Oscillatory Bursting as a Mechanism for Temporal Coupling and Information Coding

Abstract

Even the simplest cognitive processes involve interactions between cortical regions. To study these processes, we usually rely on averaging across several repetitions of a task or across long segments of data to reach a statistically valid conclusion. Neuronal oscillations reflect synchronized excitability fluctuations in ensembles of neurons and can be observed in electrophysiological recordings in the presence or absence of an external stimulus. Oscillatory brain activity has been viewed as sustained increase in power at specific frequency bands. However, this perspective has been challenged in recent years by the notion that oscillations may occur as transient burst-like events that occur in individual trials and may only appear as sustained activity when multiple trials are averaged together. In this review, we examine the idea that oscillatory activity can manifest as a transient burst as well as a sustained increase in power. We discuss the technical challenges involved in the detection and characterization of transient events at the single trial level, the mechanisms that might generate them and the features that can be extracted from these events to study single-trial dynamics of neuronal ensemble activity.

Keywords: bursts; methods; oscillations; single trial; timing; transients.

Figure 1

Noise influence on time-frequency profiles. The top panel shows a trace of a pure sinusoid at 110 Hz with a constant amplitude lasting for a duration of ~400 ms. The bottom panels show the time-frequency representation of the signal (top insets) and the time-domain trace (bottom insets) after adding different levels of white Gaussian noise. As the noise level increases (lower SNR), the time-frequency profile contains more gaps and becomes more “bursty” in appearance. In cases of low SNR, such as the ones that might be observed in single trial or ongoing data, sustained oscillations might be broken down into isolated peaks and thus might be considered as transient bursts that might even slightly vary in frequency as a result of the noise structure.

Figure 2

Examples of ongoing oscillatory activity. (A) Top and middle panels show a trace of invasive recordings from sEEG electrodes showing low-frequency oscillations (0.5–4 Hz, top; 8–12 Hz, middle). (Bottom) Laminar probe recording from a non-human primate showing bursts of oscillatory activity at a frequency of 13–30 Hz. Gray rectangles indicate the detection of transient oscillatory bursts. (B) Descriptive analysis of the duration of oscillatory activity at each frequency band. The duration of oscillations (top) was estimated as the period of time in which the power at that frequency band exceeded the 95th percentile of the theoretical χ² distribution of wavelet power values. The number of cycles (bottom) were calculated by multiplying the duration (in seconds) by the peak frequency of the oscillation (in Hz). Lower frequency oscillations (i.e., 0.5–4 and 8–12 Hz) tend to show longer durations compared with the higher frequency oscillations (13–30 Hz) in which most oscillatory bursts consisted of ~3.5 cycles. The inset shows a zoomed version to visualize the differences between 0.5–4 and 13–30 Hz. Error bars indicate standard error of the mean.

Figure 3

Neocortical circuit model used to simulate bursty gamma oscillation events through a weak PING mechanism. (A) Raster spiking plot of neuronal firing times from single trial of weak PING simulation. Top panel shows histogram of low-frequency, noisy Poisson inputs used to drive pyramidal neurons and interneurons in the model. Bottom panel shows population color-coded firing times of individual neurons. (B) Single trial current dipole signal (top) and Morlet wavelet spectrogram from dipole signal (bottom) from weak PING model. (C) Simulation of one hundred trials of weak PING model produces one hundred current dipole signals (top). Although gamma oscillation events occur at different times in each trial, averaging the wavelet spectrogram across trials (bottom) produces an appearance of a sustained gamma oscillation [adapted with permission from Figure 10 of Neymotin et al. (2020b) under the license: https://creativecommons.org/licenses/by/4.0/].