Department of Neuroscience, Physiology and Pharmacology, University College London, London,
The synapse is the fundamental element of neuronal communication. However, how sensory stimuli are representated at the level of single synapses, and how much information the activity at a single central synapse contains about a sensory stimulus has not been determined. Here we address these questions by taking advantage of cerebellar granule cells (GCs) as they provide an ideal model system for two reasons. First, they receive on average only 4 excitatory inputs from mossy fibres (MFs), making it possible to distinguish single inputs. Second, MF-GC synapses in the flocculus receive prominent vestibular input, which constitutes a well-defined sensory stimulus with a low dimensional stimulus space.
To characterise the representation of vestibular stimuli at the MF-GC synapse, we performed in vivo whole-cell voltage-clamp recordings at -70 mV from GCs in the cerebellar flocculus of ketamine/xylazine anaesthetised mice. In the absence of vestibular stimulation, vestibular-sensitive GCs showed spontaneously occuring excitatory postsynaptic currents (EPSCs) with an average frequency of 13.0±2.4 Hz. Vestibular stimulation resulted in a bidirectional modulation of EPSC frequency, which correlated linearly with the angular velocity, rather than position or acceleration. A lack of synaptic short-term dynamics over the observed frequencies ensured that the velocity representation was linear not only in terms of frequency, but also in terms of excitatory charge transfer at the GC membrane. In a subset of cells distinct inputs could be reliably distinguished based on their EPSC amplitude. In those cases, only a subset of inputs was modulated by vestibular stimulation while the remaining inputs where insensitve to horizontal rotation, suggesting that despite receiving very few inputs GCs can have multi-dimensional and/or multi-modal receptive fields. Using a Bayesian approach we examined the capacity of GC EPSC trains to report the stimulus velocity that evoked it, in order to estimate the size of the population of MF inputs/GCs to reach a behaviourally relevant resolution. While activity in a single GC could report the presence and direction of movement, the error in the stimulus reconstruction was large. However, with increasing GC number the velocity estimates improved in accuracy and reliability in a logarithmic fashion, and an ensemble of as few as 100 GCs, i.e. less than 400 synapses, provided an accuracy of 4.8∘/s, approaching the psychophysical limit. Thus the representation of angular velocity requires synaptic signals that are distributed over many tens of GCs but that within an individual cell may be integrated with other stimulus features.