We generated mice that overexpress protein targeting to glycogen (PTG) in the liver (PTGOE), which results in an increase in liver glycogen. When fed a high-fat diet (HFD), these animals reduced their food intake. The resulting effect was a lower body weight, decreased fat mass, and reduced leptin levels. Furthermore, PTG overexpression reversed the glucose intolerance and hyperinsulinemia caused by the HFD and protected against HFD-induced hepatic steatosis. Of note, when fed an HFD, PTGOE mice did not show the decrease in hepatic ATP content observed in control animals and had lower expression of neuropeptide Y and higher expression of proopiomelanocortin in the hypothalamus. Additionally, after an overnight fast, PTGOE animals presented high liver glycogen content, lower liver triacylglycerol content, and lower serum concentrations of fatty acids and β-hydroxybutyrate than control mice, regardless of whether they were fed an HFD or a standard diet. In conclusion, liver glycogen accumulation caused a reduced food intake, protected against the deleterious effects of an HFD, and diminished the metabolic impact of fasting. Therefore, we propose that hepatic glycogen content be considered a potential target for the pharmacological manipulation of diabetes and obesity.
Liver glycogen acts as an energy store in times of nutritional sufficiency for use in times of need. The metabolism of this polysaccharide in the liver is controlled by the activities of two key enzymes: glycogen synthase (GS) and glycogen phosphorylase (GP) (1). GS is phosphorylated at multiple sites, which induces its inactivation, whereas GP is activated by phosphorylation at a single site. Both enzymes are also regulated allosterically (2,3).
Glycogen-targeting subunits bind to glycogen and protein phosphatase 1 (PP1) and facilitate the dephosphorylation of GS and GP, thus activating the former and inactivating the latter. Six genes encode glycogen-targeting subunits (4). Among these, protein targeting to glycogen (PTG) (PPP1R3C or PPP1R5), which is expressed in many tissues, has been shown to control glycogen stores in various animal models (5–7).
Adenoviral PTG overexpression in the liver of normal rats increases glycogen and improves glucose tolerance without perturbing lipid metabolism (8). In a diabetic-focused approach, Yang and Newgard (9) showed that adenoviral expression of PTG in the liver of STZ-diabetic rats increased glycogen content and reversed hyperglycemia and hyperphagia. Through a different approach, we reported that hepatic adenoviral expression of an active form of liver GS (LGS), which also increases glycogen content, in STZ-diabetic rats reduced food intake and hyperglycemia (10).
Russek (11) was the first to propose a hepatostatic theory of food intake, which was further redefined as a glycogenostatic model by Flatt (12). This model predicts that individuals consume food to a level that maintains glycogen levels in the body (12). In fact, many lines of experimental evidence establish a correlation between the size of liver glycogen stores and food intake (13,14); however, other results have argued against this hypothesis (15–17). The results in Yang and Newgard (9) and Ros et al. (10) support the notion that liver glycogen is a factor controlling food intake in hyperphagic type 1 diabetic animals. However, these studies had the constraint of using adenovirus transduction in animal models, which limit the experimental period to 1 week. In the present study, we examined whether a sustained increase in liver glycogen stores regulates food intake. For this purpose, we generated mice that overexpress PTG specifically in the liver (PTGOE), which results in a sustained increase in hepatic glycogen, and fed them either a standard diet or a high-fat diet (HFD). The HFD-fed animal is a suitable model for studying impaired glucose tolerance and type 2 diabetes (18), and prolonged ingestion of an HFD is associated with overconsumption and obesity (19).
We show that when fed an HFD, PTGOE animals had a reduced food intake and presented a lower body weight and fat mass than control animals. These results identify liver glycogen stores as a regulator of food intake in a model of hyperphagia and obesity and suggest that strategies to increase liver glycogen may provide a treatment option for diabetes and obesity.
Research Design and Methods
To generate mice that overexpress PTG, targeting-vector construction and a site-directed transgenic strategy were designed and performed by genOway (Lyon, France). Briefly, the PTG cDNA under the control of the ubiquitous CAG promoter (cytomegalovirus immediate early enhancer/chicken β-actin promoter fusion) was introduced into an innocuous locus by homologous recombination. A loxP-flanked transcription stop cassette was included between the CAG promoter and the PTG cDNA to allow the expression to depend on the action of a Cre recombinase. The resulting mouse line was bred with an albumin promoter Cre recombinase–expressing animal (The Jackson Laboratory), which drove the expression of PTG specifically to the liver. All mice studied were littermates. Mice were maintained on a 12:12-h light-dark cycle with free access to water and fed a standard diet (Harlan Laboratories) or an HFD (45% kcal fat; catalog #D12451; Research Diets) for 16 weeks, starting at 6 weeks of age.
Blood and Liver Biochemical Analysis
Liver glycogen content was determined by an amyloglucosidase-based assay, as described elsewhere (20). LGS activity was determined as previously described in the presence or absence of Glc-6-P (10). GS activity measured in the presence of saturating Glc-6-P [(+) Glc-6-P] corresponds to the total amount of enzyme, whereas measurement in its absence [(−) Glc-6-P] is an indication of the active (unphosphorylated) GS form. The (−) Glc-6-P/(+) Glc-6-P activity ratio (GS activity ratio) is an estimation of the activation state of the enzyme. The intracellular concentration of ATP was measured from perchloric acid extracts of livers using a previously described fluorimetric method (21). Triglycerides in liver were quantified using a triacylglyceride kit (Sigma-Aldrich) in 3 mol/L KOH and 65% ethanol extracts based on the method described by Salmon and Flatt (22) for liver saponification. Blood glucose levels were measured using a glucometer (Ascensia Breeze2; Bayer HealthCare). Serum concentrations of β-hydroxybutyrate (Sigma-Aldrich), triacylglycerides (Sigma-Aldrich), and nonesterified fatty acids (Wako) were measured spectrophotometrically. Plasma insulin and leptin were analyzed by ELISA (Crystal Chem).
Glucose and Insulin Tolerance Tests
For glucose tolerance tests (GTTs), overnight-fasted (16 h) mice were injected with glucose 2 g/kg i.p. Whole blood was drawn from the tail tip for glucose measurements. In vivo glucose-stimulated insulin secretion was determined in separate experiments. For insulin tolerance tests (ITTs), mice fasted for 6 h were injected with insulin 0.75 units/kg i.p., and glycemia was measured from tail blood taken at the indicated times after injection.
RNA Preparations and Quantitative RT-PCR
Tissue preparation, RNA extraction, RT-PCR, and quantitative real-time PCR analyses were performed as described (23). The following TaqMan primer/probe sets (Applied Biosystems, Madrid, Spain) were used for quantitative RT-PCR: PTG (Mm01204084_m1), Hprt (Mm00446968_m1), pyruvate kinase (Pklr) (Mm00443090_m1), fatty acid synthase (Fasn) (Mm00662319_m1), acetyl CoA carboxylase (Acc1α) (Mm01304257_m1), SREBP1 (Mm00550338_m1), peroxisome proliferator–activated receptor (PPAR) γ (Mm01184322_g1), monoacylglycerol O-acyltransferase 1 (MGAT1) (Mm00503358_m1), glucokinase (GK) (Mm00439129), Ucp1 (Mm01244861_m1), Ucp2 (Mm00495907_g1), Ucp3 (Mm00494077_m1), PPARα (Mm00440939_m1), Fgf21 (Mm00840165_g1), neuropeptide Y (NPY) (Mm03048253), proopiomelanocortin (POMC) (Mm 00435874_m1), carnitine palmitoyltransferase 1α (Cpt1α) (Mm01231183_m1), acyl-CoA oxidase (Acox1) (Mm01246834_m1), and Ppia (Mm02342429_g1). Ppia was used as a housekeeping gene in the liver. Hprt was used as a housekeeping gene in the hypothalamus and brown adipose tissue.
Indirect Calorimetry, Food Intake, and Body Temperature
Indirect calorimetry was performed using an eight-chamber Oxymax system (Columbus Instruments) to measure heat production, which was calculated from oxygen consumption and CO2 production. Mice were allowed to acclimate to the cages for 2 days before one or two cycles of 24-h measurements. Energy expenditure was calculated as (3.185 + 1.232 × RER) × VO2 (24), where the RER (respiratory exchange ratio) = VCO2/VO2. Glucose oxidation (in g/min/kg0.75 = [(4.545 × VCO2) − (3.205 × VO2)] / 1,000) and lipid oxidation (in g/min/kg0.75 = [1.672 × (VO2 − VCO2)] / 1,000) were calculated. Ambulatory and total locomotor activity was monitored by an infrared photocell beam interruption method. Body temperature was determined using an animal rectal probe thermometer (Cibertec). To monitor food intake, mice were housed individually and acclimatized for 1 week before study. Food intake was measured daily for 5 consecutive days. Epididymal adipose tissue was removed and prepared in paraffin after fixation in 10% phosphate-buffered formalin. Hematoxylin-eosin stains were then performed. To measure the size of the adipocytes, total adipocyte area was traced manually and analyzed. White adipocyte areas were measured in ≥100 cells per mouse in each group.
Oil Red O Staining
Liver tissues were embedded in optimal cutting temperature compound (Sakura Finetek) and frozen. Frozen tissues were cut into 5-μm-thick cryosections and stained with Oil Red O (Sigma-Aldrich).
Data are expressed as mean ± SEM. P values were calculated using unpaired Student t test, two-way ANOVA, or three-way ANOVA with post hoc Tukey tests as appropriate. P < 0.05 was considered significant.
All procedures were approved by Barcelona Science Park’s Animal Experimentation Committee and were carried out in accordance with the European Community Council Directive and the National Institutes of Health guidelines for the care and use of laboratory animals.
Generation and Characterization of Mice With Liver-Specific PTG Overexpression
To increase the accumulation of glycogen in the liver, we generated mice that overexpress PTG specifically in the liver (PTGOE) (see research design and methods). The mRNA level of PTG in the livers of these mice was 12-fold greater than that of control animals (Fig. 1A). As expected, LGS was activated because PTG, by targeting PP1 to the glycogen particle, maintains GS and GP in a dephosphorylated state (7). As expected, LGS was activated in these mice (Fig. 1B).
Mice were fed a standard diet or an HFD. When fed, the PTGOE mice showed an approximately twofold increase in liver glycogen content compared with control animals (Fig. 1C) regardless of whether they received a standard diet or an HFD (Fig. 1C).
After an overnight fast, PTGOE mice showed decreased hepatic glycogen content compared with the fed state. This observation indicates net mobilization of liver glycogen, although this glycogen was not depleted to the same extent as in control mice (Fig. 1C). Moreover, when fasted, the control mice receiving the HFD showed a higher hepatic glycogen content than the fasted control animals fed the standard diet, as previously reported (25) (Fig. 1C). Skeletal muscle glycogen content was similar between the groups and, thus, consistent with the idea that glycogen synthesis in nonhepatic tissues was not altered (Fig. 1D).
HFD-Fed PTGOE Mice Have a Lower Food Intake and Reduced Obesity
Mice fed an HFD present hyperphagia and obesity (26,27). In the current experiments, mice of both genotypes fed a standard diet had similar body weights (Fig. 2A). Those on an HFD became obese compared with their counterparts fed a standard diet (Fig. 2A). However, the increase in body weight in PTGOE animals fed an HFD was smaller than that detected in control mice on the same diet (Fig. 2A). After 16 weeks on an HFD, PTGOE animals showed a reduction in epididymal and subcutaneous fat weight, whereas when fed a standard diet, they had a similar fat weight as their control littermates (Fig. 2B and C). Control animals on an HFD had larger adipocytes than PTGOE mice on the same diet (Fig. 2D). Consistent with the reduced adiposity of PTGOE mice fed an HFD, serum leptin concentration was significantly lower in these animals compared with control mice on the same diet, whereas no differences in serum leptin were found between the two genotypes when fed a standard diet (Fig. 2E). The decreased adiposity and reduced HFD-induced obesity could have resulted from reduced food intake, increased energy expenditure, or fat malabsorption. PTGOE mice on an HFD showed a reduced daily food intake of ∼20% compared with control mice fed the same diet (Fig. 2F). The hypothalamus is the master regulator of energy intake, energy expenditure, and body weight (28). In the hypothalamus of PTGOE mice fed an HFD, the expression of the satiating neuropeptide POMC was higher, whereas that of the orexigenic peptide NPY was lower than in control mice fed the same diet (Fig. 2G).
We also determined the total daily energy expenditure in the two genotypes by measuring VO2 and VCO2. When fed a standard diet, control and PTGOE mice had a similar oxygen consumption (Fig. 3A), energy expenditure (standardized for body weight) (Fig. 3C), locomotor activity (Fig. 3E and G), and RER (Fig. 4A). No significant difference in oxygen consumption (Fig. 4B), energy expenditure (Fig. 4D), or locomotor activity (Fig. 3B and D) was observed between the HFD-fed groups. To examine the influence of liver glycogen content on thermogenesis, mRNA expression of Ucp1, a marker of thermogenesis, was analyzed in brown adipose tissue. No difference was found in the levels of Ucp1 mRNA between control and PTGOE mice fed an HFD (Fig. 4G). Other genes involved in energy expenditure, such as Ucp2 and Ucp3, were also analyzed, and no differences were found between the groups (Fig. 4G). No changes in core body temperature were detected in the two genotypes (Fig. 4H). Moreover, there were no differences in stool lipid content (data not shown).
During the feeding period (dark phase), the RER was slightly increased in HFD-fed PTGOE mice, thereby indicating that these animals used more carbohydrates as an energy source than the control group at night (Fig. 4B). These results were confirmed by calculating the amount of glucose oxidized, which was increased in PTGOE mice fed an HFD (Fig. 4D). However, PTGOE fed a standard diet oxidized the same amount of glucose as control mice (Fig. 4C). No change in lipid oxidation was found in PTGOE mice fed a standard diet (Fig. 4E) or an HFD (Fig. 4F). Because glucose oxidation was higher in PTGOE fed an HFD, we addressed hepatic ATP content. It is known that this parameter is reduced in the livers of HFD-induced diabetic mice (29). The ATP content in the livers of HFD-fed mice was significantly reduced, and PTG overexpression resulted in an ATP content similar to that of mice fed a standard diet (Fig. 2H).
Effects of Liver PTG Overexpression on Blood Glucose, Insulin Levels, Glucose Tolerance, and Insulin Sensitivity
The liver plays a key role in the clearance of blood glucose in the postprandial state (30). Fed PTGOE animals had lower blood glucose levels than control littermates, regardless of the diet received (Fig. 5A). Blood glucose levels and plasma insulin concentration decreased in control animals when they were deprived of food for 16 h (Fig. 5A and B). However, fasted PTGOE mice had similar glucose and insulin levels as fed PTGOE mice. This effect was observed regardless of the diet given (Fig. 5A and B). Moreover, HFD-fed PTGOE mice showed lower levels of insulin in the fed state compared with control mice on the same diet (Fig. 5B).
We next performed an intraperitoneal GTT on all four experimental groups. PTGOE mice fed a standard diet had better glucose tolerance, with a 40% decrease in the area under the curve (AUC) compared with control mice. When subjected to an HFD, these mice also showed reduced glucose intolerance and presented a 25% decrease in the AUC compared with control littermates (Fig. 5C). Remarkably, when fed an HFD, PTGOE mice presented a glucose tolerance curve analogous to that of control animals fed a standard diet (Fig. 5C).
We also measured insulin secretion during the intraperitoneal GTT. In response to an HFD, PTGOE mice presented a reduction in glucose-stimulated insulin release compared with control mice (Fig. 5D). Of note, the insulin release in the former animals was similar to that of PTGOE animals fed a standard diet (Fig. 5D).
Next, an ITT was performed after a 6-h fast. In these conditions, PTGOE mice fed an HFD already had significantly lower blood glucose concentration than the HFD-fed control mice (14 ± 1.7 vs. 10 ± 0.3 mmol/L), which made the analysis of the results of the ITT difficult to compare (Fig. 5E). However, when the ITT was expressed as the percentage of the initial values, the curve for HFD-fed animals was similar in both genotypes (Fig. 5F). Of note, PTGOE mice fed a standard diet presented higher blood glucose levels 60 min after the insulin injection. This finding suggests that these animals had a faster recovery from the hypoglycemia induced by insulin than their control littermates (Fig. 5E and F).
Liver PTG Overexpression Reduces HFD-Induced Hepatic Steatosis
We also analyzed the effect of PTG overexpression on the storage of liver triacylglycerides. When fed a standard diet, PTGOE mice presented a similar liver triacylglycerol content as their control littermates (Fig. 6A and B), suggesting that PTG is not associated with lipid metabolism under these circumstances. Moreover, the expression of genes related to de novo lipogenesis were not modified in PTGOE animals fed a standard diet (Fig. 6C). However, when fed an HFD, these animals showed a lower hepatic triglyceride content (Fig. 6A and B), which was associated with the downregulation of SREBP1, GK, PPARγ, and MGAT1 gene expression (Fig. 6C and D). As has been previously described (31), we confirmed that PPARγ and MGAT1 expression was very low in normal liver but was highly express in fatty liver (Fig. 6D). There were no statistically significant differences in the expression of lipogenic genes, such as Pklr, Fasn, and Acc1α, between the HFD-fed groups (Fig. 6C). Furthermore, the expression of genes related to lipid oxidation was evaluated. No differences between genotypes were found in the expression of PPARα, Cpt1α, or Acox1 in the liver (Fig. 6E).
Liver PTG Overexpression Diminishes the Metabolic Impact of Fasting
Many metabolic changes take place during fasting. As previously mentioned, PTGOE mice fed either a standard diet or an HFD did not show reduced levels of blood glucose or insulin after an overnight fast (Fig. 5A and B). Moreover, serum nonesterified fatty acids (Fig. 7A) and β-hydroxybutyrate (Fig. 7B) were lower in fasted PTGOE animals. It is well-known that after an overnight fast, hepatic triacylglycerol content increases in mice (32,33). However, PTGOE mice presented a lower fasting liver triacylglycerol content compared with control mice (Fig. 7C and D). During fasting, the expression of the hepatokine Fgf21 increased in the liver, but no differences were found in the expression of Fgf21 between genotypes (data not shown).
Using mice that overexpress PTG specifically in the liver, we examined the impact of liver glycogen on food intake. The overexpression of this protein caused an increase in hepatic glycogen stores in mice. When fed an HFD, these animals decreased their food intake and had a lower body weight and decreased fat mass. Changes in key regulators of food intake in the hypothalamus support the decrease in appetite observed in these animals. Expression of POMC, an anorexigenic signal, increased, whereas that of orexigenic NPY decreased. These data support the idea that liver glycogen stores regulate food intake, thus reinforcing the glycogenostatic theory (12). However, in the present study, this effect was limited to hyperphagic conditions, such as HFD. Friedman (34) proposed that changes in glycogen stores do not necessarily signal changes in food intake; rather, the partitioning of carbohydrates in and out of glycogen affects eating behavior by altering fuel fluxes, and, by analogy to fat fuels, shifts between oxidation and storage of carbohydrate fuels influence food intake (34). PTGOE mice used more carbohydrates as an energy source than control animals during the dark phase, which is when mice typically eat more. This increase in carbohydrate oxidation was observed in animals fed an HFD but not a standard diet. These data highlight the role of liver glycogen stores in modulating energy substrate utilization in response to an HFD. ATP is a final product of the oxidation of glucose and fatty acids. It has been proposed that a decrease in the amount of hepatic ATP is a metabolic stimulus that triggers feeding behavior (35,36). We found that consumption of an HFD decreases hepatic ATP levels. Of note, PTGOE mice maintained hepatic ATP content when fed an HFD. We propose that increased liver glycogen stores, through the maintenance of liver energy status, contribute to decreased appetite and adiposity. This effect is probably triggered by signals from the liver that are carried to the brain by vagal sensory neurons, as previously reported (37,38).
In addition, PTGOE mice fed a standard diet showed improved glucose tolerance. Consistent with this notion, overexpression of PTG induced by adenovirus in normal rats resulted in a modest improvement of glucose tolerance; however, these animals failed to show lower glycogen levels in response to fasting (8). In our model of liver PTGOE mice, the animals degraded glycogen in response to fasting, although they were not able to completely deplete the stores of this polysaccharide after a 16-h fasting period. More importantly, PTG overexpression reversed HFD-induced glucose intolerance and hyperinsulinemia. Similar studies showed that expression—using adenovirus—of other targeting subunit isoforms of PP1, such as the truncated version of muscle isoform termed “GMΔC,” ameliorated glucose intolerance in rats fed an HFD but did not reduce the high fasting insulin levels of these animals (39).
Also noteworthy was the effect of PTG overexpression in decreasing hepatic steatosis induced by HFD. The reduction in feeding observed in PTGOE mice fed an HFD may account for the lower levels of hepatic triacylglycerol in the fed condition. This decrease in hepatic steatosis was associated with a decrease in the expression of PPARγ and MGAT1. PPARγ is a transcriptional factor that participates in hepatic steatosis in rodents (40,41), and PPARγ-regulated MGAT1 expression has been described as responsible for lipid accumulation in diet-induced hepatic steatosis (31). SREBP1 and GK expression were downregulated in the HFD-fed PTGOE group. SREBP1 and GK expression is stimulated by insulin (42–44), and both genes contribute to hepatic steatosis (45,46). Because PTG overexpression reversed HFD-induced hyperinsulinemia, the lower insulin levels in PTGOE mice may account for the downregulation of SREBP1 and GK. The expression of some of the main lipogenic genes regulated by SREBP1, such as Acc1α and Fasn (47), as well as genes related to lipid oxidation was not statistically different between the genotypes. We therefore conclude that neither changes in de novo lipogenesis nor changes in lipid oxidation contributed to the decrease in hepatic lipid content in the HFD-fed PTGOE mouse group.
A recent study revealed that glycogen shortage in the liver triggers the liver-brain-adipose tissue neural axis independently of glucose and insulin/glucagon levels, thus playing a key role in switching the fuel source from glycogen to triglycerides under prolonged fasting conditions (48). The present results support these observations because overnight-fasted PTGOE mice had higher glycogen levels than control mice. Consequently, many of the changes in metabolism that occur during fasting when the glycogen supply dwindles were attenuated in these animals. PTGOE mice showed a markedly lower hepatic lipid content during fasting, which could be attributable to a lower flux of fatty acids arriving from adipose tissue to the liver. These animals also showed lower levels of fatty acids and β-hydroxybutyrate in serum under fasting conditions, suggesting that lipolysis in adipose tissue is attenuated, as is ketogenesis in the liver. Although the impact of increased PTG or increased hepatic glycogen is consistent throughout the literature, the effects of decreased PTG appear to be more complex. In a first report (49), it was shown that heterozygous deletion of PTG in mice reduced glycogen levels and induced glucose intolerance and insulin resistance with age. These observations agree with other studies of animal models where glycogen stores are decreased by other means, such as depletion of LGS (50). However, a recent article reported that total ablation of PTG reduced fasting glucose and insulin levels in obese mice while improving insulin sensitivity and blocking hepatic steatosis during fasting in HFD-fed mice (25). Changes in the background of the animals used may explain these differences and may indicate that the consequences of decreased PTG are greatly influenced by other factors. It is also important to stress that in that study (25), homozygous deletion of PTG was constitutive, and the effects caused by PTG depletion were not circumscribed to the liver. Therefore, PTG depletion in other tissues, such as adipose tissue, may affect the whole picture. In contrast, we generated mice with a liver-specific overexpression of PTG to avoid interferences with the effect of PTG in other tissues.
The present results demonstrate that liver glycogen accumulation prevents HFD-induced glucose intolerance, decreases food intake, and lowers body weight. In conclusion, the results point to hepatic glycogen content as a potential target for the pharmacological manipulation of diabetes and obesity.
Acknowledgments. The authors thank Ramon Gomis, Rosa Gasa, Marc Schneeberger, and Marc Claret, Institut d’Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS) (Barcelona, Spain), for useful suggestions. They also thank the following members of IRB Barcelona: Mar García Rocha for help and advice; Manuel Gris, Emma Veza, Natalia Plana, and Nuno Vasconcelos for technical assistance; Antonio Berenguer for advice on the statistical analysis of the data; and Tanya Yates for correcting the English version of the manuscript.
Funding. This study was supported by grants from the Spanish Ministry of Economy and Competitiveness (BFU2011-30554) and the CIBERDEM (Instituto de Salud Carlos III, Ministerio de Ciencia e Innovación).
None of the supporting agencies had any role in establishing the work or in writing the manuscript.
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. I.L.-S. contributed to the study design, data research, and writing and revision of the manuscript. D.Z. contributed to the study design, data research, and revision of the manuscript. J.D. designed the PTGOE mice. A.A. contributed to the data research and revision of the manuscript. J.C. contributed to the study design and revision of the manuscript. J.J.G. contributed to the study design and writing and revision of the manuscript. J.J.G. 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.