The effect of physical exercise on circulating brain‐derived neurotrophic factor in healthy subjects: A meta‐analysis of randomized controlled trials

Abstract Objective To investigate how physical exercise (PE) would affect brain‐derived neurotrophic factor (BDNF) in randomized controlled trials (RCTs) of healthy subjects. Methods Seven databases (PubMed, Web of Science, Cochrane, Embase, PsycINFO, CINAHL, SPORTDiscus) were searched for RCTs assessing the effects of PE on serum and/or plasma BDNF until December 18, 2021. Meta‐analysis was performed by random‐effects method with standardized mean difference (SMD) and 95% confidence intervals (CIs). Subgroup analysis and meta‐regression analysis were conducted to investigate the potential source of heterogeneity. Trim and fill method, and leave‐one‐out cross‐validation were conducted. Results Eventually, 21 articles, involving 809 participants, were included in the meta‐analysis. Overall, both acute (5 trials, SMD: 1.20, 95% CI: 0.36 to 2.04, p = .005) and long‐term (17 trials, SMD: 0.68, 95% CI: 0.27 to 1.08, p = .001) PE had significant positive effects on BDNF levels. Via subgroup analysis, studies of long‐term PE with larger sample sizes, female participants, participants older than 60 years, and aerobic exercise contributed to a more pronounced improvement on BDNF levels than that found when all studies were combined. Conclusion Both acute and long‐term PE had significant positive effects on circulating BDNF in healthy subjects. This review suggests that acute exercise and long‐term aerobic exercise are powerful forms of PE to enhance neurotrophic effect, especially for female subjects or subjects over 60 years.


INTRODUCTION
There is a tremendous amount of evidence that physical exercise (PE) can improve neurological function and counteract the risk of dementia (Larson et al., 2006;Voss et al., 2019). Among the potential mechanisms of PE-enhancing cognitive effects, neurotrophic molecules and survival of neurons (Park & Poo, 2013) as well as immunity and tissue repair (Kerschensteiner et al., 1999). BDNF is released by many tissues, including skeletal muscle (J. J. Walsh et al., 2015) in addition to the brain. BDNF plays an essential role in the structure and function of the brain via protecting cells and DNA from damage by down-regulating oxidative stress (Hacioglu et al., 2016), modulating neurogenesis (Brown et al., 2003), promoting axonal and dendritic growth (Gonçalves et al., 2016), and modulating synaptic plasticity (Zenke et al., 2015). The bulk of available evidences have found that BDNF could improve the cognitive ability of both animal (Vaynman et al., 2004) and human models (Leckie et al., 2014). It was also found that BDNF improved cell signal transduction and restored learning and memory through amyloid-independent mechanisms in rodent and primate models of Alzheimer's disease (Nagahara et al., 2009).
Existing randomized controlled trials (RCTs) of humans showed that the effects of PE on BDNF are inconsistent, with some finding increases (Rentería et al., 2020;Schmolesky et al., 2013) in BDNF after PE, while most of the others reporting no change (Arrieta et al., 2020;Baird et al., 2018;Forti et al., 2014) in circulating BDNF. This variability may be due to differences in dose parameters, such as type, intensity, and duration of PE. Meta-analysis (Dinoff et al., 2017) based on pre-post design showed that an acute bout of PE increases circulating BDNF transiently, while the effect of long-term PE on neurotrophic molecules is still uncertain. Previous meta-analyses (Dinoff et al., 2016;Szuhany et al., 2015) on the effect of exercise training on resting concentrations of BDNF in humans found that regular exercise training could enhance the response of BDNF to acute PE. Likewise, their analysis also focused only on estimating the association between exercise and BDNF concentrations through the change of BDNF levels from preexercise to post-exercise (non-RCT), which is inefficient in clarifying the actual effect. In addition, one (Szuhany et al., 2015) of these two meta-analyses included people with diseases that are known to have lower basal BDNF (i.e., Parkinson's disease, obesity, and metabolic syndrome). A recent review by E. I. Walsh et al. (2020) concluded that high-intensity short-term activities might effectively promote BDNF response but specific PE type and dose for optimal BDNF release is unclear. Consequently, in the changing context of gradual decline of physical and cognitive abilities in the normal aging process, the magnitude of the actual effects of PE on peripheral BDNF concentrations is still uncertain.
Considering the extensive attention of PE on BDNF and cognition, it is of great significance to collect the existing RCTs for a comprehensive meta-analysis to determine: (1) the specific role of PE on BDNF under physiological condition for healthy subjects and (2) how do training protocols and characteristics of subjects influence the outcomes.

Literature search
The systematic search was carried out following the guidelines of Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) in this meta-analysis (Moher et al., 2009 ("Brain-Derived Neurotrophic Factor" OR BDNF) AND "randomized controlled trial." Detailed search strategy was shown in Table S1.

Study selection
Two researchers (YHW and HHZ) independently screened titles and abstracts, then reviewed full-text for eligibility. The third researcher (SDC) arbitrated any discrepancies to reach consensus. We also con-

Data extraction and quality assessment
The study selection, data extraction, and quality assessment were undertaken independently by two investigators (YHW and HHZ) with standard form. Eligible studies were reviewed and the following data were extracted: The first author's surname, publication year, study design, study location, sample size, participants age and gender, baseline body mass index (BMI) of participants, PE intervention (duration, type, intensity, frequency), and reported circulating BDNF levels. If one study contained two or more independent intervention strata (e.g., different types, intensity, frequency or duration of PE), it was treated as separate trials for analysis. All types of PE (i.e., aerobic exercise, resistance exercise, and multicomponent exercise) were included in this review. Intensity of exercise was classified by maximal heart rate (HRmax), maximal oxygen uptake (VO 2 peak), and repetition maximum (RM) according to American College of Sports Medicine (Haskell et al., 2007). Advanced data extraction was performed using Adobe Photoshop for studies that did not directly provide data but present their data in a graphic format, according to the protocol proposed by Gheibi et al. (2019). The methodological quality of selected studies was evaluated by the PEDro scale (Maher et al., 2003). The Grading of Recommendation, Assessment, Development and Evaluation (GRADE) system was used to assess the evidence level of each outcome (Goldet & Howick, 2013). According to the specific regulations of GRADE guidelines, study design dictates baseline quality of the evidence (RCTs are initially defined as high quality) but other factors could decrease (e.g., unexplained heterogeneity) or increase (e.g., a large magnitude of effect) the quality level (Goldet & Howick, 2013).
Discrepancies were resolved by discussing with the third reviewer (SDC) to reach consensus.

Statistical analysis
The pooled effect sizes were defined as the standardized mean difference (SMD) with 95% confidence intervals (CIs) of net changes of the concentrations of BDNF. The heterogeneity among studies was evaluated using I 2 and Cochrane's Q test and there is heterogeneity when I 2 > 50% and p-value < .1 for Q test (Higgins et al., 2003). Considering the existing heterogeneity between studies, the random effects model was used in pooling estimates of net changes. For studies in which either baseline or final mean and standard deviation (SD) of outcomes were not provided directly, advanced data extraction using the reported method proposed by Wan et al. (2014) was conducted.
Firstly, a primary meta-analysis was conducted to establish the overall effect. Then subgroup analyses were performed to investigate the potential source of heterogeneity based on the sample size, region, gender, age, baseline BMI of participants, duration of intervention, and type and intensity of exercise (only conducted if more than five trials reported the same outcomes). Differences between groups and sources of heterogeneity were tested by meta-regression analysis, with p-value < .1 as statistically significant.
Both Begg's and Egger's regression tests as well as funnel plots were utilized to assess the publication bias, with a p-value < .1 suggesting the presence of bias (Egger et al., 1997). If publication bias was encountered, the trim and fill method was performed (Duval & Tweedie, 2000). Sensitivity analysis using leave-one-out method was performed to investigate key studies that have substantial impact on the heterogeneity between studies (Serban et al., 2015), using p < .1 as the criterion. All analyses were performed using STATA version 11.0 (Stata Corp, College Station, TX, USA), with double data input to avoid input errors. p < .05 was deemed as statistically significant unless specified elsewhere.

Flow of study selection
The detailed flowchart of literature search and study selection is presented in Figure 1

Qualities of included studies and outcome measure evidences by GRADE
The methodological qualities of one study (Matura et al., 2017) was rated as excellent quality according to the PEDro scale (scores ≥ 9), while the remaining 20 studies were good quality (6-8 scores) mostly because of lacking blindness (Table 1). The evidence quality of acute PE on BDNF levels was at a moderate level, while the quality of long-term PE on BDNF levels was at a low level according to the GRADE system (Table S2). Abbreviations: BDNF, brain-derived neurotrophic factor; HRmax, maximal heart rate; kcal, kilocalories; km, kilometer; min, minute; RC, randomized crossover; RM, repetition maximum; RP, randomized-parallel; s, second; Sb, single blind; SB, single-blinded; VO2max, maximal oxygen uptake; wk, week.

Study
a Better methodological quality is indicated by a higher PEDro score (9-10: excellent; 6-8: good; 4-5: fair; < 4: poor). b The study examined both acute and chronic effects of PE on BDNF levels. Note: p 1 value for net change; p 2 value for heterogeneity in the subgroup; p 3 value for heterogeneity between groups with meta-regression, analyzed as categorical variables; p 4 value for publication bias (conducted only when trials > 5); significant p-values are highlighted in bold prints. Abbreviations: BDNF, brain-derived neurotrophic factor; BMI, body mass index.

F I G U R E 3
The forest plot of long-term PE intervention on circulating BDNF levels

3.5.2
The results of meta-regression Meta-regression analysis was conducted to explore the potential sources of heterogeneity. Among selected covariates, including sample size, region, gender, age, baseline BMI, duration of PE, type of PE, and intensity of PE, the results of meta-regression analysis revealed that gender of participants was a potential confounder of the effect of longterm PE intervention on the BDNF levels, with adjusted R 2 of 59.01% (p = .081) ( Table 2).

Publication bias
Publication bias was suggested by Begg's test (p = .030), but not by Egger's test (p = .473) in the primary meta-analysis (Table 2). Evidences also showed publication bias in the subgroup results of sample size > 20, age ≥ 60 years, baseline BMI ≥ 25, and aerobic exercise ( Table 2). As shown in Table 3, for those results with publication bias indicated by Begg's and Egger's tests, the pooling estimates were recalculated using Duval and Tweedie's trim and fill method. The results of meta-analyses remained unchanged or still statistically significant after being adjusted by trim and fill method, which confirmed and strengthened the evidence-base regarding the effects of long-term PE on BDNF (Table 3).

DISCUSSION
This is the first meta-analysis based on RCTs but not pre-post design to estimate the effect of PE (both acute and long-term) on the levels of circulating BDNF among healthy subjects. Overall, our present results mainly revealed that both acute and long-term PE significantly increased BDNF levels, although study heterogeneity was very high for both acute and long-term effects. Furthermore, studies of long-term PE with larger sample sizes, female participants, participants older than 60 years, and aerobic exercise contributed to a more pronounced improvement on BDNF levels than that found when all studies were combined.
Acute PE is an effective stimulating factor to increase peripheral BDNF. A bulk of available evidence has reported that acute PE is associated with increased circulating BDNF (Dinoff et al., 2017;Szuhany et al., 2015). H. C. Cho et al. (2012) reported that progressive, maximum intensity treadmill exercise increased the entire peripheral BDNF levels including serum, plasma, and platelets. The release of BDNF from platelets can be changed by allergic airway inflammation (Lommatzsch, Schloetcke, et al., 2005), so acute PE may affect the release of BDNF from platelets by stimulating inflammatory response (Scheffer & Latini, 2020). Moreover, the transient BDNF response to acute exercise may

F I G U R E 4
The subgroup analysis of long-term PE intervention on circulating BDNF levels stratified by sample size (A), gender (B), age (C), and baseline BMI (D)

F I G U R E 5
The subgroup analysis of long-term PE intervention on circulating BDNF levels stratified by duration of exercise (A) and type of exercise (B) induce a series of neuronal responses to improve cognitive function (Bechara, Lyne, & Kelly, 2014).
Overall, and in accordance with other studies (Dinoff et al., 2016;Szuhany et al., 2015), we observed a significant elevated effect of longterm PE on BDNF levels. One of the possible mechanisms by which long-term PE induced brain plasticity and cognitive enhancement is via stimulating an increase in the concentration of BDNF (Gligoroska & Manchevska, 2012). Usually, 99% of BDNF in circulation binds to platelets (E. I. Walsh et al., 2020), which are stored in the spleen for later release. After BDNF is released into plasma, it can bind with specific neural receptors (E. I. Walsh et al., 2020). PE can stimulate the release of BDNF from brain, skeletal muscle, platelets, and other tissues (J. J. Walsh & Tschakovsky, 2018), mainly by increasing blood circulation throughout the body and the release of platelets from the spleen (E. I. Walsh et al., 2020). Due to differences in the sample size of the included studies (ranged from 12 to 88), we stratified the results according to the sample size. Our subgroup analysis revealed that only studies with larger sample sizes displayed significant elevation in BDNF levels post long-term PE. Small sample size studies may have sampling error and instability, and are more likely to draw false negative conclusions, namely type II error (Akobeng, 2016). However, after exclud- ing studies with small sample sizes, there are still high heterogeneity and publication bias among large sample studies. However, the results of subgroup analysis did not change after adjustment for publication bias by trim and fill method. Notably, through leave-one-out analysis, we found that 6 of the 7 literatures contributing most to heterogeneity were from large-sample studies. These indicated that future largescale and well-designed RCTs are still required to further examine our main findings.
In our study, we found that women were more likely to benefit from long-term PE through subgroup analysis and meta-regression. However, previous meta-analysis (Szuhany et al., 2015) reported that the effect of long-term regular exercise on peripheral BDNF levels was negatively correlated to the proportion of women included in the studies. Meanwhile, we also found that the elevation effect of PE on BDNF was only reflected in those older than 60 years. There is no doubt that gender and age are two important factors affecting the levels of BDNF. In addition, women and the elderly are also high-risk groups of Alzheimer's disease and other diseases related to abnormal BDNF levels (Beam et al., 2018;Riedel et al., 2016). Weisbrod et al. (2019) also found that BDNF levels decreased in female rats after exposure to stress, but not in male mice. Our meta-analysis could suggest that long-term PE intervention might be effective for improving BDNF levels for people over 60-years-old and the female population. However, Abbreviations: BDNF, brain-derived neurotrophic factor; BMI, body mass index; CI, confidence interval.

SMD (95% CI) p
more females than males (n = 112/80) and more people over 60years-old than under 60-years-old (n = 505/168) were included in the analyses, so future studies with equal sex and age ratios need to be replicated.
Based on current evidences, aerobic exercise has been proved to be successful in improving circulating BDNF (Cassilhas et al., 2012;Dinoff et al., 2016), while strength training seems to be mostly ineffective (Huang et al., 2014;Knaepen et al., 2010). A meta-analysis also showed that aerobic exercise may contribute to increased levels of BDNF in neurological populations (Mackay et al., 2017). Consistent with previous studies, we also found that aerobic exercise, but not resistance training, increased circulating BDNF. Aerobic exercise is related to the improvement of endothelial function, insulin resistance, metabolic function, and cerebral blood flow, which are all associated with the increase of BDNF (Lemos et al., 2016;Zembron-Lacny et al., 2016).
In our current study, we found that intensity of exercise did not influence the levels of BDNF. Most of studies admitted that the intensity of exercise is positively correlated with the increase of BDNF circulation levels. A systematic review reported that BDNF levels increase in an intensity-dependent manner (Knaepen et al., 2010). Higher intensity of exercise is associated with hyperthermia, splenic response (Brunelli et al., 2012;Stewart et al., 2003), increased blood-brain barrier permeability (Roh et al., 2017), and hypoxia. These are all related to the increase of BDNF release. Soya et al. found that acute treadmill running at low intensity (15 m/min) increased BDNF levels in the hippocampus of Wistar rats, but no increase was observed at moderate intensity (20 or 25 m/min) (Soya et al., 2007). Similarly, Gilder et al. (2014) found that serum BDNF increased by approximately 48% at 78% VO 2 max, but decreased at maximal exertion trial in healthy young men. Hence, the relationship between intensity of exercise and BDNF levels needs to be further investigated as a key point in clarifying the effect of PE on BDNF.

Limitations
There exist several limitations in this study. Firstly, 9 of the 21 included studies did not claim to adopt blinding, which resulted in the possibility of bias. The heterogeneity across studies and the limited sample size in some of the subgroups also made the interpretation of the results requiring to be cautious. In addition, some results showed publication bias, which might threaten the validity and interpretation of the effect; however, most original analyses remained unchanged after adjustment via trim and fill analysis. Lastly, only three included studies (Matura et al., 2017;Nilsson et al., 2020;Seifert et al., 2010) reported the results of fitness levels, making it impossible to explore the relationship between fitness levels and BDNF levels.

CONCLUSION
Taken together, both acute and long-term PE significantly elevated circulating BDNF levels in healthy subjects. Long-term aerobic exercise can lead to a more pronounced neurotrophic effect especially for female subjects or subjects over 60 years. Future large-scale and highquality RCTs focusing on more detailed divisions of PE prescriptions and the importance of carefully considering the physiological response to PE will be of great necessity.

ACKNOWLEDGMENT
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

CONFLICT OF INTEREST
The authors declare that they have no competing interests.

AUTHOR CONTRIBUTIONS
All authors have contributed to the work in a meaningful way. Ya-Hai Wang, Huan-Huan Zhou, and Sidong Cui were involved in the conceptualization and design of the methodology. Ya-Hai Wang and Huan-Huan Zhou conducted the research, analyzed the data. Ya-Hai Wang, Huan-Huan Zhou, and Sidong Cui wrote the initial draft. Qiang Luo and Sidong Cui critically reviewed the manuscript. All authors agree with publication of the final version of the manuscript.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1002/brb3.2544