David DiGregorio, PhD
Professor
University of Colorado Medicine Endowed Chair

Synaptic basis of behavior: A paramount challenge of Neuroscience research in the 21st century is to understand how the cells in the brain (neurons) use their specialized contacts (synapses) to route and transform information to perceive the world around us and, in turn, drive behaviors. One fascinating function of the nervous system is its ability to keep track of time. Sensations, thoughts, and actions are dynamic events that require the brain to encode the passage of time. For many tasks, such as playing music or sports, accurate execution requires the precise estimation of time intervals in the range of milliseconds to seconds. But how neuronal elements within brain circuits represent “time” is not understood. Synaptic connections between neurons change their strength dynamically during brief bouts of activity. We hypothesize that they could act like “cellular timers” and thus be a substrate for encoding time within neural networks to generate precise behaviors.

A specialized brain region, the cerebellum, learns precise temporal details of our internal and external sensory world to fine-tune motor and cognitive behaviors. Indeed, deficits of cerebellar function could account for altered sensory responses in schizophrenia or autism. Fortunately, the cerebellar circuit architecture is relatively simple and has only a handful of well-defined neuron types. This makes it uniquely tractable to establish the role of each neuron type and its synaptic connections in generating precisely timed actions.

Hypothesis: The laboratory of Synapse and Circuit Dynamics (SCD) has made seminal discoveries about the various functions of synapses in the cerebellum and the molecular organization within nerve terminals driving this diversity. Subsequently, we developed a mathematical model that predicts how synaptic diversity is a substrate for circuit computations underlying animal behavior (Figure 1, Barri et al. 2022). The principal hypothesis is that dynamic changes in synaptic strength are necessary to generate a distributed representation of time, which can be used as a mathematical basis to learn arbitrary patterns of output activity. This distributed representation of time could enable the cerebellum to precisely time actions.

Approach: The SCD laboratory has implemented a multi-scale and multidisciplinary research program that links macromolecular organization at synapses to neural circuit function that drives well-timed behaviors. Projects in the laboratory include microscopy development, patch-clamp, and two-photon dynamic imaging in acute brain slices, super-resolution imaging of synaptic macromolecular complexes, high-speed random access 2-photon imaging of neuronal population activity, and single-unit recordings using high-density Neuropixels probes in awake behaving mice. Statistical and numerical methods are used to compare experimental results to mathematically formalized hypotheses.

Figure 1. Simulating PC pauses during Eye-lid conditioning (Barri et al. 2023). a) Scheme of eyelid conditioning. CS: conditioned stimulus (red). US: unconditioned stimulus (violet). After experiencing CS and US delivered at a fixed temporal contingency over many trials, the animal learns to close its eyelid before the US is delivered (green). A trough in PC activity(blue) precedes the eyelid closure (target time, grey dashed line). b) Scheme of cerebellar cortex rate model. MFs are classed according to synapse types from Chabrol et al. 2015. Percentages indicate relative frequency of MF groups. Insets: firing rate distributions for different MF groups. c) Representation of the CS by an instantaneous change in MF firing rate. Top: 100 MF sorted according to synapse type. MF firing rates are color-coded and drawn according to the specific firing rate distributions. Bottom: two sample MF rates per synaptic group. d) Model GC responses to the CS. Top: 1000 GC sorted according to average firing rate after CS onset. Firing rates are color coded. Bottom: steady-state subtracted and individually normalized GC transient responses. e) Example of eyelid learning over 4000 trials for a 200 ms delay. The dashed line represents the target signal used in the supervised learning procedure. Without STP-induced GC transients, no PC trough can be learned (pink line). f) Eyelid learning for different target times. Different colors indicate PC responses after 4000 learning steps of different simulations and corresponding target times (dashed lines).