Nonuniform high-gamma (60-500 Hz) power changes dissociate cognitive task and anatomy in human cortex

Abstract

High-gamma-band (>60 Hz) power changes in cortical electrophysiology are a reliable indicator of focal, event-related cortical activity. Despite discoveries of oscillatory subthreshold and synchronous suprathreshold activity at the cellular level, there is an increasingly popular view that high-gamma-band amplitude changes recorded from cellular ensembles are the result of asynchronous firing activity that yields wideband and uniform power increases. Others have demonstrated independence of power changes in the low- and high-gamma bands, but to date, no studies have shown evidence of any such independence above 60 Hz. Based on nonuniformities in time-frequency analyses of electrocorticographic (ECoG) signals, we hypothesized that induced high-gamma-band (60-500 Hz) power changes are more heterogeneous than currently understood. Using single-word repetition tasks in six human subjects, we showed that functional responsiveness of different ECoG high-gamma sub-bands can discriminate cognitive task (e.g., hearing, reading, speaking) and cortical locations. Power changes in these sub-bands of the high-gamma range are consistently present within single trials and have statistically different time courses within the trial structure. Moreover, when consolidated across all subjects within three task-relevant anatomic regions (sensorimotor, Broca's area, and superior temporal gyrus), these behavior- and location-dependent power changes evidenced nonuniform trends across the population. Together, the independence and nonuniformity of power changes across a broad range of frequencies suggest that a new approach to evaluating high-gamma-band cortical activity is necessary. These findings show that in addition to time and location, frequency is another fundamental dimension of high-gamma dynamics.

Figure 1.

Schematic representations of power change paradigms. A, A typical set of power spectral densities. P_Task(f) represents the task under observation, and P_ITI(f) represents the corresponding intertrial interval that is the basis for comparison. B, The normalized power spectrum defined in equation form and illustrated schematically. This method of normalization shows the direction and magnitude of power change with reference to a resting state over a range of frequencies. C, Schematic normalized spectra illustrating the hypothesis that high-gamma power change is uniform in nature. Low frequencies (μ, β, <30 Hz) tend to show power decreases for cognitive task while high frequencies have power increases. D, Schematic normalized spectra illustrating the hypothesis that high-frequency power change is nonuniform. Both spectra have power changes in specific bands that distinguish one cognitive task from another.

Figure 2.

Experimental setup and time-frequency analysis of ECoG signals. A, Photograph of implanted ECoG electrodes and corresponding localization on the MNI model brain. Filled electrodes represent electrodes of interest in C. B, The timing of the two different experimental paradigms. Single-word stimuli were presented either aurally or visually. Analysis windows for hearing and reading were cued to stimulus onset, preparation analysis windows were cued to stimulus offset, and windows for speaking were cued to voice onset detected from microphone signal (see Materials and Methods for time delay and duration). C, Typical time-frequency plots for the auditory repeat paradigm showing statistically significant (p < 0.001) coefficient of determination (R²) values. This region of 12 electrodes is highlighted as an exemplar region. All nonsignificant R² values are grayed out. Six electrodes of interest are numbered and correspond to the filled electrodes in A. The colored rectangles highlight notional analysis windows with interesting nonuniform power change patterns. Keys for analysis window rectangles and time-frequency plot template are shown to the right.

Figure 3.

Exemplar normalized spectra illustrate that high-gamma-band power changes are nonuniform and distinct between tasks. The blue center line is the mean normalized spectra across 71 trials. The shaded area encapsulates the 95% confidence intervals. Vertical dashed lines at 60, 100, and 250 Hz outline typical gamma-band analysis boundaries. Frequencies with normalized spectra greater than zero indicate behavior-induced power increases, whereas values less than zero reflect power decreases. Note that for all six cognitive tasks, the patterns of spectral power change are unique across a wide range of frequencies. Each cognitive task has different bandwidths of frequencies that are statistically different from rest, and in some bands the direction (sign) of power change between cognitive task reverses (e.g., 60–120 Hz hearing vs speaking after visual cue). The sharp downward spikes are the result of environmental noise components that do not change in magnitude between cognitive task and ITI.

Figure 4.

The dynamics of power in dissociation bands. A, Exemplar mean raw PSDs for rest and two cognitive tasks for N = 216 trials. H, Hearing; SA, speaking after auditory cue. Yellow bands around 80 and 288 Hz highlight high-gamma frequency bands that dissociate the two tasks. B, Mean normalized spectra (solid lines) with 99.9% confidence intervals (shaded areas) show differences in the spectral power change patterns as averaged over all trials. Nonoverlapping confidence intervals and a reversal in the relationship between power levels in the two highlighted frequency bands illustrate a pair of dissociation bands. Note also that behaviorally induced power change is significantly different from rest as high as 500 Hz. C, Bar plot with 99.9% confidence intervals illustrates the statistical significance of the difference in normalized power between the two cognitive tasks for the dissociation bands in B. This format is used for other subjects in Figure 5. D, Single-trial normalized spectra for the single electrode and two cognitive tasks in A–C illustrate the consistency of power change in the dissociation bands across trials. These plots show that normalized spectra in the dissociation bands are not dominated by outliers in any single trial. The color intensity outside of the dissociation bands is subdued to highlight activity in the bands under study. E, Time courses of power in the dissociation bands from A–C. The mean downsampled and normalized power level for each frequency with 95% confidence intervals shows that during over the course of the experiment, power levels in the two dissociation bands reverse.

Figure 5.

Exemplar dissociation bands for all six subjects. See Figure 4 for the derivation of individual bar plots. The electrode(s) of interest, confidence intervals (CIs), associated BA labels, and cortical location on the MNI model brain are shown above each bar plot for reference. Subject 2 did not have single electrodes with dissociation bands, and therefore the exemplar shows power change reversals between two electrodes during the same cognitive task. H, Hearing; PV, preparation after visual cue; SA, speaking after auditory cue; SV, speaking after visual cue.

Figure 6.

Summary of quantities of dissociation bands across the subject population. A, B, Each bar shows the number of single electrodes (A) or electrode pairs (B) with dissociation bands for each subject by p value. A, This chart quantifies the number of single electrodes with significant power changes in different high-gamma frequencies that dissociated two or more cognitive tasks. B, This chart quantifies the number of electrode pairs in which different high-gamma frequencies dissociated anatomic locations during the same cognitive task.

Figure 7.

Consolidated cortical activation plots for the population of six subjects. Positive numbers indicate percentage of electrodes with statistically significant (p < 0.001, FDR corrected for multiple comparisons) power increases. Negative numbers correspond to power decreases. Rows of activation plots correspond to cortical regions, and columns to cognitive tasks. Markers at 60, 100, and 250 Hz are typical gamma or high-gamma analysis boundaries. All subject electrodes for each cortical region of interest are plotted on the MNI model brain for reference. Multiple peaks per plot, shifts in percentage of cortex with significant power changes across frequency bands, and changes in bandwidths with significant power changes within cortical populations are all evidence of nonuniform power modulation in high-gamma bands (60–500 Hz).