The canonical model of glucose-stimulated insulin secretion (GSIS) by pancreatic β-cells predicts a glucose-induced rise in the cytosolic ATP/ADP ratio. Such bioenergetic sensitivity to metabolic fuel is unusual as it implies that ATP flux is governed, to a significant extent, by ATP supply, while it is predominantly demand-driven in other cell types. Metabolic control is generally shared between different processes, but potential control of ATP consumption over β-cell bioenergetics has been largely ignored to date. The present paper offers a brief overview of experimental evidence that demonstrates ATP flux control by glucose-fuelled oxidative phosphorylation. Based on old and new data, it is argued that ATP supply does not hold exclusive control over ATP flux, but shares it with ATP demand, and that the distribution of control is flexible. Quantification of the bioenergetic control distribution will be important from basic and clinical perspectives, but precise measurement of the cytosolic ATP/ADP ratio is complicated by adenine nucleotide compartmentalisation. Metabolic control analysis of β-cell bioenergetics will likely clarify the mechanisms by which glucose and fatty acids amplify and potentiate GSIS, respectively. Moreover, such analysis may offer hints as to how ATP flux control shifts from ATP supply to ATP demand during the development of type 2 diabetes, and why prolonged sulfonylurea treatment causes β-cell deterioration.
Glucose-stimulated insulin secretion
Glucose homeostasis is maintained through biochemical interplay between multiple organs. When the blood glucose concentration increases after the consumption of food, pancreatic β-cells secrete insulin, which adjusts the metabolism of various tissues — including the skeletal muscle, liver, white adipose, and brain — to the risen nutrient availability so that blood glucose is restored to its pre-prandial level. Classical glucose-stimulated insulin secretion (GSIS) by β-cells depends on oxidative fuel catabolism, as a glucose-induced rise in the cytosolic ATP/ADP ratio sets of a series of events that eventually trigger insulin release . An elevated blood glucose concentration stimulates glucose-fuelled oxidative phosphorylation and thus raises the ATP/ADP ratio. This rise leads sequentially to closure of ATP-sensitive potassium (KATP) channels, plasma membrane depolarisation, activation of voltage-dependent calcium channels, an increased cytosolic-free calcium level, and exocytosis (Figure 1A). Moreover, β-cells rely entirely on oxidative nutrient breakdown to generate the phosphorylation potential needed to drive ATP-demanding processes, because they do not express sufficient lactate dehydrogenase to facilitate anaerobic glycolysis . In addition to general cellular upkeep, the β-cell energy expenditure includes synthesis and processing of pro-insulin, as well as vesicle trafficking and exocytosis to secrete the produced insulin.
Glucose stimulation of the cytosolic ATP/ADP ratio is the key signal for canonical GSIS.
The glucose responsiveness of the β-cell cytosolic ATP/ADP ratio has been reviewed recently by Nicholls . The cytosolic ATP/ADP ratio is not generally glucose-sensitive in well-characterised cells such as muscle and liver, and the unusual bioenergetics that are likely responsible for the sensitivity in β-cells are not fully understood. Cellular glucose-sensing ability is often attributed to the kinetic characteristics of the hexokinase-4 isozyme expressed in β-cells, which is also known as glucokinase . Since glucokinase has a relatively low substrate affinity, β-cells do not metabolise glucose below the circulating concentration set point of 4.5–5 mM under fasting conditions. In addition, glucokinase has near-complete control over glycolysis , and its cooperative glucose kinetics indeed mirror the sigmoidal glucose-dependency of insulin secretion. It is unlikely, however, that canonical GSIS is controlled by glucokinase alone, as hepatocytes express the same isozyme  without bioenergetic glucose sensitivity.
To explain how canonical GSIS can be amplified  and potentiated  by glucose and fatty acids, respectively, additional nutrient–secretion coupling mechanisms have been reported that are not linked to changes in the cytosolic ATP/ADP ratio. Universal acceptance of these non-canonical mechanisms, however, requires conclusive identification of coupling factors . To exclude that GSIS amplification and potentiation are ‘mere’ bioenergetic phenomena — which would make them special cases of the canonical model — quantification of cytosolic phosphorylation potential is crucial, but measurement of the relevant free adenine nucleotide levels is complicated by their compartmentalisation and tendency to bind to protein .
Determination of ATP/ADP ratios
Similar to other cell types, β-cells possess distinct cytosolic and mitochondrial adenine nucleotide pools (Figure 1A). Notably, β-cells also carry adenine nucleotides in their insulin secretory granules with both ATP and ADP present at ∼4 pg/µg granular protein . These adenine nucleotides are co-released with insulin and are thought to regulate insulin secretion via autocrine signalling . β-cells each hold over 10 000 insulin granules, which constitute 10–20% of total β-cell volume , and vesicular adenine nucleotides may account for up to 50% of the β-cell's total adenine nucleotide content . The vesicular ATP/ADP ratio is rather unresponsive to metabolic cues and thus dampens glucose effects on cytosolic phosphorylation potential . By inference, bioenergetic effects of non-canonical metabolic coupling factors may go unnoticed for the same reason. The proportionally high equimolar levels of vesicular ATP and ADP are largely responsible for the relatively low whole-cell ATP/ADP ratios that have been reported, with values not much higher than 10–20 (see references cited in ref. ) even when β-cells or islets are exposed to stimulatory glucose levels. In accordance, the whole-cell ATP/ADP ratios we measure in INS-1E insulinoma cells (a widely used β-cell model ) fall within this range (Figure 1B) and, as expected, ATP/ADP ratios decrease when oxidative phosphorylation is uncoupled with FCCP (carbonyl cyanide-p-trifluoromethoxyphenylhydrazone). The values we see in cells incubated at a non-stimulatory glucose level are higher than those generally reported under such basal conditions (cf. ) and are a likely additional reason for the apparently low glucose sensitivity we detect (Figure 1B). Whole-cell ATP/ADP ratios also tend to be relatively low because of mitochondrial adenine nucleotides. Mitochondrial phosphorylation potential is generally much lower than cytosolic potential because of the electrogenic nature of ATP and ADP exchange and because of the pH-driven import of inorganic phosphate .
It is difficult to estimate how much the cytosolic ATP/ADP ratio is underestimated by other adenine nucleotide pools, and compartment-specific measurements of adenine nucleotides not bound to protein are needed for precise quantification. Cytosolic ATP/ADP ratios can be determined specifically with fluorescent fusion constructs [13,14], but the dynamic range of such constructs is presently insufficient to detect the high cytosolic ratios that are, for example, predicted (cf. the legend of Figure 1) by the phosphocreatine (PCr) and creatine (Cr) levels reported for skeletal muscle . Moreover, a live-cell imaging approach does not allow quantification of absolute cytosolic ATP/ADP ratios unless signals can be calibrated . Pancreatic β-cells contain a PCr/Cr pool . This pool is entirely cytoplasmic and not bound to protein, and may thus be used to monitor the ratio of free cytosolic ATP and ADP because the reaction catalysed by creatine kinase (Cr + ATP ⇔ PCr + ADP) is at equilibrium . When derived this way in β-HC9 β-cells [17,18], ATP/ADP ratios are obtained with values that are indeed similar to the high cytosolic phosphorylation potentials seen in skeletal muscle. When cultured with 3 mM Cr, we detected a PCr pool in INS-1E cells that responds to glucose and FCCP as predicted by the canonical GSIS model. Cytosolic ATP/ADP ratios calculated from measured PCr and Cr levels (Figure 1C) agree with published β-HC9 values . The precise role of the PCr/Cr pool in β-cells remains to be established conclusively, but a PCr/Cr shuttle has been suggested to facilitate transmission of phosphorylation potential between mitochondria and plasma membrane KATP channels .
Control of ATP flux by ATP supply
The ATP/ADP ratio of mammalian cells is generally fairly stable. Most cell types aim to keep a constant phosphorylation potential far removed from thermodynamic equilibrium, so that it can drive endergonic processes under a wide range of energetic demands . For instance, the skeletal muscle respiratory activity may fluctuate three orders of magnitude without discernible effect on the cytosolic ATP/ADP ratio. This relative stability is a direct consequence of the strong control exerted by energy expenditure over myocellular ATP flux. Demand-driven ATP flux ensures that the cytosolic ATP/ADP ratio is kept under tight homeostatic control . In contrast, the apparent lack of such homeostasis in β-cells implies that total ATP flux in these cells must be largely controlled by ATP supply. This implication, indeed, follows from a GSIS model in which full glucose breakdown via oxidative phosphorylation induces an increase in the cytosolic ATP/ADP ratio (Figure 1A).
β-cell respiratory analyses generally offer clear evidence for the strong control of ATP flux by ATP supply: basal oxygen uptake in insulinoma cells as well as in rodent and human islets is increased 1.5- to 3-fold by glucose [21–36]. From oligomycin-sensitive respiration, we calculated that glucose raises the ATP supply flux in fuel-deprived INS-1E cells up to almost six times (Figure 2A). The glucose effect on ATP supply is comparably strong because most of the respiratory rise is coupled to phosphorylation [25,36], while the proportion of basal respiration used to make ATP is exceptionally low [28,33,37–39]. Glucose increases ATP synthesis-coupled respiration more than it raises respiration linked to mitochondrial proton leak, which means that it stimulates both the rate and coupling efficiency of oxidative phosphorylation [26,36].
Fuel stimulation of oxidative phosphorylation.
Glucose stimulation of mitochondrial coupling efficiency is decreased by uncoupling protein-2 , a carrier protein that contributes to the very high proton leak of β-cells . Attenuated bioenergetic glucose sensitivity is mirrored by dampened GSIS , which suggests a role for mitochondrial proton leak in the regulation of glucose-sensing. A recent metabolic control analysis has, indeed, demonstrated that proton leak exerts significant control over both the magnitude and homeostasis of the mitochondrial membrane potential in insulinoma cells incubated at low glucose levels . Low mitochondrial ATP synthesis efficiency thus appears to ensure that the membrane potential and the mitochondrial and cytosolic ATP/ADP ratios are kept relatively low when insulin secretion is not required. Interestingly, in this respect, ATPase inhibitory factor 1 has recently also been shown to dampen insulin secretion from insulinoma cells at basal glucose concentrations through inhibition of ATP synthesis .
At stimulatory glucose levels, homeostatic control of the mitochondrial membrane potential is almost exclusively exerted by glucose oxidation, a process that is regulated itself by factors that arise downstream from ATP synthesis . These factors have not yet been identified, but positive feedback reinforces bioenergetic responsiveness to glucose and supports the notion that glucose-sensing by β-cells is not accounted for by glucokinase alone (cf. the section ‘Glucose-stimulated insulin secretion’). This notion is further supported by the β-cell responsiveness to pyruvate [33,36,41] that can enter insulinoma cells when added as sodium pyruvate and primary β-cells when given as methyl-pyruvate. We find that INS-1E cells increase their respiratory activity to the same extent in response to 5 mM pyruvate as to 20 mM glucose (Figure 2B). This bioenergetic response is reflected by pyruvate-stimulated insulin secretion (not shown) and suggests that nutrient-sensing is at least partly controlled by mechanisms downstream from glycolysis. The stimulation of respiration by metabolic fuels is amplified when oxidative phosphorylation is uncoupled (Figure 2C), which suggests that ATP demand also exerts some control over fuel sensitivity of respiration, and by extension ATP flux, under coupled conditions.
Control of ATP flux by ATP demand
Approximately 50% of the total protein synthesis rate in β-cells can be attributed to pro-insulin , which implies that β-cells need much energy to make insulin. Moreover, trafficking, priming, and exocytosis of insulin secretory vesicles are processes that require energy [43,44]. Possible control of ATP demand over cellular bioenergetics has not been quantified in β-cells, although indirect qualitative evidence for such control exists [21,23,24]. Cells in which total ATP flux is demand-driven, allow allocation of ATP supply to energy-demanding processes to be approximated from sensitivity of oligomycin-sensitive respiration to experimental inhibition of these processes . Acute cycloheximide effects on oxidative phosphorylation, mediated through increased cytosolic phosphorylation potential, have thus shown that 20–35% of ATP supply is used for protein synthesis in rat thymocytes , human embryonic stem cells , as well as in rat and human myocytes . It is worth noting that the reported percentages may be slight overestimations due to non-specific inhibitor effects .
Cycloheximide also appears to lower the respiration of β-cells, a system in which ATP flux is significantly supply-driven as predicted by the canonical GSIS model (cf. the section ‘Control of ATP flux by ATP supply’). Glucose-stimulated oxygen uptake by INS-1E cells (Figure 3A) and mouse islets (Figure 3B) is inhibited acutely after the addition of 5 µM cycloheximide. This inhibition is reversed when oxidative phosphorylation is uncoupled, which supports the notion that it is mediated via the cytosolic ATP/ADP ratio and excludes the possibility that it arises from cycloheximide effects on fuel oxidation per se. From the inhibitory cycloheximide effects on oligomycin-sensitive respiration, we estimate that as much as 40% of total ATP supply (cf.Figure 2A) is used to make protein in INS-1E cells and ∼30% in mouse islets. Perhaps unsurprisingly, the absolute ATP flux coupled to protein synthesis is stimulated by glucose (Figure 3C), which is likely owing to lowered AMP-activated protein kinase activity that follows from the glucose-induced increase in cytosolic phosphorylation potential .
Oxidative phosphorylation linked to protein synthesis.
Acute indirect inhibition of oxidative phosphorylation by cycloheximide demonstrates that energy expenditure linked to protein synthesis exerts significant control over total ATP flux in β-cells. Therefore, ATP flux control is not exclusively derived from glucose-fuelled ATP supply, as might be expected from the bioenergetic glucose sensitivity of β-cells, but also arises from ATP demand. This shared flux control between ATP supply and ATP demand is consistent with differential effects of glucose and sulfonylureas (antidiabetic drugs that act on KATP channels ) on β-cell bioenergetics reported by others [21,23,24]. In accordance with flux control from ATP supply, glucose-induced increases in oxygen uptake are concomitant with increased phosphorylation potential. The increase in oxygen consumption provoked by sulfonylureas (tolbutamide or glibenclamide), on the other hand, coincides with decreased phosphorylation potential [21,23,24], which suggests that the respiratory stimulation is secondary to increased ATP demand. Such increased demand likely emerges from the insulin secretion provoked by sulfonylureas through their inhibition of KATP channels .
Why quantitative insight in bioenergetic control is important
The limited information that is available on shared ATP flux control between fuel breakdown and energy expenditure in β-cells is entirely qualitative at present. It will be important from basic and clinical science perspectives to quantify the relative importance of ATP supply and ATP demand for bioenergetic control in pancreatic β-cells. Besides the ATP flux control discussed above, such quantification should also focus on the concentration control exerted by ATP-producing and ATP-consuming processes over the cytosolic ATP/ADP ratio, as it is this ratio that is crucial for nutrient–secretion coupling. It is worth stressing that in the context of this quantitative analysis, the term ‘control’ differs from ‘regulation’ as it reflects a potential to regulate . Distributed control between ATP supply and ATP demand would thus give the β-cell the potential to regulate ATP flux and GSIS by altering either demand or supply.
Modular kinetic and metabolic control analyses have provided quantitative information on oxidative phosphorylation in β-cells  and in isolated β-cell mitochondria . Extending such analyses to cellular energy metabolism surrounding the cytosolic ATP/ADP ratio might offer insight in some unresolved issues in β-cell biology. For instance, amplification and potentiation of GSIS by glucose and fatty acids, respectively, may be owing to an increased ATP supply capacity that would be better able to meet the high energy demand from prolonged insulin secretion. Indeed, sulfonylurea-provoked insulin secretion, which can be amplified by glucose, is reflected by bioenergetic changes that are indicative of increased ATP demand [21,23,24]. This model aligns well with classical GSIS where bioenergetic glucose sensitivity alone mediates nutrient–secretion coupling, and would explain why non-canonical coupling factors remain to be identified conclusively.
Bioenergetic glucose sensitivity implies that total ATP flux is controlled by glucose-fuelled ATP supply (Figure 4A), and the loss of this sensitivity in type 2 diabetes (T2D) [53,54] indicates that ATP flux control by ATP supply has decreased . The loss of glucose sensitivity in T2D can be managed therapeutically with sulfonylureas that induce insulin release downstream from mitochondria (Figure 4B). This management shows that glucose-unresponsive β-cells remain bioenergetically competent as ATP is still supplied at the rate necessitated by the high energy demand from insulin synthesis and insulin secretion. It thus appears that control over ATP flux in β-cells shifts from ATP supply to ATP demand during the development of T2D (Figure 4C) . Metabolic control analysis should clarify whether or not this apparent flux control shift is accompanied by a change in concentration control over the ATP/ADP ratio. Although measured ATP flux and phosphorylation potential will likely be temporal averages in such analysis, and may thus be assumed at steady state as a first approximation, it will be important to account for known oscillatory behaviour of β-cells in due course.
Does ATP flux control shift during the development of type 2 diabetes?
Bioenergetic differences between pancreatic islets from healthy individuals and those with T2D have been shown [38,55–57], but it is unclear which differences account for the apparent shift in ATP flux control. Notably, however, the loss of bioenergetic glucose responsiveness in islets from individuals with T2D can be restored by glucokinase activators, and impaired GSIS is consequently rescued . This effect underscores the importance of glucokinase for GSIS, but also leaves open the possibility that this enzyme is the endpoint of the recently discovered positive feedback loop that is affected by a factor that emerges downstream from mitochondrial ATP synthesis . In this model, bioenergetic glucose sensitivity would be lost in T2D because of failing feedback. More generally, quantitative insight in the control of β-cell bioenergetics will probably help rationalising therapeutic drug targets.
Finally, it is worth considering that the apparent bioenergetic control shift in dysfunctional β-cells is caused by the persistently high functional demand on β-cells in the insulin-resistant state that leads up to T2D . In this respect, sulfonylurea management of T2D will likely exacerbate the situation as these drugs induce insulin secretion constitutively with a chronic increase in energy demand as result. The inability to meet such demand possibly explains why β-cell function deteriorates during prolonged sulfonylurea treatment .
C.A. conceived the original work included in this review and wrote the manuscript. B.A., J.B., J.E.C., and A.G.W. designed, executed, and analysed the experiments reported in Figures 2A, 3A,C (B.A.), 2B,C, 3B (J.B.), 1B (J.E.C.), and 1B (A.G.W.). All authors commented on drafts of the manuscript and approved its final version.
Research described in this review that originates from C.A.'s laboratory was funded by the Medical Research Council (New Investigator Research grant number G1100165 to C.A.), the Daphne Jackson Trust (fellowship to A.G.W.), and the University of Plymouth (PhD studentship to J.B.). Neither Medical Research Council nor the Daphne Jackson Trust and the University of Plymouth were involved in the writing of the manuscript and the design, execution, analysis, and interpretation of the included original experiments. It was the decision of the authors only to submit the manuscript for publication.
We thank the anonymous reviewers for constructive feedback on the original version of the present paper.
The Authors declare that there are no competing interests associated with the manuscript.
Present address: Brighton and Sussex Medical School, University of Sussex, Falmer, BN1 9PX Brighton, U.K.
Present address: School of Sport, Exercise and Rehabilitation Sciences, University of Birmingham, R25, 142 Edgbaston Park Road, B15 2TT Birmingham, U.K.