In Amateur Athletes With Type 1 Diabetes, a 9-Day Period of Cycling at Moderate-to-Vigorous Intensity Unexpectedly Increased the Time Spent in Hyperglycemia, Which Was Associated With Impairment in Heart Rate Variability

In type 1 diabetes, cardiac autonomic neuropathy results from dysfunction of sympathetic and/or parasympathetic nervous system activity and is associated with an increased risk of ventricular arrhythmia and cardiovascular morbidity and mortality (1). Long before the appearance of autonomic neuropathy clinical signs, subtle cardiac autonomic dysfunction can manifest as a decrease in heart rate variability (HRV) and its components (2). A large body of literature describes an altered parasympathetic tone in individuals with uncomplicated type 1 diabetes compared with healthy control subjects, resulting in relative sympathetic overactivity (2).

In an 11-year follow-up study of 83 subjects with type 1 diabetes, Mäkimattila et al. (2) showed that chronic hyperglycemia (high HbA_1c) was a strong predictor of a lower HRV. Chronic hyperglycemia might be attenuated by interventions such as exercise training (3), which has indeed been suggested as a way to improve HRV in type 1 diabetes (4,5).

However, aerobic exercise, particularly when prolonged, intense, and/or unusual, may also trigger glycemic variability (6). Hypoglycemic episodes are common due to the increased muscle glucose disposal associated with high peripheral insulin concentrations, while nondecreased insulin levels in the portal vein prevent glucose release from the liver. Transient hyperglycemic episodes may also occur, for example, during early recovery from intense exercise performed in a postabsorptive state. Notably, it is not only sedentary or inactive patients who are prone to these exercise-induced glycemic fluctuations but also the increasing number of sports enthusiasts with type 1 diabetes engaging in outdoor ultra-endurance events.

Interestingly, outside the context of exercise, it has been suggested that acute glycemic excursions impair cardiac autonomic activity. Thus, Nguyen et al. (7) provided pilot data in six subjects with type 1 diabetes, showing that periods of naturally occurring hyperglycemia (measured over one night) were associated with an impaired global HRV and parasympathetic tone, compared with the nonhyperglycemic periods. Besides, Koivikko et al. (8) showed a reduction in HRV and parasympathetic tone in response to hyperinsulinemic-hypoglycemic clamp as compared with euglycemic clamp in subjects with type 1 diabetes. Additionally, a greater glycemic variability toward low blood glucose values, registered over a regular 5-day period, was associated with impaired HRV in adults with type 1 diabetes (9). However, the literature offers no data about cardiac autonomic activity changes accompanying exercise-induced glycemic fluctuations, even though exercise-induced hypoglycemic episodes may appear long (24 h) after the exercise session.

The aim of this observational study was to explore, in riders with uncomplicated type 1 diabetes, the impact of a 9-day cycling tour on HRV, taking into consideration concomitant exercise-induced glycemic excursions and their influencing factors (i.e., diet and insulin).

Twenty-three riders agreed to participate in this investigation, traveling the 1,456 km that separates Brussels and Geneva over 10 days (mHealth Grand Tour, 3–12 September 2015), including a recovery day (day 4) (Supplementary Table 1). The inclusion criteria were age ≥18 years, a history of type 1 diabetes for >1 year, an HbA_1c (dating back no more than 3 months) <9% (75 mmol ⋅ mol⁻¹), and to have already experienced a 1-day ride >160 km as well as rides of 100 km on consecutive days. All participants were free from overt micro- and macrovascular complications, except one who suffered from arteriopathy; thus, the latter was excluded from the analyses. Written informed consent was obtained, and data collection was granted approval by Commission Nationale de l’Informatique et des Libertés (CNIL) (MMS/TDG/ALU/AE151191). Usual physical activity was assessed using the short version of the International Physical Activity Questionnaire. Additionally, 10 riders had undergone an incremental maximal exercise test (VO_2max) as part of independent medical monitoring of athletes. Whether participants suffered from hypoglycemia unawareness was also reported. According to VO_2max or training status, the participants were recreationally trained or trained cyclists (10,11).

Throughout the 10 days of the tour, subjects wore a CGM (Dexcom G4 Platinum) (three or more calibrations per day by capillary finger-stick measurement with TAPcheck glucometer) to evaluate glycemic excursions and variability. They concomitantly wore a heart rate monitor (Polar H7) to assess HRV at night plus during time spent at different exercise intensities during cycling (12).

Helped by the onsite dietitian, riders reported their self-estimation of carbohydrates consumed at every meal (breakfast, lunch, and dinner). The day prior to the start of the tour, riders were interviewed by the dietitian to assess their ability to accurately count carbohydrates. Those who were less accustomed to this practice benefited from a closer follow-up by the dietitian throughout the tour. Riders also reported the exact times and types of snacks consumed. For better standardization, they were encouraged to use the gels, bars, and recovery drinks provided by the staff. Among the 13 individuals treated with continuous subcutaneous insulin infusion (CSII), 8 made their insulin pump data available for the study analyses. Every morning of the tour, just before breakfast, blood pressure and body composition (bioelectric impedance) were noted. Data from the heart rate monitor, CGMs, carbohydrate intake, capillary blood glucose, and symptomatic (awareness) episodes of hypoglycemia were gathered via Bluetooth on smartphones and thereafter downloaded with specific software for further analyses.

The HRV analysis was performed with Kubios HRV software in accordance with the Task Force of the European Society of Cardiology and the North American Society for Pacing and Electrophysiology (13). HRV was analyzed during a standardized calm (sleeping) period between midnight and 4:00 a.m. throughout the 9 days of cycling. We analyzed time domain parameters (SD of normal to normal R-R [SDNN]), (percentage of differences >50 ms between successive NN intervals [pNN50]), and the root mean square of differences of successive NN intervals [RMSSD]) as well as frequency domains of HRV by the Fast Fourier Transform (high frequency [HF], 0.15–0.40 Hz, and low frequency [LF], 0.04–0.15 Hz).

Glycemic excursions and variability were calculated from CGM recordings over several specific periods: 1) from midnight to 4:00 a.m., concomitant with the period of HRV analysis; 2) the day before, from the beginning of breakfast to 2 h postdinner; 3) over periods of 24 h; 4) during the cycling periods excluding the lunch break; and 5) during early and late recovery (2 and 6 h following the cycling periods). Glycemic excursions considered were the percentage of time spent in range (between 70 and 180 mg ⋅ dL⁻¹ [3.9 and 10 mmol ⋅ L⁻¹]), below range (level 1 hypoglycemia, <70 mg ⋅ dL⁻¹ [<3.9 mmol ⋅ L⁻¹]; level 2 hypoglycemia, <54 mg ⋅ dL⁻¹ [<3.0 mmol ⋅ L⁻¹]), and above range (level 1 hyperglycemia, >180 mg ⋅ dL⁻¹; level 2 hyperglycemia, >250 mg ⋅ dL⁻¹; and hyperglycemia >300 mg ⋅ dL⁻¹ [14] [10, 13.9, and 16.7 mmol ⋅ L⁻¹]) levels (15). Glycemic variability was assessed through coefficient of variation (%CV) (15), SD, mean amplitude of glycemic excursions (MAGE), continuous overlapping net glycemic action 1&2 (CONGA 1&2), and average daily risk ratio (ADRR) indexes (16).

All statistical analyses were performed with SPSS software (version 19; IBM). The quantitative data are described as the mean ± SD. Normality was checked with the Shapiro-Wilk test. A logarithm transformation was applied to data with a non-Gaussian distribution. In all of the following models, covariates were added as fixed effects and subjects as random effects to consider between- and within-participant variability.

In the 1st set of models, HRV parameters were studied as dependent variables in linear mixed models with time (i.e., days 1–10, except for day 4, which was a recovery resting day) as covariates. In a 2nd set of models, we then successively analyzed the effects of subjects and exercise characteristics during the tour, with time kept as a covariate. In a 3rd set of models, the effects of time as well as glycemic variability and excursions were tested as covariates, in addition to exercise characteristics if significant in the 2nd set of models.

In a 4th set of models, glycemic variability and excursions were studied as dependent variables in mixed models or multinomial and binary logistic regressions with time (i.e., days 1–10, except for day 4) and circadian (i.e., night vs. day periods, only for analyses over the periods of 24 h) effects as covariates. Multinomial or binary logistic regressions were specifically used to assess the percentage of time spent in hypo- and hyperglycemia (see details in the legend of Table 3). In the 5th and 6th sets of models, with glycemic variability and excursions as dependent variables, we tested the effects of exercise characteristics and subject characteristics, respectively, with time and, for the 24-h periods, circadian effects as covariates. Subsequently, we tested the effects of carbohydrates (grams) ingested (7th set of models) and the effects of carbohydrates ingested and insulin administered (n = 8 subjects on CSII [8th set of models]) on glycemic variability and excursions, considering the time effect and, when significant in the 5th set of models, the exercise characteristics.

A P value <0.05 was considered statistically significant. All results are expressed as the mean estimation “e.” A particular focus was also given to the difference magnitude in addition to inferential statistical tests expressed using P values.

Technical problems were encountered in the collection of nighttime beat-to-beat heart rate and/or interstitial glucose (e.g., disconnection of sensor, synchronization problems) for three subjects. Thus, 20 subjects were included in the final analyses. Their characteristics are displayed in Table 1.

Carbohydrate data were processed for 19 of the 20 riders because one of them incorrectly completed the food questionnaire. Among these 19 riders, 15 were considered as having mastered advanced carbohydrate counting including appropriate carbohydrate gram estimation. The other four participants benefited from closer assistance from the dietitian with counting carbohydrates at every meal throughout the tour. Exercise intensities and exact duration of the cycling periods (Supplementary Table 1) were obtained only for 46 (i.e., 25.6%) full days over the 180 (i.e., 9 days for 20 riders) days of cycling analyzed because of problems with transient disconnection between the belt and heart rate watch monitor. Riders spent a large part of the cycling period at moderate (i.e., 160.0 ± 38.3 min per day) and vigorous (i.e., 155.1 ± 27.0 min per day) intensities. Morning body mass, percent of fat mass, muscular mass, and hydration did not change over the 9-day period. While systolic blood pressure remained unchanged throughout the tour, morning diastolic blood pressure and mean blood pressure decreased significantly (main time effect e −0.43 and e −0.39, respectively, P < 0.05). Anthropometric, demographic, and diabetes-related variables were not significantly related to blood pressure throughout the tour.

The results of HRV are presented in 16 individuals because 4 riders did not wear their heart rate monitor correctly at night. The results of mixed models for the association of time, subject and exercise characteristics, and glycemic excursions, with temporal and frequency HRV domains displaying significant main effects, are presented in Table 2.

Parasympathetic tone parameters (i.e., HF 910.4 ± 1,440.2 ms², pNN50 17.2 ± 20.2%, RMSSD 43.5 ± 32.5 ms, mean over the 9 days) and sympathetic-vagal balance (i.e., LF-to-HF ratio 4.1 ± 2.2, mean over the 9 days) did not significantly change with time throughout the tour and were not altered by the characteristics of the exercise performed each day. However, global HRV (as reflected by SDNN, 101.5 ± 39.0 ms, over the 9 days) tended to decrease with the number of kilometers traveled the day before, without the influence of exercise intensity or time.

The 12 men had a higher sympathetic-vagal balance than the 4 women. Age associated with decreased parasympathetic tone. Aerobic fitness (VO_2max) was positively associated with global HRV and parasympathetic tone.

Global HRV was not linked with glycemic excursions. A decrease in sympathetic-vagal balance and an increase in parasympathetic tone during the night were associated with longer time spent in level 1 hypoglycemia during the previous day and the 6-h postexercise recovery period. A decrease in parasympathetic tone during the night was associated with a longer time spent in a state of level 2 hyperglycemia during the concomitant night, as well as with hyperglycemia >300 mg ⋅ dL⁻¹ during the previous day.

Change in time spent in hypo-, normo-, and hyperglycemia throughout the tour is presented in Fig. 1 (for the 24-h periods) and Supplementary Fig. 1 (for cycling and early and late recovery periods). Glycemic variability is reported in Supplementary Fig. 2. Factors influencing glycemic outcomes are presented in Table 3.

Hypoglycemia unawareness (subjectively reported) did not influence hypoglycemic excursions. A higher HbA_1c level was associated with a longer time spent above range but lower glycemic variability. Neither mode of insulin therapy nor the habit of using CGM influenced glycemic outcomes.

Notably, while the number of hypoglycemic episodes of which subjects were aware did not change during the tour (0–1 episode/day among the riders over the 9 days), the percentage of time spent below range (hypoglycemia levels 1 and 2) was decreased in riders (time effect), as measured over all the periods studied. However, this decrease was at the expense of glycemic variability and hyperglycemic excursions (as measured over all the periods studied), which worsened throughout the tour (time effect) and were more frequent on the days a greater distance was ridden (for hyperglycemia >300 mg ⋅ dL⁻¹, MAGE and SD), without a significant effect of exercise intensity. This was accompanied by a decrease in time in range. Throughout the tour, subjects experienced less time in range and more time above range during the daytime compared with the nighttime.

Throughout the tour, the riders progressively decreased the daily carbohydrate content ingested via daytime snacks (e −10.29, P < 0.001) but did not change the carbohydrate content of the nighttime snacks or the three meals (Fig. 1C). In this context of sustained repeated exercise, neither the carbohydrates from meals nor the carbohydrates from nighttime or daytime snacks (which decreased with time) were significantly associated with glycemic excursions throughout the tour (7th set of models). In the sample of subjects providing insulin pump data, bolus and basal rates were not significantly changed throughout the tour (8th set of models) (Fig. 1D). Notably, in comparison of insulin doses used during normal daily life (data obtained from six of the eight subjects during a usual week before the tour) with those administered during the tour, no significant difference appeared (paired t test) (insulin basal rate [mean ± SD] 14.9 ± 7.5 vs. 20.9 ± 10.0 units ⋅ day⁻¹, bolus 18.8 ± 10.7 vs. 23.9 ± 13.0 units ⋅ day⁻¹ throughout the 9 days of the tour vs. during 1 week of daily life, respectively). Nevertheless, during the tour, the larger the daily amount of insulin bolus (either in units or in units ⋅ kg⁻¹) was, the greater the extent of time spent below range (level 2 hypoglycemia, e increase of 8.87 or 0.11, respectively, P < 0.05) experienced by riders from breakfast to 2 h postdinner. Glycemic outcomes during the night (between midnight and 4:00 a.m.) were not significantly associated with the concomitant insulin basal rate throughout the tour.

Our study of 20 riders with type 1 diabetes highlighted, for the first time, that the repetition of long-duration exercise bouts at moderate-to-vigorous intensity over 9 days may trigger hyperglycemic excursions, which were negatively associated with parasympathetic tone, while time spent in hypoglycemia decreased. The type of statistical model used was chosen to ensure that the observed relationships (e.g., between hyperglycemia, a covariate, and parasympathetic tone, the dependent end point) were not due to simultaneous changes in time or in exercise characteristics but appeared for any given value of these two outcomes, which were added as covariates in the model.

While demographic characteristics such as age (17) and sex (2) are well-known predictors of sympathetic-vagal balance, as confirmed in our study, the impact of exercise training on cardiac autonomic function in type 1 diabetes is documented less frequently. Interestingly, in agreement with a recent study (18), we found that VO_2max, as a reflection of regular aerobic training level, was positively associated with cardiac autonomic function (parasympathetic tone in our study).

However, nighttime beat-to-beat heart rate recordings throughout the tour revealed that unusual multiday sustained moderate-to-vigorous exercise, without appropriate recovery, may conversely trigger impairment of HRV in type 1 diabetes. The longer the distance ridden during the day, the lower the global HRV tended to be during the subsequent night. This phenomenon could reflect a state of overreaching as already observed in recreationally trained runners without diabetes (19), who displayed a significant reduction in indexes of HRV (including SDNN) up to 24 h after an ultramarathon (64 km distance, 1,572 m accumulative altitude change). It should, however, be noted that studies on the effect of overreaching on HRV in healthy athletes remain few and far between, sometimes with the finding of no significant change in HRV (20,21).

In our study, it is worth noting that some cardiac autonomic function parameters were actually linked with glycemic excursions.

Thus, a longer time spent at low glucose levels (corresponding to both level 1 and 2 hypoglycemia) (15) during late recovery (i.e., the 6-h period postexercise) was associated with increased parasympathetic activity during the subsequent night. To the best of our knowledge, the only studies that have explored the link between hypoglycemia and cardiac autonomic balance specifically in the context of physical exercise (only one session) have considered either the changes occurring during the 60 min before the hypoglycemic episode (22) or the changes occurring concomitant with the hypoglycemic period (23,24) and showed either a decrease or an increase in parasympathetic tone, respectively. Further studies are needed to confirm the increase in nocturnal parasympathetic tone associated with postexercise hypoglycemic periods and to understand its clinical implications.

While time spent in a state of hypoglycemia decreased throughout the tour, we observed a surprisingly significant decrease in time spent in a state of euglycemia due to a considerable increase in time spent in a state of hyperglycemia. While the deleterious impact of chronic hyperglycemia (HbA_1c) on HRV and parasympathetic tone is already well-documented in youth with type 1 diabetes (25), this study is the first to show a link between cardiac autonomic imbalance and acute hyperglycemic periods in the context of physical exercise. To our knowledge, only two studies, with quite controversial results, have attempted to explore the possible acute impact of hyperglycemia on HRV in individuals with type 1 diabetes but without involving concomitant physical exercise and based only on a limited number of glycemic values (i.e., only one measure taken in a fasting state and the other 30 min after a regular meal [26] or one measure every 30 min during 1 night in only six subjects [7]).

As impaired cardiac vagal control is associated with higher cardiac mortality (27), it appears to be crucial to elucidate the factors involved in the worsening of hyperglycemia observed during the tour. Although reducing insulin administration and/or increasing carbohydrate intake are commonly recommended for avoiding hypoglycemic episodes around physical exercise in type 1 diabetes (6), these measures may not be needed in athletes for whom insulin doses and diet are already well adjusted to their usual intensive exercise training. Accordingly, in our work, the cyclists did not increase carbohydrate intake throughout the tour and presumably did not change their usual insulin dose (as verified among the individuals using an insulin pump). Thus, while insulin and diet might not be the direct cause of exercise-induced hyperglycemia worsening, the characteristics of the multiday exercise, i.e., prolonged and including a significant portion of vigorous intensity, may play a fundamental role. Intense exercise (>85% VO_2max) to exhaustion is known to induce an increase in glycemia during the early recovery period because plasma catecholamines and glucagon take time (∼30 min and 30–50 min, respectively) to return to resting concentrations. In subjects without diabetes, the increase in glycemia during early recovery following intense exercise is counteracted by a twofold increase in plasma insulin, whereas individuals with type 1 diabetes are prone to transient hyperglycemia unless a bolus of insulin is delivered (28). Consistent with this theory, the only cyclist in our study who did not experience significant postexercise hyperglycemia received a frequent insulin correction bolus in the hours following exercise.

Additionally, when exercise becomes extremely prolonged, glucose metabolism might be disrupted long after the end of exercise, as revealed in a study in subjects without diabetes (29). Equally of interest, in line with this result, we noticed a positive association of the number of kilometers traveled during the day with time spent in a state of hyperglycemia as well as with glycemic variability during the surrounding 24 h. To the best of our knowledge, blood metabolite and hormonal response to multiday long-distance sports events has only been the topic of one single case report in an athlete with type 1 diabetes, showing progressive increase in markers of inflammation (CRP) and muscle damage (creatine kinase), which are two factors of insulin resistance (30,31), while changes in cortisol, an activator of hepatic gluconeogenesis, did not exactly follow trends of hyperglycemic excursions (32). Future studies on larger sample sizes will be needed to ascertain the cause of the observed persistent hyperglycemia. Data from a study using nonrepeated sustained exercise, i.e., a single marathon, suggest that persistence of free fatty acid oxidation long after exercise may suppress carbohydrate oxidation (33). Additionally, in subjects without diabetes, the exercise-induced increase in lactatemia has been shown to be enhanced throughout the week following a marathon (29), and lactate can then serve as an alternative muscle substrate for sparing blood glucose (34).

Finally, it is noteworthy that in addition to the negative link between glycemic excursions and cardiac autonomic activity during the 9-day repeated moderate-to-vigorous exercise, such activity was associated with a progressive decrease in morning diastolic blood pressure. This phenomenon has already been observed in response to long and strenuous events in healthy athletes and was suggested to be due to metabolic vasodilatation and/or ineffective transduction of sympathetic outflow from arterial smooth muscle (35). Therefore, better attention should be paid to vulnerability to orthostatic challenges after multiday prolonged exercise in athletes with type 1 diabetes.

While our results are particularly relevant given the growing popularity of long-distance or multiday running or cycling events, including for persons with type 1 diabetes, the findings should be interpreted in the context of the study limitations. The nonstandardization of the resting day and the limited data obtained during riders’ daily life prevent drawing comparisons between control resting periods and the multiday cycling challenge. We obtained insulin data only from a small proportion of the participants, making generalization difficult. In addition, the accuracy of CGMs may be reduced during exercise bouts due to multiple factors, such as subcutaneous dehydration, temperature variations, rapid decreases in glycemia, etc. For addressing these limitations, the CGMs were calibrated at least three times daily by capillary finger-prick testing and body hydration was controlled every morning.

Given the observational study design, we cannot draw conclusions about causality. However, the current cross-sectional study was a necessary first step toward future implementation of interventional randomized control trials. Based on the current results, these future trials could contribute to reducing hyperglycemia induced by outdoor endurance sports, instead of focusing only on hypoglycemia, and provide observations of the subsequent impact on HRV. This will make it possible to partition off the possible impact of hyperglycemia from that of overreaching, the latter having already been reported in athletes without diabetes. As far as we are aware, our study is the first to confirm empirical data (i.e., based on patients’ narrative and on a single case report (36)) on the increased risk of hyperglycemia triggered by multiday ultra-endurance events in amateur athletes with type 1 diabetes, with concomitant careful consideration of exercise intensity and carbohydrate intake. Notably, two other recent reports are available (37,38) but on professional cyclists, whose adaptations to elite-stage races are certainly different from those of amateur athletes. In addition to offering widely generalizable data, the unique nature of our study lies above all in the novel way the putative links between sustained exercise-induced glycemic excursions and cardiac autonomic activity are examined. Further studies are needed to understand the mechanisms involved in the hyperglycemic effect of such multiday moderate-to-vigorous prolonged exercise events and to gain more insight into the health consequences of accompanying cardiac autonomic imbalance.

In conclusion, this study on amateur athletes with type 1 diabetes demonstrates that multiday prolonged exercise at moderate-to-vigorous intensity increased the time spent in hyperglycemia, and this was negatively associated with parasympathetic cardiac tone. These results are important considering the putative consequences on future cardiac mortality risk. In this context, beyond research on hypoglycemia prevention strategies, future work on understanding and managing exercise-induced hyperglycemia should be promising.

Acknowledgments. The authors thank all members of the Orange HealthCare team, particularly J. Braive, V.-D. Dam, O. Graille, M. Montaner-Gomez, and P. Chopineau for their help in providing all the material needed for linking, via a connected phone, glycemic data with other data (carbohydrate intake, HRV, body composition, arterial pressure, etc.). Data extraction was facilitated by O. Beltoise (Glooko Diasend) and by B. Klinkenbijl and S. Guerra from ClinInfo teams. The authors are also thankful for the crucial logistical support of A. Denton from Hydon Consulting Society, as well as C. Hugues (diabetes dietitian) for collecting diet data throughout the tour, M. Gonnet (Another Brain Society), S. Tagougui (Lille University), and P. Morel (University of the Littoral Opale Coast) for data analysis assistance, J.-F. Gautier (Department of Diabetes and Endocrinology, Lariboisiere Hospital, Paris, France) for his contribution to the study design, L. Canipel (Société Française de Télémédecine) for her help with ethics approval, P. Fontaine (Endocrinology Unit, Lille University Hospital) and M. Garcia Vigueras for their scientific advice, A. Bocock (Melun), A. Denton (Hydon Consulting Society), and B. Heyman (University of Rennes) for revising the English, and A. Bertrand (Statistical Methodology and Computing Service, Université catholique de Louvain) for checking the statistical analyses. All the above mentioned support was supplied at no charge.

Funding. This study was supported by a donation from Linde Homecare France, which allowed recruitment of E.L. as a postdoctoral researcher for 1 year.

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

Author Contributions. E.H. and M.D. conceived the experiments. E.L. and E.H. performed the experiments, collected and analyzed data, and wrote the manuscript. O.B. analyzed insulin data. J.H. created algorithms for analyses of glycemic variability. B.P. contributed to statistical analyses. F.-X.G. and S.B. contributed to the HRV analyses. O.B., J.H., B.P., S.B., F.-X.G., and M.D. contributed to the performance of the experiments and reviewed the manuscript. E.H. 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.

Prior Presentation. Parts of this study were presented in abstract form at the second congress of Société de Physiologie et Biologie Intégrative, Lille, France, 27–29 June 2018.