Low-frequency alternating current stimulation rhythmically suppresses stimulus-induced gamma-band oscillations in visual cortex and impairs perceptual performance

Alpha oscillations (8-12 Hz) are hypothesized to rhythmically gate sensory processing, reflected by activity in the 40-100 Hz gamma band, via the mechanism of pulsed inhibition. We applied transcranial alternating current stimulation (TACS) at individual alpha frequency (IAF) and flanking frequencies (IAF-4 Hz, IAF+4 Hz) to the occipital cortex of healthy human volunteers during concurrent magnetoencephalography (MEG), while participants performed a visual detection task inducing strong gamma-band responses. Occipital (but not frontal) TACS phasically suppressed stimulus-induced gamma oscillations in the visual cortex and impaired target detection, with stronger phase-to-amplitude coupling predicting behavioral impairments. Frontal control TACS ruled out retino-thalamo-cortical entrainment resulting from (subthreshold) retinal stimulation. All TACS frequencies tested were effective, suggesting that visual gamma-band responses can be modulated by a range of low frequency oscillations. We propose that TACS-induced cortical excitability fluctuations mimic the mechanism of pulsed inhibition, which mediates the function of alpha oscillations in gating sensory processing.


37
Primarily, we aimed to test the specific hypothesis that the well-described stimulus-induced increase in 38 gamma-band power in the visual cortex, associated with bottom-up visual processing (Bastos et al., 2015; 39 Fries, 2015), can be actively modulated by the phase of slower oscillations, particularly in the alpha band. 40 While correlational data from MEG studies in humans (Osipova et al., 2008) and intralaminar recordings 41 in monkeys (Spaak et al., 2012) has revealed coupling between alpha phase and gamma amplitude, the 42 causal role of alpha oscillations in modulating gamma-band power remains unresolved. We therefore 43 applied transcranial alternating current stimulation (TACS) at individual alpha frequency (IAF) to the 44 visual cortex (Oz-Cz montage) in human volunteers performing a visual detection task to mimic the 45 impact of alpha phase-related cortical excitability fluctuations on endogenous gamma activity during 46 visual stimulus processing. A second goal of this study was to test (i) whether TACS is capable of 47 modulating behaviorally relevant neuronal activity in the human brain at commonly used stimulation 48 intensities, an assumption recently called into question by modelling work (Opitz et al., 2016) and 49 cadaver studies (Underwood, 2016), and (ii) whether its effect can be attributed to transcranial as opposed 50 to mere retinal stimulation (Schutter, 2015). While simultaneous TACS-EEG recordings (Helfrich et al.,

69
Participants performed a forced-choice visual discrimination task in which they had to report the rotation 70 direction of a foveally presented asterisk inside an inward moving high-contrast grating (Figure 1A in the center of the screen and were allowed to blink, until 1500 ms later the white dot turned grey to indicate the 81 end of the blink period. At 3500 ms an inward-moving grating appeared around the fixation dot, which contained a 82 slowly rotating asterisk in its center. Participants had to report the direction of rotation, by button-press, as soon as 83 the visual stimulus disappeared at 5900 ms and before the next trial started at 6900 ms. TACS was turned on 500 ms 84 into the blink period and turned off 2400 ms after visual stimulus onset and 1000 ms before visual stimulus offset,

134
Occipital TACS phase rhythmically modulated gamma power 135 To test whether the net suppression actually resulted from a phasic modulation, or more specifically, a 136 rhythmic suppression of gamma-band power by TACS phase, we calculated TACS peak-locked TFRs 137 ( Figure 2B; Figure S3). Visually induced gamma-band power was observed to decrease at particular 138 points in the phase cycle, but not to increase at any point (pFDR < 0.05). In addition, there was a rhythmic 139 increase around 40 Hz, as well as between 70 -100 Hz for occipital, but not for frontal TACS between -140 pi and 0 (pFDR < 0.05). Although activity in the 40 Hz range is also visible in the non-peak-locked TFRs 141 (Figure 2A

301
Participants performed a rotation detection task in which they had to indicate by button-press the rotation 302 direction of an asterisk in the center of an inward moving high-contrast grating ( Figure 1A). Participants 303 were instructed to fixate a white dot on grey background in the center of the screen. Each trial started with 304 a 1.5 s period in which participants were allowed to blink. Participants were asked to refrain from 305 blinking as soon as the fixation dot turned grey. A 'baseline' period of two seconds followed after which 306 an inward-moving (0.8 degree/second) black-and-white high contrast grating with concentric circles (2.5 307 cycles/degree) appeared on screen covering 8 degrees of visual angle (adapted from Hoogenboom et al., 308 2006a) for 3.4 seconds. In the center of the inward moving grating an asterisk was present that slowly 309 rotated either clock-wise or counter-clock wise. The rotation rate was continuously updated after each 310 trial using an adaptive-staircase procedure (Watson and Pelli, 1983) so that participants were roughly 311 80% correct in detecting rotation direction. The goal of the task was to keep the participants fixated and 312 assure a stable level of attention throughout the experiment, as well as to assess TACS effects on foveal 313 detection accuracy.

316
TACS was applied using a battery-driven NeuroConn DC+ stimulator (neuroConn GmbH, Ilmenau, 317 Germany) connected to three 5 x 5 cm conductive, non-ferromagnetic rubber electrodes that were 318 attached to the scalp following the international 10-20 system, creating an occipital montage (Oz-Cz) and    tracker channels between 1 and 15 Hz (4 th order, two-pass, Butterworth). Trials that exceeded a z-score of 385 5 were rejected. Second, trials that included SQUID-jumps were detected by first high-pass filtering the 386 data at 30 Hz (4 th order, two-pass, Butterworth) to attenuate the stimulation artifact. Trials of which the 387 first-order temporal derivative exceeded a z-score of 25 were rejected. This resulted in an average of 9% 388 ± 8% (Mean ± SD) rejected trials per subject. The data were down-sampled to 600 Hz after epoching into 389 trials from -3.4 to 3.6 seconds after onset of the gamma-inducing visual stimulus.

392
A single-shell head model (Nolte, 2003) was created from the individual MRIs. Next, an equally-spaced 393 grid with 0.5 mm 3 based on a standard MNI template MRI with 0.1 mm 3 resolution was created. This 394 template grid was warped to each subject's individual anatomy to easily average and compare voxels 395 across subjects. Then, a spatial filter was designed to maximize the sensitivity to the expected gamma- . From these data, we created a new grid with 10 voxels. After bandpass-filtering (40 -70 Hz) to maximize spatial filter sensitivity to the gamma-band, the covariance matrix was calculated for epochs 417 from -2.3 to 2.3 s relative to visual stimulus onset on Sham trials, and spatial filters were calculated using 418 5% regularization. The raw sensor-level data was multiplied by the resulting spatial filters to obtain  surrogates. As the number of epochs depends on the number of cycles (being higher for higher 506 frequencies), we applied a random subsampling approach to create unbiased averages. We first 507 determined the stimulation frequency with the smallest number of epochs and then averaged 500 508 randomly drawn subsamples of that size per condition. To allow direct comparison between TACS 509 frequencies and averaging across subjects (with individualized IAF), we transformed the time-axis to 510 radians by adjusting the step size during TFR calculation accordingly. Importantly, TACS peak-locked 511 TFRs were calculated for both baseline and visual stimulation period. Since the baseline period does not 512 contain any visually-induced gamma-band responses, but may contain residual TACS artifacts, it serves 513 as an excellent control against TACS artifact-related spurious phase-amplitude coupling. Respective 514 Sham TFRs were then subtracted from the TACS peak-locked TFRs, and individual gamma power values 515 were compared for each phase angle (radians) with two-sided one-sampled t-tests against zero using FDR 516 correction for multiple comparisons ( Figure 2B and Figure S3).

519
To more formally quantify and compare the extent to which TACS phasically modulates the gamma-band 520 response, we estimated phase-amplitude coupling (PAC) using Tort's Modulation Index (MI) by 521 calculating the normalized Kullback-Leibler (KL) divergence of the histogram of TACS phase-binned 522 gamma amplitude to a uniform distribution (Tort et al., 2010). In case of significant PAC, the histogram 523 diverges from a uniform distribution. To this end, the gamma-band amplitude was determined by 524 convolving the virtual channel data between 0.5 and 1.5 s after visual stimulus onset with a 5-cycle 525 moving time window multiplied with a Hanning taper for frequencies from 30 Hz to 100 Hz in steps of 526 1 Hz, while a 1 second time window was used for estimating the phase of the TACS signal similarly to 527 the gamma-band magnitude (Jiang et al., 2015). The phase-difference between the gamma-band power 528 envelope and TACS signal was subsequently calculated and compared to a uniform distribution using 529 Tort's Modulation Index (MI). As for TACS peak-locked TFRs, we randomly subsampled the data for 530 each condition 500 times using a sample size equal to the lowest number of trials across conditions, to 531 prevent any bias due to unequal trial numbers between conditions. MIs were calculated for each random 532 subsample and then averaged. As a control, PAC was also estimated for surrogate data, for which the 533 phase-providing TACS signal was randomly phase-shifted to create frequency-specific surrogate PAC 534 values for each TACS condition. We used one-sample t-tests to test for significant PAC after subtracting 535 PAC values at baseline before visual stimulus onset and respective visual-stimulation induced changes in 536 surrogate PAC values. We used repeated-measures ANOVAs (no correction for non-sphericity was 537 necessary) on the baseline-and TACS-corrected data, followed by paired-sample t-tests were appropriate, 538 to compare PAC between TACS montage and frequency conditions.