We used rs-fcMRI to examine DMN and DMN-CEN functional connectivity in individuals from a longitudinal cohort study of families at high and low risk for depression. Based on prior literature (Sheline et al, 2009; Posner et al, 2013), we hypothesized that compared with individuals at low familial risk, those at high risk for depression would demonstrate increased connectivity within the DMN, as well as decreased negative connectivity (ie, reduced anticorrelations) between the DMN and CEN. Using diffusion MRI, we then explored whether measures of DMN and DMN-CEN functional connectivity could be corroborated with diffusion tractography. Last, we used path analyses to explore associations between familiar risk for depression, connectivity measures, and affective symptoms and impulsivity.
The Institutional Review Board of the New York State Psychiatric Institute (NYSPI) approved the study procedures. Adult participants provided informed consent; minors provided informed assent, and a parent/guardian provided consent.
The familial depression study began in 1982; complete details on study design, sample selection, and assessments are reported elsewhere (Weissman et al, 2005). Briefly, risk status for depression was defined based on the first generation (G1), such that offspring (generations 2 and 3, G2 and G3) were defined as high risk if G1 had a history of MDD, and were otherwise defined as low risk. The high- and low-risk families have been followed prospectively for up to 30 years and up to six time points (‘Waves’) affording exceptional psychiatric assessments of all study participants (Table 1). The current study is based on data collection at Wave 6. Diagnostic interviews were conducted using the Schedule for Affective Disorders and Schizophrenia-Lifetime Version (the adult version for participants over age 18 years, and the child version for participants 6-17 years of age) at up to six time points over up to 30 years. (See Supplementary Materials for further details on assessment procedures.)
We obtained MRI scans from 111 descendants of G1 families, aged 11-60 years. MRI scans from 7 individuals were excluded because of excessive head motion and/or imaging artifacts, leaving 104 individuals available for group comparisons. Of these, 57 participants comprised the high-risk group and 47 the low-risk group. Participants were group matched on sex and age (Table 1). All participants were Caucasian and G1 participants were all drawn from the same community. Exclusion criteria consisted of psychotic symptoms, pregnancy, and MRI contraindications.
Images were acquired on a GE Signa 3.0 T whole-body scanner using an 8-channel head coil. During resting state acquisition, participants were instructed to remain still with their eyes closed and to let their minds wander freely. Two 9-min resting state scans were obtained for each participant. Diffusion MR images were acquired in two runs with diffusion weighting along 15 non-collinear directions. (Further details on MRI pulse sequences are provided in the Supplementary Materials.)
As described elsewhere (Posner et al, 2013; Posner et al, 2014), standard image preprocessing methods were used, employing SPM8 software (http://www.fil.ion.ucl.ac.uk/spm/). Briefly, functional images were slice time and motion corrected, coregistered with a high-resolution anatomical scan, normalized to Montreal Neurological Institute space, resampled at 3 mm, and smoothed with a Gaussian kernel of 8 mm FWHM (Friston et al, 1995). Connectivity processing consisted of independent component analysis (ICA) and a hierarchical partner matching algorithm (Wang et al, 2011) to isolate a network of regions corresponding to the DMN. Partner matching is a clustering algorithm that identifies ICA-derived independent components that share spatial properties across subjects (Wang and Peterson, 2008).
To examine the confounding influence of head motion upon connectivity measures, we calculated root mean square and peak/average (across volumes) framewise displacement (FD), which is based on each individual’s head alignment parameters from SPM’s realignment procedure. We differentiated the six head realignment parameters across frames and then calculated instantaneous head motion as a scalar in each frame using the following formula: FDi=|Δd|+|Δd|+|Δd|+|Δα|+|Δβ|+|Δγ|, where Δd=d — d, and similarly for the other rigid body parameters [d d d α β γ]. We converted rotational displacements from degrees to millimeters by calculating displacement on the surface of a sphere of radius 50 mm (Power et al, 2012). Group differences in the motion parameters were tested using the non-parametric Kolmogorov-Smirnov test given non-normality and existence of potential outliers. There were no group differences in any of these parameters (Supplementary Figure 2). Including the motion parameters as covariates in hypothesis testing did not alter the study findings.
Each participant’s DMN functional connectivity map (ie, the DMN component from ICA), served as the dependent variable in second-level, random effects factorial models with group (ie, high- vs low-risk) as the independent variable. Age, sex, generation, familial relatedness based on a kinship coefficient (Blumenthal and Cannon-Albright, 2008), prior medication exposure, and history of depression, anxiety, or substance use disorder were included as covariates. Regions with positive functional connectivity with the DMN component were indexed within DMN connectivity; conversely, we examined connection strength between the DMN component and the CEN based on predefined masks (Seeley et al, 2007). To control for multiple statistical comparisons, for any cluster to be considered statistically significant, the cluster had to contain at least 25 neighboring voxels, with each voxel in the cluster meeting an α of 0.01. The combined application of a voxel level statistical threshold and cluster filter minimizes the false-positive identification of regions at any given threshold (Forman et al, 1995) because clustering can distinguish between true connectivity between regions and noise that has less tendency to cluster (Posner et al, 2014).
To examine structural connectivity of the DMN and DMN-CEN circuitry, we performed probabilistic tractography on diffusion MRI data, as described elsewhere (Cha et al, 2015). Diffusion MRI data were processed with the FMRIB’s Diffusion Toolbox (Smith et al, 2004) in FSL 5.0. The preprocessing pipeline includes skull stripping, eddy current correction, B-matrix rotation, affine registration of the T1-weighted, diffusion-weighted images, and FreeSurfer segmentation and parcellation images. Multi-fiber probabilistic diffusion modeling was performed next using a Bayesian Estimation of Diffusion Parameters Obtained using Sampling Techniques with Crossing Fibers modeling (bedpostx) program (Behrens et al, 2007). To maximize accuracy of Markov Chain Monte Carlo sampling, a burn-in sample size was set conservatively to 1000. White matter tracts were then estimated using Probabilistic Tracking with Crossing Fibers (probtrackx2) in FSL 5.0 (Behrens et al, 2003). We used each individual’s precuneus segmentation mask (conducted with FreeSurfer) as a seed (each hemisphere separately) and two dorsal prefrontal ROIs as target regions. We paid special attention to avoid false positive tracts in our tractography analyses. Specifically, we first created an exclusion mask from each individual’s segmentation masks consisting of the cerebral spinal fluid and all the ventricles. In addition, to effectively account for gyral and sulcal configurations when estimating tracts going through different cortices, we created a binary mask by inverting each individual’s cortical parcellation in the FreeSurfer suite. Any tracts arriving at this exclusion mask were discarded. Five thousand streamline samples were generated for each tractography run from each ROI voxel to build a connectivity distribution. We calculated the number of streamline samples from the seed mask, successfully arriving at the target ROI mask proportional to the total number of samples. We first used individual’s cortical parcellation masks as the target ROIs and then excluded ROIs whose probabilistic measures did not reach a threshold of 0.02% of the total estimated streamlines; this is a commonly used threshold in prior studies (Forstmann et al, 2011; Li et al, 2012; Chowdhury et al, 2013). Thus, the connectivity measures represent ROI-to-ROI probabilistic connectivity. Tractography analyses were performed on a Linux-based high-performance computing system at Columbia University’s Advanced Research Computing Services. For group-wise comparisons, we used factorial models analogous to those used in our hypothesis testing and restricted the tractography analysis to connections in which group differences were detected during hypothesis testing. We used false discovery rate (FDR) to correct for multiple comparisons.
We used path analysis to test two models: (i) DMN connectivity mediating a relationship between familiar risk for depression and depressive symptoms and (ii) DMN-CEN connectivity mediating a relationship between familiar risk for depression and impulsivity. Following established methods for mediation (Rucker et al, 2011), the path analyses were conducted using a series of linear regression models. For DMN connectivity, we tested whether familial history of depression (independent variable) influenced DMN connectivity (dependent variable). In a second regression model, we then tested whether DMN connectivity (independent variable) influenced depressive symptoms (dependent variable) while controlling for family history. For DMN-CEN connectivity, the same analytic approach was used except DMN-CEN connectivity was the dependent variable in the first regression model and impulsivity was the dependent variable in the second regression model. Impulsivity was assessed with the Continuous Performance Task II commission errors (CPT-II, Commissions; Conners and Staff, 2000). A complete description of the CPT-II is provided in the Supplementary Materials. The path analyses were conducted using SPSS (SPSS Inc., Chicago, IL). DMN and DMN-CEN connectivity were extracted from SPM (http://www.fil.ion.ucl.ac.uk/spm/). Statistical significance was determined using the Sobel test. Age and sex were included as covariates.
Potential confounds of the study were considered. First, anxiety symptoms were greater in the adults within the high- vs low-risk group (Table 1). Second, although all study participants were examined using the same MRI platform, 11 of the study participants (6 from the high-risk group and 5 from the low-risk group) were scanned at a different site because of renovations at the study’s primary site (although both sites used a GE Signa 3.0 T, whole-body scanner, 8-channel head coil). Third, the number of ICA-components generated for each individual could influence our hypothesis testing. To address the first potential confounds, we added the following covariates to our hypothesis testing: anxiety and depressive symptom severity. Anxiety symptoms were assessed with the Hamilton Anxiety Rating Scale and the Revised Children’s Manifest Anxiety Scale for adults and children, respectively. Depression symptoms were assessed with the Hamilton Rating Scale for Depression (Hamilton, 1960) and the Children’s Depression Rating Scale-Revised (Poznanski and Mokros, 1996) for adults and children, respectively. Because the assessment measures for depressive and anxiety symptoms differed in adults vs children, symptom severity scores were transformed into z-scores. Second, to exclude the possibility that the study findings were confounded by differences in MRI scanners, we excluded the 11 participants who were scanned at the alternate site. These sensitivity analyses did not meaningfully alter the study findings (Supplementary Materials). Third, the total number of ICA-components generated for each individual did not differ between the two groups (p=0.6). Moreover, subgroup analysis excluding participants with ICA-components outside of the 95% CI of the group mean did not alter our hypothesis testing (Supplementary Materials).
