Task induced modulation of neural oscillations in electrophysiological brain networks
Highlights
► Non-invasive characterisation of task positive electrical brain networks using MEG ► Novel extension to ICA allowing network characterisation across frequency bands ► Time–frequency characterisation of task induced change in network oscillations ► Results show modulation of network oscillations with memory load. ► Modulation of network oscillations with stimulus relevance also shown
Introduction
In recent years, one of the most important findings in the field of systems neuroscience has been the identification and characterisation of large scale distributed brain networks. Growing evidence from functional imaging confirms that the majority of communication in the brain is internal communication between brain regions rather than communication with the external world. Resting state Blood Oxygenation Level Dependent (BOLD) Functional Magnetic Resonance Imaging (fMRI) (Ogawa et al., 1993) reveals that a relatively small set of networks consistently exhibit correlated activity during rest (Beckmann et al., 2005, Smith et al., 2009) whilst meta-analysis of fMRI data acquired during task performance demonstrates that very similar networks are engaged during a wide variety of tasks (Smith et al., 2009). This evidence suggests that the recruitment of a functionally specific set of neurons to achieve a specified computation is facilitated by a background of communication between large neural assemblies that establish a favourable setting for the specific computation. These ‘background assemblies’ form networks of brain areas that have previously been associated with specific functionality, some of which involve simple sensory action (e.g. the ‘motor network’) whilst others appear to modulate with higher level cognitive processes (e.g. the ‘dorsal attention network’). However, the evidence suggests that networks do not merely function as monolithic entities within which the relationships between nodes are fixed, but rather consist of sub-networks each with its own pattern of correlations and anti-correlations (Smith et al., 2012). Such an arrangement allows overlap, such that a region's activity might reflect one network's activity under some circumstances, and another network's activity under other circumstances.
To date, non-invasive delineation of these large scale networks has been mainly based on fMRI. Unfortunately however, since BOLD is an indirect mechanism reliant on haemodynamics with limited temporal resolution, neither the electrical activity underlying haemodynamic effects nor the rapid temporal dynamics can be assessed using fMRI alone. Recently, electrophysiological imaging techniques, in particular magnetoencephalography (MEG) (Cohen, 1972), have been used to independently identify similar resting state network topography (de Pasquale et al., 2010, Liu et al., 2010, Brookes et al., in 2011a, Brookes et al., 2011b, Luckhoo et al., 2012). This work confirms an electrophysiological origin of the fMRI networks, adding weight to the argument that electrical oscillatory activity is a manifestation of cortical connectivity (Goldman et al., 2001, Laufs et al., 2003, Mantini et al., 2007, Brookes et al., 2011b). In the present paper, our aim is to investigate the existence of large scale pan-spectral networks in MEG data that are robustly observed during performance of multiple cognitive tasks.
Oscillatory electrical activity in cell assemblies has been observed since early EEG studies (Berger, 1929). These effects occur across a wide range of frequency bands ranging from δ (1–4 Hz) to γ (30–200 Hz) and a great deal of previous work has begun to characterise the nature of these oscillations and their response to sensory (for examples see Pfurtscheller and Aranibar, 1979, Pfurtscheller, 1992, Tesche and Hari, 1993, Salmelin and Hari, 1994a, Salmelin and Hari, 1994b, Jensen and Vanni, 2002, Hall et al., 2011, Stevenson et al., 2011) and cognitive (for examples see Jensen et al., 2002, Jensen and Tesche, 2002, Meltzer et al., 2007, Meltzer et al., 2008, Ossandón et al., 2011) tasks. This large body of work now suggests strongly that oscillations are an integral part of brain function (Singer, 1999, Schnitzler and Gross, 2005). The relationship between oscillations and the BOLD haemodynamic response has also been probed by a number of studies (Singh et al., 2002, Brookes et al., 2005, Zumer et al., 2010, Stevenson et al., 2011) with a close spatial correspondence shown between these disparate effects. However, despite general agreement, there is no simple ‘one-to-one’ relation between activity at any particular frequency and the BOLD response (Winterer et al., 2007, Zumer et al., 2010). More recently, oscillatory processes have also been linked closely with neurochemistry (Muthukumaraswamy et al., 2009) and, most importantly in this context, they appear to be involved in long range communication across the cortex.
Theoretical work suggesting the involvement of oscillations in connectivity has been supported by a growing number of imaging studies: Concurrent EEG/fMRI recordings have shown that oscillations measured using EEG correlate with BOLD network timecourses from fMRI (Mantini et al., 2007). Invasive electrocorticography (ECoG) has shown that the envelope of high γ band oscillations in the default mode network is modulated by task performance (Ossandón et al., 2011). Despite these successes, many questions still remain regarding the nature of distributed networks. In particular, how do constituent (spatially separate) brain areas synchronise and interact to form networks, and in what way are network oscillations induced, evoked, and/or modulated by task or behaviour?
MEG is non-invasive, has whole brain coverage and excellent temporal resolution. Furthermore, although spatial resolution is limited by the ill posed ‘inverse problem’ (reconstructing brain space current densities using field recordings made outside the head) it is more spatially precise than EEG since MEG detects magnetic field due to primary currents (and not extracellular volume currents that are dependent on conductivity). Technical development in MEG source localisation (Robinson and Vrba, 1998, Gross et al., 2001, Zumer et al., 2007, Wipf et al., 2010) has enabled neural oscillatory effects to be well localised, making it an attractive modality to elucidate the role of oscillations in networks. A rich history of MEG research has shown its ability to capture functional connectivity between spatially separate brain areas (Tass et al., 1998, Ioannides et al., 2000, Gross et al., 2001, Gross et al., 2002, Nolte et al., 2004, Nolte et al., 2008, Ramnania et al., 2004, Schlögl and Supp, 2006, Jerbi et al., 2007, Gow et al., 2008, Schoffelen and Gross, 2009, Brookes et al., in 2011a, Brookes et al., 2012, Luckhoo et al., 2012). Our recent work (Brookes et al., 2011b) applied independent component analysis (ICA) to slow (< 1 Hz) fluctuations in the amplitude envelope of neural oscillations at separate brain space voxels to show that the networks observable in fMRI could also be measured using resting state MEG. The nature of resting state measurement means however that no functional link can be made between oscillatory power fluctuations and behaviour. In contrast, measurements made during task-performance allow for the time-locking of network fluctuations to task stimuli (Luckhoo et al., 2012). Further, this time-locking offers the potential for the fine temporal resolution of MEG to be utilised to delineate the temporal evolution of event-related network activity. Hence, in this work we use task based MEG to investigate event-related network behaviour.
In what follows we describe MEG recordings made in healthy volunteers during three cognitive tasks: 1) a visual Sternberg task; 2) an N-back task and 3) a novel target detection paradigm in which task-relevance of stimuli is manipulated, referred to as the Relevance Modulation (RM) task. MEG data were recorded during these tasks and processed in order to extract the spatial signature of distributed robust electrical brain networks. We extend previous MEG ICA work (Hyvärinen et al., 2010, Brookes et al., 2011b, Luckhoo et al., 2012) by introducing a novel extension to the beamformer-ICA approach (Brookes et al., 2011b, Luckhoo et al., 2012) in which source space independent components computed within individual frequency bands are combined to generate maps of pan-spectral network topography. We use this approach to elucidate a number of physiologically meaningful networks. Further, source space MEG data are reconstructed for each network to derive the event-related time–frequency (TF) signature of neural oscillations with high (millisecond) temporal resolution, showing that oscillatory power is significantly modulated by working memory (WM) maintenance, WM load, and the task-relevance of stimuli. Below, the Methods section presents, in detail, the methodology used for network characterisation including the details of beamforming and independent component analysis. The Results section presents the main findings of the study and the discussion places these findings in the context of other related work.
Section snippets
Methods
10 subjects took part in both the Sternberg and RM tasks (4 male, 3 left handed, age range 29 ± 5 years (mean ± std), all had normal or corrected to normal vision). 8 subjects took part in the N-back task (4 male, 2 left handed, age range 28 ± 3 years, all had normal or corrected to normal vision). The cohort comprised postgraduate students and postdoctoral researchers. For all tasks, visual stimuli were presented centrally in the visual field. The visual angle subtended by the screen was 35° vertical
Results
The spatial signatures of 8 of the 10 networks extracted from our second level ICA are shown in Fig. 3 along with their maximum contribution from each frequency band. Networks homologous with those found in fMRI studies (Beckmann et al., 2005, Seeley et al., 2007, Smith et al., 2009) were observed including the primary visual areas (A), bilateral fronto-parietal network (B) bilateral insula (C) bilateral temporo-parietal junction (TPJ) (D), right motor cortex (E), left motor cortex (F), lateral
Discussion
This paper has shown that large scale electrical brain networks are identifiable using ICA of MEG data collected across multiple subjects during performance of three separate cognitive tasks. Further, significant event-related temporal modulation of neural oscillatory activity was noted for all networks, in all of the tasks studied, confirming the functional relevance of the brain areas involved. More specifically, we have shown that neural oscillations in these networks are modulated by WM
Conclusion
This paper has employed an ICA approach to data analysis in order to derive a set of electrodynamic brain networks that are identifiable in MEG data collected across subjects during performance of multiple tasks. We have introduced two methodological extensions to previous ICA approaches: 1) we show how to derive pan-spectral networks that combine independent components computed within individual frequency bands. 2) We show how to measure the temporal evolution of each network with millisecond
Acknowledgments
The authors are grateful to Ms Lesley Martin for her help in collecting subject information and Tom Allan for proof reading the manuscript. We thank the Leverhulme Trust for an Early Career Fellowship awarded to MJB. We also acknowledge support from the Dr. Hadwen Trust, the Medical Research Council and The University of Nottingham. The Dr. Hadwen Trust for Humane Research is the UK leading medical research charity that funds and promotes exclusively human-relevant research that encourages the
References (68)
- et al.
GLM-beamformer method demonstrates stationary field, alpha ERD and gamma ERS co-localisation with fMRI BOLD response in visual cortex
NeuroImage
(2005) - et al.
Optimising experimental design for MEG beamformer imaging
NeuroImage
(2008) - et al.
Measuring functional connectivity using MEG: methodology and comparison with fcMRI
NeuroImage
(2011) - et al.
Measuring functional connectivity in MEG: a multivariate approach insensitive to linear source leakage
NeuroImage
(2012) - et al.
What does delta band tell us about cognitive processes: a mental calculation study
Neurosci. Lett.
(2010) - et al.
Relating MEG measured motor cortical oscillations to resting γ-aminobutyric acid (GABA) concentration
NeuroImage
(2011) - et al.
Lexical influences on speech perception: a Granger causality analysis of MEG and EEG source estimates
NeuroImage
(2008) - et al.
Effect of hypercapnia on resting and stimulus induced MEG signals
NeuroImage
(2011) - et al.
Independent component analysis of short-time Fourier transforms for spontaneous EEG/MEG analysis
NeuroImage
(2010) - et al.
EEG-correlated fMRI of human alpha activity
NeuroImage
(2003)