• Aucun résultat trouvé

Investigating the human brain with electromagnetic signals

N/A
N/A
Protected

Academic year: 2022

Partager "Investigating the human brain with electromagnetic signals"

Copied!
141
0
0

Texte intégral

(1)

Thesis

Reference

Investigating the human brain with electromagnetic signals

GUGGISBERG, Adrian

Abstract

The human brain, heart, and muscles emit electromagnetic signals which can be measured non-invasively. By assessing specific patterns and features of these signals, one can gain insight into many aspects on human brain function and disease. This thesis summarizes the meaning and validity of electromagnetic markers allowing the investigation of autonomous nervous system activity, vigilance, neural processing during cognitive tasks, and network communication. Furthermore, it provides examples for applications in clinical and cognitive research.

GUGGISBERG, Adrian. Investigating the human brain with electromagnetic signals. Thèse de privat-docent : Univ. Genève, 2012

DOI : 10.13097/archive-ouverte/unige:26388

Available at:

http://archive-ouverte.unige.ch/unige:26388

Disclaimer: layout of this document may differ from the published version.

1 / 1

(2)

Clinical Medicine Section

Department of Clinical Neurosciences Division of Neurorehabilitation

"INVESTIGATING THE HUMAN BRAIN WITH ELECTROMAGNETIC SIGNALS"

Thesis submitted to the Medical School of the University of Geneva

for the degree of Privat-Docent by

Adrian GUGGISBERG

Geneva

2012

(3)

1

I NVESTIGATING THE H UMAN B RAIN

WITH E LECTROMAGNETIC S IGNALS

Thèse d’habilitation au titre de Privat-Docent à la Faculté de Médecine de Genève

Adrian G. Guggisberg Service de Neurorééducation Hôpitaux Universitaires de Genève

Genève 2012

(4)

2

A BSTRACT

The human brain, heart, and muscles emit electromagnetic signals which can be measured non- invasively. By assessing specific patterns and features of these signals, one can gain insight into many aspects on human brain function and disease. This thesis summarizes the meaning and validity of electromagnetic markers allowing the investigation of autonomous nervous system activity, vigilance, neural processing during cognitive tasks, and network communication. Furthermore, it provides examples for applications in clinical and cognitive research.

A CKNOWLEDGMENTS

I would like to thank the following persons, who have greatly contributed to the work presented here:

- My teachers and mentors, Professors Armin Schnider, Christian W. Hess, Johannes Mathis, and Srikantan S. Nagarajan

- My collaborators, especially Sarang Dalal, Johanna Zumer, Sviatlana Dubovik, and Leighton Hinkley

- My wife Roxana and my children Noah and Camilo for their patience and emotional support

(5)

3

I NDEX

ABSTRACT 2

ACKNOWLEDGMENTS 2

ABBREVIATIONS 5

1 INTRODUCTION 6

2 GENERATION AND RECORDING OF HUMAN ELECTROMAGNETIC SIGNALS 7

3 ASSESSING AUTONOMOUS NERVOUS ACTIVITY 8

3.1 Periodic Leg Movements in Sleep 8

3.2 Spectral Analysis of Heart-Rate-Variability 9

3.3 Improving the Temporal Resolution 10

3.4 Time-Resolved Assessment of Sympathetic and Vagal Activity During PLMS 10

3.5 Future Directions 10

4 ASSESSING THE AROUSAL LEVEL 11

4.1 Yawning 11

4.2 Delta and Theta Power 11

4.3 Frequency and Topography of Alpha Oscillations 12

4.4 Yawning and Vigilance 12

4.5 Future Directions 13

5 TRACKING THE DYNAMICS OF NEURAL COGNITIVE PROCESSING 13

5.1 Decision-Making and Conscious Awareness 13

5.2 The Bereitschaftspotential 14

5.3 The Lateralized Readiness Potential 16

5.4 fMRI Movement Predictors 17

5.5 High-Gamma Oscillations 17

5.6 Neural Dynamics of Human Decision-Making 19

(6)

4

5.7 Neural Mechanisms of Introspection 20

5.8 Summary and Conclusions 22

5.9 Future Directions 22

6 INVESTIGATING NEURAL NETWORKS 22

6.1 Functional Connectivity 23

6.2 Resting-state Networks in fMRI 23

6.3 Functional Brain Networks in MEG and EEG 24

6.4 Clinical Applications 25

6.5 Alpha-band Coherence 25

6.6 Future Directions 27

7 REFERENCES 28

8 APPENDICES: REPRINTS OF PUBLICATIONS 35

8.1 Appendix I: The significance of the sympathetic nervous system in the pathophysiology

of periodic leg movements in sleep 35

8.2 Appendix II: The functional relationship between yawning and vigilance 47 8.3 Appendix III: Five-dimensional neuroimaging: Localization of the time–frequency

dynamics of cortical activity 55

8.4 Appendix IV: High-frequency oscillations in distributed neural networks reveal

the dynamics of human decision making. 70

8.5 Appendix V: The neural basis of event-time introspection 81

8.6 Appendix VI: Localization of cortico-peripheral coherence with electroencephalography 98 8.7 Appendix VII: Mapping functional connectivity in patients with brain lesions 108 8.8 Appendix VIII: Resting functional connectivity in patients with brain tumors in eloquent areas 119 8.9 Appendix IX: The behavioral significance of coherent resting-state oscillations after stroke 131

(7)

5

A BBREVIATIONS

ABC Alpha-band coherence AD Alzheimer’s disease BP Bereitschaftspotential

CNV Contingent negative variation DC Direct current

ECG Electrocardiography EEG Electroencephalography EMG Electromyography FC Functional connectivity

fMRI Functional magnetic resonance imaging HRV Heart rate variability

HF High frequency component of HRV IC Imaginary component of coherence IES Intraoperative electrical stimulation LF Low frequency component of HRV LRP Lateralized readiness potential MEG Magnetoencephalography NREM Non rapid eye movement sleep PET Positron emission tomography PLMS Periodic leg movements in sleep SPN Stimulus preceding negativity (SPN) SWA Slow wave activity in sleep

Voxel Volume element

(8)

6

1 I NTRODUCTION

The human brain, heart, and muscles emit electromagnetic signals which can be measured non- invasively and which contain rich information about human physiology and disease.

Electroencephalograms (EEG), electrocardiograms (ECG), and electromyograms (EMG) have therefore played a crucial role for diagnosis in medicine for decades. The advent of digital recording technology has additionally enabled advanced quantitative analyses of these signals and revealed crucial insights into the working human brain that could not have been obtained by pure visual inspection of the waveforms.

Human electromagnetic signals contain numerous features which can be assessed in countless different ways. A signal feature is defined here as any pattern of the recorded signal that is quantifiable by one or several analysis algorithms. Typical features are for instance the oscillation power in a certain frequency band or the phase locking between the waveforms of two signals.

In order to use these features for addressing a given scientific or clinical question, one has to find out how they are related to, and what information they provide about, the brain processes of interest.

The application of electromagnetic signals in research and clinical practice therefore requires a validation of a given feature as biomarker for brain processes of interest. This should be based on several of the following elements.

1. Knowledge about the underlying physiology generating the recorded signals.

2. Systematic co-occurrence of the feature with the process of interest.

3. Demonstration of population differences between patients and healthy controls.

4. Correlation with other, previously validated markers 5. Correlation with clinical or behavioral variables.

6. Intervention studies demonstrating changes in the signal feature induced by manipulation of the brain processes of interest.

7. Prospective studies demonstrating the sensitivity, specificity, and predictive values of a biomarker.

Research in the past decades has introduced and validated numerous different electromagnetic features as markers allowing the study of many different aspects of brain function. The present thesis aims to summarize some of the existing electromagnetic biomarkers and their application to clinical and cognitive research.

My research is dedicated to the development, validation, and application of analysis methods for human electromagnetic signals in order to obtain new insights into brain physiology and pathology.

The following chapters will give some examples from different fields of research. The main focus lies on signals from the brain itself (EEG and magnetoencephalography, MEG), but analyses of ECG and EMG signals have also been useful for studying brain function and disease. The discussion of my work is integrated into the text in the corresponding context.

Sometimes, it is useful to refine existing and previously validated signal analysis algorithms for a given research question. Chapter 3 will give an example of how we could improve the temporal

(9)

7

resolution of the analysis which allowed us to obtain information about the timing of brain processes of interest. Chapter 5 will show that by optimizing resolution in both space and time, we were able to begin examining highly transient neural processes underlying human consciousness. Finally, chapter 6 will describe our ongoing efforts to validate a new electromagnetic biomarker of cortical network function.

2 G ENERATION AND R ECORDING OF H UMAN E LECTROMAGNETIC S IGNALS

This chapter does not intend to give an in-depth review on the basic physics of electromagnetic signals but to give a short introduction as far as it necessary to understand the subsequent chapters.

Several different structures can generate currents in the human body including neurons, synapses, and muscle fibers. Cell membranes of these structures maintain a resting potential difference between intra- and extracellular media with active ion pumps and channels. Membrane activation leads to a brief depolarization of the transmembrane potential followed by a repolarization. The differential occurrence of transmembrane currents in different areas of the brain or muscles additionally creates extracellular potential differences. All currents superimpose at a given point and generate a potential relative to a reference at another. The high resistivity of the extracellular space leads to a wide-spread field of potential gradients that can be detected at the skull surface with EEG, ECG, and EMG (Zschokke 2002; Buzsaki et al. 2012).

The main contributors to EEG and MEG signals are excitatory and inhibitory postsynaptic currents.

They are generated when incoming synapses activate ion channels in the subsynaptic membrane which leads to a potential difference with adjacent membrane parts of the neuron. Postsynaptic currents therefore reflect the neuronal input at a given location. When numerous neurons and synapses are excited or inhibited synchronously, we can detect potential fluctuations at the surface (Zschokke 2002).

It was long thought that neuronal action potentials reflecting the neuronal output are invisible to surface recordings, because the spikes they generate are of short duration and synchronization of numerous neurons in such short time windows – which would be required for detection at the surface – rarely occurs under physiological conditions. However, it has recently been found that synchronous spiking is associated with an increase in high frequency oscillations (>100 Hz) in the extracellular medium (Rasch et al. 2008; Whittingstall and Logothetis 2009). High-frequency EEG power can therefore be used as an index of neuronal spiking and output activity (Buzsaki et al. 2012).

The origin of different spectral components in the EEG is incompletely understood, but it seems to depend on local cortical architecture, thalamocortical interactions and ascending afferences from the brain stem (Moruzzi and Magoun 1949; Lopes da Silva et al. 1973; Zschokke 2002).

(10)

8

Physical laws described by the Maxwell equations dictate that current flow necessarily produces not only electrical but also magnetic fields. The brain and heart therefore also generate tiny magnetic fields which can be detected with highly sensitive superconducting coils used in MEG or magnetocardiography. Since electric and magnetic fields are highly interdependent, MEG and EEG generally capture similar information about neural processes with similar precision (Malmivuo 2012).

However, there are 3 notable differences between the 2 recording techniques. First, MEG captures preferentially currents flowing tangentially to the skull surface, whereas EEG is most sensitive to radial currents flowing perpendicularly to the surface. MEG therefore provides complementary information to the EEG in that it adds sensitivity to current directions that are largely undetected by EEG (and vice versa). Second, the inhomogeneities and anisotropies of the brain and the surrounding tissue affect magnetic fields much less than the currents recorded by EEG (Barkley 2004). MEG is therefore especially of advantage in patients with skull damage, and it depends less on accurate models of the head configuration for reconstruction of current sources. Third, the temporal scaling properties (that is, the attenuation of oscillation power at higher frequencies) of EEG and MEG often show differences because MEG is less influenced by capacitive properties of skin and skull (Buzsaki et al. 2012). This suggests that MEG can more likely detect fast gamma and high-gamma oscillations at the surface, although this still needs to be demonstrated empirically.

3 A SSESSING A UTONOMOUS N ERVOUS A CTIVITY

Deep brain structures are notoriously difficult to assess with surface electromagnetic signals emitted by the brain. This is because the amplitude of a recorded signal drops off proportionally to the distance between electrodes and the source current. Centers of the autonomous nervous system are located in the hypothalamus and are typical examples of deep brain structures that cannot be investigated with surface EEG or MEG. Fortunately, their output activity is manifested in another electromagnetic signal which can be easily recorded at the surface: the ECG. The following chapter gives an example of how ECG markers of autonomous activity can improve our understanding of disease mechanisms.

3.1 PERIODIC LEG MOVEMENTS IN SLEEP

Periodic leg movements in sleep (PLMS) consist of repetitive, periodic extensions of the big toe and dorsiflexions of the ankle during sleep. While they are rare in young persons, they can be found in over 40% of people aged 65 years or more. PLMS may occur in isolation without any subjective complaints, but they may also be associated with insomnia, disturbed sleep, and daytime sleepiness.

They are particularly frequent in patients with the so-called restless legs syndrome, in which case they may also occur during wakefulness (Montplaisir et al. 2000).

Despite the high prevalence of this condition, its influence on sleep quality and general healthy remains poorly understood. Although numerous studies have observed transient autonomic activation and arousing reactions in association with PLMS (Parrino et al. 1996; Sforza et al. 1999;

(11)

9

Karadeniz et al. 2000; Sforza et al. 2002; Ferrillo et al. 2004; Lavoie et al. 2004; Hornyak et al. 2005;

Sforza et al. 2005), their significance for the genesis of PLMS and for sleep quality is controversial (Mendelson 1996; Nicolas et al. 1998; Karadeniz et al. 2000; Chervin 2001; Hornyak et al. 2004;

Carrier et al. 2005).

This situation motivated Johannes Mathis, Christian Hess, and me at the Inselspital in Berne to look further into the mechanisms of PLMS by examining their associated electromagnetic signals. In particular, we wanted to elucidate the role of the autonomous nervous system which was repeatedly found to be activated during PLMS in previous studies. A transient increase in heart rate had been observed starting at the onset of the leg movements, followed by a decrease of the heart rate (Sforza et al. 1999; Sforza et al. 2002; Sforza et al. 2003; Ferrillo et al. 2004; Lavoie et al. 2004), as well as an increase in blood pressure (Ali et al. 1991). In order to investigate whether these changes might be relevant for the genesis of PLMS, we asked in particular the following two questions: does autonomic activation precede or follow PLMS? And is the magnitude of sympathetic or vagal autonomous activation correlated with PLMS severity?

3.2 SPECTRAL ANALYSIS OF HEART-RATE-VARIABILITY

It has been known since the 18th century that the heart rate varies with respiration and blood pressure (Hales 1733; Billman 2011). As the heart rate depends on the interaction between sympathetic and vagal efferent activities, it has been hypothesized that a quantitative assessment of its variability (HRV) might provide non-invasive information about autonomic regulation systems.

Importantly, HRV can be inferred conveniently from non-invasive ECG recordings. Several validation studies in the eighties have confirmed this and demonstrated that sympathetic and vagal activities are each associated with differential oscillatory frequencies in the HRV. A low frequency (LF) component at ~0.1 Hz is driven mainly by sympathetic efferences, and a high-frequency component at ~0.25 Hz driven mainly by vagal efferences. A LF/HF power ratio is therefore used for indicating a shift towards sympathetic tone. The usage of these components as biomarkers for sympathetic and vagal activity is supported by the following observations (Akselrod et al. 1981; Pagani et al. 1986).

1. Vagal blockade with an antimuscarinic drug (glycopyrollate) abolished the HF component in dogs.

2. Selective blockade of the renin-angiotensin system abolished the LF component in dogs.

3. Sympathetic activation with nitroglycerine infusions increased LF power in dogs. This response disappeared after bilateral stellectomy (removal of the main sympathetic relay station).

4. Combined β-sympathetic and vagal blockade abolished all heart-rate fluctuation in dogs.

5. Stimulating the sympathetic system with a tilt-test during which humans are rapidly brought from a supine to an upright position increases LF power and the LF/HF ratio.

6. β-adrenergic receptor blockage with propranolol given during the tilt-test reduced the LF response in humans.

7. Increases in blood pressure during the tilt test correlated with LF power increases.

These markers are therefore now frequently used to assess efferent sympathetic or vagal activity. It is noteworthy that the LF and HF components can also be influenced by confounding effects, in particular by the respiration rate (Pagani et al. 1986), the tidal volume, and drugs, which should be controlled during studies (Eckberg 1997; Billman 2011).

(12)

10 3.3 IMPROVING THE TEMPORAL RESOLUTION

The algorithms for calculating the HRV spectrogram can be based on Fourier transforms (Akselrod et al. 1981), autoregressive models (Pagani et al. 1986), or wavelet transforms (Pichot et al. 1999). All of them require a sample of about 10 minutes or more of recorded heart rate fluctuations to estimate the spectral components. This is obviously much too long to answer our question whether autonomic activation precedes or follows the onset of PLMS, for which we would need a time resolution of seconds. We have therefore tried to dramatically improve the temporal resolution of spectral HRV markers by using a similar approach as is commonly used in EEG analyses (Zygierewicz et al. 2005).

Rather than averaging the spectra over continuous time, we can average across repeated PLMS while keeping the time of each HRV data sample relative to the onset of PLMS constant. This allows us to study how spectral HRV markers evolve on average during a fixed period before and after the movements.

3.4 TIME-RESOLVED ASSESSMENT OF SYMPATHETIC AND VAGAL ACTIVITY DURING PLMS

Appendix I contains a reprint of our article (Guggisberg et al. 2007a) in which we show in more detail the procedures of this “time-frequency decomposition” of HRV and its application to our questions on the mechanisms of PLMS. In short, it was found that LF components of HRV indicating sympathetic activation preceded PLMS by up to 6 seconds. LF power and the LF/HF power ratio were significantly greater before PLMS than before control leg movements in sleep. Furthermore, the magnitude of sympathetic activation correlated with the severity of PLMS. This provides first evidence, although still incomplete, that the sympathetic nervous system might be the main trigger of PLMS. Furthermore, it suggests that our time-frequency decomposed HRV markers capture functionally relevant processes of sympathetic and vagal centers.

3.5 FUTURE DIRECTIONS

The time-resolved assessment of sympathetic and vagal activity might be useful also for numerous other research questions. It would provide advantages over other techniques which are currently used for this purpose. Compared to the measurement of pupil diameters (Granholm and Steinhauer 2004), it is not sensitive to external light intensity, which is of particular interest for cognitive studies requiring presentation of visual stimuli. Compared to skin conductance (Arunodaya and Taly 1995), it provides much better time resolution. It would therefore be desirable to further validate this approach.

With regards to the mechanisms of PLMS, future studies should investigate the effect of sympatholythic drugs (α- or β-blockers) on PLMS, sleep quality, and time-frequency components of HRV.

(13)

11

4 A SSESSING THE A ROUSAL L EVEL

Extracellular electromagnetic oscillations as recorded with EEG or MEG depend not only on cortical current generators but also on thalamocortical interactions and on ascending afferences from the brain stem (Zschokke 2002). Indeed, thalamic and brain stem processes can have profound but rather diffuse influences on the EEG. A typical example is the effect of vigilance. Vigilance is controlled in particular by the reticular formation in the brain stem (Moruzzi and Magoun 1949;

Steriade et al. 1991) and the posterior hypothalamus (Goutagny et al. 2004). The following chapter gives an example of EEG features allowing an objective measurement of vigilance.

4.1 YAWNING

Everybody yawns all the time. Humans do it from fetal intrauterine stages to old age. And almost all vertebrate animals yawn, from reptiles to mammals. Even fishes open their jaws in ways resembling the yawns of land-living species. Yet nobody knows why we actually do it. From an evolutionary perspective, the fact that yawning has survived evolution in numerous species strongly suggests that it must provide some useful advantage for survival. Yet, although many hypotheses have been forwarded, few have been tested with adequate experimental studies and none is generally accepted (see our review in Guggisberg et al. (2010) for more details).

One of the most widely held hypotheses proposes that yawning serves as homeostatic regulator of vigilance: when we become sleepy, we yawn in order to increase our arousal level (Baenninger 1997;

Walusinski and Deputte 2004; Matikainen and Elo 2008). Behavioral studies indeed suggest that yawning occurs mostly when subjects feel sleepy, i.e., in the early morning hours and before going to sleep (Provine et al. 1987; Greco et al. 1993; Giganti et al. 2007; Zilli et al. 2007; Zilli et al. 2008).

However, does the act of yawn itself influence vigilance? We addressed this question using EEG markers of vigilance and arousal level.

4.2 DELTA AND THETA POWER

Several lines of evidence consistently suggest that the power of slow EEG oscillations in the delta (0-4 Hz) and theta (4-7 Hz) frequency ranges are related to sleep pressure, sleepiness, and fatigue.

1. Slow wave activity (SWA) is defined as slow oscillations in the delta frequency range during human non-rapid eye movement (NREM) sleep. It is greatest at the beginning of the night and declines during subsequent sleep cycles. Sleep deprivation increases SWA in subsequent recovery sleep and this increase correlates with the duration of wakefulness (Borbely et al.

1981). Based on these observations, Borbely (1982) introduced SWA as sleep EEG marker of sleep pressure from which he derived a now widely used model of sleep homeostasis.

2. Subsequent studies have shown that delta, theta, and low alpha power in the waking EEG are also related to the duration of wakefulness and sleep pressure (Torsvall and Akerstedt 1988;

Lorenzo et al. 1995; Tinguely et al. 2006), with theta rhythms showing the most prominent and consistent effect during wakefulness.

(14)

12

3. Delta and theta activity in the waking EEG increase progressively before falling asleep (Tanaka et al. 2000; De Gennaro et al. 2001a).

4. Prolonged mental activity and fatigue increase delta and theta power in the waking EEG (Kiroy et al. 1996; Lal and Craig 2002)

5. Theta power correlates inversely with behavioral performance and alertness (Makeig and Jung 1996).

On the other hand, delta and theta oscillations also occur in focal brain pathologies (Zschokke 2002), brain immaturity, or motivational processes (Knyazev 2012), and can therefore not be considered as specific. Yet, when used to measure relative changes over time in healthy adult subjects they provide useful quantitative and objective information about fluctuations in sleep propensity and fatigue.

4.3 FREQUENCY AND TOPOGRAPHY OF ALPHA OSCILLATIONS

The influence of vigilance on alpha oscillations (~8-13 Hz) is more complex than for delta and theta frequencies. Global alpha power is therefore not as clearly related to sleepiness as slower oscillations. Nevertheless, systematic vigilance-related changes in the alpha pattern have been described since the early days of EEG (Loomis et al. 1937; Roth 1961). More recent research has introduced quantitative alpha-band correlates of vigilance.

1. From studies assessing EEG power changes during the transition from wakefulness to sleep, it is known that alpha activity decreases and moves in an anterior direction along the midline of the scalp with increasing drowsiness (Tanaka et al. 1997; De Gennaro et al. 2001a; De Gennaro et al. 2001b; Tsuno et al. 2002).

2. Oral ingestion of caffeine induces an attenuation and acceleration of alpha frequencies (Barry et al. 2005).

Hence, both sleepiness and vigilance reduce global alpha power, but sleepiness induces a slowing and anteriorization of alpha rhythms, whereas increased arousal levels are associated with an acceleration of alpha activity.

There is an extensive literature about the influences of many other factors influencing alpha rhythms such as task-induced cortical activity (Pfurtscheller 1992), attention, and working memory (Kolev et al. 2001; Jensen et al. 2002; Cooper et al. 2003), all of which superimpose on fluctuations in vigilance.

Nevertheless, when using rather long segments of EEG data and adequate controls, fast posterior alpha power can give useful information about the cortical arousal level.

4.4 YAWNING AND VIGILANCE

The paper reproduced in Appendix II (Guggisberg et al. 2007b) describes our study in which we have used these spectral EEG markers to investigate the functional relationship between yawning and vigilance. In short, delta power was found to be greater before and after yawns than before and after other control movements of the limbs or body of the same subjects. This suggests that the study participants were sleepier when they yawned than when they made other movements, consistent with the idea that yawns are triggered by sleepiness. However, there were no signs of increased vigilance after yawning. Delta power remained increased to the same extent as before the yawns.

Furthermore, alpha rhythms decreased, decelerated, and shifted in an anterior direction after yawning, as compared to the data segments before yawning, thus suggesting even decreased rather

(15)

13

than increased vigilance. Conversely, we did observe EEG markers of increased arousal levels after simple postural adjustments. Alpha rhythms became faster and smaller after body movements.

Hence, if yawning had an arousing effect – even if it were as small as the effect of simple postural adjustments – we would have detected it with our EEG analyses. Instead, we observed signs of progressive drowsiness after yawning.

Our results therefore confirm with objective and quantitative measures that we yawn when we are tired. However, we did not find evidence for an arousing effect of yawning itself.

4.5 FUTURE DIRECTIONS

Yawning has long been neglected by science and has only recently received renewed interest. In our recent review, we recapitulate the current state and propose future research directions (Guggisberg et al. 2010).

5 T RACKING THE D YNAMICS OF N EURAL C OGNITIVE

P ROCESSING

Non-invasive brain imaging methods have allowed unprecedented access to long-standing questions about how the brain deals with the many different cognitive problems we are confronted with every day, from basic visual perception to higher-order mind-reading. Traditionally, there have been two groups of imaging modalities available: positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) for precise localization of brain processes with poor time resolution, or EEG and MEG for precise measurement of their temporal evolution with low spatial resolution.

However, it has been impossible to obtain good resolution in both space and time. Many cognitive abilities involving highly parallel and transient neural processes in specialized brain areas throughout the brain have therefore been difficult to study. Recent advances in EEG and MEG signal processing further push the limits of combined spatio-temporal resolution in non-invasive imaging. They now allow addressing long-standing questions on the dynamics of neural cognitive processing which have previously remained elusive with traditional markers of neural activity.

5.1 DECISION-MAKING AND CONSCIOUS AWARENESS

We repeatedly make different kinds of decisions such as, among others, moving a limb when we want to (Deiber et al. 1991; Deiber et al. 1996; Ball et al. 1999), choosing our preferred ice cream flavor or even the future profession (Northoff and Bermpohl 2004; Northoff et al. 2006).

We have the strong and intuitive conviction that we can make these decisions and actions according to our own reflections, preferences, beliefs, and feelings. In other words, we take it for granted that the content of our consciousness somehow influences our decisions and actions. However, this assumption has been seriously challenged by a seminal study of Benjamin Libet published in the

(16)

14

eighties (Libet et al. 1983). Healthy human participants had to make repeated, self-paced finger movements while their brain activity was recorded with EEG. In addition, they were asked to watch a rapidly rotating clock and to report, after each movement, the clock hand position at the moment they had consciously “felt the urge to move”. It was found that a so-called Bereitschaftspotential (BP), a brain potential related to voluntary movements, starts several hundred milliseconds before the participants reported to be conscious about their decision to move. This result, which has been reproduced and extended by independent groups (Haggard and Eimer 1999; Trevena and Miller 2002; Soon et al. 2008), seems to indicate that consciousness about a movement decision arises only after the decision has been taken by unconscious neural processes.

The study and its replications have had massive impact on current concepts of volition and continue to be used as one of the main arguments against the existence of a freedom to choose. It seems to demonstrate that our actions are determined by unconscious neural processes whereas consciousness is merely a late byproduct of neural processing with no influence on its own.

However, this conclusion may have negative consequences for our conceptions of freedom and responsibility and hence for the functioning of society. Indeed, first studies have demonstrated that healthy volunteers reading texts coming to conclusions of this type immediately have a greater tendency to cheat (Vohs and Schooler 2008). It is therefore worth to examine the corresponding findings more closely. The conclusion is derived from a comparison of markers which are used to determine the onset of brain processes for movement preparation to subjective intention times. Let us examine whether these markers are valid and adequate.

5.2 THE BEREITSCHAFTSPOTENTIAL

Libet used the onset of the so-called BP as marker for the onset of neural movement preparation.

The BP is a slow, surface negative potential that can be obtained by recording EEG epochs before repeated self-paced movements and averaging them time-locked to movement onset (Kornhuber and Deecke 1965; Shibasaki et al. 1980). At least 2 different components can be distinguished: The first component starts about 2 sec prior to movement onset and is bilateral symmetrically distributed with a maximum over the supplementary motor area. The second component starts about 0.4 sec before movement onset and has an asymmetrical distribution with a maximum recorded above the contralateral primary motor cortex (Shibasaki et al. 1980; Shibasaki and Hallett 2006).

The usage of the early component of the BP as neural marker of internal movement generation has been justified by the following findings (Shibasaki and Hallett 2006).

1. The BP occurs only before self-paced but not before externally-paced movements (Libet et al.

1982; Papa et al. 1991). It is also usually absent before pathological and involuntary movements such as periodic leg movements (Trenkwalder et al. 1993), tics (Obeso et al.

1981), or myoclonus (Shibasaki and Kuroiwa 1975), although there are several exceptions (Keller and Heckhausen 1990; Karp et al. 1996; Oga et al. 2000).

2. Intracerebral depth recordings showed that the BP occurs mostly in regions belonging to the sensorimotor system, i.e., the supplementary motor area, primary motor and sensory cortices, premotor cortex and the cingulate gyrus, but not in temporal or parietal regions (Shibasaki and Hallett 2006; Sochurkova et al. 2006).

(17)

15

3. The early component of the BP has greater amplitude when a selection between 2 or more movements is made, compared to self-paced movements without selection (Dirnberger et al.

1998), and it increases with the level of intention (Shibasaki and Hallett 2006).

4. The BP is greater before sequential than before single movements (Kitamura et al. 1993).

However, the following arguments suggest that the BP is in fact not an adequate marker for actual movement decisions. Rather, it seems to represent a diffuse and non-specific preparation of the cortex for future tasks.

1. The BP belongs to the family of direct current (DC) potentials, the origin and generators of which are incompletely understood and a matter of debate. There is evidence for neuronal mechanisms generating DC potentials, including excitatory postsynaptic potentials at apical dendrites due to synaptic input from unspecific thalamic afferences and axonal collaterals (Caspers et al. 1980; Birbaumer et al. 1990), and hyperpolarization of the neuronal membrane following sustained and coordinated spiking of nearby neurons (Buzsaki et al.

2012). In addition, non-neural current sources such as glial cells, extracellular potassium concentrations and potential differences across epithelia of the blood-brain barrier seem to contribute as least as much to DC potential shifts at the surface (Voipio et al. 2003). Negative shifts in DC potentials can be recorded under many different circumstances including during the transition from sleep to wakefulness, sensory stimulation, attention shifts, hypoxia, and epileptic seizures (Caspers and Schulze 1959; Caspers et al. 1980; Birbaumer et al. 1990).

Given these mechanisms and circumstances, DC potentials likely represent unspecific regional workload, attention and vigilance rather than specific neural computation.

2. DC potentials similar to the BP can also occur before expected external stimuli without movements (Brunia 1988) or even before subjects decide not to move (Trevena and Miller 2010). Due to the different context of occurrence, these DC potentials are not named BP but contingent negative variation (CNV) or stimulus preceding negativity (SPN). Nevertheless, they share common mechanisms and configurations (Brunia 1988). In any case, CNV and SPN demonstrate that expecting a future event is sufficient to trigger DC potentials and hence movements may not be necessary for the generation of BP either.

3. Not all self-paced movements are preceded by a BP. Hence, the BP is not necessary for internal movement generation (Pockett and Purdy 2010).

4. The amplitude and configuration of the BP can superimpose on other DC potentials related to concomitant tasks such as spatial attention or visual processing (Lang et al. 1984), in which case it is impossible to distinguish between motor and non-motor processes. In this regard, the concomitant clock-task used for determination of the onset of consciousness in Libet’s experiment is of particular concern, as it may have contaminated the onset of the BP. Several studies have found that this is indeed the case: instructing the subjects to pay attention to their decisions leads to an earlier onset of the BP (Keller and Heckhausen 1990; Miller et al.

2011). Hence, the mere fact that the subjects in Libet’s experiment had to determine the onset of their conscious urge to move seems to have biased the marker used for assessing neural movement preparation.

5. Even random fluctuations in neural activity can produce a potential similar to the BP if averaged time-locked to movement onset (Schurger et al. 2012).

(18)

16

6. Decisions are likely based on complex overlapping stages of evaluations and information processing. The BP does not provide useful information about the different stages of the neural decision process but merely shows a steady potential increase in the motor system.

7. Lesions to the posterior parietal cortex (Sirigu et al. 2004) and the cerebellum (Kitamura et al.

1999) lead to an important reduction or even disappearance of the BP, thus suggesting that it depends not only on the motor system but on very distributed neural networks.

8. The onset of the early component of the BP does not co-vary with the reported time of conscious awareness. This provides some evidence against Libet’s conclusion that subconscious neural processes underlying the BP cause subsequent consciousness of motor decisions (Haggard and Eimer 1999).

Hence, although Libet used the best available marker of neural motor processing of his time, it must be considered insufficient for the purpose of his study today. The fact that the BP preceded conscious decisions in Libet's study merely shows that the study participants prepared for the upcoming movement decisions in each trial. It does however not demonstrate that unconscious brain processes actually took the decision to move before it became conscious. To make this point clear, we have to consider 2 critical events which occurred in each trial of Libet’s paradigm. First, the subjects prepared for and expected the upcoming task. Second, they “felt an urge to move” and decided to move precisely now. The first event occurred at the very beginning of each trial, the second at some later time shortly before the movement. The subjects were specifically instructed to report the conscious onset of the second event. Conversely, the BP onset does not mark the second but the first event and, in addition, is possibly biased by concurring neural processes such as the clock-task. Since the first event does not seem to influence the precise timing of the second event (Haggard and Eimer 1999) (beyond the obvious determination that it must occur within the same trial), the BP is not a valid marker of neural processes underlying event 2. Libet’s study, although ingeniously conceived, turns out to compare apples with oranges, i.e., markers for two essentially independent event times. It is noteworthy that Libet himself proposed that voluntary acts are generated by at least two different processes (Libet et al. 1982). He tried to focus on the second event only by using only certain subtypes of BP, which he thought to be more specifically related to the actual movement decision.

Yet, the question of how and when conscious considerations influence decisions and actions remains an important one. Several groups have therefore tried to find a better marker for investigating neural decision processes, which might enable us to address Libet’s original question more appropriately.

5.3 THE LATERALIZED READINESS POTENTIAL

After Libet’s paper, psychologists have introduced a variant of the BP as more specific marker for movement preparation: the lateralized readiness potential (LRP) (Coles 1989). It is obtained by subtracting the BP of the central electrode located over the hemisphere of the same side as the moved limb from the BP of the electrode on the opposite side. It therefore measures the extent to which the contralateral motor cortex mediating a movement is more active than the ipsilateral motor cortex. It depends mostly on the late component of the BP which is more directly related to actual movement preparation than the early component (Shibasaki et al. 1980). Hence, although much less information about the specificity and the precise meaning of the LRP is available, it seems to avoid many of the problems of the BP.

(19)

17

Haggard et al. (1999) used the LRP in a slight modification of Libet’s original paradigm. Rather than just performing repetitive, self-paced movements with one hand, the participants of their study additionally had to choose between left and right hand movement in each trial. This modification allowed them to use the onset of the LRP as the moment at which the brain decided to prepare the movement of the contralateral rather than the ipsilateral hand. This time was then again compared to the time of conscious decisions that the participants reported using the clock paradigm developed by Libet. They observed that subjects reported to be conscious about their decisions on average 350 ms before the movements, whereas the LRP started already about 800 ms before the movements.

Hence, the participants reported to be conscious about their choice only about 450 ms after the brain had started to prepare the chosen action. This confirms Libet’s original finding with a more robust marker of neural motor preparation. Yet, the LRP is still derived from an unspecific DC potential.

5.4 FMRIMOVEMENT PREDICTORS

When using sophisticated classifiers to learn and decode the intentions of human participants from fMRI signals, researchers were able to predict, with a small but significant probability of 55 to 60%, future decisions already up to 7 seconds before the subjects reported to become aware of their decisions and up to 8 second before the actual movements. The onset of conscious decisions was measured using a variant of Libet’s original clock paradigm optimized for fMRI. As in Libet’s original study, these findings again suggested that “a network of high-level control areas begins to prepare an upcoming decision long before it enters awareness” (Soon et al. 2008). This time, the predictors of future decisions were validated with state of the art techniques.

5.5 NEURAL FIRING RATE

Fried et al. (2011) recorded activity of neurons in human subjects while they performed self-paced finger movements. They observed a progressive recruitment of neurons in the supplementary motor area over about 1500 ms before subjects reported to be conscious about their decisions. Hence, changes in firing rate of individual neurons showed a similar time course as the BP recorded at the surface. More importantly, they were able to predict the time point of future movements from the firing rate of a population of neurons in the supplementary motor area, already 700 ms before subjects reported to be aware of their decisions. The predictive value of firing rate for motor decisions was verified with crossvalidation techniques.

5.6 HIGH-GAMMA OSCILLATIONS

Traditionally, the analysis of the EEG has been limited to frequencies below about 40-70 Hz, because faster rhythms can be contaminated by cranial muscle activity and environment noise. However, more recent studies using intracranial and surface EEG recordings have consistently demonstrated that fast neural oscillations in the so-called gamma and high-gamma frequency range (~40-200 Hz) are in fact reliable and specific markers of local neural processing that outperform traditional EEG/MEG and fMRI markers in combined spatiotemporal resolution.

1. Like other EEG/MEG rhythms, gamma and high-gamma oscillations depend on postsynaptic currents and therefore on the synaptic input (Buzsaki et al. 2012).

(20)

18

2. In contrast to other EEG/MEG rhythms, they also correlate with the spiking rate of nearby neurons (Rasch et al. 2008; Whittingstall and Logothetis 2009). Hence they also contain information about the output of local neural computation.

3. Unlike other EEG/MEG rhythms, they correlate positively with the fMRI hemodynamic response (Logothetis et al. 2001; Brovelli et al. 2005; Niessing et al. 2005). Hence they reflect local neural activity while having much better time resolution than fMRI.

4. They are task-specific, occurring only during tasks which activate the local brain area, for instance only during word production in Broca’s area.

5. They are spatially more focal than slow neural oscillations and event-related potentials (Brovelli et al. 2005; Edwards et al. 2005; Crone et al. 2006; Canolty et al. 2007).

6. In intracranial recordings, they have a sufficiently high signal-to-noise ratio to allow tracking even the time course of neural processing in single trials (Edwards et al. 2010).

Intracranial recordings of high-gamma oscillations therefore clearly outperform other indices of neural activity for assessing the dynamics of cortical processing. However, they have the disadvantage of being invasive and limited to a small brain region covered by an electrode grid.

Furthermore, they are currently only available from patients with epilepsy undergoing presurgical evaluation. We therefore tried to reconstruct them non-invasively and from the entire brain with EEG or MEG.

Compared to the reconstruction of evoked potentials and slower rhythms (Michel et al. 2004), non- invasive imaging of high-gamma oscillations poses the additional problem that the signal-to-noise ratio of high-gamma rhythms at the surface is generally low and lower than for other rhythms. This probably has several reasons (Jerbi et al. 2009; Buzsaki et al. 2012). As oscillations in the high-gamma range are generally more focal than in the lower frequencies, less signal can be summated for reaching the scalp. Moreover, higher frequencies exhibit larger relative phase variability than low frequencies, and the summation of these polyphasic sources lead to lower amplitude at the surface (Pfurtscheller and Cooper 1975). The morphological configuration of the dendrites of pyramidal neurons also seems to act as low-pass filter. It is a matter of debate whether the capacitive and inductive properties of the brain tissue additionally cause high-frequency attenuation. Finally, surface signals can be contaminated by EMG from oculomotor and head muscles with spectral characteristics similar to that of gamma-band activity (Yuval-Greenberg et al. 2008).

In collaboration with bioengineers at the University of California, San Francisco, we have optimized existing source localization algorithms for specific frequencies of interest. Appendix III contains the reprint of the paper (Dalal et al. 2008), in which we describe and validate the so-called time- frequency beamformer, an inverse solution allowing the localization of oscillations in all frequencies, and especially also in high-gamma frequencies, with excellent spatial and temporal resolution. Given sufficient repetitions of a task (at least ~100 trials), this technique can reliably reconstruct high- gamma oscillations from MEG (and possibly EEG) surface recordings.

Gamma and high-gamma power modulations occur in a broad frequency band from roughly 40 to 200 Hz, but task-induced power modulations typically occur only in some of these frequencies.

Furthermore, different subjects and different brain regions tend to present different patterns of high-gamma frequencies that are increased during a given task (Hoogenboom et al. 2006). This potentially leads to problems when analyzing activations across different subjects and in different

(21)

19

brain regions. If we look at the broad high-gamma band, we risk losing information due to noise and fluctuations in silent frequencies. If, on the other hand, we use narrow bands only, we risk missing activations in neighboring bands. In order to overcome this problem, we statistically combine multiple, relatively narrow high-gamma frequency bands to a single broadband time-series of neural activation. Appendix IV shows the reprint of the paper in which we describe this procedure in more detail (Guggisberg et al. 2008a).

5.7 NEURAL DYNAMICS OF HUMAN DECISION-MAKING

The new tool for assessing high-gamma activity enabled us to watch the brain decide, i.e., to look into the dynamic neural processes underlying human decision-making. The paper in Appendix IV describes and discusses the sequence of cortex activation during a so-called forced choice task, during which the participants of the study were required to choose, in each trial, one of two options (Guggisberg et al. 2008a). The observed overlapping activations can conceptually be divided into 4 processing stages corresponding to 4 cognitive steps of choice. Similar dynamics have also been described in other studies (eg. Siegel et al. 2011). For an illustrative example, assume that an individual chooses his preferred ice cream.

1. Perception of the options mediated by primary and secondary sensory cortices; e.g., the individual sees several kinds of ice cream.

2. Retrieval of information regarding the value of the options mediated by several different and specialized brain areas; e.g., our ice-cream lover checks appearance and price of the different kinds of ice cream, recalls memories about earlier experiences with the same ice cream, and analyses his personal preferences.

3. Attribution of an overall value to each option and formation of an intention corresponding to the option with the highest current value, mediated by the parietal cortex; e.g., the person opts for the chocolate flavor.

4. Execution of the action corresponding to the choice, mediated by the motor cortex; e.g., the individual grabs the chosen ice cream.

Videos of the reconstructed brain activations during different types of choice can be watched at http://www.mrsc.ucsf.edu/~aguggis/frontiers/suppinfo.html.

To come back to Libet’s question, how then do these neural stages of choice compare to the time of conscious awareness reported by the participants? We addressed this in the paper reprinted in Appendix V using our high-gamma markers of neural processing together with the clock method introduced by Libet (Guggisberg et al. 2011a). We found that subjects report to be conscious at decision stage 3, i.e., at the time when brain regions responsible for the formation of intention become active [see Figure 6 in Guggisberg et al. (2011a)], about 250 ms before onset of the movement of the chosen side. This is also the time point at which high-gamma activity in the motor and parietal cortex contralateral to the moved finger starts to increase more than high-gamma activity in the ipsilateral hemisphere (see Supplementary Figures 3 and 4) indicating that the brain starts preparing the chosen action. Hence, when using high-gamma activity of the motor cortex as specific marker of cortical movement preparation, we did not find evidence for a delayed onset of conscious awareness in forced choice tasks.

(22)

20

If we leave aside Libet’s original study using a problematic marker for determining the onset of neural movement preparation, the studies using improved markers yield conflicting results. How can we explain the divergences between our study using high-gamma oscillations in the primary motor cortex and previous studies using fMRI predictors, neural firing rate, or the LRP? The highly variable onset times of neural events predicting decisions strongly suggest that the different markers capture different stages in the dynamic process of decision-making. As described above, human decision- making involves several consecutive but overlapping stages of option perception, evaluation, intention, and action execution. Each stage may be conscious, but the study participants were explicitly instructed to report the time when they “make their conscious motor decision” or when they “feel the urge to move”. It is likely that the participants interpreted this as having to indicate the time when a decision was reached and a corresponding intention made and not as the time when they started considering the options. Indeed, when measuring high-gamma oscillations we found that the times reported by the subjects match the activation times in brain regions responsible for intention formation in decision stage 3 (Guggisberg et al. 2011a). Now, the onset of the high-gamma marker for movement preparation in the motor cortex was also in the same intention stage of decision-making which explains the close match between neural and introspective measures of intention onset, about 250 ms before action execution. Conversely, the onset of LRP and especially of the fMRI predictors was considerably earlier, probably during decision stages of option evaluation.

During rational decision-making, choices are based on values attributed to the available options (Sugrue et al. 2004; Sanfey et al. 2006). Hence, the option evaluation stages must determine, at least for rational choices, the decision made in subsequent intention stages such that the option with greatest final value is chosen. It is therefore not surprising that one can decode with a certain probability future decisions from early neural processes underlying option evaluation and value representation. The fact that the study participants did not yet report their decision at this time does, however, not mean that it was unconsciously initiated, but simply that it was not final yet. If the subjects had been instructed to report the earliest time when they consciously start considering the options, they would almost certainly have reported much earlier onset times that might even have matched the times obtained with the fMRI classifiers.

5.8 NEURAL MECHANISMS OF INTROSPECTION

We have wondered what actually happens in the subjects’ mind and brain when they introspect the onset time of their conscious decisions. A straightforward answer proposed already by the philosopher Brentano (1874) would be that introspection arises directly and automatically from ongoing conscious processing. Subjects perform the tasks of the Libet paradigm with different thoughts entering and leaving consciousness and as soon as the content of consciousness is their decision, they can automatically memorize and report it. However, several lines of evidence suggest that it is not as simple as this. The phenomenologist tradition which is particularly devoted to the examination of the mind with introspection has long pointed out that ongoing consciousness is not the same thing as introspection (Husserl 1984; Gallagher and Zahavi 2010). Ongoing primary consciousness is defined as the continuous stream of contents in consciousness, whereas introspection is defined as an additional intermittent monitoring and reassessment of the contents of primary consciousness. Evaluating different ice cream flavors and choosing one of them would be an example of content in primary consciousness, whereas realizing that I have just now made a decision about my preferred ice cream is an instance of introspection. More recent research has experimentally corroborated that there are dissociations between basic ongoing consciousness and

(23)

21

intermittent introspection and that introspection can lead to transformations and misrepresentations of the original primary conscious experience (Schooler 2002; Marcel 2003).

Our new tool for imaging high-gamma oscillations allowed us to examine the neural processing underlying event-time introspection during Libet’s clock paradigm (Guggisberg et al. 2011a) (see the paper in Appendix V). Event-time introspection is defined as the process during which subjects re- represent primarily conscious events such as their movement intentions in order to determine their onset time. We found that event-time introspection was associated with specific neural activity at the time of subjective event onset which was spatially distinct from activity induced by the decision itself (see Figure 5 in (Guggisberg et al. 2011a)). Hence, healthy humans recruit specific introspection networks to access relevant neural processes in separate decision-making networks. For a thorough discussion of the implications of this finding see the paper in Appendix V. In short, this means that subjective event times do not directly reflect the original conscious experience of the participants (which is mediated by decision-making networks) but depend on interactions with additional introspection processes.

Introspection processes can thereby distort the original conscious experience. Since the neural mechanisms of introspection have resource constraints, they are susceptible to interference with other ongoing neural activity. Hence, introspection may be disturbed by concurring tasks that require neural resources in close temporal proximity to the act of introspection, as it has indeed been observed in several studies (Moutoussis and Zeki 1997; Eagleman and Sejnowski 2000; Haggard et al.

2002; Stetson et al. 2006; Corallo et al. 2008). The resulting error can be in the order of ~50ms.

Moreover, introspective reports are additionally influenced by perceptual information obtained after the decision and the action. Thus, it is possible to manipulate the reported decision times by presenting a deceptive auditory signal just after the hand movements chosen by the subjects (Banks and Isham 2009), or by transiently disturbing the neural processing just after the movements with a magnetic pulse applied to the supplementary area (Lau et al. 2007). Finally, introspection can misrepresent the original experience for instance when one verbally reports non-verbal or ambiguous experiences (Schooler 2002).

Taken together, evidence from several lines of research suggests the existence of two distinct types of consciousness: a primary consciousness and an intermittent introspective consciousness which is also named meta-consciousness (Schooler 2002). Subjective reports obtained in the Libet paradigm arise in meta-consciousness (Guggisberg et al. 2011a). Hence, even though early decision stages seem to be “meta-unconscious”, i.e., ignored by or inaccessible to introspection/meta- consciousness, they may still be primarily conscious.

Evidence that this is indeed the case comes from a study which used an alternative method to the Libet clock for determining the onset of conscious movement decisions. Subjects were presented tones in random intervals before self-paced movements. Each tone prompted them to decide immediately (rather than using post-hoc recall as in the Libet paradigm), whether or not there was an intention to move. The onset of conscious decisions measured with this alternative method turned out to be much earlier that when measured with the Libet clock, i.e., on average 1.4 s before the movements (Matsuhashi and Hallett 2008).

(24)

22 5.9 SUMMARY AND CONCLUSIONS

The preceding sections may have shown that the marker used in Libet’s original study for assessing neural movement preparation is insufficient for supporting far-reaching conclusions about the influence of our consciousness during decision-making. We have introduced an improved marker for assessing the dynamics of neural processing, which demonstrates that neural and introspective times of intention onset do in fact match closely. Significant predictors of movement decisions can be decoded from the brain already before neural and introspective intention stages. The question whether these earlier neural predictors are conscious is currently unresolved and requires more refined methods for assessing subjective consciousness. In the meantime, we should avoid far- reaching conclusions with potentially negative consequences for society.

5.10 FUTURE DIRECTIONS

Non-invasive imaging of high-gamma activity opens exciting perspectives that will likely be useful also for other research questions. Future studies will have to examine whether reconstructions of cortical high-gamma activity, which were validated for MEG recordings (Dalal et al. 2008; Dalal et al.

2009), can also be obtained from EEG recordings.

Like the assessment of neural activity, the measurement of consciousness turns out to have its pitfalls and to require special care. However, there is no reason to go back to behaviorist reasoning that any scientific study using introspection and subjective reports is futile. On the contrary, future efforts should take into account current knowledge about the structure of consciousness and the validity of different methods for measuring consciousness. Now that we have available valid markers of neural processing, we should try to develop and use refined markers also for the assessment of the contents of consciousness. Several promising alternatives to Libet’s clock task are already available (Marcel 2003; Matsuhashi and Hallett 2008; Fahle et al. 2011; Fox et al. 2011). In particular, it seems to be important to use immediate and continuous measures to study the timing of the contents of consciousness.

6 I NVESTIGATING N EURAL N ETWORKS

Several decades of research with electromagnetic brain signals, but also with fMRI and PET, have consistently demonstrated that specialized brain regions respond to external stimulation and tasks.

These so-called activations can consist of potential changes, or local increases in oscillatory power, metabolism, and oxygen availability. They are obtained by subtracting control or baseline conditions from an active task condition of interest. The success of this approach has led to a view of the brain as a collection of specialized areas which are specifically recruited during, and responsible for, certain specific tasks. Systematic studies of the consequences of acquired brain lesions in patients have further corroborated this concept by showing that circumscribed brain damage can lead to relatively specific and reproducible behavioral deficits. For instance, speech production leads to an activation

(25)

23

of the left posterior inferior frontal gyrus and damage to this area leads to a so-called Broca aphasia consisting of difficulties with speech production.

However, this view emphasizing the functional segregation of brain areas neglects the fact that these areas are massively and reciprocally interconnected (Sporns et al. 2004). Brain lesions therefore not only induce loss of local neural function, but also distant changes in a larger network of brain interactions (Honey and Sporns 2008; Alstott et al. 2009). Similarly, task-induced brain activations are rarely limited to a single brain region but typically comprise an interacting network. It has long remained unknown how the massive structural connectivity of the brain influences brain function, but advances in signal analysis in the last 10 years have enabled new technical tools and led to a paradigm shift in our understanding of brain function and disease.

6.1 FUNCTIONAL CONNECTIVITY

While the term structural connectivity refers to neural fiber tracts connecting brain regions, the term functional connectivity (FC) describes interregional neural communication. Interregional neural communication is thought to be accompanied by a synchronization of oscillations between different brain regions (Aertsen et al. 1989; Gray et al. 1989; Gray and Singer 1989; Engel et al. 1992). FC can therefore be quantified with measures of similarity and synchronization between activities in different brain regions. If two or more regions show similar and synchronous activity they are considered to be interacting and communicating, i.e., to be functionally connected. Numerous algorithms for calculating FC have been introduced (e.g. Nunez et al. 1997; Lachaux et al. 1999; Gross et al. 2001; Stam et al. 2003; Nolte et al. 2004; Pascual-Marqui et al. 2011) with each having its advantages and disadvantages depending on the context and the question asked.

EEG researchers have been the first to try to characterize functional network interactions by measuring coherence between EEG electrodes (Lopes da Silva et al. 1973; Thatcher et al. 1986).

However, indices of functional interaction calculated between surface electrodes cannot be attributed to brain regions of interest or to structural lesions. Field spread leads to a wide representation of sources in many sensors, which makes the interpretation of functional connectivity measures between sensor pairs difficult (Srinivasan et al. 2007; Schoffelen and Gross 2009). Only when fMRI studies started to look at functional network interactions, it became clear that they have a distinct and consistent spatial configuration.

6.2 RESTING-STATE NETWORKS IN FMRI

When fMRI is recorded at rest, it contains spontaneous slow signal fluctuations. During many years, these resting-state fluctuations were treated as noise and subtracted out in the analyses. However, Biswal et al. (1995) observed that they were in fact highly correlated among regions belonging to the motor network, but not between motor and non-motor areas. Subsequent studies confirmed that spontaneous fluctuations of activity in the resting brain are highly organized and coherent within specific neuro-anatomical systems (Greicius et al. 2003; Fox et al. 2005; Damoiseaux et al. 2006). For instance, the bilateral intraparietal sulcus and frontal eye fields, which are thought to mediate spatial attention, show highly correlated fluctuations among each other but not with other unrelated areas.

These patterns of coherence are task-independent and can also be studied at rest (Greicius et al.

2003) for which reason they are named resting-state networks. The topography of resting-state networks matches the topography of brain activations induced by corresponding tasks (Vincent et al.

Références

Documents relatifs

examine interrelationships between disease duration, fear of dyspnea/ physical activity, and GMV in brain areas where patients demonstrated reduced GMV (using the gray scale

In this paper we proposed a new algorithm to track the functional brain connectivity dynamics and segment hr- EEG signals recorded during picture recognition and

part of our efforts to develop methods of chemical synthesis in flow, we now report a simple and expeditious means for conducting a variety of useful copper- promoted reactions 6

tmRNA Decreases the Bactericidal Activity of Aminoglycosides and the Susceptibility to Inhibitors of Cell Wall

Model results from the Dense Emerge + Water Depth model for nest site selection of Least Bittern nests monitored during May-August 2013–2019 at Emiquon Preserve, Fulton

Detection covariates are as follows: none – only an intercept was included; time – hours after sunrise for the day of survey; year – categorical effect of year of survey, which

Bayesian p-values for half-normal and negative-exponential formulations of winter and spring community detection models when model included no other detection covariates..

D_Arb = density of trees, D_arb = density of shrubs, A_Arb = tree height, DBH = diameter at breast height, PA= patch area (ha), SI= Shape index, FD= fractal dimension, DistLoc=