Fig. 5: Distinct roles of dPC1 and dPC2 during probe presentation.

a,b, The activity of the first two dPCs (dPC1 (a) and dPC2 (b)) during probe presentation (entire dataset), projected onto cued and other conditions, divided based on the presence or absence of the preferred item as a probe in the stream of distractors during the first visual search. The activity of dPC1 increases during all presentations of the preferred item, both when it was previously cued (cued preferred in) and when it was not part of the current memory set but was presented as a lure (other preferred in). dPC2 shows increased activity only when the preferred item was previously cued and presented as the current target. Shaded areas represent ±s.e.m. across trials. c, The accuracy of two decoders trained to classify trials where the preferred item was present versus absent in RSVP separately for the cued and other conditions. When applied to the activity of dPC1, both decoders performed above chance. However, when applied to the activity of dPC2, performance was above chance only for the cued condition. All significance levels for all analyses were determined using one-sided permutation tests assessing whether decoding accuracy was greater than expected under the null hypothesis. Reported P values are FDR corrected. d, Interaction of trial type (cued versus other) and component (dPC1 and dPC2). The difference in decoding accuracy between the cued and other conditions was greater for dPC2 than for dPC1, exceeding the expectations of the null distribution. Decoding accuracies obtained from image-selective cells are represented with stem plots. Violin plots represent null distributions (estimated using scrambled data) and black lines indicate the mean. Chance levels for decoders were at 50%, and for the interaction at 0%. Significance of the interaction was assessed similarly as in c but results were compared with the null distribution of differences.