Figure 5. Repetition effects on GC are strongest for gamma in the feedforward direction.

(a) Bivariate GC spectra between areas V1 and V4 (FF = feedforward, i.e. V1-to-V4, FB = feedback, i.e. V4-to-V1). GC was separately computed for early repetitions (trials 11–50, i.e. after the early gamma decrease) and late repetitions (trials 81–120). Error regions reflect 95% CIs. Inferential statistics are based on a non-parametric permutation test cluster-corrected for multiple comparisons across frequencies (Maris and Oostenveld, 2007). Horizontal green bar indicates significant cluster for FF GC. (b) Same analysis as in (a), but for areas V4 and IPS1, with feedforward being V4-to-IPS1 and feedback being IPS1-to-V4. (c) Total number of per-frequency significant differences between late and early repetition GC spectra between all areas (green = feedforward, gray = feedback). (d) All areas used for the analysis, plotted onto a semi-inflated average cortical surface. Area and surface definitions were taken from the HCP MMP1.0 atlas (Glasser et al., 2016a). (e) Changes in gamma GC from early to late trials, separately for the feedforward direction (upper matrix half, enclosed by green triangle) and the feedback direction (lower matrix half, enclosed by gray triangle). Non-significant matrix entries are gray masked. To be considered significant, matrix entries had to pass a t_max-corrected paired permutation test including time-reversal testing (Vinck et al., 2015). Inset right: Changes in gamma power for each brain area from early to late repetitions (significance based on a t_max-corrected paired permutation test; non-significant areas are gray masked). (f) The analysis of (e) was repeated per subject, and for the individually significant matrix entries, GC changes were averaged, separately for the feedforward (x-axis) and feedback (y-axis) direction; each dot corresponds to one subject. Across subjects, repetition-related GC changes were larger in the feedforward than the feedback direction (${\displaystyle p<0.001}$).

Figure 5—figure supplement 1. Conceptual replication of Michalareas et al., 2016.

(a) Replotting of the data previously published in Figure 6 of Michalareas et al., 2016. The graphs show the Spearman correlation coefficient, across pairs of brain areas, between a metric of GC asymmetry (DAI, per frequency) and a metric of the feedforward character of the corresponding anatomical projection (SLN). The DAI metric is the directed influence asymmetry index from brain area A to brain area B, defined as (DAI_A-to-B = [GC_A-to-B – GC_B-to-A]/[GC_A-to-B + GC_B-to-A]). The SLN metric is the supragranular labeled neuron proportion, a graded anatomical metric (defined in the macaque) of the degree to which an anatomical projection is of feedforward character (i.e., originating in supragranular layers; Barone et al., 2000). DAI was positively related with SLN in the subject-defined gamma band, and negatively related to SLN in the subject-defined alpha/beta band. Lines show the average over subjects, and error bands represent standard error of the mean across subjects. Significance was computed using cluster-based nonparametric testing against zero (Maris and Oostenveld, 2007). (b) Similar analysis for the dataset recorded here, using the same seven brain areas. The same positive relation was found in the subject-specific gamma band. A similar negative relation is visible in the subject-specific alpha band, but did not reach significance. Lines show the average over subjects, and error bands represent the bootstrapped 95% confidence interval, significance was computed using a t_max-corrected permutation test. SLN values from Chaudhuri et al., 2015.