BDNF indicates brain-derived neurotrophic factor.
Differences in blood levels of brain-derived neurotrophic factor are shown between children with autism spectrum disorder (ASD) and healthy controls. The sizes of the squares are proportional to study weights. Diamond marker indicates pooled effect size.
Pooled results compare blood levels of brain-derived neurotrophic factor between children with autism spectrum disorder (ASD) and healthy controls. Study design is stratified into neonate and nonneonate studies. The sizes of the squares are proportional to study weights. Diamond markers indicate pooled effect sizes.
Pooled results compare blood levels of brain-derived neurotrophic factor between children with autism spectrum disorder (ASD) and healthy controls. Sample source is stratified into serum and plasma samples in the nonneonate group. The sizes of the squares are proportional to study weights. Diamond markers indicate pooled effect sizes.
Publication bias in studies comparing brain-derived neurotrophic factor levels between children with autism spectrum disorder and healthy controls. The plots describe the effect size (Hedges g statisic) of studies against their precision (inverse of SE). Data markers indicate individual studies. Diamond marker indicates pooled effect.
eTable. Characteristics of Included Studies Measuring Peripheral Blood BDNF Concentrations
eFigure 1. Sensitivity Analysis
eFigure 2. Meta-regression in All Studies
eFigure 3. Meta-regression in Nonneonate Group
Customize your JAMA Network experience by selecting one or more topics from the list below.
Qin X, Feng J, Cao C, Wu H, Loh YP, Cheng Y. Association of Peripheral Blood Levels of Brain-Derived Neurotrophic Factor With Autism Spectrum Disorder in Children: A Systematic Review and Meta-analysis. JAMA Pediatr. 2016;170(11):1079–1086. doi:10.1001/jamapediatrics.2016.1626
Are peripheral blood levels of brain-derived neurotrophic factor (BDNF) altered in children with autism spectrum disorder (ASD)?
In this meta-analysis of 19 studies with 2896 unique participants, children with ASD demonstrated significantly increased blood levels of BDNF compared with healthy control children.
This study provides a novel perspective into the etiology of ASD, and future investigations into BDNF levels for early diagnosis of the disease may be warranted.
Accumulating evidence suggests that brain-derived neurotrophic factor (BDNF) may be implicated in the developmental outcomes of children with autism spectrum disorder (ASD).
To use meta-analysis to determine whether children with ASD have altered peripheral blood levels of BDNF.
A systematic search of PubMed, PsycINFO, and Web of Science was performed for English-language literature through February 7, 2016. The search terms included brain-derived neurotrophic factor or BDNF in combination with autism, without year restriction. Two additional records were retrieved after a review of the reference lists of selected articles.
Studies were included if they provided data on peripheral blood levels of BDNF in children with ASD and healthy control children. Studies that included adults or with overlapping samples were excluded.
Data Extraction and Synthesis
Data were extracted by 2 independent observers from 19 included studies. Data were pooled using a random-effects model with Comprehensive Meta-analysis software.
Main Outcomes and Measures
Blood levels of BDNF in children with ASD compared with healthy controls. Altered levels of BDNF were hypothesized to be related to ASD.
This meta-analysis included 19 studies with 2896 unique participants. Random-effects meta-analysis of all 19 studies showed that children with ASD had significantly increased peripheral blood levels of BDNF compared with healthy controls (Hedges g, 0.490; 95% CI, 0.185-0.794; P = .002). Subgroup analyses in 4 studies revealed that neonates diagnosed with ASD later in life had no association with blood levels of BDNF (Hedges g, 0.384; 95% CI, −0.244 to 1.011; P = .23), whereas children in the nonneonate ASD group (15 studies) demonstrated significantly increased BDNF levels compared with healthy controls (Hedges g, 0.524; 95% CI, 0.206 to 0.842; P = .001). Further analysis showed that children in the nonneonate ASD group had increased BDNF levels in serum (10 studies) (Hedges g, 0.564; 95% CI, 0.168 to 0.960; P = .005) but not in plasma (5 studies) (Hedges g, 0.436; 95% CI, −0.176 to 1.048; P = .16). Meta-regression analyses revealed that sample size had a moderating effect on the outcome of the meta-analysis in the nonneonate group. In addition, no publication bias was found in the meta-analysis.
Conclusions and Relevance
Children with ASD have increased peripheral blood levels of BDNF, strengthening the clinical evidence of an abnormal neurotrophic factor profile in this population.
Autism spectrum disorder (ASD) consists of a group of complex neurodevelopmental disorders characterized by impairments in social interaction, verbal and nonverbal communication, and repetitive behaviors.1 According to the DSM-5,2 ASD includes autistic disorder, childhood disintegrative disorder, pervasive developmental disorder not otherwise specified, and Asperger syndrome. According to the Centers for Disease Control and Prevention,3 about 1 in 68 children in the United States have been diagnosed with ASD as of 2014, and the disease is 4 to 5 times more common in male than in female children.4 The prevalence of ASD has increased dramatically during the past several decades, and it affected 21.7 million people globally as of 2013.5 No explanation has been established for this continued increase, although improved diagnosis and environmental influences are considered to be 2 reasons.6
Despite the fact that the research into ASD has greatly increased during the past several decades, the exact etiology of ASD remains poorly understood. However, the complexity of the disease is considered to arise from the interactions between genetic mutations and environmental influences.7 In addition, findings from analyses of postmortem and peripheral tissue and molecular genetic studies led to the hypothesis that neurotrophins—as crucial moderators of neurodevelopment and neuroplasticity—affect the pathophysiologic features of ASD.8-11
A prominent and widely expressed neurotrophin family member is brain-derived neurotrophic factor (BDNF), which plays a key role in neuronal survival and growth, cell differentiation, synapse formation, synaptic plasticity, and cognitive functions.12-17 In addition, skewed BDNF expression has been found to be associated with several neuropsychiatric diseases, including depression,18 schizophrenia,19 and Alzheimer disease.20 In ASD, a substantial number of studies measured peripheral blood levels of BDNF in children with ASD as a window to understand what occurs in the brain because peripheral blood levels of BDNF are known to be highly correlated with BDNF levels in the brain.21,22 However, studies of the association between blood BDNF levels and ASD yielded inconsistent findings. Some studies10,23-25 demonstrated that children with ASD had significantly increased peripheral blood levels of BDNF compared with healthy control individuals, whereas other studies26,27 found no association between blood levels of BDNF and ASD. Furthermore, some groups reported decreased blood levels of BDNF in children with ASD.9,28 Given the inconsistent findings, a meta-analysis on this subject is warranted. In this study, we undertook what is, to our knowledge, the first systematic review with meta-analysis of studies measuring peripheral blood levels of BDNF in children with ASD compared with healthy controls and performed subgroup and meta-regression analyses to adjust for potential confounders affecting the outcome of the meta-analysis.
Two independent investigators (X.-Y.Q. and C.C.) performed a systematic search of peer-reviewed English-language articles using PubMed, PsycINFO, and Web of Science through February 7, 2016. The search terms used for the database search included brain-derived neurotrophic factor or BDNF in combination with autism, without year restriction. Original human studies that reported data on peripheral blood levels of BDNF in patients with ASD and healthy controls were included. Exclusion criteria were (1) adult study participants and (2) samples that overlapped with another study. This meta-analysis adhered to the guidelines recommended by Preferred Reporting Items for Systematic Reviews and Meta-analysis (the PRISMA statement).29
Both investigators independently extracted the data from the included studies. Sample sizes, mean BDNF concentrations, SDs, and P values were extracted as primary outcomes to generate effect sizes. Data on sex distribution, mean patient age, publication year, sample source (serum or plasma sample or dried blood spot), and assay type were also extracted. Demographic and clinical characteristics of the included studies are summarized in the eTable in the Supplement.
All statistical analyses were performed using Comprehensive Meta-analysis software (version 2; Biostat Inc). Most of the effect sizes were generated by sample sizes, mean BDNF concentrations, and SDs; the rest were generated by sample sizes and P values. One study reported P values as an inequality (P < .0001) rather than an exact value, and the P value was rounded down to the nearest .00009 to allow compatibility with the meta-analysis software. Effect sizes were calculated as standardized mean differences in BDNF levels between children with ASD and healthy controls and converted to the Hedges g statistic, which provides an unbiased effect size adjusted for sample size.30 The statistical difference of the pooled effect size was assessed using 95% CIs. We used a random-effects model for the meta-analysis because within-study and between-study confounders were hypothesized to result in a difference of the true effect size.31 We performed sensitivity analysis by removing 1 study at a time to assess whether a single study influenced the outcomes of the meta-analysis.
We used the Cochran Q test32 to assess the statistical difference of heterogeneity among studies, and P < .10 was considered to be statistically significant. The impact of heterogeneity that reflects the inconsistent levels among studies was determined by the I2 index. An I2 index of 0.25, 0.50, and 0.75 would indicate small, moderate, and high levels of heterogeneity, respectively.33 We performed unrestricted maximum-likelihood random-effects meta-regressions of effect sizes34 to evaluate whether the theoretically relevant covariates, including sample size, sex distribution (proportion of male), mean age of the children with ASD, and publication year of each study served as confounders to influence the effect size. Under the random-effects model, study weight is the inverse of the total variance (within-study variance plus between-study variance, τ2 statistic).
Publication bias was first visually inspected by funnel plots generated by plotting the effect sizes against the precision (inverse of SE) for each study. The Egger test35 was assessed to determine the statistical significance of publication bias, which indicates the degree of funnel plot asymmetry. We further investigated publication bias using the classic fail-safe N method,36 which computes the number of missing studies (with a mean effect of zero) that would need to be added to the analysis to yield a statistically nonsignificant overall effect. P < .05 was considered statistically significantly different in all analyses of this study except where noted.
We first performed a systematic search, which produced 166 records from PubMed, 105 records from PsycINFO, 338 records from Web of Science, and 2 additional records identified from the reference lists of retrieved articles. After reading the titles and abstracts, 28 appropriate articles were identified for full-text analysis. We scrutinized the 28 articles and excluded 9 articles for lack of necessary data on BDNF levels,37,38 patient samples that overlapped with another study,39,40 adult study participants,41,42 data reporting BDNF messenger RNA levels,43,44 and no data reported on healthy controls.45 Thus, 19 studies with 2896 unique participants9,10,23-28,46-56 met criteria for inclusion in the present meta-analysis (Figure 1).
We performed a random-effects meta-analysis on the extracted 19 studies encompassing 1411 children with ASD and 1485 healthy controls. The results showed that peripheral blood levels of BDNF were significantly increased in children with ASD when compared with levels in healthy controls (Hedges g = 0.490; 95% CI, 0.185-0.794; P = .002) (Figure 2). Sensitivity analysis indicated that no individual study significantly influenced the significant difference on blood BDNF levels between children with ASD and healthy controls (eFigure 1 in the Supplement). However, significant heterogeneity among studies was found in our meta-analysis (Q18 = 211.853; I2 = 91.504; P < .001).
To investigate the potential sources that explained the high levels of heterogeneity, we performed subgroup analyses. Four of 19 studies25,26,46,49 in the meta-analysis analyzed dried blood spot samples from neonates diagnosed with ASD later in life; the other 15 studies9,10,23,27,28,47,48,50-56 analyzed blood samples (serum or plasma) from children with ASD after they were diagnosed (nonneonate ASD group). Therefore, we performed a meta-analysis based on whether blood samples were obtained from neonates. As shown in Figure 3, the nonneonate ASD group had significantly increased blood BDNF levels compared with healthy controls (Hedges g = 0.524; 95% CI, 0.206-0.842; P = .001). In contrast, neonates diagnosed with ASD later in life did not show a difference in blood BDNF levels compared with healthy controls (Hedges g = 0.384; 95% CI, −0.244 to 1.011; P = .23). However, high levels of heterogeneity among studies were still found in the neonate (Q3 = 57.526; I2 = 94.785; P < .001) and nonneonate (Q14 = 94.338; I2 = 85.160; P < .001) studies, although the heterogeneity was slightly reduced in the nonneonate group. Further analysis of the nonneonate studies showed that children with ASD had significantly increased BDNF levels in serum samples (Figure 4) (10 studies9,10,24,47,51-56) (Hedges g = 0.564; 95% CI, 0.168-0.960; P = .005) but not in plasma samples (5 studies23,27,28,48,50) (Hedges g = 0.436; 95% CI, −0.176 to 1.048; P = .16) compared with healthy controls. Again, we found high levels of heterogeneity among studies measuring serum (Q9 = 58.236; I2 = 84.546; P < .001) or plasma (Q4 = 36.049; I2 = 88.904; P < .001) BDNF levels.
We next performed meta-regression analyses to test whether the continuous variables, including age and sex of the patients, sample size, and publication year of each study, could explain the high levels of heterogeneity among studies. These tested variables did not show moderating effects on the outcome of the meta-analysis (eFigure 2 in the Supplement) in the included 19 studies.9,10,23-28,46-56 Meta-regression in the nonneonate studies showed that sample size positively correlated with the effect sizes (eFigure 3A in the Supplement), indicating that sample size had moderating effects on the outcome of the meta-analysis in the nonneonate studies. In addition, we found a nonsignificant negative correlation between age and effect size in the nonneonate studies (eFigure 3B in the Supplement), suggesting that age might be a confounding factor for studies analyzing BDNF levels in children with ASD. Sex (eFigure 3C in the Supplement) and publication year (eFigure 3D in the Supplement) did not have moderating effects in the nonneonate studies.
Visual inspection of funnel plots (Figure 5) showed no publication bias in this meta-analysis, which was confirmed by results of the Egger test (t17 = 1.64; P = .12). In addition, the classic fail-safe N method indicated that 466 missing studies would be required to make P > .05 for the meta-analysis, which further confirmed that the positive results of our meta-analysis are unlikely to be caused by publication bias.
The strength of this work is that the meta-analysis included a large body of studies and demonstrated that peripheral blood levels of BDNF are significantly increased in children with ASD compared with healthy controls. Through sensitivity analysis, we conclude that our results were not unduly influenced by a particular study. In addition, we found no indication that the outcome of the present meta-analysis was caused by publication bias. To the best of our knowledge, this meta-analysis is the first to be performed on this subject. Because the role of circulating BDNF levels in children with ASD has been controversial for more than a decade, and therefore their significance with respect to the etiology of ASD remains elusive, this study provides clinical evidence that children with ASD have increased peripheral blood levels of BDNF. This finding offers a novel perspective into a potential molecular pathway that contributes to the developmental outcomes of ASD.
Although high levels of between-study heterogeneity were found in the meta-analysis, the strength of this work is that we used subgroup and meta-regression analyses to adjust for potential moderators. Subgroup analyses showed that neonates diagnosed with ASD later in life did not have altered blood BDNF levels, whereas children in the nonneonate ASD group manifested increased peripheral blood levels of BDNF. Although the levels of between-study heterogeneity were only slightly reduced in the nonneonate studies, meta-regression analysis indicated that age might have a moderating effect on the outcome of the meta-analysis, thus raising awareness that future work analyzing BDNF in children with ASD would need to consider this potential confounding factor. Meta-regression analysis also revealed that sample size was positively correlated with the effect size in the nonneonate group, which is reasonable because larger sampling of cases and controls would be a stronger approach for clinical studies.
Most of the studies included in the meta-analysis analyzed serum BDNF levels, which provides us a sufficient number of studies to conclude that children with ASD are accompanied by the increased serum BDNF levels. Evidence suggests that the changes in serum BDNF levels may reflect what occurs in the brain. In particular, reduced serum BDNF concentrations in several neuropsychiatric disorders, such as schizophrenia19 and Alzheimer disease,57,58 were consistently observed by biochemical analyses of BDNF levels in the brains.20,59,60 In addition, a recent meta-analysis demonstrated that serum BDNF levels are low in untreated patients with depression and restored by antidepressant treatments,18 which is also consistent with what occurs in the brain.61,62 Therefore, the increased serum BDNF levels in children with ASD in our meta-analysis very likely reflects what occurs in the brains of children with ASD. In support of this theory, 1 study8 showed that BDNF levels were significantly increased in the postmortem brains from patients with ASD compared with controls.
Notwithstanding its significant strengths, this meta-analysis has some inherent limitations. First, the meta-analysis of peripheral blood levels of BDNF in children with ASD compared with healthy controls provides us a pooled result originating from cross-sectional studies. Therefore, whether an increase in BDNF levels is a cause for ASD development or an epiphenomenon, such as a counterbalance consequence of ASD development, remains unclear. However, the hypothesis that the hyperactivity of BDNF contributes to the development of ASD is plausible in view of the early postnatal neocortical overgrowth that has been proposed as a core mechanism of ASD,63 and BDNF is a well-known potent stimulator of neuronal growth.13 The idea that increased BDNF levels affect the neuropathologic features of ASD is further supported by the fact that children with ASD have excess synapses64 because a major function of BDNF is to enhance synapse formation.12 In addition, in vitro and in vivo studies suggest that BDNF contributes to epileptogenesis.65 Thus, the hyperactivity of BDNF may provide a common molecular pathway for the long-standing association between ASD and epilepsy.66
Second, the numbers of studies that analyzed dried blood spot levels of BDNF from neonates and plasma BDNF levels from nonneonates were small. However, only 1 study25 demonstrated that neonates diagnosed with ASD later in life had increased peripheral blood BDNF levels compared with healthy controls, whereas the other 3 studies26,46,49 did not find any differences. Therefore, we can reasonably conclude that neonates diagnosed with ASD later in life did not have the altered blood levels of BDNF. Although the random-effects meta-analysis did not show a significant association between plasma BDNF levels and children with ASD, after excluding an outlier28 (which had a small sample size) from the 5 studies measuring plasma BDNF levels, the meta-analysis showed that children with ASD had significantly increased plasma levels of BDNF compared with healthy controls (Hedges g = 0.803; 95% CI, 0.398-1.209; P < .001). Therefore, children with ASD likely have increased plasma levels of BDNF, but further studies on plasma BDNF levels in this population are necessary to substantiate this idea.
Third, cross-sectional studies do not indicate the developmental stage at which the manifestation of increased BDNF in children with ASD starts, and longitudinal studies on BDNF levels may be necessary to address this question. Finally, most commercially available kits do not differentiate between pro-BDNF and mature BDNF, as in the studies included herein. One exception is a study that showed that total BDNF levels, but not pro-BDNF levels, were significantly increased in children with ASD.23 Given the proposed opposing effects of pro-BDNF and mature BDNF (a major function of pro-BDNF is generally considered to be induction of apoptosis67), a study on whether the ratio of pro-BDNF to mature BDNF levels is altered in children with ASD would be interesting.
Our meta-analysis demonstrated increased peripheral blood levels of BDNF as a manifestation of ASD in children, which strengthens the clinical evidence of an abnormal neurotrophic factor profile in children with ASD. Thus, further investigations into BDNF levels as a potential key to early diagnosis and therapeutic target of ASD are warranted.
Corresponding Author: Yong Cheng, PhD, Section on Cellular Neurobiology, Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health, 49 Convent Dr, Bldg 49, Room 6C80, Bethesda, MD 20892 (firstname.lastname@example.org).
Accepted for Publication: May 19, 2016.
Published Online: September 19, 2016. doi:10.1001/jamapediatrics.2016.1626
Author Contributions: Drs Feng and Qin contributed equally to this work. Dr Cheng 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.
Study concept and design: Qin, Feng, Cheng.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Feng, Cheng.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Qin, Cheng.
Obtained funding: Qin, Feng, Cheng.
Administrative, technical, or material support: Qin, Cheng.
Study supervision: Cheng.
Conflict of Interest Disclosures: None reported.
Funding/Support: This study was supported by the Intramural Research Program of the Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health; grants YLDX01013 and 2015MDTD13C from the Minzu University 985 Academic Team-building Fund; and grant B08044 from the 111 Project of China.
Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.