Fig 1.
Analytic distinction of periodic and aperiodic temporal structures in simulations.
(A) Simulation of a sine wave with different frequency and amplitude peaks. Left: Strong correlation detected between timescales and peak frequency. Right: No correlation observed between timescales and peak amplitude. (B) Simulation of aperiodic 1/f noise with variables exponents and amplitude. Left: Timescales and 1/f exponent were strongly correlated. Right: No correlation observed between 1/f noise amplitude and timescales. (C) Simulation of several timeseries that exhibit both periodic and aperiodic temporal structures with different 1/f exponents, frequencies, and amplitudes. Upper left: No statistically significant correlation was observed between timescales and peak frequency of 1,000 timeseries with different exponents and oscillations (r = 0.01, p = 0.8305, BF10 = 0.045). Upper right: Strong significant correlation was identified between timescales and 1/f noise (r = −0.46, p < 0.0001). Lower left: A moderate correlation between timescales and peak amplitude was observed (r = −0.12, p = 0.0001). Lower right: Bootstrapped and subsample correlation coefficients (100 iterations) reveal a significant difference between the parameters amplitude, frequency and 1/f exponent (F = 406.74, p < 0.0001). (D) Simulation of random walk signal that contains no periodic activity patterns (lower left), but exhibits a characteristic timescale that can be inferred from the autocorrelation function (lower right; red for represents timescale). Abbreviations: Freq: Frequency. Amp: Amplitude. Exp: Exponent. PSD: Power spectral density. FFT: Fast Fourier Transform. ACF: Autocorrelation Function. The individual values for panel C are included in S1 Data.
Fig 2.
Behavioral timescales increase when multiple locations are sampled.
(A) Schematic of task designs. Participants fixated a central cross and were presented with a cue, which indicated the location participants should covertly attend to. After a variable cue-target interval a target appeared in either the cued or non-cued location and participants responded with a button press. In the first task participants only had to sample two locations, while in the second task, participants had to sample four locations. (B) Left: demeaned, time-resolved RTs as a function of the cue-target interval for one exemplary participant (two locations: red; four locations blue). Middle: power spectrums with different peak frequencies. Right: the autocorrelation function and the respective timescales. (C) Correlation between timescales and behavior power per frequency (four location task). (D) Behavior power did not correlate with behavioral timescales (no cluster identified; red lines represent significance threshold; four location task; comparable results for two location task). (E) Timescales are longer as well as more variable when multiple locations need to be sampled in comparison to only two locations (p = 0.0359). The individual values for panel E are included in S2 Data, Experiment 1 sheet.
Fig 3.
Neural timescales predict behavior.
(A) Behavioral timescales do not differ between different groups of participants doing the same task (two locations) (p = 0.8501). (B) Grand-averaged target-locked neural timescales (mean ± SEM; shaded gray area represents cue-target interval) throughout the trial for four example channels. Topography of timescales during cue-target interval averaged across all subjects, channels Fz, Cz, Pz, and Oz highlighted. (C) Neuronal timescales correlate with reaction times (averaged target-locked timescales across the cue-target interval (from −1 s to 0 s), error bars correspond to channels; the gray line highlights the linear regression; pcluster = 0.0040). Inset: topography indicates spatial extent of the effect (black dots indicate cluster electrodes). The individual values for panel A are included in S2 Data, Experiment 1 and 2 sheets.
Fig 4.
ECoG timescales decrease during spatial attention.
(A) Top: behavioral timescales do not differ between different groups of participants (exp. 1 and 3/4) doing the same task; Bottom: Illustration of ECoG electrode placement for all subjects for both two location (left) and four location (right) tasks for the frontal, motor and parietal regions. (B) Grand-averaged target-locked neural timescales (mean ± SEM; shaded gray area represents cue-target interval) for three ROIs. Bottom left: Topography of timescales during the cue-target interval (white dots indicate the ROIs). (C) Top: example of selective channel for left hemifield-cued trials to illustrate analysis approach to identify spatially selective channels. If power was above the threshold (z > 1.96) for a minimum of 10% of the time window (500 ms) after the cue, the channel was considered hemifield-selective. Bottom: timescales significantly decrease during covert attention relative to the attend-out condition (two locations: p = 0.0244; four locations: p < 0.0001; mean ± SEM; whiskers indicate maximum and minimum; dots correspond to individual electrodes). The individual values for panel A and C are included in S2 Data, Experiment 1, 3 and 4 sheets and S3 Data, respectively.
Fig 5.
Neural timescales increase with cognitive demands.
(A) Schematic task design. Patients were instructed to search for a target triangle (“sample”, highlighted by dashed circle) of a given color and orientation. Once they identified the target triangle, they responded with a button press indicating whether its location was on the left or right half of the screen. (B) Reaction times during pop-out condition are shorter than during visual search condition (p < 0.0001). Individual dots correspond to trials. (C) Example of autocorrelation functions for each condition calculated on response-locked data. We hypothesized pop-out condition would show shorter timescales than the search condition. (D) Left: Grand-averaged response-locked neural timescales across frontal channels. Timescales in the period following search display onset (red dashed line) and after button press (black dashed line) decrease in pop-out condition relative to visual search condition returning to baseline around 1s (black bar represents significant time points p < 0.05). Right top: Topography of iEEG electrode placement in frontal ROI. Right bottom: Averaged timescales in the 500 ms period between search display onset and button press (gray shaded area) significantly differ between the conditions (p < 0.0001). The individual values for panel B and D are included in S2 Data, Experiment 5 sheet and S4 Data, respectively.