Experimental paradigm and behavioral results.

(A) Participants had to remember two motion directions (S1 and S2) per trial (motion directions are indicated here by gray arrows for illustration only). The stimuli were either red or green. After a short delay, a retro-cue indicated which of the two motion directions (red or green) they had to recall by adjusting a randomly oriented line. (B) Response errors in the current trial were biased towards the target direction of the previous trial (pink curve), but not towards the previous non-target direction (grey curve). The x-axis shows the motion direction difference between the previous and the current target, with positive values indicating that the previous target direction was more clockwise than the current target direction. The response error on the current trial is displayed on the y-axis, with positive values indicating a clockwise deviation from the true target direction. Therefore, shifts of the response errors to the lower left and upper right quadrant indicate an attraction whereas shifts to the upper left and lower right quadrant indicate a repulsion. The asterisk marks significant attractive serial dependence towards the previous target direction.

Reconstruction, fidelity and systematic shifts of currently relevant motion direction.

(A) The reconstructed direction of the currently relevant motion direction across the course of a trial. The left panel shows the reconstruction of the S1 direction, the middle panel the reconstruction of the S2 direction and the right panel the reconstruction of the cued direction (either S1 or S2). Higher reconstruction strength is depicted in yellow and lower strength in dark blue. Time points (x-axis) with direction information show a pronounced yellow area around the common center set to zero (y-axis) and darker areas with increasing distance from this center. (B) Fidelity of reconstructed direction of the currently relevant motion direction across the course of a trial. The left panel shows the fidelity of the S1 direction, the middle panel depicts the fidelity of the S2 direction and the right panel shows the fidelity of the cued direction (either S1 or S2). Small circles show the fidelity of each participant. Colored circles indicate time points with a fidelity significantly greater than zero (cluster-based permutation test within each epoch). (C) The reconstructed direction during all time points with a significant reconstruction fidelity was averaged within each epoch and the maximum of the reconstructed direction was compared to the common center set to zero. A negative shift of the reconstructed mean indicates a repulsion from the target direction of the previous trial whereas a positive shift indicates an attraction. For a more detailed visualization, the range of direction shifts from −40° to +80° is magnified and reconstructed shift values for each participant are indicated by dots. The upper row shows the bootstrapping distributions (see section on MEG analysis below for details). Significant shifts are marked with an asterisk.

Reconstruction, fidelity and systematic shifts of S1 direction during the S2 epoch.

(A) The reconstructed direction of the S1 motion direction during the S2 epoch. (B) Fidelity of reconstructed S1 motion direction across the course of the S2 epoch. Small circles show the fidelity values for each participant. Colored circles indicate time points with a fidelity significantly greater than zero. (C) The reconstructed direction for all time points with a significant reconstruction fidelity was averaged within the S2 epoch to indicate attractive and repulsive shifts in relation to the target of the previous trials. Significant shifts are marked with an asterisk. For details, refer to Figure 2.

Correlation between single-trial shift of reconstruction and upcoming response.

For each time point, the shift of the maximum of the reconstructed direction in a single trial was correlated with the response deviation from the target direction. Small circles show the r-values of each participant. Time points where the shift of the reconstructed direction significantly correlated with the upcoming response deviation are indicated by colored circles (cluster-based permutation test within each epoch).

Reconstruction and fidelity of previous target direction.

(A) The target direction of the previous trial was reconstructed throughout the last second of the inter-trial interval (ITI) and the whole following trial. (B) Fidelity of reconstructed direction of the previous target. Small circles show the fidelity of each participant. Colored circles indicate time points with a fidelity significantly greater than zero. For details, refer to Figure 2.

Relation between the individual a (upper panels) and w (lower panels) parameters of a DoG serial dependence fit (x-axis) and the neural reconstruction shift (y-axis).

Colored circles show individual values for the S1 (left panels), S2 (middle panels) and retro-cue epochs (right panels). Solid lines depict significant correlations, dashed grey lines mark a reconstruction shift of 0°.

Cross-validated reconstruction fidelity of previous direction, model trained on currently relevant direction.

Fidelity of reconstructed direction of the previous target using a model that was trained on the currently relevant direction. The left panel shows the fidelity of the previous target during the S1 epoch, the middle panel depicts the fidelity during the S2 epoch and the right panel shows the fidelity of the cued direction (either S1 or S2). Small circles show the fidelity of each participant. Colored circles indicate time points with a fidelity significantly greater than zero (cluster-based permutation test within each epoch).

Systematic eye movements and reconstruction time courses were uncorrelated.

(a) Amount of stimulus-related gaze direction (indicated by variance reduction) during each epoch as a function of time with error bars indicating standard errors and colored circles indicating time points with a variance reduction significantly greater than zero (cluster-based permutation test within each epoch; p<.05). (b) Time courses of individual stimulus-related gaze directions. Dark red indicates the subject with the highest average variance reduction in a given epoch while green indicates the one with the smallest average variance reduction, with color gradients in between according to the individual rank of average variance reduction. (c) Time courses of individual reconstruction fidelities. Lines are colored as in (b), reflecting the stimulus-related gaze direction rank of a subject in a given epoch. (d) Time courses of p-values (uncorrected for multiple comparisons) of time point-by-time point Spearman rank correlations between the systematic gaze directions as shown in (b) and the reconstruction fidelity as shown in (c).