Glucose-Stimulated Insulin Secretion Fundamentally Requires H₂O₂ Signaling by NADPH Oxidase 4

Insulin, which is released from pancreatic β-cells, controls the blood glucose level in healthy individuals, and insulin release is impaired in those with diabetes (1–4). An understanding of the pathophysiology of the insulin release mechanism is indispensable for clinical innovation. The consensus mechanism of glucose sensing in pancreatic β-cells involves the elevation of ATP synthesis by mitochondria upon increased glucose metabolism enabled by the human GLUT1– or rodent GLUT2–mediated glucose uptake and fast glycolysis, which generates pyruvate, leading to ATP production by oxidative phosphorylation (OXPHOS) (1).

The increased ATP-to-ADP ratio in the subplasmalemmal cytosol in β-cells should cause the K_ATP channel to close (4–8), depolarizing the plasma membrane and activating voltage-gated L-type Ca²⁺ (Ca_L) channels. The resulting Ca²⁺ entry stimulates Ca²⁺-dependent exocytosis of the insulin-containing secretory granules (8).

Glucose-stimulated insulin secretion (GSIS) is facilitated by an increase in cytosolic NADPH (4,9–12), as various redox shuttles affecting metabolism generate cytosolic NADPH upon glucose intake, at the expense of NADH in the mitochondrial matrix (4,9–12). Also, the first enzyme of the pentose phosphate pathway, glucose-6-phosphate dehydrogenase (G6PDH), might contribute to the NADPH surplus upon GSIS (13–15). The resulting NADPH may be supplied to NADPH oxidase (NOX) isoforms 1 (NOX1) and 2 (NOX2) (16,17), but only if they are properly assembled. Nevertheless, the only constitutively expressed and assembled isoform is NOX4, and it is the only one that produces H₂O₂ directly (18,19). Inhibition of an unidentified NOX isoform was reported to attenuate GSIS when an antisense p47^PHOX oligonucleotide (20), the nonspecific NOX (plus complex I) inhibitor diphenyleneiodonium (DPI) (21,22), or an inhibitor of two isoforms (23) was used. Hence, if any NOX isoform participates in GSIS, it must be definitively recognized.

Only reactive oxygen species (ROS) of mitochondrial origin, such as ROS resulting from the addition of mono-oleoyl-glycerol (23), have been suggested to modulate insulin secretion (24). An effect of antioxidants has been reported, with glutathione being decreased by diethylmaleate in INS-1(823/13) cells and acting as an unspecified link between GSIS and external H₂O₂ (25). However, it has never been considered that elevated ATP is insufficient for GSIS, nor that any redox signaling may be essential for GSIS. Thus, the elevation of the ATP-to-ADP ratio was thought to close K_ATP channels without any additional requirement for parallel redox signaling. A rather weak antioxidant defense and low redox buffer capacity, but high thioredoxin and peroxiredoxin content (26), provide an ideal, delicate ROS homeostasis in β-cells, although this homeostasis might be disturbed by a relatively weak insult and may spread within the cytosol (2,27).

This work revisited a mechanism of GSIS, and this article describes a novel mechanism of NOX4-mediated redox stimulation of insulin secretion, which, together with increased ATP, is essential for GSIS. Moreover, the absence of this redox signaling affects insulin sensitivity in peripheral tissues.

Experiments were approved by the committee of the Institute of Molecular Genetics (Prague, Czech Republic) and complied with the 2010/63/EU directive (U.S. National Institutes of Health publication no. 85-23, revised in 1996) and the guidelines laid out in Animal Research: Reporting of In Vivo Experiments.

NOX4-null (knockout) (NOX4KO) mice were generated through targeted deletion of the translation initiation site and exons 1 and 2 of the gene (Research Resource Identifiers [RRID]: Mouse Genome Informatics [MGI]:4838554) (28,29). Control (“backcrossed”) mice were obtained by intercrossing NOX4-null progeny with C57BL/6 mice for 10 generations. NOX2-null (knockout) (NOX2KO) mice were from The Jackson Laboratory (Bar Harbor, ME) (B6.129S-Cybbtm1Din/J, RRID:IMSR_JAX:002365).

The β-cell–specific NOX4KO (NOX4βKO) mouse was created by crossing NOX4^Flox/Flox mice with B6.Cg-Tg(Ins2-Cre)25Mgn/J mice (RRID:IMSR_JAX:003573) (30,31). Tail genotyping selected congenic NOX4Flox^+/+Cre^+/+ (B6.Cg-Nox4<tm1Ams>Tg[Ins2-cre]25Mgn) mice (NOX4βKO). NOX4 deletion, specifically in mouse β-cells, was verified (Supplementary Fig. 3A and B). At least 64% of β-cells exhibited the β-knockout alleles (∼80% of the β-cell fraction of pancreatic islet [PI] cells). Peripheral insulin resistance was evaluated as described by Alán et al. (32).

An intraperitoneal (i.p.) glucose tolerance test and a parallel insulin assay were performed after the i.p. injection of glucose (1 mg/g body weight; ∼111 μmol glucose per mouse) into mice deprived of food overnight. Blood was taken from a blood vessel in the ocular plexus. Tests were also performed with 2-oxoisocaproate (OIC) (1 mg/g body weight; ∼153 μmol OIC per mouse). Blood glucose was determined with a glucometer (Roche, Basel, Switzerland); insulin was determined with an ELISA kit (Mercodia, Uppsala, Sweden). Samples were taken at two or three time points for both insulin and glucose estimates from each mouse. Mean data from 10–27 mice (sex is specified in the various figure legends) enabled the construction of time dependencies.

Two male and two female mice with each genotype were anesthetized. The pancreases were digested with collagenase, and the PIs were isolated subsequently on a Ficoll gradient (33). The yield was 100–200 islets per mouse.

Dynamic insulin release from PIs was determined by using perifusion. Approximately 100 islets were placed in a column with a flow adaptor (1 × 7 cm Econo-Column; Bio-Rad, Hercules, CA) attached, and the islets were immobilized with Bio-Gel P4 (Bio-Rad). PIs were washed for 60 min in a continuous flow of glucose-free Krebs-Ringer HEPES (KRH) buffer (135 mmol/L NaCl, 3.6 mmol/L KCl, 10 mmol/L HEPES, 0.5 mmol/L MgCl₂, 1.5 mmol/L CaCl₂, 0.5 NaH₂PO₄, 0.1% BSA [pH 7.4]). Insulin secretion was stimulated with KRH buffer containing stimulatory compounds. The perfusate was collected at rates of 0.5 ± 0.1 mL/min. Insulin was detected with an Insulin Mouse ELISA High Sensitive kit (BioVendor, Brno, Czech Republic). Islets were lysed, and their DNA content was quantified by using a PicoGreen Assay (Thermo Fisher Scientific, Eugene, OR).

PIs were seeded on laminin-covered 96-well plates (BioLamina, Stockholm, Sweden). NOX4 was overexpressed in PIs isolated from control (NOX4^Flox/Flox) and NOX4βKO mice by using 72-h lentiviral transduction with LentiORF particles (OriGene, Rockville, MD) that contain pLenti-C-mGFP as a control, or pLenti-mouse NOX4-C-mGFP for NOX4 overexpression. Insulin release was assayed after 6 min.

Rat insulinoma INS-1E cells (C0018009; AddexBio, San Diego, CA) (31) were routinely cultivated with 11 mmol/L glucose in RPMI 1640 medium supplemented with 10 mmol/L HEPES, 1 mmol/L pyruvate, 5% (v/v) FCS, 50 μmol/L mercaptoethanol, 50 IU/mL penicillin, and 50 μg/mL streptomycin. The 1- and 15-h incubations were also performed with 3 mmol/L glucose (34).

Cells were transfected twice in 24 h with Oligofectamine (Thermo Fisher Scientific) with two predesigned Ambion siRNAs against rat NOX4 (exons 14 and 10; Thermo Fisher Scientific) in a serum- and antibiotic-free medium. Their 5′ to 3′ sequences were CCGUUUGCAUCGAUACUAATT and CAUUAUCUCAGUAAUCAAUTT. Also, different pairs of predesigned Ambion siRNAs against rat SUR1 (Abcc8), G6PDH, and branched-chain α-ketoacid dehydrogenase (BCKDH) complex E1α (Thermo Fisher Scientific) were used. The pZeoSV2(+) vector encoding human catalase (donated by C. Glorieux, Université Catholique de Louvain, Belgium) was also used for transfection (2.5 mL lipofectamine/mg DNA; Thermo Fisher Scientific).

One day before 48-h transfections, cells were seeded (0.2 × 10⁶ cells/well) in 12-well plates coated with poly-l-lysine. Insulin was assayed in KRH (with or without 2 mmol/L glutamine) by using a Rat Insulin ELISA Kit (U-E type; Shibayagi Co., Shibukawa, Japan) after a 5-min preincubation phase before glucose (25 mmol/L) was added.

H₂O₂ monitoring with TCS-SP2 or TCS-SP8 confocal microscopes (Leica) used an overexpressed H₂O₂-selective fluorescent protein, either HyPer-C (35) or heat shock protein–fluorescence resonance energy transfer (HSP-FRET) (36). A sample chamber (37°C) was supplied with 5% CO₂. INS-1E cells were transfected with an X-tremeGene DNA reagent (Roche), and the coexpression of NOX4 siRNAs was facilitated by using Lipofectamine 2000 in a serum-free and antibiotic-free medium for 24 h. An ATP Bioluminescence Assay kit HS II (Roche) was used to quantify ATP.

A FluxOR Potassium Ion Channel Assay (Thermo Fisher Scientific) was used to assay Tl⁺ influx rates. INS-1E cells were preloaded with a “stimulus buffer” containing Tl⁺. The emission of FluxOR was monitored at 525 nm on an RF5301 spectrofluorometer (Shimadzu). Fluorescent Ca²⁺ in the cell cytosol was monitored with Fura-2.

Patch pipettes made of borosilicate glass (Harvard Apparatus, Kent, U.K.), with a resistance of 15–20 mol/LΩ, were filled with an isotonic solution (150 mmol/L KCl, 10 mmol/L HEPES, 1.2 mmol/L CaCl₂ [pH 7.2]). Cells were incubated in a solution without glucose (134 mmol/L NaCl, 6 mmol/L KCl, 10 mmol/L HEPES, 1.2 mmol/L MgCl₂, 1.2 mmol/L CaCl₂ [pH 7.2]) for 2 h. Single-channel currents with no glucose (130 mmol/L KCl, 20 mmol/L HEPES, 10 mmol/L EGTA, 4 mmol/L MgCl₂, 2.4 mmol/L CaCl₂ [pH 7.15]) were recorded in the cell-attached mode with an EPC-7 amplifier (HEKA Electronics, Lambrecht, Germany); these currents were low-pass filtered at a corner frequency of 0.5 kHz and sampled at a frequency of 2.5 kHz by using Clampex software version 9.2. Analyses were done in Clampfit software version 9.2 (Axon Instruments, Foster City, CA).

Biological (N) and experimental (n) replicates are listed. ANOVA (t test) used the Tukey and Holm-Sidak tests (for P values, see the Supplementary Material) on the prevalidated data (normality test). Calculations were performed in SigmaStat software version 3.1 (Systat Software, San Jose, CA).

Data are available from the corresponding author upon request.

To evaluate the effects of NOX4-produced H₂O₂, two NOX4 siRNAs were validated (Supplementary Fig. 1A); these siRNAs were always tested in combination in INS-1E cells. When stimulated with 25 mmol/L glucose, a linear, steady release of insulin (2.9 ± 0.8 ng ⋅ min⁻¹ ⋅ 10⁻⁶ cells [N = 10]) was observed in nontransgenic (ntg) cells (scrambled siRNA), whereas rates were 4% ± 3% without glucose (positive and negative glucose differences represent the rate of GSIS). In NOX4-silenced cells, GSIS rates were 16% ± 12% (N = 10) (Fig. 1A and B); they were 30% ± 20% after preincubation with 3 mmol/L glucose for 15 h (N = 5) (Fig. 1B). NOX activity (evaluated with the nitroblue tetrazolium assay; Fig. 1C) and the amount of NOX4 in protein (Fig. 1D) were predominantly reduced in silenced cells. GSIS rates (N = 3) were inhibited to 10% ± 8% with stigmatellin (which blocks mitochondrial respiration and hence ATP synthesis), to 8% ± 5% with DPI (which nonspecifically inhibits NOX), and to 38% ± 20% with 6-aminonicotin-amide (which blocks the pentose phosphate pathway).

Transfections with SUR1 siRNAs, disassembling K_ATP (37), decreased GSIS rates to 20% ± 1% (N = 3) (Fig. 1A and Supplementary Fig. 1B). Silencing G6PDH (Supplementary Fig. 2C), thereby diminishing NADPH, reduced GSIS by ∼50% (N = 3) (Fig. 1E). Catalase overexpression (Fig. 1F and Supplementary Fig. 1F) and the NOX4-selective inhibitor GKT-137831 (Fig. 1G) almost completely blocked GSIS (N = 3). Besides the ntg cells’ and NOX4-silenced cells’ equal responses to KCl, the ability of NOX4-silenced cells to release insulin was demonstrated by using glibenclamide to close K_ATP channels, which restored insulin secretion (N = 3) (Fig. 1G).

Next we demonstrated that NOX4 predominantly participates in accelerated cytosolic ROS/H₂O₂ release upon GSIS. Rates and levels of ROS release in the cytosol were significantly less elevated in NOX4-silenced INS-1E cells (Fig. 2A–H). Independent of NOX4 absence, mitochondrial ROS declined (Fig. 2I), whereas cytosolic ATP was elevated (Fig. 1H).

To confirm that NOX4 is essential for GSIS in vivo, NOX4KO mice (Fig. 3A) and NOX4βKO mice (Fig. 3B) were studied. (For genotype verification and RIP-Cre effects (30,31), see Supplementary Figs. 3 and 4.) Mice (12 weeks old) were starved overnight, blood was collected from the ocular plexus, and then an i.p. glucose tolerance test was performed using that blood, while also assaying for secreted insulin. Samples from two or three time points for each mouse (Supplementary Fig. 5) allowed for constructions of average time dependencies. In both NOX4KO mice (N = 34 time courses) and NOX4βKO mice (N = 37 time courses), the initial fast phase of insulin release into the blood circulation was abolished; this was not the case in age-matched controls: the “backcrossed” mice (N = 19 time courses) (Fig. 3A) and NOX4^Flox/Flox mice (N = 30 time courses) (Fig. 3B), respectively. Concomitant glycemia was reduced more slowly in knockout mice, manifesting impaired glucose tolerance (IGT) (Fig. 3C, D, K, and L and Supplementary Figs. 6 and 7). Substantial insulin was released upon stimulation with glibenclamide without glucose in both knockout mouse strains (four time courses each, accompanied by glycemia, which was probably elevated because of stress; Fig. 3E–H) and upon glucose stimulation in NOX2KO mice (N = 3 time courses; N = 5 time courses for backcrossed controls) (Fig. 3I and J).

Perifusion of PIs (the typically predominant β-cell population; Supplementary Fig. 8) isolated from knockout mice reflected profoundly diminished GSIS in its first fast phase (N = 5) (Fig. 4A–D), sensitive to GKT-137831 (N = 3) (Fig. 4D). However, significant responses to glibenclamide (N = 3) were found when it was added alone (Fig. 4E and F) and with glucose (Fig. 4C and D). Glucose-induced release of ROS in cytosol was attenuated in NOX4-deficient PIs (N = 3) (Fig. 4I–K). Insulin release was not suppressed when induced by fatty acids with 5 mmol/L glucose (N = 2) (Fig. 4G and H and Supplementary Fig. 9), because the majority of fatty acid–stimulated insulin secretion (FASIS) is dependent on GPR40 and independent from K_ATP (2,4,27) (Fig. 6G).

To investigate the possibility of restoring GSIS, mouse NOX4 was overexpressed in PIs by using lentiviral transduction (N = 3) (Fig. 4L). This overexpression rescued GSIS at the 6th minute in NOX4βKO PIs, whereas in the control cells (NOX4^Flox/Flox PIs), it increased GSIS by 1.4-fold (Fig. 4L). GSIS could also be rescued in PIs from knockout mice by perifusing with H₂O₂ for 10 min (Fig. 4M and N). In conclusion, our data show NOX4-mediated H₂O₂ release to be an essential comediator of GSIS, in parallel with ATP (Fig. 4O), and thus both must be elevated for GSIS to occur (Fig. 8A).

In addition to IGT, both NOX4KO strains developed insulin resistance, which was evaluated as insulin-dependent ¹⁴C-glucose uptake into glycogen in the diaphragm (Fig. 3K and Supplementary Fig. 10A and B) and into lipids in epididymal adipose tissue (N = 7) (Fig. 3L and Supplementary Fig. 10C). The initial stress signal originating from β-cells lacking an ability to check redox identity (38) can be predicted regarding the source of such insulin resistance caused by the ablation of a single gene (Fig. 8B and C).

Patch clamp recordings in the cell-attached mode elucidated single-channel currents through the K_ATP channel (Fig. 5A–D). INS-1E cells were depleted of glucose in modified glucose-free Ringer solution for 2 h to increase the probability that the K_ATP channel would be open in membrane patches attached to ntg cells (Fig. 5A). Whereas in ntg cells glucose addition caused significant K_ATP channel closure (that is, reducing the probability that the channel is entirely open [NP_o]), the K_ATP channel was only partially closed in NOX4-silenced cells (Fig. 5C and D). The fully open state (L3) of 73 pS contrasted with the partially open K_ATP channel states (L2 and L1). In ntg cells, there was a profound shift from the larger (L3, L2) to smaller (L1) states, whereas in the NOX4-silenced cells, the shift to smaller states was much less pronounced (Fig. 5E–G).

The holding potential was 0 mV at similar K⁺ concentrations on both sides of the membrane. Hence, the K_ATP current was driven by a resting potential of −60 mV, resulting in inward currents (downward deflections). It was possible to inhibit all three K_ATP open states (L1–L3) with glibenclamide. The single-channel conductance of the fully open state (L3) did not change by adding glucose or silencing NOX4 (Fig. 5G). The fully open state (L3) in the ntg cells started to close immediately after the addition of glucose. This closure was completed during the 3rd minute of continuous recording. The NOX4-silenced cells exhibited an increased probability of each K_ATP state being open (P_o), even without glucose. With glucose, the fully open state (L3) remained open similar to L2 and L1 in the 3rd minute.

Cromakalim or diazoxide, which open K_ATP channels, were ineffective in redox (H₂O₂)-stimulated insulin secretion (RSIS) (N = 11) (Fig. 6A and B); these vasodilators decreased the insulin release (GSIS) rate to 6% ± 5% and 13% ± 5%, respectively (Fig. 6C). Diazoxide neither activated nor inhibited residual GSIS at silenced NOX4.

Probing K_ATP activity, defined as the rate of glibenclamide-sensitive Tl⁺ influx (Tl⁺ being a surrogate for K⁺) into ntg-INS-1E cells, we found that the Tl⁺ influx was inhibited to ∼20%, indicating a characteristic closure of the K_ATP channel with 25 mmol/L glucose (N =4) (Fig. 6D and G). Predominant K_ATP participation was verified by the rate acceleration that manifested upon application of the K_ATP channel openers cromakalim and diazoxide, and inhibition of that acceleration by glibenclamide (to 10% ± 13%). However, NOX4-silenced or catalase-overexpressing INS-1E cells did not respond to glucose and still exhibited 70–95% of the maximum rates of glibenclamide-sensitive Tl⁺ influx.

A requirement for both NOX4-produced H₂O₂ and a higher ATP-to-ADP ratio (OXPHOS) was also suggested by the effects of oligomycin, which left ∼50% of the K_ATP channel open in control cells but up to 95% of it open in NOX4-silenced cells (Fig. 6D), with a concomitant blockage of GSIS (Fig. 6C). The SUR1-silenced INS-1E cells exhibited largely suppressed Tl⁺ influx rates (N = 3) (Fig. 6E), a result of the incorrect assembly of the remaining Kir6.2, which lacks the SUR1 partner (34). This result is consistent with inhibited GSIS (Fig. 1A and Fig. 6B).

Next, fluorescence monitoring with a plasma membrane potential indicator demonstrated a profound reduction of depolarization in the plasma membrane of ntg cells after the addition of glucose but not in NOX4-silenced or catalase-overexpressing INS-1E cells (N = 2) (Fig. 6F).

Nimodipine, inhibiting the Ca_L channel, decreased the rate of GSIS to 12% ± 11% (Fig. 6C). In ntg-INS-1E cells, oscillations of nimodipine-sensitive Ca²⁺ influx reflected the opening of the Ca_L channel after glucose was set to 25 mmol/L (N = 5) (Fig. 6H–M). In NOX4-silenced INS-1E cells, the rates of nimodipine-sensitive Ca²⁺ influx fell to 19% ± 3%, and the cells exhibited fewer Ca²⁺ oscillations (Fig. 6K and L). The first oscillation peaked (after a delay) after glucose was added (Fig. 6M). Typically, cromakalim prevented Ca²⁺ influx in controls and blocked the remaining Ca²⁺ oscillations in NOX4-silenced INS-1E cells (Fig. 6H–J). Hence, the K_ATP channel was indeed affected by the NOX4 deficiency, whereas the Ca_L channel was affected only indirectly. Rates of oligomycin-sensitive Ca²⁺ influx remained at 22% without the addition of glucose.

Insulin secretion rates after a single external H₂O₂ bolus were blocked in catalase-overexpressing cells (Fig. 6A). Independent of glucose, the effect of exogenous H₂O₂ was only partially blocked by NOX4 siRNA (Fig. 6A and B); it was fully blocked by nimodipine (Fig. 6C). Consequently, the H₂O₂ doses used do not directly stimulate the K_ATP-independent exocytosis of insulin granules. RSIS depolarized the plasma membrane (39) (Fig. 6F). At 90 μmol/L H₂O₂, rates of RSIS reached 120% of GSIS rates after 60 min; these RSIS rates were not significantly potentiated by glucose (Fig. 6B). Responses to 40–100 μmol/L H₂O₂ led to K_ATP channel closure, independent of the presence of NOX4 (Fig. 6D); this explains the “H₂O₂ rescue” observed in PIs (Fig. 4M and N). This K_ATP channel closure was overcome by diazoxide (Fig. 6D) and cromakalim.

Also, the Ca²⁺ influx into ntg-INS-1E cells at 90 μmol/L H₂O₂ reached 115% of the fluxes responding to glucose alone. There were concomitant H₂O₂-induced, cromakalim-blocked Ca²⁺ oscillations in both ntg and NOX4-silenced INS-1E cells, reflecting the opening of the Ca_L channel (Fig. 6H–J).

When stimulated by OIC, a well-described distinct secretagogue, the secretion of insulin was highly inhibited by the mitochondrial matrix–targeted antioxidant SkQ1 in ntg cells, but it was preserved in NOX4-silenced INS-1E cells (N = 3) (Fig. 6G and Fig. 7A–C and L) and in NOX4KO mice (N = 35 time courses); it was, however, partially abolished in NOX4βKO mice (N = 35 time courses) (Fig. 7E and F). The same pattern was found in PIs (typical results from N = 3 isolations are shown in Fig. 7I and J). OIC administration to mice increased glucose above fasting levels (Fig. 7G and H).

Note that GSIS itself was not sensitive to SkQ1 (Fig. 7K). With OIC, only a negligible fraction of H₂O₂ comes from NOX4; this H₂O₂ is supplied with NADPH by the basal activities of redox shuttles (otherwise activated by glucose). In contrast, a source of mitochondrial superoxide (transformed to H₂O₂) is essential for OIC-stimulated insulin secretion, as inferred from its nearly complete inhibition with SkQ1, which was paralleled by a decreasing release of superoxide in the matrix (N = 3) (Fig. 7C and D).

Mitochondrial H₂O₂ originates from a superoxide converted from manganese-dependent superoxide dismutase; this superoxide arises at the E_F site of electron-transferring flavoprotein:Q-oxidoreductase (ETF:QOR) (40) upon the oxidation of isovaleryl-CoA. Isovaleryl-CoA is produced from OIC by the BCKDH complex. Indeed, BCKDH E1α silencing led to ∼70% suppression of OIC-stimulated insulin release (N = 4) (Fig. 7B and C and Supplementary Fig. 1E), and it prevented the accompanying redox signal (Fig. 7D). Aminooxyacetate, an aminotransferase inhibitor upstream of BCKDH, did not affect the OIC-induced release of superoxide in the matrix (Fig. 7D), but it did slightly inhibit OIC-stimulated insulin secretion (Fig. 7B and C). Thus, OIC metabolism in INS-1E cells provides both ATP (from OXPHOS) and H₂O₂ (Fig. 7M) to stimulate insulin release.

In conclusion, the essential requirement of redox signaling for insulin secretion in vivo can be provided either by NOX4 (GSIS) or by mitochondrial sources (Fig. 8A and C).

We found that the elevation of ATP (ATP-to-ADP ratio) alone is insufficient for the fast GSIS phase in pancreatic β-cells. Parallel redox signaling must also occur. Coupling of glucose stimulus and secretion requires H₂O₂-mediated redox signaling that originates from NOX4. The essential part played by H₂O₂ in facilitating K_ATP-dependent insulin exocytosis (Fig. 8A) calls for an extensive revision of the GSIS mechanism (1–4,27,41). Our findings stimulate a rethinking of the origins of impaired insulin secretion in pancreatic β-cells and of strategies for treating type 2 diabetes. Essential dual signaling by the increased ATP-to-ADP ratio around the plasma membrane and elevated H₂O₂ is required to close the K_ATP channel and induce glucose- and OIC-stimulated insulin secretion. Such dual signaling allows fine-tuned regulation of insulin release. We demonstrated that the main target of redox signaling is K_ATP, but currently we cannot exclude parallel participation of redox-sensitive kinase signaling. Redox signaling might also target K_ATP via peroxiredoxins (26).

Consequently, not only a cell’s metabolic status (reflected by the increased ATP-to-ADP ratio) but also its redox status, determined by existing rates of H₂O₂ release into the cytosol, are linked to insulin exocytosis. Constitutively expressed and assembled NOX4 provides such an H₂O₂ branch of the mechanism of GSIS stimulation. G6PDH allows the glycolytic flux to increasingly branch out into the pentose phosphate shuttle (42).

The K_ATP channel has been suggested to be regulated by redox in vascular smooth muscle cells (43), but this has not yet been suggested in pancreatic β-cells. Plausible proposals related to a Ca²⁺-induced (44,45) or H₂O₂-induced exocytosis of insulin granules suggested the activation of depolarizing TRPM2 channels (46,47). Our results excluded the Ca²⁺-independent, H₂O₂-induced exocytosis of insulin granules. Hence, if a TRPM2-dependent mechanism exists, it contributes only marginally to GSIS. We demonstrated that increases in only ATP or H₂O₂ cannot close the K_ATP channel and initiate the membrane potential events that lead to Ca_L activation and subsequent insulin exocytosis.

Another mode of redox signaling, coming from mitochondria, was found for the ketoacid OIC, a leucine metabolite. Also, intramitochondrial redox signaling was reported previously for FASIS (4,39). In both cases, H₂O₂ is produced by manganese-dependent superoxide dismutase from the superoxide that is probably formed by ETF:QOR (40) upon OIC oxidation or β-oxidation (Fig. 8A). Branched-chain α-ketoacid dehydrogenase (BCKDH) can also be a source of superoxide (40) (Fig. 7M). Mitochondria also contribute to a portion of the increase of cytosolic NADPH through thoroughly described redox shuttles (4,9). Their operation is reflected by the insulin release that remains when a partial blockage of GSIS occurs upon G6PDH silencing.

However, the GPR40-stimulated pathway or the glycerol/fatty acid cycle (whereby monoacylglycerol activates the exocytosis-promoting protein Munc13-1), which might be independent of K_ATP, could predominate in FASIS (3,4,48,49). Upon FASIS, the H₂O₂ supplied by ETF:QOR/superoxide dismutase simultaneously activates the redox-activated phospholipase iPLA₂γ, cleaving both saturated and unsaturated fatty acids from mitochondrial phospholipids (27,39). GPR40 stimulation and the resulting K_ATP-independent insulin secretion are subsequently amplified by these mitochondrial free fatty acids migrating to the plasma membrane (27,39). A minor portion of FASIS is stimulated via the K_ATP channel as a result of increased OXPHOS by fatty acid β-oxidation, which also provides H₂O₂ (Fig. 8A). When fatty acids and glucose are metabolized in β-cells to various extents, the impact of glucose is diminished as GPR40 stimulation increases.

The IGT/insulin resistance phenotype of NOX4KO and NOX4βKO mice might have translational potential, as these two strains may represent models of early (pre)diabetes. Even though the human growth hormone minigene (transferred from RIP-Cre mice) (30,31) amplifies the insufficiency of GSIS in NOX4βKO mice (more profoundly on a systemic level) (Supplementary Fig. 4), the extensive inhibition is caused by a lack of NOX4 activity.

Surprisingly, ablation of just a single gene generates the onset of insulin resistance. We can speculate that pancreatic β-cells must emit an as yet unknown stress signal, either directly or via the immune system (Fig. 8B and C), thereby inducing peripheral insulin resistance (Fig. 3K and L). We hypothesize that such a putative stress signal is induced by the insufficient identity checking or autocrine self-maintenance of β-cells in NOX4βKO or NOX4KO mice. β-Cell identity checking can be mediated by the same redox signaling that acts upon K_ATP-dependent insulin exocytosis (38). However, Swisa et al. (38) did not know the source of such redox signaling. Here, we suggest that its source is H₂O₂ produced by NOX4 upon GSIS. This may hypothetically contribute to the “correct” β-cell identity-checking signal, which also primarily maintains sufficient insulin gene expression (38). For NOX4KO mice, diet over time stimulates insulin release via FASIS and OIC or other secretagogues; hence, these secretagogues might be sufficient. However, because the GSIS/NOX4-mediated redox signaling is impaired, such mice lack the “correct” NOX4-induced β-cell identity-checking signal. We speculate that the lack of this signal evokes an as yet unknown stress signal for the periphery (Fig. 8B and C).

Because antioxidant defense is diminished in pancreatic β-cells, the NOX4-produced H₂O₂ during GSIS in vivo, which can be repeated, could be gradually transformed into oxidative stress, reflecting high β-cell vulnerability. This could potentially contribute to diabetes (50). In light of our findings, cytosol-targeted antioxidant therapy, which should inevitably suppress GSIS, seems to be irrelevant in the early stages of diabetes. Tuning down the essential release of H₂O₂ during GSIS would amplify symptoms of prediabetes instead of preventing them. In contrast, we predict that mitochondria-targeted antioxidants would not harm physiological redox signaling (except that of oxoacids) and might avoid the premature oxidative stress in the matrix at the prediabetes stage. Also, the described repeating H₂O₂ burst upon GSIS might add to the oxidative stress resulting from the attack of macrophages recruited to the pancreas.

Acknowledgments. The authors thank Keith D. Garlid (Portland State University, Portland, OR) for providing fruitful discussion; Mark A. Magnuson (Vanderbilt University), C. Glorieux, and J.B. Verrax (Université Catholique de Louvain, Belgium) for donating plasmids; Dr. Radislav Sedláček and Dr. Inken Beck for providing the GMO mice facility (IMG and Czech Centre for Phenogenomics, Biocev, Prague, Czech Republic) for housing our mice; and all colleagues from the Institute of Physiology, Prague, such as Jitka Špačková for providing help designing the siRNA, Pavla Průchová for help with islet perifusion, Hana Engstová for analyzing images, and Jan Krůšek for monitoring calcium. The authors also thank Tomáš Špaček and Lenka Josková for providing excellent technical assistance with cell cultivation, Jana Vaicová with insulin assays, and Ludmila Šimečková with mice experiments.

Funding. The project was supported by the Grantová Agentura České Republiky (grant 16-06700S to L.P.-H. and grants 20-00408S and 17-01813S to P.J.) and by the Czech government’s support (RV0:679885623) to the Institute of Physiology, Prague, Czech Republic.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. L.P.-H., M.J., B.H., J.T., V.P., and D.S. performed the investigations. L.P.-H., M.J., B.H., Z.B., M.C., and D.S. developed the methodology. L.P.-J., B.H., K.S., and R.P.B. provided resources. L.P.-H., B.H., and P.J. validated the data and handled project administration. L.P.-H. and P.J. conceptualized and supervised the study and acquired funding. P.J. formally analyzed and curated the data, wrote the first draft of the manuscript, reviewed and edited the manuscript, and provided visualizations. P.J. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.