The Effect of Exercise Training on Resting Concentrations of Peripheral Brain-Derived Neurotrophic Factor (BDNF): A Meta-Analysis

Background The mechanisms through which physical activity supports healthy brain function remain to be elucidated. One hypothesis suggests that increased brain-derived neurotrophic factor (BDNF) mediates some cognitive and mood benefits. This meta-analysis sought to determine the effect of exercise training on resting concentrations of BDNF in peripheral blood. Methods MEDLINE, Embase, PsycINFO, SPORTDiscus, Rehabilitation & Sports Medicine Source, and CINAHL databases were searched for original, peer-reviewed reports of peripheral blood BDNF concentrations before and after exercise interventions ≥ 2 weeks. Risk of bias was assessed using standardized criteria. Standardized mean differences (SMDs) were generated from random effects models. Risk of publication bias was assessed using funnel plots and Egger’s test. Potential sources of heterogeneity were explored in subgroup analyses. Results In 29 studies that met inclusion criteria, resting concentrations of peripheral blood BDNF were higher after intervention (SMD = 0.39, 95% CI: 0.17–0.60, p < 0.001). Subgroup analyses suggested a significant effect in aerobic (SMD = 0.66, 95% CI: 0.33–0.99, p < 0.001) but not resistance training (SMD = 0.07, 95% CI: -0.15–0.30, p = 0.52) interventions. No significant difference in effect was observed between males and females, nor in serum vs plasma. Conclusion Aerobic but not resistance training interventions increased resting BDNF concentrations in peripheral blood.


Data Sources
The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [73] were followed for this meta-anlysis (S1 Text). English-language literature was searched using MEDLINE, Embase, PsycINFO, SPORTDiscus, Rehabilitation & Sports Medicine Source, and Cumulated Index to Nursing and Allied Health Literature (CINAHL) databases. One non-English language study was included (Koichiro et al. 2015), as a translator was available to extract information from that study. Databases were searched up to February 2016 for original reports of BDNF changes after exercise. A sample search strategy (MEDLINE) is presented in S2 Text. Reference lists of retrieved studies were searched for additional reports.

Study Selection
Inclusion criteria were as follows: 1) measured serum, plasma, or whole blood BDNF concentration; 2) BDNF measured before and after an exercise intervention; 3) intervention ! 2 weeks; 4) exercise intensity ! 50% of peak oxygen uptake (VO 2Peak ), or if exercise intensity was not reported, exercise described as running, cycling, or resistance training. Exclusion criteria were as follows: 1) study included a diseased population (e.g. diabetes, Parkinson's disease, multiple sclerosis, etc.) or a psychiatric population (e.g. depression, schizophrenia, etc.); 2) study population consisted of children below the age of 18; or 3) study had significant co-interventions likely to impact the effect of exercise on BDNF concentrations (e.g. military training [74], restricted sleep, etc.) as these groups have altered BDNF concentrations which may modify the effect of exercise training on BDNF concentrations [33, [75][76][77][78][79]. Some studies that met these initial eligibility criteria were not included in this meta-analysis as the data were not extractable (e.g. standard deviation not reported) or the exercise intensity description implied that the intervention intensity was < 50% VO 2Peak (e.g. yoga or easy walking).

Data Extraction
Two independent raters examined each article for eligibility. Disagreements regarding inclusion were settled by consensus with a third rater. Data on pre-and post-intervention mean BDNF concentrations and standard deviations [picograms/millilitre], population characteristics, exercise intervention characteristics, risk of bias items, and other study details were extracted into a pre-formatted spreadsheet by two raters. Missing data were requested from the corresponding authors. Exercise intensity prescriptions for percentage of maximum heart rate were converted to percentage of maximum VO 2Peak as described by the National Council on Strength & Fitness [80]. In studies with multiple exercise intervention groups, groups were combined for the overall analysis. Studies were additionally categorized by whether the exercise intervention was aerobic or resistance training. In studies with an exercise intervention consisting of both aerobic and resistance training, interventions in which >50% of the time was spent performing aerobic exercise were considered aerobic and vice-versa.

Statistical Analyses
Standardized mean differences (SMD) and 95% confidence intervals (CI) were calculated using random-effects models [81]. SMDs were chosen because of variability in absolute BDNF concentrations between assays used by different laboratories and between measures of BDNF in different components of blood [82]. Random-effects models are preferred if significant heterogeneity is expected, as they account for variable underlying effects in estimates of uncertainty, including both within-and between-study variance. Heterogeneity across studies was summarized by Q statistics calculated in Chi-square analysis and I 2 indices were calculated to investigate inconsistencies among results of the included studies [83]. Heterogeneity was further explored via subgroup analysis. In one study that measured BDNF in both serum and plasma [41], serum measurements were used in all analyses except subgroup analysis of serum versus plasma (in which both measurements were used), as serum BDNF measurements were more common across studies than plasma measurements. Inverse variance-weighted metaregression analyses were used to investigate associations between SMD and population characteristics and intervention characteristics. Risk of publication bias was assessed visually using funnel plots and quantitatively with Egger's test [84]. Study quality was assessed using criteria adapted from the Cochrane Collaboration's Risk of Bias tool as done previously [85]. Analyses were conducted using Review Manager Version 5.3 (Cochrane Collaboration, Oxford, UK) and Stata (Release 14.1; StataCorp, College Station, TX).

Population Characteristics
Twenty-nine studies met inclusion criteria and presented sufficient data to be included in this meta-analysis (Fig 1). Disagreement on whether a study met inclusion criteria arose for one study, and consensus was reached. Reasons for exclusion are shown in Fig 1. 910 participants (61.3% male, mean age 42.2 ± 22.4, mean BMI 25.8 ± 2.3) were included. Included studies ranged in size from 7 to 304 participants (Table 1) completing an exercise intervention. Mean exercise prescriptions were 50.9 ± 16.3 (20-90) minutes for 3.1 ± 1.0 (2-7) sessions per week for 12.4 ± 8.7 (5-52) weeks (Table 2).

Comparison of Pre-and Post-Intervention BDNF Concentration
Resting concentrations of peripheral blood BDNF were significantly higher after an exercise training intervention (Fig 2). Neither a Funnel plot (S1 Fig) nor Egger's test (p = 0.18) revealed significant risk of small-study effects. A Q 2 value of 112.28 and I 2 index of 75% signify considerable heterogeneity and inconsistency, respectively, among included studies. Qualitatively, nine of twenty-nine studies (31.0%) reported a significant increase in resting peripheral BDNF concentrations, no studies reported a significant decrease in resting peripheral BDNF concentrations, and twenty studies (69.0%) reported no significant change. Twenty-eight of the twenty-nine included studies were deemed to have high methodological quality (S1 Table).

Investigations of Heterogeneity
Aerobic Exercise vs Resistance Training. In eighteen studies in which the exercise intervention consisted of entirely or mostly (>50% of the time) of aerobic exercise, a greater increase in resting peripheral BDNF concentration was found (SMD = 0.66, 95% CI: 0.33-0.99, p < 0.001) than that found when all studies were combined. Substantial heterogeneity and inconsistency were present in this subgroup (Q 2 = 86.26, I 2 = 80%). In contrast, in twelve studies in which the exercise intervention consisted entirely or mostly of resistance training, there was no change in resting peripheral BDNF concentration after exercise (SMD = 0.07, 95% CI: -0.15-0.30, p = 0.52, Q 2 = 19.12, I 2 = 42%). The difference in effect size between these two subgroups was significant (Fig 3). Of note, one study [104] included both an aerobic exercise group and a resistance training group and these separate groups were included in their respective subgroups.
Other  which exercise intensity was reported in percentage of VO 2Peak or percentage of maximum heart rate (n = 15), there was no association between exercise intensity and changes in resting BDNF concentration (β = 1.478, p = 0.16, df = 14). The median and modal exercise intervention duration was 12 weeks. Effect sizes in studies with an intervention duration of less than 12 weeks (n = 13) were not significantly different from effect sizes in studies with an intervention duration of 12 weeks or greater (Fig 4). Significant heterogeneity and inconsistency were observed in both subgroups (Fig 4).
Serum vs Plasma. BDNF was measured more commonly in serum than in plasma (Fig 5). Fig 5 shows subgroup differences in change in BDNF concentration after exercise between serum and plasma. The difference between these two subgroups was not significant.
Gender. Nine studies included only men in their study population and four studies consisted entirely of females. Subgroup analysis showed no significant difference between effect sizes in studies that included only males and studies that included only females (Fig 6). Among studies in which the number of male and female participants was reported, no association was observed between percentage of male study participants and changes in resting BDNF concentration (β = -0.070, p = 0.95, df = 25). Age and Body Mass Index (BMI). The mean age of the study population was not associated with change in resting BDNF concentration (β = -1.182, p = 0.25, df = 27). Among studies reporting mean baseline BMI of participants, BMI was not associated with changes in resting BDNF concentration (β = 0.046, p = 0.96, df = 23).

Discussion
Collectively, the evidence suggests an increase in resting peripheral BDNF concentrations after aerobic exercise training interventions but not after resistance training interventions. The increase in resting peripheral BDNF concentrations after an exercise intervention is heterogeneous and was found to have a small effect size [110]. This is important, as an increase in BDNF is proposed to be a mechanism through which physical activity enhances cognition and alleviates psychiatric symptoms [24,25,[111][112][113]. This finding strengthens the evidence that aerobic exercise and resistance training constitute different physiological stimuli with respect to neurotrophin concentrations. For brain health, this finding suggests a basis for possible differential benefits from different exercise modalities. Furthermore, since BDNF is significantly lower in individuals with psychiatric disorders [26, 28-31], an increase in BDNF via exercise may confer clinical benefit by ameliorating this abnormality.
These results are similar to those reported in a previous meta-analysis on the effect of exercise on resting peripheral BDNF concentrations [72]. This current meta-analysis differs from that one due to the inclusion of 21 additional reports that were included due to a different search strategy and a later search period (February 2016). The larger number of included studies allowed for a more robust exploration of heterogeneity across studies and a stronger evaluation of current evidence.
As suggested by high Q 2 and I 2 values, heterogeneity across studies, as opposed to random sampling error, contributed considerably to variability in effect estimates. Inconsistency in effect estimates may be due to differences in study populations, exercise intervention characteristics, measurement techniques, and study quality. Our explorations of heterogeneity suggested that inconsistency was not significantly related to gender or age. This is important as exercise is recommended in elderly to promote brain health [114][115][116]. Similarly, exercise intervention duration, intensity and session time were not associated with degree of change in resting peripheral BDNF concentrations. As such, the current evidence could not determine an ideal exercise prescription for increasing resting peripheral BDNF concentrations.
No significant differences were found between studies measuring BDNF in serum versus plasma, although this subgroup analysis was limited by the relatively small number of studies measuring BDNF in plasma. As BDNF is stored in platelets, which are activated and release their contents in serum, serum concentrations of BDNF are substantially higher than plasma BDNF concentrations [67,117]. Plasma BDNF concentration represents freely-floating, not stored, BDNF and thus may have a different physiological role from serum BDNF [118].
It is likely that there is a combination of different mechanisms through which physical activity confers benefits to neural health [12][13][14]. These may include increases in cerebral blood flow, changes in neuroendocrine responses, changes in endocannabinoid and neurotransmitter release, and structural changes in the central nervous system (CNS) [12,113]. Increased BDNF is one proposed mechanism by physical activity confers these benefits [119]. Increased resting peripheral BDNF concentrations may indicate an increase in central BDNF production [120][121][122][123]. Increased BDNF production in the CNS may result in enhanced synaptogenesis and neuronal survival, resulting in structural changes and enhanced cognition [20,22,[124][125][126][127]. While the findings support the hypothesis that BDNF concentration is increased in the blood after an exercise training intervention, it cannot be inferred that BDNF concentration is increased in the brain. Although one human study suggested an association between peripheral and central BDNF concentrations [121], the evidence in humans is limited. Since the human BBB is structurally and functionally different from those in animal models, we cannot infer from those models that BDNF can cross the human BBB [128,129]. The roles of BDNF in the periphery are not well characterized and may involve regulation of energy homeostasis, modification of insulin activity, and modification of neuronal function in certain neuronal subpopulations in the peripheral nervous system [130][131][132].
Interestingly, a single nucleotide polymorphism (SNP) in the gene encoding BDNF, resulting in an amino acid substitution from valine to methionine, may impact the effect of exercise on BDNF concentrations [37,133]. This SNP is present in approximately 30% of the global population and is associated with altered secretion of BDNF as well as a possible increase in serum BDNF concentration relative to those without the SNP [134][135][136]. The presence of this polymorphism may alter the relationship between fitness and brain outcomes such as cognition, mood, and response to mood treatments [133,137,138]. Few studies have assessed the potential impact of this SNP on the effect of exercise training on peripheral BDNF and brain outcomes. Findings from these reports suggest an increase in BDNF and improvements in memory after exercise only in those without the SNP [37, 133,137]. However, those findings are limited by the small number of studies. Future work assessing the potential impact of this polymorphism on this effect and on the relationship between exercise and brain outcomes is required to determine whether exercise interacts equally with BDNF and brain outcomes in those with and without the polymorphism.
The large number of studies assessing the impact of exercise on resting concentrations of peripheral blood BDNF indicates the importance of this topic and allowed for a robust exploration of this effect. Nevertheless, this meta-analysis was limited by the heterogeneity in study populations and exercise interventions prescribed across different studies. A variety of populations and exercise interventions were included in this report, which we attempted to address via subgroup and meta-regression analyses. In addition, variable adherence to the intervention may have detracted from the quality of evidence.
Overall, this meta-analysis provides evidence for an increase in resting concentrations of peripheral blood BDNF after an exercise intervention. This effect is heterogeneous and may not be present in all individuals, suggesting the importance of further research to elucidate predictors of response. Interestingly, an increase in BDNF concentration may occur after aerobic exercise interventions but not resistance training interventions. Future studies in humans to determine whether peripheral concentrations of BDNF reflect central BDNF concentrations and whether BDNF can cross the human BBB are needed to assess the importance of these findings. In addition, further work to determine if changes in blood BDNF concentrations mediate clinical benefits, such as those on mood, cognition and other psychiatric disorders, would help to determine the usefulness of peripheral BDNF as a putative biomarker.