Semantic Tuning of Single Neurons in the Human Medial Temporal Lobe

The Medial Temporal Lobe (MTL) is key to human cognition, supporting memory, emotional processing, navigation, and semantic coding. Rare direct human MTL recordings revealed concept cells, which were proposed as episodic memory building blocks due to their context- and modality-invariant response. However, there is a long-standing debate whether concept cells encode information in a discrete, all-or-none fashion, or as a graded, continuous function of semantic dimensions. To resolve this, we developed a closed-loop protocol that analyzes neuronal spiking in real time and adaptively presents new stimuli based on semantic similarity to response-eliciting previous ones. We found that human concept cells show graded responses, following semantic tuning curves. Furthermore, the tuning width and steepness of MTL neurons vary, with the hippocampus exhibiting the steepest tuning. Regional analysis of these tuning curves yields region-specific categorical response profiles (e.g., food responses in the amygdala). Finally, we show that at the population level, MTL activity is best captured by continuous semantic vector models, which outperform categorical and network-based metrics. Together, these findings establish that concept cells exhibit semantic tuning curves, providing new mechanistic insights into the neural organization of semantic information in the human brain.