Distinct neuronal populations in the human brain combine content and context
The medial temporal lobe, and particularly the hippocampus, has been proposed to encode items in context1,2. Although hippocampal memory representations are largely context-dependent in rodents3,4, concept cells in humans appear to be context-invariant5. However, it remains unknown how item and context information are combined to form or retrieve integrated item-in-context memories at the single-neuron level in humans. Here we show that coordinated activity of distinct neuronal populations supports item-in-context memory. In a context-dependent picture-comparison task, we recorded 3,109 neurons from 16 neurosurgical patients, identifying 597 stimulus-modulated (pre-screened) and 200 context-modulated neurons (2.95% in the amygdala, 7.68% in the parahippocampal cortex, 5.68% in the entorhinal cortex and 9.42% in the hippocampus). Their co-firing combined different comparison questions (contexts) with two subsequent pictures (stimuli) through neuronal reinstatement of question contexts. Both populations were largely separate, generalized across the preferred dimension of each other, and covaried with behavioural performance. Following experimental pairing of stimuli and context, firing of entorhinal stimulus neurons predicted that of hippocampal context neurons after tens of milliseconds. Overall, synaptic modifications and co-firing of stimulus and context neurons could contribute to item-in-context memory, specify which stimulus memories need to be retrieved, and even generalize memories through mutual reinstatement of largely separate, orthogonal representations. By contrast, only 50 stimulus–context neurons represented specific picture–question combinations, consistent with limited pattern separation in the human medial temporal lobe, favouring flexible generalization over rigid conjunctive coding.
