Dissociating external and internal attentional selection

Abstract

Just as attention can shift externally toward relevant objects in the visual environment, it can shift internally toward relevant items within Visual Working Memory (VWM). Recent work has shown that spatial attention is automatically directed toward the previous location of an attended memory item, as it is to locations of perceived stimuli. When attending memory items, however, there is no sensory information to be processed at the previous location. Thus, we asked whether internal attention-akin to external attention-modulates sensory processing. In two EEG experiments, we compared location-specific sensory enhancements during attentional selection of external (perceived) versus internal (memorized) stimuli. Alpha-power and gaze-position biases confirmed an inherent spatial organization within VWM. However, Rapid Invisible Frequency Tagging (RIFT) revealed sensory modulation only during external attentional selection. Thus, VWM is not blindly recruiting existing mechanisms of external attention, but instead uses space as an organizational principle to store and select memories.

Keywords: Cognitive neuroscience; Psychology.

Graphical abstract

Figure 1

Similarities and differences between internal and external attention Similarities between the representation of location during internal (blue region) and external (purple region) prioritization exist in high-level visual areas (left), and behavioral performance at the corresponding location improves for both (right). It is currently unknown whether early visual processing behaves similarly across both. This may differ because of the different functions involved - encode novel stimuli vs. highlight memorized information. A distinction at the early processing level could indicate whether internal attention is vestigially using spatial mechanisms that evolved for processing the external world, or that this spatial layout within VWM serves a functional purpose.

Figure 2

Task design Two uniquely colored and oriented stimuli were presented on either side of fixation. This was followed by (retro-cue experiment) or preceded by (pre-cue experiment) a red or blue circle (the cue), indicating which of the two oriented stimuli (i.e., the red or blue item) would be probed. After a delay, participants indicated whether a memory probe was tilted clockwise or counterclockwise relative to the cued item. The highlighted locations flickered at 60 or 64 Hz (See Tagging Manipulation) throughout the trial (Note: not to scale; memory probe did not overlap with flickering regions in actual display. No actual outlines around flickering regions were displayed).

Figure 3

Average RIFT response (A) Time-frequency plot averaged across participants and top 6 channels with highest coherence (selected individually per participant) showing clear peaks at 60 Hz and 64 Hz following flicker onset. (B) Topographical distribution of average RIFT coherence over the interval during which both gratings were on screen. Black plus marks the POz electrode. Note that separate y axis limits are used for the coherence traces and topographies in each experiment, see text under section “RIFT responses to prioritized locations are enhanced … ”.

Figure 4

RIFT responses to prioritized locations are enhanced only during external and not internal selection RIFT coherence from frequencies corresponding to cued vs. uncued stimuli (blue) locations and their difference (red) in the (A) retro-cue experiment (B) pre-cue experiment (shaded region—95% bootstrapped CIs). Note that separate y axis limits are used for the coherence traces over both figures to convey equivalent variances visually, see text under section “RIFT responses to prioritized locations are enhanced … ”.

Figure 5

Alpha oscillations reflect the location (left/right) of the prioritized item during both internal and external selection (A) Average difference between alpha power (8–13.5 Hz) in left and right item cued trials upon presentation of retro-cue/stimuli, respectively (channel dots indicate 95% bootstrapped CIs excluding 0, positive = red or negative = blue), (B) Comparison across experiments. Shaded patch indicates kernel density estimation of respective scatterplots; dashed lines indicate respective means; gray indicates distribution of bootstrapped mean differences; black bar indicates 95% Cis.

Figure 6

Gaze position also reflects the location (left/right) of the prioritized item during both internal and external selection Difference in gaze position between right cued and left cued trials (top) and difference (bottom) in the (A) retro-cue experiment (B) pre-cue experiment (shaded region—95% bootstrapped CIs). Deviation is reported in terms of dva and percentage of distance from fixation to horizontal eccentricity of stimulus center.

Figure 7

Gaze position bias does not drive other metrics of attentional modulation Linear mixed effects model predicting top: the RIFT response and bottom: alpha lateralization in the (A) retro-cue experiment (B) pre-cue experiment (shaded region—95% bootstrapped CIs). Linear coefficients (β) are reported with error bars representing 95% confidence intervals. The trial-wise gaze position bias does not explain the trial-wise attentional modulation in the RIFT response or alpha lateralization.

Figure 8

Control experiment with a spatial cue confirms that attentional modulations of the RIFT response can also be measured in the absence of visible stimuli (A) In a spatial-cue experiment similar to the earlier pre-cue experiment, we allowed participants to shift attention before the items were displayed to confirm that RIFT modulation could be observed in the absence of stimuli. (B) The enhanced RIFT response from the cued location was still present both before and after the items were displayed. We also replicated the effects of (C) alpha lateralization and (D) gaze position bias toward the cued location as seen earlier in the retro-cue and pre-cue experiments. (E) Lastly, with Linear mixed effects modeling we (1) confirmed that for this experiment as well gaze position was not driving the RIFT response, and (2) replicated the attentional modulation effect of coherence using trial-wise Hilbert-magnitudes.