ADAPTIVE ENCODING OF CONTEXTUAL VARIABLES BY SINGLE NEURONS IN THE HUMAN TEMPORAL LOBE

The Medial temporal lobe (MTL) is critical for memory and for linking perception with contextual and emotional information. Neurons in the human MTL were shown to encode concepts while context appears to be represented through population-level representational geometries. However, it remains unclear how MTL neurons encode dynamic contextual features, and whether such encoding depends on emotional context.
We analysed single-unit activity from 500 MTL neurons recorded in 19 epilepsy patients undergoing presurgical monitoring. Participants viewed a video with alternating emotionally charged face clips and neutral landscape scenes, where three visual features varied dynamically: trial type (faces vs. landscapes), size of the main object, and its movement speed. Single-neuron selectivity was characterized using generalized linear model (GLM) predicting firing rates from visual features, combined with a stringent model selection procedure. Population-level representations were examined using GLM-based decoding to predict visual features from neuronal activity, and model weights were interpreted to compare contributions across MTL subregions.
At the single-neuron level, largest fraction of neurons encoded trial type (14%), while smaller percentages encoded object size (5%) and movement speed (6%). At the population level, visual features were robustly decoded from neuronal activity, with higher decoding accuracy during emotional trials, suggesting enhanced encoding of behaviourally relevant information. While all MTL subregions contributed to feature processing, amygdala neurons showed stronger differentiation between emotional faces versus landscape trials, while hippocampus and entorhinal cortex showed stronger selectivity for movement speed. Overall, these results demonstrate distributed and context-dependent encoding of dynamic visual information in the human MTL.