The behavioral significance of coherent resting-state oscillations after stroke

Abstract

Stroke lesions induce not only loss of local neural function, but disruptions in spatially distributed areas. However, it is unknown whether they affect the synchrony of electrical oscillations in neural networks and if changes in network coherence are associated with neurological deficits. This study assessed these questions in a population of patients with subacute, unilateral, ischemic stroke.

Spontaneous cortical oscillations were reconstructed from high-resolution electroencephalograms (EEG) with adaptive spatial filters. Maps of functional connectivity (FC) between brain areas were created and correlated with patient performance in motor and cognitive scores.

In comparison to age matched healthy controls, stroke patients showed a selective disruption of FC in the alpha frequency range. The spatial distribution of alpha band FC reflected the pattern of motor and cognitive deficits of the individual patient: network nodes that participate normally in the affected functions showed local decreases in FC with the rest of the brain. Interregional FC in the alpha band, but not in delta, theta, or beta frequencies, was highly correlated with motor and cognitive performance. In contrast, FC between contralesional areas and the rest of the brain was negatively associated with patient performance.

Alpha oscillation synchrony at rest is a unique and specific marker of network function and linearly associated with behavioral performance. Maps of alpha synchrony computed from a single resting-state EEG recording provide a robust and convenient window into the functionality and organization of cortical networks with numerous potential applications.

Introduction

Therapy of brain disease could potentially be improved if we had better knowledge about how brain organization is disturbed after brain damage and how lost functions can be restored. Functional imaging studies can provide important insights into brain physiology and pathology (Cramer, 2008), but the activation and lesion mapping techniques used traditionally have inherent limitations for the study of changes in brain function induced by lesions. Brain lesions induce not only loss of local neural function, but also changes in remote networks (Alstott et al., 2009, Honey and Sporns, 2008) which likely remain undetected when analyzing task-induced activations or lesions of circumscribed brain areas. Moreover, many patients with brain lesions are unable to perform the tasks required for inducing activation of the brain area of interest, because they are impaired exactly in those functions that have to be studied (Weiller et al., 1993).

These limitations can be overcome by imaging techniques which assess the functional connectivity (FC) between different brain regions rather than measuring task-induced activations. Studies using fMRI in healthy humans have shown that spontaneous fluctuations of activity in the resting brain are highly organized and coherent within specific neuro-anatomical systems (Damoiseaux et al., 2006, Fox et al., 2005, Greicius et al., 2003). These patterns of coherence are task-independent and can also be studied at rest (Greicius et al., 2003). The topography of coherence between brain regions observed at rest matches the topography of brain activations induced by corresponding tasks (Vincent et al., 2007, Zhao et al., 2011) and the strength of coupling in a given network is linearly related to behavioral performance in tasks relying on this network (Koyama et al., 2011, Wang et al., 2010). Thus, a careful analysis of coherence (or FC) among neural assemblies gives access to functional brain organization. This approach has the crucial advantage that it takes into account the network character of brain activity. Furthermore, it does not depend on patient cooperation or specific activation paradigms, and is therefore suitable for studies on brain dysfunction and recovery of patients with neurological deficits who are often unable to perform tasks. Indeed, recent studies have confirmed that fMRI measurements of interhemispheric FC reflect motor and cognitive performance in patients with stroke (Carter et al., 2010, He et al., 2007).

The concept of FC can be applied not only to fMRI, but also to magnetencephalography (MEG) (Schoffelen and Gross, 2009) or EEG (Guggisberg et al., 2011). It is particularly fruitful when FC is not calculated between sensor or electrode pairs, but between reconstructions of cortical oscillations obtained with inverse solutions. This approach allows for instance to localize the cortical generators of cortico-muscular interaction (Gross et al., 2001, Guggisberg et al., 2011, Schoffelen et al., 2008), as well as to reconstruct cortical network interactions. Initial direct comparisons between resting-state networks obtained with fMRI and networks reconstructed from MEG suggested a more lateralized configuration of MEG networks (de Pasquale et al., 2010), but this difference disappeared in studies controlling for distortions of MEG/EEG FC due to spatial leakage of the inverse solutions (Brookes et al., 2011a, Brookes et al., 2011b). Simultaneous recordings of resting-state fMRI and EEG have shown that if the time course of spontaneous EEG oscillations or microstate changes is convolved with a hemodynamic response function and used as regressor for fMRI analysis, the resulting BOLD responses localize to similar distributed networks as found in fMRI network studies (Britz et al., 2010, Jann et al., 2009, Musso et al., 2010). Interestingly, the closest spatial match to fMRI networks was found with EEG/MEG networks reconstructed from slow cortical potentials and beta-/gamma power fluctuations (Brookes et al., 2011b, He et al., 2008). Several recent studies have described possible mechanisms for the striking topographical similarity between rapid EEG/MEG interactions and much slower cortical potentials or fMRI fluctuations. One candidate mechanism (Raichle, 2011) might be the scale-free (fractal) dynamics of neural activity (He et al., 2010, Van de Ville et al., 2010) with cross-frequency phase coupling from very slow cortical potentials [which possibly underlie BOLD fluctuations (He and Raichle, 2009)] to fast high-gamma rhythms (Monto et al., 2008, Osipova et al., 2008, Palva et al., 2005a).

Unlike fMRI, EEG and MEG can capture a rich spectrum of neural oscillations and might therefore reveal additional information on the intrinsic brain organization. However, in contrast to the rather well studied functional meaning of fMRI resting-state networks, the behavioral and clinical significance of FC in the different EEG/MEG frequencies is largely unknown. Previous MEG studies in patients with brain tumors showed that local decreases in alpha band (~ 7.5 to 12.5 Hz) FC between a given tumor part and the rest of the brain were associated with dysfunction of this part, whereas normal or increased alpha band FC were found in functional tumor tissue (Guggisberg et al., 2008, Martino et al., 2011). These observations suggested that the magnitude of interregional MEG/EEG FC might be related to function and hence to behavioral performance or deficits of the patients. Moreover, the behavioral relevance of FC seemed to be dependent on the oscillation frequency. This study therefore aimed to assess the clinical and behavioral significance of the spectrum of EEG FC. Based on previous studies, we hypothesized that stroke lesions disrupt synchronous electrical oscillations at rest and that this disruption is linearly correlated with neurological deficits of the patients. To test this, we obtained high-resolution resting-state EEG recordings in a population of 20 patients with unilateral ischemic stroke exhibiting various degrees of motor and cognitive deficits. Quantitative scores of motor as well as frontal, left- and right-hemispheric cognitive functions were compared to indices of EEG FC of the entire cortex.