Investigating the dynamic role of fluctuations in ongoing activity in the human brain

Contents

Declaration

Abstract

Chapter 1 General introduction
1.1 Spontaneous and evoked activity
1.2 The study of spontaneous activity
1.2.1 Electrophysiological research of ongoing activity
1.2.1.1 Cortical states and response variability
1.2.1.2 Predictive coding and predictive timing
1.2.2 Neuroimaging research of ongoing activity
1.2.2.1 Resting state fluctuations
1.2.2.2 Vascular basis
1.2.2.3 Neural basis
1.2.2.4 Functional networks
1.3 The functional role of spontaneous activity
1.3.1 Perceptual domain
1.3.2 Motor domain
1.3.3 Cognitive domain
1.4 Conclusions
1.5 This thesis

Chapter 2 Methods - measuring spontaneous activity
2.1 Introduction
2.2 Group versus inter-individual differences versus lesion studies
2.3 Functional Magnetic Resonance Imaging (fMRI)
2.3.1 Overview
2.3.2 The BOLD response
2.3.3 How to collect resting state data
2.3.4 Pre-processing and noise correction
2.3.5 Functional connectivity analyses of resting state data
2.3.6 DCM - or: going beyond functional connectivity
2.3.6.1 Effective connectivity
2.3.6.2 Deterministic dynamic causal modelling
2.3.6.3 Stochastic DCM
2.3.6.4 Model selection
2.4 Electroencephalography (EEG)
2.4.1 Event-related potentials (ERPs)
2.4.2 Time-frequency analyses (TFAs)

Chapter 3 Early visual learning induces long-lasting connectivity changes during rest in the human brain
3.1 Introduction
3.2 Materials and methods
3.2.1 Participants
3.2.2 Stimuli and task design
3.2.3 Experimental procedure
3.2.4 Behavioural analysis
3.2.5 fMRI data acquisition
3.2.6 fMRI data analysis
3.2.6.1 Perceptual learning session
3.2.6.2 Psychophysiological interaction analysis
3.2.6.3 Dynamic causal modelling
3.3 Results
3.3.1 Participants showed early rapid learning of the motion task
3.3.2 Motion task activated visual, frontal and parietal areas
3.3.3 Early learning-related modulation of hippocampal activity during task performance
3.3.4 Learning-related changes in connectivity during rest
3.3.5 Dynamic causal modelling
3.4 Discussion
3.5 Conclusion

Chapter 4 The role of prestimulus activity in visual extinction
4.1 Introduction
4.1.1 The phenomenon of visual extinction
4.1.2 How does visual extinction relate to spatial neglect?
4.1.3 Mechanisms of visual extinction
4.1.4 Prestimulus activity affects perception
4.1.5 Can I analyse visual extinction using prestimulus activity?
4.2 Materials and methods
4.2.1 Participant
4.2.2 Design and procedure
4.2.2.1 Neuropsychological testing
4.2.2.2 fMRI paradigms
4.2.2.2.1 Extinction paradigm (event related design)
4.2.2.2.2 Stimulus localiser (block design)
4.2.2.2.3 Stimuli
4.2.3 fMRI data acquisition
4.2.4 Data analysis
4.2.4.1 Behavioural data
4.2.4.2 fMRI data
4.2.4.2.1 Extinction paradigm
4.2.4.2.2 Stimulus localiser
4.2.4.2.3 Peristimulus time histograms (PSTH)
4.2.4.2.4 Dynamic causal modelling (DCM)
4.3 Results
4.3.1 Patient showed signs of visual extinction
4.3.2 Stimulus localiser activated visual areas
4.3.3 Extinction paradigm produced unseen trials
4.3.4 Prestimulus activity in visually responsive areas affects perception
4.3.5 Time-course of responses to seen and unseen trials
4.3.6 Perception depends on the coupling between visual areas
4.4 Discussion
4.4.1 Prestimulus activity in visual areas affects stimulus perception
4.4.2 Prestimulus activity in other brain areas might play a role
4.4.3 Mechanisms behind visual extinction
4.4.4 Limitations of the study
4.4.5 Methodological aspects
4.5 Conclusion

Chapter 5 Effects of ongoing cortical state on ambiguous perception
5.1 Introduction
5.2 Materials and methods
5.2.1 Participants and apparatus
5.2.1.1 Stimuli
5.2.1.2 Training and thresholding
5.2.1.3 Behavioural task during EEG
5.2.1.4 EEG data acquisition
5.2.1.5 fMRI data acquisition and analysis
5.2.2 EEG data analysis
5.2.2.1 Pre-processing
5.2.2.2 ERP analysis
5.2.2.3 Prestimulus analysis
5.3 Results
5.3.1 Behavioural results
5.3.1.1 Performance and response patterns
5.3.1.2 Reaction times
5.3.2 Event-related potentials
5.3.3 Time frequency analysis of prestimulus activity
5.4 Discussion
5.4.1 ERP results
5.4.2 Alpha band oscillations
5.4.3 Beta band oscillations
5.4.4 Gamma band oscillations
5.4.5 Conclusion and future direction

Chapter 6 The relationship between mind-wandering, creativity and neuronal coupling
6.1 Introduction
6.2 Materials and methods
6.2.1 Participants
6.2.2 Stimuli and task design
6.2.3 Experimental procedure
6.2.4 Behavioural analysis
6.2.4.1 UUT
6.2.4.2 Thought probes
6.2.4.3 Target detection
6.2.5 fMRI data acquisition
6.2.6 fMRI data analysis
6.2.6.1 Pre-processing
6.2.6.2 Block task
6.2.6.3 Incubation task
6.2.6.4 ROIs
6.2.6.5 Dynamic causal modelling
6.3 Results
6.3.1 Behavioural results
6.3.2 Imaging results
6.3.2.1 Task-active regions
6.3.2.2 DMN regions
6.3.2.3 Stochastic DCM
6.4 Discussion
6.4.1 Behavioural results
6.4.2 Imaging results
6.4.3 Limitations
6.5 Conclusion

Chapter 7 General discussion
7.1 Overview of findings
7.2 Implications of this research
7.2.1 Ongoing activity predicts perception
7.2.2 Ongoing activity is modulated by learning and trait variables
7.2.3 Cause and effect: the interplay between ongoing and evoked activity
7.3 Outstanding questions and conclusion
7.3.1 Timescale of changes in ongoing activity and its relation to structural changes
7.3.2 Origin and scale of ongoing activity
7.3.3 Conclusion or: The function of ongoing activity

References

Abstract

Traditionally, the focus in cognitive neuroscience has been on so-called evoked neural activity in response to certain stimuli or experiences. However, most of the brain’s activity is actually spontaneous and therefore not ascribed to the processing of a certain task or stimulus - or in other words, uncoupled to overt stimuli or motor outputs. In this thesis I investigated the functional role of spontaneous activity with a focus on its role in contextual changes ranging from recent experiences of individuals to trial-by-trial variability in a certain task. I studied the nature of ongoing activity from two perspectives: One looking at changes in the ongoing activity due to learning, and the other one looking at the predictive role of prestimulus activity using different methodologies, i.e. EEG and fMRI. Finally, I ventured into the realm of inter-individual differences and mind-wandering to investigate the relationship between ongoing activity, certain behavioural traits and neuronal connectivity.

List of Figures

Figure 3-1 Experimental paradigm

Figure 3-2 Behavioural learning and hippocampal activation

Figure 3-3 Winning model and summed (group) log evidence for all models

Figure 3-4 Parameter estimates and model fitting reflected consolidation

Figure 4-1 The extinction paradigm

Figure 4-2 Right parietal lesion

Figure 4-3 Stimulus localiser activated visual areas in both hemispheres

Figure 4-4 Behavioural results of the extinction paradigm

Figure 4-5 Visually responsive areas are more active before bilateral seen trials

Figure 4-6 Peristimulus time courses show difference before stimulus onset

Figure 4-7 Differences in effective connectivity before bilateral seen trials

Figure 5-1 Random dot motion stimulus

Figure 5-2 Response pattern across participants

Figure 5-3 Response pattern over blocks

Figure 5-4 Response repetitions for random and coherent percepts

Figure 5-5 Grand averages comparing correct subliminal and supraliminal trials

Figure 5-6 Grand averages comparing periliminal trials

Figure 5-7 Low frequency prestimulus analysis

Figure 5-8 High frequency prestimulus analysis

Figure 5-9 High frequency prestimulus analysis

Figure 6-1 Experimental paradigm

Figure 6-2 Interparticipant differences in mind-wandering and awareness

Figure 6-3 Task activations at fusiform gyrus

Figure 6-4 Correlation between creativity and brain connectivity

List of Tables

Table 3-1 Main effect of the motion learning task compared to baseline

Table 3-2 Main effect of the motion learning task compared to the static control task

Table 4-1 Stimulus localiser activated visual areas

Table 4-2 Activity differences during the extinction paradigm for the baseline period testing for areas that show higher activity before BS compared to BU trials

Table 5-1 Individual MT coordinates

Table 6-1 Individual coordinates of task related activation in the fusiform gyrus

Chapter 1 General introduction

1.1 Spontaneous and evoked activity

Traditionally, the focus in cognitive neuroscience has been on so-called evoked neural activity in response to certain stimuli or experiences. However, most of the brain’s activity is actually spontaneous and therefore not ascribed to the processing of a certain task or stimulus - or in other words, uncoupled to overt stimuli or motor outputs. Possibly, the existence of ongoing intrinsic activity was first noted by Hans Berger when he introduced electroencephalography for humans in 1929 (Berger, 1929), asking whether “it [is] possible to demonstrate the influence of intellectual work upon the human electroencephalogram, insofar as it has been reported here?” to conclude subsequently that “[o]f course, one should not at first entertain too high hopes with regard to this, because mental work, as I explained elsewhere, adds only a small increment to the cortical work which is going on continuously and not only in the waking state”. Four years later, Bishop (1933) reported the potential physiological significance of the ongoing activity describing his experiments with rabbits. He observed cyclic changes in the excitability in visual cortex during stimulation of the optic nerve. Summarising his findings, he stated that “[…] we would look upon the cortex as being in constant activity, the physiological activity of the whole network of neurons bearing some direct relationship to the ‘present state’ of the animal’s complex behavio[u]r which is sometimes referred to as his ‘mental state’”.

Indeed, ongoing activity occurs throughout the brain and its existence is manifested in the variability of cortical responses in repeated responses to physically identical conditions or stimuli. In the past, this variability had simply been labelled as noise and scientists got rid of it by averaging over repeated trials (Gerstein, 1960; Zohary et al., 1994). However, during the last two decades an increasing number of neuroscientists recognised that ongoing neural activity is not mere noise, but plays a fundamental role in stimulus-driven processing (Arieli et al., 1996; Tsodyks et al., 1999) and behavioural variability indeed (Hesselmann, Kell, Eger, et al., 2008; Coste et al., 2011; Kleinschmidt et al., 2012).

I investigated the characteristics of the ongoing brain activity¹ focusing on its functional role and its role in contextual changes, where contextual changes can be differences in the experience of individuals (e.g. learning-related changes) or can be related to trial-by-trial variability.

1.2 The study of spontaneous activity

Why study ongoing brain activity? Contrary to the focus on evoked activity in neuroscience, spontaneous neural activity dominates the brain’s energy consumption (Attwell and Laughlin, 2001; Mintun et al., 2001; Attwell and Iadecola, 2002). The energy consumption during rest exceeds task-related increases in neural metabolism, which are usually < 5 % (Raichle and Mintun, 2006). Thus, the majority of neuroscientific studies are focused on a minor component of brain activity. Maybe it is time for an adjustment or alteration in the neurosciences, shifting towards an experimental approach that is indeed focusing on the factor that uses the lion’s share of the brain’s energy, namely ongoing or spontaneous neural activity.

Although ongoing brain activity has been studied using electrophysiological and neuroimaging methods, its physiological origin and cognitive consequences are not yet fully understood. Crucially, any clarification is difficult by its very nature, because any study that addresses the functional significance of spontaneous fluctuations inevitably requires a primary task-context in order to probe perceptual and / or behavioural consequences of the fluctuations (Hesselmann, Kell, and Kleinschmidt, 2008). Attributed roles of ongoing brain activity span processes at different levels of neural activity and range from the traditional view of “intrinsic noise” over low-level physiological processes and uncontrolled mental activity to a monitoring of the environment (Mantini and Vanduffel, 2013). In conclusion, one of the most intriguing questions in the neurosciences might be related to the functional significance of the brain’s “intrinsic noise”.

1.2.1 Electrophysiological research of ongoing activity

The brain is a noisy system whose processing parts - the neurons - receive a large number of fluctuating inputs which in turn generate spike patterns. These often appear very irregular and much of the activity is spontaneous.

1.2.1.1 Cortical states and response variability

Cortical states are determined by the states of individual neurons and the states of individual neurons are in turn related to the state of their neighbours. Possibly the ground-breaking study investigating spontaneous activity and its relation to the large variability of evoked responses to repeated presentations of the same stimulus, is the work by Arieli et al. (1996). They showed that single trial responses in cat visual cortex can be predicted by the linear summation of the deterministic response and the preceding ongoing activity, concluding that ongoing activity plays a noteworthy role in cortical function and for cognitive processes. Given the observation that ongoing fluctuations influence cortical processing, one might wonder how these two types of brain processes interact. In order to answer this question, the relation between the activity of single neurons and the dynamics of the network in which they are embedded has been explored. Using single-unit recordings and real-time imaging, Tsodyks et al. (1999) showed that the firing rate of a spontaneously active single neocortical neuron depends on the instantaneous spatial pattern of ongoing population activity of a larger cortical area. The spatial patterns of population activity recorded during spontaneous firing and those when driven by the optimal input were very similar. Moreover, the correspondence between evoked neural activity and the structure of an input signal seems to mature with age due to a shift in the dynamics of spontaneous activity (Fiser et al., 2004). These results suggest that evoked neural activity in response to sensory stimulation might represent modulated and triggered ongoing neural circuit dynamics, instead of the structure of the input itself. On a more theoretical level, aspects of neuronal responses that are often considered as “noise” might be essential components of the way in which information is propagated or represented in neurons (Ermentrout et al., 2008).

Furthermore, the spatio-temporal structure of spontaneous activity has been shown to be highly coherent over different networks and in multiple species (Chiu and Weliky, 2001). This coherent spontaneous activity of neurons involves a set of dynamically switching cortical states similar to the well-known orientation maps (Kenet et al., 2003).

Taken together, the traditional belief that neural activity is primarily driven by sensory input from the environment might be out-dated. Instead, spontaneous activity of single neurons and their networks seems to play a crucial role for the cortical processing following a sensory stimulus. Correspondingly, this observation might explain the well-established response variability observed after the repeated presentation of physically identical stimuli (Henry et al., 1973; Vogels et al., 1989; Azouz and Gray, 2008). The study of response variability is an important tool used to examine the role of spontaneous activity related to the question of whether it carries predictions about sensory stimuli. Specifically, the investigation of so-called prestimulus activity occurring prior to stimulus presentation has become a standard approach, based on the assumption that predictions - if any - are expressed shortly before stimulus onset.

However, the linear relationship between spontaneous and evoked activity as proposed by some studies (Arieli et al., 1996; Azouz and Gray, 2008) has been challenged using neuroimaging methods (e.g. Hesselmann, Kell, & Kleinschmidt, 2008; Schölvinck et al., 2012) and might indeed differ across the brain. Recently, it has been shown that evoked activity across a large part of the human cortex interacts negatively with ongoing activity (He, 2013). Consequently, a higher prestimulus baseline resulted in less activation - or more deactivation - leading to a decreased trial-to-trial variability of cortical activity after stimulus onset. Thus, measuring across-trial variability might provide a new approach to interpret neuroimaging experiments using trial-based data.

1.2.1.2 Predictive coding and predictive timing

Independent of how spontaneous activity interacts with evoked responses, its existence and functional influence as described above begs the question of why it is there in the first place. One possible explanation for its existence is the concept of predictive coding. The basic idea is that the brain possesses internalised representations of the world which it uses to “predict” what happens in the environment thereby inferring the most likely cause of sensory events (Friston, 2005). In other words, the brain might generate hypotheses about the potential causes of upcoming sensory events and compares these hypotheses with incoming sensory information (Summerfield et al., 2006). The difference between the two - the so- called prediction error - is then propagated forward throughout the cortical hierarchy; internal representations might be adjusted subsequently. Lately, the theory around predictive coding has been extended by the notion of predictive timing (Schroeder and Lakatos, 2009; Arnal and Giraud, 2012), which exploits the temporal regularities or associative contingencies (e.g. the temporal relation between two inputs) to infer the occurrence of forthcoming sensory events. In short, predictive timing is the process by which uncertainty about the temporal occurrence of events is minimised such that their processing and detection are facilitated (Jones et al., 2002; Nobre et al., 2007). During the last decade, both accounts - predictive coding and predictive timing - have been supported by several studies providing evidence for the idea that the hypotheses - or predictions - generated by the brain might be embodied in the spatial and temporal structure of spontaneous activity.

One proposed neurophysiological substrate of the operations described in the framework of predictive coding (i.e. directional message-passing) are cortical oscillations, i.e. those observed as ongoing activity. Both, gamma and beta oscillations (see 2.4 for a description of frequency bands) have been shown to play a role in predictive coding. For instance, gamma activity scales with prediction errors (Arnal et al., 2011) and has been implicated in the evaluation of sensory predictions, possibly depending on the match between bottom-up input and expectations (Herrmann et al., 2004). Beta activity has been associated with error-related effects as well, but in a different direction of processing than gamma, i.e. downstream from prediction error generation (Fujioka et al., 2009; Iversen et al., 2009). Combining these findings, gamma and beta could underlie the information flow in opposite directions, i.e. forward versus backward (for a review see Wang, 2010). Thus, prediction errors might be transmitted in a feed-forward manner using the gamma frequency channel, while predictions and their reconsiderations could be propagated by the beta channel in a backward direction (Chen et al., 2009; Wang, 2010).

With regard to predictive timing, low and mid-frequency oscillations have been identified as important. Their interactions during temporal expectations support a functional cooperation between these oscillations in predictive timing (Saleh et al., 2010). For instance, the anticipation of sensory events resets the phase of slow, delta- theta activity before stimulus presentation and the predictive alignment of these oscillations (in an optimised excitability phase) results in faster stimulus detection (Lakatos et al., 2008). Also, reduced early sensory responses are observed in response to stimuli that are implicitly expected based on their temporal regularity (Costa-Faidella et al., 2011). Furthermore, neural and perceptual responses for temporally unpredictable stimuli are modulated by the phase of alpha oscillations at which stimulation occurs (Busch et al., 2009). However, how the oscillations at the different low frequency bands interact during predictive timing of sensory events remains to be solved.

In summary, there are several lines of electrophysiological research providing evidence which supports the idea that spontaneous activity has certain predictive power for subsequent neural responses to upcoming sensory stimuli.

1.2.2 Neuroimaging research of ongoing activity

The crucial difference between electrophysiological methods and neuroimaging using functional MRI lies in the temporal resolution: while dynamically switching cortical states can be accessed with single cell recordings for example, functional MRI is restricted to a much broader timescale. Nevertheless, the technique has been used to study spontaneous activity for almost twenty years based on the pioneering work of Biswal et al. (1995), who first observed correlations of low frequency fluctuations in the resting brain between different regions involved in a simple motor task. He concluded that these might be a manifestation of functional connectivity in the brain, contrary to the traditional view of the brain as being driven by transient environmental demands. Similar to other neuroscientific methods, functional neuroimaging has spent most of its infancy with studies of evoked responses to sensory, cognitive and motor events (Posner and Raichle, 1994). Therefore, it took several years before the examination of ongoing activity became a major topic among neuroscientists using functional MRI.

1.2.2.1 Resting state fluctuations

The presence of low-frequency fluctuations in the blood oxygen level dependent (BOLD) signal (see 2.3.2) of functional MRI is a well-established finding in neuroimaging. Recent studies have identified these fluctuations as a potentially important manifestation of spontaneous neuronal activity, which is indeed organised in distinct patterns - often referred to as resting state networks (De Luca et al., 2005; Fox and Raichle, 2007; Corbetta, 2012), which will be described below (see 1.2.2.4). Several observations made about resting state fluctuations appear to become known principles. First, their spatial organisation is preserved in humans under anaesthesia (Greicius et al., 2008) and during early stages of sleep (Larson-Prior et al., 2009), as well as in monkeys (Vincent et al., 2007) and rats (Lu et al., 2007). More than that, the fluctuations in ongoing activity show a high consistency and reproducibility across participants and sessions (Damoiseaux et al., 2006; Chen et al., 2008; Meindl et al., 2010; Wang et al., 2010), a high test-retest reliability (Shehzad et al., 2009; Wang et al., 2010) and a high reproducibility across different analytic approaches (Long et al., 2008; Franco et al., 2009). Second, the strength of coherence between different regions exhibiting ongoing activity varies with different parameters among which are age (Fair et al., 2008), experience (Lewis et al., 2009) and disease (Zhang and Raichle, 2010). For instance, it has been argued that disruption in the resting state coherence between different nodes might be a sensitive early biomarker for certain diseases, such as Alzheimer’s (Zhang and Raichle, 2010). Third, the slow global fluctuations that are observed in the BOLD signal at rest can be of the same magnitude as signal changes in response to task-related paradigms (Damoiseaux et al., 2006). Fourth - related to the previous aspect and extending the electrophysiological findings discussed above in 1.2.1 - they contribute significantly to the variability in evoked signals (Fox et al., 2006) and the variability of the associated behavioural response (Fox et al., 2007). Last - but certainly not least - the frequency distribution of the spontaneous fluctuations is significantly different from BOLD fluctuations observed in water phantoms (Zarahn et al., 1997).

Taken together, the aforementioned studies and findings have provided insight into the intrinsic functional architecture of the brain. However, the mechanisms underlying these resting-state fluctuations are still controversial.

1.2.2.2 Vascular basis

A possible explanation for the observed global anatomy of spontaneous activity - given its reproducibility - which is not linked to neural architecture, could be based on vascular mechanisms. All voxels in the brain show a global level of coherence between each other. Thus, the origin of the observed spontaneous fluctuations might lie in changes of blood flow. One possible explanation is linked to so-called draining veins, which exhibit a form of “blood stealing” throughout the tissue whereby active regions generate blood flow increases at the expense of other nearby regions. Alternatively, further poorly understood mechanisms of vascular regulation could play a role (Buckner et al., 2008). Indeed, changes in respiration and intracranial pressure are known to influence blood flow and blood oxygenation in the brain; the BOLD signal measured during functional MRI is based on hemodynamic measures of blood flow and only indirectly linked to neural activity (see 2.3.2). Variations in heart rate and respiration are known to contribute to fMRI resting-state fluctuations (Wise et al., 2004; Birn et al., 2006). The highest power of spontaneous fluctuations is below 0.1 Hz (Cordes et al., 2001), which is beneath normal breathing frequency. Nevertheless, the aforementioned variations in respiration rate and depth are observed at these low frequencies (Birn et al., 2006) and correlate with the BOLD signal acquired during rest. In particular, these correlations overlap with the areas characterised as the default mode network (see 1.2.2.4 for a detailed description of this particular resting state network) (Birn et al., 2006).

Finally, with regard to the potential role of vascular coupling for spontaneous fluctuations, two aspects are worth mentioning. First, resting state networks can also be identified using measures of resting glucose metabolism, i.e. without relying on vascular coupling at all (Vogt et al., 2006). Second, confounds in resting-state data due to cardiac and respiratory effects are now addressed by sophisticated methods (Glover et al., 2000; Birn et al., 2008), which will be described in 2.3.4.

1.2.2.3 Neural basis

Knowing that part of the spontaneous fluctuations depend on non-neuronal physiological factors, it needs to be clarified what the neural basis of the signals is. Indeed, an effort has been made to determine the electrical correlates of the fMRI BOLD signal (see Khader, Schicke, Röder, & Rösler, 2008; Logothetis, 2008 for summaries of this work involving different perspectives). The conclusion drawn from this work is that the BOLD signal is best correlated with local field potentials (LFPs). Probably the most cited study in this line of research observed that the power of LFPs recorded in monkey visual cortex fluctuates at a similar timescale to the one of spontaneous fluctuations measured by fMRI, i.e. the previously mentioned < 0.1 Hz (Leopold et al., 2003). Similar to the long-distance coherences observed in resting-state activity, the fluctuations observed by Leopold et al. (2003) show a high coherence across electrode pairs without any changes due to cortical distance. Comparable fluctuations have been observed using intracranial electrocorticography (ECoG, i.e. surface electrodes) recordings in humans (He et al., 2008; Nir et al., 2008). Importantly, recordings of neuronal electrical activity either from the scalp with electroencephalography (EEG) or from the surface of the brain with ECoG always encompass a summation of a population of LFPs. One possibility to analyse fMRI resting-state fluctuations and its underlying neural activity is the simultaneous recording of EEG and fMRI (Goldman et al., 2002; Laufs et al., 2003; Mantini et al., 2007). Conventionally, LFPs are described according to their frequency components and two of these components have been associated with spontaneous BOLD fluctuations: First, fluctuations in the power of higher frequencies (Leopold et al., 2003), which are associated with cognition (Donner et al., 2009; Uhlhaas et al., 2010), and second, fluctuations that approximate those present in the spontaneous BOLD signal, commonly summarised as slow cortical potentials (SCPs) (Rockstroh et al., 1989; He et al., 2008; Raichle, 2011). Increasing evidence supports the idea that spontaneous BOLD fluctuations are most correlated with LFP activity in the SCP range (Lu et al., 2007; He et al., 2008; He and Raichle, 2009), while the spatial patterns of coherence are maintained across different levels of consciousness ranging from wakefulness to rapid eye movement (REM) sleep and slow wave sleep (He et al., 2008). For instance, alpha power (see 2.4 for a description of frequency bands) is negatively correlated with spontaneous BOLD fluctuations in occipital cortex, as well as in inferior frontal and superior parietal regions (Goldman et al., 2002; Laufs et al., 2003; Moosmann et al., 2003). The latter regions are commonly active during rest and are known to be involved in the default mode network (see 1.2.2.4 for a detailed description of this network). The less persistent correlation of power in the higher frequencies (i.e. the spatial correlation with the BOLD signal is only observed during wakefulness and REM sleep) is in line with the known role of these frequencies for consciousness awareness (Fries, 2009).

One way to analyse simultaneously acquired EEG and fMRI data is to use so-called independent component analysis (ICA). It is used for BOLD data in order to identify different networks and to correlate these subsequently with different frequency bands of the EEG data. In doing so, Mantini et al. (2007) found multiple resting-state networks in the BOLD data, which correlated with different frequency bands of the EEG data.

Finally, Shmuel and Leopold (2008) used intracortical neurophysiological recordings in combination with fMRI to investigate the relationship between spontaneous fMRI and LFP signals as directly as possible. They found strong correlations between the spiking rates (of the neurons close to the recording electrode) and slow fluctuations in the fMRI signal, as well as with slow power changes in the multi-unit activity (> 100 Hz) and LFP band of higher frequencies (25 - 80 Hz).

In summary, a large number of studies using different methodological approaches including simultaneous measurements of fMRI and neuronal activity have established the presence of a coupling between slow fluctuations in the BOLD signal measured with fMRI and underlying fluctuations in the neural activity across multiple regions, frequencies and states of consciousness. Thus, modulations in both signals possibly share the same origin - which might be subcortical in nature. Future studies are required to determine the origin the slow fluctuations in spontaneous activity.

1.2.2.4 Functional networks

Independent of the underlying mechanisms of spontaneous brain activity, the organisation of the commonly observed low-frequency fluctuations of spontaneous activity organised into specific and distinct patterns is a robust finding. A number of so-called resting state networks has been discovered since the pioneering study of Biswal et al. (1995). Thus, the large-scale sensory-motor network reported back then was only the first one of multiple networks observed during rest and exhibiting a high similarity to task-activated networks. Comparable relationships have been found for other modalities, such as visual, auditory and language processing networks (Lowe et al., 1998; Hampson et al., 2002; van de Ven et al., 2004). For instance, regions in intraparietal sulcus, frontal eye field, and middle temporal cortex that are commonly recruited during visuospatial attention or oculomotor tasks also form a functional network at rest different from the classical “visual” network including V1 to V4. Probably the most famous of the these networks is the so-called default mode network (DMN), which had been first described by Raichle et al. (2001) and which is most consistently found across experiments (Anticevic et al., 2012; Mantini and Vanduffel, 2013). Raichle’s et al. (2001) idea of a baseline - or default state of the brain - stems from a common finding in positron emission tomography (PET) and functional MRI studies, i.e. the observation that a certain set of brain regions shows decreases in activity independent of a particular task and with little variation in their location across a wide range of tasks (Shulman et al., 1997). Shulman et al. (1997) provided the first formal characterisation of task-induced activity decreases. Based on their finding, the idea of a default state of brain function distinct from the sates involved in sensory and executive functions (i.e. in attention-demanding, goal- directed behaviours) has been established (Buckner et al., 2008). By use of a generally accepted quantitative circulatory and metabolic definition of brain activation, the authors established criteria for a baseline state characterised by the absence of activation. In the meantime, the default mode network has been studied extensively, see Buckner et al. (2008) for a review. Anatomically, it is considered to encompass certain key regions among which are posterior cingulate cortex (PCC), left and right inferior parietal cortex, and ventromedial prefrontal cortex (vmPFC). In addition, lateral temporal cortex, dorsal medial prefrontal cortex (dmPFC) and the hippocampal formation have been associated with the default mode network (Buckner et al., 2008).

In general, resting state patterns of coherence characterised as “resting state networks” take patterns of anatomical connectivity in the human (Zhang et al., 2008) and the monkey brain (Vincent et al., 2007) into account, but are not restricted to these anatomical connections. Therefore, the absence of monosynaptic connections between certain brain regions does not exclude the existence of functional connectivity as for instance expressed in the networks of resting state coherence (Raichle, 2011).

1.3 The functional role of spontaneous activity

Having discussed the existence and potential neurophysiological basis of spontaneous fluctuations and their networks observed during rest, I will now focus on their functional role. As mentioned earlier, traditionally, ongoing activity has been often considered as “intrinsic noise” that can be averaged over trials in order to get to the essence of the “real signal”. Even in view of the structural regularities, i.e. the organisation into networks, it could still be possible that intrinsic functional connectivity merely reflects some “noise” that happens to have a non-random structural connectivity carrying the described form of spatial patterns. In support of this notion, computational and empirical studies have shown a correspondence between intrinsic functional and anatomical connectivity (Sporns et al., 2000; Greicius et al., 2009). However, at the level of the entire brain and involving a systematic quantitative analysis, the match is not impeccable. Honey et al. (2009) found that structural connectivity could predict the empirically observed functional connectivity only to a limited extend. Leaving potential limitations during data collection and analysis - which could cause the deficiencies of predicting functional from structural connectivity - aside, an alternative hypothesis has been put forward: If structural connectivity merely forms the backbone of functional connectivity, it has to be influenced and shaped by additional context-dependant modulations (Sadaghiani, Hesselmann, et al., 2010).

Although this hypothesis might be contradictory given the high consistency of spatial patterns of ongoing activity across different levels of consciousness and context as mentioned in 1.2.2.3, several lines of research have accumulated evidence supporting the hypothesis that ongoing brain activity is indeed context sensitive. Thus, even though functional connectivity patterns are persevered qualitatively in a range of states, including light and deep sleep (Horovitz et al., 2008, 2009; Nir et al., 2008), as well as severe disorders of consciousness such as vegetative state (Boly et al., 2009), they do express quantitative changes. For example, functional connectivity in the DMN decreases with the degree of consciousness in healthy individuals (i.e. between wakefulness, deep sleep and a state of physiological unconsciousness (Horovitz et al., 2009), as well as in patients (i.e. between minimally conscious state, vegetative state and coma) (Vanhaudenhuyse et al., 2010). As a last counter- argument to the idea of ongoing activity as mere “noise”, I refer to the finding that the reduction in connectivity between regions of the DMN, i.e. frontal and posterior areas, during sleep is anatomically selective. Therefore, it is very unlikely that intrinsic connectivity only reflects a change in the level of “noise” that is propagated through an anatomically connected network (Sadaghiani, Scheeringa, et al., 2010).

In the following, I focus on the different lines of research supporting the idea that ongoing activity is context-sensitive and also can influence behaviour.

1.3.1 Perceptual domain

The perceptual impact of prestimulus activity fluctuations has been investigated in different perceptual paradigms. For example, Boly et al. (2007) used a somatosensory detection task and found that prestimulus activity in a large distributed network determined whether stimuli close to perceptual threshold were detected or not on a single trial basis. Regions involved in the predictive system included the thalamus, dorsal anterior cingulate cortex (dACC), parieto-frontal areas and inferior frontal gyrus. These are all regions that are commonly active in response to a wide range of (cognitive) tasks (Smith et al., 2009a; Corbetta, 2012). In contrast, prestimulus activity in regions that have been coined as “task-negative” belonging to the DMN, i.e. posterior cingulate (PCC), parahippocampal and lateral parietal components, was higher for trials that were missed by participants. These results support the concept of a simple dichotomy between “task-positive” and “task- negative” networks whose activity either facilitate or deteriorate perception. However, more recent findings revealed a more complicated picture highlighting that context plays an important role predicting whether the ongoing activity in a certain brain area plays a role for subsequent perception of a stimulus - or not. Sadaghiani et al. (2009) used a free-response, auditory detection task and examined whether prestimulus activity was predictive of when participants perceived auditory stimuli at individual detection threshold. A rather complex picture was observed: comparing detections and misses, the former ones were preceded by higher activity in early auditory cortex as well as in a network including thalamus, anterior insula and dACC; misses were preceded by higher activity in the so-called dorsal attention system, including frontal and parietal areas. Thus, two task-positive networks showed opposite effects. In addition, regions of the DMN were more active prior to hits, not misses. The time courses of prestimulus effects in the different networks were distinct, underlining the idea that the two observations were probably not mere mirrored effects. These results highlight the complexity of ongoing activity - which is not only organised in spatially defined networks. Mora than that, these networks are independently organised and context sensitive. In other words, the context - including specific sensory features and cognitive faculties - of a perceptual decision impacts to what extent ongoing activity in different networks determines stimulus processing and eventually perception.

In line with this observation, several researchers have speculated that a perceptual task involving choices - compared to all-or-none detection tasks - might affect prestimulus activity in a more localised way, i.e. restricted to a certain area that is task-relevant compared to more general brain networks related to attention or memory. Two linked fMRI studies have provided evidence for this idea. First, increased prestimulus activity in fusiform face area (FFA), a region that has been previously related to the processing of faces, was observed when perceiving a face compared to a vase in a paradigm using Rubin’s ambiguous vase-face figure (Rubin, 1915; Hesselmann, Kell, Eger, et al., 2008). Second, prestimulus activity in human MT+, an area crucial for motion processing located in the middle temporal cortex, was higher in trials when participants detected coherence in a random dot stimulus paradigm compared to trials when they did not (Hesselmann, Kell, and Kleinschmidt, 2008). The difference was found comparing trials that were physically identical, i.e. at a coherence level that resembled the individual detection threshold of coherence (see Chapter 5 for a more detailed description of these findings that were used as the basis of the study described there).

In addition to the studies using fMRI, several EEG and MEG studies have provided additional support for the functional role of ongoing brain activity for perception. Suffering from poorer spatial resolution, these studies provide the benefit of identifying specific oscillations bands. For instance, using MEG, visual discriminability has been shown to decrease with an increase of certain low oscillations, i.e. in the so-called alpha band (see 2.4 for a more detailed description of frequency bands), in occipital-parietal cortex (van Dijk et al., 2008). Comparable to the findings in the fMRI literature, a free-response task revealed a rather complicated picture of different frequency bands being important for perceptual outcome. Using EEG and a somatosensory threshold detection task, Linkenkaer-Hansen et al. (2004) found that prestimulus fluctuations at medium power levels of several frequency bands over somatosensory cortex resulted in highest detection rates and shortest reaction times. In contrast, over parietal electrodes the best behavioural performance was associated with the highest power in the same frequency bands. In addition to power analyses of oscillations, the phase of slow oscillations has been shown to have certain predictive power for perceptual outcome, e.g. in visual threshold detection tasks (Thut et al., 2006; Busch et al., 2009; Mathewson et al., 2009). Thus, the cortex might go through different states of excitability - so-called microstates - which can differ in speed, depending on the oscillation whose phase is important (Monto et al., 2008; Busch et al., 2009; Mathewson et al., 2009).

The predictive nature of prestimulus activity with regard to perception has also been shown looking at electrode recordings in monkeys (Supèr et al., 2003). Recording from primary visual cortex, it has been observed that reported stimuli were preceded by higher and more correlated neural activity compared to not-reported ones. The authors concluded that the strength of neural activity and the functional connectivity between different neurons in primary sensory areas contributes to perceptual processing. More precisely, visual cortex needs to be in a suitable state to result in subsequent stimulus detection.

In conclusion, numerous lines of research using different techniques and methodological approaches have presented supporting evidence to the idea that variability in perceptual performance can be - partly - explained by the variability in intrinsic - ongoing - processes in the brain; different signals measuring brain activity directly or indirectly can be used to forecast perception in the human and primate brain.

1.3.2 Motor domain

The functional role of ongoing activity has also been investigated with regard to motor activity and behaviour. In a series of two studies, Fox et al. (2007, 2006) showed first that trial-to-trial variability of finger movement-related activity in motor cortex can be largely attributed to fluctuations in ongoing activity (Fox et al., 2006), and second, that variability in behaviour depends on spontaneous activity as well (Fox et al., 2007). In order to do so, they used a simple button press task with the right index finger and a trick in order to disentangle evoked from ongoing activity. They exploited the fact that right and left somatomotor cortex exhibit correlated spontaneous activity (Biswal et al., 1995; Cordes et al., 2000) and that button presses with one hand do not result in evoked responses in ipsilateral motor cortex. More precisely, they used activity in the right motor cortex as a proxy of spontaneous activity in the left motor cortex (activated by the right hand button presses). In doing so, they showed that 40 % of the trial-to-trial variability in the BOLD response in left motor cortex can be ascribed to spontaneous activity. Based on the additional observation that the trial-by-trial evoked activity did not depend on whether the spontaneous activity in a given trial was high or low, they concluded that both types of activity are superimposed in a linear way. The second study made use of the same paradigm, but distinguished button presses according to their strength, i.e. trials were rated as either soft or hard button presses. Subtracting right motor cortex estimates of spontaneous activity from the activity measured in left motor cortex eliminated the apparent difference between responses to soft and hard button presses. In doing so, the study showed that 74 % of the relationship between spontaneous force variability and BOLD activity in left motor cortex can be explained by spontaneous activity fluctuations. In sum, spontaneous activity influences trial-to-trial variability on the neuronal and behaviour level in response to a simple motor task.

On a different level, Ramot et al. (2011) confirmed a link between resting state activity and spontaneously emerging subconscious oculomotor behaviour. The behaviour they looked at, is a type of eye movement that occurs spontaneously and subliminally whenever we close our eyes (Allik et al., 1981; Collewijn et al., 1985). However, the neuronal mechanisms and functionality of these spontaneous eye movements are largely unknown. The findings that spontaneous fluctuations in the BOLD signal were correlated to the amplitude and velocity of these eye movements and that the neuronal activity was linked to coordinated motor programs (involving oculomotor neurons and muscles), provide further evidence for the idea that neuronal activity related to movement and associated behaviour is influenced by spontaneous activity in the brain.

1.3.3 Cognitive domain

The distinction between perceptual, motor and cognitive tasks is not straight- forward, because usually experimental tasks involve all three domains to a certain and varying degree. Often, the cognitive level is considered as the “highest” one, involving specific brain regions, so-called higher cognitive brain areas compared to primary sensory and motor areas. Given the difficulty of separating domains, I will introduce this paragraph with some studies bridging the gap between perception, movement and cognition using inhibitory control. Inhibitory control refers to the ability to suppress planned or ongoing processes, which might be related to movements or cognition. Using a monotonous task and MEG, Mazaheri et al. (2009)

showed that false alarms were preceded by higher alpha band power in visual areas and bilateral somatosensory cortices compared to correct withholds. An EEG study that looked at single-trial EEG topographic analyses to avoid averaging out effects that might get lost in the more typical ERP analyses (see 2.4.1), found supporting evidence for the idea that fluctuations in the ongoing activity immediately preceding stimulus presentation contribute to behavioural outcomes in an inhibitory control task (Chavan et al., 2013). They used an auditory spatial go no-go task and observed that a specific configuration of the EEG voltage field manifested itself before correct rejections, but not before false alarms. Using source estimation on the EEG topography, a fronto-parietal network was identified. These results support the idea that prestimulus brain activity also influences behavioural outcomes in an inhibitory control task. Furthermore, the identification and involvement of the fronto-parietal network suggests that the state-dependency of sensory-cognitive processing includes high-order, top-down inhibitory control mechanisms.

Until now, there are only a few “purely” cognitive control studies investigating the relation between ongoing brain activity fluctuations and inter-trial variability in evoked responses. However, the same picture emerges, confirming the crucial role of ongoing activity for evoked neural responses as well as behavioural outcome. For instance, Coste et al. (2011) used the Stroop task with colour-word interference and fMRI to show that prestimulus activity in several task-relevant brain regions (including dorsal anterior cingulate, dorsolateral prefrontal cortex and ventral visual areas) predicted subsequent response speed. Furthermore, this effect scaled with the Stroop effect size, i.e. it was only significant in participants who exhibited behavioural interference.

Another approach in the realm of the study of ongoing activity is focused on the relation between clinical phenomenon and changed patterns in resting state connectivity. For instance, a recent review identified 16 fMRI studies that investigated the use of resting-state fMRI in major depression (MD) and concludes that this research has yielded insight into the pathophysiology of depressive symptoms. Foremost, the role of the cortico-limbic mood regulating circuit and the interaction between task-positive and -negative networks in MD are emphasised (L et al., 2012).

Yet another methodological approach looks at the functional role of intrinsic connectivity on cognition by examining specific patterns and their capability to change in response to certain cognitive experiences. As soon as after one scanning session using fMRI, i.e. a time span that most likely does not include any gross anatomical changes, intrinsic functional connectivity has been reported to be sensitive to visuo-motor learning (Albert, Robertson, and Miall, 2009), episodic memory (Tambini et al., 2010), as well as language tasks (Waites et al., 2005; Hasson et al., 2009). To conclude, the functional context of a task seems to interact with the appearance of intrinsic activity and motivates further experimental investigation of the functional significance of ongoing activity and associated changes of it.

However, the aforementioned studies are commonly criticised for the possibility that they might confuse “true intrinsic” activity with echoed traces of the previous experiences, i.e. opening the possibility that the observed changes are some kind of memory trace. Naturally, this leads to the question what “true intrinsic” activity might be - or whether it exists at all. More precisely, this criticism can be applied to any measurement of ongoing activity: whenever resting state activity is measured - be it during wakefulness, during sleep or during anaesthesia - it will always include the so-called task-unrelated mind-wandering (Mason et al., 2007; see Chapter 6 for a more detailed description of mind-wandering) that occurs either with or without awareness of the individual (Smallwood, McSpadden, et al., 2007). Therefore, the very nature of ongoing activity cannot be considered to lack context (Sadaghiani, Hesselmann, et al., 2010). Accordingly, the only possibility to isolate “pure” ongoing activity would require that it possesses a unique spatial and temporal form. As outlined in 1.2.2.4 this does not appear to be physiologically plausible, i.e. no such qualities have been recognised with confidence. More than that, this debate and the question of whether the disputed dissociation of ongoing activity from “other” brain activity is indeed justified or reasonable, leads to the ultimate question about the function of ongoing activity.

Given the observation that ongoing activity does not simply represent unconstrained, spontaneous cognition - either called daydreaming, mind-wandering or stimulus- independent thought (Antrobus, 1968; McGuire et al., 1996; Mason et al., 2007) - it seems to reflect a more fundamental or intrinsic property of the brain’s functional organisation. In particular, the observation that spatially coherent, spontaneous BOLD activity is present under anaesthesia and during sleep renders it unlikely that the observed patterns of coherence organised in functional networks are solely the result of unconstrained, conscious cognition or mental activity (Christoff et al., 2009; Raichle, 2011). Recently, it has been shown that patterns of low-frequency oscillations in the BOLD signal are even modulated by the content - or nature - of free thought during rest (Doucet et al., 2012).

Given the research results described in this chapter, one possible function of ongoing activity is the facilitation of responses to external stimuli. Thus, a balance on the global level - similar to the balance of excitatory and inhibitory inputs at the single neuron level that determine the responsiveness of neurons - might be present on a more global level of brain function: opposing forces could enhance the precision of a wide range of processes (Raichle and Snyder, 2007). Indeed, some of these more global effects that involve balance have been reported. For example, the so-called Sprague effect has been first demonstrated in the visual system of the cat (Sprague, 1966).

A more progressive and more expanded view on the functional role of ongoing activity is the notion of predictive coding (see 1.2.1.2) in the context of the experimental investigation of spontaneous fluctuations in the brain. Combining the mentioned electrophysiological findings and the neuroimaging results outlined thereafter, the proposal of the brain a Bayesian interference machine that generates predictions about the future (Olshausen, 2003; Kersten et al., 2004) appears plausible. Most simplistically, the suggestion entails that the brain is shaped by experiences (i.e. stimuli) to represent a best guess, i.e. prior, about states of the environment and - on a cognitive level which holds for humans and some other species - to make predictions about future states of the environment.²

In sum, I propose that the function of ongoing activity is closely related to cognition and that this relation is present during “rest” and during “active states”, i.e. it is characteristic for the brain.

1.4 Conclusions

Although the majority of neuroscientists still focus much of their research activities on evoked responses, a growing community of researchers is investigating or taking into account the role of spontaneous or ongoing brain activity. Across all neuroscientific disciplines ranging from single cell recordings in the cat to clinical studies in humans, the importance of spontaneous neural activity is now appreciated and the results add to a rapidly expanding body of research aimed to understand how the brain instantiates behaviour³. Raichle and Snyder (2007) even mention the “requirement to establish a framework upon which the study of intrinsic brain activity is incorporated into the work devoted to evoked activity”.

Methodologically, the study of slow fluctuations (i.e. < 1 Hz) in neural membrane polarisation has been shown to be of particular interest. In particular, these frequencies correspond to the ones of spontaneous fluctuations in the BOLD signal and their functional consequences seem to be relevant for the understanding of the well-known variability in task-evoked activity, as well as behavioural performance variability.

1.5 This thesis

My interest within the study of the functional role of spontaneous activity is focused on its role in contextual changes ranging from recent experiences of individuals to trial-by-trial variability in a certain task. Therefore, I studied the nature of ongoing activity from two perspectives: One looking at changes in the ongoing activity due to learning, and the other one looking at the predictive role of prestimulus activity using complementary methodologies, i.e. EEG and fMRI. Finally, I ventured into the realm of inter-individual differences and mind-wandering to investigate the relationship between ongoing activity, certain behavioural traits and neuronal connectivity.

Chapter 2 Methods - measuring spontaneous activity

2.1 Introduction

As mentioned in Chapter 1, several different methodological approaches and techniques can be used to acquire and analyse ongoing brain activity and some of these have been employed in this thesis. This chapter describes the central issues in recording and analysing the data presented, including fMRI and EEG. Other techniques that have not been used here, like electrophysiological recordings in non- human species, are not described to avoid confusion. First, the difference between group studies, lesion studies and inter-individual difference studies is given. Second, a concise overview of the two techniques, i.e. fMRI and EEG, is given and more specific issues related to measuring spontaneous activity are described. With regard to fMRI, this includes a description of the use of stochastic dynamic causal modelling for resting state data.

2.2 Group versus inter-individual differences versus lesion studies

The most common approach to investigate a neuroscientific question in humans using either EEG or fMRI is to perform a group study. In case of the simplest design this entails that healthy volunteers who participate in the study are either randomly assigned to a group, i.e. in case of a between subject design, or do perform two or more experimental tasks or conditions, i.e. in case of a within subject design. For instance, a group study testing for the benefits of a certain drug might randomly assign half of the sample to a group that first takes the drug and then a placebo, and the other half to the reverse order. The gold standard is the so-called double blind design, where both participant and the experimenter (who interacts with the participant) are “blind” to the experimental condition, i.e. drug versus placebo. A combined between- and within-subject design is also possible. This allows testing for an effect of different dependant variables between and within participants.

One particular form of a group study involves the study of so-called experts, i.e. one group is then comprised of individuals that can perform the studied task(s) particularly well (due to genetic advantages and/or due to intense training).

In comparison to group studies in healthy individuals, clinical studies investigate a sample of patients that show a certain pattern of disease or malfunctioning. Usually, these are then matched, e.g. for variables like age, gender, education and lifestyle, with healthy control participants.

A complementary experimental approach is the study of inter-individual differences. In contrast to group studies, which are based on the assumption that the taken sample is representative of the underlying population and which therefore treat differences between individuals as a source of “noise” that needs to be averaged out, these differences are the main interest for this approach. In the field of differential psychology this is a - or the - standard approach to study topics like personality and different types of intelligence. However, only more recently this approach has gained popularity in neuroimaging studies, often linking brain structure to behaviour or personality traits (Kanai and Rees, 2011). More precisely, inter-individual variability from basic to higher cognitive functions including perception, motor control, memory, aspects of consciousness and the ability to introspect can be predicted from structural MRI studies, using voxel-based morphometry (VBM) (Irle et al., 2010;

Kanai et al., 2010; Schwarzkopf et al., 2011) and diffusion tensor imaging (DTI) (Forstmann et al., 2010), as well as from neural activity measured with fMRI (Wig et al., 2008; Bishop, 2009), EEG (Klimesch, 1999) or positron emission tomography (PET) (Gerretsen et al., 2010). Therefore, it has been proposed that these differences can be used to link human behaviour and cognition to brain anatomy and function (e.g. Kanai and Rees, 2011).

Probably the most debated aspect with regard to group and inter-individual difference studies is the group - or sample size respectively - that is needed for a study, i.e. required to show a certain effect. Even though this question is fundamental for any experiment carried out - not only in neuroscience - there are not many publications about the topic (Lenth, 2001). The so-called effect size is a measure of the strength of the deviation from the null hypothesis and usually refers to the estimate of an unknown true effect size based on the collected data (Friston, 2012). One common way to classify effect sizes is to rank them as small, medium and large (Cohen, 1988). Friston (2012) extended this classification to include trivial effect sizes. Essentially, these refer to statistically highly significant effects that are however grounded in “an uninformed overpowered hypothesis test”. Based on an analysis of effect sizes in classical inference - which is most often used to report results in neuroscience - Friston (2012) suggests that the optimal size for a sample is between 16 and 32 participants. This argues against the more recent trend of group studies in functional neuroimaging to increase sample sizes due to both editorial requirements and large cohort studies (e.g. Lohrenz, McCabe, Camerer, & Montague, 2007).

Last, there is the option of a case study which is commonly used in psychology and clinical medicine. Sigmund Freud conducted some of the most famous and detailed ones at the beginning of the 20th century, including Little Hans (Freud, 1909a) and The Rat Man (Freud, 1909b). In human neuroscience, case studies are mostly lesion studies, i.e. an individual shows a certain and very specific brain anomaly that cannot be easily compared to a healthy brain. Probably the most famous case is HM and the study of human memory. HM had both of his medial temporal lobes removed and subsequently suffered from intense amnesia (Penfield and Milner, 1958). The benefits, e.g. potential causal inferences can be drawn, and shortcomings, e.g. experimental control, of the lesion study approach have been outlined elsewhere (e.g. Kosslyn & Intriligator, 1992).

2.3 Functional Magnetic Resonance Imaging (fMRI)

2.3.1 Overview

Magnetic resonance imaging (MRI) is a non-invasive method used to create detailed images of the body including the brain by using a strong magnetic field and radio frequency pulses. Instead of creating images of organs and tissues, functional MRI measures blood flow in the brain to detect areas that are active. The technique detects changes in blood oxygenation and flow, which are a consequence of neural activity.

To understand functional MRI, one needs to know how MRI works and how it uses the magnetic properties of tissue: Everywhere in the brain are hydrogen atoms acting as small magnets and rotating around their own axis.

[...]

---

¹ In the literature, different terms have been used to describe ongoing neural activity - as compared to evoked responses - among which are “resting state activity”, “endogenous activity”, “spontaneous activity”, and “autonomous activity”. I use the term “ongoing activity” and “spontaneous fluctuations” interchangeably and refer to activity not evoked by an external stimulus or task.

² The question about the initial set of priors equipped with at birth - or even before that - is another interesting one, but shall not be discussed here.

³ Here, “behaviour” refers to any mental expression, i.e. a thought is also considered to be a certain type of behaviour.