Chronic Intermittent Hypoxia Impairs Insulin Sensitivity but Improves Whole-Body Glucose Tolerance by Activating Skeletal Muscle AMPK

Obstructive sleep apnea (OSA) syndrome is characterized by recurrent pharyngeal collapses during sleep leading to transient apnea with respiratory efforts, sleep fragmentation, and intermittent hypoxia (IH). The prevalence of OSA is increasing worldwide, mostly reflecting the parallel rise in obesity, a major risk factor for the disease (1). Half of obese people are indeed suffering from OSA, and a strong link has been established between apnea occurrence and metabolic disorders such as insulin resistance and type 2 diabetes (2). However, among the consequences of OSA, IH has been independently associated with metabolic dysfunction (3,4), and acute exposure to IH was shown to decrease insulin sensitivity in healthy volunteers (5).

Several groups have used an experimental model of OSA to investigate the effect of acute or chronic IH (cIH) on whole-body insulin sensitivity and glucose homeostasis in lean and obese rodents. Although cIH generally impaired insulin sensitivity in rats (6,7) and mice (8–15), the exact underlying molecular mechanisms remain mostly unknown to date, notably regarding the respective contribution of the main peripheral metabolic organs to systemic insulin resistance. Among the potential mechanisms, cIH-induced hepatic and/or adipose tissue inflammation has been suggested to be involved in alterations of tissue-specific insulin sensitivity (14–18). However, some conflicting results were reported on the effects of IH on whole-body glucose homeostasis. For instance, whole-body glucose tolerance was impaired in lean and obese mice after acute or chronic exposure to IH (12). In contrast, it was reported to be improved in mice subjected to 5 days of cIH (8) and in rats acutely exposed to IH (6). Overall, these discrepancies might reflect differences in experimental protocols and in rodent models, but they also contribute to the confusion regarding the effect of cIH on glucose tolerance.

Our aim in the current study was to investigate the effect of 2 weeks of cIH exposure on whole-body insulin sensitivity and glucose tolerance in lean C57BL/6J mice and to dissect the underlying molecular mechanisms in the liver, white adipose tissue (WAT), and skeletal muscle, the main organs involved in the regulation of whole-body metabolic homeostasis.

All experiments were performed in accordance with the Institute for Laboratory Animal Research “Guide for the Care and Use of Laboratory Animals” and received approval (B3851610006) from the Université Grenoble Alpes Ethical Committee.

In this study, 8-week-old male C57BL/6J (Janvier Labs, Saint-Berthevin, France), 8- to 10-week-old male C57BL/6J AMPKα2^−/−, and muscle-specific AMPKα1α2^−/− (mdKO) mice (a gift from Dr. Viollet, Institut Cochin, Paris, France), and 8- to 10-week-old male Swiss × Sv129 HIF-1α^+/− (oxygen-dependent α-subunit of hypoxia-inducible factor 1 [HIF-1]) mice (a gift from Dr. Carmeliet, KUL, Leuven, Belgium) were housed under standard conditions in conventional cages with ad libitum access to standard chow diet (RM1; Special Diets Services) and water. Before the start of the experiments, mice were randomized by body weight into two groups that were next exposed to normoxia or cIH for 2 weeks. cIH was achieved in experimental cages as previously described (19). Briefly, the animals were exposed in their housing cages to 60-s IH cycles (30 s at 5% fraction of inspired oxygen [Fio₂] and 30 s at 21% Fio₂) for 8 h between 6 a.m. and 2 p.m. Fio₂ was monitored with a gas analyzer (ML206; ADInstruments) throughout the experiment. In parallel, control animals were exposed to similar air-air cycles (21% Fio₂) to reproduce equivalent levels of noise and air turbulences related to gas circulation. Ambient temperature was maintained at 20–22°C.

Blood samples were obtained in mice fasted for 6 h (food withdrawn at 6 a.m.) via tail vein. Blood glucose levels were determined using a OneTouch Ultra glucometer (LifeScan). Plasma total cholesterol (TC), triglycerides (TG), leptin, and insulin levels were measured using commercially available enzymatic kits (Roche), ELISA leptin (Crystal Chem), and insulin kits (Millipore), respectively. The homeostasis model assessment of insulin resistance (HOMA-IR) adjusted to rodents (20) was calculated as: ([glucose (mg/dL)/18] × [insulin (ng/mL)/0.0347])/108.24.

Whole-body glucose tolerance and insulin sensitivity were assessed immediately after the last exposure to normoxia or IH (i.e., from ∼12 p.m.) by intraperitoneal glucose tolerance test (IPGTT) and insulin tolerance test (IPITT) tests, respectively. After initial blood collection, intraperitoneal (i.p.) injections of glucose (2 g/kg total body weight) or insulin (0.5 units/kg total body weight) (NovoRapid; Novo Nordisk) were performed in conscious mice fasted for 6 h (food withdrawn at 6 a.m., just before the start of the last exposure). Blood glucose levels were next measured by tail bleeding at different times. A pyruvate tolerance test was also performed after an i.p. injection of sodium pyruvate (2 g/kg of total body weight) in conscious mice fasted for 16 h. The area under the curve (AUC) for the glucose excursion was calculated using trapezoidal integration.

Liver lipids were extracted as previously described (21). Liver TG and TC concentrations were measured using the commercial kits described above and expressed as milligrams per milligram of total protein content using the Bradford protein assay kit (Sigma-Aldrich).

Snap-frozen liver, skeletal muscle (gastrocnemius) and epididymal WAT (eWAT) samples (∼50 mg) were lysed in ice-cold buffer containing 50 mmol/L HEPES (pH 7.6), 50 mmol/L NaF, 50 mmol/L KCl, 5 mmol/L NaPPi, 1 mmol/L EDTA, 1 mmol/L EGTA, 1 mmol/L dithiothreitol, 5 mmol/L β-glycerophosphate, 1 mmol/L sodium vanadate, 1% NP40, and protease inhibitors cocktail (Complete; Roche). Western blots were performed as previously described (21). The primary antibodies used are listed in Supplementary Table 1. Bands were visualized by enhanced chemiluminescence and quantified using ImageJ software.

RNA was extracted from snap-frozen liver, skeletal muscle, or eWAT samples (∼25 mg) using the TRI Reagent RNA Isolation protocol (Sigma-Aldrich). Total RNA (0.5 µg) was reverse transcribed, and quantitative real-time PCR was performed with a SYBR Green Core Kit on a MyIQ thermal cycler (Bio-Rad). mRNA expression was normalized to Gusb mRNA content and expressed as fold change compared with control mice using the ΔΔCt method. The primer sequences are listed in Supplementary Table 2.

Skeletal muscle samples were lysed in ice-cold HClO₄-EDTA (5%-25 mmol/L) and centrifuged at 13,000g for 2 min. The supernatant fractions were immediately neutralized, and determination of adenine nucleotides was performed by high-performance liquid chromatography, as previously described (22).

All data are expressed as mean ± SEM. Statistical analysis was performed using GraphPad Prism 6 software with two-tailed unpaired Student tests or two-way ANOVA with multiple comparisons, followed by the post hoc Fisher least significant difference tests. Differences between groups were considered statistically significant at P < 0.05.

Male 8-week-old C57BL/6J mice were exposed to cIH or normoxia 8 h daily (from 6 a.m. to 2 p.m.) for 2 weeks. At the end of the experimental period, blood hematocrit was significantly higher in cIH mice than in normoxic mice (+16%) (Fig. 1A), validating the efficacy of cIH exposure. Body weight was significantly decreased by cIH (−11% after 2 weeks) (Fig. 1B), associated with a reduction in liver weight, whereas eWAT and skeletal muscle mass was not affected (Fig. 1C). Total food intake was significantly reduced in mice exposed to cIH (−21%, data not shown), an effect exclusively caused by complete inhibition of food intake during cIH exposure (−95% from 6 a.m. to 2 p.m.) (Fig. 1D). Food intake was similar between groups during the normoxic period (i.e., from 2 p.m. to 6 a.m.), without any compensatory overfeeding in cIH-exposed mice.

cIH did not affect fasting plasma TC, TG, or glucose levels (Fig. 1E–G) but significantly increased the plasma insulin level (+51%) (Fig. 1H) and HOMA-IR (+42%) (Fig. 1J), suggesting systemic insulin resistance. The fasting plasma leptin level was also increased by cIH (+276%) (Fig. 1I). To further investigate the effect of cIH on insulin sensitivity, an IPITT was performed after 14 days of cIH exposure (Fig. 1K). In line with HOMA-IR, cIH mice exhibited impaired whole-body insulin sensitivity compared with normoxic mice (Fig. 1L and M). In separate experiments, tissue-specific insulin sensitivity was assessed in the liver, eWAT, and skeletal muscle from cIH and normoxic mice 10 min after an acute i.p. injection of PBS or insulin. Although insulin receptor-β (IRβ) protein content was significantly lower in eWAT only, the insulin-induced phosphorylation of protein kinase B (PKB), on both Thr308 and Ser473 regulatory residues and its downstream targets proline-rich Akt substrate of 40 kDa (PRAS40) and glycogen synthase kinase 3 (GSK3; data not shown), were reduced in the liver, eWAT, and skeletal muscle from cIH mice (Fig. 2A–F), indicating impairment of insulin sensitivity in all of these metabolic organs.

Mechanistically, dysregulation at the level of insulin receptor substrate 1 (IRS1) signaling, which is regulated through a complex mechanism involving phosphorylation of multiple tyrosine and serine/threonine residues (23), has been proposed to be involved in cIH-induced insulin resistance (24,25). In our conditions, we found that the insulin-induced phosphorylation of some of these regulatory residues was significantly reduced by cIH but in a tissue-specific manner (Supplementary Fig. 1A–F), suggesting different underlying mechanism depending on the metabolic organ. An increase in HIF-1 signaling was also suggested to be involved in insulin resistance in various metabolic tissues (26,27). Although HIF-1α was not detectable in the skeletal muscle from either group, we found a significant increase in HIF-1α protein content in both the liver and eWAT of mice exposed to cIH (Fig. 3A and B). However, we showed that cIH exposure was still able to severely impair systemic insulin sensitivity in HIF-1α^+/− mice (Fig. 3C–F), suggesting that the role of this transcription factor may not be determinant in cIH-induced insulin resistance. Of note, an upregulation of some genes involved in hepatic glucose oxidation (Supplementary Fig. 2A) was observed in cIH mice, whereas neither the genes involved in glycogen metabolism nor the hepatic glycogen content were affected (data not shown). The expression of key lipogenic genes, but not of those involved in fatty acid oxidation, was also significantly upregulated in the liver of cIH mice (Supplementary Fig. 2A). Hepatic TG and TC contents were not significantly affected (Supplementary Fig. 2B and C), although a trend for an increase was observed for TC. In eWAT, neither the gene expression of proinflammatory M1 macrophage markers (Supplementary Fig. 2D) nor the activity of the nuclear factor-κB and c-Jun N-terminal kinase (JNK) pathways (Supplementary Fig. 3A–F) were changed in response to cIH.

An IPGTT was performed in the mice to investigate whether the detrimental effect of cIH observed on insulin sensitivity had an impact on glucose homeostasis (Fig. 4A). Unexpectedly, we found that cIH significantly improved whole-body glucose tolerance (Fig. 4B and C), whereas plasma insulin levels at the peak of the glucose excursion curve were similar between groups (Fig. 4D). Of note, as a result of the higher baseline insulin levels, the glucose-induced insulin response assessed by the t₁₅-on-t₀ insulin ratio was reduced in cIH mice (Fig. 4E), in line with previous studies (10,13,28). Importantly, improvements in whole-body glucose tolerance were still observed after prolonged (6 weeks) exposure to cIH (Supplementary Fig. 4A–C) and when the GTT was performed after the same absolute amount of glucose was injected in both groups (i.e., not adjusted to body weight), therefore excluding any potential bias caused by lower body weight in cIH mice (Supplementary Fig. 5A–C). Similarly, glucose tolerance was still increased in cIH mice compared with normoxic mice subjected to pair-feeding by removing food from 6 a.m. to 2 p.m. (Supplementary Fig. 5D–F), indicating that this effect on glucose homeostasis was independent of the inhibition of food intake during cIH exposure. Of note, cIH did not affect the glucose excursion curve during the i.p. pyruvate tolerance test (Supplementary Fig. 6A–C) or the expression of key gluconeogenic genes in the liver (Supplementary Fig. 6D), indicating that the effect on glucose tolerance is unlikely related to changes in hepatic gluconeogenesis.

To elucidate the molecular mechanisms underlying the enhanced whole-body glucose tolerance in cIH mice, the main signaling pathways involved in the regulation of glucose uptake were studied (Fig. 4F–K). For this, liver, eWAT, and skeletal muscle samples were collected at the peak of the GTT curve (i.e., 15 min after glucose administration), and the insulin and AMPK pathways were analyzed. The phosphorylation state of PKB was not affected by cIH during the GTT in any of the tissues (Fig. 4F–K). Interestingly, although not affected in the eWAT and reduced in the liver, the phosphorylation of AMPKα on Thr172 was significantly increased in the skeletal muscle of cIH mice. Phosphorylation of AMPK downstream targets acetyl-CoA carboxylase on Ser79, Akt substrate of 160 kDa (AS160, also known as TBC1D4)/TBC1 domain family member 1 (TBC1D1) on Ser/Thr residues using anti-PAS antibody, and TBC1D1 on Ser237 was also increased in skeletal muscle (Fig. 4J and K). Altogether, our results suggest that the improvement of whole-body glucose tolerance by cIH could be caused by an increase in skeletal muscle glucose uptake secondary to tissue-specific activation of the AMPK-TBC1D1 axis. However, neither the adenine nucleotide concentrations (Supplementary Fig. 7A and B) nor the protein expression of the various AMPK subunits (Supplementary Fig. 7C and D) were affected, indicating that the cIH-induced activation of skeletal muscle AMPK occurs independently of changes in cellular energy homeostasis or in the heterotrimeric composition of the kinase. The phosphorylation states of Ser485/491-AMPKα, Ser108-AMPKβ2, and Ser182-AMPKβ2, which are other residues involved in the regulation of AMPK activity, were also not affected by cIH (Supplementary Fig. 7C and D).

To assess the contribution of skeletal muscle AMPK activation to the enhanced whole-body glucose tolerance, AMPKα1α2 mdKO mice were exposed to cIH or normoxia. As previously observed in wild-type mice, 2 weeks of cIH exposure increased blood hematocrit (Fig. 5A), decreased food intake and body weight (Fig. 5B and C), impaired HOMA-IR (Fig. 5D), and increased circulating leptin levels (Fig. 5E) in the mdKO mice when compared to normoxia. Remarkably, however, cIH did not improve whole-body glucose tolerance (Fig. 5F and G) and skeletal muscle phosphorylation of Ser237-TBC1D1 (Fig. 5J and K) in the mdKO mice, whereas the cIH-induced alteration in glucose-induced insulin response (Fig. 5H and I) and upregulation of hepatic glycolytic genes (Supplementary Fig. 8) were still present. Of note, similar results were obtained in whole-body AMPKα2-deficient mice, the main isoform expressed in skeletal muscle (Supplementary Fig. 9A–L).

OSA and its associated episodes of systemic IH is a risk factor for the development of insulin resistance. Studies in rodents and humans have consistently reported that exposure to cIH affects whole-body metabolic homeostasis, but the underlying mechanisms and the metabolic organs involved remained unclear. In the current study, we show that cIH, although impairing insulin sensitivity in most of the metabolic tissues, unexpectedly improved whole-body glucose tolerance by promoting skeletal muscle–specific activation of the AMPK-AS160/TBC1D1 axis involved in the control of glucose uptake.

We found that cIH induced a significant decrease in body weight, in line with previous studies (11,14,15,28–30). In our experimental conditions, this effect is associated with a reduction in mean daily calorie intake exclusively as a result of a complete inhibition of food intake during the 8 h of the hypoxic challenge. Feeding behavior is tightly controlled by the central nervous system (CNS) through complex regulatory networks involving afferent neural inputs from the periphery and circulating metabolites and hormones acting directly on specific neurons from key hypothalamic regions. The increase in plasma leptin levels in response to cIH has been suggested as one of the mechanisms underlying its anorexigenic effect (31). However, this might reflect systemic and/or hypothalamic leptin resistance, and direct cIH-induced alterations of some specific oxygen-sensing CNS pathways controlling feeding behavior could also be involved (32). Further studies are required to clarify this point.

Remarkably, even if the mice exhibited reduced body weight, they also presented classic features of systemic insulin resistance, with increased fasting plasma insulin levels and impaired glucose response during the IPITT, as previously reported (8–10,12–14,33,34). Interestingly, we demonstrate that cIH alters the canonical insulin-signaling pathway in all of the main metabolic tissues (i.e., WAT, skeletal muscle, and liver) by lowering the insulin-induced phosphorylation of PKB and of some of its downstream targets.

Inflammation is among the pathways that have been suggested to underlie alterations of insulin signaling in WAT (2,17). Previous studies have indeed reported that long-term exposure to cIH promotes infiltration of monocytes and/or polarization of resident macrophages toward a proinflammatory phenotype, leading to increased production of proinflammatory cytokines and systemic insulin resistance (15,35,36). In the present study, we did not find any changes in the gene expression of the main macrophage markers and proinflammatory cytokines in WAT from cIH mice in the current study, a discrepancy that might eventually be explained by the shorter duration of cIH exposure and/or by the mouse strain used.

cIH has been shown to substantially reduce WAT oxygen tension (29) and increase HIF-1α expression (37), an effect that resembles the one observed in WAT from high fat diet–fed obese mice, where a tissue-specific decrease in IRβ and reduction in canonical insulin signaling are also evidenced (38). Interestingly, we found that cIH led to a significant increase in HIF-1α content associated with lower IRβ protein expression in WAT, as also recently reported (39). This is also in line with in vitro data showing that hypoxia can directly inhibit insulin signaling and insulin-stimulated glucose uptake in adipocytes by a HIF-1–dependent mechanism (24,38). However, although the deletion of HIF-1α was only partial in this model, our results showing that systemic insulin sensitivity was still severely impaired by cIH in heterozygous HIF-1α^+/− mice suggest that its role may not be determinant in cIH-induced insulin resistance. Altogether, downregulation of IRβ might be one of the explanations, but the exact mechanisms by which cIH contributes to WAT insulin resistance remain unclear and deserve additional investigation.

In the liver, cIH was suggested to promote oxidative stress, inflammation, and/or hepatic steatosis, inducing tissue-specific insulin resistance and increased hepatic glucose production, this latter at least partly secondary to enhanced gluconeogenesis (2). Our study did not investigate oxidative stress per se, and we did not find evidence for activation of proinflammatory pathways or enhanced hepatic gluconeogenesis by cIH. However, an increase in lipogenic genes expression was observed in the livers from cIH mice, although not associated with significant differences in TG content, indicating that enhanced hepatic steatosis was unlikely contributing to the cIH-induced insulin resistance in our conditions. These results contrast with previous observations showing that prolonged exposure to cIH is associated with hepatic steatosis and detrimental effect on whole-body metabolic homeostasis (33,40). Altogether, this suggests that cIH duration might be critical and that prolonged exposure is required to induce significant hepatic steatosis.

cIH was also found to impair insulin sensitivity in skeletal muscle, a tissue-specific effect that was, in our hands, neither caused by inflammation nor to elevated endoplasmic reticulum stress (data not shown). Some other relevant pathways involved in skeletal muscle insulin resistance, such as those triggered by defects in mitochondrial oxidative phosphorylation or increased intramyocellular lipid content, were not investigated and would definitively require further dedicated investigation.

The other important outcome of this study was the counterintuitive evidence that, despite tissue-specific and systemic insulin resistance, cIH could substantially improve whole-body glucose tolerance by a mechanism that appeared to be independent of changes in body weight and glucose-induced insulin response. Our data are consistent with previous reports showing that 5 days or 5 weeks of cIH both led to a lower glucose excursion and reduced AUC during IPGTT in lean C57BL/6J mice (8,11) but contrast with some others (13,28,41). These discrepancies might potentially be explained by the fact that control (normoxic) mice were often weight-matched to cIH-exposed mice by food restriction in these latter studies, a procedure that might significantly affect whole-body glucose metabolism. Other subtle differences in experimental settings, such as duration and/or nocturnal versus diurnal period of IH exposure, cannot be excluded.

At the molecular level, AMPK activity and phosphorylation of its direct downstream target, the Rab GTPase-activating protein TBC1D1, was specifically increased in skeletal muscle from cIH mice, whereas insulin signaling was not affected. AMPK, a heterotrimeric protein kinase that acts as an intracellular energy sensor, was shown to control insulin-independent glucose uptake by promoting AS160/TBC1D1-mediated translocation of GLUT4 to the plasma membrane (42). Importantly, activation of this muscle-specific AMPK pathway and improvement in whole-body glucose tolerance by cIH were abolished in both whole-body AMPKα2 and muscle-specific AMPKα1α2 KO mice, demonstrating the central role played by the kinase in this effect. Among the possible mechanisms underlying skeletal muscle AMPK activation by cIH, a decrease in energy status was first considered but no change was observed in tissue adenine nucleotide levels or in the AMP-to-ATP ratio.

AMPK activity can also be modulated by modifications in the composition of its subunits and/or by changes in other posttranslational regulatory mechanisms. However, no differences in protein expression of the α-catalytic and β-/γ-regulatory subunits and in the phosphorylation state of the inhibitory Ser485/491 residues on AMPKα were observed.

Finally, one of the mechanisms by which cIH could selectively increase skeletal muscle AMPK activity might be through leptin-mediated activation of the hypothalamic-sympathetic nervous system axis. Indeed, intrahypothalamic injection of leptin, which is increased in response to IH ([43]; this study), was reported to increase skeletal muscle AMPKα2 activity, an effect lost after pharmacologic α-adrenergic blockage or surgical sympathetic denervation (44). Furthermore, neuron-specific deletion of protein tyrosine phosphatase 1B, which is involved in both insulin and leptin signaling, also led to increased skeletal muscle AMPKα2 activity (45). Taken together, the elucidation of the exact mechanism by which cIH promotes AMPK activation in skeletal muscle and the role played by the CNS in this effect definitely remain an intriguing area of future research.

The control of peripheral nutrient metabolism by the CNS plays an important role in the regulation of whole-body metabolic homeostasis, and selective activation of specific neurons in the medial hypothalamus, notably by leptin, has been shown to promote glucose uptake in skeletal muscle (46). In our conditions, we cannot exclude that part of the effect of cIH on glucose tolerance might be mediated by CNS-driven activation of skeletal muscle AMPK and the subsequent tissue-specific increase in glucose uptake. Further studies are required to extensively investigate the brain-mediated control of peripheral nutrient metabolism in response to cIH. In contrast, we can rule out that the increase in glycolytic genes observed in the livers from cIH mice is involved in the improvement of whole-body glucose tolerance by promoting hepatic glucose flux through the oxidative pathway. Indeed, the contribution of the liver to glucose clearance during GTT is rather marginal ([47]; C. Moro, personal communication), and the upregulation of glycolytic genes was still present in the livers from AMPK KO mice while the improvement of glucose tolerance was completely abolished.

In summary, our findings show that cIH impaired systemic insulin sensitivity in a HIF-1–independent manner by altering insulin signaling in various metabolic organs. However, the exact molecular mechanisms underlying insulin resistance specific to these tissues remain to be elucidated. Further studies are required, including studying the possible contribution of the CNS to impaired peripheral insulin sensitivity. Alternatively, cIH promoted whole-body glucose tolerance by a mechanism implying specific activation of skeletal muscle AMPK and downstream recruitment of the Rab GTPase-activating protein TBC1D1 involved in the control of glucose uptake. Altogether, one can speculate that this counterintuitive effect of cIH on glucose homeostasis might be a compensatory adaptation to systemic insulin resistance. The cIH rodent model has been validated by numerous previous publications (48) but has some limitations in immediate translation to humans. cIH is the hallmark of OSA, but the cardiometabolic consequences of sleep apnea are also related to intrathoracic pressure swings and sleep fragmentation and are not recapitulated in our model. Whether our current findings have any relevance for the pathophysiology of OSA in humans remains therefore to be determined.

Funding. This work was supported by Fonds de Dotation “Recherche en Santé Respiratoire,” Fondation du Souffle, APMC: Fond de dotation pour les maladies chroniques, Fondation de France, and the French National Research Agency in the framework of the “Investissements d'avenir” program (ANR-15-IDEX-02) (to A.T., E.B., P.L., J.-L.P., and D.G.-R.). S.H. and L.B. were supported by grants from Fonds National de la Recherche Scientifique et Médicale, Belgium, and the Action de Recherche Concertée de la Communauté Wallonie-Bruxelles, Belgium. S.H. is a Research Associate and L.B. is a Senior Research Associate of Fonds National de la Recherche Scientifique et Médicale, Belgium.

Duality of Interest. S.H. and L.B. were supported by unrestricted grants from AstraZeneca. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.T. performed experiments, analyzed data, and drafted the manuscript. E.B. performed experiments, contributed to discussion, and drafted the manuscript. S.M. and G.C.v.d.Z. performed experiments and analyzed data. S.H., B.V., and L.B. provided critical materials, contributed to discussion, and critically reviewed the manuscript. P.L., J.-L.P., and D.G.-R. supervised the project, contributed to discussion, and critically reviewed the manuscript. B.G. conceptualized and supervised the project, performed experiments, analyzed data, and wrote and edited the manuscript. B.G. is the guarantor of this work and, as such, has full access to all the data generated in the framework of the study and takes responsibility for their integrity and the accuracy of their analysis.