An agreed-upon consensus model of glucose-stimulated insulin secretion from healthy β-cells is essential for understanding diabetes pathophysiology. Since the discovery of the KATP channel in 1984, an oxidative phosphorylation (OxPhos)–driven rise in ATP has been assumed to close KATP channels to initiate insulin secretion. This model lacks any evidence, genetic or otherwise, that mitochondria possess the bioenergetics to raise the ATP/ADP ratio to the triggering threshold, and conflicts with genetic evidence demonstrating that OxPhos is dispensable for insulin secretion. It also conflates the stoichiometric yield of OxPhos with thermodynamics, and overestimates OxPhos by failing to account for established features of β-cell metabolism, such as leak, anaplerosis, cataplerosis, and NADPH production that subtract from the efficiency of mitochondrial ATP production. We have proposed an alternative model, based on the spatial and bioenergetic specializations of β-cell metabolism, in which glycolysis initiates insulin secretion. The evidence for this model includes that 1) glycolysis has high control strength over insulin secretion; 2) glycolysis is active at the correct time to explain KATP channel closure; 3) plasma membrane–associated glycolytic enzymes control KATP channels; 4) pyruvate kinase has favorable bioenergetics, relative to OxPhos, for raising ATP/ADP; and 5) OxPhos stalls before membrane depolarization and increases after. Although several key experiments remain to evaluate this model, the 1984 model is based purely on circumstantial evidence and must be rescued by causal, mechanistic experiments if it is to endure.
Introduction
The importance of glucose metabolism for triggering insulin secretion was well established by the 1960s, when glucokinase was identified as the β-cell glucose sensor. The requirement for electrical excitability followed shortly after, and the 1984 discovery of the β-cell KATP channel provided a metabolic actuator for plasma membrane electrical activity that is sensitive to the ATP/ADP ratio and initiates Ca2+ influx (1). In the subsequent four decades, mitochondrial oxidative phosphorylation (OxPhos) has been positioned as the exclusive, intervening step that raises cytosolic ATP/ADP (ATP/ADPc) to close KATP channels and initiate insulin secretion (1). While the roles of glucokinase and KATP channels have been codified by gain- and loss-of-function mutations in humans (causing either hypoglycemia or neonatal diabetes), is there anything more than circumstantial, correlative evidence for the hypothesis that a glucose-dependent increase in mitochondrial ATP production closes KATP channels? As reviewed by Rutter and Sweet (2), the answer is no, whereas causative mechanistic experiments suggest that glycolysis has preferential, if not exclusive, control over KATP channel closure.
β-Cell glucose is almost completely oxidized in the mitochondria due to the absence of pyruvate/lactate transporters. As such, often-cited evidence for the canonical model is that mitochondrial poisons (rotenone, oligomycin, etc.) prevent KATP channel closure (3). Rutter and Sweet (2) point out a flaw in this argument: since glycolytic and mitochondrial metabolism are functionally entangled, perturbations of one will impact the other. For instance, poisoning mitochondria hinders not only glycolysis but also other nonoxidative mitochondrial fluxes (e.g., the phosphoenolpyruvate [PEP] cycle, which contributes glycolytic metabolites). However, the tight coupling between glycolysis and mitochondria has been used to argue that the number of ATPs produced by glycolysis is dwarfed by OxPhos in order to support the canonical view.
We do not dispute in β-cells that 1) glucose elevation increases OxPhos, 2) mitochondria are essential to stimulus-secretion coupling, and 3) mitochondrial health and dysfunction in diabetes should be aggressively studied. We simply argue that an OxPhos-independent increase in ATP/ADP, driven by glycolysis, closes KATP channels. This postulate is supported by recent evidence demonstrating that deletion of the respiratory chain complexes severely impairs OxPhos but does not impact insulin secretion (4). What, then, is the relevant mechanism? We propose that the spatial privilege and favorable bioenergetics of the pyruvate kinase (PK) reaction are what matter for KATP closure (Fig. 1).
Spatial Regulation of β-Cell Stimulus-Secretion Coupling
The primary evidence in support of the canonical model is that a rise in NAD(P)H, ATP/ADPc, and O2 consumption precedes Ca2+ influx (5–7). This evidence not only is corollary but also is commonly obtained by rescuing ATP/ADPc from a fuel-starved state to a fuel-saturated one. This situation never occurs in vivo. In both the fed and fasted states, insulin secretion oscillates with a 5- to 10-min period; glucose increases the pulse amplitude rather than “switching on” secretion (8,9). When islet metabolic signaling is observed during such physiological oscillations, multiple approaches suggest that OxPhos turns off before KATP channels close and reactivates after membrane depolarization and Ca2+ influx (5,6,10,11). While these temporal experiments are correlative, not causal, they led to our search for an alternative ATP generator that is fuel dependent but active at the time KATP channels close.
In the last step of glycolysis, PK converts ADP and PEP to ATP and pyruvate. There are three PK isoforms in β-cells: PKm1 is constitutively active, while PKm2 and PKL require allosteric activation by fructose 1,6-bisphosphate (FBP), which is produced earlier in glycolysis by the phosphofructokinase reaction. Phosphofructokinase has strong control over Ca2+ oscillations (12), and PKm2 activity peaks at the time KATP channels close (13). PK activation was found to increase the frequency of Ca2+ oscillations, and subsequently, recordings from excised patches of human and mouse β-cell plasma membranes demonstrated that the associated PK raises the plasma membrane ATP/ADP ratio (ATP/ADPpm) to close KATP channels (12). β-Cell deletion of PKm1 and PKm2 revealed isoform redundancy for KATP channel regulation, provided that FBP (or a pharmacologic PK activator) was present to activate PKm2 (14). This work also overturns the canonical model by demonstrating PK is essential for KATP channel closure while ATP/ADPc is not. Mitochondrial PEP, generated by mitochondrial PCK2 in the PEP cycle, was found to accelerate glucose-dependent KATP channel closure but is not essential (14,15). Finally, in human and mouse β-cells, PK is part of a plasma membrane–associated glycolytic metabolon that regulates KATP channel opening and closure via ATP/ADPpm (16) (Fig. 2). These data support earlier biochemical studies, including in MIN6 cells, showing that KATP channel subunits interact with the enzymes of upper and lower glycolysis (17–19).
Alternative explanations for compartmentalized KATP channel closure have been proposed. 1) KATP channels were closed in excised patch experiments by the failure to adjust the pH of the solutions after adding PEP (3); knockout of PK excludes this possibility (14). Importantly, the bath solution pH was adjusted to 7.2 after adding metabolites in all our studies (11,14,16) and in prior studies that demonstrated glycolytic regulation of KATP in cardiac myocytes (17,18,20–22). Taken together, PEP and ADP closed plasma membrane KATP channels in eight of these nine electrophysiology studies (five of five cardiomyocyte studies and three of four β-cell studies, with the exception of Corradi et al. [3]). 2) PEP could directly close KATP channels without its metabolism; knockout of PK also excludes this possibility (14). 3) KATP channels were closed by OxPhos via stray mitochondria associated with excised patches; this possibility was excluded because pyruvate and ADP do not close KATP channels in excised patches, and oligomycin does not prevent KATP channel closure by PK substrates (16). 4) Excised patches are recessed into the pipette, which could exaggerate local KATP channel regulation by membrane-associated glycolytic enzymes. Most likely, this last issue will remain unresolved until experiments conclusively test whether the location of PK is important for KATP channel regulation. A key question is whether ATP must diffuse into the microvilli, which blanket the β-cell plasma membrane (22) and are even smaller than the patch pipette. If KATP channels are located at the microvillus tip, rather than the base, the kinetic limitation on nucleotide diffusion would favor local KATP regulation by glycolysis.
To clarify our position, our published studies have not argued that glycolysis preferentially regulates KATP channels because ATP is unable to diffuse from mitochondria to the plasma membrane, as in the canonical model, just that OxPhos cannot make ATP at the time of KATP closure (due to the underlying bioenergetics, described below) (11). That said, the conclusion of Rutter and Sweet (2) that ATP/ADPc and ATP/ADPpm are similar if not identical should be reevaluated. First, the available experiments used untargeted Perceval sensors (23) that cannot distinguish plasma membrane versus cytosol, or luciferase (24) that detects ATP rather than ATP/ADP. Second, they make the unproven and unlikely assumption that PK does not contribute to the dynamics of ATP/ADPc. Given that mitochondrial PK activity locally inhibits OxPhos (11), it would similarly shape ATP/ADPc. Indeed, it would be surprising if ATP/ADPc and ATP/ADPpm were identical, since the plasma membrane is full of Ca2+-ATPases and Na+/K+-ATPases (25). Until proven otherwise, it remains plausible that mitochondrially derived ATP is inefficient for signaling to KATP channels relative to mitochondrially derived PEP, which has few interactions (limited to enolase, PCK2, and PK) and demonstrably signals to the plasma membrane (14,15).
Another source of confusion is that mice lacking either PKm1 or PKm2 in β-cells have mild phenotypes. For example, β-cell deletion of PKm1 reduces islet PK activity by ∼90% but does not perturb glucose-stimulated Ca2+ oscillations or glucose-stimulated insulin secretion (14). How, then, can Foster et al. (14) claim that PK is essential for KATP closure? To test this hypothesis, the redundancy of PKm1 and PKm2 at elevated glucose concentrations was avoided by using mitochondria (rather than glycolysis) as the sole source of PEP for PK. At low glucose (2 mmol/L), leucine stimulates mitochondria to make PEP from glutamine without making FBP. Mitochondrial PEP can only be utilized by constitutively active PKm1, not by PKm2 or PKL, which both require FBP. Consequently, amino acids do not stimulate KATP closure, Ca2+ influx, and insulin secretion following β-cell deletion of PCK2 (which prevents mitochondrial PEP production) or PKm1 (the only FBP-independent isoform) (14,15). These experiments show that PK is essential for KATP closure and membrane depolarization. The knockouts were not sick—pharmacologic activation of PKm2 (a stand-in for glycolytic FBP) immediately restored the ability of amino acids to close KATP channels in β-cells lacking PKm1 (so does glucose, which explains the mild phenotype of the mice). Importantly, the amino acid–stimulated rise in ATP/ADPc in PKm1-deficient β-cells is comparable to that of controls, yet KATP channels did not close, indicating that ATP/ADPc is insufficient to close KATP channels.
With this new knowledge of KATP channel regulation, it makes sense that a recent report (3) examined KATP channel activity and found no impact of a PKm2 inhibitor (because PKm1 is also present). They also found no impact of a PKm2/L activator on KATP channel activity at low glucose (because there is no PEP) or at high glucose (because FBP is already present). Importantly, PKm2/L activation accelerates Ca2+ oscillations and increases insulin secretion from mouse and human islets (11,14,16). These assays were not performed by Corradi et al. (3). Our experiments (11,14,16) also explain why mitochondrial fuels (e.g., glutamine/leucine, α-ketoisocaproate, and in insulinomas, pyruvate) increase O2 consumption, raise ATP/ADPc, close KATP, and initiate Ca2+ influx and insulin release even when glucose is low. Since mitochondrial PCK2 synthesizes PEP, none of these mitochondrial substrates bypass the PK reaction (14). The newly described PCK2-PK-KATP pathway provides a hypothesis to explain how activating mutations in glutamate dehydrogenase causes hyperinsulinemic hypoglycemia in response to protein-rich meals (even without carbohydrates) (26).
Nature provides redundancy for critical processes, as shown for KATP regulation by PKm1 and PKm2 (14). A fair question, raised by the field, is what happens when both isoforms are deleted in β-cells? Rest assured, the phenotype is impressive, but it was important to first demonstrate that PK is necessary and sufficient for KATP closure in healthy β-cells (i.e., in the absence of hyperglycemia) (11,14,16). That <10% of PKm1 protein can facilitate normal glucose-stimulated insulin secretion (14) supports the argument that metabolic compartmentation is highly effective.
Bioenergetic Regulation of β-Cell Stimulus-Secretion Coupling
To the extent that the ATP/ADP ratio influences the KATP channel, the biochemical and bioenergetic constraints that determine this ratio are crucial. Our working model is that at rest and following depolarization, OxPhos maintains a nonstimulatory ATP/ADPc ratio. The model considers that β-cell mitochondria are inefficient ATP generators lacking sufficient thermodynamic strength to raise ATP/ADPc to closing thresholds. This oxidative inefficiency together with amplification by the PEP cycle further augments the PK reaction, a reaction that has favorable bioenergetics to raise ATP/ADPpm to KATP-triggering levels.
β-Cell Mitochondria Are Low-Yield ATP Generators
In a highly coupled tissue, like muscle, for every glycolytic substrate (e.g., one-half glucose, glyceraldehyde, or dihydroxyacetone) that is fully oxidized, a mere one ATP comes from the PK reaction compared with the ∼14 ATPs arising from OxPhos, calculated using modern phosphate-to-oxygen ratios (27) (Fig. 3). If this situation were imposed on β-cells, it would be easy to discount the glycolytic contribution in favor of the high OxPhos yield. However, the established metabolic features of β-cells make them different than muscle and diminishes the stoichiometric dominance of OxPhos (28). For instance, up to half of cellular PEP arises from the PEP cycle, which also consumes one ATP molecule and one GTP molecule in the mitochondrial matrix (29–32). Thus, PEP cycling doubles the yield of PK-derived ATP while undermining the OxPhos yield (Fig. 3). Furthermore, β-cell mitochondria are remarkably uncoupled, such that more than half of the proton motive force (PMF) does not contribute to ATP synthesis (30,33). Taken together, the yield of ATP for the relative PK-to-OxPhos ratio is about 1:3.4. In addition, there is substantial evidence for NADPH biosynthesis in β-cells via isoforms of malic enzyme, isocitrate dehydrogenase, and/or nicotinamide nucleotide transhydrogenase (34). Each NADP+ molecule reduced to NADPH by these enzymes prevents oxidation of 1 NADH molecule at a cost of at least 3.7 ATP molecules. This lowers the yield to about 1:2.7 if a single NADPH molecule is generated. The PK-to-OxPhos ratio further increases by the transient diversion of pyruvate from oxidation into enlarging the cytosolic citrate pool (1:1.3) and glutamate pool (1:2) (11,35). Likewise, mitochondrial Ca2+ homeostasis has a large energetic cost, where cycling of one Ca2+ in and out of the mitochondria could have been used to make 0.8 ATP molecules. (Ca2+ activates matrix dehydrogenases by lowering the substrate Km and not by increasing Vmax. In a cycle, this will just change the relative concentrations of metabolites.) Thus, depending on the amount of leak (approaching 100% at peak PMF), PEP cycling, cataplerosis, and Ca2+ cycling, the level of PK-derived ATP approaches or potentially exceeds that of OxPhos (Fig. 3). The mitochondrion-to-PK ratio is likely an underestimate of the PK contribution during the electrically silent phase prior to depolarization. As such, considering stoichiometric yield alone, β-cell OxPhos is an overestimated source of ATP.
OxPhos Is Hobbled in β-Cells
Stoichiometric yield arguments have little relevance to the bioenergetics that actually determine the ATP/ADP ratio (27). Regardless of the initial source of ATP, cytosolic ATP is buffered by rapid equilibration with creatine kinase and holds the millimolar ATP concentration nearly constant in the cytosol, while ADP is ∼2–3 orders of magnitude more dilute. As such, ATP/ADPc is dominantly influenced by lowering the small amount of ADP, not by adding more ATP (27,34). As ATP/ADPc gets lower, the energetic requirement becomes steeper (28). Thus, increasing the ATP/ADPc to triggering levels is a function of the bioenergetics (ΔGp) rather than stoichiometric yield arguments (Fig. 1). In this regard, ΔGp is set by the PMF and defines the maximal ATP/ADP that OxPhos can generate. The maximum PMF that β-cell mitochondria reach is an embarrassing −140 to −145 mV compared with muscle’s −180 to −200 mV (Fig. 4A). In addition, when mitochondria are hyperpolarized, they apparently jettison reduced cytochrome c to the cytosol, further impairing the respiratory chain capacity (36). In this regard, β-cell mitochondria are severely disadvantaged by sizable leak currents that lower the maximal PMF and throttle back the ADP-lowering potential (ΔGp).
O2 Consumption Does Not Equal OxPhos
The increased islet respiration following a glucose increase is often interpreted by proponents of the canonical model as substrate “pushing” OxPhos to raise the ATP/ADP ratio, but this logic is backwards. Following a sudden increase in glucose, or plasma membrane depolarization, 1) ADP is increased (from increased glucose phosphorylation or other ATP-hydrolyzing work), 2) ATP synthase then makes ATP from ADP, which 3) lowers the PMF, which 4) increases O2 consumption (Fig. 4B–D). That is to say, O2 consumption increases as ATP/ADP levels decrease, not vice versa. In contrast, as the ATP/ADPc ratio approaches the PMF allowed maximum (−145 mV), ATP synthase slows to a stop while mitochondrial leak proportionately increases ohmic proton conductance, increasing O2 consumption (37). At these voltages, O2 consumption correlates positively with leak and inversely with OxPhos and therefore does not support the canonical model.
Modifying Mitochondrial Metabolism Breaks the Canonical Model
The canonical model predicts substrate supply–driven OxPhos increases ATP/ADPc, so perturbations that increase substrate supply, mitochondrial membrane potential (ΔΨm; a close surrogate of PMF), and matrix ATP synthesis, or decrease ATP-consuming pathways, all should enhance insulin secretion. Substantial work directly challenges this view. For instance, acutely closing mitochondrial leaks, reducing cytosolic ATP consumption, decreasing anaplerosis and cataplerosis, increasing NAD+ reduction of pyruvate flux, and increasing matrix substrate–level ATP synthesis all decreased insulin secretion (15,30–33,37,38). Whereas increased mitochondrial leak increased basal insulin secretion, decreased substrate-level matrix ATP synthesis increased insulin secretion (30,31,33,38). Notably, the changes with insulin secretion associated with PEP cycling did not correlate with O2 consumption, ΔΨm, or NAD(P)H reduction (30–32). Each of the above treatments is explainable by changes in PK flux but inversely correlate with OxPhos. As a potential final blow to the canonical model, recent data dismiss the role of the electron transport chain for glucose-stimulated insulin secretion (4).
β-Cell OxPhos Is Not Supply Sided
The canonical model is built on the premise that β-cell mitochondria are regulated by substrate supply, as opposed to the mitochondria of all other tissues, where demand controls OxPhos (27). Demand is generated from work-related hydrolysis of ATP to ADP together with Ca2+, which signals OxPhos to activate (39). Ca2+ also may have a more important role in activating glycolysis to increase the pyruvate supply for anaplerotic, cataplerotic, or oxidative metabolism rather than activating OxPhos per se (40). For demand mitochondria, the ATP/ADPc ratio equilibrium when ATP consumption and production are matched is set near the maximum PMF voltage. An increase in cellular work drops the ATP/ADPc ratio, which increases OxPhos and lowers PMF to activate respiration and explains why we breathe harder with exercise than after drinking a soda. In contrast, a supply regimen presupposes that a glucose-induced increase in pyruvate pushes mitochondria to raise the PMF to increase OxPhos. This model assumes that while at baseline, β-cell mitochondria are deprived of substrate, which lowers the PMF. A prediction of such a model is that increased demand from further lowering the ATP/ADPc ratio (e.g., via mitochondrial uncoupling) could not increase respiration, since, being in a fuel-deficient state, there is no excess fuel to oxidize. Contrary to this canonical prediction, uncoupling β-cell mitochondria even at low glucose dramatically increases respiration, even more than substrate can stimulate it, and formally rules out supply-sided regulation (11). Using canonical reasoning, pharmacologic activation of distal glycolysis (e.g., PK activation) could not increase OxPhos to increase insulin secretion, since glucokinase controls the pyruvate supply. The opposite is observed, namely, PK activation dramatically enhances insulin secretion (11).
There is a third possibility, where mitochondria remain demand regulated while PK is substrate supply regulated, which may explain how glucose is actually sensed. In this model, OxPhos defends the resting ATP/ADPc using the same demand principles common to all mitochondria. However, because the mitochondria are hobbled by substantial proton leak, the equilibrium ATP/ADPc is held lower, below the threshold to close KATP channels. As glucose is increased, PK uses the augmented PEP supply (coming from glucokinase and PCK2) to raise the ATP/ADPpm ratio to KATP channel–triggering levels. This is a testable model that does not require a complete reimagining of mitochondrial function; however, definitive quantitative characterization of these bioenergetic constraints is needed.
Phosphoenolpyruvate Has More Pep
Consequently, two mitochondrial ATP equivalents power the synthesis of one PEP equivalent, leaving it with more favorable bioenergetics than OxPhos to lower ADPc. As such, the PEP cycle acts like a step-up transformer that supercharges the cytosolic ADP–lowering potential of PK (Fig. 2B). This also explains how PEP induces mitochondrial hyperpolarization by ADP privation and generates triggering ATP/ADPpm ratios locally at KATP channels.
PK Controls OxPhos and the ATP/ADPpm Ratio
Glycolysis has been portrayed as a passive conduit with the sole purpose of delivering pyruvate to the mitochondria while generating insignificant ATP. In β-cells, these assumptions are inaccurate and should be replaced with a model where PEP hydrolysis by PK lowers ADP, which closes KATP channels and inhibits OxPhos prior to membrane depolarization. β-Cell mitochondria have evolved to be extremely leaky, a feature that keeps the ATP/ADPc ratio relatively low but increases flexibility to augment PK flux when glucose levels are increased. PK harnesses the free energy of PEP hydrolysis to deprive mitochondria and KATP channels of ADP. ADP privation in hyperpolarized mitochondria feeds forward to induce PEP cycling to further amplify ADP lowering. Importantly, in this model OxPhos is still extremely important and mitochondria still retain responsiveness to ADP following plasma membrane depolarization to support rapid ATP synthesis, but OxPhos cannot affect KATP channel closure.
Conclusions and Future Directions
In the end, the 1984 model lacks mechanistic support and should be replaced whether or not the details of our spatial and bioenergetic arguments prove correct. Given the 1960s observations that glycolysis has strong control strength over insulin secretion, our hypothesized model of glycolytic KATP channel regulation is plausible and offers better paradigms to understand and treat β-cell dysfunction. Nonetheless, important experiments remain to be performed:
- 1.
While we have provided evidence that PK is essential for KATP channel closure, we have not yet tested whether the location of glycolysis on the plasma membrane (or other organelles) is necessary for efficient β-cell stimulus-secretion coupling. Notably, the location of PK may not matter if PK outperforms OxPhos bioenergetically; either argument overturns the canonical model.
- 2.
There is a KATP channel–independent or amplifying effect of PK, revealed by exocytosis experiments that included PK activators in the patch pipette or experiments conducted in intact islets treated with diazoxide and KCl (11). Consistent with this finding, PEP plus ADP, but not pyruvate and ADP, stimulate biphasic insulin release in permeabilized islets (even in the presence of rotenone) (43). While the mechanistic basis of this amplifying effect remains to be clarified, these findings do not argue against KATP channel regulation by PK, as claimed by Rutter and Sweet (2).
- 3.
Our working model is that β-cell mitochondria evolved to be incapable of raising the ATP/ADPc ratio sufficiently to close KATP channels via OxPhos alone, and play a supportive role that expands after depolarization. PK, fueled by glycolysis and the PEP cycle, is responsive to changes in nutrients and has favorable thermodynamics to close KATP channels as well as to deprive mitochondria of ADP. While gaps in this model remain, quantitative bioenergetic studies are underway.
This article is featured in a podcast available at diabetesjournals.org/diabetes/pages/diabetesbio.
Article Information
Funding. This work received support from the National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases, from a joint award to M.J.M. and R.G.K. (R01DK127637). M.J.M. also acknowledges support from the National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases (R01DK113103), and the U.S. Department of Veterans Affairs Biomedical Laboratory Research and Development Service (I01BX005113).
Duality of Interest. No potential conflicts of interest relevant to this article were reported.