CEREBELLAR EXPANSION RECODING DURING A MOUSE FORELIMB REACHING TASK

The Marr–Albus framework proposes that the cerebellar input layer transforms overlapping mossy fibre (MF) inputs into sparser, decorrelated and higher-dimensional granule cell representations, supporting pattern separation for sensorimotor learning. However, direct tests of how MF population activity is transformed in granule cells during behaviour, remain limited.

To address this, we simultaneously recorded population activity in MFs in the granular layer and parallel fibres (PFs) in the molecular layer of cerebellar lobule 4/5, using 3D acousto-optic lens two-photon microscopy with real-time motion correction. Head-fixed mice performed a forelimb reaching task to either a single, or three locations, for a water reward.

We are currently investigating expansion recoding by comparing the lifetime and population sparseness, pairwise correlations as well as cross-validated dimensionality of the GCaMP8m activity in the MF and PF populations during the reaching task. To understand how population activity is related to behaviour, we quantified the forelimb trajectory with high speed video recordings and deep learning methods, and applied linear decoding.

Our preliminary results are consistent with some aspects of Marr-Albus theory such as sparsification and decorrelation. Moreover, we could decode the reach target using either the MF or PF activity as predictors. We are currently exploring how the embedding dimensionality of the MF and PF activity spaces differ during distinct movements.