|Home | About | Journals | Submit | Contact Us | Français|
In order to understand the working brain as a network, it is essential to identify the mechanisms by which information is gated between regions. We here propose that information is gated by inhibiting task-irrelevant regions, thus routing information to task-relevant regions. The functional inhibition is reflected in oscillatory activity in the alpha band (8–13Hz). From a physiological perspective the alpha activity provides pulsed inhibition reducing the processing capabilities of a given area. Active processing in the engaged areas is reflected by neuronal synchronization in the gamma band (30–100Hz) accompanied by an alpha band decrease. According to this framework the brain could be studied as a network by investigating cross-frequency interactions between gamma and alpha activity. Specifically the framework predicts that optimal task performance will correlate with alpha activity in task-irrelevant areas. In this review we will discuss the empirical support for this framework. Given that alpha activity is by far the strongest signal recorded by EEG and MEG, we propose that a major part of the electrophysiological activity detected from the working brain reflects gating by inhibition.
The human brain is a complex processing device which rapidly adapts to the environment. This requires fast changes in the functional architecture of the brain network which must rely on dynamic interactions. Network questions are often addressed by asking how different regions become functionally coupled depending on the task at hand (Friston et al., 1997). This can be conceptualized as the question of how information is gated from a sending region to one of two receiving regions. This could be achieved by fast changes in synaptic strength, for instance due to changes in presynaptic vesicle release (Figure (Figure1A).1A). It has also been proposed that functional connectivity between task-relevant regions is established by oscillatory phase coupling between the sending and receiving region (Bressler, 1995; Varela et al., 2001; Fries, 2005) (Figure (Figure1B).1B). We here promote the idea that information is routed by functionally blocking off the task-irrelevant pathways: gating by inhibition (Figure (Figure1C).1C). Importantly, this inhibition is reflected by oscillatory activity in the alpha band. This functional inhibition might be a consequence of GABAergic inhibition provided by the interneuronal network. We will later address this issue in detail. Nevertheless,we propose that the alpha activity is a consequence pulsed inhibition which reduces the processing capabilities of a given area.
Oscillatory activity in the alpha and beta bands was first discovered in the late 1920s by Hans Berger and is by far the strongest electrophysiological signal measured from the human waking brain (Berger, 1929; da Silva et al., 1973; Niedermeyer and Lopes da Silva, 1999). Alpha activity has for a long period been interpreted to reflect “cortical idling” (Pfurtscheller et al., 1996). This interpretation was based on the finding that alpha activity increases when subjects are awake but not engaged in any task. Recently strong evidence has accumulated against the idling hypothesis and it is becoming clear that the functional role of alpha activity has been highly underappreciated.
The recent insight into the functional role of alpha activity has been brought about by high-density EEG and MEG recordings. The improved spatial resolution of these techniques has facilitated the functional interpretation of the measured activity. One strong piece of evidence demonstrating that alpha activity cannot reflect idling, stems from working memory experiments in which subjects were presented items to be remembered for a shorter period. Using these types of paradigms several studies have now demonstrated a robust increase in alpha power with memory load during the retention interval (Jensen et al., 2002; Tuladhar et al., 2007; Scheeringa et al., 2009). The alpha increase has been suggested to either reflect active processing related to memory maintenance (Palva and Palva, 2007; Palva et al., 2007) or inhibition of posterior regions not required for the task (Klimesch et al., 2007).
In order to conclusively determine if the alpha activity is reflecting functional inhibition, experiments are required which can tell whether the alpha activity is related to the engagement or disengagement of specific brain regions. For instance it has been demonstrated that alpha activity over visual areas increases in motor tasks and vice versa (Pfurtscheller, 1992). Studies of spatial attention have shown that when covert attention is directed to one hemifield (e.g., the left), then alpha decreases in the contralateral (right) but increases in the ipsilateral (left) hemisphere (Worden et al., 2000; Thut et al., 2006; Rihs et al., 2007; Kelly et al., 2009; van Gerven and Jensen, 2009; Handel et al., 2010). Interestingly, this hemispheric alpha lateralization had consequences for detection performance even when it was attenuated by moving-dot kinematograms (Handel et al., 2010). The modulation of the posterior alpha activity with covert attention is so robust even in single trials that it can be used as a control signal for brain–computer interfaces (van Gerven and Jensen, 2009; Bahramisharif et al., 2010). A similar pattern was observed in working memory tasks in which subjects were presented visual stimuli to be remembered in either the left or right visual hemifield (Van Der Werf et al., 2008; Grimault et al., 2009; Sauseng et al., 2009). In a cross-modal EEG study subjects were asked to attend either to visual or auditory stimuli (Fu et al., 2001). The posterior alpha activity increased only when subjects attended to the auditory stimuli. The suppressive nature of alpha is not only present in visuo-spatial paradigms. In a recent MEG study subjects were asked to maintain either the identity or orientation of presented faces (Jokisch and Jensen, 2007). A clear increase in alpha activity over the dorsal stream was observed when the ventral stream was selectively engaged by the face identity task (Figure (Figure2).2). In a somatosensory working memory task, the alpha activity decreased in the primary sensorimotor cortex contralateral to the engaged hand while it increased in the ipsilateral hemisphere (Haegens et al., 2010). The functional role of alpha activity in defining the brain state has also been observed in auditory studies. A recent study found that alpha power at parieto-occipital sites was lateralized by the direction of auditory attention to continuous speech (Kerlin et al., 2010). The alpha lateralization might serve to engage the processing of visual input associated with auditory input from the same hemifield. In a working memory study on maintaining pitches, alpha activity from left superior temporal areas increased during the retention interval (van Dijk et al., 2010). This temporal alpha activity, possibly being the tau-rhythm (Lehtela et al., 1997), likely reflected inhibition of the left auditory cortex to allocate resources to the right auditory cortex involved in pitch processing. In sum, these studies make the case that alpha activity decreases in engaged regions, while it increases in disengaged regions.
In some of our recent studies we have been able to demonstrate cases where the increase in alpha activity in task-irrelevant areas correlates with task performance. This suggests that alpha activity in task-irrelevant regions is required for optimal performance. Such an effect was first shown in a somatosensory working memory task in which stimuli were delivered to the right hand (Haegens et al., 2010). This resulted in alpha power increases over the task-irrelevant right somatosensory areas during the retention interval (Figure (Figure3).3). Furthermore, this increase correlated with performance, i.e., more errors were made when the right hemisphere alpha activity was not sufficiently strong. Interestingly, the alpha decrease in the left hemisphere did not correlate with performance; nor did the left hemisphere gamma increase. This effect is not specific to working memory tasks. Recently we conducted a long-term memory experiment in which subjects were asked to learn sets of lists three words long. We then characterized the brain activity during encoding with respect to lists that were later remembered and forgotten. We found a dramatic increase in alpha activity over occipital regions for later remembered lists (Meeuwissen et al., in press). Interestingly, it was possible to predict if a given list was later remembered on a single trial basis from the alpha activity alone. Given that the occipital regions are not required for the task, we suggest that successful long-term memory encoding is improved by active inhibition of occipital regions. Importantly, the strongest subsequent memory effect was found in the alpha band when compared to effects in the gamma band. These two studies show that performance is impaired if functional inhibition reflected by alpha band activity fails in task-irrelevant regions. This is consistent with the notion that optimal task performance is dependent on shutting down task-irrelevant regions in order to allocate resources to talk-relevant regions.
It is becoming increasingly clear that ongoing oscillatory activity prior to an event has a strong impact on subsequent processing. Indeed it has been demonstrated that posterior pre-stimulus alpha power modulates both perception of visual and somatosensory stimuli (Linkenkaer-Hansen et al., 2004; Babiloni et al., 2006; Hanslmayr et al., 2007; van Dijk et al., 2008; Zhang et al., 2008; Kelly et al., 2009; Mazaheri et al., 2009; Schubert et al., 2009). These findings extend beyond perceptual tasks. Recently, we demonstrated that errors made in a Go–noGo task were predicted by posterior and central alpha activity (Mazaheri et al., 2009). Interestingly, after an error was made, the posterior alpha activity was reduced as if to prevent subsequent errors. This is related to a recent study in which visuo-motor performance has been shown to be predicted by alpha activity in posterior and sensorimotor regions (Del Percio et al., 2009). Studies combining TMS and EEG have confirmed the functional inhibitory role of pre-stimulus oscillatory alpha activity (Thut and Miniussi, 2009). For instance, the group of Thut et al. has demonstrated that phosphenes evoked by TMS pulses are depressed by strong alpha activity (Romei et al., 2008; Thut and Miniussi, 2009).
While the studies mentioned above suggest a purely inhibitory role of the alpha activity, there are also studies pointing a parabolic (“inverted-U”) relationship. Linkenkaer-Hansen et al. (2004) demonstrated that weak somatosensory stimuli were best detected at medium levels of the 10Hz sensorimotor rhythm (but see Schubert et al., 2009). This finding was later complemented by the demonstration that the magnitude of the somatosensory N1 also is related to the 10Hz sensorimotor rhythm according to a parabolic relationship (Zhang and Ding, 2010). A similar relationship has been demonstrated between the posterior alpha activity and visual evoked responses (Rajagovindan and Ding, 2010). In sum, these studies suggest that pre-stimulus alpha activity is likely to be involved in shaping the functional architecture of the brain network when preparing for upcoming processing.
It has been proposed that oscillatory alpha activity operates in a phasic manner (Varela et al., 1981; VanRullen and Koch, 2003), and it has been demonstrated that the phase of pre-stimulus alpha oscillations modulates visual detection (Busch et al., 2009; Mathewson et al., 2009). On certain phases of the alpha cycle visual detection ability was high, whereas it was lower at other phases. Taken together with the inhibitory role of the alpha activity, this supports the notion of “pulsed inhibition,” namely that the alpha activity produces bouts of inhibition repeated every 100ms (Mazaheri and Jensen, 2010). This is consistent with a recent studies demonstrating that the gamma activity is modulated by alpha phase (Osipova et al., 2008; Voytek et al., 2010). This led us to propose that an increase in alpha activity is a consequence of an increase in the magnitude of inhibitory bouts, which serve to break the ongoing gamma activity. While it remains to be tested, it is possible that rhythmic GABAergic input from the interneuronal network is a key mechanism for producing the pulsed inhibition. The stronger the alpha activity, the shorter the “duty-cycle” of gamma activity (Figure (Figure4).4). By this principle the alpha activity can provide graded inhibition by blocking the ongoing gamma activity in a phasic manner: If the alpha activity is strong, the time-window of processing is short (the duty-cycle). As alpha activity decreases in magnitude, the time-windows become longer.
How might the alpha activity serve to provide pulsed inhibition from a physiological perspective? GABAergic feedback from interneurons has been strongly implicated in the physiological mechanism generating the alpha rhythm (Jones et al., 2000; Lorincz et al., 2009). It should be mentioned that GABAergic interneurons also have been strongly implicated in the generation of the much faster 30–100Hz gamma rhythm (Traub et al., 1999; Bartos et al., 2007; Mann and Paulsen, 2007; Cardin et al., 2009; Sohal et al., 2009). However, it is conceivable that GABAergic interneurons can be involved in producing rhythms at different temporal scales. For instance, the mechanism for the alpha activity proposed by Jones et al. (2000) relies amongst others on low-threshold T-type voltage-gated calcium channels. This type of channel produces a current with a dynamics slow enough to support the alpha generation. Interestingly, a class of somatostatin-expressing inhibitory interneurons have been identified which can produce persistent oscillations at 3–10Hz. These cells were identified in layer 2/3 of the mouse somatosensory cortex (Fanselow et al., 2008). The GABAergic feedback could serve to directly silence processing in pyramidal neurons or reducing the efficacy of excitatory input by shunting inhibition (Mann and Paulsen, 2007). While mechanism have been identified for how oscillatory activity in the alpha band can be produced alone, the alpha rhythm is likely to be modulated by thalamic rhythm generators (da Silva et al., 1973; Suffczynski et al., 2001; Hughes and Crunelli, 2005; Jones et al., 2009; Lorincz et al., 2009). This drive could be implemented through phasic inhibition.
It should be mentioned that pulsed inhibition is consistent with the notion of “amplitude asymmetry.” Amplitude asymmetry refers non-sinusoidal properties of the ongoing oscillatory activity resulting in the mean of the signal being biased by its magnitude (Nikulin et al., 2007; Mazaheri and Jensen, 2008; Mazaheri and Jensen, 2010). This is a consequence of the peaks of the oscillations being modulated over time rather than the troughs, i.e., it can be conceived of as rhythmic pulsing. Importantly, systematic modulation of the alpha activity can create slow event-related responses. It was recently demonstrated that such a mechanism can account for slow visual evoked responses and the sustained evoked responses observed in a working memory task (Mazaheri and Jensen, 2008; van Dijk et al., 2010). Thus the notion of pulsed inhibition holds the promise of providing insight into the physiological substrate of certain types of slow evoked responses. This is extensively discussed in the review (Mazaheri and Jensen, 2010).
Numerous studies, including research on attention, working memory, long-term memory, language and motor tasks suggest that activity in the gamma range reflects engagement/processing. This was shown by intracranial, EEG and MEG recordings in animals and humans (Paller et al., 1987; Kaiser and Lutzenberger, 2005; Crone et al., 2006; Fries et al., 2007; Jensen et al., 2007; Womelsdorf and Fries, 2007; Jerbi et al., 2009). Typically gamma increases are accompanied by an alpha decrease. Since there is a strong consensus on this point, we will here only provide one of the key argument for why neuronal synchronization in the gamma band is bound to be involved in neuronal communication (Konig et al., 1996; Fries et al., 2007). Consider a group of excitatory neurons projecting to a down-stream area. Given that the time course of excitatory post-synaptic events is 10–15ms, these neurons will have a strong impact on the down-stream area if they synchronize at a rate of 60–100Hz. This implies that excitatory neuronal firing synchronized in the gamma band will be communicated to other regions. Note, that while GABAergic inhibitory interneurons are likely to be involved in setting up the gamma band synchronization, this does not equate functional inhibition. Rather, gamma band activity seems to reflect neuronal processing and communication between regions.
In order to study the working brain as a network it is crucial to understand not only how the task-relevant regions are engaged but also how the task-irrelevant regions are inhibited. Since these processes are reflected in different frequency bands, cross-frequency interactions must be considered. There are several principles for cross-frequency interactions (Jensen and Colgin, 2007). On possibility is phase-to-phase coupling (Palva et al., 2005), e.g., the phase of gamma activity in one regions is coupled to the phase of the alpha activity in another (n:m coupling, e.g., four gamma cycle per alpha cycle). Another possibility is phase-to-power coupling. For instance, it was found that the gamma activity in CA1 is coupled to the phase of the theta activity in the entorhinal cortex (Colgin et al., 2009). Finally, power-to-power coupling remains a viable possibility (Bruns et al., 2000). Power-to-power correlations might reflect coupling over longer distances where the timing accuracy required for phase coupling is difficult to maintain. It should be mentioned that these measures of cross-frequency couplings are not mutually exclusive, e.g., a phase-to-power correlations might be reflected in power-to-power correlations as well. Recent MEG studies have demonstrated robust cross-frequency power-to-power coupling. One example showed that occipital gamma activity correlates negatively on a trial-by-trial basis with beta suppression in motor cortical areas during a mental rotation task of hands (de Lange et al., 2008). This functional coupling was suggested to reflect the interaction between motor cortical and visual areas while the rotation task was performed. Another example showed that frontal theta activity increased while posterior alpha activity was suppressed after button press responses in Go–noGo tasks (Mazaheri et al., 2009). Importantly the trial-by-trial anti-correlation between theta and alpha power was greatest after erroneous button press responses. This could reflect a top-down signal engaging the visual areas stronger in order to prevent subsequent errors.
In future work it will be important to apply these measures of cross-frequency interactions between gamma and alpha activity. In particular we predict a positive correlation between gamma power in task-relevant regions and alpha power in task-irrelevant regions. This correlation is expected to increase with task performance. Given the notion of pulsed inhibition we would also expect that alpha phase in one region will correlate with gamma power in another region. For instance, when a region is inhibited by alpha activity some processing can occur between the pulses of inhibition. This processing would have impact on down-stream regions and might be detected as gamma activity in the down-stream region phasically modulated by the alpha activity in sensory regions.
Multiple studies combining fMRI and EEG have found that the BOLD signal correlates negatively with alpha activity (Goldman et al., 2002; Laufs et al., 2003; Feige et al., 2005; Scheeringa et al., 2009) whereas other studies have demonstrated a positive correlation between the BOLD signal and gamma activity (Logothetis et al., 2001; Niessing et al., 2005). These findings are consistent with the notion that alpha activity reflects functional inhibition. While most fMRI studies on cognitive processing focus on task-relevant regions, it would be interesting to investigate if BOLD decreases in task-irrelevant regions also are predictive of performance. This has been found in some studies. For instance, BOLD decreases in an extended network was shown to be predictive of memory formation (Daselaar et al., 2004). We propose to conduct further neuroimaging studies to investigate to what extend the deactivation of task-irrelevant regions reflected in a BOLD decrease correlates with performance. Are such decreases likewise correlated with an alpha increase? If it turns out to be a general phenomenon, it would suggest that our framework of gating by inhibition could be applied in fMRI studies as well.
We have presented a working hypothesis postulating that information is gated through the brain by functional inhibition of task-irrelevant areas. Further, the functional inhibition is reflected by oscillatory activity in the alpha band. Our hypothesis provides a general framework for understanding how the functional architecture of the brain is shaped. Our hypothesis provides a putative explanation for why the alpha activity is the strongest electrophysiological signal measured from the human brain: optimal performance requires an effective inhibition of task-irrelevant regions. While several studies support this framework more empirical work is required in order to substantiate the principle of gating by inhibition.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
This research was supported by The Netherlands Organization for Scientific Research (NWO): Innovational Research Incentive Schemes, VICI grant number: 453-09-002; ALW Open Competition, grant number 817.02.010; and a Rubicon grant.