Deciphering the logic of representational drift and its conversion to stable readouts.

We perceive the world around us as constant. When we engage in the same action over days and weeks, for instance when we walk down a familiar road, it feels the same every day. Neurons build a model (or representation) of the external world that is thought to guide behaviour. However, neural representations change over time, leading to a restructured model of the world. This phenomenon, called representational drift, has been identified across a range of brain regions and behaviours, and poses a conceptual challenge: How can changing neural correlates of external variables be reconciled with perceptual stability? This proposal aims to decipher this discrepancy. We recently showed that the medial prefrontal cortex (mPFC) displays little representational drift when mice are engaged in an olfaction-guided memory task. Interestingly, upstream sensory regions relaying olfactory information to the mPFC show prominent drift. These results suggest that drifting codes early in signal processing are converted to a more stable code in the mPFC. Using the primary olfactory cortex to mPFC signal chain as model system, DRIFT-rules aims to systematically map out the extent of representational drift in the mPFC and its upstream partners under conditions of varying computational demands for the mPFC circuitry. Using 1- and 2-photon calcium imaging and chronic electrophysiological recordings with Neuropixels probes in awake mice, we will determine drift during different behaviours, in which the animals rely on olfactory cues to solve distinct cognitive problems. Combining large-scale neuronal activity measurements with novel computational techniques, we will reveal how mechanisms on the level of single cells, microcircuits and neuronal populations contribute to the conversion of drifting to stable codes. This research program will redefine our understanding of how consistent perception and cognitive behaviour are possible in the context of drifting neuronal representations.