Event:

Understanding how complex information is represented and accessed in working memory remains a central challenge in cognitive neuroscience. While classical models emphasize a serial scanning of the elements held in memory, much less is known about how the strength of individual memory representations contributes to behaviour.

We recorded single-neuron activity from the medial temporal lobe of epilepsy patients performing a verbal working memory task with letter strings. Using population-level decoding, we show that ensembles of hippocampal neurons represent the identity of individual letters held in working memory. Different from most human single neuron studies, where participants memorize visual stimuli, our task activates the phonological loop where participants memorize phonemes.

At the behavioural level, response time increases with set size, consistent with a serial memory scanning process. In addition, we identified a complementary mechanism: letters with higher decoding accuracy are associated with faster response times. Using a trial-by-trial linear mixed-effects model, we find that decoding accuracy correlates with representation strength and predicts response time. We thereby link population-level neural representations in the human hippocampus to behaviour and show how neural representation quality modulates working memory performance. This suggests that these neurons organise a space for phonemes, which are at the basis of language and central to human social interaction.