Rapid, automated and experimenter-free assessment of cognitive flexibility reveals learning impairments following recovery from activity-based anorexia in female rats

Anorexia nervosa (AN) has among the highest mortality rates of any psychiatric disorder and is characterised by cognitive inflexibility that persists after weight recovery and contributes to the low rates of recovery. What remains unknown is whether cognitive inflexibility predisposes individuals to AN, a question that is difficult to determine from human studies. Our previous work using the most well-established animal model of AN, known as activity-based anorexia (ABA) identified a neurobiological link between cognitive inflexibility and susceptibility to pathological weight loss in female rats. However, testing flexible learning prior to exposure to ABA in the same animals has been thus far impossible due to the length of training required and the necessity of daily handling, which can itself influence the development of ABA. Here we describe experiments that validate and optimise the first fully-automated and experimenter-free touchscreen cognitive testing system for rats (n=20) and use this novel system to examine the reciprocal links between reversal learning (an assay of cognitive flexibility) and weight loss in the ABA model (n=60). Firstly, we show substantially reduced testing time and increased throughput compared to conventional touchscreen testing methods because animals engage in test sessions at their own direction and can complete multiple sessions per day without experimenter involvement. We also show that, contrary to expectations, cognitive inflexibility does not predispose rats to pathological weight loss in ABA but instead that rats subsequently susceptible to weight loss performed better on the reversal learning task. Intriguingly, we show reciprocal links between ABA exposure and cognitive flexibility, with ABA exposed (but weight recovered) rats performing much worse that ABA naive rats on the reversal learning task. On the other hand, animals that had been trained on reversal learning were better able to resist weight loss upon subsequent exposure to the ABA model. We also uncovered some stable behavioural differences between ABA susceptible versus resistant rats during touchscreen test sessions using machine learning tools that highlight possible predictors of anorectic phenotypes. These findings shed new light on the relationship between cognitive inflexibility and pathological weight loss and provide a robust target for future studies using the ABA model to investigate potential novel pharmacotherapies for AN.