Developing a neural prosthesis for hippocampus: proof-of-concept using the in vitro slice

Restricted until 5 Nov. 2009.We are developing a neural prosthesis that can replace regions of the hippocampal brain area that have been damaged by disease or insult. We used the hippocampal slice preparation as the first step in developing such a prosthesis. The major intrinsic circuitry of the hippocampus consists of an excitatory cascade involving the dentate gyrus (DG), CA3, and CA1 subregions; it can be maintained in a transverse slice preparation. Our goal is to develop a biologically realistic model to replace the CA3 region; as such the model transforms DG signals into the corresponding output signals, reinstating the activity in CA1 region. The path to proof our concept involves: i) developing a slice-prosthesis interface specifically for the trisynaptic pathway, ii) formulating nonlinear neuronal models of the hippocampal trans-synaptic transformation, and iii) designing experimental paradigms to validate our prosthesis model.; In this thesis, a conformal multi-electrode array that spatially matched with the trisynaptic pathway was fabricated and used as the slice-prosthesis interface. The recorded data were used to characterize the hippocampal nonlinear properties using Laguerre-Volterra kernel modeling approach. Two experimental paradigms were developed to validate our in vitro prosthesis model. The first paradigm explicitly replaced CA3 computational function. A CA3 model was built to predict CA3 output based on DG input, and was further implemented into a VLSI device. In the replacement experiment, DG signals transmitted to the VLSI-CA3 model. Output from the VLSI device, which represents the equivalent responses from the CA3, provided the activation of synaptic inputs to the CA1. Results show the propagation of patterns of activity from DG-VLSI-CA1 reproduces the activity observed in DG-CA3-CA1 circuitry.; The second experimental paradigm was developed to optimize our first prosthesis model; it directly controlled CA1 dynamics. A DG-CA1 trajectory model and a CA1 plant model were built, and an inverse CA1 plant model was formulated. The desired CA1 output was first predicted by the DG-CA1 trajectory model and then converted to the desired stimulation through an inverse CA1 plant model. The derived desired stimuli were used to evoke CA1 activity. Results show improved accuracy in CA1 responses using this modeling-control paradigm