Temporal Expectations Guide Dynamic Prioritization in Visual Working Memory through Attenuated α Oscillations

Abstract

Although working memory is generally considered a highly dynamic mnemonic store, popular laboratory tasks used to understand its psychological and neural mechanisms (such as change detection and continuous reproduction) often remain relatively "static," involving the retention of a set number of items throughout a shared delay interval. In the current study, we investigated visual working memory in a more dynamic setting, and assessed the following: (1) whether internally guided temporal expectations can dynamically and reversibly prioritize individual mnemonic items at specific times at which they are deemed most relevant; and (2) the neural substrates that support such dynamic prioritization. Participants encoded two differently colored oriented bars into visual working memory to retrieve the orientation of one bar with a precision judgment when subsequently probed. To test for the flexible temporal control to access and retrieve remembered items, we manipulated the probability for each of the two bars to be probed over time, and recorded EEG in healthy human volunteers. Temporal expectations had a profound influence on working memory performance, leading to faster access times as well as more accurate orientation reproductions for items that were probed at expected times. Furthermore, this dynamic prioritization was associated with the temporally specific attenuation of contralateral α (8-14 Hz) oscillations that, moreover, predicted working memory access times on a trial-by-trial basis. We conclude that attentional prioritization in working memory can be dynamically steered by internally guided temporal expectations, and is supported by the attenuation of α oscillations in task-relevant sensory brain areas.

Significance statement: In dynamic, everyday-like, environments, flexible goal-directed behavior requires that mental representations that are kept in an active (working memory) store are dynamic, too. We investigated working memory in a more dynamic setting than is conventional, and demonstrate that expectations about when mnemonic items are most relevant can dynamically and reversibly prioritize these items in time. Moreover, we uncover a neural substrate of such dynamic prioritization in contralateral visual brain areas and show that this substrate predicts working memory retrieval times on a trial-by-trial basis. This places the experimental study of working memory, and its neuronal underpinnings, in a more dynamic and ecologically valid context, and provides new insights into the neural implementation of attentional prioritization within working memory.

Keywords: attention; neuronal oscillations; temporal attention; working memory; α oscillations.

Figure 1.

Task. During encoding, two randomly oriented bars were presented for 250 ms. Bars were always positioned to the left and right of fixation, and one bar was always yellow, whereas the other was blue (color-side mapping was randomly determined for each trial). After a delay of either 1250 or 2500 ms, either of the items was probed by a central probe stimulus. Color of the handles of the central probe (as illustrated above the early and late probe displays in the schematic) represents the to-be-reproduced item. The key manipulation was that the probability of being probed about either the blue or the yellow item was varied over time (with color-interval mappings being counterbalanced across participants). Schematic represents a case in which yellow is expected early and blue late, as indicated by the percentages above the early and late probe displays. Participants reproduced the orientation of the probed item using the mouse. After the probe display appeared, participants were given unlimited time to retrieve the item from working memory and decide what to report. However, once they started to move the mouse, they were given limited time (2500 ms) to complete their report. Elapsed time was displayed under the probe.

Figure 2.

Behavioral results. a, Mean reproduction errors (i.e., [abs(reported orientation − target orientation)]) in degrees, as a function of when an item was probed (early/late) and when it was expected to be probed (early/late). Bar graph represents performance for items probed at expected times (valid, red) or unexpected times (invalid, gray). Error bars indicate SEM, calculated across participants. Rightmost panel, Distributions of response deviations (relative to target orientation) for valid and invalid trials. b, Same conventions as in a, except for the dependent variable: decision time.

Figure 3.

Posterior α modulation during dynamic prioritization in working memory. a, Channel selections for the left and right visual areas, as derived from an independent visual localizer (see Materials and Methods). Color coding represents the percentage of participants for which a given channel was selected to be part of the left (top) or right (bottom) posterior channel clusters. b, Time-frequency plot of the normalized difference in power contralateral to the item expected early versus the item expected late (i.e., [(early − late)/(early + late)] × 100). Only data segments were included in which the probe had not yet occurred (see Materials and Methods). The transparency mask highlights the significant time-frequency cluster (see Materials and Methods). c, Topography of the difference in α power in the late interval, for late expected items that were presented on the left versus on the right at encoding.

Figure 4.

Gaze during dynamic prioritization in working memory and its independence of the α modulation. a, Gaze on the horizontal axis, expressed as a percentage of the interitem distance (as calibrated using an eye-tracker localizer; see Materials and Methods). Data are expressed as the difference in gaze between trials in which the item expected early was on the left versus on the right (which is equivalent to right vs left in the late interval). Red curve indicates the original data and reveals a slight bias of gaze in the direction of the side of the expected item at encoding. Shading represents ± 1 SEM, calculated across participants. Critically, after we removed trials to the point where this gaze bias was reversed (blue curve; see Materials and Methods), the α modulation in Figure 3b remained virtually identical, as depicted in b.

Figure 5.

Trialwise correlation between preprobe contralateral α power and working memory access times. a, Time-frequency map of the trialwise correlation between preprobe and postprobe power and decision times, separately for channels contralateral (leftmost) and ipsilateral (middle) to the location of the probed item at encoding, as well as for the difference in correlation between the contralateral and ipsilateral channels (rightmost plot). Transparency masks highlight the significant time-frequency cluster (see Materials and Methods). Only valid trials (in which the sides of the expected item and the probed item were the same) were included in the analysis. b, Topography plots of the trialwise correlation between preprobe α power and decision times, separately for probed items that were on the right and on the left at encoding. c, Bar graph represents the trialwise correlations of interest between preprobe α power and decision times, separately for early and late probes.