A wavelet-based method for measuring the oscillatory dynamics of resting-state functional connectivity in MEG
Research Highlights
► How to analyze resting-state functional connectivity (RSFC) in MEG is not established. ► Wavelet-based phase-locking can help determine the dynamics of RSFC MEG data. ► An important artifact arises in RSFC data that is reduced by use of empty room data. ► Using this method, the auditory network showed RSFC dynamics in the alpha band.
Introduction
Measuring how neural regions interact is critical for understanding the dynamics of the normal and disordered brain. “Functional connectivity” is thought to reflect these interactions and is defined as “the correlation between spatially remote neurophysiological events” (Friston et al., 1993). Correlation is used as a measure of functional connectivity based on the principle that if two neuronal populations fire together, they are likely to be part of the same functional circuit. Traditionally, changes in the correlations between neural populations are measured across tasks or cognitive states. Recently, it was discovered that the activity in regions forming task-critical networks (for example the networks associated with visual, auditory, memory, and sensorimotor functions Biswal et al., 1995, Cordes et al., 2000, Vincent et al., 2006); correlate even when those tasks are not being performed (when the subject is at “rest,” in light sleep, or even sedated; see Boly et al., 2008, Fox and Raichle, 2007 for reviews). Resting-state functional connectivity has garnered a great deal of interest as a method for examining functional networks in a “natural” state. This interest arises from two aspects of resting-state functional connectivity. First, because these networks arise without being driven by a task, this method has the potential to illustrate fundamental aspects of the brain's intrinsic functional organization (Fox and Raichle, 2007). Second, this lack of task can allow us to examine functional connectivity in clinical populations where behavioral responses may be abnormal (Greicius, 2008).
The majority of resting-state functional connectivity studies have been performed using functional magnetic resonance imaging (fMRI). Because of the relatively poor temporal resolution of fMRI (~.5–1 Hz and below) these studies have been restricted to examining correlations in slow resting-state oscillations. These studies have demonstrated that there are strong correlations within functional networks in the very low frequency (<.1 Hz) aspects of the resting-state activity (Fox and Raichle, 2007). Correlations in this low frequency band are surprising since most electrophysiological aspects of neural activity occur at a much faster time scale. One likely possibility that has emerged is that these very slow fluctuations in large part reflect slow changes in underlying higher frequency neural activity (He et al., 2008, Leopold et al., 2003, Mantini et al., 2007, Nir et al., 2008). However, little is known regarding the exact nature of correlated activity in these higher frequencies. To explore these faster aspects of resting-state functional connectivity requires methods to examine interactions using electrophysiological measures of neural activity that have a higher temporal resolution, such as magnetoencephalography (MEG).
Methods for measuring functional connectivity in non-invasive electrophysiological data have generally examined task/condition differences in the correlation of the time–frequency response of the neural activity arising from disparate brain regions (Jerbi et al., 2007, Lin et al., 2004, Tass et al., 2003). These studies were designed to allow researchers to make inferences about whether the task or cognitive state modifies functional connectivity. For example, the spectrum of interregional correlations has been examined before and after learning (Duzel et al., 2005, Ghuman et al., 2008), compared across attentional states (Gross et al., 2004), and for different visual stimuli (Bar et al., 2006, Kaiser et al., 2004). Many of these approaches were adapted from methods that have been used successfully to examine functional connectivity using electrode recordings in animals (Engel et al., 2001, Roelfsema et al., 1997, Varela et al., 2001).
There are many methodological issues that must be considered in resting-state functional connectivity before adapting task-related approaches to resting-state data. These issues appear because of the lack of a within-subjects comparison condition in resting-state studies. For example, when examining resting-state functional connectivity between brain areas, one must take into account the fact that some of the activity projected to each area originates from common sensors. In task-based studies, crosstalk is somewhat mitigated because it is present to some degree in all conditions and is reduced when a comparison across conditions is performed.
Here we introduce a novel wavelet-based method for measuring resting-state phase-locking between electrophysiological signals that are measured non-invasively, but mapped onto the human brain, and introduce normalization to reduce a major crosstalk artifact. Wavelet-based analyses have the advantage of not requiring the data to be stationary (Percival and Walden, 2000) and therefore are likely to be more appropriate for non-stationary neural data than Fourier-based methods. It should be noted that while we primarily discuss this method for MEG, in principle it could be used for any non-invasive electrophysiological measure of brain activity, such as electroencephalography (EEG). After describing the method, we use simulations to test the spatial sensitivity and specificity of the functional connectivity on the cortical surface. We use these simulations to examine a key artifact that appears in non-invasive resting-state functional connectivity analyses of electrophysiological data and demonstrate a procedure for reducing much of this problem. Finally, we apply the method to examine the spectral functional connectivity in the left (LH) and right hemisphere (RH) auditory network using MEG with eight subjects. While these MEG results are primarily used to validate the method, this is also the first demonstration of connectivity between the LH and RH auditory cortices in a true resting state. Previous studies have used fMRI to examine connectivity in this network (Cordes et al., 2000), but fMRI cannot be considered a true resting state for the auditory cortex because of the noise the MRI machine produces; in contrast, MEG is silent.
Section snippets
Methods
The method for calculating resting-state functional connectivity was adapted from the dynamic statistical parametric mapping method developed by Lin et al. (2004). Specifically, the process involves six steps: 1) artifact removal 2) selection of “seed” region of interest (ROI) 3) calculation of the inverse solution and projection onto the brain 4) wavelet transformation of the signal 5) calculation of the phase-locking values (PLVs; Lachaux et al., 1999) between the seed ROI and every other
Spatial sensitivity and specificity simulations
One major concern when employing a linear inverse solution with electrophysiological data is that most inverse solutions act as a spatial filter where the data projected onto each point in the brain is a combination of the data derived from each sensor location (Hämäläinen et al., 1993). Therefore, some artifactual phase-locking will be introduced because each source location shares some data with each other source location. An example of artifactual phase-locking introduced by the imperfect
Conclusion
We have described a method for detecting and describing the oscillatory dynamics of functional connectivity in MEG resting-state data projected onto the brain. We examined the spatial sensitivity and specificity of this phase-locking technique using simulated data and showed how to use empty room data to account for much of the crosstalk that arises due to the imperfect inverse solution. We then applied this technique to show that the auditory network displays resting-state functional
Acknowledgments
We thank Rebecca van den Honert for assistance with data collection and analysis, Steve Gotts, Richard Coppola, and for insightful comments, Tom Holroyd and the staff NIMH MEG core for assistance with data collection, Gang Chen for assistance with statistics, Zhongming Liu for assistance with the cardiac artifact removal procedure, and Ziad Saad for assistance with MRI processing. This work was supported by NIMH-DIRP.
References (41)
- et al.
Gamma, alpha, delta, and theta oscillations govern cognitive processes
Int. J. Psychophysiol.
(2001) - et al.
Dynamic statistical parametric mapping: combining fMRI and MEG for high-resolution imaging of cortical activity
Neuron
(2000) - et al.
EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis
J. Neurosci. Meth.
(2004) - et al.
Analysis of fMRI time-series revisited
Neuroimage
(1995) - et al.
Magnetoencephalographic gamma-band responses to illusory triangles in humans
Neuroimage
(2004) - et al.
Evidence for reactive magnetic 10-Hz rhythm in the human auditory cortex
Neurosci. Lett.
(1997) - et al.
Spectral spatiotemporal imaging of cortical oscillations and interactions in the human brain
Neuroimage
(2004) - et al.
Assessing and improving the spatial accuracy in MEG source localization by depth-weighted minimum-norm estimates
Neuroimage
(2006) - et al.
Large-scale spontaneous fluctuations and correlations in brain electrical activity observed with magnetoencephalography
Neuroimage
(2010) - et al.
Nonparametric statistical testing of EEG- and MEG-data
J. Neurosci. Meth.
(2007)
A possible role of evoked alpha in primary sensory processing: common properties of cat intracranial recordings and human EEG and MEG
Int. J. Psychophysiol.
Magnetoencephalographic 10-Hz rhythm from the human auditory cortex
Neurosci. Lett.
Top-down facilitation of visual recognition
Proc. Natl Acad. Sci. USA
Functional connectivity in the motor cortex of resting human brain using echo-planar MRI
Magn. Reson. Med.
Intrinsic brain activity in altered states of consciousness: how conscious is the default mode of brain function?
Ann. NY Acad. Sci.
Mapping functionally related regions of brain with functional connectivity MR imaging
AJNR Am. J. Neuroradiol.
The oscillatory dynamics of recognition memory and its relationship to event-related responses
Cereb. Cortex
Dynamic predictions: oscillations and synchrony in top-down processing
Nat. Rev. Neurosci.
Statistical analysis of circular data
Spontaneous fluctuations in brain activity observed with functional magnetic resonance imaging
Nat. Rev. Neurosci.
Cited by (55)
Non-negative Matrix Factorization Reveals Resting-State Cortical Alpha Network Abnormalities in the First-Episode Schizophrenia Spectrum
2020, Biological Psychiatry: Cognitive Neuroscience and NeuroimagingBrain networks and their relevance for stroke rehabilitation
2019, Clinical Neurophysiology