Pathological intrinsic brain activity specific to major depressive disorder (MDD) could be encoded in temporal dynamics of BOLD resting-state networks (RSNs). Resting-state is not a stable state but undergoes dynamic state changes at different time scales¹. Previous research indicated a different temporal dynamic of RSN in MDD, with default-mode network (DMN) activity lasting longer for MDD². This study investigated temporal fluctuations of multiple RSNs in multiple frequencies using wavelet variance analysis to further elucidate differences in RSN temporal dynamics between MDD and healthy control (HC). Additionally we aim to identify a brain activity associated with RSNs state change and abnormal low fluctuations in DMN in MDD.

Forty-five HC (12 male, age 21–55) and 44 MDD (12 male, 20–55, diagnosed by DSM-IV-TR) participated in 7m30s resting-state fMRI session with eyes open. Whole-brain EPI (TR/TE=2000/30ms, FA=90°, FOV/slice=240/3.2mm×41 axial slices, matrix=128×128, SENSE acceleration=2) was acquired with 3T GE MR750 scanner. FMRI analysis using AFNI includes; discarding the first three volumes, despiking, RETROICOR/RVT physiological noise correction, slice-timing and motion correction, normalize to MNI template, smoothing with 4mm-FWHM kernel, scaling to percent change, regressing out mean CSF signal, motion parameters and 3rd-order polynomials. Group independent component analysis (ICA) was applied to the processed images using melodic software in FSL³. Forty-four ICs were extracted and 18 ICs were identified as RSNs (visual inspection). Activation time courses for the RSNs in individual subject were reconstructed by dual regression⁴. Wavelet variance analysis (WVA) was applied to the extracted activation time series. WVA indicates how much the signal time course fluctuated at a certain frequency. Up to fourth scales of wavelet variance, corresponding to 0.125-0.25Hz, 0.0625-0.125Hz, 0.0313-0.0625Hz, and 0.0156-0.0313Hz, were extracted with maximum overlap discrete wavelet transformation and Symlet4 wavelet filter⁵. Log-transformed wavelet variances for each RSN in each scale were entered into linear mixed-effect model (LMM) analysis with fixed effects of diagnosis, RSN, and frequency along with age, gender, and motion size covariates. Post-hoc analysis comparing MDD to HC was also performed for each RSN in each frequency scale.

We also investigated a brain activation associated with a state change of RSN activity. Local maxima and minima in the wavelet coefficient time course were extracted as event onsets (Fig. 1). Brain activation correlated with these events was extracted with a general linear model analysis. The design matrix included modeled responses for the onset of local maxima and minima (onset times were shifted 4.5s to account for hemodynamic delay) with a canonical hemodynamic response function and parametric modulation by extremum size, which could discount the effect of small fluctuations. The activations associated with local maxima and minima were compared between MDD and HC. This analysis was performed only for the RSNs/frequencies that showed significant difference in wavelet variance between MDD and HC.

Fig. 2 shows the ICs/ frequencies that showed significant difference in wavelet variance between MDD and HC (corrected p<0.05). Fluctuations of IC3 and IC6 were significantly lower at 0.031-0.125Hz for MDD. Fluctuations of IC20 and IC25 were significantly higher in 0.125-0.25Hz and in 0.0156-0.031HZ for MDD. Fig 3 shows spatial pattern of these ICs. The ICs correspond to the high visual (IC3), default-mode (IC6), auditory/language (IC20), and cerebellum (IC25) networks, respectively. The analysis for a state-change-related activation revealed significantly higher precuneus activity in MDD at minimum point of IC6 (DMN) in 0.0313-0.0625Hz (voxel-wise p<0.005 and cluster-size corrected p<0.05).MDD subjects showed less fluctuation in DMN activity during rest, consistent with previous reports showing longer DMN dominance time for MDD². This suggests that MDD subjects tend to be locked in the default-mode state at rest. This dominance of default-mode activity could be associated with high precuneus activity at the minimum point of DMN activity. Precuneus has been suggested as a hub region switching between task-positive and default-mode network activity^6,7. Our finding indeed suggests that high precuneus activity at minimum points of DMN activation might serve as trigger to increase and to switch to DMN activity, and that might cause subjects with MDD to stay longer in default-mode state at rest.Novel analysis of temporal dynamics of resting-state activity in MDD suggests that high precuneus activity can trigger a switch to default-mode activity and prevent breaking out of such activity. This finding sheds light on brain network mechanisms of MDD pathophysiology.