......................................................................

NJIT Mathematical Biology Seminar

Tuesday, March 22, 2011, 2:30pm
Cullimore Hall 611
New Jersey Institute of Technology

......................................................................

Thalamic Transformation of Retinal Input: Information Transfer and LGN Modeling

Alex Casti

Center for Molecular and Behavioral Neuroscience, Rutgers University

The lateral geniculate nucleus (LGN) is the thalamic conduit through which visual information from the retina is passed to the visual cortex. Single-cell electrical recordings in the LGN that capture simultaneously the retinal ganglion cell input to an LGN relay cell and its output spikes reveal that, typically, the LGN transmits less than half of the spikes it receives. Since each LGN relay cell is driven primarily by a single ganglion cell and is silent without retinal input, the deletion of spikes at the thalamic level presents an interesting question: how does the LGN communicate the incoming visual information with fewer spikes?

We attack this issue by exploring various coding strategies that the LGN might employ in its process of deciding which retinal spikes to transmit and which to reject. By comparing the actual LGN information transmitted (Shannon information) by the recorded cell with that of spike trains generated by artificial coding schemes, we gain insight into the patterns of retinal input that the thalamus deems worthy of passage to the cortex. In addition, we ask whether the LGN transmits information in an optimal manner, in the sense of transmitting the most information it can given its retinal input.

A related issue is the extent to which non-retinal inputs contribute to LGN spiking. Each thalamic cell is embedded in a complex circuit that includes the primary retinal drive as well as interneuronal inhibition, feedback from the adjacent thalamic reticular nucleus, feedback from primary visual cortex, and input from the brainstem. The importance of these "extra-retinal" inputs may be addressed implicitly by quantifying the predictability of LGN spike trains given knowledge of the retinal inputs only. We explore this with optimized Hodgkin-Huxley and generalized linear models (GLMs).