Oscillatory default mode network coupling in concussion

Background Concussion is a common form of mild traumatic brain injury (mTBI). Despite the descriptor ‘mild’, a single injury can leave long-lasting and sustained alterations to brain function, including changes to localised activity and large-scale interregional communication. Cognitive complaints are thought to arise from such functional deficits. We investigated the impact of injury on neurophysiological and functionally-specialised resting networks, known as intrinsic connectivity networks (ICNs), using MEG. Methods We assessed neurophysiological connectivity in 40 males, 20 with concussion, 20 without, using MEG. Regions-of-interest that comprise nodes of ICNs were defined, and their time courses derived using a beamformer approach. Pairwise fluctuations and covariations in band-limited amplitude envelopes were computed reflecting measures of functional connectivity. Intra-network connectivity was compared between groups using permutation testing, and correlated with symptoms. Results We observed increased resting spectral connectivity in the default mode and motor networks in our concussion group when compared with controls, across alpha through gamma ranges. Moreover, these differences were not explained by power spectrum density (absolute changes in the spectral profiles within the ICNs). Furthermore, this increased coupling was significantly associated with symptoms in the DMN and MOT networks – but once accounting for comorbid symptoms (including, depression, anxiety, and ADHD) only the DMN continued to be associated with symptoms. Conclusion The DMN network plays a critical role in shifting between cognitive tasks. These data suggest even a single concussion can perturb the intrinsic coupling of functionally-specialised networks in the brain and may explain persistent and wide-ranging symptomatology.


Introduction
scan; and no previous incidence of concussion. Every participant in the mTBI group was able to 138 tolerate enclosed space for MR brain imaging; be English speaking; be able to comply with 139 instructions to complete tasks during MEG and MR scans; be able to give informed consent. The 140 control group had no history of TBI (mild, moderate or severe) or neurological disorders. 141 Exclusion criteria for both groups included ferrous metal inside the body that might be classified 142 as MRI contraindications, or items that might interfere with MEG data acquisition; presence of 143 implanted medical devices; seizures or other neurological disorders, or active substance abuse; 144 certain ongoing medications (anticonvulsants, benzodiazepines, and/or GABA antagonists) 145 known to directly or significantly influence electroencephalographic (EEG) findings. 146 All participants underwent brief cognitive-behavioural testing in addition to the MEG 147 resting-state scan. These assessments included: estimates of IQ from the Wechsler Abbreviated 148 Scale of Intelligence (WASI (Wechsler, 1999)); Conner's Attention-Deficit Hyperactivity 149 Disorder Test; the Generalized Anxiety Disorder 7 test (GAD-7); Patient Health Questionnaire 150 (PHQ9); and the Sports Concussion Assessment Tool 2 (SCAT2 (McCrory, 2009)). 151

Procedure and MEG data acquisition 152
Resting-state MEG data were collected whilst participants were lying supine, and 153 instructed to rest with eyes open and maintain visual fixation on an X within a circle on a screen 154 60 cm from the eyes. MEG data were collected inside a magnetically-shielded room on a CTF 155 Omega 151 channel system (CTF Systems, Inc., Coquitlam, Canada) at The Hospital for Sick 156 Children, at 600 Hz for 300 seconds. Throughout the scan, head position was continuously 157 recorded by three fiducial coils placed on the nasion, and left and right pre-auricular points. 158 After the MEG session, anatomical MRI images were acquired using the 3T MRI scanner 159 (Magnetom Tim Trio, Siemens AG, Erlangen, Germany) in a suite adjacent to the MEG. and Visual network VIS). Figure 1A shows the node locations, Figure 1B shows the analysis 177 pipeline. 178  Hz). A beamformer is a spatial filter used to suppress signals from extrinsic neural and noise 193 sources, whilst maintaining unit gain for activity in a target brain location (in this case, the 194 defined seed locations). Individual weight vectors are applied to each sensor measurement and 195 summated to give estimated source activity at target seed locations. This type of spatial filter is 196 also effective at suppressing ocular artefacts generated by eye movements, and non-ocular 197 artefacts, such as cardiac and muscle activity (Cheyne, Bostan, Gaetz, & Pang, 2007;198 Muthukumaraswamy, 2013). 199

Statistical analysis 200
Each of the analyzed frequency ranges from each subject were then submitted to a 201 functional connectivity analysis, by computing amplitude envelope correlations (AEC) across 202 each of the 10 second epochs, based on the instantaneous amplitude estimate of each sample 203 from the filtered time-series calculated using the Hilbert transform. The magnitude of AEC 204 between all pairwise combinations of the seeds varied between 1 (perfect correlation) and -1 205 (perfect anti-correlation). These values quantify the time-varying correlation in the envelope 206 between any two sources, referred to henceforth as functional connectivity. 207 Then, adjacency matrices with AEC values acting as edge weights for all sources pairs 208 were constructed, which resulted in a matrix of weighted undirected graphs in each analysed 209 frequency band for each participant. Connectivity weights between seeds within an intrinsic 210 network were averaged to characterise the magnitude of spontaneous intra-network coupling. 211 These were then either averaged over the 12 epochs (2-minute run) to derive time-averaged 212 connectivity, or the standard deviation was calculated to define temporal dynamism (Koelewijn 213 et al., 2015;Muthukumaraswamy, 2013), albeit in network connectivity, rather than local source 214 oscillatory amplitude. These adjacency matrices were then divided into the respective groups and 215 inferential statistics investigating group differences for mean edge weight were implemented 216 using non-parametric permutation testing (20,000 iterations), which do not require the data 217 distributions to be normal. False positives due to multiple comparisons were controlled using 218 Bonferroni-correction across frequency-bands. Cognitive-behavioural correlation analyses were 219 conducted using the MATLAB Statistics Toolbox (The Mathworks, Inc.). Networks were 220 visualized and figures produced using BrainNet Viewer (Xia, Wang, & He, 2013). 221

Resting-state spectral power does not explain differences in ICN coupling 258
To determine the extent to which changes in intra-network coupling were dependent on 259 oscillatory power, the mean internal power spectrum for each of ICNs was calculated and 260 divided into canonical frequency ranges -qualitative assessment of the spectrum show an 261 al. ta he apparent increase in low-frequency power/an alpha peak shift towards low-frequencies, 262 particularly within the DMN (Figure 3). Mixed ANOVAs on each of the ICNs independently 263 revealed a main effect of band (p < 0.001), as expected, but not of group or any interaction (p > 264 0.05). Post-hoc contrasts between groups within bands revealed no significant differences (all p's 265 > 0.05). This suggests connectivity in the bands exhibiting between group differences is not 266 explained by raw spectral power. correlations and least-squares fit line; test statistics for full and partial correlations are given in 277 Table 2. 278 In addition to the main MEG findings, we also conducted follow-up analyses of the 279 relation between network connectivity and concussion symptom number ( Figure 4 shows scatter 280 plots of connectivity versus symptom presence, and test-statistics for full and partial correlations 281 are detailed in Table 2). Specifically, we examined brain-behaviour relations in the DMN and 282 MOT networks across frequency ranges where differences were observed in between groups 283 contrasts; the alpha, beta and gamma band (Bonferroni-corrected across frequencies). Significant 284 correlations (non-parametric Spearman's Rho) with symptoms were observed in the DMN and 285 MOT across alpha, beta and low gamma ranges; however, partial correlations with comorbidity 286 symptom scores (Conner's, GAD7, and PHQ9) entered covariates revealed that the variance in 287 MOT connectivity was not solely driven by concussion symptoms. Factoring in these covariates 288 reduced the full correlation coefficients, such that they were found to no longer be significant -289 al.  In this study, we used MEG to investigate frequency-specific interactions within 297 functional ICNs in adults with a single concussion compared to a matched control group. In 298 contradiction to our initial predictions, a single concussion was associated with increased 299 functional connectivity mediated via band-limited AEC -specifically, elevated coupling within 300 DMN and motor networks, and importantly, in the absence of canonical band power spectrum 301 differences. Intra-DMN connectivity was positively associated with concussion symptoms, even 302 when controlling for secondary/comorbid outcomes (depression, anxiety and attentional 303 problems), which suggests that the internal coupling of this task-negative (Raichle et al., 2001) 304 and dynamic 'cortical-core' network (de Pasquale et al., 2012) is particularly prone to the effects 305 of even relatively mild traumatic brain injuries, which are to a degree independent of 306 comorbidities that follow. Additionally, elevated MOT connectivity was also associated with 307 symptoms and secondary sequelae. 308 We found altered oscillatory-mediated connectivity in the DMN in concussion patients. 310 These results appear in opposition to previous fMRI literature examining BOLD

Relations between functional connectivity and structural architecture 423
Cartography of the human 'connectome' has seen a surge in recent years, but progress 424 has been slow in elucidating the association between electrophysiological connectivity, and 425 structural connectivity, as in cortical oligodendrocytes and myelin content. Preclinical work 426 using optogenetics to stimulate neuronal (electrical) activity has shown that this promotes 427 adaptive myelination (Gibson et al., 2014), and a recent human study has proposed that there is a 428 strong link between neural connectonomics, their dynamics and cortical white matter structure 429 (or 'myeloarchitecture' (Hunt et al., 2016)). It seems reasonable to posit, given these findings,