Directional gaze biases by past and future locations during mnemonic selection overlap in time.

a) Task schematic. Participants memorised two oriented gratings with different colours presented either vertically or horizontally. Following a delay, a colour change of the central fixation dot prompted participants to select the colour-matching item from working memory to report its orientation later. After another delay, two test gratings appeared transiently and participants compared the cued memory grating to the relevant test grating (clockwise/counter-clockwise judgment) that was determined by the ‘future rule’ that was stable within each block. After a response, feedback (0: wrong, 1: correct) was presented at the side of the relevant test grating. Dash lines serve to explain the association between the encoding and test locations and were never presented in the actual experiment. b-c) Time courses of gaze shift rates (number of saccades per second) for shifts toward and away from the encoded (panel b) and to-be-tested (panel c) locations. d) Overlays and comparisons of the difference in gaze-shift rates (toward minus away) for the past (encoded) location and the future (to-be-tested) location. Horizontal lines indicate significant temporal clusters (cluster-based permutation test, P < 0.001). Data are presented as mean values with shading reflecting 95% confidence intervals, calculated across participants (n = 25).

Individual saccades are jointly biased to past and future memory attributes.

a) The distribution of the direction of the first saccades we detected after cue onset relative to past (horizontal) and future (vertical) locations. Data from the different sessions were rotated to match a common coordinate frame. b) The bias toward the past (x-axis) as a function of saccade direction with regard to the future (y-axis). c) The bias toward the future (y-axis) as a function of saccade direction with regard to the past (x-axis). In b-c, The bold black line indicates the significant temporal cluster (cluster-based permutation test, P < 0.001). Data are presented as mean values with shading indicating 95% confidence intervals, calculated across participants (n = 25). d) The percentage of identified first saccades toward or away from the future location as a function of whether the same saccades were also biased toward or away from the past location. Error bars in panel d indicate ±1 SEM calculated across participants (n = 25).

Four possible associations between encoding and test locations.

Early saccade biases by past and future memory attributes are predominantly driven by microsaccades.

Difference in gaze-shift rates toward minus away relative to past location (panel a) or future location (panel b), as a function of saccade size (y axes). For reference, dashed horizontal lines indicate 1° visual angle. Additionally, for each panel, we separately show the difference in gaze-shift rates (toward minus away) in time course at the top (collapsed over all depicted saccade sizes) and the difference in number of gaze-shift as a function of gaze-shift magnitude to the right (collapsed over all depicted times).

Trials with vertical or horizontal configurations show similar joint consideration of past and future memory attributes.

Conventions as in Main Figure 1b-d, separately for trials that encoded items horizontally and tested items vertically (panel a) and trials that encoded items vertically and tested items horizontally (panel b).

Gaze biases in extended time window as a complement to Figure 1 and Supplementary Figure 2.

This extended analysis reveals that while the gaze bias towards the past location disappears around 600 ms after cue onset, the gaze bias towards the future location persists (panel a) and that while the early (joint) future bias occurs predominantly in the microsaccade range below 1 degree visual angle, the later bias to the future location incorporates larger eye movement that likely involve preparing for optimally perceiving the anticipated test stimulus (panel b).

Distribution of saccade directions relative to the future rule from encoding onset.

(Top panel) The spatial layouts in the four future rules. (Middle panel) Polar distributions of saccades during 0 to 1500 ms after encoding onset (i.e., the period between encoding onset and cue onset). The purple quadrants represent the axis of the future rule and the grey quadrants the orthogonal axis. (Bottom panel) Time courses of saccades along the above two axes. We did not observe any sign of a bias along the axis of the future rule itself.