Incretin-Modulated Beta Cell Energetics in Intact Islets of Langerhans

Incretins such as glucagon-like peptide 1 (GLP-1) are released from the gut and potentiate insulin release in a glucose-dependent manner. Although this action is generally believed to hinge on cAMP and protein kinase A signaling, up-regulated beta cell intermediary metabolism may also play a role in incretin-stimulated insulin secretion. By employing recombinant probes to image ATP dynamically in situ within intact mouse and human islets, we sought to clarify the role of GLP-1-modulated energetics in beta cell function. Using these techniques, we show that GLP-1 engages a metabolically coupled subnetwork of beta cells to increase cytosolic ATP levels, an action independent of prevailing energy status. We further demonstrate that the effects of GLP-1 are accompanied by alterations in the mitochondrial inner membrane potential and, at elevated glucose concentration, depend upon GLP-1 receptor-directed calcium influx through voltage-dependent calcium channels. Lastly, and highlighting critical species differences, beta cells within mouse but not human islets respond coordinately to incretin stimulation. Together, these findings suggest that GLP-1 alters beta cell intermediary metabolism to influence ATP dynamics in a species-specific manner, and this may contribute to divergent regulation of the incretin-axis in rodents and man.

T ype 2 diabetes is a socioeconomically costly disease state usually characterized by pancreatic beta cell decompensation in the face of increased resistance to circulating insulin (1). The resulting glucose intolerance leads to undesirable sequelae including neuropathy, renal failure, cardiac disease, and increased cancer risk. Under nor-mal conditions, the secretion of insulin is primarily driven by the aerobic glycolysis of glucose, raising cytosolic ATP/ ADP ratios [ATP/ADP] cyto . This leads to the closure of hyperpolarizing ATP-sensitive K ϩ channels (K ATP ) and calcium (Ca 2ϩ )-dependent exocytosis due to Ca 2ϩ influx through voltage-dependent Ca 2ϩ channels (VDCC). Se-cretion is further augmented by "amplifying" pathways (2,3), which may involve intracellular signaling cascades such as those mediated by cAMP, acting upstream of exchange protein activated by cAMP (Epac) (4) and protein kinase A (5), as well as AMP-activated protein kinase (6), protein kinase C (7) and MAPK (8,9).
In addition to glucose, a number of alternative fuels and circulating factors regulate insulin secretion. Notably, gut-derived incretins including glucagon-like peptide 1 (GLP-1) and glucose-dependent insulinotropic polypeptide are liberated from entero-endocrine cells in response to bile acid and nutrient flux (10,11), and act to potentiate insulin release in a glucose-dependent manner (12,13). Due to the latter and other properties, incretin-based analogs are becoming mainstays of type 2 diabetes treatment. Although the effects of GLP-1 upon adenylate cyclase (AC) activity, [cAMP] i oscillations, Epac signaling and exocytosis are increasingly well characterized (4,14), whether incretins are able to alter beta cell intermediary metabolism to influence insulin secretion remains controversial. Thus, whereas dynamic luciferase-based studies by us have demonstrated increased free cytosolic [ATP] in GLP-1-stimulated MIN6 immortalized beta cells (15), others have failed to detect any effect of the GLP-1 mimetic, exendin-4, on mitochondrial ATP levels in primary rodent islets (16). Nonetheless, the latter studies did report a significant increase in glucose utilization in response to GLP-1 at elevated glucose concentration, although glucose oxidation was not changed by the incretin at the time points studied.
Further suggesting that incretin-stimulated insulin secretion may in part involve altered metabolism are the observations that: 1) GLP-1 and exendin-4 stimulate large oscillations in intracellular Ca 2ϩ concentrations ([Ca 2ϩ ] i ), both in rodent and human beta cells (16 -18); 2) ATP-consuming processes are required for cytoplasmic Ca 2ϩ removal and intracellular store refilling (19); 3) mitochondrial Ca 2ϩ uptake activates citrate cycle dehydrogenases, augmenting ATP production (20); 4) excessive mitochondrial Ca 2ϩ uptake may depolarize the inner mitochondrial membrane, resulting in a temporary cessation of ATP synthesis (21); and 5) Ca 2ϩ stimulates energyconsuming processes such as exocytosis (22). Correspondingly, we (19) and others (21) have recently demonstrated that glucose-dependent oscillations in intracellular ATP are strongly influenced by Ca 2ϩ in pancreatic beta cells.
To further investigate a role for incretin in beta cell energetics, we used a recombinant strategy to direct expression of the GFP-based ATP-binding protein Perceval (23) throughout the first few layers of mouse and human primary islets (21). Although previously deployed in dis-sociated beta cells (24 -26) and small numbers of cells in intact islets (21), this technique has not been employed to investigate ATP/ADP dynamics across a large population, nor have responses to incretins been examined in islets.
Using this approach, we show that GLP-1 modulates ATP dynamics, resulting in [ATP/ADP] cyto rises. These effects were independent of prevailing cell metabolic status because GLP-1 was still able to induce ATP oscillations in the presence of low (3 mM) glucose concentration, which is nonpermissive for insulin secretion. Moreover, GLP-1-stimulated ATP dynamics at elevated glucose concentration were reliant upon engagement of the GLP-1-receptor (GLP-1R) and ensuing Ca 2ϩ influx because they could be blocked reversibly using exendin 9-39 and verapamil, respectively. Lastly, species differences in GLP-1-regulated "metabolic connectivity" were present, with mouse but not human islets responding to stimulus with synchronous ATP dynamics. Thus, GLP-1 affects beta cell intermediary metabolism through alterations to ATP dynamics in a species-specific manner, and this may play an important role in incretin-modulated beta cell function at both low and elevated glucose concentrations.

Adenovirus-harboring Perceval is tropic for beta cells and reports ATP changes in intact islets
Specific immunohistochemical analyses detected Perceval expression predominantly in the insulin immunopositive cell population, confirming the reported (27)(28)(29) tropism of adenoviruses for beta ( Figure 1A) over alpha ( Figure 1B) and other endocrine cells within mouse islets of Langerhans. Nipkow spinning disk microscopy (30) was therefore used to capture the effects of glucose and other secretagogues on cytoplasmic ATP/ADP dynamics throughout the population of beta cells residing within the first few layers of intact islets. Elevation of glucose from 3-17 mM induced a multiphasic response typified by an initial increase in [ATP/ADP] cyto (F/Fmin), followed by the development of superimposed oscillations, as previously reported (Figure 1, C and D) (⌬ATP/ADP ϭ 0.13 Ϯ 0.02 AU, n ϭ 14 recordings from six animals) (21). Just under half of the imaged population (44%) responded to glucose with increases in apparent [ATP] cyto ( Figure 1E), suggesting that a subpopulation of beta cells may act to integrate metabolic information before propagating this throughout the syncytium as Ca 2ϩ waves (31,32).
To determine whether cAMP could interfere with the Perceval ATP-binding site, HEK293 cells expressing the  Figure 1F).

GLP-1 induces oscillations in cytosolic ATP/ADP
In the presence of high (17 mM) glucose concentration, the addition of 20 nM GLP-1 to Perceval-expressing mouse islets stimulated increases in fluorescence, which were superseded by the appearance of low-frequency oscillations ( Figure 2A). This dose of incretin has previously been shown to maximally stimulate intracellular Ca 2ϩ rises in islets (17). Importantly, in our hands, GLP-1 evoked a barely detectable decrease in intracellular pH (ϳ0.05 pH unit) (23,26), indicating that the fluctuations in Perceval fluorescence intensity are likely to result from oscillations in [ATP/ADP] cyto , as opposed to acidification/alkalinization. Following incubation at nonpermissive (3 mM) glucose concentration, GLP-1 was similarly able to induce oscillatory increases in Perceval fluorescence ( Figure 2B), although the incretin was unable to alter Ca 2ϩ responses under these conditions (81.1 Ϯ 6.0 vs 8.9 Ϯ 2.3% GLP-1-responsive cells, G3 ϩ GLP-1 vs G11 ϩ GLP-1, respectively; n ϭ 6 recordings from three animals, P Ͻ .01). As observed for the effects of glucose ( Figure 1E), the action of GLP-1 involved the recruitment of a subnetwork of beta cells. Thus, ϳ40% of the recorded population responded to GLP-1 in the above manner ( Figure 2, C-E). Intriguingly, sharp drops in apparent [ATP/ADP] cyto were seen between peaks ( Figure 2, A and B). These deflections seemed to be part of normal stochastic behavior because they were not influenced by prevailing glucose concentration ( Figure 2F) and persisted in the presence of cyclosporin A (CysA), an inhibitor of the mitochondrial permeability transition pore ( Figure 2G) (33). Although GLP-1 tended to stimulate smaller [ATP/ADP] cyto rises than 17 mM glucose, this was not significant (⌬ATP/ADP ϭ 0.13 Ϯ 0.02 vs 0.08 Ϯ 0.01 AU, 17 mM glucose vs GLP-1 applied at 3 mM glucose, respectively; n ϭ 13-14 recordings from six animals, nonsignificant, NS) ( Figure 2H). Notably, the incretin was still able to augment [ATP] cyto even in the continued presence of high (17 mM) glucose ( Figure 2H). Similar results were obtained using conventional luciferase-based detection of ATP (Table 1).

GLP-1 drives increases in inner mitochondrial membrane potential, which are dependent on glucose
At elevated glucose concentration, both enhanced supply of glycolytically derived pyruvate and Ca 2ϩ activation of key dehydrogenases (20) accelerate citrate cycle flux in beta cell mitochondria, elevating the inner mitochondrial membrane potential (⌬⌿ m ) (26). The consequent activation of the F 1 /F 0 ATP-synthase then results in accelerated ATP production, potentially raising [ATP/ADP] cyto and countering decreases in the latter driven by cytosolic ATPconsuming reactions. Conversely, excessive Ca 2ϩ uptake, and hence an accumulation of positive charge, may exert a direct depolarizing effect to lower ⌬⌿ m. To determine whether GLP-1-induced increases in [ATP/ADP] cyto may be due at least in part to increases in ⌿ m , we dynamically tracked fluorescence of the potential-sensitive fluorescent probe tetramethyl rhodamine ethylester (TMRE) in beta cells within whole islets. In response to elevated (17 mM) glucose, cells demonstrated rapid and large increases in TMRE intensity, indicative of mitochondrial hyperpolarization ( Figure 3A), as expected (19,34,35). By comparison, GLP-1 at both low (3 mM) and high (17 mM) glucose concentrations induced slower and smaller increases in TMRE fluorescence ( Figure 3D).
To assess whether the effects of GLP-1 on [ATP/ADP] cyto were likely to involve the generation of intracellular cAMP, islets were exposed to forskolin (FSK), a cAMP-raising agent (5). As for GLP-1, FSK elicited a step-change in [ATP/ADP] cyto (⌬ATP/ADP ϭ 0.09 Ϯ 0.02, n ϭ 13 recordings), although this was notable by the

Calcium-dependency of GLP-1-evoked ATP increases
Because GLP-1-induced ATP increases may depend upon or provoke intracellular Ca 2ϩ rises due to effects on K ATP , dual imaging experiments were performed in isolated beta cells. This preparation was used to minimize photobleaching and interference between signals derived from the two probes that complicated measurements in intact islets. Multiparametric recordings using Perceval and Fura Red revealed that the onset of the ATP/ADP response to GLP-1 preceded any increases in cytosolic Ca 2ϩ (Figure 4, A and B and Supplemental Figure 1), indicating that initiation of the latter is likely to be a downstream consequence rather than the cause of incretin-induced [ATP/ADP] cyto, increases. In this case, tolbutamide was used as a control to stimulate large Ca 2ϩ rises, which precede net ATP/ADP consumption ( Figure 4C). As for experiments using intact islets, FSK was able to evoke increases in ATP in dissociated cells and this could also be mimicked using the phosphodiesterase inhibitor isobutyl methyl xanthine (IBMX) ( Figure 4D).
Mitochondrial Ca 2ϩ sequestration stimulates citrate cycle dehydrogenases (20,37) to increase ATP production. Therefore, the effects of extracellular Ca 2ϩ chelation were examined to delineate whether continued Ca 2ϩ influx through VDCCs was required for the actions of GLP-1 on [ATP/ADP] cyto at elevated glucose concentration, or whether intracellular pathways were relatively more important. Islets perifused with buffer containing zero added Ca 2ϩ plus 1 mM ethylene glycol tetraacetic acid (EGTA) failed to display any changes in [ATP/ADP] cyto following application of GLP-1 (Figure 4, E and F). Because the removal of external Ca 2ϩ may lead to depletion of internal Ca 2ϩ stores or islet dissociation, Ca 2ϩ influx through L-type VDCC was blocked using 10 M verapamil. In line with the above observations, verapamil inhibited the effects of GLP-1 on ATP dynamics (0.03 Ϯ 0.01 vs 0.14 Ϯ 0.02 events/minutes, during and after verapamil, respectively; n ϭ 9 recordings, P Ͻ .01) (Figure 4, G and H).

GLP-1 modulates ATP dynamics in a species-specific manner
We have recently shown that incretins augment insulin secretion in human islets by boosting beta cell cooperativity in a process termed "incretin-regulated cell connectivity" (17). Given that similar effects are largely absent in mouse islets, we wondered whether incretin could differentially affect ATP/ADP dynamics to influence the metabolism thought to drive ionic oscillations (38,39). To investigate this, large-scale mapping of cell-cell correlations was used to determine the effects of GLP-1 on the population organization of metabolic oscillations in mouse and human tissue over a 30 -40-minute period (17,40). During GLP-1 application, the responsive subpopulation within mouse islets mounted synchronous deflections in [ATP/ADP] cyto ( Figure 5A). By contrast, human islets responded to the same challenge with largely asynchronous ATP oscillations ( Figure 5A). In line with this, correlated activity was much higher in mouse compared with human islets (Figure 5B), as evidenced by lower levels of metabolic connectivity in the latter species (90.7 Ϯ 4.6 vs 55.6 Ϯ 4.7% significantly correlated cell pairs, mouse vs human tissue, respectively; n ϭ 8 recordings from three donors and four animals, P Ͻ .01) ( Figure 5C).

Discussion
The literature surrounding the action of incretin on beta cell ATP/ADP ratios is contentious, with the existence of conflicting reports regarding the effects of the incretins upon cellular metabolism and energetics (15,16). By combining recombinant expression of the ATP sensor Perceval with in situ imaging of ATP/ADP dynamics in intact mouse islets, we demonstrate that GLP-1 likely influences beta cell intermediary metabolism at both low and high glucose concentrations. Mechanistically, these effects involved changes in mitochondrial potential, GLP-1R engagement, and ATP-triggered Ca 2ϩ influx through VDCC, and could be mimicked using the cAMP-elevating compounds FSK or IBMX (see Figure 6 for a schematic).
Our observations with Perceval reflect steady-state [ATP/ADP] cyto and, as such, report the balance between the rates of ATP synthesis and degradation. However, GLP-1 seems unlikely to inhibit ATP-consuming processes under the conditions used here because it acutely acts to enhance ionic fluxes and insulin secretion (17). Thus, increases in [ATP/ADP] cyto in response to the incretin seem more likely to reflect accelerated ATP synthesis.
However, we cannot formally exclude the possibility that factors other than enhanced metabolism, such as a redistribution of ATP between different intracellular pools (eg, secretory granules or the endoplasmic reticulum) (41), contribute to the observed actions of GLP-1 on the cytosolic ATP/ADP ratio. In response to elevated glucose, and extending our earlier findings, which used the less sensitive photoprotein  firefly luciferase to image cytosolic and mitochondrialfree ATP in islets (19), mouse beta cells responded by mounting oscillations in ATP/ADP that were coordinated across the imaged population. This may reflect a bidirectional interplay between metabolic and ionic signals that is phase-set by negative feedback emanating at the level of Ca 2ϩ , and which drives the slow oscillations in K ATP conductance and Ca 2ϩ influx (21,39). Notably, only ϳ50% of the Perceval-expressing population displayed [ATP/ ADP] cyto rises/oscillations following exposure to high glucose, raising the intriguing possibility that a metabolically coupled subnetwork of beta cells orchestrates the global Ca 2ϩ dynamics known to underlie insulin secretion in mouse islets (31,42). Although electrotonic coupling via gap junctions would allow nonmetabolically active beta cells to contribute to islet-wide Ca 2ϩ oscillations (31,43), previous studies have shown that nicotinamide adenine dinucleotide phosphate (NAD(P)H) increases are ob-served in 90% of beta cells within islets (44). Potential explanations for these discrepancies include the existence of functional heterogeneity between beta cell subpopulations including differences in ATP generation (45), compartmentalization of ATP responses into discrete domains (eg, in the subplasma membrane space) that remain undetectable at the resolutions employed here (21), and probe saturation returning values under the 20% threshold for inclusion as responsive cells. Of note, GLP-1 was able to influence ATP/ADP dynamics within a matter of minutes, principally by altering both levels and patterning, and the former effect was confirmed using static biochemical assays. The significance of the reported changes to ATP dynamics remains unknown. Given that GLP-1 increases intracellular Ca 2ϩ load (17), the pronounced and rapid downward deflections in [ATP/ ADP] cyto may reflect either Ca 2ϩ -dependent ATP-consuming processes (eg, exocytosis, Ca 2ϩ store refilling, etc.), or alternatively, a mechanism to prevent mitochondrial Ca 2ϩ toxicity, both of which are rapidly balanced by augmented metabolism. In terms of the latter mechanism, the uptake of ATP into the mitochondrial matrix via the Ca 2ϩ -activated ATP-Mg/Pi transporter would buffer intramitochondrial [Ca 2ϩ ], preventing mitochondrial permeability transition and cell death at the expense of [ATP/ADP] cyto .
Although mitochondrial Ca 2ϩ uptake stimulates NADH production to drive respiratory chain activity and ATP synthesis (37), an exaggerated flux of Ca 2ϩ across the inner mitochondrial membrane can constrain ATP production by reducing the electrochemical gradient that powers proton pumping and F 1 /F 0 ATP-synthase activity. We therefore used TMRE to monitor whether GLP-1 was able to alter m to evoke [ATP/ADP] cyto rises. Whereas glucose exerted a rapid and large hyperpolarizing influence upon m , as previously reported (26), this was much less pronounced in incretin-stimulated islets. The relatively larger increases in ATP detected in response to GLP-1 per unit decrease in m is consistent with a higher rate of Ca 2ϩ entry and ATP consumption in response to glucose than incretin (eg, to drive increased secretory granule dynamics, ion pumping, and protein synthesis).
In line with our previous reports using MIN6 beta cells (15), GLP-1 was able to modulate ATP/ADP dynamics even under conditions of low glucose. Although changes in [ATP/ADP] cyto would be expected to close hyperpolarizing K ATP channels, leading to depolarization and Ca 2ϩ influx, GLP-1 was unable to increase cytosolic free Ca 2ϩ in beta cells exposed to nonpermissive glucose concentration. Thus, other glucose-derived signals may be required to translate GLP-1-induced oscillations in [ATP/ADP] cyto into Ca 2ϩ rises and Ca 2ϩ -dependent insulin secretion. In- deed, glucose and GLP-1 engage distinct ACs (46), and summation and/or cell compartmentalization of the ensuing changes to cAMP-Epac dynamics may therefore be required to fully sensitize K ATP to GLP-1-stimulated alterations to beta cell metabolism (47,48). Although the mechanisms underlying GLP-1 effects at low glucose remain unknown, they may implicate a role for mitochondrial m that was altered by the incretin even in the presence of nonpermissive levels of the sugar.
Simultaneous recordings of Perceval and Fura Red revealed that GLP-1 elicited rises in [ATP/ADP] cyto before those of [Ca 2ϩ ] i , supporting the notion that metabolism and K ATP are required to initiate GLP-1-stimulated Ca 2ϩ influx. It is worthwhile to note that the response time for Perceval (seconds) is much slower than that for Fura Red (milliseconds) (23,49), meaning that the lag between the onset of [ATP/ADP] cyto and Ca 2ϩ increases was likely underestimated. Continued effects of GLP-1 upon beta cell metabolism were dependent upon Ca 2ϩ influx through L-type VDCCs, because the incretin was unable to stimulate [ATP/ADP] cyto rises in cells pretreated with EGTA and verapamil. This is unsurprising given our recent findings that mitochondrial Ca 2ϩ uptake through the mitochondrial Ca 2ϩ uniporter is critical for rendering beta cells glucose competent (24,26). This effect is most likely achieved through the Ca 2ϩ -stimulated up-regulation of mitochondrial dehydrogenase activity, which supplies reducing equivalents to the respiratory chain, leading to enhanced ATP production (50,51). Although it could be argued that the observed effects were due to blockade of glucose actions, which then reappeared following washout, it should be noted that GLP-1 was unable to alter [ATP/ADP] cyto during antagonist application, making an effect of the incretin on beta cell metabolism via a nonextracellular Ca 2ϩ -linked pathway unlikely. We observed that, following washout of both verapamil and exendin 9-39, the increases in [ATP/ADP] cyto were smaller than those observed under control conditions; this may reflect residual blockade of GLP-1R/VDCC or alternatively slow dissociation kinetics due to use of a perfusion system.
A key observation here ( Figure  4B) was that the onset of GLP-1-induced increases in [ATP/ADP] cyto occurred before detectable changes in cytosolic free Ca 2ϩ . What, therefore, may be the mechanisms through which GLP-1 leads to an apparent direct stimulation of ATP synthesis? A recent report (52) has suggested that the GLP-1 receptor agonist geniposide may stimulate pyruvate carboxylase activity in beta cells to promote anaplerosis into the citrate cycle and hence ATP synthesis (53). These observations are in line with the present findings and provide one possible mechanism through which elevated cAMP may enhance ATP production. Other possibilities include stimulation of glucokinase (54), enhanced protein kinase A-dependent breakdown of fuel stores such as glycogen or triglycerides (55) and, conceivably, activation of glycolytic enzymes including phosphofructokinase (56). Finally we note, intriguingly, that intramitochondrial cAMP levels may fluctuate independently of those in the cytosol (57) to regulate intramitochondrial ATP synthesis, though whether GLP-1 is able to engage this pathway is unclear given the apparent impermeability of the mitochondrial inner membrane to cAMP.
We have recently shown that human islets mount poorly coordinated Ca 2ϩ responses to glucose, but high levels of correlated activity between beta cells can be driven by GLP-1 in a process termed "incretin-regulated connectivity." By contrast, mouse islets already display high levels of coordinated activity in response to glucose, and GLP-1 acts to increase time spent in the active state while maintaining synchronicity within the beta cell syncytium (17,58). We therefore sought to clarify whether these species differences in incretin potentiation of glucose-stimulated insulin secretion were accompanied by alterations to metabolism. Interestingly, whereas GLP-1induced [ATP/ADP] cyto dynamics were highly correlated across the responsive beta cell subpopulation in mice, they were more stochastic in human islets. Thus, GLP-1 is a poor orchestrator of metabolic communications be- tween human beta cells, and such uncoupling of metabolic and ionic oscillations may contribute to speciesspecific regulation of the incretin axis. Although the mechanisms remain unexplored, they may reflect divergent islet architecture, paracrine/autocrine signaling circuits, and gap junction function in human vs mouse tissue (58,59). We cannot exclude, however, a role for speciesdifferences in beta cell physiology introduced by islet isolation procedures, cold ischemia time, the relative immaturity of mouse compared with human islets, and the more varied nature of human islet material in terms of sex, age, and body mass index (see references (60,61) for useful discussion). Finally, although we refer to measurement of [ATP/ ADP] cyto throughout the present study, in the case of Perceval, this requires ADP levels commensurate with probe affinity (23). Because calculation of free ATP and ADP cannot readily be inferred from total levels due to nucleotide sequestration in organelles and cytoskeleton binding (41,62), the actual parameter under measure may vary. Similarly, our own measurements of total ATP, while most easily explained through changes in phosphorylation potential and ATP/ADP ratio (63), could conceivably reflect a change in the concentration of ATP alone. Future studies using probes with a range of affinity values will therefore be required to accurately quantify the effects of glucose and incretin on intracellular [ATP/ADP] cyto .
In summary, we show that GLP-1 is able to influence beta cell intermediary metabolism through alterations to the extent and patterning of intracellular ATP/ADP increases. This requires GLP-1R signaling, changes to m and extracellular Ca 2ϩ influx, and displays marked species divergence in "metabolic connectivity," as human beta cells fail to properly coordinate [ATP/ADP] cyto oscillations. Thus, alterations to beta cell metabolism may contribute to the diverse and glucose-dependent actions of incretin, including potentiation of insulin secretion and prevention of apoptosis.

Mouse islet isolation
Animals were maintained in individually ventilated cages in a specific pathogen-free facility under a 12-h light-dark cycle with ad libitum access to water and food. Mice (8 -12 weeks old) were euthanized by cervical dislocation and pancreatic islets isolated by collagenase digestion, as described (64). Animal procedures were approved by the Home Office according to the Animals (Scientific Procedures) Act 1986 of the United Kingdom (PPL 70/7349).

Human islet isolation
Human islets (donor age range, 34 -52 years) were isolated at transplantation facilities in Oxford, Geneva, and Pisa with the relevant national and local ethical permissions, including consent from next of kin where required. Islets were cultured as previously described (17). All studies involving human tissue were approved by the National Research Ethics Committee London (Fulham) "Signal Transduction in isolated human islets: regulation by glucose and other stimuli" REC No. 07/H0711/114.

Adenoviral delivery of Perceval
Complementary DNA encoding the ATP/ADP sensor Perceval (a kind gift from Professor Gary Yellen) was cloned and packaged into adenovirus as described (25,26). Forty-eighthour incubation with virus was sufficient to express Perceval throughout the first few islet cell layers.

Immunohistochemistry
Islets were fixed overnight at 4 C in paraformaldehyde (4%, wt/vol) before application of guinea pig anti-insulin 1:200 and mouse anti-glucagon 1:1000 antibodies, and processing as previously described (17). Uniform linear adjustments were applied to contrast/brightness to improve image quality for presentation.

ATP/ADP and Ca 2؉ imaging
Perceval-expressing islets were placed in a custom-manufactured 36 C chamber (Digital Pixel) and perifused with a HEPESbicarbonate buffer (120 mM NaCl, 4.8 mM KCl, 24 mM NaHCO 3 , 0.5 mM Na 2 HPO 4 , 5 mM HEPES, 2.5 mM CaCl 2 , and 1.2 mM MgCl 2 ) saturated with 95% O 2 /5% CO 2 and adjusted to pH 7.4. A solid-state 491-nm laser was passed through a Nipkow spinning-disk head (Yokogawa CSU-10) coupled to ϫ10 -ϫ20/0.3-0.5NA objectives (EC Plan-Neofluar, Zeiss). Emitted signals (510 -540 nm) were subsequently captured at a frame rate of 0.2 Hz using a highly-sensitive 16-bit 512ϫ512pixel electron multiplying charge-coupled device (Hamamatsu). Perceval-expressing cells were manually delineated using a region of interest and intensity over time traces extracted. Signals were normalized using the function F/Fmin where F is fluorescence at a given time point and Fmin is the minimum recorded fluorescence.

Biochemical detection of ATP
Batches of 10 islets were treated as indicated for 30 minutes before removal of supernatant and extraction of lysate using either percholoric acid or distilled boiling H 2 O followed by sonication and storage on ice (65,66). ATP concentration in the supernatant was immediately assayed using a luciferase-based detection kit according to the manufacturer's instructions (ATP Determination Kit, Life Technologies).

TMRE imaging
Islets were incubated in 20 nM TMRE for 30 minutes before imaging as above, but using excitation and emission wavelengths of 563 nm and 600 nm, respectively. Treatments were applied as indicated and at the end of each recording, carbonyl cyanide 4-(trifluoromethoxy)phenylhydrazone was added at 2 M.

Correlation and wavelet analyses
Correlation analyses were performed using the Pearson r coefficient as previously detailed (P Ͻ .05) (17,67). Phase maps were compiled by converting the normalized intensity of each cell to a value between 1-100% and assigning this to a color. To depict the contribution of period to [ATP/ADP] cyto dynamics, the frequency and time components of the mean Perceval fluorescence trace were extracted using bias-corrected wavelet analysis.

Statistical analysis
Data distribution was determined using the D'Agostino omnibus test. Pairwise comparisons were performed using Mann-Whitney U test or Student unpaired and paired t tests. Interactions between multiple treatments were assessed using one-way ANOVA followed by Bonferonni's post hoc test. A sigmoidal fit was used to calculate the EC50 of normalized and log-transformed dose-response curves. Analyses were conducted using R (R-project), Graphpad Prism (Graphpad Software) and IgorPro (Wavemetrics), and results deemed significant at P Ͻ .05.