Patient recruitment
Research was carried out following approved protocols by the Institutional Review Board at Baylor College of Medicine. Participants included patients diagnosed with drug-resistant temporal lobe epilepsy, all slated for anteromesial temporal lobectomy aimed at controlling their seizures. Each patient gave written consent and was made fully aware that their participation was voluntary, which wouldn’t impact their clinical care. The age of participants ranged from 25 to 54, averaging at about 39.6, comprised of three females and four males. Four of the surgeries were on the left side while three were on the right. In one case, recordings were taken from the middle temporal lobe before the tissue was removed. Notably, none of the patients remembered any intraoperative events when they were asked in the post-operative care unit the next day.
For comparison, we also included a cohort of awake patients who were listening to podcast audio. These patients were sourced from individuals undergoing invasive recordings in the EMU at Baylor St Luke’s Hospital. Methods for this patient group were detailed in previous reports.
Neuropixels data acquisition set-up and intraoperative recordings
We used Neuropixels 1.0-S probes with 384 recording channels (total of 960 contacts; 384 of which were usable). The probes were individually sterilized and our data acquisition system consisted of a custom rig with specialized hardware for capturing neuronal signals. The setup was verified by Biomedical Engineering at Baylor St Luke’s Medical Center, and recordings occurred during surgery using high-performance computers with open-source software. Neural signals were sampled at various rates for both action potentials and local field potentials.
Audio was delivered through a separate computer, with the sound captured through a coaxial cable splitter to speakers near the patient. Synchronization of audio and neural data was ensured through both online and offline methods using appropriate software tools. Intraoperative recordings were executed in tissues designated for removal, monitored using a robotic arm and specialized visualization equipment.
Data was collected under general anaesthesia for a maximum of 30 minutes. Each patient received intravenous anaesthesia, primarily with propofol, and inhaled anaesthetics were limited to induction. The anaesthetist managed drug infusion rates carefully throughout the procedure, maintaining specific monitor values indicative of anaesthesia depth.
During passive auditory stimulus presentation, we recorded neuronal activity for selected patients. Various auditory sequences were employed, including pure tones and oddball stimuli, interspersed with washout periods to maintain balance in stimulus presentation.
Specific patient sessions involved listening to engaging podcast stories. One patient followed three stories from The Moth Radio Hour, while another listened to an educational video with rich content aimed at sustaining interest.
Micro-CT
Post-recording, a small tissue cube was excised from the operative area for analysis, following defined methods. The samples were prepared and preserved before imaging them with high-resolution micro-computed tomography. This imaging provided detailed back-projection images, subsequently analyzed for any abnormalities, none of which were found.
Neuronal data processing
Patients did not have seizures during the operations, likely due to anaesthesia, so no related data cleaning was necessary. This was confirmed by a neurologist reviewing the signals.
Motion correction
Motion artifacts resulting from brain movement were addressed through previously established algorithms. Motion estimates were derived using specific software packages, and corrections were made based on interpolation methods, ensuring spikes and local field potentials were properly aligned for further analysis.
Unit extraction and classification
Spike detection and clustering were automated and refined through a combination of software tools, taking care to ensure optimal unit quality before proceeding with further analytical steps.
LFP data
Local field potential data underwent bandpass filtering and alignment to task events to facilitate the extraction of response characteristics.
Neuronal data analysis
All analyses were conducted using custom MATLAB scripts, focusing particularly on the corrected motion data. The overall amount of movement was quantified, and tone-responsive neurons were identified through systematic statistical methods.
Neuronal response learning dynamics
We explored how statistical learning impacted oddball detection by examining response patterns across trials. This involved creating a vector representation of neuronal activity and analyzing subsequent shifts in neural responses.
Mixed-effects models
Mixed effects models were utilized to investigate how various task conditions affected neuronal responses, taking into account the complexity of the hierarchical data structure.
Continuous-rate RNN model
A continuous-rate recurrent neural network was developed to mirror the oddball detection task in our experiments. The network’s parameters were optimized using gradient-based methods, targeting specific performance accuracy before advancing through various trial contexts.
Neuronal data analysis: natural language stimuli
Natural language statistics
The podcast audio was systematically transcribed and checked for accuracy. Statistical analyses were based on word frequencies drawn from a broad corpus of movie subtitles, while further calculations addressed the emotional and semantic implications of the content.
Word embedding and part of speech classification
Using advanced NLP tools, we analyzed the dataset for part-of-speech categorization, leveraging a pre-existing toolkit to classify the words present in the narratives.
Probe localization
Intraoperative navigation tools were employed to label the probe entry site after its removal, helping to accurately place recordings in the MNI standard space for subsequent analyses.
Ethics statement
All experiments followed Institutional Review Board protocols to ensure compliance and ethical considerations during the study.
Reporting summary
Further details on research design are accessible in the linked reporting summary.
