supplemental data brain oscillatory substrates of … data brain oscillatory ... (order of splines:...

15
Current Biology, Volume 19 Supplemental Data Brain Oscillatory Substrates of Visual Short-Term Memory Capacity Paul Sauseng, Wolfgang Klimesch, Kirstin F. Heise, Walter R. Gruber, Elisa Holz, Ahmed A. Karim, Mark Glennon, Christian Gerloff, Niels Birbaumer, and Friedhelm C. Hummel Supplemental Experimental Procedures Experiment 1 Experimental Paradigm A visuospatial short-term memory task as used by Vogel and Machizawa [1] was run. Figure S1 depicts a single trial. A fixation cross was presented throughout the whole trial. Subjects were instructed to fixate the center of the screen. For 200 ms a cue was presented, as either an arrow to the left or right. This was followed by a 100 ms presentation of the memory array – colored squares in the left and in the right visual hemifield (gaze had to be directed to a fixation cross in the middle of the array throughout the trial; fixation was controlled with horizontal and vertical electrooculogram EOG). Only the colors of the squares presented in the hemifield which was previously cued by the arrow had to be retained for the following retention interval of 900 ms. All other squares had to be ignored. After the retention interval a probe item was presented and subjects had to indicate by button press with the index finger of the right hand whether the color of the attended items from the memory array was the same as in the probe or whether the color of any of these squares had changed. The position of the squares never changed from array to probe. Unattended items also could change their color, but this had to be ignored. Subjects EEG was recorded from the scalp of 21 volunteers (15 females; all right handed). Five participants had to be excluded from further data analysis because their EEG was contaminated by artifacts due to eye blinks and horizontal eye movements. Mean age of the remaining sample of 16 participants was 23.9 years (SD = 4.4). The study was approved by the review board of the University of Tübingen Medical School. All participants gave written informed consent. Behavioral Data Analysis Cowan’s K [2] was calculated for individual subjects taking into account the hit and false alarm rate and the number of relevant items in the memory array: K = (hit rate – false alarm rate) * set size The maximal K over all load conditions was used as estimate for individual memory capacity for further analysis.

Upload: vuongdang

Post on 02-May-2018

214 views

Category:

Documents


1 download

TRANSCRIPT

Current Biology, Volume 19

Supplemental Data

Brain Oscillatory Substrates

of Visual Short-Term Memory Capacity Paul Sauseng, Wolfgang Klimesch, Kirstin F. Heise, Walter R. Gruber, Elisa Holz, Ahmed A. Karim, Mark Glennon, Christian Gerloff, Niels Birbaumer, and Friedhelm C. Hummel Supplemental Experimental Procedures Experiment 1 Experimental Paradigm A visuospatial short-term memory task as used by Vogel and Machizawa [1] was run. Figure S1 depicts a single trial. A fixation cross was presented throughout the whole trial. Subjects were instructed to fixate the center of the screen. For 200 ms a cue was presented, as either an arrow to the left or right. This was followed by a 100 ms presentation of the memory array – colored squares in the left and in the right visual hemifield (gaze had to be directed to a fixation cross in the middle of the array throughout the trial; fixation was controlled with horizontal and vertical electrooculogram EOG). Only the colors of the squares presented in the hemifield which was previously cued by the arrow had to be retained for the following retention interval of 900 ms. All other squares had to be ignored. After the retention interval a probe item was presented and subjects had to indicate by button press with the index finger of the right hand whether the color of the attended items from the memory array was the same as in the probe or whether the color of any of these squares had changed. The position of the squares never changed from array to probe. Unattended items also could change their color, but this had to be ignored. Subjects EEG was recorded from the scalp of 21 volunteers (15 females; all right handed). Five participants had to be excluded from further data analysis because their EEG was contaminated by artifacts due to eye blinks and horizontal eye movements. Mean age of the remaining sample of 16 participants was 23.9 years (SD = 4.4). The study was approved by the review board of the University of Tübingen Medical School. All participants gave written informed consent. Behavioral Data Analysis Cowan’s K [2] was calculated for individual subjects taking into account the hit and false alarm rate and the number of relevant items in the memory array:

K = (hit rate – false alarm rate) * set size The maximal K over all load conditions was used as estimate for individual memory capacity for further analysis.

EEG Data Acquisition. Electrical brain activity was recorded with a sampling rate of 500 Hz within 0.16 and 70 Hz from 28 scalp sites positioned according to the international 10-20-system using a BrainAmp (Brain Products) amplifier and Ag-AgCl-ring electrodes mounted in an Easycap (FMS). In addition, EOG was recorded for vertical and horizontal eye movements. During recording a common reference at the tip of the nose was used. Data were later re-referenced to digitally linked earlobes. EEG Data Analysis Preprocessing. After re-referencing of EEG data to digitally linked ear-lobes data were segmented into 1000 ms epochs containing the 900 ms retention interval. Data were then visually inspected for artifacts due to eye movements (vertical or horizontal), eye blinks, muscular activity, etc. Segments containing artifacts were removed from further analysis. For left as well as right hemifield load 5 conditions, significantly less artifact-free trials remained for analysis than in the other conditions, and two subjects showed less than 20 artifact-free trials for load 5 conditions. Therefore memory load 5 conditions were removed from all further analyses. Only results for conditions with memory load 2, 3, 4 and 6 are reported. Mean number of analyzed trials in the other conditions was 26.4 (sd=3.0). Laplacian current source density was calculated for EEG channels (order of splines: 4; max. deg. of Legendre polynomials: 10; lambda: 1 e-5) in order to attenuate effects due to volume conduction. To avoid influence of stimulus-evoked EEG activity on oscillatory activity, the event-related potential was subtracted from each artifact-free single trial used for frequency analyses. This is particularly important for cross-frequency interaction estimates since subtraction of the event-related potential in every single trial ensures that the observed effects cannot stem from filter-artifacts to an evoked response. Phase-Locked Power Modulation Gabor expansion was applied to artifact-free segmented data using Matlab 7.0.1 (Mathworks) to obtain single-trial instantaneous phase and amplitude information for frequencies between 0 and 70 Hz with a distance of 1 Hz between center frequencies. Gabor expansion was applied to transform a signal x(t) from the time domain into a complex time-frequency signal y(fn, t) which is obtained from the inverse Fourier transform F-1 of Y(fn, f) with

( ) ( ) ( )00;0

0;,2, Nn

ffffGfF

ffY nxn ∈

⎩⎨⎧

≤>⋅⋅

= (1)

with Fx(f) being the Fourier-transformed signal x(t) and ( ) ( ){ }f,fYFt,fy n

1n

−= , (2) where G(fn, f) is a window filter in the frequency domain defined as:

( ) ( )22

, nffn effG −−= α

, (3) with α2 as the parameter that controls the time-frequency resolution. The Gabor filter with the Gaussian window (3) represents a linear time-frequency-representation with optimal time-frequency resolution. From (2) the instantaneous phase of the signal may be obtained as a function of frequency and time:

[ ] [ ]{ })t,f(yIm,)t,f(yRearg)t,f( nnn =Φ , (4) where arg{.} denotes the generalized arctan, Re[.], Im[.] the real and imaginary component of the complex representation and n the n’th frequency.

Instantaneous phase values for delta (2 Hz), theta (6 Hz) and alpha (10 Hz), as well as instantaneous amplitude values for 1 Hz frequency bins (each normalized within trials) between 20 and 70 Hz from all artifact-free trials in every experimental condition were concatenated. Thereafter, amplitude values between 20 and 70 Hz were sorted according to instantaneous phase values (from –pi to +pi) for delta, theta or alpha. 20 to 70 Hz amplitude values were then averaged for 50 segments of 2*pi/50 (=7.2°) in respect to the (sorted) delta, theta or alpha cycle (see Figure S2). A similar approach to study (lower frequency) phase-dependent (higher frequency) amplitude modulation was proposed by Demiralp et al. [3].

As reported by Penny et al. [4] a highly reliable method for testing cross-frequency phase-amplitude coupling is calculating a phase locking value (PLV) as suggested by Mormann et al. [5]. Therefore, a cross-frequency PLV was analyzed between theta phase and amplitude envelope of frequencies between 20 and 70 Hz. This was repeated for surrogate data in which the theta phase values were randomly shifted in each trial. Then for each frequency bin the PLV from real data was compared to the PLV from surrogate data using paired t-tests corrected for multiple comparisons.

For statistical analysis four-way repeated measures ANOVAs with factors MEMORY LOAD (2,3,4 and 6), VISUAL HEMIFIELD (left, right), HEMISPHERE (average of left posterior sites P7, P3, PO3, O1 vs. average of right posterior sites P8, P4, PO4, O2) and PHASE SEGMENT (-pi to +pi in 50 equally large segments) for high gamma (50 to 70 Hz; the frequency band with strongest low-frequency phase-locked amplitude modulation, see Fig. 2) amplitude phase-locked to delta, theta, and alpha were run. Huynh-Feldt correction was applied where necessary. Theta/Delta Phase-Locked Gamma Phase Synchronization Instantaneous phase values of the high gamma band (50 to 70 Hz) were sorted in respect to low frequency phase angles (see Figure S3). Then for bins of 2*pi/250 (=1.44°) of the low frequency cycle phase synchronization of high gamma activity was calculated. This was obtained by averaging of sine and cosine values for instantaneous high gamma phases within each 2*pi/250 segment of the low frequency. The resulting length of the mean vector was then used as estimate for low frequency phase-locked gamma phase synchronization.

This estimate describes strict phase locking between low frequencies and gamma activity independent of amplitude modulation and independent of the actual instantaneous phase of gamma activity. It is, however, possible to estimate to which phase angle of the low frequency gamma phase is synchronized to. Only delta-locked and theta-locked high gamma phase synchronization was calculated as alpha phase did not significantly modulate gamma amplitude (see results section). Furthermore, for statistical analysis only delta-locked and theta-locked high gamma phase synchronization was used within a broad delta or theta phase segment showing high gamma amplitude (see Fig. 1). Therefore, repeated measures ANOVAs were only calculated for averaged delta-locked high gamma phase synchronization within 64.8° around the zero crossing of the positive slope within the delta cycle (segment of strongest delta phase-locked gamma activity; Figure S7a), and for a 64.8° segment following the negative peak of the theta cycle for theta-locked gamma phase synchronization (see Fig. 1a). Two-way repeated measures ANOVAs with factors MEMORY LOAD (2,3,4 and 6) and HEMISPHERE (average of left posterior sites P7, P3, PO3, O1 vs. average of right posterior sites P8, P4, PO4, O2) were run for delta-locked and for theta-locked high gamma phase synchronization differences between

right and left hemifield conditions. Huynh-Feldt correction was applied where necessary, and Scheffé post-hoc tests were calculated. EEG Amplitude To analyze amplitude of EEG-alpha activity Fast Fourier Transformation was applied to artifact-free current source density data during the retention interval (with the evoked response subtracted from each trial). The Fourier transformed data were then averaged over trials for each condition separately and amplitude values between 8 and 12 Hz at the same electrode sites as used for cross-frequency interaction estimates were entered into statistical analysis. Two-way repeated measures ANOVAs with factors MEMORY LOAD (2,3,4 and 6) and HEMISPHERE (average of left posterior sites P7, P3, PO3, O1 vs. average of right posterior sites P8, P4, PO4, O2) were run for spectral alpha amplitude value differences between left and right hemifield conditions. Scheffé post-hoc tests were calculated. Experiment 2 Experimental Paradigm A visuospatial short-term memory task was run in which colored squares were presented to both visual hemifields. Unlike in experiment 1 the number of items was not always equal in the two hemifields. There were four different possibilities: (i) two items left and two items right, (ii) two items left but four items right, (iii) four items left but two items right, or (iv) four items left and four items right. The temporal structure of a single trial was as follows (depicted in Figure S6). A fixation cross was shown in the middle of the screen throughout the experiment. Subjects had to keep their gaze at the fixation cross during the whole trial (fixation was monitored using EOG). In each trial first a cue consisting of one arrow above and one arrow below the fixation cross, both pointing either to the left or to the right, was presented for 200 ms. The cue was immediately followed by a memory array consisting of colored squares within both visual hemifields presented for only 100 ms. Participants were told to retain the color of only those squares which were shown in the visual hemifield cued by the arrows. In order to keep the properties of visual stimulation for both hemifields relatively comparable, we always added two gray filler items if in one hemifield there were only two colored items and in the other four squares. Participants were told that the filler items did not contain any information. After a retention period of 1000 ms a probe item was presented for 2000 ms, and subjects had to indicate by button press whether the color of any of the retained items had changed or not. Subjects EEG was recorded from the scalp of 15 volunteers (12 females; all but one right handed). One participant had to be excluded from further data analysis because her EEG was contaminated by artifacts due to eye blinks and horizontal eye movements. Mean age of the remaining sample of 14 participants was 24.7 years (SD = 3.5). All participants gave written informed consent. EEG Data Analysis Preprocessing. EEG data pre-processing was identical to experiment 1. Mean number of artifact-free trials was 33.8 (SD = 10.4).

Theta-Locked Gamma Phase Synchronization Theta-locked gamma phase synchronization was analyzed in the same way as described for experiment 1. For statistical analysis a three-way repeated measures ANOVA with factors RELEVANT_INFORMATION (2 vs. 4 items in the cued hemifield), IRRELEVANT_INFORMATION (2 vs. 4 items in the uncued hemifield) and HEMISPHERE (average of left posterior sites P7, P3, PO3, O1 vs. average of right posterior sites P8, P4, PO4, O2) was run. EEG Amplitude Alpha amplitude was analyzed in the same way as implemented in experiment 1. Similar to theta-locked gamma phase synchronization a three-way repeated measures ANOVA with factors RELEVANT_INFORMATION (2 vs. 4 items in the cued hemifield), IRRELEVANT_INFORMATION (2 vs. 4 items in the uncued hemifield) and HEMISPHERE (average of left posterior sites P7, P3, PO3, O1 vs. average of right posterior sites P8, P4, PO4, O2) was calculated. Experiment 3 Subjects Seven volunteers participated in this experiment (6 females; all right handed; mean age = 29.6 years; SE = 3.6). The study was approved by the review board of the University of Tuebingen Medical School. All participants gave written informed consent. Experimental and Stimulation Paradigm The same experimental paradigm was run as in experiment 1 with the exception that only trials with memory array size four were used. During the retention interval a train consisting of nine TMS pulses at a rate of 10 Hz was delivered (biphasic pulses with posterior-anterior current direction). Stimulation sites were 10-20-electrode sites P3 and P4 (left and right posterior parietal), representative sites among recording sites showing memory capacity-related effects in the EEG experiment. Orientation of the coil was with the handle pointing in a posterior direction. Control conditions were rTMS with the coil tilted by 45° (reducing stimulation of the underlying cortex, but mimicking coil clicking and sensation of magnetic stimulation; so the effect is similar to a 90° tilted coil usually used as sham condition, however, sensation of stimulation is better imitated [6]) applied at sites P3 and P4 and real rTMS (verum rTMS) delivered at the vertex (to control for discomfort of stimulation). Stimulation was applied with a Magstim rapid stimulator and a figure of eight coil with 5 cm coil diameter. Stimulation intensity was 110% of resting motor threshold. All stimulation parameters were within safety-guidelines for the use of repetitive TMS [7, 8]. Memory capacity was calculated according to Cowan [2] separately for trials with rTMS contralateral or magnetic stimulation ipsilateral to the squares which had to be retained in memory in the verum and in the control conditions and for rTMS at the vertex. A one-way repeated measures ANOVA with factor STIMULATION (verum rTMS contralateral, verum rTMS ipsilateral, control rTMS contralateral, control rTMS ipsilateral, verum rTMS vertex) was run. Huynh-Feldt correction was applied, and Scheffé post-hoc tests were calculated.

Experiment 4 Subjects 13 volunteers participated in this experiment (8 females; all right handed; mean age = 29.3 years; SE = 5.3). All participants gave written informed consent. Experimental and Stimulation Paradigm The same experimental paradigm was run as in experiment 3. During the retention interval a train consisting of nine TMS pulses at a rate of 10 Hz or consisting of 14 TMS pulses at a rate of 15 Hz was delivered (10 and 15 Hz rTMS are considered to have a similar facilitory effect on brain excitability [9-11]). Stimulation sites were 10-20-electrode sites P3 and P4 (left and right posterior parietal), representative sites among recording sites showing memory capacity-related effects in the EEG experiment, and as control sites, centro-parietal sites 3 cm anterior and 1 cm medial to the parietal stimulation sites. rTMS was applied with a Magstim rapid stimulator and a figure of eight coil with 7 cm coil diameter. Stimulation intensity was 110% of resting motor threshold. All stimulation parameters were within safety-guidelines for the use of repetitive TMS [7, 8]. Memory capacity was calculated according to Cowan [2] for trials with rTMS contralateral or magnetic stimulation ipsilateral to the squares which had to be retained in memory for parietal and centro-parietal stimulation at 10 Hz and 15 Hz separately. A three-way repeated measures ANOVA with factors SITE (parietal, centroparietal), FREQUENCY (10 Hz, 15 Hz) and HEMIFIELD (ipsilateral, contralateral) was run. Scheffé post-hoc tests were calculated.

Supplemental Results Behavioral Results In addition to estimation of Cowan’s K as a behavioral measure error rates were also analyzed for the different load conditions from experiment 1. We compared rates of false alarms (when the color of the probe was identical with the memory set but participants indicated that color of the probe was different) and missed trials (when there was a change of color which participants did not detect). A two-way ANOVA with factors ERROR TYPE (false alarms, miss) and LOAD (2 to 6; note, for behavioral analysis we also included memory load 5 conditions which could not be analyzed with the EEG due to artifacts) was implemented. Both main effects were significant (ERROR TYPE: F1/15 = 109.94, p < 0.000; LOAD: F4/60 = 126.69, p < 0.000). In general there were more false alarms than misses and error rate increased with higher memory load. The interaction between ERROR TYPE and LOAD was significant (F4/60 = 54.51, p < 0.000). As evident in Figure S5, the false alarm rate shows a much stronger load-dependent increase than the miss rate. Note also that the miss rate shows a linear-like increase with memory load whereas the false alarm rate exhibits a rather sigmoidal function.

The fact that false alarm rates respond primarily to increased memory load might suggest that the drop in performance with higher load is more likely due to poor suppression of distracting information than deficient retention of relevant items. The basis for this assumption is that if items from the distracting visual hemifield are not suppressed efficiently and change their color in the probe, participants will be more prone to indicate that there was a change in color although relevant items have not changed, resulting in false alarms. However, only in half of the trials with congruent cued visual field did the distracting items actually change color, rendering this interpretation rather implausible, as increased false alarm rates would only be expected if in all matching target trials the distracting items were incongruent. So it is more likely that subjects tended to indicate that colors had changed when they had to guess, resulting in higher false alarm than miss rates. Impact of 10 Hz rTMS on Error Rates In experiments 3 and 4 the effect of rTMS on memory capacity estimate K (which is based on hit and false alarm rates) was investigated. Here, we present additional analysis where the direct influence of magnetic stimulation on error rates is considered. Data from all participants from experiment 3 and 4 were used. For parietal rTMS at 10 Hz (these experimental conditions were used in both experiments), a two-way ANOVA with factors ERROR (false alarm vs. miss) and HEMISPHERE (ipsilateral vs. contralateral) was run on error rates. As depicted in Figure S6, with ipsilateral stimulation, error rates were decreased in general, resulting in a significant main effect for HEMISPHERE (F1/19 = 69.69, p < 0.000). In general there was again a smaller miss than false alarm rate (F1/19 = 5.49, p = 0.030). The interaction between factors ERROR and HEMIFIELD only slightly missed significance (F1/19 = 3.15, p = 0.091). This indicates that there was a trend towards stronger reduction of false alarm rate after ipsilateral 10 Hz stimulation than for miss rate. However, when one takes into account that in general subjects tended to indicate that the color of the squares have changed (increasing the number of false alarms; see above), this can be relativized. The decrease of error rates after ipsilateral rTMS compared to contralateral rTMS was 40.4 % for false alarms and 37.3 % for miss trials. Therefore, it appears rather implausible that parietal 10 Hz rTMS had greater impact on false alarm rate, in particular

if (as stated above) it is considered that only in half of the matching trials distractors changed their color. Effect of rTMS Experimental Block Order For rTMS experiments a block design was used in which blocks were presented in a pseudo-randomized order. To exclude the possibility of an influence of block order on the results, the repeated measures ANOVA from experiment 3 was also conducted as a co-variance analysis, controlling for the effect of stimulation block order. The co-variance analysis yielded a nearly identical result as the original ANOVA (co-variance analysis: F4/24 = 4.047, p = 0.013; ANOVA: F4/24 = 4.418, p = 0.008). This suggests that after controlling for the effect of block order, rTMS still had a significant effect on memory capacity (p = 0.013). Delta-Locked and Alpha-Locked Gamma Amplitude and Phase Synchronization In experiment 1, which showed gamma amplitude and gamma phase being locked to theta phase, cross-frequency phase-amplitude and phase-phase coupling was also analyzed for the frequency combinations delta:gamma and alpha:gamma. There was significant modulation of high gamma amplitude by delta phase (Figure S7a) as indicated via a significant main effect for delta phase angle on high gamma amplitude (F49/735 = 16.04, p = 0.000) with a similar topography to theta-gamma interaction. However, in contrast to theta-locked gamma, there was increased gamma amplitude around the zero-crossing in the positive slope of the delta cycle and more moderately also at the zero crossing in the negative slope. Thus, gamma amplitude seemed to fluctuate at a faster frequency than delta resulting in two cycles of gamma waxing and waning within one delta period, namely at theta frequency (about the double frequency of delta). Possibly this was due to a too poor frequency resolution of the filtering to resolve neighbouring frequency bins, and thus also reflects the significant effect of theta phase to gamma amplitude coupling. An alternative interpretation is that this pattern might resemble a hierarchical structure of coupled oscillations in the delta, theta, and gamma frequency range as proposed by Lakatos et al. [12]. It is possible that delta and theta oscillations were synchronized and delta:gamma coupling thus only represents a spurious relation. Alpha phase, on the other hand, (Figure S7b) did not modulate high gamma amplitude significantly (F49/735 = 2.30, p = 0.097). Neither delta nor alpha phase-locked gamma amplitude showed any effects on memory load or hemifield. As reported for theta:gamma coupling, delta-locked gamma phase synchronization was also calculated. But in contrast to theta phase-locked gamma synchronization it showed no memory load dependent or task specific effect. For alpha, no phase-locked gamma synchronization was analyzed as alpha did not significantly modulate gamma amplitude. Additional Theta:Gamma Cross-Frequency Phase Synchronization Results As reported in the results section of the manuscript theta:gamma phase synchronization was calculated and statistically evaluated for the theta phase segment at which significant theta-locked gamma amplitude had been obtained (starting 7.2° prior and ending 57.6° following the negative peak of theta). Within this theta phase segment the reported memory load dependent effects were found. Although we have argued that this estimate of theta:gamma phase synchronization seems to be most valid when significant gamma activity (i.e. amplitude) locked to theta occurs, as a control analysis we also calculated theta:gamma phase synchronization for the theta phase segment where no significant phase-locked gamma amplitude was obtained. This

control analysis should investigate whether memory specific theta:gamma cross-frequency phase synchronization is dependent on the theta phase-locked gamma amplitude or a more general phenomenon. A two-way ANOVA with factors HEMISPHERE and MEMORY LOAD was run and there was no significant main effect or interaction (all p>.180). This strengthens the argument that theta:gamma phase synchronization is most valid if there is also significant gamma amplitude obtained. EEG Amplitude at Delta, Theta, and Gamma Frequency The same analyses for posterior alpha amplitude in experiment 1 were also carried out for the delta band (0.16-4 Hz), theta (4-8 Hz) and high gamma frequency (50-70 Hz). The difference between left and right posterior amplitude was entered into a two-way ANOVA with factors MEMORY LOAD (2, 3, 4, 6) and HEMISPHERE (ipsilateral vs. contralateral) for each frequency band separately. For delta no main effect or interaction was significant. Theta amplitude only showed a significant main effect for HEMISPHERE (F1/15 = 8.30, p = 0.011) indicating stronger theta power contralateral to the relevant visual hemifield. However, this effect was independent of memory load. Also for gamma amplitude a significant main effect for the factor HEMISPHERE was obtained (F1/15 = 21.11, p < 0.000), but this time with ipsilateral sites showing stronger amplitude. Again, this effect was independent from MEMORY LOAD. Unlike alpha amplitude, lateralized amplitude increase from memory load two to four of none of the three frequency bands could predict individual memory capacity (all p > 0.65). Effects of Eye Movements Yuval-Greenberg et al. [13] argued that saccadic eye movements can generate spurious gamma activity in human scalp EEG. In the current experiment subjects were instructed to keep their gaze at a central fixation cross. As mentioned above fixation was monitored using EOG recordings, and trials containing eye movements were excluded from further analysis. Still, there might be effects on oscillatory EEG activity by saccadic eye movements which were not detected in the analysis. As we used CSD estimates instead of raw EEG for analysis it however is unlikely that such effects should have had an influence on the present results (see [14] for a comprehensive discussion). Nevertheless, we implemented the same EEG analyses on the horizontal EOG to ensure that the current results are not due to eye movements. For rhythmical activity at alpha frequency there was no main effect or interaction for the factors HEMIFIELD (left, right) and MEMORY LOAD (2, 3, 4 and 6) significant (all p>.130) at the EOG channel. Also theta phase to gamma amplitude coupling did not show any significant effect (all main effects and interaction p>.148). For the theta phase interval in which there was a short-term memory specific effect on theta:gamma cross-frequency phase synchronization in the EEG we also calculated cross-frequency phase synchronization for the EOG. On this measure we could only find a significant main effect on the factor HEMIFIELD (left, right) in the EOG (F1/15 = 6.60, p = 0.021). All other main effects and interactions were not significant (p > 0.129). These results are evidence against the assumption that the effects on EEG alpha amplitude, theta phase-locked gamma amplitude, or theta:gamma cross-frequency phase synchronization reported in the present paper were influenced by saccadic eye movements.

Supplemental References 1. Vogel, E.K., and Machizawa, M.G. (2004). Neural activity predicts individual differences in

visual working memory capacity. Nature 428, 748-751. 2. Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental

storage capacity. Behavioral and Brain Sciences 24, 87-114. 3. Demiralp, T., Bayraktaroglu, Z., Lenz, D., Junge, S., Busch, N.A., Maess, B., Ergen, M., and

Herrmann, C.S. (2007). Gamma amplitudes are coupled to theta phase in human EEG during visual perception. International Journal of Psychophysiology 64, 24-30.

4. Penny, W.D., Duzel, E., Miller, K.J., and Ojemann, J.G. (2008). Testing for nested oscillation. Journal of Neuroscience Methods 174, 50-61.

5. Mormann, F., Fell, J., Axmacher, N., Weber, B., Lehnertz, K., Elger, C.E., and Fernandez, G. (2005). Phase/amplitude reset and theta-gamma interaction in the human medial temporal lobe during a continuous word recognition memory task. Hippocampus 15, 890-900.

6. Lisanby, S.H., Gutman, D., Luber, B., Schroeder, C., and Sackeim, H.A. (2001). Sham TMS: Intracerebral measurement of the induced electrical field and the induction of motor-evoked potentials. Biological Psychiatry 49, 460-463.

7. Chen, R., Gerloff, C., Classen, J., Wassermann, E.M., Hallett, M., and G. Cohen, L. (1997). Safety of different inter-train intervals for repetitive transcranial magnetic stimulation and recommendations for safe ranges of stimulation parameters. Electroencephalography and Clinical Neurophysiology - Electromyography and Motor Control 105, 415-421.

8. Wassermann, E.M. (1998). Risk and safety of repetitive transcranial magnetic stimulation: Report and suggested guidelines from the International Workshop on the Safety of Repetitive Transcranial Magnetic Stimulation, June 5-7, 1996. Electroencephalography and Clinical Neurophysiology - Evoked Potentials 108, 1-16.

9. Berardelli, A., Inghilleri, M., Rothwell, J.C., Romeo, S., Currà , A., Gilio, F., Modugno, N., and Manfredi, M. (1998). Facilitation of muscle evoked responses after repetitive cortical stimulation in man. Experimental Brain Research 122, 79-84.

10. Pascual-Leone, A., Tormos, J.M., Keenan, J., Tarazona, F., Cañete, C., and Català, M.D. (1998). Study and modulation of human cortical excitability with transcranial magnetic stimulation. Journal of Clinical Neurophysiology 15, 333-343.

11. Maeda, F., Keenan, J.P., Tormos, J.M., Topka, H., and Pascual-Leone, A. (2000). Interindividual variability of the modulatory effects of repetitive transcranial magnetic stimulation on cortical excitability. Experimental Brain Research 133, 425-430.

12. Lakatos, P., Shah, A.S., Knuth, K.H., Ulbert, I., Karmos, G., and Schroeder, C.E. (2005). An oscillatory hierarchy controlling neuronal excitability and stimulus processing in the auditory cortex. Journal of Neurophysiology 94, 1904-1911.

13. Yuval-Greenberg, S., Tomer, O., Keren, A.S., Nelken, I., and Deouell, L.Y. (2008). Transient Induced Gamma-Band Response in EEG as a Manifestation of Miniature Saccades. Neuron 58, 429-441.

14. Melloni, L., Schwiedrzik, C.M., Rodriguez, E., and Singer, W. (2009). (Micro)Saccades, corollary activity and cortical oscillations. Trends in Cognitive Sciences 13, 239-245.

Figure S1. Scheme of a Single Trial from Experiment 1

Figure S2. Idealized Depiction of Theta-Locked Gamma Amplitude Modulation Theta oscillations are transformed into phase values. For gamma, band amplitude is estimated (amplitude envelope). Gamma amplitude values from all trials are sorted according to the appropriate theta phase value. Last, sorted gamma amplitude values are averaged within segments of 7.2° of a theta cycle.

Figure S3. Idealized Depiction of Theta-Locked Gamma Phase Synchronization First, phase information is gained for theta and gamma oscillations. Then instantaneous gamma phase values from all trials are sorted according to the appropriate instantaneous theta phase value. For segments of 1.44° of a theta cycle the length of the mean vector (result of averaging the sine and cosine of single trial gamma phase values from the segment) is calculated as estimate for theta-locked gamma phase synchronization.

Figure S4. Temporal Structure of a Single Trial from Experiment 2

Figure S5. False Alarm Rate Increases as a Sigmoidal Function with Memory Load Miss rate is in general lower than false alarm rate and shows only a moderate rather linear increase with memory load. Error bars represent standard error.

Figure S6. Ipsilateral rTMS Decreases Error Rates in General, i.e., Both False Alarm and Miss Rate Error bars represent standard error.

Figure S7. Delta and Alpha Phase-Locked Gamma Amplitude In contrast to alpha (b), delta phase modulates gamma amplitude (a).