Increased perceptual reliability reduces membrane potential variability in cortical neurons

Uncertainty is omnipresent. While humans and other animals take uncertainty into account during decision making, it remains unclear how it is represented in cortex. Recent theoretical work on uncertainty computation in cortical neurons predicts a stimulus-triggered decrease of the within-trial membrane potential variability. Yet, testing this prediction in experimental data is uniquely challenging as it would require a large number of intra-cellular recordings in-vivo. We thus leverage simulation-based inference to gain insights about the membrane potential statistics underlying single unit spiking activity. This allows us to investigate the effect of stimulus reliability on membrane potential variability in posterior parietal cortex (PPC) neurons while male mice performed a multisensory change detection task. The inferred membrane potential statistics show that neurons decrease their membrane potential variability in response to task relevant stimuli. In particular, more perceptually reliable stimuli lead to larger decreases in membrane potential variability than less reliable ones, in line with theoretical predictions. These findings suggest that cortical neurons track uncertainty, providing Bayesian benefits for downstream computations. Significance Statement The mechanisms underlying uncertainty representation in the brain remain elusive. Recent theoretical models predict signatures of Bayes-optimal computation at the level of neuronal membrane potentials. More specifically, more perceptually reliable inputs are expected to cause a decreases in membrane potential variability. Testing this hypothesis in experimental data requires a large number of intra-cellular recordings in-vivo which are challenging to obtain. We thus systematically infer the membrane potential statistics from single trial, single unit activity using simulation based inference. As predicted, we find that the within-trial membrane potential variability decreases with increasing perceptual reliability. These results suggest an intimate link between the quenching of within-trial response variability and Bayesian computation in the brain.