The Effect of a Single Bout of Continuous Aerobic Exercise on Glucose, Insulin and Glucagon Concentrations Compared to Resting Conditions in Healthy Adults: A Systematic Review, Meta-Analysis and Meta-Regression

This Review and Meta-analysis was registered at PROSPERO (registration number: CRD42020191345). PRISMA guidelines were followed throughout the preparation of this manuscript [26].

CENTRAL, CINAHL, Embase, Global Health, HMIC, Medline, PubMed, PsycINFO, ScienceDirect, Scopus and Web of Science databases were searched from inception to May 2020. Searches were undertaken between March 2020 and May 2020 using the following concepts and search terms (parentheses): 1. Intervention (‘exercise’, ‘run’, ‘running’, ‘cycle’, ‘cycling’, ‘walk’, ‘walking’), 2. Comparator (‘rest’, ‘resting’, ‘control’, ‘ctrl’, ‘no exercise’, ‘sedentary’), 3. Outcomes (‘glucose’, ‘insulin’, ‘glucagon’), and 4. Study Design (‘crossover’, ‘cross-over’, ‘counterbalanced’). These were then joined (1 AND 2 AND 3 AND 4) to provide the final set of search results. No limits were used during any database search. Full details of the search strategy are provided in Electronic Supplementary Material Appendix S1.

Database results were imported into Covidence systematic review software (Veritas Health Innovation, Australia). Titles and abstracts were independently reviewed by all authors and classified as ‘yes’, ‘no’ or ‘maybe’. Papers classified as ‘yes’ or ‘maybe’ proceeded to the full-text screening stage. Full-text papers were then classified as ‘yes’ or ‘no’ independently by two authors (J.F. and E.S.C), with those classified as ‘yes’ proceeding to the data extraction stage. Any disagreements in paper classification were examined by all authors before coming to an agreement regarding the eligibility of the paper.

Data were extracted by a single author (J.F.) into an electronic spreadsheet (Excel 2016, Microsoft Corporation, USA) under the following columns: author; year of publication; sample size; participant characteristics; intervention characteristics; pre- and post-exercise concentrations of glucose and/or insulin and/or glucagon concentrations; and corresponding concentrations of glucose and/or insulin and/or glucagon concentrations in the resting control arm. WebPlotDigitizer Version 4.2 (Ankit Rohatgi, USA) was used to extract data from articles that only presented data in graphical form. If not all error bars were presented, homoscedasticity was assumed and the variance from the timepoint within the same experimental arm was imputed. All extracted data were checked for accuracy by a second author (E.S.C).

Following data extraction, glucose, insulin and glucagon values were converted to SI units (glucose: mmol/L; insulin: pmol/L; glucagon: ng/L). If standard errors or 95% confidence intervals were provided, these were converted to a standard deviation. For each outcome, change scores for exercise and resting arms were calculated by subtracting pre-exercise concentrations from post-exercise concentrations. Mean differences (MDs) between resting and exercise arms were then calculated by subtracting the resting change score from the exercise change score. A positive MD represented an increase in the outcome with exercise, whereas a negative MD represented a decrease with exercise. When within-participant correlation coefficients were not available, a correlation coefficient of 0.5 was assumed to calculate variance and standard error [27]. Sensitivity analyses were performed using correlation coefficients of 0.3, 0.7 and 0.9 to assess the robustness of findings to this assumption.

Studies which reported participants undertaking multiple exercise interventions but only one resting arm were combined into a single change score [28]. Exercise intervention characteristics (duration and intensity) were then averaged. Studies which did not report exercise intensity relative to maximal aerobic capacity (\(\dot{V}\)O₂ max) were converted using equations reported previously [29, 30].

Risk of bias was assessed using the Revised Cochrane Risk of Bias Tool for Randomized trials (RoB 2.0) with additional considerations for cross-over trials. These additional considerations include carryover effects, period effects, and concerns that trials may report only analyses based on the first period. The risk of bias was assessed using the following domains: bias arising from the randomization process; bias due to deviations from intended intervention; bias due to missing outcome data; bias in the measurement of the outcome; and bias in the selection of the reported result. No studies were excluded based on the risk of bias assessment.

Data were entered into Stata 16 (StataCorp, USA) for analysis. Data included: participant characteristics (metabolic state [postprandial or fasted], sample size, % males, age, BMI, \(\dot{V}\)O₂ max), exercise characteristics (mode [cycle ergometer or treadmill], duration, intensity), mean difference and corresponding standard error. Postprandial exercise was defined as exercise performed within 6 h of meal ingestion. Fasted exercise was defined as the exercise performed 6 h after last meal ingestion.

Simple effect sizes for each outcome were calculated using a random-effects model and with the Sidik–Jonkman approach being employed [31]. All simple effect sizes were presented as (unstandardised) MDs and using SI units to facilitate interpretability of results. A random-effects model was chosen over a fixed-effects model to account for differences in participant characteristics and methodology between studies [32]. Heterogeneity was assessed using the chi-squared (Q) and I² statistic. A Q value above the degrees of freedom (df) for the estimate and an I² statistic > 50% indicated large heterogeneity between studies. To investigate the influence of participant characteristics and methodology on MDs, sub-group meta-analyses (categorical covariates) and random-effects meta-regression (continuous covariates) were performed. For the random-effects meta-regression, a positive coefficient indicated that an increase in the covariate was associated with an increase in glucose/insulin/glucagon concentrations with exercise. A negative coefficient indicated that an increase in the covariate was associated with a decrease in glucose/insulin/glucagon concentrations with exercise. Metabolic state and exercise mode were analysed as categorical covariates; % males, age, BMI, \(\dot{V}\)O₂ max, exercise duration and exercise intensity were analysed as continuous covariates. Publication bias was assessed using visual inspection of contour-enhanced funnel plots [33] and statistically by Egger’s regression test. Trim and fill analyses were used when publication bias was suspected to explore its impact on MDs. Statistical significance was set at P < 0.05 in a Z test analysis. Z tests were used to examine if MDs were significantly different from zero. Results are displayed as overall MDs with 95% confidence intervals (CI).

The quality of evidence was assessed using the strategy recommended by the Grading of Recommendations Assessment Development and Evaluation (GRADE) working group [34]. The quality of evidence was assessed using the following domains: risk of bias; inconsistency; indirectness; imprecision; and publication bias. The estimated effect for each outcome was then classified as very low (true effect is probably markedly different from the estimated effect), low (true effect might be markedly different from the estimated effect), moderate (true effect is probably close to the estimated effect) or high quality (true effect is similar to the estimated effect).

A risk of bias summary table is presented in Electronic Supplementary Material Appendix S2.

Most studies measuring glucose (93%) and insulin (97%) concentrations were classified as possessing an unclear risk of bias overall (Electronic Supplementary Material Figure S1a and 1b). All studies measuring glucagon concentrations were classified as having an unclear risk of bias overall (Electronic Supplementary Material Figure S1c).

The effect estimates for insulin, glucagon, and glucose outcomes were all categorised as moderate quality. Insulin, glucagon, and glucose were all downgraded by one level due to inconsistency of results, as the large heterogeneity observed for all three outcomes could not be explained by sub-group analyses or meta-regression. A summary of findings is presented in Table 2.

Overall, acute aerobic exercise appeared to result in no meaningful change in glucose concentrations compared to resting conditions. However, when accounting for the metabolic state (postprandial or fasted), postprandial aerobic exercise resulted in a significant reduction in glucose concentrations. Acute aerobic exercise thus appears to be an effective method to reduce glucose concentrations in the postprandial state. This reduction is likely due to the induction of glucose transporter translocation and glucose transporter activity in skeletal muscle by exercise [69]. The upregulation of glucose transporter translocation and activity may not be secondary to insulin action, as exercise-stimulated glucose uptake has been demonstrated to occur independently of insulin [70], and glucose concentrations decreased in the context of decreasing insulin concentrations during postprandial aerobic exercise. Alternatively, exercise can increase insulin-dependent glucose uptake, possibly via a reduction in intramuscular glycogen and/or increase in AS160 phosphorylation [71]. The decrease in insulin concentrations (despite a reduction in glucose concentrations) may therefore reflect an increase in insulin sensitivity instead. Regardless of the mechanism responsible, this reduction is likely facilitated by the increase in microvascular recruitment and blood flow to skeletal muscle (thus increasing glucose delivery) caused by exercise [70]. The reduction in glucose concentrations is nevertheless small in magnitude and therefore the clinical significance of this finding is questionable. In contrast, no significant change in glucose concentrations during fasted aerobic exercise was detected. This is likely due to glucose concentrations already being low following an overnight fast [23], and that participants were individuals without diabetes, therefore making any further reduction difficult. The overall absence of a large decrease or increase in glucose concentrations does, however, highlight the high degree to which glucose concentrations are homeostatically regulated in non-diabetic populations.

The change in glucose concentrations during acute aerobic exercise was also influenced by exercise modality. Glucose concentrations were significantly reduced following acute aerobic exercise performed on a cycle ergometer, but non-significantly increased following aerobic exercise performed on a treadmill. Differences in glucose concentrations between exercise modalities have been reported previously [72, 73] and likely relate to underlying physiological differences between cycle ergometer and treadmill exercise [74], in particular the differences in carbohydrate metabolism between these exercise modalities [75]. Alternatively, this difference may be an artefact of the proportion of studies conducted in the postprandial and fasted state. Studies using a cycle ergometer were predominantly conducted in the postprandial state (~ 60% postprandial, ~ 40% fasted), whereas studies using a treadmill were predominantly conducted in the fasted state (~ 30% postprandial, ~ 70% fasted).

A higher participant \(\dot{V}\)O₂ max and longer exercise duration were both associated with a smaller decrease in glucose concentrations. \(\dot{V}\)O₂ max is positively correlated with insulin sensitivity [76], and therefore individuals that possess higher \(\dot{V}\)O₂ max values are likely to have lower glucose concentrations in both the fed and fasted state, diminishing the extent to which glucose concentrations can be lowered by an intervention such as acute aerobic exercise. This also suggests that individuals possessing a lower level of cardiorespiratory fitness (and by inference a lower degree of insulin sensitivity) will see greater reductions in glucose concentrations with acute aerobic exercise. This is important considering individuals that possess a low level of cardiorespiratory fitness are at a greater risk of developing type 2 diabetes [77, 78]. The mechanism underlying the effect of exercise duration on glucose concentrations is unclear, but longer exercise durations may provide a larger window for homeostasis to be achieved following the initial disruption by aerobic exercise commencement. Nevertheless, the moderating effect of participant \(\dot{V}\)O₂ max and exercise duration on glucose concentrations appears small when accounting for the magnitude of the overall effect of acute exercise on glucose concentrations.

Acute aerobic exercise resulted in a significant reduction in circulating insulin concentrations relative to resting conditions. This reduction in insulin concentrations may partly reflect the decrease in glucose concentrations observed with postprandial aerobic exercise, which may be due to the stimulation of insulin-independent glucose uptake pathways in skeletal muscle by exercise [13]. Alternatively or additively, the reduction in insulin concentrations with exercise may be caused by an increase in insulin clearance [79], or an increase in insulin delivery (blood flow x blood insulin concentration) as a result of exercise-induced increases in skeletal muscle perfusion, decreasing insulin requirements and thus output [80]. When acute aerobic exercise is performed in the postprandial state, the effect of exercise on insulin concentrations is considerably greater. The present analysis therefore highlights that acute aerobic exercise is a potent tool for reducing postprandial insulin concentrations. In contrast, acute aerobic exercise undertaken in the fasted state resulted in a non-significant reduction in insulin concentrations. Short-term fasting (< 24 h) is well known to decrease insulin concentrations [23] and insulin levels would likely be at their lowest following > 6 h of fasting in non-diabetic individuals. Therefore, aerobic exercise performed in the fasted state is unlikely to prompt further reductions, especially when compared to fasted resting conditions. Changes in insulin concentrations were not significantly moderated by any covariate included in the meta-regression. This could be considered unexpected given that changes in glucose concentrations were moderated by participant \(\dot{V}\)O₂ max and exercise duration. However, this finding is consistent with previous meta-regression analyses investigating the effect of acute exercise on the concentrations of other circulating hormones [81]. This meta-analysis reported no significant effect of sex, participant \(\dot{V}\)O₂ max, exercise duration, or exercise intensity on the acyl-ghrelin or peptide YY response to acute exercise, suggesting that the hormonal milieu in response to acute exercise is comparable across individuals irrespective of key characteristics (e.g. age, sex, exercise intensity).

To our knowledge, this is the first review to quantify the changes in glucagon concentrations during exercise using a meta-analytical approach. The results from our analysis showed that acute aerobic exercise increased glucagon concentrations relative to resting conditions. Importantly, all five studies reported an increase in glucagon concentrations independent of metabolic state. This increase may be necessary to stimulate hepatic gluconeogenesis to provide substrate for contracting muscles and maintain euglycaemia [82], thus facilitating the absence of any large deviations in glucose concentrations. Despite the consistency of the glucagon response to acute aerobic exercise, the findings from this analysis should be treated with caution due to the small number of studies included. Future work should explore the effect of metabolic state (postprandial vs fasted) and exercise modality on changes in glucagon concentrations during acute aerobic exercise considering the limited number of studies currently available.

In addition to its glucoregulatory role, data from animal and human studies demonstrate that glucagon can decrease appetite [83] and therefore may be a key mechanism underlying exercise-induced anorexia [84]. However, glucagon concentrations are rarely measured during acute exercise studies, and consequently there are no reports of its association with appetite during or post-exercise. Future work investigating acute exercise and appetite should look to prioritize glucagon measurements (considering the consistent glucagon response to acute aerobic exercise) to evaluate the role of glucagon in exercise-induced anorexia.

There are several limitations to the present review and meta-analysis. Firstly, the application of these results is restricted to individuals who possess the same characteristics as those defined by the inclusion and exclusion criteria (healthy adults). The glucose, insulin and glucagon response to acute aerobic exercise in other patient populations cannot be assumed from our findings. Likewise, the results cannot be applied to other exercise modalities, such as high-intensity interval training or resistance training. The large heterogeneity observed in all three analyses is another limitation of the current review and meta-analysis but was expected considering the diversity in participant characteristics and experimental methodology used in acute exercise studies. Glucose, insulin and glucagon outcomes were consequently downgraded by one level using the GRADE approach and classified as moderate quality due to the large heterogeneity observed. Meal characteristics (e.g. timing and energy content) were not included in the meta-regression due to concerns regarding overfitting the model and therefore the influence of these properties on the study outcomes is unknown. The present review is also limited by the small number of studies measuring glucagon concentrations, which prevented sub-group meta-analyses and meta-regression from being performed, and thus the investigation of the influence of participant and intervention characteristics on this response. A further limitation is the use of pre- and post-exercise measurements to summarise the effect of acute exercise. While these measures represent the effect of acute exercise on glucose, insulin and glucagon concentrations at the point of exercise completion, they do not account for the temporal changes during exercise. Furthermore, these measures do not consider the effect of acute exercise on glucose, insulin and glucagon concentrations in the post-exercise period. Lastly, the majority of studies measuring glucose and/or insulin and/or glucagon concentrations were classified as having an unclear risk of bias overall. This was largely due to inadequate reporting of the randomisation process. Future investigations in this field should therefore report methodology in sufficient detail as described in the recent Proper Reporting of Evidence in Sport and Exercise Nutrition Trials (PRESENT) checklist proposed by Betts et al. [85].