Glutamine-Elicited Secretion of Glucagon-Like Peptide 1 Is Governed by an Activated Glutamate Dehydrogenase

Glucagon-like peptide 1 (GLP-1), secreted from intestinal L cells, potentiates insulin secretion from the pancreatic β-cells (1). Importantly, this potentiation is glucose dependent (i.e., GLP-1 stimulates insulin secretion when blood glucose levels are elevated) (2). Hence, GLP-1 is unlikely to induce severe hypoglycemia, a dreaded complication of insulin treatment. Moreover, GLP-1 delays gastric emptying, leading to reduced hunger and food intake (1). These features of GLP-1 underlie the increasing use of GLP-1 analogs in treatment of type 2 diabetes.

Currently, therapeutic actions of GLP-1 in vivo are mediated by administration of GLP-1 analogs or dipeptidyl peptidase 4 (DPP-4) inhibitors. However, DPP-4 has important physiological roles other than degradation of GLP-1 (3), and indiscriminate inhibitors may also affect functions of other biological processes, resulting in a potentially wide range of physiological effects. For instance, DPP-4 has been shown to suppress tumor growth and inhibit malignancies (4); its inhibition may thus increase cancer risk. L cell function is relatively well preserved in type 2 diabetes (5), as opposed to that of the pancreatic β-cell, which gradually deteriorates during progression of the disease (6). In fact, studies have revealed that the GLP-1 secretagogue glutamine potentiates GLP-1 secretion in healthy subjects as well as in those with type 2 diabetes (5). Notably, glutamine does not elicit secretion of insulin from the pancreatic β-cell (7). Levels of glutamine are tightly regulated by glutaminase, glutamine synthetase, and glutamate dehydrogenase (GDH). The amino acid functions as both a modulator of ammonia levels and a substrate for energy production (8). An increased understanding of stimulus-secretion coupling in the L cell could enable nutritional or pharmacological enhancement of GLP-1 secretion.

L cells are difficult to isolate from the intestine due to their relatively low abundance. Moreover, survival of L cells in purified cultures is poor (9). β-Cell stimulus-secretion coupling, in contrast, has been extensively studied in isolated pancreatic islets, sorted β-cells, and multiple clonal cell lines (10). Only few immortalized L cell lines are available, with the GLUTag cell line being recognized as a useful and relevant model of the primary L cell (11).

Studies of stimulus-secretion coupling in the β-cell have revealed two major pathways underlying secretion of insulin. In both of these, increased blood glucose levels result in an increase in glucose uptake and phosphorylation followed by glycolysis and complete oxidation of the sugar in mitochondria. The ATP generated thereby acts on the ATP-sensitive K⁺ channel, leading to its closure, membrane depolarization, opening of voltage-gated Ca²⁺ channels, and finally exocytosis of insulin granules (12). In addition to this triggering pathway, an amplifying pathway has been shown to account for a large proportion of secreted insulin. Multiple metabolites have been implicated in the amplifying pathway, but a functional role for most of them has been questioned due to conflicting results from later studies (12,13). Studies on stimulus-secretion coupling in the L cell have revealed similar mechanisms as in the pancreatic β-cell (14). However, in contrast to the β-cell, which expresses Glut1/2, the L cells also express electrogenic sugar and amino acid transporters (15). Hence, Na⁺-coupled nutrient uptake has been suggested to account for a large proportion of GLP-1 secretion (14). Moreover, the L cell has also been shown to respond to stimulation with peptones (16).

In the current study, we examined stimulus-secretion coupling in GLUTag cells and used the widely studied β-cell model INS-1 832/13 as a reference. Results from these studies were validated in vivo in mice.

GLUTag cells were cultured in DMEM containing 5.6 mmol/L glucose and supplemented with 10% FBS at 37°C in a humidified atmosphere containing 95% air and 5% CO₂. INS-1 832/13 cells were cultured in RPMI 1640 containing either 5.6 or 11.1 mmol/L glucose and supplemented with 10% FBS, 10 mmol/L HEPES, 2 mmol/L glutamine, 1 mmol/L sodium pyruvate, and 50 μmol/L 2-mercaptoethanol at 37°C in a humidified atmosphere containing 95% air and 5% CO₂. Cells were seeded in 24-well tissue-culture plates for 24 or 48 h prior to experiments.

GLUTag and INS-1 832/13 cells were incubated as previously described in detail (14,17). Briefly, GLUTag cells were washed twice with 0.5 mL PBS and then incubated in 0.5 mL Hanks’ balanced salt solution (HBSS) containing a DPP-4 inhibitor (0.1 mmol/L diprotin A; Sigma-Aldrich, St. Louis, MO), fatty acid–free BSA (0.2%, weight for weight), and various concentrations of glucose, glutamine, dimethylglutamate (DMG), and leucine for 2 h. INS-1 832/13 cells were washed with 0.5 mL PBS and preincubated in 0.5 mL HBSS with 2.8 mmol/L glucose for 2 h. Then, cells were incubated in 0.5 mL HBSS containing the same secretagogues as used for the GLUTag-cells for 1 h. An epigallocatechin (EGCG) stock solution was prepared by dissolving EGCG (4 mmol/L) in water containing 1 mmol/L ascorbic acid. EGCG and ascorbic acid were then added to both preincubation and incubation media at 20 μmol/L and 0.5 mmol/L, respectively (18). Secreted GLP-1 and insulin were determined in the supernatant after centrifugation using an ELISA to active GLP-1 (7–36) (EMD Millipore, Billerica, MA) or human insulin (Mercodia, Stockholm, Sweden), respectively, according to manufacturer’s instructions. Secretion of hormones was normalized to protein levels, determined by the bicinchoninic acid assay.

Cells from the hormone secretion assays were swiftly washed with 1 mL ice-cold PBS prior to quenching of metabolism by adding 300 μL methanol at −80°C. Metabolites were extracted and derivatized as previously described (19,20). The derivatized metabolites were analyzed on an Agilent 6890N gas chromatograph (Agilent Technologies, Atlanta, GA) equipped with an Agilent 7683B autosampler (Agilent Technologies) and coupled to a LECO Pegasus III TOFMS electron impact time-of-flight mass spectrometer (LECO Corp., St. Joseph, MI), as previously described (21), and on a 5973 inert GC/MS system (Agilent Technologies) in single ion-monitoring mode.

Oxygen consumption rate (OCR) was measured by the XF24 Extracellular Flux Analyzer (Seahorse Bioscience, North Billerica, MA) as previously described (22). Cells were preincubated for 1 h at basal glucose levels (0 mmol/L for GLUTag cells and 2.8 mmol/L for INS-1 832/13 cells) at 37°C in air, after which respiration was measured in the absence of glucose, followed by addition of either leucine or glutamine. Oligomycin, carbonyl cyanide-p-trifluoromethoxy-phenylhydrazone, and rotenone were injected as described previously (22).

GLUTag and INS-1 832/13 cells were cultured in 24-well plates and incubated in glucose-free HBSS (GLUTag) or HBSS supplemented with 2.8 mmol/L glucose (INS-1 832/13). Subsequently, the incubation buffer was removed, and cells were washed in PBS and lysed by addition of 100 μL lysis buffer. Finally, cells were snap-frozen on dry ice/ethanol and levels of ATP determined using a luciferase-based luminescence assay (BioThema, Handen, Sweden).

Nucleotides were also analyzed by high-performance liquid chromatography with ultraviolet detection at 254 nm using an XBridge Amide column (4.6 × 150 mm, 3.5 μm). Briefly, cells were deproteinized by adding 1.2 mol/L HClO₄, followed by centrifugation and removal of lipids from the supernatant by extraction with CHCl₃. Samples were neutralized by addition of 2 mol/L K₂CO₃, filtered, and diluted fourfold in acetonitrile. Nucleotides were eluted using a linear gradient composed of A: acetonitrile and B: 2 mmol/L KH₂PO₄ starting at 75% A and ending at 62% A in 10 min.

Single-cell ATP/ADP ratio measurements were carried out using the pericam-based genetically encoded ATP biosensor Perceval HR (23). Cells were seeded onto poly-d-lysine–coated eight-well chambered cover glasses (Lab-Tek; Thermo Scientific, Waltham, MA) at a density of 70,000 cells/cm². At 24 h after seeding, cells at ∼50% confluency were transfected with 1 µg plasmid encoding Perceval HR (Addgene ID: 49083). On the day of imaging, 48 h after transfection, cells were preincubated at 37°C in 400 μL buffer P (135 mmol/L NaCl, 3.6 mmol/L KCl, 1.5 mmol/L CaCl₂, 0.5 mmol/L MgSO₄, 0.5 mmol/L Na₂HPO₄, 10 mmol/L HEPES, and 5 mmol/L NaHCO₃, pH 7.4) containing 0 or 2.8 mmol/L glucose for GLUTag and INS-1 832/13 cells, respectively. After 1.5 h of incubation, cells were imaged with 490-nm excitation and 535-nm emission filter settings on a Zeiss LSM510 inverted confocal fluorescence microscope (Carl Zeiss, Oberkochen, Germany).

Total RNA was extracted from GLUTag and INS-1 832/13 cells using the RNAeasy RNA purification kit (Qiagen, Hilden, Germany) according to the manufacturer’s protocol. RNA concentrations were determined using a NanoDrop Spectrophotometer (Thermo Scientific). Equal quantities of total RNA were reversely transcribed using the RevertAid First-Strand cDNA synthesis kit (Fermentas, Vilnius, Lithuania) in reactions containing 500 ng total RNA. Quantitative real-time PCR was performed using the TaqMan gene expression assay (GDH isoform GLUD1, regulator of GDH activity sirtuin-4 [SIRT4], and mitochondrial transcription factor A) using a 7900HT Fast Real-Time System (Applied Biosystems, Foster City, CA). The quantitative real-time PCR was carried out as previously described (24). Gene expression was quantified by the comparative threshold cycle method, in which the amount of target is expressed as 2^−∆∆Ct using hypoxanthine guanine phosphoribosyl transferase as reference gene. GDH activity was measured as previously described in detail (25).

Fasted (2 h) female C57BL/6 mice (n = 24; Janvier Laboratories, Rennes, France) were anesthetized with midazolam (0.4 mg/mouse; Dormicum; Hoffmann-La Roche, Basel, Switzerland), fluanisone (0.9 mg/mouse), and fentanyl (0.02 mg/mouse; Hypnorm; Janssen, Beerse, Belgium). Mice were placed on a heating pad to maintain body temperature and divided into three groups. Retro-orbital blood samples were taken at time zero with a Luer capillary glass pipette rinsed in EDTA. Thereafter, mice were infused colorectally (26) with 1 mL physiological NaCl (9 mg/L; control), 10 mmol/L glutamine, or 10 mmol/L DMG in physiological NaCl. The concentration of secretagogues in our study are in the same range as postprandial intestinal concentrations of the amino acid (27). EGCG was infused at 1 mmol/L in 0.5 mmol/L ascorbic acid diluted in physiological NaCl. The infusion rate was 1 mL/min. After 10 min, an additional blood sample was taken. Blood samples from both time points were centrifuged, and plasma was separated from blood cells. Plasma samples were immediately assayed for total GLP-1 (EMD Millipore; Merck Millipore, Billerica, MA).

All data are presented as means ± SEM for the indicated number of experiments. Metabolite data were log2-transformed prior to assessment of differences between groups by the paired Student t test or ANOVA with Bonferroni test post hoc when more than two groups were compared. Orthogonal projection to latent structures discriminant analysis (OPLS-DA) was performed in SIMCA 13.0 (Umetrics, Umeå, Sweden) on mean-centered and unit variance scaled data.

Glucose elicited secretion of GLP-1 and insulin from GLUTag and INS-1 832/13 cells, respectively (Fig. 1). To further examine to which extent metabolism of the hexose is involved in this process, we profiled changes in metabolism elicited by glucose using gas chromatography/mass spectrometry. To this end, we generated data on levels of amino acids, fatty acids, and glycolytic, pentose phosphate pathway, and tricarboxylic acid (TCA) cycle intermediates. Data were analyzed by OPLS-DA; one model was calculated for each of the cell types, with the glucose level as discriminating variable. Data were then visualized in a shared and unique structures–like plot (Fig. 2A), derived from the loadings, scaled as correlations, of the OPLS-DA models (28). The significance of the loadings, obtained by jackknifing, highlighted metabolites, the levels of which were uniquely increased by glucose in GLUTag or INS-1 832/13 cells, as well as changes in metabolite levels common to both cell types. Clearly, glucose elicited similar changes in metabolism in GLUTag and INS-1 832/13 cells. Thus, levels of glycolytic and TCA cycle intermediates increased in both cell types (Fig. 2B and C). Glutamate levels increased 5- (P < 0.001) and 1.3-fold (P < 0.05) in GLUTag and INS-1 832/13 cells, respectively. Aspartate levels decreased in both cell types: 0.6-fold (P < 0.001) in GLUTag cells and 0.5-fold (P < 0.001) in INS-1 832/13 cells. Glycerol-3-phosphate levels increased after glucose stimulation only in the INS-1 832/13 cells (1.5-fold; P < 0.001). Levels of glutamine were unaltered in both cell types.

After having established that mitochondrial metabolism is activated by glucose in both GLUTag and INS-1 832/13 cells, we investigated whether the cell lines responded similarly also to glutamine. Glutamine alone was found to potently stimulate GLP-1 secretion from GLUTag cells, whereas it was ineffective in stimulating insulin secretion from INS-1 832/13 cells (Fig. 3A). Activation of GDH with leucine did not affect glutamine-induced GLP-1 secretion, but permitted insulin secretion in response to glutamine. Leucine alone did not induce either GLP-1 secretion from GLUTag cells or insulin secretion from INS-1 832/13. These differences between cells were not caused by the differing culturing conditions used, as INS-1 832/13 cells cultured for 48 h at 5.6 mmol/L glucose showed a similar pattern of insulin secretion as those cultured at 11.1 mmol/L glucose (Fig. 3B).

Metabolite profiling revealed that addition of glutamine alone increased levels of TCA cycle intermediates in GLUTag cells. Notably, levels of these intermediates were unaffected by leucine (Fig. 3C). In INS-1 832/13 cells, in contrast, glutamine was largely ineffective in stimulating mitochondrial metabolism in the absence of leucine (Fig. 3D). Hence, these results suggest that GDH is active in the GLUTag cells even in the absence of the allosteric activator leucine, which is required for glutamine-activated TCA metabolism and hormone secretion in β-cells.

To further investigate whether glutamine-elicited hormone secretion has a metabolic component, we stimulated GLUTag and INS-1 832/13 cells with the membrane-permeable glutamate analog DMG, which bypasses Na⁺-coupled plasma membrane transporters. Similar to glutamine, DMG induced GLP-1 secretion from GLUTag cells independent of leucine, whereas leucine was required for DMG to provoke insulin secretion from INS-1 832/13 cells (Fig. 4A).

The metabolic response elicited by either glutamine or DMG in the GLUTag cells was similar with regard to GLP-1 secretion and the increase in levels of TCA cycle intermediates, with the exception of aspartate (Fig. 4B and Supplementary Fig. 1); the level of this metabolite increased to a greater extent when stimulated with glutamine (2.4-fold by glutamine vs. 1.7-fold by DMG; P < 0.001). In INS-1 832/13 cells, in contrast, glutamine and DMG failed to raise levels of TCA cycle intermediates in the absence of the allosteric GDH activator leucine (Fig. 4C and Supplementary Fig. 1).

Thus far, our results indicate that GDH activity differs between GLUTag and INS-1 832/13 cells. To examine this further, we assessed GDH activity in lysed cells. To reflect the conditions used in our previous experiments, we used glutamine as substrate, which, via deamination by glutaminase, produces glutamate and ammonia. Then we determined NADH generated by the GDH-catalyzed oxidation of glutamate to α-ketoglutarate. However, these analyses showed no difference in activity between GLUTag and INS-1 832/13 cells, estimated as the glutamine-elicited area under the curve (AUC) (Fig. 5A).

Next, we aimed to determine enzyme activity in intact cells. As a measure of GDH activity linked to TCA cycle metabolism and oxidative phosphorylation, we monitored the OCR after stimulation with glutamine followed by leucine or vice versa. In GLUTag cells, leucine was ineffective in stimulating OCR, whereas glutamine potently increased OCR, which did not increase further upon subsequent addition of leucine (Fig. 5B). In contrast, both leucine and glutamine were required to elicit a robust increase in OCR in INS-1 832/13 cells (Fig. 5C). These results further support that GDH exists in an enhanced activity state in GLUTag cells.

GDH activity is regulated both allosterically by, for example, amino acids and nucleotides and covalently by ADP ribosylation, catalyzed by SIRT4 (7,29). Unexpectedly, the GLUD1/SIRT4 expression ratio was 55% (P < 0.05) lower in GLUTag compared with INS-1 832/13 cells (Fig. 6A). Moreover, basal levels of ATP, reflecting the energy status of the cell, were 2.9-fold (P < 0.05) higher in GLUTag cells (Fig. 6B). We also determined cytosolic ATP/ADP ratio using Perceval HR (Fig. 6C). The maximal response (Fig. 6D) and AUC of the ATP/ADP ratio (Fig. 6E) in response to glutamine stimulation were higher in GLUTag cells as compared with INS-1 832/13. Addition of leucine did not affect the ATP/ADP ratio in either of the cell types. Levels of ADP, AMP, guanosine monophosphate, guanosine diphosphate, and guanosine triphosphate did not differ between cell types. Of note, basal levels of leucine were 7.4-fold (P < 0.01) higher in GLUTag cells than in INS-1 832/13 cells (Fig. 6F). Given leucine’s role as allosteric activator of GDH, its higher basal level may explain the increased GDH activity in the GLUTag cells.

To further examine the role of GDH in glutamine-elicited GLP-1 secretion, we inhibited the enzyme using EGCG. Secretion of GLP-1 in response to both glutamine and DMG was significantly reduced in the presence of EGCG (Fig. 7A). Notably, glutamine was still effective in eliciting GLP-1 secretion in presence of the inhibitor (P < 0.001), whereas DMG was not. Reduced secretion of GLP-1 in response to the secretagogues in presence of EGCG was paralleled by an almost complete abolishment of the glutamine- and DMG-elicited increases in levels of TCA cycle intermediates (Fig. 7B and C).

To gain physiological support for our findings, we examined whether the membrane-permeable glutamate analog DMG could affect GLP-1 secretion in vivo in mice. Colorectal infusion of either glutamine or DMG yielded a 2.1-fold (P < 0.05) and a 2.0-fold (P < 0.05), respectively, stronger stimulation of GLP-1 secretion as compared with the control (Fig. 7D). Inhibition of GDH by EGCG abrogated secretion of GLP-1 in response to DMG (Fig. 7E). A schematic depiction of differences in stimulus-secretion coupling between the L cell and β-cell is shown in Fig. 8.

Metabolic coupling in nutrient-stimulated insulin secretion from the pancreatic β-cell has been comprehensively studied. These studies have highlighted mitochondria as key in the generation of signals that trigger and amplify secretion of the hormone. The L cell, in contrast, is less well characterized, but studies have revealed mechanisms similar to those in the β-cell (14), as well as mechanisms that may be unique to the L cell (15). As opposed to the β-cell, secretion of GLP-1 from the L cell has also been shown to be governed by sodium-coupled nutrient uptake, implying an action essentially independent of intracellular metabolism of the secretagogue (14).

In the current study, we investigated the metabolic component of stimulus-secretion coupling in the L cell. To facilitate these analyses, we compared the L cell model GLUTag with the well-established β-cell model INS-1 832/13. Conditions were selected from the literature (14,17) and hence differ between GLUTag and INS-1 832/13 cells, but allow comparison with previous studies in these cells. Secretion of insulin in response to glucose, glutamine, and leucine was not altered when INS-1 832/13 cells were precultured for 48 h at the lower glucose levels that were used for the GLUTag cells. It must, however, be acknowledged that metabolism in these immortalized cell lines may not exactly mirror metabolism in vivo.

We found that exposure of GLUTag cells or L cells in vivo to nutrients and stimuli acting upstream and downstream of electrogenic transporters yielded qualitatively similar results. Hence, our data support previous studies that have indicated a metabolic component, involving ATP production and K_ATP channel closure, analogous to that observed in the β-cell, in L cell glutamine (30), glucose (9,14), and fructose sensing (14). It needs to be taken into account that glutamine has also been suggested to act via a cAMP-dependent nonelectrogenic sensing mechanism (31).

It is possible that different sensing mechanisms might govern GLP-1 secretion from different parts of the gastrointestinal tract. GLUTag cells are derived from colonic tumors, and the in vivo administration of glutamine and DMG mainly targeted the more distal segments of the gastrointestinal tract. Oral glucose tolerance tests conducted in sodium-glucose linked transporter 1 knockout mice have revealed a blunted early secretion of GLP-1 (32) but increased intestinal levels of glucose to associate with an exaggerated late secretion of the hormone (33). Hence, our data support that nonelectrogenic nutrient uptake and sensing may play a more important role in the more distal sections of the gastrointestinal tract.

Our results also provide evidence that glutamine-elicited GLP-1 secretion requires an active GDH. When the enzyme was inhibited, both mitochondrial metabolism and GLP-1 secretion became insensitive to DMG. However, whereas GLP-1 secretion was significantly reduced in response to glutamine, the amino acid was still effective in eliciting secretion of the hormone. This was paralleled by an almost abolished response in mitochondrial metabolism, suggesting that glutamine may act via additional mechanisms. This contrasts with the β-cell, in which GDH activation by, for example, leucine is necessary for glutamine-stimulated insulin secretion (18,34–38). In fact, mutations resulting in escape from nucleotide inhibition and constitutive activation of GDH cause hyperinsulinemic hypoglycemia (7). Moreover, GDH activity is also regulated by SIRT4 that inhibits GDH activity via ADP ribosylation (29). Hence, silencing of SIRT4 in β-cells results in glutamine-stimulated insulin secretion (39,40). We could not find any differences in SIRT4 mRNA levels between GLUTag and INS-1 832/13 cells. Rather the relative levels of SIRT4 to GLUD1 were higher in the former, as was the overall energy state, reflected by total cellular ATP levels. Instead, metabolite profiling revealed higher basal levels of leucine in the GLUTag cells. It is possible that this elevation of basal leucine levels serves to keep GDH in an active state in GLUTag cells, thereby promoting secretion of GLP-1.

The ATP/ADP ratio changed only slowly upon stimulation of GLUTag cells with glutamine and did not increase significantly upon addition of leucine to INS-1 832/13 cells. Such slow responses are consistently observed and likely due to consumption of ATP by Ca²⁺-ATPases (41) and, in this case, also leucine-elicited energy using protein synthesis (42).

Whereas low GDH activity is a prerequisite for normal β-cell function, the opposite holds true for astrocytes (43). Astrocytes rely on GDH activity to enable glutamate clearance and energy production required for glutamate uptake (44). In humans, but not rodents, these cells express the GTP-insensitive GDH isoform GLUD2, enabling these pivotal functions to also operate at a high-energy state. Notably, the neurotransmitter glutamate has been shown to be secreted from GLUTag cells (45), as well as α- (46,47) and β-cells (48). Indeed, enteroendocrine cells and cells of the nervous system share multiple features, which previously were misinterpreted because of both cell lineages being derived from the neural crest (49).

In conclusion, our data suggest that nonelectrogenic nutrient uptake and metabolism play an equally important role in nutrient sensing in colonic L cells as they do in the pancreatic β-cells. An activated state of GDH is essential in the L cells, whereas activity of this enzyme needs to be inactivated in the β-cell in the absence of glucose. Our results support that glutamine and related amino acid analogs may offer a means of increasing GLP-1 levels without affecting insulin secretion directly in vivo.

Acknowledgments. The authors thank Dr. Daniel J. Drucker, Mount Sinai Hospital, Toronto, Ontario, Canada, for permitting use of the GLUTag cells.

Funding. This work was supported by grants from Vetenskapsrådet (Swedish Research Council), Novo Nordisk Foundation, Swedish Diabetes Foundation, Crafoordska Stiftelsen, Stiftelsen Lars Hiertas Minne, Fredrik och Ingrid Thurings Stiftelse, O.E. och Edla Johanssons Vetenskapliga Stiftelse, Åke Wibergs Stiftelse, Direktör Albert Påhlssons Stiftelse, Magnus Bergvalls Stiftelse, Inga and John Hain Foundation, Hjelt Foundation, and Kungliga Fysiografiska Sällskapet i Lund (Royal Physiographic Society of Lund). An equipment grant from Knut och Alice Wallenbergs Stiftelse was also received.

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

Author Contributions. L.E.A., L.S., M.A.-M., N.V., and M.F. designed and performed experiments, analyzed data, interpreted results, and edited the manuscript. C.B.A. and J.A.C. designed and performed experiments and edited the manuscript. C.B.W., H.M., and N.W. provided intellectual guidance and cowrote the paper. P.S. conceived and directed the project, analyzed data, interpreted results, and wrote the first draft of the paper together with L.E.A. P.S. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.