Experiment procedure and behavioral performance.

(A) Attention tasks in the No-Ping and Ping sessions. Note that only long-delay trials are shown. A small proportion of short-delay trials (20%, with a delay of 1.5 s or 3.5 s) were included to create temporal uncertainty and encourage consistent active preparation during the delay. Both component gratings were flickering at 10 Hz between white and black, so that luminance could not confound either the task strategy (e.g., attending to luminance) or neural measures. The inset shows two sets of color-orientation mapping, which were reversed halfway through the experiment to minimize the impact of cue-induced sensory difference on neural activity. A high-contrast impulse was presented during preparation period in the Ping session. (B) Perception task. Similar to the attention task, the single-orientation grating also flickered at 10 Hz between white and black. (C) Behavioral accuracy in the attention tasks in the No-Ping and Ping sessions. Each dot represents one subject’s data. Error bars denote standard error of the means (SEM).

MVPA for the No-Ping session and Ping session.

(A) Schematic illustration of the decoding of attended orientation (attend leftward vs. attend rightward) in the attention task (left panel) and the cross-task generalization analysis from perception task to the attention task (right panel). The four regions are shown on a representative right hemisphere as colored areas: V1 is marked in red, EVC in yellow, IPS in cyan and PFC in purple. (B) Decoding accuracy during preparation and stimulus selection periods across regions in the No-Ping and (D) Ping session. (C) Cross-task generalization performance from the perception task to the preparatory periods and the stimulus selection periods across regions in the No-Ping and (E) Ping session. The dashed lines represent the theoretical chance level (0.5). Each dot represents one subject’s data. Error bars denote SEM.

Orientation-selective attentional modulations on neural pattern distances during preparation.

(A) Schematic illustration of the representational distance (mean Mahalanobis distances) between each of the attention conditions and each of the perception conditions. Colored arrows indicate measures of the pair-wise Mahalanobis distance. The right panel shows two attention trials (red indicates attend-to-leftward and green indicates attend-to-rightward) to the distribution of each perception condition (shown in a cloud of light-colored dots). (B) Mahalanobis distance between preparatory attention condition and perceived orientation condition in the No-Ping and (C) Ping sessions. Error bars denote SEM. (D) Correlations between attentional modulation index (AMI) and reaction time (RT) in the No-Ping and (E) Ping sessions. Each dot represents one subject’s data. The shaded area represents the confidence intervals of the regressed lines. * p < 0.05, ** p < 0.01.

Information connectivity analysis.

(A) Schematic illustration of the procedure for the information connectivity analysis in the space of two hypothetical voxels. For each region, we calculated the Mahalanobis distance of the attention trial (from one left-out run) from two attention distributions (all trials from remaining runs). Red and green dots indicate activity patterns from two trials (right panel). The brain image shows an example pair of intercortical information connectivity between V1 and PFC. The time series (lower-left panel) consisted of attentional modulation index (AMI) based on the Mahalanobis distance. (B) Between-region information connectivity in the No-Ping and Ping sessions. (C) The differences in connectivity between the Ping and No-Ping sessions. * p < 0.05, ** p < 0.01.

Univariate BOLD results.

(A). Mean fMRI time course from four regions (V1, EVC, IPS, PFC) for No-Ping session and (B) Ping session. The gray triangle indicates the onset of superimposed gratings. The gray area indicates a slightly elevated response in the presence of visual impulse. Error bars denote standard error of the means.

Sensory decoding in the perception task.

(A) Decoding of orientations in No-Ping and (B) Ping sessions. The dashed lines represent theoretical chance level (0.5). Each dot represents one participant. Error bars denote standard error of the means.

Descriptive statistics of Mahalanobis distance during preparatory attention

Relationship between attentional modulation on Mahalanobis distance and RT.

(A) Behavioral performance (as indexed by RT) for “strong modulation” (high AMI) trials and “weak modulation” (low AMI) trials, sorted by V1 preparatory activity in the No-Ping and (B) Ping sessions. Each dot represents one subject’s data. Error bars denote SEM. * p < 0.05

Mahalanobis distance during stimulus selection period.

(A) Attentional modulation on Mahalanobis distance in No-Ping and (B) Ping sesions. Each dot represents one participant. Error bars denote standard error of the means. † p < 0.06, * p < 0.05, ** p < 0.01, *** p < 0.001.

Relationship between information connectivity and RT.

(A) Correlations between V1-PFC connectivity and RTs in the No-Ping and (B) Ping sessions. Each dot represents one subject’s data. The shaded area represents the confidence intervals of the regressed lines.

Functional connectivity analysis.

(A) Connectivity based on multivoxel patterns during stimulus selection period: between-region connectivity in the No-Ping session and Ping session based on cross-validated Mahalanobis distance (left panel) and the differences in connectivity between the two sessions (right panel). (B) Connectivity based on mean BOLD responses over time: between-region connectivity in the No-Ping session and Ping session based on raw time series (left panel) and the differences in connectivity between the two sessions (right panel).

Analysis of eye positions.

(A) Group-level horizontal eye position and (B) vertical eye position for the preparation period during behavioral training in No-Ping and Ping sessions. Each dot represents one participant. Error bars denote standard error of the means.