Fig. 3: A backbone of feedforward activity flow during perceptual maintenance.

A Granger causality was calculated separately for each direction between a pair of electrodes residing in different lobes. To assess significance, the difference in Granger causality between the two directions (‘asymmetry’) was compared with a null distribution created by shuffling the electrode labels 1000 times for each electrode pair. B To aggregate the results across many electrode pairs, a bias measure was calculated by comparing the number of significant inter-lobe connections in each direction using a two-sided binomial test, separately for each percept. C Significant (p < 0.05, uncorrected) biases (as assessed in B) in inter-lobe connections, separately assessed for 2 images × 2 percepts. Lobes were assigned a level in the cortical hierarchy (bottom: occipital; middle: temporal, parietal; top: frontal) and each directed inter-lobe connection between levels was defined as feedforward (red) or feedback (blue). Line width indicates the strength of significance. D Frequency-domain inter-lobe biases for the face-vase image during the preferred percept (green) and non-preferred percept (magenta). Positive and negative values correspond to feedforward and feedback biases, respectively. Horizontal bars: two-sided binomial test p < 0.05, cluster-corrected. Corresponding results for the cube images are shown in Supplementary Fig. 3. Source data are provided as a Source Data file.