83.7/LL16

Monitoring plasticity through changes in functional connectivity

San Diego Convention Center: Halls B-H

Saturday, Nov 03, 2007, 3:00 PM - 4:00 PM

*J. M. REBESCO¹, I. STEVENSON¹, S. A. SOLLA^1,2, L. E. MILLER^1,3;
¹Dept. of Physiology, Northwestern Univ., Chicago, IL; ²Dept. of Physics and Astronomy, ³Dept. of Biomed. Engin., Northwestern Univ., Evanston, IL

Bidirectional brain-machine interfaces offer the opportunity to directly monitor information processing within small networks of neurons and to characterize the effective functional connectivity of the circuits involved in the mapping from afferent to efferent signals. To determine the functional connectivity in these networks one needs both stable recordings and stable measures of functional connectivity. Once in place, these tools can also be used to monitor changes in synaptic strength induced through learning. Here we present experimental and computational approaches to these issues.
To track recorded neurons across time, we monitor both the shape of the waveforms (Bar-Hillel, Spiro, and Stark 2006) and the spiking statistics of individual neurons. Datasets of population activity ranging from one to four hours in length are subdivided into 10 minute epochs. The data are assumed to be stationary within each epoch, and mixture-of-Gaussians models are used to cluster and sort the recorded spikes based on waveform shape. A Bayesian network is then used to attribute clusters from different epochs to a common neural source. Interspike interval distributions computed within each epoch are parameterized and compared. Estimates of consistent waveform and spiking statistics are then combined to track neurons across epochs.
Given stable recordings, numerous methods exist in the literature for assessing the functional connectivity of small networks of neurons. Among these, maximum-likelihood estimates over stochastic point processes have been used extensively. These models rely on a concave likelihood function with a unique global minimum. However, in practical applications the likelihood function is usually quite flat around the global minimum. Thus, small statistical fluctuations in the input data, even from a stationary source, can cause large fluctuations in the inferred connectivity. This problem becomes acute in the case of interest to us: that of characterizing the effective connectivity among a relatively small number of observed neurons embedded in a large network of unobserved neurons. We simulate this scenario through a network of O(10000) neurons, of which only O(10) are observable. In order to overcome the connectivity ambiguity, we combine successive connectivity estimations with a prior over previously estimated model parameters. This novel approach allows us to generate stable models and to correctly detect changes in connectivity resulting from induced plastic changes in the simulated network. We are currently applying these techniques to chronic, multielectrode recordings from the sensorimotor cortex of rats.

 J.M. Rebesco, None; I. Stevenson, None; S.A. Solla, None; L.E. Miller, None.

NINDS NS048845

NSF-IGERT DGE-9987577