If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Poitras Center for Affective Disorders Research at the McGovern Institute for Brain Research and Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, CambridgeDepartment of Psychiatry, Harvard Medical School, BostonClinical and Research Program in Pediatric Psychopharmacology, Massachusetts General Hospital, BostonInstitute for Medical Engineering and Science, Massachusetts Institute of Technology, Cambridge, Massachusetts
Neuroimaging studies of patients with major depression have revealed abnormal intrinsic functional connectivity measured during the resting state in multiple distributed networks. However, it is unclear whether these findings reflect the state of major depression or reflect trait neurobiological underpinnings of risk for major depression.
We compared resting-state functional connectivity, measured with functional magnetic resonance imaging, between unaffected children of parents who had documented histories of major depression (at-risk, n = 27; 8–14 years of age) and age-matched children of parents with no lifetime history of depression (control subjects, n = 16).
At-risk children exhibited hyperconnectivity between the default mode network and subgenual anterior cingulate cortex/orbital frontal cortex, and the magnitude of connectivity positively correlated with individual symptom scores. At-risk children also exhibited 1) hypoconnectivity within the cognitive control network, which also lacked the typical anticorrelation with the default mode network; 2) hypoconnectivity between left dorsolateral prefrontal cortex and subgenual anterior cingulate cortex; and 3) hyperconnectivity between the right amygdala and right inferior frontal gyrus, a key region for top-down modulation of emotion. Classification between at-risk children and control subjects based on resting-state connectivity yielded high accuracy with high sensitivity and specificity that was superior to clinical rating scales.
Children at familial risk for depression exhibited atypical functional connectivity in the default mode, cognitive control, and affective networks. Such task-independent functional brain measures of risk for depression in children could be used to promote early intervention to reduce the likelihood of developing depression.
Neuroimaging in patients with major depression (major depressive disorder [MDD]) has revealed abnormal activation patterns in multiple brain networks, including the default mode network (DMN) and cognitive control and affective networks. The DMN, anchored in the medial prefrontal cortex (mPFC) and posterior cingulate cortex (PCC), is suppressed in healthy adults during tasks that demand external attention but does not show the typical pattern of task-induced deactivation in adults and adolescents with MDD (
). The cognitive control network, including the dorsolateral prefrontal cortex (DLPFC), which is typically activated during cognitively demanding tasks, has shown decreased activation in adults with MDD (
Mirroring these brain activation abnormalities, patients of different ages with MDD have shown abnormal intrinsic functional connectivity of the brain measured via resting-state functional magnetic resonance imaging (rs-fMRI) (
) with MDD and is thought to reflect deficits in emotion regulation.
Despite evidence of abnormal functional connectivity across distributed brain networks in patients with MDD, it is unclear whether these differences reflect the state of current depression versus neurobiological traits that predispose individuals to be at risk for MDD. One approach to distinguishing between current state and predisposing traits is the study of unaffected individuals at heightened risk for MDD, such as unaffected children at familial risk for MDD by virtue of having a parent with MDD. Such familial history increases the risk of MDD in offspring by threefold to fivefold (
). Understanding whether rs-fMRI findings represent trait or state markers of MDD in the young can lead to the identification of informative neural biomarkers of risk for mood and anxiety disorders and help develop early intervention strategies to mitigate this risk. Resting-state fMRI also possesses significant translational strengths in its short duration of scanning and the lack of task performance demands that can complicate interpretation of activations.
In the present study, we examined rs-fMRI in unaffected children at familial risk for MDD and other mood and anxiety disorders by virtue of being offspring of parents with MDD (at-risk group) and compared them with age-matched children who were offspring of parents with no lifetime history of any mood disorder (control group). Two previous studies examining at-risk children and adolescents found decreased connectivity between the amygdala and frontal-parietal network in unaffected children of depressed mothers and in children with early-onset depression (
Based on previous functional connectivity results in patients with MDD, we focused on functional connectivity differences between at-risk and control children in the DMN, the cognitive control network, and the affective network, using a seed-based functional connectivity approach. We examined connectivity differences from the two midline anchor regions of the DMN (mPFC and PCC), which are associated with self-referential processing (
), and from seed regions in left and right DLPFC and amygdala. We tested 1) whether unaffected at-risk children exhibit patterns of abnormal functional connectivity similar to those reported in patients with MDD, and 2) whether connectivity of DMN-sgACC is related to symptom scores in at-risk children. To further test whether resting-state connectivity can be a useful neural biomarker for risk for MDD, we built classification models based on resting-state data to discriminate at-risk versus control children.
Methods and Materials
We initially recruited 38 offspring, 8 to 14 years of age, of parents with a lifetime history of MDD (at-risk group) and 30 age-matched offspring of parents with no lifetime mood disorder (control group). The study was approved by the Institutional Review Boards at Massachusetts General Hospital and at Massachusetts Institute of Technology. Parents provided written informed consent for their and their child’s participation, and youths provided written assent. Exclusion criteria included the presence of acute psychosis or suicidality in a parent or a child, the presence at any point in the life span of bipolar disorder in the parent, autism in the child, or a lifetime history of a traumatic brain injury or neurological disorder in the child.
The final sample included in the analyses consisted of 27 at-risk and 16 control participants with no prior history of depression or current clinical-range symptom scores. Participants who did not complete the scan, had excessive head movement during the scan, or had a history of depression or clinical-range symptom scores were excluded. See Supplement for details.
At enrollment for the present study, each child and both parents in each family were assessed for current and lifetime mood disorders (MDD, bipolar disorder, and dysthymia), using structured clinical interviews in which the mother was the informant. Interviews about parents used the depression, mania, dysthymia, and psychosis modules from the Structured Interview for DSM-IV (
) and those about the child used the depression, mania, dysthymia, and psychosis modules from the Schedule of Affective Disorders and Schizophrenia for School-Aged Children–Epidemiological Version for DSM-IV (
) (see Supplement for details) about all children. The CBCL includes a total problems score, as well as scores reflecting internalizing (affective and anxiety) and externalizing symptoms (attentional problems and disruptive behavior). T-scores of 70 and above have been shown to discriminate clinical-range from nonclinical-range children (
) to all children. See Supplement for details of the CDI.
Children in the at-risk and control groups did not differ significantly in age, gender distribution, or IQ (ps > .3) (Table 1). The at-risk group had marginally higher CBCL total (p = .05), internalizing (p = .096), and anxiety (p = .08) scores but did not differ significantly in CBCL external problem scores (p = .34). None of the children had clinical-range CBCL scores (>70). CDI total scores did not differ significantly between the two groups (p = .26). Additionally, by parent report, the children were largely prepubertal (with the exception of four at-risk and three control children).
Table 1Participant Demographic and Clinical Information
Control (n = 16)
At-risk (n = 27)
11.3 ± 2.14
11.2 ± 1.67
t41 = 0.17, p = .86
8 F, 8 M
13 F, 14 M
χ2 = 0.14, p = .9
117 ± 10.5
120.6 ± 12.0
t41 = 0.99, p = .33
Both Parents Affected
41.0 ± 11.8
48.8 ± 10.0
t35 = 2.07, p = .046
44.3 ± 8.50
50.1 ± 9.83
t35 = 1.71, p = .096
45.1 ± 10.5
47.8 ± 9.30
t35 = 0.96, p = .34
51.5 ± 2.78
55.2 ± 6.56
t35 = 1.79, p = .08
4.33 ± 5.54
6.57 ± 4.64
t35 = 1.16, p = .26
Values are mean ± SD where appropriate.
CBCL, Child Behavior Checklist; CDI, total score on the Child Depression Inventory; F, female; KBIT, Kaufman Brief Intelligence Test; M, male; p, between-group test p value; t(df), between-group t statistic and degrees of freedom.
Data were acquired on a 3T TrioTim Siemens scanner (Siemens, Erlagen, Germany) using a 32-channel head coil. T1-weighted whole-brain anatomical images (magnetization prepared rapid acquisition gradient-echo sequence, 256 × 256 voxels, 1 × 1.3 mm in-plane resolution, 1.3-mm slice thickness) were acquired. After the anatomical scan, participants underwent a resting fMRI scan in which participants were instructed to keep their eyes open and the screen was blanked. Resting scan images were obtained in 67 2-mm-thick transverse slices, covering the entire brain (interleaved echo planar image sequence, T2*-weighted images; repetition time = 6 seconds, echo time = 30 ms, flip angle = 90, 2 × 2 × 2 mm voxels). The resting scan lasted 6.2 minutes (62 volumes). Online prospective acquisition correction was applied to the echo planar image sequence (
) (Supplement). Two dummy scans were included at the start of the sequence.
Functional Connectivity Analysis
Resting-state fMRI data were first preprocessed in SPM8 (Wellcome Trust Center for Neuroimaging, University College London, United Kingdom), using standard spatial preprocessing steps. Images were slice-time corrected, realigned to the first image of the resting scan, resampled such that they matched the first image of the resting scan voxel for voxel, normalized in Montreal Neurological Institute space, and smoothed with a 6-mm kernel (full width at half maximum). Functional connectivity analysis was performed using a seed-driven approach with in-house, custom software CONN (
). We performed seed-voxel correlations by estimating maps showing temporal correlations between the blood oxygen level–dependent signal from our a priori regions of interest (ROIs) and that at every brain voxel. We performed resting-state connectivity analysis from the DMN seeds (mPFC, PCC), cognitive control network seeds (bilateral DLPFC), and bilateral amygdala seeds (Figure 1). The DMN and DLPFC seeds were defined as 6-mm spheres around peak coordinates from Fair et al. (
). See Supplement for details on the aCompCor. A temporal band-pass filter of .008 Hz to .083 Hz was applied simultaneously to all regressors in the model. Residual head motion parameters (three rotation and three translation parameters plus another six parameters representing their first-order temporal derivatives) were regressed out. Artifact/outlier scans (average intensity deviated more than 3 standard deviations from the mean intensity in the session or composite head movement exceeded 1 mm from the previous image) were also regressed out. Head displacement across the resting scan did not differ significantly between the two groups for either frame-to-frame translations in x, y, z directions (at-risk group: mean = .19 mm ± .11; control group: mean = .16 mm ± .11; p = .33) or frame-to-frame rotations (at-risk group: mean = .0044 ± .002; control group: mean = .004 ± .003; p = .66). The number of outliers also did not differ significantly between the groups (range: 0 to 9; at-risk group: mean = 2.7 ± 2.2; control group: mean = 2.1 ± 3.1; p = .47). Outlier images were modeled as nuisance covariates. Each outlier image was represented by a single regressor in the general linear model, with a 1 for the outlier time point and 0 elsewhere.
Time series of all the voxels within each seed were averaged, and first-level correlation maps were produced by extracting the residual blood oxygen level–dependent time course from each seed and computing Pearson correlation coefficients between that time course and the time course of all other voxels. Correlation coefficients were converted to normally distributed Z scores using the Fisher transformation to allow for second-level general linear model analyses. DMN connectivity was calculated from the averages of the time series from mPFC and PCC seeds (
), given their similar connectivity patterns. Functional connectivity of left and right DLPFC were analyzed separately, as were left and right amygdala, due to evidence of differential roles in emotion processing (
). First-level connectivity maps for each participant were entered into a between-group t test to determine connectivity differences for each seed between groups. Cluster-level threshold was set at p < .05 using false discovery rate correction for multiple comparisons (
), with voxelwise t value threshold of 2.42 (df = 41; p < .01). Bonferroni correction was applied to the false discovery rate-corrected cluster-level p values to correct for multiple comparisons of the five a priori seeds tested (DMN, left and right DLPFC, and left and right amygdala). Regions that showed significant connectivity differences between groups were further examined for their connectivity values (significantly above or below zero) using one-sample t tests in each group. Based on prior evidence of DMN-sgACC hyperconnectivity in MDD and its implication in depressive rumination (
), we examined the within-group correlations between DMN-sgACC connectivity values and CBCL scores. Given the higher CBCL total scores in the at-risk group, we retested group differences by including CBCL total scores as a covariate.
Classification Models of At-Risk Children and Control Subject Discrimination
We trained two linear classification models using logistic regression, implemented in machine learning software Weka (University of Waikato, Hamilton, Waikato, New Zealand) (
), to categorize individual participants to the at-risk or control groups based on their rs-fMRI or behavioral data. To create robust prediction models that can be generalized to new cases, we performed leave-one-out cross-validation so that each individual was classified on the basis of data from the other individuals. Specifically, data from all participants except one were used as the training set to build a classification model, and the remaining participant was classified with the model and used as the validation case. This procedure was iterated for each participant and used to estimate specificity/sensitivity from the out-of-sample predictions. In the first model, we used anatomically defined ROIs that were independent from the regions that showed between-group connectivity differences. Connectivity values between the five a priori seeds and 116 clusters defined by the Automated Anatomical Labeling atlas (Université de Caen, Université de Paris 5, France) (
) were estimated and used in the prediction model. We constructed a second model based on CBCL scores (total, internalizing, externalizing, anxiety) to compare with classification accuracies from the model based on rs-fMRI data in anatomically defined ROIs.
Increased Connectivity Between DMN and sgACC/Orbitofrontal Cortex in At-Risk Children
Compared with the control group, the at-risk group exhibited increased positive DMN connectivity with a cluster in the sgACC extending into medial orbitofrontal cortex (OFC) bilaterally (Figure 2A, B; Table 2). Among the at-risk children, connectivity between the DMN and sgACC/OFC correlated significantly and positively with CBCL internalizing scores (at-risk: r = .53, p = .003; Figure 2C) and CBCL total scores (at-risk: r = .39, p = .04); there was no such correlation among the control children. Connectivity strengths within the DMN did not differ significantly between groups.
Table 2Between-Group Connectivity Differences From Default Mode Network, Right Dorsolateral Prefrontal Cortex, Left Dorsolateral Prefrontal Cortex, and Right Amygdala
x, y, z
Default Mode Network Connectivity
At-risk > control
−8, 22, −20
R supramarginal gyrus
64, −40, 26
R mid cingulum
18, −34, 38
Control > at-risk
Right Dorsolateral Prefrontal Cortex Connectivity
At-risk > control
Control > at-risk
42, 28, 22
R inferior parietal lobule
46, −50, 58
Left Dorsolateral Prefrontal Cortex Connectivity
At-risk > control
Medial frontal gyrus
0, −2, 48
Control > at-risk
10, 18, −18
L inferior parietal lobule
−50, −56, 54
L lingual gyrus
−14, −82, −14
R lingual gyrus
32, −70, −14
R superior frontal gyrus
14, 34, 60
R inferior temporal gyrus
60, −14, −20
L inferior temporal gyrus
−60, −14, −20
Right Amygdala Connectivity
At-risk > control
R inferior frontal gyrus
44, 40, 4
42, −40, 16
Control > at-risk
All reported clusters survived Bonferroni correction of p < .05 for the number of seeds tested (five).
Peak coordinates (x, y, z) based on Montreal Neurologic Institute brain.
BA, Brodmann area; DLPFC, dorsolateral prefrontal cortex; k, cluster size in mm3; L, left; OFC, orbitofrontal cortex; p Value, false discovery rate-corrected cluster-level p value; R, right; sgACC, subgenual anterior cingulate cortex; SMG, supramarginal gyrus; STG, superior temporal gyrus; t, peak t value from the cluster (degrees of freedom = 41).
Decreased Anticorrelation Between DMN and Inferior Parietal Lobule in At-Risk Children
Compared with the control group, the at-risk group exhibited higher positive connectivity between the DMN and the right inferior parietal lobule (IPL) (Figure 3; Table 2). Instead of the anticorrelation exhibited in the control group (t15 = −5.99, p = .004), the at-risk group exhibited a positive correlation between the DMN and the right IPL (t29 = 2.25, p = .03).
Decreased Connectivity Within Cognitive Control Network in At-Risk Children
Compared with the control group, the at-risk group exhibited decreased positive connectivity between the right DLPFC seed and the right frontal-parietal control network regions, including the right IPL and the right DLPFC (Brodmann area 46) (Figure 4; Table 2) and decreased connectivity between left DLPFC seed and the left IPL (Table 2).
Decreased Connectivity Between Left DLPFC and sgACC in At-Risk Children
Compared to the control group, the at-risk group exhibited decreased connectivity between the left DLPFC seed and sgACC (bilateral), right lingual gyrus, right superior frontal gyrus, and bilateral inferior temple gyri and increased connectivity between left DLPFC and supplementary motor cortex (Table 2). Left DLPFC and sgACC were anticorrelated in at-risk children only (t29 = −3.36, p = .002; Figure 5).
Increased Connectivity Between Amygdala and Inferior Frontal Gyrus in At-Risk Children
Compared with the control group, the at-risk group exhibited increased connectivity between the right amygdala and both the right inferior frontal gyrus (IFG) and the right supramarginal gyrus (Figure 6; Table 2). Instead of the negative correlations exhibited in the control group, the at-risk group exhibited positive correlations between right amygdala and right IFG (control group: t15 = −3.54, p = .003; at-risk group: t29 = 4.67, p < .001) and between right amygdala and right supramarginal gyrus (control group: t15 = −2.53, p = .02; at-risk group: t29 = 4.53, p < .001). Connectivity from the left amygdala did not differ between the two groups.
Group Differences After Controlling for Symptom Scores
After controlling for CBCL total scores, differences between the at-risk and control groups remained largely similar to the above reported results (Supplemental Table S1).
Classification of At-Risk Children and Control Subjects
The classification model based on connectivity data in ROIs defined from the Automated Anatomical Labeling atlas (Université de Caen, Université de Paris 5) yielded 79% accuracy, 81% sensitivity, and 78% specificity. The model based on CBCL scores yielded only 64% accuracy with 80% sensitivity and 27% specificity.
We found differential intrinsic functional connectivity patterns in unaffected children with familial risk for MDD compared with children without such familial risk in the DMN, the cognitive control network, and the amygdala. At-risk children showed hyperconnectivity between the DMN and the sgACC/OFC. Furthermore, although none of the at-risk children were clinically depressed, DMN-sgACC/OFC connectivity was positively correlated with individual CBCL scores among those children. At-risk children also showed hypoconnectivity within the cognitive control network, lacked the typical anticorrelation between the DMN and the right parietal region, and exhibited lower connectivity between left DLPFC and sgACC. In addition, at-risk children showed hyperconnectivity between amygdala and the right IFG. Finally, classification between at-risk children and control subjects based on resting-state connectivity yielded high sensitivity and specificity. These findings appear to identify trait neurobiological underpinnings of risk for major depression in the absence of the state of depression.
Increased connectivity between DMN and sgACC in at-risk children and the positive correlation between DMN-sgACC connectivity and current symptom scores are consistent with findings reported in adult (
) patients with MDD. The fact that these findings were observed in unaffected children at familial risk for MDD suggests that hyperconnectivity with sgACC is not a consequence or manifestation of MDD but instead may be a biomarker of predisposed risk for MDD. The at-risk children also exhibited an atypical anticorrelation between sgACC and left DLPFC. In line with our finding, stimulation of the sgACC resulted in attenuation of hyperactivation in sgACC and increased activation in previously underactive DLPFC in adults with MDD (
). This dynamic interplay between DMN and cognitive control networks in MDD was examined in a task-based connectivity study. During an external attention condition, adults with MDD exhibited increased DMN connectivity and decreased cognitive control network connectivity (
). The present study suggests that an imbalance between DMN and cognitive control networks is a developmental risk factor for MDD.
With regard to decreased connectivity within the cognitive control regions in at-risk children, a previous study of adolescents with familial risk for depression also reported reduced connectivity between cognitive control regions (
). In that study, lower connectivity in the control network was associated with more severe parental depression symptoms. These results in at-risk children and adolescents are consistent with findings from depressed adults of reduced connectivity in attention control regions including the DLPFC (
). The atypically high level of connectivity between amygdala and emotion regulation and cognitive control regions might reflect emotion dysregulation in MDD.
To test whether intrinsic functional organization of the brain, as measured by rs-fMRI, can be a potential biomarker for risk for depression in children, we performed a classification analysis to discriminate children in the at-risk group and control group based on their resting-state functional connectivity data. This classification based on functional connectivity yielded high accuracy, sensitivity, and specificity in discriminating between children at risk for MDD and control subjects compared with classification based on CBCL scores. Importantly, the rs-fMRI classification was based on analyses that, at the level of each individual child, were independent of the group differences in functional connectivity. Such generalizable and individually robust classification is important if brain measures are to be used for early identification (
). Future prospective and longitudinal studies can determine whether such biomarkers predict which high-risk children progress to MDD and whether early intervention reduces the likelihood of developing MDD. Also, perhaps such biomarkers may be helpful in identifying children at risk for developing depression independent of parental histories of depression.
Our findings need to be viewed in light of some methodological limitations. First, we did not exclude children born prematurely, and premature births can lead to neurological complications. However, we did exclude children with known developmental delays such as autism and intellectual disability. Second, because parental MDD confers a spectrum of risk to offspring (
), the at-risk children were also at risk for anxiety and other disorders. Parents with MDD also have higher rates of comorbid anxiety than the general population. Thus, we cannot rule out that the brain differences we found were due to the risk for anxiety and other disorders in these children. Third, although our sample size of at-risk children (n = 27) was moderate, the control group was small (n = 16). Lastly, our resting-state scans were acquired with a repetition time (TR) of 6 seconds, which is longer than most resting-state fMRI studies so that we could acquire high-resolution whole-brain data (2-mm isotropic voxels) without the use of parallel imaging. A previous study found there was no significant difference in correlation strengths within and between resting-state functional networks when comparing resting scans of TR = 2.5 seconds and TR = 5 seconds and that correlation strengths stabilized with acquisition time of 5 minutes (TR = 5) (
), we observed the typical resting-state network patterns observed in other studies. Nonetheless, an additional issue of the long TR is that cognitive and emotional processes internally initiated at the beginning and the end of each scan can be different. We cannot rule out the possibility that the group difference observed here might be, in part, due to systematic differences in chronometry between the two groups.
The present study consisted of a sample of preadolescent children who were at familial risk for depression but not currently affected with depression, and therefore, functional connectivity differences cannot reflect an expression of depression as could be the case in patients with ongoing MDD. Rather, the differences in intrinsic functional brain architecture likely reflect neural traits that predispose children toward MDD or related disorders. Importantly, we demonstrated that discrimination between at-risk and control children occurred with high sensitivity and specificity based on resting-state functional connectivity. Future studies that track the development of children at familial risk for MDD and determine which children develop MDD or other mood and anxiety disorders are needed to build predictive models based on findings from the present study so as to identify high-risk individuals for early intervention.
Acknowledgments and Disclosures
This work was supported by the Tommy Fuss Fund , the Poitras Center for Affective Disorders Research , the Massachusetts General Hospital Pediatric Psychopharmacology Council Fund , and the National Institutes of Health (NIH) grants to JB (Grant Nos. R01 HD036317 and R01MH050657 ). The original participant recruitment was supported by NIH grants to DHB (Grant Nos. R01 MH636833 and R01 MH/CHD076923 ), and Grant No. R01 MH47077 (PI: Rosenbaum). The funding sources had no involvement in study design; in the collection, analysis, and interpretation of data; in the writing of the report; or in the decision to submit the article for publication.
We thank Gretchen Reynolds, Daniel O’Young, and Jiahe Zhang for their help with collecting imaging data. This research was carried out at the Athinoula A. Martinos Imaging Center at the McGovern Institute for Brain Research at the Massachusetts Institute of Technology and Massachusetts General Hospital (MGH).
Dr. Joseph Biederman is currently receiving research support from the following sources: The Department of Defense , American Academy of Child and Adolescent Psychiatry , Alcobra , Forest Research Institute , Ironshore , Lundbeck , Magceutics Inc. , Merck , PamLab , Pfizer , Shire Pharmaceuticals Inc. , SPRITES , Sunovion , Vaya Pharma/Enzymotec , and National Institutes of Health . In 2014, Dr. Joseph Biederman received honoraria from the MGH Psychiatry Academy for tuition-funded continuing medical education courses. He has a US Patent application pending (Provisional Number #61/233,686) through MGH corporate licensing on a method to prevent stimulant abuse. Dr. Biederman received departmental royalties from a copyrighted rating scale used for attention-deficit/hyperactivity disorder diagnoses, paid by Ingenix, Prophase, Shire, Bracket Global, Sunovion, and Theravance; these royalties were paid to the Department of Psychiatry at MGH. All other authors report no biomedical financial interests or potential conflicts of interest.