Glucose-Stimulated Oscillations in Free Cytosolic ATP Concentration Imaged in Single Islet β-Cells: Evidence for a Ca²⁺-Dependent Mechanism Free

Islets of Langerhans were isolated from 12- to 16-week-old male CD1 strain mice using collagenase digestion (30). Briefly, the pancreases were injected in situ via the bile duct with a solution of 0.5 mg/ml collagenase (Serva, Heidelberg, Germany) in serum-free Dulbecco’s modified Eagle’s medium (DMEM) (Sigma, Poole, Dorset, U.K.). The pancreases were removed, finely chopped, and further digested in the above collagenase solution for 20 min at 37°C. The pancreases were transferred to DMEM containing 10% fetal bovine serum to stop the digestion and then vigorously shaken and filtered through a nylon mesh with an approximate pore size of 500 μm. The samples were washed three times in serum-containing (10% vol/vol) medium before hand-picking of islets. Islets were incubated overnight in DMEM containing 10% (vol/vol) fetal bovine serum and 11 mmol/l glucose before dissociation into individual cells.

Islets were dissociated by incubation in 25 mg/ml trypsin (Sigma) for 20 min at 37°C. After pelleting by brief centrifugation (1,000g for 3 min), the cells were resuspended in serum-containing medium and spotted onto glass coverslips that had been precoated with 0.35 mg/ml Cell-Tak cell adhesion protein (Collaborative Biomedical Products, Bedford, MA) in 50 μmol/l NaHCO₃. The cells were maintained in culture medium (11 mmol/l glucose) for 1–2 days before incubating in medium containing 3 mmol/l glucose 12–16 h before use.

Human islets were obtained from Dr. Roger F.L. James (University of Leicester, Leicester, U.K.) and were dissociated after overnight incubation as described above. The islets were then fixed to plastic Petri dishes as opposed to glass coverslips. Dissociated islet cells were infected with adenovirus and treated before experimentation as described for mouse cells.

MIN6 cells (31) were cultured essentially as described (27) and were used between passage 25 and 32. Briefly, cells were routinely cultured in DMEM containing 25 mmol/l glucose and 15% fetal bovine serum. For experimental use, cells were grown on poly-l-lysine–treated glass coverslips and incubated in medium containing 3 mmol/l glucose overnight before measurements.

Luciferase expression was achieved by infection of cells with an adenoviral vector encoding cytoplasmic (AdCMVcLuc)-targeted luciferase under the control of the cytomegalovirus major-late promoter (18). Infection was carried out by incubation for 4 h with a multiplicity of infection of 100 viral particles per cell for isolated β-cells and 30 particles per cell for MIN6 cells.

Time-resolved photon-counting imaging of individual cells expressing cytoplasmic or mitochondrial luciferases was performed during static incubation in Krebs-Ringer bicarbonate (KRB) medium. The medium comprised 125 mmol/l NaCl, 3.5 mmol/l KCl, 1.5 mmol/l CaCl₂, 0.5 mmol/l MgSO₄, 0.5 mmol/l KH₂ PO₄, 2.5 mmol/l NaHCO₃, and 10 mmol/l Hepes-Na⁺, pH 7.4, containing initially 3 mmol/l glucose and equilibrated with 95:5 O₂/CO₂ and supplemented with 1 mmol/l luciferin on the thermostated (37°C) stage of an Olympus IX70 inverted microscope (×10 objective, 0.4 NA) linked to a three-stage intensified charge-coupled device camera (Photek; Lewes, East Sussex, U.K.) as previously described (27,32).

For perifusion of MIN6 cell populations, luciferase luminescence was monitored continuously using a photomultiplier tube (ThornEMI Electron Tubes, Ruislip, U.K.) and perifusion system (33) in KRB medium (37°C) plus 5 μmol/l luciferin and other additions as given (18). Bioluminescence was recorded at 1-s intervals and averaged over 5-s intervals. Light output was normalized to that observed under basal conditions (i.e., 3 mmol/l glucose) to give the traces shown in Figs. 2–4. Where EGTA (0.1 mmol/l) was added, CaCl₂ was omitted from KRB.

Changes in [Ca²⁺]_cyt were measured dynamically at 37°C with entrapped fura-2 (16,32) using a Leica DM-IRBE inverted optics epifluorescence microscope (×40 objective) and a Hamamatsu C4742-995 charge-coupled device camera driven by OpenLab (Improvision, Coventry, U.K.) software. Primary β-cells were loaded by incubation for 20 min in KRB medium supplemented with 2 μmol/l fura-2-AM (Sigma) and 0.05% Pluronic F-127 (BASF, Mount Olive, NJ). MIN6 cells were loaded similarly by incubation for 40 min with 5 μmol/l fura-2-AM and 0.05% Pluronic F-127. Changes in [Ca²⁺]_cyt are reported as changes in the ratio of the fluorescence emission (510 nm) during excitation with light at 340 or 380 nm (32).

Three separate statistical tests were performed on the data from the single-cell luciferase light output and fura-2 fluorescence assays to identify oscillatory behavior in [ATP]_cyt and [Ca²⁺]_cyt, respectively. Discrete Fourier transformation (DFT) was performed on the data to highlight the principal resonant frequencies present and the approximate amplitude of the oscillations at those frequencies. Mathematical analysis was carried out with a Microsoft Visual Basic script written in-house. Calculations were based on those described by Briggs and Henson (34). The presence of oscillations at a specific frequency was implicated by a high value for the amplitude of the Fourier transform at that frequency.

Autocorrelation of data were performed as described (35). The Pearson product-moment correlation coefficient was determined for the data set paired with itself but was offset by a specific time period. If the data displayed oscillations, then it was expected that a positive correlation would be found when the offset time period was equal to the wavelength of oscillation. Conversely, a negative correlation would be expected when the data were offset by half the wavelength of oscillation. The significance of the correlation coefficient was tested using an F test (36).

The fit of the data to a sine curve was assessed by nonlinear least-squares regression analysis. The amplitude, frequency, and phase were fitted by minimizing the sum of squares of the residuals using the “Solver” Add-in within Microsoft Excel, and the mean value of the sine curve was determined as the mean of the data set. The significance of the fit of the sine curve was determined from an F test comparing the residual sum of squares from the curve-fit to the total sum of squares. In effect, this tested the significance of the improvement of fit for the sine curve compared with a single mean value for the data set.

Data and images were generated from a 20-s integration of the light output. This did not enable reliable detection of oscillations of periods <60 s. Reanalysis of the data from individual cells with a 10-s integration window gave better temporal resolution, yet in most cases, gave essentially the same best-fit sine curve. However, the significance of fit was reduced in this case because of increased noise.

The significance of a difference between two data sets was assessed using a t test. Data are given as means ± SE. All calculations and statistical tests were carried out in Microsoft Excel 1997 running under Microsoft Windows NT, version 4.0.

The existence of oscillations in [ATP]_cyt in pancreatic β-cells was investigated using primary isolated β-cells expressing recombinant firefly luciferase (27,37). An adenoviral vector (18) was used here to introduce luciferase cDNA into the primary cells because this achieved both efficient transformation and a high level of luciferase expression in individual cells. Whereas we did not attempt to correct for the presence of non–β-cells in these cultures, β-cells could usually be identified by their morphology (being somewhat larger than the majority of non–β-cells).

The response of a typical single human islet β-cell to a stepped increase in glucose concentration is shown in Fig. 1A. An increase in light output occurred upon incubation in high (15 mmol/l) glucose concentrations compared with incubation in low (3 mmol/l) glucose concentrations, equivalent to 9.0% of the average light output under basal conditions for the cell in Fig. 1A. [ATP]_cyt was observed to oscillate in human β-cells at both high and low glucose concentrations, albeit around a raised average point at 15 mmol/l glucose (Fig. 1A). Images of the light output from the cell at the peak and trough of these oscillations at 3 and 15 mmol/l glucose are shown in Fig. 1B.

To confirm that the apparent oscillations in luciferase luminescence reported genuine oscillations in [ATP]_cyt, three different statistical tests were performed. This was considered important, given the intrinsically low signal-to-noise ratio entailed by the low output of photon events (typically 3–10 photons · s⁻¹ · cell⁻¹) from the luciferase enzyme expressed in a single cell (27). DFT of the data (34) was carried out first to highlight the likely frequency with which [ATP]_cyt may be oscillating. The results of this analysis of the data in Fig. 1A are shown in Fig. 1C. From comparison of the amplitudes of the DFT over the frequency range, this analysis suggested that [ATP]_cyt displayed oscillations of an approximate period of 67 s (from the amplitude peak at a frequency of 15 mHz) at 3 mmol/l glucose and 133 s (7.5 mHz) at 15 mmol/l glucose, both with an amplitude of ∼9%.

Autocorrelation was used as a second indication of the approximate wavelength of oscillations (see research design and methods) (35). Autocorrelation of the data (Fig. 1D) obtained at 3 mmol/l glucose demonstrated a significant negative correlation when the data were offset by either 40 or 120 s (P < 5%; F test) and a positive correlation when the data were offset by 80 s (consistent with oscillation of a wavelength of ∼80 s [12.5 mHz]). Similar analysis of the data during incubation of this cell at 15 mmol/l glucose showed a significant negative correlation when offset by 60 s (P < 1%; F test) and a positive correlation when offset by ∼140 s (consistent with oscillation of a wavelength of ∼140 s [7.1 mHz]).

Using the information gained from DFT and autocorrelation, it was possible to refine the fit of a sine curve to the data set using nonlinear least-squares regression analysis (see research design and methods). The refined sine curves, fitted to the data obtained during incubation at 3 or 15 mmol/l glucose in Fig. 1A, are shown as bold (blue or red) traces. Both fitted sine curves showed a statistical improvement of fit compared with a horizontal line that was equal to the mean of the data (P = 1.5% for both traces, F test). For the data obtained during incubation in 3 mmol/l glucose, the fitted sine curve had a wavelength of 78 s (12.8 mHz) and amplitude of 11%. The corresponding data during incubation with 15 mmol/l glucose had a wavelength of 141 s (7.1 mHz) and amplitude of 9%. Thus, it can be seen that these parameters are fully consistent with both the DFT and autocorrelation analyses (above) and confirm the presence of oscillations in [ATP]_cyt.

Using the above analyses, significant [ATP]_cyt oscillations (i.e., P < 5% for fitted sine curve) were observed in 4 of 20 human β-cells during incubation at 3 mmol/l glucose and 9 (of 20) cells during incubation at 15 mmol/l glucose (separate experiments on islet cells from two different donors). The mean period of [ATP]_cyt oscillation showed a significant increase from 66 ± 6 to 167 ± 32 s upon elevation of glucose (P < 5%, unpaired t test). However, the amplitude of oscillations was unchanged from 13 ± 2 to 11 ± 1% at low and high glucose, respectively.

Oscillations in free [ATP]_cyt were also seen in primary β-cells isolated from mouse islets. In contrast to human islet cells, oscillations were not observed at 3 mmol/l glucose, i.e., only one cell displayed significant oscillatory behavior (P < 5%) of 27 cells analyzed (five separate experiments). During incubation at 16 mmol/l glucose, however, 9 of 27 cells gave significant oscillations. The normalized light output during incubation at 3 and 16 mmol/l glucose for a typical cell is given in Fig. 1E, and pseudo-color images of the cell at low glucose and during an oscillation are given in Fig. 1F. As with human (above) and rat (27) islet β-cells, the elevation of glucose concentration caused a clear increase in the average light output from mouse cells and hence in [ATP]_cyt. DFT and autocorrelation analyses of the data obtained at 16 mmol/l glucose are given in Fig. 1G and H, respectively. Least-squares fitting of a sine curve to the data gave a wavelength of 147 s (6.8 mHz) and amplitude of 15% (bold trace, Fig. 1E; P = 0.1%, F test). The average wavelength of the seven cells displaying significant oscillations was 112 ± 9 s, and the amplitude was 14 ± 1%. Oscillations of similar frequency were observed in cells expressing mitochondrially targeted luciferase (27) expressed via an adenoviral vector (18) (not shown).

We then explored the temporal relationship between oscillations in [ATP]_cyt and oscillations in free [Ca²⁺]_cyt to assess whether the two may be causally linked. In these experiments, the response to glucose of [Ca²⁺]_cyt was imaged in single β-cells using the ratiometric dye fura-2. Similar to the oscillations in [ATP]_cyt, oscillations in [Ca²⁺]_cyt were observed in 6 of 22 cells analyzed at 16 mmol/l glucose (eight separate experiments), whereas no significant [Ca²⁺]_cyt oscillations were seen during incubation at 3 mmol/l glucose. The mean period of the oscillations observed in [Ca²⁺]_cyt was 113 ± 17 s—the same period as that typically seen for oscillations in [ATP]_cyt (Fig. 1B). However, the oscillations in [Ca²⁺]_cyt appeared with a delay of 217 ± 29 s after elevation of glucose (assessed as the time to the beginning of the first down-slope of the oscillation). In contrast, the time of onset for oscillations in [ATP]_cyt was significantly shorter than that of the first [Ca²⁺]_cyt oscillations, at 136 ± 17 s (P < 5%; two-tailed unpaired t test).

The next series of experiments investigated the control that [Ca²⁺]_cyt exerted over free [ATP]_cyt and examined whether oscillations in [Ca²⁺]_cyt could form part of a feed-forward/feedback mechanism that could contribute to oscillations in [ATP]_cyt (8,38). To achieve improved temporal resolution and sensitivity compared with that possible when imaging single primary β-cells, the response of [ATP]_cyt to various regulators of [Ca²⁺]_cyt was measured in populations of the murine β-cell line MIN6. In this approach, MIN6 cells (∼300,000 per experiment) were infected with AdCMVcLuc and perifused with luciferin (5 μmol/l) to enable changes in light output (and hence [ATP]_cyt or mitochondrial [ATP]) to be detected at a level of <1% (18).

Increasing the glucose concentration from 3 to 30 mmol/l caused a time-dependent increase in intracellular free [Ca²⁺]_cyt, measured in single fura-2–loaded MIN6 cells (Fig. 2A and E). In cells infected with AdCMVcLuc (Fig. 2B and F), this step in glucose concentration caused a clear increase in luminescence, similar to that previously reported in single MIN6 and primary rat β-cells microinjected with the corresponding cDNAs (27). The onset of this increase in [ATP]_cyt occurred within 5 s of the increase in glucose concentration and before any detectable increase in [Ca²⁺]_cyt (not detectable before 30 s). At 3 mmol/l glucose, depolarization of the plasma membrane with 70 mmol/l KCl caused a dramatic biphasic increase in [Ca²⁺]_cyt and an increase in apparent [ATP]_cyt, which was stable for ∼100 s, but then decreased with time (Fig. 2D). Little or no time lag was observed between the initiation of these two changes upon KCl addition (Fig. 2C and D). By contrast, addition of 70 mmol/l KCl to cells in which [Ca²⁺]_cyt had previously been elevated by incubation for 5 min in the presence of 30 mmol/l glucose caused a brief “spiking” in light output followed by a clear decrease in [ATP]_cyt (Fig. 2F).

Removal of external Ca²⁺ ions with EGTA (39,40) decreased the [Ca²⁺]_cyt and also the magnitude of the [ATP]_cyt increase after exposure to 30 mmol/l glucose to <4% over basal (Fig. 3A and B) compared with a mean increase of 14.2 ± 0.9% (n = 12 separate experiments) seen in untreated cells. Thus, increases in [Ca²⁺]_cyt appeared to be necessary for effective glucose metabolism. However, in cells incubated at either 3 or 30 mmol/l glucose, addition of EGTA alone led to a drop in [Ca²⁺]_cyt (Fig. 3A and C) but caused a significant increase in luciferase luminescence (Fig. 3B and D). Thus, the final increase in [ATP]_cyt over basal achieved when glucose was increased to 30 mmol/l before the addition of EGTA was 60% compared with 33% when EGTA was added before the increase in glucose. Identical results were obtained by closure of l-type Ca²⁺ channels with verapamil as opposed to EGTA treatment (not shown).

The observation (Fig. 3B and D) that the lowering of [Ca²⁺]_cyt by incubation in Ca²⁺-free medium resulted in an ∼20% increase in light output from unstimulated MIN6 cells (Fig. 2B and F) indicated that lowering [Ca²⁺] suppressed basal ATP consumption. Under these conditions, changes in the activity of ATP-consuming processes, rather than in mitochondrial ATP synthesis, seem likely to underlie the increase in [ATP]_cyt because the change in [Ca²⁺]_cyt upon the blockade of Ca²⁺ influx occurred below the range (i.e., <100 nmol/l) at which mitochondrial oxidative metabolism is sensitive to [Ca²⁺]_cyt (19).

We considered the SERCA pumps as a principal candidate for the Ca²⁺-sensitive ATP-consuming processes (29). The contribution of these Ca²⁺ pumps to cellular ATP turnover was investigated by measuring the effect of their inhibition with 10 μmol/l cyclopiazonic acid (CPA) on light output in luciferase-expressing MIN6 cells.

Figure 4A and B show the responses of [Ca²⁺]_cyt and [ATP]_cyt to incubation with CPA of MIN6 cells under basal (3 mmol/l glucose) conditions. The responses to CPA during stimulation with either 30 mmol/l glucose or 70 mmol/l KCl are shown in Fig. 4C and D. CPA caused a transient elevation of [Ca²⁺]_cyt (Fig. 4A), presumably due to the release of Ca²⁺ from stores in the endoplasmic reticulum and a large increase in [ATP]_cyt even under basal conditions (Fig. 4B). Increases in [ATP]_cyt were also evident upon CPA incubation in the presence of both high KCl and glucose concentrations. Overall, these results suggest that the SERCA pump is very active even in unstimulated cells, representing an important Ca²⁺-dependent ATP-consuming process.

Using adenovirus-driven expression of the ATP-dependent bioluminescent enzyme, firefly luciferase (27,37), we were able to image oscillations in free [ATP]_cyt in individual primary β-cells from mouse and human islets. In both cases, [ATP]_cyt displayed oscillatory changes that were affected by glucose concentration (Fig. 1). In human β-cells, [ATP]_cyt oscillations were evident during incubation at either low (3 mmol/l) or high (15 mmol/l) glucose concentrations, whereas in mouse β-cells, [ATP]_cyt oscillations were only seen at elevated glucose concentrations. The origin of this difference may lie in the respective sensitivities of β-cells from these two species to glucose. The normal blood glucose concentration in humans (4–5 mmol/l) is substantially lower than that in mice (8–10 mmol/l) and thus, 3 mmol/l glucose may be below a threshold for which oscillations occur in mouse β-cells but above the threshold for human β-cells. In contrast to mouse and human β-cells (this study), we have previously reported that oscillations in [ATP]_cyt were not apparent in dissociated rat islet β-cells (27) or whole islets (18) at either high or low glucose concentrations, consistent with the failure of others (41) to detect [Ca²⁺]_cyt oscillations in rat islets (although low-frequency oscillations in mitochondrial [Ca²⁺] were evident in our own studies) (18). These observations suggest that important differences exist in the mechanisms responsible for secretory oscillations in different mammalian species.

Upon elevation of glucose concentration, the period of [ATP]_cyt oscillations in human β-cells displayed a clear increase from 60 to 170 s, but there was no apparent change in amplitude of the oscillations. Interestingly, these oscillations are much faster than those of insulin secretion measured in living subjects (12–15 min) (9,35), suggesting that some entrainment of the [ATP]_cyt pulses may occur in the context of the intact islet, or within the pancreas, by neural or other means. Nevertheless, a small number of cells displayed oscillations in [ATP]_cyt at 3 mmol/l glucose, as has been reported for [Ca²⁺]_cyt oscillations measured in dissociated human islet β-cells ex vivo (42). Furthermore, the increase in wavelength of [ATP]_cyt oscillations in response to elevated glucose is similar to the behavior of [Ca²⁺]_cyt oscillations observed in human islets (28). However, it must be appreciated that isolation and culture conditions can profoundly influence the glucose-induced oscillatory behavior of β-cells (43).

In mouse β-cells, the period of [ATP]_cyt and [Ca²⁺]_cyt oscillations at elevated glucose concentrations was identical, at ∼120 s, in cells from identical preparations. The period of these oscillations in [ATP]_cyt was compatible with the “slow wave” oscillations seen in islet [Ca²⁺]_cyt and insulin secretion (1,2). Thus, it appears that the structure of the β-cell oscillator is inherent to individual cells and is not a property of the islet, although the organization of cells within the islet may influence oscillatory behavior (1).

Despite the high levels of luciferase expression achieved with the adenoviral vector, it was not possible to observe dynamic changes in light output with sufficient temporal resolution to differentiate rapid oscillations, of a period of 5–20 s, as seen in [Ca²⁺]_cyt and oxygen consumption (43,44). However, the present results demonstrate the coherent behavior of oscillations in both [ATP]_cyt and [Ca²⁺]_cyt in the β-cell and the likely interdependence of each.

Given the strong regulatory influence that ATP exerts over Ca²⁺ influx into the β-cell (by modulating K_ATP channel activity), it is possible that an oscillator that solely governed ATP production, such as a proposed glycolytic oscillator (45), would also be able to drive oscillations in [Ca²⁺]_cyt. However, it is widely acknowledged that, in a variety of cell types, [Ca²⁺]_cyt is a potent activator of mitochondrial metabolism (19,21). It therefore seems more likely that oscillations in β-cell [Ca²⁺]_cyt influence ATP production and thus form part of the basis of the β-cell oscillatory mechanism.

To explore the regulation by Ca²⁺ of mitochondrial ATP production, we investigated the effect of modulations in [Ca²⁺]_cyt on [ATP]_cyt in populations of MIN6 cells during stimulation with high glucose concentrations. Elevation of glucose concentration led to increases in [ATP]_cyt and [Ca²⁺]_cyt in MIN6 cells. An increase in [Ca²⁺]_cyt at low glucose concentrations (3 mmol/l) in response to membrane depolarization with KCl also led to a transient elevation of [ATP]_cyt (Fig. 3A–D). Furthermore, blockade of Ca²⁺ influx with verapamil, or removal of external Ca²⁺, inhibited the activation of ATP generation in response to 30 mmol/l glucose by >75% (Fig. 4B and F). Thus, these data demonstrate the requirement for Ca²⁺ influx for the full activation of glucose metabolism, as previously suggested by measurements of NAD(P)H fluorescence in mouse islets (40). Full [ATP]_cyt increases in response to subsaturating concentrations of two substrates of mitochondrial metabolism, leucine and methylsuccinate, also required Ca²⁺ influx (results not shown). Elevated [Ca²⁺]_cyt increased the apparent sensitivity of mitochondrial oxidation for both substrates but not the maximal apparent ATP production (not shown). The requirement for maintained intracellular Ca²⁺ concentrations for the metabolism of these substrates, as well as glucose, therefore probably reflects the activation of mitochondrial ATP production by Ca²⁺ (19,23,46).

In contrast, in cells where [Ca²⁺]_cyt (and hence mitochondrial ATP synthesis) had already been elevated by incubation with 30 mmol/l glucose, a further dramatic increase in [Ca²⁺]_cyt, imposed by cell depolarization with KCl, led to a fall in [ATP]_cyt (Fig. 2F). This result, which is consistent with previous measurements of total cellular ATP and ADP in intact islets (38), indicates that excessive accumulation of Ca²⁺ by mitochondria exerts either no effect or an inhibitory effect (24,25,47) on mitochondrial ATP synthesis, i.e., the ATP production subsystem is effectively saturated with Ca²⁺. Interestingly, such an inhibitory effect seems unlikely to involve opening of the mitochondrial permeability transition pore (26) because it was unaffected by the potent inhibitor of mitochondrial permeability transition pore opening, cyclosporin A (not shown).

In addition to the above effects on mitochondrial function, the [Ca²⁺]_cyt rise induced by KCl in the presence of an elevated glucose concentration is also likely to stimulate ATP-consuming events in the cytosol. These events seem likely to include Ca²⁺ pumping into the endoplasmic reticulum (Fig. 4) and possibly out of the cell. We have previously considered the possibility that the activation of insulin synthesis, as well secretory vesicle movement (48) and exocytosis (49), may contribute significantly to the consumption of ATP in response to increases in [Ca²⁺]_cyt. Surprisingly, however, inhibition of secretion with clonidine (an α₂-adrenoreceptor agonist) (18) or diazoxide (50) had little effect on [ATP]_cyt, indicating that ion pumping may represent that the more important ATP sink under these conditions. Whatever the mechanism(s) responsible for the observed [ATP]_cyt changes, the ability of an increase in [Ca²⁺]_cyt (and/or mitochondrial [Ca²⁺]) to reduce [ATP]_cyt supports the proposal of Detimary et al. (38) and Krippeit-Drews et al. (47) that increases in [Ca²⁺]_cyt may act in a feedback loop to decrease [ATP]_cyt. This is then likely to lead to the opening of K_ATP channels and thus initiate the down-stroke of electrical and [Ca²⁺]_cyt oscillations in islet β-cells.

Overall, Ca²⁺-regulated processes, including Ca²⁺ ion pumping, appear to contribute a considerable proportion of total cellular ATP turnover. Thus, lowering cytosolic [Ca²⁺] apparently diminishes mitochondrial ATP synthesis to a much smaller extent than the decrease in extramitochondrial ATP consumption, giving a net increase in [ATP]_cyt (Fig. 3). It should be pointed out that the lowering in basal [Ca²⁺]_cyt by verapamil in MIN6 cells may not be typical of primary β-cells (51) and may reflect partially altered Ca²⁺ homeostasis in the derived cell line. The fact that the increase in [ATP]_cyt upon Ca²⁺ removal was greater after incubation with 30 mmol/l glucose (Fig. 3D vs. 3B) may be indicative of the increased activity of the Ca²⁺-dependent ATP-consuming reactions during elevation of [Ca²⁺]_cyt by glucose (although it should be stressed that this cannot be proven by measurement of [ATP]_cyt alone).

We propose that the negative regulation of [ATP]_cyt by Ca²⁺ may be due to SERCA pump activity, since we found the latter to be a major ATP-consuming process in these cells judged by the large positive effect of CPA on [ATP]_cyt and the fact that treatment with verapamil severely diminished the effect of CPA on [ATP]_cyt (not shown). In contrast, vanadate ions, at a concentration reported to inhibit plasma membrane Ca²⁺-ATPase (10 μmol/l) (52), was without effect on [ATP]_cyt at resting and glucose-stimulated conditions. Only under KCl stimulation did vanadate evoke an increase in [ATP]_cyt (not shown), suggesting that under most conditions, the plasma membrane Ca²⁺-ATPase does not significantly control [ATP]_cyt. However, it should be mentioned that the efficacy of vanadate in inhibiting plasma membrane Ca²⁺-ATPases has been questioned in the β-cell type (53).

A possible cycle of events that may lead to oscillations in both [ATP]_cyt and [Ca²⁺]_cyt is outlined in Fig. 5. In this scheme, which was developed from the schemes of Jung et al. (8) and Detimary et al. (38), an increase in glucose concentration leads to a limited increase in [ATP]_cyt, which then triggers K_ATP channel closure and a small influx of Ca²⁺ into the cell (1 in Fig. 5). The resulting increase in [Ca²⁺]_cyt begins to activate mitochondrial metabolism, permitting the accelerated metabolism of glucose carbon atoms by mitochondria, thus creating a feed-forward loop that further enhances the increase in [ATP]_cyt (2 in Fig. 5). As [Ca²⁺]_cyt increases still further, insulin secretion will be stimulated but the activation of mitochondrial ATP production is saturated. At this point, further influx of [Ca²⁺]_cyt results in a decrease in [ATP]_cyt due to increased ATP consumption (3 in Fig. 5). This will result in [ATP]_cyt falling to sufficiently low levels that Ca²⁺ influx is reduced, and [Ca²⁺]_cyt decreases because of net Ca²⁺ efflux (4 in Fig. 5). The now-reduced activity of Ca²⁺-dependent ATP-consuming reactions means that ATP production outweighs ATP consumption, leading to increasing [ATP]_cyt, and the cycle starts again.

We propose that the critical decrease in [ATP]_cyt during oscillations is due to increased ATP consumption activated by Ca²⁺. In agreement with Detimary et al. (38), we suggest that increased ATP consumption is due to increased ion cycling rather than secretion alone. Ion cycling involving the SERCA pumps is a major contributor to ATP consumption (Fig. 4B) and is therefore a strong candidate to be the Ca²⁺-dependent ATP-consuming process capable of driving [ATP]_cyt oscillations. Nevertheless, it must be stressed that the above conclusion does not preclude the contribution of other cytosolic ATP-consuming processes nor the contribution of changes in cytosolic [ATP]/[ADP] to the regulation of glycolysis (8,15). However, a simple glycolytic oscillator, of the type proposed by Tornheim (15), would seem alone insufficient to explain the metabolic oscillations that presumably underlie pulsatile insulin secretion.

After reviewing the mechanism outlined in Fig. 5, we propose that oscillations in [ATP]_cyt and [Ca²⁺] are out of phase with each other, as suggested by others (8), on the basis of measurements of oxygen and glucose consumption by intact islets. Crucially, we observed that the appearance of an [ATP]_cyt increase in response to elevated glucose concentration preceded that of [Ca²⁺]_cyt by approximately half the wavelength of the oscillations in single mouse β-cells. This finding is consistent with earlier measurements of total extractable adenine nucleotides (38), measurements of the onset of NAD(P)H increases and [Ca²⁺]_cyt oscillations (6,40), and K_ATP channel activity (54). Further investigations will require the development of new techniques to investigate the relative phase of oscillations in [Ca²⁺] and [ATP]_cyt, as well as oxygen and glucose consumption, and insulin secretion in the same single β-cell or islet. Such studies should be particularly instructive in assessing the relative strengths of the regulatory mechanisms described here and thus to provide a fuller picture of the β-cell oscillator.

This research was funded by project grants to G.A.R. from the Wellcome Trust (U.K.), the Biotechnology and Biological Sciences Research Council, the Medical Research Council (MRC) (U.K.), and Diabetes U.K. E.K.A. was the recipient of a Research Training Fellowship of the MRC (Ref. G81/431).

We thank the MRC for providing an Infrastructure Award to establish the School of Medical Sciences Cell Imaging Facility and Alan Leard for his assistance. We would also like to thank Dr. Roger F.L. James (University of Leicester, Leicester, U.K.) for the provision of human islets and Dr. Aniko Varadi for comments on the manuscript.