Supporting Information

Supporting Information Daitch et al. 1.173/pnas.13794711 SI Experimental Procedures Subject-Specific Methods. Patients 3 and 4. Patients 3 and 4 had vagal nerve stimulators, which are incompatible with the MRI scanner, so presurgical MRIs were not obtained. The CT from each subject (with electrodes) was coregistered with an atlas-representative magnetization-prepared rapid gradient-echo (MP-RAGE) target, and average functional connectivity maps over a group of 25 control subjects were used to define task-relevant and -irrelevant electrodes. Patient 6. A postsurgical CT was not obtained from patient 6, so electrode coordinates were estimated based on skull landmarks taken from saggital and coronal skull X-rays, using the LOC (location on cortex) localization package (1). Electrocorticography Data Collection/Stimulus Presentation. The BCI2 platform (2) was used to synchronize data collection, stimulus presentation, and behavioral response monitoring (e.g., mouse click) during the task. We asked subjects to maintain fixation on a central crosshair throughout each run of the task, and eye movements were monitored with a Tobii T6 integrated infrared eye tracker, which was also synchronized with stimulus presentation and electrocorticography (ECoG) data collection using BCI2. Visual stimuli were presented on a monitor set up by the patient s hospital bed. We collected between three and four runs of the task from each subject, each run consisting of 12 trials. Each subject was implanted with an electrode grid (8 8or6 8 configuration) or electrode strips (1 4, 1 6, or 1 8), all placed subdurally facing the cortical surface. Electrodes, manufactured by PMT, were made of platinum, each 4 mm in diameter with 2.3 mm exposed to the cortical surface and spaced apart by 1 cm. A separate 1 4 strip was placed subdurally facing the skull to use as the amplifier s ground and reference. Signals were acquired at 1,2 Hz using optically isolated 16-channel g. USBamp amplifiers (Guger Technologies) and a Dell Precision 69 Quad Core computer (Dell) and converted to MATLAB files for processing and analysis. ECoG Signal Analysis. All signal processing scripts were custom written in MATLAB, unless otherwise noted. The signal at each electrode was first rereferenced to the common mean of all electrodes (excluding noisy electrodes) to minimize common sources of noise from the signals. Spectral decomposition of the rereferenced signals was then accomplished using Gabor wavelet filtering (between 1 Hz and 512 Hz), which yields instantaneous amplitude and phase estimates at each time point for each frequency, and tailors the temporal resolution for each frequency. Gabor filtering was performed on the entire signal recorded during the task (i.e., before it was divided into individual trials), such that the amplitude and phase of low frequencies could be estimated accurately. The Gabor output was then divided into trials for the event-related analyses. The intertrial coherence (ITC) was computed at each time point within a trial and at all considered frequencies to determine what events reset the phase of ongoing oscillations and at what frequencies. ITC reflects the consistency of the phase of the oscillation of a particular frequency at a particular time point within a trial, across trials (Fig. S2 A and B). ITC is equivalent to the magnitude of the mean resultant vector of the oscillatory phase across trials (computed separately for each time point and frequency) (3). A Rayleigh uniformity test determined the significance of each ITC calculation against the null hypothesis that the phases across all trials came from a uniform circular distribution (MATLAB Circular Statistics toolbox). To measure the phase clustering between multiple electrodes (e.g., electrodes within a functional network), we first computed the average phase of each electrode across trials (at a single time point and frequency), then computed the mean phase across electrodes, weighting each electrode s phase by its ITC at that time point and frequency. Average phase differences between networks were calculated as the absolute value of the difference between the mean phase of each network. Phase-locking values (PLV) were computed between each pair of electrodes over different epochs in each subject of the task by concatenating the signal from each 5-ms epoch in a trial, across trials. The PLV is then equal to the magnitude of the mean resultant vector of the phase difference between two electrodes, averaged across all time points within the concatenated signal. We performed paired t tests to determine in which network pairs there was a significant difference between the PLVs in different task epochs and the ITI, correcting for number of epochs and number of network pairs. Statistical evaluation of differences between physiological measurements (e.g., ITC) in two experimental conditions (e.g., ipsilateral vs. contralateral cue; invalid vs. valid targets) was accomplished with permutation tests, which calculated the likelihood of obtaining the given difference in metrics between the two experimental conditions if the two conditions were not in fact different. To conduct the permutation test, the data were resampled 1, times, each time shuffling the labels for the two conditions and recomputing the test statistic with these shuffled datasets. The P value was determined by the percentile of the test statistic using the correct labels along the distribution of test statistics using the shuffled labels. When the test statistic related to the variance of a measure (e.g., ITC, which is related to the variance of phases across trials), the distributions from the two conditions were first shifted such that they had the same means before conducting the permutation test. Functional MRI Acquisition and Analysis. All scans were collected on a Siemens 3T Tim-Trio scanner. Structural scans consisted of a saggital MP-RAGE T1-weighted image (TR = 1,95 ms, TE = 2.26 ms, flip angle = 9, voxel size = 1. 1. 1. mm) and a transverse turbo spin-echo T2-weighted image (TR = 2,5 ms, TE = 435 ms, voxel size = 1. 1. 1. mm). Blood-oxygen level-dependent (BOLD) contrast was measured with a gradient echo echo-planar imaging (EPI) sequence (TR = 2, ms, TE = 27 ms, 32 contiguous 4-mm slices, 4 4 in-plane resolution). The resting state functional MRI (fmri) scans involved between four and eight 6.7-min scans (2 frames per scan) during which subjects fixated on a centrally presented plus sign. Preprocessing consisted of the following steps: (i) Asynchronous slice acquisition was compensated by sinc interpolation to align all slices. (ii) Elimination of odd/even slice intensity differences resulting from interleaved acquisition. (iii) Whole-brain normalization corrected for changes in signal intensity across scans. (iv) Data were realigned within and across scans to correct for head movement. (v) EPI data were coregistered to the subject s T2-weighted anatomical image, which in turn was coregistered with the T1-weighted MP-RAGE, in both cases using a crossmodal procedure based on alignment of image gradients (4). The MP-RAGE was then transformed to an atlas-space (5) representative target using a 12-parameter affine transformation. Movement correction and atlas transformation were accomplished Daitch et al. www.pnas.org/cgi/content/short/13794711 1of8

in one resampling step to minimize blur and noise. (vi) Temporal filtering that eliminates frequencies above.8 Hz. (vi) Removal by linear regression of (a) six parameters for head movement, (b) the signal averaged over the whole brain, excluding the ventricles, (c) the signal over a ventricular region, and (d) the signal from a white matter region. Temporal derivatives of these regressors are also included in the linear model, accounting for time-shifted versions of spurious variance. Functional connectivity (FC) analyses defined the dorsal attention network (), ventral attention network (VAN), sensorimotor network (), and default-mode network (DMN) within each subject, using the resting state fmri data. The first four frames of each scan were eliminated to allow steady-state magnetization and the remaining frames were concatenated. For each network a set of 6-mm radius spherical seed regions of interst (ROIs) (8, 4, 12, and 1 ROIs, respectively, for, VAN,, and DMN) were first identified from meta-analyses that we have previously conducted of task-based activation experiments (refs. 6, 7. and 8 as reanalyzed by ref. 9). Then, for each seed ROI, a voxel-wise FC map was computed indicating the correlation of the timeseries of each voxel in the brain with the seed timeseries. The Fisher z-transform was applied to each FC map, the FC maps were averaged, and the average map was thresholded at Fisher z =.3. The resulting map indicated the voxels that were most strongly and consistently related to each network and was used to define task-relevant or -irrelevant electrodes (see Fig. S1A for variability of functional connectivity maps across subjects). Realignment of Subdural Grids with MRI. CT images were acquired before removal of the grid. Preoperative MP-RAGEs were acquired using standard clinical protocols. CTs were transformed to atlas space using a cross-modal procedure based on alignment of image gradients (4) in which the CT image is aligned to the individual subject MP-RAGE, and the MP-RAGE is then transformed to an atlas-space (5) representative target using a 12- parameter affine transformation. Electrodes were segmented in the CT image by thresholding. Center-of-mass coordinates from clusters of face-contiguous voxels were isolated using an in-house clustering algorithm. Because of subdural hygroma after the surgery filling the intracranial space, the locations of the electrodes at the time of CT acquisition are generally displaced inward relative to the location of the subject s cortical surface at the time of MRI acquisition. To correct for this displacement, electrode coordinates were projected to the surface of the brain along a path normal to the cortical surface. The surface anatomy used in this procedure was extracted using the average of the first four frames of each BOLD run, which was then thresholded and blurred modestly (5 mm) such that electrodes arrive at a location reflecting the smoothed convexity of the brain. 1. Miller KJ, et al. (27) Cortical electrode localization from X-rays and simple mapping for electrocorticographic research: The Location on Cortex (LOC) package for MATLAB. J Neurosci Methods 162(1-2):33 38. 2. Schalk G, McFarland DJ, Hinterberger T, Birbaumer N, Wolpaw JR (24) BCI2: A general-purpose brain-computer interface (BCI) system. IEEE Trans Biomed Eng 51(6): 134 143. 3. Lakatos P, Chen CM, O Connell MN, Mills A, Schroeder CE (27) Neuronal oscillations and multisensory interaction in primary auditory cortex. Neuron 53(2):279 292. 4. Rowland DJ, Garbow JR, Laforest R, Snyder AZ (25) Registration of [18F]FDG micropet and small-animal MRI. Nucl Med Biol 32(6):567 572. 5. Talairach J, Tournoux P (1988) Co-Planar Stereotaxic Atlas of the Human Brain, trans Rayport M (Thieme Medical, New York), pp 122. 6. He BJ, et al. (27) Breakdown of functional connectivity in frontoparietal networks underlies behavioral deficits in spatial neglect. Neuron 53(6):95 918. 7. Astafiev SV, et al. (23) Functional organization of human intraparietal and frontal cortex for attending, looking, and pointing. J Neurosci 23(11):4689 4699. 8. Shulman GL, et al. (1997) Common blood flow changes across visual tasks: II. Decreases in cerebral cortex. J Cogn Neurosci 9(5):648 663. 9. Fox MD, et al. (25) The human brain is intrinsically organized into dynamic, anticorrelated functional networks. Proc Natl Acad Sci USA 12(27):9673 9678. A VAN DMN 4 subjects 3 subjects 2 subjects 1 subject B Right hemisphere Left hemisphere PT1 PT2 PT3 PT4 PT5 PT6 Fig. S1. (A) Conjunction functional connectivity maps. Shown is the degree of overlap between the functional connectivity maps of the four subjects who received fmri scans. Note that although the general structure is similar across subjects, there is much individual variability. (B) Patients grid locations. Shown are the approximate electrode locations for the patients from this study, drawn on atlas-representative brains using the LOC localization package (1). Daitch et al. www.pnas.org/cgi/content/short/13794711 2of8

Fig. S2. Cognitive events associated with synchronization of ongoing oscillations in task-relevant regions/networks. (A) Simulated local field potential oscillations at a single site. The yellow vertical line represents an event during the trial that resets the phase of the some oscillations. Before this event, oscillations exhibit low ITC (shaded in red) and after it, they exhibit high ITC (shaded in blue). We hypothesized that task-relevant regions would become selectively phasereset by particular events, to modulate the excitability of single sites and the effective connectivity between multiple sites, which are simultaneously phasereset. (B) ITC measures the consistency of phase values across trials at different points in time. It is equivalent to the magnitude of the mean resultant vector of the oscillatory phase across trials (at a single time point and frequency. (C) Phase-resetting may allow two regions to become in-phase (Top and Middle traces), possibly facilitating communication between them, or to become out of phase (Middle and Bottom traces), inhibiting communication. Daitch et al. www.pnas.org/cgi/content/short/13794711 3of8

A Delta (1-3 Hz) Theta (3-7 Hz) E Cue period Delay period Cue period Mean normalized power.25 -.25.3 ITC at electrodes with significant ITC Delta (1-3 Hz) Cue onset Target onset B Norm. power at electrodes with significant ITC.5 -.5 ITC or norm. power VAN DMN Beta (15-32 Hz) 5ms.3 -.3 Norm. power Norm. power Norm. power C Distribution of norm. power 1.5 -.5-1 1.5 -.5-1 1.5 -.5-1 Theta (3-7 Hz) 5 6 2 ITC ITC ITC # occurrences Low gamma (32-55 Hz).3.3 -.3 D.8.6.4.2.8.6.4.2.8.6.4.2 Norm. power vs. ITC r = -.4265, p =.64163 r = -.89636, p =.21875 r =.59557, p = 8.1588e-7-1 1 Normalized power (relative to ITI) Alpha (7-15 Hz) High gamma (7-15 Hz).3 -.3 -.3 -.3 Fig. S3. (A) Topography and strengh of ITC at electrodes with significant ITC in the δ-band during the cue or delay, or in the θ-band during the cue period. (B) Normalized power [relative to intertrial interval (ITI)] in the corresponding frequency band and task epoch as in the left column, at the same electrodes with significant ITC in that frequency band and epoch. Note that many of the electrodes exhibiting significant δ or θ ITC exhibit a power decrease (blue) in that same frequency/epoch. (C) Distribution of normalized power at the electrodes in the second column. (D) Scatter plot showing the relationship between the normalized power and ITC at the electrodes in the first and second column, in the respective frequency band and task epoch. (E) Time-courses of spectral power modulations (normalized by power during ITI) throughout the Posner task trial, averaged across electrodes within each functional network. Daitch et al. www.pnas.org/cgi/content/short/13794711 4of8

A.5 PT4 PT5 PT6 Horizontal eye gaze B Right targets Horizontal eye gaze Left targets Horizontal eye gaze -.5.5 -.1.1 -.5 Right cue Left cue Right cue Left cue Right cue Valid targets Invalid targets Valid targets Invalid targets Valid targets Left cue Invalid targets Fig. S4. (A) Horizontal eye gaze during right vs. left cues. In each trial, the maximum eye gaze deviation (from the center of the monitor) within the 5-ms cue period was recorded, and these maximum deviations were averaged across trials. Right and left cue trials were separated to see if subjects saccade to the cued location. Error bars represent SEMs. There was no significant difference between the eye gaze during right versus left cues in any of the three subjects (unpaired t test, P >.5). (B) Horizontal eye gaze during valid vs. invalid targets. In each trial, the maximum eye gaze deviation (from the center of the monitor) within the 5 ms following target onset was recorded, and these maximum deviations were averaged across trials. Trials were separated by those with right and left targets, and within each of these groups, by invalid and valid targets. Error bars represent SEMs. There were no significant differences between the gaze during invalid vs. valid targets (right valid vs. right invalid and left valid vs. left invalid; unpaired t test, P >.5), except for right valid vs. invalid targets in patient 4 (P =.253); however, in this case, the deviation was slightly greater for valid than invalid targets, which is the opposite of what would be expected if subjects saccade to unexpected (i.e., invalid) targets. Fig. S5. Right hemisphere dominance of reorienting response. ITC plots, locked to target onset (dotted line), during valid trials, invalid trials, and the difference between the two conditions, averaged across electrodes in each functional network and across subjects, separately for electrodes over the right hemisphere (Upper; three subjects) and left hemisphere (Lower; three subjects). Daitch et al. www.pnas.org/cgi/content/short/13794711 5of8

Fig. S6. ITC by frequency. Shown are curves of the average ITC at each frequency, averaged separately over electrodes in each functional network. Note that in different epochs, ITC is more prominent at different frequencies. For example, during the delay period, electrodes across all networks exhibit a peak in ITC around 2 Hz. Fig. S7. Spatial distribution of 2-Hz phase-locking. Shown are 2-Hz phase-locking maps from two example subjects, with seed regions marked in white. Note that during the ITI, phase-locking is mostly local, whereas during the trial, seed regions become phase-locked with electrodes over many other regions of interest. FEF, frontal eye field; IPS, intraparietal sulcus; MT, middle temporal region. Daitch et al. www.pnas.org/cgi/content/short/13794711 6of8

A VAN DMN VAN DMN Delta (1-3 Hz) Theta (3-7 Hz) Alpha (7-15 Hz) Cue - ITI Delay - ITI Target - ITI Response - ITI Beta (15-32 Hz) Low gamma (32-55 Hz) High gamma (7-15 Hz) % PLV change from ITI 2% -2% B All electrode pairs Within Within VAN Within MOT Within DMN C All electrode pairs Cue Delay Target Response 1 2 3 4 5 6 7 8 9111 Inter-electrode distance (mm) % PLV change from ITI 2 1-1 -2 Cue Delay Target Response/ITI r =.44514, p = 9.6988e-7 2 4 6 8 1 12 Inter-electrode distance (mm) 2 1-1 -2 r =.42528, p = 2.8834e-6 2 4 6 8 1 12 2 1-1 -2 * p<.1, Bonferroni corrected r =.15564, p =.86351 2 4 6 8 1 12 2 1-1 -2 r = -.1452, p =.1136 2 4 6 8 1 12 % PLV change from ITI 1-1 Within 2 1-1 -2 r =.16292, p =.75286 2 4 6 8 1 12 2 1-1 -2 r =.2593, p =.69433 2 4 6 8 1 12 2 1-1 -2 r =.11912, p =.51418 2 4 6 8 1 12 2 1-1 -2 r =.3492, p =.61921 2 4 6 8 1 12 Within VAN 2 1-1 -2 r = -.96592, p =.222 2 4 6 8 1 12 2 1-1 -2 r = -.1584, p =.36181 2 4 6 8 1 12 2 1-1 -2 r = -.15482, p =.426 2 4 6 8 1 12 2 1-1 -2 r = -.57554, p =.4482 2 4 6 8 1 12 Within 2 1-1 -2 r =.11369, p =.22228 2 4 6 8 1 12 2 1-1 -2 r =.11344, p =.22331 2 4 6 8 1 12 2 1-1 -2 r =.26336, p =.41192 2 4 6 8 1 12 2 1-1 -2 r =.2999, p =.1537 2 4 6 8 1 12 Within DMN 2 1-1 -2 r =.12419, p =.11885 2 4 6 8 1 12 2 1-1 -2 r =.12348, p =.1299 2 4 6 8 1 12 2 1-1 -2 r =.12824, p =.172 2 4 6 8 1 12 2 1-1 -2 r =.15653, p =.48791 2 4 6 8 1 12 Fig. S8. (A) Average phase-locking values within and between networks. Percent change in the average PLV between pairs of electrodes in different networks in different epochs of the task relative to the ITI. Starred squares represent network pairs where there was a significant difference between phase locking in that epochs versus the ITI (paired t test <.1, corrected for number of frequency bands, number of epochs, and number of epoch pairs). (B) Average PLV change from the ITI to each task epoch, as a function of interelectrode distance, separated by task epoch and network (only electrode pairs within the same network are shown here). Note that many of the largest PLV increases occur between more distant electrode pairs. (C) Relationship between task-related PLV change and interelectrode distance (raw data shown here that is used to compute average PLV by interelectrode distance in B). Top/first row: Scatterplots showing, for all electrode pairs, the relationship between the percent change in PLV in each 5-ms trial epoch relative to the ITI and interelectrode distance. Legend continued on following page Daitch et al. www.pnas.org/cgi/content/short/13794711 7of8

Bottom four rows: Scatterplots showing the relationship between percent PLV change from ITI and interelectrode distance, considering only electrode pairs within each of the four functional brain networks studied here. The line of best fit for each scatterplot is displayed in red, and the r and P values for each correlation (between percent PLV change from ITI and interelectrode distance) are shown at the bottom of each scatterplot. Mean ITC.2.1 Cue onset const. delay var.delay Delta (1-3 Hz) 5 ms const. delay var.delay Fig. S9. δ ITC averaged over electrodes in the and following the cue in the constant-delay version of Posner (light lines) and the variable-delay version (dark lines). For the variable-delay trials, only the first 5 ms of the delay is plotted here. Note that when the delay is variable, δ ITC is not sustained following the cue, as it is when the delay is constant. Shaded regions represent SEMs. Table S1. Posner task accuracy across subjects Subject % Correct (valid trials) % Correct (invalid trials) % Correct (all trials) PT1 9.8 86.7 9. PT2 94.8 91.2 94.2 PT3 85.2 88.4 85.8 PT4 93.4 93.1 93.3 PT5 96.9 98.6 97.2 PT6 94.5 95.9 94.8 PT, patient. Table S2. Posner task reaction times across subjects Patient RT (ms; valid trials) RT (ms; invalid trials) RT (ms; all trials) PT1 933 978 942 PT2 872 941 886 PT3 755 892 795 PT4 818 865 827 PT5 95 96 952 PT6 85 81 86 PT, patient; RT, reaction time. Daitch et al. www.pnas.org/cgi/content/short/13794711 8of8