Structured sampling of molecularly classified mossy fiber inputs by cerebellar granule cells

The cerebellar granule cell layer receives mossy fiber inputs from diverse brain regions, yet the principles governing how individual granule cells sample distinct types of inputs remain poorly understood. Using a volumetric correlated light and electron microscopy (vCLEM) dataset from an adult female mouse cerebellum, in which VGluT1-positive and VGluT1-negative mossy fiber terminals are molecularly distinguished, we reconstructed granule cell and mossy fiber connectivity to examine input selection rules. To test whether connectivity can be explained by spatial proximity alone, we developed distance-limited random sampling null models based on empirical cell spatial arrangement, simulating adult and developmental sampling regimes. Granule cell-centered analyses showed that granule cells shared less innervation from the same mossy fiber than expected by chance, indicating that structured sampling cannot be explained by distance-constrained random connectivity alone. Moreover, subpopulations of granule cells preferentially sample either VGluT1-positive or VGluT1-negative mossy fibers. In contrast, mossy fiber–centered analysis showed that individual terminals distributed their outputs across granule cells in a pattern broadly consistent with random sampling. However, sampling in the adult model was more selective than in the model that reflects developmental processes. Together, our findings demonstrated structured, non-random sampling of cerebellar VGluT1-positive and VGluT1-negative mossy fiber inputs and provide insight into how granule cells integrate molecularly distinct inputs to support cerebellar computation.