Theodore W. Berger - US grants

The research supported by this award will focus on neural plasticity that develops in the hippocampus as a result of behavioral learning, using classical conditioning of eyeblink in rabbit as a model system. Three specific issues with respect to this plasticity will be investigated. First, we will determine the multi-synaptic anatomical pathways through which learning-induced changes in the activity of hippocampal pyramidal neurons affects the cerebellum--a brain structure known to be involved in the formation of the conditioned eyeblink response. Second, we will use nonlinear systems analytic techniques to characterize functional properties of the hippocampus expressed only at the network level, i.e., properties emerging from the coordinated activity of all its subpopulations of neurons acting as a system. We then will investigate how those system properties are altered during eyeblink conditioning. Finally, we will investigate the contribution of brainstem noradrenergic and serotonergic inputs to changes in pyramidal cell activity that develop during classical conditioning.

The goals of this proposal are to further develop and apply a combined theoretical and experimental approach, utilizing principles of nonlinear systems theory, for achieving a biologically-based model of the hippocampal formation. In this approach, the functional properties resulting from interaction among the elements of a neural network are quantitatively characterized as input/output functions, i.e., the kernels of a functional power series. The linear and higher-order nonlinear components of the input/output relationship are determined experimentally by stimulating afferents to the network with random inputs to generate a wide range of interactions among the network elements, simultaneously recording activity of the output neurons, and estimating the kernels using cross-correlation or other techniques. The hippocampal formation consists of five subsystems (entorhinal cortex, dentate gyrus, the CA3/4 and CA1/2 pyramidal cell regions of Ammons' horn, and the subicular cortex) interconnected through feedforward and feedback pathways. Studies will focus on entorhinal input to the dentate gyrus, and will extend to CA3/4. Entorhinal afferents to the dentate will be activated with a train of electrical impulses having randomly determined (Poisson) inter-impulse intervals; evoked responses will be recorded electrophysiologically from dentate granule cells. Cross-correlation and a novel Laguerre expansion techniques will be used to estimate the kernels. The identical procedures will be repeated for progressively simplified in vivo and in vitro preparations in which the dentate (and later, other subsystems) are isolated from the remaining network granule cells isolated from intrinsic interneurons, and ultimately, feedback mechanisms intrinsic to granule cells (e.g., voltage-dependent conductances) isolated from the synaptic currents generated in response to the randomized input. In this manner, the biological mechanisms responsible for the nonlinearities expressed by the intact system can be identified. Models of single cells and circuits characterized experimentally will be developed using multi-dimensional Laplace transforms, allowing a progressively more complex representation of the global hippocampal system as a composite of the input/output functions of its subsystems. Our ultimate objective is to utilize such a model is to identify the functional dynamics of the hippocampus expressed at a systems level, and to investigate the relationship between those dynamics and learning-related changes in hippocampal activity recorded in behaving animals and humans.

technology /technique development; psychology; nervous system; biomedical resource; bioengineering /biomedical engineering; model design /development; behavioral /social science research tag;

The second Advance Short Course, entitled "Modeling and Data Analysis in Pharmacokinetics and Pharmacodynamics Using ADAPT II", organized by Professor D'Argenio will be held on August 5-6, 1998 at the National Institutes of Health. Professor D'Argenio will be assisted by two NCI scientists experienced in using the ADAPT Package (Dr. Steve Piscitelli and Dr. William Figg). Over 20 NCI scientists are scheduled to attend this two-day meeting, with each participant accessing ADAPT through their individual notebook computers. The Short Course will focus on the use of the ADAPT software packages for modeling, simulation, parameter estimation, and design of experiments in pharmacokinetics and pharmacodynamics. Case studies will be presented illustrating the application of the ADAPT software for solving a variety of modeling, estimation, and experiment design problems. The Case Studies involves hands-on computer work and will cover the following topics: pharmacodynamic modeling; Bayesian estimation, maximum likelihood and generalized least squares estimation; estimation with multiple output models; sample schedule design; physiological model simulation; models with time delays; and modeling with covariates.

EIA-130898-John Granacki-University of Southern California- Architectures for Neuromimetic Information System

Biologically inspired computing modules performing spatio-temporal pattern recognition will be a key aspect of future computing systems. Current prototypes of biomimetic circuit models have demonstrated real-time performance at very low power levels and high physical density. Further, these simple structures are capable of recognizing signal patterns masked by significant noise levels. These properties are difficult to mimic in digital signal processing algorithms, which require substantial modification when adapting to new applications.

We propose to implement second-order dynamic synapse neural models with digital equivalents that use features extracted from software simulations. Arrays of these digital equivalents can be readily implemented in dense re-useable commercial field-programmable gate arrays (FPGAs), enabling very high-speed operation. Additional features, such as writeable registers for user tuning of dynamics, will accelerate the evaluation of new kernels as well as enable investigation of high-speed "super-time" learning.

EIA-0130883-Theodore W. Berger-University of Southern California-Title: Neurobiological Nonlinear Dynamics for Biomimetic Signal Processing-Title-The fundamental goal of the proposed research is to derived a new generation of temporal and spatio-temporal pattern recognition systems based on the nonlinear dynamics, network architecture, and synaptic plasticity properties of the hippocampus, a cortical brain system responsible for the formation of new pattern recognition memories. From a neurobiological perspective, the proposed experimental/modeling work promises to generate (1) first-characterizations of high-order nonlinearities of cortical brain tissue, i.e., predictive models of the input/output transformations in spatio-temporal activity performed by individual hippocampal neurons, and to (2) investigate the increasingly likely possibility of dynamic neural "learning rules", i.e., requisite conditions for the induction of synaptic plasticity that depend on the past history of activity. In addition, the proposed research will investigate (3) the role of known hippocampal network topology in neurobiological signal processing and hierarchical feature extraction. From a theoretical/computational perspective, the proposed work is designed to (4) develop novel methodologies essential for characterizing nonlinearities of neurobiological systems, as well as to (5) further expand a newly developed paradigm for biologically realistic neural system modeling (the "dynamic synapse neural network architecture") that has already demonstrated a heretofore unmatched capability for identifying optimal feature sets for temporally and spatio-temporally coded information.

DESCRIPTION (provided by applicant): The end goal of this multiscale modeling research is to bridge the gap existing between three-dimensional, full- wave, macro-modeling of electrical and magnetic biointeractions (global modeling) and cellular-level modeling strategies. Our research team is composed of engineers, neuroscientists, biophysicists, surgeons, and computer scientists that are experts in all computational and experimental aspects necessary to fill the existing gaps in multi-scale modeling. This new multi-university effort to predict spatio-temporal distributions of active neurons based on current densities created by multi-electrode electrical stimulation depends on having a set of core models of molecular (receptor-channel kinetics), synaptic, neuron, and multi-neuron activity. These models and their inputs and outputs must be integrated into a global model of the extracellular media/matrix including relevant multi-electrode arrays. Successful modeling at these levels will allow hypotheses about space-time patterns of electrical stimulation to produce predictions about the number and distribution of activated inputs (based on known spatial distributions of afferent axons). The linked molecular, synaptic, neuron, multi-neuron, and global model will provide the basis for emerging predictions of the spatio-temporal distribution of active neurons and thus, the spatio-temporal distributions of spike train activity that encode all information in the nervous system. Our research effort will capitalize on our accomplishments in the realm of retinal and cortical prostheses, and use these as test beds for the multiscale predictive modeling methods that we will develop within the proposed activity.

Biomedical engineering, particularly as it applies to neuroscience, has reached a stage of development at which further understanding of complex neural systems, such as the hippocampus and other cortical regions that underlie cognition and higher thought processes, will depend on mathematical modeling as a means to organize experimental data that is known, and to systematically explore the unknown. The research objectives of Core Project #4 are to further develop and apply methodologies based on principles of nonlinear systems theory for experimentally-based, mathematical modeling of neurons and neural systems. This approach leads to what are commonly termed non-parametric or input-output models, i.e., functional properties that emerge as a consequence of interactions among the internal components of the system - without necessarily describing the internal components themselves. In contrast, parametric models represent the mechanistic properties of the system, with parameters that can be interpreted directly with respect to those underlying mechanisms. We will explore parametric modeling of the hippocampus both in the context of a glutamatergic synaptic model (EONS) continued from previous work, and a new project: a large-scale, compartmental neuron model (10[6] neurons, 10[10] synapses) of hippocampus that incorporates much of the quantitative neuroanatomy, synaptic physiology, and topographic connectivity available for that structure. Ultimately our goal is to establish means for the synergistic use of non-parametric and parametric modeling methods, in the context of accelerating multi-scale modeling, to further our understanding of how global system dynamics underlying cognition, and specifically memory, derive from molecular and synaptic mechanisms.

Project Abstract The end goal of this multiscale modeling research is to bridge the gap existing between three-dimensional, full- wave, macro-modeling of electrical and magnetic biointeractions (global modeling) and cellular-level modeling strategies. Our research team is composed of engineers and neuroscientists that are experts in all computational and experimental aspects necessary to fill the existing gaps in multi-scale modeling. This multi-university effort to predict spatio-temporal distributions of active neurons based on current densities created by multi-electrode electrical stimulation depends on having a set of core models of molecular (receptor-channel kinetics), synaptic, neuron, and multi-neuron activity. These models and their inputs and outputs must be integrated into a global model of the extracellular media/matrix including relevant multi- electrode arrays. Successful modeling at these levels will allow hypotheses about space-time patterns of electrical stimulation to produce predictions about the number and distribution of activated inputs (based on known spatial distributions of afferent axons). The linked molecular, synaptic, neuron, multi-neuron, and global model will provide the basis for emerging predictions of the spatio-temporal distribution of active neurons and thus, the spatio-temporal distributions of spike train activity that encode all information in the nervous system. Further, we believe the proposed multiscale modeling framework constitutes an ideal platform capable of generating novel insights into the pathogenic mechanisms precipitating abnormal hippocampal function. Although the proposed research is focused on the hippocampal system, our effort will capitalize on our multiscale modeling accomplishments during the performance period of our original multiscale modeling U01 grant, in the realm of both retinal and cortical prostheses.