Abstract
The brain-derived neurotrophic factor (BDNF) is critical for brain development, and the functional BDNF Val66Met polymorphism is implicated in risk for mood disorders. The objective of this study was to determine how the Val66Met polymorphism influences amygdala–cortical connectivity during neurodevelopment and assess the relevance for mood disorders. Age- and sex-specific effects of the BDNF Val66Met polymorphism on amygdala–cortical connectivity were assessed by examining covariance of amygdala volumes with thickness throughout the cortex in a sample of Caucasian youths ages 8–22 that were part of the Philadelphia Neurodevelopmental Cohort (n = 339). Follow-up analyses assessed corresponding BDNF genotype effects on resting-state functional connectivity (n = 186) and the association between BDNF genotype and major depressive disorder (MDD) (n = 2749). In adolescents, amygdala–cortical covariance was significantly stronger in Met allele carriers compared with Val/Val homozygotes in amygdala–cortical networks implicated in depression; these differences were driven by females. In follow-up analyses, the Met allele was also associated with stronger resting-state functional connectivity in adolescents and increased likelihood of MDD in adolescent females. The BDNF Val66Met polymorphism may confer risk for mood disorders in females through effects on amygdala–cortical connectivity during adolescence, coinciding with a period in the lifespan when onset of depression often occurs, more commonly in females.
Introduction
It is increasingly recognized that normal development of the “male brain” or “female brain” occurs in a manner that may predispose males or females to certain mental illnesses. Mood and anxiety disorders are more common in females than in males (Beesdo et al. 2009). Onset of depression typically occurs in adolescence when subcortical–cortical connections in the brain mature.
Common variation in plasticity-related genes may influence the maturation of brain circuits during development. Brain-derived neurotrophic factor (BDNF) protein is widely expressed throughout the developing and adult human brain and is a key regulator of neural circuit development and function (Park and Poo 2013). Decreased BDNF serum levels have been associated with depression (Castren and Rantamaki 2010) and the expression of BDNF is augmented by antidepressant treatment (Duman and Monteggia 2006). The BDNF Val66Met polymorphism may be a risk factor for mood disorders, though results directly linking it to diagnoses are mixed (Notaras et al. 2015); evidence suggests that complex interactions between age, sex, environmental factors, and other genetic variants may modify the effects of this variant on disease-related phenotypes (Boulle et al. 2012). These interactions may explain in part the lack of association observed in genome-wide association studies.
On the other hand, a number of studies have consistently shown relationships between BDNF Val66Met and alterations in the structure and function of the amygdala and cortical regions (Pezawas et al. 2004; Soliman et al. 2010). Some studies have gone further, finding that this polymorphism interacts with age or sex to predict such alterations (Voineskos et al. 2011). Given the prominent role of the amygdala and cortical regions in emotion processing and regulation (Fusar-Poli et al. 2009) as well as their altered functioning in depression (Kerestes et al. 2014) these neural substrates may serve as intermediate biological markers for depression. However, a drawback of regional analyses is that brain regions do not function in isolation. Structural covariance patterns can be assessed to probe the interactivity of brain regions (Alexander-Bloch et al. 2013), have been shown to reflect their coordinated maturation (Raznahan et al. 2011) and are largely genetically determined (Schmitt et al. 2010; Eyler et al. 2011). Inverse relationships between amygdala volume and frontal and inferior parietal cortices have been demonstrated in a developmental sample (Albaugh et al. 2013) and there is evidence for links between amygdala medial frontal structural relationships and normal variation in negative affect (Holmes et al. 2012). Therefore, the structural covariance approach may be particularly useful for the identification of biomarkers of altered development that contributes to the emergence of mood disorders.
The aim of this study was to examine the influence of the BDNF Val66Met polymorphism on amygdala–cortical circuits in a sample of children, adolescents, and young adults to assess a developmentally timed potential risk mechanism for mood and anxiety disorders. We also assessed sex-specific effects of this potential risk mechanism. We hypothesized that, during development, variation in BDNF Val66Met would: (1) influence amygdala volume-cortical thickness covariance most prominently during adolescence and (2) that these altered relationships would be more pronounced in females. We also sought to explore whether our main findings would be supported via association with functional connectivity in similar brain circuits, and via occurrence of major depressive disorder (MDD) symptoms.
Materials and Methods
Sample
Sample Data
Participants were part of the Philadelphia Neurodevelopmental Cohort (PNC), a large study of child and adolescent development (ages 8–22) (Satterthwaite et al. 2015) from which data have been made publicly available for 8719 individuals. All participants submitted samples for genetic analysis and were assessed clinically with a structured interview. Genome-wide genotyping was performed on Affymetrix (6.0 Genechip and Axiom) and Illumina (Human610, HumanHap550 v1.0, and HumanHap550 v3.0) platforms. Psychopathology was assessed using a computerized, structured interview (GOASSESS) (Calkins et al. 2015). A subset of 1000 of these participants also underwent multimodal neuroimaging. All imaging data were collected on the same 3T Siemens TIM Trio whole-body scanner (Satterthwaite et al. 2014a).
Genetic Data Processing
Genetic data were processed as described previously (Voineskos et al. 2016) and is described in detail in the Supplementary Methods. All nonCaucasian participants, ethnic outliers, and highly related individuals were excluded from statistical analysis. Genotyping at rs6265 was extracted to assess BDNF Val66Met genotype.
Summary of Analyses
Individuals were included in BDNF Val66Met-neuroimaging analyses if they were Caucasian, had neuroimaging and genetic data that passed quality control, and did not have a history of moderate or severe medical conditions that may impact brain development. This resulted in a sample of 339 unrelated individuals for “BDNF Val66Met-structural covariance analysis” (number of participants who met exclusion criteria: nonCaucasian ethnicity N = 485, failed neuroimaging quality control N = 66, missing genetic data N = 24, ethnic outlier N = 31, related participants N = 227, medical condition N = 58; some participants met multiple exclusion criteria). Based on our structural covariance results, we performed guided follow-up analyses of functional connectivity and MDD. Of those included in the structural covariance analysis, 153 out of 339 subjects were lost to quality assurance for subsequent “follow-up BDNF Val66Met-functional connectivity analysis,” resulting in n = 186. Participants were suitable for the “follow-up BDNF Val66Met-MDD analysis” if they met the same inclusion criteria without the neuroimaging requirement (n = 2749). Risk associated with individuals carrying 2 copies of the minor Met allele was assessed in this larger sample available for the MDD analysis, something that despite high interest was not feasible in the neuroimaging analysis due to negligible numbers of Met allele homozygotes (9/339). The frequency of the Met allele varies by ethnic background ranging from 44% in Asia to 20% in Europe to 1% in Sub-Saharan Africa (Petryshen et al. 2010). This sample contained only a small proportion of participants of Asian decent (1%) and while there was a sizeable proportion of African Americans (35%), the small number of Met allele carriers (8%) left us underpowered to perform well-powered analyses in each of these ethnic groups. See Figure 1 for a summary of analysis groups and sample sizes.
Summary of PNC groups used in separate analyses. Primary analysis of BDNF Val66Met-structural covariance was followed up with BDNF Val66Met-functional connectivity analysis and BDNF Val666Met-MDD analysis in groups of PNC participants selected based on available data and specific inclusion/exclusion criteria described in the Materials and Methods section and summarized here. PNC, Philadelphia neurodevelopmental cohort; QC, quality control; MDD, major depressive disorder.
BDNF Val66Met-Structural Covariance Analysis
Image Acquisition—T1-Weighted Magnetic Resonance Imaging
A magnetization-prepared, rapid acquisition gradient-echo T1-weighted structural image was acquired, using the following parameters: repetition time (TR), 1810 ms; echo time (TE), 3.51 ms; field of view (FOV), 180 × 240 mm; matrix, 256 × 192; 160 slices; TI, 1100 ms; flip angle, 9°; effective voxel resolution, 0.9 × 0.9 × 1 mm.
Image Processing—Cortical Thickness
T1-weighted images were processed through the CIVET processing pipeline (version 1.1.10; Montreal Neurological Institute). Images were registered to a nonlinear template (ICBM152 nonlinear sixth-generation template with a 9-parameter linear transformation), inhomogeneity corrected (Sled et al. 1998) and tissue classified. Deformable models were used to create white and gray matter surfaces for each hemisphere resulting in 4 surfaces of 40 962 vertices each, white to gray matter surface distances were then determined using the t-link metric (Lerch and Evans 2005). The thickness data were subsequently blurred using a 20-mm surface-based diffusion blurring kernel and nonlinearly registered to a template surface. Quality control criteria for each subject were generated by examining metrics associated with the quality of image registration and surface extraction. Cortical thickness, unlike cortical and amygdala volume is not strongly associated with total brain volume (Sowell et al. 2008) and was, therefore, not adjusted by dividing by total brain volume.
Image Processing—Amygdala Volume
Left and right amygdalae were automatically identified on T1-weighted images using the MAGeT Brain multiatlas segmentation tool (Chakravarty et al. 2013). Briefly, 5 high-resolution (300 μm isotropic voxels) in vivo atlases of the amygdala were used for the segmentation pipeline (Winterburn et al. 2013). The atlases were customized to a subset of the data (21 participants selected to match the age and sex distribution of the data), using a nonlinear transformation estimated in a region-of-interest defined around the structure (Chakravarty et al. 2009). This then acted as a set of templates to which all scans in the study were warped, producing 21 candidate amygdala segmentations for each scan. The final segmentation was determined by retaining the label occurring most frequently at a specific location. To rule out the presence of segmentation errors, a quality control image file was produced for each scan, allowing detailed visual inspection. Total brain volume was computed as part of the CIVET processing pipeline. The volume of the amygdala and other subcortical structures tend to be proportional to total brain volume; however, here we are interested in the covariance of amygdala volume with cortical thickness, which is independent of total brain volume (Barnes et al. 2010). Therefore, amygdala volumes were adjusted by dividing by total brain volume.
Statistical Analyses—BDNF Val66Met and Structural Covariance
Individuals with 2 copies of the Val allele (Val/Val) were compared with individuals with either 1 or 2 copies of the Met allele (Met carriers). Overall amygdala–cortical covariance in each group was assessed by computing a partial correlation (adjusting for the effects of age and sex) between total amygdala volume (sum of left and right amygdala) and mean thickness across the cortex using Pearson's r. This was performed within the whole sample and subsequently within age quartile groups (childhood [8–11], early adolescence [12–14], late adolescence [15–18] and young adulthood [19–22]) in order to assess if the effects were specific to a particular developmental period. Cortical thickness shows complex developmental trajectories across the age range examined here (Shaw et al. 2008). Because we are interested in the age-specific genotypic effects of amygdala cortical covariance, we adjusted for the effects of age on thickness before computing the amygdala–cortical correlations. This helps ensure that our findings are independent of age effects on absolute cortical thickness. Where age group differences were detected in the overall analysis, regionally specific amygdala–cortical covariance was examined by computing partial correlations (adjusting for the effects of age and sex) between amygdalae volumes and vertex-wise cortical thickness, using the mapping of anatomical correlations across the cerebral cortex (Lerch et al. 2006) toolbox in R (v3.1.1). To assess the relative sex-specific contributions to these amygdala–cortical correlation patterns the vertex-wise analysis was repeated in separate groups of males and females (adjusting for effects of age only). For all of these analyses differences in correlations between BDNF genotype groups were assessed with a linear interaction model that tested for differences in slope using the student's t-statistic. Supplementary analyses computed partial correlations that adjusted for additional potential confounding factors in the model including nonlinear age effects, population stratification, and total brain volume (see Supplemental Methods). To account for multiple comparisons in the vertex-wise analysis, the statistical threshold was determined by application of a 5% false discovery rate (FDR) correction (Genovese et al. 2002). Correlations between amygdala volume and individual vertices within significant clusters were converted with the Fisher's z transform and averaged in order to report mean values within clusters. A supplementary analysis examined the relationships between cortical regions by examining pairwise correlations between mean thickness in cortical clusters where we saw differences in amygdala cortical covariance between genotypes (see Supplemental Methods).
Follow-Up BDNF Val66Met-Functional Connectivity Analysis
Image Acquisition—Functional MRI
Resting-state blood oxygen level-dependent (BOLD) functional magnetic resonance imaging (fMRI) was acquired using a whole-brain, single-shot, multislice, gradient-echo echoplanar sequence with the following parameters: 124 volumes, TR 3000 ms, TE 32 ms, flip angle 90°, FOV 192 × 192 mm, matrix 64 × 64, slice thickness 3 mm, slice gap 0 mm, effective voxel resolution 3.0 × 3.0 × 3.0 mm. During the resting-state scan, a fixation cross was displayed as images were acquired. Participants were instructed to remain still, stay awake, keep their eyes open and fixate on the displayed crosshair.
Image Processing—Functional MRI
Functional MRI preprocessing was performed as described in previous analyses of resting-state functional imaging data from this cohort (Satterthwaite Wolf et al. 2014b) and is described in detail in the Supplemental Methods.
Statistical Analyses: BDNF Val66Met and Functional Connectivity
Mean time series were extracted from the amygdalae of each subject using individual masks generated by the amygdala volume-processing pipeline. The cortical time series were projected onto a CIVET surface using analysis of functional neuroimaging (AFNI) tools to provide direct correspondence between functional connectivity and cortical thickness measures. Cortical surface region-of interest masks was derived from cortical regions where the structural covariance analysis detected significant differences in amygdala–cortical covariance between groups at a 5% FDR corrected level. Because some of the regions of interest contained very few vertices, masks in these regions were inflated to include vertices that showed significant structural differences at 10% FDR. The mean time series were extracted from each of these cortical regions. Where significant vertex-wise structural covariance differences based on BDNF Val66Met genotype were found Pearson's correlation coefficients were calculated between the mean time series of the amygdala and the mean time series of each cortical region-of interest. Fisher's z transformation was used to convert the resulting correlation coefficients into z-values which were used in general linear model analyses with BDNF Val66Met genotype group as the between group factor and age and sex as covariates. Supplementary analyses included additional covariates in the model to account for population stratification, nonlinear age effects and in scanner motion and compared alternate methods for assessing functional connectivity between brain regions (see Supplementary Methods). Mean z-values within groups are reported here. Consistent with the follow-up nature of these analyses a P < 0.05 was considered meaningful. A supplementary analysis examined the relationships between cortical regions by examining pairwise correlations in BOLD signal time series in cortical clusters where we saw differences in resting-state functional connectivity between genotypes (see Supplementary Methods).
Follow-Up BDNF Val66Met-MDD Analysis
MDD Assessment
The structured interview assessed relevant symptoms and diagnostic and statical manual of mental disorders (DSM-4) criteria for MDD along with associated distress and impairment on separate 11-point scales ranging from no bother/problems to extremely serious bother/problems. Participants met criteria for MDD if they reported 5 or more DSM-4 criteria for MDD that were accompanied by significant distress or impairment.
Statistical Analyses—BDNF Val66Met and MDD
In age groups where significant vertex-wise structural covariance differences based on BDNF Val66Met genotype were found logistic regression was used to assess the association of BDNF Val66Met genotype with MDD psychopathology in male and female groups separately. In each group, 3 genotype models (additive, dominant, and recessive) were tested predicting the binomial outcome of occurrence, co-varying for age at assessment and the top 3 principle components from multidimensional scaling analysis (to account for population stratification). P-values were corrected within groups using the P(ACT) method (Conneely and Boehnke 2007), which accounts for correlations between models, then between groups (males vs. females) using the Bonferroni method.
Results
Sample Characterization
For all analyses, BDNF Val66Met genotype groups compared did not differ by age, sex, or Wide Range Assessment Test score. Additionally, the imaging analysis subsamples did not differ from the larger genetic analysis sample in any of these factors (Supplementary Tables 1 and 2). Allele frequencies did not show deviation from Hardy–Weinberg equilibrium (structural covariance analysis chi-squared P = 0.51; MDD analysis chi-squared P = 0.83).
BDNF Val66Met-Structural Covariance Analysis
Age-Specific Effects of BDNF Val66Met on Structural Covariance (Age 8–22)
First, examining the whole sample, covariance between total amygdala volume and mean cortical thickness differed by genotype at trend-level of significance (Met Carrier r = 0.09, Val/Val r = −0.13, interaction P = 0.07). Second, we tested whether these differences were specific to a particular developmental period by dividing the sample into quartiles based on age. Covariance between total amygdala volume and mean cortical thickness differed by genotype in the middle 2 age quartiles (Q2 age 12–14, Met Carrier r = 0.28, Val/Val r = −0.22, interaction P = 0.04; Q3 age 15–18, Met Carrier r = 0.24, Val/Val r = −0.23, interaction P = 0.04), but not the lower or upper quartile (interaction P > 0.7) (Fig. 2).
Overall amygdala–cortical covariance by BDNF Val66Met genotype in separate age groups. Correlations are plotted between total amygdalae volume (x-axis) and mean cortical thickness (y-axis) in separate age quartile groups for Val/Val (blue) and Met carrier (red) individuals. Residuals after regressing for age and sex are plotted so that plotted points reflect relative values. The shaded area around the regression lines represent 95% confidence intervals. Significant differences between genotype groups were detected in the middle 2 age groups that corresponding to adolescence (Age 12–14; Age 15–18, P < 0.05).
Regional Effects of BDNF Val66Met on Structural Covariance Among All Adolescents (Age 12–18)
Regional localization of cortical differences was examined in the combined middle 2 age groups where overall differences in amygdala cortical covariance were detected (i.e., adolescence; age 12–18). Correlations between the right amygdala and vertex-wise cortical thickness differed by BDNF genotype with most cortical regions, demonstrating weak negative correlations with amygdala volume in the Val/Val group and positive correlations with the amygdala in Met carriers (Fig. 3A). The significant differences were localized to the left insula (Met Carrier z = 0.43, Val/Val z = −0.18), left cuneus (Met Carrier z = 0.41, Val/Val z = −0.19), left middle temporal gyrus (MTG) (Met Carrier z = 0.48, Val/Val z = −0.12), right subgenual cingulate (Met Carrier z = 0.56, Val/Val z = −0.16), and right precuneus (Met Carrier z = 0.46, Val/Val z = −0.22) (Fig. 3B; 5% FDR correction). Results were similar in a model that additionally adjusted for nonlinear age effects, population stratification and total brain volume (Supplementary Fig. 1). Strong pairwise correlations in thickness between the left MTG and left insula suggest that these regions may be part of a common network (Supplementary Fig. 2).
Amygdala–cortical covariance patterns and differences by BDNF Val66Met genotype in adolescents. (A) Vertex-wise covariance patterns in the Val/Val group (top) and Met carrier group (bottom) within the middle 2 age quartiles that correspond to adolescence (age 12–18). Colors mapped onto the cortex correspond to strength of correlation of that cortical vertex with the right amygdala with warm colors reflecting a positive correlation and cool colors reflecting a negative correlation (see color bar). (B) The location of significantly different vertex-wise cortical relationships with the amygdala are shown in purple (5% FDR corrected). Right amygdala volume (x-axes) is plotted against mean thickness in significantly different clusters localized to the cuneus, insula, MTG, sACC, and precuneus (y-axes). Residuals after regressing for age and sex are plotted so that plotted points reflect relative values. Shaded areas around the regression lines represent 95% confidence intervals. sACC, subgenual anterior cingulate cortex. MTG, middle temporal gyrus.
Sex-Specific Effects of BDNF Val66Met on Structural Covariance in Adolescent Females (Age 12–18)
To evaluate distinct contributions of males and females to these divergent amygdala–cortical covariance patterns, genotype-specific correlations between right amygdala volume and vertex-wise cortical thickness were assessed separately in males and females age 12–18 (Fig. 4A). Significant differences between BDNF genotype groups were observed in females for correlations between the amygdala and the left cuneus (Met Carrier z = 0.62, Val/Val z = −0.25), left precuneus (Met Carrier z = 0.52, Val/Val z = −0.29), left insula (Met Carrier z = 0.60, Val/Val z = −0.26), left lateral frontal cortex (Met Carrier z = 0.42, Val/Val z = −0.40), and left superior temporal gyrus (STG) (Met Carrier z = 0.48, Val/Val z = −0.33) (Fig. 4B; 5% FDR correction). In males no differences were detected between genotype groups (minimum FDR corrected P = 0.91).
Amygdala–cortical covariance patterns and differences by BDNF genotype in male and female adolescents. (A) Vertex-wise covariance patterns in males (top panel) and females (bottom panel) in the Val/Val (top row) and Met carrier (bottom row) groups within the middle 2 age quartiles that correspond to adolescence (age 12–18). Colors mapped onto the cortex correspond to strength of correlation of that cortical vertex with the right amygdala with warm colors reflecting a positive correlation and cool colors reflecting a negative correlation (see color bar). (B) The location of significantly different vertex-wise cortical relationships with the amygdala in females are shown in purple (5% FDR corrected). Right amygdala volume (x-axes) is plotted against mean thickness in significantly different clusters localized to the cuneus, precuneus, insula, frontal cortex, and STG (y-axes). Residuals after regressing for age and sex are plotted so that plotted points reflect relative values. Shaded areas around the regression lines represent 95% confidence intervals. STG, superior temporal gyrus.
Follow-Up BDNF Val66Met-Functional Connectivity Analysis
Effects of BDNF Val66Met on Functional Connectivity in Similar Circuitry in Adolescents (Age 12–18)
Resting-state functional connectivity between the right amygdala and 5 cortical regions where structural covariance differences were identified was examined in the subset of adolescent participants age 12–18 with high quality functional imaging data (Met Carrier n = 32, Val/Val n = 68). Functional connectivity differed based on BDNF genotype group between the amygdala and 3 of the 5 cortical regions examined (Fig. 5). Significantly stronger connectivity was seen in the Met allele carriers compared to the Val/Val individuals between the right amygdala and the right subgenual cingulate (Met Carrier mean z = 0.17, Val/Val mean z = 0.01, P = 0.01), the left insula (Met Carrier mean z = 0.21, Val/Val mean z = 0.10, P = 0.05), and the left MTG (Met Carrier mean z = 0.40, Val/Val mean z = 0.26, P = 0.02). No significant differences in functional connectivity were observed between the amygdala and the left cuneus and right precuneus (P > 0.50) or when females and males were examined separately (all P > 0.20). Results did not change when covariates accounting for nonlinear age effects, population stratification, and in scanner motion were included in the statistical model (Supplementary Table 3). Comparing alternate methods for assessing functional connectivity between brain regions demonstrated that the most consistent difference between genotypes was in functional connectivity between the right amygdala and the right subgenual cingulate (Supplementary Table 4). As observed with structural covariance strong BOLD signal correlations between the left MTG and left insula suggest that these regions may be part of a common network (Supplementary Fig. 3).
Amygdala–cortical functional connectivity in adolescents. Functional connectivity (z-values) in adolescent Met allele carriers (red) and Val/Val individuals (blue) is plotted along with mean values (horizontal line). The group of Met allele carriers showed statistically stronger functional connectivity than the Val/Val group in the insula, MTG, and sACC. MTG, middle temporal gyrus; sACC, subgenual cingulate cortex.
Follow-Up BDNF Val66Met-MDD Analysis
BDNF Val66Met Associated MDD Risk in Adolescent Females Versus Males (Age 12–18)
Relationship of the BDNF Val66Met with MDD was assessed in 1315 adolescent participants age 12–18 (Met/Met n = 43, Val/Met = 396, Val/Val n = 876; Supplementary Fig. 4). In the female subgroup, BDNF genotype was associated with MDD occurrence at trend-level significance, according to the recessive model (P = 0.07) but not the additive or dominant models (n = 686). On average, female BDNF Met homozygotes were more likely to meet criteria for MDD than Val allele carriers (OR = 3.2, 95% CI = [1.24, 8.46]). No genotype effects observed in males approached significance (n = 629; all P > 0.9).
Discussion
Examination of amygdala–cortical covariance based on the functional Val66Met polymorphism in the BDNF gene revealed age- and sex-specific differences in a depression vulnerability network during development. Increased amygdala–cortical coupling was present in Met allele carriers compared with individuals with 2 copies of the Val allele, and this difference was most pronounced in adolescent females. In adolescents, these differences were specific to relationships between the amygdala and the insula, cuneus, temporal gyrus, subgenual cingulate, and precuneus with additional cortical differences in the lateral frontal cortex when examining females only. Follow-up analyses revealed that: (1) increased amygdala–cortical covariance corresponded to enhanced functional connectivity with the amygdala in adolescent Met allele carriers in the subgenual cingulate, insula and MTG, and (2) having 2 copies of the Met allele increased the likelihood that adolescent females met criteria for MDD. These findings suggest that the BDNF Val66Met influences the development of amygdala–cortical circuits in a sex-specific and developmentally timed manner.
Plasticity processes in the brain are especially dynamic during adolescence, a period of development when the onset of mood disorders often occurs (Beesdo et al. 2009). In this study, genotype-specific differences were most pronounced and significant at a vertex-wise level when examining adolescents age 12–18. The brain undergoes significant structural changes between childhood and adulthood with continued growth of subcortical regions including the amygdala (Wierenga et al. 2014) and thinning of the cortex (Shaw et al. 2008). Altered structural covariance emerging during this period may reflect diverging patterns of coordinated development between the amygdala and cortex mediated by differing connectivity based relationships. The Met allele substitution in the BDNF protein results in impairment in the dendritic trafficking and synaptic localization of the protein as well as a reduction in activity-dependent BDNF secretion (Egan et al. 2003). Increased covariance and functional connectivity associated with the “reduced plasticity” version of the BDNF protein seen here suggests that perhaps these individual's brains have a reduced ability to develop regional specificity that may be protective against mood disorders. Consistent with the developmentally specific effects of this polymorphism it has been shown that the Met allele alters contextual fear learning during adolescence in mice (Dincheva et al. 2014).
These results suggest that genotype-specific amygdala–cortical structural coupling differences were more pronounced in females than in males. Healthy brain development differs between the sexes; white matter, the cortex, and subcortical structures, particularly ones that have a relatively high density of sex steroid receptor such as the amygdala, have been shown to differ in morphology and trajectory of development (Giedd et al. 2012). Recently, sex differences during development in the structural connectome have been shown (Ingalhalikar et al. 2014). However, for a different perspective that calls the extent of some of these sex-based differences into question, see Joel (2011); Joel et al. (2015). Common genetic variation implicated in brain plasticity may differentially affect these developmental patterns contributing to the predisposition of males or females to certain mental illnesses. A previously published factor analysis of psychopathology domains in the same developmental sample demonstrated that being female had a high association with the “anxious-misery” factor (Calkins et al. 2015), consistent with studies that demonstrate that by mid puberty females are more likely to develop mood disorders than males (Angold et al. 1998). While there is some data suggesting a relationship between depression and the Met allele in males (Verhagen et al. 2010), others have shown that the Met allele is associated with depression among females (Lavebratt et al. 2010). Additionally, female mice with 2 copies of the Met allele from human Val66Met BDNF polymorphism show increased anxiety-like behaviors as well as significant fluctuations in anxiety-like behaviors over the estrous cycle (Bath et al. 2012). This sex-specific vulnerability may be explained by interactions between hormonal fluctuations during the estrous cycle and BDNF. In response to estrogen, the expression or release of BDNF is significantly increased (Begliuomini et al. 2007) and there is evidence that BNDF and estradiols have interactive effects on brain function (Sato et al. 2007), as well as mood and cognition (Cubeddu et al. 2011). Therefore, the BDNF Val66Met polymorphism may mediate differences in hormone-related behavioral changes during adolescence when hormone cycling begins with onset of puberty.
Dysfunction of the amygdala–cortical networks identified in this study have been demonstrated in populations of adolescents with MDD. Among the cortical regions identified, the subgenual cingulate and the insula—2 of the regions where the structural covariance results were supported by corresponding functional connectivity differences—are the most reproducibly found cortical regions to be disrupted in MDD. These are also 2 of the cortical regions with the highest levels of BDNF expression (Hawrylycz et al. 2012). The subgenual cingulate has reciprocal cortical connections with the amygdala, is involved in the production of sad emotions and antidepressant treatment response (Ressler and Mayberg 2007) and has clinical importance as a target site for deep brain stimulation (Holtzheimer and Mayberg 2012). The insula also has extensive reciprocal connections with the amygdala, and the anterior ventral portion that is identified in this study is specifically involved in emotion and social emotional tasks (Kurth et al. 2010). Several recent task-based fMRI studies have interrogated functioning of circuits during emotion processing paradigms in depressed adolescents showing altered amygdala-subgenual cingulate functional connectivity (Ho et al. 2014) and increased amygdala–anterior insula connectivity (Henje Blom et al. 2015). Resting-state fMRI studies have also shown altered functioning of these circuits in depressed adolescents (Connolly et al. 2013). Studies have additionally implicated altered structure (Korgaonkar et al. 2014) and functioning (Cullen et al. 2014) of the default mode network (in which the precuneus identified in this study is a prominent node) in adolescent MDD. In this the current study, structural covariance results in the cuneus and precuneus were not supported by functional connectivity differences in these same circuits, however, this may be due to the smaller sample size for the follow-up analysis. Despite the increased prevalence of MDD in females previous studies have not demonstrated sex-specific effects possibly due to limiting sample sizes. Dysfunction of the identified circuits in depressed adolescents provides evidence that BDNF Val66Met exerts neural risk for MDD.
Some limitations to our analysis and its interpretation should be noted. Notably BDNF plays widespread complex roles in plasticity-related processes in many brain regions including hippocampal circuits and the Val66Met polymorphism has been implicated in different brain disorders. Moreover the allele (Met vs. Val) that confers risk may change across development, differing in trajectory for various disorders. Furthermore, other candidate genes (e.g., 5-HTTLPR and COMT) have been shown to affect amygdala cortical functions. The specificity of the results to adolescence is consistent with the transition through puberty; though in the future the impact of pubertal status on these relationships needs to be further understood. An additional limitation to the interpretation of these results is that they could be attributed to a locus in high linkage disequilibrium with the BDNF locus or hidden population stratification. In the follow-up analysis of corresponding functional connectivity differences between genotype groups, alternate methods for describing function connectivity between regions produced some divergent results. It should also be noted that participants with moderate or severe, but not mild, medical co-morbidity were excluded from the analyses, which should be considered given the potential impact of medical conditions and medications on brain structure and function. While multiple comparison correction was employed within phenotypes (e.g., for number of cortical thickness vertices), we did not correct for tests across phenotypes, and our results should be taken in the context of this limitation. Moreover, the imaging analyses were performed in subsamples of the larger community sample. Ideally, where feasible, future studies could attempt to acquire all measures in each participant to allow for stronger conclusions across phenotypic analyses. Additionally, as in all studies that examine a single sample, replication of these effects in an independent sample would provide greater confidence in their reliability. Lastly, the frequency of Met allele homozygotes in the imaging sample was too low to examine these individuals as a separate group and so they were grouped with the heterozygote individuals, which may have obscured some important effects. We would have preferred to analyze differences among each genotypic group in the imaging sample, particularly given data from both human and animal studies that support met allele homozygotes as a group with particularly strong impact on imaging (Egan et al. 2003) and behavioral phenotypes (Schofield et al. 2009). It is not necessarily surprising that the observed effects were stronger on the intermediary imaging phenotype than in the behavioral phenotype. Nor was it unexpected that 2 Met alleles were required to demonstrate increased risk for depression, since the effects of functional polymorphisms may be more penetrant in relation to brain phenotypes versus behavior (Meyer-Lindenberg 2010; Voineskos et al. 2011).
Our results implicate the BDNF Val66Met polymorphism in the development of the adolescent “female brain” that is also associated with increased risk for a mood disorder. The novelty of our work lies in the different genetically mediated networks present between males and females at the sensitive, high-risk period of adolescent brain development. The disruption of the identified amygdala-based brain circuits in MDD suggests that this common genetic influence on the brain and disease may be linked. Thus genetic susceptibility for mood disorders during adolescence may be associated with the genotype-related alterations in anatomy and function of amygdala–cortical circuits important for emotion processing.
Supplementary Material
Supplementary material is available at Cerebral Cortex online.
Funding
A.L.W. and A.N.V. are supported by The Canadian Institute of Health Research and the Brain and Behavior Research Foundation A.N.V. is supported by the Canada Foundation for Innovation, the National Institute of Mental Health (R01MH099167 and R01MH102324), and the Ontario Ministry of Research and Innovation Support for the collection of the data sets was provided by grant RC2MH089983 awarded to Raquel Gur and RC2MH089924 awarded to Hakon Hakonarson.
Notes
Conflict of Interest: None declared. All participants were recruited through the Center for Applied Genomics at The Children's Hospital in Philadelphia. dbGaP Study Accession: phs000607.v1.p1.
References
