Shaping Information Processing: The Role of Oscillatory Dynamics in a Working Memory Task

Abstract

Neural oscillations are thought to reflect low-level operations that can be used for higher-level cognitive functions. Here, we investigated the role of brain rhythms in the 1-30 Hz range by recording MEG in human participants performing a visual delayed match-to-sample paradigm in which orientation or spatial frequency of sample and probe gratings had to be matched. A cue occurring before or after sample presentation indicated the to-be-matched feature. We demonstrate that alpha/beta power decrease tracks the presentation of the informative cue and indexes faster responses. Moreover, these faster responses coincided with an augmented phase alignment of slow oscillations, as well as phase-amplitude coupling between slow and fast oscillations. Importantly, stimulus decodability was boosted by both low alpha power and high beta power. In summary, we provide support for a comprehensive framework in which different rhythms play specific roles: slow rhythms control input sampling, while alpha (and beta) gates the information flow, beta recruits task-relevant circuits, and the timing of faster oscillations is controlled by slower ones.

Keywords: MEG; alpha rhythm; beta rhythm; brain oscillations; working memory.

Figure 1.

Paradigm and behavioral results. A, Schematic of the four trial types: attend orientation (two solid lines) or attend frequency (two dotted lines) for precue and retro-cue conditions (first vs second cue is informative; uninformative cues consist of a solid and dotted lines). B, C, Participants were slower (B) and more accurate (C) matching the spatial frequency (gray bars) of the visual gratings compared with orientation (orange). This effect was mainly driven by the precue condition. Within each boxplot, the horizontal line represents the median, the box delineates the area between the first and third quartiles (interquartile range). D, Average accuracy across RT bins. Error bars represent SEM. **p < 0.01, ***p < 0.001.

Figure 2.

Alpha and beta peak detection. A, Grand-averaged ERF of sensors (highlighted in sensor space) displaying maximal postgrating ERFs within 200 ms following the grating onset. Gray shaded area around the curve denotes between-participants SE. Light gray shaded box highlights the time period used to select channels with maximal postgrating ERFs. B, Power spectra (averaged over same sensors marked in A; t = −1 to 0 s relative to first cue onset) showing alpha and beta peaks.

Figure 3.

Decodability of task features. A, B, Temporal sample-by-sample decoding of cue features (attend frequency vs attend orientation) for precues (A) and retro-cues (B). C, D, Decoding performance for grating spatial frequency (low vs high; C) and grating orientation (clockwise vs counterclockwise; D). Gray bars indicate significance of decoding accuracy (t test vs chance).

Figure 4.

Cue-related power modulations. A, Grand average ERFs for the precue (green) and retro-cue conditions (purple). Shaded areas denote between-participants SE. Gray bars indicate significant differences between conditions. Topographies show statistical and power distributions of these significant differences in sensor and source space, respectively. B, Intertrial phase coherence (averaged between 1 and 6 Hz) for the precue and retro-cue conditions. C, Time course of oscillatory power averaged within the alpha band. D, Same as C but for the beta band. For all time courses, all sensors displaying significant differences as highlighted by the topographies were included in the plot.

Figure 5.

Correlates of behavioral performance: power and phase. A, Grand average ERFs (occipital sensors) for the slowest (red) and fastest (blue) RT bins. Shaded areas denote between-participants SE. Gray bars indicate significant differences between conditions. Topographies show statistical and power distributions of these significant differences in sensor and source space, respectively. B, Time course of intertrial phase coherence (averaged between 1 and 6 Hz) for the slowest and fastest RT bins. C, Time course of oscillatory power averaged within the alpha band. Dashed lines represent mean power (i.e., normalized power = 1). D, Same as C but for the beta band.

Figure 6.

Correlates of behavioral performance: ITC and PAC. A, TFR of statistical differences in PAC between the slowest and fastest RT bins (masked at p < 0.05) in the intraparietal sulcus (as highlighted in green on brain surface). Negative values (blue) indicate higher PAC for the fastest RT bin. B, Same as A for the superior parietal lobe.

Figure 7.

Correlates of behavioral performance: ERFs. A, Source-level topographies of the statistical differences between ERF and ITC for the cueing contrast (precure vs retro-cue; t = 3.6–4.1 s; masked at p < 0.05). B, Same as A for the RT contrast (slow vs fast RT bin; t = 4.4–5.1 s).

Figure 8.

Effects of prestimulus oscillatory power on decodability. A, Temporal sample-by-sample decoding of precue (left) and retro-cue (right) instruction for the bins of lowest and highest occipital alpha power preceding cue onset. B, Same as A for the beta band. C, Averaged time-generalization matrices (over the first 0.3 s of training time) of decoding the orientation of the sample (left) and frequency (right) for the bins of lowest and highest beta power preceding sample onset. Gray bars indicate significant differences between low-power and high-power bins.