BDNF Val66Met and Cognition: All, None, or Some? A Meta-Analysis of the Genetic Association

Abstract

The Val66Met, G196A (rs6265) polymorphism in the brain-derived neurotrophic factor gene, BDNF, located at 11p13, has been associated with a wide range of cognitive functions. Yet, the pattern of results is complex and conflicting. In the current study we conducted a meta-analysis that included 23 publications containing 31 independent samples comprised of 7,095 individuals. The phenotypes that were examined in this analysis covered a wide variety of cognitive functions and included indicators of general cognitive ability, memory, executive function, visual processing skills, and cognitive fluency. The meta-analysis did not establish significant genetic associations between the Val66Met polymorphism and any of the phenotypes that were included.

Introduction

The last decade has yielded a myriad of studies attempting to investigate connections between specific genes (or, more often, specific genetic variants) and complex behavior traits. Five characteristics of these types of studies appear to be common. First, the results of such studies are often inconsistent, which raises the question of the replicability of these results. Second, individually, they are typically characterized by relatively small sample sizes. Third, there is a noticeable variability in the phenotypes utilized in this research. Fourth, there is relatively little ethnic diversity among the studied sample, with the overwhelming majority of them being sampled from populations of European descent. Finally, the field has been focused on a relatively small number of genes (or variants within these genes), whose proteins appear to be functionally highly important and highly pleiotropic, thus generating voluminous research depicting complex connections between these genes/variants and a multiple behavior traits.

One such prime candidate gene that, due its function, has generated much attention in the field is the gene encoding the brain-derived neurotrophic factor (the BDNF gene), located at 11p13 (Houlihan et al. 2009). BDNF belongs to the neurotrophin superfamily (Matsuo et al. 2009), members of which include growth factors that play a role in the proliferation, differentiation, and fate of neuronal cells and, consequently, in the regulation of synaptic plasticity and connectivity in the brain. BDNF is highly expressed in the central nervous system and is known to educe a plethora of functions, particularly in the hippocampus (Lau et al. 2010). It has also been demonstrated that BDNF impacts the function of the key neurotransmitter systems: dopaminergic (Guillin et al. 2001), glutamatergic (Carvalho et al. 2008), and serotonergic (Mossner et al. 2000). Aberrations in BDNF function are associated with numerous neuropsychiatric disorders, although these associations are not consistent (Gong et al. 2009; Strauss et al. 2004; Verhagen et al. 2008, 2010). Moreover, variation in the BDNF genes has been studied as a source of individual differences in such traits as brain structure (Karnik et al. 2010) and function—e.g., as ascertained through EEG (Beste et al. 2010) and fMRI (Soliman et al. 2010)—intelligence (Hansell et al. 2007), and personality (Lang et al. 2005).

BDNF is synthesized as a precursor protein (proBDNF) that is proteolytically cleaved to produce mature BDNF (Greenberg et al. 2009). These two different types of BDNF preferentially bind to functionally different groups of receptors (Lee et al. 2001; Teng et al. 2005); proBDNF is bound by p75^NTR—a member of the tumor necrosis factor receptor family, and mature BNDF is bound by the TrkB receptor tyrosine kinase. The former triggers biochemical reaction leading to apoptosis, axonal retraction, and pruning dendritic spines (Roux & Barker 2002); the latter affects the cell-cycle, neurite outgrowth, and synaptic plasticity (Chao et al. 2006; Cowley et al. 1994; Mazzucchelli et al. 2002; Rosenblum et al. 2002).

The human BDNF gene is complex; it contains 10 exons that are alternatively spliced encoding for >30 mRNA transcripts (Pruunsild et al. 2007). BDNF is differentially expressed in diverse tissues and, within the brain, in various regions (Bishop et al. 1994; Timmusk et al. 1993). Moreover, transcript expression can be altered by environmental factors, such as stress (Fuchikami et al. 2009; Marmigere et al. 2003), so that the ratio of pro/mature BDNF may change (Tognoli et al. 2010). In addition to multiple isoforms, the gene has numerous polymorphisms. One of these polymorphisms is the G196A polymorphism (rs6265) in the pro-region of BDNF resulting in a substitution of a valine for methionine residue at position 66 (Val66Met). It has been reported as having a variable frequency of the derivative (A) allele ranging from 0 or near 0 in Africa to up to 60% in Asia, and a frequency of 17% in Caucasians (www.ncbi.nlm.nih.gov). This polymorphism modifies the intracellular packaging of pro-BDNF and impacts activity-dependent secretion of the mature BDNF (Chen et al. 2004). The molecular and cellular characteristics of this polymorphism have been investigated in a number of in vitro and in vivo model systems, demonstrating that the Val66Met differentially impacts BDNF availability, neuronal survival and morphology, and altered neuronal function (Frielingsdorf et al. 2010). In humans, this polymorphism has also been extensively studied in association with a variety of cognitive traits in a number of different typical and disordered samples. It is the purpose of this article to review the findings obtained in association studies of the Val66Met and human complex cognitive traits and subject them to a meta-analysis.

Materials and Methods

Literature Search

The primary literature search was conducted up to the end of December 2010 using the Medline, PsycINFO, and Academic Search Premier research databases. The search terms that were used were “BDNF”, “brain-derived neurotrophic factor,” “intelligence,” “cognition,” and “memory” in various combinations, and using a number of the search options in the aforementioned databases.

Selection of Literature and Data Extraction

The retrieved literature was examined for extractable data. The criteria for inclusion in the analysis were that the articles contain genotypic information as well as the means and standard deviations of various cognitive measures used in the published reports. Also extracted were the participants’ country of origin, the number of participants in each study, the proportion of male and female subjects, the clinical status (and diagnosis) of the participants, and the mean age of the participants. If an article did not have extractable data, the authors were contacted and this information was requested; all but one author released their data in a format that allowed for its inclusion in the analysis. There were 23 articles with 31 independent samples; all these publications were included in the meta-analysis. When articles contained more than one sample, each sample was treated as an independent sample. Similarly, when articles contained reports on more than one phenotypic category (see below), the indicators were treated as independent. Papers that included unknown cognitive tasks or data in formats other than means and standard deviations were excluded.

Phenotypic Categories

After the extraction of the data was completed, all of the measures used in the literature were presented to a group of clinical and developmental psychologists (n ~ 20). These psychologists grouped all of the measures and reached consensus on the groupings, which resulted in the establishment of the 5 phenotypic categories. This process was completed twice to ensure reliability. The agreement between the first and second sorting was over 90%. The resulting phenotypic categories were general cognitive ability, memory, executive function, visual processing skills, and cognitive fluency. Each of these phenotypic categories included a variety of assessments. A full listing of these assessments can be found in the legends for Figures 1–5. Only one assessment indicator from each sample was included in the analysis of each phenotypic category. When there was more than one indicator reported that fit into a phenotypic category, the phenotype that appeared most frequently in the literature was selected.

Figure 1.

Indicators of General Cognitive Ability

Note:

Val: Val/Val homozygotes, Met: Met allele carriers

Boxes: the effect estimates, the size of the boxes represents the weighting of the study in the analysis

Lines: the confidence interval for the finding of the study

Arrow: the findings are beyond the scale set for the figure

Diamond: the overall effect of the meta-analysis

Unspecified IQ test:

• 1.a, 1.b, 1.c (Egan et al. 2003)

The Wechsler Adult Intelligence Scale (WAIS):

• 2 (Tsai et al. 2004); 3 (Oroszi et al. 2006); 4.a,4.b (Ho et al. 2006); 5.a, 5.b (Cerasa et al. 2010); 6.a, 6.b (Chung et al. 2010)

The Wechsler Intelligence Scale for Children (WISC)

• 7 Verbal IQ (Strauss et al. 2004)

Test of Nonverbal Intelligence

• 8 (Echeverria et al. 2005)

Kaufman Brief Intelligence Test

• 9 (Erickson et al. 2008)

Heim Intelligence Test

• 10 (Miyajima et al. 2008a)

Estimated IQ (by Spot Word Test)

• 11 (Schofield et al. 2009)

Cognitive Abilities Screening Instrument, Chinese Version

• 12 (Tsai et al. 2008)

g-factor

• 13(Houlihan et al. 2009)

Mini Mental State Examination

• 14 (Nacmias et al. 2004); 15.a, 15.b (Yu et al. 2008)

Problem Solving

• 16 (Ho et al. 2007)

National Adult Reading Test

• 17 (Baig et al. 2010)

Figure 5.

Indicators of Cognitive Fluency

Note:

Val: Val/Val homozygotes, Met: Met allele carriers

Boxes: the effect estimates, the size of the boxes represents the weighting of the study in the analysis

Lines: the confidence interval for the finding of the study

Arrow: the findings are beyond the scale set for the figure

Diamond: the overall effect of the meta-analysis

Letter Fluency

• 1 (Nacmias et al. 2004); 2 (Miyajima et al. 2008a)

Vocabulary/Verbal Fluency

• 3 (Echeverria et al. 2005); 4 (Harris et al. 2006); 5 (Houlihan et al. 2009)

List-generating Fluency

• 6 (Tsai et al. 2008)

Category Fluency

• 7.a, 7.b (Yu et al. 2008)

Analyses

Genotypes were grouped according to the presence or absence of the Met allele (Val/Val vs. Val/Met or Met/Met). All of the analyses were modeled after published work (Barnett et al. 2008) and conducted using Comprehensive Meta-Analyses Software Comprehensive Meta-Analysis Version 2.0 (Biostat Inc., Englewood, New Jersey). Specifically, the data were analyzed using fixed or random effects approaches based on the results of the homogeneity test. If there was no evidence of heterogeneity, an assumption was made that the effect of genotype was constant across studies and whatever variance was encountered that differentiated studies was attributed to chance or random variation. If there was evidence of heterogeneity, it was assumed that the variance that differentiated the samples was due not only to chance or random variation, but also to individual study effects. Effect sizes (Cohen’s d, calculated through mean differences) for each sample and each phenotype were pooled to generate summative effect sizes and 95% confidence intervals (95% CI); the significance of these summative effect sizes was established via a z test. A spreadsheet with all of the data that was included in the analysis is available from the authors. In addition, sensitivity and publication bias analyses were conducted. The sensitivity analysis examines whether the inclusion of a particular study has an effect on the final results of the meta-analysis. This analysis reruns the entire analysis k-times excluding one study on each pass to be able to assess the contribution of the particular study. Funnel plots were produced to examine the possibility of publication bias. Publication bias which can skew the results of the analysis, can occur because of the nature of the literature to preferentially publish significant rather than null results.

Results

The results of the analyses are presented below, in Table 2, and in Figures 1–5 (these figures are illustrating the standard difference in means and 95% confidence intervals for the difference).

Table 2.

Results of the Meta-Analysis

		Cohen's D Effect size and 95% CI			Test of null (2-Tail)		Heterogeneity
	N	Point estimate	Lower limit	Upper limit	Z- value	P-value	Q-value	P-value
General Cognitive Ability	23	0.023	−0.040	0.085	0.714	0.475	33.354	0.057
Memory	22	0.047	−0.016	0.110	1.460	0.144	39.964	0.008
Executive Functioning	17	0.030	−0.071	0.130	0.578	0.563	32.051	0.010
Visual	15	0.019	−0.071	0.109	0.406	0.685	18.728	0.176
Cognitive Fluency	8	0.054	−0.024	0.133	1.355	0.175	13.340	0.064

Note:

N=number of studies analyzed

Effect-size Analyses

General Cognitive Ability

Seventeen published reports with 23 independent samples were included in the analysis (Baig et al. 2010; Cerasa et al. 2010; Chung et al. 2010; Echeverria et al. 2005; Egan et al. 2003; Erickson et al. 2008; Ho et al. 2007; Houlihan et al. 2009; Miyajima et al. 2008a; Nacmias et al. 2004; Oroszi et al. 2006; Schofield et al. 2009; Strauss et al. 2004; Tsai et al. 2008; Tsai et al. 2004; Yu et al. 2008). The summative result showed no association between Val66Met and general cognitive ability (d=.023, 95% CI .040-.085, z=.714, p=.475; see Table 2 and Figure 1).

Memory

Fifteen published reports with 22 independent samples were included in the analysis (Cathomas et al. 2010; Cerasa et al. 2010; Chung et al. 2010; Echeverria et al. 2005; Egan et al. 2003; Harris et al. 2006; Ho et al. 2007; Ho et al. 2006; Houlihan et al. 2009; Nacmias et al. 2004; Oroszi et al. 2006; Strauss et al. 2004; Tan et al. 2005; Tsai et al. 2008; Yu et al. 2008). The meta-analysis revealed no association between indicators of memory and the Val66Met polymorphism (d=.047, 95% CI .016-.110, z= 1.460, p=0.144; see Table 2 and Figure 2).

Figure 2.

Indicators of Memory

Note:

Val: Val/Val homozygotes, Met: Met allele carriers

Boxes: the effect estimates, the size of the boxes represents the weighting of the study in the analysis

Lines: the confidence interval for the finding of the study

Arrow: the findings are beyond the scale set for the figure

Diamond: the overall effect of the meta-analysis

Working Memory

• 1.a, 1.b, 1.c (Egan et al. 2003)

Wechsler Memory Scale

• 2 (Strauss et al. 2004); 3 (Tan et al. 2005); 4 (Harris et al. 2006); 5.a, 5.b (Ho et al. 2006)

Digit Span

• 6 (Nacmias et al. 2004) 7 (Echeverria et al. 2005);8.a, 8.b (Yu et al. 2008); 9 (Houlihan et al. 2009); 10.a, 10.b (Cerasa et al. 2010)

Verbal memory

• 11 (Ho et al. 2007)

Delayed free recall

• 12 (Oroszi et al. 2006)

Rey-Kim Assessment (a Korean memory assessment)

• 13.a, 13.b (Chung et al. 2010)

Short delay

• 14 (Cathomas et al. 2010),

Long delay

• 15 (Tsai et al. 2008); 16 (Cathomas et al. 2010)

Executive Functioning

Eleven published reports with 17 independent samples were included in the analysis (Cerasa et al. 2010; Chung et al. 2010; Echeverria et al. 2005; Ho et al. 2007; Ho et al. 2006; Nacmias et al. 2004; Oroszi et al. 2006; Rybakowski et al. 2003; Rybakowski et al. 2006; Tsai et al. 2008; Yu et al. 2008). The results revealed a suggestive association, with p = .07 (d=.030, 95% CI .071-.130, z=.578, p=.563) (see Table 2 and Figure 3).

Figure 3.

Indicators of Executive Functioning

Note:

Val: Val/Val homozygotes, Met: Met allele carriers

Boxes: the effect estimates, the size of the boxes represents the weighting of the study in the analysis

Lines: the confidence interval for the finding of the study

Arrow: the findings are beyond the scale set for the figure

Diamond: the overall effect of the meta-analysis

Wisconsin Card Storing Task

• 1 (Rybakowski et al. 2003); 2.a, 2.b, 2.c (Rybakowski et al. 2006); 3.a, 3.b (Cerasa et al. 2010); 4.a, 4.b (Chung et al. 2010)

Trails

• 5 (Echeverria et al. 2005); 6 (Oroszi et al. 2006); 7.a, 7.b (Yu et al. 2008)

Attention

• 8.a, 8.b (Ho et al. 2006); 9 (Ho et al. 2007); 10 (Tsai et al. 2008)

Random letter test

• 11 (Miyajima et al. 2008a)

Information-Memory-Concentration Test

• 12 (Nacmias et al. 2004)

Visual

Eleven published reports with 15 independent samples were analyzed (Cerasa et al. 2010; Chung et al. 2010; Echeverria et al. 2005; Hansell et al. 2007; Ho et al. 2007; Ho et al. 2006; Nacmias et al. 2004; Oroszi et al. 2006; Strauss et al. 2004; Tsai et al. 2008; Yu et al. 2008). No association was detected (d=.019, 95% CI .071-.109, z=.406, p=.685) (see Table 2 and Figure 4).

Figure 4.

Indicators of Visual Phenotype

Note:

Val: Val/Val homozygotes, Met: Met allele carriers

Boxes: the effect estimates, the size of the boxes represents the weighting of the study in the analysis

Lines: the confidence interval for the finding of the study

Arrow: the findings are beyond the scale set for the figure

Diamond: the overall effect of the meta-analysis

Rey-Osterrieth Complex Figure Test

• 1 (Strauss et al. 2004); 2 (Oroszi et al. 2006); 3.a, 3.b (Ho et al. 2006); 4 (Ho et al. 2007); 5.a, 5.b (Yu et al. 2008); 6.a, 6.b (Cerasa et al. 2010); 7.a, 7.b (Chung et al. 2010)

Copying Drawing

• 8 (Nacmias et al. 2004)

Visual Reproduction/Construction

• 9 (Echeverria et al. 2005); 10 (Tsai et al. 2008)

Trials correct (i.e., correct identification of a visual target in a computerized visuo-spatial delayed-response task)

• 11(Hansell et al. 2007)

Cognitive Fluency

Seven published reports with 8 independent samples were included in the analysis (Echeverria et al. 2005; Harris et al. 2006; Houlihan et al. 2009; Miyajima et al. 2008a; Nacmias et al. 2004; Tsai et al. 2008; Yu et al. 2008). Similar to the analyses presented above, no association was established (d=.054, 95% CI .024-.133, z=1.355, p=.175) (see Table 2 and Figure 5).

Sensitivity and Publication Bias Analyses

The results of the sensitivity analysis did not differ from the results presented above; the corresponding forest plots of the sensitivity analysis can be found in Figures 6–10. Funnel plots to assess publication bias can be found in Figures 11–15. Figures 6–15 can be found in the supplementary materials.

Discussion

The present meta-analysis of the association of the BDNF Val66Met polymorphism with a number of indicators of cognition included 7,095 individuals, both healthy and challenged by a variety of mental health conditions (see Table 1). The data for this meta-analysis were drawn from 23 publications that included 31 independent samples. As far as we know, this is the first meta-analysis carried out to investigate the relationship between the BDNF Val66Met polymorphism and cognition.

Table 1.

Reference Numbers and Sample Information

First Author	Year	Country	Participants	Gender	Clinical	Status	Age
Egan a	2003	US	203	N/A	Y	Schizophrenia	Mean of groups =35
Egan b	2003	US	305	N/A	Y	Siblings of Individuals with Schizophrenia	Mean of groups = 35.9
Egan c	2003	US	133	N/A	N	Control	Mean of groups = 36.1
Rybakowski	2003	Poland	54	M-18 F-36	Y	Bipolar I Affective Illness	M= 46
Nacmias	2004	Italy	83	M-25 F-58	Y	Probable Alzheimer’s Disease	M=72.2
Strauss	2004	US	63	M-52% F-48%	Y	Major Depression or Dysthymic Disorder	M = 18.4
Tsai	2004	China	114	F-114	N	Healthy	M=19–21
Echeverria	2005	USA	427	M- 194 F-233	N	Mercury Exposure	M= M-49.0 F-36.0
Tan	2005	China	108	M- 70 F-38	Y	Schizophrenia	M=47.7
Harris et al.	2006	UK	460	N/A	N	Healthy	M=79.1
Ho a	2006	USA	144	M-45.1%	N	Healthy	M= 27.90
Ho b	2006	USA	293	M -72.7%	Y	Schizophrenia spectrum disorders	M=27.39
Oroszi et al.	2006	USA	59	M-13 F-46	Y	Systemic Lupus Erythematosus	M=41
Rybakowski a	2006	Poland	111	M-37 F-74	Y	Bipolar Mood Disorder	M= 43
Rybakowski b	2006	Poland	129	M-66 F-63	Y	Schizophrenic	M= 27
Rybakowski c	2006	Poland	92	M-34 F-58	N	Healthy	M =31
Hansell	2007	Australia	785	M- 385 F- 400	N	Healthy	M=16.4
Ho	2007	USA	119	M-83 F-36	Y	Schizophrenia Spectrum Disorders	M=26.4
Erickson et al	2008	USA	53	M-14F-39	N	Healthy	M=75.5
Miyajima	2008	England	722	N/A	N	Healthy	M=63
Tsai	2008	China	161	M – All	N	Normal Functioning	M=78.7
Yu a	2008	China	99	M-62 F-37	Y	Amnestic Mild Cognitive Impairment	M = 73.25
Yu b	2008	China	99	M-50 F-49	N	Healthy	M = 68.82
Houlihan	2009	Scotland/UK	1031	M-519 F-512	N	Healthy	Mean 69.5
Schofield et al	2009	Australia	475	M-49% F-51%	N	Healthy	M= 32.4
Baig	2010	Scotland	58	M-23 F-35	Y	Enhanced genetic risk of schizophrenia	M=26.90 F=25.91 (3.43)
Cathomas	2010	Switzerland	333	M-101 F-232	N	Healthy	M=22.8
Cerasa a	2010	Italy	29	M-14 F-15	Y	Multiple Sclerosis	M=28.9
Cerasa b	2010	Italy	32	M-14 F-18	N	Healthy	M=30.5
Chung a	2010	South Korea	51	M-51	Y	Homicidal Schizophrenic	M=37.9
Chung b	2010	South Korea	50	M-50	Y	Non-Homicidal Schizophrenic	M=38.5

Note.

M-Male, F- Female, Y-Yes, N-No, M-Mean

Overall, this meta-analysis did not detect consistent significant genetic association between the BDNF Val66Met polymorphism and any of the investigated cognitive phenotypes. As evident from the graphical representations of results from individual studies (see Figure 1–5), this summary is, perhaps, of no surprise, because the patterns of findings in the literature are, indeed, diverse, pointing in both directions, but regardless of the direction, indicating modest effect sizes. Of note also is that sensitivity and publication bias analyses did not change the overall pattern of results.

The conflicting findings in the literature may reflect any one of at least five different situations. First, it could be that the genetic variant (i.e., the BDNF Val66Met polymorphism) is truly not associated with the five clusters of cognitive indicators captured in this analysis. Second, it is possible that the genetic variant is associated with some intermediate phenotype (e.g., brain-based phenotype), which, in turn, is related to some but not all of the cognitive phenotypes examined here. Third, given the multitude and variability of BDNF isoforms and the diversity of its transcripts in different brain areas, it is possible that cognitive phenotypes should be grouped not by their behavioral similarities, but by similarities in the brain activation pathways that underlie these phenotypes. Fourth, it might be that this meta-analysis was hindered by reduced statistical power due to the limited sample sizes in some of the reported studies (see Table 1). Fifth, the results might vary because of some hidden sources of variation that stratify the findings and were not properly reflected in our analysis. Indeed, we did register between-study heterogeneity for two out of our five clusters of phenotypes. The obvious candidates for such sources of stratification are demographic characteristics such as gender, age, and ethnicity as well as the type of the sample (i.e., inclusion of typical vs. disordered individuals).

It is important to note that each of these stratifying sources is present in the literature on the BDNF Val66Met polymorphism. Specifically, with regard to gender, the literature contains reports of a dimorphic association of the polymorphism and depression, in both allelic and genotypic analyses (Verhagen et al. 2010). Similarly, there is also evidence of differential age-based effects of the Val66Met polymorphism (Verhagen et al. 2010). Unfortunately, the data that substantiated the current meta-analysis did not permit the analyses of either gender or age stratifying effects on the association between the BDNF Val66Met polymorphism and cognition. Likewise, we were not able to investigate the potential impact of ethnicity, although there is a clear reason to do so. Particularly, the Met allele frequency is higher in Asians compared to Caucasians (Shimizu et al. 2004). Yet, studies used in this meta-analysis were carried out, predominantly, with individuals of European descent, as is typical in the field of genetic association studies in general (Rosenberg et al. 2010), although there are selected Chinese and Korean studies. Clearly, the potential stratifying impact of ethnic background on the putative association between the BDNF Val66Met polymorphism and cognition needs to be investigated further. Finally, as is obvious from Table 1, the samples that were utilized in these analyses include both healthy and disordered individuals. Moreover, there is a wide spectrum of disorders and diseases that are represented in these studies. Once again, there is evidence in the literature that health status can be another stratifying variable, For example, opposite effects of the Val66Met polymorphism were reported for cognitive indicators in patients with multiple sclerosis and controls (Cerasa et al. 2010). Yet, although all of these considerations are relevant, as it stands now, the field has not generated enough data to allow a close investigation of the majority of potential stratifying factors; there is just not enough power to appraise their importance at this time.

In addition, our meta-analysis raised a number of other considerations. Specifically, it is striking how many different tasks have been used in the field to assess, arguably, similar (or the same) cognitive functions. Although often reflective of the theoretical and methodological specifics of the research question at hand, this diversity also signifies a lack of convergence in the methods employed to assess similar (the same) cognitive functions. Clearly, the presence of more homogeneity in the assessment methods for similar (or the same) underlying cognitive function would have been advantageous for detecting the presence or absence of the association with the BDNF Val66Met polymorphism more unequivocally. Thus, more orchestrated approaches to phenotyping in designing genetic and allelic association studies of BDNF and other candidate genes for cognition are highly desired and are expected to be fruitful. It is our hope that this summary of the phenotypes that have been used in this context might be of assistance in making decisions on what assessment should be included for particular cognitive functions. Moreover, anatomically and physiologically, BDNF is thought to be especially prominent in the hippocampus, therefore the impact of the Val66Met polymorphisms could have been more pertinent in hippocampus-specific, fine-grained tasks (Carlson et al. 2010). Yet, the present list of phenotypes utilized in the literature on the association studies of the BDNF Val66Met polymorphism does not present an opportunity for such fine-grained analyses. Finally, as this polymorphism has been reported to be associated, although with a certain degree of inconsistency in results, with a variety of neuropsychiatric disorders (e.g., eating disorders, bipolar disorder and/or depression, anxiety, and schizophrenia), it is possible that the connection between BDNF and cognition is indirect and modulated by some other factors.

Of note also is that in the analysis presented here we focused on a single, albeit functionally important, missense polymorphism of the BDNF gene. As mentioned in the Introduction, this gene contains numerous polymorphic sites and the frequencies of ancestral alleles at many of these sites are highly variable around the world; there is also evidence of a substantial amount of linkage disequilibrium between the polymorphic sites of BDNF and of positive selection at the BDNF locus, at least in pedigrees of European origin (Petryshen et al. 2010). All these phenomena underline the importance of carrying out not only single-polymorphism, but also haplotype analyses to reveal associations not only with particular alleles, but also with combinations of alleles (i.e., haplotypes) and disorders (Borroni et al. 2009; Suchanek et al. 2011; You et al. 2010), as well as cognitive and intermediate (Juckel et al. 2010) phenotypes. As the nature of genotyping changes and single-polymorphism studies are being replaced by targeted dense genotyping of a genetic region of interest or whole-genome genotyping, more research that investigates broad patterns of association between variations in BDNF and complex behavior can be carried out. In addition, it is possible that, for the effect to be consistently detectable in numerous samples, it has to be coupled with effects of other genes in the BDNF pathway, either additive or interactive (Lin et al. 2009; Miyajima et al. 2008b; Ribases et al. 2008). Clearly, the BDNF gene is complex structurally, functionally, and evolutionarily, and this complexity should be taken into consideration both at the design and result interpretation stages of association studies of BDNF, as they can influence strongly the likelihood of the detection of the susceptibility alleles for specific phenotypes, whether holistic (i.e., cognitive function or disorder) or intermediate (i.e., brain-based), and the appraisal of the magnitude of their effects.

In conclusion, although the Val66Met BDNF polymorphism remains an object of curiosity and investigative activities of researchers interested in the etiology of individual differences in cognition, this meta-analysis did not reveal any evidence of association between this polymorphism and the five studied indicators of cognition. This is, of course, disappointing, as both the gene and the polymorphism are, theoretically, excellent candidates for association with cognition. It is also disappointing because the literature on the connection between BDNF and complex human behavior spans a large spectrum of phenotypes and samples, contains numerous interesting insights, and yet is rife with contradictory findings. Nevertheless, although the results of this meta-analysis are not encouraging, the factors discussed above, uniquely and collectively, can explain these results. Also, as the evidence accumulates and the collective understanding of the connection between genes and cognition grows, such meta-analyses as the one presented here might appear to be helpful, providing a summative, although perhaps not decisive, given their own limitations, overview of the field.