Our Research Approach
Our goal is to discover general principles of the organization and operation of neural systems. Our approach is to study the eye movements of rhesus monkeys so that we can take advantage of the ease of controlling, measuring, and quantifying eye movement behavior while we monitor the activity of one or more neurons in the circuit that generates the movements. We study smooth pursuit eye movements because they comprise a motor system with the same cortical and sub-cortical architecture as other sensory-motor systems. Thus, we can learn how the brain transforms sensory inputs into coordinated movements, through analysis of how it moves the eyes.
The basic circuit for pursuit is known. Sensory inputs arise from the middle temporal visual area (MT). The “motor cortex” for pursuit is the smooth eye movement region of the frontal eye fields (FEFsem). Other well studied areas include the floccular complex of the cerebellum, the pontine nuclei that connect cortex to cerebellum, and the final brainstem motor circuits.
The multiple parallel pathways in the pursuit system decode the representation of visual motion in area MT and transform it into commands for eye velocity. We think of sensory-motor decoding as having two major components. One converts the population response in area MT into estimates of target speed and direction. The other, through FEFsem, uses the population response in MT to estimate motion reliability and generates a signal that modulates the strength of visual-motor transmission. Downstream, neural integration converts a transient visual motion signal into a sustained eye velocity command to allow perfect tracking during steady-state pursuit. We think that these simple building blocks work together to allow pursuit to be based on a reliability-weighted combination of sensory data and priors based on past experience and expectations, that is “Bayesian inference”.
Current efforts include expanding our knowledge of the physiology of the pursuit circuit through new recordings with multi-contact probes in the basal ganglia, cerebellum, and cerebral cortex. We also are embarking on a large-scale computer simulation that will use a biologically-motivated circuit model and operate dynamically on a millisecond time scale. We will determine the local and long-range neural computations that allow model neurons to mimic the first- and second-order statistics of the responses of real neurons.
Our research on motor learning in pursuit eye movements has led to a theory of the principles of operation of a learning neural circuit. (i) Early, fast learning occurs in the cerebellar cortex based on depression of synapses from parallel fibers to Purkinje cells guided by errors signaled on climbing fiber inputs. (ii) Later, slow learning occurs in the cerebellar nucleus, transferred from and instructed by the learned changes in simple-spike firing of Purkinje cells. (iii) Feedback from the cerebellar nucleus to the inferior olive reduces climbing fiber inputs and limits the amount of learning in the cerebellar cortex. Current research uses multi-contract probes to record from all cell types in the cerebellar cortex and map more precisely the multiple sites and mechanisms of learning. A large-scale computer simulation will create a model of the cerebellar circuit that learns autonomously. We also will explore long-term learning in the FEFsem.
We rely on exquisite control over a very precise behavior, quantitative and accurate measurements of eye movements, and recordings from isolated neurons throughout the pursuit circuit. Increasingly, our recordings use multi-contact probes that allow us to analyze local connectivity and record from multiple neuron types. We analyze our data quantitatively, frequently using methods that are derived from theoretical approaches. I am open to the possibility of deploying optogenetic approaches strategically in non-human primates.