Computational Neuroscience in Epilepsy

Ivan Soltesz, Kevin Staley

• Publisher: Academic Press
• Year: 2011
• ISBN:
  9780080559537
• Topic: Sciences biologiques
• Subtopic: Biologie

Epilepsy is a neurological disorder that affects millions of patients worldwide and arises from the concurrent action of multiple pathophysiological processes. The power of mathematical analysis and computational modeling is increasingly utilized in basic and clinical epilepsy research to better understand the relative importance of the multi-faceted, seizure-related changes taking place in the brain during an epileptic seizure. This groundbreaking book is designed to synthesize the current ideas and future directions of the emerging discipline of computational epilepsy research. Chapters address relevant basic questions (e.g., neuronal gain control) as well as long-standing, critically important clinical challenges (e.g., seizure prediction). Computational Neuroscience in Epilepsy should be of high interest to a wide range of readers, including undergraduate and graduate students, postdoctoral fellows and faculty working in the fields of basic or clinical neuroscience, epilepsy research, computational modeling and bioengineering.

• Covers a wide range of topics from molecular to seizure predictions and brain implants to control seizures
• Contributors are top experts at the forefront of computational epilepsy research
• Chapter contents are highly relevant to both basic and clinical epilepsy researchers

Part I

Computational Modeling Techniques and Databases in Epilepsy Research

Outline

Chapter 1: SIMULATION OF LARGE NETWORKS: TECHNIQUE AND PROGRESS

Chapter 2: THE NEURON SIMULATION ENVIRONMENT IN EPILEPSY RESEARCH

Chapter 3: THE COCODAT DATABASE: SYSTEMATICALLY ORGANIZING AND SELECTING QUANTITATIVE DATA ON SINGLE NEURONS AND MICROCIRCUITRY

Chapter 4: VALIDATING MODELS OF EPILEPSY

Chapter 5: USING NEUROCONSTRUCT TO DEVELOP AND MODIFY BIOLOGICALLY DETAILED 3D NEURONAL NETWORK MODELS IN HEALTH AND DISEASE

Chapter 6: COMPUTATIONAL NEUROANATOMY OF THE RAT HIPPOCAMPUS: IMPLICATIONS AND APPLICATIONS TO EPILEPSY

1

SIMULATION OF LARGE NETWORKS: TECHNIQUE AND PROGRESS

WILLIAM W. LYTTON, MARK STEWART and MICHAEL HINES

ABSTRACT

Computer models of epilepsy and seizures require simulation of large networks in order to produce emergent phenomenology, but also require inclusion of details of cellular and synaptic physiology to retain connection with pharmacological effectors for therapeutic intervention. The former constraint suggests the use of simplifying approaches: integrate-and-fire models or mean-field approximations. The latter constraint pushes one towards multi-compartment models. Compromise is required. We have developed highly-simplified event-driven complex artificial-cell models that utilize a series of rules to emulate critical underlying phenomena likely important for seizure generation and patterning. These include depolarization blockade, voltage-dependent NMDA activation, afterhyperpolarization and several others. Using these units, we can readily run simulations of 1–2 million units on a single processor. Here, we present results of data-mining 138,240 simulations of a 3,500-unit network. Parameter explorations can then relate single-unit dynamical ‘structure’ to network function. We find that the networks are prone to latch-up, reminiscent of a seizure tonic phase. This can alternate with a slow repetitive clonic-like activity. Parameter dependence of epileptiform features was typically complex. For example, massive initial network response increased with greater unit excitability only in the context of particular AMPA and NMDA strength settings. This consequence of network complexity fits with the multi-site activity of endogenous neuroeffectors and may help explain why ‘dirty’ drugs, those acting at multiple sites, might be particularly effective for seizure control.

GOALS OF COMPUTER MODELING FOR CLINICAL DISEASE

We use the fruits of computer simulation every day when we consult a weather report. Although weather reporting accuracy remains low for periods of greater than 10 days from the present, the improvement over weather reporting of 50 years ago is striking. The big difference is size: both the size of the data set available for initial conditions and the size of the computers for running the massive simulations that are now used. In the next 50 years we will likely see comparable progress in our ability to:

1. predict seizures

2. understand the genesis of seizures based on alterations in the underlying neurobiological substrate

3. intervene in directed ways to prevent a single seizure or to alter the substrate and prevent seizures from occurring.

Unlike weather simulation, where the underlying factors involved (temperature, wind, humidity) are well characterized and where the interactions are fully understood, the nervous system remains full of under- or uncharacterized neurotransmitters, receptors, second and third messengers and electrical interactions. Also unlike weather, many locations where we would like to measure chemical concentrations or currents are inaccessible. For these reasons, accurate simulation will require growth not only in simulation but also in measuring capabilities.

The hope and expectation is that accurate simulation will lead to rational therapeutics, whether prophylactic or acute, whether pharmacological, electrical or surgical. Acute electrical intervention prior to a seizure is becoming a readily foreseeable scenario with the increasing use of implanted electrodes for other brain diseases. It seems likely that practical efforts in this direction will result from direct analysis of brain signals. Here, modeling will assist by permitting rapid assessment of many possible stimulation protocols. With regard to prophylaxis, genetic analysis and molecular biology can identify an underlying substrate that predisposes to seizures but cannot trace the dynamical path whereby an abnormality results in a seizure. Characterization of this dynamical pathway, the chain of causality, can identify critical loci where interventions are most practical.

Clinically useful seizure simulations require that connection be made both with clinical observables – the seizure and its electroencephalographic manifestation – and with therapeutic approaches – whether pharmacological or surgical. For this reason, we and others have largely focused on detailed compartmental models which provide explicit modeling of dendritic regions, of voltage-sensitive ion-channel populations and of various synaptic receptor subtypes. These receptors and channels are the targets of pharmacotherapeutic manipulations and can be comparably manipulated in these models. However, simplified alternatives to compartmental models are proving useful for investigation as well.

DETAILED VERSUS SIMPLIFYING MODELING

There is a tension between the desire to model as accurately as possible and the desire to simplify, both due to practical limitations of computer size and due to the need to isolate the critical factors and associations that lead to understanding. In the neural modeling literature this tension arises where the study of dynamical systems meets biology.

The most common approach to mathematical modeling of neural systems has involved explicit modeling of individual neurons. However, there is also a long tradition of mean field approaches, utilizing the insights of statistical mechanics. Here, the individual neurons are assumed to be indistinguishable particles whose interactions produce bulk properties, just as the interactions of molecules in a liquid produce the bulk properties familiar from thermodynamics (Wilson and Cowan, 1972). These approaches have lately been applied to epilepsy as well (Wilson et al., 2006).

Moving up toward slightly greater complexity, many followers of the simplifying ethic utilize leaky integrate-and-fire models. These models forego all biological details except for a threshold and a leaky membrane. When the threshold is reached, the cell discharges. The simulated membrane allows inputs to summate while the leakiness permits membrane potential gradually to return to rest after a subthreshold input. At the synapse level, these simulations may utilize instantaneous or brief depolarizing or hyperpolarizing synapses to drive the cell.

N. Brunel and colleagues have been pioneers in studying the dynamics of large networks of leaky integrate-and-fire cells, also using mean field methods to enhance understanding of observed activity patterns (Brunel, 2000; Brunel and Wang, 2003). They identified several different characteristic patterns of firing: asynchronous irregular, asynchronous regular, synchronous irregular and synchronous regular, demonstrating the parameter associations for each of these regimes. In general, firing patterns depended on the strength of network driving and the strength of excitatory connections within the network. They demonstrated that low inhibition states gave fast firing with some coordination and that higher levels of inhibition produced irregular firing that, nonetheless, coordinated in global population oscillations that would produce a high amplitude field potential.

Leaping from these highly simplified models to the other end of the complexity spectrum brings us to the models of Traub and associates (Traub and Wong, 1982; Traub et al., 2005). They pioneered simulations of large networks with detailed compa...

Table of contents

• Cover image
• Title page
• Table of Contents
• CONTRIBUTORS
• FOREWORD: RISE OF THE MACHINES – ON THE THRESHOLD OF A NEW ERA IN EPILEPSY RESEARCH
• INTRODUCTION: APPLICATIONS AND EMERGING CONCEPTS OF COMPUTATIONAL NEUROSCIENCE IN EPILEPSY RESEARCH
• Part I: Computational Modeling Techniques and Databases in Epilepsy Research
• Part II: Epilepsy and Altered Network Topology
• Part III: Destabilization of Neural Networks
• Part IV: Homeostasis and Epilepsy
• Part V: Mechanisms of Synchronization
• Part VI: Interictal to Ictal Transitions
• Part VII: Seizure Dynamics
• Part VIII: Towards Computer-Aided Therapy
• INDEX