Working-memory load did not robustly change hippocampal SEEG band power during maintenance

We analyzed hippocampal bipolar band power during the 3-s maintenance period of a verbal working-memory task with set sizes of 4, 6, or 8 letters. The source was OpenNeuro ds004752 v1.0.1, with 15 patients and two sessions analyzed separately; session 2 was declared as a replication before session 1 group results were seen. Correct, artifact-free trials were analyzed using Welch log power in five bands and subject-level sign-flip tests with Holm correction. The load manipulation affected behavior: in session 1, accuracy fell from 0.97 to 0.84 and response time increased with set size in 14 of 15 participants. In hippocampal maintenance data from 14 participants and 99 pairs, no band survived Holm correction. High gamma (70–150 Hz) showed a small decrease (mean rho −0.064, p = 0.035, Holm p = 0.175), absent in the fixation control. In session 2, all bands were null; the high-gamma association was −0.014 (p = 0.80), so the session 1 trend did not replicate. Effect sizes were small (|mean rho| < 0.07); about 40 trials per session gave power only for group effects around |rho| > 0.08. Correct-only trial selection may weaken load effects, and power was not corrected for the 1/f slope. The analysis does not test the phase-synchrony measures in earlier reports.

working memory · hippocampus · SEEG · intracranial EEG · band power · replication · OpenNeuro