretreatment with oligomycin A for just 3-5 min abrogated increases in [ATP] c when glutamine-pretreated islets were stimulated with leucine. Meanwhile, [Ca 2ϩ ] c gradually increased in parallel with the decrease in [ATP] c and did not rapidly increase after stimulation with leucine (Fig. 7G). These data strongly support the idea that leucine stimulation induces the activation of glutamate dehydrogenase, thereby enhancing the TCA cycle and ATP production in mitochondria, which results in Ca 2ϩ influx in islets.

Correlation between [ATP] c and Oscillating [Ca 2ϩ ] c in High Glucose Conditions-It is well known in GSIS that [Ca 2ϩ
] c , following its initial elevation, decreases and then starts to oscillate, which in turn induces oscillatory secretion of insulin (6). However, the mechanism underlying the [Ca 2ϩ ] oscillation in GSIS remains controversial. One of the candidates is the oscillation of [ATP] c . In an earlier study, oscillation of [ATP] c in pancreatic islets was implied by the observation that single islets expressing firefly luciferase, which requires ATP for light emission, showed oscillatory luminescence in both low and high glucose conditions (28). However, in our study, as well as in other recent studies (14,15), oscillations were not observed in low glucose conditions. Because luminescence by firefly luciferase is susceptible to various factors, including oxygen, acetyl-CoA, and availability of luciferin, it is difficult to exclude the possibility that oscillatory luminescence in luciferase-expressing islets resulted from perturbation by the above factors. A recent study showed that mitochondrial energy metabolism does not oscillate in INS-1 832/13 insulinoma cells (29). However, it is still possible that the oscillation of glycolysis drives [ATP] c in pancreatic ␤-cells. We compared the dynamics of [ATP] c with [Ca 2ϩ ] c in glucose-stimulated single islets. After stimulation with high glucose, [ATP] c increased sharply and then remained at high levels. In contrast, [Ca 2ϩ ] c started to oscillate following the first burst phase. We could not observe clear oscillatory behaviors in [ATP] c , even when [Ca 2ϩ ] c was oscillating (Fig.  8A). At the dispersed single ␤-cell level, we always see that 25 mM glucose caused a sustained rise in ATP levels with no clear oscillations over a period of 10 min (Fig. 2G), which is similar to what we observed in whole islets. Moreover, cross-correlation analyses for [ATP] c and [Ca 2ϩ ] c in individual glucose-stimulated islets did not reveal any correlation between [ATP] c and [Ca 2ϩ ] c dynamics during a period of Ca 2ϩ oscillation (Fig. 8B). If the oscillation of [ATP] c occurred with the same frequency as that of [Ca 2ϩ ] c , the frequency for [ATP] c should be typically less than 0.5/min. We have to note that GO-ATeam1 has enough dynamic sensitivity to detect oscillations of [ATP] c , if they were to exist (17). The FRET signal of GO-ATeam1 was not saturated in islets undergoing [Ca 2ϩ ] c oscillations at 25 mM glucose because further increasing the glucose concentration to 42 mM resulted in a further elevation of the FRET signal (Fig. 8C). Next, to exclude the possibility that oscillation of [ATP] c is hindered by the nonphysiological, abrupt increase in glucose (from 2.8 to 25 mM) in the above experiments, the glucose concentration in the medium was gradually increased in a stepwise manner. Whereas [Ca 2ϩ ] c began to oscillate at 11 mM glucose, [ATP] c gradually increased with increasing glucose concentration without significant oscillations (Fig. 8D). [Ca 2ϩ ] c oscillated in the second phase of GSIS, [ATP] c was sustained at high levels. We investigated whether sustained high [ATP] c is necessary for the oscillation of [Ca 2ϩ ] c . Treatment of islets that were pretreated with high glucose, with carbonylcyanide 3-chlorophenylhydrazone, an uncoupler of mitochondrial membrane potential, resulted in the rapid cessation of [Ca 2ϩ ] c oscillation (Fig. 9A). Likewise, [Ca 2ϩ ] c oscillations also stopped immediately after the glucose concentration was lowered from 25 to 2.8 mM (Fig. 9B). In both cases, [Ca 2ϩ ] c oscillation stopped as soon as [ATP] c began to decrease. It is most likely that sustained high [ATP] c and/or high energy metabolism is required for islets to continue to exhibit [Ca 2ϩ ] c oscillations in GSIS.

Dynamics of [ATP] c and [ATP] m upon Depletion of Either Extracellular or Intracellular
Ca 2ϩ -Finally, we investigated the role of Ca 2ϩ on glucose-induced intracellular ATP elevation. Ca 2ϩ is known to enhance the activities of several mitochondrial dehydrogenases in the TCA cycle (30). Indeed, buffering of mitochondrial Ca 2ϩ resulted in reduced insulin secretion in isolated rat islets (31). Several molecules have been implicated in Ca 2ϩ transport to the mitochondria (32)(33)(34)(35). Of these, a Ca 2ϩ -sensitive mitochondrial uniporter called MCU is involved in mitochondrial Ca 2ϩ homeostasis in pancreatic islets because its silencing impairs mitochondrial Ca 2ϩ uptake in isolated islets (9). In the particular context of glucose stimulation, however, it has been unclear where the mitochondrial Ca 2ϩ comes from. First, we investigated the role of extracellular Ca 2ϩ . We depleted Ca 2ϩ from the medium and monitored ATP levels. When the glucose concentration in the medium was increased from 2.8 to 20 mM, the corresponding increases in both [ATP] m (Fig. 10A) and [ATP] c (Fig. 10C) were similar to the increases in [ATP] m and [ATP] c observed in islets cultured with normal Ca 2ϩ -containing medium (Fig. 10, B and D). To test the requirement of the intracellular Ca 2ϩ pool, BAPTA-AM, a chelator of intracellular Ca 2ϩ , was added. Pretreatment of islets with BAPTA-AM almost completely suppressed the increases in both [ATP] m (Fig. 10E) and [ATP] c (Fig. 10F) after glucose stimulation. Moreover, treatment of islets cultured in low glucose medium with BAPTA-AM decreased [ATP] m (Fig.  10G). These data indicate that intracellular Ca 2ϩ , rather than extracellular Ca 2ϩ , is required for glucose-induced ATP elevation and for maintaining basal intracellular ATP levels.

DISCUSSION
In this paper, we fluorescently imaged the dynamics of both [ATP] c and [Ca 2ϩ ] c in single isolated mouse pancreatic islets, using a genetically encoded FRET-based ATP biosensor, GO-ATeam1, and a fluorescent Ca 2ϩ dye, fura-2. We demonstrated that stimulation of islets with glucose, MP, or leucine/glutamine all induced rapid increase of [ATP] c , followed by that of [Ca 2ϩ ] c . These results are almost consistent with a recent report that glucose-induced ATP/ADP ratio elevation is followed by depolarization of plasma membrane and [Ca 2ϩ ] c rise (14). Pharmacological inhibition of energy metabolism blocked increases of both [ATP] c and [Ca 2ϩ ] c in all situations, indicating that increase in ATP production or energy metabolism is the fundamental mechanism that triggers the [Ca 2ϩ ] c burst in the initial phase of GSIS, as well as both MP-stimulated and leucine/glutamine-stimulated insulin secretion. Oscillation of [Ca 2ϩ ] c in the second phase of GSIS is a typical feature of islets or ␤-cells (6 -8) and is assumed to be required for pulsatile insulin secretions (6). For the generation of such Ca 2ϩ oscillation, oscillatory activities of the K ATP channel have been implicated by the observation that islets of Kir6.2 Ϫ/Ϫ mice exhibited high nonoscillatory intracellular Ca 2ϩ levels after glucose or tolbutamide stimulation (36). To explain the oscillatory activities of the K ATP channel, it has long been believed that intracellular ATP levels also oscillate during the insulin secretion. In contrast to this generally believed view, however, we did not observe any significant oscillations of [ATP] c in isolated mouse pancreatic islets and in single islet cells during GSIS. Consistent with this, it has been also shown that mitochondrial bioenergetic activities do not oscillate in glucose-or pyruvatestimulated INS-1 832/13 cells (29). Conversely, a recent study has reported that [Ca 2ϩ ] oscillation in glucose-stimulated isolated mouse islet cells induces small oscillation of ATP levels near the plasma membrane ([ATP] pm ) (15). This small local oscillation of ATP level was, however, suggested to be a result of enhanced local ATP consumption by increased Ca 2ϩ (15). It is notable that GO-ATeam1 is able to detect [ATP] c changes of ϳ1 mM in insulin-secreting cells (Fig. 1). Thus, the amplitude of [ATP] c oscillation would be much less than 1 mM, even if it exists. Given that mitochondria supply large amount of ATP to bulk cytosolic space, it is not surprising that the bulk cytosolic ATP levels inside pancreatic ␤-cells are almost constant in the second phase of GSIS even when [ATP] pm oscillates.
Alternation of FRET signal of GO-ATeam1 by both glucose stimulation and metabolic inhibitors indicates that [ATP] c of pancreatic ␤-cell is within the dynamic range of the biosensor, which is ϳ2-20 mM ( Fig. 1 and Ref. 17). Apparently, this high concentration of cytosolic ATP could not regulate the K ATP channel in ␤-cells, because the channel is blocked by the Mg 2ϩunbound form of ATP with an approximate K i value of 10 M (3, 37). However, the Mg 2ϩ -unbound form of ATP exists at quite low levels inside cells, because of the high affinity of ATP to Mg 2ϩ . Indeed, most of K ATP channel activity is blocked by physiological ATP levels in the presence of Mg 2ϩ (2). Only slight increases of ATP levels thus could be sufficient for switching off the activity of the K ATP channel.
Our data, presented in this study, collectively supported the possibility that sustained high levels of [ATP] c via energy metabolism, rather than its oscillation, are required for maintaining the oscillation of [Ca 2ϩ ] c of pancreatic ␤-cells. However, we could not currently exclude the possibility that slight oscillation of free ATP levels is causative for the production of oscillatory K ATP channel activities, and thus further examinations are required to fully answer the question as to how oscillatory K ATP channel activities are created. Alternation of energy metabolism will also affect intracellular ADP level. It was reported that Mg 2ϩ -bound ADP antagonizes the ATP binding of K ATP channels (38). Thus, it is also possible that not only ATP levels but also ADP levels (or ATP/ADP ratio) are involved in the regulation of K ATP channels in GSIS. However, because no technique to specifically monitor ADP in living culture cells is currently available, it is quite difficult to know ADP dynamics in islet. If a genetically encoded biosensor for ADP is estab-lished, it will contribute to more detailed understanding of the mechanism of opening and closing of K ATP channels.
Another notable finding is that the uncoupling of mitochondria or reduction of medium glucose levels immediately arrested oscillation of [Ca 2ϩ ] c with a slight drop in [ATP] c (Fig.  9, A and B), suggesting that ␤-cells are able to sense a small decrease in energy supply, namely a blood glucose levels, probably sensing a slight reduction in [ATP] c . This system would be suitable for maintaining the energy homeostasis of the whole body. If complete depletion of [ATP] c to the basal level were required to halt oscillation of [Ca 2ϩ ] c , a significant amount of insulin would continue to be secreted even at low blood glucose levels, leading to hypoglycemia.
The dysfunction of ␤-cells is one of the hallmarks of type 2 diabetes, in which insulin secretion is attenuated even after glucose stimulation (39 -43). According to our imaging data, maintaining elevated [ATP] c must be necessary for insulin secretion, so it is conceivable that the dynamics of [ATP] c is impaired in islets of diabetic mouse models or human patients.