Mouse behaviour on the trial-unique non-matching-to-location (TUNL) touchscreen task reflects a mixture of distinct working memory codes and response biases

Abstract

The trial-unique non-matching to location (TUNL) touchscreen task shows promise as a translational assay of working memory deficits in disorders including autism, ADHD, and schizophrenia. Although it is commonly assumed that the TUNL task predominantly measures spatial working memory in rodents, this proposition has not previously been tested. In this project, we used computational modelling of behaviour from mice performing the TUNL task (total N = 163 mice across three datasets; 158,843 total trials). Contrary to common assumptions, behaviour on the TUNL task did not exclusively reflect spatial working memory. Instead, choice behaviour was explained as a mixture of both retrospective (spatial) working memory and prospective working memory for an intended behavioural response, as well as animal-specific response biases. We suggest that these findings can be understood within a resource-rational cognitive framework, and propose several task-design principles that we predict will maximise spatial working memory and minimise alternative behavioural strategies.