On the relationship between amplitude asymmetry in the alpha
band and evoked responses: an overview.
Tara van Viegen1,2
1University of Birmingham, the United Kingdom 2University of Amsterdam, the Netherlands
Abstract
Neurons are the primary processing units of the brain. Synchronized current flow across the dendrites of millions of pyramidal neurons, result in an electrical potential, which can be picked up with an electroencephalogram (EEG) or magnetoencephalogram (MEG). One way in which EEG has been used to link brain activity to cognition has been the event-related averaging approach. This approach entails epoching many segments of the EEG activity around the onset of an experimental event followed by averaging of those epochs. The assumption here is that the averaging cancels out activity not precisely phase locked to the experimental event, which ultimately yields the event related potential (ERP): the brain’s transient phase-locked response to the event. The temporal precision of the ERP makes it a powerful and robust tool to study cognitive processes in the brain. ERPs yielded insights into the mechanisms underlying important cognitive processes like error processing, working memory and sentence comprehension, yet despite the utility of the event-related averaging approach as a tool to study cognition, the mechanistic origin of the ERP still remains controversial. In addition to the ERP, the ongoing EEG signal contains oscillations, generated by the synchronization of large neural ensembles. These oscillations were thought to be noise in the earlier years of neuroscience and averaging in the time-domain (to obtain the ERP), would cancel these non-phase-locked components out. Over the last two decades evidence has been heaping up on the functional importance of this oscillatory activity in the EEG and many frequency bands have been shown to be involved in cognitive processes. To explain the relationship between ERPs and ongoing oscillatory activity three historically influential models have been put forward. The additive model suggests an added signal to the background oscillatory activity, whereas the phase-resetting model suggests a (partial) phase-reset of the background oscillatory activity. Finally, the amplitude asymmetry model suggests asymmetric modulation of the background oscillatory activity. The focus of this review will be to critically review the literature supporting these different models and to specifically examine the relationship between cognitive ERPs, e.g. the P300 and the contralateral delayed activation (CDA), and oscillatory asymmetry in the alpha band. Alpha oscillations are the most prominent oscillations in the EEG and interestingly these oscillations can show an event-related desynchronization (ERD) or synchronization (ERS), whereas the other frequency bands mostly show ERSs. Elucidating the relationship between ongoing oscillations and the evoked response is important, because ERPs have a long history in neuroscience and conceptually they are easier to understand than oscillations, but their exact underlying neural mechanisms remain largely unknown. Broadening our understanding of the relationship between oscillatory activity and evoked responses may help elucidate the neural mechanism underlying ERPs.
Introduction
Neurons are the primary processing units of the brain. Information flows in and out of neurons via the passage of current through dendrites (input) and axons (output), respectively. Synchronized current flow across the dendrites of millions of pyramidal neurons, result in (1) an electrical potential that can be picked up by electroencephalogram (EEG) electrodes, and (2) magnetic fields that can be sensed with magnetoencephalogram (MEG) superconducting quantum interference devices (SQUIDS). Continuous M/EEG activity is epoched around the onset of an experimental event, e.g. stimulus presentation, and averaging many epochs yields the event related potential (ERP; event related field or ERF for MEG), which reflects the brain’s transient phase-locked response to that event. Although the spatial resolution (i.e. where the source of the activity is in the brain) of ERPs compared to other neuroimaging techniques such as functional magnetic resonance imaging (fMRI) is inferior, there are important characteristics of the ERP, which make it a suited tool to examine cognitive processes. Firstly, the temporal resolution of the ERP is on the scale of milliseconds. Importantly, many cognitive processes, like perception and attention, function on this timescale. Secondly and relatedly, EEG reflects the post-synaptic potential changes of neurons, so unlike hemodynamic imaging techniques, it’s a direct real-time measure of
brain activity (Woodman, 2010). Therefore, ERPs provide a direct link between underlying neuronal activity and behaviour.
Despite the numerous advantages of ERPs to study neural activity, the mechanistic origin of the ERP remains controversial. The objective of this review is to examine the relationship between ongoing oscillations and cognitive ERPs like the P300, which seems to reflect the allocation of attentional resources during information processing, the contralateral delayed activity (CDA), a slow negative evoked response which was found to correlate with individual differences in working memory capacity and other sustained ERPs like the contingent negative variation (CNV), which seems to play a role in time perception and consciousness.
Historical overview
In the early years of electrophysiology researchers referred to the averaged recordings as the evoked potential (EP), because it was believed that the signal captured the brain’s response to the presentation of the stimulus and that the rest of the signal was background noise (Rugg & Coles, 1995). Typically, the amplitude of the background EEG varies between -100 and +100 µV and changes in the amplitude of the signal upon stimulus presentation constitute the ERP. The ERP however is at least one order of magnitude smaller than the amplitude of the ongoing EEG signal and typically ranges from 0.5 to 10 µV. When averaging in the time-domain to obtain the ERP, this order of magnitude difference is relatively ignored, because averaging non-phase-locked signals will lead to the signals cancelling each other out (see Table 1).
As early as in the 1970’s, research suggested that sensory stimuli could disrupt the phase of ongoing oscillatory activity, leading to a so-called phase reset (Sayers et al., 1974). The ongoing background activity of the EEG contains oscillatory activity generated by the synchronization of large neuronal ensembles (Pfurtscheller & Da Silva, 1999). These oscillations were divided in distinct frequency bands (see below), which change their power according to experimental events. Increased states of synchrony are addressed as event-related synchronizations (ERS) or power (amplitude squared) increases, whereas decreased states of synchrony are referred to as event-related desynchronizations (ERD) or power decreases. Different types of frequency analyses are available to study ERSs and ERDs, which can be divided in classical frequency bands, like delta (0.5 - 4 Hz), theta (4 - 8 Hz), alpha (8 - 13 Hz), beta (15 - 25 Hz) and gamma (>30 Hz). ERSs and ERDs in these frequency bands have been found to correlate with cognitive brain processes. The theta band, for example, has been related to cognitive control and navigational memory (Cavanagh & Frank, 2014; Kahana et al., 1999), alpha band decreases have been related to attention (Snyder & Foxe, 2010) and the gamma band with object representation (Tallon-Baudry & Bertrand, 1999). Extracting time-frequency representations with e.g. Fourier transformations and wavelet decomposition showed that, in general, lower frequencies span larger neuronal networks than higher frequencies (Pfurtscheller & Da Silva, 1999).
The induced response was hypothesized to be related to the psychological demands of the situation, and it was this realization that led to the more neutral term ‘event-related’ instead of evoked response (Rugg & Coles, 1995). The induced response is time-locked to an event but not necessarily phase-locked, and a linear method like averaging will therefore not detect any changes in the signal, but those changes may be detected with frequency analyses (Pfurtscheller & Da Silva, 1999). Now, for almost two decades the information in the ongoing signal is combined with information obtained from the ERPs, in so called event-related brain dynamics (Penny et al., 2002).
Table 1. Changes in M/EEG after an event
Evoked Phase-locked Time domain averaging
Oscillatory (induced)
Not phase-locked Averaging the event-related spectra, typically studied in the frequency domain
Thus, the onset of an experimental event results in both evoked responses, as well as changes to ongoing oscillations (induced responses), in the EEG (Table 1). Disentangling ongoing oscillatory and stimulus-evoked activity, is a rather challenging endeavour given the confounds in the quantification of both phenomena. Firstly, analysing the ongoing activity in a time-frequency representation creates a time-frequency trade-off, where increasing frequency precision leads to a decrease in temporal precision and vice versa (e.g. see Cohen, 2014). A direct consequence of the time-frequency approach is smearing of a large transient response spanning several frequencies, thereby affecting the ongoing oscillatory activity in its time-frequency environment. Finally, high correlations in phase measures from decomposed signals (e.g., wavelets) may indicate correlated phases in the ongoing activity, but can also represent the coherent phase of an additive, evoked response (Laufs et al., 2003). These methodological issues obscure the relationship between induced and evoked brain signals.
So, why is it important to elucidate the relationship between oscillatory activity and the event related potentials, introduced above? This importance comes about most clearly in the field of working memory, because the first evidence of a neural mechanism for working memory stems from the ERP literature, not from ongoing oscillations (Vogel & Machizawa, 2004). The CDA was the first neural correlate to working memory capacity. Moreover, ERPs are conceptually easy to understand and the methodology to acquire the ERP is straightforward. And as emphasized before, the ERP has a long history in neuroscience.
In addition, there are some similarities between ERPs and ongoing oscillations. The P300 for example has been related to attention and memory processes during information processing, but the P300 has also been coined an inhibitory wave. Like the P300, alpha oscillations have been related to attention processes and inhibitory processes in task-related and task unrelated brain regions. These similarities beg for a critical investigation of the underlying mechanisms. Moreover, the CDA was related to asymmetric modulation in the alpha band, by Van Dijk and colleagues in 2010 but a recent paper from Fukuda et al. (2015) challenged those findings. This review aims to give an overview of the different models explaining the relationship between induced and evoked responses and focuses on studies on amplitude asymmetry in particular.
There are three historically, influential models on the relation between ongoing oscillatory and stimulus-evoked activity that will be discussed and reviewed in the remainder of this paper (Table 2). The additive and phase-resetting models are more focused on the relation between ongoing oscillations and transient evoked responses, whereas the amplitude asymmetry model largely relates to slow evoked responses. Transient responses are short-lived, phase-locked responses to external events, whereas slow evoked responses are time-locked responses that are more related to higher cognitive processes, like memory and consciousness. Understanding the interaction between event-related neural responses and ongoing oscillations is important, because ongoing oscillations are hypothesized to be a means of communication between different neuronal sub-populations (Fries, 2005), and elucidating the role of these oscillations may eventually lead to a neural understanding of complex behavior. This review attempts to examine the most influential models and review empirical evidence in the light of the amplitude asymmetry model. Therefore, the next sections will review the different models, whereafter the relationship between two different
sustained responses, the P300 and the CDA, and oscillatory activity, specifically in the alpha band, will be examined.
Table 2. Most influential models on the relation between ongoing oscillations and evoked responses.
Ongoing activity Added signal Method Evidence Additive Model Background noise Necessary Linear Averaging Mäkinen et al. (2005) Phase-resetting Model Partial reset causes the ERP to appear Unnecessary Frequency Analysis Tiesinga et al. (2001); Makeig et al. (2002) Amplitude asymmetry Model Amplitude asymmetry causes ERP to appear Unnecessary Asymmetry of amplitude fluctuation index Nikulin et al. (2007); Mazaheri & Jensen (2008)
ADDITIVE MODEL
The additive model of ERPs largely relies on the idea that background oscillations are noise and that the ERP is measured due to the synchronization of post-synaptic currents after an experimental event. Averaging trials will decrease the noise component and increase the ERP of interest. The event, e.g. a stimulus, adds a fixed-polarity, fixed-latency signal to the background noise present in the EEG. Therefore, averaging of the signal merely cancels out the non phase-locked oscillations. The most important implication of this model is that it justifies the extensive literature where ERP components are linked to perceptual and cognitive functioning, their anatomical substrate (Burgess, 2012) and neuropsychological disorders (e.g., Burwell, Malone, Bernat, & Iacono, 2014). The phase-resetting model proposes that all ERP-like phenomena can be constituted by a phase-resetting mechanism (see below), making the interpretation of the extensive ERP literature less straightforward than previously assumed.
Although the additive model has been influential for decades, one limitation is that the additive model is unable to explain the shape of the ERP wave. The ERP’s frequency decreases over time and the modulation of the amplitude of the signal is inversely proportional to the frequency (Burgess, 2012). This relation between amplitude and frequency suggests that ERPs have an underlying structure in the frequency domain, for which the additive model cannot easily account (Burgess, 2012). Interestingly, the additive model predicts more post-stimulus power in the EEG signal, because that is where the evoked response would be riding atop of the ongoing oscillations. This prediction, however, was not found to be fulfilled by Sayers et al. (1974), which led Sayers and his colleagues to hypothesize that the ERP stems from a realignment of the phase in the ongoing EEG and hence they postulated the phase-resetting model. In summary, the additive model is unable to account for the shape of the ERP wave, nor for the absence of a post-stimulus power increase.
The additive model proposes that the ongoing EEG and the ERP are largely independent neuronal events. The ERP should appear in the frequency domain as a short transient modulation in amplitude, therefore this model is also referred to as the amplitude-modulation theory. Next, we will review the evidence in favour of the phase-resetting model, but we will get back to evidence in favour of the additive model when we directly compare the models further and present evidence from the additive model opposing the phase-resetting model.
The phase-resetting model hypothesizes that the ERP stems from a (partial) phase alignment of the ongoing EEG to the stimulus. The phase alignment causes positive and negative peaks in the ongoing oscillations to summate into an ERP, which makes an evoked, additive signal unnecessary (Burgess, 2012). This model is supported by extensive evidence relating oscillatory activity in the alpha (8-13 Hz) frequency band to the amplitude of the ERP (Mäkinen, Tiitinen, & May, 2005; Rahn & Başar, 1993; Tiesinga et al., 2001), which may suggest that amplitude changes in these ongoing oscillations underlie the generation of the ERP. Opponents of the additive model however would argue that a latent variable might be responsible for the link between ongoing oscillatory activity and the amplitude of the ERP, e.g. the vigilance of the subject is known to modulate both the ERP and ongoing oscillations (Mäkinen et al., 2005).
Evidence in favor of the phase-resetting model was provided by Makeig et al. (2002), who showed that partial phase resetting in the absence of an increase in energy explained at least some important aspects of the ERP. An immediate consequence of this phase-resetting model would be that the interpretation of ERP components can not provide insights at the functional level, because the peaks and troughs of the ERP are simply artifacts, due to the phase-resetting of the ongoing oscillations (Burgess, 2012). In that regard, the additive and phase-resetting model are each other’s opposites.
Researchers supporting the additive model over the phase-resetting model claim that a functional role of phase alignment would be problematic, because firstly, ongoing oscillations may phase-synchronize sporadically which would then cause an illusory percept and secondly, when an ongoing oscillation is already in the state wherein it would be set by a stimulus, the stimulus presentation would not be able to influence brain activity (Mäkinen et al., 2005). Moreover, the reasoning of Sayers and colleagues (1974), where a post-stimulus increase in energy was hypothesized if the additive model were to be true, has been interpreted as premature in the presence of sufficient alpha activity (Becker, Ritter, & Villringer, 2008).
It can be concluded that the phase-resetting model is not easily teased apart from the additive model, because adding an ERP waveform to ongoing oscillations with random phases, as suggested by the additive model, will increase phase-locking measures. Because the added activity will be phase-locked to some extent (Mazaheri & Picton, 2005). Phase synchronization during the ERP, therefore, cannot provide definitive proof in favor of the phase-resetting model (Fell et al., 2004). Furthermore, contrasting results have been found when examining phase resets in evoked responses. Makeig and his colleagues (2002) found support for the phase-resetting model for early evoked responses, whereas Fell and his colleagues (2004) did not find support for the phase-resetting model for transient evoked responses. These discrepancies suggest that the additive and phase-resetting model may interact, or may explain slightly different aspects of the ERP. Finally, there are more difficulties when directly comparing the additive and phase-resetting model, which will be discussed next.
How resetting at the microscopic level may underlie both the additive and the phase-resetting model
According to the models explored so far, a stimulus presentation either modulates the synchrony of an already active population or activates neurons that were relatively quiet before stimulus onset (Telenczuk, Nikulin, & Curio, 2010). As previously explained, one of the hallmarks of the additive model is the post-stimulus increase in power. The underlying mechanism is thought to be the activation of quiescent neurons, without interfering with ongoing oscillatory activity. It is the activation of these neurons that will lead to the hypothesized power increase. Importantly, when inspecting the macroscopic response, it is impossible to distinguish the additive from the phase-resetting model, because the added activity will be phase-locked to the stimulus. The phase-reset model suggests that a microscopic phase-reset is underlying the evoked response. The realignment of the phases of neuronal populations at the microscopic scale increases synchrony and hence increases coherence, which at the macroscopic level will be recorded as oscillations. Once more, the
models cannot be distinguished, because the increase in oscillations will be measured as a power increase. Furthermore, for the phase-locking model there are two scenarios: the phase-reset can take place in an asynchronous or in a synchronous population. The microscopic and macroscopic effects on an asynchronous population have been discussed above. However, when synchronous neuronal activity is phase-reset, the macroscopic measurements will not show increased synchrony, nor increased power. All in all, it is difficult to disentangle the microscopic mechanism from a macroscopic measure.
Shared neural generators
The first explanation for the contradictory findings in the literature came from Mazaheri & Picton (2005), where shared neuronal generators were proposed. In this hypothesis most neurons are involved in both the generation of the stimulus-evoked ERP and in the ongoing oscillations in a particular region of the brain as recorded with EEG. The shared neuronal generator hypothesis predicts that a stimulus changes several things (Mazaheri & Picton, 2005). Firstly, relatively quiescent cells may become activated by a stimulus and this activation can be relatively time-locked or not. When the activation is largely time-locked, the activation will appear as a distinct evoked ERP, but when the activation is variably time-locked activity will appear as an event related synchronization (ERS) or a power increase in the EEG rhythm. Secondly, the stimulus can cause phase-resetting of neurons, which are already active. Thirdly, event related desynchronization (ERD) can be explained by either a general inhibition of oscillatory activity, or several ERSs can lead to activity cancelling each other out, which appears as a decrease in power or ERD.
The shared generator hypothesis can account for some experimental findings, that neither the additive nor the resetting model can account for. The additive and the phase-resetting model have difficulties explaining an ERD in the ongoing oscillations taking place simultaneously with an evoked response (Mazaheri & Picton, 2005). The shared generator hypothesis can account for these findings, because neurons that were involved in generating the ongoing oscillations before stimulus onset may become involved in the generation of the ERP after stimulus onset, leading to an ERD at the same time as the ERP. The shared generator hypothesis also accounts for the discrepancies between both models. It explains why when the ongoing activity is low, an ERP manifests itself as an added response (in favor of the additive model), whereas when the ongoing activity is high, an ERP does not come about as an added response (in favor of the phase-resetting model). The same neural generators are involved in both responses, so when the background activity is high, an ERP will not show additive characteristics. Finally, the intimate link between the phase of the ongoing EEG and the ERP may be explained by the relative excitability or state the neurons are in before they are recruited to participate in the ERP (Mazaheri & Picton, 2005).
Firefly model
More recently, the firefly model of synchronization through cross-frequency phase modulation was coined by Burgess (2012). This model also accounts for certain differences between the phase-resetting and additive model and in addition accounts for cross-frequency information processing. Burgess (2012) suggested that a phase realignment will be accompanied by a change in frequency, because one cannot occur without the other even if only for a very short period of time. This would allow differentiation between the additive and the phase-resetting model, because the added signal would lead to instantaneous phase changes that would lead to a transient infinite frequency. This infinite frequency is likely to get lost in the signal due to averaging or filtering (Burgess, 2012). For the phase-resetting account however, changes in phase would be gradual over time and they would therefore be picked up as frequency changes. With empirical evidence Burgess (2012) showed that there is indeed a frequency change and proposed the firefly model, which will be explained next.
Before stimulus onset different networks oscillate at different preferred frequencies (Burgess, 2012). Some networks will oscillate at the same frequency but their phases are not aligned. When a stimulus is presented, the networks transiently synchronize their phases by
adjusting their frequency slightly. According to the firefly model, it is this frequency slowing, within different networks of the same preferred oscillation frequency, that gives rise to the ERP (Burgess, 2012). The shifting of the frequency bands has important implications, because if the firefly model is correct, it means that frequency bands are not fixed entities but dynamic in themselves making classical Fourier analysis and wavelet decomposition unsuitable for M/EEG data analysis. Instead, assumptions of the firefly model should be tested with Empirical Mode Decomposition (EMD). EMD uses so-called Intrinsic Mode Functions (IMFs), which can change amplitude and frequency over time. On the IMFs a Hilbert transform can be applied allowing the extraction of instantaneous power, phase and frequency. This combination is known as the Hilbert-Huang transform and some empirical papers have started implementing this method (Burgess, 2012; Cong et al., 2009; Gross, 2014). However, more empirical evidence is needed to reliably test the predictions posed by the firefly model.
AMPLITUDE ASYMMETRY MODEL
Conventionally (hence in all the models discussed so far), the amplitude of oscillatory activity is viewed as symmetric around zero. Amplitude asymmetry, however, can be found in many systems around us (Mazaheri & Jensen, 2010a). Amplitude asymmetry refers to the observation that the troughs or the peaks of a signal are more affected than the other component, e.g. when the peaks are modulated to a large extent, the trough values are more or less unaffected. The classical example given by Mazaheri & Jensen (2010a) states that the amount of light in your office at 12AM will show stronger modulations on a day to day basis, than the amount of light in your office at 12PM. If amplitude asymmetry is present in ongoing oscillations, event-related activity will not average out over trials, but will lead to the formation of sustained responses instead (Mazaheri & Jensen, 2010a). In other words, amplitude asymmetry causes oscillations to take the shape of an ERP after averaging those oscillations over many trials.
“Non-zero mean” properties in oscillations were first linked to event related responses by Nikulin and colleagues (2007). A consequence of these non-sinusoidal properties is that the signal distribution is skewed so that the mean covaries with the magnitude of the signal (van Dijk, van der Werf, Mazaheri, Medendorp, & Jensen, 2010). As a result, the mean of the signal will be different before and after the modulation. Mazaheri & Jensen (2008) created the asymmetry of amplitude fluctuation index (AFAindex). This index compares the variance in
the peaks and the troughs by taking the normalized difference between the two variances (Grothe & Plöchl, 2008; Mazaheri & Jensen, 2008). Mazaheri & Jensen (2008) indeed showed that the peaks and troughs of alpha oscillations are modulated differently.
At the physiological level Mazaheri & Jensen (2008; 2010a) proposed that asymmetric amplitudes stem from a disbalance in the inward and outward flowing currents. It is well known that dendritic currents in pyramidal cells underlie M/EEG signals (Hämäläinen et al., 1993). Excitatory synaptic inputs and after-hyperpolarization currents are most likely to mechanistically explain forward propagating dendritic currents (Mazaheri & Jensen, 2010a). And although both back-propagating dendritic currents and outward dendritic currents, that are caused by depolarizations around the soma, exist, it seems unlikely that these currents exactly match the forward propagating currents (Mazaheri & Jensen, 2010a). Furthermore, the inhibition from the alpha oscillations is hypothesized to come about through GABAergic feedback from interneurons, which are paced by neocortical or thalamic rhythm generators (Mazaheri & Jensen, 2010a). The GABAergic feedback can then silence information processing in the pyramidal neurons or decrease the excitability of the neuron via shunted inhibition (Mazaheri & Jensen, 2010a). Moreover, with modeling data Mazaheri & Jensen (2010a; their Fig. 9) showed that the power in the alpha band is related to cortical excitability, so that when alpha power is low, cortical excitability is high and firing is not alpha phase-dependent, but when alpha power is high, cortical excitability is low and firing is alpha phase-dependent.
More recently, a framework called function-through-biased oscillations (FBO) was coined where the author synthesized two theories (Schalk, 2015), namely communication through coherence (CTC) and gating-by-inhibition (GBI). CTC was posed by Fries (2005), where he suggested that the functional connectivity between two different neuronal populations comes about by increasing the coherence between oscillatory activities from the populations. Communication is then established preferentially during the troughs of these signals allowing the phase of an oscillation to execute a crucial role (Fries, 2005). GBI was posed by Mazaheri & Jensen (2010b) and in contrast to CTC relies on oscillatory power. They suggested that task-unrelated regions are functionally inhibited by increased oscillatory power in the alpha band (Schalk, 2015; Mazaheri & Jensen, 2010b). Schalk (2015) suggested that the relation between oscillatory power and phase and its relation to cortical excitability remains to be elucidated and proposed the FBO hypothesis. This hypothesis rests on two principles, where the first principle states that the instantaneous voltage amplitude, rather than the power or phase, reflects cortical excitability closest. And the second principle suggests that biased oscillations determine cortical functioning (Schalk, 2015). This second principle heavily relies on peak-to-peak amplitude and phase of the biased oscillations and it is here that the FBO hypothesis shows large overlap with the rhythmic pulsing account of Mazaheri & Jensen (2010b), whereas the first principle is conceptually the same as the notion of amplitude asymmetry. Therefore, it remains to be clarified how the FBO hypothesis differentiates itself from the rhythmic pulsing account. However, both accounts suggest that asymmetry in the signal plays a larger role than previously thought.
In summary, the amplitude asymmetry model directly provides an explanation for the interdependencies of ongoing oscillations and ERPs, allowing slow, sustained ERPs to be produced by asymmetric modulations of ongoing oscillatory activity especially in the alpha band. The next step therefore is to review the evidence linking amplitude asymmetry and sustained ERPs. Different sustained ERPs will be compared to ERDs in the alpha band, starting with the P300, but first a short, historical overview of alpha oscillations will be provided.
Alpha oscillations
The alpha rhythm from human scalp EEG was first described by Berger (1929). The famous Berger effect refers to his discovery that alpha oscillations seemingly disappeared from the EEG upon opening the eyes or when performing a cognitive task (Bastiaansen et al., 2012). Traditionally, the alpha rhythm was seen as an idling rhythm because of its most pronounced presence during rest (Pfurtscheller & Lopes da Silva, 1999). This task of alpha oscillations as passive inhibition was challenged by findings that showed more alpha power over regions in the hemisphere contralateral to the location where participants were planning to saccade to, compared to alpha power in the ipsilateral hemisphere (Medendorp et al., 2007). Moreover, power increases were also observed during working memory maintenance (Klimesch, Doppelmayr, Schwaiger, Auinger, & Winkler, 1999). These findings together led to the hypothesis that alpha oscillations may serve functional inhibition. It is also important to emphasize here that alpha oscillations receive special attention when comparing evoked and induced responses, because alpha oscillations are the most prominent oscillations in the EEG (e.g. Makeig et al., 2002). Moreover, in contrast to other frequency bands, the alpha band oscillations can both synchronize or desynchronize (ERS or ERD) in response to a stimulus or task demands (Klimesch, 2012), whereas the other bands typically only respond with synchronization.
ERSs in the alpha band are seen as active inhibitory mechanisms, e.g. seen over ipsilateral hemisphere when attending one hemifield, whereas an ERD in the alpha band over the contralateral hemisphere can be viewed as a release of inhibition. These (de)synchronizations are reflected in power measures, whereas alpha phase also has consequences for ERPs (e.g. Hanslmayr et al., 2006). Mazaheri & Jensen (2010b) extended their rhythmic pulsing account of ERP generation and alpha oscillations towards the gating-by-inhibition hypothesis, as mentioned above. In this review the alpha oscillations are
interpreted in line with the gating-by-inhibition hypothesis, where both power and phase are mechanistically meaningful. Due to the implications of alpha oscillations in inhibitory processes, the relationship between the P300, which is an ERP hypothesized to be related to inhibitory processes, will be examined next.
P300 and alpha oscillations in an oddball paradigm
Introduction to the P300
The P300 is a positive going ERP component peaking at about 300 ms after stimulus-onset. The P300 can also be referred to as the P3, indicating that it is the third positive wave in the averaged EEG trace. A P300 is elicited by a traditional two-stimulus oddball paradigm, when participants detect infrequent targets in a stream of more frequent, standard stimuli (Ritter & Vaughan, 1969). The difficulty to distinguish the infrequent target from the frequent standards modulates the amplitude and latency of the P300 (Kok, 2001). This modulation has been ascribed to the attentional demands, which are greater when the stimuli are more difficult to disentangle (Katayama & Polich, 1998). Although the underlying neural generators are not precisely known, the P300 is most clearly seen over centroparietal regions at the scalp level (Huang, Chen, & Zhang, 2015). Importantly, the P300 was also visible for infrequent stimuli in the absence of a task and therefore this wave is now referred to as the P3a, to distinguish it from the task-relevant P3b.
The first studies on the neural underpinnings of the P300 focused on recordings from the medial temporal lobe with depth electrodes in the hippocampi of epileptic patients. These recordings suggested that the P3b is generated in the hippocampus, but further research showed no difference in either P300 amplitude or latency, between patients with bilateral hippocampal lesions and healthy controls (Molnár, 1994). P300 amplitude is influenced by the presence of the temporal-parietal junction (Polich, 2007). Rare or physically different stimuli have been shown to activate the frontal lobe as indicated with both fMRI and ERP studies (Huang et al., 2015). In addition, patients with prefrontal damage showed diminished P3a amplitudes, but the P3b was unaffected (Knight et al., 1995). Combined, these findings suggest that the P3a and P3b are involved in a network including the frontal and temporal/parietal brain areas.
Although the P300 can be generated reliably with the oddball tasks and this ERP has been studied for 50 years now (Sutton, Braren, Zubin, & John, 1965), it remains to be elucidated how and why the brain generates a P300. The P300 seems to reflect the allocation of attentional resources during information processing (Humphrey et al., 1994; Yordanova, Kolev, & Polich, 2001), which relates the P300 to both attentional and memory related processes. Because of the P300’s involvement in attention and memory-processes, the P300 has been hypothesized to be closely related to inhibitory processes (Polich, 2007). The P300 is thought to suppress ongoing activity to allow communication of task-related information between frontal (P3a) and temporal-parietal (P3b) sources. It are these features of the P300 that suggest an intimate relationship between alpha ERD and the P300.
However, comparing the P300 and alpha ERD comes with methodological challenges related to the ones previously discussed. First, it is important to keep in mind that the P300 stems from phase-locked activity whereas alpha oscillations are non-phase-locked. This difference may suggest that distinct neural events underlie these measures (Yordanova et al., 2001), but the P300 and alpha ERD can still reflect the same cognitive process, which would cause task-effects of the two measures to correlate in the spatio-temporal domain. Moreover, individual differences should affect both measures equally (Yordanova et al., 2001). Second, P300 and alpha ERD both depend on background alpha EEG in several ways: e.g. Price (1997) found that ongoing alpha activity modifies the P300 and similar reports have been made for alpha ERD and prestimulus alpha activity (Feige et al., 1996). The amplitude asymmetry model suggests that the averaging of ongoing activity leads to ERP-like waves because of non-sinusoidal properties in the signal. It is important to note that the studies reviewed here did not specifically look at the non-sinusoidal properties of the ongoing EEG trace. This hampers the interpretation of the results, because we cannot directly compare the neural generators of the P300 and the alpha amplitude. However, these
papers do address an important and related question, namely: what is the relationship between the P300 and alpha ERD?
Alpha ERD and P300 are closely related
First, a study by Yordanova et al. (2001) will be discussed shortly in which participants performed an auditory oddball task condition and a passive listening condition. The alpha ERD was subdivided in a low (7-10 Hz) and a high (10-14 Hz) alpha band. The major finding showed a similar pattern for both alpha ERD and P300 for both task and scalp distribution (Yordanova et al., 2001). They found no clear P300 and an absence of alpha ERD in the passive listening condition and went on to conclude that alpha ERD, like the P300, was involved in attention resource allocation in relation to memory updating. However, when they zoomed in on the temporal aspects of the task condition, they found that the major ERD activity took place much later than P300 peak latency. This suggests that alpha ERD and P300 are not identical because they emerged at different times (Yordanova et al., 2001). The researchers went on to suggest that their findings showed that the P300 predicted alpha ERD strength and timing. More specifically, they found that larger and shorter P300s were related to delayed and enhanced lower alpha band ERDs, but to inhibit higher alpha band ERDs. This last finding also suggested that separating the alpha band in a low and high band seems sensible, because the P300 affected both bands differently (Yordanova et al., 2001).
The second study that will be discussed examines the causality in the association between the P300 and alpha ERD (Peng, Hu, Zhang, & Hu, 2012). The authors correctly observed a discrepancy in the literature, where the study above from Yordanova and colleagues (2001) is compared to a review from Polich (2007). Yordanova et al. showed that alpha ERD was guided by internal events indexed by the P300, Polich on the other hand suggested in his review that the amplitude and latency of the P300 could stem from alpha ERD (Peng et al., 2012). Peng and colleagues (2012) used an oddball paradigm in four different modalities, which allowed them to study modality dependency of the alpha ERD during cognitive processing. In addition, the researchers examined effective connectivity at the source level with Granger causality, which allowed them to investigate the discrepancy mentioned here. Granger causality allows inferences of the directionality in functional connectivity between one area and the other. The rationale is that if Yordanova and colleagues are right, the sources underlying the P300 will drive communication to the sources of alpha ERD, but if Polich is correct, the alpha ERD sources will drive the P300 sources. Over the different blocks the same stimuli were used as frequent non-target stimuli (i.e. in the first block) and as infrequent targets (i.e. in the second block), which allowed the researchers to investigate P300 and alpha ERD characteristics to both target and non-target processing.
For the target condition the researchers reported that for the different modalities the topographic distribution and the cortical sources of the P300 and alpha ERD were highly similar and located in the posterior cingulate cortex and in the occipital lobes. For the non-target condition the alpha ERD showed a similar occipital pattern for auditory and visual stimuli, but a contralateral distribution for the somatosensory and pain modalities. Based on these findings the researchers go on to conclude that the P300 and alpha ERD reflected task-related parameters like attention, in the target condition, and that these measures were modality independent. The alpha ERD for non-target processing however, seemed modality dependent and therefore mainly reflected sensory perception and judgment (Peng et al., 2012). Effective connectivity analyses based on Granger causality were performed on target stimuli only and Peng and colleagues found that cortical information systematically flowed from alpha ERD sources to P300 sources between 300-500 ms after stimulus onset in the 2-4 Hz frequency range.
Based on these findings the researchers suggested that the P300 is largely modality independent and reflects high cognitive activation and attention during target processing. The alpha ERD followed the P300 pattern for target processing, which suggests that during target processing the alpha ERD was caused by internal mental events. However, for non-target processing the alpha ERD showed modality dependent differences, which suggests
that these alpha ERDs were more related to external sensory stimuli. Based on the connectivity analyses the researchers went on to conclude that in line with the findings of Yordanova et al. (2001) the P300 and alpha ERD are functionally associated, but that information flows from bilateral alpha ERD sources to P300 sources during target processing in a modality independent manner. This last finding contrasts sharply with the findings of Yordanova et al. (2001) who found that the P300 is predictive of alpha ERD, which is difficult to reconcile if the alpha sources are driving the P300 sources.
Conclusion
The research reviewed here showed that alpha ERD and the P300 are intimately linked. However, the results of Yordanova et al. (2001) suggested that the P300 is predictive of the low and high alpha band ERDs in a different manner, which suggests that the P300 drives alpha oscillations, whereas the results of Peng et al. (2012) suggested that the alpha ERD sources are leading the P300 sources. These discrepancies may largely be related to methodological differences. First, Yordanova and colleagues divided the alpha ERD in a lower (7-10 Hz) and higher (10-14 Hz) alpha band, whereas Peng and colleagues treated the alpha band as one band (8-13 Hz). Second, Peng and colleagues used Granger causality at the source level to study the effective connectivity between the P300 and alpha ERD. Yordanova et al. (2001) based their results on stepwise multiple regression analysis at the electrode level. The approach of Yordanova and colleagues (2001) is therefore correlational, whereas Peng et al. (2012) used a causal measure. Third, the non-target conditions differed for both tasks, because Yordanova’s study used a passive listening task, whereas Peng et al. (2012) used the same task but flipped the infrequent target from the first block to be a frequent non-target in the second block. Both approaches come with disadvantages, because with a passive listening condition you introduce differences in vigilance within participants. Flipping the stimulus meaning from one block to the other, however also has implications, because the signal-to-noise ratio is much better for the often presented non-target stimuli compared to the infrequent target-stimuli (160 vs. 40 trials, respectively), which Peng and colleagues did not control for. Finally, it should be emphasized that none of the discussed studies examined the asymmetry of the alpha oscillations and none of the studies directly examined the relationship between alpha oscillations and the P300 in a systematic way. Also, the P300 was studied as one phenomenon and there was no differentiation made between the P3a and P3b. In summary, both studies were indicative of an intimate relation between the two measures and the studies discussed here are not able to exclude the possibility that amplitude asymmetry in the alpha oscillations underlies the P300, or part of the P300 (e.g. P3a or P3b).
Working memory delay activity and alpha oscillations
Working memory and slow evoked responses
Working memory is a process, which is used to keep sensory information online for short-term memory storage and manipulation, and acts on the order of seconds (Ma et al., 2014). It has been considered a core cognitive process that is highly correlated to other cognitive tasks like long-term memory, fluid intelligence and attention control (Unsworth, Fukuda, Awh, & Vogel, 2014; Unsworth & Robison, 2015). According to the classical view working memory is limited to keeping a small number of items online, although this dogma is challenged by some more recent evidence (see Ma et al., 2014 for a review). A slow negative evoked response called the contralateral delay activity (CDA) was found to correlate with individual differences in working memory capacity (Vogel & Machizawa, 2004). The CDA can scale linearly with increased items to remember, or show an inverted U-shape response (Ma et al., 2014). Interestingly, two studies on the relationship between the slow evoked responses and alpha oscillations appeared more recently with opposing results. Next, both studies will be reviewed and their differences will be highlighted.
Conflicting results when comparing negative slow waves and alpha oscillations in working memory paradigms
First, the study that found that amplitude asymmetry of alpha oscillations can explain slow evoked fields, like the CDA, will be discussed. Participants performed a spatial working memory task, where a target location was cued and participants were instructed to keep the location in mind after the disappearance of the target. After a 1.6 s delay, participants were instructed to make a saccade to the location where the target was presented (van Dijk et al., 2010). With this paradigm and magnetoencephalography (MEG), the researchers found that the CDA was reliably evoked and showed a contralateral distribution. In other words, sensors over the left hemisphere showed higher amplitudes for target locations in the left hemifield than in the right hemifield and the reverse was true for sensors over the right hemisphere. The same lateralized pattern was found for alpha power (van Dijk et al., 2010). Moreover, the CDA and alpha lateralization showed a similar scalp topography as well as temporal pattern, which suggested that both measures reflect the same mechanism. The authors then continued to calculate the AFAindex showing significant amplitude asymmetry in
the alpha and beta band, with a distribution over posterior sensors for the alpha band. Finally, the AFAindex was shown to correlate with the ERFs (MEG equivalent of ERPs) over
subjects. Based on these findings the authors concluded that slow evoked fields related to working memory maintenance might be explained by amplitude asymmetry in alpha oscillations (van Dijk et al., 2010).
Next, the study that found that the slow wave negativity and alpha suppression showed qualitatively dissociable patterns during working memory retention (Fukuda, Mance, & Vogel, 2015) will be discussed. The authors started with highlighting a shortcoming in the traditional designs used to study working memory, where stimuli are displayed across the entire hemifield and a cue tells participants which hemifield to attend. The authors correctly pointed out that as the stimuli that need to-be-remembered on the cued side increase, there is also an increase in the to-be-ignored stimuli in the uncued hemifield. The authors therefore used a whole hemifield display to present their stimuli without information that needed to be ignored over the retention interval (Fukuda et al., 2015). Visual working memory capacity for each set size was estimated in each individual by using Cowan’s K, according to the formula: K = set size ✕ (hit rate - false alarm rate). This approach has been widely used in working memory paradigms, although recent evidence seems to challenge this traditional K approach (Ma et al., 2014).
In accordance with previous literature the researchers showed that working memory capacity increased up to set size 4 and then plateaued (Fukuda et al., 2015). In accordance, the study found that alpha power decreases up to set size 3 and then plateaus as well. The alpha band was defined by the authors based on the inflection point of different set sizes and a band of 7-9 Hz was selected (Fukuda et al., 2015). The alpha band suppression was found to correlate with individual differences in visual working memory capacity. Specifically, the study showed that high-capacity individuals showed more alpha suppression for supracapacity set sizes (set sizes higher than the inflection point), than subcapacity set sizes, but that this pattern was absent in low-capacity individuals (Fukuda et al., 2015). These findings suggested that alpha suppression is a neural correlate of working memory capacity. For the negative slow evoked response Fukuda and colleagues reported very similar findings to the CDA and go on to conclude that the negative slow wave is a neural correlate of working memory capacity limit.
The researchers used alpha phase bins to examine the modulation of the peaks and troughs and showed that the peaks were modulated stronger by the set size effect than the troughs, which is in accordance with the amplitude asymmetry model. However, correlational results showed that the negative slow wave and the alpha suppression explained different aspects of individual differences in visual working memory capacity (Fukuda et al., 2015). Therefore, the researchers did an additional experiment to discriminate between both measures by elongating the retention interval, using only set sizes one and three. They showed that the negative slow wave only showed a significant set-size effect for the first time bin 400-1000 ms, whereas the alpha suppression showed a significant set-size effect up to the 1600-2200
ms bin. These results indicated that the negative slow wave and the alpha suppression are qualitatively different, because their timings are different. The researchers speculated that the negative slow wave might reflect attentional control mechanisms active during encoding, whereas alpha suppression might reflect the recruitment of a synchronized idle population.
Conclusion
Van Dijk and colleagues (2010) presented correlational evidence that amplitude asymmetry in alpha oscillations underlies the CDA during the retention period of a working memory paradigm, whereas Fukuda et al. (2015) showed that the negative slow wave and alpha suppression represent qualitatively different mechanisms over longer retention periods. The paper of Fukuda and colleagues used an indirect measure of amplitude asymmetry and found that the peaks of the signal were more strongly modulated than the troughs. They explained this asymmetry with a DC offset in the signal, rather than a true amplitude asymmetry in their data. Their measure may have been sensitive to the DC offset, making this a sensible explanation. Fukuda et al. (2015) go on to show that the alpha suppression and the negative slow wave are not correlated over different set sizes and therefore explain unique variances of the data. Moreover, they showed that the alpha suppression outlasts the negative slow wave, which suggests that the alpha suppression and the negative slow wave are qualitatively different signals. These findings make it unlikely that the alpha suppression and the negative slow wave are manifestations of the same neural events. A systematic overview of the different set sizes, however is lacking. Especially, when focusing on Figure 6 of the paper of Fukuda et al. (2015, p. 14013), it becomes clear that the modulation for set sizes 1 and 3 is less asymmetrical than the modulation of peaks and troughs for set sizes 2 and 4. It would have been interesting to see a similar figure for the data obtained with longer retention intervals. However, the decision to choose different set sizes is unlikely to change the interpretation of the results.
Interestingly, the study of Van Dijk et al. (2010) also measured amplitude asymmetry in their data, but used a direct measure (AFAindex) to measure the presence of amplitude asymmetry.
Although, Nikulin et al. (2010) showed that in some cases the AFAindex is sensitive to
harmonics at 20 Hz in the data and can falsely report amplitude asymmetry at 10 Hz, this does not explain the relation between the alpha suppression and the CDA described by Van Dijk and colleagues (Mazaheri & Jensen, 2010a). More likely explanations for the differences between both studies should be sought in the different methodologies used by both studies. First of all, Van Dijk et al. (2010) used MEG, whereas Fukuda et al. (2015) used EEG. However, this difference is unlikely to underlie the differences between both studies because the amplitude asymmetry model suggests that amplitude asymmetry would largely affect both signals (MEG or EEG) equally (Mazaheri & Jensen, 2010a; their Fig. 8). Second, the working memory task was different and therefore the measured negative slow waves were different between both studies. Participants in the study of Van Dijk et al. (2010) were instructed to keep a target location in mind over the retention interval, whereas participants in the study of Fukuda et al. (2015) were instructed to hold a memory array online during the retention interval. After the retention interval the tasks also differed wildly. These methodological differences may contribute to the differences highlighted here. Where one memory array is lateralized (van Dijk et al., 2010), the other is not (Fukuda et al., 2015). And where in one task the response is a match-to-sample (Fukuda et al., 2015), the other is not (van Dijk et al., 2010). Therefore, the negative slow waves evoked by the paradigms are qualitatively different, because of the different paradigms used (lateralized vs. whole field). Finally, it is interesting to consider the role of individual differences in alpha peak oscillations, because the study of Fukuda et al. used a fixed 7-9 Hz alpha band across participants whereas the study of Van Dijk et al. used 9-13 Hz. Neither of the studies took into account the individual variability in alpha peak frequency (Haegens, Cousijn, Wallis, Harrison, & Nobre, 2014), which also dynamically changes from rest to task activity.
In addition to these electrophysiological results, fMRI studies showed that decoding working memory representations can be done in the ipsilateral hemisphere, in addition to the contralateral hemisphere (Ester, Serences, & Awh, 2009; Pratte & Tong, 2014). This
suggests that the ipsilateral hemisphere is also involved in working memory and that indeed the CDA and the alpha oscillations are not as intimately linked as previously thought. As discussed, Fukuda and colleagues recently followed up on these fMRI findings with electrophysiological data, which suggested that the CDA indexed the contralateral visual working memory representations and that alpha oscillations indexed the spatially global visual working memory representations. These findings suggest that the CDA and alpha oscillations underlie different aspects of working memory. However, because of the experimental differences these findings can not explain the findings of Van Dijk et al. (2010). Moreover, a very recent fMRI multi-voxel pattern analysis (MVPA) study showed that decoding is possible from the superior IPS and the parietal region but not from the occipital regions (Bettencourt & Xu, in press). The electrodes used in the study of Fukuda et al. (2015) span the occipital and parietal regions, whereas the study of Van Dijk et al. (2010) only looked at parietal electrodes.
At this point, it would be interesting to use a combined EEG/fMRI approach to examine the underlying sources of the sustained ERP and the alpha oscillations during the encoding and retention of stimuli for working memory purposes. Although, source-level reconstruction in combination with a causal measure for functional connectivity may also shine light on the underlying neural sources. Elucidating the relationship between these sustained responses and alpha oscillations are especially important in the context of working memory, because of the intimate link between other higher cognitive functions and working memory.
Other slow waves and other frequencies
The amplitude asymmetry found in alpha oscillations has been compared to the P300, which seems to reflect attentional allocation, and to working memory related potentials like the CDA, above. However, there are other sustained responses that might come about from averaging asymmetric oscillations, which will be discussed here briefly. It is important to note that amplitude asymmetry might also be present in other frequency bands and therefore a relation between amplitude asymmetry in other bands, than the alpha band, and sustained ERPs should be considered. ERPs related to cognitive control and timing mechanisms will be discussed shortly.
A widely used paradigm to study conflict processing is the Stroop task, where a word representing a color is either presented in the same color (congruent stimuli, blue) or in a different color (incongruent stimuli, red) and participants are asked to report the color of the word. Resolving the conflict of the automatic activation of the semantic color and the font color is a core cognitive control function. The N450, a negative deflection in the time window of 350-500 ms upon stimulus onset, and the late positive component (LPC) or the late positive potential, a positive deflection in the 600-900 ms window are ERPs associated to the conflict resolution (Zhao et al., 2015). It is noteworthy that both ERPs scale with congruency, so that the N450 is more negative and the LPC is more positive for incongruent stimuli (Ergen et al., 2014). Moreover, with an adjusted form of the traditional Stroop task, Zhao and colleagues (2015) showed that the N450 is more involved in response conflict and that the LPC is more involved in stimulus conflict.
Time-frequency results of this study showed theta band power changes were associated with both types of conflict (Zhao et al., 2015). Interestingly, the N450 and the theta band change associated to response conflict resolution and the LPC and the theta band change associated to stimulus induced conflict showed similar time scales. Moreover, in accordance with previous studies the N450 and the theta band change for response resolution showed a similar scalp distribution. Both the N450 and the theta band effect have been previously localized to the anterior cingulate cortex (ACC; Hanslmayr et al., 2008; Liotti, Woldorff, Perez, & Mayberg, 2000). These findings suggest that the N450 and the theta band may have shared neural generators in the ACC. However, it is important to note that not all studies with a similar paradigm to Zhao and colleagues found exclusive modulation of the N450 of response induced conflict and modulation of the N450 by stimulus induced conflict has been reported (e.g. Killikelly & Szűcs, 2013).
After suggesting a possible relationship between theta oscillations and conflict resolution, another interesting venue of research may lie in timing and expectation. Temporal expectations have been found to affect early stages of information processing. Moreover, Doherty and colleagues (2005) found that when orthogonally manipulating spatial and temporal expectations, temporal expectations exerted the strongest effect. More recently, Rohenkohl & Nobre (2011) showed a relationship between alpha oscillatory activity and temporal expectations. The contingent negative variation (CNV) may be a component related to temporal expectations, which makes this a logical candidate to examine with respect to asymmetric amplitude modulations, the exact role of the CNV however is widely debated (Kononowicz & van Rijn, 2014).
General Conclusions & Discussion
The literature reviewed here examined whether evoked and induced responses are mechanistically different, or whether they both reflect the same underlying neural process but come about through different analysis procedures. The implications of the different models are wide reaching, because they pose different theses about the importance of ERPs (see Table 1). Although various attempts have been made, empirical evidence seems inconclusive about the relationship between ERPs and ongoing oscillations. Most pronounced discrepancies were found for working memory paradigms, where one study found correlational evidence for a relation between evoked and induced responses (van Dijk et al., 2010), whereas a more recent study found direct evidence that the amplitude asymmetry in the data was not related to a slow negative evoked response (Fukuda et al., 2015).
Amplitude asymmetry may be caused by other aspects picked up in the EEG signal, like a DC offset, but this explanation is not able to account for the presence of amplitude asymmetry as measured with the AFAindex in the study of Van Dijk et al. (2010). The
amplitude asymmetry in that study is most likely caused by a disbalance between forward and backward propagating currents (Mazaheri & Jensen, 2010a). When asymmetric oscillations are averaged, they will give rise to sustained responses (Mazaheri & Jensen, 2010a). Therefore, it seems premature to conclude that alpha asymmetry does not underlie sustained ERPs, especially because the relationship between alpha oscillations and sustained ERPs is studied even less systematic outside the field of working memory.
The studies reviewed here studying the relationship between the P300 and alpha oscillations found opposite results, where Yordanova et al. (2001) found that the P300 predicted the alpha ERD and that the P300 was expressed earlier in time than the alpha ERD, whereas Peng et al. (2012) found that alpha sources were driving the P300 sources when studied with Granger causality. Both studies found an intimate link between the P300 and alpha oscillations, but a systematic examination of the link between alpha oscillations and aspects of the P300, like the P3a or P3b, is lacking.
Finally, other slow evoked potentials and their relation to alpha oscillations were examined, but again the evidence is premature. Moreover, amplitude asymmetry may be present in other frequency bands than the alpha band, like the theta band, which may have a more intimate relationship to the N450 than previously assumed, because they seem to originate in the same region. However, more research investigating the relationship between induced and evoked responses is necessary, because understanding the link between these phenomena will broaden our understanding of neural functioning. ERPs have a long history in neuroscience and conceptually they are easily understood, giving them an important advantage over oscillations. Oscillations, on the other hand, have been related to many cognitive processes and are hypothesized to play an important role in communication between different regions (Fries, 2005). Understanding the relationship between both phenomena is therefore crucial.
Bastiaansen, M. C. M., Mazaheri, A., & Jensen, O. (2012). Beyond ERPs: Oscillatory
neuronal dynamics. In S. J. Luck & E. S. Kappenman (Eds.), The Oxford handbook of
event-related potential components (pp. 31–50). New York: Oxford University Press.
Becker, R., Ritter, P., & Villringer, A. (2008). Influence of ongoing alpha rhythm on the visual evoked potential. NeuroImage, 39(2), 707–716.
http://doi.org/10.1016/j.neuroimage.2007.09.016
Berger, H. (1929). Über das Elektroencephalogramm des Menschen. Archiv Psychiatrie Nerven.-Krankheith., 87, 527–570.
Bettencourt, K.C., & Xu, Y. (in press). Decoding the content of visual short-term memory under distraction in occipital and parietal areas.
Burgess, A. P. (2012). Towards a Unified Understanding of Event-Related Changes in the EEG: The Firefly Model of Synchronization through Cross-Frequency Phase Modulation.
PLoS ONE, 7(9), e45630. http://doi.org/10.1371/journal.pone.0045630
Burwell, S. J., Malone, S. M., Bernat, E. M., & Iacono, W. G. (2014). Does electroencephalogram phase variability account for reduced P3 brain potential in externalizing disorders? Clinical Neurophysiology, 125(10), 2007–2015.
http://doi.org/10.1016/j.clinph.2014.02.020
Cavanagh, J. F., & Frank, M. J. (2014). Frontal theta as a mechanism for cognitive control.
Trends in Cognitive Sciences, 18(8), 414–421. http://doi.org/10.1016/j.tics.2014.04.012
Cohen, M.X. (2014). Anaylzing neural time series data. MIT Press: Cambridga, Massachusetts.
Cong, F., Sipola, T., Huttunen-Scott, T., Xu, X., Ristaniemi, T., & Lyytinen, H. (2009). Hilbert-Huang versus Morlet wavelet transformation on mismatch negativity of children in
uninterrupted sound paradigm. Nonlinear Biomedical Physics, 3(1), 1. http://doi.org/10.1186/1753-4631-3-1
Doherty, J. R. (2005). Synergistic Effect of Combined Temporal and Spatial Expectations on Visual Attention. Journal of Neuroscience, 25(36), 8259–8266.
http://doi.org/10.1523/JNEUROSCI.1821-05.2005
Ergen, M., Saban, S., Kirmizi-Alsan, E., Uslu, A., Keskin-Ergen, Y., & Demiralp, T. (2014). Time–frequency analysis of the event-related potentials associated with the Stroop test.
International Journal of Psychophysiology, 94(3), 463–472.
http://doi.org/10.1016/j.ijpsycho.2014.08.177
Ester, E. F., Serences, J. T., & Awh, E. (2009). Spatially Global Representations in Human Primary Visual Cortex during Working Memory Maintenance. Journal of Neuroscience,
29(48), 15258–15265. http://doi.org/10.1523/JNEUROSCI.4388-09.2009
Feige, B., Kristeva-Feige, R., Rossi, S., Pizzella, V., & Rossini, P. M. (1996). Neuromagnetic study of movement-related changes in rhythmic brain activity. Brain Research, 734, 252– 260.
Fell, J., Dietl, T., Grunwald, T., Kurthen, M., Klaver, P., Trautner, P., … Fernández, G. (2004). Neural bases of cognitive ERPs: more than phase reset. Journal of Cognitive
Fries, P. (2005). A mechanism for cognitive dynamics: neuronal communication through neuronal coherence. Trends in Cognitive Sciences, 9(10), 474–480.
http://doi.org/10.1016/j.tics.2005.08.011
Fukuda, K., Mance, I., & Vogel, E. K. (2015). Power Modulation and Event-Related Slow Wave Provide Dissociable Correlates of Visual Working Memory. Journal of Neuroscience,
35(41), 14009–14016. http://doi.org/10.1523/JNEUROSCI.5003-14.2015
Gross, J. (2014). Analytical methods and experimental approaches for electrophysiological studies of brain oscillations. Journal of Neuroscience Methods, 228, 57–66.
http://doi.org/10.1016/j.jneumeth.2014.03.007
Grothe, I., & Plöchl, M. (2008). Amplitude Asymmetry: A Direct Link between Ongoing Oscillatory Activity and Event-Related Potentials? Journal of Neuroscience, 28(49), 13025– 13027. http://doi.org/10.1523/JNEUROSCI.4670-08.2008
Haegens, S., Cousijn, H., Wallis, G., Harrison, P. J., & Nobre, A. C. (2014). Inter- and intra-individual variability in alpha peak frequency. NeuroImage, 92, 46–55.
http://doi.org/10.1016/j.neuroimage.2014.01.049
Hämäläinen, M.S., Hari, R., Ilmoniemi, R.J., Knuutila, J., & Lounasmaa (1993).
Magnetoencephalography. Theory, instrumentation and applications to the noninvasive study of brain funciton. Rev Mod Phys, 65, 413-497.
Hanslmayr, S., Klimesch, W., Sauseng, P., Gruber, W., Doppelmayr, M., Freunberger, R., & Birbaumer, N. (2006). Alpha Phase Reset Contributes to the Generation of ERPs. Cerebral
Cortex, 17(1), 1–8. http://doi.org/10.1093/cercor/bhj129
Hanslmayr, S., Pastötter, B., Bäuml, K.-H., Gruber, S., Wimber, M., & Klimesch, W. (2008). The electrophysiological dynamics of interference during the Stroop task. Cognitive
Neuroscience, Journal of, 20(2), 215–225.
Huang, W. J., Chen, W. W., & Zhang, X. (2015). The neurophysiology of P 300–an integrated review. European Review for Medical and Pharmacological Sciences, 19(8), 1480–1488.
Humphrey, D.G., Kramer, A.F., & Stanny, R.R. (1994). Influence of extended wakefulness on automatic and nonautomatic processing. Hum Factors, 36, 4, 652-69.
Kahana, M.J., Sekuler, R., Caplan, J.B., Kirschen, M., & Madsen, J.R. (1999). Human theta oscillations exhibit task dependence during virtual maze navigation. Nature, 399, 781-784. Katayama, J., & Polich, J. (1998). Stimulus context determines P3a and P3b.
Psychophysiology, 35(1), 23–33.
Killikelly, C., & Szűcs, D. (2013). Asymmetry in stimulus and response conflict processing across the adult lifespan: ERP and EMG evidence. Cortex, 49(10), 2888–2903.
http://doi.org/10.1016/j.cortex.2013.08.017
Klimesch, W. (2012). Alpha-band oscillations, attention, and controlled access to stored information. Trends in Cognitive Sciences, 16(12), 606–617.
http://doi.org/10.1016/j.tics.2012.10.007
Klimesch, W., Doppelmayr, M., Schwaiger, J., Auinger, P., & Winkler, T. (1999). Paradoxical alpha synchronization in a memory task. Cognitive Brain Research, 7(4), 493–501.
