Seven-Day Caloric and Saturated Fat Restriction Increases Myocardial Dietary Fatty Acid Partitioning in Impaired Glucose-Tolerant Subjects

In industrialized and developing countries, reductions in dietary quality and physical activity strongly contribute to the development of chronic diseases such as obesity, type 2 diabetes, and cardiovascular disorders (1,2). While physical activity has decreased in the past few years, the energy intake, principally from fat and refined carbohydrate, has markedly increased. Abnormal fatty acid metabolism plays a major role in the pathophysiology of type 2 diabetes by altering insulin sensibility and glucose homeostasis (3,4). Adipose tissue samples from prediabetic and diabetic subjects show inefficient storage of dietary fatty acids (DFAs) (4,5), increasing the “spillover” of triglycerides (TGs) and nonesterified fatty acid (NEFA) in circulation during the postprandial state. This fatty acid spillover appears to be directly related to abdominal obesity in subjects with type 2 diabetes (6), resulting in an increased exposition of some, but not all, lean tissues to DFA (7).

There is strong experimental support for a deleterious effect of fatty acids on insulin sensitivity and β-cell function in humans (3). Short-term dietary trials in humans (8–10) have shown that saturated fatty acid (SFA) substitution for monounsaturated fatty acids (MUFAs) or polyunsaturated fatty acids (PUFAs) may lead to insulin resistance and more β-cell dysfunction. A change in myocardial fatty acid metabolism is also associated with diabetic cardiomyopathy (11). In animal and human dietary trials, the replacement of TG-raising carbohydrates with monounsaturated or n3 polyunsaturated fats (12) improved cardiac function. Recently, using the noninvasive method of positron emission tomography (PET) imaging with the oral administration of 14(R,S)-[¹⁸F]-fluoro-6-thia-heptadecanoic acid ([¹⁸F]-FTHA), a long-chain fatty acid analog (13), we have shown that subjects with impaired glucose tolerance (IGT) display an increase in myocardial DFA uptake and partitioning (defined as the relative biodistribution of DFA among all organs) that is associated with reduced left ventricular function (14). These metabolic and functional cardiac abnormalities were corrected after a 1-year lifestyle intervention with modest weight loss (15). As this lifestyle intervention included an individually tailored change in diet and physical activity, it was not possible to determine whether weight loss, change in insulin sensitivity, change in diet, or change in physical activity was responsible for correction of the cardiac DFA metabolic abnormalities.

The aim of the current study was to determine the effect of a 7-day caloric restriction with reduction of SFA intake, an intervention known to significantly improve insulin sensitivity (9), on cardiac and whole-body DFA metabolism. The hypothesis was that a short-term, low-calorie, low-SFA diet, with improvement in insulin sensitivity, when compared with an isocaloric (ISOCAL) diet, would be associated with a reduction in cardiac DFA uptake and partitioning and improvement in white adipose tissue DFA partitioning in individuals with IGT.

Fifteen Caucasian subjects (seven men and eight women between 44 and 65 years of age) with glucose intolerance, which was defined as having results from a 2-h post–75-g oral glucose tolerance test, undertaken after a 12-h overnight fast, of between 7.8 and 11.1 mmol/L on two occasions, underwent two postprandial metabolic studies using a randomized crossover design. Subjects with a history or clinical evidence of any cardiac disorder; any evidence of kidney, liver, or thyroid dysfunction; or any uncontrolled medical or surgical condition were excluded. Subjects who had received any antidiabetic medication (except metformin), β-blockers, or fibrates; had a history of any dietary or severe past allergic reaction; or had participated in any research trial involving radiation exposure within the past 12 months were also excluded. One male patient was removed from postprandial analyses because of technical problems with postprandial data acquisition during one of the two protocols. Those subjects who had dyslipidemia and were receiving treatment with a statin (three participants) or had hypertension and were treated with antihypertensive agents (four participants) had to stop taking these medications 3 weeks and 7 days, respectively, prior to the metabolic assessments. No women involved in this study were receiving hormonal contraceptive agents or menopausal hormone therapy. All metabolic studies performed in premenopausal women were performed during the follicular phase of the menstrual cycle.

The subjects followed two different diets: an ISOCAL diet (0% alcohol, 17% proteins, 50% carbohydrates, and 33% fats [34% saturated fat, 66% monounsaturated fat and polyunsaturated fat]) and a hypocaloric diet low in saturated fat (LOWCAL diet) (<7% of total energy compared with ∼10% for the ISOCAL diet; 500 kcal/day less than the ISOCAL diet; 0% alcohol, 17% proteins, 50% carbohydrates, and 33% fats [20% saturated fat, 80% monounsaturated fat and polyunsaturated fat]) for 7 days before undergoing each experimental protocol. Diets were administered in random order, and a washout period of 3–4 weeks was observed between the two diets. An experienced research nutritionist prescribed and supervised dietary intake every other day during the ISOCAL and LOWCAL interventions. The participants were asked to report food intake using a food diary with a scale to weigh the food ingested. Nutritional characteristics of the ISOCAL and LOWCAL diets were determined using Nutrific software (https://nutrific.fsaa.ulaval.ca). The subjects were instructed to maintain their regular physical activities and avoid strenuous exercise during the dietary intervention. During the 7 days of the ISOCAL and LOWCAL diets, each participant wore an accelerometer (GT3X+; ActiGraph, Pensacola, FL) to control for any variation in their physical activity.

On arrival at the test center on the morning of both meal tests, after a 12-h overnight fast, body weight, height, and waist circumference were measured, and lean body mass was determined with a Body Composition Analyzer (TBF 300A; Tanita Corporation of America, Arlington Heights, IL). An intravenous catheter was placed in one forearm for infusions, and another was placed in a distal vein in the contralateral arm and was maintained in a heating pad (55°C) for blood sampling.

Each participant underwent a 6-h postprandial experimental protocol after both the 7-day LOWCAL and ISOCAL diets, in which a standard liquid meal prepared as described (6) was consumed over 20 min (400 mL for a total of 906 kcal, 33 g or 33% as fat, 34 g or 17% as proteins, and 101 g or 50% as carbohydrates). Blood samples were collected in tubes containing Na₂ EDTA and orlistat (30 mg/mL; Roche, Mississauga, Ontario, Canada) to prevent in vitro TG lipolysis.

From time −60 min in the fasting state of the protocol to time 360 min after meal intake, a constant infusion of [7,7,8,8-D₄]-potassium palmitate (0.01 µmol ⋅ L⁻¹ ⋅ kg⁻¹ ⋅ min⁻¹ in 100 mL 25% human serum albumin; Cambridge Isotopes Laboratories Inc.) and [1,1,2,3,3-²H]-glycerol (0.05 µmol ⋅ L⁻¹ ⋅ kg⁻¹ ⋅ min⁻¹ in saline; Cambridge Isotopes Laboratories Inc.) were administered, preceded by an intravenous bolus of [1,1,2,3,3-²H]-glycerol (1.6 µmol ⋅ L⁻¹ ⋅ kg⁻¹) and NaHCO₃ (1.2 µmol ⋅ L⁻¹ ⋅ kg⁻¹; Cambridge Isotopes Laboratories Inc.) to prime the bicarbonate pool. In the liquid meal, [U-¹³C]-palmitate (10.8 µmol ⋅ L⁻¹ ⋅ kg⁻¹; Cambridge Isotopes Laboratories Inc.) was added.

Each subject was placed in the supine position and scanned using a 16-slice PET/computed tomography (CT) scanner (Gemini TF; Philips, Eindhoven, the Netherlands) (13). At time −60 min before meal intake, [¹¹C]-acetate (∼185 MBq) was administered intravenously, and electrocardiogram gating was performed, allowing the application of PET ventriculography to determine cardiac function, as previously published (14,15).

We used our novel oral [¹⁸F]-FTHA method (13) to determine whole-body DFA partitioning. [¹⁸F]-FTHA given orally is absorbed through chylomicron-TG, delivered into the circulation from the thoracic duct, and redistributes normally into other circulating fatty acid pools (13). This method assumes [¹⁸F]-FTHA-containing chylomicron-TG hydrolysis and uptake into local tissue microcirculation, which are similar to those of chylomicron-TGs not containing [¹⁸F]-FTHA. At time 0 min, ∼70 MBq [¹⁸F]-FTHA, produced as previously described (16), was mixed into Intralipid 20% (Baxter, Mississauga, Ontario, Canada), incorporated into gel capsules (T U B Enterprises), and given orally with the liquid meal. A dynamic list-mode PET scan centered on the thoracoabdominal segment was performed between time 90 and 120 min, and a whole-body PET/CT acquisition at time 360 min was performed as previously described (13). The maximal gastrointestinal tract radioactivity exposure for oral [¹⁸F]-FTHA administration was estimated at 2.35 mSv in the stomach. The total radioactivity exposure to the participants was 19.54 mSv, including the two protocols. All tracers were tested for sterility and pyrogenicity.

For dynamic PET acquisitions, myocardial and hepatic fractional DFA uptakes (K_i) were determined using Patlak linearization (17) as previously described (14,15). For whole-body scans, the mean value of pixels (mean standard uptake value [SUV]) for all tissues of interest was recorded (14,15). Liver volume was extrapolated by drawing the liver surface on each 5 mm slice of the whole-body CT scan at time 360 min using OsiriX 5.8.2 software (Pixmeo SARL, Bernex, Switzerland).

The cardiac blood flow index (K₁ in min⁻¹ of [¹¹C]-acetate, with correction for first-pass tissue extraction [18]) and oxidative metabolism index (K₂ in min⁻¹ of [¹¹C]-acetate) were estimated from [¹¹C]-acetate using a three-compartment model (19), as previously published (7). For analysis of ventricular function, PET data from [¹¹C]-acetate images were analyzed as previously described (14,15).

Glucose, insulin, C-peptide, total NEFA, and TG levels were measured as previously described (6). Chylomicrons and plasma lipids were separated by ultracentrifugation and assayed for [¹⁸F] activity and TG concentration (13). The Matsuda index was determined as previously described (20).

The plasma palmitate appearance rate (Ra_palmitate), plasma NEFA appearance rate (Ra_NEFA), plasma glycerol appearance rate (Ra_glycerol), fractional palmitate oxidation (FOx_palmitate), plasma palmitate oxidation (Ox_palmitate), and net total body carbohydrate oxidation (CHOox) and fatty acid oxidation (FA_ox) were calculated as previously described (21). Total energy expenditure was estimated by indirect calorimetry during the postprandial state.

Data are expressed as the mean ± SEM. A Wilcoxon test, Mann-Whitney U test, and two-way ANOVA for repeated measures were used when appropriate. Spearman correlations were performed to examine the association between variables. A two-tailed P value <0.05 was considered significant. All analyses were performed with Prism version 6.0 for Macintosh (GraphPad, San Diego, CA).

Informed written consent was obtained, in accordance with the Declaration of Helsinki, and the protocol received approval from the Human Ethics Committee of the Centre de Recherche du Centre Hospitalier Universitaire de Sherbrooke.

The aim of this study was to induce a decrease in energy intake of ∼500 kcal/day during 1 week with a saturated fat intake of <7% of energy (Table 1). The participants reduced their energy intake by 546.8 ± 20.8 kcal/day (P < 0.0001) and their total postprandial energy expenditure by 34.9 ± 14.8 kcal ⋅ 370 min⁻¹ (P < 0.02) in the LOWCAL diet versus the ISOCAL diet. Saturated fat intake was 6.6 ± 0.1% during the 7-day LOWCAL diet (−4.5 ± 0.2% vs. ISOCAL diet, P < 0.0001). Although we attempted to keep the percentage of energy from proteins, carbohydrates, and total fats the same between the LOWCAL and ISOCAL diets, there was a marginal but statistically significant increase in the percentage of energy from carbohydrates during the LOWCAL diet (0.6 ± 0.2%, P < 0.004). The decrease in the percentage of energy from saturated fats was compensated by an increase in the percentage of energy from MUFAs and PUFAs (MUFA+PUFA; 4.2 ± 0.3%, P < 0.0001) during the LOWCAL diet. There was also a significant decrease in fiber content during the LOWCAL diet (−3.1 ± 0.6 g/day, P < 0.0009).

After the LOWCAL diet, the participants had lost a significant amount of weight (−0.9 ± 0.3 kg, P < 0.002) and BMI (−0.3 ± 0.1 kg ⋅ m⁻², P < 0.002), but had no significant variation in their waist circumference (Table 2). The participants lost almost as much lean mass (−0.4 ± 0.2 kg, P < 0.08) as nonlean mass (−0.5 ± 0.2 kg, P = 0.14). Furthermore, IGT subjects had a significant reduction in their liver volume after the LOWCAL diet (−124.5 ± 49.3 cm³, P < 0.04). We found no variation in fasting plasma glucose, NEFA, and adiponectin levels. However, we found a significant decrease in fasting levels of insulin (−8.1 ± 4.1 pmol ⋅ L⁻¹, P < 0.05), TG (−0.3 ± 0.1 mmol ⋅ L⁻¹, P < 0.05), and leptin (−2.8 ± 0.8 ng ⋅ mL⁻¹, P < 0.002). Postprandial excursion levels of plasma glucose, insulin, C-peptide, NEFA, TG, and chylomicron-TG are shown in Fig. 1. During the postprandial state, we found a significant decrease in postprandial area under the curve for insulin at 60–360 minutes (AUC_60–360 insulin) (−22,144 ± 7,458 pmol ⋅ L⁻¹ ⋅ 300 min⁻¹, P < 0.02) and a trend toward reduction in AUC_60–360 C-peptide (−85.2 ± 87.8 ng ⋅ mL⁻¹ ⋅ 300 min⁻¹, P < 0.06). We also found a significant decrease in AUC_60–360 TG (−77.3 ± 24.3 mmol ⋅ L⁻¹ ⋅ 300 min⁻¹, P < 0.02) and a trend toward a reduction in AUC_120–360 chylomicron-TG (−15.8 ± 7.8 mmol ⋅ L⁻¹ ⋅ 240 min⁻¹, P < 0.07). After the LOWCAL diet, no variation was found for the total postprandial plasma [¹⁸F] activity, [¹⁸F]-NEFA activity, total [¹⁸F]-TG activity, and [¹⁸F]-chylomicron activity (Fig. 2A–D). Finally, the participants did not change their physical activity (−22.8 ± 35.8 kcal ⋅ day⁻¹, P = 0.78).

After the LOWCAL diet, we found a significant 16.4% increase in myocardial partitioning (0.32 ± 0.14 SUV, P < 0.04) (Fig. 3C) and a significant 28.6% increase in skeletal muscle (quadriceps femoris) partitioning (0.11 ± 0.05 SUV, P < 0.05) (Fig. 3E). The increase in myocardial DFA partitioning with LOWCAL tended to be higher in women (ΔSUV 0.037 ± 0.272 vs. 0.537 ± 0.099 in men, P = 0.13) (Supplementary Fig. 2A). No variations in DFA partitioning were found in the liver (Fig. 3D) and in subcutaneous (Fig. 3F and G) and perirenal (Fig. 3H) adipose tissue depots. No variation in myocardial DFA fractional uptake was found between diets (Fig. 4A), but hepatic DFA fractional uptake tended to be higher (0.017 ± 0.009 min⁻¹, P < 0.09) (Fig. 4B) after the LOWCAL diet. No gender difference was observed in the response of myocardial and hepatic DFA fractional uptake between the two diets (Supplementary Fig. 3A and B).

After the LOWCAL diet, we found a significant decrease in AUC_0–360 Ra_palmitate (−10,290 ± 6,577 µmol ⋅ min⁻¹ ⋅ 360 min⁻¹, P < 0.04) (Supplementary Fig. 1A). No variation was found for Ra_NEFA, Ra_glycerol, palmitate concentration, Fox_palmitate, Ox_palmitate, CHO_ox, and FA_ox with two-way ANOVA or Mann-Whitney U test for AUC (Supplementary Fig. 1B–H).

Left ventricular ejection fraction (LVEF) was significantly decreased after the LOWCAL diet when compared with the ISOCAL diet (−2.9 ± 0.5%, P < 0.008) (Table 3). We also found a trend toward an increase in end-systolic volume after the LOWCAL diet (2.4 ± 1.1 mL, P < 0.07) and a trend toward decrease in cardiac output (−384 ± 156 mL ⋅ min⁻¹, P < 0.06), cardiac index (−0.19 ± 0.08 mL ⋅ min⁻¹ ⋅ m⁻², P < 0.06), and diastolic blood pressure (−2.8 ± 1.3 mmHg, P < 0.06). No variations were found in blood flow index and oxidative metabolism.

We found a strong negative correlation between the change in MUFA+PUFA, the change in dietary fiber intake or leptin level, and the change in myocardial DFA level partitioning (Table 4). We also found that the reduction in liver radio-density correlated with the increase in myocardial DFA partitioning of DFA.

We previously found that subjects with IGT display an increase in myocardial DFA uptake that is associated with decreased left ventricular function (14,22), an abnormality that was corrected by modest weight loss after a 1-year individually tailored lifestyle intervention (15). Here, we found that the improvement of insulin resistance with a 7-day hypocaloric, low-SFA diet did not change early 2-h postprandial myocardial DFA uptake in subjects with IGT. In contrast to our initial hypothesis, we found that a LOWCAL diet increased cardiac DFA partitioning over 6 h after meal intake despite significant improvement in insulin sensitivity, a small but significant degree of weight loss, and excellent compliance of the subjects with the dietary interventions. DFA partitioning in white adipose tissue also did not change significantly with the LOWCAL diet, which was reflected by the unchanged postprandial total plasma NEFA concentration and Ra_NEFA, and the unchanged oxidation rate, although the plasma Ra_palmitate was slightly reduced.

The mechanism for the LOWCAL diet–induced increase in cardiac DFA partitioning is not clear, but we can rule out an increase in cardiac chylomicron-TG uptake and an increase in NEFA spillover as possible causes. Taken together, the observation of a higher early liver fractional DFA uptake without an increase in liver DFA partitioning at 6 h suggests an increase in hepatic turnover of DFA. As fasting plasma TG levels were significantly reduced, this may suggest an increase in VLDL-TG clearance that is associated with an increased secretion rate. This could in turn lead to increased transport of DFA to the heart via VLDL-TG uptake (23). As insulin is known to suppress VLDL-TG secretion (24), it is possible that lower postprandial insulin levels led to an increase in VLDL-TG secretion. However, we found no correlation between either the change in insulin level or the change in insulin sensibility (by Matsuda index) and the change in cardiac DFA partitioning. Unfortunately, the transport of [¹⁸F]-FTHA into VLDL-TG was not measured in the current study. In contrast, the increase in cardiac DFA partitioning was significantly correlated with reduced leptin level. Leptin is known to increase cardiac fatty acid oxidation in animals (25). On the other hand, in db/db mice, a murine model of nonfunctional leptin receptor, TGs accumulate in myocardial cells (26) through an increase of peroxisome proliferator–activated receptor α target genes, including lipoprotein lipase (27). [¹⁸F]-FTHA accumulates in oxidative and nonoxidative metabolic pathways in the heart (16). Therefore, a shift toward nonoxidative fatty acid metabolism would not be detectable by our method.

Diabetes is an important risk factor for heart failure: each 1% elevation of hemoglobin A_1c is associated with a 15% increase in the risk of heart failure (28,29). At least part of this risk can be attributed to diabetic cardiomyopathy that is characterized by reduced myocardial energetic efficiency (17,30) (i.e., increased oxidative metabolism in the face of a similar or reduced mechanical work). Type 2 diabetes is associated with increased cardiac NEFA uptake and oxidation during the fasting state both in men (31) and in women (32). Prediabetic and diabetic subjects are also characterized by an increase in intramyocardial TG content (33–36). An increase in myocardial fatty acid use and oxidation and a decrease in myocardial energetic efficiency can predispose subjects to ischemic damage and cardiomyocyte apoptosis (30,35). Cardiac lipoprotein lipase–mediated lipolysis is also thought to play a role in diabetic and metabolic cardiomyopathies in animals (37–39). In the current study, the increase in cardiac DFA partitioning was associated with reduced LVEF and a trend toward reduction in cardiac output. This is consistent with our previous studies showing an association between increased cardiac DFA partitioning and reduced left ventricular function in humans (14,15) and rats (17). However, diastolic blood pressure was also reduced together with reduced postprandial energy expenditure, a likely explanation for the reduction in cardiac output. Furthermore, a reduction in plasma NEFA levels with acipimox treatment over 1 week was shown to reduce left ventricular function (40). In subjects with type 2 diabetes, a 3-day very low-calorie diet (471 kcal/day, but with SFA >13%) induced an increase in myocardial TG levels and reduced diastolic function, but with no change in LVEF measured by MRI (41). Fatty acid uptake by the heart is essential for intramyocellular TG repletion and the maintenance of cardiac function (42,43). It is therefore premature to postulate that enhanced cardiac DFA partitioning can be mechanistically linked to a reduction in LVEF in the current study. The small 2–3% reduction in LVEF within the normal range that we observed in the current study is likely not clinically significant. Whether these changes may be clinically relevant in patients with cardiomyopathies remains to be determined.

The dietary fiber content was not exactly the same between the LOWCAL and ISOCAL diets. Dietary fibers are plant constituents that are resistant to digestion and absorption by the human small intestine with complete or partial fermentation in the large intestine and are classified into the following two types: water soluble, which represents ∼20% of total dietary fiber intake, and water insoluble, which represents the remaining 80% (44). In the current study, we found that a reduction in total dietary fibers intake was associated with increased cardiac DFA partitioning. Water-soluble fibers are associated with a decrease in cardiovascular events by lowering LDL cholesterol levels (44,45). Serum TG concentration is not modified by water-soluble fibers (45), but a study (46) in type 2 diabetes subjects has shown that high intake of water-soluble plus insoluble fibers could decrease serum TG concentration. However, the amount of total dietary fiber intake that was administered in the latter study was twice the amount that our subject consumed during the dietary interventions, and plasma TG levels fell concomitantly with the reduction in total fiber intake. Bacterial fermentation of dietary fibers in the colon induces biosynthesis of short-chain fatty acids such as acetate, propionate, and butyrate (47). Butyrate decreases the synthesis of adipose TG lipase (48), which contributes to myocardial uptake and the oxidation of fatty acids (49). An increase in acetate production by the intestinal microbiota could induce an increase in the production of acetyl-CoA in myocardial cells, which in turn could induce an increase in malonyl-CoA, an allosteric inhibitor of carnitine palmitoyltransferase 1, the rate-limiting enzyme of β-oxidation (49). We did not measure plasma or stool levels of butyrate and acetate in the current study. Future studies will be needed to address the effect of the change in dietary fiber intake, microbiota metabolism, and cardiac DFA metabolism.

An important limitation of our work is the small number of participants. This limitation is due to the complexity and costs of the metabolic protocols used. Despite the small number of participants, the a priori power of our study for most outcomes ranged between 70% and 95%. The small number of men and women limits our power to detect sex differences in the contribution of adipose tissues versus direct chylomicron-TG uptake to cardiac DFA partitioning (22). Another limitation of our study is that the PET ventriculography method may not be as sensitive as cardiac MRI or echocardiography to detect changes in left ventricular diastolic function.

In conclusion, the current study demonstrates that a short-term, moderate caloric restriction with low SFA levels and improvement of insulin resistance induces an increase in cardiac DFA partitioning, which is associated with a decrease in LVEF in individuals with IGT. Neither the change in direct chylomicron-TG uptake nor the change in NEFA spillover from white adipose tissue can explain these findings. Therefore, short-term caloric and SFA dietary restrictions do not lead to the same changes in cardiac and adipose tissue DFA metabolism as those associated with weight loss induced by lifestyle changes in subjects with IGT. Lower total fiber intake was closely associated with increased cardiac DFA partitioning in the current study. Future studies will be needed to test the effect of dietary fiber intake on cardiac DFA metabolism.

Funding. This work was supported by a grant from the Canadian Institutes of Health Research (MOP 53094 to A.C.C.) and was performed at Centre de Recherche du Centre Hospitalier Universitaire de Sherbrooke, a research center funded by Fonds de Recherche du Québec–Santé. C.N. is the recipient of a Canadian Institutes of Health Research postdoctoral fellowship.

Duality of Interest. The Canadian Institutes of Health Research-GlaxoSmithKline Chair in Diabetes, which is held by A.C.C., was created in part through a donation of $1 million by GlaxoSmithKline to the Université de Sherbrooke. A.C.C. holds the Canadian Institutes of Health Research–GlaxoSmithKline Chair in Diabetes. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. C.N. was responsible for the collection, analysis, and interpretation of the data and drafted the article or revised it critically for important intellectual content. M.K., F.F., L.B., S.D., F.J.-D., S.P., S.C.C., B.G., and E.E.T. were responsible for the collection, analysis, and interpretation of the data. A.C.C. was responsible for the collection, analysis, and interpretation of the data; drafted the article or revised it critically for important intellectual content; and was responsible for the conception and design of the experiments. A.C.C. 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.