Two signatures of visual responses measured with ECoG.

(a) Mapping stimuli. Three examples of the bar stimuli and one image of the blank stimulus during the mapping experiment (brown and black outlines, respectively). The contrast within the bar is increased here for visibility. (b) A close-up view of the right occipital cortex of Patient 2. The surface colors show several visual field maps derived from an atlas, with V1 in red, V2 in orange, and V3 in yellow. Black circles indicate ECoG electrode locations. (See Supplementary Figure 1-1 for all patients) (c) The voltage signal during a single trial with a stimulus present for electrode Oc17. The thick and thin lines are voltages before and after regressing out the ERP, respectively. Note the sharp high frequency oscillations in both traces, and the lack of a clear alpha oscillation. (d) Same but for a single blank trial. Here, the two signals are nearly identical since there is no ERP during a blank trial. Note the strong alpha oscillation. (See Supplementary Figure 1-2 for the effect of ERP removal processing in pRF estimation.) (e) The power spectral density (PSD) was computed based on the regressed signal, averaged across all blank trials (black) or all stimulus trials (brown). The stimulus causes a broadband power increase, spanning frequencies from 1 Hz to 150 Hz. The dotted black rectangle indicates the frequency range shown in zoom in panel f. (f) During the blank, but not during visual stimulation, there is a peak in the spectrum at around 13 Hz (alpha oscillation, dashed vertical line). Shading represents the standard error across trials (320 trials for stimulus, 128 trials for blank). See makeFigure1.m.

Interaction between the alpha rhythm and broadband power.

(Left column) PSDs from 3 electrodes in Patient 2 without baseline correction show that the stimulus-related alpha power can slightly increase (Oc18), slightly decrease (Oc17) or substantially decrease (Oc25) relative to a blank stimulus. (Right column) After correcting for the broadband shift, it is clear that for all three electrodes, the alpha oscillation is suppressed by visual stimulation. The dashed vertical lines indicate the alpha peak frequency. Shading represents the standard error across trials (320 trials for stimulus, 128 trials for blank). See Methods and makeFigure2.m.

Example time series and pRF fits.

Representative data from a V3 electrode (‘GB103’) from Patient 8. (a) The computation of broadband and alpha summary metrics for an example stimulus location. The power spectra were computed over 500 ms of stimulus presentations or blanks. The dashed brown line in the lower panel indicates spectral power for the stimulus after correcting for broadband shift. Shading represents the 68% bootstrapped confidence intervals across the 6 runs for STIMULUS and across 384 trials for BLANK, each with 1,000 resamples. (b) The time series of the broadband and alpha summary metrics averaged across six 190-s mapping runs, during which bar stimuli swept the visual field 8 times (with 4 blank periods indicated by the gray bars). Each data point (black circle) is a summary value —either broadband or alpha— for one stimulus position (Equations 1–3). Predicted responses from a pRF model are shown as solid lines. Model predictions below baseline in broadband and above baseline in alpha were possible because of the Difference of Gaussians (DoG) pRF model (see Supplementary Figure 3-1). (c) The location of the pRFs fitted to the broadband and alpha times series. The circles show the 1-sigma pRF bootstrapped 100 times across the 6 repeated runs. See makeFigure3.m.

Prediction accuracy of alpha pRF models in V1–V3.

The variance explained by alpha pRF models was higher in visually selective than non-visually selective electrodes. The left panel shows the results pooled across patients. The right side shows results from each patient separately. Two patients who didn’t have electrodes in V1– V3 are not displayed. Visually selective electrodes were defined as those whose broadband pRF model accurately predicted the broadband time course (see Supplementary Figure 4-1). A similar pattern was observed for Dorsolateral maps beyond V1–V3 (Supplementary Figure 4-2). Electrodes were assigned to visual field maps probabilistically across 5,000 samples. Data points and error bars represent means ±1 standard deviation of the mean across the 5000 samples. An open dot includes only one electrode. See makeFigure4_4S1_4S2.m.

Relationship between alpha and broadband pRFs in V1–V3.

(a) pRF locations in V1–V3 displayed for each electrode, as in Figure 3. For visualization purposes, only electrodes that are more likely to be assigned to V1–V3 than any other map are plotted, even though the probabilistic assignment (e.g. for panel c) included additional electrodes. (b) pRFs were normalized by rotation (subtracting the broadband pRF polar angle from both the broadband and alpha pRF) and then scaling (dividing the eccentricity and size of both pRFs by the broadband pRF eccentricity). This puts the broadband pRF center at (0,1) for all electrodes, indicated by an ‘x’. Within this normalized space, the average pRF across electrodes was computed 5,000 times (bootstrapping across electrodes). 100 of these averages are indicated as dotted lines, and the average across all 5,000 bootstraps is indicated as a solid line. (c) Each panel shows the relation between broadband and alpha pRF parameters in 2-D histograms. 53 electrodes were randomly sampled with replacement 5,000 times (bootstrapping), and each time they were selected they were probabilistically assigned to a visual area. The colormap indicates the frequency of pRFs in that bin (see Comparison of alpha and broadband pRFs in Methods 4.7). Note that the phases in the horizontal and vertical axes are wrapped to show their circular relations. The middle and right panels show the relation of eccentricity and pRF sizes, respectively. Yellow ellipses indicate the 1-sd covariance ellipses. (d) The plot shows pRF size vs eccentricity, both for broadband (blue) and alpha (red). As in panel c, the 53 electrodes were bootstrapped 5,000 times. The regression lines are the average across bootstraps, and the shaded regions are the 16th to 84th percentile evaluated at each eccentricity. Electrodes that are more likely to be assigned to V1–V3 than any other maps are plotted as dots. (see Supplementary Figure 5-1 for electrodes in dorsolateral maps). See makeFigure5_5S1.m.

PRF sizes in broadband and alpha.

Violin plots of pRF sizes with 5,000 bootstraps, estimated from high-frequency broadband, low-frequency broadband, and alpha. Black horizontal lines and dark color regions within violins indicate means ±1 standard deviation of the mean across the 5000 samples. A dotted horizontal line indicates the mean of high-frequency broadband pRF sizes. For the full power spectra showing the broadband and alpha responses, see Supplementary Figure 6-1. See makeFigure6_6S2.m.

The advantage of baseline correction on alpha pRFs.

We compared pRF solutions for alpha using our spectral model (“baseline corrected”) vs by computing power within the alpha band (‘no correction”). Left: Variance explained. The white dots are the median and the gray bars are the interquartile range. Each colored dot is one electrode and the shaded regions are smoothed histograms. Middle: Response gain, plotted in the same manner as variance explained. Gain is quantified as the maximum or minimum value in the pRF fitted time series (maximal for electrodes with positive gain, minimal for electrodes with negative gain). A value above 1 means that the response increased relative to baseline (positive gain). A value below 1 means the response decreased (negative gain). Right: pRF size vs eccentricity. Electrodes are colored by the sign of the gain (filled for negative, open for positive). These tendencies are consistent across individual patients (Supplementary Figure 7-1). The shadings represent 68% bootstrapped confidence intervals with 1,000 resamples. See makeFigure7_7S1.m.

Inter-electrode coherence across distances in high-density grids.

Coherence between electrode pairs was computed within each 1-s epoch and then averaged epochs. Seed electrodes were those which had a reliable pRF fit, and these seed electrodes were compared to all other electrodes on the grid. The curved lines indicate the average fit of an exponential decay function across 5,000 bootstraps; shading indicates ±1 standard deviation of bootstraps; dashed lines indicate the baselines the coherence converges to. See Supplementary Figure 8-3 for coherence as a function of frequency. See makeFigure8.m.

Schematics of neural excitation and excitability changes induced by visual stimulation.

The upper panels show stimulus position relative to the broadband and alpha pRFs for a given electrode. The lower panel shows predicted alpha and broadband responses from that electrode, with stimulus onset indicated by the black horizontal bar. In the left and middle panels, the stimulus causes the alpha oscillations to be suppressed and cortex to become more excitable (red shading). In the right panel, the alpha oscillation increases slightly and cortex becomes less excitable (gray shading).

Asymmetric profile of alpha pRF and asymmetric effect of spatial attention.

The left panel shows an asymmetric profile of alpha pRF across normalized eccentricity in V1–V3. The right panel shows asymmetric effects of attention on normalized cue–target locations. Black line represents response time changes induced by endogenous attention from Downing and Pinker 30, relative to the response time without any attentional cue. The gold line represents response time changes induced by exogenous attention from Shulman et al. 29, relative to the response time when targets appeared at the attended locations. Results from the Shulman et al. paper with foveal cue locations at 0.5 degree eccentricity were excluded from the plot because after normalizing, the data from this condition has no overlap with the other conditions perhaps because at such a small eccentricity, a minor error in attended location makes a large difference on a normalized scale. The experimental results for each cue location are shown in Supplementary Figure 10-1. See makeFigure10.m.

Overview of patient data included in this dataset.

The patient number and subject code for BIDS do not always match because the 9 subjects with pRF data are a subset of 11 subjects from a larger BIDS dataset (https://openneuro.org/datasets/ds004194). Two of the 11 patients did not take part in pRF experiments.

Schematics of model-based alpha computation.

The left panel shows the average power spectra elicited by the pRF mapping stimuli (brown) and blanks (black) from a representative electrode. The middle panel shows the ratio between the two power spectra (green). This ratio, evaluated on log-log axes, is modeled (dashed red) as the sum of a broadband elevation (right panel, top) and an alpha suppression (right panel, bottom). The magnitude of alpha suppression is the coefficient of the Gaussian bump (βalpha, black arrow). See Equation 1 and makeFigure11.m.

Overview of electrodes and patients per visual area and figure.

Electrodes in visual areas (rows 1-3) indicate electrodes with > 0% probability of being in an either V1 to V3 or dorsolateral maps, irrespective of the accuracy of pRF fits. Electrodes with broadband pRFs (rows 4-6) indicate the subset of electrodes from rows 1-3 exceeding a threshold variance explained by the broadband pRF model. Electrodes with broadband and alpha pRFs (rows 7-9) are the subset of electrodes from rows 4-6 satisfying selection criteria both for broadband and alpha pRFs (exceeding threshold variance explained and a pRF center within the stimulus aperture). The totals for rows “V1–V3 or dorsolateral” are greater than the sum of “V1-V3” and “dorsolateral” because some electrodes have greater than 0% probability of being included in both groups (see Electrode Localization, section 4.6). Any electrode with greater than 95% chance of being assigned to no visual area in the atlas was excluded from the table entirely.

Electrode locations and visual areas.

The electrode locations and visual areas on brain surfaces for each patient. The retinotopic atlas is modified from Wang et al. (2015). Only the cortices with electrodes are shown: left cortex for Patient 1, 4, 5, and 9; right cortex for Patient 2, 3, 6, and 8; both cortices for Patient 7. See makeFigure1S1.m

pRF model comparisons with and without ERP-Regression Removal.

(a) The broadband and alpha response amplitudes, along with the estimated pRF locations, are shown for three representative electrodes depicted in Figures 1 and 2. These summary metrics are little affected by ERP regression. (b) Except for a handful of electrodes in Dorsolateral maps, the cross-validated variance explained in the pRF models is nearly identical for broadband with vs without ERP regression. When the regression has a large impact, it tends to increase variance explained. There is a larger effect of ERP regression on the alpha pRF, but on average, the variance explained is the same without vs without regression. See makeFigure1S2.m

PRF model comparisons between two Gaussian models.

Cross-validated variance explained in pRF models are almost the same between two Gaussian models (Difference of Gaussians VS One Gaussian) both in V1–V3 (blue circle) and dorsolateral (purple triangle), but the Difference of Gaussians shows advantage in some electrodes. Each dot indicates electrode which is randomly selected according to the visual area probabilities in 5,000 bootstraps. See makeFigure3S1.m

Distributions of Variance Explained in alpha and broadband pRFs.

The distributions of variance explained in alpha and broadband pRFs from all visual areas show rightward tendency compared to shuffled distributions. (A negative variance explained is common for cross-validation when making predictions with an inappropriate model.) The medians shown as black arrows have larger values in the raw distributions than the shuffled distributions (24% vs −21% in broadband and 5% vs −18% in alpha), suggesting the prediction accuracy of the pRF model fit is meaningful both in broadband and alpha. The false positive criterions were set at 95% of the values in this null distribution (dashed line: 31% for broadband and 22% for alpha). The electrodes shown in the right side of the false positive criterion (dark gray bars) were included for the comparison between broadband and alpha pRFs. in broadband pRFs are the visually selected electrodes. The electrodes shown in the right side of the false positive criterion in alpha pRFs are considered as the alpha sensitive electrodes. See makeFigure4_4S1_4S2.m.

Prediction accuracy of alpha pRF model in dorsolateral.

The average two-fold cross-validated variance explained by alpha pRF models was higher in the visually selective (blue) than in the non-selective electrodes (red) except for individual patients with only 1 to 2 visually selective electrodes. The left side of the plot shows the results pooled across all patients. The right side shows the results from each patient separately. Visually selective electrodes were defined as those whose broadband pRF model accurately predicted the broadband time course. Dorsolateral includes the V3A/B, LO-1/2, TO-1/2, and IPS maps. were assigned to visual field maps probabilistically across 5,000 samples. Data points and error bars represent means ±1 standard deviation of the mean across the 5000 samples. Open dots include only one electrode each. See makeFigure4_4S1_4S2.m.

Relations of alpha and broadband pRFs in dorsolateral.

(a) pRF locations in dorsolateral maps. For visualization purposes, only electrodes that are more likely to be assigned to dorsolateral than to V1 to V3 are plotted, even though the probabilistic assignment included additional electrodes that are not displayed here. (b) pRFs were normalized by rotation (subtracting the broadband pRF polar angle from both the broadband and alpha pRF) and then scaling (dividing the eccentricity and size of both pRFs by the broadband pRF eccentricity). This puts the broadband pRF center at (0,1) for all electrodes, indicated by an ‘x’. Within this normalized space, the average pRF across electrodes was computed 5,000 times (bootstrapping across electrodes). 100 of these averages are plotted as dotted lines, and the average across all 5,000 bootstraps is plotted as a solid line. (c) Each panel shows the relation between broadband and alpha pRF parameters in 2-D histograms. 53 electrodes were randomly sampled with replacement 5,000 times (bootstrapping), and each time they were selected they were probabilistically assigned to a visual area. The colormap indicates the frequency of pRFs in that bin (see Comparison of alpha and broadband pRFs in Methods 4.7). Note that the phases in the horizontal and vertical axes are wrapped to show their circular relations. The middle and right panels show the relation of eccentricity and pRF sizes, respectively. Yellow ellipses indicate the 1-sd covariance ellipses. (d) The plot shows pRF size vs eccentricity, both for broadband (blue) and alpha (red). As in panel c, the 53 electrodes were bootstrapped 5,000 times. The regression lines are the average across bootstraps, and the shaded regions are the 16th to 84th percentile evaluated at each eccentricity. Only electrodes that are more likely to be assigned to V1–V3 are plotted. See makeFigure5_5S1.m.

Estimated pRF size versus cross-validated variance explained.

(a) The plot illustrates the relationship between pRF size and variance explained by the pRF model for V1 to V3 maps. The red dots are for the alpha pRF model, and the red line is the best fit regression line. Broadband pRF models are shown by blue dots (high frequency broadband) and green dots (low frequency broadband). The blue regression line was fit to the combination of high and low frequency broadband. The plot shows that pRF size is larger for alpha than for broadband, irrespective of variance explained. Had the larger pRFs in the alpha model been due to greater noise, than a single regression line would fit all the data points. Error shading represents the 16th to 84th percentiles derived from 5,000 bootstrapping iterations. (b) Same as A except for dorsolateral maps. Here, low frequency broadband models are not included because there was little low frequency broadband signal in these electrodes. See makeFigure5S2.m

Power in the broadband response and alpha oscillation.

Each panel shows the spectral power changes averaged across all electrodes between BLANK and in-pRF stimulation (STIMULUS) conditions in V1–V3 and dorsolateral. (a) PSD for in-pRF (brown line) is higher than for BLANK (black line) and across all frequencies including not only high frequencies (blue region) but also low frequencies (green region) in V1–V3. In dorsolateral, on the other hand, PSD for in-pRF is higher than for BLANK only in high frequencies. Gaussian bumps at peak alpha frequencies (red dashed lines) are observed only in BLANK condition. (b) The ratio of PSDs between in-pRF and BLANK is positive in high-frequency broadband and negative at alpha frequency consistently in V1–V3 and dorsolateral. The ratio in low-frequency broadband is also positive in V1–V3 but is almost zero in dorsolateral. See makeFigure6S1.m.

Low broadband model comparisons.

(a) PRFs for low frequency broadband (3–26 Hz) and high frequency broadband (70–180 Hz) in V1–V3 electrodes. Plotted as in Figure 5A. (b) PRF parameters are highly correlated between the high-frequency broadband estimates (x-axis) and the low-frequency broadband estimates (y-axis) in V1–V3. Yellow ellipses indicate the 1-sd covariance ellipses. See makeFigure6_6S2.m.

A comparison of alpha pRFs with and without baseline correction in individual patients.

We compared alpha pRF solutions that were baseline corrected (using our spectral model) or not baseline corrected (computing power within the alpha band). Each row is a separate participant. Left: Variance explained. The white dots are the median and the gray bars are the interquartile range. Each colored dot is one electrode and the shaded regions are smoothed histograms. Middle: Response gain, plotted in the same manner as variance explained. Gain is quantified as the maximum or minimum value in the pRF fitted time series (maximal for electrodes with positive gain, minimal for electrodes with negative gain). A value above 1 means that the response increased relative to baseline (positive gain). A value below 1 means the response decreased (negative gain). The baseline correction resulted in the expected negative gain. Right: pRF size vs eccentricity. Electrodes are colored by the sign of the gain (filled for negative, open for positive gain). The baseline corrected data shows the expected pattern of increasing pRF size with eccentricity. See makeFigure7_7S1.m.

PRF parameters from high-density grid in Patient 8. (Upper left)

The electrode locations and visual areas on the brain surface from Patient 8. (Upper right) The visual area assignments of electrodes from the high-density grid. White spaces indicate absence of electrodes. (Second row) Two-fold cross-validated variance explained in pRF fitting for broadband and alpha across electrodes in the high-density grid. Electrode locations which were excluded in the preprocessing are shown as white. (Third and Fourth rows) Polar angles and eccentricities of estimated pRFs. Electrodes which had low variance explained (under 31% and 22% for broadband and alpha, respectively) are shown as white. See makeFigure8S1_8S2.m.

PRF parameters from high-density grid in Patient 9. (Upper left)

The electrode locations and visual areas on the brain surface from Patient 9. (Upper right) The visual area assignments of electrodes from the high-density grid. White spaces indicate absence of electrodes. Gray spaces indicate electrodes which were not assigned to any visual areas. (Second row) Two-fold cross-validated variance explained in pRF fitting for broadband and alpha across electrodes in the high-density grid. Electrode locations which were excluded in the preprocessing are shown as white. (Third and Fourth rows) Polar angles and eccentricities of estimated pRFs. Electrodes which had low variance explained (under 31% and 22% for broadband and alpha, respectively) are shown as white. See makeFigure8S1_8S2.m.

Inter-electrode coherence in the high-density grids across 1–200 Hz.

Each curve indicates the coherence averaged across electrode-pairs, binned by inter-electrode distance. The coherence for electrode pairs more than 2 cm apart was used as a baseline, and subtracted from each curve. For both patients, the coherence is highest around 10 Hz, the frequency of the alpha oscillation. See makeFigure8S3.m.

Response time costs in previous spatial attention experiments.

Left panels represent response time changes induced by endogenous attention from Downing and Pinker (1985), relative to the response time without any attentional cue. Right panels represent response time changes induced by exogenous attention from Shulman et al. (1986), relative to the response time when targets appeared at the attended locations. Lower panels are the same results as top panels after normalizing target eccentricities by dividing by the cue eccentricities. Results from the Shulman et al. paper with foveal cue locations at 0.5 degree eccentricity were excluded from the plot because after normalizing, the data from this condition has no overlap with the other conditions. See makeFigure10S1.m.