Background

Major depressive disorder (MDD) is one of the leading causes of disability worldwide, affecting more than 150 million people [1]. In addition, the efficacy of the current treatment for MDD is suboptimal, and only about 30% of patients will achieve adequate remission after optimal treatment according to consensus guidelines [2]. Efforts to improve treatment strategies are, among others, hindered by the insufficient knowledge about the underlying pathophysiological mechanisms [3].

Over the past years, molecular studies of depression have moved beyond the monoamine hypothesis. Novel observations led to the “neurotrophic hypothesis of depression”, highlighting the functional significance of alterations in neurotrophic factors, particularly the brain-derived neurotrophic factor (BDNF) [4]. BDNF regulates neural development in early stages of life and plays a critical role in neural differentiation, and neurite and synaptic growth in adult brain [5]. In addition, by regulating synaptic plasticity and neurogenesis in different brain regions, BDNF seems to have a pivotal role in memory acquisition and consolidation [6].

Findings from multiple individual studies as well as meta-analyses suggest that depressed patients are characterized by decreased peripheral concentrations of BDNF compared to healthy controls [7,8,9,10]. Moreover, the functional significance of reduced BDNF levels in MDD was demonstrated in studies showing that this blunting was associated with neural atrophy and loss in key limbic regions such as hippocampus and prefrontal cortex [4, 11]. These alterations are in line with the observed reduced performance of depressed patients in cognitive tasks [11]. In contrast, increased BDNF levels in brain regions involved in mesolimbic dopaminergic circuit, such as the ventral tegmental area (VTA) and nucleus accumbens (NA), were shown to induce stress-related depressive-like behaviour in animal studies [12, 13]. Mesolimbic BDNF signalling seems to play a role in a range of motivation and reward-related behaviours [14] and has therefore been implicated in the pathophysiology of MDD, particularly with regard to loss of pleasure or anhedonia [15].

A growing body of research emphasizes the significance of the interplay between genetic vulnerability, reflected in the existence of single-nucleotide polymorphisms (SNPs), and epigenetic regulation of the BDNF gene in psychopathology. The BDNF gene has a complex structure, containing 11 exons and 9 have functional promoters that are brain region- and tissue-specific [16]. In particular, a large number of studies focused on investigating the rs6265 SNP as a potential risk biomarker in MDD [17]. This SNP is located at exon IX and causes guanine (G) to adenosine (A) replacement at codon 66, resulting in substitution of valine (Val) to methionine (Met) and a consequent impairment of the activity-dependent release of BDNF [18]. Although there is some evidence of the role of rs6265 in depression, the findings are inconsistent, especially in terms of identifying which of the two allele (Val or Met) indicates higher vulnerability [17].

One of the possible reasons for this inconsistency is the potential moderating effect of epigenetic regulation on the effect of the polymorphism on clinical outcomes. Epigenetic regulation of BDNF in depression has been increasingly studied, particularly DNA methylation [19]. Briefly, DNA methylation involves transfer of a methyl group to the CpG nucleotides, which mainly leads to gene silencing when located in the gene promoter region [20]. In the context of BDNF, DNA methylation of promoters of exon I and IV has been the most extensively studied [21, 22]. Increased BDNF methylation in depressed patients compared to healthy subjects has been reported in several studies, mainly in promoter of exon I and IV [23,24,25,26]. Most of these studies assessed purely differences in BDNF methylation between depressed patients and healthy controls. In addition, the interplay between Val66Met and BDNF methylation has been somewhat studied in the general population, showing that Val66Met affects DNA methylation of different individual CpGs in the BDNF gene [27, 28].

In contrast, the moderating and mediating roles of BDNF methylation on the association of Val66Met and specific functional outcomes corresponding to differential BDNF pathways in MDD remain largely unexplored. Despite the well-established role of BDNF in anhedonia and reward circuits on the one hand and cognitive performance on the other hand, the effect of interplay between the rs6265 SNP and BDNF methylation on these outcomes in MDD remains poorly understood. To the best of our knowledge, there are no studies investigating genetic and epigenetic regulation of BDNF in the context of anhedonia and reward learning in MDD. Regarding cognitive performance, only one study focused on the effect of rs6265 and methylation of BDNF exon I and IV [29], but without assessing methylation of exon IX, which contains the polymorphism. Previously, we found methylation of exon IX to be highly correlated with rs6265 and serum BDNF concentration [30] and changes in methylation of this exon were previously associated with changes in BDNF expression in MDD patients [31]. Therefore, in the present study, we aimed to investigate the genetic (rs6265) and epigenetic regulation of BDNF exon I, IV and IX in MDD and their impact on two major outcomes: (1) anhedonia and reward learning and (2) cognitive performance.

Methods

Study population

Eighty depressed patients and 58 healthy control subjects, matched for age and gender, were included. All patients were hospitalized at the University Psychiatric Centre of the University of Leuven in Belgium. A detailed recruitment procedure and inclusion/exclusion criteria were described previously [32, 33]. Briefly, the MDD diagnosis and psychiatric comorbidity were evaluated by a psychiatrist using the Structured Clinical Interview for DSM-IV-TR (SCID-I) [34]. Patients with other mood spectrum disorders, substance abuse or unstable medical conditions were excluded. Almost all patients started antidepressant treatment prior to admission to the hospital. During follow-up, treatment was not standardized and patients were treated with psychopharmacology and/or psychotherapy. Healthy participants did not meet criteria for any current or past psychiatric disorder or unstable medical conditions and were assessed only at baseline.

The study was approved by the UZ Leuven Medical Ethics Committee and all participants signed an informed consent.

Clinical and neuropsychological assessment

During the psychiatric interview, the 17-item Hamilton Rating Scale for Depression (HDRS) [35] was used to assess the severity of depression and the semi-structured trauma questionnaire (STI) was used to assess early life stress (ELS) [36]. Anhedonia was assessed using self-reported Snaith–Hamilton Pleasure Scale (SHAPS) [37] and positive and negative affect were measured using the Positive and Negative Affect Schedule (PANAS) [38].

To assess different aspects of cognitive performance, a battery of neuropsychological tests was administered. This included: (1) executive functioning/cognitive interference: Stroop Color and Word Test (SCWT) [39], (2) processing speed: Trial Making Test A (TMT-A) [40], (3) attention: Digit Span Forward (DS-F) [41] and (4) executive control: Trial Making Test B (TMT-B) [40] and Digit Span Backward (DS-B) [41]. All scores were transformed to percentiles, based on age, gender and education level, according to the appropriate reference categories [40, 42].

Reward task

A computerized reward learning task was used to measure participants’ ability to modulate behaviour as a function of reward [43]. The task has been previously described in detail [33, 44] and has been shown to objectively measure reward responsiveness in healthy individuals and depressed patients [43]. Briefly, the task lasted approximately 25 min and included 300 trials, divided in 3 blocks of 100 trials. In each trial, a cartoon face with either short or long mouth appeared for 100 ms. Participants were instructed to indicate which type of mouth was presented, and a monetary award was given for approximately 40 correct answers. An asymmetric reinforcement ratio was used to induce a response bias, meaning that subjects received a reward 3 times more frequently for correct identification of one mouth (the “rich stimulus”) than for correct identification of the other mouth (the “lean stimulus”). Before the task, participants were instructed that the goal of the task is to win as much money as possible. Response bias (RB) towards the rich stimulus was used as a measure of reward learning and was calculated as the difference in RB across Blocks (1, 2 and 3).

Genetic and epigenetic analysis

DNA was extracted from whole blood and the quantity and purity of DNA were determined using a NanoDrop spectrophotometer. Next, DNA samples were bisulphite-converted using the EZ-96 DNA Methylation-Gold™ Kit (#D5008, Zymo Research), according to the manufacturer’s protocol. Polymerase chain reaction (PCR) was used to amplify the bisulphite-converted DNA sequences selected for genetic and epigenetic analyses. For this, 1 μL of converted DNA was amplified by PCR in a total volume of 25 μL containing 0.2 μM of primers and 2 × Qiagen PyroMark PCR Master Mix (#978703, Qiagen). Pyrosequencing was performed using Pyro Gold reagents (#970802, Qiagen) on the PyroMark Q24 instrument (Qiagen) following the manufacturer’s instructions. Pyrosequencing results were analysed using the PyroMark analysis 2.0.7 software (Qiagen). The analysed sequences include 8 CpGs in the promoter of exon I (4 in promoter Ia and 4 in promoter Ib), 7 CpGs in promoter of exon IV and 5 CpGs in a coding region in exon IX (including Val66Met polymorphism) of the BDNF gene. We focused on DNA methylation of these specific regions due to their relevance for BDNF expression as well as the fact that these transcripts are highly abundant in both blood (leucocytes) and the brain tissue (hippocampus) [45]. Primer validation was performed through gel electrophoresis using MultiNA (Shimadzu Benelux BV, Belgium). Control DNA was used to perform gel electrophoresis and to validate each pyrosequencing analysis. All samples were randomized prior to DNA methylation analysis.

A detailed protocol with all analysed amplicons, PCR and sequencing primers is provided in the Additional file 1 and has been previously described in detail [46].

Statistical analysis

Comparisons of socio-demographic and clinical variables between the control and the depression group were performed using an independent sample t-test for continuous variables and Chi-Square test for categorical variables. Genotype differences between the groups were tested using Chi-Square test. The effect of polymorphism on DNA methylation as well as DNA methylation differences between groups was tested using Mann–Whitney U test due to data skewness. Methylation was analysed as average methylation percent across different CpGs within each assay to avoid multiple testings. CpG-specific analyses were only performed in the between-group comparison and to test the effect of polymorphism on DNA methylation of specific CpGs. As these analyses were explorative in nature, no additional correction for multiple testing was applied.

Explorative associations between DNA methylation and clinical variables (anhedonia, depression severity and cognitive performance) were tested using Spearman’s correlation. The observed correlations were further tested in univariate linear models, with appropriate covariates. The effect of the polymorphism on clinical variables (anhedonia, depression severity, cognitive performance) was tested either using an independent sample t-test or Mann–Whitney U test, depending on the distribution of the outcome variable. In case we observed an association between both the polymorphism and DNA methylation and any of the cognitive performance outcomes, we tested potential mediation effect of DNA methylation on this association, by using the PROCESS macro of SPSS version 21.0, developed by Hayes [10]. The mediation effect was interpreted using the bias-corrected bootstrap confidence intervals, whereby the mediation effect was considered significant when the bias-corrected 95% confidence intervals did not contain zero.

To test the association between polymorphism/DNA methylation and response bias, we applied a repeated measures ANOVA with response bias as the outcome variable and with Block (1, 2, 3) and polymorphism/DNA methylation as predictors. This model was also used to test for potential interactions between polymorphism and DNA methylation on response bias. The Greenhouse–Geisser correction was used when the sphericity assumption was violated.

All statistical analyses were performed using SPSS software package, version 26.0. All tests were two-sided, and the significance level was set at 0.05.

Results

Population characteristics

Socio-demographic and clinical characteristics of all participants are presented in Table 1. Socio-demographic features were comparable between the groups, except from education level, which differed significantly (X2 = 19.3, p < 0.001). As expected, depressed patients had higher levels of anhedonia and more frequently reported early life stress events (ps < 0.001). Regarding the cognitive performance tests, relative to the healthy controls, depressed patients had significantly poorer performance in SCWT, TMT-A and TMT-B (all ps < 0.001) but not in DS-F (p = 0.098) nor DS-B (p = 0.411).

Table 1 Socio-demographic and clinical characteristics of participants at baseline
Full size table

Val66Met Polymorphism and depression

Evaluating the polymorphism data, depressed patients and healthy subjects did not differ significantly in the number of Met carriers (36.2% in the MDD group compared to 32.8% in the control group, X2 = 0.181, p = 0.720). In addition, Val66Met polymorphism was not associated with any clinical features (depression severity, anhedonia), neither in the whole sample nor in the depression group separately (all ps > 0.05, data not shown). I n terms of the effect of Val66Met on BDNF methylation in the whole sample (both groups), the presence of a Met allele had a negative effect on the average methylation of the exon IX (mean difference 8.78%, p < 0.001), as well as 4 out of 5 individual CpGs in this region (Additional file 1: Table S2). In addition, Met carriers had lower average methylation of exon Ia (mean difference 0.35%, p = 0.006) and exon Ib (mean difference 0.28%, p = 0.014) as well as several individual CpGs in these two regions (presented in Additional file 1: Table S2) in the overall sample. For exon IV, only association with one CpG was found (CpG5: mean difference 0.34%, p = 0.011). Group (depression vs. control) was not a moderator of any of these associations (all ps > 0.05).

BDNF methylation and depression

When comparing the average methylation levels of different BDNF regions between the depression and the control group (Fig. 1; Additional file 1: Table S3), we observed significantly decreased methylation of BDNF exon Ia in the depression group (mean difference 0.30%, p = 0.042), as well as 2 individual CpGs (CpG1: mean difference 0.29%, p = 0.040, CpG2: mean difference 0.46%, p = 0.013). The group effect on average methylation of exon Ia remained significant when controlling for the Val66Met polymorphism (B = 0.282, p = 0.043, ηp2 = 0.03).

Fig. 1
figure 1

Overview of BDNF methylation in the control group (grey) and the depression group (black). DNA methylation is presented as mean and standard error of the mean for each of the analysed regions: promoter Ia (A), promoter Ib (B), promoter IV (C) and exon IX (D). Methylation of each individual CpG as well as the average methylation levels across all CpGs are presented for each analysed region. Significance levels are derived from Mann–Whitney U test (*p < 0.05, **p < 0.01)

Full size image

As a confirmatory analysis, we tested whether the observed methylation differences were associated dimensionally with clinical features such as anhedonia and depression severity (Additional file 1: Table S4). In the overall sample, average methylation of BDNF exon Ia was negatively associated with anhedonia (Spearman's Rho = − 0.180, p = 0.036) and negative affect (Spearman's Rho = − 0.198, p = 0.021). In the depression group, no associations were observed (all ps > 0.05).

For exon Ib, IV and IX, there were no significant group differences (all ps > 0.05, results presented in Fig. 1 and Additional file 1: Table S3). In the dimensional analysis (Additional file 1: Table S4), average methylation of exon Ib was negatively associated with anhedonia in the overall sample (Spearman's Rho = − 0.177, p = 0.039) and in the MDD group separately (Spearman's Rho = − 0.225, p = 0.047) and was positively associated with positive affect in the depression group (Spearman's Rho = 0.232, p = 0.040). For exon IV, no associations were observed in the overall sample, but in the MDD group, average methylation was negatively associated with depression severity (Spearman's Rho = − 0.245, p = 0.031) and positively with positive affect (Spearman's Rho = 0.288, p = 0.011).

None of the average methylation levels differed between subjects with and without early life stress, neither in the overall sample nor among depressed patients (all ps > 0.05). Similarly, type of antidepressant did not have an impact on DNA methylation (all ps > 0.05).

Val66Met, BDNF methylation and reward learning

In the overall sample, there was no significant effect of polymorphism on response bias (F(2,250) = 1.843, p = 0.136, ηp2 = 0.015). However, DNA methylation of BDNF promoter Ib had a significant positive effect on reward bias (F(2,248) = 3.804, p = 0.026, ηp2 = 0.030), indicating that participants with lower BDNF promoter Ib methylation showed lower response bias during the computerized reward learning task. More specifically, this association was driven by the effect of DNA methylation of this region on response bias in Block 1 (B = 0.040, p = 0.043) and 2 (B = 0.071, p = 0.008), but not Block 3 (B = − 0.004, p = 0.882). In addition, we did not observe a significant interaction effect between polymorphism and BDNF promoter Ib methylation on response bias (F(2,244) = 0.131, p = 0.865, ηp2 = 0.001). DNA methylation of other BDNF regions (promoter Ia, promoter IV and exon IX) did not have a significant effect on the performance on the reward learning task (all ps > 0.05, data not shown).

In contrast, in the depression group, we observed a significant association between polymorphism and response bias (F(2,146) = 3.061, p < 0.050, ηp2 = 0.040), whereby Met carriers had significantly lower response bias than Val/Val homozygotes, as illustrated in Fig. 2. Similar as in the overall sample, DNA methylation of BDNF promoter Ib had a significant positive effect on response bias (F(2,144) = 3.301, p = 0.040, ηp2 = 0.044). More specifically, depressed patients with lower DNA methylation of promoter Ib showed lower response bias in Block 2 (B = 0.077, p = 0.015) but not in Block 1 and 3 (all ps > 0.05). When both predictors (polymorphism and promoter Ib methylation) were added to the model, only the effect of DNA methylation remained significant (p = 0.047) whereas the effect of polymorphism was lost (p = 0.072), indicating a potential indirect (mediating) effect of polymorphism on reward bias, via DNA methylation. When testing the interaction effect between polymorphism and DNA methylation on the performance on the reward learning task, no significant effect was found (F(2,140) = 1.635, p = 0.199, ηp2 = 0.023). Type of antidepressant did not have a moderation effect on any of the observed associations (all ps > 0.05). DNA methylation of other BDNF regions did not have a significant effect on reward bias in the depression group (all ps > 0.05). The effect of polymorphism and BDNF promoter Ib methylation on response bias is presented in Table 2.

Fig. 2
figure 2

Comparison of reward bias as a function of blocks, as measured by the computerized reward learning task, between Val/Val homozygotes (orange) and Met carriers (blue) in the depression group. Data are presented as mean and standard error of the mean

Full size image
Table 2 Overview of the effect of polymorphism and BDNF promoter Ib methylation on reward bias analysed using repeated measures ANOVA
Full size table

Val66Met, BDNF methylation and cognitive performance

Regarding the effect of Val66Met on cognitive performance, there were no significant associations in the overall sample (all ps > 0.05). However, in the depression group, Met carriers had better performance on the SCWT (Z = 2.71, p = 0.007) and DS-B (Z = 2.62, p = 0.009), as presented in Fig. 3.

Fig. 3
figure 3

Comparison of cognitive performance between Val/Val homozygotes (orange) and Met carriers (blue) in the depression group. Data are presented as mean and standard error of the mean. Significance levels are derived from Mann–Whitney U test (*p < 0.05, **p < 0.01). SCWT: Stroop Color and Word Test; TMT-A: Trial Making Test part A; TMT-B: Trial Making Test part B; DS-F: Digit Span Forward; DS-B: Digit Span Backward

Full size image

In the overall sample, average methylation of different assays was not significantly associated with cognitive performance (all ps > 0.05). However, in depressed patients, average methylation of exon IX (containing the polymorphism) was negatively associated with performance of the SCWT (Spearman's Rho = − 0.228, p = 0.046) and DS-B (Spearman's Rho = − 0.334, p = 0.002), as illustrated in Fig. 4.

Fig. 4
figure 4

Correlation matrix (Spearman correlation) between BDNF methylation of different regions and cognitive performance in the depression group. Positive correlations are presented in blue whereas negative correlations are given in red and the intensity of colours corresponds to the values of Spearman’s Rho correlation coefficient. Significance: *p < 0.05, **p < 0.001. SCWT: Stroop Color and Word Test; TMT-A: Trial Making Test part A; TMT-B: Trial Making Test part B; DS-F: Digit Span Forward; DS-B: Digit Span Backward

Full size image

In addition, mediation analysis revealed that exon IX methylation was a significant mediator of the effect of Val66Met on DS-B (Effect 20.05, CI [1.47, 50.20]) but not SCWT (Effect 0.83 CI [− 29.88, 22.17]). The mediation effect of exon IX methylation on the association between Val66Met and DS-B is presented in Additional file 1: Fig. S5. Type of antidepressant did not moderate any of the observed associations in the depression group (all ps > 0.05).

Discussion

In the present study, we probed the role of genetic and epigenetic regulation of BDNF in major depression, by focusing on anhedonia, reward learning and cognitive performance as the main outcomes. Our results demonstrated that lower BDNF methylation of promoter I contributed to more pronounced anhedonia and impaired reward learning in depressed patients. The presence of Met allele in the common BDNF polymorphism (Val66Met) had a negative effect on reward learning, but this seemed to be an indirect effect, via DNA methylation of promoter I. On the other hand, Met allele exhibited a protective function for cognitive performance (executive functioning in particular), through the mediating effect of reduced methylation in the BDNF coding region (exon IX) in the same population of depressed patients.

To the best of our knowledge, our study was the first to test the association between genetic and epigenetic control of BDNF in reward learning in MDD. Even though there are no other similar studies to compare these findings, the observed impairing effect of decreased BDNF methylation seems to be in line with the growing data demonstrating the depressogenic and anhedonic effects of BDNF in the mesolimbic reward circuitries in animals exposed to chronic social defeat stress [14, 15, 47]. In addition, the fact that we found a negative association between self-reported anhedonia and promoter I methylation as well as decreased methylation of this region in between-group comparison with a healthy population makes our conclusions stronger and more convincing. Interestingly, increased BDNF plasma levels have been previously associated with decreased reward learning in patients with bulimia nervosa [48]. Even though this is a different clinical and nosological phenotype from MDD, these data together with our findings provide evidence for the role of BDNF in reward circuitries disruption that extend beyond diagnostic categories. In the present study, we did not assess mRNA expression or serum/plasma BDNF levels and therefore our conclusions on the functional consequences of decreased BDNF promoter I methylation are limited. However, our previous work [46] together with other research [49] has shown a negative correlation between promoter I methylation and BDNF expression. In addition, we also observed association between depression severity and reduced positive affect with decreased promoter IV methylation, which was previously shown to correlate negatively with BDNF expression (16). This might be an additional indicator of increased BDNF expression in our population of depressed patients, even though direct association with anhedonia or reward learning was not found.

In our study, we also find a significant effect of genotype (Val66Met) on reward learning, whereby Met carriers showed a more blunted reward bias in the reward learning task than Val/Val homozygotes. We speculate that our findings are consistent with research indicating that the Met allele is a vulnerability allele for depression (and mental disease in general) due to its impairing effect on BDNF translocation and secretion [18]. However, as mentioned before, this conclusion failed to be replicated [17, 50, 51], perhaps indicating that this polymorphism determines vulnerability to environment-induced epigenetic changes rather than being a risk factor per se. Indeed, in our study, Val66Met was not associated with any of the outcomes in healthy participants, but only in the MDD group. In addition, this effect was lost after we controlled for DNA methylation of BDNF promoter I, indicating the predominant effect of epigenetic regulation and a potentially indirect (mediating) effect of the polymorphism. The importance of environment-induced epigenetic regulation of BDNF is also supported by evidence showing that BDNF signalling pathway does not play a role in reward circuitries in normal conditions, but gains importance once these are disrupted [52].

In contrast, among the MDD group, we found increased BDNF exon IX methylation to be associated with poorer executive functioning, particularly in the executive domain (STROOP and DS-B). To the best of our knowledge, methylation of this region has never been not tested before with respect to cognitive performance in depressed patients. Previously, Ferrer et al. [29] found methylation of exon I and IV to be associated with visual learning and memory, but reported no association with cognitive function tested, which is in line with our findings. Exon IX has been previously shown to be particularly abundant in the hippocampus [45] and de-methylation of this region was shown to induce adult hippocampal neurogenesis [53]. In addition, in our study, exon IX methylation was a mediator of the effect of Val66Met on the executive functions, whereby the presence of the Met allele appeared to be protective, which is somewhat surprising. Specifically, previous meta-analysis [54] have shown that Met carriers had lower hippocampal volume, but also provided evidence for publication bias and small effect sizes, questioning whether the observed effect is subject to a winner's curse. Moreover, some recent studies indicate that the Met allele can be protective for executive functions in the presence of depressive symptoms [55] and can promote recovery of executive functioning after PTSD [56], indicating a potentially differential effect of Val66Met on cognitive function in healthy individuals and in the presence of psychopathology. In our study, the effect of Val66Met on executive function was predominantly indirect, through DNA methylation of BDNF exon IX. Therefore, our results indicate that Val66Met is a vulnerability factor susceptible to epigenetic regulation and future studies should account for DNA methylation when further exploring the complex role of this polymorphism in different neural processes and psychopathology.

Several limitations need to be taken into account when interpreting the results of our study. First, almost all depressed patients were undergoing antidepressant treatment at the time of inclusion, which might have influenced BDNF methylation levels [25, 57]. Due to high collinearity with the MDD diagnosis, we did not include medication as a covariate in the analysis; however, we did test the effect of different types of medication on BDNF methylation levels and found no significant differences. Medication could have influenced the between-group differences in BDNF methylation, but it is highly unlikely that antidepressants could explain the observed correlations with anhedonia, reward learning and cognitive performance found in the MDD group. In addition, the effect of antidepressants on BDNF methylation remains unclear. Whereas studies show that antidepressants can restore reduced serum BDNF protein levels in depressed patients [8, 58], methylation studies show conflicting findings, demonstrating that antidepressants were associated with both increase and decrease in BDNF methylation in this patient population [25, 59, 60]. Moreover, some studies reported the absence of difference between BDNF methylation at the baseline and at the end of antidepressant treatment in depressed patients [61]. Therefore, we believe that methylation studies on both medication-free MDD patients as well as patients taking antidepressants are important complementary source of data as they allow comparing which epigenetic effects are potentially a result of medication intake and which reflect phenotypic changes linked to MDD itself. In the context of BDNF, this is still to be elucidated as we do not completely understand the effect of medication (and psychotherapy) on BDNF methylation patterns. Next, in the present study, we did not collect RNA samples and therefore mRNA expression analysis could not be performed. This is another limitation of the study as these data would help get a clearer picture of the direct effects of the observed DNA methylation changes on the gene expression and the overall interaction between genetic, epigenetic regulation and gene expression of BDNF. Previous data showed that both Val66Met and BDNF methylation of different exons could have an impact on BDNF expression [28, 30, 46] but further conclusions on this could not be drawn in our patient population. Next, in our study, we analysed DNA methylation using targeted approach (pyrosequencing), which allowed mapping DNA methylation patterns only in one part of the BDNF gene (20 individual CpGs). Other studies show Val66Met to have an influence on DNA methylation of other CpG position in the BDNF gene [27], and therefore, we might have failed to capture other genotype-DNA methylation interaction effects of relevance of the outcomes analysed here (reward learning and cognitive performance) in the population of clinically depressed patients. Finally, all patients included in the MDD group were hospitalized and therefore might be more prone to the chronic course of illness, which might limit generalizability of our findings.

Conclusions

To conclude, our data suggest that Val66Met is associated with DNA methylation of different BDNF exons, which may in turn contribute to impaired reward learning and executive functioning in patients with MDD. We hypothesize that these genetic and epigenetic patterns affect reward learning and executive functioning in depressed patients via changes in the expression of different BDNF exons; however, this was not directly assessed in the present study and needs to be further explored. Therefore, we recommend future studies to simultaneously assess Val66Met, DNA methylation and possibly mRNA expression of different BDNF transcripts in order to learn more about their role in different pathways of relevance for MDD.