Diego Lozano-Soldevilla’s research while affiliated with French National Centre for Scientific Research and other places

Alpha oscillations play a special role in vision. During sensory processing, reverse-correlation techniques revealed that white-noise luminance sequences elicit a robust occipital ∼10 Hz response that periodically reverberates the input sequence for up to 1 s. These perceptual echoes constitute the impulse response function of the visual system. However, the spatial dimension of perceptual echoes remains unknown: do they reverberate across the cortex simultaneously? Does stimulation over multiple visual coordinates evoke multiple synchronized echoes, or do they show consistent phase differences? Here, we tested the spatial dimension of perceptual echoes in two electroencephalogram (EEG) experiments manipulating the location of the visual stimulation. When a single disc flickered a white-noise luminance sequence in the upper visual field, we observed a single “echo wave” originating in posterior sensors and spatially propagating towards frontal ones (i.e. periodic travelling wave). The presentation of two independent flickering discs in separate visual hemifields produced two simultaneous and superimposed echo waves propagating in opposite directions, one in response to each stimulus. Strikingly, at many electrode sites, the phase of the two echoes differed, with a phase advance for the contralateral stimulus location. EEG source reconstruction tentatively located the waves within contralateral parieto-occipital cortex. In conclusion, the alpha rhythm processes stimulus information as a travelling wave that propagates across the cortical representation of retinotopic space in the human brain. In line with the “cortical scanning” hypothesis (Pitts & McCulloch, 1947), these results suggest the existence of an additional spatial dimension embedded in the phase of the alpha rhythm. Significance statement How does the spatial dimension of sensory processing relate to the temporal dimension of brain rhythms? Using correlation techniques, we characterized perceptual echoes, the average electroencephalogram response induced by visual stimuli that change luminance randomly. We found that perceptual echoes are actually periodic waves that travel through human visual cortex. Strikingly these periodic waves show consistent phase differences across the visual field, processing screen locations sequentially across distinct phases of the cycle following basic retinotopy. These results suggest the existence of an additional “hidden” spatial dimension in sensory cortex, encoded in the phase of the alpha oscillatory cycle. This could mean that perceptual echoes behave like sweeps of a sonar, processing the visual field in cycles of ∼100 ms duration.

Visual information is encoded periodically at alpha frequency (~10Hz): for example, a robust alpha component was found by cross-correlating visually presented white-noise (random) luminance fluctuations with corresponding EEG responses over posterior sensors (VanRullen & Macdonald, 2012). This oscillatory impulse response function can be interpreted as a ~10Hz perceptual echo that reverberates the input sequence periodically. Here, we explored the spatial dimension of these perceptual echoes. Two independent random luminance white-noise sequences were simultaneously displayed in two discs on the left and right of fixation. For each EEG channel, two echo functions could thus be computed, by cross-correlating each random luminance sequence (left and right discs) with the simultaneously acquired EEG activity, and subsequently averaging these location-specific visual responses across trials. Multiple posterior sites gave rise to two sizeable echo functions. Interestingly, the echo phase for a given screen location was about 10-12ms earlier on contralateral than ipsilateral electrodes. This travelling wave propagating across the scalp was highly consistent across participants (n=10). Likewise, for each posterior sensor, we found systematic phase differences between locations, such that the echo in response to the ipsilateral disc always lagged by 10–12ms relative to the echo in response to the contralateral disc. In other words, echo functions also behaved like travelling waves across the visual field, sequentially propagating from contra- to ipsi-lateral screen locations. This reveals that occipital cortex, beyond its standard encoding of retinotopic spatial dimensions, also encodes at least one more spatial dimension –but in the temporal domain, i.e. in the phase of alpha reverberations. These results constitute the first direct experimental evidence for Pitts & McCulloch's (1947) scanning hypothesis: "the alpha rhythm performs a temporal 'scanning' of the cortex which thereby gains, at the cost of time, the equivalent of another spatial dimension in its neural manifold". Meeting abstract presented at VSS 2016

LZP modulations in the alpha (8–12 Hz) and beta (15–30 Hz) bands during rest. (A) Grand mean topographic representation of alpha power during EO (left) and EC (right). Color code represents relative power changes with respect to the mean of the power spectrum (5–40 Hz). The rows reflect the drug levels: Placebo, 0.5 and 1.5 mg LZP. Sensors marked in bold display statistical significant drug main effect (repeated-measures ANOVA) using a cluster-based nonparametric permutation test. (B) Grand average of power spectra [sensors marked white in (A)] estimated for the EO (top) and EC (bottom) and all drug sessions. Inserts show a repeated-measures ANOVA confirming that the occipital alpha power decreased with LZP. Error bars show the SEM. (C) Occipito-parietal sources reflected the alpha power decrease under 1.5 mg LZP dosage relative to placebo (EO and EC averaged). (D) Gran mean topographic representation of beta power during EO (left) and EC (right) during rest [same conventions as in (A)]. (E) Grand average of power spectra [sensors marked white (D)] estimated for the EO (top) and EC (bottom) and all drug sessions. Inserts showed a repeated-measures ANOVA confirming that the sensorimotor beta power increased with LZP dosage; same conventions as in Figure 1B. Error bars show the SEM. (F) The sources in the sensorimotor cortex reflected the beta power increase under 1.5 mg LZP dosage relative to placebo (EO and EC averaged) [same conventions as in (C)]. The peak of the beta source lied out of the brain due to inaccuracies after the individual head model normalization to the MNI space. †p < 1 × 10−6; ***p < 1 × 10−5; **p < 0.001; *p < 0.05.

Neuronal oscillations support cognitive processing. Modern views suggest that neuronal oscillations do not only reflect coordinated activity in spatially distributed networks, but also that there is interaction between the oscillations at different frequencies. For example, invasive recordings in animals and humans have found that the amplitude of fast oscillations (>40 Hz) occur non-uniformly within the phase of slower oscillations, forming the so-called cross-frequency coupling (CFC). However, the CFC patterns might be influenced by features in the signal that do not relate to underlying physiological interactions. For example, CFC estimates may be sensitive to spectral correlations due to non-sinusoidal properties of the alpha band wave morphology. To investigate this issue, we performed CFC analysis using experimental and synthetic data. The former consisted in a double-blind magnetoencephalography pharmacological study in which participants received either placebo, 0.5 or 1.5 mg of lorazepam (LZP; GABAergic enhancer) in different experimental sessions. By recording oscillatory brain activity with during rest and working memory (WM), we were able to demonstrate that posterior alpha (8–12 Hz) phase was coupled to beta-low gamma band (20–45 Hz) amplitude envelope during all sessions. Importantly, bicoherence values around the harmonics of the alpha frequency were similar both in magnitude and topographic distribution to the cross-frequency coherence (CFCoh) values observed in the alpha-phase to beta-low gamma coupling. In addition, despite the large CFCoh we found no significant cross-frequency directionality (CFD). Critically, simulations demonstrated that a sizable part of our empirical CFCoh between alpha and beta-low gamma coupling and the lack of CFD could be explained by two-three harmonics aligned in zero phase-lag produced by the physiologically characteristic alpha asymmetry in the amplitude of the peaks relative to the troughs. Furthermore, we showed that periodic signals whose waveform deviate from pure sine waves produce non-zero CFCoh with predictable CFD. Our results reveal the important role of the non-sinusoidal wave morphology on state of the art CFC metrics and we recommend caution with strong physiological interpretations of CFC and suggest basic data quality checks to enhance the mechanistic understanding of CFC.

LZP did not modulate alpha-phase to beta-low gamma amplitude coupling during WM delay. (A) Delayed match to sample WM task. Participants had to keep fixations while covertly attend to the cued visual hemifield to encode the target items. The task consisted to compare the cued items, compare them with a probe and decide whether they matched or not. (B) Grand mean topographic representation of alpha-phase to beta (16–22 Hz; left) and to beta-low gamma amplitude coupling (25–45 Hz; right) during WM delay (0.5 to 1.5 s). Color code represents magnitude squared coherence. Note that topographies have different scale. The strongest coherence values appeared over occipito-parietal sensors. Grand mean frequency-by-frequency coherence comodulogram (middle) of the occipital axial gradiometers of interest as in Figure S2A. Black rectangles indicate the x-y frequency selections used in the statistical analysis and topographic representations (8–12 Hz; 16–22 Hz; 25–45 Hz). Comodulograms have the same coherence scale as alpha—beta topography (left). (C,D) are the same as (B) but for 0.5 and 1.5 mg LZP, respectively.

LZP did not significantly change CFD during EC. (A) Grand mean topographic representation of the CFD (8–12 Hz; 20–45 Hz) for the different drug conditions. White sensors were significantly modulated by LZP in Figure 1. The color code indicates the normalized Ψ~ weighted average of the CFD slope (Ψ = Ψ~/std(Ψ~)). (B) Grand mean normalized CFD index averaged over the occipital sensors marked in topographies (A) for each drug session. Note the low sigma values we obtained (± ~0.3; ± 2 was considered significant). (C) CFD weighted average (Ψ~) and (D) standard deviation of the slope (std(Ψ~)) averaged over the occipital sensors marked in topographies for each drug session.

LZP did not change CFD during WM delay. (A) Grand mean topographic representation of alpha-beta CFD (16–22 Hz) during WM delay for each drug separately. Color code indicates the normalized Ψ~ weighted average of the CFD slope [Ψ = Ψ~/std(Ψ~); color scale same as in (B)]. (B) Grand mean normalized CFD index taken over the occipital sensors of interest chosen from Figure S2A. Note the low sigma values we obtained (~ ±0.6; ±2 is considered significant). (C) CFD weighted average (Ψ~) and (D) standard deviation of the slope (std(Ψ~)) taken over the same occipital sensors for each drug session respectively. Drug sessions are displayed in different columns.

Beta-low gamma peak-triggered average during rest. (A) Grand mean of event-related fields locked to the beta-low gamma (20–45 Hz) peaks. MEG axial gradiometers from left and right hemispheres [marked white in (B)] during EO for all drug sessions. Vertical dashed lines at ± 0.05 s indicates the lower limit of the bandpass filter used for peak detection (1/0.05 s = 20 Hz). Beyond the ± 0.05 s limit, the underlying alpha oscillations were highly reduced if any. (B) Grand mean topographic representation of beta-low gamma peak-triggered average taken around the peak period (–0.015 to 0.015 s). Marked bold sensors of interest based on Figure S2A. Panels (C,D) are the same as (A,B) but for EC, respectively with same figure conventions.