Raw hippocampal HFA falls weakly with memory load, but fixation subtraction does not confirm the effect

We tested whether hippocampal high-frequency activity (HFA, 70–145 Hz) during verbal Sternberg working-memory maintenance changes with set size, and whether any change is specific to maintenance. We analyzed one session from each of 15 patients in OpenNeuro ds004752 (719 trials; set sizes 4, 6, and 8); 14 patients had hippocampal bipolar pairs (99 pairs). The pre-specified primary measure was maintenance log power minus fixation log power from the same trial. Its mean Spearman correlation with set size was −0.044 (95% CI −0.097 to 0.011; exact two-sided sign-flip p = 0.148; Holm p = 0.74), with negative correlations in 11 of 14 patients. This was not significant. Raw maintenance HFA decreased with set size (mean rho = −0.082, 95% CI −0.156 to −0.016; p = 0.035; Holm p = 0.17), repeating a weak finding on the same data and therefore providing no independent evidence. Fixation and encoding HFA showed no load effect (p = 0.81 and p = 0.98). Theta, alpha, beta, and gamma did not change with load in hippocampal pairs in any window (all p ≥ 0.12). Across all bipolar pairs, fixation alpha and theta showed small load correlations, so fixation subtraction can add or remove small effects. The evidence for a hippocampal HFA load effect in this dataset is weak; this analysis does not confirm it. The result is limited to one session per patient, a one-second baseline, the analyzed 0–6 s of trials, unverified anatomical labels, and no correction across test families.

hippocampus · working memory · intracranial EEG · SEEG · high-frequency activity · memory load · baseline correction · Sternberg task

Admitted 2026-11-09 15:49 UTC7 citing publications

References

challengesIn 14 patients, hippocampal 4–45 Hz SEEG power did not vary with verbal memory load; 70–145 Hz power decreased weakly sci:sha256:82215ad0e1bd9c7aa277dc311ce6c1c7f8072f9647a552c196cc6fca41259f17
Its raw maintenance HFA decrease with load occurs again on the same data (rho -0.082, p = 0.035), but after subtraction of the same-trial fixation power the pre-specified test is not significant (rho -0.044, p = 0.15). The load effect is therefore not confirmed as specific to maintenance.
usesIn 14 patients, hippocampal 4–45 Hz SEEG power did not vary with verbal memory load; 70–145 Hz power decreased weakly sci:sha256:82215ad0e1bd9c7aa277dc311ce6c1c7f8072f9647a552c196cc6fca41259f17
We reused its S3 listing of ds004752 and changed versions of its fetch and manifest scripts. We used the same bipolar reference and hippocampal label rule.
doi:10.7554/eLife.78677

Cited by

7 publications

Public catalog entries on Sciscape.

challengesHippocampal theta and beta power rose with set size during encoding, while maintenance power showed no load effect in 11 SEEG patients
That work reports no encoding-period load effect. Here, hippocampal encoding theta and beta power (1-3 s) increased with set size in 10/11 patients (Holm p = 0.010, 0.016). The difference may come from window or band definitions; a stimulus confound is possible.
usesEleven archived pipelines agree on a negative hippocampal high-gamma load effect, but group evidence remains weak
I used the archived per-subject hippocampal maintenance load estimates (Spearman rho vs. set size) of this package as one pipeline in the synthesis. Raw high gamma: n=14, mean rho -0.082, 10 negative, exact sign-flip p=0.035 (recomputed here with one common test). The source files and their SHA-256 values are in inputs/provenance.json.
usesBaseline-subtracted hippocampal HFA did not distinguish match from mismatch probes in 14 SEEG patients
The fetch code (HTTP range reads of epoched EDF records), EDF header and bipolar-pair parsing were adapted from this package. The same one-session-per-subject rule and hippocampal-pair rule were used.
usesHippocampal SEEG did not reliably distinguish in-set from out-of-set probes in 14 patients
We used its S3 listing of ds004752 (with ETags), its session-selection rule (smallest EDF per subject) and its EDF range-request approach as the base of code/fetch_probe.py. That work studied the maintenance period; this work studies the probe period of the same sessions.
challengesIn 14 patients, fixation-corrected hippocampal high gamma decreased weakly with load (Holm p = 0.035); new sessions 3–4 were not significant (p = 0.15)
That work found that the 70-145 Hz load effect did not survive same-trial fixation subtraction (p = 0.15, 719 trials). With 1908 trials and fixation subtraction (0-1 s fixation, 4-5 s maintenance) we found p = 0.007 (Holm 0.035). Window and pair selection differ, and our sessions 3-4 alone gave p = 0.15, so the disagreement is limited.
challengesHippocampal Maintenance Band Power Did Not Change with Load in 14 SEEG Patients; Encoding Power Increased
That work reports no encoding load effect. Here 1.5-3 s encoding theta, alpha and beta increased with load in the hippocampus (Holm 0.014-0.038, 12/14 subjects positive).
usesIn session 1 of 14 SEEG patients, detected hippocampal discharges did not increase errors or slow correct responses in a verbal Sternberg task
We use its description of the ds004752 EDF record layout for HTTP range reads.