Dynamic Analysis of Hippocampal Information Encoding

Emery N. Brown

Department of Anesthesia and Critical Care

Massachusetts General Hospital

Harvard Medical School/ MIT Division of Health Science and Technology

Neural systems represent information about stimuli from the outside world in the stochastic structure of their firing patterns. Accurate characterization of this stochastic structure is crucial for deciphering how neural systems encode and transmit information. We use the question of spatial information encoding by ensemble firing patterns of hippocampal place cells recorded from rats freely foraging in an open environment to develop a statistical paradigm to study neural information representation. The research has three components. First, we are developing a class of statistical models, the inhomogeneous general inverse gaussian (IGIG) probability densities, to model the dependence of place cell firing on the position of the animal in the environment. This model class has the inhomogeneous Poisson, gamma and inverse Gaussian probability densities as special cases. Q-Q and K-S plot goodness-of-fit methods based on the time-rescaling theorem show that the IGIG model offers significant improvement relative to Poisson in describing the spiking activity of individual place cell neurons. Second, we are using neural spike train decoding algorithms based on nonlinear recursive moment estimation to quantify the extent to which the improved description of individual place cell spiking activity leads to a better description of the ensemble representation of place information. Third, the place receptive fields of hippocampal neurons are dynamic, i.e. the fields evolve over the course of an experiment even when the animal is in a familiar environment. We have developed an adaptive estimation algorithm based on instantaneous steepest descent to track in real-time the spatio-temporal dynamics of place field evolution. These methods are some of the statistical tools we are developing to help understand how hippocampal neurons encode spatial information in short-term memory.