The cognitive neurosciences have had high frequency oscillations on the brain: so called “gamma-waves”, as recorded on the scalp, have been linked to working memory processes (via their interaction with slower “theta waves”), to cognitive insight, and even to consciousness. (I think there’s an unwritten rule that whenever someone mentions consciousness, they’ll be made to look foolish by a subsequent paper). In the midst of these “inflationary accounts” of the role of gamma oscillations, a debate has emerged: could these oscillations (at least, as recorded on the scalp) reflect simply the movements of the eyes, as now appears to largely be the case? More interestingly, is that necessarily incompatible with the more lofty interpretations of gamma? And what is the relationship between these movement-related artifacts on scalp recordings, and the well-established importance of gamma waves as recorded on or inside the brain?
(UPDATE: first para edited to emphasize scalp- vs. i-EEG distinction; thanks Alex & farraway).
At the center of this debate is the 2008 Neuron article by Yuval-Greenberg et al, met with intense discussion on the Neuron site, followed by commentary from Singer and colleagues, followed by another missive from the original authors, followed by additional articles from both groups, an editorial outlining 5 new papers on the topic in Brain Topography, and a general state of confusion about who to believe and what to make of the indisputably-bizarre phenomenon that is gamma.
Here’s a quick chronology of the confusion:
Yuval-Greenberg et al noted that spikes in gamma-band activity have a large amount of variability from trial to trial. When this activity was displayed on a topographic map that included eye channels, it became clear that much of the activity was centered around the eyes. They therefore replicated an experiment showing this effect while tracking the movements of the eyes with a video-based system in combination with a saccade detection algorithm. Interestingly, the correlation between the latency of the gamma-band power increase and the latency of the saccades were above .8 for each of three subjects (with p-values of less than 10 to the negative 27th power). Perhaps more disturbingly, no gamma band activity was observed on trials without any saccades whatsoever, indicating that the gamma band activity observed at the scalp might solely reflect saccade-related muscle activity. Using source analysis software, 98% of the observed saccade-aligned scalp activity could be explained by two electrical dipoles centered in the middle of each eye, even though the average saccade was quite small (most of less than one degree of visual angle – approximately the width of the index finger held at arm’s length).
Obviously, some people with more lofty interpretations of gamma weren’t happy with this demonstration. In a series of comments on the Neuron website, a reply by Yuval-Greenberg et al to Tallon-Baudry sticks out:
“In summary, Tallon-Baudry suggests that there are two separate responses: one related to microsaccades and the other to neural gamma oscillations. These two responses seem, however, to occur at the same latency (200-300 ms), the same distribution (occipito-parietal with nose reference), the same frequency spectrum (peak at 30-40 Hz with a broadband tail) and could be affected by the same cognitive manipulations (e.g. familiarity enhances both). While we cannot exclude such coincidence, […] our paper implies that studying the contribution of a neural component to [the gamma-band] requires at least the explicit consideration of small saccades and control of their effects. In our hands, this control seems to abolish the transient [gamma-band] signal. ”
Then, much more recently, Melloni et al replied to point out that a number of shortcomings complicate Yuval-Greenberg et al’s claims:
1) Yuval-Greenberg referenced their EEG data to the voltage recorded at the nose, which is known to be prone to oculomotor artifacts (perhaps exaggerating the influence of saccades, and strengthening their case)
2) Correlation does not mean causation – just because gamma-band activity coincides with saccades does not mean that both microsaccades and gamma-band activity might index some underlying cognitive process of interest.
3) The spike potential generated by saccades is broadband in nature, but many studies report changes in particular sub-bands – a finding that’s incompatible with the claim that all gamma activity reflects is these microsaccades.
4) By aligning the electrical activity to the onset of saccades, Yuval-Greenberg might have “averaged out” gamma-band variance timelocked to some other cognitive event. Basically, this means it was an unfair test.
Yuval-Greenberg replied to say that many studies fail to control for saccade related activity, and that most published studies fail to show data from the electrodes positioned near the eyes, raising the possibility that many of the posterior gamma-band effects have their source in a dipole whose other end projects to the eye channels. They also note that the use of electrodes on the nose as a reference is common, and that the use of other references does not resolve the underlying issue: much of what is reported as cortical gamma could reflect a peripheral source in the eyes, and that while not all findings on gamma are suspect, many are.
The next reply by Melloni et al occurs in a newTiCS article, and Melloni et al now take a different tack. They first develop a theory about saccades and their role in cognitive processing; given that saccades punctuate the flow of visual information into the brain, they argue that the brain likely has a mechanism for maximizing the extraction of information between saccades while minimizing the extraction of information during saccades. While the latter mechanism is well known (“saccadic suppression“), there has been little attention to the former – so if gamma reliably follows microsaccades, then it could reflect exactly this type of process. In addition, this complicates the exclusion of trials with microsaccades as these are precisely those trials which should have the most observable gamma.
One might think that math can save the day – for example, by using spatial, temporal or spectral independent components analysis to separate the muscle-related activity from the cortical activity. An interesting paper in press at Brain Topography shows that ICA is no panacea – by separating groups into those with high and low muscle-related activity and performing ICA-based correction, Shackman et al gamma band power remained higher for the high muscle-related activity group. Of course, by Melloni et al’s account, this increase in power might reflect the underlying trait that puts those subjects in the high muscle-related activity group in the first place.
But there is one sure-fire panacea: paralyze your subjects. Also as reported in Brain Topography, Pope et al administered the neuromuscular blocker cisatracurium to show that even in the absence of muscle movement, the EEG may be contaminated by sustained muscle tension (eliminated only with paralysis.) Of course, this kind of study requires the use of ventilators, because the subjects cannot breathe on their own. Such an experience may itself change the way people are cognitively processing their environment, and so may not be a great technique for investigating cognitive processing. Nonetheless, the Pope et al results show that in the absence of muscle activity, gamma band activity may in some cases become more apparent.
My take on all of this is that Yuval-Greenberg et al are clearly correct – saccade-related activity is a big problem for time-frequency analysis and needs to be considered more centrally both in the reporting of time-frequency results and their interpretation. Melloni et al are also right in suggesting that saccade-related activity is not necessarily trivial, and that there are likely a whole host of important cognitive processes that precede, coincide, and follow microsaccades – any one of which could be important for understanding the role of gamma. Finally, I think Shackman et al and Pope et al take the most pragmatic approach, which is to assess current techniques for their ability to cleanly separate the different sources of variance in scalp EEG (even intracranial EEG is affected by saccades, although to a much lesser extent than scalp EEG). Until better techniques are developed, or until some combination of existing techniques is definitively shown to cleanly separate muscle-related from brain-related activity, lofty interpretations of scalp-recorded gamma in terms of consciousness (etc) are very clearly premature.