The human hippocampus demonstrated neural activity consistent with oddball discrimination, semantic processing, and contextual language encoding during propofol-based general anesthesia, according to a study published in Nature.
Kalman A. Katlowitz, PhD, of Baylor College of Medicine, and colleagues performed intraoperative recordings using high-density Neuropixels microelectrodes in seven patients with drug-resistant temporal lobe epilepsy undergoing anteromesial temporal lobectomy. Recordings were obtained in the anterior hippocampal body following resection of the lateral temporal cortex and before resection of mesial temporal structures. Across all recordings, the researchers isolated 651 neuronal units.
Three patients completed an auditory oddball task in which 100-ms pure tones were presented in an 80/20 probability distribution, with less frequent “oddball” tones interspersed among standard tones. In two patients, the researchers counterbalanced which tone served as the oddball stimulus. Most recorded units showed tone-evoked responses, and about 25% of units in the counterbalanced analysis significantly encoded oddball tones within the first 300 ms following stimulus onset.
A support vector machine decoder identified tone identity with accuracies ranging from 61% to 70% across patients and signal types, while oddball identity was decoded above chance in pooled analyses. Local field potential recordings also demonstrated oddball-related responses, including negative evoked-response deflections and increased gamma-band activity.
Among 43 oddball-selective units, oddball decoding accuracy increased over the approximately 10-minute experiment, while tone-identity encoding decreased. The inverse relationship suggested a redistribution of population-level representations from tone identity toward oddball context.
Additional analyses showed progressive divergence between standard-tone and oddball neural population vectors using both Euclidean distance and cosine-angle metrics. The researchers interpreted these findings as evidence that anesthesia did not abolish hippocampal representational plasticity.
To explore potential mechanisms underlying these findings, the investigators developed a recurrent neural network model trained on a similar tone-discrimination task. Although the model was trained only to distinguish tone identity, it also decoded oddball context above chance and reproduced the divergence patterns observed in the human recordings. The results suggest such representational divergence may emerge from local computations rather than requiring input from upstream networks. Lesioning analyses within the model further suggested inhibitory connections were important for encoding both tone identity and oddball context.
In four additional patients, the researchers recorded hippocampal activity while patients listened to 10 to 20 minutes of podcast audio during anesthesia. Neural activity was aligned to word onset and offset across 962 to 3,024 spoken words per patient.
Using semantic word embeddings and cross-validated linear models, the researchers found that neural activity encoded meaning carried by spoken words. The average correlation between predicted and observed firing rates was approximately 0.40 across 368 units. When analyses were restricted to unique words, about 75% of units still demonstrated statistically significant semantic generalization.
The researchers compared these findings with recordings from a separate cohort of awake patients undergoing epilepsy monitoring and reported comparable semantic-encoding performance between anesthetized and awake cohorts.
Nearly 86% of recorded units demonstrated selectivity for semantic categories, and 80% encoded part-of-speech information. Semantic-category decoding accuracy averaged approximately 60%, compared with 57% for part-of-speech classification.
Neural responses also corresponded to semantic information from preceding and upcoming words. In many unites, firing rates were modulated by surprisal, a measure reflecting the probability of a word based on preceding linguistic context. However, the researchers cautioned that decoding of future-word information “does not necessarily imply active prediction beyond contextualization.”
The study also identified semantic and syntactic encoding in multiple local field potential frequency bands.
The investigators emphasized several limitations. The study involved a small cohort of patients with epilepsy undergoing surgery, used propofol-based total intravenous anesthesia, and may not generalize to other anesthetics or nonconscious states such as sleep or coma. The researchers also noted that the observed processes may not be unique to the hippocampus and that the study could not determine whether changes in tone-identity encoding reflected compensatory plasticity or independent adaptive effects.
“These results highlight the robust coding of cognitive variables in the hippocampus in the absence of consciousness,” the researchers wrote. “Such task-correlated patterns of neural activity are typically thought of as neural correlates of their corresponding cognitive processes, so our observations raise the possibility that those processes may occur in the absence of consciousness.”
Disclosures: Sameer A. Sheth, MD, PhD, reported consulting relationships with Boston Scientific, Abbott, Koh Young, Neuropace, and Zimmer Biomet, and is co-founder of Motif Neurotech. The other researchers reported no competing interests. The study was funded in part by NIH grant U01 NS121472, the McNair Foundation, and the Gordon and Mary Cain Pediatric Neurology Research Foundation.
Source: Nature