Serotonin regulates working memory function non-monotonically in a computational network model: implications for schizophrenia
Independently, both serotonin and working memory (WM) have been associated with the prefrontal cortex. On the one hand, serotonergic neurons in the raphe nucleus project directly to prefrontal cortex, where serotonergic receptors are particularly enriched. On the other hand, electrophysiological and neuroimaging studies associate prefrontal cortex activations to successful WM performance. Schizophrenia provides another link between serotonin and WM: many antipsychotics target dominantly serotonin receptors, and WM impairment is considered to be a core deficit in schizophrenia. However, a direct association between serotonin and WM has proved elusive in psycho-pharmacological studies. I will present a computational network model for spatial WM in the prefrontal cortex that reveals a direct relationship between serotonin and WM. We found that serotonin modulated the network's WM performance non-monotonously, following an inverted U-shape. This could partly explain that only weak behavioral effects of serotonin treatment were found in previous WM studies. Our simulations showed that WM errors committed with low and high tonic serotonin were due to different network dynamics instabilities, suggesting that these two conditions could be distinguished experimentally based on the response confidence declared in error trials, and on prefrontal activation contrasts in neuroimaging studies. Finally, we tested the antipsychotic effects of serotonin receptor manipulations in network models with synaptic imbalances suggested by the GABAergic and glutamatergic hypotheses for schizophrenia. We found that WM deficits were attenuated by acting on serotonergic receptors, but re-emerged in high dosage treatments. This non-monotonicity was again resolved in our model by separating WM errors based on the declared response confidence. Our study underscores the relevance of identifying different types of error trials in WM tasks in order to reveal the association between neuromodulatory systems and WM, and the benefits of serotonergic treatments for cognitive deficits in schizophrenia.