Fig. 4: Seizure bursts become directional as they diverge from baseline bursts.

a For each of the same four example seizure bursts (mean population spiking rate in dark blue trace; micro-LFP in maroon traces; copied from Fig. 1a), we mapped the timing of peak MUA activity onto the spatial layout of the MEA. Each micro-electrode is colored according to the time of peak MUA activity during that seizure burst (blue: early MUA; yellow: late MUA). We fit a plane to the spatial organization of each spiking sequence, generating a measure of goodness of fit (R², indicated by length of black arrow) and a direction (indicated by direction of black arrow). b Distribution of R² values for all bursts in the same example seizure (purple, mean 0.12 ± 0.09) compared to IED bursts (red, mean 0.16 ± 0.14), baseline bursts (cyan, mean 0.09 ± 0.08), and null bursts in which the order of spiking electrodes were scrambled (gray, 0.03 ± 0.03; see Methods). c Average R² over all seizure (sz; 0.15 ± 0.05), IED (0.10 ± 0.04), baseline (bl; 0.07 ± 0.02), and null bursts (0.04 ± 0.02) in every participant and array. Each dot represents a single participant and array. R² for seizure bursts and IED bursts was significantly greater than for null (p = 0.002 and p = 0.02, respectively; two-tailed p-values adjusted using the Holm–Bonferroni correction for multiple comparisons). d Timecourse of the goodness of fit (R²) for all seizure bursts in this example seizure (three-second moving average). e Correlation between distance from baseline centroid and R² for each seizure burst (ρ = 0.60). f Weighted average correlations between distance from baseline centroid and R², for all seizure bursts across all seizures in each participant, for n = 6 arrays. For each array in each participant, the average correlation is shown (dark blue) with the 95% confidence interval (light blue). Average correlation across all participants and arrays is shown below (dark blue vertical line and error bar).  × indicates participant 1.