Alterations in resting-state network dynamics along the Alzheimer’s disease continuum: a combined MEG-PET/MR approach

Abstract

Human brain activity is intrinsically organized into resting-state networks (RSNs) that transiently activate or deactivate at the sub-second timescale. Few neuroimaging studies have addressed how Alzheimer's disease (AD) affects these fast temporal brain dynamics, and how they relate to the cognitive, structural and metabolic abnormalities characterizing AD.We aimed at closing this gap by investigating both brain structure and function using magnetoencephalography (MEG) and hybrid positron emission tomography-magnetic resonance (PET/MR) in 10 healthy elders, 10 patients with Subjective Cognitive Decline (SCD), 10 patients with amnestic Mild Cognitive Impairment (aMCI) and 10 patients with typical Alzheimer's disease with dementia (AD). The fast activation/deactivation state dynamics of RSNs were assessed using hidden Markov modeling (HMM) of power envelope fluctuations at rest measured with MEG. HMM patterns were related to participants' cognitivetest scores, whole hippocampal grey matter volume and regional brain glucose metabolism.The posterior default-mode network (DMN) was less often activated and for shorter durations in AD patients than matched healthy elders. No significant difference was found in patients with SCD or aMCI. The time spent by participants in the activated posterior DMN state did not correlate significantly with cognitive scores. However, it correlated positively with the whole hippocampal volume and regional glucose consumption in the right temporo-parietal junctions and dorsolateral prefrontal cortex, and negatively with glucose consumption in the cerebellum.In AD patients, alterations of posterior DMN power activation dynamics at rest correlate with structural and neurometabolic abnormalities. These findings represent an additional electrophysiological correlate of AD-related synaptic and neural dysfunction.

Microstates and power envelope hidden Markov modeling probe bursting brain activity at different timescales

“…States 3 MEG and 4 MEG identified pDMN activation and deactivation, respectively. Of note, similar states involving precuneus activity were also identified in previous works (Coquelet et al, 2020b; Puttaert et al, 2020) and not in others (e.g., (Baker et al, 2014; Brookes et al, 2018)) due to different choices of source projection (for details, see (Sjøgård et al, 2019)). State 5 MEG corresponded to the activation of the SMN in isolation (rather than in competition with the pDMN/VoN, as in states 1 MEG and 2 MEG ).…”

“…One side result noteworthy of mention is that the HMM of sensor power signals leads to network-level states similar to the HMM of reconstructed source power considered in the seminal paper of (Baker et al, 2014) and subsequent MEG studies (Brookes et al, 2018; Coquelet et al, 2020b; Puttaert et al, 2020; Quinn et al, 2018; Sitnikova et al, 2018; Van Schependom et al, 2019; Vidaurre et al, 2018). The HMM of electrophysiological signals can thus be performed in a computationally less cumbersome way than previously done, for similar results.…”

“…The pDMN state pairs 3 MEG /3 EEG and 4 MEG /4 EEG exhibited a substantial overlap of their activation periods, but the others lacked such strong temporal correspondence. The pDMN states were also the most stable (Coquelet et al, 2020b; Puttaert et al, 2020), suggesting that state co-occurrence rate increases with state stability. The difficulty of short-lived states to co-activate explains in particular the poor cross-modal temporal correlation for the quasi-instantaneous microstates.…”

“…Only the imaging of state brain maps would depend on these choices. This is particularly interesting with regard to the contribution of precuneus activity to HMM state dynamics, as it was identified from MEG source power HMMs when using minimum norm estimation (Coquelet et al, 2020b; Puttaert et al, 2020) but not when using a beamformer (Baker et al, 2014; Brookes et al, 2018; Vidaurre et al, 2018) due to a suppression effect (Sjøgård et al, 2019). The states 1 MEG , 3 MEG /3 EEG and 4 MEG /4 EEG obtained in this study show that sensor-level HMM is sensitive to precuneus activity, independently of source reconstruction biases.…”

“…The HMM inference applied to MEG power envelope signals has typically been used to identify 6 or 8 states disclosing a spatial distribution reminiscent of brain functional networks as well as mean lifetimes ranging from 50 to 200 ms (Baker et al, 2014; Quinn et al, 2018). Temporal properties of HMM states are also altered by physiological processes such as ageing (Brookes et al, 2018; Coquelet et al, 2020b) as well as brain disorders such as Alzheimer’s disease (Puttaert et al, 2020; Sitnikova et al, 2018) or multiple sclerosis (Van Schependom et al, 2019).…”